1 @c Copyright (C) 1988,89,92,93,94,96,97,1998 Free Software Foundation, Inc.
2 @c This is part of the GCC manual.
3 @c For copying conditions, see the file gcc.texi.
6 @chapter Target Description Macros
7 @cindex machine description macros
8 @cindex target description macros
9 @cindex macros, target description
10 @cindex @file{tm.h} macros
12 In addition to the file @file{@var{machine}.md}, a machine description
13 includes a C header file conventionally given the name
14 @file{@var{machine}.h}. This header file defines numerous macros
15 that convey the information about the target machine that does not fit
16 into the scheme of the @file{.md} file. The file @file{tm.h} should be
17 a link to @file{@var{machine}.h}. The header file @file{config.h}
18 includes @file{tm.h} and most compiler source files include
22 * Driver:: Controlling how the driver runs the compilation passes.
23 * Run-time Target:: Defining @samp{-m} options like @samp{-m68000} and @samp{-m68020}.
24 * Storage Layout:: Defining sizes and alignments of data.
25 * Type Layout:: Defining sizes and properties of basic user data types.
26 * Registers:: Naming and describing the hardware registers.
27 * Register Classes:: Defining the classes of hardware registers.
28 * Stack and Calling:: Defining which way the stack grows and by how much.
29 * Varargs:: Defining the varargs macros.
30 * Trampolines:: Code set up at run time to enter a nested function.
31 * Library Calls:: Controlling how library routines are implicitly called.
32 * Addressing Modes:: Defining addressing modes valid for memory operands.
33 * Condition Code:: Defining how insns update the condition code.
34 * Costs:: Defining relative costs of different operations.
35 * Sections:: Dividing storage into text, data, and other sections.
36 * PIC:: Macros for position independent code.
37 * Assembler Format:: Defining how to write insns and pseudo-ops to output.
38 * Debugging Info:: Defining the format of debugging output.
39 * Cross-compilation:: Handling floating point for cross-compilers.
40 * Misc:: Everything else.
44 @section Controlling the Compilation Driver, @file{gcc}
46 @cindex controlling the compilation driver
48 @c prevent bad page break with this line
49 You can control the compilation driver.
52 @findex SWITCH_TAKES_ARG
53 @item SWITCH_TAKES_ARG (@var{char})
54 A C expression which determines whether the option @samp{-@var{char}}
55 takes arguments. The value should be the number of arguments that
56 option takes--zero, for many options.
58 By default, this macro is defined as
59 @code{DEFAULT_SWITCH_TAKES_ARG}, which handles the standard options
60 properly. You need not define @code{SWITCH_TAKES_ARG} unless you
61 wish to add additional options which take arguments. Any redefinition
62 should call @code{DEFAULT_SWITCH_TAKES_ARG} and then check for
65 @findex WORD_SWITCH_TAKES_ARG
66 @item WORD_SWITCH_TAKES_ARG (@var{name})
67 A C expression which determines whether the option @samp{-@var{name}}
68 takes arguments. The value should be the number of arguments that
69 option takes--zero, for many options. This macro rather than
70 @code{SWITCH_TAKES_ARG} is used for multi-character option names.
72 By default, this macro is defined as
73 @code{DEFAULT_WORD_SWITCH_TAKES_ARG}, which handles the standard options
74 properly. You need not define @code{WORD_SWITCH_TAKES_ARG} unless you
75 wish to add additional options which take arguments. Any redefinition
76 should call @code{DEFAULT_WORD_SWITCH_TAKES_ARG} and then check for
79 @findex SWITCH_CURTAILS_COMPILATION
80 @item SWITCH_CURTAILS_COMPILATION (@var{char})
81 A C expression which determines whether the option @samp{-@var{char}}
82 stops compilation before the generation of an executable. The value is
83 boolean, non-zero if the option does stop an executable from being
84 generated, zero otherwise.
86 By default, this macro is defined as
87 @code{DEFAULT_SWITCH_CURTAILS_COMPILATION}, which handles the standard
88 options properly. You need not define
89 @code{SWITCH_CURTAILS_COMPILATION} unless you wish to add additional
90 options which affect the generation of an executable. Any redefinition
91 should call @code{DEFAULT_SWITCH_CURTAILS_COMPILATION} and then check
92 for additional options.
94 @findex SWITCHES_NEED_SPACES
95 @item SWITCHES_NEED_SPACES
96 A string-valued C expression which enumerates the options for which
97 the linker needs a space between the option and its argument.
99 If this macro is not defined, the default value is @code{""}.
103 A C string constant that tells the GNU CC driver program options to
104 pass to CPP. It can also specify how to translate options you
105 give to GNU CC into options for GNU CC to pass to the CPP.
107 Do not define this macro if it does not need to do anything.
109 @findex NO_BUILTIN_SIZE_TYPE
110 @item NO_BUILTIN_SIZE_TYPE
111 If this macro is defined, the preprocessor will not define the builtin macro
112 @code{__SIZE_TYPE__}. The macro @code{__SIZE_TYPE__} must then be defined
113 by @code{CPP_SPEC} instead.
115 This should be defined if @code{SIZE_TYPE} depends on target dependent flags
116 which are not accessible to the preprocessor. Otherwise, it should not
119 @findex NO_BUILTIN_PTRDIFF_TYPE
120 @item NO_BUILTIN_PTRDIFF_TYPE
121 If this macro is defined, the preprocessor will not define the builtin macro
122 @code{__PTRDIFF_TYPE__}. The macro @code{__PTRDIFF_TYPE__} must then be
123 defined by @code{CPP_SPEC} instead.
125 This should be defined if @code{PTRDIFF_TYPE} depends on target dependent flags
126 which are not accessible to the preprocessor. Otherwise, it should not
129 @findex SIGNED_CHAR_SPEC
130 @item SIGNED_CHAR_SPEC
131 A C string constant that tells the GNU CC driver program options to
132 pass to CPP. By default, this macro is defined to pass the option
133 @samp{-D__CHAR_UNSIGNED__} to CPP if @code{char} will be treated as
134 @code{unsigned char} by @code{cc1}.
136 Do not define this macro unless you need to override the default
141 A C string constant that tells the GNU CC driver program options to
142 pass to @code{cc1}. It can also specify how to translate options you
143 give to GNU CC into options for GNU CC to pass to the @code{cc1}.
145 Do not define this macro if it does not need to do anything.
149 A C string constant that tells the GNU CC driver program options to
150 pass to @code{cc1plus}. It can also specify how to translate options you
151 give to GNU CC into options for GNU CC to pass to the @code{cc1plus}.
153 Do not define this macro if it does not need to do anything.
157 A C string constant that tells the GNU CC driver program options to
158 pass to the assembler. It can also specify how to translate options
159 you give to GNU CC into options for GNU CC to pass to the assembler.
160 See the file @file{sun3.h} for an example of this.
162 Do not define this macro if it does not need to do anything.
164 @findex ASM_FINAL_SPEC
166 A C string constant that tells the GNU CC driver program how to
167 run any programs which cleanup after the normal assembler.
168 Normally, this is not needed. See the file @file{mips.h} for
171 Do not define this macro if it does not need to do anything.
175 A C string constant that tells the GNU CC driver program options to
176 pass to the linker. It can also specify how to translate options you
177 give to GNU CC into options for GNU CC to pass to the linker.
179 Do not define this macro if it does not need to do anything.
183 Another C string constant used much like @code{LINK_SPEC}. The difference
184 between the two is that @code{LIB_SPEC} is used at the end of the
185 command given to the linker.
187 If this macro is not defined, a default is provided that
188 loads the standard C library from the usual place. See @file{gcc.c}.
192 Another C string constant that tells the GNU CC driver program
193 how and when to place a reference to @file{libgcc.a} into the
194 linker command line. This constant is placed both before and after
195 the value of @code{LIB_SPEC}.
197 If this macro is not defined, the GNU CC driver provides a default that
198 passes the string @samp{-lgcc} to the linker unless the @samp{-shared}
201 @findex STARTFILE_SPEC
203 Another C string constant used much like @code{LINK_SPEC}. The
204 difference between the two is that @code{STARTFILE_SPEC} is used at
205 the very beginning of the command given to the linker.
207 If this macro is not defined, a default is provided that loads the
208 standard C startup file from the usual place. See @file{gcc.c}.
212 Another C string constant used much like @code{LINK_SPEC}. The
213 difference between the two is that @code{ENDFILE_SPEC} is used at
214 the very end of the command given to the linker.
216 Do not define this macro if it does not need to do anything.
220 Define this macro to provide additional specifications to put in the
221 @file{specs} file that can be used in various specifications like
224 The definition should be an initializer for an array of structures,
225 containing a string constant, that defines the specification name, and a
226 string constant that provides the specification.
228 Do not define this macro if it does not need to do anything.
230 @code{EXTRA_SPECS} is useful when an architecture contains several
231 related targets, which have various @code{..._SPECS} which are similar
232 to each other, and the maintainer would like one central place to keep
235 For example, the PowerPC System V.4 targets use @code{EXTRA_SPECS} to
236 define either @code{_CALL_SYSV} when the System V calling sequence is
237 used or @code{_CALL_AIX} when the older AIX-based calling sequence is
240 The @file{config/rs6000/rs6000.h} target file defines:
243 #define EXTRA_SPECS \
244 @{ "cpp_sysv_default", CPP_SYSV_DEFAULT @},
246 #define CPP_SYS_DEFAULT ""
249 The @file{config/rs6000/sysv.h} target file defines:
253 "%@{posix: -D_POSIX_SOURCE @} \
254 %@{mcall-sysv: -D_CALL_SYSV @} %@{mcall-aix: -D_CALL_AIX @} \
255 %@{!mcall-sysv: %@{!mcall-aix: %(cpp_sysv_default) @}@} \
256 %@{msoft-float: -D_SOFT_FLOAT@} %@{mcpu=403: -D_SOFT_FLOAT@}"
258 #undef CPP_SYSV_DEFAULT
259 #define CPP_SYSV_DEFAULT "-D_CALL_SYSV"
262 while the @file{config/rs6000/eabiaix.h} target file defines
263 @code{CPP_SYSV_DEFAULT} as:
266 #undef CPP_SYSV_DEFAULT
267 #define CPP_SYSV_DEFAULT "-D_CALL_AIX"
270 @findex LINK_LIBGCC_SPECIAL
271 @item LINK_LIBGCC_SPECIAL
272 Define this macro if the driver program should find the library
273 @file{libgcc.a} itself and should not pass @samp{-L} options to the
274 linker. If you do not define this macro, the driver program will pass
275 the argument @samp{-lgcc} to tell the linker to do the search and will
276 pass @samp{-L} options to it.
278 @findex LINK_LIBGCC_SPECIAL_1
279 @item LINK_LIBGCC_SPECIAL_1
280 Define this macro if the driver program should find the library
281 @file{libgcc.a}. If you do not define this macro, the driver program will pass
282 the argument @samp{-lgcc} to tell the linker to do the search.
283 This macro is similar to @code{LINK_LIBGCC_SPECIAL}, except that it does
284 not affect @samp{-L} options.
286 @findex MULTILIB_DEFAULTS
287 @item MULTILIB_DEFAULTS
288 Define this macro as a C expression for the initializer of an array of
289 string to tell the driver program which options are defaults for this
290 target and thus do not need to be handled specially when using
291 @code{MULTILIB_OPTIONS}.
293 Do not define this macro if @code{MULTILIB_OPTIONS} is not defined in
294 the target makefile fragment or if none of the options listed in
295 @code{MULTILIB_OPTIONS} are set by default.
296 @xref{Target Fragment}.
298 @findex RELATIVE_PREFIX_NOT_LINKDIR
299 @item RELATIVE_PREFIX_NOT_LINKDIR
300 Define this macro to tell @code{gcc} that it should only translate
301 a @samp{-B} prefix into a @samp{-L} linker option if the prefix
302 indicates an absolute file name.
304 @findex STANDARD_EXEC_PREFIX
305 @item STANDARD_EXEC_PREFIX
306 Define this macro as a C string constant if you wish to override the
307 standard choice of @file{/usr/local/lib/gcc-lib/} as the default prefix to
308 try when searching for the executable files of the compiler.
310 @findex MD_EXEC_PREFIX
312 If defined, this macro is an additional prefix to try after
313 @code{STANDARD_EXEC_PREFIX}. @code{MD_EXEC_PREFIX} is not searched
314 when the @samp{-b} option is used, or the compiler is built as a cross
317 @findex STANDARD_STARTFILE_PREFIX
318 @item STANDARD_STARTFILE_PREFIX
319 Define this macro as a C string constant if you wish to override the
320 standard choice of @file{/usr/local/lib/} as the default prefix to
321 try when searching for startup files such as @file{crt0.o}.
323 @findex MD_STARTFILE_PREFIX
324 @item MD_STARTFILE_PREFIX
325 If defined, this macro supplies an additional prefix to try after the
326 standard prefixes. @code{MD_EXEC_PREFIX} is not searched when the
327 @samp{-b} option is used, or when the compiler is built as a cross
330 @findex MD_STARTFILE_PREFIX_1
331 @item MD_STARTFILE_PREFIX_1
332 If defined, this macro supplies yet another prefix to try after the
333 standard prefixes. It is not searched when the @samp{-b} option is
334 used, or when the compiler is built as a cross compiler.
336 @findex INIT_ENVIRONMENT
337 @item INIT_ENVIRONMENT
338 Define this macro as a C string constant if you wish to set environment
339 variables for programs called by the driver, such as the assembler and
340 loader. The driver passes the value of this macro to @code{putenv} to
341 initialize the necessary environment variables.
343 @findex LOCAL_INCLUDE_DIR
344 @item LOCAL_INCLUDE_DIR
345 Define this macro as a C string constant if you wish to override the
346 standard choice of @file{/usr/local/include} as the default prefix to
347 try when searching for local header files. @code{LOCAL_INCLUDE_DIR}
348 comes before @code{SYSTEM_INCLUDE_DIR} in the search order.
350 Cross compilers do not use this macro and do not search either
351 @file{/usr/local/include} or its replacement.
353 @findex SYSTEM_INCLUDE_DIR
354 @item SYSTEM_INCLUDE_DIR
355 Define this macro as a C string constant if you wish to specify a
356 system-specific directory to search for header files before the standard
357 directory. @code{SYSTEM_INCLUDE_DIR} comes before
358 @code{STANDARD_INCLUDE_DIR} in the search order.
360 Cross compilers do not use this macro and do not search the directory
363 @findex STANDARD_INCLUDE_DIR
364 @item STANDARD_INCLUDE_DIR
365 Define this macro as a C string constant if you wish to override the
366 standard choice of @file{/usr/include} as the default prefix to
367 try when searching for header files.
369 Cross compilers do not use this macro and do not search either
370 @file{/usr/include} or its replacement.
372 @findex STANDARD_INCLUDE_COMPONENT
373 @item STANDARD_INCLUDE_COMPONENT
374 The ``component'' corresponding to @code{STANDARD_INCLUDE_DIR}.
375 See @code{INCLUDE_DEFAULTS}, below, for the description of components.
376 If you do not define this macro, no component is used.
378 @findex INCLUDE_DEFAULTS
379 @item INCLUDE_DEFAULTS
380 Define this macro if you wish to override the entire default search path
381 for include files. For a native compiler, the default search path
382 usually consists of @code{GCC_INCLUDE_DIR}, @code{LOCAL_INCLUDE_DIR},
383 @code{SYSTEM_INCLUDE_DIR}, @code{GPLUSPLUS_INCLUDE_DIR}, and
384 @code{STANDARD_INCLUDE_DIR}. In addition, @code{GPLUSPLUS_INCLUDE_DIR}
385 and @code{GCC_INCLUDE_DIR} are defined automatically by @file{Makefile},
386 and specify private search areas for GCC. The directory
387 @code{GPLUSPLUS_INCLUDE_DIR} is used only for C++ programs.
389 The definition should be an initializer for an array of structures.
390 Each array element should have four elements: the directory name (a
391 string constant), the component name, and flag for C++-only directories,
392 and a flag showing that the includes in the directory don't need to be
393 wrapped in @code{extern @samp{C}} when compiling C++. Mark the end of
394 the array with a null element.
396 The component name denotes what GNU package the include file is part of,
397 if any, in all upper-case letters. For example, it might be @samp{GCC}
398 or @samp{BINUTILS}. If the package is part of the a vendor-supplied
399 operating system, code the component name as @samp{0}.
402 For example, here is the definition used for VAX/VMS:
405 #define INCLUDE_DEFAULTS \
407 @{ "GNU_GXX_INCLUDE:", "G++", 1, 1@}, \
408 @{ "GNU_CC_INCLUDE:", "GCC", 0, 0@}, \
409 @{ "SYS$SYSROOT:[SYSLIB.]", 0, 0, 0@}, \
416 Here is the order of prefixes tried for exec files:
420 Any prefixes specified by the user with @samp{-B}.
423 The environment variable @code{GCC_EXEC_PREFIX}, if any.
426 The directories specified by the environment variable @code{COMPILER_PATH}.
429 The macro @code{STANDARD_EXEC_PREFIX}.
432 @file{/usr/lib/gcc/}.
435 The macro @code{MD_EXEC_PREFIX}, if any.
438 Here is the order of prefixes tried for startfiles:
442 Any prefixes specified by the user with @samp{-B}.
445 The environment variable @code{GCC_EXEC_PREFIX}, if any.
448 The directories specified by the environment variable @code{LIBRARY_PATH}
449 (native only, cross compilers do not use this).
452 The macro @code{STANDARD_EXEC_PREFIX}.
455 @file{/usr/lib/gcc/}.
458 The macro @code{MD_EXEC_PREFIX}, if any.
461 The macro @code{MD_STARTFILE_PREFIX}, if any.
464 The macro @code{STANDARD_STARTFILE_PREFIX}.
473 @node Run-time Target
474 @section Run-time Target Specification
475 @cindex run-time target specification
476 @cindex predefined macros
477 @cindex target specifications
479 @c prevent bad page break with this line
480 Here are run-time target specifications.
483 @findex CPP_PREDEFINES
485 Define this to be a string constant containing @samp{-D} options to
486 define the predefined macros that identify this machine and system.
487 These macros will be predefined unless the @samp{-ansi} option is
490 In addition, a parallel set of macros are predefined, whose names are
491 made by appending @samp{__} at the beginning and at the end. These
492 @samp{__} macros are permitted by the ANSI standard, so they are
493 predefined regardless of whether @samp{-ansi} is specified.
495 For example, on the Sun, one can use the following value:
498 "-Dmc68000 -Dsun -Dunix"
501 The result is to define the macros @code{__mc68000__}, @code{__sun__}
502 and @code{__unix__} unconditionally, and the macros @code{mc68000},
503 @code{sun} and @code{unix} provided @samp{-ansi} is not specified.
505 @findex extern int target_flags
506 @item extern int target_flags;
507 This declaration should be present.
509 @cindex optional hardware or system features
510 @cindex features, optional, in system conventions
512 This series of macros is to allow compiler command arguments to
513 enable or disable the use of optional features of the target machine.
514 For example, one machine description serves both the 68000 and
515 the 68020; a command argument tells the compiler whether it should
516 use 68020-only instructions or not. This command argument works
517 by means of a macro @code{TARGET_68020} that tests a bit in
520 Define a macro @code{TARGET_@var{featurename}} for each such option.
521 Its definition should test a bit in @code{target_flags}; for example:
524 #define TARGET_68020 (target_flags & 1)
527 One place where these macros are used is in the condition-expressions
528 of instruction patterns. Note how @code{TARGET_68020} appears
529 frequently in the 68000 machine description file, @file{m68k.md}.
530 Another place they are used is in the definitions of the other
531 macros in the @file{@var{machine}.h} file.
533 @findex TARGET_SWITCHES
534 @item TARGET_SWITCHES
535 This macro defines names of command options to set and clear
536 bits in @code{target_flags}. Its definition is an initializer
537 with a subgrouping for each command option.
539 Each subgrouping contains a string constant, that defines the option
540 name, and a number, which contains the bits to set in
541 @code{target_flags}. A negative number says to clear bits instead;
542 the negative of the number is which bits to clear. The actual option
543 name is made by appending @samp{-m} to the specified name.
545 One of the subgroupings should have a null string. The number in
546 this grouping is the default value for @code{target_flags}. Any
547 target options act starting with that value.
549 Here is an example which defines @samp{-m68000} and @samp{-m68020}
550 with opposite meanings, and picks the latter as the default:
553 #define TARGET_SWITCHES \
554 @{ @{ "68020", 1@}, \
559 @findex TARGET_OPTIONS
561 This macro is similar to @code{TARGET_SWITCHES} but defines names of command
562 options that have values. Its definition is an initializer with a
563 subgrouping for each command option.
565 Each subgrouping contains a string constant, that defines the fixed part
566 of the option name, and the address of a variable. The variable, type
567 @code{char *}, is set to the variable part of the given option if the fixed
568 part matches. The actual option name is made by appending @samp{-m} to the
571 Here is an example which defines @samp{-mshort-data-@var{number}}. If the
572 given option is @samp{-mshort-data-512}, the variable @code{m88k_short_data}
573 will be set to the string @code{"512"}.
576 extern char *m88k_short_data;
577 #define TARGET_OPTIONS \
578 @{ @{ "short-data-", &m88k_short_data @} @}
581 @findex TARGET_VERSION
583 This macro is a C statement to print on @code{stderr} a string
584 describing the particular machine description choice. Every machine
585 description should define @code{TARGET_VERSION}. For example:
589 #define TARGET_VERSION \
590 fprintf (stderr, " (68k, Motorola syntax)");
592 #define TARGET_VERSION \
593 fprintf (stderr, " (68k, MIT syntax)");
597 @findex OVERRIDE_OPTIONS
598 @item OVERRIDE_OPTIONS
599 Sometimes certain combinations of command options do not make sense on
600 a particular target machine. You can define a macro
601 @code{OVERRIDE_OPTIONS} to take account of this. This macro, if
602 defined, is executed once just after all the command options have been
605 Don't use this macro to turn on various extra optimizations for
606 @samp{-O}. That is what @code{OPTIMIZATION_OPTIONS} is for.
608 @findex OPTIMIZATION_OPTIONS
609 @item OPTIMIZATION_OPTIONS (@var{level}, @var{size})
610 Some machines may desire to change what optimizations are performed for
611 various optimization levels. This macro, if defined, is executed once
612 just after the optimization level is determined and before the remainder
613 of the command options have been parsed. Values set in this macro are
614 used as the default values for the other command line options.
616 @var{level} is the optimization level specified; 2 if @samp{-O2} is
617 specified, 1 if @samp{-O} is specified, and 0 if neither is specified.
619 @var{size} is non-zero if @samp{-Os} is specified and zero otherwise.
621 You should not use this macro to change options that are not
622 machine-specific. These should uniformly selected by the same
623 optimization level on all supported machines. Use this macro to enable
624 machine-specific optimizations.
626 @strong{Do not examine @code{write_symbols} in
627 this macro!} The debugging options are not supposed to alter the
630 @findex CAN_DEBUG_WITHOUT_FP
631 @item CAN_DEBUG_WITHOUT_FP
632 Define this macro if debugging can be performed even without a frame
633 pointer. If this macro is defined, GNU CC will turn on the
634 @samp{-fomit-frame-pointer} option whenever @samp{-O} is specified.
638 @section Storage Layout
639 @cindex storage layout
641 Note that the definitions of the macros in this table which are sizes or
642 alignments measured in bits do not need to be constant. They can be C
643 expressions that refer to static variables, such as the @code{target_flags}.
644 @xref{Run-time Target}.
647 @findex BITS_BIG_ENDIAN
648 @item BITS_BIG_ENDIAN
649 Define this macro to have the value 1 if the most significant bit in a
650 byte has the lowest number; otherwise define it to have the value zero.
651 This means that bit-field instructions count from the most significant
652 bit. If the machine has no bit-field instructions, then this must still
653 be defined, but it doesn't matter which value it is defined to. This
654 macro need not be a constant.
656 This macro does not affect the way structure fields are packed into
657 bytes or words; that is controlled by @code{BYTES_BIG_ENDIAN}.
659 @findex BYTES_BIG_ENDIAN
660 @item BYTES_BIG_ENDIAN
661 Define this macro to have the value 1 if the most significant byte in a
662 word has the lowest number. This macro need not be a constant.
664 @findex WORDS_BIG_ENDIAN
665 @item WORDS_BIG_ENDIAN
666 Define this macro to have the value 1 if, in a multiword object, the
667 most significant word has the lowest number. This applies to both
668 memory locations and registers; GNU CC fundamentally assumes that the
669 order of words in memory is the same as the order in registers. This
670 macro need not be a constant.
672 @findex LIBGCC2_WORDS_BIG_ENDIAN
673 @item LIBGCC2_WORDS_BIG_ENDIAN
674 Define this macro if WORDS_BIG_ENDIAN is not constant. This must be a
675 constant value with the same meaning as WORDS_BIG_ENDIAN, which will be
676 used only when compiling libgcc2.c. Typically the value will be set
677 based on preprocessor defines.
679 @findex FLOAT_WORDS_BIG_ENDIAN
680 @item FLOAT_WORDS_BIG_ENDIAN
681 Define this macro to have the value 1 if @code{DFmode}, @code{XFmode} or
682 @code{TFmode} floating point numbers are stored in memory with the word
683 containing the sign bit at the lowest address; otherwise define it to
684 have the value 0. This macro need not be a constant.
686 You need not define this macro if the ordering is the same as for
689 @findex BITS_PER_UNIT
691 Define this macro to be the number of bits in an addressable storage
692 unit (byte); normally 8.
694 @findex BITS_PER_WORD
696 Number of bits in a word; normally 32.
698 @findex MAX_BITS_PER_WORD
699 @item MAX_BITS_PER_WORD
700 Maximum number of bits in a word. If this is undefined, the default is
701 @code{BITS_PER_WORD}. Otherwise, it is the constant value that is the
702 largest value that @code{BITS_PER_WORD} can have at run-time.
704 @findex UNITS_PER_WORD
706 Number of storage units in a word; normally 4.
708 @findex MIN_UNITS_PER_WORD
709 @item MIN_UNITS_PER_WORD
710 Minimum number of units in a word. If this is undefined, the default is
711 @code{UNITS_PER_WORD}. Otherwise, it is the constant value that is the
712 smallest value that @code{UNITS_PER_WORD} can have at run-time.
716 Width of a pointer, in bits. You must specify a value no wider than the
717 width of @code{Pmode}. If it is not equal to the width of @code{Pmode},
718 you must define @code{POINTERS_EXTEND_UNSIGNED}.
720 @findex POINTERS_EXTEND_UNSIGNED
721 @item POINTERS_EXTEND_UNSIGNED
722 A C expression whose value is nonzero if pointers that need to be
723 extended from being @code{POINTER_SIZE} bits wide to @code{Pmode} are to
724 be zero-extended and zero if they are to be sign-extended.
726 You need not define this macro if the @code{POINTER_SIZE} is equal
727 to the width of @code{Pmode}.
730 @item PROMOTE_MODE (@var{m}, @var{unsignedp}, @var{type})
731 A macro to update @var{m} and @var{unsignedp} when an object whose type
732 is @var{type} and which has the specified mode and signedness is to be
733 stored in a register. This macro is only called when @var{type} is a
736 On most RISC machines, which only have operations that operate on a full
737 register, define this macro to set @var{m} to @code{word_mode} if
738 @var{m} is an integer mode narrower than @code{BITS_PER_WORD}. In most
739 cases, only integer modes should be widened because wider-precision
740 floating-point operations are usually more expensive than their narrower
743 For most machines, the macro definition does not change @var{unsignedp}.
744 However, some machines, have instructions that preferentially handle
745 either signed or unsigned quantities of certain modes. For example, on
746 the DEC Alpha, 32-bit loads from memory and 32-bit add instructions
747 sign-extend the result to 64 bits. On such machines, set
748 @var{unsignedp} according to which kind of extension is more efficient.
750 Do not define this macro if it would never modify @var{m}.
752 @findex PROMOTE_FUNCTION_ARGS
753 @item PROMOTE_FUNCTION_ARGS
754 Define this macro if the promotion described by @code{PROMOTE_MODE}
755 should also be done for outgoing function arguments.
757 @findex PROMOTE_FUNCTION_RETURN
758 @item PROMOTE_FUNCTION_RETURN
759 Define this macro if the promotion described by @code{PROMOTE_MODE}
760 should also be done for the return value of functions.
762 If this macro is defined, @code{FUNCTION_VALUE} must perform the same
763 promotions done by @code{PROMOTE_MODE}.
765 @findex PROMOTE_FOR_CALL_ONLY
766 @item PROMOTE_FOR_CALL_ONLY
767 Define this macro if the promotion described by @code{PROMOTE_MODE}
768 should @emph{only} be performed for outgoing function arguments or
769 function return values, as specified by @code{PROMOTE_FUNCTION_ARGS}
770 and @code{PROMOTE_FUNCTION_RETURN}, respectively.
772 @findex PARM_BOUNDARY
774 Normal alignment required for function parameters on the stack, in
775 bits. All stack parameters receive at least this much alignment
776 regardless of data type. On most machines, this is the same as the
779 @findex STACK_BOUNDARY
781 Define this macro if you wish to preserve a certain alignment for
782 the stack pointer. The definition is a C expression
783 for the desired alignment (measured in bits).
785 @cindex @code{PUSH_ROUNDING}, interaction with @code{STACK_BOUNDARY}
786 If @code{PUSH_ROUNDING} is not defined, the stack will always be aligned
787 to the specified boundary. If @code{PUSH_ROUNDING} is defined and specifies a
788 less strict alignment than @code{STACK_BOUNDARY}, the stack may be
789 momentarily unaligned while pushing arguments.
791 @findex FUNCTION_BOUNDARY
792 @item FUNCTION_BOUNDARY
793 Alignment required for a function entry point, in bits.
795 @findex BIGGEST_ALIGNMENT
796 @item BIGGEST_ALIGNMENT
797 Biggest alignment that any data type can require on this machine, in bits.
799 @findex MINIMUM_ATOMIC_ALIGNMENT
800 @item MINIMUM_ATOMIC_ALIGNMENT
801 If defined, the smallest alignment, in bits, that can be given to an
802 object that can be referenced in one operation, without disturbing any
803 nearby object. Normally, this is @code{BITS_PER_UNIT}, but may be larger
804 on machines that don't have byte or half-word store operations.
806 @findex BIGGEST_FIELD_ALIGNMENT
807 @item BIGGEST_FIELD_ALIGNMENT
808 Biggest alignment that any structure field can require on this machine,
809 in bits. If defined, this overrides @code{BIGGEST_ALIGNMENT} for
810 structure fields only.
812 @findex ADJUST_FIELD_ALIGN
813 @item ADJUST_FIELD_ALIGN (@var{field}, @var{computed})
814 An expression for the alignment of a structure field @var{field} if the
815 alignment computed in the usual way is @var{computed}. GNU CC uses
816 this value instead of the value in @code{BIGGEST_ALIGNMENT} or
817 @code{BIGGEST_FIELD_ALIGNMENT}, if defined, for structure fields only.
819 @findex MAX_OFILE_ALIGNMENT
820 @item MAX_OFILE_ALIGNMENT
821 Biggest alignment supported by the object file format of this machine.
822 Use this macro to limit the alignment which can be specified using the
823 @code{__attribute__ ((aligned (@var{n})))} construct. If not defined,
824 the default value is @code{BIGGEST_ALIGNMENT}.
826 @findex DATA_ALIGNMENT
827 @item DATA_ALIGNMENT (@var{type}, @var{basic-align})
828 If defined, a C expression to compute the alignment for a variables in
829 the static store. @var{type} is the data type, and @var{basic-align} is
830 the alignment that the object would ordinarily have. The value of this
831 macro is used instead of that alignment to align the object.
833 If this macro is not defined, then @var{basic-align} is used.
836 One use of this macro is to increase alignment of medium-size data to
837 make it all fit in fewer cache lines. Another is to cause character
838 arrays to be word-aligned so that @code{strcpy} calls that copy
839 constants to character arrays can be done inline.
841 @findex CONSTANT_ALIGNMENT
842 @item CONSTANT_ALIGNMENT (@var{constant}, @var{basic-align})
843 If defined, a C expression to compute the alignment given to a constant
844 that is being placed in memory. @var{constant} is the constant and
845 @var{basic-align} is the alignment that the object would ordinarily
846 have. The value of this macro is used instead of that alignment to
849 If this macro is not defined, then @var{basic-align} is used.
851 The typical use of this macro is to increase alignment for string
852 constants to be word aligned so that @code{strcpy} calls that copy
853 constants can be done inline.
855 @findex EMPTY_FIELD_BOUNDARY
856 @item EMPTY_FIELD_BOUNDARY
857 Alignment in bits to be given to a structure bit field that follows an
858 empty field such as @code{int : 0;}.
860 Note that @code{PCC_BITFIELD_TYPE_MATTERS} also affects the alignment
861 that results from an empty field.
863 @findex STRUCTURE_SIZE_BOUNDARY
864 @item STRUCTURE_SIZE_BOUNDARY
865 Number of bits which any structure or union's size must be a multiple of.
866 Each structure or union's size is rounded up to a multiple of this.
868 If you do not define this macro, the default is the same as
869 @code{BITS_PER_UNIT}.
871 @findex STRICT_ALIGNMENT
872 @item STRICT_ALIGNMENT
873 Define this macro to be the value 1 if instructions will fail to work
874 if given data not on the nominal alignment. If instructions will merely
875 go slower in that case, define this macro as 0.
877 @findex PCC_BITFIELD_TYPE_MATTERS
878 @item PCC_BITFIELD_TYPE_MATTERS
879 Define this if you wish to imitate the way many other C compilers handle
880 alignment of bitfields and the structures that contain them.
882 The behavior is that the type written for a bitfield (@code{int},
883 @code{short}, or other integer type) imposes an alignment for the
884 entire structure, as if the structure really did contain an ordinary
885 field of that type. In addition, the bitfield is placed within the
886 structure so that it would fit within such a field, not crossing a
889 Thus, on most machines, a bitfield whose type is written as @code{int}
890 would not cross a four-byte boundary, and would force four-byte
891 alignment for the whole structure. (The alignment used may not be four
892 bytes; it is controlled by the other alignment parameters.)
894 If the macro is defined, its definition should be a C expression;
895 a nonzero value for the expression enables this behavior.
897 Note that if this macro is not defined, or its value is zero, some
898 bitfields may cross more than one alignment boundary. The compiler can
899 support such references if there are @samp{insv}, @samp{extv}, and
900 @samp{extzv} insns that can directly reference memory.
902 The other known way of making bitfields work is to define
903 @code{STRUCTURE_SIZE_BOUNDARY} as large as @code{BIGGEST_ALIGNMENT}.
904 Then every structure can be accessed with fullwords.
906 Unless the machine has bitfield instructions or you define
907 @code{STRUCTURE_SIZE_BOUNDARY} that way, you must define
908 @code{PCC_BITFIELD_TYPE_MATTERS} to have a nonzero value.
910 If your aim is to make GNU CC use the same conventions for laying out
911 bitfields as are used by another compiler, here is how to investigate
912 what the other compiler does. Compile and run this program:
931 printf ("Size of foo1 is %d\n",
932 sizeof (struct foo1));
933 printf ("Size of foo2 is %d\n",
934 sizeof (struct foo2));
939 If this prints 2 and 5, then the compiler's behavior is what you would
940 get from @code{PCC_BITFIELD_TYPE_MATTERS}.
942 @findex BITFIELD_NBYTES_LIMITED
943 @item BITFIELD_NBYTES_LIMITED
944 Like PCC_BITFIELD_TYPE_MATTERS except that its effect is limited to
945 aligning a bitfield within the structure.
947 @findex ROUND_TYPE_SIZE
948 @item ROUND_TYPE_SIZE (@var{struct}, @var{size}, @var{align})
949 Define this macro as an expression for the overall size of a structure
950 (given by @var{struct} as a tree node) when the size computed from the
951 fields is @var{size} and the alignment is @var{align}.
953 The default is to round @var{size} up to a multiple of @var{align}.
955 @findex ROUND_TYPE_ALIGN
956 @item ROUND_TYPE_ALIGN (@var{struct}, @var{computed}, @var{specified})
957 Define this macro as an expression for the alignment of a structure
958 (given by @var{struct} as a tree node) if the alignment computed in the
959 usual way is @var{computed} and the alignment explicitly specified was
962 The default is to use @var{specified} if it is larger; otherwise, use
963 the smaller of @var{computed} and @code{BIGGEST_ALIGNMENT}
965 @findex MAX_FIXED_MODE_SIZE
966 @item MAX_FIXED_MODE_SIZE
967 An integer expression for the size in bits of the largest integer
968 machine mode that should actually be used. All integer machine modes of
969 this size or smaller can be used for structures and unions with the
970 appropriate sizes. If this macro is undefined, @code{GET_MODE_BITSIZE
971 (DImode)} is assumed.
973 @findex CHECK_FLOAT_VALUE
974 @item CHECK_FLOAT_VALUE (@var{mode}, @var{value}, @var{overflow})
975 A C statement to validate the value @var{value} (of type
976 @code{double}) for mode @var{mode}. This means that you check whether
977 @var{value} fits within the possible range of values for mode
978 @var{mode} on this target machine. The mode @var{mode} is always
979 a mode of class @code{MODE_FLOAT}. @var{overflow} is nonzero if
980 the value is already known to be out of range.
982 If @var{value} is not valid or if @var{overflow} is nonzero, you should
983 set @var{overflow} to 1 and then assign some valid value to @var{value}.
984 Allowing an invalid value to go through the compiler can produce
985 incorrect assembler code which may even cause Unix assemblers to crash.
987 This macro need not be defined if there is no work for it to do.
989 @findex TARGET_FLOAT_FORMAT
990 @item TARGET_FLOAT_FORMAT
991 A code distinguishing the floating point format of the target machine.
992 There are three defined values:
995 @findex IEEE_FLOAT_FORMAT
996 @item IEEE_FLOAT_FORMAT
997 This code indicates IEEE floating point. It is the default; there is no
998 need to define this macro when the format is IEEE.
1000 @findex VAX_FLOAT_FORMAT
1001 @item VAX_FLOAT_FORMAT
1002 This code indicates the peculiar format used on the Vax.
1004 @findex UNKNOWN_FLOAT_FORMAT
1005 @item UNKNOWN_FLOAT_FORMAT
1006 This code indicates any other format.
1009 The value of this macro is compared with @code{HOST_FLOAT_FORMAT}
1010 (@pxref{Config}) to determine whether the target machine has the same
1011 format as the host machine. If any other formats are actually in use on
1012 supported machines, new codes should be defined for them.
1014 The ordering of the component words of floating point values stored in
1015 memory is controlled by @code{FLOAT_WORDS_BIG_ENDIAN} for the target
1016 machine and @code{HOST_FLOAT_WORDS_BIG_ENDIAN} for the host.
1018 @findex DEFAULT_VTABLE_THUNKS
1019 @item DEFAULT_VTABLE_THUNKS
1020 GNU CC supports two ways of implementing C++ vtables: traditional or with
1021 so-called ``thunks''. The flag @samp{-fvtable-thunk} chooses between them.
1022 Define this macro to be a C expression for the default value of that flag.
1023 If @code{DEFAULT_VTABLE_THUNKS} is 0, GNU CC uses the traditional
1024 implementation by default. The ``thunk'' implementation is more efficient
1025 (especially if you have provided an implementation of
1026 @code{ASM_OUTPUT_MI_THUNK}, see @ref{Function Entry}), but is not binary
1027 compatible with code compiled using the traditional implementation.
