1 @c Copyright (C) 1988,1989,1992,1993,1994,1995,1996,1997,1998,1999,2000,2001,
2 @c 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010
3 @c Free Software Foundation, Inc.
4 @c This is part of the GCC manual.
5 @c For copying conditions, see the file gcc.texi.
8 @chapter Target Description Macros and Functions
9 @cindex machine description macros
10 @cindex target description macros
11 @cindex macros, target description
12 @cindex @file{tm.h} macros
14 In addition to the file @file{@var{machine}.md}, a machine description
15 includes a C header file conventionally given the name
16 @file{@var{machine}.h} and a C source file named @file{@var{machine}.c}.
17 The header file defines numerous macros that convey the information
18 about the target machine that does not fit into the scheme of the
19 @file{.md} file. The file @file{tm.h} should be a link to
20 @file{@var{machine}.h}. The header file @file{config.h} includes
21 @file{tm.h} and most compiler source files include @file{config.h}. The
22 source file defines a variable @code{targetm}, which is a structure
23 containing pointers to functions and data relating to the target
24 machine. @file{@var{machine}.c} should also contain their definitions,
25 if they are not defined elsewhere in GCC, and other functions called
26 through the macros defined in the @file{.h} file.
29 * Target Structure:: The @code{targetm} variable.
30 * Driver:: Controlling how the driver runs the compilation passes.
31 * Run-time Target:: Defining @samp{-m} options like @option{-m68000} and @option{-m68020}.
32 * Per-Function Data:: Defining data structures for per-function information.
33 * Storage Layout:: Defining sizes and alignments of data.
34 * Type Layout:: Defining sizes and properties of basic user data types.
35 * Registers:: Naming and describing the hardware registers.
36 * Register Classes:: Defining the classes of hardware registers.
37 * Old Constraints:: The old way to define machine-specific constraints.
38 * Stack and Calling:: Defining which way the stack grows and by how much.
39 * Varargs:: Defining the varargs macros.
40 * Trampolines:: Code set up at run time to enter a nested function.
41 * Library Calls:: Controlling how library routines are implicitly called.
42 * Addressing Modes:: Defining addressing modes valid for memory operands.
43 * Anchored Addresses:: Defining how @option{-fsection-anchors} should work.
44 * Condition Code:: Defining how insns update the condition code.
45 * Costs:: Defining relative costs of different operations.
46 * Scheduling:: Adjusting the behavior of the instruction scheduler.
47 * Sections:: Dividing storage into text, data, and other sections.
48 * PIC:: Macros for position independent code.
49 * Assembler Format:: Defining how to write insns and pseudo-ops to output.
50 * Debugging Info:: Defining the format of debugging output.
51 * Floating Point:: Handling floating point for cross-compilers.
52 * Mode Switching:: Insertion of mode-switching instructions.
53 * Target Attributes:: Defining target-specific uses of @code{__attribute__}.
54 * Emulated TLS:: Emulated TLS support.
55 * MIPS Coprocessors:: MIPS coprocessor support and how to customize it.
56 * PCH Target:: Validity checking for precompiled headers.
57 * C++ ABI:: Controlling C++ ABI changes.
58 * Named Address Spaces:: Adding support for named address spaces
59 * Misc:: Everything else.
62 @node Target Structure
63 @section The Global @code{targetm} Variable
65 @cindex target functions
67 @deftypevar {struct gcc_target} targetm
68 The target @file{.c} file must define the global @code{targetm} variable
69 which contains pointers to functions and data relating to the target
70 machine. The variable is declared in @file{target.h};
71 @file{target-def.h} defines the macro @code{TARGET_INITIALIZER} which is
72 used to initialize the variable, and macros for the default initializers
73 for elements of the structure. The @file{.c} file should override those
74 macros for which the default definition is inappropriate. For example:
77 #include "target-def.h"
79 /* @r{Initialize the GCC target structure.} */
81 #undef TARGET_COMP_TYPE_ATTRIBUTES
82 #define TARGET_COMP_TYPE_ATTRIBUTES @var{machine}_comp_type_attributes
84 struct gcc_target targetm = TARGET_INITIALIZER;
88 Where a macro should be defined in the @file{.c} file in this manner to
89 form part of the @code{targetm} structure, it is documented below as a
90 ``Target Hook'' with a prototype. Many macros will change in future
91 from being defined in the @file{.h} file to being part of the
92 @code{targetm} structure.
95 @section Controlling the Compilation Driver, @file{gcc}
97 @cindex controlling the compilation driver
99 @c prevent bad page break with this line
100 You can control the compilation driver.
102 @defmac DRIVER_SELF_SPECS
103 A list of specs for the driver itself. It should be a suitable
104 initializer for an array of strings, with no surrounding braces.
106 The driver applies these specs to its own command line between loading
107 default @file{specs} files (but not command-line specified ones) and
108 choosing the multilib directory or running any subcommands. It
109 applies them in the order given, so each spec can depend on the
110 options added by earlier ones. It is also possible to remove options
111 using @samp{%<@var{option}} in the usual way.
113 This macro can be useful when a port has several interdependent target
114 options. It provides a way of standardizing the command line so
115 that the other specs are easier to write.
117 Do not define this macro if it does not need to do anything.
120 @defmac OPTION_DEFAULT_SPECS
121 A list of specs used to support configure-time default options (i.e.@:
122 @option{--with} options) in the driver. It should be a suitable initializer
123 for an array of structures, each containing two strings, without the
124 outermost pair of surrounding braces.
126 The first item in the pair is the name of the default. This must match
127 the code in @file{config.gcc} for the target. The second item is a spec
128 to apply if a default with this name was specified. The string
129 @samp{%(VALUE)} in the spec will be replaced by the value of the default
130 everywhere it occurs.
132 The driver will apply these specs to its own command line between loading
133 default @file{specs} files and processing @code{DRIVER_SELF_SPECS}, using
134 the same mechanism as @code{DRIVER_SELF_SPECS}.
136 Do not define this macro if it does not need to do anything.
140 A C string constant that tells the GCC driver program options to
141 pass to CPP@. It can also specify how to translate options you
142 give to GCC into options for GCC to pass to the CPP@.
144 Do not define this macro if it does not need to do anything.
147 @defmac CPLUSPLUS_CPP_SPEC
148 This macro is just like @code{CPP_SPEC}, but is used for C++, rather
149 than C@. If you do not define this macro, then the value of
150 @code{CPP_SPEC} (if any) will be used instead.
154 A C string constant that tells the GCC driver program options to
155 pass to @code{cc1}, @code{cc1plus}, @code{f771}, and the other language
157 It can also specify how to translate options you give to GCC into options
158 for GCC to pass to front ends.
160 Do not define this macro if it does not need to do anything.
164 A C string constant that tells the GCC driver program options to
165 pass to @code{cc1plus}. It can also specify how to translate options you
166 give to GCC into options for GCC to pass to the @code{cc1plus}.
168 Do not define this macro if it does not need to do anything.
169 Note that everything defined in CC1_SPEC is already passed to
170 @code{cc1plus} so there is no need to duplicate the contents of
171 CC1_SPEC in CC1PLUS_SPEC@.
175 A C string constant that tells the GCC driver program options to
176 pass to the assembler. It can also specify how to translate options
177 you give to GCC into options for GCC to pass to the assembler.
178 See the file @file{sun3.h} for an example of this.
180 Do not define this macro if it does not need to do anything.
183 @defmac ASM_FINAL_SPEC
184 A C string constant that tells the GCC driver program how to
185 run any programs which cleanup after the normal assembler.
186 Normally, this is not needed. See the file @file{mips.h} for
189 Do not define this macro if it does not need to do anything.
192 @defmac AS_NEEDS_DASH_FOR_PIPED_INPUT
193 Define this macro, with no value, if the driver should give the assembler
194 an argument consisting of a single dash, @option{-}, to instruct it to
195 read from its standard input (which will be a pipe connected to the
196 output of the compiler proper). This argument is given after any
197 @option{-o} option specifying the name of the output file.
199 If you do not define this macro, the assembler is assumed to read its
200 standard input if given no non-option arguments. If your assembler
201 cannot read standard input at all, use a @samp{%@{pipe:%e@}} construct;
202 see @file{mips.h} for instance.
206 A C string constant that tells the GCC driver program options to
207 pass to the linker. It can also specify how to translate options you
208 give to GCC into options for GCC to pass to the linker.
210 Do not define this macro if it does not need to do anything.
214 Another C string constant used much like @code{LINK_SPEC}. The difference
215 between the two is that @code{LIB_SPEC} is used at the end of the
216 command given to the linker.
218 If this macro is not defined, a default is provided that
219 loads the standard C library from the usual place. See @file{gcc.c}.
223 Another C string constant that tells the GCC driver program
224 how and when to place a reference to @file{libgcc.a} into the
225 linker command line. This constant is placed both before and after
226 the value of @code{LIB_SPEC}.
228 If this macro is not defined, the GCC driver provides a default that
229 passes the string @option{-lgcc} to the linker.
232 @defmac REAL_LIBGCC_SPEC
233 By default, if @code{ENABLE_SHARED_LIBGCC} is defined, the
234 @code{LIBGCC_SPEC} is not directly used by the driver program but is
235 instead modified to refer to different versions of @file{libgcc.a}
236 depending on the values of the command line flags @option{-static},
237 @option{-shared}, @option{-static-libgcc}, and @option{-shared-libgcc}. On
238 targets where these modifications are inappropriate, define
239 @code{REAL_LIBGCC_SPEC} instead. @code{REAL_LIBGCC_SPEC} tells the
240 driver how to place a reference to @file{libgcc} on the link command
241 line, but, unlike @code{LIBGCC_SPEC}, it is used unmodified.
244 @defmac USE_LD_AS_NEEDED
245 A macro that controls the modifications to @code{LIBGCC_SPEC}
246 mentioned in @code{REAL_LIBGCC_SPEC}. If nonzero, a spec will be
247 generated that uses --as-needed and the shared libgcc in place of the
248 static exception handler library, when linking without any of
249 @code{-static}, @code{-static-libgcc}, or @code{-shared-libgcc}.
253 If defined, this C string constant is added to @code{LINK_SPEC}.
254 When @code{USE_LD_AS_NEEDED} is zero or undefined, it also affects
255 the modifications to @code{LIBGCC_SPEC} mentioned in
256 @code{REAL_LIBGCC_SPEC}.
259 @defmac STARTFILE_SPEC
260 Another C string constant used much like @code{LINK_SPEC}. The
261 difference between the two is that @code{STARTFILE_SPEC} is used at
262 the very beginning of the command given to the linker.
264 If this macro is not defined, a default is provided that loads the
265 standard C startup file from the usual place. See @file{gcc.c}.
269 Another C string constant used much like @code{LINK_SPEC}. The
270 difference between the two is that @code{ENDFILE_SPEC} is used at
271 the very end of the command given to the linker.
273 Do not define this macro if it does not need to do anything.
276 @defmac THREAD_MODEL_SPEC
277 GCC @code{-v} will print the thread model GCC was configured to use.
278 However, this doesn't work on platforms that are multilibbed on thread
279 models, such as AIX 4.3. On such platforms, define
280 @code{THREAD_MODEL_SPEC} such that it evaluates to a string without
281 blanks that names one of the recognized thread models. @code{%*}, the
282 default value of this macro, will expand to the value of
283 @code{thread_file} set in @file{config.gcc}.
286 @defmac SYSROOT_SUFFIX_SPEC
287 Define this macro to add a suffix to the target sysroot when GCC is
288 configured with a sysroot. This will cause GCC to search for usr/lib,
289 et al, within sysroot+suffix.
292 @defmac SYSROOT_HEADERS_SUFFIX_SPEC
293 Define this macro to add a headers_suffix to the target sysroot when
294 GCC is configured with a sysroot. This will cause GCC to pass the
295 updated sysroot+headers_suffix to CPP, causing it to search for
296 usr/include, et al, within sysroot+headers_suffix.
300 Define this macro to provide additional specifications to put in the
301 @file{specs} file that can be used in various specifications like
304 The definition should be an initializer for an array of structures,
305 containing a string constant, that defines the specification name, and a
306 string constant that provides the specification.
308 Do not define this macro if it does not need to do anything.
310 @code{EXTRA_SPECS} is useful when an architecture contains several
311 related targets, which have various @code{@dots{}_SPECS} which are similar
312 to each other, and the maintainer would like one central place to keep
315 For example, the PowerPC System V.4 targets use @code{EXTRA_SPECS} to
316 define either @code{_CALL_SYSV} when the System V calling sequence is
317 used or @code{_CALL_AIX} when the older AIX-based calling sequence is
320 The @file{config/rs6000/rs6000.h} target file defines:
323 #define EXTRA_SPECS \
324 @{ "cpp_sysv_default", CPP_SYSV_DEFAULT @},
326 #define CPP_SYS_DEFAULT ""
329 The @file{config/rs6000/sysv.h} target file defines:
333 "%@{posix: -D_POSIX_SOURCE @} \
334 %@{mcall-sysv: -D_CALL_SYSV @} \
335 %@{!mcall-sysv: %(cpp_sysv_default) @} \
336 %@{msoft-float: -D_SOFT_FLOAT@} %@{mcpu=403: -D_SOFT_FLOAT@}"
338 #undef CPP_SYSV_DEFAULT
339 #define CPP_SYSV_DEFAULT "-D_CALL_SYSV"
342 while the @file{config/rs6000/eabiaix.h} target file defines
343 @code{CPP_SYSV_DEFAULT} as:
346 #undef CPP_SYSV_DEFAULT
347 #define CPP_SYSV_DEFAULT "-D_CALL_AIX"
351 @defmac LINK_LIBGCC_SPECIAL_1
352 Define this macro if the driver program should find the library
353 @file{libgcc.a}. If you do not define this macro, the driver program will pass
354 the argument @option{-lgcc} to tell the linker to do the search.
357 @defmac LINK_GCC_C_SEQUENCE_SPEC
358 The sequence in which libgcc and libc are specified to the linker.
359 By default this is @code{%G %L %G}.
362 @defmac LINK_COMMAND_SPEC
363 A C string constant giving the complete command line need to execute the
364 linker. When you do this, you will need to update your port each time a
365 change is made to the link command line within @file{gcc.c}. Therefore,
366 define this macro only if you need to completely redefine the command
367 line for invoking the linker and there is no other way to accomplish
368 the effect you need. Overriding this macro may be avoidable by overriding
369 @code{LINK_GCC_C_SEQUENCE_SPEC} instead.
372 @defmac LINK_ELIMINATE_DUPLICATE_LDIRECTORIES
373 A nonzero value causes @command{collect2} to remove duplicate @option{-L@var{directory}} search
374 directories from linking commands. Do not give it a nonzero value if
375 removing duplicate search directories changes the linker's semantics.
378 @defmac MULTILIB_DEFAULTS
379 Define this macro as a C expression for the initializer of an array of
380 string to tell the driver program which options are defaults for this
381 target and thus do not need to be handled specially when using
382 @code{MULTILIB_OPTIONS}.
384 Do not define this macro if @code{MULTILIB_OPTIONS} is not defined in
385 the target makefile fragment or if none of the options listed in
386 @code{MULTILIB_OPTIONS} are set by default.
387 @xref{Target Fragment}.
390 @defmac RELATIVE_PREFIX_NOT_LINKDIR
391 Define this macro to tell @command{gcc} that it should only translate
392 a @option{-B} prefix into a @option{-L} linker option if the prefix
393 indicates an absolute file name.
396 @defmac MD_EXEC_PREFIX
397 If defined, this macro is an additional prefix to try after
398 @code{STANDARD_EXEC_PREFIX}. @code{MD_EXEC_PREFIX} is not searched
399 when the compiler is built as a cross
400 compiler. If you define @code{MD_EXEC_PREFIX}, then be sure to add it
401 to the list of directories used to find the assembler in @file{configure.in}.
404 @defmac STANDARD_STARTFILE_PREFIX
405 Define this macro as a C string constant if you wish to override the
406 standard choice of @code{libdir} as the default prefix to
407 try when searching for startup files such as @file{crt0.o}.
408 @code{STANDARD_STARTFILE_PREFIX} is not searched when the compiler
409 is built as a cross compiler.
412 @defmac STANDARD_STARTFILE_PREFIX_1
413 Define this macro as a C string constant if you wish to override the
414 standard choice of @code{/lib} as a prefix to try after the default prefix
415 when searching for startup files such as @file{crt0.o}.
416 @code{STANDARD_STARTFILE_PREFIX_1} is not searched when the compiler
417 is built as a cross compiler.
420 @defmac STANDARD_STARTFILE_PREFIX_2
421 Define this macro as a C string constant if you wish to override the
422 standard choice of @code{/lib} as yet another prefix to try after the
423 default prefix when searching for startup files such as @file{crt0.o}.
424 @code{STANDARD_STARTFILE_PREFIX_2} is not searched when the compiler
425 is built as a cross compiler.
428 @defmac MD_STARTFILE_PREFIX
429 If defined, this macro supplies an additional prefix to try after the
430 standard prefixes. @code{MD_EXEC_PREFIX} is not searched when the
431 compiler is built as a cross compiler.
434 @defmac MD_STARTFILE_PREFIX_1
435 If defined, this macro supplies yet another prefix to try after the
436 standard prefixes. It is not searched when the compiler is built as a
440 @defmac INIT_ENVIRONMENT
441 Define this macro as a C string constant if you wish to set environment
442 variables for programs called by the driver, such as the assembler and
443 loader. The driver passes the value of this macro to @code{putenv} to
444 initialize the necessary environment variables.
447 @defmac LOCAL_INCLUDE_DIR
448 Define this macro as a C string constant if you wish to override the
449 standard choice of @file{/usr/local/include} as the default prefix to
450 try when searching for local header files. @code{LOCAL_INCLUDE_DIR}
451 comes before @code{SYSTEM_INCLUDE_DIR} in the search order.
453 Cross compilers do not search either @file{/usr/local/include} or its
457 @defmac SYSTEM_INCLUDE_DIR
458 Define this macro as a C string constant if you wish to specify a
459 system-specific directory to search for header files before the standard
460 directory. @code{SYSTEM_INCLUDE_DIR} comes before
461 @code{STANDARD_INCLUDE_DIR} in the search order.
463 Cross compilers do not use this macro and do not search the directory
467 @defmac STANDARD_INCLUDE_DIR
468 Define this macro as a C string constant if you wish to override the
469 standard choice of @file{/usr/include} as the default prefix to
470 try when searching for header files.
472 Cross compilers ignore this macro and do not search either
473 @file{/usr/include} or its replacement.
476 @defmac STANDARD_INCLUDE_COMPONENT
477 The ``component'' corresponding to @code{STANDARD_INCLUDE_DIR}.
478 See @code{INCLUDE_DEFAULTS}, below, for the description of components.
479 If you do not define this macro, no component is used.
482 @defmac INCLUDE_DEFAULTS
483 Define this macro if you wish to override the entire default search path
484 for include files. For a native compiler, the default search path
485 usually consists of @code{GCC_INCLUDE_DIR}, @code{LOCAL_INCLUDE_DIR},
486 @code{SYSTEM_INCLUDE_DIR}, @code{GPLUSPLUS_INCLUDE_DIR}, and
487 @code{STANDARD_INCLUDE_DIR}. In addition, @code{GPLUSPLUS_INCLUDE_DIR}
488 and @code{GCC_INCLUDE_DIR} are defined automatically by @file{Makefile},
489 and specify private search areas for GCC@. The directory
490 @code{GPLUSPLUS_INCLUDE_DIR} is used only for C++ programs.
492 The definition should be an initializer for an array of structures.
493 Each array element should have four elements: the directory name (a
494 string constant), the component name (also a string constant), a flag
495 for C++-only directories,
496 and a flag showing that the includes in the directory don't need to be
497 wrapped in @code{extern @samp{C}} when compiling C++. Mark the end of
498 the array with a null element.
500 The component name denotes what GNU package the include file is part of,
501 if any, in all uppercase letters. For example, it might be @samp{GCC}
502 or @samp{BINUTILS}. If the package is part of a vendor-supplied
503 operating system, code the component name as @samp{0}.
505 For example, here is the definition used for VAX/VMS:
508 #define INCLUDE_DEFAULTS \
510 @{ "GNU_GXX_INCLUDE:", "G++", 1, 1@}, \
511 @{ "GNU_CC_INCLUDE:", "GCC", 0, 0@}, \
512 @{ "SYS$SYSROOT:[SYSLIB.]", 0, 0, 0@}, \
519 Here is the order of prefixes tried for exec files:
523 Any prefixes specified by the user with @option{-B}.
526 The environment variable @code{GCC_EXEC_PREFIX} or, if @code{GCC_EXEC_PREFIX}
527 is not set and the compiler has not been installed in the configure-time
528 @var{prefix}, the location in which the compiler has actually been installed.
531 The directories specified by the environment variable @code{COMPILER_PATH}.
534 The macro @code{STANDARD_EXEC_PREFIX}, if the compiler has been installed
535 in the configured-time @var{prefix}.
538 The location @file{/usr/libexec/gcc/}, but only if this is a native compiler.
541 The location @file{/usr/lib/gcc/}, but only if this is a native compiler.
544 The macro @code{MD_EXEC_PREFIX}, if defined, but only if this is a native
548 Here is the order of prefixes tried for startfiles:
552 Any prefixes specified by the user with @option{-B}.
555 The environment variable @code{GCC_EXEC_PREFIX} or its automatically determined
556 value based on the installed toolchain location.
559 The directories specified by the environment variable @code{LIBRARY_PATH}
560 (or port-specific name; native only, cross compilers do not use this).
563 The macro @code{STANDARD_EXEC_PREFIX}, but only if the toolchain is installed
564 in the configured @var{prefix} or this is a native compiler.
567 The location @file{/usr/lib/gcc/}, but only if this is a native compiler.
570 The macro @code{MD_EXEC_PREFIX}, if defined, but only if this is a native
574 The macro @code{MD_STARTFILE_PREFIX}, if defined, but only if this is a
575 native compiler, or we have a target system root.
578 The macro @code{MD_STARTFILE_PREFIX_1}, if defined, but only if this is a
579 native compiler, or we have a target system root.
582 The macro @code{STANDARD_STARTFILE_PREFIX}, with any sysroot modifications.
583 If this path is relative it will be prefixed by @code{GCC_EXEC_PREFIX} and
584 the machine suffix or @code{STANDARD_EXEC_PREFIX} and the machine suffix.
587 The macro @code{STANDARD_STARTFILE_PREFIX_1}, but only if this is a native
588 compiler, or we have a target system root. The default for this macro is
592 The macro @code{STANDARD_STARTFILE_PREFIX_2}, but only if this is a native
593 compiler, or we have a target system root. The default for this macro is
597 @node Run-time Target
598 @section Run-time Target Specification
599 @cindex run-time target specification
600 @cindex predefined macros
601 @cindex target specifications
603 @c prevent bad page break with this line
604 Here are run-time target specifications.
606 @defmac TARGET_CPU_CPP_BUILTINS ()
607 This function-like macro expands to a block of code that defines
608 built-in preprocessor macros and assertions for the target CPU, using
609 the functions @code{builtin_define}, @code{builtin_define_std} and
610 @code{builtin_assert}. When the front end
611 calls this macro it provides a trailing semicolon, and since it has
612 finished command line option processing your code can use those
615 @code{builtin_assert} takes a string in the form you pass to the
616 command-line option @option{-A}, such as @code{cpu=mips}, and creates
617 the assertion. @code{builtin_define} takes a string in the form
618 accepted by option @option{-D} and unconditionally defines the macro.
620 @code{builtin_define_std} takes a string representing the name of an
621 object-like macro. If it doesn't lie in the user's namespace,
622 @code{builtin_define_std} defines it unconditionally. Otherwise, it
623 defines a version with two leading underscores, and another version
624 with two leading and trailing underscores, and defines the original
625 only if an ISO standard was not requested on the command line. For
626 example, passing @code{unix} defines @code{__unix}, @code{__unix__}
627 and possibly @code{unix}; passing @code{_mips} defines @code{__mips},
628 @code{__mips__} and possibly @code{_mips}, and passing @code{_ABI64}
629 defines only @code{_ABI64}.
631 You can also test for the C dialect being compiled. The variable
632 @code{c_language} is set to one of @code{clk_c}, @code{clk_cplusplus}
633 or @code{clk_objective_c}. Note that if we are preprocessing
634 assembler, this variable will be @code{clk_c} but the function-like
635 macro @code{preprocessing_asm_p()} will return true, so you might want
636 to check for that first. If you need to check for strict ANSI, the
637 variable @code{flag_iso} can be used. The function-like macro
638 @code{preprocessing_trad_p()} can be used to check for traditional
642 @defmac TARGET_OS_CPP_BUILTINS ()
643 Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional
644 and is used for the target operating system instead.
647 @defmac TARGET_OBJFMT_CPP_BUILTINS ()
648 Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional
649 and is used for the target object format. @file{elfos.h} uses this
650 macro to define @code{__ELF__}, so you probably do not need to define
654 @deftypevar {extern int} target_flags
655 This variable is declared in @file{options.h}, which is included before
656 any target-specific headers.
659 @hook TARGET_DEFAULT_TARGET_FLAGS
660 This variable specifies the initial value of @code{target_flags}.
661 Its default setting is 0.
664 @cindex optional hardware or system features
665 @cindex features, optional, in system conventions
667 @hook TARGET_HANDLE_OPTION
668 This hook is called whenever the user specifies one of the
669 target-specific options described by the @file{.opt} definition files
670 (@pxref{Options}). It has the opportunity to do some option-specific
671 processing and should return true if the option is valid. The default
672 definition does nothing but return true.
674 @var{decoded} specifies the option and its arguments. @var{opts} and
675 @var{opts_set} are the @code{gcc_options} structures to be used for
676 storing option state, and @var{loc} is the location at which the
677 option was passed (@code{UNKNOWN_LOCATION} except for options passed
681 @hook TARGET_HANDLE_C_OPTION
682 This target hook is called whenever the user specifies one of the
683 target-specific C language family options described by the @file{.opt}
684 definition files(@pxref{Options}). It has the opportunity to do some
685 option-specific processing and should return true if the option is
686 valid. The arguments are like for @code{TARGET_HANDLE_OPTION}. The
687 default definition does nothing but return false.
689 In general, you should use @code{TARGET_HANDLE_OPTION} to handle
690 options. However, if processing an option requires routines that are
691 only available in the C (and related language) front ends, then you
692 should use @code{TARGET_HANDLE_C_OPTION} instead.
695 @hook TARGET_OBJC_CONSTRUCT_STRING_OBJECT
697 @hook TARGET_STRING_OBJECT_REF_TYPE_P
699 @hook TARGET_CHECK_STRING_OBJECT_FORMAT_ARG
701 @hook TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE
702 This target function is similar to the hook @code{TARGET_OPTION_OVERRIDE}
703 but is called when the optimize level is changed via an attribute or
704 pragma or when it is reset at the end of the code affected by the
705 attribute or pragma. It is not called at the beginning of compilation
706 when @code{TARGET_OPTION_OVERRIDE} is called so if you want to perform these
707 actions then, you should have @code{TARGET_OPTION_OVERRIDE} call
708 @code{TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE}.
711 @defmac C_COMMON_OVERRIDE_OPTIONS
712 This is similar to the @code{TARGET_OPTION_OVERRIDE} hook
713 but is only used in the C
714 language frontends (C, Objective-C, C++, Objective-C++) and so can be
715 used to alter option flag variables which only exist in those
719 @hook TARGET_OPTION_OPTIMIZATION_TABLE
720 Some machines may desire to change what optimizations are performed for
721 various optimization levels. This variable, if defined, describes
722 options to enable at particular sets of optimization levels. These
723 options are processed once
724 just after the optimization level is determined and before the remainder
725 of the command options have been parsed, so may be overridden by other
726 options passed explicitly.
728 This processing is run once at program startup and when the optimization
729 options are changed via @code{#pragma GCC optimize} or by using the
730 @code{optimize} attribute.
733 @hook TARGET_OPTION_INIT_STRUCT
735 @hook TARGET_OPTION_DEFAULT_PARAMS
738 This hook is called in response to the user invoking
739 @option{--target-help} on the command line. It gives the target a
740 chance to display extra information on the target specific command
741 line options found in its @file{.opt} file.
744 @defmac SWITCHABLE_TARGET
745 Some targets need to switch between substantially different subtargets
746 during compilation. For example, the MIPS target has one subtarget for
747 the traditional MIPS architecture and another for MIPS16. Source code
748 can switch between these two subarchitectures using the @code{mips16}
749 and @code{nomips16} attributes.
751 Such subtargets can differ in things like the set of available
752 registers, the set of available instructions, the costs of various
753 operations, and so on. GCC caches a lot of this type of information
754 in global variables, and recomputing them for each subtarget takes a
755 significant amount of time. The compiler therefore provides a facility
756 for maintaining several versions of the global variables and quickly
757 switching between them; see @file{target-globals.h} for details.
759 Define this macro to 1 if your target needs this facility. The default
763 @node Per-Function Data
764 @section Defining data structures for per-function information.
765 @cindex per-function data
766 @cindex data structures
768 If the target needs to store information on a per-function basis, GCC
769 provides a macro and a couple of variables to allow this. Note, just
770 using statics to store the information is a bad idea, since GCC supports
771 nested functions, so you can be halfway through encoding one function
772 when another one comes along.
774 GCC defines a data structure called @code{struct function} which
775 contains all of the data specific to an individual function. This
776 structure contains a field called @code{machine} whose type is
777 @code{struct machine_function *}, which can be used by targets to point
778 to their own specific data.
780 If a target needs per-function specific data it should define the type
781 @code{struct machine_function} and also the macro @code{INIT_EXPANDERS}.
782 This macro should be used to initialize the function pointer
783 @code{init_machine_status}. This pointer is explained below.
785 One typical use of per-function, target specific data is to create an
786 RTX to hold the register containing the function's return address. This
787 RTX can then be used to implement the @code{__builtin_return_address}
788 function, for level 0.
790 Note---earlier implementations of GCC used a single data area to hold
791 all of the per-function information. Thus when processing of a nested
792 function began the old per-function data had to be pushed onto a
793 stack, and when the processing was finished, it had to be popped off the
794 stack. GCC used to provide function pointers called
795 @code{save_machine_status} and @code{restore_machine_status} to handle
796 the saving and restoring of the target specific information. Since the
797 single data area approach is no longer used, these pointers are no
800 @defmac INIT_EXPANDERS
801 Macro called to initialize any target specific information. This macro
802 is called once per function, before generation of any RTL has begun.
803 The intention of this macro is to allow the initialization of the
804 function pointer @code{init_machine_status}.
807 @deftypevar {void (*)(struct function *)} init_machine_status
808 If this function pointer is non-@code{NULL} it will be called once per
809 function, before function compilation starts, in order to allow the
810 target to perform any target specific initialization of the
811 @code{struct function} structure. It is intended that this would be
812 used to initialize the @code{machine} of that structure.
814 @code{struct machine_function} structures are expected to be freed by GC@.
815 Generally, any memory that they reference must be allocated by using
816 GC allocation, including the structure itself.
820 @section Storage Layout
821 @cindex storage layout
823 Note that the definitions of the macros in this table which are sizes or
824 alignments measured in bits do not need to be constant. They can be C
825 expressions that refer to static variables, such as the @code{target_flags}.
826 @xref{Run-time Target}.
828 @defmac BITS_BIG_ENDIAN
829 Define this macro to have the value 1 if the most significant bit in a
830 byte has the lowest number; otherwise define it to have the value zero.
831 This means that bit-field instructions count from the most significant
832 bit. If the machine has no bit-field instructions, then this must still
833 be defined, but it doesn't matter which value it is defined to. This
834 macro need not be a constant.
836 This macro does not affect the way structure fields are packed into
837 bytes or words; that is controlled by @code{BYTES_BIG_ENDIAN}.
840 @defmac BYTES_BIG_ENDIAN
841 Define this macro to have the value 1 if the most significant byte in a
842 word has the lowest number. This macro need not be a constant.
845 @defmac WORDS_BIG_ENDIAN
846 Define this macro to have the value 1 if, in a multiword object, the
847 most significant word has the lowest number. This applies to both
848 memory locations and registers; GCC fundamentally assumes that the
849 order of words in memory is the same as the order in registers. This
850 macro need not be a constant.
853 @defmac FLOAT_WORDS_BIG_ENDIAN
854 Define this macro to have the value 1 if @code{DFmode}, @code{XFmode} or
855 @code{TFmode} floating point numbers are stored in memory with the word
856 containing the sign bit at the lowest address; otherwise define it to
857 have the value 0. This macro need not be a constant.
859 You need not define this macro if the ordering is the same as for
863 @defmac BITS_PER_UNIT
864 Define this macro to be the number of bits in an addressable storage
865 unit (byte). If you do not define this macro the default is 8.
868 @defmac BITS_PER_WORD
869 Number of bits in a word. If you do not define this macro, the default
870 is @code{BITS_PER_UNIT * UNITS_PER_WORD}.
873 @defmac MAX_BITS_PER_WORD
874 Maximum number of bits in a word. If this is undefined, the default is
875 @code{BITS_PER_WORD}. Otherwise, it is the constant value that is the
876 largest value that @code{BITS_PER_WORD} can have at run-time.
879 @defmac UNITS_PER_WORD
880 Number of storage units in a word; normally the size of a general-purpose
881 register, a power of two from 1 or 8.
884 @defmac MIN_UNITS_PER_WORD
885 Minimum number of units in a word. If this is undefined, the default is
886 @code{UNITS_PER_WORD}. Otherwise, it is the constant value that is the
887 smallest value that @code{UNITS_PER_WORD} can have at run-time.
891 Width of a pointer, in bits. You must specify a value no wider than the
892 width of @code{Pmode}. If it is not equal to the width of @code{Pmode},
893 you must define @code{POINTERS_EXTEND_UNSIGNED}. If you do not specify
894 a value the default is @code{BITS_PER_WORD}.
897 @defmac POINTERS_EXTEND_UNSIGNED
898 A C expression that determines how pointers should be extended from
899 @code{ptr_mode} to either @code{Pmode} or @code{word_mode}. It is
900 greater than zero if pointers should be zero-extended, zero if they
901 should be sign-extended, and negative if some other sort of conversion
902 is needed. In the last case, the extension is done by the target's
903 @code{ptr_extend} instruction.
905 You need not define this macro if the @code{ptr_mode}, @code{Pmode}
906 and @code{word_mode} are all the same width.
909 @defmac PROMOTE_MODE (@var{m}, @var{unsignedp}, @var{type})
910 A macro to update @var{m} and @var{unsignedp} when an object whose type
911 is @var{type} and which has the specified mode and signedness is to be
912 stored in a register. This macro is only called when @var{type} is a
915 On most RISC machines, which only have operations that operate on a full
916 register, define this macro to set @var{m} to @code{word_mode} if
917 @var{m} is an integer mode narrower than @code{BITS_PER_WORD}. In most
918 cases, only integer modes should be widened because wider-precision
919 floating-point operations are usually more expensive than their narrower
922 For most machines, the macro definition does not change @var{unsignedp}.
923 However, some machines, have instructions that preferentially handle
924 either signed or unsigned quantities of certain modes. For example, on
925 the DEC Alpha, 32-bit loads from memory and 32-bit add instructions
926 sign-extend the result to 64 bits. On such machines, set
927 @var{unsignedp} according to which kind of extension is more efficient.
929 Do not define this macro if it would never modify @var{m}.
932 @hook TARGET_PROMOTE_FUNCTION_MODE
933 Like @code{PROMOTE_MODE}, but it is applied to outgoing function arguments or
934 function return values. The target hook should return the new mode
935 and possibly change @code{*@var{punsignedp}} if the promotion should
936 change signedness. This function is called only for scalar @emph{or
939 @var{for_return} allows to distinguish the promotion of arguments and
940 return values. If it is @code{1}, a return value is being promoted and
941 @code{TARGET_FUNCTION_VALUE} must perform the same promotions done here.
942 If it is @code{2}, the returned mode should be that of the register in
943 which an incoming parameter is copied, or the outgoing result is computed;
944 then the hook should return the same mode as @code{promote_mode}, though
945 the signedness may be different.
947 The default is to not promote arguments and return values. You can
948 also define the hook to @code{default_promote_function_mode_always_promote}
949 if you would like to apply the same rules given by @code{PROMOTE_MODE}.
952 @defmac PARM_BOUNDARY
953 Normal alignment required for function parameters on the stack, in
954 bits. All stack parameters receive at least this much alignment
955 regardless of data type. On most machines, this is the same as the
959 @defmac STACK_BOUNDARY
960 Define this macro to the minimum alignment enforced by hardware for the
961 stack pointer on this machine. The definition is a C expression for the
962 desired alignment (measured in bits). This value is used as a default
963 if @code{PREFERRED_STACK_BOUNDARY} is not defined. On most machines,
964 this should be the same as @code{PARM_BOUNDARY}.
967 @defmac PREFERRED_STACK_BOUNDARY
968 Define this macro if you wish to preserve a certain alignment for the
969 stack pointer, greater than what the hardware enforces. The definition
970 is a C expression for the desired alignment (measured in bits). This
971 macro must evaluate to a value equal to or larger than
972 @code{STACK_BOUNDARY}.
975 @defmac INCOMING_STACK_BOUNDARY
976 Define this macro if the incoming stack boundary may be different
977 from @code{PREFERRED_STACK_BOUNDARY}. This macro must evaluate
978 to a value equal to or larger than @code{STACK_BOUNDARY}.
981 @defmac FUNCTION_BOUNDARY
982 Alignment required for a function entry point, in bits.
985 @defmac BIGGEST_ALIGNMENT
986 Biggest alignment that any data type can require on this machine, in
987 bits. Note that this is not the biggest alignment that is supported,
988 just the biggest alignment that, when violated, may cause a fault.
991 @defmac MALLOC_ABI_ALIGNMENT
992 Alignment, in bits, a C conformant malloc implementation has to
993 provide. If not defined, the default value is @code{BITS_PER_WORD}.
996 @defmac ATTRIBUTE_ALIGNED_VALUE
997 Alignment used by the @code{__attribute__ ((aligned))} construct. If
998 not defined, the default value is @code{BIGGEST_ALIGNMENT}.
1001 @defmac MINIMUM_ATOMIC_ALIGNMENT
1002 If defined, the smallest alignment, in bits, that can be given to an
1003 object that can be referenced in one operation, without disturbing any
1004 nearby object. Normally, this is @code{BITS_PER_UNIT}, but may be larger
1005 on machines that don't have byte or half-word store operations.
1008 @defmac BIGGEST_FIELD_ALIGNMENT
1009 Biggest alignment that any structure or union field can require on this
1010 machine, in bits. If defined, this overrides @code{BIGGEST_ALIGNMENT} for
1011 structure and union fields only, unless the field alignment has been set
1012 by the @code{__attribute__ ((aligned (@var{n})))} construct.
1015 @defmac ADJUST_FIELD_ALIGN (@var{field}, @var{computed})
1016 An expression for the alignment of a structure field @var{field} if the
1017 alignment computed in the usual way (including applying of
1018 @code{BIGGEST_ALIGNMENT} and @code{BIGGEST_FIELD_ALIGNMENT} to the
1019 alignment) is @var{computed}. It overrides alignment only if the
1020 field alignment has not been set by the
1021 @code{__attribute__ ((aligned (@var{n})))} construct.
1024 @defmac MAX_STACK_ALIGNMENT
1025 Biggest stack alignment guaranteed by the backend. Use this macro
1026 to specify the maximum alignment of a variable on stack.
1028 If not defined, the default value is @code{STACK_BOUNDARY}.
1030 @c FIXME: The default should be @code{PREFERRED_STACK_BOUNDARY}.
1031 @c But the fix for PR 32893 indicates that we can only guarantee
1032 @c maximum stack alignment on stack up to @code{STACK_BOUNDARY}, not
1033 @c @code{PREFERRED_STACK_BOUNDARY}, if stack alignment isn't supported.
1036 @defmac MAX_OFILE_ALIGNMENT
1037 Biggest alignment supported by the object file format of this machine.
1038 Use this macro to limit the alignment which can be specified using the
1039 @code{__attribute__ ((aligned (@var{n})))} construct. If not defined,
1040 the default value is @code{BIGGEST_ALIGNMENT}.
1042 On systems that use ELF, the default (in @file{config/elfos.h}) is
1043 the largest supported 32-bit ELF section alignment representable on
1044 a 32-bit host e.g. @samp{(((unsigned HOST_WIDEST_INT) 1 << 28) * 8)}.
1045 On 32-bit ELF the largest supported section alignment in bits is
1046 @samp{(0x80000000 * 8)}, but this is not representable on 32-bit hosts.
1049 @defmac DATA_ALIGNMENT (@var{type}, @var{basic-align})
1050 If defined, a C expression to compute the alignment for a variable in
1051 the static store. @var{type} is the data type, and @var{basic-align} is
1052 the alignment that the object would ordinarily have. The value of this
1053 macro is used instead of that alignment to align the object.
1055 If this macro is not defined, then @var{basic-align} is used.
1058 One use of this macro is to increase alignment of medium-size data to
1059 make it all fit in fewer cache lines. Another is to cause character
1060 arrays to be word-aligned so that @code{strcpy} calls that copy
1061 constants to character arrays can be done inline.
1064 @defmac CONSTANT_ALIGNMENT (@var{constant}, @var{basic-align})
1065 If defined, a C expression to compute the alignment given to a constant
1066 that is being placed in memory. @var{constant} is the constant and
1067 @var{basic-align} is the alignment that the object would ordinarily
1068 have. The value of this macro is used instead of that alignment to
1071 If this macro is not defined, then @var{basic-align} is used.
1073 The typical use of this macro is to increase alignment for string
1074 constants to be word aligned so that @code{strcpy} calls that copy
1075 constants can be done inline.
1078 @defmac LOCAL_ALIGNMENT (@var{type}, @var{basic-align})
1079 If defined, a C expression to compute the alignment for a variable in
1080 the local store. @var{type} is the data type, and @var{basic-align} is
1081 the alignment that the object would ordinarily have. The value of this
1082 macro is used instead of that alignment to align the object.