1028 If you are writing a new ports, define @code{DEFAULT_VTABLE_THUNKS} to 1.
1030 If you do not define this macro, the default for @samp{-fvtable-thunk} is 0.
1034 @section Layout of Source Language Data Types
1036 These macros define the sizes and other characteristics of the standard
1037 basic data types used in programs being compiled. Unlike the macros in
1038 the previous section, these apply to specific features of C and related
1039 languages, rather than to fundamental aspects of storage layout.
1042 @findex INT_TYPE_SIZE
1044 A C expression for the size in bits of the type @code{int} on the
1045 target machine. If you don't define this, the default is one word.
1047 @findex MAX_INT_TYPE_SIZE
1048 @item MAX_INT_TYPE_SIZE
1049 Maximum number for the size in bits of the type @code{int} on the target
1050 machine. If this is undefined, the default is @code{INT_TYPE_SIZE}.
1051 Otherwise, it is the constant value that is the largest value that
1052 @code{INT_TYPE_SIZE} can have at run-time. This is used in @code{cpp}.
1054 @findex SHORT_TYPE_SIZE
1055 @item SHORT_TYPE_SIZE
1056 A C expression for the size in bits of the type @code{short} on the
1057 target machine. If you don't define this, the default is half a word.
1058 (If this would be less than one storage unit, it is rounded up to one
1061 @findex LONG_TYPE_SIZE
1062 @item LONG_TYPE_SIZE
1063 A C expression for the size in bits of the type @code{long} on the
1064 target machine. If you don't define this, the default is one word.
1066 @findex MAX_LONG_TYPE_SIZE
1067 @item MAX_LONG_TYPE_SIZE
1068 Maximum number for the size in bits of the type @code{long} on the
1069 target machine. If this is undefined, the default is
1070 @code{LONG_TYPE_SIZE}. Otherwise, it is the constant value that is the
1071 largest value that @code{LONG_TYPE_SIZE} can have at run-time. This is
1074 @findex LONG_LONG_TYPE_SIZE
1075 @item LONG_LONG_TYPE_SIZE
1076 A C expression for the size in bits of the type @code{long long} on the
1077 target machine. If you don't define this, the default is two
1078 words. If you want to support GNU Ada on your machine, the value of
1079 macro must be at least 64.
1081 @findex CHAR_TYPE_SIZE
1082 @item CHAR_TYPE_SIZE
1083 A C expression for the size in bits of the type @code{char} on the
1084 target machine. If you don't define this, the default is one quarter
1085 of a word. (If this would be less than one storage unit, it is rounded up
1088 @findex MAX_CHAR_TYPE_SIZE
1089 @item MAX_CHAR_TYPE_SIZE
1090 Maximum number for the size in bits of the type @code{char} on the
1091 target machine. If this is undefined, the default is
1092 @code{CHAR_TYPE_SIZE}. Otherwise, it is the constant value that is the
1093 largest value that @code{CHAR_TYPE_SIZE} can have at run-time. This is
1096 @findex FLOAT_TYPE_SIZE
1097 @item FLOAT_TYPE_SIZE
1098 A C expression for the size in bits of the type @code{float} on the
1099 target machine. If you don't define this, the default is one word.
1101 @findex DOUBLE_TYPE_SIZE
1102 @item DOUBLE_TYPE_SIZE
1103 A C expression for the size in bits of the type @code{double} on the
1104 target machine. If you don't define this, the default is two
1107 @findex LONG_DOUBLE_TYPE_SIZE
1108 @item LONG_DOUBLE_TYPE_SIZE
1109 A C expression for the size in bits of the type @code{long double} on
1110 the target machine. If you don't define this, the default is two
1113 @findex WIDEST_HARDWARE_FP_SIZE
1114 @item WIDEST_HARDWARE_FP_SIZE
1115 A C expression for the size in bits of the widest floating-point format
1116 supported by the hardware. If you define this macro, you must specify a
1117 value less than or equal to the value of @code{LONG_DOUBLE_TYPE_SIZE}.
1118 If you do not define this macro, the value of @code{LONG_DOUBLE_TYPE_SIZE}
1121 @findex DEFAULT_SIGNED_CHAR
1122 @item DEFAULT_SIGNED_CHAR
1123 An expression whose value is 1 or 0, according to whether the type
1124 @code{char} should be signed or unsigned by default. The user can
1125 always override this default with the options @samp{-fsigned-char}
1126 and @samp{-funsigned-char}.
1128 @findex DEFAULT_SHORT_ENUMS
1129 @item DEFAULT_SHORT_ENUMS
1130 A C expression to determine whether to give an @code{enum} type
1131 only as many bytes as it takes to represent the range of possible values
1132 of that type. A nonzero value means to do that; a zero value means all
1133 @code{enum} types should be allocated like @code{int}.
1135 If you don't define the macro, the default is 0.
1139 A C expression for a string describing the name of the data type to use
1140 for size values. The typedef name @code{size_t} is defined using the
1141 contents of the string.
1143 The string can contain more than one keyword. If so, separate them with
1144 spaces, and write first any length keyword, then @code{unsigned} if
1145 appropriate, and finally @code{int}. The string must exactly match one
1146 of the data type names defined in the function
1147 @code{init_decl_processing} in the file @file{c-decl.c}. You may not
1148 omit @code{int} or change the order---that would cause the compiler to
1151 If you don't define this macro, the default is @code{"long unsigned
1154 @findex PTRDIFF_TYPE
1156 A C expression for a string describing the name of the data type to use
1157 for the result of subtracting two pointers. The typedef name
1158 @code{ptrdiff_t} is defined using the contents of the string. See
1159 @code{SIZE_TYPE} above for more information.
1161 If you don't define this macro, the default is @code{"long int"}.
1165 A C expression for a string describing the name of the data type to use
1166 for wide characters. The typedef name @code{wchar_t} is defined using
1167 the contents of the string. See @code{SIZE_TYPE} above for more
1170 If you don't define this macro, the default is @code{"int"}.
1172 @findex WCHAR_TYPE_SIZE
1173 @item WCHAR_TYPE_SIZE
1174 A C expression for the size in bits of the data type for wide
1175 characters. This is used in @code{cpp}, which cannot make use of
1178 @findex MAX_WCHAR_TYPE_SIZE
1179 @item MAX_WCHAR_TYPE_SIZE
1180 Maximum number for the size in bits of the data type for wide
1181 characters. If this is undefined, the default is
1182 @code{WCHAR_TYPE_SIZE}. Otherwise, it is the constant value that is the
1183 largest value that @code{WCHAR_TYPE_SIZE} can have at run-time. This is
1186 @findex OBJC_INT_SELECTORS
1187 @item OBJC_INT_SELECTORS
1188 Define this macro if the type of Objective C selectors should be
1191 If this macro is not defined, then selectors should have the type
1192 @code{struct objc_selector *}.
1194 @findex OBJC_SELECTORS_WITHOUT_LABELS
1195 @item OBJC_SELECTORS_WITHOUT_LABELS
1196 Define this macro if the compiler can group all the selectors together
1197 into a vector and use just one label at the beginning of the vector.
1198 Otherwise, the compiler must give each selector its own assembler
1201 On certain machines, it is important to have a separate label for each
1202 selector because this enables the linker to eliminate duplicate selectors.
1206 A C constant expression for the integer value for escape sequence
1211 @findex TARGET_NEWLINE
1214 @itemx TARGET_NEWLINE
1215 C constant expressions for the integer values for escape sequences
1216 @samp{\b}, @samp{\t} and @samp{\n}.
1224 C constant expressions for the integer values for escape sequences
1225 @samp{\v}, @samp{\f} and @samp{\r}.
1229 @section Register Usage
1230 @cindex register usage
1232 This section explains how to describe what registers the target machine
1233 has, and how (in general) they can be used.
1235 The description of which registers a specific instruction can use is
1236 done with register classes; see @ref{Register Classes}. For information
1237 on using registers to access a stack frame, see @ref{Frame Registers}.
1238 For passing values in registers, see @ref{Register Arguments}.
1239 For returning values in registers, see @ref{Scalar Return}.
1242 * Register Basics:: Number and kinds of registers.
1243 * Allocation Order:: Order in which registers are allocated.
1244 * Values in Registers:: What kinds of values each reg can hold.
1245 * Leaf Functions:: Renumbering registers for leaf functions.
1246 * Stack Registers:: Handling a register stack such as 80387.
1247 * Obsolete Register Macros:: Macros formerly used for the 80387.
1250 @node Register Basics
1251 @subsection Basic Characteristics of Registers
1253 @c prevent bad page break with this line
1254 Registers have various characteristics.
1257 @findex FIRST_PSEUDO_REGISTER
1258 @item FIRST_PSEUDO_REGISTER
1259 Number of hardware registers known to the compiler. They receive
1260 numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first
1261 pseudo register's number really is assigned the number
1262 @code{FIRST_PSEUDO_REGISTER}.
1264 @item FIXED_REGISTERS
1265 @findex FIXED_REGISTERS
1266 @cindex fixed register
1267 An initializer that says which registers are used for fixed purposes
1268 all throughout the compiled code and are therefore not available for
1269 general allocation. These would include the stack pointer, the frame
1270 pointer (except on machines where that can be used as a general
1271 register when no frame pointer is needed), the program counter on
1272 machines where that is considered one of the addressable registers,
1273 and any other numbered register with a standard use.
1275 This information is expressed as a sequence of numbers, separated by
1276 commas and surrounded by braces. The @var{n}th number is 1 if
1277 register @var{n} is fixed, 0 otherwise.
1279 The table initialized from this macro, and the table initialized by
1280 the following one, may be overridden at run time either automatically,
1281 by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by
1282 the user with the command options @samp{-ffixed-@var{reg}},
1283 @samp{-fcall-used-@var{reg}} and @samp{-fcall-saved-@var{reg}}.
1285 @findex CALL_USED_REGISTERS
1286 @item CALL_USED_REGISTERS
1287 @cindex call-used register
1288 @cindex call-clobbered register
1289 @cindex call-saved register
1290 Like @code{FIXED_REGISTERS} but has 1 for each register that is
1291 clobbered (in general) by function calls as well as for fixed
1292 registers. This macro therefore identifies the registers that are not
1293 available for general allocation of values that must live across
1296 If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler
1297 automatically saves it on function entry and restores it on function
1298 exit, if the register is used within the function.
1300 @findex CONDITIONAL_REGISTER_USAGE
1302 @findex call_used_regs
1303 @item CONDITIONAL_REGISTER_USAGE
1304 Zero or more C statements that may conditionally modify two variables
1305 @code{fixed_regs} and @code{call_used_regs} (both of type @code{char
1306 []}) after they have been initialized from the two preceding macros.
1308 This is necessary in case the fixed or call-clobbered registers depend
1311 You need not define this macro if it has no work to do.
1313 @cindex disabling certain registers
1314 @cindex controlling register usage
1315 If the usage of an entire class of registers depends on the target
1316 flags, you may indicate this to GCC by using this macro to modify
1317 @code{fixed_regs} and @code{call_used_regs} to 1 for each of the
1318 registers in the classes which should not be used by GCC. Also define
1319 the macro @code{REG_CLASS_FROM_LETTER} to return @code{NO_REGS} if it
1320 is called with a letter for a class that shouldn't be used.
1322 (However, if this class is not included in @code{GENERAL_REGS} and all
1323 of the insn patterns whose constraints permit this class are
1324 controlled by target switches, then GCC will automatically avoid using
1325 these registers when the target switches are opposed to them.)
1327 @findex NON_SAVING_SETJMP
1328 @item NON_SAVING_SETJMP
1329 If this macro is defined and has a nonzero value, it means that
1330 @code{setjmp} and related functions fail to save the registers, or that
1331 @code{longjmp} fails to restore them. To compensate, the compiler
1332 avoids putting variables in registers in functions that use
1335 @findex INCOMING_REGNO
1336 @item INCOMING_REGNO (@var{out})
1337 Define this macro if the target machine has register windows. This C
1338 expression returns the register number as seen by the called function
1339 corresponding to the register number @var{out} as seen by the calling
1340 function. Return @var{out} if register number @var{out} is not an
1343 @findex OUTGOING_REGNO
1344 @item OUTGOING_REGNO (@var{in})
1345 Define this macro if the target machine has register windows. This C
1346 expression returns the register number as seen by the calling function
1347 corresponding to the register number @var{in} as seen by the called
1348 function. Return @var{in} if register number @var{in} is not an inbound
1354 If the program counter has a register number, define this as that
1355 register number. Otherwise, do not define it.
1359 @node Allocation Order
1360 @subsection Order of Allocation of Registers
1361 @cindex order of register allocation
1362 @cindex register allocation order
1364 @c prevent bad page break with this line
1365 Registers are allocated in order.
1368 @findex REG_ALLOC_ORDER
1369 @item REG_ALLOC_ORDER
1370 If defined, an initializer for a vector of integers, containing the
1371 numbers of hard registers in the order in which GNU CC should prefer
1372 to use them (from most preferred to least).
1374 If this macro is not defined, registers are used lowest numbered first
1375 (all else being equal).
1377 One use of this macro is on machines where the highest numbered
1378 registers must always be saved and the save-multiple-registers
1379 instruction supports only sequences of consecutive registers. On such
1380 machines, define @code{REG_ALLOC_ORDER} to be an initializer that lists
1381 the highest numbered allocable register first.
1383 @findex ORDER_REGS_FOR_LOCAL_ALLOC
1384 @item ORDER_REGS_FOR_LOCAL_ALLOC
1385 A C statement (sans semicolon) to choose the order in which to allocate
1386 hard registers for pseudo-registers local to a basic block.
1388 Store the desired register order in the array @code{reg_alloc_order}.
1389 Element 0 should be the register to allocate first; element 1, the next
1390 register; and so on.
1392 The macro body should not assume anything about the contents of
1393 @code{reg_alloc_order} before execution of the macro.
1395 On most machines, it is not necessary to define this macro.
1398 @node Values in Registers
1399 @subsection How Values Fit in Registers
1401 This section discusses the macros that describe which kinds of values
1402 (specifically, which machine modes) each register can hold, and how many
1403 consecutive registers are needed for a given mode.
1406 @findex HARD_REGNO_NREGS
1407 @item HARD_REGNO_NREGS (@var{regno}, @var{mode})
1408 A C expression for the number of consecutive hard registers, starting
1409 at register number @var{regno}, required to hold a value of mode
1412 On a machine where all registers are exactly one word, a suitable
1413 definition of this macro is
1416 #define HARD_REGNO_NREGS(REGNO, MODE) \
1417 ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) \
1421 @findex ALTER_HARD_SUBREG
1422 @item ALTER_HARD_SUBREG (@var{tgt_mode}, @var{word}, @var{src_mode}, @var{regno})
1423 A C expression that returns an adjusted hard register number for
1426 (subreg:@var{tgt_mode} (reg:@var{src_mode} @var{regno}) @var{word})
1429 This may be needed if the target machine has mixed sized big-endian
1430 registers, like Sparc v9.
1432 @findex HARD_REGNO_MODE_OK
1433 @item HARD_REGNO_MODE_OK (@var{regno}, @var{mode})
1434 A C expression that is nonzero if it is permissible to store a value
1435 of mode @var{mode} in hard register number @var{regno} (or in several
1436 registers starting with that one). For a machine where all registers
1437 are equivalent, a suitable definition is
1440 #define HARD_REGNO_MODE_OK(REGNO, MODE) 1
1443 You need not include code to check for the numbers of fixed registers,
1444 because the allocation mechanism considers them to be always occupied.
1446 @cindex register pairs
1447 On some machines, double-precision values must be kept in even/odd
1448 register pairs. You can implement that by defining this macro to reject
1449 odd register numbers for such modes.
1451 The minimum requirement for a mode to be OK in a register is that the
1452 @samp{mov@var{mode}} instruction pattern support moves between the
1453 register and other hard register in the same class and that moving a
1454 value into the register and back out not alter it.
1456 Since the same instruction used to move @code{word_mode} will work for
1457 all narrower integer modes, it is not necessary on any machine for
1458 @code{HARD_REGNO_MODE_OK} to distinguish between these modes, provided
1459 you define patterns @samp{movhi}, etc., to take advantage of this. This
1460 is useful because of the interaction between @code{HARD_REGNO_MODE_OK}
1461 and @code{MODES_TIEABLE_P}; it is very desirable for all integer modes
1464 Many machines have special registers for floating point arithmetic.
1465 Often people assume that floating point machine modes are allowed only
1466 in floating point registers. This is not true. Any registers that
1467 can hold integers can safely @emph{hold} a floating point machine
1468 mode, whether or not floating arithmetic can be done on it in those
1469 registers. Integer move instructions can be used to move the values.
1471 On some machines, though, the converse is true: fixed-point machine
1472 modes may not go in floating registers. This is true if the floating
1473 registers normalize any value stored in them, because storing a
1474 non-floating value there would garble it. In this case,
1475 @code{HARD_REGNO_MODE_OK} should reject fixed-point machine modes in
1476 floating registers. But if the floating registers do not automatically
1477 normalize, if you can store any bit pattern in one and retrieve it
1478 unchanged without a trap, then any machine mode may go in a floating
1479 register, so you can define this macro to say so.
1481 The primary significance of special floating registers is rather that
1482 they are the registers acceptable in floating point arithmetic
1483 instructions. However, this is of no concern to
1484 @code{HARD_REGNO_MODE_OK}. You handle it by writing the proper
1485 constraints for those instructions.
1487 On some machines, the floating registers are especially slow to access,
1488 so that it is better to store a value in a stack frame than in such a
1489 register if floating point arithmetic is not being done. As long as the
1490 floating registers are not in class @code{GENERAL_REGS}, they will not
1491 be used unless some pattern's constraint asks for one.
1493 @findex MODES_TIEABLE_P
1494 @item MODES_TIEABLE_P (@var{mode1}, @var{mode2})
1495 A C expression that is nonzero if a value of mode
1496 @var{mode1} is accessible in mode @var{mode2} without copying.
1498 If @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and
1499 @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are always the same for
1500 any @var{r}, then @code{MODES_TIEABLE_P (@var{mode1}, @var{mode2})}
1501 should be nonzero. If they differ for any @var{r}, you should define
1502 this macro to return zero unless some other mechanism ensures the
1503 accessibility of the value in a narrower mode.
1505 You should define this macro to return nonzero in as many cases as
1506 possible since doing so will allow GNU CC to perform better register
1509 @findex AVOID_CCMODE_COPIES
1510 @item AVOID_CCMODE_COPIES
1511 Define this macro if the compiler should avoid copies to/from @code{CCmode}
1512 registers. You should only define this macro if support fo copying to/from
1513 @code{CCmode} is incomplete.
1516 @node Leaf Functions
1517 @subsection Handling Leaf Functions
1519 @cindex leaf functions
1520 @cindex functions, leaf
1521 On some machines, a leaf function (i.e., one which makes no calls) can run
1522 more efficiently if it does not make its own register window. Often this
1523 means it is required to receive its arguments in the registers where they
1524 are passed by the caller, instead of the registers where they would
1527 The special treatment for leaf functions generally applies only when
1528 other conditions are met; for example, often they may use only those
1529 registers for its own variables and temporaries. We use the term ``leaf
1530 function'' to mean a function that is suitable for this special
1531 handling, so that functions with no calls are not necessarily ``leaf
1534 GNU CC assigns register numbers before it knows whether the function is
1535 suitable for leaf function treatment. So it needs to renumber the
1536 registers in order to output a leaf function. The following macros
1540 @findex LEAF_REGISTERS
1541 @item LEAF_REGISTERS
1542 A C initializer for a vector, indexed by hard register number, which
1543 contains 1 for a register that is allowable in a candidate for leaf
1546 If leaf function treatment involves renumbering the registers, then the
1547 registers marked here should be the ones before renumbering---those that
1548 GNU CC would ordinarily allocate. The registers which will actually be
1549 used in the assembler code, after renumbering, should not be marked with 1
1552 Define this macro only if the target machine offers a way to optimize
1553 the treatment of leaf functions.
1555 @findex LEAF_REG_REMAP
1556 @item LEAF_REG_REMAP (@var{regno})
1557 A C expression whose value is the register number to which @var{regno}
1558 should be renumbered, when a function is treated as a leaf function.
1560 If @var{regno} is a register number which should not appear in a leaf
1561 function before renumbering, then the expression should yield -1, which
1562 will cause the compiler to abort.
1564 Define this macro only if the target machine offers a way to optimize the
1565 treatment of leaf functions, and registers need to be renumbered to do
1569 @findex leaf_function
1570 Normally, @code{FUNCTION_PROLOGUE} and @code{FUNCTION_EPILOGUE} must
1571 treat leaf functions specially. It can test the C variable
1572 @code{leaf_function} which is nonzero for leaf functions. (The variable
1573 @code{leaf_function} is defined only if @code{LEAF_REGISTERS} is
1575 @c changed this to fix overfull. ALSO: why the "it" at the beginning
1576 @c of the next paragraph?! --mew 2feb93
1578 @node Stack Registers
1579 @subsection Registers That Form a Stack
1581 There are special features to handle computers where some of the
1582 ``registers'' form a stack, as in the 80387 coprocessor for the 80386.
1583 Stack registers are normally written by pushing onto the stack, and are
1584 numbered relative to the top of the stack.
1586 Currently, GNU CC can only handle one group of stack-like registers, and
1587 they must be consecutively numbered.
1592 Define this if the machine has any stack-like registers.
1594 @findex FIRST_STACK_REG
1595 @item FIRST_STACK_REG
1596 The number of the first stack-like register. This one is the top
1599 @findex LAST_STACK_REG
1600 @item LAST_STACK_REG
1601 The number of the last stack-like register. This one is the bottom of
1605 @node Obsolete Register Macros
1606 @subsection Obsolete Macros for Controlling Register Usage
1608 These features do not work very well. They exist because they used to
1609 be required to generate correct code for the 80387 coprocessor of the
1610 80386. They are no longer used by that machine description and may be
1611 removed in a later version of the compiler. Don't use them!
1614 @findex OVERLAPPING_REGNO_P
1615 @item OVERLAPPING_REGNO_P (@var{regno})
1616 If defined, this is a C expression whose value is nonzero if hard
1617 register number @var{regno} is an overlapping register. This means a
1618 hard register which overlaps a hard register with a different number.
1619 (Such overlap is undesirable, but occasionally it allows a machine to
1620 be supported which otherwise could not be.) This macro must return
1621 nonzero for @emph{all} the registers which overlap each other. GNU CC
1622 can use an overlapping register only in certain limited ways. It can
1623 be used for allocation within a basic block, and may be spilled for
1624 reloading; that is all.
1626 If this macro is not defined, it means that none of the hard registers
1627 overlap each other. This is the usual situation.
1629 @findex INSN_CLOBBERS_REGNO_P
1630 @item INSN_CLOBBERS_REGNO_P (@var{insn}, @var{regno})
1631 If defined, this is a C expression whose value should be nonzero if
1632 the insn @var{insn} has the effect of mysteriously clobbering the
1633 contents of hard register number @var{regno}. By ``mysterious'' we
1634 mean that the insn's RTL expression doesn't describe such an effect.
1636 If this macro is not defined, it means that no insn clobbers registers
1637 mysteriously. This is the usual situation; all else being equal,
1638 it is best for the RTL expression to show all the activity.
1641 @findex PRESERVE_DEATH_INFO_REGNO_P
1642 @item PRESERVE_DEATH_INFO_REGNO_P (@var{regno})
1643 If defined, this is a C expression whose value is nonzero if correct
1644 @code{REG_DEAD} notes are needed for hard register number @var{regno}
1647 You would arrange to preserve death info for a register when some of the
1648 code in the machine description which is executed to write the assembler
1649 code looks at the death notes. This is necessary only when the actual
1650 hardware feature which GNU CC thinks of as a register is not actually a
1651 register of the usual sort. (It might, for example, be a hardware
1654 It is also useful for peepholes and linker relaxation.
1656 If this macro is not defined, it means that no death notes need to be
1657 preserved, and some may even be incorrect. This is the usual situation.
1660 @node Register Classes
1661 @section Register Classes
1662 @cindex register class definitions
1663 @cindex class definitions, register
1665 On many machines, the numbered registers are not all equivalent.
1666 For example, certain registers may not be allowed for indexed addressing;
1667 certain registers may not be allowed in some instructions. These machine
1668 restrictions are described to the compiler using @dfn{register classes}.
1670 You define a number of register classes, giving each one a name and saying
1671 which of the registers belong to it. Then you can specify register classes
1672 that are allowed as operands to particular instruction patterns.
1676 In general, each register will belong to several classes. In fact, one
1677 class must be named @code{ALL_REGS} and contain all the registers. Another
1678 class must be named @code{NO_REGS} and contain no registers. Often the
1679 union of two classes will be another class; however, this is not required.
1681 @findex GENERAL_REGS
1682 One of the classes must be named @code{GENERAL_REGS}. There is nothing
1683 terribly special about the name, but the operand constraint letters
1684 @samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is
1685 the same as @code{ALL_REGS}, just define it as a macro which expands
1688 Order the classes so that if class @var{x} is contained in class @var{y}
1689 then @var{x} has a lower class number than @var{y}.
1691 The way classes other than @code{GENERAL_REGS} are specified in operand
1692 constraints is through machine-dependent operand constraint letters.
1693 You can define such letters to correspond to various classes, then use
1694 them in operand constraints.
1696 You should define a class for the union of two classes whenever some
1697 instruction allows both classes. For example, if an instruction allows
1698 either a floating point (coprocessor) register or a general register for a
1699 certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS}
1700 which includes both of them. Otherwise you will get suboptimal code.
1702 You must also specify certain redundant information about the register
1703 classes: for each class, which classes contain it and which ones are
1704 contained in it; for each pair of classes, the largest class contained
1707 When a value occupying several consecutive registers is expected in a
1708 certain class, all the registers used must belong to that class.
1709 Therefore, register classes cannot be used to enforce a requirement for
1710 a register pair to start with an even-numbered register. The way to
1711 specify this requirement is with @code{HARD_REGNO_MODE_OK}.
1713 Register classes used for input-operands of bitwise-and or shift
1714 instructions have a special requirement: each such class must have, for
1715 each fixed-point machine mode, a subclass whose registers can transfer that
1716 mode to or from memory. For example, on some machines, the operations for
1717 single-byte values (@code{QImode}) are limited to certain registers. When
1718 this is so, each register class that is used in a bitwise-and or shift
1719 instruction must have a subclass consisting of registers from which
1720 single-byte values can be loaded or stored. This is so that
1721 @code{PREFERRED_RELOAD_CLASS} can always have a possible value to return.
1724 @findex enum reg_class
1725 @item enum reg_class
1726 An enumeral type that must be defined with all the register class names
1727 as enumeral values. @code{NO_REGS} must be first. @code{ALL_REGS}
1728 must be the last register class, followed by one more enumeral value,
1729 @code{LIM_REG_CLASSES}, which is not a register class but rather
1730 tells how many classes there are.
1732 Each register class has a number, which is the value of casting
1733 the class name to type @code{int}. The number serves as an index
1734 in many of the tables described below.
1736 @findex N_REG_CLASSES
1738 The number of distinct register classes, defined as follows:
1741 #define N_REG_CLASSES (int) LIM_REG_CLASSES
1744 @findex REG_CLASS_NAMES
1745 @item REG_CLASS_NAMES
1746 An initializer containing the names of the register classes as C string
1747 constants. These names are used in writing some of the debugging dumps.
1749 @findex REG_CLASS_CONTENTS
1750 @item REG_CLASS_CONTENTS
1751 An initializer containing the contents of the register classes, as integers
1752 which are bit masks. The @var{n}th integer specifies the contents of class
1753 @var{n}. The way the integer @var{mask} is interpreted is that
1754 register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1.
1756 When the machine has more than 32 registers, an integer does not suffice.
1757 Then the integers are replaced by sub-initializers, braced groupings containing
1758 several integers. Each sub-initializer must be suitable as an initializer
1759 for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}.
1761 @findex REGNO_REG_CLASS
1762 @item REGNO_REG_CLASS (@var{regno})
1763 A C expression whose value is a register class containing hard register
1764 @var{regno}. In general there is more than one such class; choose a class
1765 which is @dfn{minimal}, meaning that no smaller class also contains the
1768 @findex BASE_REG_CLASS
1769 @item BASE_REG_CLASS
1770 A macro whose definition is the name of the class to which a valid
1771 base register must belong. A base register is one used in an address
1772 which is the register value plus a displacement.
1774 @findex INDEX_REG_CLASS
1775 @item INDEX_REG_CLASS
1776 A macro whose definition is the name of the class to which a valid
1777 index register must belong. An index register is one used in an
1778 address where its value is either multiplied by a scale factor or
1779 added to another register (as well as added to a displacement).
1781 @findex REG_CLASS_FROM_LETTER
1782 @item REG_CLASS_FROM_LETTER (@var{char})
1783 A C expression which defines the machine-dependent operand constraint
1784 letters for register classes. If @var{char} is such a letter, the
1785 value should be the register class corresponding to it. Otherwise,
1786 the value should be @code{NO_REGS}. The register letter @samp{r},
1787 corresponding to class @code{GENERAL_REGS}, will not be passed
1788 to this macro; you do not need to handle it.
1790 @findex REGNO_OK_FOR_BASE_P
1791 @item REGNO_OK_FOR_BASE_P (@var{num})
1792 A C expression which is nonzero if register number @var{num} is
1793 suitable for use as a base register in operand addresses. It may be
1794 either a suitable hard register or a pseudo register that has been
1795 allocated such a hard register.
1797 @findex REGNO_MODE_OK_FOR_BASE_P
1798 @item REGNO_MODE_OK_FOR_BASE_P (@var{num}, @var{mode})
1799 A C expression that is just like @code{REGNO_OK_FOR_BASE_P}, except that
1800 that expression may examine the mode of the memory reference in
1801 @var{mode}. You should define this macro if the mode of the memory
1802 reference affects whether a register may be used as a base register. If
1803 you define this macro, the compiler will use it instead of
1804 @code{REGNO_OK_FOR_BASE_P}.
1806 @findex REGNO_OK_FOR_INDEX_P
1807 @item REGNO_OK_FOR_INDEX_P (@var{num})
1808 A C expression which is nonzero if register number @var{num} is
1809 suitable for use as an index register in operand addresses. It may be
1810 either a suitable hard register or a pseudo register that has been
1811 allocated such a hard register.
1813 The difference between an index register and a base register is that
1814 the index register may be scaled. If an address involves the sum of
1815 two registers, neither one of them scaled, then either one may be
1816 labeled the ``base'' and the other the ``index''; but whichever
1817 labeling is used must fit the machine's constraints of which registers
1818 may serve in each capacity. The compiler will try both labelings,
1819 looking for one that is valid, and will reload one or both registers
1820 only if neither labeling works.
1822 @findex PREFERRED_RELOAD_CLASS
1823 @item PREFERRED_RELOAD_CLASS (@var{x}, @var{class})
1824 A C expression that places additional restrictions on the register class
1825 to use when it is necessary to copy value @var{x} into a register in class
1826 @var{class}. The value is a register class; perhaps @var{class}, or perhaps
1827 another, smaller class. On many machines, the following definition is
1831 #define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS
1834 Sometimes returning a more restrictive class makes better code. For
1835 example, on the 68000, when @var{x} is an integer constant that is in range
1836 for a @samp{moveq} instruction, the value of this macro is always
1837 @code{DATA_REGS} as long as @var{class} includes the data registers.
1838 Requiring a data register guarantees that a @samp{moveq} will be used.
1840 If @var{x} is a @code{const_double}, by returning @code{NO_REGS}
1841 you can force @var{x} into a memory constant. This is useful on
1842 certain machines where immediate floating values cannot be loaded into
1843 certain kinds of registers.
1845 @findex PREFERRED_OUTPUT_RELOAD_CLASS
1846 @item PREFERRED_OUTPUT_RELOAD_CLASS (@var{x}, @var{class})
1847 Like @code{PREFERRED_RELOAD_CLASS}, but for output reloads instead of
1848 input reloads. If you don't define this macro, the default is to use
1849 @var{class}, unchanged.
1851 @findex LIMIT_RELOAD_CLASS
1852 @item LIMIT_RELOAD_CLASS (@var{mode}, @var{class})
1853 A C expression that places additional restrictions on the register class
1854 to use when it is necessary to be able to hold a value of mode
1855 @var{mode} in a reload register for which class @var{class} would
1858 Unlike @code{PREFERRED_RELOAD_CLASS}, this macro should be used when
1859 there are certain modes that simply can't go in certain reload classes.
1861 The value is a register class; perhaps @var{class}, or perhaps another,
1864 Don't define this macro unless the target machine has limitations which
1865 require the macro to do something nontrivial.
1867 @findex SECONDARY_RELOAD_CLASS
1868 @findex SECONDARY_INPUT_RELOAD_CLASS
1869 @findex SECONDARY_OUTPUT_RELOAD_CLASS
1870 @item SECONDARY_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
1871 @itemx SECONDARY_INPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
1872 @itemx SECONDARY_OUTPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
1873 Many machines have some registers that cannot be copied directly to or
1874 from memory or even from other types of registers. An example is the
1875 @samp{MQ} register, which on most machines, can only be copied to or
1876 from general registers, but not memory. Some machines allow copying all
1877 registers to and from memory, but require a scratch register for stores
1878 to some memory locations (e.g., those with symbolic address on the RT,
1879 and those with certain symbolic address on the Sparc when compiling
1880 PIC). In some cases, both an intermediate and a scratch register are
1883 You should define these macros to indicate to the reload phase that it may
1884 need to allocate at least one register for a reload in addition to the
1885 register to contain the data. Specifically, if copying @var{x} to a
1886 register @var{class} in @var{mode} requires an intermediate register,
1887 you should define @code{SECONDARY_INPUT_RELOAD_CLASS} to return the
1888 largest register class all of whose registers can be used as
1889 intermediate registers or scratch registers.
1891 If copying a register @var{class} in @var{mode} to @var{x} requires an
1892 intermediate or scratch register, @code{SECONDARY_OUTPUT_RELOAD_CLASS}
1893 should be defined to return the largest register class required. If the
1894 requirements for input and output reloads are the same, the macro
1895 @code{SECONDARY_RELOAD_CLASS} should be used instead of defining both
1898 The values returned by these macros are often @code{GENERAL_REGS}.
1899 Return @code{NO_REGS} if no spare register is needed; i.e., if @var{x}
1900 can be directly copied to or from a register of @var{class} in
1901 @var{mode} without requiring a scratch register. Do not define this
1902 macro if it would always return @code{NO_REGS}.
1904 If a scratch register is required (either with or without an
1905 intermediate register), you should define patterns for
1906 @samp{reload_in@var{m}} or @samp{reload_out@var{m}}, as required
1907 (@pxref{Standard Names}. These patterns, which will normally be
1908 implemented with a @code{define_expand}, should be similar to the
1909 @samp{mov@var{m}} patterns, except that operand 2 is the scratch
1912 Define constraints for the reload register and scratch register that
1913 contain a single register class. If the original reload register (whose
1914 class is @var{class}) can meet the constraint given in the pattern, the
1915 value returned by these macros is used for the class of the scratch
1916 register. Otherwise, two additional reload registers are required.
1917 Their classes are obtained from the constraints in the insn pattern.
1919 @var{x} might be a pseudo-register or a @code{subreg} of a
1920 pseudo-register, which could either be in a hard register or in memory.
1921 Use @code{true_regnum} to find out; it will return -1 if the pseudo is
1922 in memory and the hard register number if it is in a register.
1924 These macros should not be used in the case where a particular class of
1925 registers can only be copied to memory and not to another class of
1926 registers. In that case, secondary reload registers are not needed and
1927 would not be helpful. Instead, a stack location must be used to perform
1928 the copy and the @code{mov@var{m}} pattern should use memory as a
1929 intermediate storage. This case often occurs between floating-point and
1932 @findex SECONDARY_MEMORY_NEEDED
1933 @item SECONDARY_MEMORY_NEEDED (@var{class1}, @var{class2}, @var{m})
1934 Certain machines have the property that some registers cannot be copied
1935 to some other registers without using memory. Define this macro on
1936 those machines to be a C expression that is non-zero if objects of mode
1937 @var{m} in registers of @var{class1} can only be copied to registers of
1938 class @var{class2} by storing a register of @var{class1} into memory
1939 and loading that memory location into a register of @var{class2}.
1941 Do not define this macro if its value would always be zero.
1943 @findex SECONDARY_MEMORY_NEEDED_RTX
1944 @item SECONDARY_MEMORY_NEEDED_RTX (@var{mode})
1945 Normally when @code{SECONDARY_MEMORY_NEEDED} is defined, the compiler
1946 allocates a stack slot for a memory location needed for register copies.
1947 If this macro is defined, the compiler instead uses the memory location
1948 defined by this macro.
1950 Do not define this macro if you do not define
1951 @code{SECONDARY_MEMORY_NEEDED}.
1953 @findex SECONDARY_MEMORY_NEEDED_MODE
1954 @item SECONDARY_MEMORY_NEEDED_MODE (@var{mode})
1955 When the compiler needs a secondary memory location to copy between two
1956 registers of mode @var{mode}, it normally allocates sufficient memory to
1957 hold a quantity of @code{BITS_PER_WORD} bits and performs the store and
1958 load operations in a mode that many bits wide and whose class is the
1959 same as that of @var{mode}.
1961 This is right thing to do on most machines because it ensures that all
1962 bits of the register are copied and prevents accesses to the registers
1963 in a narrower mode, which some machines prohibit for floating-point
1966 However, this default behavior is not correct on some machines, such as
1967 the DEC Alpha, that store short integers in floating-point registers
1968 differently than in integer registers. On those machines, the default
1969 widening will not work correctly and you must define this macro to
1970 suppress that widening in some cases. See the file @file{alpha.h} for
1973 Do not define this macro if you do not define
1974 @code{SECONDARY_MEMORY_NEEDED} or if widening @var{mode} to a mode that
1975 is @code{BITS_PER_WORD} bits wide is correct for your machine.
1977 @findex SMALL_REGISTER_CLASSES
1978 @item SMALL_REGISTER_CLASSES
1979 Normally the compiler avoids choosing registers that have been
1980 explicitly mentioned in the rtl as spill registers (these registers are
1981 normally those used to pass parameters and return values). However,
1982 some machines have so few registers of certain classes that there
1983 would not be enough registers to use as spill registers if this were
1986 Define @code{SMALL_REGISTER_CLASSES} to be an expression with a non-zero
1987 value on these machines. When this macro has a non-zero value, the
1988 compiler allows registers explicitly used in the rtl to be used as spill
1989 registers but avoids extending the lifetime of these registers.
1991 It is always safe to define this macro with a non-zero value, but if you
1992 unnecessarily define it, you will reduce the amount of optimizations
1993 that can be performed in some cases. If you do not define this macro
1994 with a non-zero value when it is required, the compiler will run out of
1995 spill registers and print a fatal error message. For most machines, you
1996 should not define this macro at all.
1998 @findex CLASS_LIKELY_SPILLED_P
1999 @item CLASS_LIKELY_SPILLED_P (@var{class})
2000 A C expression whose value is nonzero if pseudos that have been assigned
2001 to registers of class @var{class} would likely be spilled because
2002 registers of @var{class} are needed for spill registers.