1084 If this macro is not defined, then @var{basic-align} is used.
1086 One use of this macro is to increase alignment of medium-size data to
1087 make it all fit in fewer cache lines.
1089 If the value of this macro has a type, it should be an unsigned type.
1092 @defmac STACK_SLOT_ALIGNMENT (@var{type}, @var{mode}, @var{basic-align})
1093 If defined, a C expression to compute the alignment for stack slot.
1094 @var{type} is the data type, @var{mode} is the widest mode available,
1095 and @var{basic-align} is the alignment that the slot would ordinarily
1096 have. The value of this macro is used instead of that alignment to
1099 If this macro is not defined, then @var{basic-align} is used when
1100 @var{type} is @code{NULL}. Otherwise, @code{LOCAL_ALIGNMENT} will
1103 This macro is to set alignment of stack slot to the maximum alignment
1104 of all possible modes which the slot may have.
1106 If the value of this macro has a type, it should be an unsigned type.
1109 @defmac LOCAL_DECL_ALIGNMENT (@var{decl})
1110 If defined, a C expression to compute the alignment for a local
1111 variable @var{decl}.
1113 If this macro is not defined, then
1114 @code{LOCAL_ALIGNMENT (TREE_TYPE (@var{decl}), DECL_ALIGN (@var{decl}))}
1117 One use of this macro is to increase alignment of medium-size data to
1118 make it all fit in fewer cache lines.
1120 If the value of this macro has a type, it should be an unsigned type.
1123 @defmac MINIMUM_ALIGNMENT (@var{exp}, @var{mode}, @var{align})
1124 If defined, a C expression to compute the minimum required alignment
1125 for dynamic stack realignment purposes for @var{exp} (a type or decl),
1126 @var{mode}, assuming normal alignment @var{align}.
1128 If this macro is not defined, then @var{align} will be used.
1131 @defmac EMPTY_FIELD_BOUNDARY
1132 Alignment in bits to be given to a structure bit-field that follows an
1133 empty field such as @code{int : 0;}.
1135 If @code{PCC_BITFIELD_TYPE_MATTERS} is true, it overrides this macro.
1138 @defmac STRUCTURE_SIZE_BOUNDARY
1139 Number of bits which any structure or union's size must be a multiple of.
1140 Each structure or union's size is rounded up to a multiple of this.
1142 If you do not define this macro, the default is the same as
1143 @code{BITS_PER_UNIT}.
1146 @defmac STRICT_ALIGNMENT
1147 Define this macro to be the value 1 if instructions will fail to work
1148 if given data not on the nominal alignment. If instructions will merely
1149 go slower in that case, define this macro as 0.
1152 @defmac PCC_BITFIELD_TYPE_MATTERS
1153 Define this if you wish to imitate the way many other C compilers handle
1154 alignment of bit-fields and the structures that contain them.
1156 The behavior is that the type written for a named bit-field (@code{int},
1157 @code{short}, or other integer type) imposes an alignment for the entire
1158 structure, as if the structure really did contain an ordinary field of
1159 that type. In addition, the bit-field is placed within the structure so
1160 that it would fit within such a field, not crossing a boundary for it.
1162 Thus, on most machines, a named bit-field whose type is written as
1163 @code{int} would not cross a four-byte boundary, and would force
1164 four-byte alignment for the whole structure. (The alignment used may
1165 not be four bytes; it is controlled by the other alignment parameters.)
1167 An unnamed bit-field will not affect the alignment of the containing
1170 If the macro is defined, its definition should be a C expression;
1171 a nonzero value for the expression enables this behavior.
1173 Note that if this macro is not defined, or its value is zero, some
1174 bit-fields may cross more than one alignment boundary. The compiler can
1175 support such references if there are @samp{insv}, @samp{extv}, and
1176 @samp{extzv} insns that can directly reference memory.
1178 The other known way of making bit-fields work is to define
1179 @code{STRUCTURE_SIZE_BOUNDARY} as large as @code{BIGGEST_ALIGNMENT}.
1180 Then every structure can be accessed with fullwords.
1182 Unless the machine has bit-field instructions or you define
1183 @code{STRUCTURE_SIZE_BOUNDARY} that way, you must define
1184 @code{PCC_BITFIELD_TYPE_MATTERS} to have a nonzero value.
1186 If your aim is to make GCC use the same conventions for laying out
1187 bit-fields as are used by another compiler, here is how to investigate
1188 what the other compiler does. Compile and run this program:
1207 printf ("Size of foo1 is %d\n",
1208 sizeof (struct foo1));
1209 printf ("Size of foo2 is %d\n",
1210 sizeof (struct foo2));
1215 If this prints 2 and 5, then the compiler's behavior is what you would
1216 get from @code{PCC_BITFIELD_TYPE_MATTERS}.
1219 @defmac BITFIELD_NBYTES_LIMITED
1220 Like @code{PCC_BITFIELD_TYPE_MATTERS} except that its effect is limited
1221 to aligning a bit-field within the structure.
1224 @hook TARGET_ALIGN_ANON_BITFIELD
1225 When @code{PCC_BITFIELD_TYPE_MATTERS} is true this hook will determine
1226 whether unnamed bitfields affect the alignment of the containing
1227 structure. The hook should return true if the structure should inherit
1228 the alignment requirements of an unnamed bitfield's type.
1231 @hook TARGET_NARROW_VOLATILE_BITFIELD
1232 This target hook should return @code{true} if accesses to volatile bitfields
1233 should use the narrowest mode possible. It should return @code{false} if
1234 these accesses should use the bitfield container type.
1236 The default is @code{!TARGET_STRICT_ALIGN}.
1239 @defmac MEMBER_TYPE_FORCES_BLK (@var{field}, @var{mode})
1240 Return 1 if a structure or array containing @var{field} should be accessed using
1243 If @var{field} is the only field in the structure, @var{mode} is its
1244 mode, otherwise @var{mode} is VOIDmode. @var{mode} is provided in the
1245 case where structures of one field would require the structure's mode to
1246 retain the field's mode.
1248 Normally, this is not needed.
1251 @defmac ROUND_TYPE_ALIGN (@var{type}, @var{computed}, @var{specified})
1252 Define this macro as an expression for the alignment of a type (given
1253 by @var{type} as a tree node) if the alignment computed in the usual
1254 way is @var{computed} and the alignment explicitly specified was
1257 The default is to use @var{specified} if it is larger; otherwise, use
1258 the smaller of @var{computed} and @code{BIGGEST_ALIGNMENT}
1261 @defmac MAX_FIXED_MODE_SIZE
1262 An integer expression for the size in bits of the largest integer
1263 machine mode that should actually be used. All integer machine modes of
1264 this size or smaller can be used for structures and unions with the
1265 appropriate sizes. If this macro is undefined, @code{GET_MODE_BITSIZE
1266 (DImode)} is assumed.
1269 @defmac STACK_SAVEAREA_MODE (@var{save_level})
1270 If defined, an expression of type @code{enum machine_mode} that
1271 specifies the mode of the save area operand of a
1272 @code{save_stack_@var{level}} named pattern (@pxref{Standard Names}).
1273 @var{save_level} is one of @code{SAVE_BLOCK}, @code{SAVE_FUNCTION}, or
1274 @code{SAVE_NONLOCAL} and selects which of the three named patterns is
1275 having its mode specified.
1277 You need not define this macro if it always returns @code{Pmode}. You
1278 would most commonly define this macro if the
1279 @code{save_stack_@var{level}} patterns need to support both a 32- and a
1283 @defmac STACK_SIZE_MODE
1284 If defined, an expression of type @code{enum machine_mode} that
1285 specifies the mode of the size increment operand of an
1286 @code{allocate_stack} named pattern (@pxref{Standard Names}).
1288 You need not define this macro if it always returns @code{word_mode}.
1289 You would most commonly define this macro if the @code{allocate_stack}
1290 pattern needs to support both a 32- and a 64-bit mode.
1293 @hook TARGET_LIBGCC_CMP_RETURN_MODE
1294 This target hook should return the mode to be used for the return value
1295 of compare instructions expanded to libgcc calls. If not defined
1296 @code{word_mode} is returned which is the right choice for a majority of
1300 @hook TARGET_LIBGCC_SHIFT_COUNT_MODE
1301 This target hook should return the mode to be used for the shift count operand
1302 of shift instructions expanded to libgcc calls. If not defined
1303 @code{word_mode} is returned which is the right choice for a majority of
1307 @hook TARGET_UNWIND_WORD_MODE
1308 Return machine mode to be used for @code{_Unwind_Word} type.
1309 The default is to use @code{word_mode}.
1312 @defmac ROUND_TOWARDS_ZERO
1313 If defined, this macro should be true if the prevailing rounding
1314 mode is towards zero.
1316 Defining this macro only affects the way @file{libgcc.a} emulates
1317 floating-point arithmetic.
1319 Not defining this macro is equivalent to returning zero.
1322 @defmac LARGEST_EXPONENT_IS_NORMAL (@var{size})
1323 This macro should return true if floats with @var{size}
1324 bits do not have a NaN or infinity representation, but use the largest
1325 exponent for normal numbers instead.
1327 Defining this macro only affects the way @file{libgcc.a} emulates
1328 floating-point arithmetic.
1330 The default definition of this macro returns false for all sizes.
1333 @hook TARGET_MS_BITFIELD_LAYOUT_P
1334 This target hook returns @code{true} if bit-fields in the given
1335 @var{record_type} are to be laid out following the rules of Microsoft
1336 Visual C/C++, namely: (i) a bit-field won't share the same storage
1337 unit with the previous bit-field if their underlying types have
1338 different sizes, and the bit-field will be aligned to the highest
1339 alignment of the underlying types of itself and of the previous
1340 bit-field; (ii) a zero-sized bit-field will affect the alignment of
1341 the whole enclosing structure, even if it is unnamed; except that
1342 (iii) a zero-sized bit-field will be disregarded unless it follows
1343 another bit-field of nonzero size. If this hook returns @code{true},
1344 other macros that control bit-field layout are ignored.
1346 When a bit-field is inserted into a packed record, the whole size
1347 of the underlying type is used by one or more same-size adjacent
1348 bit-fields (that is, if its long:3, 32 bits is used in the record,
1349 and any additional adjacent long bit-fields are packed into the same
1350 chunk of 32 bits. However, if the size changes, a new field of that
1351 size is allocated). In an unpacked record, this is the same as using
1352 alignment, but not equivalent when packing.
1354 If both MS bit-fields and @samp{__attribute__((packed))} are used,
1355 the latter will take precedence. If @samp{__attribute__((packed))} is
1356 used on a single field when MS bit-fields are in use, it will take
1357 precedence for that field, but the alignment of the rest of the structure
1358 may affect its placement.
1361 @hook TARGET_DECIMAL_FLOAT_SUPPORTED_P
1362 Returns true if the target supports decimal floating point.
1365 @hook TARGET_FIXED_POINT_SUPPORTED_P
1366 Returns true if the target supports fixed-point arithmetic.
1369 @hook TARGET_EXPAND_TO_RTL_HOOK
1370 This hook is called just before expansion into rtl, allowing the target
1371 to perform additional initializations or analysis before the expansion.
1372 For example, the rs6000 port uses it to allocate a scratch stack slot
1373 for use in copying SDmode values between memory and floating point
1374 registers whenever the function being expanded has any SDmode
1378 @hook TARGET_INSTANTIATE_DECLS
1379 This hook allows the backend to perform additional instantiations on rtl
1380 that are not actually in any insns yet, but will be later.
1383 @hook TARGET_MANGLE_TYPE
1384 If your target defines any fundamental types, or any types your target
1385 uses should be mangled differently from the default, define this hook
1386 to return the appropriate encoding for these types as part of a C++
1387 mangled name. The @var{type} argument is the tree structure representing
1388 the type to be mangled. The hook may be applied to trees which are
1389 not target-specific fundamental types; it should return @code{NULL}
1390 for all such types, as well as arguments it does not recognize. If the
1391 return value is not @code{NULL}, it must point to a statically-allocated
1394 Target-specific fundamental types might be new fundamental types or
1395 qualified versions of ordinary fundamental types. Encode new
1396 fundamental types as @samp{@w{u @var{n} @var{name}}}, where @var{name}
1397 is the name used for the type in source code, and @var{n} is the
1398 length of @var{name} in decimal. Encode qualified versions of
1399 ordinary types as @samp{@w{U @var{n} @var{name} @var{code}}}, where
1400 @var{name} is the name used for the type qualifier in source code,
1401 @var{n} is the length of @var{name} as above, and @var{code} is the
1402 code used to represent the unqualified version of this type. (See
1403 @code{write_builtin_type} in @file{cp/mangle.c} for the list of
1404 codes.) In both cases the spaces are for clarity; do not include any
1405 spaces in your string.
1407 This hook is applied to types prior to typedef resolution. If the mangled
1408 name for a particular type depends only on that type's main variant, you
1409 can perform typedef resolution yourself using @code{TYPE_MAIN_VARIANT}
1412 The default version of this hook always returns @code{NULL}, which is
1413 appropriate for a target that does not define any new fundamental
1418 @section Layout of Source Language Data Types
1420 These macros define the sizes and other characteristics of the standard
1421 basic data types used in programs being compiled. Unlike the macros in
1422 the previous section, these apply to specific features of C and related
1423 languages, rather than to fundamental aspects of storage layout.
1425 @defmac INT_TYPE_SIZE
1426 A C expression for the size in bits of the type @code{int} on the
1427 target machine. If you don't define this, the default is one word.
1430 @defmac SHORT_TYPE_SIZE
1431 A C expression for the size in bits of the type @code{short} on the
1432 target machine. If you don't define this, the default is half a word.
1433 (If this would be less than one storage unit, it is rounded up to one
1437 @defmac LONG_TYPE_SIZE
1438 A C expression for the size in bits of the type @code{long} on the
1439 target machine. If you don't define this, the default is one word.
1442 @defmac ADA_LONG_TYPE_SIZE
1443 On some machines, the size used for the Ada equivalent of the type
1444 @code{long} by a native Ada compiler differs from that used by C@. In
1445 that situation, define this macro to be a C expression to be used for
1446 the size of that type. If you don't define this, the default is the
1447 value of @code{LONG_TYPE_SIZE}.
1450 @defmac LONG_LONG_TYPE_SIZE
1451 A C expression for the size in bits of the type @code{long long} on the
1452 target machine. If you don't define this, the default is two
1453 words. If you want to support GNU Ada on your machine, the value of this
1454 macro must be at least 64.
1457 @defmac CHAR_TYPE_SIZE
1458 A C expression for the size in bits of the type @code{char} on the
1459 target machine. If you don't define this, the default is
1460 @code{BITS_PER_UNIT}.
1463 @defmac BOOL_TYPE_SIZE
1464 A C expression for the size in bits of the C++ type @code{bool} and
1465 C99 type @code{_Bool} on the target machine. If you don't define
1466 this, and you probably shouldn't, the default is @code{CHAR_TYPE_SIZE}.
1469 @defmac FLOAT_TYPE_SIZE
1470 A C expression for the size in bits of the type @code{float} on the
1471 target machine. If you don't define this, the default is one word.
1474 @defmac DOUBLE_TYPE_SIZE
1475 A C expression for the size in bits of the type @code{double} on the
1476 target machine. If you don't define this, the default is two
1480 @defmac LONG_DOUBLE_TYPE_SIZE
1481 A C expression for the size in bits of the type @code{long double} on
1482 the target machine. If you don't define this, the default is two
1486 @defmac SHORT_FRACT_TYPE_SIZE
1487 A C expression for the size in bits of the type @code{short _Fract} on
1488 the target machine. If you don't define this, the default is
1489 @code{BITS_PER_UNIT}.
1492 @defmac FRACT_TYPE_SIZE
1493 A C expression for the size in bits of the type @code{_Fract} on
1494 the target machine. If you don't define this, the default is
1495 @code{BITS_PER_UNIT * 2}.
1498 @defmac LONG_FRACT_TYPE_SIZE
1499 A C expression for the size in bits of the type @code{long _Fract} on
1500 the target machine. If you don't define this, the default is
1501 @code{BITS_PER_UNIT * 4}.
1504 @defmac LONG_LONG_FRACT_TYPE_SIZE
1505 A C expression for the size in bits of the type @code{long long _Fract} on
1506 the target machine. If you don't define this, the default is
1507 @code{BITS_PER_UNIT * 8}.
1510 @defmac SHORT_ACCUM_TYPE_SIZE
1511 A C expression for the size in bits of the type @code{short _Accum} on
1512 the target machine. If you don't define this, the default is
1513 @code{BITS_PER_UNIT * 2}.
1516 @defmac ACCUM_TYPE_SIZE
1517 A C expression for the size in bits of the type @code{_Accum} on
1518 the target machine. If you don't define this, the default is
1519 @code{BITS_PER_UNIT * 4}.
1522 @defmac LONG_ACCUM_TYPE_SIZE
1523 A C expression for the size in bits of the type @code{long _Accum} on
1524 the target machine. If you don't define this, the default is
1525 @code{BITS_PER_UNIT * 8}.
1528 @defmac LONG_LONG_ACCUM_TYPE_SIZE
1529 A C expression for the size in bits of the type @code{long long _Accum} on
1530 the target machine. If you don't define this, the default is
1531 @code{BITS_PER_UNIT * 16}.
1534 @defmac LIBGCC2_LONG_DOUBLE_TYPE_SIZE
1535 Define this macro if @code{LONG_DOUBLE_TYPE_SIZE} is not constant or
1536 if you want routines in @file{libgcc2.a} for a size other than
1537 @code{LONG_DOUBLE_TYPE_SIZE}. If you don't define this, the
1538 default is @code{LONG_DOUBLE_TYPE_SIZE}.
1541 @defmac LIBGCC2_HAS_DF_MODE
1542 Define this macro if neither @code{DOUBLE_TYPE_SIZE} nor
1543 @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is
1544 @code{DFmode} but you want @code{DFmode} routines in @file{libgcc2.a}
1545 anyway. If you don't define this and either @code{DOUBLE_TYPE_SIZE}
1546 or @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 64 then the default is 1,
1550 @defmac LIBGCC2_HAS_XF_MODE
1551 Define this macro if @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is not
1552 @code{XFmode} but you want @code{XFmode} routines in @file{libgcc2.a}
1553 anyway. If you don't define this and @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE}
1554 is 80 then the default is 1, otherwise it is 0.
1557 @defmac LIBGCC2_HAS_TF_MODE
1558 Define this macro if @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is not
1559 @code{TFmode} but you want @code{TFmode} routines in @file{libgcc2.a}
1560 anyway. If you don't define this and @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE}
1561 is 128 then the default is 1, otherwise it is 0.
1568 Define these macros to be the size in bits of the mantissa of
1569 @code{SFmode}, @code{DFmode}, @code{XFmode} and @code{TFmode} values,
1570 if the defaults in @file{libgcc2.h} are inappropriate. By default,
1571 @code{FLT_MANT_DIG} is used for @code{SF_SIZE}, @code{LDBL_MANT_DIG}
1572 for @code{XF_SIZE} and @code{TF_SIZE}, and @code{DBL_MANT_DIG} or
1573 @code{LDBL_MANT_DIG} for @code{DF_SIZE} according to whether
1574 @code{DOUBLE_TYPE_SIZE} or
1575 @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 64.
1578 @defmac TARGET_FLT_EVAL_METHOD
1579 A C expression for the value for @code{FLT_EVAL_METHOD} in @file{float.h},
1580 assuming, if applicable, that the floating-point control word is in its
1581 default state. If you do not define this macro the value of
1582 @code{FLT_EVAL_METHOD} will be zero.
1585 @defmac WIDEST_HARDWARE_FP_SIZE
1586 A C expression for the size in bits of the widest floating-point format
1587 supported by the hardware. If you define this macro, you must specify a
1588 value less than or equal to the value of @code{LONG_DOUBLE_TYPE_SIZE}.
1589 If you do not define this macro, the value of @code{LONG_DOUBLE_TYPE_SIZE}
1593 @defmac DEFAULT_SIGNED_CHAR
1594 An expression whose value is 1 or 0, according to whether the type
1595 @code{char} should be signed or unsigned by default. The user can
1596 always override this default with the options @option{-fsigned-char}
1597 and @option{-funsigned-char}.
1600 @hook TARGET_DEFAULT_SHORT_ENUMS
1601 This target hook should return true if the compiler should give an
1602 @code{enum} type only as many bytes as it takes to represent the range
1603 of possible values of that type. It should return false if all
1604 @code{enum} types should be allocated like @code{int}.
1606 The default is to return false.
1610 A C expression for a string describing the name of the data type to use
1611 for size values. The typedef name @code{size_t} is defined using the
1612 contents of the string.
1614 The string can contain more than one keyword. If so, separate them with
1615 spaces, and write first any length keyword, then @code{unsigned} if
1616 appropriate, and finally @code{int}. The string must exactly match one
1617 of the data type names defined in the function
1618 @code{init_decl_processing} in the file @file{c-decl.c}. You may not
1619 omit @code{int} or change the order---that would cause the compiler to
1622 If you don't define this macro, the default is @code{"long unsigned
1626 @defmac PTRDIFF_TYPE
1627 A C expression for a string describing the name of the data type to use
1628 for the result of subtracting two pointers. The typedef name
1629 @code{ptrdiff_t} is defined using the contents of the string. See
1630 @code{SIZE_TYPE} above for more information.
1632 If you don't define this macro, the default is @code{"long int"}.
1636 A C expression for a string describing the name of the data type to use
1637 for wide characters. The typedef name @code{wchar_t} is defined using
1638 the contents of the string. See @code{SIZE_TYPE} above for more
1641 If you don't define this macro, the default is @code{"int"}.
1644 @defmac WCHAR_TYPE_SIZE
1645 A C expression for the size in bits of the data type for wide
1646 characters. This is used in @code{cpp}, which cannot make use of
1651 A C expression for a string describing the name of the data type to
1652 use for wide characters passed to @code{printf} and returned from
1653 @code{getwc}. The typedef name @code{wint_t} is defined using the
1654 contents of the string. See @code{SIZE_TYPE} above for more
1657 If you don't define this macro, the default is @code{"unsigned int"}.
1661 A C expression for a string describing the name of the data type that
1662 can represent any value of any standard or extended signed integer type.
1663 The typedef name @code{intmax_t} is defined using the contents of the
1664 string. See @code{SIZE_TYPE} above for more information.
1666 If you don't define this macro, the default is the first of
1667 @code{"int"}, @code{"long int"}, or @code{"long long int"} that has as
1668 much precision as @code{long long int}.
1671 @defmac UINTMAX_TYPE
1672 A C expression for a string describing the name of the data type that
1673 can represent any value of any standard or extended unsigned integer
1674 type. The typedef name @code{uintmax_t} is defined using the contents
1675 of the string. See @code{SIZE_TYPE} above for more information.
1677 If you don't define this macro, the default is the first of
1678 @code{"unsigned int"}, @code{"long unsigned int"}, or @code{"long long
1679 unsigned int"} that has as much precision as @code{long long unsigned
1683 @defmac SIG_ATOMIC_TYPE
1689 @defmacx UINT16_TYPE
1690 @defmacx UINT32_TYPE
1691 @defmacx UINT64_TYPE
1692 @defmacx INT_LEAST8_TYPE
1693 @defmacx INT_LEAST16_TYPE
1694 @defmacx INT_LEAST32_TYPE
1695 @defmacx INT_LEAST64_TYPE
1696 @defmacx UINT_LEAST8_TYPE
1697 @defmacx UINT_LEAST16_TYPE
1698 @defmacx UINT_LEAST32_TYPE
1699 @defmacx UINT_LEAST64_TYPE
1700 @defmacx INT_FAST8_TYPE
1701 @defmacx INT_FAST16_TYPE
1702 @defmacx INT_FAST32_TYPE
1703 @defmacx INT_FAST64_TYPE
1704 @defmacx UINT_FAST8_TYPE
1705 @defmacx UINT_FAST16_TYPE
1706 @defmacx UINT_FAST32_TYPE
1707 @defmacx UINT_FAST64_TYPE
1708 @defmacx INTPTR_TYPE
1709 @defmacx UINTPTR_TYPE
1710 C expressions for the standard types @code{sig_atomic_t},
1711 @code{int8_t}, @code{int16_t}, @code{int32_t}, @code{int64_t},
1712 @code{uint8_t}, @code{uint16_t}, @code{uint32_t}, @code{uint64_t},
1713 @code{int_least8_t}, @code{int_least16_t}, @code{int_least32_t},
1714 @code{int_least64_t}, @code{uint_least8_t}, @code{uint_least16_t},
1715 @code{uint_least32_t}, @code{uint_least64_t}, @code{int_fast8_t},
1716 @code{int_fast16_t}, @code{int_fast32_t}, @code{int_fast64_t},
1717 @code{uint_fast8_t}, @code{uint_fast16_t}, @code{uint_fast32_t},
1718 @code{uint_fast64_t}, @code{intptr_t}, and @code{uintptr_t}. See
1719 @code{SIZE_TYPE} above for more information.
1721 If any of these macros evaluates to a null pointer, the corresponding
1722 type is not supported; if GCC is configured to provide
1723 @code{<stdint.h>} in such a case, the header provided may not conform
1724 to C99, depending on the type in question. The defaults for all of
1725 these macros are null pointers.
1728 @defmac TARGET_PTRMEMFUNC_VBIT_LOCATION
1729 The C++ compiler represents a pointer-to-member-function with a struct
1736 ptrdiff_t vtable_index;
1743 The C++ compiler must use one bit to indicate whether the function that
1744 will be called through a pointer-to-member-function is virtual.
1745 Normally, we assume that the low-order bit of a function pointer must
1746 always be zero. Then, by ensuring that the vtable_index is odd, we can
1747 distinguish which variant of the union is in use. But, on some
1748 platforms function pointers can be odd, and so this doesn't work. In
1749 that case, we use the low-order bit of the @code{delta} field, and shift
1750 the remainder of the @code{delta} field to the left.
1752 GCC will automatically make the right selection about where to store
1753 this bit using the @code{FUNCTION_BOUNDARY} setting for your platform.
1754 However, some platforms such as ARM/Thumb have @code{FUNCTION_BOUNDARY}
1755 set such that functions always start at even addresses, but the lowest
1756 bit of pointers to functions indicate whether the function at that
1757 address is in ARM or Thumb mode. If this is the case of your
1758 architecture, you should define this macro to
1759 @code{ptrmemfunc_vbit_in_delta}.
1761 In general, you should not have to define this macro. On architectures
1762 in which function addresses are always even, according to
1763 @code{FUNCTION_BOUNDARY}, GCC will automatically define this macro to
1764 @code{ptrmemfunc_vbit_in_pfn}.
1767 @defmac TARGET_VTABLE_USES_DESCRIPTORS
1768 Normally, the C++ compiler uses function pointers in vtables. This
1769 macro allows the target to change to use ``function descriptors''
1770 instead. Function descriptors are found on targets for whom a
1771 function pointer is actually a small data structure. Normally the
1772 data structure consists of the actual code address plus a data
1773 pointer to which the function's data is relative.
1775 If vtables are used, the value of this macro should be the number
1776 of words that the function descriptor occupies.
1779 @defmac TARGET_VTABLE_ENTRY_ALIGN
1780 By default, the vtable entries are void pointers, the so the alignment
1781 is the same as pointer alignment. The value of this macro specifies
1782 the alignment of the vtable entry in bits. It should be defined only
1783 when special alignment is necessary. */
1786 @defmac TARGET_VTABLE_DATA_ENTRY_DISTANCE
1787 There are a few non-descriptor entries in the vtable at offsets below
1788 zero. If these entries must be padded (say, to preserve the alignment
1789 specified by @code{TARGET_VTABLE_ENTRY_ALIGN}), set this to the number
1790 of words in each data entry.
1794 @section Register Usage
1795 @cindex register usage
1797 This section explains how to describe what registers the target machine
1798 has, and how (in general) they can be used.
1800 The description of which registers a specific instruction can use is
1801 done with register classes; see @ref{Register Classes}. For information
1802 on using registers to access a stack frame, see @ref{Frame Registers}.
1803 For passing values in registers, see @ref{Register Arguments}.
1804 For returning values in registers, see @ref{Scalar Return}.
1807 * Register Basics:: Number and kinds of registers.
1808 * Allocation Order:: Order in which registers are allocated.
1809 * Values in Registers:: What kinds of values each reg can hold.
1810 * Leaf Functions:: Renumbering registers for leaf functions.
1811 * Stack Registers:: Handling a register stack such as 80387.
1814 @node Register Basics
1815 @subsection Basic Characteristics of Registers
1817 @c prevent bad page break with this line
1818 Registers have various characteristics.
1820 @defmac FIRST_PSEUDO_REGISTER
1821 Number of hardware registers known to the compiler. They receive
1822 numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first
1823 pseudo register's number really is assigned the number
1824 @code{FIRST_PSEUDO_REGISTER}.
1827 @defmac FIXED_REGISTERS
1828 @cindex fixed register
1829 An initializer that says which registers are used for fixed purposes
1830 all throughout the compiled code and are therefore not available for
1831 general allocation. These would include the stack pointer, the frame
1832 pointer (except on machines where that can be used as a general
1833 register when no frame pointer is needed), the program counter on
1834 machines where that is considered one of the addressable registers,
1835 and any other numbered register with a standard use.
1837 This information is expressed as a sequence of numbers, separated by
1838 commas and surrounded by braces. The @var{n}th number is 1 if
1839 register @var{n} is fixed, 0 otherwise.
1841 The table initialized from this macro, and the table initialized by
1842 the following one, may be overridden at run time either automatically,
1843 by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by
1844 the user with the command options @option{-ffixed-@var{reg}},
1845 @option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}.
1848 @defmac CALL_USED_REGISTERS
1849 @cindex call-used register
1850 @cindex call-clobbered register
1851 @cindex call-saved register
1852 Like @code{FIXED_REGISTERS} but has 1 for each register that is
1853 clobbered (in general) by function calls as well as for fixed
1854 registers. This macro therefore identifies the registers that are not
1855 available for general allocation of values that must live across
1858 If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler
1859 automatically saves it on function entry and restores it on function
1860 exit, if the register is used within the function.
1863 @defmac CALL_REALLY_USED_REGISTERS
1864 @cindex call-used register
1865 @cindex call-clobbered register
1866 @cindex call-saved register
1867 Like @code{CALL_USED_REGISTERS} except this macro doesn't require
1868 that the entire set of @code{FIXED_REGISTERS} be included.
1869 (@code{CALL_USED_REGISTERS} must be a superset of @code{FIXED_REGISTERS}).
1870 This macro is optional. If not specified, it defaults to the value
1871 of @code{CALL_USED_REGISTERS}.
1874 @defmac HARD_REGNO_CALL_PART_CLOBBERED (@var{regno}, @var{mode})
1875 @cindex call-used register
1876 @cindex call-clobbered register
1877 @cindex call-saved register
1878 A C expression that is nonzero if it is not permissible to store a
1879 value of mode @var{mode} in hard register number @var{regno} across a
1880 call without some part of it being clobbered. For most machines this
1881 macro need not be defined. It is only required for machines that do not
1882 preserve the entire contents of a register across a call.
1886 @findex call_used_regs
1889 @findex reg_class_contents
1890 @hook TARGET_CONDITIONAL_REGISTER_USAGE
1891 This hook may conditionally modify five variables
1892 @code{fixed_regs}, @code{call_used_regs}, @code{global_regs},
1893 @code{reg_names}, and @code{reg_class_contents}, to take into account
1894 any dependence of these register sets on target flags. The first three
1895 of these are of type @code{char []} (interpreted as Boolean vectors).
1896 @code{global_regs} is a @code{const char *[]}, and
1897 @code{reg_class_contents} is a @code{HARD_REG_SET}. Before the macro is
1898 called, @code{fixed_regs}, @code{call_used_regs},
1899 @code{reg_class_contents}, and @code{reg_names} have been initialized
1900 from @code{FIXED_REGISTERS}, @code{CALL_USED_REGISTERS},
1901 @code{REG_CLASS_CONTENTS}, and @code{REGISTER_NAMES}, respectively.
1902 @code{global_regs} has been cleared, and any @option{-ffixed-@var{reg}},
1903 @option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}
1904 command options have been applied.
1906 @cindex disabling certain registers
1907 @cindex controlling register usage
1908 If the usage of an entire class of registers depends on the target
1909 flags, you may indicate this to GCC by using this macro to modify
1910 @code{fixed_regs} and @code{call_used_regs} to 1 for each of the
1911 registers in the classes which should not be used by GCC@. Also define
1912 the macro @code{REG_CLASS_FROM_LETTER} / @code{REG_CLASS_FROM_CONSTRAINT}
1913 to return @code{NO_REGS} if it
1914 is called with a letter for a class that shouldn't be used.
1916 (However, if this class is not included in @code{GENERAL_REGS} and all
1917 of the insn patterns whose constraints permit this class are
1918 controlled by target switches, then GCC will automatically avoid using
1919 these registers when the target switches are opposed to them.)
1922 @defmac INCOMING_REGNO (@var{out})
1923 Define this macro if the target machine has register windows. This C
1924 expression returns the register number as seen by the called function
1925 corresponding to the register number @var{out} as seen by the calling
1926 function. Return @var{out} if register number @var{out} is not an
1930 @defmac OUTGOING_REGNO (@var{in})
1931 Define this macro if the target machine has register windows. This C
1932 expression returns the register number as seen by the calling function
1933 corresponding to the register number @var{in} as seen by the called
1934 function. Return @var{in} if register number @var{in} is not an inbound
1938 @defmac LOCAL_REGNO (@var{regno})
1939 Define this macro if the target machine has register windows. This C
1940 expression returns true if the register is call-saved but is in the
1941 register window. Unlike most call-saved registers, such registers
1942 need not be explicitly restored on function exit or during non-local
1947 If the program counter has a register number, define this as that
1948 register number. Otherwise, do not define it.
1951 @node Allocation Order
1952 @subsection Order of Allocation of Registers
1953 @cindex order of register allocation
1954 @cindex register allocation order
1956 @c prevent bad page break with this line
1957 Registers are allocated in order.
1959 @defmac REG_ALLOC_ORDER
1960 If defined, an initializer for a vector of integers, containing the
1961 numbers of hard registers in the order in which GCC should prefer
1962 to use them (from most preferred to least).
1964 If this macro is not defined, registers are used lowest numbered first
1965 (all else being equal).
1967 One use of this macro is on machines where the highest numbered
1968 registers must always be saved and the save-multiple-registers
1969 instruction supports only sequences of consecutive registers. On such
1970 machines, define @code{REG_ALLOC_ORDER} to be an initializer that lists
1971 the highest numbered allocable register first.
1974 @defmac ADJUST_REG_ALLOC_ORDER
1975 A C statement (sans semicolon) to choose the order in which to allocate
1976 hard registers for pseudo-registers local to a basic block.
1978 Store the desired register order in the array @code{reg_alloc_order}.
1979 Element 0 should be the register to allocate first; element 1, the next
1980 register; and so on.
1982 The macro body should not assume anything about the contents of
1983 @code{reg_alloc_order} before execution of the macro.
1985 On most machines, it is not necessary to define this macro.
1988 @defmac HONOR_REG_ALLOC_ORDER
1989 Normally, IRA tries to estimate the costs for saving a register in the
1990 prologue and restoring it in the epilogue. This discourages it from
1991 using call-saved registers. If a machine wants to ensure that IRA
1992 allocates registers in the order given by REG_ALLOC_ORDER even if some
1993 call-saved registers appear earlier than call-used ones, this macro
1997 @defmac IRA_HARD_REGNO_ADD_COST_MULTIPLIER (@var{regno})
1998 In some case register allocation order is not enough for the
1999 Integrated Register Allocator (@acronym{IRA}) to generate a good code.
2000 If this macro is defined, it should return a floating point value
2001 based on @var{regno}. The cost of using @var{regno} for a pseudo will
2002 be increased by approximately the pseudo's usage frequency times the
2003 value returned by this macro. Not defining this macro is equivalent
2004 to having it always return @code{0.0}.
2006 On most machines, it is not necessary to define this macro.
2009 @node Values in Registers
2010 @subsection How Values Fit in Registers
2012 This section discusses the macros that describe which kinds of values
2013 (specifically, which machine modes) each register can hold, and how many
2014 consecutive registers are needed for a given mode.
2016 @defmac HARD_REGNO_NREGS (@var{regno}, @var{mode})
2017 A C expression for the number of consecutive hard registers, starting
2018 at register number @var{regno}, required to hold a value of mode
2019 @var{mode}. This macro must never return zero, even if a register
2020 cannot hold the requested mode - indicate that with HARD_REGNO_MODE_OK
2021 and/or CANNOT_CHANGE_MODE_CLASS instead.
2023 On a machine where all registers are exactly one word, a suitable
2024 definition of this macro is
2027 #define HARD_REGNO_NREGS(REGNO, MODE) \
2028 ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) \
2033 @defmac HARD_REGNO_NREGS_HAS_PADDING (@var{regno}, @var{mode})
2034 A C expression that is nonzero if a value of mode @var{mode}, stored
2035 in memory, ends with padding that causes it to take up more space than
2036 in registers starting at register number @var{regno} (as determined by
2037 multiplying GCC's notion of the size of the register when containing
2038 this mode by the number of registers returned by
2039 @code{HARD_REGNO_NREGS}). By default this is zero.
2041 For example, if a floating-point value is stored in three 32-bit
2042 registers but takes up 128 bits in memory, then this would be
2045 This macros only needs to be defined if there are cases where
2046 @code{subreg_get_info}
2047 would otherwise wrongly determine that a @code{subreg} can be
2048 represented by an offset to the register number, when in fact such a
2049 @code{subreg} would contain some of the padding not stored in
2050 registers and so not be representable.
2053 @defmac HARD_REGNO_NREGS_WITH_PADDING (@var{regno}, @var{mode})
2054 For values of @var{regno} and @var{mode} for which
2055 @code{HARD_REGNO_NREGS_HAS_PADDING} returns nonzero, a C expression
2056 returning the greater number of registers required to hold the value
2057 including any padding. In the example above, the value would be four.
2060 @defmac REGMODE_NATURAL_SIZE (@var{mode})
2061 Define this macro if the natural size of registers that hold values
2062 of mode @var{mode} is not the word size. It is a C expression that
2063 should give the natural size in bytes for the specified mode. It is
2064 used by the register allocator to try to optimize its results. This
2065 happens for example on SPARC 64-bit where the natural size of
2066 floating-point registers is still 32-bit.
2069 @defmac HARD_REGNO_MODE_OK (@var{regno}, @var{mode})
2070 A C expression that is nonzero if it is permissible to store a value
2071 of mode @var{mode} in hard register number @var{regno} (or in several
2072 registers starting with that one). For a machine where all registers
2073 are equivalent, a suitable definition is
2076 #define HARD_REGNO_MODE_OK(REGNO, MODE) 1
2079 You need not include code to check for the numbers of fixed registers,
2080 because the allocation mechanism considers them to be always occupied.
2082 @cindex register pairs
2083 On some machines, double-precision values must be kept in even/odd
2084 register pairs. You can implement that by defining this macro to reject
2085 odd register numbers for such modes.
2087 The minimum requirement for a mode to be OK in a register is that the
2088 @samp{mov@var{mode}} instruction pattern support moves between the
2089 register and other hard register in the same class and that moving a
2090 value into the register and back out not alter it.
2092 Since the same instruction used to move @code{word_mode} will work for
2093 all narrower integer modes, it is not necessary on any machine for
2094 @code{HARD_REGNO_MODE_OK} to distinguish between these modes, provided
2095 you define patterns @samp{movhi}, etc., to take advantage of this. This
2096 is useful because of the interaction between @code{HARD_REGNO_MODE_OK}
2097 and @code{MODES_TIEABLE_P}; it is very desirable for all integer modes
2100 Many machines have special registers for floating point arithmetic.
2101 Often people assume that floating point machine modes are allowed only
2102 in floating point registers. This is not true. Any registers that
2103 can hold integers can safely @emph{hold} a floating point machine
2104 mode, whether or not floating arithmetic can be done on it in those
2105 registers. Integer move instructions can be used to move the values.
2107 On some machines, though, the converse is true: fixed-point machine
2108 modes may not go in floating registers. This is true if the floating
2109 registers normalize any value stored in them, because storing a
2110 non-floating value there would garble it. In this case,
2111 @code{HARD_REGNO_MODE_OK} should reject fixed-point machine modes in
2112 floating registers. But if the floating registers do not automatically
2113 normalize, if you can store any bit pattern in one and retrieve it
2114 unchanged without a trap, then any machine mode may go in a floating
2115 register, so you can define this macro to say so.
2117 The primary significance of special floating registers is rather that
2118 they are the registers acceptable in floating point arithmetic
2119 instructions. However, this is of no concern to
2120 @code{HARD_REGNO_MODE_OK}. You handle it by writing the proper
2121 constraints for those instructions.
2123 On some machines, the floating registers are especially slow to access,
2124 so that it is better to store a value in a stack frame than in such a
2125 register if floating point arithmetic is not being done. As long as the
2126 floating registers are not in class @code{GENERAL_REGS}, they will not
2127 be used unless some pattern's constraint asks for one.
2130 @defmac HARD_REGNO_RENAME_OK (@var{from}, @var{to})
2131 A C expression that is nonzero if it is OK to rename a hard register
2132 @var{from} to another hard register @var{to}.
2134 One common use of this macro is to prevent renaming of a register to
2135 another register that is not saved by a prologue in an interrupt
2138 The default is always nonzero.
2141 @defmac MODES_TIEABLE_P (@var{mode1}, @var{mode2})
2142 A C expression that is nonzero if a value of mode
2143 @var{mode1} is accessible in mode @var{mode2} without copying.
2145 If @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and
2146 @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are always the same for
2147 any @var{r}, then @code{MODES_TIEABLE_P (@var{mode1}, @var{mode2})}
2148 should be nonzero. If they differ for any @var{r}, you should define
2149 this macro to return zero unless some other mechanism ensures the
2150 accessibility of the value in a narrower mode.
2152 You should define this macro to return nonzero in as many cases as
2153 possible since doing so will allow GCC to perform better register
2157 @hook TARGET_HARD_REGNO_SCRATCH_OK
2158 This target hook should return @code{true} if it is OK to use a hard register
2159 @var{regno} as scratch reg in peephole2.
2161 One common use of this macro is to prevent using of a register that
2162 is not saved by a prologue in an interrupt handler.
2164 The default version of this hook always returns @code{true}.
2167 @defmac AVOID_CCMODE_COPIES
2168 Define this macro if the compiler should avoid copies to/from @code{CCmode}
2169 registers. You should only define this macro if support for copying to/from
2170 @code{CCmode} is incomplete.
2173 @node Leaf Functions
2174 @subsection Handling Leaf Functions
2176 @cindex leaf functions
2177 @cindex functions, leaf
2178 On some machines, a leaf function (i.e., one which makes no calls) can run
2179 more efficiently if it does not make its own register window. Often this
2180 means it is required to receive its arguments in the registers where they
2181 are passed by the caller, instead of the registers where they would
2184 The special treatment for leaf functions generally applies only when
2185 other conditions are met; for example, often they may use only those
2186 registers for its own variables and temporaries. We use the term ``leaf
2187 function'' to mean a function that is suitable for this special
2188 handling, so that functions with no calls are not necessarily ``leaf
2191 GCC assigns register numbers before it knows whether the function is
2192 suitable for leaf function treatment. So it needs to renumber the
2193 registers in order to output a leaf function. The following macros
2196 @defmac LEAF_REGISTERS
2197 Name of a char vector, indexed by hard register number, which
2198 contains 1 for a register that is allowable in a candidate for leaf
2201 If leaf function treatment involves renumbering the registers, then the
2202 registers marked here should be the ones before renumbering---those that
2203 GCC would ordinarily allocate. The registers which will actually be
2204 used in the assembler code, after renumbering, should not be marked with 1
2207 Define this macro only if the target machine offers a way to optimize
2208 the treatment of leaf functions.
2211 @defmac LEAF_REG_REMAP (@var{regno})
2212 A C expression whose value is the register number to which @var{regno}
2213 should be renumbered, when a function is treated as a leaf function.
2215 If @var{regno} is a register number which should not appear in a leaf
2216 function before renumbering, then the expression should yield @minus{}1, which
2217 will cause the compiler to abort.
2219 Define this macro only if the target machine offers a way to optimize the
2220 treatment of leaf functions, and registers need to be renumbered to do
2224 @findex current_function_is_leaf
2225 @findex current_function_uses_only_leaf_regs
2226 @code{TARGET_ASM_FUNCTION_PROLOGUE} and
2227 @code{TARGET_ASM_FUNCTION_EPILOGUE} must usually treat leaf functions
2228 specially. They can test the C variable @code{current_function_is_leaf}
2229 which is nonzero for leaf functions. @code{current_function_is_leaf} is
2230 set prior to local register allocation and is valid for the remaining
2231 compiler passes. They can also test the C variable
2232 @code{current_function_uses_only_leaf_regs} which is nonzero for leaf
2233 functions which only use leaf registers.
2234 @code{current_function_uses_only_leaf_regs} is valid after all passes
2235 that modify the instructions have been run and is only useful if
2236 @code{LEAF_REGISTERS} is defined.
2237 @c changed this to fix overfull. ALSO: why the "it" at the beginning
2238 @c of the next paragraph?! --mew 2feb93
2240 @node Stack Registers
2241 @subsection Registers That Form a Stack
2243 There are special features to handle computers where some of the
2244 ``registers'' form a stack. Stack registers are normally written by
2245 pushing onto the stack, and are numbered relative to the top of the
2248 Currently, GCC can only handle one group of stack-like registers, and
2249 they must be consecutively numbered. Furthermore, the existing
2250 support for stack-like registers is specific to the 80387 floating
2251 point coprocessor. If you have a new architecture that uses
2252 stack-like registers, you will need to do substantial work on
2253 @file{reg-stack.c} and write your machine description to cooperate
2254 with it, as well as defining these macros.
2257 Define this if the machine has any stack-like registers.
2260 @defmac STACK_REG_COVER_CLASS
2261 This is a cover class containing the stack registers. Define this if
2262 the machine has any stack-like registers.
2265 @defmac FIRST_STACK_REG
2266 The number of the first stack-like register. This one is the top
2270 @defmac LAST_STACK_REG
2271 The number of the last stack-like register. This one is the bottom of
2275 @node Register Classes
2276 @section Register Classes
2277 @cindex register class definitions
2278 @cindex class definitions, register
2280 On many machines, the numbered registers are not all equivalent.
2281 For example, certain registers may not be allowed for indexed addressing;
2282 certain registers may not be allowed in some instructions. These machine
2283 restrictions are described to the compiler using @dfn{register classes}.
2285 You define a number of register classes, giving each one a name and saying
2286 which of the registers belong to it. Then you can specify register classes
2287 that are allowed as operands to particular instruction patterns.
2291 In general, each register will belong to several classes. In fact, one
2292 class must be named @code{ALL_REGS} and contain all the registers. Another
2293 class must be named @code{NO_REGS} and contain no registers. Often the
2294 union of two classes will be another class; however, this is not required.
2296 @findex GENERAL_REGS
2297 One of the classes must be named @code{GENERAL_REGS}. There is nothing
2298 terribly special about the name, but the operand constraint letters
2299 @samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is
2300 the same as @code{ALL_REGS}, just define it as a macro which expands
2303 Order the classes so that if class @var{x} is contained in class @var{y}
2304 then @var{x} has a lower class number than @var{y}.
2306 The way classes other than @code{GENERAL_REGS} are specified in operand
2307 constraints is through machine-dependent operand constraint letters.
2308 You can define such letters to correspond to various classes, then use
2309 them in operand constraints.
2311 You should define a class for the union of two classes whenever some
2312 instruction allows both classes. For example, if an instruction allows
2313 either a floating point (coprocessor) register or a general register for a
2314 certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS}
2315 which includes both of them. Otherwise you will get suboptimal code,
2316 or even internal compiler errors when reload cannot find a register in the
2317 the class computed via @code{reg_class_subunion}.
2319 You must also specify certain redundant information about the register
2320 classes: for each class, which classes contain it and which ones are
2321 contained in it; for each pair of classes, the largest class contained
2324 When a value occupying several consecutive registers is expected in a
2325 certain class, all the registers used must belong to that class.
2326 Therefore, register classes cannot be used to enforce a requirement for
2327 a register pair to start with an even-numbered register. The way to
2328 specify this requirement is with @code{HARD_REGNO_MODE_OK}.
2330 Register classes used for input-operands of bitwise-and or shift
2331 instructions have a special requirement: each such class must have, for
2332 each fixed-point machine mode, a subclass whose registers can transfer that
2333 mode to or from memory. For example, on some machines, the operations for
2334 single-byte values (@code{QImode}) are limited to certain registers. When
2335 this is so, each register class that is used in a bitwise-and or shift
2336 instruction must have a subclass consisting of registers from which
2337 single-byte values can be loaded or stored. This is so that
2338 @code{PREFERRED_RELOAD_CLASS} can always have a possible value to return.
2340 @deftp {Data type} {enum reg_class}
2341 An enumerated type that must be defined with all the register class names
2342 as enumerated values. @code{NO_REGS} must be first. @code{ALL_REGS}
2343 must be the last register class, followed by one more enumerated value,
2344 @code{LIM_REG_CLASSES}, which is not a register class but rather
2345 tells how many classes there are.
2347 Each register class has a number, which is the value of casting
2348 the class name to type @code{int}. The number serves as an index
2349 in many of the tables described below.
2352 @defmac N_REG_CLASSES
2353 The number of distinct register classes, defined as follows:
2356 #define N_REG_CLASSES (int) LIM_REG_CLASSES
2360 @defmac REG_CLASS_NAMES
2361 An initializer containing the names of the register classes as C string
2362 constants. These names are used in writing some of the debugging dumps.
2365 @defmac REG_CLASS_CONTENTS
2366 An initializer containing the contents of the register classes, as integers
2367 which are bit masks. The @var{n}th integer specifies the contents of class
2368 @var{n}. The way the integer @var{mask} is interpreted is that
2369 register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1.
2371 When the machine has more than 32 registers, an integer does not suffice.
2372 Then the integers are replaced by sub-initializers, braced groupings containing
2373 several integers. Each sub-initializer must be suitable as an initializer
2374 for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}.
2375 In this situation, the first integer in each sub-initializer corresponds to
2376 registers 0 through 31, the second integer to registers 32 through 63, and
2380 @defmac REGNO_REG_CLASS (@var{regno})
2381 A C expression whose value is a register class containing hard register
2382 @var{regno}. In general there is more than one such class; choose a class
2383 which is @dfn{minimal}, meaning that no smaller class also contains the
2387 @defmac BASE_REG_CLASS
2388 A macro whose definition is the name of the class to which a valid
2389 base register must belong. A base register is one used in an address
2390 which is the register value plus a displacement.
2393 @defmac MODE_BASE_REG_CLASS (@var{mode})
2394 This is a variation of the @code{BASE_REG_CLASS} macro which allows
2395 the selection of a base register in a mode dependent manner. If
2396 @var{mode} is VOIDmode then it should return the same value as
2397 @code{BASE_REG_CLASS}.
2400 @defmac MODE_BASE_REG_REG_CLASS (@var{mode})
2401 A C expression whose value is the register class to which a valid
2402 base register must belong in order to be used in a base plus index
2403 register address. You should define this macro if base plus index
2404 addresses have different requirements than other base register uses.
2407 @defmac MODE_CODE_BASE_REG_CLASS (@var{mode}, @var{outer_code}, @var{index_code})
2408 A C expression whose value is the register class to which a valid
2409 base register must belong. @var{outer_code} and @var{index_code} define the
2410 context in which the base register occurs. @var{outer_code} is the code of
2411 the immediately enclosing expression (@code{MEM} for the top level of an
2412 address, @code{ADDRESS} for something that occurs in an
2413 @code{address_operand}). @var{index_code} is the code of the corresponding
2414 index expression if @var{outer_code} is @code{PLUS}; @code{SCRATCH} otherwise.
2417 @defmac INDEX_REG_CLASS
2418 A macro whose definition is the name of the class to which a valid
2419 index register must belong. An index register is one used in an
2420 address where its value is either multiplied by a scale factor or
2421 added to another register (as well as added to a displacement).
2424 @defmac REGNO_OK_FOR_BASE_P (@var{num})
2425 A C expression which is nonzero if register number @var{num} is
2426 suitable for use as a base register in operand addresses.
2429 @defmac REGNO_MODE_OK_FOR_BASE_P (@var{num}, @var{mode})
2430 A C expression that is just like @code{REGNO_OK_FOR_BASE_P}, except that
2431 that expression may examine the mode of the memory reference in
2432 @var{mode}. You should define this macro if the mode of the memory
2433 reference affects whether a register may be used as a base register. If
2434 you define this macro, the compiler will use it instead of
2435 @code{REGNO_OK_FOR_BASE_P}. The mode may be @code{VOIDmode} for
2436 addresses that appear outside a @code{MEM}, i.e., as an
2437 @code{address_operand}.
2440 @defmac REGNO_MODE_OK_FOR_REG_BASE_P (@var{num}, @var{mode})
2441 A C expression which is nonzero if register number @var{num} is suitable for
2442 use as a base register in base plus index operand addresses, accessing
2443 memory in mode @var{mode}. It may be either a suitable hard register or a
2444 pseudo register that has been allocated such a hard register. You should
2445 define this macro if base plus index addresses have different requirements
2446 than other base register uses.
2448 Use of this macro is deprecated; please use the more general
2449 @code{REGNO_MODE_CODE_OK_FOR_BASE_P}.
2452 @defmac REGNO_MODE_CODE_OK_FOR_BASE_P (@var{num}, @var{mode}, @var{outer_code}, @var{index_code})
2453 A C expression that is just like @code{REGNO_MODE_OK_FOR_BASE_P}, except
2454 that that expression may examine the context in which the register
2455 appears in the memory reference. @var{outer_code} is the code of the
2456 immediately enclosing expression (@code{MEM} if at the top level of the
2457 address, @code{ADDRESS} for something that occurs in an
2458 @code{address_operand}). @var{index_code} is the code of the
2459 corresponding index expression if @var{outer_code} is @code{PLUS};
2460 @code{SCRATCH} otherwise. The mode may be @code{VOIDmode} for addresses
2461 that appear outside a @code{MEM}, i.e., as an @code{address_operand}.
2464 @defmac REGNO_OK_FOR_INDEX_P (@var{num})
2465 A C expression which is nonzero if register number @var{num} is
2466 suitable for use as an index register in operand addresses. It may be
2467 either a suitable hard register or a pseudo register that has been
2468 allocated such a hard register.
2470 The difference between an index register and a base register is that
2471 the index register may be scaled. If an address involves the sum of
2472 two registers, neither one of them scaled, then either one may be
2473 labeled the ``base'' and the other the ``index''; but whichever
2474 labeling is used must fit the machine's constraints of which registers
2475 may serve in each capacity. The compiler will try both labelings,
2476 looking for one that is valid, and will reload one or both registers
2477 only if neither labeling works.
2480 @hook TARGET_PREFERRED_RENAME_CLASS
2482 @hook TARGET_PREFERRED_RELOAD_CLASS
2483 A target hook that places additional restrictions on the register class
2484 to use when it is necessary to copy value @var{x} into a register in class
2485 @var{rclass}. The value is a register class; perhaps @var{rclass}, or perhaps
2486 another, smaller class.
2488 The default version of this hook always returns value of @code{rclass} argument.
2490 Sometimes returning a more restrictive class makes better code. For
2491 example, on the 68000, when @var{x} is an integer constant that is in range
2492 for a @samp{moveq} instruction, the value of this macro is always
2493 @code{DATA_REGS} as long as @var{rclass} includes the data registers.
2494 Requiring a data register guarantees that a @samp{moveq} will be used.
2496 One case where @code{TARGET_PREFERRED_RELOAD_CLASS} must not return
2497 @var{rclass} is if @var{x} is a legitimate constant which cannot be
2498 loaded into some register class. By returning @code{NO_REGS} you can
2499 force @var{x} into a memory location. For example, rs6000 can load
2500 immediate values into general-purpose registers, but does not have an
2501 instruction for loading an immediate value into a floating-point
2502 register, so @code{TARGET_PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when
2503 @var{x} is a floating-point constant. If the constant can't be loaded
2504 into any kind of register, code generation will be better if
2505 @code{LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead
2506 of using @code{TARGET_PREFERRED_RELOAD_CLASS}.
2508 If an insn has pseudos in it after register allocation, reload will go
2509 through the alternatives and call repeatedly @code{TARGET_PREFERRED_RELOAD_CLASS}
2510 to find the best one. Returning @code{NO_REGS}, in this case, makes
2511 reload add a @code{!} in front of the constraint: the x86 back-end uses
2512 this feature to discourage usage of 387 registers when math is done in
2513 the SSE registers (and vice versa).
2516 @defmac PREFERRED_RELOAD_CLASS (@var{x}, @var{class})
2517 A C expression that places additional restrictions on the register class
2518 to use when it is necessary to copy value @var{x} into a register in class
2519 @var{class}. The value is a register class; perhaps @var{class}, or perhaps
2520 another, smaller class. On many machines, the following definition is
2524 #define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS
2527 Sometimes returning a more restrictive class makes better code. For
2528 example, on the 68000, when @var{x} is an integer constant that is in range
2529 for a @samp{moveq} instruction, the value of this macro is always
2530 @code{DATA_REGS} as long as @var{class} includes the data registers.
2531 Requiring a data register guarantees that a @samp{moveq} will be used.
2533 One case where @code{PREFERRED_RELOAD_CLASS} must not return
2534 @var{class} is if @var{x} is a legitimate constant which cannot be
2535 loaded into some register class. By returning @code{NO_REGS} you can
2536 force @var{x} into a memory location. For example, rs6000 can load
2537 immediate values into general-purpose registers, but does not have an
2538 instruction for loading an immediate value into a floating-point
2539 register, so @code{PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when
2540 @var{x} is a floating-point constant. If the constant can't be loaded
2541 into any kind of register, code generation will be better if
2542 @code{LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead
2543 of using @code{PREFERRED_RELOAD_CLASS}.
2545 If an insn has pseudos in it after register allocation, reload will go
2546 through the alternatives and call repeatedly @code{PREFERRED_RELOAD_CLASS}
2547 to find the best one. Returning @code{NO_REGS}, in this case, makes
2548 reload add a @code{!} in front of the constraint: the x86 back-end uses
2549 this feature to discourage usage of 387 registers when math is done in
2550 the SSE registers (and vice versa).
2553 @defmac PREFERRED_OUTPUT_RELOAD_CLASS (@var{x}, @var{class})
2554 Like @code{PREFERRED_RELOAD_CLASS}, but for output reloads instead of
2555 input reloads. If you don't define this macro, the default is to use
2556 @var{class}, unchanged.
2558 You can also use @code{PREFERRED_OUTPUT_RELOAD_CLASS} to discourage
2559 reload from using some alternatives, like @code{PREFERRED_RELOAD_CLASS}.
2562 @hook TARGET_PREFERRED_OUTPUT_RELOAD_CLASS
2563 Like @code{TARGET_PREFERRED_RELOAD_CLASS}, but for output reloads instead of
2566 The default version of this hook always returns value of @code{rclass}
2569 You can also use @code{TARGET_PREFERRED_OUTPUT_RELOAD_CLASS} to discourage
2570 reload from using some alternatives, like @code{TARGET_PREFERRED_RELOAD_CLASS}.
2573 @defmac LIMIT_RELOAD_CLASS (@var{mode}, @var{class})
2574 A C expression that places additional restrictions on the register class
2575 to use when it is necessary to be able to hold a value of mode
2576 @var{mode} in a reload register for which class @var{class} would
2579 Unlike @code{PREFERRED_RELOAD_CLASS}, this macro should be used when
2580 there are certain modes that simply can't go in certain reload classes.
2582 The value is a register class; perhaps @var{class}, or perhaps another,
2585 Don't define this macro unless the target machine has limitations which
2586 require the macro to do something nontrivial.
2589 @hook TARGET_SECONDARY_RELOAD
2590 Many machines have some registers that cannot be copied directly to or
2591 from memory or even from other types of registers. An example is the
2592 @samp{MQ} register, which on most machines, can only be copied to or
2593 from general registers, but not memory. Below, we shall be using the
2594 term 'intermediate register' when a move operation cannot be performed
2595 directly, but has to be done by copying the source into the intermediate
2596 register first, and then copying the intermediate register to the
2597 destination. An intermediate register always has the same mode as
2598 source and destination. Since it holds the actual value being copied,
2599 reload might apply optimizations to re-use an intermediate register
2600 and eliding the copy from the source when it can determine that the
2601 intermediate register still holds the required value.
2603 Another kind of secondary reload is required on some machines which
2604 allow copying all registers to and from memory, but require a scratch
2605 register for stores to some memory locations (e.g., those with symbolic
2606 address on the RT, and those with certain symbolic address on the SPARC
2607 when compiling PIC)@. Scratch registers need not have the same mode
2608 as the value being copied, and usually hold a different value than
2609 that being copied. Special patterns in the md file are needed to
2610 describe how the copy is performed with the help of the scratch register;
2611 these patterns also describe the number, register class(es) and mode(s)
2612 of the scratch register(s).
2614 In some cases, both an intermediate and a scratch register are required.
2616 For input reloads, this target hook is called with nonzero @var{in_p},
2617 and @var{x} is an rtx that needs to be copied to a register of class
2618 @var{reload_class} in @var{reload_mode}. For output reloads, this target
2619 hook is called with zero @var{in_p}, and a register of class @var{reload_class}
2620 needs to be copied to rtx @var{x} in @var{reload_mode}.
2622 If copying a register of @var{reload_class} from/to @var{x} requires
2623 an intermediate register, the hook @code{secondary_reload} should
2624 return the register class required for this intermediate register.
2625 If no intermediate register is required, it should return NO_REGS.
2626 If more than one intermediate register is required, describe the one
2627 that is closest in the copy chain to the reload register.
2629 If scratch registers are needed, you also have to describe how to
2630 perform the copy from/to the reload register to/from this
2631 closest intermediate register. Or if no intermediate register is
2632 required, but still a scratch register is needed, describe the
2633 copy from/to the reload register to/from the reload operand @var{x}.
2635 You do this by setting @code{sri->icode} to the instruction code of a pattern
2636 in the md file which performs the move. Operands 0 and 1 are the output
2637 and input of this copy, respectively. Operands from operand 2 onward are
2638 for scratch operands. These scratch operands must have a mode, and a
2639 single-register-class
2640 @c [later: or memory]
2643 When an intermediate register is used, the @code{secondary_reload}
2644 hook will be called again to determine how to copy the intermediate
2645 register to/from the reload operand @var{x}, so your hook must also
2646 have code to handle the register class of the intermediate operand.
2648 @c [For later: maybe we'll allow multi-alternative reload patterns -
2649 @c the port maintainer could name a mov<mode> pattern that has clobbers -
2650 @c and match the constraints of input and output to determine the required
2651 @c alternative. A restriction would be that constraints used to match
2652 @c against reloads registers would have to be written as register class
2653 @c constraints, or we need a new target macro / hook that tells us if an
2654 @c arbitrary constraint can match an unknown register of a given class.
2655 @c Such a macro / hook would also be useful in other places.]
2658 @var{x} might be a pseudo-register or a @code{subreg} of a
2659 pseudo-register, which could either be in a hard register or in memory.
2660 Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is
2661 in memory and the hard register number if it is in a register.
2663 Scratch operands in memory (constraint @code{"=m"} / @code{"=&m"}) are
2664 currently not supported. For the time being, you will have to continue
2665 to use @code{SECONDARY_MEMORY_NEEDED} for that purpose.
2667 @code{copy_cost} also uses this target hook to find out how values are
2668 copied. If you want it to include some extra cost for the need to allocate
2669 (a) scratch register(s), set @code{sri->extra_cost} to the additional cost.
2670 Or if two dependent moves are supposed to have a lower cost than the sum
2671 of the individual moves due to expected fortuitous scheduling and/or special
2672 forwarding logic, you can set @code{sri->extra_cost} to a negative amount.
2675 @defmac SECONDARY_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2676 @defmacx SECONDARY_INPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2677 @defmacx SECONDARY_OUTPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2678 These macros are obsolete, new ports should use the target hook
2679 @code{TARGET_SECONDARY_RELOAD} instead.
2681 These are obsolete macros, replaced by the @code{TARGET_SECONDARY_RELOAD}
2682 target hook. Older ports still define these macros to indicate to the
2683 reload phase that it may
2684 need to allocate at least one register for a reload in addition to the
2685 register to contain the data. Specifically, if copying @var{x} to a
2686 register @var{class} in @var{mode} requires an intermediate register,
2687 you were supposed to define @code{SECONDARY_INPUT_RELOAD_CLASS} to return the
2688 largest register class all of whose registers can be used as
2689 intermediate registers or scratch registers.
2691 If copying a register @var{class} in @var{mode} to @var{x} requires an
2692 intermediate or scratch register, @code{SECONDARY_OUTPUT_RELOAD_CLASS}
2693 was supposed to be defined be defined to return the largest register
2694 class required. If the
2695 requirements for input and output reloads were the same, the macro
2696 @code{SECONDARY_RELOAD_CLASS} should have been used instead of defining both
2699 The values returned by these macros are often @code{GENERAL_REGS}.
2700 Return @code{NO_REGS} if no spare register is needed; i.e., if @var{x}
2701 can be directly copied to or from a register of @var{class} in
2702 @var{mode} without requiring a scratch register. Do not define this
2703 macro if it would always return @code{NO_REGS}.
2705 If a scratch register is required (either with or without an
2706 intermediate register), you were supposed to define patterns for
2707 @samp{reload_in@var{m}} or @samp{reload_out@var{m}}, as required
2708 (@pxref{Standard Names}. These patterns, which were normally
2709 implemented with a @code{define_expand}, should be similar to the
2710 @samp{mov@var{m}} patterns, except that operand 2 is the scratch
2713 These patterns need constraints for the reload register and scratch
2715 contain a single register class. If the original reload register (whose
2716 class is @var{class}) can meet the constraint given in the pattern, the
2717 value returned by these macros is used for the class of the scratch
2718 register. Otherwise, two additional reload registers are required.
2719 Their classes are obtained from the constraints in the insn pattern.
2721 @var{x} might be a pseudo-register or a @code{subreg} of a
2722 pseudo-register, which could either be in a hard register or in memory.
2723 Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is
2724 in memory and the hard register number if it is in a register.
2726 These macros should not be used in the case where a particular class of
2727 registers can only be copied to memory and not to another class of
2728 registers. In that case, secondary reload registers are not needed and
2729 would not be helpful. Instead, a stack location must be used to perform
2730 the copy and the @code{mov@var{m}} pattern should use memory as an
2731 intermediate storage. This case often occurs between floating-point and
2735 @defmac SECONDARY_MEMORY_NEEDED (@var{class1}, @var{class2}, @var{m})
2736 Certain machines have the property that some registers cannot be copied
2737 to some other registers without using memory. Define this macro on
2738 those machines to be a C expression that is nonzero if objects of mode
2739 @var{m} in registers of @var{class1} can only be copied to registers of
2740 class @var{class2} by storing a register of @var{class1} into memory
2741 and loading that memory location into a register of @var{class2}.
2743 Do not define this macro if its value would always be zero.
2746 @defmac SECONDARY_MEMORY_NEEDED_RTX (@var{mode})
2747 Normally when @code{SECONDARY_MEMORY_NEEDED} is defined, the compiler
2748 allocates a stack slot for a memory location needed for register copies.
2749 If this macro is defined, the compiler instead uses the memory location
2750 defined by this macro.
2752 Do not define this macro if you do not define
2753 @code{SECONDARY_MEMORY_NEEDED}.
2756 @defmac SECONDARY_MEMORY_NEEDED_MODE (@var{mode})
2757 When the compiler needs a secondary memory location to copy between two
2758 registers of mode @var{mode}, it normally allocates sufficient memory to
2759 hold a quantity of @code{BITS_PER_WORD} bits and performs the store and
2760 load operations in a mode that many bits wide and whose class is the
2761 same as that of @var{mode}.
2763 This is right thing to do on most machines because it ensures that all
2764 bits of the register are copied and prevents accesses to the registers
2765 in a narrower mode, which some machines prohibit for floating-point
2768 However, this default behavior is not correct on some machines, such as
2769 the DEC Alpha, that store short integers in floating-point registers
2770 differently than in integer registers. On those machines, the default
2771 widening will not work correctly and you must define this macro to
2772 suppress that widening in some cases. See the file @file{alpha.h} for
2775 Do not define this macro if you do not define
2776 @code{SECONDARY_MEMORY_NEEDED} or if widening @var{mode} to a mode that
2777 is @code{BITS_PER_WORD} bits wide is correct for your machine.
2780 @hook TARGET_CLASS_LIKELY_SPILLED_P
2781 A target hook which returns @code{true} if pseudos that have been assigned
2782 to registers of class @var{rclass} would likely be spilled because
2783 registers of @var{rclass} are needed for spill registers.
2785 The default version of this target hook returns @code{true} if @var{rclass}
2786 has exactly one register and @code{false} otherwise. On most machines, this
2787 default should be used. Only use this target hook to some other expression
2788 if pseudos allocated by @file{local-alloc.c} end up in memory because their
2789 hard registers were needed for spill registers. If this target hook returns
2790 @code{false} for those classes, those pseudos will only be allocated by
2791 @file{global.c}, which knows how to reallocate the pseudo to another
2792 register. If there would not be another register available for reallocation,
2793 you should not change the implementation of this target hook since
2794 the only effect of such implementation would be to slow down register
2798 @defmac CLASS_MAX_NREGS (@var{class}, @var{mode})
2799 A C expression for the maximum number of consecutive registers
2800 of class @var{class} needed to hold a value of mode @var{mode}.
2802 This is closely related to the macro @code{HARD_REGNO_NREGS}. In fact,
2803 the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})}
2804 should be the maximum value of @code{HARD_REGNO_NREGS (@var{regno},
2805 @var{mode})} for all @var{regno} values in the class @var{class}.
2807 This macro helps control the handling of multiple-word values
2811 @defmac CANNOT_CHANGE_MODE_CLASS (@var{from}, @var{to}, @var{class})
2812 If defined, a C expression that returns nonzero for a @var{class} for which
2813 a change from mode @var{from} to mode @var{to} is invalid.
2815 For the example, loading 32-bit integer or floating-point objects into
2816 floating-point registers on the Alpha extends them to 64 bits.
2817 Therefore loading a 64-bit object and then storing it as a 32-bit object
2818 does not store the low-order 32 bits, as would be the case for a normal
2819 register. Therefore, @file{alpha.h} defines @code{CANNOT_CHANGE_MODE_CLASS}
2823 #define CANNOT_CHANGE_MODE_CLASS(FROM, TO, CLASS) \
2824 (GET_MODE_SIZE (FROM) != GET_MODE_SIZE (TO) \
2825 ? reg_classes_intersect_p (FLOAT_REGS, (CLASS)) : 0)
2829 @node Old Constraints
2830 @section Obsolete Macros for Defining Constraints
2831 @cindex defining constraints, obsolete method
2832 @cindex constraints, defining, obsolete method
2834 Machine-specific constraints can be defined with these macros instead
2835 of the machine description constructs described in @ref{Define
2836 Constraints}. This mechanism is obsolete. New ports should not use
2837 it; old ports should convert to the new mechanism.
2839 @defmac CONSTRAINT_LEN (@var{char}, @var{str})
2840 For the constraint at the start of @var{str}, which starts with the letter
2841 @var{c}, return the length. This allows you to have register class /
2842 constant / extra constraints that are longer than a single letter;
2843 you don't need to define this macro if you can do with single-letter
2844 constraints only. The definition of this macro should use
2845 DEFAULT_CONSTRAINT_LEN for all the characters that you don't want
2846 to handle specially.
2847 There are some sanity checks in genoutput.c that check the constraint lengths
2848 for the md file, so you can also use this macro to help you while you are
2849 transitioning from a byzantine single-letter-constraint scheme: when you
2850 return a negative length for a constraint you want to re-use, genoutput
2851 will complain about every instance where it is used in the md file.
2854 @defmac REG_CLASS_FROM_LETTER (@var{char})
2855 A C expression which defines the machine-dependent operand constraint
2856 letters for register classes. If @var{char} is such a letter, the
2857 value should be the register class corresponding to it. Otherwise,
2858 the value should be @code{NO_REGS}. The register letter @samp{r},
2859 corresponding to class @code{GENERAL_REGS}, will not be passed
2860 to this macro; you do not need to handle it.
2863 @defmac REG_CLASS_FROM_CONSTRAINT (@var{char}, @var{str})
2864 Like @code{REG_CLASS_FROM_LETTER}, but you also get the constraint string
2865 passed in @var{str}, so that you can use suffixes to distinguish between
2869 @defmac CONST_OK_FOR_LETTER_P (@var{value}, @var{c})
2870 A C expression that defines the machine-dependent operand constraint
2871 letters (@samp{I}, @samp{J}, @samp{K}, @dots{} @samp{P}) that specify
2872 particular ranges of integer values. If @var{c} is one of those
2873 letters, the expression should check that @var{value}, an integer, is in
2874 the appropriate range and return 1 if so, 0 otherwise. If @var{c} is
2875 not one of those letters, the value should be 0 regardless of
2879 @defmac CONST_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str})
2880 Like @code{CONST_OK_FOR_LETTER_P}, but you also get the constraint
2881 string passed in @var{str}, so that you can use suffixes to distinguish
2882 between different variants.
2885 @defmac CONST_DOUBLE_OK_FOR_LETTER_P (@var{value}, @var{c})
2886 A C expression that defines the machine-dependent operand constraint
2887 letters that specify particular ranges of @code{const_double} values
2888 (@samp{G} or @samp{H}).
2890 If @var{c} is one of those letters, the expression should check that
2891 @var{value}, an RTX of code @code{const_double}, is in the appropriate
2892 range and return 1 if so, 0 otherwise. If @var{c} is not one of those
2893 letters, the value should be 0 regardless of @var{value}.
2895 @code{const_double} is used for all floating-point constants and for
2896 @code{DImode} fixed-point constants. A given letter can accept either
2897 or both kinds of values. It can use @code{GET_MODE} to distinguish
2898 between these kinds.
2901 @defmac CONST_DOUBLE_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str})
2902 Like @code{CONST_DOUBLE_OK_FOR_LETTER_P}, but you also get the constraint
2903 string passed in @var{str}, so that you can use suffixes to distinguish
2904 between different variants.
2907 @defmac EXTRA_CONSTRAINT (@var{value}, @var{c})
2908 A C expression that defines the optional machine-dependent constraint
2909 letters that can be used to segregate specific types of operands, usually
2910 memory references, for the target machine. Any letter that is not
2911 elsewhere defined and not matched by @code{REG_CLASS_FROM_LETTER} /
2912 @code{REG_CLASS_FROM_CONSTRAINT}
2913 may be used. Normally this macro will not be defined.
2915 If it is required for a particular target machine, it should return 1
2916 if @var{value} corresponds to the operand type represented by the
2917 constraint letter @var{c}. If @var{c} is not defined as an extra
2918 constraint, the value returned should be 0 regardless of @var{value}.
2920 For example, on the ROMP, load instructions cannot have their output
2921 in r0 if the memory reference contains a symbolic address. Constraint
2922 letter @samp{Q} is defined as representing a memory address that does
2923 @emph{not} contain a symbolic address. An alternative is specified with
2924 a @samp{Q} constraint on the input and @samp{r} on the output. The next
2925 alternative specifies @samp{m} on the input and a register class that
2926 does not include r0 on the output.
2929 @defmac EXTRA_CONSTRAINT_STR (@var{value}, @var{c}, @var{str})
2930 Like @code{EXTRA_CONSTRAINT}, but you also get the constraint string passed
2931 in @var{str}, so that you can use suffixes to distinguish between different
2935 @defmac EXTRA_MEMORY_CONSTRAINT (@var{c}, @var{str})
2936 A C expression that defines the optional machine-dependent constraint
2937 letters, amongst those accepted by @code{EXTRA_CONSTRAINT}, that should
2938 be treated like memory constraints by the reload pass.
2940 It should return 1 if the operand type represented by the constraint
2941 at the start of @var{str}, the first letter of which is the letter @var{c},
2942 comprises a subset of all memory references including
2943 all those whose address is simply a base register. This allows the reload
2944 pass to reload an operand, if it does not directly correspond to the operand
2945 type of @var{c}, by copying its address into a base register.
2947 For example, on the S/390, some instructions do not accept arbitrary
2948 memory references, but only those that do not make use of an index
2949 register. The constraint letter @samp{Q} is defined via
2950 @code{EXTRA_CONSTRAINT} as representing a memory address of this type.
2951 If the letter @samp{Q} is marked as @code{EXTRA_MEMORY_CONSTRAINT},
2952 a @samp{Q} constraint can handle any memory operand, because the
2953 reload pass knows it can be reloaded by copying the memory address
2954 into a base register if required. This is analogous to the way
2955 an @samp{o} constraint can handle any memory operand.
2958 @defmac EXTRA_ADDRESS_CONSTRAINT (@var{c}, @var{str})
2959 A C expression that defines the optional machine-dependent constraint
2960 letters, amongst those accepted by @code{EXTRA_CONSTRAINT} /
2961 @code{EXTRA_CONSTRAINT_STR}, that should
2962 be treated like address constraints by the reload pass.
2964 It should return 1 if the operand type represented by the constraint
2965 at the start of @var{str}, which starts with the letter @var{c}, comprises
2966 a subset of all memory addresses including
2967 all those that consist of just a base register. This allows the reload
2968 pass to reload an operand, if it does not directly correspond to the operand
2969 type of @var{str}, by copying it into a base register.
2971 Any constraint marked as @code{EXTRA_ADDRESS_CONSTRAINT} can only
2972 be used with the @code{address_operand} predicate. It is treated
2973 analogously to the @samp{p} constraint.
2976 @node Stack and Calling
2977 @section Stack Layout and Calling Conventions
2978 @cindex calling conventions
2980 @c prevent bad page break with this line
2981 This describes the stack layout and calling conventions.
2985 * Exception Handling::
2990 * Register Arguments::
2992 * Aggregate Return::
2997 * Stack Smashing Protection::
3001 @subsection Basic Stack Layout
3002 @cindex stack frame layout
3003 @cindex frame layout
3005 @c prevent bad page break with this line
3006 Here is the basic stack layout.
3008 @defmac STACK_GROWS_DOWNWARD
3009 Define this macro if pushing a word onto the stack moves the stack
3010 pointer to a smaller address.
3012 When we say, ``define this macro if @dots{}'', it means that the
3013 compiler checks this macro only with @code{#ifdef} so the precise
3014 definition used does not matter.
3017 @defmac STACK_PUSH_CODE
3018 This macro defines the operation used when something is pushed
3019 on the stack. In RTL, a push operation will be
3020 @code{(set (mem (STACK_PUSH_CODE (reg sp))) @dots{})}
3022 The choices are @code{PRE_DEC}, @code{POST_DEC}, @code{PRE_INC},
3023 and @code{POST_INC}. Which of these is correct depends on
3024 the stack direction and on whether the stack pointer points
3025 to the last item on the stack or whether it points to the
3026 space for the next item on the stack.
3028 The default is @code{PRE_DEC} when @code{STACK_GROWS_DOWNWARD} is
3029 defined, which is almost always right, and @code{PRE_INC} otherwise,
3030 which is often wrong.
3033 @defmac FRAME_GROWS_DOWNWARD
3034 Define this macro to nonzero value if the addresses of local variable slots
3035 are at negative offsets from the frame pointer.
3038 @defmac ARGS_GROW_DOWNWARD
3039 Define this macro if successive arguments to a function occupy decreasing
3040 addresses on the stack.
3043 @defmac STARTING_FRAME_OFFSET
3044 Offset from the frame pointer to the first local variable slot to be allocated.
3046 If @code{FRAME_GROWS_DOWNWARD}, find the next slot's offset by
3047 subtracting the first slot's length from @code{STARTING_FRAME_OFFSET}.
3048 Otherwise, it is found by adding the length of the first slot to the
3049 value @code{STARTING_FRAME_OFFSET}.
3050 @c i'm not sure if the above is still correct.. had to change it to get
3051 @c rid of an overfull. --mew 2feb93
3054 @defmac STACK_ALIGNMENT_NEEDED
3055 Define to zero to disable final alignment of the stack during reload.
3056 The nonzero default for this macro is suitable for most ports.
3058 On ports where @code{STARTING_FRAME_OFFSET} is nonzero or where there
3059 is a register save block following the local block that doesn't require
3060 alignment to @code{STACK_BOUNDARY}, it may be beneficial to disable
3061 stack alignment and do it in the backend.
3064 @defmac STACK_POINTER_OFFSET
3065 Offset from the stack pointer register to the first location at which
3066 outgoing arguments are placed. If not specified, the default value of
3067 zero is used. This is the proper value for most machines.
3069 If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
3070 the first location at which outgoing arguments are placed.
3073 @defmac FIRST_PARM_OFFSET (@var{fundecl})
3074 Offset from the argument pointer register to the first argument's
3075 address. On some machines it may depend on the data type of the
3078 If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
3079 the first argument's address.
3082 @defmac STACK_DYNAMIC_OFFSET (@var{fundecl})
3083 Offset from the stack pointer register to an item dynamically allocated
3084 on the stack, e.g., by @code{alloca}.
3086 The default value for this macro is @code{STACK_POINTER_OFFSET} plus the
3087 length of the outgoing arguments. The default is correct for most
3088 machines. See @file{function.c} for details.
3091 @defmac INITIAL_FRAME_ADDRESS_RTX
3092 A C expression whose value is RTL representing the address of the initial
3093 stack frame. This address is passed to @code{RETURN_ADDR_RTX} and
3094 @code{DYNAMIC_CHAIN_ADDRESS}. If you don't define this macro, a reasonable
3095 default value will be used. Define this macro in order to make frame pointer
3096 elimination work in the presence of @code{__builtin_frame_address (count)} and
3097 @code{__builtin_return_address (count)} for @code{count} not equal to zero.
3100 @defmac DYNAMIC_CHAIN_ADDRESS (@var{frameaddr})
3101 A C expression whose value is RTL representing the address in a stack
3102 frame where the pointer to the caller's frame is stored. Assume that
3103 @var{frameaddr} is an RTL expression for the address of the stack frame
3106 If you don't define this macro, the default is to return the value
3107 of @var{frameaddr}---that is, the stack frame address is also the
3108 address of the stack word that points to the previous frame.
3111 @defmac SETUP_FRAME_ADDRESSES
3112 If defined, a C expression that produces the machine-specific code to
3113 setup the stack so that arbitrary frames can be accessed. For example,
3114 on the SPARC, we must flush all of the register windows to the stack
3115 before we can access arbitrary stack frames. You will seldom need to
3119 @hook TARGET_BUILTIN_SETJMP_FRAME_VALUE
3120 This target hook should return an rtx that is used to store
3121 the address of the current frame into the built in @code{setjmp} buffer.
3122 The default value, @code{virtual_stack_vars_rtx}, is correct for most
3123 machines. One reason you may need to define this target hook is if
3124 @code{hard_frame_pointer_rtx} is the appropriate value on your machine.
3127 @defmac FRAME_ADDR_RTX (@var{frameaddr})
3128 A C expression whose value is RTL representing the value of the frame
3129 address for the current frame. @var{frameaddr} is the frame pointer
3130 of the current frame. This is used for __builtin_frame_address.
3131 You need only define this macro if the frame address is not the same
3132 as the frame pointer. Most machines do not need to define it.
3135 @defmac RETURN_ADDR_RTX (@var{count}, @var{frameaddr})
3136 A C expression whose value is RTL representing the value of the return
3137 address for the frame @var{count} steps up from the current frame, after
3138 the prologue. @var{frameaddr} is the frame pointer of the @var{count}
3139 frame, or the frame pointer of the @var{count} @minus{} 1 frame if
3140 @code{RETURN_ADDR_IN_PREVIOUS_FRAME} is defined.