2004 The default value of this macro returns 1 if @var{class} has exactly one
2005 register and zero otherwise. On most machines, this default should be
2006 used. Only define this macro to some other expression if pseudos
2007 allocated by @file{local-alloc.c} end up in memory because their hard
2008 registers were needed for spill registers. If this macro returns nonzero
2009 for those classes, those pseudos will only be allocated by
2010 @file{global.c}, which knows how to reallocate the pseudo to another
2011 register. If there would not be another register available for
2012 reallocation, you should not change the definition of this macro since
2013 the only effect of such a definition would be to slow down register
2016 @findex CLASS_MAX_NREGS
2017 @item CLASS_MAX_NREGS (@var{class}, @var{mode})
2018 A C expression for the maximum number of consecutive registers
2019 of class @var{class} needed to hold a value of mode @var{mode}.
2021 This is closely related to the macro @code{HARD_REGNO_NREGS}. In fact,
2022 the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})}
2023 should be the maximum value of @code{HARD_REGNO_NREGS (@var{regno},
2024 @var{mode})} for all @var{regno} values in the class @var{class}.
2026 This macro helps control the handling of multiple-word values
2029 @item CLASS_CANNOT_CHANGE_SIZE
2030 If defined, a C expression for a class that contains registers which the
2031 compiler must always access in a mode that is the same size as the mode
2032 in which it loaded the register.
2034 For the example, loading 32-bit integer or floating-point objects into
2035 floating-point registers on the Alpha extends them to 64-bits.
2036 Therefore loading a 64-bit object and then storing it as a 32-bit object
2037 does not store the low-order 32-bits, as would be the case for a normal
2038 register. Therefore, @file{alpha.h} defines this macro as
2042 Three other special macros describe which operands fit which constraint
2046 @findex CONST_OK_FOR_LETTER_P
2047 @item CONST_OK_FOR_LETTER_P (@var{value}, @var{c})
2048 A C expression that defines the machine-dependent operand constraint
2049 letters (@samp{I}, @samp{J}, @samp{K}, @dots{} @samp{P}) that specify
2050 particular ranges of integer values. If @var{c} is one of those
2051 letters, the expression should check that @var{value}, an integer, is in
2052 the appropriate range and return 1 if so, 0 otherwise. If @var{c} is
2053 not one of those letters, the value should be 0 regardless of
2056 @findex CONST_DOUBLE_OK_FOR_LETTER_P
2057 @item CONST_DOUBLE_OK_FOR_LETTER_P (@var{value}, @var{c})
2058 A C expression that defines the machine-dependent operand constraint
2059 letters that specify particular ranges of @code{const_double} values
2060 (@samp{G} or @samp{H}).
2062 If @var{c} is one of those letters, the expression should check that
2063 @var{value}, an RTX of code @code{const_double}, is in the appropriate
2064 range and return 1 if so, 0 otherwise. If @var{c} is not one of those
2065 letters, the value should be 0 regardless of @var{value}.
2067 @code{const_double} is used for all floating-point constants and for
2068 @code{DImode} fixed-point constants. A given letter can accept either
2069 or both kinds of values. It can use @code{GET_MODE} to distinguish
2070 between these kinds.
2072 @findex EXTRA_CONSTRAINT
2073 @item EXTRA_CONSTRAINT (@var{value}, @var{c})
2074 A C expression that defines the optional machine-dependent constraint
2075 letters (@samp{Q}, @samp{R}, @samp{S}, @samp{T}, @samp{U}) that can
2076 be used to segregate specific types of operands, usually memory
2077 references, for the target machine. Normally this macro will not be
2078 defined. If it is required for a particular target machine, it should
2079 return 1 if @var{value} corresponds to the operand type represented by
2080 the constraint letter @var{c}. If @var{c} is not defined as an extra
2081 constraint, the value returned should be 0 regardless of @var{value}.
2083 For example, on the ROMP, load instructions cannot have their output in r0 if
2084 the memory reference contains a symbolic address. Constraint letter
2085 @samp{Q} is defined as representing a memory address that does
2086 @emph{not} contain a symbolic address. An alternative is specified with
2087 a @samp{Q} constraint on the input and @samp{r} on the output. The next
2088 alternative specifies @samp{m} on the input and a register class that
2089 does not include r0 on the output.
2092 @node Stack and Calling
2093 @section Stack Layout and Calling Conventions
2094 @cindex calling conventions
2096 @c prevent bad page break with this line
2097 This describes the stack layout and calling conventions.
2105 * Register Arguments::
2107 * Aggregate Return::
2114 @subsection Basic Stack Layout
2115 @cindex stack frame layout
2116 @cindex frame layout
2118 @c prevent bad page break with this line
2119 Here is the basic stack layout.
2122 @findex STACK_GROWS_DOWNWARD
2123 @item STACK_GROWS_DOWNWARD
2124 Define this macro if pushing a word onto the stack moves the stack
2125 pointer to a smaller address.
2127 When we say, ``define this macro if @dots{},'' it means that the
2128 compiler checks this macro only with @code{#ifdef} so the precise
2129 definition used does not matter.
2131 @findex FRAME_GROWS_DOWNWARD
2132 @item FRAME_GROWS_DOWNWARD
2133 Define this macro if the addresses of local variable slots are at negative
2134 offsets from the frame pointer.
2136 @findex ARGS_GROW_DOWNWARD
2137 @item ARGS_GROW_DOWNWARD
2138 Define this macro if successive arguments to a function occupy decreasing
2139 addresses on the stack.
2141 @findex STARTING_FRAME_OFFSET
2142 @item STARTING_FRAME_OFFSET
2143 Offset from the frame pointer to the first local variable slot to be allocated.
2145 If @code{FRAME_GROWS_DOWNWARD}, find the next slot's offset by
2146 subtracting the first slot's length from @code{STARTING_FRAME_OFFSET}.
2147 Otherwise, it is found by adding the length of the first slot to the
2148 value @code{STARTING_FRAME_OFFSET}.
2149 @c i'm not sure if the above is still correct.. had to change it to get
2150 @c rid of an overfull. --mew 2feb93
2152 @findex STACK_POINTER_OFFSET
2153 @item STACK_POINTER_OFFSET
2154 Offset from the stack pointer register to the first location at which
2155 outgoing arguments are placed. If not specified, the default value of
2156 zero is used. This is the proper value for most machines.
2158 If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
2159 the first location at which outgoing arguments are placed.
2161 @findex FIRST_PARM_OFFSET
2162 @item FIRST_PARM_OFFSET (@var{fundecl})
2163 Offset from the argument pointer register to the first argument's
2164 address. On some machines it may depend on the data type of the
2167 If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
2168 the first argument's address.
2170 @findex STACK_DYNAMIC_OFFSET
2171 @item STACK_DYNAMIC_OFFSET (@var{fundecl})
2172 Offset from the stack pointer register to an item dynamically allocated
2173 on the stack, e.g., by @code{alloca}.
2175 The default value for this macro is @code{STACK_POINTER_OFFSET} plus the
2176 length of the outgoing arguments. The default is correct for most
2177 machines. See @file{function.c} for details.
2179 @findex DYNAMIC_CHAIN_ADDRESS
2180 @item DYNAMIC_CHAIN_ADDRESS (@var{frameaddr})
2181 A C expression whose value is RTL representing the address in a stack
2182 frame where the pointer to the caller's frame is stored. Assume that
2183 @var{frameaddr} is an RTL expression for the address of the stack frame
2186 If you don't define this macro, the default is to return the value
2187 of @var{frameaddr}---that is, the stack frame address is also the
2188 address of the stack word that points to the previous frame.
2190 @findex SETUP_FRAME_ADDRESSES
2191 @item SETUP_FRAME_ADDRESSES
2192 If defined, a C expression that produces the machine-specific code to
2193 setup the stack so that arbitrary frames can be accessed. For example,
2194 on the Sparc, we must flush all of the register windows to the stack
2195 before we can access arbitrary stack frames. You will seldom need to
2198 @findex BUILTIN_SETJMP_FRAME_VALUE
2199 @item BUILTIN_SETJMP_FRAME_VALUE
2200 If defined, a C expression that contains an rtx that is used to store
2201 the address of the current frame into the built in @code{setjmp} buffer.
2202 The default value, @code{virtual_stack_vars_rtx}, is correct for most
2203 machines. One reason you may need to define this macro is if
2204 @code{hard_frame_pointer_rtx} is the appropriate value on your machine.
2206 @findex RETURN_ADDR_RTX
2207 @item RETURN_ADDR_RTX (@var{count}, @var{frameaddr})
2208 A C expression whose value is RTL representing the value of the return
2209 address for the frame @var{count} steps up from the current frame, after
2210 the prologue. @var{frameaddr} is the frame pointer of the @var{count}
2211 frame, or the frame pointer of the @var{count} @minus{} 1 frame if
2212 @code{RETURN_ADDR_IN_PREVIOUS_FRAME} is defined.
2214 The value of the expression must always be the correct address when
2215 @var{count} is zero, but may be @code{NULL_RTX} if there is not way to
2216 determine the return address of other frames.
2218 @findex RETURN_ADDR_IN_PREVIOUS_FRAME
2219 @item RETURN_ADDR_IN_PREVIOUS_FRAME
2220 Define this if the return address of a particular stack frame is accessed
2221 from the frame pointer of the previous stack frame.
2223 @findex INCOMING_RETURN_ADDR_RTX
2224 @item INCOMING_RETURN_ADDR_RTX
2225 A C expression whose value is RTL representing the location of the
2226 incoming return address at the beginning of any function, before the
2227 prologue. This RTL is either a @code{REG}, indicating that the return
2228 value is saved in @samp{REG}, or a @code{MEM} representing a location in
2231 You only need to define this macro if you want to support call frame
2232 debugging information like that provided by DWARF 2.
2234 @findex INCOMING_FRAME_SP_OFFSET
2235 @item INCOMING_FRAME_SP_OFFSET
2236 A C expression whose value is an integer giving the offset, in bytes,
2237 from the value of the stack pointer register to the top of the stack
2238 frame at the beginning of any function, before the prologue. The top of
2239 the frame is defined to be the value of the stack pointer in the
2240 previous frame, just before the call instruction.
2242 You only need to define this macro if you want to support call frame
2243 debugging information like that provided by DWARF 2.
2246 @node Stack Checking
2247 @subsection Specifying How Stack Checking is Done
2249 GNU CC will check that stack references are within the boundaries of
2250 the stack, if the @samp{-fstack-check} is specified, in one of three ways:
2254 If the value of the @code{STACK_CHECK_BUILTIN} macro is nonzero, GNU CC
2255 will assume that you have arranged for stack checking to be done at
2256 appropriate places in the configuration files, e.g., in
2257 @code{FUNCTION_PROLOGUE}. GNU CC will do not other special processing.
2260 If @code{STACK_CHECK_BUILTIN} is zero and you defined a named pattern
2261 called @code{check_stack} in your @file{md} file, GNU CC will call that
2262 pattern with one argument which is the address to compare the stack
2263 value against. You must arrange for this pattern to report an error if
2264 the stack pointer is out of range.
2267 If neither of the above are true, GNU CC will generate code to periodically
2268 ``probe'' the stack pointer using the values of the macros defined below.
2271 Normally, you will use the default values of these macros, so GNU CC
2272 will use the third approach.
2275 @findex STACK_CHECK_BUILTIN
2276 @item STACK_CHECK_BUILTIN
2277 A nonzero value if stack checking is done by the configuration files in a
2278 machine-dependent manner. You should define this macro if stack checking
2279 is require by the ABI of your machine or if you would like to have to stack
2280 checking in some more efficient way than GNU CC's portable approach.
2281 The default value of this macro is zero.
2283 @findex STACK_CHECK_PROBE_INTERVAL
2284 @item STACK_CHECK_PROBE_INTERVAL
2285 An integer representing the interval at which GNU CC must generate stack
2286 probe instructions. You will normally define this macro to be no larger
2287 than the size of the ``guard pages'' at the end of a stack area. The
2288 default value of 4096 is suitable for most systems.
2290 @findex STACK_CHECK_PROBE_LOAD
2291 @item STACK_CHECK_PROBE_LOAD
2292 A integer which is nonzero if GNU CC should perform the stack probe
2293 as a load instruction and zero if GNU CC should use a store instruction.
2294 The default is zero, which is the most efficient choice on most systems.
2296 @findex STACK_CHECK_PROTECT
2297 @item STACK_CHECK_PROTECT
2298 The number of bytes of stack needed to recover from a stack overflow,
2299 for languages where such a recovery is supported. The default value of
2300 75 words should be adequate for most machines.
2302 @findex STACK_CHECK_MAX_FRAME_SIZE
2303 @item STACK_CHECK_MAX_FRAME_SIZE
2304 The maximum size of a stack frame, in bytes. GNU CC will generate probe
2305 instructions in non-leaf functions to ensure at least this many bytes of
2306 stack are available. If a stack frame is larger than this size, stack
2307 checking will not be reliable and GNU CC will issue a warning. The
2308 default is chosen so that GNU CC only generates one instruction on most
2309 systems. You should normally not change the default value of this macro.
2311 @findex STACK_CHECK_FIXED_FRAME_SIZE
2312 @item STACK_CHECK_FIXED_FRAME_SIZE
2313 GNU CC uses this value to generate the above warning message. It
2314 represents the amount of fixed frame used by a function, not including
2315 space for any callee-saved registers, temporaries and user variables.
2316 You need only specify an upper bound for this amount and will normally
2317 use the default of four words.
2319 @findex STACK_CHECK_MAX_VAR_SIZE
2320 @item STACK_CHECK_MAX_VAR_SIZE
2321 The maximum size, in bytes, of an object that GNU CC will place in the
2322 fixed area of the stack frame when the user specifies
2323 @samp{-fstack-check}.
2324 GNU CC computed the default from the values of the above macros and you will
2325 normally not need to override that default.
2329 @node Frame Registers
2330 @subsection Registers That Address the Stack Frame
2332 @c prevent bad page break with this line
2333 This discusses registers that address the stack frame.
2336 @findex STACK_POINTER_REGNUM
2337 @item STACK_POINTER_REGNUM
2338 The register number of the stack pointer register, which must also be a
2339 fixed register according to @code{FIXED_REGISTERS}. On most machines,
2340 the hardware determines which register this is.
2342 @findex FRAME_POINTER_REGNUM
2343 @item FRAME_POINTER_REGNUM
2344 The register number of the frame pointer register, which is used to
2345 access automatic variables in the stack frame. On some machines, the
2346 hardware determines which register this is. On other machines, you can
2347 choose any register you wish for this purpose.
2349 @findex HARD_FRAME_POINTER_REGNUM
2350 @item HARD_FRAME_POINTER_REGNUM
2351 On some machines the offset between the frame pointer and starting
2352 offset of the automatic variables is not known until after register
2353 allocation has been done (for example, because the saved registers are
2354 between these two locations). On those machines, define
2355 @code{FRAME_POINTER_REGNUM} the number of a special, fixed register to
2356 be used internally until the offset is known, and define
2357 @code{HARD_FRAME_POINTER_REGNUM} to be the actual hard register number
2358 used for the frame pointer.
2360 You should define this macro only in the very rare circumstances when it
2361 is not possible to calculate the offset between the frame pointer and
2362 the automatic variables until after register allocation has been
2363 completed. When this macro is defined, you must also indicate in your
2364 definition of @code{ELIMINABLE_REGS} how to eliminate
2365 @code{FRAME_POINTER_REGNUM} into either @code{HARD_FRAME_POINTER_REGNUM}
2366 or @code{STACK_POINTER_REGNUM}.
2368 Do not define this macro if it would be the same as
2369 @code{FRAME_POINTER_REGNUM}.
2371 @findex ARG_POINTER_REGNUM
2372 @item ARG_POINTER_REGNUM
2373 The register number of the arg pointer register, which is used to access
2374 the function's argument list. On some machines, this is the same as the
2375 frame pointer register. On some machines, the hardware determines which
2376 register this is. On other machines, you can choose any register you
2377 wish for this purpose. If this is not the same register as the frame
2378 pointer register, then you must mark it as a fixed register according to
2379 @code{FIXED_REGISTERS}, or arrange to be able to eliminate it
2380 (@pxref{Elimination}).
2382 @findex RETURN_ADDRESS_POINTER_REGNUM
2383 @item RETURN_ADDRESS_POINTER_REGNUM
2384 The register number of the return address pointer register, which is used to
2385 access the current function's return address from the stack. On some
2386 machines, the return address is not at a fixed offset from the frame
2387 pointer or stack pointer or argument pointer. This register can be defined
2388 to point to the return address on the stack, and then be converted by
2389 @code{ELIMINABLE_REGS} into either the frame pointer or stack pointer.
2391 Do not define this macro unless there is no other way to get the return
2392 address from the stack.
2394 @findex STATIC_CHAIN_REGNUM
2395 @findex STATIC_CHAIN_INCOMING_REGNUM
2396 @item STATIC_CHAIN_REGNUM
2397 @itemx STATIC_CHAIN_INCOMING_REGNUM
2398 Register numbers used for passing a function's static chain pointer. If
2399 register windows are used, the register number as seen by the called
2400 function is @code{STATIC_CHAIN_INCOMING_REGNUM}, while the register
2401 number as seen by the calling function is @code{STATIC_CHAIN_REGNUM}. If
2402 these registers are the same, @code{STATIC_CHAIN_INCOMING_REGNUM} need
2403 not be defined.@refill
2405 The static chain register need not be a fixed register.
2407 If the static chain is passed in memory, these macros should not be
2408 defined; instead, the next two macros should be defined.
2410 @findex STATIC_CHAIN
2411 @findex STATIC_CHAIN_INCOMING
2413 @itemx STATIC_CHAIN_INCOMING
2414 If the static chain is passed in memory, these macros provide rtx giving
2415 @code{mem} expressions that denote where they are stored.
2416 @code{STATIC_CHAIN} and @code{STATIC_CHAIN_INCOMING} give the locations
2417 as seen by the calling and called functions, respectively. Often the former
2418 will be at an offset from the stack pointer and the latter at an offset from
2419 the frame pointer.@refill
2421 @findex stack_pointer_rtx
2422 @findex frame_pointer_rtx
2423 @findex arg_pointer_rtx
2424 The variables @code{stack_pointer_rtx}, @code{frame_pointer_rtx}, and
2425 @code{arg_pointer_rtx} will have been initialized prior to the use of these
2426 macros and should be used to refer to those items.
2428 If the static chain is passed in a register, the two previous macros should
2433 @subsection Eliminating Frame Pointer and Arg Pointer
2435 @c prevent bad page break with this line
2436 This is about eliminating the frame pointer and arg pointer.
2439 @findex FRAME_POINTER_REQUIRED
2440 @item FRAME_POINTER_REQUIRED
2441 A C expression which is nonzero if a function must have and use a frame
2442 pointer. This expression is evaluated in the reload pass. If its value is
2443 nonzero the function will have a frame pointer.
2445 The expression can in principle examine the current function and decide
2446 according to the facts, but on most machines the constant 0 or the
2447 constant 1 suffices. Use 0 when the machine allows code to be generated
2448 with no frame pointer, and doing so saves some time or space. Use 1
2449 when there is no possible advantage to avoiding a frame pointer.
2451 In certain cases, the compiler does not know how to produce valid code
2452 without a frame pointer. The compiler recognizes those cases and
2453 automatically gives the function a frame pointer regardless of what
2454 @code{FRAME_POINTER_REQUIRED} says. You don't need to worry about
2457 In a function that does not require a frame pointer, the frame pointer
2458 register can be allocated for ordinary usage, unless you mark it as a
2459 fixed register. See @code{FIXED_REGISTERS} for more information.
2461 @findex INITIAL_FRAME_POINTER_OFFSET
2462 @findex get_frame_size
2463 @item INITIAL_FRAME_POINTER_OFFSET (@var{depth-var})
2464 A C statement to store in the variable @var{depth-var} the difference
2465 between the frame pointer and the stack pointer values immediately after
2466 the function prologue. The value would be computed from information
2467 such as the result of @code{get_frame_size ()} and the tables of
2468 registers @code{regs_ever_live} and @code{call_used_regs}.
2470 If @code{ELIMINABLE_REGS} is defined, this macro will be not be used and
2471 need not be defined. Otherwise, it must be defined even if
2472 @code{FRAME_POINTER_REQUIRED} is defined to always be true; in that
2473 case, you may set @var{depth-var} to anything.
2475 @findex ELIMINABLE_REGS
2476 @item ELIMINABLE_REGS
2477 If defined, this macro specifies a table of register pairs used to
2478 eliminate unneeded registers that point into the stack frame. If it is not
2479 defined, the only elimination attempted by the compiler is to replace
2480 references to the frame pointer with references to the stack pointer.
2482 The definition of this macro is a list of structure initializations, each
2483 of which specifies an original and replacement register.
2485 On some machines, the position of the argument pointer is not known until
2486 the compilation is completed. In such a case, a separate hard register
2487 must be used for the argument pointer. This register can be eliminated by
2488 replacing it with either the frame pointer or the argument pointer,
2489 depending on whether or not the frame pointer has been eliminated.
2491 In this case, you might specify:
2493 #define ELIMINABLE_REGS \
2494 @{@{ARG_POINTER_REGNUM, STACK_POINTER_REGNUM@}, \
2495 @{ARG_POINTER_REGNUM, FRAME_POINTER_REGNUM@}, \
2496 @{FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM@}@}
2499 Note that the elimination of the argument pointer with the stack pointer is
2500 specified first since that is the preferred elimination.
2502 @findex CAN_ELIMINATE
2503 @item CAN_ELIMINATE (@var{from-reg}, @var{to-reg})
2504 A C expression that returns non-zero if the compiler is allowed to try
2505 to replace register number @var{from-reg} with register number
2506 @var{to-reg}. This macro need only be defined if @code{ELIMINABLE_REGS}
2507 is defined, and will usually be the constant 1, since most of the cases
2508 preventing register elimination are things that the compiler already
2511 @findex INITIAL_ELIMINATION_OFFSET
2512 @item INITIAL_ELIMINATION_OFFSET (@var{from-reg}, @var{to-reg}, @var{offset-var})
2513 This macro is similar to @code{INITIAL_FRAME_POINTER_OFFSET}. It
2514 specifies the initial difference between the specified pair of
2515 registers. This macro must be defined if @code{ELIMINABLE_REGS} is
2518 @findex LONGJMP_RESTORE_FROM_STACK
2519 @item LONGJMP_RESTORE_FROM_STACK
2520 Define this macro if the @code{longjmp} function restores registers from
2521 the stack frames, rather than from those saved specifically by
2522 @code{setjmp}. Certain quantities must not be kept in registers across
2523 a call to @code{setjmp} on such machines.
2526 @node Stack Arguments
2527 @subsection Passing Function Arguments on the Stack
2528 @cindex arguments on stack
2529 @cindex stack arguments
2531 The macros in this section control how arguments are passed
2532 on the stack. See the following section for other macros that
2533 control passing certain arguments in registers.
2536 @findex PROMOTE_PROTOTYPES
2537 @item PROMOTE_PROTOTYPES
2538 Define this macro if an argument declared in a prototype as an
2539 integral type smaller than @code{int} should actually be passed as an
2540 @code{int}. In addition to avoiding errors in certain cases of
2541 mismatch, it also makes for better code on certain machines.
2543 @findex PUSH_ROUNDING
2544 @item PUSH_ROUNDING (@var{npushed})
2545 A C expression that is the number of bytes actually pushed onto the
2546 stack when an instruction attempts to push @var{npushed} bytes.
2548 If the target machine does not have a push instruction, do not define
2549 this macro. That directs GNU CC to use an alternate strategy: to
2550 allocate the entire argument block and then store the arguments into
2553 On some machines, the definition
2556 #define PUSH_ROUNDING(BYTES) (BYTES)
2560 will suffice. But on other machines, instructions that appear
2561 to push one byte actually push two bytes in an attempt to maintain
2562 alignment. Then the definition should be
2565 #define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1)
2568 @findex ACCUMULATE_OUTGOING_ARGS
2569 @findex current_function_outgoing_args_size
2570 @item ACCUMULATE_OUTGOING_ARGS
2571 If defined, the maximum amount of space required for outgoing arguments
2572 will be computed and placed into the variable
2573 @code{current_function_outgoing_args_size}. No space will be pushed
2574 onto the stack for each call; instead, the function prologue should
2575 increase the stack frame size by this amount.
2577 Defining both @code{PUSH_ROUNDING} and @code{ACCUMULATE_OUTGOING_ARGS}
2580 @findex REG_PARM_STACK_SPACE
2581 @item REG_PARM_STACK_SPACE (@var{fndecl})
2582 Define this macro if functions should assume that stack space has been
2583 allocated for arguments even when their values are passed in
2586 The value of this macro is the size, in bytes, of the area reserved for
2587 arguments passed in registers for the function represented by @var{fndecl}.
2589 This space can be allocated by the caller, or be a part of the
2590 machine-dependent stack frame: @code{OUTGOING_REG_PARM_STACK_SPACE} says
2592 @c above is overfull. not sure what to do. --mew 5feb93 did
2593 @c something, not sure if it looks good. --mew 10feb93
2595 @findex MAYBE_REG_PARM_STACK_SPACE
2596 @findex FINAL_REG_PARM_STACK_SPACE
2597 @item MAYBE_REG_PARM_STACK_SPACE
2598 @itemx FINAL_REG_PARM_STACK_SPACE (@var{const_size}, @var{var_size})
2599 Define these macros in addition to the one above if functions might
2600 allocate stack space for arguments even when their values are passed
2601 in registers. These should be used when the stack space allocated
2602 for arguments in registers is not a simple constant independent of the
2603 function declaration.
2605 The value of the first macro is the size, in bytes, of the area that
2606 we should initially assume would be reserved for arguments passed in registers.
2608 The value of the second macro is the actual size, in bytes, of the area
2609 that will be reserved for arguments passed in registers. This takes two
2610 arguments: an integer representing the number of bytes of fixed sized
2611 arguments on the stack, and a tree representing the number of bytes of
2612 variable sized arguments on the stack.
2614 When these macros are defined, @code{REG_PARM_STACK_SPACE} will only be
2615 called for libcall functions, the current function, or for a function
2616 being called when it is known that such stack space must be allocated.
2617 In each case this value can be easily computed.
2619 When deciding whether a called function needs such stack space, and how
2620 much space to reserve, GNU CC uses these two macros instead of
2621 @code{REG_PARM_STACK_SPACE}.
2623 @findex OUTGOING_REG_PARM_STACK_SPACE
2624 @item OUTGOING_REG_PARM_STACK_SPACE
2625 Define this if it is the responsibility of the caller to allocate the area
2626 reserved for arguments passed in registers.
2628 If @code{ACCUMULATE_OUTGOING_ARGS} is defined, this macro controls
2629 whether the space for these arguments counts in the value of
2630 @code{current_function_outgoing_args_size}.
2632 @findex STACK_PARMS_IN_REG_PARM_AREA
2633 @item STACK_PARMS_IN_REG_PARM_AREA
2634 Define this macro if @code{REG_PARM_STACK_SPACE} is defined, but the
2635 stack parameters don't skip the area specified by it.
2636 @c i changed this, makes more sens and it should have taken care of the
2637 @c overfull.. not as specific, tho. --mew 5feb93
2639 Normally, when a parameter is not passed in registers, it is placed on the
2640 stack beyond the @code{REG_PARM_STACK_SPACE} area. Defining this macro
2641 suppresses this behavior and causes the parameter to be passed on the
2642 stack in its natural location.
2644 @findex RETURN_POPS_ARGS
2645 @item RETURN_POPS_ARGS (@var{fundecl}, @var{funtype}, @var{stack-size})
2646 A C expression that should indicate the number of bytes of its own
2647 arguments that a function pops on returning, or 0 if the
2648 function pops no arguments and the caller must therefore pop them all
2649 after the function returns.
2651 @var{fundecl} is a C variable whose value is a tree node that describes
2652 the function in question. Normally it is a node of type
2653 @code{FUNCTION_DECL} that describes the declaration of the function.
2654 From this you can obtain the DECL_MACHINE_ATTRIBUTES of the function.
2656 @var{funtype} is a C variable whose value is a tree node that
2657 describes the function in question. Normally it is a node of type
2658 @code{FUNCTION_TYPE} that describes the data type of the function.
2659 From this it is possible to obtain the data types of the value and
2660 arguments (if known).
2662 When a call to a library function is being considered, @var{fundecl}
2663 will contain an identifier node for the library function. Thus, if
2664 you need to distinguish among various library functions, you can do so
2665 by their names. Note that ``library function'' in this context means
2666 a function used to perform arithmetic, whose name is known specially
2667 in the compiler and was not mentioned in the C code being compiled.
2669 @var{stack-size} is the number of bytes of arguments passed on the
2670 stack. If a variable number of bytes is passed, it is zero, and
2671 argument popping will always be the responsibility of the calling function.
2673 On the Vax, all functions always pop their arguments, so the definition
2674 of this macro is @var{stack-size}. On the 68000, using the standard
2675 calling convention, no functions pop their arguments, so the value of
2676 the macro is always 0 in this case. But an alternative calling
2677 convention is available in which functions that take a fixed number of
2678 arguments pop them but other functions (such as @code{printf}) pop
2679 nothing (the caller pops all). When this convention is in use,
2680 @var{funtype} is examined to determine whether a function takes a fixed
2681 number of arguments.
2684 @node Register Arguments
2685 @subsection Passing Arguments in Registers
2686 @cindex arguments in registers
2687 @cindex registers arguments
2689 This section describes the macros which let you control how various
2690 types of arguments are passed in registers or how they are arranged in
2694 @findex FUNCTION_ARG
2695 @item FUNCTION_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
2696 A C expression that controls whether a function argument is passed
2697 in a register, and which register.
2699 The arguments are @var{cum}, which summarizes all the previous
2700 arguments; @var{mode}, the machine mode of the argument; @var{type},
2701 the data type of the argument as a tree node or 0 if that is not known
2702 (which happens for C support library functions); and @var{named},
2703 which is 1 for an ordinary argument and 0 for nameless arguments that
2704 correspond to @samp{@dots{}} in the called function's prototype.
2706 The value of the expression is usually either a @code{reg} RTX for the
2707 hard register in which to pass the argument, or zero to pass the
2708 argument on the stack.
2710 For machines like the Vax and 68000, where normally all arguments are
2711 pushed, zero suffices as a definition.
2713 The value of the expression can also be a @code{parallel} RTX. This is
2714 used when an argument is passed in multiple locations. The mode of the
2715 of the @code{parallel} should be the mode of the entire argument. The
2716 @code{parallel} holds any number of @code{expr_list} pairs; each one
2717 describes where part of the argument is passed. In each @code{expr_list},
2718 the first operand can be either a @code{reg} RTX for the hard register
2719 in which to pass this part of the argument, or zero to pass the argument
2720 on the stack. If this operand is a @code{reg}, then the mode indicates
2721 how large this part of the argument is. The second operand of the
2722 @code{expr_list} is a @code{const_int} which gives the offset in bytes
2723 into the entire argument where this part starts.
2725 @cindex @file{stdarg.h} and register arguments
2726 The usual way to make the ANSI library @file{stdarg.h} work on a machine
2727 where some arguments are usually passed in registers, is to cause
2728 nameless arguments to be passed on the stack instead. This is done
2729 by making @code{FUNCTION_ARG} return 0 whenever @var{named} is 0.
2731 @cindex @code{MUST_PASS_IN_STACK}, and @code{FUNCTION_ARG}
2732 @cindex @code{REG_PARM_STACK_SPACE}, and @code{FUNCTION_ARG}
2733 You may use the macro @code{MUST_PASS_IN_STACK (@var{mode}, @var{type})}
2734 in the definition of this macro to determine if this argument is of a
2735 type that must be passed in the stack. If @code{REG_PARM_STACK_SPACE}
2736 is not defined and @code{FUNCTION_ARG} returns non-zero for such an
2737 argument, the compiler will abort. If @code{REG_PARM_STACK_SPACE} is
2738 defined, the argument will be computed in the stack and then loaded into
2741 @findex MUST_PASS_IN_STACK
2742 @item MUST_PASS_IN_STACK (@var{mode}, @var{type})
2743 Define as a C expression that evaluates to nonzero if we do not know how
2744 to pass TYPE solely in registers. The file @file{expr.h} defines a
2745 definition that is usually appropriate, refer to @file{expr.h} for additional
2748 @findex FUNCTION_INCOMING_ARG
2749 @item FUNCTION_INCOMING_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
2750 Define this macro if the target machine has ``register windows'', so
2751 that the register in which a function sees an arguments is not
2752 necessarily the same as the one in which the caller passed the
2755 For such machines, @code{FUNCTION_ARG} computes the register in which
2756 the caller passes the value, and @code{FUNCTION_INCOMING_ARG} should
2757 be defined in a similar fashion to tell the function being called
2758 where the arguments will arrive.
2760 If @code{FUNCTION_INCOMING_ARG} is not defined, @code{FUNCTION_ARG}
2761 serves both purposes.@refill
2763 @findex FUNCTION_ARG_PARTIAL_NREGS
2764 @item FUNCTION_ARG_PARTIAL_NREGS (@var{cum}, @var{mode}, @var{type}, @var{named})
2765 A C expression for the number of words, at the beginning of an
2766 argument, must be put in registers. The value must be zero for
2767 arguments that are passed entirely in registers or that are entirely
2768 pushed on the stack.
2770 On some machines, certain arguments must be passed partially in
2771 registers and partially in memory. On these machines, typically the
2772 first @var{n} words of arguments are passed in registers, and the rest
2773 on the stack. If a multi-word argument (a @code{double} or a
2774 structure) crosses that boundary, its first few words must be passed
2775 in registers and the rest must be pushed. This macro tells the
2776 compiler when this occurs, and how many of the words should go in
2779 @code{FUNCTION_ARG} for these arguments should return the first
2780 register to be used by the caller for this argument; likewise
2781 @code{FUNCTION_INCOMING_ARG}, for the called function.
2783 @findex FUNCTION_ARG_PASS_BY_REFERENCE
2784 @item FUNCTION_ARG_PASS_BY_REFERENCE (@var{cum}, @var{mode}, @var{type}, @var{named})
2785 A C expression that indicates when an argument must be passed by reference.
2786 If nonzero for an argument, a copy of that argument is made in memory and a
2787 pointer to the argument is passed instead of the argument itself.
2788 The pointer is passed in whatever way is appropriate for passing a pointer
2791 On machines where @code{REG_PARM_STACK_SPACE} is not defined, a suitable
2792 definition of this macro might be
2794 #define FUNCTION_ARG_PASS_BY_REFERENCE\
2795 (CUM, MODE, TYPE, NAMED) \
2796 MUST_PASS_IN_STACK (MODE, TYPE)
2798 @c this is *still* too long. --mew 5feb93
2800 @findex FUNCTION_ARG_CALLEE_COPIES
2801 @item FUNCTION_ARG_CALLEE_COPIES (@var{cum}, @var{mode}, @var{type}, @var{named})
2802 If defined, a C expression that indicates when it is the called function's
2803 responsibility to make a copy of arguments passed by invisible reference.
2804 Normally, the caller makes a copy and passes the address of the copy to the
2805 routine being called. When FUNCTION_ARG_CALLEE_COPIES is defined and is
2806 nonzero, the caller does not make a copy. Instead, it passes a pointer to the
2807 ``live'' value. The called function must not modify this value. If it can be
2808 determined that the value won't be modified, it need not make a copy;
2809 otherwise a copy must be made.
2811 @findex CUMULATIVE_ARGS
2812 @item CUMULATIVE_ARGS
2813 A C type for declaring a variable that is used as the first argument of
2814 @code{FUNCTION_ARG} and other related values. For some target machines,
2815 the type @code{int} suffices and can hold the number of bytes of
2818 There is no need to record in @code{CUMULATIVE_ARGS} anything about the
2819 arguments that have been passed on the stack. The compiler has other
2820 variables to keep track of that. For target machines on which all
2821 arguments are passed on the stack, there is no need to store anything in
2822 @code{CUMULATIVE_ARGS}; however, the data structure must exist and
2823 should not be empty, so use @code{int}.
2825 @findex INIT_CUMULATIVE_ARGS
2826 @item INIT_CUMULATIVE_ARGS (@var{cum}, @var{fntype}, @var{libname}, @var{indirect})
2827 A C statement (sans semicolon) for initializing the variable @var{cum}
2828 for the state at the beginning of the argument list. The variable has
2829 type @code{CUMULATIVE_ARGS}. The value of @var{fntype} is the tree node
2830 for the data type of the function which will receive the args, or 0
2831 if the args are to a compiler support library function. The value of
2832 @var{indirect} is nonzero when processing an indirect call, for example
2833 a call through a function pointer. The value of @var{indirect} is zero
2834 for a call to an explicitly named function, a library function call, or when
2835 @code{INIT_CUMULATIVE_ARGS} is used to find arguments for the function
2838 When processing a call to a compiler support library function,
2839 @var{libname} identifies which one. It is a @code{symbol_ref} rtx which
2840 contains the name of the function, as a string. @var{libname} is 0 when
2841 an ordinary C function call is being processed. Thus, each time this
2842 macro is called, either @var{libname} or @var{fntype} is nonzero, but
2843 never both of them at once.
2845 @findex INIT_CUMULATIVE_INCOMING_ARGS
2846 @item INIT_CUMULATIVE_INCOMING_ARGS (@var{cum}, @var{fntype}, @var{libname})
2847 Like @code{INIT_CUMULATIVE_ARGS} but overrides it for the purposes of
2848 finding the arguments for the function being compiled. If this macro is
2849 undefined, @code{INIT_CUMULATIVE_ARGS} is used instead.
2851 The value passed for @var{libname} is always 0, since library routines
2852 with special calling conventions are never compiled with GNU CC. The
2853 argument @var{libname} exists for symmetry with
2854 @code{INIT_CUMULATIVE_ARGS}.
2855 @c could use "this macro" in place of @code{INIT_CUMULATIVE_ARGS}, maybe.
2856 @c --mew 5feb93 i switched the order of the sentences. --mew 10feb93
2858 @findex FUNCTION_ARG_ADVANCE
2859 @item FUNCTION_ARG_ADVANCE (@var{cum}, @var{mode}, @var{type}, @var{named})
2860 A C statement (sans semicolon) to update the summarizer variable
2861 @var{cum} to advance past an argument in the argument list. The
2862 values @var{mode}, @var{type} and @var{named} describe that argument.
2863 Once this is done, the variable @var{cum} is suitable for analyzing
2864 the @emph{following} argument with @code{FUNCTION_ARG}, etc.@refill
2866 This macro need not do anything if the argument in question was passed
2867 on the stack. The compiler knows how to track the amount of stack space
2868 used for arguments without any special help.
2870 @findex FUNCTION_ARG_PADDING
2871 @item FUNCTION_ARG_PADDING (@var{mode}, @var{type})
2872 If defined, a C expression which determines whether, and in which direction,
2873 to pad out an argument with extra space. The value should be of type
2874 @code{enum direction}: either @code{upward} to pad above the argument,
2875 @code{downward} to pad below, or @code{none} to inhibit padding.
2877 The @emph{amount} of padding is always just enough to reach the next
2878 multiple of @code{FUNCTION_ARG_BOUNDARY}; this macro does not control
2881 This macro has a default definition which is right for most systems.
2882 For little-endian machines, the default is to pad upward. For
2883 big-endian machines, the default is to pad downward for an argument of
2884 constant size shorter than an @code{int}, and upward otherwise.
2886 @findex FUNCTION_ARG_BOUNDARY
2887 @item FUNCTION_ARG_BOUNDARY (@var{mode}, @var{type})
2888 If defined, a C expression that gives the alignment boundary, in bits,
2889 of an argument with the specified mode and type. If it is not defined,
2890 @code{PARM_BOUNDARY} is used for all arguments.
2892 @findex FUNCTION_ARG_REGNO_P
2893 @item FUNCTION_ARG_REGNO_P (@var{regno})
2894 A C expression that is nonzero if @var{regno} is the number of a hard
2895 register in which function arguments are sometimes passed. This does
2896 @emph{not} include implicit arguments such as the static chain and
2897 the structure-value address. On many machines, no registers can be
2898 used for this purpose since all function arguments are pushed on the
2901 @findex LOAD_ARGS_REVERSED
2902 @item LOAD_ARGS_REVERSED
2903 If defined, the order in which arguments are loaded into their
2904 respective argument registers is reversed so that the last
2905 argument is loaded first. This macro only effects arguments
2906 passed in registers.
2911 @subsection How Scalar Function Values Are Returned
2912 @cindex return values in registers
2913 @cindex values, returned by functions
2914 @cindex scalars, returned as values
2916 This section discusses the macros that control returning scalars as
2917 values---values that can fit in registers.
2920 @findex TRADITIONAL_RETURN_FLOAT
2921 @item TRADITIONAL_RETURN_FLOAT
2922 Define this macro if @samp{-traditional} should not cause functions
2923 declared to return @code{float} to convert the value to @code{double}.
2925 @findex FUNCTION_VALUE
2926 @item FUNCTION_VALUE (@var{valtype}, @var{func})
2927 A C expression to create an RTX representing the place where a
2928 function returns a value of data type @var{valtype}. @var{valtype} is
2929 a tree node representing a data type. Write @code{TYPE_MODE
2930 (@var{valtype})} to get the machine mode used to represent that type.
2931 On many machines, only the mode is relevant. (Actually, on most
2932 machines, scalar values are returned in the same place regardless of
2935 The value of the expression is usually a @code{reg} RTX for the hard
2936 register where the return value is stored. The value can also be a
2937 @code{parallel} RTX, if the return value is in multiple places. See
2938 @code{FUNCTION_ARG} for an explanation of the @code{parallel} form.
2940 If @code{PROMOTE_FUNCTION_RETURN} is defined, you must apply the same
2941 promotion rules specified in @code{PROMOTE_MODE} if @var{valtype} is a
2944 If the precise function being called is known, @var{func} is a tree
2945 node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
2946 pointer. This makes it possible to use a different value-returning
2947 convention for specific functions when all their calls are
2950 @code{FUNCTION_VALUE} is not used for return vales with aggregate data
2951 types, because these are returned in another way. See
2952 @code{STRUCT_VALUE_REGNUM} and related macros, below.
2954 @findex FUNCTION_OUTGOING_VALUE
2955 @item FUNCTION_OUTGOING_VALUE (@var{valtype}, @var{func})
2956 Define this macro if the target machine has ``register windows''
2957 so that the register in which a function returns its value is not
2958 the same as the one in which the caller sees the value.
2960 For such machines, @code{FUNCTION_VALUE} computes the register in which
2961 the caller will see the value. @code{FUNCTION_OUTGOING_VALUE} should be
2962 defined in a similar fashion to tell the function where to put the
2965 If @code{FUNCTION_OUTGOING_VALUE} is not defined,
2966 @code{FUNCTION_VALUE} serves both purposes.@refill
2968 @code{FUNCTION_OUTGOING_VALUE} is not used for return vales with
2969 aggregate data types, because these are returned in another way. See
2970 @code{STRUCT_VALUE_REGNUM} and related macros, below.
2972 @findex LIBCALL_VALUE
2973 @item LIBCALL_VALUE (@var{mode})
2974 A C expression to create an RTX representing the place where a library
2975 function returns a value of mode @var{mode}. If the precise function
2976 being called is known, @var{func} is a tree node
2977 (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
2978 pointer. This makes it possible to use a different value-returning
2979 convention for specific functions when all their calls are
2982 Note that ``library function'' in this context means a compiler
2983 support routine, used to perform arithmetic, whose name is known
2984 specially by the compiler and was not mentioned in the C code being
2987 The definition of @code{LIBRARY_VALUE} need not be concerned aggregate
2988 data types, because none of the library functions returns such types.
2990 @findex FUNCTION_VALUE_REGNO_P
2991 @item FUNCTION_VALUE_REGNO_P (@var{regno})
2992 A C expression that is nonzero if @var{regno} is the number of a hard
2993 register in which the values of called function may come back.
2995 A register whose use for returning values is limited to serving as the
2996 second of a pair (for a value of type @code{double}, say) need not be
2997 recognized by this macro. So for most machines, this definition
3001 #define FUNCTION_VALUE_REGNO_P(N) ((N) == 0)
3004 If the machine has register windows, so that the caller and the called
3005 function use different registers for the return value, this macro
3006 should recognize only the caller's register numbers.
3008 @findex APPLY_RESULT_SIZE
3009 @item APPLY_RESULT_SIZE
3010 Define this macro if @samp{untyped_call} and @samp{untyped_return}
3011 need more space than is implied by @code{FUNCTION_VALUE_REGNO_P} for
3012 saving and restoring an arbitrary return value.
3015 @node Aggregate Return
3016 @subsection How Large Values Are Returned
3017 @cindex aggregates as return values
3018 @cindex large return values
3019 @cindex returning aggregate values
3020 @cindex structure value address
3022 When a function value's mode is @code{BLKmode} (and in some other
3023 cases), the value is not returned according to @code{FUNCTION_VALUE}
3024 (@pxref{Scalar Return}). Instead, the caller passes the address of a
3025 block of memory in which the value should be stored. This address
3026 is called the @dfn{structure value address}.
3028 This section describes how to control returning structure values in
3032 @findex RETURN_IN_MEMORY
3033 @item RETURN_IN_MEMORY (@var{type})
3034 A C expression which can inhibit the returning of certain function
3035 values in registers, based on the type of value. A nonzero value says
3036 to return the function value in memory, just as large structures are
3037 always returned. Here @var{type} will be a C expression of type
3038 @code{tree}, representing the data type of the value.
3040 Note that values of mode @code{BLKmode} must be explicitly handled
3041 by this macro. Also, the option @samp{-fpcc-struct-return}
3042 takes effect regardless of this macro. On most systems, it is
3043 possible to leave the macro undefined; this causes a default
3044 definition to be used, whose value is the constant 1 for @code{BLKmode}
3045 values, and 0 otherwise.
3047 Do not use this macro to indicate that structures and unions should always
3048 be returned in memory. You should instead use @code{DEFAULT_PCC_STRUCT_RETURN}
3051 @findex DEFAULT_PCC_STRUCT_RETURN
3052 @item DEFAULT_PCC_STRUCT_RETURN
3053 Define this macro to be 1 if all structure and union return values must be
3054 in memory. Since this results in slower code, this should be defined
3055 only if needed for compatibility with other compilers or with an ABI.
3056 If you define this macro to be 0, then the conventions used for structure
3057 and union return values are decided by the @code{RETURN_IN_MEMORY} macro.
3059 If not defined, this defaults to the value 1.
3061 @findex STRUCT_VALUE_REGNUM
3062 @item STRUCT_VALUE_REGNUM
3063 If the structure value address is passed in a register, then
3064 @code{STRUCT_VALUE_REGNUM} should be the number of that register.
3066 @findex STRUCT_VALUE
3068 If the structure value address is not passed in a register, define
3069 @code{STRUCT_VALUE} as an expression returning an RTX for the place
3070 where the address is passed. If it returns 0, the address is passed as
3071 an ``invisible'' first argument.
3073 @findex STRUCT_VALUE_INCOMING_REGNUM
3074 @item STRUCT_VALUE_INCOMING_REGNUM
3075 On some architectures the place where the structure value address
3076 is found by the called function is not the same place that the
3077 caller put it. This can be due to register windows, or it could
3078 be because the function prologue moves it to a different place.
3080 If the incoming location of the structure value address is in a
3081 register, define this macro as the register number.
3083 @findex STRUCT_VALUE_INCOMING
3084 @item STRUCT_VALUE_INCOMING
3085 If the incoming location is not a register, then you should define
3086 @code{STRUCT_VALUE_INCOMING} as an expression for an RTX for where the
3087 called function should find the value. If it should find the value on
3088 the stack, define this to create a @code{mem} which refers to the frame
3089 pointer. A definition of 0 means that the address is passed as an
3090 ``invisible'' first argument.
3092 @findex PCC_STATIC_STRUCT_RETURN
3093 @item PCC_STATIC_STRUCT_RETURN
3094 Define this macro if the usual system convention on the target machine
3095 for returning structures and unions is for the called function to return
3096 the address of a static variable containing the value.
3098 Do not define this if the usual system convention is for the caller to
3099 pass an address to the subroutine.
3101 This macro has effect in @samp{-fpcc-struct-return} mode, but it does
3102 nothing when you use @samp{-freg-struct-return} mode.
3106 @subsection Caller-Saves Register Allocation
3108 If you enable it, GNU CC can save registers around function calls. This
3109 makes it possible to use call-clobbered registers to hold variables that
3110 must live across calls.
3113 @findex DEFAULT_CALLER_SAVES
3114 @item DEFAULT_CALLER_SAVES
3115 Define this macro if function calls on the target machine do not preserve
3116 any registers; in other words, if @code{CALL_USED_REGISTERS} has 1
3117 for all registers. This macro enables @samp{-fcaller-saves} by default.
3118 Eventually that option will be enabled by default on all machines and both
3119 the option and this macro will be eliminated.
3121 @findex CALLER_SAVE_PROFITABLE
3122 @item CALLER_SAVE_PROFITABLE (@var{refs}, @var{calls})
3123 A C expression to determine whether it is worthwhile to consider placing
3124 a pseudo-register in a call-clobbered hard register and saving and
3125 restoring it around each function call. The expression should be 1 when
3126 this is worth doing, and 0 otherwise.
3128 If you don't define this macro, a default is used which is good on most
3129 machines: @code{4 * @var{calls} < @var{refs}}.
3132 @node Function Entry
3133 @subsection Function Entry and Exit
3134 @cindex function entry and exit
3138 This section describes the macros that output function entry
3139 (@dfn{prologue}) and exit (@dfn{epilogue}) code.
3142 @findex FUNCTION_PROLOGUE
3143 @item FUNCTION_PROLOGUE (@var{file}, @var{size})
3144 A C compound statement that outputs the assembler code for entry to a
3145 function. The prologue is responsible for setting up the stack frame,
3146 initializing the frame pointer register, saving registers that must be
3147 saved, and allocating @var{size} additional bytes of storage for the
3148 local variables. @var{size} is an integer. @var{file} is a stdio
3149 stream to which the assembler code should be output.
3151 The label for the beginning of the function need not be output by this
3152 macro. That has already been done when the macro is run.
3154 @findex regs_ever_live
3155 To determine which registers to save, the macro can refer to the array
3156 @code{regs_ever_live}: element @var{r} is nonzero if hard register
3157 @var{r} is used anywhere within the function. This implies the function
3158 prologue should save register @var{r}, provided it is not one of the
3159 call-used registers. (@code{FUNCTION_EPILOGUE} must likewise use
3160 @code{regs_ever_live}.)
3162 On machines that have ``register windows'', the function entry code does
3163 not save on the stack the registers that are in the windows, even if
3164 they are supposed to be preserved by function calls; instead it takes
3165 appropriate steps to ``push'' the register stack, if any non-call-used
3166 registers are used in the function.
3168 @findex frame_pointer_needed
3169 On machines where functions may or may not have frame-pointers, the
3170 function entry code must vary accordingly; it must set up the frame
3171 pointer if one is wanted, and not otherwise. To determine whether a
3172 frame pointer is in wanted, the macro can refer to the variable
3173 @code{frame_pointer_needed}. The variable's value will be 1 at run
3174 time in a function that needs a frame pointer. @xref{Elimination}.
3176 The function entry code is responsible for allocating any stack space
3177 required for the function. This stack space consists of the regions
3178 listed below. In most cases, these regions are allocated in the
3179 order listed, with the last listed region closest to the top of the
3180 stack (the lowest address if @code{STACK_GROWS_DOWNWARD} is defined, and
3181 the highest address if it is not defined). You can use a different order
3182 for a machine if doing so is more convenient or required for
3183 compatibility reasons. Except in cases where required by standard
3184 or by a debugger, there is no reason why the stack layout used by GCC
3185 need agree with that used by other compilers for a machine.
3189 @findex current_function_pretend_args_size
3190 A region of @code{current_function_pretend_args_size} bytes of
3191 uninitialized space just underneath the first argument arriving on the
3192 stack. (This may not be at the very start of the allocated stack region
3193 if the calling sequence has pushed anything else since pushing the stack
3194 arguments. But usually, on such machines, nothing else has been pushed
3195 yet, because the function prologue itself does all the pushing.) This
3196 region is used on machines where an argument may be passed partly in
3197 registers and partly in memory, and, in some cases to support the
3198 features in @file{varargs.h} and @file{stdargs.h}.
3201 An area of memory used to save certain registers used by the function.
3202 The size of this area, which may also include space for such things as
3203 the return address and pointers to previous stack frames, is
3204 machine-specific and usually depends on which registers have been used
3205 in the function. Machines with register windows often do not require
3209 A region of at least @var{size} bytes, possibly rounded up to an allocation
3210 boundary, to contain the local variables of the function. On some machines,
3211 this region and the save area may occur in the opposite order, with the
3212 save area closer to the top of the stack.
3215 @cindex @code{ACCUMULATE_OUTGOING_ARGS} and stack frames
3216 Optionally, when @code{ACCUMULATE_OUTGOING_ARGS} is defined, a region of
3217 @code{current_function_outgoing_args_size} bytes to be used for outgoing
3218 argument lists of the function. @xref{Stack Arguments}.
3221 Normally, it is necessary for the macros @code{FUNCTION_PROLOGUE} and
3222 @code{FUNCTION_EPILOGUE} to treat leaf functions specially. The C
3223 variable @code{leaf_function} is nonzero for such a function.
3225 @findex EXIT_IGNORE_STACK
3226 @item EXIT_IGNORE_STACK
3227 Define this macro as a C expression that is nonzero if the return
3228 instruction or the function epilogue ignores the value of the stack
3229 pointer; in other words, if it is safe to delete an instruction to
3230 adjust the stack pointer before a return from the function.
3232 Note that this macro's value is relevant only for functions for which
3233 frame pointers are maintained. It is never safe to delete a final
3234 stack adjustment in a function that has no frame pointer, and the
3235 compiler knows this regardless of @code{EXIT_IGNORE_STACK}.
3237 @findex EPILOGUE_USES
3238 @item EPILOGUE_USES (@var{regno})
3239 Define this macro as a C expression that is nonzero for registers are
3240 used by the epilogue or the @samp{return} pattern. The stack and frame
3241 pointer registers are already be assumed to be used as needed.
3243 @findex FUNCTION_EPILOGUE
3244 @item FUNCTION_EPILOGUE (@var{file}, @var{size})
3245 A C compound statement that outputs the assembler code for exit from a
3246 function. The epilogue is responsible for restoring the saved
3247 registers and stack pointer to their values when the function was
3248 called, and returning control to the caller. This macro takes the
3249 same arguments as the macro @code{FUNCTION_PROLOGUE}, and the
3250 registers to restore are determined from @code{regs_ever_live} and
3251 @code{CALL_USED_REGISTERS} in the same way.
3253 On some machines, there is a single instruction that does all the work
3254 of returning from the function. On these machines, give that
3255 instruction the name @samp{return} and do not define the macro
3256 @code{FUNCTION_EPILOGUE} at all.
3258 Do not define a pattern named @samp{return} if you want the
3259 @code{FUNCTION_EPILOGUE} to be used. If you want the target switches
3260 to control whether return instructions or epilogues are used, define a
3261 @samp{return} pattern with a validity condition that tests the target
3262 switches appropriately. If the @samp{return} pattern's validity
3263 condition is false, epilogues will be used.
3265 On machines where functions may or may not have frame-pointers, the
3266 function exit code must vary accordingly. Sometimes the code for these
3267 two cases is completely different. To determine whether a frame pointer
3268 is wanted, the macro can refer to the variable
3269 @code{frame_pointer_needed}. The variable's value will be 1 when compiling
3270 a function that needs a frame pointer.
3272 Normally, @code{FUNCTION_PROLOGUE} and @code{FUNCTION_EPILOGUE} must
3273 treat leaf functions specially. The C variable @code{leaf_function} is
3274 nonzero for such a function. @xref{Leaf Functions}.
3276 On some machines, some functions pop their arguments on exit while
3277 others leave that for the caller to do. For example, the 68020 when
3278 given @samp{-mrtd} pops arguments in functions that take a fixed
3279 number of arguments.
3281 @findex current_function_pops_args
3282 Your definition of the macro @code{RETURN_POPS_ARGS} decides which
3283 functions pop their own arguments. @code{FUNCTION_EPILOGUE} needs to
3284 know what was decided. The variable that is called
3285 @code{current_function_pops_args} is the number of bytes of its
3286 arguments that a function should pop. @xref{Scalar Return}.
3287 @c what is the "its arguments" in the above sentence referring to, pray
3288 @c tell? --mew 5feb93
3290 @findex DELAY_SLOTS_FOR_EPILOGUE
3291 @item DELAY_SLOTS_FOR_EPILOGUE
3292 Define this macro if the function epilogue contains delay slots to which
3293 instructions from the rest of the function can be ``moved''. The
3294 definition should be a C expression whose value is an integer
3295 representing the number of delay slots there.
3297 @findex ELIGIBLE_FOR_EPILOGUE_DELAY
3298 @item ELIGIBLE_FOR_EPILOGUE_DELAY (@var{insn}, @var{n})
3299 A C expression that returns 1 if @var{insn} can be placed in delay
3300 slot number @var{n} of the epilogue.
3302 The argument @var{n} is an integer which identifies the delay slot now
3303 being considered (since different slots may have different rules of
3304 eligibility). It is never negative and is always less than the number
3305 of epilogue delay slots (what @code{DELAY_SLOTS_FOR_EPILOGUE} returns).
3306 If you reject a particular insn for a given delay slot, in principle, it
3307 may be reconsidered for a subsequent delay slot. Also, other insns may
3308 (at least in principle) be considered for the so far unfilled delay
3311 @findex current_function_epilogue_delay_list
3312 @findex final_scan_insn
3313 The insns accepted to fill the epilogue delay slots are put in an RTL
3314 list made with @code{insn_list} objects, stored in the variable
3315 @code{current_function_epilogue_delay_list}. The insn for the first
3316 delay slot comes first in the list. Your definition of the macro
3317 @code{FUNCTION_EPILOGUE} should fill the delay slots by outputting the
3318 insns in this list, usually by calling @code{final_scan_insn}.
3320 You need not define this macro if you did not define
3321 @code{DELAY_SLOTS_FOR_EPILOGUE}.
3323 @findex ASM_OUTPUT_MI_THUNK
3324 @item ASM_OUTPUT_MI_THUNK (@var{file}, @var{thunk_fndecl}, @var{delta}, @var{function})
3325 A C compound statement that outputs the assembler code for a thunk
3326 function, used to implement C++ virtual function calls with multiple
3327 inheritance. The thunk acts as a wrapper around a virtual function,
3328 adjusting the implicit object parameter before handing control off to
3331 First, emit code to add the integer @var{delta} to the location that
3332 contains the incoming first argument. Assume that this argument
3333 contains a pointer, and is the one used to pass the @code{this} pointer
3334 in C++. This is the incoming argument @emph{before} the function prologue,
3335 e.g. @samp{%o0} on a sparc. The addition must preserve the values of
3336 all other incoming arguments.
3338 After the addition, emit code to jump to @var{function}, which is a
3339 @code{FUNCTION_DECL}. This is a direct pure jump, not a call, and does
3340 not touch the return address. Hence returning from @var{FUNCTION} will
3341 return to whoever called the current @samp{thunk}.
3343 The effect must be as if @var{function} had been called directly with
3344 the adjusted first argument. This macro is responsible for emitting all
3345 of the code for a thunk function; @code{FUNCTION_PROLOGUE} and
3346 @code{FUNCTION_EPILOGUE} are not invoked.
3348 The @var{thunk_fndecl} is redundant. (@var{delta} and @var{function}
3349 have already been extracted from it.) It might possibly be useful on
3350 some targets, but probably not.
3352 If you do not define this macro, the target-independent code in the C++
3353 frontend will generate a less efficient heavyweight thunk that calls
3354 @var{function} instead of jumping to it. The generic approach does
3355 not support varargs.
3359 @subsection Generating Code for Profiling
3360 @cindex profiling, code generation
3362 These macros will help you generate code for profiling.
3365 @findex FUNCTION_PROFILER
3366 @item FUNCTION_PROFILER (@var{file}, @var{labelno})
3367 A C statement or compound statement to output to @var{file} some
3368 assembler code to call the profiling subroutine @code{mcount}.
3369 Before calling, the assembler code must load the address of a
3370 counter variable into a register where @code{mcount} expects to
3371 find the address. The name of this variable is @samp{LP} followed
3372 by the number @var{labelno}, so you would generate the name using
3373 @samp{LP%d} in a @code{fprintf}.
3376 The details of how the address should be passed to @code{mcount} are
3377 determined by your operating system environment, not by GNU CC. To
3378 figure them out, compile a small program for profiling using the
3379 system's installed C compiler and look at the assembler code that
3382 @findex PROFILE_BEFORE_PROLOGUE
3383 @item PROFILE_BEFORE_PROLOGUE
3384 Define this macro if the code for function profiling should come before
3385 the function prologue. Normally, the profiling code comes after.
3387 @findex FUNCTION_BLOCK_PROFILER
3388 @vindex profile_block_flag
3389 @item FUNCTION_BLOCK_PROFILER (@var{file}, @var{labelno})
3390 A C statement or compound statement to output to @var{file} some
3391 assembler code to initialize basic-block profiling for the current
3392 object module. The global compile flag @code{profile_block_flag}
3393 distinguishes two profile modes.
3396 @findex __bb_init_func
3397 @item profile_block_flag != 2
3398 Output code to call the subroutine @code{__bb_init_func} once per
3399 object module, passing it as its sole argument the address of a block
3400 allocated in the object module.
3402 The name of the block is a local symbol made with this statement:
3405 ASM_GENERATE_INTERNAL_LABEL (@var{buffer}, "LPBX", 0);
3408 Of course, since you are writing the definition of
3409 @code{ASM_GENERATE_INTERNAL_LABEL} as well as that of this macro, you
3410 can take a short cut in the definition of this macro and use the name
3411 that you know will result.
3413 The first word of this block is a flag which will be nonzero if the
3414 object module has already been initialized. So test this word first,
3415 and do not call @code{__bb_init_func} if the flag is
3416 nonzero. BLOCK_OR_LABEL contains a unique number which may be used to
3417 generate a label as a branch destination when @code{__bb_init_func}
3420 Described in assembler language, the code to be output looks like:
3430 @findex __bb_init_trace_func
3431 @item profile_block_flag == 2
3432 Output code to call the subroutine @code{__bb_init_trace_func}
3433 and pass two parameters to it. The first parameter is the same as
3434 for @code{__bb_init_func}. The second parameter is the number of the
3435 first basic block of the function as given by BLOCK_OR_LABEL. Note
3436 that @code{__bb_init_trace_func} has to be called, even if the object
3437 module has been initialized already.
3439 Described in assembler language, the code to be output looks like:
3442 parameter2 <- BLOCK_OR_LABEL
3443 call __bb_init_trace_func
3447 @findex BLOCK_PROFILER
3448 @vindex profile_block_flag
3449 @item BLOCK_PROFILER (@var{file}, @var{blockno})
3450 A C statement or compound statement to output to @var{file} some
3451 assembler code to increment the count associated with the basic
3452 block number @var{blockno}. The global compile flag
3453 @code{profile_block_flag} distinguishes two profile modes.
3456 @item profile_block_flag != 2
3457 Output code to increment the counter directly. Basic blocks are
3458 numbered separately from zero within each compilation. The count
3459 associated with block number @var{blockno} is at index
3460 @var{blockno} in a vector of words; the name of this array is a local
3461 symbol made with this statement:
3464 ASM_GENERATE_INTERNAL_LABEL (@var{buffer}, "LPBX", 2);
3467 @c This paragraph is the same as one a few paragraphs up.
3468 @c That is not an error.
3469 Of course, since you are writing the definition of
3470 @code{ASM_GENERATE_INTERNAL_LABEL} as well as that of this macro, you
3471 can take a short cut in the definition of this macro and use the name
3472 that you know will result.
3474 Described in assembler language, the code to be output looks like:
3477 inc (LPBX2+4*BLOCKNO)
3481 @findex __bb_trace_func
3482 @item profile_block_flag == 2
3483 Output code to initialize the global structure @code{__bb} and
3484 call the function @code{__bb_trace_func}, which will increment the
3487 @code{__bb} consists of two words. In the first word, the current
3488 basic block number, as given by BLOCKNO, has to be stored. In
3489 the second word, the address of a block allocated in the object
3490 module has to be stored. The address is given by the label created
3491 with this statement:
3494 ASM_GENERATE_INTERNAL_LABEL (@var{buffer}, "LPBX", 0);
3497 Described in assembler language, the code to be output looks like:
3499 move BLOCKNO -> (__bb)
3500 move LPBX0 -> (__bb+4)
3501 call __bb_trace_func
3505 @findex FUNCTION_BLOCK_PROFILER_EXIT
3506 @findex __bb_trace_ret
3507 @vindex profile_block_flag
3508 @item FUNCTION_BLOCK_PROFILER_EXIT (@var{file})
3509 A C statement or compound statement to output to @var{file}
3510 assembler code to call function @code{__bb_trace_ret}. The
3511 assembler code should only be output
3512 if the global compile flag @code{profile_block_flag} == 2. This
3513 macro has to be used at every place where code for returning from
3514 a function is generated (e.g. @code{FUNCTION_EPILOGUE}). Although
3515 you have to write the definition of @code{FUNCTION_EPILOGUE}
3516 as well, you have to define this macro to tell the compiler, that
3517 the proper call to @code{__bb_trace_ret} is produced.
3519 @findex MACHINE_STATE_SAVE
3520 @findex __bb_init_trace_func
3521 @findex __bb_trace_func
3522 @findex __bb_trace_ret
3523 @item MACHINE_STATE_SAVE (@var{id})
3524 A C statement or compound statement to save all registers, which may
3525 be clobbered by a function call, including condition codes. The
3526 @code{asm} statement will be mostly likely needed to handle this
3527 task. Local labels in the assembler code can be concatenated with the
3528 string @var{id}, to obtain a unique lable name.
3530 Registers or condition codes clobbered by @code{FUNCTION_PROLOGUE} or
3531 @code{FUNCTION_EPILOGUE} must be saved in the macros
3532 @code{FUNCTION_BLOCK_PROFILER}, @code{FUNCTION_BLOCK_PROFILER_EXIT} and
3533 @code{BLOCK_PROFILER} prior calling @code{__bb_init_trace_func},
3534 @code{__bb_trace_ret} and @code{__bb_trace_func} respectively.
3536 @findex MACHINE_STATE_RESTORE
3537 @findex __bb_init_trace_func
3538 @findex __bb_trace_func
3539 @findex __bb_trace_ret
3540 @item MACHINE_STATE_RESTORE (@var{id})
3541 A C statement or compound statement to restore all registers, including
3542 condition codes, saved by @code{MACHINE_STATE_SAVE}.
3544 Registers or condition codes clobbered by @code{FUNCTION_PROLOGUE} or
3545 @code{FUNCTION_EPILOGUE} must be restored in the macros
3546 @code{FUNCTION_BLOCK_PROFILER}, @code{FUNCTION_BLOCK_PROFILER_EXIT} and
3547 @code{BLOCK_PROFILER} after calling @code{__bb_init_trace_func},
3548 @code{__bb_trace_ret} and @code{__bb_trace_func} respectively.
3550 @findex BLOCK_PROFILER_CODE
3551 @item BLOCK_PROFILER_CODE
3552 A C function or functions which are needed in the library to
3553 support block profiling.
3557 @section Implementing the Varargs Macros
3558 @cindex varargs implementation
3560 GNU CC comes with an implementation of @file{varargs.h} and
3561 @file{stdarg.h} that work without change on machines that pass arguments
3562 on the stack. Other machines require their own implementations of
3563 varargs, and the two machine independent header files must have
3564 conditionals to include it.
3566 ANSI @file{stdarg.h} differs from traditional @file{varargs.h} mainly in
3567 the calling convention for @code{va_start}. The traditional
3568 implementation takes just one argument, which is the variable in which
3569 to store the argument pointer. The ANSI implementation of
3570 @code{va_start} takes an additional second argument. The user is
3571 supposed to write the last named argument of the function here.
3573 However, @code{va_start} should not use this argument. The way to find
3574 the end of the named arguments is with the built-in functions described
3578 @findex __builtin_saveregs
3579 @item __builtin_saveregs ()
3580 Use this built-in function to save the argument registers in memory so
3581 that the varargs mechanism can access them. Both ANSI and traditional
3582 versions of @code{va_start} must use @code{__builtin_saveregs}, unless
3583 you use @code{SETUP_INCOMING_VARARGS} (see below) instead.
3585 On some machines, @code{__builtin_saveregs} is open-coded under the
3586 control of the macro @code{EXPAND_BUILTIN_SAVEREGS}. On other machines,
3587 it calls a routine written in assembler language, found in
3590 Code generated for the call to @code{__builtin_saveregs} appears at the
3591 beginning of the function, as opposed to where the call to
3592 @code{__builtin_saveregs} is written, regardless of what the code is.
3593 This is because the registers must be saved before the function starts
3594 to use them for its own purposes.
3595 @c i rewrote the first sentence above to fix an overfull hbox. --mew
3598 @findex __builtin_args_info
3599 @item __builtin_args_info (@var{category})
3600 Use this built-in function to find the first anonymous arguments in
3603 In general, a machine may have several categories of registers used for
3604 arguments, each for a particular category of data types. (For example,
3605 on some machines, floating-point registers are used for floating-point
3606 arguments while other arguments are passed in the general registers.)
3607 To make non-varargs functions use the proper calling convention, you
3608 have defined the @code{CUMULATIVE_ARGS} data type to record how many
3609 registers in each category have been used so far
3611 @code{__builtin_args_info} accesses the same data structure of type
3612 @code{CUMULATIVE_ARGS} after the ordinary argument layout is finished
3613 with it, with @var{category} specifying which word to access. Thus, the
3614 value indicates the first unused register in a given category.
3616 Normally, you would use @code{__builtin_args_info} in the implementation
3617 of @code{va_start}, accessing each category just once and storing the
3618 value in the @code{va_list} object. This is because @code{va_list} will
3619 have to update the values, and there is no way to alter the
3620 values accessed by @code{__builtin_args_info}.
3622 @findex __builtin_next_arg
3623 @item __builtin_next_arg (@var{lastarg})
3624 This is the equivalent of @code{__builtin_args_info}, for stack
3625 arguments. It returns the address of the first anonymous stack
3626 argument, as type @code{void *}. If @code{ARGS_GROW_DOWNWARD}, it
3627 returns the address of the location above the first anonymous stack
3628 argument. Use it in @code{va_start} to initialize the pointer for
3629 fetching arguments from the stack. Also use it in @code{va_start} to
3630 verify that the second parameter @var{lastarg} is the last named argument
3631 of the current function.
3633 @findex __builtin_classify_type
3634 @item __builtin_classify_type (@var{object})
3635 Since each machine has its own conventions for which data types are
3636 passed in which kind of register, your implementation of @code{va_arg}
3637 has to embody these conventions. The easiest way to categorize the
3638 specified data type is to use @code{__builtin_classify_type} together
3639 with @code{sizeof} and @code{__alignof__}.
3641 @code{__builtin_classify_type} ignores the value of @var{object},
3642 considering only its data type. It returns an integer describing what
3643 kind of type that is---integer, floating, pointer, structure, and so on.
3645 The file @file{typeclass.h} defines an enumeration that you can use to
3646 interpret the values of @code{__builtin_classify_type}.
3649 These machine description macros help implement varargs:
3652 @findex EXPAND_BUILTIN_SAVEREGS
3653 @item EXPAND_BUILTIN_SAVEREGS (@var{args})
3654 If defined, is a C expression that produces the machine-specific code
3655 for a call to @code{__builtin_saveregs}. This code will be moved to the
3656 very beginning of the function, before any parameter access are made.
3657 The return value of this function should be an RTX that contains the
3658 value to use as the return of @code{__builtin_saveregs}.
3660 The argument @var{args} is a @code{tree_list} containing the arguments
3661 that were passed to @code{__builtin_saveregs}.
3663 If this macro is not defined, the compiler will output an ordinary
3664 call to the library function @samp{__builtin_saveregs}.
3666 @c !!! a bug in texinfo; how to make the entry on the @item line allow
3667 @c more than one line of text... help... --mew 10feb93
3668 @findex SETUP_INCOMING_VARARGS
3669 @item SETUP_INCOMING_VARARGS (@var{args_so_far}, @var{mode}, @var{type},
3670 @var{pretend_args_size}, @var{second_time})
3671 This macro offers an alternative to using @code{__builtin_saveregs} and
3672 defining the macro @code{EXPAND_BUILTIN_SAVEREGS}. Use it to store the
3673 anonymous register arguments into the stack so that all the arguments
3674 appear to have been passed consecutively on the stack. Once this is
3675 done, you can use the standard implementation of varargs that works for
3676 machines that pass all their arguments on the stack.
3678 The argument @var{args_so_far} is the @code{CUMULATIVE_ARGS} data
3679 structure, containing the values that obtain after processing of the
3680 named arguments. The arguments @var{mode} and @var{type} describe the
3681 last named argument---its machine mode and its data type as a tree node.
3683 The macro implementation should do two things: first, push onto the
3684 stack all the argument registers @emph{not} used for the named
3685 arguments, and second, store the size of the data thus pushed into the
3686 @code{int}-valued variable whose name is supplied as the argument
3687 @var{pretend_args_size}. The value that you store here will serve as
3688 additional offset for setting up the stack frame.
3690 Because you must generate code to push the anonymous arguments at
3691 compile time without knowing their data types,
3692 @code{SETUP_INCOMING_VARARGS} is only useful on machines that have just
3693 a single category of argument register and use it uniformly for all data
3696 If the argument @var{second_time} is nonzero, it means that the
3697 arguments of the function are being analyzed for the second time. This
3698 happens for an inline function, which is not actually compiled until the
3699 end of the source file. The macro @code{SETUP_INCOMING_VARARGS} should
3700 not generate any instructions in this case.
3702 @findex STRICT_ARGUMENT_NAMING
3703 @item STRICT_ARGUMENT_NAMING
3704 Define this macro to be a nonzero value if the location where a function
3705 argument is passed depends on whether or not it is a named argument.
3707 This macro controls how the @var{named} argument to @code{FUNCTION_ARG}
3708 is set for varargs and stdarg functions. If this macro returns a
3709 nonzero value, the @var{named} argument is always true for named
3710 arguments, and false for unnamed arguments. If it returns a value of
3711 zero, but @code{SETUP_INCOMING_VARARGS} is defined, then all arguments
3712 are treated as named. Otherwise, all named arguments except the last
3713 are treated as named.
3715 You need not define this macro if it always returns zero.
3719 @section Trampolines for Nested Functions
3720 @cindex trampolines for nested functions
3721 @cindex nested functions, trampolines for
3723 A @dfn{trampoline} is a small piece of code that is created at run time
3724 when the address of a nested function is taken. It normally resides on
3725 the stack, in the stack frame of the containing function. These macros
3726 tell GNU CC how to generate code to allocate and initialize a
3729 The instructions in the trampoline must do two things: load a constant
3730 address into the static chain register, and jump to the real address of
3731 the nested function. On CISC machines such as the m68k, this requires
3732 two instructions, a move immediate and a jump. Then the two addresses
3733 exist in the trampoline as word-long immediate operands. On RISC
3734 machines, it is often necessary to load each address into a register in
3735 two parts. Then pieces of each address form separate immediate
3738 The code generated to initialize the trampoline must store the variable
3739 parts---the static chain value and the function address---into the
3740 immediate operands of the instructions. On a CISC machine, this is
3741 simply a matter of copying each address to a memory reference at the
3742 proper offset from the start of the trampoline. On a RISC machine, it
3743 may be necessary to take out pieces of the address and store them
3747 @findex TRAMPOLINE_TEMPLATE
3748 @item TRAMPOLINE_TEMPLATE (@var{file})
3749 A C statement to output, on the stream @var{file}, assembler code for a
3750 block of data that contains the constant parts of a trampoline. This
3751 code should not include a label---the label is taken care of
3754 If you do not define this macro, it means no template is needed
3755 for the target. Do not define this macro on systems where the block move
3756 code to copy the trampoline into place would be larger than the code
3757 to generate it on the spot.
3759 @findex TRAMPOLINE_SECTION
3760 @item TRAMPOLINE_SECTION
3761 The name of a subroutine to switch to the section in which the
3762 trampoline template is to be placed (@pxref{Sections}). The default is
3763 a value of @samp{readonly_data_section}, which places the trampoline in
3764 the section containing read-only data.
3766 @findex TRAMPOLINE_SIZE
3767 @item TRAMPOLINE_SIZE
3768 A C expression for the size in bytes of the trampoline, as an integer.
3770 @findex TRAMPOLINE_ALIGNMENT
3771 @item TRAMPOLINE_ALIGNMENT
3772 Alignment required for trampolines, in bits.
3774 If you don't define this macro, the value of @code{BIGGEST_ALIGNMENT}
3775 is used for aligning trampolines.
3777 @findex INITIALIZE_TRAMPOLINE
3778 @item INITIALIZE_TRAMPOLINE (@var{addr}, @var{fnaddr}, @var{static_chain})
3779 A C statement to initialize the variable parts of a trampoline.
3780 @var{addr} is an RTX for the address of the trampoline; @var{fnaddr} is
3781 an RTX for the address of the nested function; @var{static_chain} is an
3782 RTX for the static chain value that should be passed to the function
3785 @findex ALLOCATE_TRAMPOLINE
3786 @item ALLOCATE_TRAMPOLINE (@var{fp})
3787 A C expression to allocate run-time space for a trampoline. The
3788 expression value should be an RTX representing a memory reference to the
3789 space for the trampoline.
3791 @cindex @code{FUNCTION_EPILOGUE} and trampolines
3792 @cindex @code{FUNCTION_PROLOGUE} and trampolines
3793 If this macro is not defined, by default the trampoline is allocated as
3794 a stack slot. This default is right for most machines. The exceptions
3795 are machines where it is impossible to execute instructions in the stack
3796 area. On such machines, you may have to implement a separate stack,
3797 using this macro in conjunction with @code{FUNCTION_PROLOGUE} and
3798 @code{FUNCTION_EPILOGUE}.
3800 @var{fp} points to a data structure, a @code{struct function}, which
3801 describes the compilation status of the immediate containing function of
3802 the function which the trampoline is for. Normally (when
3803 @code{ALLOCATE_TRAMPOLINE} is not defined), the stack slot for the
3804 trampoline is in the stack frame of this containing function. Other
3805 allocation strategies probably must do something analogous with this
3809 Implementing trampolines is difficult on many machines because they have
3810 separate instruction and data caches. Writing into a stack location
3811 fails to clear the memory in the instruction cache, so when the program
3812 jumps to that location, it executes the old contents.