3142 The value of the expression must always be the correct address when
3143 @var{count} is zero, but may be @code{NULL_RTX} if there is no way to
3144 determine the return address of other frames.
3147 @defmac RETURN_ADDR_IN_PREVIOUS_FRAME
3148 Define this if the return address of a particular stack frame is accessed
3149 from the frame pointer of the previous stack frame.
3152 @defmac INCOMING_RETURN_ADDR_RTX
3153 A C expression whose value is RTL representing the location of the
3154 incoming return address at the beginning of any function, before the
3155 prologue. This RTL is either a @code{REG}, indicating that the return
3156 value is saved in @samp{REG}, or a @code{MEM} representing a location in
3159 You only need to define this macro if you want to support call frame
3160 debugging information like that provided by DWARF 2.
3162 If this RTL is a @code{REG}, you should also define
3163 @code{DWARF_FRAME_RETURN_COLUMN} to @code{DWARF_FRAME_REGNUM (REGNO)}.
3166 @defmac DWARF_ALT_FRAME_RETURN_COLUMN
3167 A C expression whose value is an integer giving a DWARF 2 column
3168 number that may be used as an alternative return column. The column
3169 must not correspond to any gcc hard register (that is, it must not
3170 be in the range of @code{DWARF_FRAME_REGNUM}).
3172 This macro can be useful if @code{DWARF_FRAME_RETURN_COLUMN} is set to a
3173 general register, but an alternative column needs to be used for signal
3174 frames. Some targets have also used different frame return columns
3178 @defmac DWARF_ZERO_REG
3179 A C expression whose value is an integer giving a DWARF 2 register
3180 number that is considered to always have the value zero. This should
3181 only be defined if the target has an architected zero register, and
3182 someone decided it was a good idea to use that register number to
3183 terminate the stack backtrace. New ports should avoid this.
3186 @hook TARGET_DWARF_HANDLE_FRAME_UNSPEC
3187 This target hook allows the backend to emit frame-related insns that
3188 contain UNSPECs or UNSPEC_VOLATILEs. The DWARF 2 call frame debugging
3189 info engine will invoke it on insns of the form
3191 (set (reg) (unspec [@dots{}] UNSPEC_INDEX))
3195 (set (reg) (unspec_volatile [@dots{}] UNSPECV_INDEX)).
3197 to let the backend emit the call frame instructions. @var{label} is
3198 the CFI label attached to the insn, @var{pattern} is the pattern of
3199 the insn and @var{index} is @code{UNSPEC_INDEX} or @code{UNSPECV_INDEX}.
3202 @defmac INCOMING_FRAME_SP_OFFSET
3203 A C expression whose value is an integer giving the offset, in bytes,
3204 from the value of the stack pointer register to the top of the stack
3205 frame at the beginning of any function, before the prologue. The top of
3206 the frame is defined to be the value of the stack pointer in the
3207 previous frame, just before the call instruction.
3209 You only need to define this macro if you want to support call frame
3210 debugging information like that provided by DWARF 2.
3213 @defmac ARG_POINTER_CFA_OFFSET (@var{fundecl})
3214 A C expression whose value is an integer giving the offset, in bytes,
3215 from the argument pointer to the canonical frame address (cfa). The
3216 final value should coincide with that calculated by
3217 @code{INCOMING_FRAME_SP_OFFSET}. Which is unfortunately not usable
3218 during virtual register instantiation.
3220 The default value for this macro is
3221 @code{FIRST_PARM_OFFSET (fundecl) + crtl->args.pretend_args_size},
3222 which is correct for most machines; in general, the arguments are found
3223 immediately before the stack frame. Note that this is not the case on
3224 some targets that save registers into the caller's frame, such as SPARC
3225 and rs6000, and so such targets need to define this macro.
3227 You only need to define this macro if the default is incorrect, and you
3228 want to support call frame debugging information like that provided by
3232 @defmac FRAME_POINTER_CFA_OFFSET (@var{fundecl})
3233 If defined, a C expression whose value is an integer giving the offset
3234 in bytes from the frame pointer to the canonical frame address (cfa).
3235 The final value should coincide with that calculated by
3236 @code{INCOMING_FRAME_SP_OFFSET}.
3238 Normally the CFA is calculated as an offset from the argument pointer,
3239 via @code{ARG_POINTER_CFA_OFFSET}, but if the argument pointer is
3240 variable due to the ABI, this may not be possible. If this macro is
3241 defined, it implies that the virtual register instantiation should be
3242 based on the frame pointer instead of the argument pointer. Only one
3243 of @code{FRAME_POINTER_CFA_OFFSET} and @code{ARG_POINTER_CFA_OFFSET}
3247 @defmac CFA_FRAME_BASE_OFFSET (@var{fundecl})
3248 If defined, a C expression whose value is an integer giving the offset
3249 in bytes from the canonical frame address (cfa) to the frame base used
3250 in DWARF 2 debug information. The default is zero. A different value
3251 may reduce the size of debug information on some ports.
3254 @node Exception Handling
3255 @subsection Exception Handling Support
3256 @cindex exception handling
3258 @defmac EH_RETURN_DATA_REGNO (@var{N})
3259 A C expression whose value is the @var{N}th register number used for
3260 data by exception handlers, or @code{INVALID_REGNUM} if fewer than
3261 @var{N} registers are usable.
3263 The exception handling library routines communicate with the exception
3264 handlers via a set of agreed upon registers. Ideally these registers
3265 should be call-clobbered; it is possible to use call-saved registers,
3266 but may negatively impact code size. The target must support at least
3267 2 data registers, but should define 4 if there are enough free registers.
3269 You must define this macro if you want to support call frame exception
3270 handling like that provided by DWARF 2.
3273 @defmac EH_RETURN_STACKADJ_RTX
3274 A C expression whose value is RTL representing a location in which
3275 to store a stack adjustment to be applied before function return.
3276 This is used to unwind the stack to an exception handler's call frame.
3277 It will be assigned zero on code paths that return normally.
3279 Typically this is a call-clobbered hard register that is otherwise
3280 untouched by the epilogue, but could also be a stack slot.
3282 Do not define this macro if the stack pointer is saved and restored
3283 by the regular prolog and epilog code in the call frame itself; in
3284 this case, the exception handling library routines will update the
3285 stack location to be restored in place. Otherwise, you must define
3286 this macro if you want to support call frame exception handling like
3287 that provided by DWARF 2.
3290 @defmac EH_RETURN_HANDLER_RTX
3291 A C expression whose value is RTL representing a location in which
3292 to store the address of an exception handler to which we should
3293 return. It will not be assigned on code paths that return normally.
3295 Typically this is the location in the call frame at which the normal
3296 return address is stored. For targets that return by popping an
3297 address off the stack, this might be a memory address just below
3298 the @emph{target} call frame rather than inside the current call
3299 frame. If defined, @code{EH_RETURN_STACKADJ_RTX} will have already
3300 been assigned, so it may be used to calculate the location of the
3303 Some targets have more complex requirements than storing to an
3304 address calculable during initial code generation. In that case
3305 the @code{eh_return} instruction pattern should be used instead.
3307 If you want to support call frame exception handling, you must
3308 define either this macro or the @code{eh_return} instruction pattern.
3311 @defmac RETURN_ADDR_OFFSET
3312 If defined, an integer-valued C expression for which rtl will be generated
3313 to add it to the exception handler address before it is searched in the
3314 exception handling tables, and to subtract it again from the address before
3315 using it to return to the exception handler.
3318 @defmac ASM_PREFERRED_EH_DATA_FORMAT (@var{code}, @var{global})
3319 This macro chooses the encoding of pointers embedded in the exception
3320 handling sections. If at all possible, this should be defined such
3321 that the exception handling section will not require dynamic relocations,
3322 and so may be read-only.
3324 @var{code} is 0 for data, 1 for code labels, 2 for function pointers.
3325 @var{global} is true if the symbol may be affected by dynamic relocations.
3326 The macro should return a combination of the @code{DW_EH_PE_*} defines
3327 as found in @file{dwarf2.h}.
3329 If this macro is not defined, pointers will not be encoded but
3330 represented directly.
3333 @defmac ASM_MAYBE_OUTPUT_ENCODED_ADDR_RTX (@var{file}, @var{encoding}, @var{size}, @var{addr}, @var{done})
3334 This macro allows the target to emit whatever special magic is required
3335 to represent the encoding chosen by @code{ASM_PREFERRED_EH_DATA_FORMAT}.
3336 Generic code takes care of pc-relative and indirect encodings; this must
3337 be defined if the target uses text-relative or data-relative encodings.
3339 This is a C statement that branches to @var{done} if the format was
3340 handled. @var{encoding} is the format chosen, @var{size} is the number
3341 of bytes that the format occupies, @var{addr} is the @code{SYMBOL_REF}
3345 @defmac MD_UNWIND_SUPPORT
3346 A string specifying a file to be #include'd in unwind-dw2.c. The file
3347 so included typically defines @code{MD_FALLBACK_FRAME_STATE_FOR}.
3350 @defmac MD_FALLBACK_FRAME_STATE_FOR (@var{context}, @var{fs})
3351 This macro allows the target to add CPU and operating system specific
3352 code to the call-frame unwinder for use when there is no unwind data
3353 available. The most common reason to implement this macro is to unwind
3354 through signal frames.
3356 This macro is called from @code{uw_frame_state_for} in
3357 @file{unwind-dw2.c}, @file{unwind-dw2-xtensa.c} and
3358 @file{unwind-ia64.c}. @var{context} is an @code{_Unwind_Context};
3359 @var{fs} is an @code{_Unwind_FrameState}. Examine @code{context->ra}
3360 for the address of the code being executed and @code{context->cfa} for
3361 the stack pointer value. If the frame can be decoded, the register
3362 save addresses should be updated in @var{fs} and the macro should
3363 evaluate to @code{_URC_NO_REASON}. If the frame cannot be decoded,
3364 the macro should evaluate to @code{_URC_END_OF_STACK}.
3366 For proper signal handling in Java this macro is accompanied by
3367 @code{MAKE_THROW_FRAME}, defined in @file{libjava/include/*-signal.h} headers.
3370 @defmac MD_HANDLE_UNWABI (@var{context}, @var{fs})
3371 This macro allows the target to add operating system specific code to the
3372 call-frame unwinder to handle the IA-64 @code{.unwabi} unwinding directive,
3373 usually used for signal or interrupt frames.
3375 This macro is called from @code{uw_update_context} in @file{unwind-ia64.c}.
3376 @var{context} is an @code{_Unwind_Context};
3377 @var{fs} is an @code{_Unwind_FrameState}. Examine @code{fs->unwabi}
3378 for the abi and context in the @code{.unwabi} directive. If the
3379 @code{.unwabi} directive can be handled, the register save addresses should
3380 be updated in @var{fs}.
3383 @defmac TARGET_USES_WEAK_UNWIND_INFO
3384 A C expression that evaluates to true if the target requires unwind
3385 info to be given comdat linkage. Define it to be @code{1} if comdat
3386 linkage is necessary. The default is @code{0}.
3389 @node Stack Checking
3390 @subsection Specifying How Stack Checking is Done
3392 GCC will check that stack references are within the boundaries of the
3393 stack, if the option @option{-fstack-check} is specified, in one of
3398 If the value of the @code{STACK_CHECK_BUILTIN} macro is nonzero, GCC
3399 will assume that you have arranged for full stack checking to be done
3400 at appropriate places in the configuration files. GCC will not do
3401 other special processing.
3404 If @code{STACK_CHECK_BUILTIN} is zero and the value of the
3405 @code{STACK_CHECK_STATIC_BUILTIN} macro is nonzero, GCC will assume
3406 that you have arranged for static stack checking (checking of the
3407 static stack frame of functions) to be done at appropriate places
3408 in the configuration files. GCC will only emit code to do dynamic
3409 stack checking (checking on dynamic stack allocations) using the third
3413 If neither of the above are true, GCC will generate code to periodically
3414 ``probe'' the stack pointer using the values of the macros defined below.
3417 If neither STACK_CHECK_BUILTIN nor STACK_CHECK_STATIC_BUILTIN is defined,
3418 GCC will change its allocation strategy for large objects if the option
3419 @option{-fstack-check} is specified: they will always be allocated
3420 dynamically if their size exceeds @code{STACK_CHECK_MAX_VAR_SIZE} bytes.
3422 @defmac STACK_CHECK_BUILTIN
3423 A nonzero value if stack checking is done by the configuration files in a
3424 machine-dependent manner. You should define this macro if stack checking
3425 is required by the ABI of your machine or if you would like to do stack
3426 checking in some more efficient way than the generic approach. The default
3427 value of this macro is zero.
3430 @defmac STACK_CHECK_STATIC_BUILTIN
3431 A nonzero value if static stack checking is done by the configuration files
3432 in a machine-dependent manner. You should define this macro if you would
3433 like to do static stack checking in some more efficient way than the generic
3434 approach. The default value of this macro is zero.
3437 @defmac STACK_CHECK_PROBE_INTERVAL_EXP
3438 An integer specifying the interval at which GCC must generate stack probe
3439 instructions, defined as 2 raised to this integer. You will normally
3440 define this macro so that the interval be no larger than the size of
3441 the ``guard pages'' at the end of a stack area. The default value
3442 of 12 (4096-byte interval) is suitable for most systems.
3445 @defmac STACK_CHECK_MOVING_SP
3446 An integer which is nonzero if GCC should move the stack pointer page by page
3447 when doing probes. This can be necessary on systems where the stack pointer
3448 contains the bottom address of the memory area accessible to the executing
3449 thread at any point in time. In this situation an alternate signal stack
3450 is required in order to be able to recover from a stack overflow. The
3451 default value of this macro is zero.
3454 @defmac STACK_CHECK_PROTECT
3455 The number of bytes of stack needed to recover from a stack overflow, for
3456 languages where such a recovery is supported. The default value of 75 words
3457 with the @code{setjmp}/@code{longjmp}-based exception handling mechanism and
3458 8192 bytes with other exception handling mechanisms should be adequate for
3462 The following macros are relevant only if neither STACK_CHECK_BUILTIN
3463 nor STACK_CHECK_STATIC_BUILTIN is defined; you can omit them altogether
3464 in the opposite case.
3466 @defmac STACK_CHECK_MAX_FRAME_SIZE
3467 The maximum size of a stack frame, in bytes. GCC will generate probe
3468 instructions in non-leaf functions to ensure at least this many bytes of
3469 stack are available. If a stack frame is larger than this size, stack
3470 checking will not be reliable and GCC will issue a warning. The
3471 default is chosen so that GCC only generates one instruction on most
3472 systems. You should normally not change the default value of this macro.
3475 @defmac STACK_CHECK_FIXED_FRAME_SIZE
3476 GCC uses this value to generate the above warning message. It
3477 represents the amount of fixed frame used by a function, not including
3478 space for any callee-saved registers, temporaries and user variables.
3479 You need only specify an upper bound for this amount and will normally
3480 use the default of four words.
3483 @defmac STACK_CHECK_MAX_VAR_SIZE
3484 The maximum size, in bytes, of an object that GCC will place in the
3485 fixed area of the stack frame when the user specifies
3486 @option{-fstack-check}.
3487 GCC computed the default from the values of the above macros and you will
3488 normally not need to override that default.
3492 @node Frame Registers
3493 @subsection Registers That Address the Stack Frame
3495 @c prevent bad page break with this line
3496 This discusses registers that address the stack frame.
3498 @defmac STACK_POINTER_REGNUM
3499 The register number of the stack pointer register, which must also be a
3500 fixed register according to @code{FIXED_REGISTERS}. On most machines,
3501 the hardware determines which register this is.
3504 @defmac FRAME_POINTER_REGNUM
3505 The register number of the frame pointer register, which is used to
3506 access automatic variables in the stack frame. On some machines, the
3507 hardware determines which register this is. On other machines, you can
3508 choose any register you wish for this purpose.
3511 @defmac HARD_FRAME_POINTER_REGNUM
3512 On some machines the offset between the frame pointer and starting
3513 offset of the automatic variables is not known until after register
3514 allocation has been done (for example, because the saved registers are
3515 between these two locations). On those machines, define
3516 @code{FRAME_POINTER_REGNUM} the number of a special, fixed register to
3517 be used internally until the offset is known, and define
3518 @code{HARD_FRAME_POINTER_REGNUM} to be the actual hard register number
3519 used for the frame pointer.
3521 You should define this macro only in the very rare circumstances when it
3522 is not possible to calculate the offset between the frame pointer and
3523 the automatic variables until after register allocation has been
3524 completed. When this macro is defined, you must also indicate in your
3525 definition of @code{ELIMINABLE_REGS} how to eliminate
3526 @code{FRAME_POINTER_REGNUM} into either @code{HARD_FRAME_POINTER_REGNUM}
3527 or @code{STACK_POINTER_REGNUM}.
3529 Do not define this macro if it would be the same as
3530 @code{FRAME_POINTER_REGNUM}.
3533 @defmac ARG_POINTER_REGNUM
3534 The register number of the arg pointer register, which is used to access
3535 the function's argument list. On some machines, this is the same as the
3536 frame pointer register. On some machines, the hardware determines which
3537 register this is. On other machines, you can choose any register you
3538 wish for this purpose. If this is not the same register as the frame
3539 pointer register, then you must mark it as a fixed register according to
3540 @code{FIXED_REGISTERS}, or arrange to be able to eliminate it
3541 (@pxref{Elimination}).
3544 @defmac HARD_FRAME_POINTER_IS_FRAME_POINTER
3545 Define this to a preprocessor constant that is nonzero if
3546 @code{hard_frame_pointer_rtx} and @code{frame_pointer_rtx} should be
3547 the same. The default definition is @samp{(HARD_FRAME_POINTER_REGNUM
3548 == FRAME_POINTER_REGNUM)}; you only need to define this macro if that
3549 definition is not suitable for use in preprocessor conditionals.
3552 @defmac HARD_FRAME_POINTER_IS_ARG_POINTER
3553 Define this to a preprocessor constant that is nonzero if
3554 @code{hard_frame_pointer_rtx} and @code{arg_pointer_rtx} should be the
3555 same. The default definition is @samp{(HARD_FRAME_POINTER_REGNUM ==
3556 ARG_POINTER_REGNUM)}; you only need to define this macro if that
3557 definition is not suitable for use in preprocessor conditionals.
3560 @defmac RETURN_ADDRESS_POINTER_REGNUM
3561 The register number of the return address pointer register, which is used to
3562 access the current function's return address from the stack. On some
3563 machines, the return address is not at a fixed offset from the frame
3564 pointer or stack pointer or argument pointer. This register can be defined
3565 to point to the return address on the stack, and then be converted by
3566 @code{ELIMINABLE_REGS} into either the frame pointer or stack pointer.
3568 Do not define this macro unless there is no other way to get the return
3569 address from the stack.
3572 @defmac STATIC_CHAIN_REGNUM
3573 @defmacx STATIC_CHAIN_INCOMING_REGNUM
3574 Register numbers used for passing a function's static chain pointer. If
3575 register windows are used, the register number as seen by the called
3576 function is @code{STATIC_CHAIN_INCOMING_REGNUM}, while the register
3577 number as seen by the calling function is @code{STATIC_CHAIN_REGNUM}. If
3578 these registers are the same, @code{STATIC_CHAIN_INCOMING_REGNUM} need
3581 The static chain register need not be a fixed register.
3583 If the static chain is passed in memory, these macros should not be
3584 defined; instead, the @code{TARGET_STATIC_CHAIN} hook should be used.
3587 @hook TARGET_STATIC_CHAIN
3588 This hook replaces the use of @code{STATIC_CHAIN_REGNUM} et al for
3589 targets that may use different static chain locations for different
3590 nested functions. This may be required if the target has function
3591 attributes that affect the calling conventions of the function and
3592 those calling conventions use different static chain locations.
3594 The default version of this hook uses @code{STATIC_CHAIN_REGNUM} et al.
3596 If the static chain is passed in memory, this hook should be used to
3597 provide rtx giving @code{mem} expressions that denote where they are stored.
3598 Often the @code{mem} expression as seen by the caller will be at an offset
3599 from the stack pointer and the @code{mem} expression as seen by the callee
3600 will be at an offset from the frame pointer.
3601 @findex stack_pointer_rtx
3602 @findex frame_pointer_rtx
3603 @findex arg_pointer_rtx
3604 The variables @code{stack_pointer_rtx}, @code{frame_pointer_rtx}, and
3605 @code{arg_pointer_rtx} will have been initialized and should be used
3606 to refer to those items.
3609 @defmac DWARF_FRAME_REGISTERS
3610 This macro specifies the maximum number of hard registers that can be
3611 saved in a call frame. This is used to size data structures used in
3612 DWARF2 exception handling.
3614 Prior to GCC 3.0, this macro was needed in order to establish a stable
3615 exception handling ABI in the face of adding new hard registers for ISA
3616 extensions. In GCC 3.0 and later, the EH ABI is insulated from changes
3617 in the number of hard registers. Nevertheless, this macro can still be
3618 used to reduce the runtime memory requirements of the exception handling
3619 routines, which can be substantial if the ISA contains a lot of
3620 registers that are not call-saved.
3622 If this macro is not defined, it defaults to
3623 @code{FIRST_PSEUDO_REGISTER}.
3626 @defmac PRE_GCC3_DWARF_FRAME_REGISTERS
3628 This macro is similar to @code{DWARF_FRAME_REGISTERS}, but is provided
3629 for backward compatibility in pre GCC 3.0 compiled code.
3631 If this macro is not defined, it defaults to
3632 @code{DWARF_FRAME_REGISTERS}.
3635 @defmac DWARF_REG_TO_UNWIND_COLUMN (@var{regno})
3637 Define this macro if the target's representation for dwarf registers
3638 is different than the internal representation for unwind column.
3639 Given a dwarf register, this macro should return the internal unwind
3640 column number to use instead.
3642 See the PowerPC's SPE target for an example.
3645 @defmac DWARF_FRAME_REGNUM (@var{regno})
3647 Define this macro if the target's representation for dwarf registers
3648 used in .eh_frame or .debug_frame is different from that used in other
3649 debug info sections. Given a GCC hard register number, this macro
3650 should return the .eh_frame register number. The default is
3651 @code{DBX_REGISTER_NUMBER (@var{regno})}.
3655 @defmac DWARF2_FRAME_REG_OUT (@var{regno}, @var{for_eh})
3657 Define this macro to map register numbers held in the call frame info
3658 that GCC has collected using @code{DWARF_FRAME_REGNUM} to those that
3659 should be output in .debug_frame (@code{@var{for_eh}} is zero) and
3660 .eh_frame (@code{@var{for_eh}} is nonzero). The default is to
3661 return @code{@var{regno}}.
3666 @subsection Eliminating Frame Pointer and Arg Pointer
3668 @c prevent bad page break with this line
3669 This is about eliminating the frame pointer and arg pointer.
3671 @hook TARGET_FRAME_POINTER_REQUIRED
3672 This target hook should return @code{true} if a function must have and use
3673 a frame pointer. This target hook is called in the reload pass. If its return
3674 value is @code{true} the function will have a frame pointer.
3676 This target hook can in principle examine the current function and decide
3677 according to the facts, but on most machines the constant @code{false} or the
3678 constant @code{true} suffices. Use @code{false} when the machine allows code
3679 to be generated with no frame pointer, and doing so saves some time or space.
3680 Use @code{true} when there is no possible advantage to avoiding a frame
3683 In certain cases, the compiler does not know how to produce valid code
3684 without a frame pointer. The compiler recognizes those cases and
3685 automatically gives the function a frame pointer regardless of what
3686 @code{TARGET_FRAME_POINTER_REQUIRED} returns. You don't need to worry about
3689 In a function that does not require a frame pointer, the frame pointer
3690 register can be allocated for ordinary usage, unless you mark it as a
3691 fixed register. See @code{FIXED_REGISTERS} for more information.
3693 Default return value is @code{false}.
3696 @findex get_frame_size
3697 @defmac INITIAL_FRAME_POINTER_OFFSET (@var{depth-var})
3698 A C statement to store in the variable @var{depth-var} the difference
3699 between the frame pointer and the stack pointer values immediately after
3700 the function prologue. The value would be computed from information
3701 such as the result of @code{get_frame_size ()} and the tables of
3702 registers @code{regs_ever_live} and @code{call_used_regs}.
3704 If @code{ELIMINABLE_REGS} is defined, this macro will be not be used and
3705 need not be defined. Otherwise, it must be defined even if
3706 @code{TARGET_FRAME_POINTER_REQUIRED} always returns true; in that
3707 case, you may set @var{depth-var} to anything.
3710 @defmac ELIMINABLE_REGS
3711 If defined, this macro specifies a table of register pairs used to
3712 eliminate unneeded registers that point into the stack frame. If it is not
3713 defined, the only elimination attempted by the compiler is to replace
3714 references to the frame pointer with references to the stack pointer.
3716 The definition of this macro is a list of structure initializations, each
3717 of which specifies an original and replacement register.
3719 On some machines, the position of the argument pointer is not known until
3720 the compilation is completed. In such a case, a separate hard register
3721 must be used for the argument pointer. This register can be eliminated by
3722 replacing it with either the frame pointer or the argument pointer,
3723 depending on whether or not the frame pointer has been eliminated.
3725 In this case, you might specify:
3727 #define ELIMINABLE_REGS \
3728 @{@{ARG_POINTER_REGNUM, STACK_POINTER_REGNUM@}, \
3729 @{ARG_POINTER_REGNUM, FRAME_POINTER_REGNUM@}, \
3730 @{FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM@}@}
3733 Note that the elimination of the argument pointer with the stack pointer is
3734 specified first since that is the preferred elimination.
3737 @hook TARGET_CAN_ELIMINATE
3738 This target hook should returns @code{true} if the compiler is allowed to
3739 try to replace register number @var{from_reg} with register number
3740 @var{to_reg}. This target hook need only be defined if @code{ELIMINABLE_REGS}
3741 is defined, and will usually be @code{true}, since most of the cases
3742 preventing register elimination are things that the compiler already
3745 Default return value is @code{true}.
3748 @defmac INITIAL_ELIMINATION_OFFSET (@var{from-reg}, @var{to-reg}, @var{offset-var})
3749 This macro is similar to @code{INITIAL_FRAME_POINTER_OFFSET}. It
3750 specifies the initial difference between the specified pair of
3751 registers. This macro must be defined if @code{ELIMINABLE_REGS} is
3755 @node Stack Arguments
3756 @subsection Passing Function Arguments on the Stack
3757 @cindex arguments on stack
3758 @cindex stack arguments
3760 The macros in this section control how arguments are passed
3761 on the stack. See the following section for other macros that
3762 control passing certain arguments in registers.
3764 @hook TARGET_PROMOTE_PROTOTYPES
3765 This target hook returns @code{true} if an argument declared in a
3766 prototype as an integral type smaller than @code{int} should actually be
3767 passed as an @code{int}. In addition to avoiding errors in certain
3768 cases of mismatch, it also makes for better code on certain machines.
3769 The default is to not promote prototypes.
3773 A C expression. If nonzero, push insns will be used to pass
3775 If the target machine does not have a push instruction, set it to zero.
3776 That directs GCC to use an alternate strategy: to
3777 allocate the entire argument block and then store the arguments into
3778 it. When @code{PUSH_ARGS} is nonzero, @code{PUSH_ROUNDING} must be defined too.
3781 @defmac PUSH_ARGS_REVERSED
3782 A C expression. If nonzero, function arguments will be evaluated from
3783 last to first, rather than from first to last. If this macro is not
3784 defined, it defaults to @code{PUSH_ARGS} on targets where the stack
3785 and args grow in opposite directions, and 0 otherwise.
3788 @defmac PUSH_ROUNDING (@var{npushed})
3789 A C expression that is the number of bytes actually pushed onto the
3790 stack when an instruction attempts to push @var{npushed} bytes.
3792 On some machines, the definition
3795 #define PUSH_ROUNDING(BYTES) (BYTES)
3799 will suffice. But on other machines, instructions that appear
3800 to push one byte actually push two bytes in an attempt to maintain
3801 alignment. Then the definition should be
3804 #define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1)
3807 If the value of this macro has a type, it should be an unsigned type.
3810 @findex current_function_outgoing_args_size
3811 @defmac ACCUMULATE_OUTGOING_ARGS
3812 A C expression. If nonzero, the maximum amount of space required for outgoing arguments
3813 will be computed and placed into the variable
3814 @code{current_function_outgoing_args_size}. No space will be pushed
3815 onto the stack for each call; instead, the function prologue should
3816 increase the stack frame size by this amount.
3818 Setting both @code{PUSH_ARGS} and @code{ACCUMULATE_OUTGOING_ARGS}
3822 @defmac REG_PARM_STACK_SPACE (@var{fndecl})
3823 Define this macro if functions should assume that stack space has been
3824 allocated for arguments even when their values are passed in
3827 The value of this macro is the size, in bytes, of the area reserved for
3828 arguments passed in registers for the function represented by @var{fndecl},
3829 which can be zero if GCC is calling a library function.
3830 The argument @var{fndecl} can be the FUNCTION_DECL, or the type itself
3833 This space can be allocated by the caller, or be a part of the
3834 machine-dependent stack frame: @code{OUTGOING_REG_PARM_STACK_SPACE} says
3837 @c above is overfull. not sure what to do. --mew 5feb93 did
3838 @c something, not sure if it looks good. --mew 10feb93
3840 @defmac OUTGOING_REG_PARM_STACK_SPACE (@var{fntype})
3841 Define this to a nonzero value if it is the responsibility of the
3842 caller to allocate the area reserved for arguments passed in registers
3843 when calling a function of @var{fntype}. @var{fntype} may be NULL
3844 if the function called is a library function.
3846 If @code{ACCUMULATE_OUTGOING_ARGS} is defined, this macro controls
3847 whether the space for these arguments counts in the value of
3848 @code{current_function_outgoing_args_size}.
3851 @defmac STACK_PARMS_IN_REG_PARM_AREA
3852 Define this macro if @code{REG_PARM_STACK_SPACE} is defined, but the
3853 stack parameters don't skip the area specified by it.
3854 @c i changed this, makes more sens and it should have taken care of the
3855 @c overfull.. not as specific, tho. --mew 5feb93
3857 Normally, when a parameter is not passed in registers, it is placed on the
3858 stack beyond the @code{REG_PARM_STACK_SPACE} area. Defining this macro
3859 suppresses this behavior and causes the parameter to be passed on the
3860 stack in its natural location.
3863 @hook TARGET_RETURN_POPS_ARGS
3864 This target hook returns the number of bytes of its own arguments that
3865 a function pops on returning, or 0 if the function pops no arguments
3866 and the caller must therefore pop them all after the function returns.
3868 @var{fundecl} is a C variable whose value is a tree node that describes
3869 the function in question. Normally it is a node of type
3870 @code{FUNCTION_DECL} that describes the declaration of the function.
3871 From this you can obtain the @code{DECL_ATTRIBUTES} of the function.
3873 @var{funtype} is a C variable whose value is a tree node that
3874 describes the function in question. Normally it is a node of type
3875 @code{FUNCTION_TYPE} that describes the data type of the function.
3876 From this it is possible to obtain the data types of the value and
3877 arguments (if known).
3879 When a call to a library function is being considered, @var{fundecl}
3880 will contain an identifier node for the library function. Thus, if
3881 you need to distinguish among various library functions, you can do so
3882 by their names. Note that ``library function'' in this context means
3883 a function used to perform arithmetic, whose name is known specially
3884 in the compiler and was not mentioned in the C code being compiled.
3886 @var{size} is the number of bytes of arguments passed on the
3887 stack. If a variable number of bytes is passed, it is zero, and
3888 argument popping will always be the responsibility of the calling function.
3890 On the VAX, all functions always pop their arguments, so the definition
3891 of this macro is @var{size}. On the 68000, using the standard
3892 calling convention, no functions pop their arguments, so the value of
3893 the macro is always 0 in this case. But an alternative calling
3894 convention is available in which functions that take a fixed number of
3895 arguments pop them but other functions (such as @code{printf}) pop
3896 nothing (the caller pops all). When this convention is in use,
3897 @var{funtype} is examined to determine whether a function takes a fixed
3898 number of arguments.
3901 @defmac CALL_POPS_ARGS (@var{cum})
3902 A C expression that should indicate the number of bytes a call sequence
3903 pops off the stack. It is added to the value of @code{RETURN_POPS_ARGS}
3904 when compiling a function call.
3906 @var{cum} is the variable in which all arguments to the called function
3907 have been accumulated.
3909 On certain architectures, such as the SH5, a call trampoline is used
3910 that pops certain registers off the stack, depending on the arguments
3911 that have been passed to the function. Since this is a property of the
3912 call site, not of the called function, @code{RETURN_POPS_ARGS} is not
3916 @node Register Arguments
3917 @subsection Passing Arguments in Registers
3918 @cindex arguments in registers
3919 @cindex registers arguments
3921 This section describes the macros which let you control how various
3922 types of arguments are passed in registers or how they are arranged in
3925 @hook TARGET_FUNCTION_ARG
3926 Return an RTX indicating whether a function argument is passed in a
3927 register and if so, which register.
3929 The arguments are @var{ca}, which summarizes all the previous
3930 arguments; @var{mode}, the machine mode of the argument; @var{type},
3931 the data type of the argument as a tree node or 0 if that is not known
3932 (which happens for C support library functions); and @var{named},
3933 which is @code{true} for an ordinary argument and @code{false} for
3934 nameless arguments that correspond to @samp{@dots{}} in the called
3935 function's prototype. @var{type} can be an incomplete type if a
3936 syntax error has previously occurred.
3938 The return value is usually either a @code{reg} RTX for the hard
3939 register in which to pass the argument, or zero to pass the argument
3942 The value of the expression can also be a @code{parallel} RTX@. This is
3943 used when an argument is passed in multiple locations. The mode of the
3944 @code{parallel} should be the mode of the entire argument. The
3945 @code{parallel} holds any number of @code{expr_list} pairs; each one
3946 describes where part of the argument is passed. In each
3947 @code{expr_list} the first operand must be a @code{reg} RTX for the hard
3948 register in which to pass this part of the argument, and the mode of the
3949 register RTX indicates how large this part of the argument is. The
3950 second operand of the @code{expr_list} is a @code{const_int} which gives
3951 the offset in bytes into the entire argument of where this part starts.
3952 As a special exception the first @code{expr_list} in the @code{parallel}
3953 RTX may have a first operand of zero. This indicates that the entire
3954 argument is also stored on the stack.
3956 The last time this hook is called, it is called with @code{MODE ==
3957 VOIDmode}, and its result is passed to the @code{call} or @code{call_value}
3958 pattern as operands 2 and 3 respectively.
3960 @cindex @file{stdarg.h} and register arguments
3961 The usual way to make the ISO library @file{stdarg.h} work on a
3962 machine where some arguments are usually passed in registers, is to
3963 cause nameless arguments to be passed on the stack instead. This is
3964 done by making @code{TARGET_FUNCTION_ARG} return 0 whenever
3965 @var{named} is @code{false}.
3967 @cindex @code{TARGET_MUST_PASS_IN_STACK}, and @code{TARGET_FUNCTION_ARG}
3968 @cindex @code{REG_PARM_STACK_SPACE}, and @code{TARGET_FUNCTION_ARG}
3969 You may use the hook @code{targetm.calls.must_pass_in_stack}
3970 in the definition of this macro to determine if this argument is of a
3971 type that must be passed in the stack. If @code{REG_PARM_STACK_SPACE}
3972 is not defined and @code{TARGET_FUNCTION_ARG} returns nonzero for such an
3973 argument, the compiler will abort. If @code{REG_PARM_STACK_SPACE} is
3974 defined, the argument will be computed in the stack and then loaded into
3978 @hook TARGET_MUST_PASS_IN_STACK
3979 This target hook should return @code{true} if we should not pass @var{type}
3980 solely in registers. The file @file{expr.h} defines a
3981 definition that is usually appropriate, refer to @file{expr.h} for additional
3985 @hook TARGET_FUNCTION_INCOMING_ARG
3986 Define this hook if the target machine has ``register windows'', so
3987 that the register in which a function sees an arguments is not
3988 necessarily the same as the one in which the caller passed the
3991 For such machines, @code{TARGET_FUNCTION_ARG} computes the register in
3992 which the caller passes the value, and
3993 @code{TARGET_FUNCTION_INCOMING_ARG} should be defined in a similar
3994 fashion to tell the function being called where the arguments will
3997 If @code{TARGET_FUNCTION_INCOMING_ARG} is not defined,
3998 @code{TARGET_FUNCTION_ARG} serves both purposes.
4001 @hook TARGET_ARG_PARTIAL_BYTES
4002 This target hook returns the number of bytes at the beginning of an
4003 argument that must be put in registers. The value must be zero for
4004 arguments that are passed entirely in registers or that are entirely
4005 pushed on the stack.
4007 On some machines, certain arguments must be passed partially in
4008 registers and partially in memory. On these machines, typically the
4009 first few words of arguments are passed in registers, and the rest
4010 on the stack. If a multi-word argument (a @code{double} or a
4011 structure) crosses that boundary, its first few words must be passed
4012 in registers and the rest must be pushed. This macro tells the
4013 compiler when this occurs, and how many bytes should go in registers.
4015 @code{TARGET_FUNCTION_ARG} for these arguments should return the first
4016 register to be used by the caller for this argument; likewise
4017 @code{TARGET_FUNCTION_INCOMING_ARG}, for the called function.
4020 @hook TARGET_PASS_BY_REFERENCE
4021 This target hook should return @code{true} if an argument at the
4022 position indicated by @var{cum} should be passed by reference. This
4023 predicate is queried after target independent reasons for being
4024 passed by reference, such as @code{TREE_ADDRESSABLE (type)}.
4026 If the hook returns true, a copy of that argument is made in memory and a
4027 pointer to the argument is passed instead of the argument itself.
4028 The pointer is passed in whatever way is appropriate for passing a pointer
4032 @hook TARGET_CALLEE_COPIES
4033 The function argument described by the parameters to this hook is
4034 known to be passed by reference. The hook should return true if the
4035 function argument should be copied by the callee instead of copied
4038 For any argument for which the hook returns true, if it can be
4039 determined that the argument is not modified, then a copy need
4042 The default version of this hook always returns false.
4045 @defmac CUMULATIVE_ARGS
4046 A C type for declaring a variable that is used as the first argument
4047 of @code{TARGET_FUNCTION_ARG} and other related values. For some
4048 target machines, the type @code{int} suffices and can hold the number
4049 of bytes of argument so far.
4051 There is no need to record in @code{CUMULATIVE_ARGS} anything about the
4052 arguments that have been passed on the stack. The compiler has other
4053 variables to keep track of that. For target machines on which all
4054 arguments are passed on the stack, there is no need to store anything in
4055 @code{CUMULATIVE_ARGS}; however, the data structure must exist and
4056 should not be empty, so use @code{int}.
4059 @defmac OVERRIDE_ABI_FORMAT (@var{fndecl})
4060 If defined, this macro is called before generating any code for a
4061 function, but after the @var{cfun} descriptor for the function has been
4062 created. The back end may use this macro to update @var{cfun} to
4063 reflect an ABI other than that which would normally be used by default.
4064 If the compiler is generating code for a compiler-generated function,
4065 @var{fndecl} may be @code{NULL}.
4068 @defmac INIT_CUMULATIVE_ARGS (@var{cum}, @var{fntype}, @var{libname}, @var{fndecl}, @var{n_named_args})
4069 A C statement (sans semicolon) for initializing the variable
4070 @var{cum} for the state at the beginning of the argument list. The
4071 variable has type @code{CUMULATIVE_ARGS}. The value of @var{fntype}
4072 is the tree node for the data type of the function which will receive
4073 the args, or 0 if the args are to a compiler support library function.
4074 For direct calls that are not libcalls, @var{fndecl} contain the
4075 declaration node of the function. @var{fndecl} is also set when
4076 @code{INIT_CUMULATIVE_ARGS} is used to find arguments for the function
4077 being compiled. @var{n_named_args} is set to the number of named
4078 arguments, including a structure return address if it is passed as a
4079 parameter, when making a call. When processing incoming arguments,
4080 @var{n_named_args} is set to @minus{}1.
4082 When processing a call to a compiler support library function,
4083 @var{libname} identifies which one. It is a @code{symbol_ref} rtx which
4084 contains the name of the function, as a string. @var{libname} is 0 when
4085 an ordinary C function call is being processed. Thus, each time this
4086 macro is called, either @var{libname} or @var{fntype} is nonzero, but
4087 never both of them at once.
4090 @defmac INIT_CUMULATIVE_LIBCALL_ARGS (@var{cum}, @var{mode}, @var{libname})
4091 Like @code{INIT_CUMULATIVE_ARGS} but only used for outgoing libcalls,
4092 it gets a @code{MODE} argument instead of @var{fntype}, that would be
4093 @code{NULL}. @var{indirect} would always be zero, too. If this macro
4094 is not defined, @code{INIT_CUMULATIVE_ARGS (cum, NULL_RTX, libname,
4095 0)} is used instead.
4098 @defmac INIT_CUMULATIVE_INCOMING_ARGS (@var{cum}, @var{fntype}, @var{libname})
4099 Like @code{INIT_CUMULATIVE_ARGS} but overrides it for the purposes of
4100 finding the arguments for the function being compiled. If this macro is
4101 undefined, @code{INIT_CUMULATIVE_ARGS} is used instead.
4103 The value passed for @var{libname} is always 0, since library routines
4104 with special calling conventions are never compiled with GCC@. The
4105 argument @var{libname} exists for symmetry with
4106 @code{INIT_CUMULATIVE_ARGS}.
4107 @c could use "this macro" in place of @code{INIT_CUMULATIVE_ARGS}, maybe.
4108 @c --mew 5feb93 i switched the order of the sentences. --mew 10feb93
4111 @hook TARGET_FUNCTION_ARG_ADVANCE
4112 This hook updates the summarizer variable pointed to by @var{ca} to
4113 advance past an argument in the argument list. The values @var{mode},
4114 @var{type} and @var{named} describe that argument. Once this is done,
4115 the variable @var{cum} is suitable for analyzing the @emph{following}
4116 argument with @code{TARGET_FUNCTION_ARG}, etc.