3814 Here are two possible solutions. One is to clear the relevant parts of
3815 the instruction cache whenever a trampoline is set up. The other is to
3816 make all trampolines identical, by having them jump to a standard
3817 subroutine. The former technique makes trampoline execution faster; the
3818 latter makes initialization faster.
3820 To clear the instruction cache when a trampoline is initialized, define
3821 the following macros which describe the shape of the cache.
3824 @findex INSN_CACHE_SIZE
3825 @item INSN_CACHE_SIZE
3826 The total size in bytes of the cache.
3828 @findex INSN_CACHE_LINE_WIDTH
3829 @item INSN_CACHE_LINE_WIDTH
3830 The length in bytes of each cache line. The cache is divided into cache
3831 lines which are disjoint slots, each holding a contiguous chunk of data
3832 fetched from memory. Each time data is brought into the cache, an
3833 entire line is read at once. The data loaded into a cache line is
3834 always aligned on a boundary equal to the line size.
3836 @findex INSN_CACHE_DEPTH
3837 @item INSN_CACHE_DEPTH
3838 The number of alternative cache lines that can hold any particular memory
3842 Alternatively, if the machine has system calls or instructions to clear
3843 the instruction cache directly, you can define the following macro.
3846 @findex CLEAR_INSN_CACHE
3847 @item CLEAR_INSN_CACHE (@var{BEG}, @var{END})
3848 If defined, expands to a C expression clearing the @emph{instruction
3849 cache} in the specified interval. If it is not defined, and the macro
3850 INSN_CACHE_SIZE is defined, some generic code is generated to clear the
3851 cache. The definition of this macro would typically be a series of
3852 @code{asm} statements. Both @var{BEG} and @var{END} are both pointer
3856 To use a standard subroutine, define the following macro. In addition,
3857 you must make sure that the instructions in a trampoline fill an entire
3858 cache line with identical instructions, or else ensure that the
3859 beginning of the trampoline code is always aligned at the same point in
3860 its cache line. Look in @file{m68k.h} as a guide.
3863 @findex TRANSFER_FROM_TRAMPOLINE
3864 @item TRANSFER_FROM_TRAMPOLINE
3865 Define this macro if trampolines need a special subroutine to do their
3866 work. The macro should expand to a series of @code{asm} statements
3867 which will be compiled with GNU CC. They go in a library function named
3868 @code{__transfer_from_trampoline}.
3870 If you need to avoid executing the ordinary prologue code of a compiled
3871 C function when you jump to the subroutine, you can do so by placing a
3872 special label of your own in the assembler code. Use one @code{asm}
3873 statement to generate an assembler label, and another to make the label
3874 global. Then trampolines can use that label to jump directly to your
3875 special assembler code.
3879 @section Implicit Calls to Library Routines
3880 @cindex library subroutine names
3881 @cindex @file{libgcc.a}
3883 @c prevent bad page break with this line
3884 Here is an explanation of implicit calls to library routines.
3887 @findex MULSI3_LIBCALL
3888 @item MULSI3_LIBCALL
3889 A C string constant giving the name of the function to call for
3890 multiplication of one signed full-word by another. If you do not
3891 define this macro, the default name is used, which is @code{__mulsi3},
3892 a function defined in @file{libgcc.a}.
3894 @findex DIVSI3_LIBCALL
3895 @item DIVSI3_LIBCALL
3896 A C string constant giving the name of the function to call for
3897 division of one signed full-word by another. If you do not define
3898 this macro, the default name is used, which is @code{__divsi3}, a
3899 function defined in @file{libgcc.a}.
3901 @findex UDIVSI3_LIBCALL
3902 @item UDIVSI3_LIBCALL
3903 A C string constant giving the name of the function to call for
3904 division of one unsigned full-word by another. If you do not define
3905 this macro, the default name is used, which is @code{__udivsi3}, a
3906 function defined in @file{libgcc.a}.
3908 @findex MODSI3_LIBCALL
3909 @item MODSI3_LIBCALL
3910 A C string constant giving the name of the function to call for the
3911 remainder in division of one signed full-word by another. If you do
3912 not define this macro, the default name is used, which is
3913 @code{__modsi3}, a function defined in @file{libgcc.a}.
3915 @findex UMODSI3_LIBCALL
3916 @item UMODSI3_LIBCALL
3917 A C string constant giving the name of the function to call for the
3918 remainder in division of one unsigned full-word by another. If you do
3919 not define this macro, the default name is used, which is
3920 @code{__umodsi3}, a function defined in @file{libgcc.a}.
3922 @findex MULDI3_LIBCALL
3923 @item MULDI3_LIBCALL
3924 A C string constant giving the name of the function to call for
3925 multiplication of one signed double-word by another. If you do not
3926 define this macro, the default name is used, which is @code{__muldi3},
3927 a function defined in @file{libgcc.a}.
3929 @findex DIVDI3_LIBCALL
3930 @item DIVDI3_LIBCALL
3931 A C string constant giving the name of the function to call for
3932 division of one signed double-word by another. If you do not define
3933 this macro, the default name is used, which is @code{__divdi3}, a
3934 function defined in @file{libgcc.a}.
3936 @findex UDIVDI3_LIBCALL
3937 @item UDIVDI3_LIBCALL
3938 A C string constant giving the name of the function to call for
3939 division of one unsigned full-word by another. If you do not define
3940 this macro, the default name is used, which is @code{__udivdi3}, a
3941 function defined in @file{libgcc.a}.
3943 @findex MODDI3_LIBCALL
3944 @item MODDI3_LIBCALL
3945 A C string constant giving the name of the function to call for the
3946 remainder in division of one signed double-word by another. If you do
3947 not define this macro, the default name is used, which is
3948 @code{__moddi3}, a function defined in @file{libgcc.a}.
3950 @findex UMODDI3_LIBCALL
3951 @item UMODDI3_LIBCALL
3952 A C string constant giving the name of the function to call for the
3953 remainder in division of one unsigned full-word by another. If you do
3954 not define this macro, the default name is used, which is
3955 @code{__umoddi3}, a function defined in @file{libgcc.a}.
3957 @findex INIT_TARGET_OPTABS
3958 @item INIT_TARGET_OPTABS
3959 Define this macro as a C statement that declares additional library
3960 routines renames existing ones. @code{init_optabs} calls this macro after
3961 initializing all the normal library routines.
3964 @cindex @code{EDOM}, implicit usage
3966 The value of @code{EDOM} on the target machine, as a C integer constant
3967 expression. If you don't define this macro, GNU CC does not attempt to
3968 deposit the value of @code{EDOM} into @code{errno} directly. Look in
3969 @file{/usr/include/errno.h} to find the value of @code{EDOM} on your
3972 If you do not define @code{TARGET_EDOM}, then compiled code reports
3973 domain errors by calling the library function and letting it report the
3974 error. If mathematical functions on your system use @code{matherr} when
3975 there is an error, then you should leave @code{TARGET_EDOM} undefined so
3976 that @code{matherr} is used normally.
3978 @findex GEN_ERRNO_RTX
3979 @cindex @code{errno}, implicit usage
3981 Define this macro as a C expression to create an rtl expression that
3982 refers to the global ``variable'' @code{errno}. (On certain systems,
3983 @code{errno} may not actually be a variable.) If you don't define this
3984 macro, a reasonable default is used.
3986 @findex TARGET_MEM_FUNCTIONS
3987 @cindex @code{bcopy}, implicit usage
3988 @cindex @code{memcpy}, implicit usage
3989 @cindex @code{bzero}, implicit usage
3990 @cindex @code{memset}, implicit usage
3991 @item TARGET_MEM_FUNCTIONS
3992 Define this macro if GNU CC should generate calls to the System V
3993 (and ANSI C) library functions @code{memcpy} and @code{memset}
3994 rather than the BSD functions @code{bcopy} and @code{bzero}.
3996 @findex LIBGCC_NEEDS_DOUBLE
3997 @item LIBGCC_NEEDS_DOUBLE
3998 Define this macro if only @code{float} arguments cannot be passed to
3999 library routines (so they must be converted to @code{double}). This
4000 macro affects both how library calls are generated and how the library
4001 routines in @file{libgcc1.c} accept their arguments. It is useful on
4002 machines where floating and fixed point arguments are passed
4003 differently, such as the i860.
4005 @findex FLOAT_ARG_TYPE
4006 @item FLOAT_ARG_TYPE
4007 Define this macro to override the type used by the library routines to
4008 pick up arguments of type @code{float}. (By default, they use a union
4009 of @code{float} and @code{int}.)
4011 The obvious choice would be @code{float}---but that won't work with
4012 traditional C compilers that expect all arguments declared as @code{float}
4013 to arrive as @code{double}. To avoid this conversion, the library routines
4014 ask for the value as some other type and then treat it as a @code{float}.
4016 On some systems, no other type will work for this. For these systems,
4017 you must use @code{LIBGCC_NEEDS_DOUBLE} instead, to force conversion of
4018 the values @code{double} before they are passed.
4021 @item FLOATIFY (@var{passed-value})
4022 Define this macro to override the way library routines redesignate a
4023 @code{float} argument as a @code{float} instead of the type it was
4024 passed as. The default is an expression which takes the @code{float}
4027 @findex FLOAT_VALUE_TYPE
4028 @item FLOAT_VALUE_TYPE
4029 Define this macro to override the type used by the library routines to
4030 return values that ought to have type @code{float}. (By default, they
4033 The obvious choice would be @code{float}---but that won't work with
4034 traditional C compilers gratuitously convert values declared as
4035 @code{float} into @code{double}.
4038 @item INTIFY (@var{float-value})
4039 Define this macro to override the way the value of a
4040 @code{float}-returning library routine should be packaged in order to
4041 return it. These functions are actually declared to return type
4042 @code{FLOAT_VALUE_TYPE} (normally @code{int}).
4044 These values can't be returned as type @code{float} because traditional
4045 C compilers would gratuitously convert the value to a @code{double}.
4047 A local variable named @code{intify} is always available when the macro
4048 @code{INTIFY} is used. It is a union of a @code{float} field named
4049 @code{f} and a field named @code{i} whose type is
4050 @code{FLOAT_VALUE_TYPE} or @code{int}.
4052 If you don't define this macro, the default definition works by copying
4053 the value through that union.
4055 @findex nongcc_SI_type
4056 @item nongcc_SI_type
4057 Define this macro as the name of the data type corresponding to
4058 @code{SImode} in the system's own C compiler.
4060 You need not define this macro if that type is @code{long int}, as it usually
4063 @findex nongcc_word_type
4064 @item nongcc_word_type
4065 Define this macro as the name of the data type corresponding to the
4066 word_mode in the system's own C compiler.
4068 You need not define this macro if that type is @code{long int}, as it usually
4071 @findex perform_@dots{}
4072 @item perform_@dots{}
4073 Define these macros to supply explicit C statements to carry out various
4074 arithmetic operations on types @code{float} and @code{double} in the
4075 library routines in @file{libgcc1.c}. See that file for a full list
4076 of these macros and their arguments.
4078 On most machines, you don't need to define any of these macros, because
4079 the C compiler that comes with the system takes care of doing them.
4081 @findex NEXT_OBJC_RUNTIME
4082 @item NEXT_OBJC_RUNTIME
4083 Define this macro to generate code for Objective C message sending using
4084 the calling convention of the NeXT system. This calling convention
4085 involves passing the object, the selector and the method arguments all
4086 at once to the method-lookup library function.
4088 The default calling convention passes just the object and the selector
4089 to the lookup function, which returns a pointer to the method.
4092 @node Addressing Modes
4093 @section Addressing Modes
4094 @cindex addressing modes
4096 @c prevent bad page break with this line
4097 This is about addressing modes.
4100 @findex HAVE_POST_INCREMENT
4101 @item HAVE_POST_INCREMENT
4102 Define this macro if the machine supports post-increment addressing.
4104 @findex HAVE_PRE_INCREMENT
4105 @findex HAVE_POST_DECREMENT
4106 @findex HAVE_PRE_DECREMENT
4107 @item HAVE_PRE_INCREMENT
4108 @itemx HAVE_POST_DECREMENT
4109 @itemx HAVE_PRE_DECREMENT
4110 Similar for other kinds of addressing.
4112 @findex CONSTANT_ADDRESS_P
4113 @item CONSTANT_ADDRESS_P (@var{x})
4114 A C expression that is 1 if the RTX @var{x} is a constant which
4115 is a valid address. On most machines, this can be defined as
4116 @code{CONSTANT_P (@var{x})}, but a few machines are more restrictive
4117 in which constant addresses are supported.
4120 @code{CONSTANT_P} accepts integer-values expressions whose values are
4121 not explicitly known, such as @code{symbol_ref}, @code{label_ref}, and
4122 @code{high} expressions and @code{const} arithmetic expressions, in
4123 addition to @code{const_int} and @code{const_double} expressions.
4125 @findex MAX_REGS_PER_ADDRESS
4126 @item MAX_REGS_PER_ADDRESS
4127 A number, the maximum number of registers that can appear in a valid
4128 memory address. Note that it is up to you to specify a value equal to
4129 the maximum number that @code{GO_IF_LEGITIMATE_ADDRESS} would ever
4132 @findex GO_IF_LEGITIMATE_ADDRESS
4133 @item GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label})
4134 A C compound statement with a conditional @code{goto @var{label};}
4135 executed if @var{x} (an RTX) is a legitimate memory address on the
4136 target machine for a memory operand of mode @var{mode}.
4138 It usually pays to define several simpler macros to serve as
4139 subroutines for this one. Otherwise it may be too complicated to
4142 This macro must exist in two variants: a strict variant and a
4143 non-strict one. The strict variant is used in the reload pass. It
4144 must be defined so that any pseudo-register that has not been
4145 allocated a hard register is considered a memory reference. In
4146 contexts where some kind of register is required, a pseudo-register
4147 with no hard register must be rejected.
4149 The non-strict variant is used in other passes. It must be defined to
4150 accept all pseudo-registers in every context where some kind of
4151 register is required.
4153 @findex REG_OK_STRICT
4154 Compiler source files that want to use the strict variant of this
4155 macro define the macro @code{REG_OK_STRICT}. You should use an
4156 @code{#ifdef REG_OK_STRICT} conditional to define the strict variant
4157 in that case and the non-strict variant otherwise.
4159 Subroutines to check for acceptable registers for various purposes (one
4160 for base registers, one for index registers, and so on) are typically
4161 among the subroutines used to define @code{GO_IF_LEGITIMATE_ADDRESS}.
4162 Then only these subroutine macros need have two variants; the higher
4163 levels of macros may be the same whether strict or not.@refill
4165 Normally, constant addresses which are the sum of a @code{symbol_ref}
4166 and an integer are stored inside a @code{const} RTX to mark them as
4167 constant. Therefore, there is no need to recognize such sums
4168 specifically as legitimate addresses. Normally you would simply
4169 recognize any @code{const} as legitimate.
4171 Usually @code{PRINT_OPERAND_ADDRESS} is not prepared to handle constant
4172 sums that are not marked with @code{const}. It assumes that a naked
4173 @code{plus} indicates indexing. If so, then you @emph{must} reject such
4174 naked constant sums as illegitimate addresses, so that none of them will
4175 be given to @code{PRINT_OPERAND_ADDRESS}.
4177 @cindex @code{ENCODE_SECTION_INFO} and address validation
4178 On some machines, whether a symbolic address is legitimate depends on
4179 the section that the address refers to. On these machines, define the
4180 macro @code{ENCODE_SECTION_INFO} to store the information into the
4181 @code{symbol_ref}, and then check for it here. When you see a
4182 @code{const}, you will have to look inside it to find the
4183 @code{symbol_ref} in order to determine the section. @xref{Assembler
4186 @findex saveable_obstack
4187 The best way to modify the name string is by adding text to the
4188 beginning, with suitable punctuation to prevent any ambiguity. Allocate
4189 the new name in @code{saveable_obstack}. You will have to modify
4190 @code{ASM_OUTPUT_LABELREF} to remove and decode the added text and
4191 output the name accordingly, and define @code{STRIP_NAME_ENCODING} to
4192 access the original name string.
4194 You can check the information stored here into the @code{symbol_ref} in
4195 the definitions of the macros @code{GO_IF_LEGITIMATE_ADDRESS} and
4196 @code{PRINT_OPERAND_ADDRESS}.
4198 @findex REG_OK_FOR_BASE_P
4199 @item REG_OK_FOR_BASE_P (@var{x})
4200 A C expression that is nonzero if @var{x} (assumed to be a @code{reg}
4201 RTX) is valid for use as a base register. For hard registers, it
4202 should always accept those which the hardware permits and reject the
4203 others. Whether the macro accepts or rejects pseudo registers must be
4204 controlled by @code{REG_OK_STRICT} as described above. This usually
4205 requires two variant definitions, of which @code{REG_OK_STRICT}
4206 controls the one actually used.
4208 @findex REG_MODE_OK_FOR_BASE_P
4209 @item REG_MODE_OK_FOR_BASE_P (@var{x}, @var{mode})
4210 A C expression that is just like @code{REG_OK_FOR_BASE_P}, except that
4211 that expression may examine the mode of the memory reference in
4212 @var{mode}. You should define this macro if the mode of the memory
4213 reference affects whether a register may be used as a base register. If
4214 you define this macro, the compiler will use it instead of
4215 @code{REG_OK_FOR_BASE_P}.
4217 @findex REG_OK_FOR_INDEX_P
4218 @item REG_OK_FOR_INDEX_P (@var{x})
4219 A C expression that is nonzero if @var{x} (assumed to be a @code{reg}
4220 RTX) is valid for use as an index register.
4222 The difference between an index register and a base register is that
4223 the index register may be scaled. If an address involves the sum of
4224 two registers, neither one of them scaled, then either one may be
4225 labeled the ``base'' and the other the ``index''; but whichever
4226 labeling is used must fit the machine's constraints of which registers
4227 may serve in each capacity. The compiler will try both labelings,
4228 looking for one that is valid, and will reload one or both registers
4229 only if neither labeling works.
4231 @findex LEGITIMIZE_ADDRESS
4232 @item LEGITIMIZE_ADDRESS (@var{x}, @var{oldx}, @var{mode}, @var{win})
4233 A C compound statement that attempts to replace @var{x} with a valid
4234 memory address for an operand of mode @var{mode}. @var{win} will be a
4235 C statement label elsewhere in the code; the macro definition may use
4238 GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{win});
4242 to avoid further processing if the address has become legitimate.
4244 @findex break_out_memory_refs
4245 @var{x} will always be the result of a call to @code{break_out_memory_refs},
4246 and @var{oldx} will be the operand that was given to that function to produce
4249 The code generated by this macro should not alter the substructure of
4250 @var{x}. If it transforms @var{x} into a more legitimate form, it
4251 should assign @var{x} (which will always be a C variable) a new value.
4253 It is not necessary for this macro to come up with a legitimate
4254 address. The compiler has standard ways of doing so in all cases. In
4255 fact, it is safe for this macro to do nothing. But often a
4256 machine-dependent strategy can generate better code.
4258 @findex LEGITIMIZE_RELOAD_ADDRESS
4259 @item LEGITIMIZE_RELOAD_ADDRESS (@var{x}, @var{mode}, @var{opnum}, @var{type}, @var{ind_levels}, @var{win})
4260 A C compound statement that attempts to replace @var{x}, which is an address
4261 that needs reloading, with a valid memory address for an operand of mode
4262 @var{mode}. @var{win} will be a C statement label elsewhere in the code.
4263 It is not necessary to define this macro, but it might be useful for
4264 performance reasons.
4266 For example, on the i386, it is sometimes possible to use a single
4267 reload register instead of two by reloading a sum of two pseudo
4268 registers into a register. On the other hand, for number of RISC
4269 processors offsets are limited so that often an intermediate address
4270 needs to be generated in order to address a stack slot. By defining
4271 LEGITIMIZE_RELOAD_ADDRESS appropriately, the intermediate addresses
4272 generated for adjacent some stack slots can be made identical, and thus
4275 @emph{Note}: This macro should be used with caution. It is necessary
4276 to know something of how reload works in order to effectively use this,
4277 and it is quite easy to produce macros that build in too much knowledge
4278 of reload internals.
4281 The macro definition should use @code{push_reload} to indicate parts that
4282 need reloading; @var{opnum}, @var{type} and @var{ind_levels} are usually
4283 suitable to be passed unaltered to @code{push_reload}.
4285 The code generated by this macro must not alter the substructure of
4286 @var{x}. If it transforms @var{x} into a more legitimate form, it
4287 should assign @var{x} (which will always be a C variable) a new value.
4288 This also applies to parts that you change indirectly by calling
4291 @findex strict_memory_address_p
4292 The macro definition may use @code{strict_memory_address_p} to test if
4293 the address has become legitimate.
4296 If you want to change only a part of @var{x}, one standard way of doing
4297 this is to use @code{copy_rtx}. Note, however, that is unshares only a
4298 single level of rtl. Thus, if the part to be changed is not at the
4299 top level, you'll need to replace first the top leve
4300 It is not necessary for this macro to come up with a legitimate
4301 address; but often a machine-dependent strategy can generate better code.
4303 @findex GO_IF_MODE_DEPENDENT_ADDRESS
4304 @item GO_IF_MODE_DEPENDENT_ADDRESS (@var{addr}, @var{label})
4305 A C statement or compound statement with a conditional @code{goto
4306 @var{label};} executed if memory address @var{x} (an RTX) can have
4307 different meanings depending on the machine mode of the memory
4308 reference it is used for or if the address is valid for some modes
4311 Autoincrement and autodecrement addresses typically have mode-dependent
4312 effects because the amount of the increment or decrement is the size
4313 of the operand being addressed. Some machines have other mode-dependent
4314 addresses. Many RISC machines have no mode-dependent addresses.
4316 You may assume that @var{addr} is a valid address for the machine.
4318 @findex LEGITIMATE_CONSTANT_P
4319 @item LEGITIMATE_CONSTANT_P (@var{x})
4320 A C expression that is nonzero if @var{x} is a legitimate constant for
4321 an immediate operand on the target machine. You can assume that
4322 @var{x} satisfies @code{CONSTANT_P}, so you need not check this. In fact,
4323 @samp{1} is a suitable definition for this macro on machines where
4324 anything @code{CONSTANT_P} is valid.@refill
4326 @findex DONT_RECORD_EQUIVALENCE
4327 @item DONT_RECORD_EQUIVALENCE (@var{note})
4328 A C expression that is nonzero if the @code{REG_EQUAL} note @var{x} should not
4329 be promoted to a @code{REG_EQUIV} note.
4331 Define this macro if @var{note} refers to a constant that must be accepted
4332 by @code{LEGITIMATE_CONSTANT_P}, but must not appear as an immediate operand.
4334 Most machine descriptions do not need to define this macro.
4337 @node Condition Code
4338 @section Condition Code Status
4339 @cindex condition code status
4341 @c prevent bad page break with this line
4342 This describes the condition code status.
4345 The file @file{conditions.h} defines a variable @code{cc_status} to
4346 describe how the condition code was computed (in case the interpretation of
4347 the condition code depends on the instruction that it was set by). This
4348 variable contains the RTL expressions on which the condition code is
4349 currently based, and several standard flags.
4351 Sometimes additional machine-specific flags must be defined in the machine
4352 description header file. It can also add additional machine-specific
4353 information by defining @code{CC_STATUS_MDEP}.
4356 @findex CC_STATUS_MDEP
4357 @item CC_STATUS_MDEP
4358 C code for a data type which is used for declaring the @code{mdep}
4359 component of @code{cc_status}. It defaults to @code{int}.
4361 This macro is not used on machines that do not use @code{cc0}.
4363 @findex CC_STATUS_MDEP_INIT
4364 @item CC_STATUS_MDEP_INIT
4365 A C expression to initialize the @code{mdep} field to ``empty''.
4366 The default definition does nothing, since most machines don't use
4367 the field anyway. If you want to use the field, you should probably
4368 define this macro to initialize it.
4370 This macro is not used on machines that do not use @code{cc0}.
4372 @findex NOTICE_UPDATE_CC
4373 @item NOTICE_UPDATE_CC (@var{exp}, @var{insn})
4374 A C compound statement to set the components of @code{cc_status}
4375 appropriately for an insn @var{insn} whose body is @var{exp}. It is
4376 this macro's responsibility to recognize insns that set the condition
4377 code as a byproduct of other activity as well as those that explicitly
4380 This macro is not used on machines that do not use @code{cc0}.
4382 If there are insns that do not set the condition code but do alter
4383 other machine registers, this macro must check to see whether they
4384 invalidate the expressions that the condition code is recorded as
4385 reflecting. For example, on the 68000, insns that store in address
4386 registers do not set the condition code, which means that usually
4387 @code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such
4388 insns. But suppose that the previous insn set the condition code
4389 based on location @samp{a4@@(102)} and the current insn stores a new
4390 value in @samp{a4}. Although the condition code is not changed by
4391 this, it will no longer be true that it reflects the contents of
4392 @samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter
4393 @code{cc_status} in this case to say that nothing is known about the
4394 condition code value.
4396 The definition of @code{NOTICE_UPDATE_CC} must be prepared to deal
4397 with the results of peephole optimization: insns whose patterns are
4398 @code{parallel} RTXs containing various @code{reg}, @code{mem} or
4399 constants which are just the operands. The RTL structure of these
4400 insns is not sufficient to indicate what the insns actually do. What
4401 @code{NOTICE_UPDATE_CC} should do when it sees one is just to run
4402 @code{CC_STATUS_INIT}.
4404 A possible definition of @code{NOTICE_UPDATE_CC} is to call a function
4405 that looks at an attribute (@pxref{Insn Attributes}) named, for example,
4406 @samp{cc}. This avoids having detailed information about patterns in
4407 two places, the @file{md} file and in @code{NOTICE_UPDATE_CC}.
4409 @findex EXTRA_CC_MODES
4410 @item EXTRA_CC_MODES
4411 A list of names to be used for additional modes for condition code
4412 values in registers (@pxref{Jump Patterns}). These names are added
4413 to @code{enum machine_mode} and all have class @code{MODE_CC}. By
4414 convention, they should start with @samp{CC} and end with @samp{mode}.
4416 You should only define this macro if your machine does not use @code{cc0}
4417 and only if additional modes are required.
4419 @findex EXTRA_CC_NAMES
4420 @item EXTRA_CC_NAMES
4421 A list of C strings giving the names for the modes listed in
4422 @code{EXTRA_CC_MODES}. For example, the Sparc defines this macro and
4423 @code{EXTRA_CC_MODES} as
4426 #define EXTRA_CC_MODES CC_NOOVmode, CCFPmode, CCFPEmode
4427 #define EXTRA_CC_NAMES "CC_NOOV", "CCFP", "CCFPE"
4430 This macro is not required if @code{EXTRA_CC_MODES} is not defined.
4432 @findex SELECT_CC_MODE
4433 @item SELECT_CC_MODE (@var{op}, @var{x}, @var{y})
4434 Returns a mode from class @code{MODE_CC} to be used when comparison
4435 operation code @var{op} is applied to rtx @var{x} and @var{y}. For
4436 example, on the Sparc, @code{SELECT_CC_MODE} is defined as (see
4437 @pxref{Jump Patterns} for a description of the reason for this
4441 #define SELECT_CC_MODE(OP,X,Y) \
4442 (GET_MODE_CLASS (GET_MODE (X)) == MODE_FLOAT \
4443 ? ((OP == EQ || OP == NE) ? CCFPmode : CCFPEmode) \
4444 : ((GET_CODE (X) == PLUS || GET_CODE (X) == MINUS \
4445 || GET_CODE (X) == NEG) \
4446 ? CC_NOOVmode : CCmode))
4449 You need not define this macro if @code{EXTRA_CC_MODES} is not defined.
4451 @findex CANONICALIZE_COMPARISON
4452 @item CANONICALIZE_COMPARISON (@var{code}, @var{op0}, @var{op1})
4453 One some machines not all possible comparisons are defined, but you can
4454 convert an invalid comparison into a valid one. For example, the Alpha
4455 does not have a @code{GT} comparison, but you can use an @code{LT}
4456 comparison instead and swap the order of the operands.
4458 On such machines, define this macro to be a C statement to do any
4459 required conversions. @var{code} is the initial comparison code
4460 and @var{op0} and @var{op1} are the left and right operands of the
4461 comparison, respectively. You should modify @var{code}, @var{op0}, and
4462 @var{op1} as required.
4464 GNU CC will not assume that the comparison resulting from this macro is
4465 valid but will see if the resulting insn matches a pattern in the
4468 You need not define this macro if it would never change the comparison
4471 @findex REVERSIBLE_CC_MODE
4472 @item REVERSIBLE_CC_MODE (@var{mode})
4473 A C expression whose value is one if it is always safe to reverse a
4474 comparison whose mode is @var{mode}. If @code{SELECT_CC_MODE}
4475 can ever return @var{mode} for a floating-point inequality comparison,
4476 then @code{REVERSIBLE_CC_MODE (@var{mode})} must be zero.
4478 You need not define this macro if it would always returns zero or if the
4479 floating-point format is anything other than @code{IEEE_FLOAT_FORMAT}.
4480 For example, here is the definition used on the Sparc, where floating-point
4481 inequality comparisons are always given @code{CCFPEmode}:
4484 #define REVERSIBLE_CC_MODE(MODE) ((MODE) != CCFPEmode)
4490 @section Describing Relative Costs of Operations
4491 @cindex costs of instructions
4492 @cindex relative costs
4493 @cindex speed of instructions
4495 These macros let you describe the relative speed of various operations
4496 on the target machine.
4500 @item CONST_COSTS (@var{x}, @var{code}, @var{outer_code})
4501 A part of a C @code{switch} statement that describes the relative costs
4502 of constant RTL expressions. It must contain @code{case} labels for
4503 expression codes @code{const_int}, @code{const}, @code{symbol_ref},
4504 @code{label_ref} and @code{const_double}. Each case must ultimately
4505 reach a @code{return} statement to return the relative cost of the use
4506 of that kind of constant value in an expression. The cost may depend on
4507 the precise value of the constant, which is available for examination in
4508 @var{x}, and the rtx code of the expression in which it is contained,
4509 found in @var{outer_code}.
4511 @var{code} is the expression code---redundant, since it can be
4512 obtained with @code{GET_CODE (@var{x})}.
4515 @findex COSTS_N_INSNS
4516 @item RTX_COSTS (@var{x}, @var{code}, @var{outer_code})
4517 Like @code{CONST_COSTS} but applies to nonconstant RTL expressions.
4518 This can be used, for example, to indicate how costly a multiply
4519 instruction is. In writing this macro, you can use the construct
4520 @code{COSTS_N_INSNS (@var{n})} to specify a cost equal to @var{n} fast
4521 instructions. @var{outer_code} is the code of the expression in which
4522 @var{x} is contained.
4524 This macro is optional; do not define it if the default cost assumptions
4525 are adequate for the target machine.
4527 @findex DEFAULT_RTX_COSTS
4528 @item DEFAULT_RTX_COSTS (@var{x}, @var{code}, @var{outer_code})
4529 This macro, if defined, is called for any case not handled by the
4530 @code{RTX_COSTS} or @code{CONST_COSTS} macros. This eliminates the need
4531 to put case labels into the macro, but the code, or any functions it
4532 calls, must assume that the RTL in @var{x} could be of any type that has
4533 not already been handled. The arguments are the same as for
4534 @code{RTX_COSTS}, and the macro should execute a return statement giving
4535 the cost of any RTL expressions that it can handle. The default cost
4536 calculation is used for any RTL for which this macro does not return a
4539 This macro is optional; do not define it if the default cost assumptions
4540 are adequate for the target machine.
4542 @findex ADDRESS_COST
4543 @item ADDRESS_COST (@var{address})
4544 An expression giving the cost of an addressing mode that contains
4545 @var{address}. If not defined, the cost is computed from
4546 the @var{address} expression and the @code{CONST_COSTS} values.
4548 For most CISC machines, the default cost is a good approximation of the
4549 true cost of the addressing mode. However, on RISC machines, all
4550 instructions normally have the same length and execution time. Hence
4551 all addresses will have equal costs.
4553 In cases where more than one form of an address is known, the form with
4554 the lowest cost will be used. If multiple forms have the same, lowest,
4555 cost, the one that is the most complex will be used.
4557 For example, suppose an address that is equal to the sum of a register
4558 and a constant is used twice in the same basic block. When this macro
4559 is not defined, the address will be computed in a register and memory
4560 references will be indirect through that register. On machines where
4561 the cost of the addressing mode containing the sum is no higher than
4562 that of a simple indirect reference, this will produce an additional
4563 instruction and possibly require an additional register. Proper
4564 specification of this macro eliminates this overhead for such machines.
4566 Similar use of this macro is made in strength reduction of loops.
4568 @var{address} need not be valid as an address. In such a case, the cost
4569 is not relevant and can be any value; invalid addresses need not be
4570 assigned a different cost.
4572 On machines where an address involving more than one register is as
4573 cheap as an address computation involving only one register, defining
4574 @code{ADDRESS_COST} to reflect this can cause two registers to be live
4575 over a region of code where only one would have been if
4576 @code{ADDRESS_COST} were not defined in that manner. This effect should
4577 be considered in the definition of this macro. Equivalent costs should
4578 probably only be given to addresses with different numbers of registers
4579 on machines with lots of registers.
4581 This macro will normally either not be defined or be defined as a
4584 @findex REGISTER_MOVE_COST
4585 @item REGISTER_MOVE_COST (@var{from}, @var{to})
4586 A C expression for the cost of moving data from a register in class
4587 @var{from} to one in class @var{to}. The classes are expressed using
4588 the enumeration values such as @code{GENERAL_REGS}. A value of 2 is the
4589 default; other values are interpreted relative to that.
4591 It is not required that the cost always equal 2 when @var{from} is the
4592 same as @var{to}; on some machines it is expensive to move between
4593 registers if they are not general registers.
4595 If reload sees an insn consisting of a single @code{set} between two
4596 hard registers, and if @code{REGISTER_MOVE_COST} applied to their
4597 classes returns a value of 2, reload does not check to ensure that the
4598 constraints of the insn are met. Setting a cost of other than 2 will
4599 allow reload to verify that the constraints are met. You should do this
4600 if the @samp{mov@var{m}} pattern's constraints do not allow such copying.
4602 @findex MEMORY_MOVE_COST
4603 @item MEMORY_MOVE_COST (@var{mode}, @var{class}, @var{in})
4604 A C expression for the cost of moving data of mode @var{mode} between a
4605 register of class @var{class} and memory; @var{in} is zero if the value
4606 is to be written to memory, non-zero if it is to be read in. This cost
4607 is relative to those in @code{REGISTER_MOVE_COST}. If moving between
4608 registers and memory is more expensive than between two registers, you
4609 should define this macro to express the relative cost.
4611 If you do not define this macro, GNU CC uses a default cost of 4 plus
4612 the cost of copying via a secondary reload register, if one is
4613 needed. If your machine requires a secondary reload register to copy
4614 between memory and a register of @var{class} but the reload mechanism is
4615 more complex than copying via an intermediate, define this macro to
4616 reflect the actual cost of the move.
4618 GNU CC defines the function @code{memory_move_secondary_cost} if
4619 secondary reloads are needed. It computes the costs due to copying via
4620 a secondary register. If your machine copies from memory using a
4621 secondary register in the conventional way but the default base value of
4622 4 is not correct for your machine, define this macro to add some other
4623 value to the result of that function. The arguments to that function
4624 are the same as to this macro.
4628 A C expression for the cost of a branch instruction. A value of 1 is
4629 the default; other values are interpreted relative to that.
4632 Here are additional macros which do not specify precise relative costs,
4633 but only that certain actions are more expensive than GNU CC would
4637 @findex SLOW_BYTE_ACCESS
4638 @item SLOW_BYTE_ACCESS
4639 Define this macro as a C expression which is nonzero if accessing less
4640 than a word of memory (i.e. a @code{char} or a @code{short}) is no
4641 faster than accessing a word of memory, i.e., if such access
4642 require more than one instruction or if there is no difference in cost
4643 between byte and (aligned) word loads.
4645 When this macro is not defined, the compiler will access a field by
4646 finding the smallest containing object; when it is defined, a fullword
4647 load will be used if alignment permits. Unless bytes accesses are
4648 faster than word accesses, using word accesses is preferable since it
4649 may eliminate subsequent memory access if subsequent accesses occur to
4650 other fields in the same word of the structure, but to different bytes.
4652 @findex SLOW_ZERO_EXTEND
4653 @item SLOW_ZERO_EXTEND
4654 Define this macro if zero-extension (of a @code{char} or @code{short}
4655 to an @code{int}) can be done faster if the destination is a register
4656 that is known to be zero.
4658 If you define this macro, you must have instruction patterns that
4659 recognize RTL structures like this:
4662 (set (strict_low_part (subreg:QI (reg:SI @dots{}) 0)) @dots{})
4666 and likewise for @code{HImode}.
4668 @findex SLOW_UNALIGNED_ACCESS
4669 @item SLOW_UNALIGNED_ACCESS
4670 Define this macro to be the value 1 if unaligned accesses have a cost
4671 many times greater than aligned accesses, for example if they are
4672 emulated in a trap handler.
4674 When this macro is non-zero, the compiler will act as if
4675 @code{STRICT_ALIGNMENT} were non-zero when generating code for block
4676 moves. This can cause significantly more instructions to be produced.
4677 Therefore, do not set this macro non-zero if unaligned accesses only add a
4678 cycle or two to the time for a memory access.
4680 If the value of this macro is always zero, it need not be defined.
4682 @findex DONT_REDUCE_ADDR
4683 @item DONT_REDUCE_ADDR
4684 Define this macro to inhibit strength reduction of memory addresses.
4685 (On some machines, such strength reduction seems to do harm rather
4690 The number of scalar move insns which should be generated instead of a
4691 string move insn or a library call. Increasing the value will always
4692 make code faster, but eventually incurs high cost in increased code size.
4694 If you don't define this, a reasonable default is used.
4696 @findex NO_FUNCTION_CSE
4697 @item NO_FUNCTION_CSE
4698 Define this macro if it is as good or better to call a constant
4699 function address than to call an address kept in a register.
4701 @findex NO_RECURSIVE_FUNCTION_CSE
4702 @item NO_RECURSIVE_FUNCTION_CSE
4703 Define this macro if it is as good or better for a function to call
4704 itself with an explicit address than to call an address kept in a
4708 @item ADJUST_COST (@var{insn}, @var{link}, @var{dep_insn}, @var{cost})
4709 A C statement (sans semicolon) to update the integer variable @var{cost}
4710 based on the relationship between @var{insn} that is dependent on
4711 @var{dep_insn} through the dependence @var{link}. The default is to
4712 make no adjustment to @var{cost}. This can be used for example to
4713 specify to the scheduler that an output- or anti-dependence does not
4714 incur the same cost as a data-dependence.
4716 @findex ADJUST_PRIORITY
4717 @item ADJUST_PRIORITY (@var{insn})
4718 A C statement (sans semicolon) to update the integer scheduling
4719 priority @code{INSN_PRIORITY(@var{insn})}. Reduce the priority
4720 to execute the @var{insn} earlier, increase the priority to execute
4721 @var{insn} later. Do not define this macro if you do not need to
4722 adjust the scheduling priorities of insns.
4726 @section Dividing the Output into Sections (Texts, Data, @dots{})
4727 @c the above section title is WAY too long. maybe cut the part between
4728 @c the (...)? --mew 10feb93
4730 An object file is divided into sections containing different types of
4731 data. In the most common case, there are three sections: the @dfn{text
4732 section}, which holds instructions and read-only data; the @dfn{data
4733 section}, which holds initialized writable data; and the @dfn{bss
4734 section}, which holds uninitialized data. Some systems have other kinds
4737 The compiler must tell the assembler when to switch sections. These
4738 macros control what commands to output to tell the assembler this. You
4739 can also define additional sections.