4118 This hook need not do anything if the argument in question was passed
4119 on the stack. The compiler knows how to track the amount of stack space
4120 used for arguments without any special help.
4123 @defmac FUNCTION_ARG_OFFSET (@var{mode}, @var{type})
4124 If defined, a C expression that is the number of bytes to add to the
4125 offset of the argument passed in memory. This is needed for the SPU,
4126 which passes @code{char} and @code{short} arguments in the preferred
4127 slot that is in the middle of the quad word instead of starting at the
4131 @defmac FUNCTION_ARG_PADDING (@var{mode}, @var{type})
4132 If defined, a C expression which determines whether, and in which direction,
4133 to pad out an argument with extra space. The value should be of type
4134 @code{enum direction}: either @code{upward} to pad above the argument,
4135 @code{downward} to pad below, or @code{none} to inhibit padding.
4137 The @emph{amount} of padding is always just enough to reach the next
4138 multiple of @code{TARGET_FUNCTION_ARG_BOUNDARY}; this macro does not
4141 This macro has a default definition which is right for most systems.
4142 For little-endian machines, the default is to pad upward. For
4143 big-endian machines, the default is to pad downward for an argument of
4144 constant size shorter than an @code{int}, and upward otherwise.
4147 @defmac PAD_VARARGS_DOWN
4148 If defined, a C expression which determines whether the default
4149 implementation of va_arg will attempt to pad down before reading the
4150 next argument, if that argument is smaller than its aligned space as
4151 controlled by @code{PARM_BOUNDARY}. If this macro is not defined, all such
4152 arguments are padded down if @code{BYTES_BIG_ENDIAN} is true.
4155 @defmac BLOCK_REG_PADDING (@var{mode}, @var{type}, @var{first})
4156 Specify padding for the last element of a block move between registers and
4157 memory. @var{first} is nonzero if this is the only element. Defining this
4158 macro allows better control of register function parameters on big-endian
4159 machines, without using @code{PARALLEL} rtl. In particular,
4160 @code{MUST_PASS_IN_STACK} need not test padding and mode of types in
4161 registers, as there is no longer a "wrong" part of a register; For example,
4162 a three byte aggregate may be passed in the high part of a register if so
4166 @hook TARGET_FUNCTION_ARG_BOUNDARY
4167 This hook returns the alignment boundary, in bits, of an argument
4168 with the specified mode and type. The default hook returns
4169 @code{PARM_BOUNDARY} for all arguments.
4172 @defmac FUNCTION_ARG_REGNO_P (@var{regno})
4173 A C expression that is nonzero if @var{regno} is the number of a hard
4174 register in which function arguments are sometimes passed. This does
4175 @emph{not} include implicit arguments such as the static chain and
4176 the structure-value address. On many machines, no registers can be
4177 used for this purpose since all function arguments are pushed on the
4181 @hook TARGET_SPLIT_COMPLEX_ARG
4182 This hook should return true if parameter of type @var{type} are passed
4183 as two scalar parameters. By default, GCC will attempt to pack complex
4184 arguments into the target's word size. Some ABIs require complex arguments
4185 to be split and treated as their individual components. For example, on
4186 AIX64, complex floats should be passed in a pair of floating point
4187 registers, even though a complex float would fit in one 64-bit floating
4190 The default value of this hook is @code{NULL}, which is treated as always
4194 @hook TARGET_BUILD_BUILTIN_VA_LIST
4195 This hook returns a type node for @code{va_list} for the target.
4196 The default version of the hook returns @code{void*}.
4199 @hook TARGET_ENUM_VA_LIST_P
4200 This target hook is used in function @code{c_common_nodes_and_builtins}
4201 to iterate through the target specific builtin types for va_list. The
4202 variable @var{idx} is used as iterator. @var{pname} has to be a pointer
4203 to a @code{const char *} and @var{ptree} a pointer to a @code{tree} typed
4205 The arguments @var{pname} and @var{ptree} are used to store the result of
4206 this macro and are set to the name of the va_list builtin type and its
4208 If the return value of this macro is zero, then there is no more element.
4209 Otherwise the @var{IDX} should be increased for the next call of this
4210 macro to iterate through all types.
4213 @hook TARGET_FN_ABI_VA_LIST
4214 This hook returns the va_list type of the calling convention specified by
4216 The default version of this hook returns @code{va_list_type_node}.
4219 @hook TARGET_CANONICAL_VA_LIST_TYPE
4220 This hook returns the va_list type of the calling convention specified by the
4221 type of @var{type}. If @var{type} is not a valid va_list type, it returns
4225 @hook TARGET_GIMPLIFY_VA_ARG_EXPR
4226 This hook performs target-specific gimplification of
4227 @code{VA_ARG_EXPR}. The first two parameters correspond to the
4228 arguments to @code{va_arg}; the latter two are as in
4229 @code{gimplify.c:gimplify_expr}.
4232 @hook TARGET_VALID_POINTER_MODE
4233 Define this to return nonzero if the port can handle pointers
4234 with machine mode @var{mode}. The default version of this
4235 hook returns true for both @code{ptr_mode} and @code{Pmode}.
4238 @hook TARGET_REF_MAY_ALIAS_ERRNO
4240 @hook TARGET_SCALAR_MODE_SUPPORTED_P
4241 Define this to return nonzero if the port is prepared to handle
4242 insns involving scalar mode @var{mode}. For a scalar mode to be
4243 considered supported, all the basic arithmetic and comparisons
4246 The default version of this hook returns true for any mode
4247 required to handle the basic C types (as defined by the port).
4248 Included here are the double-word arithmetic supported by the
4249 code in @file{optabs.c}.
4252 @hook TARGET_VECTOR_MODE_SUPPORTED_P
4253 Define this to return nonzero if the port is prepared to handle
4254 insns involving vector mode @var{mode}. At the very least, it
4255 must have move patterns for this mode.
4258 @hook TARGET_SMALL_REGISTER_CLASSES_FOR_MODE_P
4259 Define this to return nonzero for machine modes for which the port has
4260 small register classes. If this target hook returns nonzero for a given
4261 @var{mode}, the compiler will try to minimize the lifetime of registers
4262 in @var{mode}. The hook may be called with @code{VOIDmode} as argument.
4263 In this case, the hook is expected to return nonzero if it returns nonzero
4266 On some machines, it is risky to let hard registers live across arbitrary
4267 insns. Typically, these machines have instructions that require values
4268 to be in specific registers (like an accumulator), and reload will fail
4269 if the required hard register is used for another purpose across such an
4272 Passes before reload do not know which hard registers will be used
4273 in an instruction, but the machine modes of the registers set or used in
4274 the instruction are already known. And for some machines, register
4275 classes are small for, say, integer registers but not for floating point
4276 registers. For example, the AMD x86-64 architecture requires specific
4277 registers for the legacy x86 integer instructions, but there are many
4278 SSE registers for floating point operations. On such targets, a good
4279 strategy may be to return nonzero from this hook for @code{INTEGRAL_MODE_P}
4280 machine modes but zero for the SSE register classes.
4282 The default version of this hook returns false for any mode. It is always
4283 safe to redefine this hook to return with a nonzero value. But if you
4284 unnecessarily define it, you will reduce the amount of optimizations
4285 that can be performed in some cases. If you do not define this hook
4286 to return a nonzero value when it is required, the compiler will run out
4287 of spill registers and print a fatal error message.
4290 @hook TARGET_FLAGS_REGNUM
4293 @subsection How Scalar Function Values Are Returned
4294 @cindex return values in registers
4295 @cindex values, returned by functions
4296 @cindex scalars, returned as values
4298 This section discusses the macros that control returning scalars as
4299 values---values that can fit in registers.
4301 @hook TARGET_FUNCTION_VALUE
4303 Define this to return an RTX representing the place where a function
4304 returns or receives a value of data type @var{ret_type}, a tree node
4305 representing a data type. @var{fn_decl_or_type} is a tree node
4306 representing @code{FUNCTION_DECL} or @code{FUNCTION_TYPE} of a
4307 function being called. If @var{outgoing} is false, the hook should
4308 compute the register in which the caller will see the return value.
4309 Otherwise, the hook should return an RTX representing the place where
4310 a function returns a value.
4312 On many machines, only @code{TYPE_MODE (@var{ret_type})} is relevant.
4313 (Actually, on most machines, scalar values are returned in the same
4314 place regardless of mode.) The value of the expression is usually a
4315 @code{reg} RTX for the hard register where the return value is stored.
4316 The value can also be a @code{parallel} RTX, if the return value is in
4317 multiple places. See @code{TARGET_FUNCTION_ARG} for an explanation of the
4318 @code{parallel} form. Note that the callee will populate every
4319 location specified in the @code{parallel}, but if the first element of
4320 the @code{parallel} contains the whole return value, callers will use
4321 that element as the canonical location and ignore the others. The m68k
4322 port uses this type of @code{parallel} to return pointers in both
4323 @samp{%a0} (the canonical location) and @samp{%d0}.
4325 If @code{TARGET_PROMOTE_FUNCTION_RETURN} returns true, you must apply
4326 the same promotion rules specified in @code{PROMOTE_MODE} if
4327 @var{valtype} is a scalar type.
4329 If the precise function being called is known, @var{func} is a tree
4330 node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
4331 pointer. This makes it possible to use a different value-returning
4332 convention for specific functions when all their calls are
4335 Some target machines have ``register windows'' so that the register in
4336 which a function returns its value is not the same as the one in which
4337 the caller sees the value. For such machines, you should return
4338 different RTX depending on @var{outgoing}.
4340 @code{TARGET_FUNCTION_VALUE} is not used for return values with
4341 aggregate data types, because these are returned in another way. See
4342 @code{TARGET_STRUCT_VALUE_RTX} and related macros, below.
4345 @defmac FUNCTION_VALUE (@var{valtype}, @var{func})
4346 This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE} for
4347 a new target instead.
4350 @defmac LIBCALL_VALUE (@var{mode})
4351 A C expression to create an RTX representing the place where a library
4352 function returns a value of mode @var{mode}.
4354 Note that ``library function'' in this context means a compiler
4355 support routine, used to perform arithmetic, whose name is known
4356 specially by the compiler and was not mentioned in the C code being
4360 @hook TARGET_LIBCALL_VALUE
4361 Define this hook if the back-end needs to know the name of the libcall
4362 function in order to determine where the result should be returned.
4364 The mode of the result is given by @var{mode} and the name of the called
4365 library function is given by @var{fun}. The hook should return an RTX
4366 representing the place where the library function result will be returned.
4368 If this hook is not defined, then LIBCALL_VALUE will be used.
4371 @defmac FUNCTION_VALUE_REGNO_P (@var{regno})
4372 A C expression that is nonzero if @var{regno} is the number of a hard
4373 register in which the values of called function may come back.
4375 A register whose use for returning values is limited to serving as the
4376 second of a pair (for a value of type @code{double}, say) need not be
4377 recognized by this macro. So for most machines, this definition
4381 #define FUNCTION_VALUE_REGNO_P(N) ((N) == 0)
4384 If the machine has register windows, so that the caller and the called
4385 function use different registers for the return value, this macro
4386 should recognize only the caller's register numbers.
4388 This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE_REGNO_P}
4389 for a new target instead.
4392 @hook TARGET_FUNCTION_VALUE_REGNO_P
4393 A target hook that return @code{true} if @var{regno} is the number of a hard
4394 register in which the values of called function may come back.
4396 A register whose use for returning values is limited to serving as the
4397 second of a pair (for a value of type @code{double}, say) need not be
4398 recognized by this target hook.
4400 If the machine has register windows, so that the caller and the called
4401 function use different registers for the return value, this target hook
4402 should recognize only the caller's register numbers.
4404 If this hook is not defined, then FUNCTION_VALUE_REGNO_P will be used.
4407 @defmac APPLY_RESULT_SIZE
4408 Define this macro if @samp{untyped_call} and @samp{untyped_return}
4409 need more space than is implied by @code{FUNCTION_VALUE_REGNO_P} for
4410 saving and restoring an arbitrary return value.
4413 @hook TARGET_RETURN_IN_MSB
4414 This hook should return true if values of type @var{type} are returned
4415 at the most significant end of a register (in other words, if they are
4416 padded at the least significant end). You can assume that @var{type}
4417 is returned in a register; the caller is required to check this.
4419 Note that the register provided by @code{TARGET_FUNCTION_VALUE} must
4420 be able to hold the complete return value. For example, if a 1-, 2-
4421 or 3-byte structure is returned at the most significant end of a
4422 4-byte register, @code{TARGET_FUNCTION_VALUE} should provide an
4426 @node Aggregate Return
4427 @subsection How Large Values Are Returned
4428 @cindex aggregates as return values
4429 @cindex large return values
4430 @cindex returning aggregate values
4431 @cindex structure value address
4433 When a function value's mode is @code{BLKmode} (and in some other
4434 cases), the value is not returned according to
4435 @code{TARGET_FUNCTION_VALUE} (@pxref{Scalar Return}). Instead, the
4436 caller passes the address of a block of memory in which the value
4437 should be stored. This address is called the @dfn{structure value
4440 This section describes how to control returning structure values in
4443 @hook TARGET_RETURN_IN_MEMORY
4444 This target hook should return a nonzero value to say to return the
4445 function value in memory, just as large structures are always returned.
4446 Here @var{type} will be the data type of the value, and @var{fntype}
4447 will be the type of the function doing the returning, or @code{NULL} for
4450 Note that values of mode @code{BLKmode} must be explicitly handled
4451 by this function. Also, the option @option{-fpcc-struct-return}
4452 takes effect regardless of this macro. On most systems, it is
4453 possible to leave the hook undefined; this causes a default
4454 definition to be used, whose value is the constant 1 for @code{BLKmode}
4455 values, and 0 otherwise.
4457 Do not use this hook to indicate that structures and unions should always
4458 be returned in memory. You should instead use @code{DEFAULT_PCC_STRUCT_RETURN}
4462 @defmac DEFAULT_PCC_STRUCT_RETURN
4463 Define this macro to be 1 if all structure and union return values must be
4464 in memory. Since this results in slower code, this should be defined
4465 only if needed for compatibility with other compilers or with an ABI@.
4466 If you define this macro to be 0, then the conventions used for structure
4467 and union return values are decided by the @code{TARGET_RETURN_IN_MEMORY}
4470 If not defined, this defaults to the value 1.
4473 @hook TARGET_STRUCT_VALUE_RTX
4474 This target hook should return the location of the structure value
4475 address (normally a @code{mem} or @code{reg}), or 0 if the address is
4476 passed as an ``invisible'' first argument. Note that @var{fndecl} may
4477 be @code{NULL}, for libcalls. You do not need to define this target
4478 hook if the address is always passed as an ``invisible'' first
4481 On some architectures the place where the structure value address
4482 is found by the called function is not the same place that the
4483 caller put it. This can be due to register windows, or it could
4484 be because the function prologue moves it to a different place.
4485 @var{incoming} is @code{1} or @code{2} when the location is needed in
4486 the context of the called function, and @code{0} in the context of
4489 If @var{incoming} is nonzero and the address is to be found on the
4490 stack, return a @code{mem} which refers to the frame pointer. If
4491 @var{incoming} is @code{2}, the result is being used to fetch the
4492 structure value address at the beginning of a function. If you need
4493 to emit adjusting code, you should do it at this point.
4496 @defmac PCC_STATIC_STRUCT_RETURN
4497 Define this macro if the usual system convention on the target machine
4498 for returning structures and unions is for the called function to return
4499 the address of a static variable containing the value.
4501 Do not define this if the usual system convention is for the caller to
4502 pass an address to the subroutine.
4504 This macro has effect in @option{-fpcc-struct-return} mode, but it does
4505 nothing when you use @option{-freg-struct-return} mode.
4508 @hook TARGET_GET_RAW_RESULT_MODE
4510 @hook TARGET_GET_RAW_ARG_MODE
4513 @subsection Caller-Saves Register Allocation
4515 If you enable it, GCC can save registers around function calls. This
4516 makes it possible to use call-clobbered registers to hold variables that
4517 must live across calls.
4519 @defmac CALLER_SAVE_PROFITABLE (@var{refs}, @var{calls})
4520 A C expression to determine whether it is worthwhile to consider placing
4521 a pseudo-register in a call-clobbered hard register and saving and
4522 restoring it around each function call. The expression should be 1 when
4523 this is worth doing, and 0 otherwise.
4525 If you don't define this macro, a default is used which is good on most
4526 machines: @code{4 * @var{calls} < @var{refs}}.
4529 @defmac HARD_REGNO_CALLER_SAVE_MODE (@var{regno}, @var{nregs})
4530 A C expression specifying which mode is required for saving @var{nregs}
4531 of a pseudo-register in call-clobbered hard register @var{regno}. If
4532 @var{regno} is unsuitable for caller save, @code{VOIDmode} should be
4533 returned. For most machines this macro need not be defined since GCC
4534 will select the smallest suitable mode.
4537 @node Function Entry
4538 @subsection Function Entry and Exit
4539 @cindex function entry and exit
4543 This section describes the macros that output function entry
4544 (@dfn{prologue}) and exit (@dfn{epilogue}) code.
4546 @hook TARGET_ASM_FUNCTION_PROLOGUE
4547 If defined, a function that outputs the assembler code for entry to a
4548 function. The prologue is responsible for setting up the stack frame,
4549 initializing the frame pointer register, saving registers that must be
4550 saved, and allocating @var{size} additional bytes of storage for the
4551 local variables. @var{size} is an integer. @var{file} is a stdio
4552 stream to which the assembler code should be output.
4554 The label for the beginning of the function need not be output by this
4555 macro. That has already been done when the macro is run.
4557 @findex regs_ever_live
4558 To determine which registers to save, the macro can refer to the array
4559 @code{regs_ever_live}: element @var{r} is nonzero if hard register
4560 @var{r} is used anywhere within the function. This implies the function
4561 prologue should save register @var{r}, provided it is not one of the
4562 call-used registers. (@code{TARGET_ASM_FUNCTION_EPILOGUE} must likewise use
4563 @code{regs_ever_live}.)
4565 On machines that have ``register windows'', the function entry code does
4566 not save on the stack the registers that are in the windows, even if
4567 they are supposed to be preserved by function calls; instead it takes
4568 appropriate steps to ``push'' the register stack, if any non-call-used
4569 registers are used in the function.
4571 @findex frame_pointer_needed
4572 On machines where functions may or may not have frame-pointers, the
4573 function entry code must vary accordingly; it must set up the frame
4574 pointer if one is wanted, and not otherwise. To determine whether a
4575 frame pointer is in wanted, the macro can refer to the variable
4576 @code{frame_pointer_needed}. The variable's value will be 1 at run
4577 time in a function that needs a frame pointer. @xref{Elimination}.
4579 The function entry code is responsible for allocating any stack space
4580 required for the function. This stack space consists of the regions
4581 listed below. In most cases, these regions are allocated in the
4582 order listed, with the last listed region closest to the top of the
4583 stack (the lowest address if @code{STACK_GROWS_DOWNWARD} is defined, and
4584 the highest address if it is not defined). You can use a different order
4585 for a machine if doing so is more convenient or required for
4586 compatibility reasons. Except in cases where required by standard
4587 or by a debugger, there is no reason why the stack layout used by GCC
4588 need agree with that used by other compilers for a machine.
4591 @hook TARGET_ASM_FUNCTION_END_PROLOGUE
4592 If defined, a function that outputs assembler code at the end of a
4593 prologue. This should be used when the function prologue is being
4594 emitted as RTL, and you have some extra assembler that needs to be
4595 emitted. @xref{prologue instruction pattern}.
4598 @hook TARGET_ASM_FUNCTION_BEGIN_EPILOGUE
4599 If defined, a function that outputs assembler code at the start of an
4600 epilogue. This should be used when the function epilogue is being
4601 emitted as RTL, and you have some extra assembler that needs to be
4602 emitted. @xref{epilogue instruction pattern}.
4605 @hook TARGET_ASM_FUNCTION_EPILOGUE
4606 If defined, a function that outputs the assembler code for exit from a
4607 function. The epilogue is responsible for restoring the saved
4608 registers and stack pointer to their values when the function was
4609 called, and returning control to the caller. This macro takes the
4610 same arguments as the macro @code{TARGET_ASM_FUNCTION_PROLOGUE}, and the
4611 registers to restore are determined from @code{regs_ever_live} and
4612 @code{CALL_USED_REGISTERS} in the same way.
4614 On some machines, there is a single instruction that does all the work
4615 of returning from the function. On these machines, give that
4616 instruction the name @samp{return} and do not define the macro
4617 @code{TARGET_ASM_FUNCTION_EPILOGUE} at all.
4619 Do not define a pattern named @samp{return} if you want the
4620 @code{TARGET_ASM_FUNCTION_EPILOGUE} to be used. If you want the target
4621 switches to control whether return instructions or epilogues are used,
4622 define a @samp{return} pattern with a validity condition that tests the
4623 target switches appropriately. If the @samp{return} pattern's validity
4624 condition is false, epilogues will be used.
4626 On machines where functions may or may not have frame-pointers, the
4627 function exit code must vary accordingly. Sometimes the code for these
4628 two cases is completely different. To determine whether a frame pointer
4629 is wanted, the macro can refer to the variable
4630 @code{frame_pointer_needed}. The variable's value will be 1 when compiling
4631 a function that needs a frame pointer.
4633 Normally, @code{TARGET_ASM_FUNCTION_PROLOGUE} and
4634 @code{TARGET_ASM_FUNCTION_EPILOGUE} must treat leaf functions specially.
4635 The C variable @code{current_function_is_leaf} is nonzero for such a
4636 function. @xref{Leaf Functions}.
4638 On some machines, some functions pop their arguments on exit while
4639 others leave that for the caller to do. For example, the 68020 when
4640 given @option{-mrtd} pops arguments in functions that take a fixed
4641 number of arguments.
4643 @findex current_function_pops_args
4644 Your definition of the macro @code{RETURN_POPS_ARGS} decides which
4645 functions pop their own arguments. @code{TARGET_ASM_FUNCTION_EPILOGUE}
4646 needs to know what was decided. The number of bytes of the current
4647 function's arguments that this function should pop is available in
4648 @code{crtl->args.pops_args}. @xref{Scalar Return}.
4653 @findex current_function_pretend_args_size
4654 A region of @code{current_function_pretend_args_size} bytes of
4655 uninitialized space just underneath the first argument arriving on the
4656 stack. (This may not be at the very start of the allocated stack region
4657 if the calling sequence has pushed anything else since pushing the stack
4658 arguments. But usually, on such machines, nothing else has been pushed
4659 yet, because the function prologue itself does all the pushing.) This
4660 region is used on machines where an argument may be passed partly in
4661 registers and partly in memory, and, in some cases to support the
4662 features in @code{<stdarg.h>}.
4665 An area of memory used to save certain registers used by the function.
4666 The size of this area, which may also include space for such things as
4667 the return address and pointers to previous stack frames, is
4668 machine-specific and usually depends on which registers have been used
4669 in the function. Machines with register windows often do not require
4673 A region of at least @var{size} bytes, possibly rounded up to an allocation
4674 boundary, to contain the local variables of the function. On some machines,
4675 this region and the save area may occur in the opposite order, with the
4676 save area closer to the top of the stack.
4679 @cindex @code{ACCUMULATE_OUTGOING_ARGS} and stack frames
4680 Optionally, when @code{ACCUMULATE_OUTGOING_ARGS} is defined, a region of
4681 @code{current_function_outgoing_args_size} bytes to be used for outgoing
4682 argument lists of the function. @xref{Stack Arguments}.
4685 @defmac EXIT_IGNORE_STACK
4686 Define this macro as a C expression that is nonzero if the return
4687 instruction or the function epilogue ignores the value of the stack
4688 pointer; in other words, if it is safe to delete an instruction to
4689 adjust the stack pointer before a return from the function. The
4692 Note that this macro's value is relevant only for functions for which
4693 frame pointers are maintained. It is never safe to delete a final
4694 stack adjustment in a function that has no frame pointer, and the
4695 compiler knows this regardless of @code{EXIT_IGNORE_STACK}.
4698 @defmac EPILOGUE_USES (@var{regno})
4699 Define this macro as a C expression that is nonzero for registers that are
4700 used by the epilogue or the @samp{return} pattern. The stack and frame
4701 pointer registers are already assumed to be used as needed.
4704 @defmac EH_USES (@var{regno})
4705 Define this macro as a C expression that is nonzero for registers that are
4706 used by the exception handling mechanism, and so should be considered live
4707 on entry to an exception edge.
4710 @defmac DELAY_SLOTS_FOR_EPILOGUE
4711 Define this macro if the function epilogue contains delay slots to which
4712 instructions from the rest of the function can be ``moved''. The
4713 definition should be a C expression whose value is an integer
4714 representing the number of delay slots there.
4717 @defmac ELIGIBLE_FOR_EPILOGUE_DELAY (@var{insn}, @var{n})
4718 A C expression that returns 1 if @var{insn} can be placed in delay
4719 slot number @var{n} of the epilogue.
4721 The argument @var{n} is an integer which identifies the delay slot now
4722 being considered (since different slots may have different rules of
4723 eligibility). It is never negative and is always less than the number
4724 of epilogue delay slots (what @code{DELAY_SLOTS_FOR_EPILOGUE} returns).
4725 If you reject a particular insn for a given delay slot, in principle, it
4726 may be reconsidered for a subsequent delay slot. Also, other insns may
4727 (at least in principle) be considered for the so far unfilled delay
4730 @findex current_function_epilogue_delay_list
4731 @findex final_scan_insn
4732 The insns accepted to fill the epilogue delay slots are put in an RTL
4733 list made with @code{insn_list} objects, stored in the variable
4734 @code{current_function_epilogue_delay_list}. The insn for the first
4735 delay slot comes first in the list. Your definition of the macro
4736 @code{TARGET_ASM_FUNCTION_EPILOGUE} should fill the delay slots by
4737 outputting the insns in this list, usually by calling
4738 @code{final_scan_insn}.
4740 You need not define this macro if you did not define
4741 @code{DELAY_SLOTS_FOR_EPILOGUE}.
4744 @hook TARGET_ASM_OUTPUT_MI_THUNK
4745 A function that outputs the assembler code for a thunk
4746 function, used to implement C++ virtual function calls with multiple
4747 inheritance. The thunk acts as a wrapper around a virtual function,
4748 adjusting the implicit object parameter before handing control off to
4751 First, emit code to add the integer @var{delta} to the location that
4752 contains the incoming first argument. Assume that this argument
4753 contains a pointer, and is the one used to pass the @code{this} pointer
4754 in C++. This is the incoming argument @emph{before} the function prologue,
4755 e.g.@: @samp{%o0} on a sparc. The addition must preserve the values of
4756 all other incoming arguments.
4758 Then, if @var{vcall_offset} is nonzero, an additional adjustment should be
4759 made after adding @code{delta}. In particular, if @var{p} is the
4760 adjusted pointer, the following adjustment should be made:
4763 p += (*((ptrdiff_t **)p))[vcall_offset/sizeof(ptrdiff_t)]
4766 After the additions, emit code to jump to @var{function}, which is a
4767 @code{FUNCTION_DECL}. This is a direct pure jump, not a call, and does
4768 not touch the return address. Hence returning from @var{FUNCTION} will
4769 return to whoever called the current @samp{thunk}.
4771 The effect must be as if @var{function} had been called directly with
4772 the adjusted first argument. This macro is responsible for emitting all
4773 of the code for a thunk function; @code{TARGET_ASM_FUNCTION_PROLOGUE}
4774 and @code{TARGET_ASM_FUNCTION_EPILOGUE} are not invoked.
4776 The @var{thunk_fndecl} is redundant. (@var{delta} and @var{function}
4777 have already been extracted from it.) It might possibly be useful on
4778 some targets, but probably not.
4780 If you do not define this macro, the target-independent code in the C++
4781 front end will generate a less efficient heavyweight thunk that calls
4782 @var{function} instead of jumping to it. The generic approach does
4783 not support varargs.
4786 @hook TARGET_ASM_CAN_OUTPUT_MI_THUNK
4787 A function that returns true if TARGET_ASM_OUTPUT_MI_THUNK would be able
4788 to output the assembler code for the thunk function specified by the
4789 arguments it is passed, and false otherwise. In the latter case, the
4790 generic approach will be used by the C++ front end, with the limitations
4795 @subsection Generating Code for Profiling
4796 @cindex profiling, code generation
4798 These macros will help you generate code for profiling.
4800 @defmac FUNCTION_PROFILER (@var{file}, @var{labelno})
4801 A C statement or compound statement to output to @var{file} some
4802 assembler code to call the profiling subroutine @code{mcount}.
4805 The details of how @code{mcount} expects to be called are determined by
4806 your operating system environment, not by GCC@. To figure them out,
4807 compile a small program for profiling using the system's installed C
4808 compiler and look at the assembler code that results.
4810 Older implementations of @code{mcount} expect the address of a counter
4811 variable to be loaded into some register. The name of this variable is
4812 @samp{LP} followed by the number @var{labelno}, so you would generate
4813 the name using @samp{LP%d} in a @code{fprintf}.
4816 @defmac PROFILE_HOOK
4817 A C statement or compound statement to output to @var{file} some assembly
4818 code to call the profiling subroutine @code{mcount} even the target does
4819 not support profiling.
4822 @defmac NO_PROFILE_COUNTERS
4823 Define this macro to be an expression with a nonzero value if the
4824 @code{mcount} subroutine on your system does not need a counter variable
4825 allocated for each function. This is true for almost all modern
4826 implementations. If you define this macro, you must not use the
4827 @var{labelno} argument to @code{FUNCTION_PROFILER}.
4830 @defmac PROFILE_BEFORE_PROLOGUE
4831 Define this macro if the code for function profiling should come before
4832 the function prologue. Normally, the profiling code comes after.
4836 @subsection Permitting tail calls
4839 @hook TARGET_FUNCTION_OK_FOR_SIBCALL
4840 True if it is ok to do sibling call optimization for the specified
4841 call expression @var{exp}. @var{decl} will be the called function,
4842 or @code{NULL} if this is an indirect call.
4844 It is not uncommon for limitations of calling conventions to prevent
4845 tail calls to functions outside the current unit of translation, or
4846 during PIC compilation. The hook is used to enforce these restrictions,
4847 as the @code{sibcall} md pattern can not fail, or fall over to a
4848 ``normal'' call. The criteria for successful sibling call optimization
4849 may vary greatly between different architectures.
4852 @hook TARGET_EXTRA_LIVE_ON_ENTRY
4853 Add any hard registers to @var{regs} that are live on entry to the
4854 function. This hook only needs to be defined to provide registers that
4855 cannot be found by examination of FUNCTION_ARG_REGNO_P, the callee saved
4856 registers, STATIC_CHAIN_INCOMING_REGNUM, STATIC_CHAIN_REGNUM,
4857 TARGET_STRUCT_VALUE_RTX, FRAME_POINTER_REGNUM, EH_USES,
4858 FRAME_POINTER_REGNUM, ARG_POINTER_REGNUM, and the PIC_OFFSET_TABLE_REGNUM.
4861 @node Stack Smashing Protection
4862 @subsection Stack smashing protection
4863 @cindex stack smashing protection
4865 @hook TARGET_STACK_PROTECT_GUARD
4866 This hook returns a @code{DECL} node for the external variable to use
4867 for the stack protection guard. This variable is initialized by the
4868 runtime to some random value and is used to initialize the guard value
4869 that is placed at the top of the local stack frame. The type of this
4870 variable must be @code{ptr_type_node}.
4872 The default version of this hook creates a variable called
4873 @samp{__stack_chk_guard}, which is normally defined in @file{libgcc2.c}.
4876 @hook TARGET_STACK_PROTECT_FAIL
4877 This hook returns a tree expression that alerts the runtime that the
4878 stack protect guard variable has been modified. This expression should
4879 involve a call to a @code{noreturn} function.
4881 The default version of this hook invokes a function called
4882 @samp{__stack_chk_fail}, taking no arguments. This function is
4883 normally defined in @file{libgcc2.c}.
4886 @hook TARGET_SUPPORTS_SPLIT_STACK
4889 @section Implementing the Varargs Macros
4890 @cindex varargs implementation
4892 GCC comes with an implementation of @code{<varargs.h>} and
4893 @code{<stdarg.h>} that work without change on machines that pass arguments
4894 on the stack. Other machines require their own implementations of
4895 varargs, and the two machine independent header files must have
4896 conditionals to include it.
4898 ISO @code{<stdarg.h>} differs from traditional @code{<varargs.h>} mainly in
4899 the calling convention for @code{va_start}. The traditional
4900 implementation takes just one argument, which is the variable in which
4901 to store the argument pointer. The ISO implementation of
4902 @code{va_start} takes an additional second argument. The user is
4903 supposed to write the last named argument of the function here.
4905 However, @code{va_start} should not use this argument. The way to find
4906 the end of the named arguments is with the built-in functions described
4909 @defmac __builtin_saveregs ()
4910 Use this built-in function to save the argument registers in memory so
4911 that the varargs mechanism can access them. Both ISO and traditional
4912 versions of @code{va_start} must use @code{__builtin_saveregs}, unless
4913 you use @code{TARGET_SETUP_INCOMING_VARARGS} (see below) instead.
4915 On some machines, @code{__builtin_saveregs} is open-coded under the
4916 control of the target hook @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. On
4917 other machines, it calls a routine written in assembler language,
4918 found in @file{libgcc2.c}.
4920 Code generated for the call to @code{__builtin_saveregs} appears at the
4921 beginning of the function, as opposed to where the call to
4922 @code{__builtin_saveregs} is written, regardless of what the code is.
4923 This is because the registers must be saved before the function starts
4924 to use them for its own purposes.
4925 @c i rewrote the first sentence above to fix an overfull hbox. --mew
4929 @defmac __builtin_next_arg (@var{lastarg})
4930 This builtin returns the address of the first anonymous stack
4931 argument, as type @code{void *}. If @code{ARGS_GROW_DOWNWARD}, it
4932 returns the address of the location above the first anonymous stack
4933 argument. Use it in @code{va_start} to initialize the pointer for
4934 fetching arguments from the stack. Also use it in @code{va_start} to
4935 verify that the second parameter @var{lastarg} is the last named argument
4936 of the current function.
4939 @defmac __builtin_classify_type (@var{object})
4940 Since each machine has its own conventions for which data types are
4941 passed in which kind of register, your implementation of @code{va_arg}
4942 has to embody these conventions. The easiest way to categorize the
4943 specified data type is to use @code{__builtin_classify_type} together
4944 with @code{sizeof} and @code{__alignof__}.
4946 @code{__builtin_classify_type} ignores the value of @var{object},
4947 considering only its data type. It returns an integer describing what
4948 kind of type that is---integer, floating, pointer, structure, and so on.
4950 The file @file{typeclass.h} defines an enumeration that you can use to
4951 interpret the values of @code{__builtin_classify_type}.
4954 These machine description macros help implement varargs:
4956 @hook TARGET_EXPAND_BUILTIN_SAVEREGS
4957 If defined, this hook produces the machine-specific code for a call to
4958 @code{__builtin_saveregs}. This code will be moved to the very
4959 beginning of the function, before any parameter access are made. The
4960 return value of this function should be an RTX that contains the value
4961 to use as the return of @code{__builtin_saveregs}.
4964 @hook TARGET_SETUP_INCOMING_VARARGS
4965 This target hook offers an alternative to using
4966 @code{__builtin_saveregs} and defining the hook
4967 @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. Use it to store the anonymous
4968 register arguments into the stack so that all the arguments appear to
4969 have been passed consecutively on the stack. Once this is done, you can
4970 use the standard implementation of varargs that works for machines that
4971 pass all their arguments on the stack.
4973 The argument @var{args_so_far} points to the @code{CUMULATIVE_ARGS} data
4974 structure, containing the values that are obtained after processing the
4975 named arguments. The arguments @var{mode} and @var{type} describe the
4976 last named argument---its machine mode and its data type as a tree node.
4978 The target hook should do two things: first, push onto the stack all the
4979 argument registers @emph{not} used for the named arguments, and second,
4980 store the size of the data thus pushed into the @code{int}-valued
4981 variable pointed to by @var{pretend_args_size}. The value that you
4982 store here will serve as additional offset for setting up the stack
4985 Because you must generate code to push the anonymous arguments at
4986 compile time without knowing their data types,
4987 @code{TARGET_SETUP_INCOMING_VARARGS} is only useful on machines that
4988 have just a single category of argument register and use it uniformly
4991 If the argument @var{second_time} is nonzero, it means that the
4992 arguments of the function are being analyzed for the second time. This
4993 happens for an inline function, which is not actually compiled until the
4994 end of the source file. The hook @code{TARGET_SETUP_INCOMING_VARARGS} should
4995 not generate any instructions in this case.
4998 @hook TARGET_STRICT_ARGUMENT_NAMING
4999 Define this hook to return @code{true} if the location where a function
5000 argument is passed depends on whether or not it is a named argument.
5002 This hook controls how the @var{named} argument to @code{TARGET_FUNCTION_ARG}
5003 is set for varargs and stdarg functions. If this hook returns
5004 @code{true}, the @var{named} argument is always true for named
5005 arguments, and false for unnamed arguments. If it returns @code{false},
5006 but @code{TARGET_PRETEND_OUTGOING_VARARGS_NAMED} returns @code{true},
5007 then all arguments are treated as named. Otherwise, all named arguments
5008 except the last are treated as named.
5010 You need not define this hook if it always returns @code{false}.
5013 @hook TARGET_PRETEND_OUTGOING_VARARGS_NAMED
5014 If you need to conditionally change ABIs so that one works with
5015 @code{TARGET_SETUP_INCOMING_VARARGS}, but the other works like neither
5016 @code{TARGET_SETUP_INCOMING_VARARGS} nor @code{TARGET_STRICT_ARGUMENT_NAMING} was
5017 defined, then define this hook to return @code{true} if
5018 @code{TARGET_SETUP_INCOMING_VARARGS} is used, @code{false} otherwise.
5019 Otherwise, you should not define this hook.
5023 @section Trampolines for Nested Functions
5024 @cindex trampolines for nested functions
5025 @cindex nested functions, trampolines for
5027 A @dfn{trampoline} is a small piece of code that is created at run time
5028 when the address of a nested function is taken. It normally resides on
5029 the stack, in the stack frame of the containing function. These macros
5030 tell GCC how to generate code to allocate and initialize a
5033 The instructions in the trampoline must do two things: load a constant
5034 address into the static chain register, and jump to the real address of
5035 the nested function. On CISC machines such as the m68k, this requires
5036 two instructions, a move immediate and a jump. Then the two addresses
5037 exist in the trampoline as word-long immediate operands. On RISC
5038 machines, it is often necessary to load each address into a register in
5039 two parts. Then pieces of each address form separate immediate
5042 The code generated to initialize the trampoline must store the variable
5043 parts---the static chain value and the function address---into the
5044 immediate operands of the instructions. On a CISC machine, this is
5045 simply a matter of copying each address to a memory reference at the
5046 proper offset from the start of the trampoline. On a RISC machine, it
5047 may be necessary to take out pieces of the address and store them
5050 @hook TARGET_ASM_TRAMPOLINE_TEMPLATE
5051 This hook is called by @code{assemble_trampoline_template} to output,
5052 on the stream @var{f}, assembler code for a block of data that contains
5053 the constant parts of a trampoline. This code should not include a
5054 label---the label is taken care of automatically.
5056 If you do not define this hook, it means no template is needed
5057 for the target. Do not define this hook on systems where the block move
5058 code to copy the trampoline into place would be larger than the code
5059 to generate it on the spot.
5062 @defmac TRAMPOLINE_SECTION
5063 Return the section into which the trampoline template is to be placed
5064 (@pxref{Sections}). The default value is @code{readonly_data_section}.
5067 @defmac TRAMPOLINE_SIZE
5068 A C expression for the size in bytes of the trampoline, as an integer.
5071 @defmac TRAMPOLINE_ALIGNMENT
5072 Alignment required for trampolines, in bits.
5074 If you don't define this macro, the value of @code{FUNCTION_ALIGNMENT}
5075 is used for aligning trampolines.
5078 @hook TARGET_TRAMPOLINE_INIT
5079 This hook is called to initialize a trampoline.
5080 @var{m_tramp} is an RTX for the memory block for the trampoline; @var{fndecl}
5081 is the @code{FUNCTION_DECL} for the nested function; @var{static_chain} is an
5082 RTX for the static chain value that should be passed to the function
5085 If the target defines @code{TARGET_ASM_TRAMPOLINE_TEMPLATE}, then the
5086 first thing this hook should do is emit a block move into @var{m_tramp}
5087 from the memory block returned by @code{assemble_trampoline_template}.
5088 Note that the block move need only cover the constant parts of the
5089 trampoline. If the target isolates the variable parts of the trampoline
5090 to the end, not all @code{TRAMPOLINE_SIZE} bytes need be copied.
5092 If the target requires any other actions, such as flushing caches or
5093 enabling stack execution, these actions should be performed after
5094 initializing the trampoline proper.
5097 @hook TARGET_TRAMPOLINE_ADJUST_ADDRESS
5098 This hook should perform any machine-specific adjustment in
5099 the address of the trampoline. Its argument contains the address of the
5100 memory block that was passed to @code{TARGET_TRAMPOLINE_INIT}. In case
5101 the address to be used for a function call should be different from the
5102 address at which the template was stored, the different address should
5103 be returned; otherwise @var{addr} should be returned unchanged.
5104 If this hook is not defined, @var{addr} will be used for function calls.
5107 Implementing trampolines is difficult on many machines because they have
5108 separate instruction and data caches. Writing into a stack location
5109 fails to clear the memory in the instruction cache, so when the program
5110 jumps to that location, it executes the old contents.
5112 Here are two possible solutions. One is to clear the relevant parts of
5113 the instruction cache whenever a trampoline is set up. The other is to
5114 make all trampolines identical, by having them jump to a standard
5115 subroutine. The former technique makes trampoline execution faster; the
5116 latter makes initialization faster.
5118 To clear the instruction cache when a trampoline is initialized, define
5119 the following macro.
5121 @defmac CLEAR_INSN_CACHE (@var{beg}, @var{end})
5122 If defined, expands to a C expression clearing the @emph{instruction
5123 cache} in the specified interval. The definition of this macro would
5124 typically be a series of @code{asm} statements. Both @var{beg} and
5125 @var{end} are both pointer expressions.
5128 The operating system may also require the stack to be made executable
5129 before calling the trampoline. To implement this requirement, define
5130 the following macro.
5132 @defmac ENABLE_EXECUTE_STACK
5133 Define this macro if certain operations must be performed before executing
5134 code located on the stack. The macro should expand to a series of C
5135 file-scope constructs (e.g.@: functions) and provide a unique entry point
5136 named @code{__enable_execute_stack}. The target is responsible for
5137 emitting calls to the entry point in the code, for example from the
5138 @code{TARGET_TRAMPOLINE_INIT} hook.
5141 To use a standard subroutine, define the following macro. In addition,
5142 you must make sure that the instructions in a trampoline fill an entire
5143 cache line with identical instructions, or else ensure that the
5144 beginning of the trampoline code is always aligned at the same point in
5145 its cache line. Look in @file{m68k.h} as a guide.
5147 @defmac TRANSFER_FROM_TRAMPOLINE
5148 Define this macro if trampolines need a special subroutine to do their
5149 work. The macro should expand to a series of @code{asm} statements
5150 which will be compiled with GCC@. They go in a library function named
5151 @code{__transfer_from_trampoline}.
5153 If you need to avoid executing the ordinary prologue code of a compiled
5154 C function when you jump to the subroutine, you can do so by placing a
5155 special label of your own in the assembler code. Use one @code{asm}
5156 statement to generate an assembler label, and another to make the label
5157 global. Then trampolines can use that label to jump directly to your
5158 special assembler code.
5162 @section Implicit Calls to Library Routines
5163 @cindex library subroutine names
5164 @cindex @file{libgcc.a}
5166 @c prevent bad page break with this line
5167 Here is an explanation of implicit calls to library routines.
5169 @defmac DECLARE_LIBRARY_RENAMES
5170 This macro, if defined, should expand to a piece of C code that will get
5171 expanded when compiling functions for libgcc.a. It can be used to
5172 provide alternate names for GCC's internal library functions if there
5173 are ABI-mandated names that the compiler should provide.
5176 @findex set_optab_libfunc
5177 @findex init_one_libfunc
5178 @hook TARGET_INIT_LIBFUNCS
5179 This hook should declare additional library routines or rename
5180 existing ones, using the functions @code{set_optab_libfunc} and
5181 @code{init_one_libfunc} defined in @file{optabs.c}.
5182 @code{init_optabs} calls this macro after initializing all the normal
5185 The default is to do nothing. Most ports don't need to define this hook.
5188 @defmac FLOAT_LIB_COMPARE_RETURNS_BOOL (@var{mode}, @var{comparison})
5189 This macro should return @code{true} if the library routine that
5190 implements the floating point comparison operator @var{comparison} in
5191 mode @var{mode} will return a boolean, and @var{false} if it will
5194 GCC's own floating point libraries return tristates from the
5195 comparison operators, so the default returns false always. Most ports
5196 don't need to define this macro.
5199 @defmac TARGET_LIB_INT_CMP_BIASED
5200 This macro should evaluate to @code{true} if the integer comparison
5201 functions (like @code{__cmpdi2}) return 0 to indicate that the first
5202 operand is smaller than the second, 1 to indicate that they are equal,
5203 and 2 to indicate that the first operand is greater than the second.
5204 If this macro evaluates to @code{false} the comparison functions return
5205 @minus{}1, 0, and 1 instead of 0, 1, and 2. If the target uses the routines
5206 in @file{libgcc.a}, you do not need to define this macro.
5209 @cindex @code{EDOM}, implicit usage
5212 The value of @code{EDOM} on the target machine, as a C integer constant
5213 expression. If you don't define this macro, GCC does not attempt to
5214 deposit the value of @code{EDOM} into @code{errno} directly. Look in
5215 @file{/usr/include/errno.h} to find the value of @code{EDOM} on your
5218 If you do not define @code{TARGET_EDOM}, then compiled code reports
5219 domain errors by calling the library function and letting it report the
5220 error. If mathematical functions on your system use @code{matherr} when
5221 there is an error, then you should leave @code{TARGET_EDOM} undefined so
5222 that @code{matherr} is used normally.
5225 @cindex @code{errno}, implicit usage
5226 @defmac GEN_ERRNO_RTX
5227 Define this macro as a C expression to create an rtl expression that
5228 refers to the global ``variable'' @code{errno}. (On certain systems,
5229 @code{errno} may not actually be a variable.) If you don't define this
5230 macro, a reasonable default is used.
5233 @cindex C99 math functions, implicit usage
5234 @defmac TARGET_C99_FUNCTIONS
5235 When this macro is nonzero, GCC will implicitly optimize @code{sin} calls into
5236 @code{sinf} and similarly for other functions defined by C99 standard. The
5237 default is zero because a number of existing systems lack support for these
5238 functions in their runtime so this macro needs to be redefined to one on
5239 systems that do support the C99 runtime.
5242 @cindex sincos math function, implicit usage
5243 @defmac TARGET_HAS_SINCOS
5244 When this macro is nonzero, GCC will implicitly optimize calls to @code{sin}
5245 and @code{cos} with the same argument to a call to @code{sincos}. The
5246 default is zero. The target has to provide the following functions:
5248 void sincos(double x, double *sin, double *cos);
5249 void sincosf(float x, float *sin, float *cos);
5250 void sincosl(long double x, long double *sin, long double *cos);
5254 @defmac NEXT_OBJC_RUNTIME
5255 Define this macro to generate code for Objective-C message sending using
5256 the calling convention of the NeXT system. This calling convention
5257 involves passing the object, the selector and the method arguments all
5258 at once to the method-lookup library function.
5260 The default calling convention passes just the object and the selector
5261 to the lookup function, which returns a pointer to the method.
5264 @node Addressing Modes
5265 @section Addressing Modes
5266 @cindex addressing modes
5268 @c prevent bad page break with this line
5269 This is about addressing modes.
5271 @defmac HAVE_PRE_INCREMENT
5272 @defmacx HAVE_PRE_DECREMENT
5273 @defmacx HAVE_POST_INCREMENT
5274 @defmacx HAVE_POST_DECREMENT
5275 A C expression that is nonzero if the machine supports pre-increment,
5276 pre-decrement, post-increment, or post-decrement addressing respectively.
5279 @defmac HAVE_PRE_MODIFY_DISP
5280 @defmacx HAVE_POST_MODIFY_DISP
5281 A C expression that is nonzero if the machine supports pre- or
5282 post-address side-effect generation involving constants other than
5283 the size of the memory operand.
5286 @defmac HAVE_PRE_MODIFY_REG
5287 @defmacx HAVE_POST_MODIFY_REG
5288 A C expression that is nonzero if the machine supports pre- or
5289 post-address side-effect generation involving a register displacement.
5292 @defmac CONSTANT_ADDRESS_P (@var{x})
5293 A C expression that is 1 if the RTX @var{x} is a constant which
5294 is a valid address. On most machines the default definition of
5295 @code{(CONSTANT_P (@var{x}) && GET_CODE (@var{x}) != CONST_DOUBLE)}
5296 is acceptable, but a few machines are more restrictive as to which
5297 constant addresses are supported.
5300 @defmac CONSTANT_P (@var{x})
5301 @code{CONSTANT_P}, which is defined by target-independent code,
5302 accepts integer-values expressions whose values are not explicitly
5303 known, such as @code{symbol_ref}, @code{label_ref}, and @code{high}
5304 expressions and @code{const} arithmetic expressions, in addition to
5305 @code{const_int} and @code{const_double} expressions.
5308 @defmac MAX_REGS_PER_ADDRESS
5309 A number, the maximum number of registers that can appear in a valid
5310 memory address. Note that it is up to you to specify a value equal to
5311 the maximum number that @code{TARGET_LEGITIMATE_ADDRESS_P} would ever
5315 @hook TARGET_LEGITIMATE_ADDRESS_P
5316 A function that returns whether @var{x} (an RTX) is a legitimate memory
5317 address on the target machine for a memory operand of mode @var{mode}.
5319 Legitimate addresses are defined in two variants: a strict variant and a
5320 non-strict one. The @var{strict} parameter chooses which variant is
5321 desired by the caller.
5323 The strict variant is used in the reload pass. It must be defined so
5324 that any pseudo-register that has not been allocated a hard register is
5325 considered a memory reference. This is because in contexts where some
5326 kind of register is required, a pseudo-register with no hard register
5327 must be rejected. For non-hard registers, the strict variant should look
5328 up the @code{reg_renumber} array; it should then proceed using the hard
5329 register number in the array, or treat the pseudo as a memory reference
5330 if the array holds @code{-1}.
5332 The non-strict variant is used in other passes. It must be defined to
5333 accept all pseudo-registers in every context where some kind of
5334 register is required.
5336 Normally, constant addresses which are the sum of a @code{symbol_ref}
5337 and an integer are stored inside a @code{const} RTX to mark them as
5338 constant. Therefore, there is no need to recognize such sums
5339 specifically as legitimate addresses. Normally you would simply
5340 recognize any @code{const} as legitimate.
5342 Usually @code{PRINT_OPERAND_ADDRESS} is not prepared to handle constant
5343 sums that are not marked with @code{const}. It assumes that a naked
5344 @code{plus} indicates indexing. If so, then you @emph{must} reject such
5345 naked constant sums as illegitimate addresses, so that none of them will
5346 be given to @code{PRINT_OPERAND_ADDRESS}.
5348 @cindex @code{TARGET_ENCODE_SECTION_INFO} and address validation
5349 On some machines, whether a symbolic address is legitimate depends on
5350 the section that the address refers to. On these machines, define the
5351 target hook @code{TARGET_ENCODE_SECTION_INFO} to store the information
5352 into the @code{symbol_ref}, and then check for it here. When you see a
5353 @code{const}, you will have to look inside it to find the
5354 @code{symbol_ref} in order to determine the section. @xref{Assembler
5357 @cindex @code{GO_IF_LEGITIMATE_ADDRESS}
5358 Some ports are still using a deprecated legacy substitute for
5359 this hook, the @code{GO_IF_LEGITIMATE_ADDRESS} macro. This macro
5363 #define GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label})
5367 and should @code{goto @var{label}} if the address @var{x} is a valid
5368 address on the target machine for a memory operand of mode @var{mode}.
5370 @findex REG_OK_STRICT
5371 Compiler source files that want to use the strict variant of this
5372 macro define the macro @code{REG_OK_STRICT}. You should use an
5373 @code{#ifdef REG_OK_STRICT} conditional to define the strict variant in
5374 that case and the non-strict variant otherwise.
5376 Using the hook is usually simpler because it limits the number of
5377 files that are recompiled when changes are made.
5380 @defmac TARGET_MEM_CONSTRAINT
5381 A single character to be used instead of the default @code{'m'}
5382 character for general memory addresses. This defines the constraint
5383 letter which matches the memory addresses accepted by
5384 @code{TARGET_LEGITIMATE_ADDRESS_P}. Define this macro if you want to
5385 support new address formats in your back end without changing the
5386 semantics of the @code{'m'} constraint. This is necessary in order to
5387 preserve functionality of inline assembly constructs using the
5388 @code{'m'} constraint.
5391 @defmac FIND_BASE_TERM (@var{x})
5392 A C expression to determine the base term of address @var{x},
5393 or to provide a simplified version of @var{x} from which @file{alias.c}
5394 can easily find the base term. This macro is used in only two places:
5395 @code{find_base_value} and @code{find_base_term} in @file{alias.c}.
5397 It is always safe for this macro to not be defined. It exists so
5398 that alias analysis can understand machine-dependent addresses.
5400 The typical use of this macro is to handle addresses containing
5401 a label_ref or symbol_ref within an UNSPEC@.
5404 @hook TARGET_LEGITIMIZE_ADDRESS
5405 This hook is given an invalid memory address @var{x} for an
5406 operand of mode @var{mode} and should try to return a valid memory
5409 @findex break_out_memory_refs
5410 @var{x} will always be the result of a call to @code{break_out_memory_refs},
5411 and @var{oldx} will be the operand that was given to that function to produce
5414 The code of the hook should not alter the substructure of
5415 @var{x}. If it transforms @var{x} into a more legitimate form, it
5416 should return the new @var{x}.
5418 It is not necessary for this hook to come up with a legitimate address.
5419 The compiler has standard ways of doing so in all cases. In fact, it
5420 is safe to omit this hook or make it return @var{x} if it cannot find
5421 a valid way to legitimize the address. But often a machine-dependent
5422 strategy can generate better code.
5425 @defmac LEGITIMIZE_RELOAD_ADDRESS (@var{x}, @var{mode}, @var{opnum}, @var{type}, @var{ind_levels}, @var{win})
5426 A C compound statement that attempts to replace @var{x}, which is an address
5427 that needs reloading, with a valid memory address for an operand of mode
5428 @var{mode}. @var{win} will be a C statement label elsewhere in the code.
5429 It is not necessary to define this macro, but it might be useful for
5430 performance reasons.
5432 For example, on the i386, it is sometimes possible to use a single
5433 reload register instead of two by reloading a sum of two pseudo
5434 registers into a register. On the other hand, for number of RISC
5435 processors offsets are limited so that often an intermediate address
5436 needs to be generated in order to address a stack slot. By defining
5437 @code{LEGITIMIZE_RELOAD_ADDRESS} appropriately, the intermediate addresses
5438 generated for adjacent some stack slots can be made identical, and thus
5441 @emph{Note}: This macro should be used with caution. It is necessary
5442 to know something of how reload works in order to effectively use this,
5443 and it is quite easy to produce macros that build in too much knowledge
5444 of reload internals.
5446 @emph{Note}: This macro must be able to reload an address created by a
5447 previous invocation of this macro. If it fails to handle such addresses
5448 then the compiler may generate incorrect code or abort.
5451 The macro definition should use @code{push_reload} to indicate parts that
5452 need reloading; @var{opnum}, @var{type} and @var{ind_levels} are usually
5453 suitable to be passed unaltered to @code{push_reload}.
5455 The code generated by this macro must not alter the substructure of
5456 @var{x}. If it transforms @var{x} into a more legitimate form, it
5457 should assign @var{x} (which will always be a C variable) a new value.
5458 This also applies to parts that you change indirectly by calling
5461 @findex strict_memory_address_p
5462 The macro definition may use @code{strict_memory_address_p} to test if
5463 the address has become legitimate.
5466 If you want to change only a part of @var{x}, one standard way of doing
5467 this is to use @code{copy_rtx}. Note, however, that it unshares only a
5468 single level of rtl. Thus, if the part to be changed is not at the
5469 top level, you'll need to replace first the top level.
5470 It is not necessary for this macro to come up with a legitimate
5471 address; but often a machine-dependent strategy can generate better code.
5474 @hook TARGET_MODE_DEPENDENT_ADDRESS_P
5475 This hook returns @code{true} if memory address @var{addr} can have
5476 different meanings depending on the machine mode of the memory
5477 reference it is used for or if the address is valid for some modes
5480 Autoincrement and autodecrement addresses typically have mode-dependent
5481 effects because the amount of the increment or decrement is the size
5482 of the operand being addressed. Some machines have other mode-dependent
5483 addresses. Many RISC machines have no mode-dependent addresses.
5485 You may assume that @var{addr} is a valid address for the machine.
5487 The default version of this hook returns @code{false}.
5490 @defmac GO_IF_MODE_DEPENDENT_ADDRESS (@var{addr}, @var{label})
5491 A C statement or compound statement with a conditional @code{goto
5492 @var{label};} executed if memory address @var{x} (an RTX) can have
5493 different meanings depending on the machine mode of the memory
5494 reference it is used for or if the address is valid for some modes
5497 Autoincrement and autodecrement addresses typically have mode-dependent
5498 effects because the amount of the increment or decrement is the size
5499 of the operand being addressed. Some machines have other mode-dependent
5500 addresses. Many RISC machines have no mode-dependent addresses.
5502 You may assume that @var{addr} is a valid address for the machine.
5504 These are obsolete macros, replaced by the
5505 @code{TARGET_MODE_DEPENDENT_ADDRESS_P} target hook.
5508 @defmac LEGITIMATE_CONSTANT_P (@var{x})
5509 A C expression that is nonzero if @var{x} is a legitimate constant for
5510 an immediate operand on the target machine. You can assume that
5511 @var{x} satisfies @code{CONSTANT_P}, so you need not check this. In fact,
5512 @samp{1} is a suitable definition for this macro on machines where
5513 anything @code{CONSTANT_P} is valid.
5516 @hook TARGET_DELEGITIMIZE_ADDRESS
5517 This hook is used to undo the possibly obfuscating effects of the
5518 @code{LEGITIMIZE_ADDRESS} and @code{LEGITIMIZE_RELOAD_ADDRESS} target
5519 macros. Some backend implementations of these macros wrap symbol
5520 references inside an @code{UNSPEC} rtx to represent PIC or similar
5521 addressing modes. This target hook allows GCC's optimizers to understand
5522 the semantics of these opaque @code{UNSPEC}s by converting them back
5523 into their original form.
5526 @hook TARGET_CANNOT_FORCE_CONST_MEM
5527 This hook should return true if @var{x} is of a form that cannot (or
5528 should not) be spilled to the constant pool. The default version of
5529 this hook returns false.
5531 The primary reason to define this hook is to prevent reload from
5532 deciding that a non-legitimate constant would be better reloaded
5533 from the constant pool instead of spilling and reloading a register
5534 holding the constant. This restriction is often true of addresses
5535 of TLS symbols for various targets.
5538 @hook TARGET_USE_BLOCKS_FOR_CONSTANT_P
5539 This hook should return true if pool entries for constant @var{x} can
5540 be placed in an @code{object_block} structure. @var{mode} is the mode
5543 The default version returns false for all constants.
5546 @hook TARGET_BUILTIN_RECIPROCAL
5547 This hook should return the DECL of a function that implements reciprocal of
5548 the builtin function with builtin function code @var{fn}, or
5549 @code{NULL_TREE} if such a function is not available. @var{md_fn} is true
5550 when @var{fn} is a code of a machine-dependent builtin function. When
5551 @var{sqrt} is true, additional optimizations that apply only to the reciprocal
5552 of a square root function are performed, and only reciprocals of @code{sqrt}
5556 @hook TARGET_VECTORIZE_BUILTIN_MASK_FOR_LOAD
5557 This hook should return the DECL of a function @var{f} that given an
5558 address @var{addr} as an argument returns a mask @var{m} that can be
5559 used to extract from two vectors the relevant data that resides in
5560 @var{addr} in case @var{addr} is not properly aligned.
5562 The autovectorizer, when vectorizing a load operation from an address
5563 @var{addr} that may be unaligned, will generate two vector loads from
5564 the two aligned addresses around @var{addr}. It then generates a
5565 @code{REALIGN_LOAD} operation to extract the relevant data from the
5566 two loaded vectors. The first two arguments to @code{REALIGN_LOAD},
5567 @var{v1} and @var{v2}, are the two vectors, each of size @var{VS}, and
5568 the third argument, @var{OFF}, defines how the data will be extracted
5569 from these two vectors: if @var{OFF} is 0, then the returned vector is
5570 @var{v2}; otherwise, the returned vector is composed from the last
5571 @var{VS}-@var{OFF} elements of @var{v1} concatenated to the first
5572 @var{OFF} elements of @var{v2}.
5574 If this hook is defined, the autovectorizer will generate a call
5575 to @var{f} (using the DECL tree that this hook returns) and will
5576 use the return value of @var{f} as the argument @var{OFF} to
5577 @code{REALIGN_LOAD}. Therefore, the mask @var{m} returned by @var{f}
5578 should comply with the semantics expected by @code{REALIGN_LOAD}
5580 If this hook is not defined, then @var{addr} will be used as
5581 the argument @var{OFF} to @code{REALIGN_LOAD}, in which case the low
5582 log2(@var{VS}) @minus{} 1 bits of @var{addr} will be considered.
5585 @hook TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_EVEN
5586 This hook should return the DECL of a function @var{f} that implements
5587 widening multiplication of the even elements of two input vectors of type @var{x}.
5589 If this hook is defined, the autovectorizer will use it along with the
5590 @code{TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_ODD} target hook when vectorizing
5591 widening multiplication in cases that the order of the results does not have to be
5592 preserved (e.g.@: used only by a reduction computation). Otherwise, the
5593 @code{widen_mult_hi/lo} idioms will be used.
5596 @hook TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_ODD
5597 This hook should return the DECL of a function @var{f} that implements
5598 widening multiplication of the odd elements of two input vectors of type @var{x}.
5600 If this hook is defined, the autovectorizer will use it along with the
5601 @code{TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_EVEN} target hook when vectorizing
5602 widening multiplication in cases that the order of the results does not have to be
5603 preserved (e.g.@: used only by a reduction computation). Otherwise, the
5604 @code{widen_mult_hi/lo} idioms will be used.
5607 @hook TARGET_VECTORIZE_BUILTIN_VECTORIZATION_COST
5608 Returns cost of different scalar or vector statements for vectorization cost model.
5609 For vector memory operations the cost may depend on type (@var{vectype}) and
5610 misalignment value (@var{misalign}).
5613 @hook TARGET_VECTORIZE_VECTOR_ALIGNMENT_REACHABLE
5614 Return true if vector alignment is reachable (by peeling N iterations) for the given type.
5617 @hook TARGET_VECTORIZE_BUILTIN_VEC_PERM
5618 Target builtin that implements vector permute.
5621 @hook TARGET_VECTORIZE_BUILTIN_VEC_PERM_OK
5622 Return true if a vector created for @code{builtin_vec_perm} is valid.
5625 @hook TARGET_VECTORIZE_BUILTIN_CONVERSION
5626 This hook should return the DECL of a function that implements conversion of the
5627 input vector of type @var{src_type} to type @var{dest_type}.
5628 The value of @var{code} is one of the enumerators in @code{enum tree_code} and
5629 specifies how the conversion is to be applied
5630 (truncation, rounding, etc.).
5632 If this hook is defined, the autovectorizer will use the
5633 @code{TARGET_VECTORIZE_BUILTIN_CONVERSION} target hook when vectorizing
5634 conversion. Otherwise, it will return @code{NULL_TREE}.
5637 @hook TARGET_VECTORIZE_BUILTIN_VECTORIZED_FUNCTION
5638 This hook should return the decl of a function that implements the
5639 vectorized variant of the builtin function with builtin function code
5640 @var{code} or @code{NULL_TREE} if such a function is not available.
5641 The value of @var{fndecl} is the builtin function declaration. The
5642 return type of the vectorized function shall be of vector type
5643 @var{vec_type_out} and the argument types should be @var{vec_type_in}.
5646 @hook TARGET_VECTORIZE_SUPPORT_VECTOR_MISALIGNMENT
5647 This hook should return true if the target supports misaligned vector
5648 store/load of a specific factor denoted in the @var{misalignment}
5649 parameter. The vector store/load should be of machine mode @var{mode} and
5650 the elements in the vectors should be of type @var{type}. @var{is_packed}
5651 parameter is true if the memory access is defined in a packed struct.
5654 @hook TARGET_VECTORIZE_PREFERRED_SIMD_MODE
5655 This hook should return the preferred mode for vectorizing scalar
5656 mode @var{mode}. The default is
5657 equal to @code{word_mode}, because the vectorizer can do some
5658 transformations even in absence of specialized @acronym{SIMD} hardware.
5661 @hook TARGET_VECTORIZE_AUTOVECTORIZE_VECTOR_SIZES
5662 This hook should return a mask of sizes that should be iterated over
5663 after trying to autovectorize using the vector size derived from the
5664 mode returned by @code{TARGET_VECTORIZE_PREFERRED_SIMD_MODE}.
5665 The default is zero which means to not iterate over other vector sizes.
5668 @node Anchored Addresses
5669 @section Anchored Addresses
5670 @cindex anchored addresses
5671 @cindex @option{-fsection-anchors}
5673 GCC usually addresses every static object as a separate entity.
5674 For example, if we have:
5678 int foo (void) @{ return a + b + c; @}
5681 the code for @code{foo} will usually calculate three separate symbolic
5682 addresses: those of @code{a}, @code{b} and @code{c}. On some targets,
5683 it would be better to calculate just one symbolic address and access
5684 the three variables relative to it. The equivalent pseudocode would
5690 register int *xr = &x;
5691 return xr[&a - &x] + xr[&b - &x] + xr[&c - &x];
5695 (which isn't valid C). We refer to shared addresses like @code{x} as
5696 ``section anchors''. Their use is controlled by @option{-fsection-anchors}.
5698 The hooks below describe the target properties that GCC needs to know
5699 in order to make effective use of section anchors. It won't use
5700 section anchors at all unless either @code{TARGET_MIN_ANCHOR_OFFSET}
5701 or @code{TARGET_MAX_ANCHOR_OFFSET} is set to a nonzero value.
5703 @hook TARGET_MIN_ANCHOR_OFFSET
5704 The minimum offset that should be applied to a section anchor.
5705 On most targets, it should be the smallest offset that can be
5706 applied to a base register while still giving a legitimate address
5707 for every mode. The default value is 0.
5710 @hook TARGET_MAX_ANCHOR_OFFSET
5711 Like @code{TARGET_MIN_ANCHOR_OFFSET}, but the maximum (inclusive)
5712 offset that should be applied to section anchors. The default
5716 @hook TARGET_ASM_OUTPUT_ANCHOR
5717 Write the assembly code to define section anchor @var{x}, which is a
5718 @code{SYMBOL_REF} for which @samp{SYMBOL_REF_ANCHOR_P (@var{x})} is true.
5719 The hook is called with the assembly output position set to the beginning
5720 of @code{SYMBOL_REF_BLOCK (@var{x})}.
5722 If @code{ASM_OUTPUT_DEF} is available, the hook's default definition uses
5723 it to define the symbol as @samp{. + SYMBOL_REF_BLOCK_OFFSET (@var{x})}.
5724 If @code{ASM_OUTPUT_DEF} is not available, the hook's default definition
5725 is @code{NULL}, which disables the use of section anchors altogether.
5728 @hook TARGET_USE_ANCHORS_FOR_SYMBOL_P
5729 Return true if GCC should attempt to use anchors to access @code{SYMBOL_REF}
5730 @var{x}. You can assume @samp{SYMBOL_REF_HAS_BLOCK_INFO_P (@var{x})} and
5731 @samp{!SYMBOL_REF_ANCHOR_P (@var{x})}.
5733 The default version is correct for most targets, but you might need to
5734 intercept this hook to handle things like target-specific attributes
5735 or target-specific sections.
5738 @node Condition Code
5739 @section Condition Code Status
5740 @cindex condition code status
5742 The macros in this section can be split in two families, according to the
5743 two ways of representing condition codes in GCC.
5745 The first representation is the so called @code{(cc0)} representation
5746 (@pxref{Jump Patterns}), where all instructions can have an implicit
5747 clobber of the condition codes. The second is the condition code
5748 register representation, which provides better schedulability for
5749 architectures that do have a condition code register, but on which
5750 most instructions do not affect it. The latter category includes
5753 The implicit clobbering poses a strong restriction on the placement of
5754 the definition and use of the condition code, which need to be in adjacent
5755 insns for machines using @code{(cc0)}. This can prevent important
5756 optimizations on some machines. For example, on the IBM RS/6000, there
5757 is a delay for taken branches unless the condition code register is set
5758 three instructions earlier than the conditional branch. The instruction
5759 scheduler cannot perform this optimization if it is not permitted to
5760 separate the definition and use of the condition code register.
5762 For this reason, it is possible and suggested to use a register to
5763 represent the condition code for new ports. If there is a specific
5764 condition code register in the machine, use a hard register. If the
5765 condition code or comparison result can be placed in any general register,
5766 or if there are multiple condition registers, use a pseudo register.
5767 Registers used to store the condition code value will usually have a mode
5768 that is in class @code{MODE_CC}.
5770 Alternatively, you can use @code{BImode} if the comparison operator is
5771 specified already in the compare instruction. In this case, you are not
5772 interested in most macros in this section.
5775 * CC0 Condition Codes:: Old style representation of condition codes.
5776 * MODE_CC Condition Codes:: Modern representation of condition codes.
5777 * Cond Exec Macros:: Macros to control conditional execution.
5780 @node CC0 Condition Codes
5781 @subsection Representation of condition codes using @code{(cc0)}
5785 The file @file{conditions.h} defines a variable @code{cc_status} to
5786 describe how the condition code was computed (in case the interpretation of
5787 the condition code depends on the instruction that it was set by). This
5788 variable contains the RTL expressions on which the condition code is
5789 currently based, and several standard flags.
5791 Sometimes additional machine-specific flags must be defined in the machine
5792 description header file. It can also add additional machine-specific
5793 information by defining @code{CC_STATUS_MDEP}.
5795 @defmac CC_STATUS_MDEP
5796 C code for a data type which is used for declaring the @code{mdep}
5797 component of @code{cc_status}. It defaults to @code{int}.
5799 This macro is not used on machines that do not use @code{cc0}.
5802 @defmac CC_STATUS_MDEP_INIT
5803 A C expression to initialize the @code{mdep} field to ``empty''.
5804 The default definition does nothing, since most machines don't use
5805 the field anyway. If you want to use the field, you should probably
5806 define this macro to initialize it.
5808 This macro is not used on machines that do not use @code{cc0}.
5811 @defmac NOTICE_UPDATE_CC (@var{exp}, @var{insn})
5812 A C compound statement to set the components of @code{cc_status}
5813 appropriately for an insn @var{insn} whose body is @var{exp}. It is
5814 this macro's responsibility to recognize insns that set the condition
5815 code as a byproduct of other activity as well as those that explicitly
5818 This macro is not used on machines that do not use @code{cc0}.
5820 If there are insns that do not set the condition code but do alter
5821 other machine registers, this macro must check to see whether they
5822 invalidate the expressions that the condition code is recorded as
5823 reflecting. For example, on the 68000, insns that store in address
5824 registers do not set the condition code, which means that usually
5825 @code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such
5826 insns. But suppose that the previous insn set the condition code
5827 based on location @samp{a4@@(102)} and the current insn stores a new
5828 value in @samp{a4}. Although the condition code is not changed by
5829 this, it will no longer be true that it reflects the contents of
5830 @samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter
5831 @code{cc_status} in this case to say that nothing is known about the
5832 condition code value.
5834 The definition of @code{NOTICE_UPDATE_CC} must be prepared to deal
5835 with the results of peephole optimization: insns whose patterns are
5836 @code{parallel} RTXs containing various @code{reg}, @code{mem} or
5837 constants which are just the operands. The RTL structure of these
5838 insns is not sufficient to indicate what the insns actually do. What
5839 @code{NOTICE_UPDATE_CC} should do when it sees one is just to run
5840 @code{CC_STATUS_INIT}.
5842 A possible definition of @code{NOTICE_UPDATE_CC} is to call a function
5843 that looks at an attribute (@pxref{Insn Attributes}) named, for example,
5844 @samp{cc}. This avoids having detailed information about patterns in
5845 two places, the @file{md} file and in @code{NOTICE_UPDATE_CC}.
5848 @node MODE_CC Condition Codes
5849 @subsection Representation of condition codes using registers
5853 @defmac SELECT_CC_MODE (@var{op}, @var{x}, @var{y})
5854 On many machines, the condition code may be produced by other instructions
5855 than compares, for example the branch can use directly the condition
5856 code set by a subtract instruction. However, on some machines
5857 when the condition code is set this way some bits (such as the overflow
5858 bit) are not set in the same way as a test instruction, so that a different
5859 branch instruction must be used for some conditional branches. When
5860 this happens, use the machine mode of the condition code register to
5861 record different formats of the condition code register. Modes can
5862 also be used to record which compare instruction (e.g. a signed or an
5863 unsigned comparison) produced the condition codes.
5865 If other modes than @code{CCmode} are required, add them to
5866 @file{@var{machine}-modes.def} and define @code{SELECT_CC_MODE} to choose
5867 a mode given an operand of a compare. This is needed because the modes
5868 have to be chosen not only during RTL generation but also, for example,
5869 by instruction combination. The result of @code{SELECT_CC_MODE} should
5870 be consistent with the mode used in the patterns; for example to support
5871 the case of the add on the SPARC discussed above, we have the pattern
5875 [(set (reg:CC_NOOV 0)
5877 (plus:SI (match_operand:SI 0 "register_operand" "%r")
5878 (match_operand:SI 1 "arith_operand" "rI"))
5885 together with a @code{SELECT_CC_MODE} that returns @code{CC_NOOVmode}
5886 for comparisons whose argument is a @code{plus}:
5889 #define SELECT_CC_MODE(OP,X,Y) \
5890 (GET_MODE_CLASS (GET_MODE (X)) == MODE_FLOAT \
5891 ? ((OP == EQ || OP == NE) ? CCFPmode : CCFPEmode) \
5892 : ((GET_CODE (X) == PLUS || GET_CODE (X) == MINUS \
5893 || GET_CODE (X) == NEG) \
5894 ? CC_NOOVmode : CCmode))
5897 Another reason to use modes is to retain information on which operands
5898 were used by the comparison; see @code{REVERSIBLE_CC_MODE} later in
5901 You should define this macro if and only if you define extra CC modes
5902 in @file{@var{machine}-modes.def}.
5905 @defmac CANONICALIZE_COMPARISON (@var{code}, @var{op0}, @var{op1})
5906 On some machines not all possible comparisons are defined, but you can
5907 convert an invalid comparison into a valid one. For example, the Alpha
5908 does not have a @code{GT} comparison, but you can use an @code{LT}
5909 comparison instead and swap the order of the operands.
5911 On such machines, define this macro to be a C statement to do any
5912 required conversions. @var{code} is the initial comparison code
5913 and @var{op0} and @var{op1} are the left and right operands of the
5914 comparison, respectively. You should modify @var{code}, @var{op0}, and
5915 @var{op1} as required.
5917 GCC will not assume that the comparison resulting from this macro is
5918 valid but will see if the resulting insn matches a pattern in the
5921 You need not define this macro if it would never change the comparison
5925 @defmac REVERSIBLE_CC_MODE (@var{mode})
5926 A C expression whose value is one if it is always safe to reverse a
5927 comparison whose mode is @var{mode}. If @code{SELECT_CC_MODE}
5928 can ever return @var{mode} for a floating-point inequality comparison,
5929 then @code{REVERSIBLE_CC_MODE (@var{mode})} must be zero.
5931 You need not define this macro if it would always returns zero or if the
5932 floating-point format is anything other than @code{IEEE_FLOAT_FORMAT}.
5933 For example, here is the definition used on the SPARC, where floating-point
5934 inequality comparisons are always given @code{CCFPEmode}:
5937 #define REVERSIBLE_CC_MODE(MODE) ((MODE) != CCFPEmode)
5941 @defmac REVERSE_CONDITION (@var{code}, @var{mode})
5942 A C expression whose value is reversed condition code of the @var{code} for
5943 comparison done in CC_MODE @var{mode}. The macro is used only in case
5944 @code{REVERSIBLE_CC_MODE (@var{mode})} is nonzero. Define this macro in case
5945 machine has some non-standard way how to reverse certain conditionals. For
5946 instance in case all floating point conditions are non-trapping, compiler may
5947 freely convert unordered compares to ordered one. Then definition may look
5951 #define REVERSE_CONDITION(CODE, MODE) \
5952 ((MODE) != CCFPmode ? reverse_condition (CODE) \
5953 : reverse_condition_maybe_unordered (CODE))
5957 @hook TARGET_FIXED_CONDITION_CODE_REGS
5958 On targets which do not use @code{(cc0)}, and which use a hard
5959 register rather than a pseudo-register to hold condition codes, the
5960 regular CSE passes are often not able to identify cases in which the
5961 hard register is set to a common value. Use this hook to enable a
5962 small pass which optimizes such cases. This hook should return true
5963 to enable this pass, and it should set the integers to which its
5964 arguments point to the hard register numbers used for condition codes.
5965 When there is only one such register, as is true on most systems, the
5966 integer pointed to by @var{p2} should be set to
5967 @code{INVALID_REGNUM}.
5969 The default version of this hook returns false.
5972 @hook TARGET_CC_MODES_COMPATIBLE
5973 On targets which use multiple condition code modes in class
5974 @code{MODE_CC}, it is sometimes the case that a comparison can be
5975 validly done in more than one mode. On such a system, define this
5976 target hook to take two mode arguments and to return a mode in which
5977 both comparisons may be validly done. If there is no such mode,
5978 return @code{VOIDmode}.
5980 The default version of this hook checks whether the modes are the
5981 same. If they are, it returns that mode. If they are different, it
5982 returns @code{VOIDmode}.
5985 @node Cond Exec Macros
5986 @subsection Macros to control conditional execution
5987 @findex conditional execution
5990 There is one macro that may need to be defined for targets
5991 supporting conditional execution, independent of how they
5992 represent conditional branches.
5994 @defmac REVERSE_CONDEXEC_PREDICATES_P (@var{op1}, @var{op2})
5995 A C expression that returns true if the conditional execution predicate
5996 @var{op1}, a comparison operation, is the inverse of @var{op2} and vice
5997 versa. Define this to return 0 if the target has conditional execution
5998 predicates that cannot be reversed safely. There is no need to validate
5999 that the arguments of op1 and op2 are the same, this is done separately.
6000 If no expansion is specified, this macro is defined as follows:
6003 #define REVERSE_CONDEXEC_PREDICATES_P (x, y) \
6004 (GET_CODE ((x)) == reversed_comparison_code ((y), NULL))
6009 @section Describing Relative Costs of Operations
6010 @cindex costs of instructions
6011 @cindex relative costs
6012 @cindex speed of instructions
6014 These macros let you describe the relative speed of various operations
6015 on the target machine.
6017 @defmac REGISTER_MOVE_COST (@var{mode}, @var{from}, @var{to})
6018 A C expression for the cost of moving data of mode @var{mode} from a
6019 register in class @var{from} to one in class @var{to}. The classes are
6020 expressed using the enumeration values such as @code{GENERAL_REGS}. A
6021 value of 2 is the default; other values are interpreted relative to
6024 It is not required that the cost always equal 2 when @var{from} is the
6025 same as @var{to}; on some machines it is expensive to move between
6026 registers if they are not general registers.
6028 If reload sees an insn consisting of a single @code{set} between two
6029 hard registers, and if @code{REGISTER_MOVE_COST} applied to their
6030 classes returns a value of 2, reload does not check to ensure that the
6031 constraints of the insn are met. Setting a cost of other than 2 will
6032 allow reload to verify that the constraints are met. You should do this
6033 if the @samp{mov@var{m}} pattern's constraints do not allow such copying.
6035 These macros are obsolete, new ports should use the target hook
6036 @code{TARGET_REGISTER_MOVE_COST} instead.
6039 @hook TARGET_REGISTER_MOVE_COST
6040 This target hook should return the cost of moving data of mode @var{mode}
6041 from a register in class @var{from} to one in class @var{to}. The classes
6042 are expressed using the enumeration values such as @code{GENERAL_REGS}.
6043 A value of 2 is the default; other values are interpreted relative to
6046 It is not required that the cost always equal 2 when @var{from} is the
6047 same as @var{to}; on some machines it is expensive to move between
6048 registers if they are not general registers.
6050 If reload sees an insn consisting of a single @code{set} between two
6051 hard registers, and if @code{TARGET_REGISTER_MOVE_COST} applied to their
6052 classes returns a value of 2, reload does not check to ensure that the
6053 constraints of the insn are met. Setting a cost of other than 2 will
6054 allow reload to verify that the constraints are met. You should do this
6055 if the @samp{mov@var{m}} pattern's constraints do not allow such copying.
6057 The default version of this function returns 2.
6060 @defmac MEMORY_MOVE_COST (@var{mode}, @var{class}, @var{in})
6061 A C expression for the cost of moving data of mode @var{mode} between a
6062 register of class @var{class} and memory; @var{in} is zero if the value
6063 is to be written to memory, nonzero if it is to be read in. This cost
6064 is relative to those in @code{REGISTER_MOVE_COST}. If moving between
6065 registers and memory is more expensive than between two registers, you
6066 should define this macro to express the relative cost.
6068 If you do not define this macro, GCC uses a default cost of 4 plus
6069 the cost of copying via a secondary reload register, if one is
6070 needed. If your machine requires a secondary reload register to copy
6071 between memory and a register of @var{class} but the reload mechanism is
6072 more complex than copying via an intermediate, define this macro to
6073 reflect the actual cost of the move.
6075 GCC defines the function @code{memory_move_secondary_cost} if
6076 secondary reloads are needed. It computes the costs due to copying via
6077 a secondary register. If your machine copies from memory using a
6078 secondary register in the conventional way but the default base value of
6079 4 is not correct for your machine, define this macro to add some other
6080 value to the result of that function. The arguments to that function
6081 are the same as to this macro.
6083 These macros are obsolete, new ports should use the target hook
6084 @code{TARGET_MEMORY_MOVE_COST} instead.
6087 @hook TARGET_MEMORY_MOVE_COST
6088 This target hook should return the cost of moving data of mode @var{mode}
6089 between a register of class @var{rclass} and memory; @var{in} is @code{false}
6090 if the value is to be written to memory, @code{true} if it is to be read in.
6091 This cost is relative to those in @code{TARGET_REGISTER_MOVE_COST}.
6092 If moving between registers and memory is more expensive than between two
6093 registers, you should add this target hook to express the relative cost.
6095 If you do not add this target hook, GCC uses a default cost of 4 plus
6096 the cost of copying via a secondary reload register, if one is
6097 needed. If your machine requires a secondary reload register to copy
6098 between memory and a register of @var{rclass} but the reload mechanism is
6099 more complex than copying via an intermediate, use this target hook to
6100 reflect the actual cost of the move.