4742 @findex TEXT_SECTION_ASM_OP
4743 @item TEXT_SECTION_ASM_OP
4744 A C expression whose value is a string containing the assembler
4745 operation that should precede instructions and read-only data. Normally
4746 @code{".text"} is right.
4748 @findex DATA_SECTION_ASM_OP
4749 @item DATA_SECTION_ASM_OP
4750 A C expression whose value is a string containing the assembler
4751 operation to identify the following data as writable initialized data.
4752 Normally @code{".data"} is right.
4754 @findex SHARED_SECTION_ASM_OP
4755 @item SHARED_SECTION_ASM_OP
4756 If defined, a C expression whose value is a string containing the
4757 assembler operation to identify the following data as shared data. If
4758 not defined, @code{DATA_SECTION_ASM_OP} will be used.
4760 @findex BSS_SECTION_ASM_OP
4761 @item BSS_SECTION_ASM_OP
4762 If defined, a C expression whose value is a string containing the
4763 assembler operation to identify the following data as uninitialized global
4764 data. If not defined, and neither @code{ASM_OUTPUT_BSS} nor
4765 @code{ASM_OUTPUT_ALIGNED_BSS} are defined, uninitialized global data will be
4766 output in the data section if @samp{-fno-common} is passed, otherwise
4767 @code{ASM_OUTPUT_COMMON} will be used.
4769 @findex SHARED_BSS_SECTION_ASM_OP
4770 @item SHARED_BSS_SECTION_ASM_OP
4771 If defined, a C expression whose value is a string containing the
4772 assembler operation to identify the following data as uninitialized global
4773 shared data. If not defined, and @code{BSS_SECTION_ASM_OP} is, the latter
4776 @findex INIT_SECTION_ASM_OP
4777 @item INIT_SECTION_ASM_OP
4778 If defined, a C expression whose value is a string containing the
4779 assembler operation to identify the following data as initialization
4780 code. If not defined, GNU CC will assume such a section does not
4783 @findex EXTRA_SECTIONS
4786 @item EXTRA_SECTIONS
4787 A list of names for sections other than the standard two, which are
4788 @code{in_text} and @code{in_data}. You need not define this macro
4789 on a system with no other sections (that GCC needs to use).
4791 @findex EXTRA_SECTION_FUNCTIONS
4792 @findex text_section
4793 @findex data_section
4794 @item EXTRA_SECTION_FUNCTIONS
4795 One or more functions to be defined in @file{varasm.c}. These
4796 functions should do jobs analogous to those of @code{text_section} and
4797 @code{data_section}, for your additional sections. Do not define this
4798 macro if you do not define @code{EXTRA_SECTIONS}.
4800 @findex READONLY_DATA_SECTION
4801 @item READONLY_DATA_SECTION
4802 On most machines, read-only variables, constants, and jump tables are
4803 placed in the text section. If this is not the case on your machine,
4804 this macro should be defined to be the name of a function (either
4805 @code{data_section} or a function defined in @code{EXTRA_SECTIONS}) that
4806 switches to the section to be used for read-only items.
4808 If these items should be placed in the text section, this macro should
4811 @findex SELECT_SECTION
4812 @item SELECT_SECTION (@var{exp}, @var{reloc})
4813 A C statement or statements to switch to the appropriate section for
4814 output of @var{exp}. You can assume that @var{exp} is either a
4815 @code{VAR_DECL} node or a constant of some sort. @var{reloc}
4816 indicates whether the initial value of @var{exp} requires link-time
4817 relocations. Select the section by calling @code{text_section} or one
4818 of the alternatives for other sections.
4820 Do not define this macro if you put all read-only variables and
4821 constants in the read-only data section (usually the text section).
4823 @findex SELECT_RTX_SECTION
4824 @item SELECT_RTX_SECTION (@var{mode}, @var{rtx})
4825 A C statement or statements to switch to the appropriate section for
4826 output of @var{rtx} in mode @var{mode}. You can assume that @var{rtx}
4827 is some kind of constant in RTL. The argument @var{mode} is redundant
4828 except in the case of a @code{const_int} rtx. Select the section by
4829 calling @code{text_section} or one of the alternatives for other
4832 Do not define this macro if you put all constants in the read-only
4835 @findex JUMP_TABLES_IN_TEXT_SECTION
4836 @item JUMP_TABLES_IN_TEXT_SECTION
4837 Define this macro to be an expression with a non-zero value if jump
4838 tables (for @code{tablejump} insns) should be output in the text
4839 section, along with the assembler instructions. Otherwise, the
4840 readonly data section is used.
4842 This macro is irrelevant if there is no separate readonly data section.
4844 @findex ENCODE_SECTION_INFO
4845 @item ENCODE_SECTION_INFO (@var{decl})
4846 Define this macro if references to a symbol must be treated differently
4847 depending on something about the variable or function named by the
4848 symbol (such as what section it is in).
4850 The macro definition, if any, is executed immediately after the rtl for
4851 @var{decl} has been created and stored in @code{DECL_RTL (@var{decl})}.
4852 The value of the rtl will be a @code{mem} whose address is a
4855 @cindex @code{SYMBOL_REF_FLAG}, in @code{ENCODE_SECTION_INFO}
4856 The usual thing for this macro to do is to record a flag in the
4857 @code{symbol_ref} (such as @code{SYMBOL_REF_FLAG}) or to store a
4858 modified name string in the @code{symbol_ref} (if one bit is not enough
4861 @findex STRIP_NAME_ENCODING
4862 @item STRIP_NAME_ENCODING (@var{var}, @var{sym_name})
4863 Decode @var{sym_name} and store the real name part in @var{var}, sans
4864 the characters that encode section info. Define this macro if
4865 @code{ENCODE_SECTION_INFO} alters the symbol's name string.
4867 @findex UNIQUE_SECTION_P
4868 @item UNIQUE_SECTION_P (@var{decl})
4869 A C expression which evaluates to true if @var{decl} should be placed
4870 into a unique section for some target-specific reason. If you do not
4871 define this macro, the default is @samp{0}. Note that the flag
4872 @samp{-ffunction-sections} will also cause functions to be placed into
4875 @findex UNIQUE_SECTION
4876 @item UNIQUE_SECTION (@var{decl}, @var{reloc})
4877 A C statement to build up a unique section name, expressed as a
4878 STRING_CST node, and assign it to @samp{DECL_SECTION_NAME (@var{decl})}.
4879 @var{reloc} indicates whether the initial value of @var{exp} requires
4880 link-time relocations. If you do not define this macro, GNU CC will use
4881 the symbol name prefixed by @samp{.} as the section name.
4885 @section Position Independent Code
4886 @cindex position independent code
4889 This section describes macros that help implement generation of position
4890 independent code. Simply defining these macros is not enough to
4891 generate valid PIC; you must also add support to the macros
4892 @code{GO_IF_LEGITIMATE_ADDRESS} and @code{PRINT_OPERAND_ADDRESS}, as
4893 well as @code{LEGITIMIZE_ADDRESS}. You must modify the definition of
4894 @samp{movsi} to do something appropriate when the source operand
4895 contains a symbolic address. You may also need to alter the handling of
4896 switch statements so that they use relative addresses.
4897 @c i rearranged the order of the macros above to try to force one of
4898 @c them to the next line, to eliminate an overfull hbox. --mew 10feb93
4901 @findex PIC_OFFSET_TABLE_REGNUM
4902 @item PIC_OFFSET_TABLE_REGNUM
4903 The register number of the register used to address a table of static
4904 data addresses in memory. In some cases this register is defined by a
4905 processor's ``application binary interface'' (ABI). When this macro
4906 is defined, RTL is generated for this register once, as with the stack
4907 pointer and frame pointer registers. If this macro is not defined, it
4908 is up to the machine-dependent files to allocate such a register (if
4911 @findex PIC_OFFSET_TABLE_REG_CALL_CLOBBERED
4912 @item PIC_OFFSET_TABLE_REG_CALL_CLOBBERED
4913 Define this macro if the register defined by
4914 @code{PIC_OFFSET_TABLE_REGNUM} is clobbered by calls. Do not define
4915 this macro if @code{PPIC_OFFSET_TABLE_REGNUM} is not defined.
4917 @findex FINALIZE_PIC
4919 By generating position-independent code, when two different programs (A
4920 and B) share a common library (libC.a), the text of the library can be
4921 shared whether or not the library is linked at the same address for both
4922 programs. In some of these environments, position-independent code
4923 requires not only the use of different addressing modes, but also
4924 special code to enable the use of these addressing modes.
4926 The @code{FINALIZE_PIC} macro serves as a hook to emit these special
4927 codes once the function is being compiled into assembly code, but not
4928 before. (It is not done before, because in the case of compiling an
4929 inline function, it would lead to multiple PIC prologues being
4930 included in functions which used inline functions and were compiled to
4933 @findex LEGITIMATE_PIC_OPERAND_P
4934 @item LEGITIMATE_PIC_OPERAND_P (@var{x})
4935 A C expression that is nonzero if @var{x} is a legitimate immediate
4936 operand on the target machine when generating position independent code.
4937 You can assume that @var{x} satisfies @code{CONSTANT_P}, so you need not
4938 check this. You can also assume @var{flag_pic} is true, so you need not
4939 check it either. You need not define this macro if all constants
4940 (including @code{SYMBOL_REF}) can be immediate operands when generating
4941 position independent code.
4944 @node Assembler Format
4945 @section Defining the Output Assembler Language
4947 This section describes macros whose principal purpose is to describe how
4948 to write instructions in assembler language--rather than what the
4952 * File Framework:: Structural information for the assembler file.
4953 * Data Output:: Output of constants (numbers, strings, addresses).
4954 * Uninitialized Data:: Output of uninitialized variables.
4955 * Label Output:: Output and generation of labels.
4956 * Initialization:: General principles of initialization
4957 and termination routines.
4958 * Macros for Initialization::
4959 Specific macros that control the handling of
4960 initialization and termination routines.
4961 * Instruction Output:: Output of actual instructions.
4962 * Dispatch Tables:: Output of jump tables.
4963 * Exception Region Output:: Output of exception region code.
4964 * Alignment Output:: Pseudo ops for alignment and skipping data.
4967 @node File Framework
4968 @subsection The Overall Framework of an Assembler File
4969 @cindex assembler format
4970 @cindex output of assembler code
4972 @c prevent bad page break with this line
4973 This describes the overall framework of an assembler file.
4976 @findex ASM_FILE_START
4977 @item ASM_FILE_START (@var{stream})
4978 A C expression which outputs to the stdio stream @var{stream}
4979 some appropriate text to go at the start of an assembler file.
4981 Normally this macro is defined to output a line containing
4982 @samp{#NO_APP}, which is a comment that has no effect on most
4983 assemblers but tells the GNU assembler that it can save time by not
4984 checking for certain assembler constructs.
4986 On systems that use SDB, it is necessary to output certain commands;
4987 see @file{attasm.h}.
4989 @findex ASM_FILE_END
4990 @item ASM_FILE_END (@var{stream})
4991 A C expression which outputs to the stdio stream @var{stream}
4992 some appropriate text to go at the end of an assembler file.
4994 If this macro is not defined, the default is to output nothing
4995 special at the end of the file. Most systems don't require any
4998 On systems that use SDB, it is necessary to output certain commands;
4999 see @file{attasm.h}.
5001 @findex ASM_IDENTIFY_GCC
5002 @item ASM_IDENTIFY_GCC (@var{file})
5003 A C statement to output assembler commands which will identify
5004 the object file as having been compiled with GNU CC (or another
5007 If you don't define this macro, the string @samp{gcc_compiled.:}
5008 is output. This string is calculated to define a symbol which,
5009 on BSD systems, will never be defined for any other reason.
5010 GDB checks for the presence of this symbol when reading the
5011 symbol table of an executable.
5013 On non-BSD systems, you must arrange communication with GDB in
5014 some other fashion. If GDB is not used on your system, you can
5015 define this macro with an empty body.
5017 @findex ASM_COMMENT_START
5018 @item ASM_COMMENT_START
5019 A C string constant describing how to begin a comment in the target
5020 assembler language. The compiler assumes that the comment will end at
5021 the end of the line.
5025 A C string constant for text to be output before each @code{asm}
5026 statement or group of consecutive ones. Normally this is
5027 @code{"#APP"}, which is a comment that has no effect on most
5028 assemblers but tells the GNU assembler that it must check the lines
5029 that follow for all valid assembler constructs.
5033 A C string constant for text to be output after each @code{asm}
5034 statement or group of consecutive ones. Normally this is
5035 @code{"#NO_APP"}, which tells the GNU assembler to resume making the
5036 time-saving assumptions that are valid for ordinary compiler output.
5038 @findex ASM_OUTPUT_SOURCE_FILENAME
5039 @item ASM_OUTPUT_SOURCE_FILENAME (@var{stream}, @var{name})
5040 A C statement to output COFF information or DWARF debugging information
5041 which indicates that filename @var{name} is the current source file to
5042 the stdio stream @var{stream}.
5044 This macro need not be defined if the standard form of output
5045 for the file format in use is appropriate.
5047 @findex OUTPUT_QUOTED_STRING
5048 @item OUTPUT_QUOTED_STRING (@var{stream}, @var{name})
5049 A C statement to output the string @var{string} to the stdio stream
5050 @var{stream}. If you do not call the function @code{output_quoted_string}
5051 in your config files, GNU CC will only call it to output filenames to
5052 the assembler source. So you can use it to canonicalize the format
5053 of the filename using this macro.
5055 @findex ASM_OUTPUT_SOURCE_LINE
5056 @item ASM_OUTPUT_SOURCE_LINE (@var{stream}, @var{line})
5057 A C statement to output DBX or SDB debugging information before code
5058 for line number @var{line} of the current source file to the
5059 stdio stream @var{stream}.
5061 This macro need not be defined if the standard form of debugging
5062 information for the debugger in use is appropriate.
5064 @findex ASM_OUTPUT_IDENT
5065 @item ASM_OUTPUT_IDENT (@var{stream}, @var{string})
5066 A C statement to output something to the assembler file to handle a
5067 @samp{#ident} directive containing the text @var{string}. If this
5068 macro is not defined, nothing is output for a @samp{#ident} directive.
5070 @findex ASM_OUTPUT_SECTION_NAME
5071 @item ASM_OUTPUT_SECTION_NAME (@var{stream}, @var{decl}, @var{name}, @var{reloc})
5072 A C statement to output something to the assembler file to switch to section
5073 @var{name} for object @var{decl} which is either a @code{FUNCTION_DECL}, a
5074 @code{VAR_DECL} or @code{NULL_TREE}. @var{reloc}
5075 indicates whether the initial value of @var{exp} requires link-time
5076 relocations. Some target formats do not support
5077 arbitrary sections. Do not define this macro in such cases.
5079 At present this macro is only used to support section attributes.
5080 When this macro is undefined, section attributes are disabled.
5082 @findex OBJC_PROLOGUE
5084 A C statement to output any assembler statements which are required to
5085 precede any Objective C object definitions or message sending. The
5086 statement is executed only when compiling an Objective C program.
5091 @subsection Output of Data
5093 @c prevent bad page break with this line
5094 This describes data output.
5097 @findex ASM_OUTPUT_LONG_DOUBLE
5098 @findex ASM_OUTPUT_DOUBLE
5099 @findex ASM_OUTPUT_FLOAT
5100 @item ASM_OUTPUT_LONG_DOUBLE (@var{stream}, @var{value})
5101 @itemx ASM_OUTPUT_DOUBLE (@var{stream}, @var{value})
5102 @itemx ASM_OUTPUT_FLOAT (@var{stream}, @var{value})
5103 @itemx ASM_OUTPUT_THREE_QUARTER_FLOAT (@var{stream}, @var{value})
5104 @itemx ASM_OUTPUT_SHORT_FLOAT (@var{stream}, @var{value})
5105 @itemx ASM_OUTPUT_BYTE_FLOAT (@var{stream}, @var{value})
5106 A C statement to output to the stdio stream @var{stream} an assembler
5107 instruction to assemble a floating-point constant of @code{TFmode},
5108 @code{DFmode}, @code{SFmode}, @code{TQFmode}, @code{HFmode}, or
5109 @code{QFmode}, respectively, whose value is @var{value}. @var{value}
5110 will be a C expression of type @code{REAL_VALUE_TYPE}. Macros such as
5111 @code{REAL_VALUE_TO_TARGET_DOUBLE} are useful for writing these
5114 @findex ASM_OUTPUT_QUADRUPLE_INT
5115 @findex ASM_OUTPUT_DOUBLE_INT
5116 @findex ASM_OUTPUT_INT
5117 @findex ASM_OUTPUT_SHORT
5118 @findex ASM_OUTPUT_CHAR
5119 @findex output_addr_const
5120 @item ASM_OUTPUT_QUADRUPLE_INT (@var{stream}, @var{exp})
5121 @itemx ASM_OUTPUT_DOUBLE_INT (@var{stream}, @var{exp})
5122 @itemx ASM_OUTPUT_INT (@var{stream}, @var{exp})
5123 @itemx ASM_OUTPUT_SHORT (@var{stream}, @var{exp})
5124 @itemx ASM_OUTPUT_CHAR (@var{stream}, @var{exp})
5125 A C statement to output to the stdio stream @var{stream} an assembler
5126 instruction to assemble an integer of 16, 8, 4, 2 or 1 bytes,
5127 respectively, whose value is @var{value}. The argument @var{exp} will
5128 be an RTL expression which represents a constant value. Use
5129 @samp{output_addr_const (@var{stream}, @var{exp})} to output this value
5130 as an assembler expression.@refill
5132 For sizes larger than @code{UNITS_PER_WORD}, if the action of a macro
5133 would be identical to repeatedly calling the macro corresponding to
5134 a size of @code{UNITS_PER_WORD}, once for each word, you need not define
5137 @findex ASM_OUTPUT_BYTE
5138 @item ASM_OUTPUT_BYTE (@var{stream}, @var{value})
5139 A C statement to output to the stdio stream @var{stream} an assembler
5140 instruction to assemble a single byte containing the number @var{value}.
5144 A C string constant giving the pseudo-op to use for a sequence of
5145 single-byte constants. If this macro is not defined, the default is
5148 @findex ASM_OUTPUT_ASCII
5149 @item ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len})
5150 A C statement to output to the stdio stream @var{stream} an assembler
5151 instruction to assemble a string constant containing the @var{len}
5152 bytes at @var{ptr}. @var{ptr} will be a C expression of type
5153 @code{char *} and @var{len} a C expression of type @code{int}.
5155 If the assembler has a @code{.ascii} pseudo-op as found in the
5156 Berkeley Unix assembler, do not define the macro
5157 @code{ASM_OUTPUT_ASCII}.
5159 @findex CONSTANT_POOL_BEFORE_FUNCTION
5160 @item CONSTANT_POOL_BEFORE_FUNCTION
5161 You may define this macro as a C expression. You should define the
5162 expression to have a non-zero value if GNU CC should output the constant
5163 pool for a function before the code for the function, or a zero value if
5164 GNU CC should output the constant pool after the function. If you do
5165 not define this macro, the usual case, GNU CC will output the constant
5166 pool before the function.
5168 @findex ASM_OUTPUT_POOL_PROLOGUE
5169 @item ASM_OUTPUT_POOL_PROLOGUE (@var{file} @var{funname} @var{fundecl} @var{size})
5170 A C statement to output assembler commands to define the start of the
5171 constant pool for a function. @var{funname} is a string giving
5172 the name of the function. Should the return type of the function
5173 be required, it can be obtained via @var{fundecl}. @var{size}
5174 is the size, in bytes, of the constant pool that will be written
5175 immediately after this call.
5177 If no constant-pool prefix is required, the usual case, this macro need
5180 @findex ASM_OUTPUT_SPECIAL_POOL_ENTRY
5181 @item ASM_OUTPUT_SPECIAL_POOL_ENTRY (@var{file}, @var{x}, @var{mode}, @var{align}, @var{labelno}, @var{jumpto})
5182 A C statement (with or without semicolon) to output a constant in the
5183 constant pool, if it needs special treatment. (This macro need not do
5184 anything for RTL expressions that can be output normally.)
5186 The argument @var{file} is the standard I/O stream to output the
5187 assembler code on. @var{x} is the RTL expression for the constant to
5188 output, and @var{mode} is the machine mode (in case @var{x} is a
5189 @samp{const_int}). @var{align} is the required alignment for the value
5190 @var{x}; you should output an assembler directive to force this much
5193 The argument @var{labelno} is a number to use in an internal label for
5194 the address of this pool entry. The definition of this macro is
5195 responsible for outputting the label definition at the proper place.
5196 Here is how to do this:
5199 ASM_OUTPUT_INTERNAL_LABEL (@var{file}, "LC", @var{labelno});
5202 When you output a pool entry specially, you should end with a
5203 @code{goto} to the label @var{jumpto}. This will prevent the same pool
5204 entry from being output a second time in the usual manner.
5206 You need not define this macro if it would do nothing.
5208 @findex CONSTANT_AFTER_FUNCTION_P
5209 @item CONSTANT_AFTER_FUNCTION_P (@var{exp})
5210 Define this macro as a C expression which is nonzero if the constant
5211 @var{exp}, of type @code{tree}, should be output after the code for a
5212 function. The compiler will normally output all constants before the
5213 function; you need not define this macro if this is OK.
5215 @findex ASM_OUTPUT_POOL_EPILOGUE
5216 @item ASM_OUTPUT_POOL_EPILOGUE (@var{file} @var{funname} @var{fundecl} @var{size})
5217 A C statement to output assembler commands to at the end of the constant
5218 pool for a function. @var{funname} is a string giving the name of the
5219 function. Should the return type of the function be required, you can
5220 obtain it via @var{fundecl}. @var{size} is the size, in bytes, of the
5221 constant pool that GNU CC wrote immediately before this call.
5223 If no constant-pool epilogue is required, the usual case, you need not
5226 @findex IS_ASM_LOGICAL_LINE_SEPARATOR
5227 @item IS_ASM_LOGICAL_LINE_SEPARATOR (@var{C})
5228 Define this macro as a C expression which is nonzero if @var{C} is
5229 used as a logical line separator by the assembler.
5231 If you do not define this macro, the default is that only
5232 the character @samp{;} is treated as a logical line separator.
5235 @findex ASM_OPEN_PAREN
5236 @findex ASM_CLOSE_PAREN
5237 @item ASM_OPEN_PAREN
5238 @itemx ASM_CLOSE_PAREN
5239 These macros are defined as C string constant, describing the syntax
5240 in the assembler for grouping arithmetic expressions. The following
5241 definitions are correct for most assemblers:
5244 #define ASM_OPEN_PAREN "("
5245 #define ASM_CLOSE_PAREN ")"
5249 These macros are provided by @file{real.h} for writing the definitions
5250 of @code{ASM_OUTPUT_DOUBLE} and the like:
5253 @item REAL_VALUE_TO_TARGET_SINGLE (@var{x}, @var{l})
5254 @itemx REAL_VALUE_TO_TARGET_DOUBLE (@var{x}, @var{l})
5255 @itemx REAL_VALUE_TO_TARGET_LONG_DOUBLE (@var{x}, @var{l})
5256 @findex REAL_VALUE_TO_TARGET_SINGLE
5257 @findex REAL_VALUE_TO_TARGET_DOUBLE
5258 @findex REAL_VALUE_TO_TARGET_LONG_DOUBLE
5259 These translate @var{x}, of type @code{REAL_VALUE_TYPE}, to the target's
5260 floating point representation, and store its bit pattern in the array of
5261 @code{long int} whose address is @var{l}. The number of elements in the
5262 output array is determined by the size of the desired target floating
5263 point data type: 32 bits of it go in each @code{long int} array
5264 element. Each array element holds 32 bits of the result, even if
5265 @code{long int} is wider than 32 bits on the host machine.
5267 The array element values are designed so that you can print them out
5268 using @code{fprintf} in the order they should appear in the target
5271 @item REAL_VALUE_TO_DECIMAL (@var{x}, @var{format}, @var{string})
5272 @findex REAL_VALUE_TO_DECIMAL
5273 This macro converts @var{x}, of type @code{REAL_VALUE_TYPE}, to a
5274 decimal number and stores it as a string into @var{string}.
5275 You must pass, as @var{string}, the address of a long enough block
5276 of space to hold the result.
5278 The argument @var{format} is a @code{printf}-specification that serves
5279 as a suggestion for how to format the output string.
5282 @node Uninitialized Data
5283 @subsection Output of Uninitialized Variables
5285 Each of the macros in this section is used to do the whole job of
5286 outputting a single uninitialized variable.
5289 @findex ASM_OUTPUT_COMMON
5290 @item ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded})
5291 A C statement (sans semicolon) to output to the stdio stream
5292 @var{stream} the assembler definition of a common-label named
5293 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
5294 is the size rounded up to whatever alignment the caller wants.
5296 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
5297 output the name itself; before and after that, output the additional
5298 assembler syntax for defining the name, and a newline.
5300 This macro controls how the assembler definitions of uninitialized
5301 common global variables are output.
5303 @findex ASM_OUTPUT_ALIGNED_COMMON
5304 @item ASM_OUTPUT_ALIGNED_COMMON (@var{stream}, @var{name}, @var{size}, @var{alignment})
5305 Like @code{ASM_OUTPUT_COMMON} except takes the required alignment as a
5306 separate, explicit argument. If you define this macro, it is used in
5307 place of @code{ASM_OUTPUT_COMMON}, and gives you more flexibility in
5308 handling the required alignment of the variable. The alignment is specified
5309 as the number of bits.
5311 @findex ASM_OUTPUT_ALIGNED_DECL_COMMON
5312 @item ASM_OUTPUT_ALIGNED_DECL_COMMON (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
5313 Like @code{ASM_OUTPUT_ALIGNED_COMMON} except that @var{decl} of the
5314 variable to be output, if there is one, or @code{NULL_TREE} if there
5315 is not corresponding variable. If you define this macro, GNU CC wil use it
5316 in place of both @code{ASM_OUTPUT_COMMON} and
5317 @code{ASM_OUTPUT_ALIGNED_COMMON}. Define this macro when you need to see
5318 the variable's decl in order to chose what to output.
5320 @findex ASM_OUTPUT_SHARED_COMMON
5321 @item ASM_OUTPUT_SHARED_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded})
5322 If defined, it is similar to @code{ASM_OUTPUT_COMMON}, except that it
5323 is used when @var{name} is shared. If not defined, @code{ASM_OUTPUT_COMMON}
5326 @findex ASM_OUTPUT_BSS
5327 @item ASM_OUTPUT_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{rounded})
5328 A C statement (sans semicolon) to output to the stdio stream
5329 @var{stream} the assembler definition of uninitialized global @var{decl} named
5330 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
5331 is the size rounded up to whatever alignment the caller wants.
5333 Try to use function @code{asm_output_bss} defined in @file{varasm.c} when
5334 defining this macro. If unable, use the expression
5335 @code{assemble_name (@var{stream}, @var{name})} to output the name itself;
5336 before and after that, output the additional assembler syntax for defining
5337 the name, and a newline.
5339 This macro controls how the assembler definitions of uninitialized global
5340 variables are output. This macro exists to properly support languages like
5341 @code{c++} which do not have @code{common} data. However, this macro currently
5342 is not defined for all targets. If this macro and
5343 @code{ASM_OUTPUT_ALIGNED_BSS} are not defined then @code{ASM_OUTPUT_COMMON}
5344 or @code{ASM_OUTPUT_ALIGNED_COMMON} or
5345 @code{ASM_OUTPUT_ALIGNED_DECL_COMMON} is used.
5347 @findex ASM_OUTPUT_ALIGNED_BSS
5348 @item ASM_OUTPUT_ALIGNED_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
5349 Like @code{ASM_OUTPUT_BSS} except takes the required alignment as a
5350 separate, explicit argument. If you define this macro, it is used in
5351 place of @code{ASM_OUTPUT_BSS}, and gives you more flexibility in
5352 handling the required alignment of the variable. The alignment is specified
5353 as the number of bits.
5355 Try to use function @code{asm_output_aligned_bss} defined in file
5356 @file{varasm.c} when defining this macro.
5358 @findex ASM_OUTPUT_SHARED_BSS
5359 @item ASM_OUTPUT_SHARED_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{rounded})
5360 If defined, it is similar to @code{ASM_OUTPUT_BSS}, except that it
5361 is used when @var{name} is shared. If not defined, @code{ASM_OUTPUT_BSS}
5364 @findex ASM_OUTPUT_LOCAL
5365 @item ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded})
5366 A C statement (sans semicolon) to output to the stdio stream
5367 @var{stream} the assembler definition of a local-common-label named
5368 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
5369 is the size rounded up to whatever alignment the caller wants.
5371 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
5372 output the name itself; before and after that, output the additional
5373 assembler syntax for defining the name, and a newline.
5375 This macro controls how the assembler definitions of uninitialized
5376 static variables are output.
5378 @findex ASM_OUTPUT_ALIGNED_LOCAL
5379 @item ASM_OUTPUT_ALIGNED_LOCAL (@var{stream}, @var{name}, @var{size}, @var{alignment})
5380 Like @code{ASM_OUTPUT_LOCAL} except takes the required alignment as a
5381 separate, explicit argument. If you define this macro, it is used in
5382 place of @code{ASM_OUTPUT_LOCAL}, and gives you more flexibility in
5383 handling the required alignment of the variable. The alignment is specified
5384 as the number of bits.
5386 @findex ASM_OUTPUT_ALIGNED_DECL_LOCAL
5387 @item ASM_OUTPUT_ALIGNED_DECL_LOCAL (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
5388 Like @code{ASM_OUTPUT_ALIGNED_DECL} except that @var{decl} of the
5389 variable to be output, if there is one, or @code{NULL_TREE} if there
5390 is not corresponding variable. If you define this macro, GNU CC wil use it
5391 in place of both @code{ASM_OUTPUT_DECL} and
5392 @code{ASM_OUTPUT_ALIGNED_DECL}. Define this macro when you need to see
5393 the variable's decl in order to chose what to output.
5396 @findex ASM_OUTPUT_SHARED_LOCAL
5397 @item ASM_OUTPUT_SHARED_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded})
5398 If defined, it is similar to @code{ASM_OUTPUT_LOCAL}, except that it
5399 is used when @var{name} is shared. If not defined, @code{ASM_OUTPUT_LOCAL}
5404 @subsection Output and Generation of Labels
5406 @c prevent bad page break with this line
5407 This is about outputting labels.
5410 @findex ASM_OUTPUT_LABEL
5411 @findex assemble_name
5412 @item ASM_OUTPUT_LABEL (@var{stream}, @var{name})
5413 A C statement (sans semicolon) to output to the stdio stream
5414 @var{stream} the assembler definition of a label named @var{name}.
5415 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
5416 output the name itself; before and after that, output the additional
5417 assembler syntax for defining the name, and a newline.
5419 @findex ASM_DECLARE_FUNCTION_NAME
5420 @item ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl})
5421 A C statement (sans semicolon) to output to the stdio stream
5422 @var{stream} any text necessary for declaring the name @var{name} of a
5423 function which is being defined. This macro is responsible for
5424 outputting the label definition (perhaps using
5425 @code{ASM_OUTPUT_LABEL}). The argument @var{decl} is the
5426 @code{FUNCTION_DECL} tree node representing the function.
5428 If this macro is not defined, then the function name is defined in the
5429 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
5431 @findex ASM_DECLARE_FUNCTION_SIZE
5432 @item ASM_DECLARE_FUNCTION_SIZE (@var{stream}, @var{name}, @var{decl})
5433 A C statement (sans semicolon) to output to the stdio stream
5434 @var{stream} any text necessary for declaring the size of a function
5435 which is being defined. The argument @var{name} is the name of the
5436 function. The argument @var{decl} is the @code{FUNCTION_DECL} tree node
5437 representing the function.
5439 If this macro is not defined, then the function size is not defined.
5441 @findex ASM_DECLARE_OBJECT_NAME
5442 @item ASM_DECLARE_OBJECT_NAME (@var{stream}, @var{name}, @var{decl})
5443 A C statement (sans semicolon) to output to the stdio stream
5444 @var{stream} any text necessary for declaring the name @var{name} of an
5445 initialized variable which is being defined. This macro must output the
5446 label definition (perhaps using @code{ASM_OUTPUT_LABEL}). The argument
5447 @var{decl} is the @code{VAR_DECL} tree node representing the variable.
5449 If this macro is not defined, then the variable name is defined in the
5450 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
5452 @findex ASM_FINISH_DECLARE_OBJECT
5453 @item ASM_FINISH_DECLARE_OBJECT (@var{stream}, @var{decl}, @var{toplevel}, @var{atend})
5454 A C statement (sans semicolon) to finish up declaring a variable name
5455 once the compiler has processed its initializer fully and thus has had a
5456 chance to determine the size of an array when controlled by an
5457 initializer. This is used on systems where it's necessary to declare
5458 something about the size of the object.
5460 If you don't define this macro, that is equivalent to defining it to do
5463 @findex ASM_GLOBALIZE_LABEL
5464 @item ASM_GLOBALIZE_LABEL (@var{stream}, @var{name})
5465 A C statement (sans semicolon) to output to the stdio stream
5466 @var{stream} some commands that will make the label @var{name} global;
5467 that is, available for reference from other files. Use the expression
5468 @code{assemble_name (@var{stream}, @var{name})} to output the name
5469 itself; before and after that, output the additional assembler syntax
5470 for making that name global, and a newline.
5472 @findex ASM_WEAKEN_LABEL
5473 @item ASM_WEAKEN_LABEL
5474 A C statement (sans semicolon) to output to the stdio stream
5475 @var{stream} some commands that will make the label @var{name} weak;
5476 that is, available for reference from other files but only used if
5477 no other definition is available. Use the expression
5478 @code{assemble_name (@var{stream}, @var{name})} to output the name
5479 itself; before and after that, output the additional assembler syntax
5480 for making that name weak, and a newline.
5482 If you don't define this macro, GNU CC will not support weak
5483 symbols and you should not define the @code{SUPPORTS_WEAK} macro.
5485 @findex SUPPORTS_WEAK
5487 A C expression which evaluates to true if the target supports weak symbols.
5489 If you don't define this macro, @file{defaults.h} provides a default
5490 definition. If @code{ASM_WEAKEN_LABEL} is defined, the default
5491 definition is @samp{1}; otherwise, it is @samp{0}. Define this macro if
5492 you want to control weak symbol support with a compiler flag such as
5495 @findex MAKE_DECL_ONE_ONLY (@var{decl})
5496 @item MAKE_DECL_ONE_ONLY
5497 A C statement (sans semicolon) to mark @var{decl} to be emitted as a
5498 public symbol such that extra copies in multiple translation units will
5499 be discarded by the linker. Define this macro if your object file
5500 format provides support for this concept, such as the @samp{COMDAT}
5501 section flags in the Microsoft Windows PE/COFF format, and this support
5502 requires changes to @var{decl}, such as putting it in a separate section.
5504 @findex SUPPORTS_ONE_ONLY
5505 @item SUPPORTS_ONE_ONLY
5506 A C expression which evaluates to true if the target supports one-only
5509 If you don't define this macro, @file{varasm.c} provides a default
5510 definition. If @code{MAKE_DECL_ONE_ONLY} is defined, the default
5511 definition is @samp{1}; otherwise, it is @samp{0}. Define this macro if
5512 you want to control one-only symbol support with a compiler flag, or if
5513 setting the @code{DECL_ONE_ONLY} flag is enough to mark a declaration to
5514 be emitted as one-only.
5516 @findex ASM_OUTPUT_EXTERNAL
5517 @item ASM_OUTPUT_EXTERNAL (@var{stream}, @var{decl}, @var{name})
5518 A C statement (sans semicolon) to output to the stdio stream
5519 @var{stream} any text necessary for declaring the name of an external
5520 symbol named @var{name} which is referenced in this compilation but
5521 not defined. The value of @var{decl} is the tree node for the
5524 This macro need not be defined if it does not need to output anything.
5525 The GNU assembler and most Unix assemblers don't require anything.
5527 @findex ASM_OUTPUT_EXTERNAL_LIBCALL
5528 @item ASM_OUTPUT_EXTERNAL_LIBCALL (@var{stream}, @var{symref})
5529 A C statement (sans semicolon) to output on @var{stream} an assembler
5530 pseudo-op to declare a library function name external. The name of the
5531 library function is given by @var{symref}, which has type @code{rtx} and
5532 is a @code{symbol_ref}.
5534 This macro need not be defined if it does not need to output anything.
5535 The GNU assembler and most Unix assemblers don't require anything.
5537 @findex ASM_OUTPUT_LABELREF
5538 @item ASM_OUTPUT_LABELREF (@var{stream}, @var{name})
5539 A C statement (sans semicolon) to output to the stdio stream
5540 @var{stream} a reference in assembler syntax to a label named
5541 @var{name}. This should add @samp{_} to the front of the name, if that
5542 is customary on your operating system, as it is in most Berkeley Unix
5543 systems. This macro is used in @code{assemble_name}.
5545 @ignore @c Seems not to exist anymore.
5546 @findex ASM_OUTPUT_LABELREF_AS_INT
5547 @item ASM_OUTPUT_LABELREF_AS_INT (@var{file}, @var{label})
5548 Define this macro for systems that use the program @code{collect2}.
5549 The definition should be a C statement to output a word containing
5550 a reference to the label @var{label}.
5553 @findex ASM_OUTPUT_INTERNAL_LABEL
5554 @item ASM_OUTPUT_INTERNAL_LABEL (@var{stream}, @var{prefix}, @var{num})
5555 A C statement to output to the stdio stream @var{stream} a label whose
5556 name is made from the string @var{prefix} and the number @var{num}.
5558 It is absolutely essential that these labels be distinct from the labels
5559 used for user-level functions and variables. Otherwise, certain programs
5560 will have name conflicts with internal labels.
5562 It is desirable to exclude internal labels from the symbol table of the
5563 object file. Most assemblers have a naming convention for labels that
5564 should be excluded; on many systems, the letter @samp{L} at the
5565 beginning of a label has this effect. You should find out what
5566 convention your system uses, and follow it.
5568 The usual definition of this macro is as follows:
5571 fprintf (@var{stream}, "L%s%d:\n", @var{prefix}, @var{num})
5574 @findex ASM_GENERATE_INTERNAL_LABEL
5575 @item ASM_GENERATE_INTERNAL_LABEL (@var{string}, @var{prefix}, @var{num})
5576 A C statement to store into the string @var{string} a label whose name
5577 is made from the string @var{prefix} and the number @var{num}.
5579 This string, when output subsequently by @code{assemble_name}, should
5580 produce the output that @code{ASM_OUTPUT_INTERNAL_LABEL} would produce
5581 with the same @var{prefix} and @var{num}.
5583 If the string begins with @samp{*}, then @code{assemble_name} will
5584 output the rest of the string unchanged. It is often convenient for
5585 @code{ASM_GENERATE_INTERNAL_LABEL} to use @samp{*} in this way. If the
5586 string doesn't start with @samp{*}, then @code{ASM_OUTPUT_LABELREF} gets
5587 to output the string, and may change it. (Of course,
5588 @code{ASM_OUTPUT_LABELREF} is also part of your machine description, so
5589 you should know what it does on your machine.)