6102 GCC defines the function @code{memory_move_secondary_cost} if
6103 secondary reloads are needed. It computes the costs due to copying via
6104 a secondary register. If your machine copies from memory using a
6105 secondary register in the conventional way but the default base value of
6106 4 is not correct for your machine, use this target hook to add some other
6107 value to the result of that function. The arguments to that function
6108 are the same as to this target hook.
6111 @defmac BRANCH_COST (@var{speed_p}, @var{predictable_p})
6112 A C expression for the cost of a branch instruction. A value of 1 is
6113 the default; other values are interpreted relative to that. Parameter
6114 @var{speed_p} is true when the branch in question should be optimized
6115 for speed. When it is false, @code{BRANCH_COST} should return a value
6116 optimal for code size rather than performance. @var{predictable_p} is
6117 true for well-predicted branches. On many architectures the
6118 @code{BRANCH_COST} can be reduced then.
6121 Here are additional macros which do not specify precise relative costs,
6122 but only that certain actions are more expensive than GCC would
6125 @defmac SLOW_BYTE_ACCESS
6126 Define this macro as a C expression which is nonzero if accessing less
6127 than a word of memory (i.e.@: a @code{char} or a @code{short}) is no
6128 faster than accessing a word of memory, i.e., if such access
6129 require more than one instruction or if there is no difference in cost
6130 between byte and (aligned) word loads.
6132 When this macro is not defined, the compiler will access a field by
6133 finding the smallest containing object; when it is defined, a fullword
6134 load will be used if alignment permits. Unless bytes accesses are
6135 faster than word accesses, using word accesses is preferable since it
6136 may eliminate subsequent memory access if subsequent accesses occur to
6137 other fields in the same word of the structure, but to different bytes.
6140 @defmac SLOW_UNALIGNED_ACCESS (@var{mode}, @var{alignment})
6141 Define this macro to be the value 1 if memory accesses described by the
6142 @var{mode} and @var{alignment} parameters have a cost many times greater
6143 than aligned accesses, for example if they are emulated in a trap
6146 When this macro is nonzero, the compiler will act as if
6147 @code{STRICT_ALIGNMENT} were nonzero when generating code for block
6148 moves. This can cause significantly more instructions to be produced.
6149 Therefore, do not set this macro nonzero if unaligned accesses only add a
6150 cycle or two to the time for a memory access.
6152 If the value of this macro is always zero, it need not be defined. If
6153 this macro is defined, it should produce a nonzero value when
6154 @code{STRICT_ALIGNMENT} is nonzero.
6157 @defmac MOVE_RATIO (@var{speed})
6158 The threshold of number of scalar memory-to-memory move insns, @emph{below}
6159 which a sequence of insns should be generated instead of a
6160 string move insn or a library call. Increasing the value will always
6161 make code faster, but eventually incurs high cost in increased code size.
6163 Note that on machines where the corresponding move insn is a
6164 @code{define_expand} that emits a sequence of insns, this macro counts
6165 the number of such sequences.
6167 The parameter @var{speed} is true if the code is currently being
6168 optimized for speed rather than size.
6170 If you don't define this, a reasonable default is used.
6173 @defmac MOVE_BY_PIECES_P (@var{size}, @var{alignment})
6174 A C expression used to determine whether @code{move_by_pieces} will be used to
6175 copy a chunk of memory, or whether some other block move mechanism
6176 will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less
6177 than @code{MOVE_RATIO}.
6180 @defmac MOVE_MAX_PIECES
6181 A C expression used by @code{move_by_pieces} to determine the largest unit
6182 a load or store used to copy memory is. Defaults to @code{MOVE_MAX}.
6185 @defmac CLEAR_RATIO (@var{speed})
6186 The threshold of number of scalar move insns, @emph{below} which a sequence
6187 of insns should be generated to clear memory instead of a string clear insn
6188 or a library call. Increasing the value will always make code faster, but
6189 eventually incurs high cost in increased code size.
6191 The parameter @var{speed} is true if the code is currently being
6192 optimized for speed rather than size.
6194 If you don't define this, a reasonable default is used.
6197 @defmac CLEAR_BY_PIECES_P (@var{size}, @var{alignment})
6198 A C expression used to determine whether @code{clear_by_pieces} will be used
6199 to clear a chunk of memory, or whether some other block clear mechanism
6200 will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less
6201 than @code{CLEAR_RATIO}.
6204 @defmac SET_RATIO (@var{speed})
6205 The threshold of number of scalar move insns, @emph{below} which a sequence
6206 of insns should be generated to set memory to a constant value, instead of
6207 a block set insn or a library call.
6208 Increasing the value will always make code faster, but
6209 eventually incurs high cost in increased code size.
6211 The parameter @var{speed} is true if the code is currently being
6212 optimized for speed rather than size.
6214 If you don't define this, it defaults to the value of @code{MOVE_RATIO}.
6217 @defmac SET_BY_PIECES_P (@var{size}, @var{alignment})
6218 A C expression used to determine whether @code{store_by_pieces} will be
6219 used to set a chunk of memory to a constant value, or whether some
6220 other mechanism will be used. Used by @code{__builtin_memset} when
6221 storing values other than constant zero.
6222 Defaults to 1 if @code{move_by_pieces_ninsns} returns less
6223 than @code{SET_RATIO}.
6226 @defmac STORE_BY_PIECES_P (@var{size}, @var{alignment})
6227 A C expression used to determine whether @code{store_by_pieces} will be
6228 used to set a chunk of memory to a constant string value, or whether some
6229 other mechanism will be used. Used by @code{__builtin_strcpy} when
6230 called with a constant source string.
6231 Defaults to 1 if @code{move_by_pieces_ninsns} returns less
6232 than @code{MOVE_RATIO}.
6235 @defmac USE_LOAD_POST_INCREMENT (@var{mode})
6236 A C expression used to determine whether a load postincrement is a good
6237 thing to use for a given mode. Defaults to the value of
6238 @code{HAVE_POST_INCREMENT}.
6241 @defmac USE_LOAD_POST_DECREMENT (@var{mode})
6242 A C expression used to determine whether a load postdecrement is a good
6243 thing to use for a given mode. Defaults to the value of
6244 @code{HAVE_POST_DECREMENT}.
6247 @defmac USE_LOAD_PRE_INCREMENT (@var{mode})
6248 A C expression used to determine whether a load preincrement is a good
6249 thing to use for a given mode. Defaults to the value of
6250 @code{HAVE_PRE_INCREMENT}.
6253 @defmac USE_LOAD_PRE_DECREMENT (@var{mode})
6254 A C expression used to determine whether a load predecrement is a good
6255 thing to use for a given mode. Defaults to the value of
6256 @code{HAVE_PRE_DECREMENT}.
6259 @defmac USE_STORE_POST_INCREMENT (@var{mode})
6260 A C expression used to determine whether a store postincrement is a good
6261 thing to use for a given mode. Defaults to the value of
6262 @code{HAVE_POST_INCREMENT}.
6265 @defmac USE_STORE_POST_DECREMENT (@var{mode})
6266 A C expression used to determine whether a store postdecrement is a good
6267 thing to use for a given mode. Defaults to the value of
6268 @code{HAVE_POST_DECREMENT}.
6271 @defmac USE_STORE_PRE_INCREMENT (@var{mode})
6272 This macro is used to determine whether a store preincrement is a good
6273 thing to use for a given mode. Defaults to the value of
6274 @code{HAVE_PRE_INCREMENT}.
6277 @defmac USE_STORE_PRE_DECREMENT (@var{mode})
6278 This macro is used to determine whether a store predecrement is a good
6279 thing to use for a given mode. Defaults to the value of
6280 @code{HAVE_PRE_DECREMENT}.
6283 @defmac NO_FUNCTION_CSE
6284 Define this macro if it is as good or better to call a constant
6285 function address than to call an address kept in a register.
6288 @defmac RANGE_TEST_NON_SHORT_CIRCUIT
6289 Define this macro if a non-short-circuit operation produced by
6290 @samp{fold_range_test ()} is optimal. This macro defaults to true if
6291 @code{BRANCH_COST} is greater than or equal to the value 2.
6294 @hook TARGET_RTX_COSTS
6295 This target hook describes the relative costs of RTL expressions.
6297 The cost may depend on the precise form of the expression, which is
6298 available for examination in @var{x}, and the rtx code of the expression
6299 in which it is contained, found in @var{outer_code}. @var{code} is the
6300 expression code---redundant, since it can be obtained with
6301 @code{GET_CODE (@var{x})}.
6303 In implementing this hook, you can use the construct
6304 @code{COSTS_N_INSNS (@var{n})} to specify a cost equal to @var{n} fast
6307 On entry to the hook, @code{*@var{total}} contains a default estimate
6308 for the cost of the expression. The hook should modify this value as
6309 necessary. Traditionally, the default costs are @code{COSTS_N_INSNS (5)}
6310 for multiplications, @code{COSTS_N_INSNS (7)} for division and modulus
6311 operations, and @code{COSTS_N_INSNS (1)} for all other operations.
6313 When optimizing for code size, i.e.@: when @code{speed} is
6314 false, this target hook should be used to estimate the relative
6315 size cost of an expression, again relative to @code{COSTS_N_INSNS}.
6317 The hook returns true when all subexpressions of @var{x} have been
6318 processed, and false when @code{rtx_cost} should recurse.
6321 @hook TARGET_ADDRESS_COST
6322 This hook computes the cost of an addressing mode that contains
6323 @var{address}. If not defined, the cost is computed from
6324 the @var{address} expression and the @code{TARGET_RTX_COST} hook.
6326 For most CISC machines, the default cost is a good approximation of the
6327 true cost of the addressing mode. However, on RISC machines, all
6328 instructions normally have the same length and execution time. Hence
6329 all addresses will have equal costs.
6331 In cases where more than one form of an address is known, the form with
6332 the lowest cost will be used. If multiple forms have the same, lowest,
6333 cost, the one that is the most complex will be used.
6335 For example, suppose an address that is equal to the sum of a register
6336 and a constant is used twice in the same basic block. When this macro
6337 is not defined, the address will be computed in a register and memory
6338 references will be indirect through that register. On machines where
6339 the cost of the addressing mode containing the sum is no higher than
6340 that of a simple indirect reference, this will produce an additional
6341 instruction and possibly require an additional register. Proper
6342 specification of this macro eliminates this overhead for such machines.
6344 This hook is never called with an invalid address.
6346 On machines where an address involving more than one register is as
6347 cheap as an address computation involving only one register, defining
6348 @code{TARGET_ADDRESS_COST} to reflect this can cause two registers to
6349 be live over a region of code where only one would have been if
6350 @code{TARGET_ADDRESS_COST} were not defined in that manner. This effect
6351 should be considered in the definition of this macro. Equivalent costs
6352 should probably only be given to addresses with different numbers of
6353 registers on machines with lots of registers.
6357 @section Adjusting the Instruction Scheduler
6359 The instruction scheduler may need a fair amount of machine-specific
6360 adjustment in order to produce good code. GCC provides several target
6361 hooks for this purpose. It is usually enough to define just a few of
6362 them: try the first ones in this list first.
6364 @hook TARGET_SCHED_ISSUE_RATE
6365 This hook returns the maximum number of instructions that can ever
6366 issue at the same time on the target machine. The default is one.
6367 Although the insn scheduler can define itself the possibility of issue
6368 an insn on the same cycle, the value can serve as an additional
6369 constraint to issue insns on the same simulated processor cycle (see
6370 hooks @samp{TARGET_SCHED_REORDER} and @samp{TARGET_SCHED_REORDER2}).
6371 This value must be constant over the entire compilation. If you need
6372 it to vary depending on what the instructions are, you must use
6373 @samp{TARGET_SCHED_VARIABLE_ISSUE}.
6376 @hook TARGET_SCHED_VARIABLE_ISSUE
6377 This hook is executed by the scheduler after it has scheduled an insn
6378 from the ready list. It should return the number of insns which can
6379 still be issued in the current cycle. The default is
6380 @samp{@w{@var{more} - 1}} for insns other than @code{CLOBBER} and
6381 @code{USE}, which normally are not counted against the issue rate.
6382 You should define this hook if some insns take more machine resources
6383 than others, so that fewer insns can follow them in the same cycle.
6384 @var{file} is either a null pointer, or a stdio stream to write any
6385 debug output to. @var{verbose} is the verbose level provided by
6386 @option{-fsched-verbose-@var{n}}. @var{insn} is the instruction that
6390 @hook TARGET_SCHED_ADJUST_COST
6391 This function corrects the value of @var{cost} based on the
6392 relationship between @var{insn} and @var{dep_insn} through the
6393 dependence @var{link}. It should return the new value. The default
6394 is to make no adjustment to @var{cost}. This can be used for example
6395 to specify to the scheduler using the traditional pipeline description
6396 that an output- or anti-dependence does not incur the same cost as a
6397 data-dependence. If the scheduler using the automaton based pipeline
6398 description, the cost of anti-dependence is zero and the cost of
6399 output-dependence is maximum of one and the difference of latency
6400 times of the first and the second insns. If these values are not
6401 acceptable, you could use the hook to modify them too. See also
6402 @pxref{Processor pipeline description}.
6405 @hook TARGET_SCHED_ADJUST_PRIORITY
6406 This hook adjusts the integer scheduling priority @var{priority} of
6407 @var{insn}. It should return the new priority. Increase the priority to
6408 execute @var{insn} earlier, reduce the priority to execute @var{insn}
6409 later. Do not define this hook if you do not need to adjust the
6410 scheduling priorities of insns.
6413 @hook TARGET_SCHED_REORDER
6414 This hook is executed by the scheduler after it has scheduled the ready
6415 list, to allow the machine description to reorder it (for example to
6416 combine two small instructions together on @samp{VLIW} machines).
6417 @var{file} is either a null pointer, or a stdio stream to write any
6418 debug output to. @var{verbose} is the verbose level provided by
6419 @option{-fsched-verbose-@var{n}}. @var{ready} is a pointer to the ready
6420 list of instructions that are ready to be scheduled. @var{n_readyp} is
6421 a pointer to the number of elements in the ready list. The scheduler
6422 reads the ready list in reverse order, starting with
6423 @var{ready}[@var{*n_readyp} @minus{} 1] and going to @var{ready}[0]. @var{clock}
6424 is the timer tick of the scheduler. You may modify the ready list and
6425 the number of ready insns. The return value is the number of insns that
6426 can issue this cycle; normally this is just @code{issue_rate}. See also
6427 @samp{TARGET_SCHED_REORDER2}.
6430 @hook TARGET_SCHED_REORDER2
6431 Like @samp{TARGET_SCHED_REORDER}, but called at a different time. That
6432 function is called whenever the scheduler starts a new cycle. This one
6433 is called once per iteration over a cycle, immediately after
6434 @samp{TARGET_SCHED_VARIABLE_ISSUE}; it can reorder the ready list and
6435 return the number of insns to be scheduled in the same cycle. Defining
6436 this hook can be useful if there are frequent situations where
6437 scheduling one insn causes other insns to become ready in the same
6438 cycle. These other insns can then be taken into account properly.
6441 @hook TARGET_SCHED_DEPENDENCIES_EVALUATION_HOOK
6442 This hook is called after evaluation forward dependencies of insns in
6443 chain given by two parameter values (@var{head} and @var{tail}
6444 correspondingly) but before insns scheduling of the insn chain. For
6445 example, it can be used for better insn classification if it requires
6446 analysis of dependencies. This hook can use backward and forward
6447 dependencies of the insn scheduler because they are already
6451 @hook TARGET_SCHED_INIT
6452 This hook is executed by the scheduler at the beginning of each block of
6453 instructions that are to be scheduled. @var{file} is either a null
6454 pointer, or a stdio stream to write any debug output to. @var{verbose}
6455 is the verbose level provided by @option{-fsched-verbose-@var{n}}.
6456 @var{max_ready} is the maximum number of insns in the current scheduling
6457 region that can be live at the same time. This can be used to allocate
6458 scratch space if it is needed, e.g.@: by @samp{TARGET_SCHED_REORDER}.
6461 @hook TARGET_SCHED_FINISH
6462 This hook is executed by the scheduler at the end of each block of
6463 instructions that are to be scheduled. It can be used to perform
6464 cleanup of any actions done by the other scheduling hooks. @var{file}
6465 is either a null pointer, or a stdio stream to write any debug output
6466 to. @var{verbose} is the verbose level provided by
6467 @option{-fsched-verbose-@var{n}}.
6470 @hook TARGET_SCHED_INIT_GLOBAL
6471 This hook is executed by the scheduler after function level initializations.
6472 @var{file} is either a null pointer, or a stdio stream to write any debug output to.
6473 @var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}.
6474 @var{old_max_uid} is the maximum insn uid when scheduling begins.
6477 @hook TARGET_SCHED_FINISH_GLOBAL
6478 This is the cleanup hook corresponding to @code{TARGET_SCHED_INIT_GLOBAL}.
6479 @var{file} is either a null pointer, or a stdio stream to write any debug output to.
6480 @var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}.
6483 @hook TARGET_SCHED_DFA_PRE_CYCLE_INSN
6484 The hook returns an RTL insn. The automaton state used in the
6485 pipeline hazard recognizer is changed as if the insn were scheduled
6486 when the new simulated processor cycle starts. Usage of the hook may
6487 simplify the automaton pipeline description for some @acronym{VLIW}
6488 processors. If the hook is defined, it is used only for the automaton
6489 based pipeline description. The default is not to change the state
6490 when the new simulated processor cycle starts.
6493 @hook TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN
6494 The hook can be used to initialize data used by the previous hook.
6497 @hook TARGET_SCHED_DFA_POST_CYCLE_INSN
6498 The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used
6499 to changed the state as if the insn were scheduled when the new
6500 simulated processor cycle finishes.
6503 @hook TARGET_SCHED_INIT_DFA_POST_CYCLE_INSN
6504 The hook is analogous to @samp{TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN} but
6505 used to initialize data used by the previous hook.
6508 @hook TARGET_SCHED_DFA_PRE_ADVANCE_CYCLE
6509 The hook to notify target that the current simulated cycle is about to finish.
6510 The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used
6511 to change the state in more complicated situations - e.g., when advancing
6512 state on a single insn is not enough.
6515 @hook TARGET_SCHED_DFA_POST_ADVANCE_CYCLE
6516 The hook to notify target that new simulated cycle has just started.
6517 The hook is analogous to @samp{TARGET_SCHED_DFA_POST_CYCLE_INSN} but used
6518 to change the state in more complicated situations - e.g., when advancing
6519 state on a single insn is not enough.
6522 @hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD
6523 This hook controls better choosing an insn from the ready insn queue
6524 for the @acronym{DFA}-based insn scheduler. Usually the scheduler
6525 chooses the first insn from the queue. If the hook returns a positive
6526 value, an additional scheduler code tries all permutations of
6527 @samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD ()}
6528 subsequent ready insns to choose an insn whose issue will result in
6529 maximal number of issued insns on the same cycle. For the
6530 @acronym{VLIW} processor, the code could actually solve the problem of
6531 packing simple insns into the @acronym{VLIW} insn. Of course, if the
6532 rules of @acronym{VLIW} packing are described in the automaton.
6534 This code also could be used for superscalar @acronym{RISC}
6535 processors. Let us consider a superscalar @acronym{RISC} processor
6536 with 3 pipelines. Some insns can be executed in pipelines @var{A} or
6537 @var{B}, some insns can be executed only in pipelines @var{B} or
6538 @var{C}, and one insn can be executed in pipeline @var{B}. The
6539 processor may issue the 1st insn into @var{A} and the 2nd one into
6540 @var{B}. In this case, the 3rd insn will wait for freeing @var{B}
6541 until the next cycle. If the scheduler issues the 3rd insn the first,
6542 the processor could issue all 3 insns per cycle.
6544 Actually this code demonstrates advantages of the automaton based
6545 pipeline hazard recognizer. We try quickly and easy many insn
6546 schedules to choose the best one.
6548 The default is no multipass scheduling.
6551 @hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD
6553 This hook controls what insns from the ready insn queue will be
6554 considered for the multipass insn scheduling. If the hook returns
6555 zero for @var{insn}, the insn will be not chosen to
6558 The default is that any ready insns can be chosen to be issued.
6561 @hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_BEGIN
6562 This hook prepares the target backend for a new round of multipass
6566 @hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_ISSUE
6567 This hook is called when multipass scheduling evaluates instruction INSN.
6570 @hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_BACKTRACK
6571 This is called when multipass scheduling backtracks from evaluation of
6575 @hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_END
6576 This hook notifies the target about the result of the concluded current
6577 round of multipass scheduling.
6580 @hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_INIT
6581 This hook initializes target-specific data used in multipass scheduling.
6584 @hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_FINI
6585 This hook finalizes target-specific data used in multipass scheduling.
6588 @hook TARGET_SCHED_DFA_NEW_CYCLE
6589 This hook is called by the insn scheduler before issuing @var{insn}
6590 on cycle @var{clock}. If the hook returns nonzero,
6591 @var{insn} is not issued on this processor cycle. Instead,
6592 the processor cycle is advanced. If *@var{sort_p}
6593 is zero, the insn ready queue is not sorted on the new cycle
6594 start as usually. @var{dump} and @var{verbose} specify the file and
6595 verbosity level to use for debugging output.
6596 @var{last_clock} and @var{clock} are, respectively, the
6597 processor cycle on which the previous insn has been issued,
6598 and the current processor cycle.
6601 @hook TARGET_SCHED_IS_COSTLY_DEPENDENCE
6602 This hook is used to define which dependences are considered costly by
6603 the target, so costly that it is not advisable to schedule the insns that
6604 are involved in the dependence too close to one another. The parameters
6605 to this hook are as follows: The first parameter @var{_dep} is the dependence
6606 being evaluated. The second parameter @var{cost} is the cost of the
6607 dependence as estimated by the scheduler, and the third
6608 parameter @var{distance} is the distance in cycles between the two insns.
6609 The hook returns @code{true} if considering the distance between the two
6610 insns the dependence between them is considered costly by the target,
6611 and @code{false} otherwise.
6613 Defining this hook can be useful in multiple-issue out-of-order machines,
6614 where (a) it's practically hopeless to predict the actual data/resource
6615 delays, however: (b) there's a better chance to predict the actual grouping
6616 that will be formed, and (c) correctly emulating the grouping can be very
6617 important. In such targets one may want to allow issuing dependent insns
6618 closer to one another---i.e., closer than the dependence distance; however,
6619 not in cases of ``costly dependences'', which this hooks allows to define.
6622 @hook TARGET_SCHED_H_I_D_EXTENDED
6623 This hook is called by the insn scheduler after emitting a new instruction to
6624 the instruction stream. The hook notifies a target backend to extend its
6625 per instruction data structures.
6628 @hook TARGET_SCHED_ALLOC_SCHED_CONTEXT
6629 Return a pointer to a store large enough to hold target scheduling context.
6632 @hook TARGET_SCHED_INIT_SCHED_CONTEXT
6633 Initialize store pointed to by @var{tc} to hold target scheduling context.
6634 It @var{clean_p} is true then initialize @var{tc} as if scheduler is at the
6635 beginning of the block. Otherwise, copy the current context into @var{tc}.
6638 @hook TARGET_SCHED_SET_SCHED_CONTEXT
6639 Copy target scheduling context pointed to by @var{tc} to the current context.
6642 @hook TARGET_SCHED_CLEAR_SCHED_CONTEXT
6643 Deallocate internal data in target scheduling context pointed to by @var{tc}.
6646 @hook TARGET_SCHED_FREE_SCHED_CONTEXT
6647 Deallocate a store for target scheduling context pointed to by @var{tc}.
6650 @hook TARGET_SCHED_SPECULATE_INSN
6651 This hook is called by the insn scheduler when @var{insn} has only
6652 speculative dependencies and therefore can be scheduled speculatively.
6653 The hook is used to check if the pattern of @var{insn} has a speculative
6654 version and, in case of successful check, to generate that speculative
6655 pattern. The hook should return 1, if the instruction has a speculative form,
6656 or @minus{}1, if it doesn't. @var{request} describes the type of requested
6657 speculation. If the return value equals 1 then @var{new_pat} is assigned
6658 the generated speculative pattern.
6661 @hook TARGET_SCHED_NEEDS_BLOCK_P
6662 This hook is called by the insn scheduler during generation of recovery code
6663 for @var{insn}. It should return @code{true}, if the corresponding check
6664 instruction should branch to recovery code, or @code{false} otherwise.
6667 @hook TARGET_SCHED_GEN_SPEC_CHECK
6668 This hook is called by the insn scheduler to generate a pattern for recovery
6669 check instruction. If @var{mutate_p} is zero, then @var{insn} is a
6670 speculative instruction for which the check should be generated.
6671 @var{label} is either a label of a basic block, where recovery code should
6672 be emitted, or a null pointer, when requested check doesn't branch to
6673 recovery code (a simple check). If @var{mutate_p} is nonzero, then
6674 a pattern for a branchy check corresponding to a simple check denoted by
6675 @var{insn} should be generated. In this case @var{label} can't be null.
6678 @hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD_SPEC
6679 This hook is used as a workaround for
6680 @samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD} not being
6681 called on the first instruction of the ready list. The hook is used to
6682 discard speculative instructions that stand first in the ready list from
6683 being scheduled on the current cycle. If the hook returns @code{false},
6684 @var{insn} will not be chosen to be issued.
6685 For non-speculative instructions,
6686 the hook should always return @code{true}. For example, in the ia64 backend
6687 the hook is used to cancel data speculative insns when the ALAT table
6691 @hook TARGET_SCHED_SET_SCHED_FLAGS
6692 This hook is used by the insn scheduler to find out what features should be
6694 The structure *@var{spec_info} should be filled in by the target.
6695 The structure describes speculation types that can be used in the scheduler.
6698 @hook TARGET_SCHED_SMS_RES_MII
6699 This hook is called by the swing modulo scheduler to calculate a
6700 resource-based lower bound which is based on the resources available in
6701 the machine and the resources required by each instruction. The target
6702 backend can use @var{g} to calculate such bound. A very simple lower
6703 bound will be used in case this hook is not implemented: the total number
6704 of instructions divided by the issue rate.
6707 @hook TARGET_SCHED_DISPATCH
6708 This hook is called by Haifa Scheduler. It returns true if dispatch scheduling
6709 is supported in hardware and the condition specified in the parameter is true.
6712 @hook TARGET_SCHED_DISPATCH_DO
6713 This hook is called by Haifa Scheduler. It performs the operation specified
6714 in its second parameter.
6718 @section Dividing the Output into Sections (Texts, Data, @dots{})
6719 @c the above section title is WAY too long. maybe cut the part between
6720 @c the (...)? --mew 10feb93
6722 An object file is divided into sections containing different types of
6723 data. In the most common case, there are three sections: the @dfn{text
6724 section}, which holds instructions and read-only data; the @dfn{data
6725 section}, which holds initialized writable data; and the @dfn{bss
6726 section}, which holds uninitialized data. Some systems have other kinds
6729 @file{varasm.c} provides several well-known sections, such as
6730 @code{text_section}, @code{data_section} and @code{bss_section}.
6731 The normal way of controlling a @code{@var{foo}_section} variable
6732 is to define the associated @code{@var{FOO}_SECTION_ASM_OP} macro,
6733 as described below. The macros are only read once, when @file{varasm.c}
6734 initializes itself, so their values must be run-time constants.
6735 They may however depend on command-line flags.
6737 @emph{Note:} Some run-time files, such @file{crtstuff.c}, also make
6738 use of the @code{@var{FOO}_SECTION_ASM_OP} macros, and expect them
6739 to be string literals.
6741 Some assemblers require a different string to be written every time a
6742 section is selected. If your assembler falls into this category, you
6743 should define the @code{TARGET_ASM_INIT_SECTIONS} hook and use
6744 @code{get_unnamed_section} to set up the sections.
6746 You must always create a @code{text_section}, either by defining
6747 @code{TEXT_SECTION_ASM_OP} or by initializing @code{text_section}
6748 in @code{TARGET_ASM_INIT_SECTIONS}. The same is true of
6749 @code{data_section} and @code{DATA_SECTION_ASM_OP}. If you do not
6750 create a distinct @code{readonly_data_section}, the default is to
6751 reuse @code{text_section}.
6753 All the other @file{varasm.c} sections are optional, and are null
6754 if the target does not provide them.
6756 @defmac TEXT_SECTION_ASM_OP
6757 A C expression whose value is a string, including spacing, containing the
6758 assembler operation that should precede instructions and read-only data.
6759 Normally @code{"\t.text"} is right.
6762 @defmac HOT_TEXT_SECTION_NAME
6763 If defined, a C string constant for the name of the section containing most
6764 frequently executed functions of the program. If not defined, GCC will provide
6765 a default definition if the target supports named sections.
6768 @defmac UNLIKELY_EXECUTED_TEXT_SECTION_NAME
6769 If defined, a C string constant for the name of the section containing unlikely
6770 executed functions in the program.
6773 @defmac DATA_SECTION_ASM_OP
6774 A C expression whose value is a string, including spacing, containing the
6775 assembler operation to identify the following data as writable initialized
6776 data. Normally @code{"\t.data"} is right.
6779 @defmac SDATA_SECTION_ASM_OP
6780 If defined, a C expression whose value is a string, including spacing,
6781 containing the assembler operation to identify the following data as
6782 initialized, writable small data.
6785 @defmac READONLY_DATA_SECTION_ASM_OP
6786 A C expression whose value is a string, including spacing, containing the
6787 assembler operation to identify the following data as read-only initialized
6791 @defmac BSS_SECTION_ASM_OP
6792 If defined, a C expression whose value is a string, including spacing,
6793 containing the assembler operation to identify the following data as
6794 uninitialized global data. If not defined, and neither
6795 @code{ASM_OUTPUT_BSS} nor @code{ASM_OUTPUT_ALIGNED_BSS} are defined,
6796 uninitialized global data will be output in the data section if
6797 @option{-fno-common} is passed, otherwise @code{ASM_OUTPUT_COMMON} will be
6801 @defmac SBSS_SECTION_ASM_OP
6802 If defined, a C expression whose value is a string, including spacing,
6803 containing the assembler operation to identify the following data as
6804 uninitialized, writable small data.
6807 @defmac TLS_COMMON_ASM_OP
6808 If defined, a C expression whose value is a string containing the
6809 assembler operation to identify the following data as thread-local
6810 common data. The default is @code{".tls_common"}.
6813 @defmac TLS_SECTION_ASM_FLAG
6814 If defined, a C expression whose value is a character constant
6815 containing the flag used to mark a section as a TLS section. The
6816 default is @code{'T'}.
6819 @defmac INIT_SECTION_ASM_OP
6820 If defined, a C expression whose value is a string, including spacing,
6821 containing the assembler operation to identify the following data as
6822 initialization code. If not defined, GCC will assume such a section does
6823 not exist. This section has no corresponding @code{init_section}
6824 variable; it is used entirely in runtime code.
6827 @defmac FINI_SECTION_ASM_OP
6828 If defined, a C expression whose value is a string, including spacing,
6829 containing the assembler operation to identify the following data as
6830 finalization code. If not defined, GCC will assume such a section does
6831 not exist. This section has no corresponding @code{fini_section}
6832 variable; it is used entirely in runtime code.
6835 @defmac INIT_ARRAY_SECTION_ASM_OP
6836 If defined, a C expression whose value is a string, including spacing,
6837 containing the assembler operation to identify the following data as
6838 part of the @code{.init_array} (or equivalent) section. If not
6839 defined, GCC will assume such a section does not exist. Do not define
6840 both this macro and @code{INIT_SECTION_ASM_OP}.
6843 @defmac FINI_ARRAY_SECTION_ASM_OP
6844 If defined, a C expression whose value is a string, including spacing,
6845 containing the assembler operation to identify the following data as
6846 part of the @code{.fini_array} (or equivalent) section. If not
6847 defined, GCC will assume such a section does not exist. Do not define
6848 both this macro and @code{FINI_SECTION_ASM_OP}.
6851 @defmac CRT_CALL_STATIC_FUNCTION (@var{section_op}, @var{function})
6852 If defined, an ASM statement that switches to a different section
6853 via @var{section_op}, calls @var{function}, and switches back to
6854 the text section. This is used in @file{crtstuff.c} if
6855 @code{INIT_SECTION_ASM_OP} or @code{FINI_SECTION_ASM_OP} to calls
6856 to initialization and finalization functions from the init and fini
6857 sections. By default, this macro uses a simple function call. Some
6858 ports need hand-crafted assembly code to avoid dependencies on
6859 registers initialized in the function prologue or to ensure that
6860 constant pools don't end up too far way in the text section.
6863 @defmac TARGET_LIBGCC_SDATA_SECTION
6864 If defined, a string which names the section into which small
6865 variables defined in crtstuff and libgcc should go. This is useful
6866 when the target has options for optimizing access to small data, and
6867 you want the crtstuff and libgcc routines to be conservative in what
6868 they expect of your application yet liberal in what your application
6869 expects. For example, for targets with a @code{.sdata} section (like
6870 MIPS), you could compile crtstuff with @code{-G 0} so that it doesn't
6871 require small data support from your application, but use this macro
6872 to put small data into @code{.sdata} so that your application can
6873 access these variables whether it uses small data or not.
6876 @defmac FORCE_CODE_SECTION_ALIGN
6877 If defined, an ASM statement that aligns a code section to some
6878 arbitrary boundary. This is used to force all fragments of the
6879 @code{.init} and @code{.fini} sections to have to same alignment
6880 and thus prevent the linker from having to add any padding.
6883 @defmac JUMP_TABLES_IN_TEXT_SECTION
6884 Define this macro to be an expression with a nonzero value if jump
6885 tables (for @code{tablejump} insns) should be output in the text
6886 section, along with the assembler instructions. Otherwise, the
6887 readonly data section is used.
6889 This macro is irrelevant if there is no separate readonly data section.
6892 @hook TARGET_ASM_INIT_SECTIONS
6893 Define this hook if you need to do something special to set up the
6894 @file{varasm.c} sections, or if your target has some special sections
6895 of its own that you need to create.
6897 GCC calls this hook after processing the command line, but before writing
6898 any assembly code, and before calling any of the section-returning hooks
6902 @hook TARGET_ASM_RELOC_RW_MASK
6903 Return a mask describing how relocations should be treated when
6904 selecting sections. Bit 1 should be set if global relocations
6905 should be placed in a read-write section; bit 0 should be set if
6906 local relocations should be placed in a read-write section.
6908 The default version of this function returns 3 when @option{-fpic}
6909 is in effect, and 0 otherwise. The hook is typically redefined
6910 when the target cannot support (some kinds of) dynamic relocations
6911 in read-only sections even in executables.
6914 @hook TARGET_ASM_SELECT_SECTION
6915 Return the section into which @var{exp} should be placed. You can
6916 assume that @var{exp} is either a @code{VAR_DECL} node or a constant of
6917 some sort. @var{reloc} indicates whether the initial value of @var{exp}
6918 requires link-time relocations. Bit 0 is set when variable contains
6919 local relocations only, while bit 1 is set for global relocations.
6920 @var{align} is the constant alignment in bits.
6922 The default version of this function takes care of putting read-only
6923 variables in @code{readonly_data_section}.
6925 See also @var{USE_SELECT_SECTION_FOR_FUNCTIONS}.
6928 @defmac USE_SELECT_SECTION_FOR_FUNCTIONS
6929 Define this macro if you wish TARGET_ASM_SELECT_SECTION to be called
6930 for @code{FUNCTION_DECL}s as well as for variables and constants.
6932 In the case of a @code{FUNCTION_DECL}, @var{reloc} will be zero if the
6933 function has been determined to be likely to be called, and nonzero if
6934 it is unlikely to be called.
6937 @hook TARGET_ASM_UNIQUE_SECTION
6938 Build up a unique section name, expressed as a @code{STRING_CST} node,
6939 and assign it to @samp{DECL_SECTION_NAME (@var{decl})}.
6940 As with @code{TARGET_ASM_SELECT_SECTION}, @var{reloc} indicates whether
6941 the initial value of @var{exp} requires link-time relocations.
6943 The default version of this function appends the symbol name to the
6944 ELF section name that would normally be used for the symbol. For
6945 example, the function @code{foo} would be placed in @code{.text.foo}.
6946 Whatever the actual target object format, this is often good enough.
6949 @hook TARGET_ASM_FUNCTION_RODATA_SECTION
6950 Return the readonly data section associated with
6951 @samp{DECL_SECTION_NAME (@var{decl})}.
6952 The default version of this function selects @code{.gnu.linkonce.r.name} if
6953 the function's section is @code{.gnu.linkonce.t.name}, @code{.rodata.name}
6954 if function is in @code{.text.name}, and the normal readonly-data section
6958 @hook TARGET_ASM_SELECT_RTX_SECTION
6959 Return the section into which a constant @var{x}, of mode @var{mode},
6960 should be placed. You can assume that @var{x} is some kind of
6961 constant in RTL@. The argument @var{mode} is redundant except in the
6962 case of a @code{const_int} rtx. @var{align} is the constant alignment
6965 The default version of this function takes care of putting symbolic
6966 constants in @code{flag_pic} mode in @code{data_section} and everything
6967 else in @code{readonly_data_section}.
6970 @hook TARGET_MANGLE_DECL_ASSEMBLER_NAME
6971 Define this hook if you need to postprocess the assembler name generated
6972 by target-independent code. The @var{id} provided to this hook will be
6973 the computed name (e.g., the macro @code{DECL_NAME} of the @var{decl} in C,
6974 or the mangled name of the @var{decl} in C++). The return value of the
6975 hook is an @code{IDENTIFIER_NODE} for the appropriate mangled name on
6976 your target system. The default implementation of this hook just
6977 returns the @var{id} provided.
6980 @hook TARGET_ENCODE_SECTION_INFO
6981 Define this hook if references to a symbol or a constant must be
6982 treated differently depending on something about the variable or
6983 function named by the symbol (such as what section it is in).
6985 The hook is executed immediately after rtl has been created for
6986 @var{decl}, which may be a variable or function declaration or
6987 an entry in the constant pool. In either case, @var{rtl} is the
6988 rtl in question. Do @emph{not} use @code{DECL_RTL (@var{decl})}
6989 in this hook; that field may not have been initialized yet.
6991 In the case of a constant, it is safe to assume that the rtl is
6992 a @code{mem} whose address is a @code{symbol_ref}. Most decls
6993 will also have this form, but that is not guaranteed. Global
6994 register variables, for instance, will have a @code{reg} for their
6995 rtl. (Normally the right thing to do with such unusual rtl is
6998 The @var{new_decl_p} argument will be true if this is the first time
6999 that @code{TARGET_ENCODE_SECTION_INFO} has been invoked on this decl. It will
7000 be false for subsequent invocations, which will happen for duplicate
7001 declarations. Whether or not anything must be done for the duplicate
7002 declaration depends on whether the hook examines @code{DECL_ATTRIBUTES}.
7003 @var{new_decl_p} is always true when the hook is called for a constant.
7005 @cindex @code{SYMBOL_REF_FLAG}, in @code{TARGET_ENCODE_SECTION_INFO}
7006 The usual thing for this hook to do is to record flags in the
7007 @code{symbol_ref}, using @code{SYMBOL_REF_FLAG} or @code{SYMBOL_REF_FLAGS}.
7008 Historically, the name string was modified if it was necessary to
7009 encode more than one bit of information, but this practice is now
7010 discouraged; use @code{SYMBOL_REF_FLAGS}.
7012 The default definition of this hook, @code{default_encode_section_info}
7013 in @file{varasm.c}, sets a number of commonly-useful bits in
7014 @code{SYMBOL_REF_FLAGS}. Check whether the default does what you need
7015 before overriding it.
7018 @hook TARGET_STRIP_NAME_ENCODING
7019 Decode @var{name} and return the real name part, sans
7020 the characters that @code{TARGET_ENCODE_SECTION_INFO}
7024 @hook TARGET_IN_SMALL_DATA_P
7025 Returns true if @var{exp} should be placed into a ``small data'' section.
7026 The default version of this hook always returns false.
7029 @hook TARGET_HAVE_SRODATA_SECTION
7030 Contains the value true if the target places read-only
7031 ``small data'' into a separate section. The default value is false.
7034 @hook TARGET_PROFILE_BEFORE_PROLOGUE
7036 @hook TARGET_BINDS_LOCAL_P
7037 Returns true if @var{exp} names an object for which name resolution
7038 rules must resolve to the current ``module'' (dynamic shared library
7039 or executable image).
7041 The default version of this hook implements the name resolution rules
7042 for ELF, which has a looser model of global name binding than other
7043 currently supported object file formats.
7046 @hook TARGET_HAVE_TLS
7047 Contains the value true if the target supports thread-local storage.
7048 The default value is false.
7053 @section Position Independent Code
7054 @cindex position independent code
7057 This section describes macros that help implement generation of position
7058 independent code. Simply defining these macros is not enough to
7059 generate valid PIC; you must also add support to the hook
7060 @code{TARGET_LEGITIMATE_ADDRESS_P} and to the macro
7061 @code{PRINT_OPERAND_ADDRESS}, as well as @code{LEGITIMIZE_ADDRESS}. You
7062 must modify the definition of @samp{movsi} to do something appropriate
7063 when the source operand contains a symbolic address. You may also
7064 need to alter the handling of switch statements so that they use
7066 @c i rearranged the order of the macros above to try to force one of
7067 @c them to the next line, to eliminate an overfull hbox. --mew 10feb93
7069 @defmac PIC_OFFSET_TABLE_REGNUM
7070 The register number of the register used to address a table of static
7071 data addresses in memory. In some cases this register is defined by a
7072 processor's ``application binary interface'' (ABI)@. When this macro
7073 is defined, RTL is generated for this register once, as with the stack
7074 pointer and frame pointer registers. If this macro is not defined, it
7075 is up to the machine-dependent files to allocate such a register (if
7076 necessary). Note that this register must be fixed when in use (e.g.@:
7077 when @code{flag_pic} is true).
7080 @defmac PIC_OFFSET_TABLE_REG_CALL_CLOBBERED
7081 A C expression that is nonzero if the register defined by
7082 @code{PIC_OFFSET_TABLE_REGNUM} is clobbered by calls. If not defined,
7083 the default is zero. Do not define
7084 this macro if @code{PIC_OFFSET_TABLE_REGNUM} is not defined.