5591 @findex ASM_FORMAT_PRIVATE_NAME
5592 @item ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number})
5593 A C expression to assign to @var{outvar} (which is a variable of type
5594 @code{char *}) a newly allocated string made from the string
5595 @var{name} and the number @var{number}, with some suitable punctuation
5596 added. Use @code{alloca} to get space for the string.
5598 The string will be used as an argument to @code{ASM_OUTPUT_LABELREF} to
5599 produce an assembler label for an internal static variable whose name is
5600 @var{name}. Therefore, the string must be such as to result in valid
5601 assembler code. The argument @var{number} is different each time this
5602 macro is executed; it prevents conflicts between similarly-named
5603 internal static variables in different scopes.
5605 Ideally this string should not be a valid C identifier, to prevent any
5606 conflict with the user's own symbols. Most assemblers allow periods
5607 or percent signs in assembler symbols; putting at least one of these
5608 between the name and the number will suffice.
5610 @findex ASM_OUTPUT_DEF
5611 @item ASM_OUTPUT_DEF (@var{stream}, @var{name}, @var{value})
5612 A C statement to output to the stdio stream @var{stream} assembler code
5613 which defines (equates) the symbol @var{name} to have the value @var{value}.
5615 If SET_ASM_OP is defined, a default definition is provided which is
5616 correct for most systems.
5618 @findex ASM_OUTPUT_DEFINE_LABEL_DIFFERENCE_SYMBOL
5619 @item ASM_OUTPUT_DEFINE_LABEL_DIFFERENCE_SYMBOL (@var{stream}, @var{symbol}, @var{high}, @var{low})
5620 A C statement to output to the stdio stream @var{stream} assembler code
5621 which defines (equates) the symbol @var{symbol} to have a value equal to
5622 the difference of the two symbols @var{high} and @var{low}, i.e.
5623 @var{high} minus @var{low}. GNU CC guarantees that the symbols @var{high}
5624 and @var{low} are already known by the assembler so that the difference
5625 resolves into a constant.
5627 If SET_ASM_OP is defined, a default definition is provided which is
5628 correct for most systems.
5630 @findex ASM_OUTPUT_WEAK_ALIAS
5631 @item ASM_OUTPUT_WEAK_ALIAS (@var{stream}, @var{name}, @var{value})
5632 A C statement to output to the stdio stream @var{stream} assembler code
5633 which defines (equates) the weak symbol @var{name} to have the value
5636 Define this macro if the target only supports weak aliases; define
5637 ASM_OUTPUT_DEF instead if possible.
5639 @findex OBJC_GEN_METHOD_LABEL
5640 @item OBJC_GEN_METHOD_LABEL (@var{buf}, @var{is_inst}, @var{class_name}, @var{cat_name}, @var{sel_name})
5641 Define this macro to override the default assembler names used for
5642 Objective C methods.
5644 The default name is a unique method number followed by the name of the
5645 class (e.g.@: @samp{_1_Foo}). For methods in categories, the name of
5646 the category is also included in the assembler name (e.g.@:
5649 These names are safe on most systems, but make debugging difficult since
5650 the method's selector is not present in the name. Therefore, particular
5651 systems define other ways of computing names.
5653 @var{buf} is an expression of type @code{char *} which gives you a
5654 buffer in which to store the name; its length is as long as
5655 @var{class_name}, @var{cat_name} and @var{sel_name} put together, plus
5656 50 characters extra.
5658 The argument @var{is_inst} specifies whether the method is an instance
5659 method or a class method; @var{class_name} is the name of the class;
5660 @var{cat_name} is the name of the category (or NULL if the method is not
5661 in a category); and @var{sel_name} is the name of the selector.
5663 On systems where the assembler can handle quoted names, you can use this
5664 macro to provide more human-readable names.
5667 @node Initialization
5668 @subsection How Initialization Functions Are Handled
5669 @cindex initialization routines
5670 @cindex termination routines
5671 @cindex constructors, output of
5672 @cindex destructors, output of
5674 The compiled code for certain languages includes @dfn{constructors}
5675 (also called @dfn{initialization routines})---functions to initialize
5676 data in the program when the program is started. These functions need
5677 to be called before the program is ``started''---that is to say, before
5678 @code{main} is called.
5680 Compiling some languages generates @dfn{destructors} (also called
5681 @dfn{termination routines}) that should be called when the program
5684 To make the initialization and termination functions work, the compiler
5685 must output something in the assembler code to cause those functions to
5686 be called at the appropriate time. When you port the compiler to a new
5687 system, you need to specify how to do this.
5689 There are two major ways that GCC currently supports the execution of
5690 initialization and termination functions. Each way has two variants.
5691 Much of the structure is common to all four variations.
5693 @findex __CTOR_LIST__
5694 @findex __DTOR_LIST__
5695 The linker must build two lists of these functions---a list of
5696 initialization functions, called @code{__CTOR_LIST__}, and a list of
5697 termination functions, called @code{__DTOR_LIST__}.
5699 Each list always begins with an ignored function pointer (which may hold
5700 0, @minus{}1, or a count of the function pointers after it, depending on
5701 the environment). This is followed by a series of zero or more function
5702 pointers to constructors (or destructors), followed by a function
5703 pointer containing zero.
5705 Depending on the operating system and its executable file format, either
5706 @file{crtstuff.c} or @file{libgcc2.c} traverses these lists at startup
5707 time and exit time. Constructors are called in reverse order of the
5708 list; destructors in forward order.
5710 The best way to handle static constructors works only for object file
5711 formats which provide arbitrarily-named sections. A section is set
5712 aside for a list of constructors, and another for a list of destructors.
5713 Traditionally these are called @samp{.ctors} and @samp{.dtors}. Each
5714 object file that defines an initialization function also puts a word in
5715 the constructor section to point to that function. The linker
5716 accumulates all these words into one contiguous @samp{.ctors} section.
5717 Termination functions are handled similarly.
5719 To use this method, you need appropriate definitions of the macros
5720 @code{ASM_OUTPUT_CONSTRUCTOR} and @code{ASM_OUTPUT_DESTRUCTOR}. Usually
5721 you can get them by including @file{svr4.h}.
5723 When arbitrary sections are available, there are two variants, depending
5724 upon how the code in @file{crtstuff.c} is called. On systems that
5725 support an @dfn{init} section which is executed at program startup,
5726 parts of @file{crtstuff.c} are compiled into that section. The
5727 program is linked by the @code{gcc} driver like this:
5730 ld -o @var{output_file} crtbegin.o @dots{} crtend.o -lgcc
5733 The head of a function (@code{__do_global_ctors}) appears in the init
5734 section of @file{crtbegin.o}; the remainder of the function appears in
5735 the init section of @file{crtend.o}. The linker will pull these two
5736 parts of the section together, making a whole function. If any of the
5737 user's object files linked into the middle of it contribute code, then that
5738 code will be executed as part of the body of @code{__do_global_ctors}.
5740 To use this variant, you must define the @code{INIT_SECTION_ASM_OP}
5743 If no init section is available, do not define
5744 @code{INIT_SECTION_ASM_OP}. Then @code{__do_global_ctors} is built into
5745 the text section like all other functions, and resides in
5746 @file{libgcc.a}. When GCC compiles any function called @code{main}, it
5747 inserts a procedure call to @code{__main} as the first executable code
5748 after the function prologue. The @code{__main} function, also defined
5749 in @file{libgcc2.c}, simply calls @file{__do_global_ctors}.
5751 In file formats that don't support arbitrary sections, there are again
5752 two variants. In the simplest variant, the GNU linker (GNU @code{ld})
5753 and an `a.out' format must be used. In this case,
5754 @code{ASM_OUTPUT_CONSTRUCTOR} is defined to produce a @code{.stabs}
5755 entry of type @samp{N_SETT}, referencing the name @code{__CTOR_LIST__},
5756 and with the address of the void function containing the initialization
5757 code as its value. The GNU linker recognizes this as a request to add
5758 the value to a ``set''; the values are accumulated, and are eventually
5759 placed in the executable as a vector in the format described above, with
5760 a leading (ignored) count and a trailing zero element.
5761 @code{ASM_OUTPUT_DESTRUCTOR} is handled similarly. Since no init
5762 section is available, the absence of @code{INIT_SECTION_ASM_OP} causes
5763 the compilation of @code{main} to call @code{__main} as above, starting
5764 the initialization process.
5766 The last variant uses neither arbitrary sections nor the GNU linker.
5767 This is preferable when you want to do dynamic linking and when using
5768 file formats which the GNU linker does not support, such as `ECOFF'. In
5769 this case, @code{ASM_OUTPUT_CONSTRUCTOR} does not produce an
5770 @code{N_SETT} symbol; initialization and termination functions are
5771 recognized simply by their names. This requires an extra program in the
5772 linkage step, called @code{collect2}. This program pretends to be the
5773 linker, for use with GNU CC; it does its job by running the ordinary
5774 linker, but also arranges to include the vectors of initialization and
5775 termination functions. These functions are called via @code{__main} as
5778 Choosing among these configuration options has been simplified by a set
5779 of operating-system-dependent files in the @file{config} subdirectory.
5780 These files define all of the relevant parameters. Usually it is
5781 sufficient to include one into your specific machine-dependent
5782 configuration file. These files are:
5786 For operating systems using the `a.out' format.
5789 For operating systems using the `MachO' format.
5792 For System V Release 3 and similar systems using `COFF' format.
5795 For System V Release 4 and similar systems using `ELF' format.
5798 For the VMS operating system.
5802 The following section describes the specific macros that control and
5803 customize the handling of initialization and termination functions.
5806 @node Macros for Initialization
5807 @subsection Macros Controlling Initialization Routines
5809 Here are the macros that control how the compiler handles initialization
5810 and termination functions:
5813 @findex INIT_SECTION_ASM_OP
5814 @item INIT_SECTION_ASM_OP
5815 If defined, a C string constant for the assembler operation to identify
5816 the following data as initialization code. If not defined, GNU CC will
5817 assume such a section does not exist. When you are using special
5818 sections for initialization and termination functions, this macro also
5819 controls how @file{crtstuff.c} and @file{libgcc2.c} arrange to run the
5820 initialization functions.
5822 @item HAS_INIT_SECTION
5823 @findex HAS_INIT_SECTION
5824 If defined, @code{main} will not call @code{__main} as described above.
5825 This macro should be defined for systems that control the contents of the
5826 init section on a symbol-by-symbol basis, such as OSF/1, and should not
5827 be defined explicitly for systems that support
5828 @code{INIT_SECTION_ASM_OP}.
5830 @item LD_INIT_SWITCH
5831 @findex LD_INIT_SWITCH
5832 If defined, a C string constant for a switch that tells the linker that
5833 the following symbol is an initialization routine.
5835 @item LD_FINI_SWITCH
5836 @findex LD_FINI_SWITCH
5837 If defined, a C string constant for a switch that tells the linker that
5838 the following symbol is a finalization routine.
5841 @findex INVOKE__main
5842 If defined, @code{main} will call @code{__main} despite the presence of
5843 @code{INIT_SECTION_ASM_OP}. This macro should be defined for systems
5844 where the init section is not actually run automatically, but is still
5845 useful for collecting the lists of constructors and destructors.
5847 @item ASM_OUTPUT_CONSTRUCTOR (@var{stream}, @var{name})
5848 @findex ASM_OUTPUT_CONSTRUCTOR
5849 Define this macro as a C statement to output on the stream @var{stream}
5850 the assembler code to arrange to call the function named @var{name} at
5851 initialization time.
5853 Assume that @var{name} is the name of a C function generated
5854 automatically by the compiler. This function takes no arguments. Use
5855 the function @code{assemble_name} to output the name @var{name}; this
5856 performs any system-specific syntactic transformations such as adding an
5859 If you don't define this macro, nothing special is output to arrange to
5860 call the function. This is correct when the function will be called in
5861 some other manner---for example, by means of the @code{collect2} program,
5862 which looks through the symbol table to find these functions by their
5865 @item ASM_OUTPUT_DESTRUCTOR (@var{stream}, @var{name})
5866 @findex ASM_OUTPUT_DESTRUCTOR
5867 This is like @code{ASM_OUTPUT_CONSTRUCTOR} but used for termination
5868 functions rather than initialization functions.
5871 If your system uses @code{collect2} as the means of processing
5872 constructors, then that program normally uses @code{nm} to scan an
5873 object file for constructor functions to be called. On certain kinds of
5874 systems, you can define these macros to make @code{collect2} work faster
5875 (and, in some cases, make it work at all):
5878 @findex OBJECT_FORMAT_COFF
5879 @item OBJECT_FORMAT_COFF
5880 Define this macro if the system uses COFF (Common Object File Format)
5881 object files, so that @code{collect2} can assume this format and scan
5882 object files directly for dynamic constructor/destructor functions.
5884 @findex OBJECT_FORMAT_ROSE
5885 @item OBJECT_FORMAT_ROSE
5886 Define this macro if the system uses ROSE format object files, so that
5887 @code{collect2} can assume this format and scan object files directly
5888 for dynamic constructor/destructor functions.
5890 These macros are effective only in a native compiler; @code{collect2} as
5891 part of a cross compiler always uses @code{nm} for the target machine.
5893 @findex REAL_NM_FILE_NAME
5894 @item REAL_NM_FILE_NAME
5895 Define this macro as a C string constant containing the file name to use
5896 to execute @code{nm}. The default is to search the path normally for
5899 If your system supports shared libraries and has a program to list the
5900 dynamic dependencies of a given library or executable, you can define
5901 these macros to enable support for running initialization and
5902 termination functions in shared libraries:
5906 Define this macro to a C string constant containing the name of the
5907 program which lists dynamic dependencies, like @code{"ldd"} under SunOS 4.
5909 @findex PARSE_LDD_OUTPUT
5910 @item PARSE_LDD_OUTPUT (@var{PTR})
5911 Define this macro to be C code that extracts filenames from the output
5912 of the program denoted by @code{LDD_SUFFIX}. @var{PTR} is a variable
5913 of type @code{char *} that points to the beginning of a line of output
5914 from @code{LDD_SUFFIX}. If the line lists a dynamic dependency, the
5915 code must advance @var{PTR} to the beginning of the filename on that
5916 line. Otherwise, it must set @var{PTR} to @code{NULL}.
5920 @node Instruction Output
5921 @subsection Output of Assembler Instructions
5923 @c prevent bad page break with this line
5924 This describes assembler instruction output.
5927 @findex REGISTER_NAMES
5928 @item REGISTER_NAMES
5929 A C initializer containing the assembler's names for the machine
5930 registers, each one as a C string constant. This is what translates
5931 register numbers in the compiler into assembler language.
5933 @findex ADDITIONAL_REGISTER_NAMES
5934 @item ADDITIONAL_REGISTER_NAMES
5935 If defined, a C initializer for an array of structures containing a name
5936 and a register number. This macro defines additional names for hard
5937 registers, thus allowing the @code{asm} option in declarations to refer
5938 to registers using alternate names.
5940 @findex ASM_OUTPUT_OPCODE
5941 @item ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr})
5942 Define this macro if you are using an unusual assembler that
5943 requires different names for the machine instructions.
5945 The definition is a C statement or statements which output an
5946 assembler instruction opcode to the stdio stream @var{stream}. The
5947 macro-operand @var{ptr} is a variable of type @code{char *} which
5948 points to the opcode name in its ``internal'' form---the form that is
5949 written in the machine description. The definition should output the
5950 opcode name to @var{stream}, performing any translation you desire, and
5951 increment the variable @var{ptr} to point at the end of the opcode
5952 so that it will not be output twice.
5954 In fact, your macro definition may process less than the entire opcode
5955 name, or more than the opcode name; but if you want to process text
5956 that includes @samp{%}-sequences to substitute operands, you must take
5957 care of the substitution yourself. Just be sure to increment
5958 @var{ptr} over whatever text should not be output normally.
5960 @findex recog_operand
5961 If you need to look at the operand values, they can be found as the
5962 elements of @code{recog_operand}.
5964 If the macro definition does nothing, the instruction is output
5967 @findex FINAL_PRESCAN_INSN
5968 @item FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands})
5969 If defined, a C statement to be executed just prior to the output of
5970 assembler code for @var{insn}, to modify the extracted operands so
5971 they will be output differently.
5973 Here the argument @var{opvec} is the vector containing the operands
5974 extracted from @var{insn}, and @var{noperands} is the number of
5975 elements of the vector which contain meaningful data for this insn.
5976 The contents of this vector are what will be used to convert the insn
5977 template into assembler code, so you can change the assembler output
5978 by changing the contents of the vector.
5980 This macro is useful when various assembler syntaxes share a single
5981 file of instruction patterns; by defining this macro differently, you
5982 can cause a large class of instructions to be output differently (such
5983 as with rearranged operands). Naturally, variations in assembler
5984 syntax affecting individual insn patterns ought to be handled by
5985 writing conditional output routines in those patterns.
5987 If this macro is not defined, it is equivalent to a null statement.
5989 @findex FINAL_PRESCAN_LABEL
5990 @item FINAL_PRESCAN_LABEL
5991 If defined, @code{FINAL_PRESCAN_INSN} will be called on each
5992 @code{CODE_LABEL}. In that case, @var{opvec} will be a null pointer and
5993 @var{noperands} will be zero.
5995 @findex PRINT_OPERAND
5996 @item PRINT_OPERAND (@var{stream}, @var{x}, @var{code})
5997 A C compound statement to output to stdio stream @var{stream} the
5998 assembler syntax for an instruction operand @var{x}. @var{x} is an
6001 @var{code} is a value that can be used to specify one of several ways
6002 of printing the operand. It is used when identical operands must be
6003 printed differently depending on the context. @var{code} comes from
6004 the @samp{%} specification that was used to request printing of the
6005 operand. If the specification was just @samp{%@var{digit}} then
6006 @var{code} is 0; if the specification was @samp{%@var{ltr}
6007 @var{digit}} then @var{code} is the ASCII code for @var{ltr}.
6010 If @var{x} is a register, this macro should print the register's name.
6011 The names can be found in an array @code{reg_names} whose type is
6012 @code{char *[]}. @code{reg_names} is initialized from
6013 @code{REGISTER_NAMES}.
6015 When the machine description has a specification @samp{%@var{punct}}
6016 (a @samp{%} followed by a punctuation character), this macro is called
6017 with a null pointer for @var{x} and the punctuation character for
6020 @findex PRINT_OPERAND_PUNCT_VALID_P
6021 @item PRINT_OPERAND_PUNCT_VALID_P (@var{code})
6022 A C expression which evaluates to true if @var{code} is a valid
6023 punctuation character for use in the @code{PRINT_OPERAND} macro. If
6024 @code{PRINT_OPERAND_PUNCT_VALID_P} is not defined, it means that no
6025 punctuation characters (except for the standard one, @samp{%}) are used
6028 @findex PRINT_OPERAND_ADDRESS
6029 @item PRINT_OPERAND_ADDRESS (@var{stream}, @var{x})
6030 A C compound statement to output to stdio stream @var{stream} the
6031 assembler syntax for an instruction operand that is a memory reference
6032 whose address is @var{x}. @var{x} is an RTL expression.
6034 @cindex @code{ENCODE_SECTION_INFO} usage
6035 On some machines, the syntax for a symbolic address depends on the
6036 section that the address refers to. On these machines, define the macro
6037 @code{ENCODE_SECTION_INFO} to store the information into the
6038 @code{symbol_ref}, and then check for it here. @xref{Assembler Format}.
6040 @findex DBR_OUTPUT_SEQEND
6041 @findex dbr_sequence_length
6042 @item DBR_OUTPUT_SEQEND(@var{file})
6043 A C statement, to be executed after all slot-filler instructions have
6044 been output. If necessary, call @code{dbr_sequence_length} to
6045 determine the number of slots filled in a sequence (zero if not
6046 currently outputting a sequence), to decide how many no-ops to output,
6049 Don't define this macro if it has nothing to do, but it is helpful in
6050 reading assembly output if the extent of the delay sequence is made
6051 explicit (e.g. with white space).
6053 @findex final_sequence
6054 Note that output routines for instructions with delay slots must be
6055 prepared to deal with not being output as part of a sequence (i.e.
6056 when the scheduling pass is not run, or when no slot fillers could be
6057 found.) The variable @code{final_sequence} is null when not
6058 processing a sequence, otherwise it contains the @code{sequence} rtx
6061 @findex REGISTER_PREFIX
6062 @findex LOCAL_LABEL_PREFIX
6063 @findex USER_LABEL_PREFIX
6064 @findex IMMEDIATE_PREFIX
6066 @item REGISTER_PREFIX
6067 @itemx LOCAL_LABEL_PREFIX
6068 @itemx USER_LABEL_PREFIX
6069 @itemx IMMEDIATE_PREFIX
6070 If defined, C string expressions to be used for the @samp{%R}, @samp{%L},
6071 @samp{%U}, and @samp{%I} options of @code{asm_fprintf} (see
6072 @file{final.c}). These are useful when a single @file{md} file must
6073 support multiple assembler formats. In that case, the various @file{tm.h}
6074 files can define these macros differently.
6076 @findex ASSEMBLER_DIALECT
6077 @item ASSEMBLER_DIALECT
6078 If your target supports multiple dialects of assembler language (such as
6079 different opcodes), define this macro as a C expression that gives the
6080 numeric index of the assembler language dialect to use, with zero as the
6083 If this macro is defined, you may use constructs of the form
6084 @samp{@{option0|option1|option2@dots{}@}} in the output
6085 templates of patterns (@pxref{Output Template}) or in the first argument
6086 of @code{asm_fprintf}. This construct outputs @samp{option0},
6087 @samp{option1} or @samp{option2}, etc., if the value of
6088 @code{ASSEMBLER_DIALECT} is zero, one or two, etc. Any special
6089 characters within these strings retain their usual meaning.
6091 If you do not define this macro, the characters @samp{@{}, @samp{|} and
6092 @samp{@}} do not have any special meaning when used in templates or
6093 operands to @code{asm_fprintf}.
6095 Define the macros @code{REGISTER_PREFIX}, @code{LOCAL_LABEL_PREFIX},
6096 @code{USER_LABEL_PREFIX} and @code{IMMEDIATE_PREFIX} if you can express
6097 the variations in assembler language syntax with that mechanism. Define
6098 @code{ASSEMBLER_DIALECT} and use the @samp{@{option0|option1@}} syntax
6099 if the syntax variant are larger and involve such things as different
6100 opcodes or operand order.
6102 @findex ASM_OUTPUT_REG_PUSH
6103 @item ASM_OUTPUT_REG_PUSH (@var{stream}, @var{regno})
6104 A C expression to output to @var{stream} some assembler code
6105 which will push hard register number @var{regno} onto the stack.
6106 The code need not be optimal, since this macro is used only when
6109 @findex ASM_OUTPUT_REG_POP
6110 @item ASM_OUTPUT_REG_POP (@var{stream}, @var{regno})
6111 A C expression to output to @var{stream} some assembler code
6112 which will pop hard register number @var{regno} off of the stack.
6113 The code need not be optimal, since this macro is used only when
6117 @node Dispatch Tables
6118 @subsection Output of Dispatch Tables
6120 @c prevent bad page break with this line
6121 This concerns dispatch tables.
6124 @cindex dispatch table
6125 @findex ASM_OUTPUT_ADDR_DIFF_ELT
6126 @item ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{body}, @var{value}, @var{rel})
6127 A C statement to output to the stdio stream @var{stream} an assembler
6128 pseudo-instruction to generate a difference between two labels.
6129 @var{value} and @var{rel} are the numbers of two internal labels. The
6130 definitions of these labels are output using
6131 @code{ASM_OUTPUT_INTERNAL_LABEL}, and they must be printed in the same
6132 way here. For example,
6135 fprintf (@var{stream}, "\t.word L%d-L%d\n",
6136 @var{value}, @var{rel})
6139 You must provide this macro on machines where the addresses in a
6140 dispatch table are relative to the table's own address. If defined, GNU
6141 CC will also use this macro on all machines when producing PIC.
6142 @var{body} is the body of the ADDR_DIFF_VEC; it is provided so that the
6143 mode and flags can be read.
6145 @findex ASM_OUTPUT_ADDR_VEC_ELT
6146 @item ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value})
6147 This macro should be provided on machines where the addresses
6148 in a dispatch table are absolute.
6150 The definition should be a C statement to output to the stdio stream
6151 @var{stream} an assembler pseudo-instruction to generate a reference to
6152 a label. @var{value} is the number of an internal label whose
6153 definition is output using @code{ASM_OUTPUT_INTERNAL_LABEL}.
6157 fprintf (@var{stream}, "\t.word L%d\n", @var{value})
6160 @findex ASM_OUTPUT_CASE_LABEL
6161 @item ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table})
6162 Define this if the label before a jump-table needs to be output
6163 specially. The first three arguments are the same as for
6164 @code{ASM_OUTPUT_INTERNAL_LABEL}; the fourth argument is the
6165 jump-table which follows (a @code{jump_insn} containing an
6166 @code{addr_vec} or @code{addr_diff_vec}).
6168 This feature is used on system V to output a @code{swbeg} statement
6171 If this macro is not defined, these labels are output with
6172 @code{ASM_OUTPUT_INTERNAL_LABEL}.
6174 @findex ASM_OUTPUT_CASE_END
6175 @item ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table})
6176 Define this if something special must be output at the end of a
6177 jump-table. The definition should be a C statement to be executed
6178 after the assembler code for the table is written. It should write
6179 the appropriate code to stdio stream @var{stream}. The argument
6180 @var{table} is the jump-table insn, and @var{num} is the label-number
6181 of the preceding label.
6183 If this macro is not defined, nothing special is output at the end of
6187 @node Exception Region Output
6188 @subsection Assembler Commands for Exception Regions
6190 @c prevent bad page break with this line
6192 This describes commands marking the start and the end of an exception
6196 @findex ASM_OUTPUT_EH_REGION_BEG
6197 @item ASM_OUTPUT_EH_REGION_BEG ()
6198 A C expression to output text to mark the start of an exception region.
6200 This macro need not be defined on most platforms.
6202 @findex ASM_OUTPUT_EH_REGION_END
6203 @item ASM_OUTPUT_EH_REGION_END ()
6204 A C expression to output text to mark the end of an exception region.
6206 This macro need not be defined on most platforms.
6208 @findex EXCEPTION_SECTION
6209 @item EXCEPTION_SECTION ()
6210 A C expression to switch to the section in which the main
6211 exception table is to be placed (@pxref{Sections}). The default is a
6212 section named @code{.gcc_except_table} on machines that support named
6213 sections via @code{ASM_OUTPUT_SECTION_NAME}, otherwise if @samp{-fpic}
6214 or @samp{-fPIC} is in effect, the @code{data_section}, otherwise the
6215 @code{readonly_data_section}.
6217 @findex EH_FRAME_SECTION_ASM_OP
6218 @item EH_FRAME_SECTION_ASM_OP
6219 If defined, a C string constant for the assembler operation to switch to
6220 the section for exception handling frame unwind information. If not
6221 defined, GNU CC will provide a default definition if the target supports
6222 named sections. @file{crtstuff.c} uses this macro to switch to the
6223 appropriate section.
6225 You should define this symbol if your target supports DWARF 2 frame
6226 unwind information and the default definition does not work.
6228 @findex OMIT_EH_TABLE
6229 @item OMIT_EH_TABLE ()
6230 A C expression that is nonzero if the normal exception table output
6233 This macro need not be defined on most platforms.
6235 @findex EH_TABLE_LOOKUP
6236 @item EH_TABLE_LOOKUP ()
6237 Alternate runtime support for looking up an exception at runtime and
6238 finding the associated handler, if the default method won't work.
6240 This macro need not be defined on most platforms.
6242 @findex DOESNT_NEED_UNWINDER
6243 @item DOESNT_NEED_UNWINDER
6244 A C expression that decides whether or not the current function needs to
6245 have a function unwinder generated for it. See the file @code{except.c}
6246 for details on when to define this, and how.
6248 @findex MASK_RETURN_ADDR
6249 @item MASK_RETURN_ADDR
6250 An rtx used to mask the return address found via RETURN_ADDR_RTX, so
6251 that it does not contain any extraneous set bits in it.
6253 @findex DWARF2_UNWIND_INFO
6254 @item DWARF2_UNWIND_INFO
6255 Define this macro to 0 if your target supports DWARF 2 frame unwind
6256 information, but it does not yet work with exception handling.
6257 Otherwise, if your target supports this information (if it defines
6258 @samp{INCOMING_RETURN_ADDR_RTX} and either @samp{UNALIGNED_INT_ASM_OP}
6259 or @samp{OBJECT_FORMAT_ELF}), GCC will provide a default definition of
6262 If this macro is defined to 1, the DWARF 2 unwinder will be the default
6263 exception handling mechanism; otherwise, setjmp/longjmp will be used by
6266 If this macro is defined to anything, the DWARF 2 unwinder will be used
6267 instead of inline unwinders and __unwind_function in the non-setjmp case.
6271 @node Alignment Output
6272 @subsection Assembler Commands for Alignment
6274 @c prevent bad page break with this line
6275 This describes commands for alignment.
6278 @findex LABEL_ALIGN_AFTER_BARRIER
6279 @item LABEL_ALIGN_AFTER_BARRIER (@var{label})
6280 The alignment (log base 2) to put in front of @var{label}, which follows
6283 This macro need not be defined if you don't want any special alignment
6284 to be done at such a time. Most machine descriptions do not currently
6288 @item LOOP_ALIGN (@var{label})
6289 The alignment (log base 2) to put in front of @var{label}, which follows
6290 a NOTE_INSN_LOOP_BEG note.
6292 This macro need not be defined if you don't want any special alignment
6293 to be done at such a time. Most machine descriptions do not currently
6297 @item LABEL_ALIGN (@var{label})
6298 The alignment (log base 2) to put in front of @var{label}.
6299 If LABEL_ALIGN_AFTER_BARRIER / LOOP_ALIGN specify a different alignment,
6300 the maximum of the specified values is used.
6302 @findex ASM_OUTPUT_SKIP
6303 @item ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes})
6304 A C statement to output to the stdio stream @var{stream} an assembler
6305 instruction to advance the location counter by @var{nbytes} bytes.
6306 Those bytes should be zero when loaded. @var{nbytes} will be a C
6307 expression of type @code{int}.
6309 @findex ASM_NO_SKIP_IN_TEXT
6310 @item ASM_NO_SKIP_IN_TEXT
6311 Define this macro if @code{ASM_OUTPUT_SKIP} should not be used in the
6312 text section because it fails to put zeros in the bytes that are skipped.
6313 This is true on many Unix systems, where the pseudo--op to skip bytes
6314 produces no-op instructions rather than zeros when used in the text
6317 @findex ASM_OUTPUT_ALIGN
6318 @item ASM_OUTPUT_ALIGN (@var{stream}, @var{power})
6319 A C statement to output to the stdio stream @var{stream} an assembler
6320 command to advance the location counter to a multiple of 2 to the
6321 @var{power} bytes. @var{power} will be a C expression of type @code{int}.
6325 @node Debugging Info
6326 @section Controlling Debugging Information Format
6328 @c prevent bad page break with this line
6329 This describes how to specify debugging information.
6332 * All Debuggers:: Macros that affect all debugging formats uniformly.
6333 * DBX Options:: Macros enabling specific options in DBX format.
6334 * DBX Hooks:: Hook macros for varying DBX format.
6335 * File Names and DBX:: Macros controlling output of file names in DBX format.
6336 * SDB and DWARF:: Macros for SDB (COFF) and DWARF formats.
6340 @subsection Macros Affecting All Debugging Formats
6342 @c prevent bad page break with this line
6343 These macros affect all debugging formats.
6346 @findex DBX_REGISTER_NUMBER
6347 @item DBX_REGISTER_NUMBER (@var{regno})
6348 A C expression that returns the DBX register number for the compiler
6349 register number @var{regno}. In simple cases, the value of this
6350 expression may be @var{regno} itself. But sometimes there are some
6351 registers that the compiler knows about and DBX does not, or vice
6352 versa. In such cases, some register may need to have one number in
6353 the compiler and another for DBX.
6355 If two registers have consecutive numbers inside GNU CC, and they can be
6356 used as a pair to hold a multiword value, then they @emph{must} have
6357 consecutive numbers after renumbering with @code{DBX_REGISTER_NUMBER}.
6358 Otherwise, debuggers will be unable to access such a pair, because they
6359 expect register pairs to be consecutive in their own numbering scheme.
6361 If you find yourself defining @code{DBX_REGISTER_NUMBER} in way that
6362 does not preserve register pairs, then what you must do instead is
6363 redefine the actual register numbering scheme.
6365 @findex DEBUGGER_AUTO_OFFSET
6366 @item DEBUGGER_AUTO_OFFSET (@var{x})
6367 A C expression that returns the integer offset value for an automatic
6368 variable having address @var{x} (an RTL expression). The default
6369 computation assumes that @var{x} is based on the frame-pointer and
6370 gives the offset from the frame-pointer. This is required for targets
6371 that produce debugging output for DBX or COFF-style debugging output
6372 for SDB and allow the frame-pointer to be eliminated when the
6373 @samp{-g} options is used.
6375 @findex DEBUGGER_ARG_OFFSET
6376 @item DEBUGGER_ARG_OFFSET (@var{offset}, @var{x})
6377 A C expression that returns the integer offset value for an argument
6378 having address @var{x} (an RTL expression). The nominal offset is
6381 @findex PREFERRED_DEBUGGING_TYPE
6382 @item PREFERRED_DEBUGGING_TYPE
6383 A C expression that returns the type of debugging output GNU CC should
6384 produce when the user specifies just @samp{-g}. Define
6385 this if you have arranged for GNU CC to support more than one format of
6386 debugging output. Currently, the allowable values are @code{DBX_DEBUG},
6387 @code{SDB_DEBUG}, @code{DWARF_DEBUG}, @code{DWARF2_DEBUG}, and
6390 When the user specifies @samp{-ggdb}, GNU CC normally also uses the
6391 value of this macro to select the debugging output format, but with two
6392 exceptions. If @code{DWARF2_DEBUGGING_INFO} is defined and
6393 @code{LINKER_DOES_NOT_WORK_WITH_DWARF2} is not defined, GNU CC uses the
6394 value @code{DWARF2_DEBUG}. Otherwise, if @code{DBX_DEBUGGING_INFO} is
6395 defined, GNU CC uses @code{DBX_DEBUG}.
6397 The value of this macro only affects the default debugging output; the
6398 user can always get a specific type of output by using @samp{-gstabs},
6399 @samp{-gcoff}, @samp{-gdwarf-1}, @samp{-gdwarf-2}, or @samp{-gxcoff}.
6403 @subsection Specific Options for DBX Output
6405 @c prevent bad page break with this line
6406 These are specific options for DBX output.
6409 @findex DBX_DEBUGGING_INFO
6410 @item DBX_DEBUGGING_INFO
6411 Define this macro if GNU CC should produce debugging output for DBX
6412 in response to the @samp{-g} option.
6414 @findex XCOFF_DEBUGGING_INFO
6415 @item XCOFF_DEBUGGING_INFO
6416 Define this macro if GNU CC should produce XCOFF format debugging output
6417 in response to the @samp{-g} option. This is a variant of DBX format.
6419 @findex DEFAULT_GDB_EXTENSIONS
6420 @item DEFAULT_GDB_EXTENSIONS
6421 Define this macro to control whether GNU CC should by default generate
6422 GDB's extended version of DBX debugging information (assuming DBX-format
6423 debugging information is enabled at all). If you don't define the
6424 macro, the default is 1: always generate the extended information
6425 if there is any occasion to.
6427 @findex DEBUG_SYMS_TEXT
6428 @item DEBUG_SYMS_TEXT
6429 Define this macro if all @code{.stabs} commands should be output while
6430 in the text section.
6432 @findex ASM_STABS_OP
6434 A C string constant naming the assembler pseudo op to use instead of
6435 @code{.stabs} to define an ordinary debugging symbol. If you don't
6436 define this macro, @code{.stabs} is used. This macro applies only to
6437 DBX debugging information format.
6439 @findex ASM_STABD_OP
6441 A C string constant naming the assembler pseudo op to use instead of
6442 @code{.stabd} to define a debugging symbol whose value is the current
6443 location. If you don't define this macro, @code{.stabd} is used.
6444 This macro applies only to DBX debugging information format.
6446 @findex ASM_STABN_OP
6448 A C string constant naming the assembler pseudo op to use instead of
6449 @code{.stabn} to define a debugging symbol with no name. If you don't
6450 define this macro, @code{.stabn} is used. This macro applies only to
6451 DBX debugging information format.
6453 @findex DBX_NO_XREFS
6455 Define this macro if DBX on your system does not support the construct
6456 @samp{xs@var{tagname}}. On some systems, this construct is used to
6457 describe a forward reference to a structure named @var{tagname}.
6458 On other systems, this construct is not supported at all.
6460 @findex DBX_CONTIN_LENGTH
6461 @item DBX_CONTIN_LENGTH
6462 A symbol name in DBX-format debugging information is normally
6463 continued (split into two separate @code{.stabs} directives) when it
6464 exceeds a certain length (by default, 80 characters). On some
6465 operating systems, DBX requires this splitting; on others, splitting
6466 must not be done. You can inhibit splitting by defining this macro
6467 with the value zero. You can override the default splitting-length by
6468 defining this macro as an expression for the length you desire.
6470 @findex DBX_CONTIN_CHAR
6471 @item DBX_CONTIN_CHAR
6472 Normally continuation is indicated by adding a @samp{\} character to
6473 the end of a @code{.stabs} string when a continuation follows. To use
6474 a different character instead, define this macro as a character
6475 constant for the character you want to use. Do not define this macro
6476 if backslash is correct for your system.
6478 @findex DBX_STATIC_STAB_DATA_SECTION
6479 @item DBX_STATIC_STAB_DATA_SECTION
6480 Define this macro if it is necessary to go to the data section before
6481 outputting the @samp{.stabs} pseudo-op for a non-global static
6484 @findex DBX_TYPE_DECL_STABS_CODE
6485 @item DBX_TYPE_DECL_STABS_CODE
6486 The value to use in the ``code'' field of the @code{.stabs} directive
6487 for a typedef. The default is @code{N_LSYM}.
6489 @findex DBX_STATIC_CONST_VAR_CODE
6490 @item DBX_STATIC_CONST_VAR_CODE
6491 The value to use in the ``code'' field of the @code{.stabs} directive
6492 for a static variable located in the text section. DBX format does not
6493 provide any ``right'' way to do this. The default is @code{N_FUN}.
6495 @findex DBX_REGPARM_STABS_CODE
6496 @item DBX_REGPARM_STABS_CODE
6497 The value to use in the ``code'' field of the @code{.stabs} directive
6498 for a parameter passed in registers. DBX format does not provide any
6499 ``right'' way to do this. The default is @code{N_RSYM}.
6501 @findex DBX_REGPARM_STABS_LETTER
6502 @item DBX_REGPARM_STABS_LETTER
6503 The letter to use in DBX symbol data to identify a symbol as a parameter
6504 passed in registers. DBX format does not customarily provide any way to
6505 do this. The default is @code{'P'}.
6507 @findex DBX_MEMPARM_STABS_LETTER
6508 @item DBX_MEMPARM_STABS_LETTER
6509 The letter to use in DBX symbol data to identify a symbol as a stack
6510 parameter. The default is @code{'p'}.
6512 @findex DBX_FUNCTION_FIRST
6513 @item DBX_FUNCTION_FIRST
6514 Define this macro if the DBX information for a function and its
6515 arguments should precede the assembler code for the function. Normally,
6516 in DBX format, the debugging information entirely follows the assembler
6519 @findex DBX_LBRAC_FIRST
6520 @item DBX_LBRAC_FIRST
6521 Define this macro if the @code{N_LBRAC} symbol for a block should
6522 precede the debugging information for variables and functions defined in
6523 that block. Normally, in DBX format, the @code{N_LBRAC} symbol comes
6526 @findex DBX_BLOCKS_FUNCTION_RELATIVE
6527 @item DBX_BLOCKS_FUNCTION_RELATIVE
6528 Define this macro if the value of a symbol describing the scope of a
6529 block (@code{N_LBRAC} or @code{N_RBRAC}) should be relative to the start
6530 of the enclosing function. Normally, GNU C uses an absolute address.
6532 @findex DBX_USE_BINCL
6534 Define this macro if GNU C should generate @code{N_BINCL} and
6535 @code{N_EINCL} stabs for included header files, as on Sun systems. This
6536 macro also directs GNU C to output a type number as a pair of a file
6537 number and a type number within the file. Normally, GNU C does not
6538 generate @code{N_BINCL} or @code{N_EINCL} stabs, and it outputs a single
6539 number for a type number.
6543 @subsection Open-Ended Hooks for DBX Format
6545 @c prevent bad page break with this line
6546 These are hooks for DBX format.
6549 @findex DBX_OUTPUT_LBRAC
6550 @item DBX_OUTPUT_LBRAC (@var{stream}, @var{name})
6551 Define this macro to say how to output to @var{stream} the debugging
6552 information for the start of a scope level for variable names. The
6553 argument @var{name} is the name of an assembler symbol (for use with
6554 @code{assemble_name}) whose value is the address where the scope begins.
6556 @findex DBX_OUTPUT_RBRAC
6557 @item DBX_OUTPUT_RBRAC (@var{stream}, @var{name})
6558 Like @code{DBX_OUTPUT_LBRAC}, but for the end of a scope level.
6560 @findex DBX_OUTPUT_ENUM
6561 @item DBX_OUTPUT_ENUM (@var{stream}, @var{type})
6562 Define this macro if the target machine requires special handling to
6563 output an enumeration type. The definition should be a C statement
6564 (sans semicolon) to output the appropriate information to @var{stream}
6565 for the type @var{type}.
6567 @findex DBX_OUTPUT_FUNCTION_END
6568 @item DBX_OUTPUT_FUNCTION_END (@var{stream}, @var{function})
6569 Define this macro if the target machine requires special output at the
6570 end of the debugging information for a function. The definition should
6571 be a C statement (sans semicolon) to output the appropriate information
6572 to @var{stream}. @var{function} is the @code{FUNCTION_DECL} node for
6575 @findex DBX_OUTPUT_STANDARD_TYPES
6576 @item DBX_OUTPUT_STANDARD_TYPES (@var{syms})
6577 Define this macro if you need to control the order of output of the
6578 standard data types at the beginning of compilation. The argument
6579 @var{syms} is a @code{tree} which is a chain of all the predefined
6580 global symbols, including names of data types.
6582 Normally, DBX output starts with definitions of the types for integers
6583 and characters, followed by all the other predefined types of the
6584 particular language in no particular order.
6586 On some machines, it is necessary to output different particular types
6587 first. To do this, define @code{DBX_OUTPUT_STANDARD_TYPES} to output
6588 those symbols in the necessary order. Any predefined types that you
6589 don't explicitly output will be output afterward in no particular order.
6591 Be careful not to define this macro so that it works only for C. There
6592 are no global variables to access most of the built-in types, because
6593 another language may have another set of types. The way to output a
6594 particular type is to look through @var{syms} to see if you can find it.
6600 for (decl = syms; decl; decl = TREE_CHAIN (decl))
6601 if (!strcmp (IDENTIFIER_POINTER (DECL_NAME (decl)),
6603 dbxout_symbol (decl);
6609 This does nothing if the expected type does not exist.
6611 See the function @code{init_decl_processing} in @file{c-decl.c} to find
6612 the names to use for all the built-in C types.
6614 Here is another way of finding a particular type:
6616 @c this is still overfull. --mew 10feb93
6620 for (decl = syms; decl; decl = TREE_CHAIN (decl))
6621 if (TREE_CODE (decl) == TYPE_DECL
6622 && (TREE_CODE (TREE_TYPE (decl))
6624 && TYPE_PRECISION (TREE_TYPE (decl)) == 16
6625 && TYPE_UNSIGNED (TREE_TYPE (decl)))
6627 /* @r{This must be @code{unsigned short}.} */
6628 dbxout_symbol (decl);
6634 @findex NO_DBX_FUNCTION_END
6635 @item NO_DBX_FUNCTION_END
6636 Some stabs encapsulation formats (in particular ECOFF), cannot handle the
6637 @code{.stabs "",N_FUN,,0,0,Lscope-function-1} gdb dbx extention construct.
6638 On those machines, define this macro to turn this feature off without
6639 disturbing the rest of the gdb extensions.
6643 @node File Names and DBX
6644 @subsection File Names in DBX Format
6646 @c prevent bad page break with this line
6647 This describes file names in DBX format.
6650 @findex DBX_WORKING_DIRECTORY
6651 @item DBX_WORKING_DIRECTORY
6652 Define this if DBX wants to have the current directory recorded in each
6655 Note that the working directory is always recorded if GDB extensions are
6658 @findex DBX_OUTPUT_MAIN_SOURCE_FILENAME
6659 @item DBX_OUTPUT_MAIN_SOURCE_FILENAME (@var{stream}, @var{name})
6660 A C statement to output DBX debugging information to the stdio stream
6661 @var{stream} which indicates that file @var{name} is the main source
6662 file---the file specified as the input file for compilation.
6663 This macro is called only once, at the beginning of compilation.
6665 This macro need not be defined if the standard form of output
6666 for DBX debugging information is appropriate.
6668 @findex DBX_OUTPUT_MAIN_SOURCE_DIRECTORY
6669 @item DBX_OUTPUT_MAIN_SOURCE_DIRECTORY (@var{stream}, @var{name})
6670 A C statement to output DBX debugging information to the stdio stream
6671 @var{stream} which indicates that the current directory during
6672 compilation is named @var{name}.
6674 This macro need not be defined if the standard form of output
6675 for DBX debugging information is appropriate.
6677 @findex DBX_OUTPUT_MAIN_SOURCE_FILE_END
6678 @item DBX_OUTPUT_MAIN_SOURCE_FILE_END (@var{stream}, @var{name})
6679 A C statement to output DBX debugging information at the end of
6680 compilation of the main source file @var{name}.
6682 If you don't define this macro, nothing special is output at the end
6683 of compilation, which is correct for most machines.
6685 @findex DBX_OUTPUT_SOURCE_FILENAME
6686 @item DBX_OUTPUT_SOURCE_FILENAME (@var{stream}, @var{name})
6687 A C statement to output DBX debugging information to the stdio stream
6688 @var{stream} which indicates that file @var{name} is the current source
6689 file. This output is generated each time input shifts to a different
6690 source file as a result of @samp{#include}, the end of an included file,
6691 or a @samp{#line} command.
6693 This macro need not be defined if the standard form of output
6694 for DBX debugging information is appropriate.
6699 @subsection Macros for SDB and DWARF Output
6701 @c prevent bad page break with this line
6702 Here are macros for SDB and DWARF output.
6705 @findex SDB_DEBUGGING_INFO
6706 @item SDB_DEBUGGING_INFO
6707 Define this macro if GNU CC should produce COFF-style debugging output
6708 for SDB in response to the @samp{-g} option.
6710 @findex DWARF_DEBUGGING_INFO
6711 @item DWARF_DEBUGGING_INFO
6712 Define this macro if GNU CC should produce dwarf format debugging output
6713 in response to the @samp{-g} option.
6715 @findex DWARF2_DEBUGGING_INFO
6716 @item DWARF2_DEBUGGING_INFO
6717 Define this macro if GNU CC should produce dwarf version 2 format
6718 debugging output in response to the @samp{-g} option.
6720 To support optional call frame debugging information, you must also
6721 define @code{INCOMING_RETURN_ADDR_RTX} and either set
6722 @code{RTX_FRAME_RELATED_P} on the prologue insns if you use RTL for the
6723 prologue, or call @code{dwarf2out_def_cfa} and @code{dwarf2out_reg_save}
6724 as appropriate from @code{FUNCTION_PROLOGUE} if you don't.
6726 @findex LINKER_DOES_NOT_WORK_WITH_DWARF2
6727 @item LINKER_DOES_NOT_WORK_WITH_DWARF2
6728 Define this macro if the linker does not work with Dwarf version 2.
6729 Normally, if the user specifies only @samp{-ggdb} GNU CC will use Dwarf
6730 version 2 if available; this macro disables this. See the description
6731 of the @code{PREFERRED_DEBUGGING_TYPE} macro for more details.
6733 @findex PUT_SDB_@dots{}
6734 @item PUT_SDB_@dots{}
6735 Define these macros to override the assembler syntax for the special
6736 SDB assembler directives. See @file{sdbout.c} for a list of these
6737 macros and their arguments. If the standard syntax is used, you need
6738 not define them yourself.
6742 Some assemblers do not support a semicolon as a delimiter, even between
6743 SDB assembler directives. In that case, define this macro to be the
6744 delimiter to use (usually @samp{\n}). It is not necessary to define
6745 a new set of @code{PUT_SDB_@var{op}} macros if this is the only change
6748 @findex SDB_GENERATE_FAKE
6749 @item SDB_GENERATE_FAKE
6750 Define this macro to override the usual method of constructing a dummy
6751 name for anonymous structure and union types. See @file{sdbout.c} for
6754 @findex SDB_ALLOW_UNKNOWN_REFERENCES
6755 @item SDB_ALLOW_UNKNOWN_REFERENCES
6756 Define this macro to allow references to unknown structure,
6757 union, or enumeration tags to be emitted. Standard COFF does not
6758 allow handling of unknown references, MIPS ECOFF has support for
6761 @findex SDB_ALLOW_FORWARD_REFERENCES
6762 @item SDB_ALLOW_FORWARD_REFERENCES
6763 Define this macro to allow references to structure, union, or
6764 enumeration tags that have not yet been seen to be handled. Some
6765 assemblers choke if forward tags are used, while some require it.
6768 @node Cross-compilation
6769 @section Cross Compilation and Floating Point
6770 @cindex cross compilation and floating point
6771 @cindex floating point and cross compilation
6773 While all modern machines use 2's complement representation for integers,
6774 there are a variety of representations for floating point numbers. This
6775 means that in a cross-compiler the representation of floating point numbers
6776 in the compiled program may be different from that used in the machine
6777 doing the compilation.
6780 Because different representation systems may offer different amounts of
6781 range and precision, the cross compiler cannot safely use the host
6782 machine's floating point arithmetic. Therefore, floating point constants
6783 must be represented in the target machine's format. This means that the
6784 cross compiler cannot use @code{atof} to parse a floating point constant;
6785 it must have its own special routine to use instead. Also, constant
6786 folding must emulate the target machine's arithmetic (or must not be done
6789 The macros in the following table should be defined only if you are cross
6790 compiling between different floating point formats.
6792 Otherwise, don't define them. Then default definitions will be set up which
6793 use @code{double} as the data type, @code{==} to test for equality, etc.
6795 You don't need to worry about how many times you use an operand of any
6796 of these macros. The compiler never uses operands which have side effects.
6799 @findex REAL_VALUE_TYPE
6800 @item REAL_VALUE_TYPE
6801 A macro for the C data type to be used to hold a floating point value
6802 in the target machine's format. Typically this would be a
6803 @code{struct} containing an array of @code{int}.
6805 @findex REAL_VALUES_EQUAL
6806 @item REAL_VALUES_EQUAL (@var{x}, @var{y})
6807 A macro for a C expression which compares for equality the two values,
6808 @var{x} and @var{y}, both of type @code{REAL_VALUE_TYPE}.
6810 @findex REAL_VALUES_LESS
6811 @item REAL_VALUES_LESS (@var{x}, @var{y})
6812 A macro for a C expression which tests whether @var{x} is less than
6813 @var{y}, both values being of type @code{REAL_VALUE_TYPE} and
6814 interpreted as floating point numbers in the target machine's
6817 @findex REAL_VALUE_LDEXP
6819 @item REAL_VALUE_LDEXP (@var{x}, @var{scale})
6820 A macro for a C expression which performs the standard library
6821 function @code{ldexp}, but using the target machine's floating point
6822 representation. Both @var{x} and the value of the expression have
6823 type @code{REAL_VALUE_TYPE}. The second argument, @var{scale}, is an
6826 @findex REAL_VALUE_FIX
6827 @item REAL_VALUE_FIX (@var{x})
6828 A macro whose definition is a C expression to convert the target-machine
6829 floating point value @var{x} to a signed integer. @var{x} has type
6830 @code{REAL_VALUE_TYPE}.
6832 @findex REAL_VALUE_UNSIGNED_FIX
6833 @item REAL_VALUE_UNSIGNED_FIX (@var{x})
6834 A macro whose definition is a C expression to convert the target-machine
6835 floating point value @var{x} to an unsigned integer. @var{x} has type
6836 @code{REAL_VALUE_TYPE}.
6838 @findex REAL_VALUE_RNDZINT
6839 @item REAL_VALUE_RNDZINT (@var{x})
6840 A macro whose definition is a C expression to round the target-machine
6841 floating point value @var{x} towards zero to an integer value (but still
6842 as a floating point number). @var{x} has type @code{REAL_VALUE_TYPE},
6843 and so does the value.
6845 @findex REAL_VALUE_UNSIGNED_RNDZINT
6846 @item REAL_VALUE_UNSIGNED_RNDZINT (@var{x})
6847 A macro whose definition is a C expression to round the target-machine
6848 floating point value @var{x} towards zero to an unsigned integer value
6849 (but still represented as a floating point number). @var{x} has type
6850 @code{REAL_VALUE_TYPE}, and so does the value.
6852 @findex REAL_VALUE_ATOF
6853 @item REAL_VALUE_ATOF (@var{string}, @var{mode})
6854 A macro for a C expression which converts @var{string}, an expression of
6855 type @code{char *}, into a floating point number in the target machine's
6856 representation for mode @var{mode}. The value has type
6857 @code{REAL_VALUE_TYPE}.
6859 @findex REAL_INFINITY
6861 Define this macro if infinity is a possible floating point value, and
6862 therefore division by 0 is legitimate.
6864 @findex REAL_VALUE_ISINF
6866 @item REAL_VALUE_ISINF (@var{x})
6867 A macro for a C expression which determines whether @var{x}, a floating
6868 point value, is infinity. The value has type @code{int}.
6869 By default, this is defined to call @code{isinf}.
6871 @findex REAL_VALUE_ISNAN
6873 @item REAL_VALUE_ISNAN (@var{x})
6874 A macro for a C expression which determines whether @var{x}, a floating
6875 point value, is a ``nan'' (not-a-number). The value has type
6876 @code{int}. By default, this is defined to call @code{isnan}.
6879 @cindex constant folding and floating point
6880 Define the following additional macros if you want to make floating
6881 point constant folding work while cross compiling. If you don't
6882 define them, cross compilation is still possible, but constant folding
6883 will not happen for floating point values.
6886 @findex REAL_ARITHMETIC
6887 @item REAL_ARITHMETIC (@var{output}, @var{code}, @var{x}, @var{y})
6888 A macro for a C statement which calculates an arithmetic operation of
6889 the two floating point values @var{x} and @var{y}, both of type
6890 @code{REAL_VALUE_TYPE} in the target machine's representation, to
6891 produce a result of the same type and representation which is stored
6892 in @var{output} (which will be a variable).
6894 The operation to be performed is specified by @var{code}, a tree code
6895 which will always be one of the following: @code{PLUS_EXPR},
6896 @code{MINUS_EXPR}, @code{MULT_EXPR}, @code{RDIV_EXPR},
6897 @code{MAX_EXPR}, @code{MIN_EXPR}.@refill
6899 @cindex overflow while constant folding
6900 The expansion of this macro is responsible for checking for overflow.
6901 If overflow happens, the macro expansion should execute the statement
6902 @code{return 0;}, which indicates the inability to perform the
6903 arithmetic operation requested.
6905 @findex REAL_VALUE_NEGATE
6906 @item REAL_VALUE_NEGATE (@var{x})
6907 A macro for a C expression which returns the negative of the floating
6908 point value @var{x}. Both @var{x} and the value of the expression
6909 have type @code{REAL_VALUE_TYPE} and are in the target machine's
6910 floating point representation.
6912 There is no way for this macro to report overflow, since overflow
6913 can't happen in the negation operation.
6915 @findex REAL_VALUE_TRUNCATE
6916 @item REAL_VALUE_TRUNCATE (@var{mode}, @var{x})
6917 A macro for a C expression which converts the floating point value
6918 @var{x} to mode @var{mode}.
6920 Both @var{x} and the value of the expression are in the target machine's
6921 floating point representation and have type @code{REAL_VALUE_TYPE}.
6922 However, the value should have an appropriate bit pattern to be output
6923 properly as a floating constant whose precision accords with mode
6926 There is no way for this macro to report overflow.
6928 @findex REAL_VALUE_TO_INT
6929 @item REAL_VALUE_TO_INT (@var{low}, @var{high}, @var{x})
6930 A macro for a C expression which converts a floating point value
6931 @var{x} into a double-precision integer which is then stored into
6932 @var{low} and @var{high}, two variables of type @var{int}.
6934 @item REAL_VALUE_FROM_INT (@var{x}, @var{low}, @var{high}, @var{mode})
6935 @findex REAL_VALUE_FROM_INT
6936 A macro for a C expression which converts a double-precision integer
6937 found in @var{low} and @var{high}, two variables of type @var{int},
6938 into a floating point value which is then stored into @var{x}.
6939 The value is in the target machine's representation for mode @var{mode}
6940 and has the type @code{REAL_VALUE_TYPE}.
6944 @section Miscellaneous Parameters
6945 @cindex parameters, miscellaneous
6947 @c prevent bad page break with this line
6948 Here are several miscellaneous parameters.
6951 @item PREDICATE_CODES
6952 @findex PREDICATE_CODES
6953 Define this if you have defined special-purpose predicates in the file
6954 @file{@var{machine}.c}. This macro is called within an initializer of an
6955 array of structures. The first field in the structure is the name of a
6956 predicate and the second field is an array of rtl codes. For each
6957 predicate, list all rtl codes that can be in expressions matched by the
6958 predicate. The list should have a trailing comma. Here is an example
6959 of two entries in the list for a typical RISC machine:
6962 #define PREDICATE_CODES \
6963 @{"gen_reg_rtx_operand", @{SUBREG, REG@}@}, \
6964 @{"reg_or_short_cint_operand", @{SUBREG, REG, CONST_INT@}@},
6967 Defining this macro does not affect the generated code (however,
6968 incorrect definitions that omit an rtl code that may be matched by the
6969 predicate can cause the compiler to malfunction). Instead, it allows
6970 the table built by @file{genrecog} to be more compact and efficient,
6971 thus speeding up the compiler. The most important predicates to include
6972 in the list specified by this macro are those used in the most insn
6975 @findex CASE_VECTOR_MODE
6976 @item CASE_VECTOR_MODE
6977 An alias for a machine mode name. This is the machine mode that
6978 elements of a jump-table should have.
6980 @findex CASE_VECTOR_SHORTEN_MODE
6981 @item CASE_VECTOR_SHORTEN_MODE (@var{min_offset}, @var{max_offset}, @var{body})
6982 Optional: return the preferred mode for an @code{addr_diff_vec}
6983 when the minimum and maximum offset are known. If you define this,
6984 it enables extra code in branch shortening to deal with @code{addr_diff_vec}.
6985 To make this work, you also have to define INSN_ALIGN and
6986 make the alignment for @code{addr_diff_vec} explicit.
6987 The @var{body} argument is provided so that teh offset_unsigned and scale
6988 flags can be updated.
6990 @findex CASE_VECTOR_PC_RELATIVE
6991 @item CASE_VECTOR_PC_RELATIVE
6992 Define this macro to be a C expression to indicate when jump-tables
6993 should contain relative addresses. If jump-tables never contain
6994 relative addresses, then you need not define this macro.
6996 @findex CASE_DROPS_THROUGH
6997 @item CASE_DROPS_THROUGH
6998 Define this if control falls through a @code{case} insn when the index
6999 value is out of range. This means the specified default-label is
7000 actually ignored by the @code{case} insn proper.
7002 @findex CASE_VALUES_THRESHOLD
7003 @item CASE_VALUES_THRESHOLD
7004 Define this to be the smallest number of different values for which it
7005 is best to use a jump-table instead of a tree of conditional branches.
7006 The default is four for machines with a @code{casesi} instruction and
7007 five otherwise. This is best for most machines.
7009 @findex WORD_REGISTER_OPERATIONS
7010 @item WORD_REGISTER_OPERATIONS
7011 Define this macro if operations between registers with integral mode
7012 smaller than a word are always performed on the entire register.
7013 Most RISC machines have this property and most CISC machines do not.
7015 @findex LOAD_EXTEND_OP
7016 @item LOAD_EXTEND_OP (@var{mode})
7017 Define this macro to be a C expression indicating when insns that read
7018 memory in @var{mode}, an integral mode narrower than a word, set the
7019 bits outside of @var{mode} to be either the sign-extension or the
7020 zero-extension of the data read. Return @code{SIGN_EXTEND} for values
7021 of @var{mode} for which the
7022 insn sign-extends, @code{ZERO_EXTEND} for which it zero-extends, and
7023 @code{NIL} for other modes.
7025 This macro is not called with @var{mode} non-integral or with a width
7026 greater than or equal to @code{BITS_PER_WORD}, so you may return any
7027 value in this case. Do not define this macro if it would always return
7028 @code{NIL}. On machines where this macro is defined, you will normally
7029 define it as the constant @code{SIGN_EXTEND} or @code{ZERO_EXTEND}.
7031 @findex SHORT_IMMEDIATES_SIGN_EXTEND
7032 @item SHORT_IMMEDIATES_SIGN_EXTEND
7033 Define this macro if loading short immediate values into registers sign
7036 @findex IMPLICIT_FIX_EXPR
7037 @item IMPLICIT_FIX_EXPR
7038 An alias for a tree code that should be used by default for conversion
7039 of floating point values to fixed point. Normally,
7040 @code{FIX_ROUND_EXPR} is used.@refill
7042 @findex FIXUNS_TRUNC_LIKE_FIX_TRUNC
7043 @item FIXUNS_TRUNC_LIKE_FIX_TRUNC
7044 Define this macro if the same instructions that convert a floating
7045 point number to a signed fixed point number also convert validly to an
7048 @findex EASY_DIV_EXPR
7050 An alias for a tree code that is the easiest kind of division to
7051 compile code for in the general case. It may be
7052 @code{TRUNC_DIV_EXPR}, @code{FLOOR_DIV_EXPR}, @code{CEIL_DIV_EXPR} or
7053 @code{ROUND_DIV_EXPR}. These four division operators differ in how
7054 they round the result to an integer. @code{EASY_DIV_EXPR} is used
7055 when it is permissible to use any of those kinds of division and the
7056 choice should be made on the basis of efficiency.@refill
7060 The maximum number of bytes that a single instruction can move quickly
7061 between memory and registers or between two memory locations.
7063 @findex MAX_MOVE_MAX
7065 The maximum number of bytes that a single instruction can move quickly
7066 between memory and registers or between two memory locations. If this
7067 is undefined, the default is @code{MOVE_MAX}. Otherwise, it is the
7068 constant value that is the largest value that @code{MOVE_MAX} can have
7071 @findex SHIFT_COUNT_TRUNCATED
7072 @item SHIFT_COUNT_TRUNCATED
7073 A C expression that is nonzero if on this machine the number of bits
7074 actually used for the count of a shift operation is equal to the number
7075 of bits needed to represent the size of the object being shifted. When
7076 this macro is non-zero, the compiler will assume that it is safe to omit
7077 a sign-extend, zero-extend, and certain bitwise `and' instructions that
7078 truncates the count of a shift operation. On machines that have
7079 instructions that act on bitfields at variable positions, which may
7080 include `bit test' instructions, a nonzero @code{SHIFT_COUNT_TRUNCATED}
7081 also enables deletion of truncations of the values that serve as
7082 arguments to bitfield instructions.
7084 If both types of instructions truncate the count (for shifts) and
7085 position (for bitfield operations), or if no variable-position bitfield
7086 instructions exist, you should define this macro.
7088 However, on some machines, such as the 80386 and the 680x0, truncation
7089 only applies to shift operations and not the (real or pretended)
7090 bitfield operations. Define @code{SHIFT_COUNT_TRUNCATED} to be zero on
7091 such machines. Instead, add patterns to the @file{md} file that include
7092 the implied truncation of the shift instructions.
7094 You need not define this macro if it would always have the value of zero.
7096 @findex TRULY_NOOP_TRUNCATION
7097 @item TRULY_NOOP_TRUNCATION (@var{outprec}, @var{inprec})
7098 A C expression which is nonzero if on this machine it is safe to
7099 ``convert'' an integer of @var{inprec} bits to one of @var{outprec}
7100 bits (where @var{outprec} is smaller than @var{inprec}) by merely
7101 operating on it as if it had only @var{outprec} bits.
7103 On many machines, this expression can be 1.
7105 @c rearranged this, removed the phrase "it is reported that". this was
7106 @c to fix an overfull hbox. --mew 10feb93
7107 When @code{TRULY_NOOP_TRUNCATION} returns 1 for a pair of sizes for
7108 modes for which @code{MODES_TIEABLE_P} is 0, suboptimal code can result.
7109 If this is the case, making @code{TRULY_NOOP_TRUNCATION} return 0 in
7110 such cases may improve things.
7112 @findex STORE_FLAG_VALUE
7113 @item STORE_FLAG_VALUE
7114 A C expression describing the value returned by a comparison operator
7115 with an integral mode and stored by a store-flag instruction
7116 (@samp{s@var{cond}}) when the condition is true. This description must
7117 apply to @emph{all} the @samp{s@var{cond}} patterns and all the
7118 comparison operators whose results have a @code{MODE_INT} mode.
7120 A value of 1 or -1 means that the instruction implementing the
7121 comparison operator returns exactly 1 or -1 when the comparison is true
7122 and 0 when the comparison is false. Otherwise, the value indicates
7123 which bits of the result are guaranteed to be 1 when the comparison is
7124 true. This value is interpreted in the mode of the comparison
7125 operation, which is given by the mode of the first operand in the
7126 @samp{s@var{cond}} pattern. Either the low bit or the sign bit of
7127 @code{STORE_FLAG_VALUE} be on. Presently, only those bits are used by
7130 If @code{STORE_FLAG_VALUE} is neither 1 or -1, the compiler will
7131 generate code that depends only on the specified bits. It can also
7132 replace comparison operators with equivalent operations if they cause
7133 the required bits to be set, even if the remaining bits are undefined.
7134 For example, on a machine whose comparison operators return an
7135 @code{SImode} value and where @code{STORE_FLAG_VALUE} is defined as
7136 @samp{0x80000000}, saying that just the sign bit is relevant, the
7140 (ne:SI (and:SI @var{x} (const_int @var{power-of-2})) (const_int 0))
7147 (ashift:SI @var{x} (const_int @var{n}))
7151 where @var{n} is the appropriate shift count to move the bit being
7152 tested into the sign bit.
7154 There is no way to describe a machine that always sets the low-order bit
7155 for a true value, but does not guarantee the value of any other bits,
7156 but we do not know of any machine that has such an instruction. If you
7157 are trying to port GNU CC to such a machine, include an instruction to
7158 perform a logical-and of the result with 1 in the pattern for the
7159 comparison operators and let us know
7161 (@pxref{Bug Reporting,,How to Report Bugs}).
7164 (@pxref{Bug Reporting,,How to Report Bugs,gcc.info,Using GCC}).
7167 Often, a machine will have multiple instructions that obtain a value
7168 from a comparison (or the condition codes). Here are rules to guide the
7169 choice of value for @code{STORE_FLAG_VALUE}, and hence the instructions
7174 Use the shortest sequence that yields a valid definition for
7175 @code{STORE_FLAG_VALUE}. It is more efficient for the compiler to
7176 ``normalize'' the value (convert it to, e.g., 1 or 0) than for the
7177 comparison operators to do so because there may be opportunities to
7178 combine the normalization with other operations.
7181 For equal-length sequences, use a value of 1 or -1, with -1 being
7182 slightly preferred on machines with expensive jumps and 1 preferred on
7186 As a second choice, choose a value of @samp{0x80000001} if instructions
7187 exist that set both the sign and low-order bits but do not define the
7191 Otherwise, use a value of @samp{0x80000000}.
7194 Many machines can produce both the value chosen for
7195 @code{STORE_FLAG_VALUE} and its negation in the same number of
7196 instructions. On those machines, you should also define a pattern for
7197 those cases, e.g., one matching
7200 (set @var{A} (neg:@var{m} (ne:@var{m} @var{B} @var{C})))
7203 Some machines can also perform @code{and} or @code{plus} operations on
7204 condition code values with less instructions than the corresponding
7205 @samp{s@var{cond}} insn followed by @code{and} or @code{plus}. On those
7206 machines, define the appropriate patterns. Use the names @code{incscc}
7207 and @code{decscc}, respectively, for the patterns which perform
7208 @code{plus} or @code{minus} operations on condition code values. See
7209 @file{rs6000.md} for some examples. The GNU Superoptizer can be used to
7210 find such instruction sequences on other machines.
7212 You need not define @code{STORE_FLAG_VALUE} if the machine has no store-flag
7215 @findex FLOAT_STORE_FLAG_VALUE
7216 @item FLOAT_STORE_FLAG_VALUE
7217 A C expression that gives a non-zero floating point value that is
7218 returned when comparison operators with floating-point results are true.
7219 Define this macro on machine that have comparison operations that return
7220 floating-point values. If there are no such operations, do not define
7225 An alias for the machine mode for pointers. On most machines, define
7226 this to be the integer mode corresponding to the width of a hardware
7227 pointer; @code{SImode} on 32-bit machine or @code{DImode} on 64-bit machines.
7228 On some machines you must define this to be one of the partial integer
7229 modes, such as @code{PSImode}.
7231 The width of @code{Pmode} must be at least as large as the value of
7232 @code{POINTER_SIZE}. If it is not equal, you must define the macro
7233 @code{POINTERS_EXTEND_UNSIGNED} to specify how pointers are extended
7236 @findex FUNCTION_MODE
7238 An alias for the machine mode used for memory references to functions
7239 being called, in @code{call} RTL expressions. On most machines this
7240 should be @code{QImode}.
7242 @findex INTEGRATE_THRESHOLD
7243 @item INTEGRATE_THRESHOLD (@var{decl})
7244 A C expression for the maximum number of instructions above which the
7245 function @var{decl} should not be inlined. @var{decl} is a
7246 @code{FUNCTION_DECL} node.
7248 The default definition of this macro is 64 plus 8 times the number of
7249 arguments that the function accepts. Some people think a larger
7250 threshold should be used on RISC machines.
7252 @findex SCCS_DIRECTIVE
7253 @item SCCS_DIRECTIVE
7254 Define this if the preprocessor should ignore @code{#sccs} directives
7255 and print no error message.
7257 @findex NO_IMPLICIT_EXTERN_C
7258 @item NO_IMPLICIT_EXTERN_C
7259 Define this macro if the system header files support C++ as well as C.
7260 This macro inhibits the usual method of using system header files in
7261 C++, which is to pretend that the file's contents are enclosed in
7262 @samp{extern "C" @{@dots{}@}}.
7264 @findex HANDLE_PRAGMA
7267 @item HANDLE_PRAGMA (@var{stream}, @var{node})
7268 Define this macro if you want to implement any pragmas. If defined, it
7269 is a C expression whose value is 1 if the pragma was handled by the function.
7270 The argument @var{stream} is the stdio input stream from which the source text
7271 can be read. @var{node} is the tree node for the identifier after the
7274 It is generally a bad idea to implement new uses of @code{#pragma}. The
7275 only reason to define this macro is for compatibility with other
7276 compilers that do support @code{#pragma} for the sake of any user
7277 programs which already use it.
7279 @findex VALID_MACHINE_DECL_ATTRIBUTE
7280 @item VALID_MACHINE_DECL_ATTRIBUTE (@var{decl}, @var{attributes}, @var{identifier}, @var{args})
7281 If defined, a C expression whose value is nonzero if @var{identifier} with
7282 arguments @var{args} is a valid machine specific attribute for @var{decl}.
7283 The attributes in @var{attributes} have previously been assigned to @var{decl}.
7285 @findex VALID_MACHINE_TYPE_ATTRIBUTE
7286 @item VALID_MACHINE_TYPE_ATTRIBUTE (@var{type}, @var{attributes}, @var{identifier}, @var{args})
7287 If defined, a C expression whose value is nonzero if @var{identifier} with
7288 arguments @var{args} is a valid machine specific attribute for @var{type}.
7289 The attributes in @var{attributes} have previously been assigned to @var{type}.
7291 @findex COMP_TYPE_ATTRIBUTES
7292 @item COMP_TYPE_ATTRIBUTES (@var{type1}, @var{type2})
7293 If defined, a C expression whose value is zero if the attributes on
7294 @var{type1} and @var{type2} are incompatible, one if they are compatible,
7295 and two if they are nearly compatible (which causes a warning to be
7298 @findex SET_DEFAULT_TYPE_ATTRIBUTES
7299 @item SET_DEFAULT_TYPE_ATTRIBUTES (@var{type})
7300 If defined, a C statement that assigns default attributes to
7301 newly defined @var{type}.
7303 @findex MERGE_MACHINE_TYPE_ATTRIBUTES
7304 @item MERGE_MACHINE_TYPE_ATTRIBUTES (@var{type1}, @var{type2})
7305 Define this macro if the merging of type attributes needs special handling.
7306 If defined, the result is a list of the combined TYPE_ATTRIBUTES of
7307 @var{type1} and @var{type2}. It is assumed that comptypes has already been
7308 called and returned 1.
7310 @findex MERGE_MACHINE_DECL_ATTRIBUTES
7311 @item MERGE_MACHINE_DECL_ATTRIBUTES (@var{olddecl}, @var{newdecl})
7312 Define this macro if the merging of decl attributes needs special handling.
7313 If defined, the result is a list of the combined DECL_MACHINE_ATTRIBUTES of
7314 @var{olddecl} and @var{newdecl}. @var{newdecl} is a duplicate declaration
7315 of @var{olddecl}. Examples of when this is needed are when one attribute
7316 overrides another, or when an attribute is nullified by a subsequent
7319 @findex DOLLARS_IN_IDENTIFIERS
7320 @item DOLLARS_IN_IDENTIFIERS
7321 Define this macro to control use of the character @samp{$} in identifier
7322 names. 0 means @samp{$} is not allowed by default; 1 means it is allowed.
7323 1 is the default; there is no need to define this macro in that case.
7324 This macro controls the compiler proper; it does not affect the preprocessor.
7326 @findex NO_DOLLAR_IN_LABEL
7327 @item NO_DOLLAR_IN_LABEL
7328 Define this macro if the assembler does not accept the character
7329 @samp{$} in label names. By default constructors and destructors in
7330 G++ have @samp{$} in the identifiers. If this macro is defined,
7331 @samp{.} is used instead.
7333 @findex NO_DOT_IN_LABEL
7334 @item NO_DOT_IN_LABEL
7335 Define this macro if the assembler does not accept the character
7336 @samp{.} in label names. By default constructors and destructors in G++
7337 have names that use @samp{.}. If this macro is defined, these names
7338 are rewritten to avoid @samp{.}.
7340 @findex DEFAULT_MAIN_RETURN
7341 @item DEFAULT_MAIN_RETURN
7342 Define this macro if the target system expects every program's @code{main}
7343 function to return a standard ``success'' value by default (if no other
7344 value is explicitly returned).
7346 The definition should be a C statement (sans semicolon) to generate the
7347 appropriate rtl instructions. It is used only when compiling the end of
7352 Define this if the target system supports the function
7353 @code{atexit} from the ANSI C standard. If this is not defined,
7354 and @code{INIT_SECTION_ASM_OP} is not defined, a default
7355 @code{exit} function will be provided to support C++.
7359 Define this if your @code{exit} function needs to do something
7360 besides calling an external function @code{_cleanup} before
7361 terminating with @code{_exit}. The @code{EXIT_BODY} macro is
7362 only needed if neither @code{HAVE_ATEXIT} nor
7363 @code{INIT_SECTION_ASM_OP} are defined.
7365 @findex INSN_SETS_ARE_DELAYED
7366 @item INSN_SETS_ARE_DELAYED (@var{insn})
7367 Define this macro as a C expression that is nonzero if it is safe for the
7368 delay slot scheduler to place instructions in the delay slot of @var{insn},
7369 even if they appear to use a resource set or clobbered in @var{insn}.
7370 @var{insn} is always a @code{jump_insn} or an @code{insn}; GNU CC knows that
7371 every @code{call_insn} has this behavior. On machines where some @code{insn}
7372 or @code{jump_insn} is really a function call and hence has this behavior,
7373 you should define this macro.
7375 You need not define this macro if it would always return zero.
7377 @findex INSN_REFERENCES_ARE_DELAYED
7378 @item INSN_REFERENCES_ARE_DELAYED (@var{insn})
7379 Define this macro as a C expression that is nonzero if it is safe for the
7380 delay slot scheduler to place instructions in the delay slot of @var{insn},
7381 even if they appear to set or clobber a resource referenced in @var{insn}.
7382 @var{insn} is always a @code{jump_insn} or an @code{insn}. On machines where
7383 some @code{insn} or @code{jump_insn} is really a function call and its operands
7384 are registers whose use is actually in the subroutine it calls, you should
7385 define this macro. Doing so allows the delay slot scheduler to move
7386 instructions which copy arguments into the argument registers into the delay
7389 You need not define this macro if it would always return zero.
7391 @findex MACHINE_DEPENDENT_REORG
7392 @item MACHINE_DEPENDENT_REORG (@var{insn})
7393 In rare cases, correct code generation requires extra machine
7394 dependent processing between the second jump optimization pass and
7395 delayed branch scheduling. On those machines, define this macro as a C
7396 statement to act on the code starting at @var{insn}.
7398 @findex MULTIPLE_SYMBOL_SPACES
7399 @item MULTIPLE_SYMBOL_SPACES
7400 Define this macro if in some cases global symbols from one translation
7401 unit may not be bound to undefined symbols in another translation unit
7402 without user intervention. For instance, under Microsoft Windows
7403 symbols must be explicitly imported from shared libraries (DLLs).
7405 @findex GIV_SORT_CRITERION
7406 @item GIV_SORT_CRITERION (@var{giv1}, @var{giv2})
7407 In some cases, the strength reduction optimization pass can produce better
7408 code if this is defined. This macro controls the order that induction
7409 variables are combined. This macro is particularly useful if the target has
7410 limited addressing modes. For instance, the SH target has only positive
7411 offsets in addresses. Thus sorting to put the smallest address first
7412 allows the most combinations to be found.
7416 A C expression that returns how many instructions can be issued at the
7417 same time if the machine is a superscalar machine. This is only used by
7418 the @samp{Haifa} scheduler, and not the traditional scheduler.
7420 @findex MAX_INTEGER_COMPUTATION_MODE
7421 @item MAX_INTEGER_COMPUTATION_MODE
7422 Define this to the largest integer machine mode which can be used for
7423 operations other than load, store and copy operations.
7425 You need only define this macro if the target holds values larger than
7426 @code{word_mode} in general purpose registers. Most targets should not define