7087 @defmac LEGITIMATE_PIC_OPERAND_P (@var{x})
7088 A C expression that is nonzero if @var{x} is a legitimate immediate
7089 operand on the target machine when generating position independent code.
7090 You can assume that @var{x} satisfies @code{CONSTANT_P}, so you need not
7091 check this. You can also assume @var{flag_pic} is true, so you need not
7092 check it either. You need not define this macro if all constants
7093 (including @code{SYMBOL_REF}) can be immediate operands when generating
7094 position independent code.
7097 @node Assembler Format
7098 @section Defining the Output Assembler Language
7100 This section describes macros whose principal purpose is to describe how
7101 to write instructions in assembler language---rather than what the
7105 * File Framework:: Structural information for the assembler file.
7106 * Data Output:: Output of constants (numbers, strings, addresses).
7107 * Uninitialized Data:: Output of uninitialized variables.
7108 * Label Output:: Output and generation of labels.
7109 * Initialization:: General principles of initialization
7110 and termination routines.
7111 * Macros for Initialization::
7112 Specific macros that control the handling of
7113 initialization and termination routines.
7114 * Instruction Output:: Output of actual instructions.
7115 * Dispatch Tables:: Output of jump tables.
7116 * Exception Region Output:: Output of exception region code.
7117 * Alignment Output:: Pseudo ops for alignment and skipping data.
7120 @node File Framework
7121 @subsection The Overall Framework of an Assembler File
7122 @cindex assembler format
7123 @cindex output of assembler code
7125 @c prevent bad page break with this line
7126 This describes the overall framework of an assembly file.
7128 @findex default_file_start
7129 @hook TARGET_ASM_FILE_START
7130 Output to @code{asm_out_file} any text which the assembler expects to
7131 find at the beginning of a file. The default behavior is controlled
7132 by two flags, documented below. Unless your target's assembler is
7133 quite unusual, if you override the default, you should call
7134 @code{default_file_start} at some point in your target hook. This
7135 lets other target files rely on these variables.
7138 @hook TARGET_ASM_FILE_START_APP_OFF
7139 If this flag is true, the text of the macro @code{ASM_APP_OFF} will be
7140 printed as the very first line in the assembly file, unless
7141 @option{-fverbose-asm} is in effect. (If that macro has been defined
7142 to the empty string, this variable has no effect.) With the normal
7143 definition of @code{ASM_APP_OFF}, the effect is to notify the GNU
7144 assembler that it need not bother stripping comments or extra
7145 whitespace from its input. This allows it to work a bit faster.
7147 The default is false. You should not set it to true unless you have
7148 verified that your port does not generate any extra whitespace or
7149 comments that will cause GAS to issue errors in NO_APP mode.
7152 @hook TARGET_ASM_FILE_START_FILE_DIRECTIVE
7153 If this flag is true, @code{output_file_directive} will be called
7154 for the primary source file, immediately after printing
7155 @code{ASM_APP_OFF} (if that is enabled). Most ELF assemblers expect
7156 this to be done. The default is false.
7159 @hook TARGET_ASM_FILE_END
7160 Output to @code{asm_out_file} any text which the assembler expects
7161 to find at the end of a file. The default is to output nothing.
7164 @deftypefun void file_end_indicate_exec_stack ()
7165 Some systems use a common convention, the @samp{.note.GNU-stack}
7166 special section, to indicate whether or not an object file relies on
7167 the stack being executable. If your system uses this convention, you
7168 should define @code{TARGET_ASM_FILE_END} to this function. If you
7169 need to do other things in that hook, have your hook function call
7173 @hook TARGET_ASM_LTO_START
7174 Output to @code{asm_out_file} any text which the assembler expects
7175 to find at the start of an LTO section. The default is to output
7179 @hook TARGET_ASM_LTO_END
7180 Output to @code{asm_out_file} any text which the assembler expects
7181 to find at the end of an LTO section. The default is to output
7185 @hook TARGET_ASM_CODE_END
7186 Output to @code{asm_out_file} any text which is needed before emitting
7187 unwind info and debug info at the end of a file. Some targets emit
7188 here PIC setup thunks that cannot be emitted at the end of file,
7189 because they couldn't have unwind info then. The default is to output
7193 @defmac ASM_COMMENT_START
7194 A C string constant describing how to begin a comment in the target
7195 assembler language. The compiler assumes that the comment will end at
7196 the end of the line.
7200 A C string constant for text to be output before each @code{asm}
7201 statement or group of consecutive ones. Normally this is
7202 @code{"#APP"}, which is a comment that has no effect on most
7203 assemblers but tells the GNU assembler that it must check the lines
7204 that follow for all valid assembler constructs.
7208 A C string constant for text to be output after each @code{asm}
7209 statement or group of consecutive ones. Normally this is
7210 @code{"#NO_APP"}, which tells the GNU assembler to resume making the
7211 time-saving assumptions that are valid for ordinary compiler output.
7214 @defmac ASM_OUTPUT_SOURCE_FILENAME (@var{stream}, @var{name})
7215 A C statement to output COFF information or DWARF debugging information
7216 which indicates that filename @var{name} is the current source file to
7217 the stdio stream @var{stream}.
7219 This macro need not be defined if the standard form of output
7220 for the file format in use is appropriate.
7223 @hook TARGET_ASM_OUTPUT_SOURCE_FILENAME
7225 @defmac OUTPUT_QUOTED_STRING (@var{stream}, @var{string})
7226 A C statement to output the string @var{string} to the stdio stream
7227 @var{stream}. If you do not call the function @code{output_quoted_string}
7228 in your config files, GCC will only call it to output filenames to
7229 the assembler source. So you can use it to canonicalize the format
7230 of the filename using this macro.
7233 @defmac ASM_OUTPUT_IDENT (@var{stream}, @var{string})
7234 A C statement to output something to the assembler file to handle a
7235 @samp{#ident} directive containing the text @var{string}. If this
7236 macro is not defined, nothing is output for a @samp{#ident} directive.
7239 @hook TARGET_ASM_NAMED_SECTION
7240 Output assembly directives to switch to section @var{name}. The section
7241 should have attributes as specified by @var{flags}, which is a bit mask
7242 of the @code{SECTION_*} flags defined in @file{output.h}. If @var{decl}
7243 is non-NULL, it is the @code{VAR_DECL} or @code{FUNCTION_DECL} with which
7244 this section is associated.
7247 @hook TARGET_ASM_FUNCTION_SECTION
7248 Return preferred text (sub)section for function @var{decl}.
7249 Main purpose of this function is to separate cold, normal and hot
7250 functions. @var{startup} is true when function is known to be used only
7251 at startup (from static constructors or it is @code{main()}).
7252 @var{exit} is true when function is known to be used only at exit
7253 (from static destructors).
7254 Return NULL if function should go to default text section.
7257 @hook TARGET_ASM_FUNCTION_SWITCHED_TEXT_SECTIONS
7259 @hook TARGET_HAVE_NAMED_SECTIONS
7260 This flag is true if the target supports @code{TARGET_ASM_NAMED_SECTION}.
7261 It must not be modified by command-line option processing.
7264 @anchor{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS}
7265 @hook TARGET_HAVE_SWITCHABLE_BSS_SECTIONS
7266 This flag is true if we can create zeroed data by switching to a BSS
7267 section and then using @code{ASM_OUTPUT_SKIP} to allocate the space.
7268 This is true on most ELF targets.
7271 @hook TARGET_SECTION_TYPE_FLAGS
7272 Choose a set of section attributes for use by @code{TARGET_ASM_NAMED_SECTION}
7273 based on a variable or function decl, a section name, and whether or not the
7274 declaration's initializer may contain runtime relocations. @var{decl} may be
7275 null, in which case read-write data should be assumed.
7277 The default version of this function handles choosing code vs data,
7278 read-only vs read-write data, and @code{flag_pic}. You should only
7279 need to override this if your target has special flags that might be
7280 set via @code{__attribute__}.
7283 @hook TARGET_ASM_RECORD_GCC_SWITCHES
7284 Provides the target with the ability to record the gcc command line
7285 switches that have been passed to the compiler, and options that are
7286 enabled. The @var{type} argument specifies what is being recorded.
7287 It can take the following values:
7290 @item SWITCH_TYPE_PASSED
7291 @var{text} is a command line switch that has been set by the user.
7293 @item SWITCH_TYPE_ENABLED
7294 @var{text} is an option which has been enabled. This might be as a
7295 direct result of a command line switch, or because it is enabled by
7296 default or because it has been enabled as a side effect of a different
7297 command line switch. For example, the @option{-O2} switch enables
7298 various different individual optimization passes.
7300 @item SWITCH_TYPE_DESCRIPTIVE
7301 @var{text} is either NULL or some descriptive text which should be
7302 ignored. If @var{text} is NULL then it is being used to warn the
7303 target hook that either recording is starting or ending. The first
7304 time @var{type} is SWITCH_TYPE_DESCRIPTIVE and @var{text} is NULL, the
7305 warning is for start up and the second time the warning is for
7306 wind down. This feature is to allow the target hook to make any
7307 necessary preparations before it starts to record switches and to
7308 perform any necessary tidying up after it has finished recording
7311 @item SWITCH_TYPE_LINE_START
7312 This option can be ignored by this target hook.
7314 @item SWITCH_TYPE_LINE_END
7315 This option can be ignored by this target hook.
7318 The hook's return value must be zero. Other return values may be
7319 supported in the future.
7321 By default this hook is set to NULL, but an example implementation is
7322 provided for ELF based targets. Called @var{elf_record_gcc_switches},
7323 it records the switches as ASCII text inside a new, string mergeable
7324 section in the assembler output file. The name of the new section is
7325 provided by the @code{TARGET_ASM_RECORD_GCC_SWITCHES_SECTION} target
7329 @hook TARGET_ASM_RECORD_GCC_SWITCHES_SECTION
7330 This is the name of the section that will be created by the example
7331 ELF implementation of the @code{TARGET_ASM_RECORD_GCC_SWITCHES} target
7337 @subsection Output of Data
7340 @hook TARGET_ASM_BYTE_OP
7341 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_HI_OP
7342 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_SI_OP
7343 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_DI_OP
7344 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_TI_OP
7345 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_HI_OP
7346 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_SI_OP
7347 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_DI_OP
7348 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_TI_OP
7349 These hooks specify assembly directives for creating certain kinds
7350 of integer object. The @code{TARGET_ASM_BYTE_OP} directive creates a
7351 byte-sized object, the @code{TARGET_ASM_ALIGNED_HI_OP} one creates an
7352 aligned two-byte object, and so on. Any of the hooks may be
7353 @code{NULL}, indicating that no suitable directive is available.
7355 The compiler will print these strings at the start of a new line,
7356 followed immediately by the object's initial value. In most cases,
7357 the string should contain a tab, a pseudo-op, and then another tab.
7360 @hook TARGET_ASM_INTEGER
7361 The @code{assemble_integer} function uses this hook to output an
7362 integer object. @var{x} is the object's value, @var{size} is its size
7363 in bytes and @var{aligned_p} indicates whether it is aligned. The
7364 function should return @code{true} if it was able to output the
7365 object. If it returns false, @code{assemble_integer} will try to
7366 split the object into smaller parts.
7368 The default implementation of this hook will use the
7369 @code{TARGET_ASM_BYTE_OP} family of strings, returning @code{false}
7370 when the relevant string is @code{NULL}.
7373 @hook TARGET_ASM_OUTPUT_ADDR_CONST_EXTRA
7374 A target hook to recognize @var{rtx} patterns that @code{output_addr_const}
7375 can't deal with, and output assembly code to @var{file} corresponding to
7376 the pattern @var{x}. This may be used to allow machine-dependent
7377 @code{UNSPEC}s to appear within constants.
7379 If target hook fails to recognize a pattern, it must return @code{false},
7380 so that a standard error message is printed. If it prints an error message
7381 itself, by calling, for example, @code{output_operand_lossage}, it may just
7385 @defmac OUTPUT_ADDR_CONST_EXTRA (@var{stream}, @var{x}, @var{fail})
7386 A C statement to recognize @var{rtx} patterns that
7387 @code{output_addr_const} can't deal with, and output assembly code to
7388 @var{stream} corresponding to the pattern @var{x}. This may be used to
7389 allow machine-dependent @code{UNSPEC}s to appear within constants.
7391 If @code{OUTPUT_ADDR_CONST_EXTRA} fails to recognize a pattern, it must
7392 @code{goto fail}, so that a standard error message is printed. If it
7393 prints an error message itself, by calling, for example,
7394 @code{output_operand_lossage}, it may just complete normally.
7397 @defmac ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len})
7398 A C statement to output to the stdio stream @var{stream} an assembler
7399 instruction to assemble a string constant containing the @var{len}
7400 bytes at @var{ptr}. @var{ptr} will be a C expression of type
7401 @code{char *} and @var{len} a C expression of type @code{int}.
7403 If the assembler has a @code{.ascii} pseudo-op as found in the
7404 Berkeley Unix assembler, do not define the macro
7405 @code{ASM_OUTPUT_ASCII}.
7408 @defmac ASM_OUTPUT_FDESC (@var{stream}, @var{decl}, @var{n})
7409 A C statement to output word @var{n} of a function descriptor for
7410 @var{decl}. This must be defined if @code{TARGET_VTABLE_USES_DESCRIPTORS}
7411 is defined, and is otherwise unused.
7414 @defmac CONSTANT_POOL_BEFORE_FUNCTION
7415 You may define this macro as a C expression. You should define the
7416 expression to have a nonzero value if GCC should output the constant
7417 pool for a function before the code for the function, or a zero value if
7418 GCC should output the constant pool after the function. If you do
7419 not define this macro, the usual case, GCC will output the constant
7420 pool before the function.
7423 @defmac ASM_OUTPUT_POOL_PROLOGUE (@var{file}, @var{funname}, @var{fundecl}, @var{size})
7424 A C statement to output assembler commands to define the start of the
7425 constant pool for a function. @var{funname} is a string giving
7426 the name of the function. Should the return type of the function
7427 be required, it can be obtained via @var{fundecl}. @var{size}
7428 is the size, in bytes, of the constant pool that will be written
7429 immediately after this call.
7431 If no constant-pool prefix is required, the usual case, this macro need
7435 @defmac ASM_OUTPUT_SPECIAL_POOL_ENTRY (@var{file}, @var{x}, @var{mode}, @var{align}, @var{labelno}, @var{jumpto})
7436 A C statement (with or without semicolon) to output a constant in the
7437 constant pool, if it needs special treatment. (This macro need not do
7438 anything for RTL expressions that can be output normally.)
7440 The argument @var{file} is the standard I/O stream to output the
7441 assembler code on. @var{x} is the RTL expression for the constant to
7442 output, and @var{mode} is the machine mode (in case @var{x} is a
7443 @samp{const_int}). @var{align} is the required alignment for the value
7444 @var{x}; you should output an assembler directive to force this much
7447 The argument @var{labelno} is a number to use in an internal label for
7448 the address of this pool entry. The definition of this macro is
7449 responsible for outputting the label definition at the proper place.
7450 Here is how to do this:
7453 @code{(*targetm.asm_out.internal_label)} (@var{file}, "LC", @var{labelno});
7456 When you output a pool entry specially, you should end with a
7457 @code{goto} to the label @var{jumpto}. This will prevent the same pool
7458 entry from being output a second time in the usual manner.
7460 You need not define this macro if it would do nothing.
7463 @defmac ASM_OUTPUT_POOL_EPILOGUE (@var{file} @var{funname} @var{fundecl} @var{size})
7464 A C statement to output assembler commands to at the end of the constant
7465 pool for a function. @var{funname} is a string giving the name of the
7466 function. Should the return type of the function be required, you can
7467 obtain it via @var{fundecl}. @var{size} is the size, in bytes, of the
7468 constant pool that GCC wrote immediately before this call.
7470 If no constant-pool epilogue is required, the usual case, you need not
7474 @defmac IS_ASM_LOGICAL_LINE_SEPARATOR (@var{C}, @var{STR})
7475 Define this macro as a C expression which is nonzero if @var{C} is
7476 used as a logical line separator by the assembler. @var{STR} points
7477 to the position in the string where @var{C} was found; this can be used if
7478 a line separator uses multiple characters.
7480 If you do not define this macro, the default is that only
7481 the character @samp{;} is treated as a logical line separator.
7484 @hook TARGET_ASM_OPEN_PAREN
7485 These target hooks are C string constants, describing the syntax in the
7486 assembler for grouping arithmetic expressions. If not overridden, they
7487 default to normal parentheses, which is correct for most assemblers.
7490 These macros are provided by @file{real.h} for writing the definitions
7491 of @code{ASM_OUTPUT_DOUBLE} and the like:
7493 @defmac REAL_VALUE_TO_TARGET_SINGLE (@var{x}, @var{l})
7494 @defmacx REAL_VALUE_TO_TARGET_DOUBLE (@var{x}, @var{l})
7495 @defmacx REAL_VALUE_TO_TARGET_LONG_DOUBLE (@var{x}, @var{l})
7496 @defmacx REAL_VALUE_TO_TARGET_DECIMAL32 (@var{x}, @var{l})
7497 @defmacx REAL_VALUE_TO_TARGET_DECIMAL64 (@var{x}, @var{l})
7498 @defmacx REAL_VALUE_TO_TARGET_DECIMAL128 (@var{x}, @var{l})
7499 These translate @var{x}, of type @code{REAL_VALUE_TYPE}, to the
7500 target's floating point representation, and store its bit pattern in
7501 the variable @var{l}. For @code{REAL_VALUE_TO_TARGET_SINGLE} and
7502 @code{REAL_VALUE_TO_TARGET_DECIMAL32}, this variable should be a
7503 simple @code{long int}. For the others, it should be an array of
7504 @code{long int}. The number of elements in this array is determined
7505 by the size of the desired target floating point data type: 32 bits of
7506 it go in each @code{long int} array element. Each array element holds
7507 32 bits of the result, even if @code{long int} is wider than 32 bits
7508 on the host machine.
7510 The array element values are designed so that you can print them out
7511 using @code{fprintf} in the order they should appear in the target
7515 @node Uninitialized Data
7516 @subsection Output of Uninitialized Variables
7518 Each of the macros in this section is used to do the whole job of
7519 outputting a single uninitialized variable.
7521 @defmac ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded})
7522 A C statement (sans semicolon) to output to the stdio stream
7523 @var{stream} the assembler definition of a common-label named
7524 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
7525 is the size rounded up to whatever alignment the caller wants. It is
7526 possible that @var{size} may be zero, for instance if a struct with no
7527 other member than a zero-length array is defined. In this case, the
7528 backend must output a symbol definition that allocates at least one
7529 byte, both so that the address of the resulting object does not compare
7530 equal to any other, and because some object formats cannot even express
7531 the concept of a zero-sized common symbol, as that is how they represent
7532 an ordinary undefined external.
7534 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7535 output the name itself; before and after that, output the additional
7536 assembler syntax for defining the name, and a newline.
7538 This macro controls how the assembler definitions of uninitialized
7539 common global variables are output.
7542 @defmac ASM_OUTPUT_ALIGNED_COMMON (@var{stream}, @var{name}, @var{size}, @var{alignment})
7543 Like @code{ASM_OUTPUT_COMMON} except takes the required alignment as a
7544 separate, explicit argument. If you define this macro, it is used in
7545 place of @code{ASM_OUTPUT_COMMON}, and gives you more flexibility in
7546 handling the required alignment of the variable. The alignment is specified
7547 as the number of bits.
7550 @defmac ASM_OUTPUT_ALIGNED_DECL_COMMON (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
7551 Like @code{ASM_OUTPUT_ALIGNED_COMMON} except that @var{decl} of the
7552 variable to be output, if there is one, or @code{NULL_TREE} if there
7553 is no corresponding variable. If you define this macro, GCC will use it
7554 in place of both @code{ASM_OUTPUT_COMMON} and
7555 @code{ASM_OUTPUT_ALIGNED_COMMON}. Define this macro when you need to see
7556 the variable's decl in order to chose what to output.
7559 @defmac ASM_OUTPUT_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{rounded})
7560 A C statement (sans semicolon) to output to the stdio stream
7561 @var{stream} the assembler definition of uninitialized global @var{decl} named
7562 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
7563 is the size rounded up to whatever alignment the caller wants.
7565 Try to use function @code{asm_output_bss} defined in @file{varasm.c} when
7566 defining this macro. If unable, use the expression
7567 @code{assemble_name (@var{stream}, @var{name})} to output the name itself;
7568 before and after that, output the additional assembler syntax for defining
7569 the name, and a newline.
7571 There are two ways of handling global BSS@. One is to define either
7572 this macro or its aligned counterpart, @code{ASM_OUTPUT_ALIGNED_BSS}.
7573 The other is to have @code{TARGET_ASM_SELECT_SECTION} return a
7574 switchable BSS section (@pxref{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS}).
7575 You do not need to do both.
7577 Some languages do not have @code{common} data, and require a
7578 non-common form of global BSS in order to handle uninitialized globals
7579 efficiently. C++ is one example of this. However, if the target does
7580 not support global BSS, the front end may choose to make globals
7581 common in order to save space in the object file.
7584 @defmac ASM_OUTPUT_ALIGNED_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
7585 Like @code{ASM_OUTPUT_BSS} except takes the required alignment as a
7586 separate, explicit argument. If you define this macro, it is used in
7587 place of @code{ASM_OUTPUT_BSS}, and gives you more flexibility in
7588 handling the required alignment of the variable. The alignment is specified
7589 as the number of bits.
7591 Try to use function @code{asm_output_aligned_bss} defined in file
7592 @file{varasm.c} when defining this macro.
7595 @defmac ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded})
7596 A C statement (sans semicolon) to output to the stdio stream
7597 @var{stream} the assembler definition of a local-common-label named
7598 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
7599 is the size rounded up to whatever alignment the caller wants.
7601 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7602 output the name itself; before and after that, output the additional
7603 assembler syntax for defining the name, and a newline.
7605 This macro controls how the assembler definitions of uninitialized
7606 static variables are output.
7609 @defmac ASM_OUTPUT_ALIGNED_LOCAL (@var{stream}, @var{name}, @var{size}, @var{alignment})
7610 Like @code{ASM_OUTPUT_LOCAL} except takes the required alignment as a
7611 separate, explicit argument. If you define this macro, it is used in
7612 place of @code{ASM_OUTPUT_LOCAL}, and gives you more flexibility in
7613 handling the required alignment of the variable. The alignment is specified
7614 as the number of bits.
7617 @defmac ASM_OUTPUT_ALIGNED_DECL_LOCAL (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
7618 Like @code{ASM_OUTPUT_ALIGNED_DECL} except that @var{decl} of the
7619 variable to be output, if there is one, or @code{NULL_TREE} if there
7620 is no corresponding variable. If you define this macro, GCC will use it
7621 in place of both @code{ASM_OUTPUT_DECL} and
7622 @code{ASM_OUTPUT_ALIGNED_DECL}. Define this macro when you need to see
7623 the variable's decl in order to chose what to output.
7627 @subsection Output and Generation of Labels
7629 @c prevent bad page break with this line
7630 This is about outputting labels.
7632 @findex assemble_name
7633 @defmac ASM_OUTPUT_LABEL (@var{stream}, @var{name})
7634 A C statement (sans semicolon) to output to the stdio stream
7635 @var{stream} the assembler definition of a label named @var{name}.
7636 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7637 output the name itself; before and after that, output the additional
7638 assembler syntax for defining the name, and a newline. A default
7639 definition of this macro is provided which is correct for most systems.
7642 @defmac ASM_OUTPUT_FUNCTION_LABEL (@var{stream}, @var{name}, @var{decl})
7643 A C statement (sans semicolon) to output to the stdio stream
7644 @var{stream} the assembler definition of a label named @var{name} of
7646 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7647 output the name itself; before and after that, output the additional
7648 assembler syntax for defining the name, and a newline. A default
7649 definition of this macro is provided which is correct for most systems.
7651 If this macro is not defined, then the function name is defined in the
7652 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
7655 @findex assemble_name_raw
7656 @defmac ASM_OUTPUT_INTERNAL_LABEL (@var{stream}, @var{name})
7657 Identical to @code{ASM_OUTPUT_LABEL}, except that @var{name} is known
7658 to refer to a compiler-generated label. The default definition uses
7659 @code{assemble_name_raw}, which is like @code{assemble_name} except
7660 that it is more efficient.
7664 A C string containing the appropriate assembler directive to specify the
7665 size of a symbol, without any arguments. On systems that use ELF, the
7666 default (in @file{config/elfos.h}) is @samp{"\t.size\t"}; on other
7667 systems, the default is not to define this macro.
7669 Define this macro only if it is correct to use the default definitions
7670 of @code{ASM_OUTPUT_SIZE_DIRECTIVE} and @code{ASM_OUTPUT_MEASURED_SIZE}
7671 for your system. If you need your own custom definitions of those
7672 macros, or if you do not need explicit symbol sizes at all, do not
7676 @defmac ASM_OUTPUT_SIZE_DIRECTIVE (@var{stream}, @var{name}, @var{size})
7677 A C statement (sans semicolon) to output to the stdio stream
7678 @var{stream} a directive telling the assembler that the size of the
7679 symbol @var{name} is @var{size}. @var{size} is a @code{HOST_WIDE_INT}.
7680 If you define @code{SIZE_ASM_OP}, a default definition of this macro is
7684 @defmac ASM_OUTPUT_MEASURED_SIZE (@var{stream}, @var{name})
7685 A C statement (sans semicolon) to output to the stdio stream
7686 @var{stream} a directive telling the assembler to calculate the size of
7687 the symbol @var{name} by subtracting its address from the current
7690 If you define @code{SIZE_ASM_OP}, a default definition of this macro is
7691 provided. The default assumes that the assembler recognizes a special
7692 @samp{.} symbol as referring to the current address, and can calculate
7693 the difference between this and another symbol. If your assembler does
7694 not recognize @samp{.} or cannot do calculations with it, you will need
7695 to redefine @code{ASM_OUTPUT_MEASURED_SIZE} to use some other technique.
7699 A C string containing the appropriate assembler directive to specify the
7700 type of a symbol, without any arguments. On systems that use ELF, the
7701 default (in @file{config/elfos.h}) is @samp{"\t.type\t"}; on other
7702 systems, the default is not to define this macro.
7704 Define this macro only if it is correct to use the default definition of
7705 @code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own
7706 custom definition of this macro, or if you do not need explicit symbol
7707 types at all, do not define this macro.
7710 @defmac TYPE_OPERAND_FMT
7711 A C string which specifies (using @code{printf} syntax) the format of
7712 the second operand to @code{TYPE_ASM_OP}. On systems that use ELF, the
7713 default (in @file{config/elfos.h}) is @samp{"@@%s"}; on other systems,
7714 the default is not to define this macro.
7716 Define this macro only if it is correct to use the default definition of
7717 @code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own
7718 custom definition of this macro, or if you do not need explicit symbol
7719 types at all, do not define this macro.
7722 @defmac ASM_OUTPUT_TYPE_DIRECTIVE (@var{stream}, @var{type})
7723 A C statement (sans semicolon) to output to the stdio stream
7724 @var{stream} a directive telling the assembler that the type of the
7725 symbol @var{name} is @var{type}. @var{type} is a C string; currently,
7726 that string is always either @samp{"function"} or @samp{"object"}, but
7727 you should not count on this.
7729 If you define @code{TYPE_ASM_OP} and @code{TYPE_OPERAND_FMT}, a default
7730 definition of this macro is provided.
7733 @defmac ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl})
7734 A C statement (sans semicolon) to output to the stdio stream
7735 @var{stream} any text necessary for declaring the name @var{name} of a
7736 function which is being defined. This macro is responsible for
7737 outputting the label definition (perhaps using
7738 @code{ASM_OUTPUT_FUNCTION_LABEL}). The argument @var{decl} is the
7739 @code{FUNCTION_DECL} tree node representing the function.
7741 If this macro is not defined, then the function name is defined in the
7742 usual manner as a label (by means of @code{ASM_OUTPUT_FUNCTION_LABEL}).
7744 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition
7748 @defmac ASM_DECLARE_FUNCTION_SIZE (@var{stream}, @var{name}, @var{decl})
7749 A C statement (sans semicolon) to output to the stdio stream
7750 @var{stream} any text necessary for declaring the size of a function
7751 which is being defined. The argument @var{name} is the name of the
7752 function. The argument @var{decl} is the @code{FUNCTION_DECL} tree node
7753 representing the function.
7755 If this macro is not defined, then the function size is not defined.
7757 You may wish to use @code{ASM_OUTPUT_MEASURED_SIZE} in the definition
7761 @defmac ASM_DECLARE_OBJECT_NAME (@var{stream}, @var{name}, @var{decl})
7762 A C statement (sans semicolon) to output to the stdio stream
7763 @var{stream} any text necessary for declaring the name @var{name} of an
7764 initialized variable which is being defined. This macro must output the
7765 label definition (perhaps using @code{ASM_OUTPUT_LABEL}). The argument
7766 @var{decl} is the @code{VAR_DECL} tree node representing the variable.
7768 If this macro is not defined, then the variable name is defined in the
7769 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
7771 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} and/or
7772 @code{ASM_OUTPUT_SIZE_DIRECTIVE} in the definition of this macro.
7775 @hook TARGET_ASM_DECLARE_CONSTANT_NAME
7776 A target hook to output to the stdio stream @var{file} any text necessary
7777 for declaring the name @var{name} of a constant which is being defined. This
7778 target hook is responsible for outputting the label definition (perhaps using
7779 @code{assemble_label}). The argument @var{exp} is the value of the constant,
7780 and @var{size} is the size of the constant in bytes. The @var{name}
7781 will be an internal label.
7783 The default version of this target hook, define the @var{name} in the
7784 usual manner as a label (by means of @code{assemble_label}).
7786 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in this target hook.
7789 @defmac ASM_DECLARE_REGISTER_GLOBAL (@var{stream}, @var{decl}, @var{regno}, @var{name})
7790 A C statement (sans semicolon) to output to the stdio stream
7791 @var{stream} any text necessary for claiming a register @var{regno}
7792 for a global variable @var{decl} with name @var{name}.
7794 If you don't define this macro, that is equivalent to defining it to do
7798 @defmac ASM_FINISH_DECLARE_OBJECT (@var{stream}, @var{decl}, @var{toplevel}, @var{atend})
7799 A C statement (sans semicolon) to finish up declaring a variable name
7800 once the compiler has processed its initializer fully and thus has had a
7801 chance to determine the size of an array when controlled by an
7802 initializer. This is used on systems where it's necessary to declare
7803 something about the size of the object.
7805 If you don't define this macro, that is equivalent to defining it to do
7808 You may wish to use @code{ASM_OUTPUT_SIZE_DIRECTIVE} and/or
7809 @code{ASM_OUTPUT_MEASURED_SIZE} in the definition of this macro.
7812 @hook TARGET_ASM_GLOBALIZE_LABEL
7813 This target hook is a function to output to the stdio stream
7814 @var{stream} some commands that will make the label @var{name} global;
7815 that is, available for reference from other files.
7817 The default implementation relies on a proper definition of
7818 @code{GLOBAL_ASM_OP}.
7821 @hook TARGET_ASM_GLOBALIZE_DECL_NAME
7822 This target hook is a function to output to the stdio stream
7823 @var{stream} some commands that will make the name associated with @var{decl}
7824 global; that is, available for reference from other files.
7826 The default implementation uses the TARGET_ASM_GLOBALIZE_LABEL target hook.
7829 @defmac ASM_WEAKEN_LABEL (@var{stream}, @var{name})
7830 A C statement (sans semicolon) to output to the stdio stream
7831 @var{stream} some commands that will make the label @var{name} weak;
7832 that is, available for reference from other files but only used if
7833 no other definition is available. Use the expression
7834 @code{assemble_name (@var{stream}, @var{name})} to output the name
7835 itself; before and after that, output the additional assembler syntax
7836 for making that name weak, and a newline.
7838 If you don't define this macro or @code{ASM_WEAKEN_DECL}, GCC will not
7839 support weak symbols and you should not define the @code{SUPPORTS_WEAK}
7843 @defmac ASM_WEAKEN_DECL (@var{stream}, @var{decl}, @var{name}, @var{value})
7844 Combines (and replaces) the function of @code{ASM_WEAKEN_LABEL} and
7845 @code{ASM_OUTPUT_WEAK_ALIAS}, allowing access to the associated function
7846 or variable decl. If @var{value} is not @code{NULL}, this C statement
7847 should output to the stdio stream @var{stream} assembler code which
7848 defines (equates) the weak symbol @var{name} to have the value
7849 @var{value}. If @var{value} is @code{NULL}, it should output commands
7850 to make @var{name} weak.
7853 @defmac ASM_OUTPUT_WEAKREF (@var{stream}, @var{decl}, @var{name}, @var{value})
7854 Outputs a directive that enables @var{name} to be used to refer to
7855 symbol @var{value} with weak-symbol semantics. @code{decl} is the
7856 declaration of @code{name}.
7859 @defmac SUPPORTS_WEAK
7860 A preprocessor constant expression which evaluates to true if the target
7861 supports weak symbols.
7863 If you don't define this macro, @file{defaults.h} provides a default
7864 definition. If either @code{ASM_WEAKEN_LABEL} or @code{ASM_WEAKEN_DECL}
7865 is defined, the default definition is @samp{1}; otherwise, it is @samp{0}.
7868 @defmac TARGET_SUPPORTS_WEAK
7869 A C expression which evaluates to true if the target supports weak symbols.
7871 If you don't define this macro, @file{defaults.h} provides a default
7872 definition. The default definition is @samp{(SUPPORTS_WEAK)}. Define
7873 this macro if you want to control weak symbol support with a compiler
7874 flag such as @option{-melf}.
7877 @defmac MAKE_DECL_ONE_ONLY (@var{decl})
7878 A C statement (sans semicolon) to mark @var{decl} to be emitted as a
7879 public symbol such that extra copies in multiple translation units will
7880 be discarded by the linker. Define this macro if your object file
7881 format provides support for this concept, such as the @samp{COMDAT}
7882 section flags in the Microsoft Windows PE/COFF format, and this support
7883 requires changes to @var{decl}, such as putting it in a separate section.
7886 @defmac SUPPORTS_ONE_ONLY
7887 A C expression which evaluates to true if the target supports one-only
7890 If you don't define this macro, @file{varasm.c} provides a default
7891 definition. If @code{MAKE_DECL_ONE_ONLY} is defined, the default
7892 definition is @samp{1}; otherwise, it is @samp{0}. Define this macro if
7893 you want to control one-only symbol support with a compiler flag, or if
7894 setting the @code{DECL_ONE_ONLY} flag is enough to mark a declaration to
7895 be emitted as one-only.
7898 @hook TARGET_ASM_ASSEMBLE_VISIBILITY
7899 This target hook is a function to output to @var{asm_out_file} some
7900 commands that will make the symbol(s) associated with @var{decl} have
7901 hidden, protected or internal visibility as specified by @var{visibility}.
7904 @defmac TARGET_WEAK_NOT_IN_ARCHIVE_TOC
7905 A C expression that evaluates to true if the target's linker expects
7906 that weak symbols do not appear in a static archive's table of contents.
7907 The default is @code{0}.
7909 Leaving weak symbols out of an archive's table of contents means that,
7910 if a symbol will only have a definition in one translation unit and
7911 will have undefined references from other translation units, that
7912 symbol should not be weak. Defining this macro to be nonzero will
7913 thus have the effect that certain symbols that would normally be weak
7914 (explicit template instantiations, and vtables for polymorphic classes
7915 with noninline key methods) will instead be nonweak.
7917 The C++ ABI requires this macro to be zero. Define this macro for
7918 targets where full C++ ABI compliance is impossible and where linker
7919 restrictions require weak symbols to be left out of a static archive's
7923 @defmac ASM_OUTPUT_EXTERNAL (@var{stream}, @var{decl}, @var{name})
7924 A C statement (sans semicolon) to output to the stdio stream
7925 @var{stream} any text necessary for declaring the name of an external
7926 symbol named @var{name} which is referenced in this compilation but
7927 not defined. The value of @var{decl} is the tree node for the
7930 This macro need not be defined if it does not need to output anything.
7931 The GNU assembler and most Unix assemblers don't require anything.
7934 @hook TARGET_ASM_EXTERNAL_LIBCALL
7935 This target hook is a function to output to @var{asm_out_file} an assembler
7936 pseudo-op to declare a library function name external. The name of the
7937 library function is given by @var{symref}, which is a @code{symbol_ref}.
7940 @hook TARGET_ASM_MARK_DECL_PRESERVED
7941 This target hook is a function to output to @var{asm_out_file} an assembler
7942 directive to annotate @var{symbol} as used. The Darwin target uses the
7943 .no_dead_code_strip directive.
7946 @defmac ASM_OUTPUT_LABELREF (@var{stream}, @var{name})
7947 A C statement (sans semicolon) to output to the stdio stream
7948 @var{stream} a reference in assembler syntax to a label named
7949 @var{name}. This should add @samp{_} to the front of the name, if that
7950 is customary on your operating system, as it is in most Berkeley Unix
7951 systems. This macro is used in @code{assemble_name}.
7954 @hook TARGET_MANGLE_ASSEMBLER_NAME
7956 @defmac ASM_OUTPUT_SYMBOL_REF (@var{stream}, @var{sym})
7957 A C statement (sans semicolon) to output a reference to
7958 @code{SYMBOL_REF} @var{sym}. If not defined, @code{assemble_name}
7959 will be used to output the name of the symbol. This macro may be used
7960 to modify the way a symbol is referenced depending on information
7961 encoded by @code{TARGET_ENCODE_SECTION_INFO}.
7964 @defmac ASM_OUTPUT_LABEL_REF (@var{stream}, @var{buf})
7965 A C statement (sans semicolon) to output a reference to @var{buf}, the
7966 result of @code{ASM_GENERATE_INTERNAL_LABEL}. If not defined,
7967 @code{assemble_name} will be used to output the name of the symbol.
7968 This macro is not used by @code{output_asm_label}, or the @code{%l}
7969 specifier that calls it; the intention is that this macro should be set
7970 when it is necessary to output a label differently when its address is
7974 @hook TARGET_ASM_INTERNAL_LABEL
7975 A function to output to the stdio stream @var{stream} a label whose
7976 name is made from the string @var{prefix} and the number @var{labelno}.
7978 It is absolutely essential that these labels be distinct from the labels
7979 used for user-level functions and variables. Otherwise, certain programs
7980 will have name conflicts with internal labels.
7982 It is desirable to exclude internal labels from the symbol table of the
7983 object file. Most assemblers have a naming convention for labels that
7984 should be excluded; on many systems, the letter @samp{L} at the
7985 beginning of a label has this effect. You should find out what
7986 convention your system uses, and follow it.
7988 The default version of this function utilizes @code{ASM_GENERATE_INTERNAL_LABEL}.
7991 @defmac ASM_OUTPUT_DEBUG_LABEL (@var{stream}, @var{prefix}, @var{num})
7992 A C statement to output to the stdio stream @var{stream} a debug info
7993 label whose name is made from the string @var{prefix} and the number
7994 @var{num}. This is useful for VLIW targets, where debug info labels
7995 may need to be treated differently than branch target labels. On some
7996 systems, branch target labels must be at the beginning of instruction
7997 bundles, but debug info labels can occur in the middle of instruction
8000 If this macro is not defined, then @code{(*targetm.asm_out.internal_label)} will be
8004 @defmac ASM_GENERATE_INTERNAL_LABEL (@var{string}, @var{prefix}, @var{num})
8005 A C statement to store into the string @var{string} a label whose name
8006 is made from the string @var{prefix} and the number @var{num}.
8008 This string, when output subsequently by @code{assemble_name}, should
8009 produce the output that @code{(*targetm.asm_out.internal_label)} would produce
8010 with the same @var{prefix} and @var{num}.
8012 If the string begins with @samp{*}, then @code{assemble_name} will
8013 output the rest of the string unchanged. It is often convenient for
8014 @code{ASM_GENERATE_INTERNAL_LABEL} to use @samp{*} in this way. If the
8015 string doesn't start with @samp{*}, then @code{ASM_OUTPUT_LABELREF} gets
8016 to output the string, and may change it. (Of course,
8017 @code{ASM_OUTPUT_LABELREF} is also part of your machine description, so
8018 you should know what it does on your machine.)
8021 @defmac ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number})
8022 A C expression to assign to @var{outvar} (which is a variable of type
8023 @code{char *}) a newly allocated string made from the string
8024 @var{name} and the number @var{number}, with some suitable punctuation
8025 added. Use @code{alloca} to get space for the string.
8027 The string will be used as an argument to @code{ASM_OUTPUT_LABELREF} to
8028 produce an assembler label for an internal static variable whose name is
8029 @var{name}. Therefore, the string must be such as to result in valid
8030 assembler code. The argument @var{number} is different each time this
8031 macro is executed; it prevents conflicts between similarly-named
8032 internal static variables in different scopes.
8034 Ideally this string should not be a valid C identifier, to prevent any
8035 conflict with the user's own symbols. Most assemblers allow periods
8036 or percent signs in assembler symbols; putting at least one of these
8037 between the name and the number will suffice.
8039 If this macro is not defined, a default definition will be provided
8040 which is correct for most systems.
8043 @defmac ASM_OUTPUT_DEF (@var{stream}, @var{name}, @var{value})
8044 A C statement to output to the stdio stream @var{stream} assembler code
8045 which defines (equates) the symbol @var{name} to have the value @var{value}.
8048 If @code{SET_ASM_OP} is defined, a default definition is provided which is
8049 correct for most systems.
8052 @defmac ASM_OUTPUT_DEF_FROM_DECLS (@var{stream}, @var{decl_of_name}, @var{decl_of_value})
8053 A C statement to output to the stdio stream @var{stream} assembler code
8054 which defines (equates) the symbol whose tree node is @var{decl_of_name}
8055 to have the value of the tree node @var{decl_of_value}. This macro will
8056 be used in preference to @samp{ASM_OUTPUT_DEF} if it is defined and if
8057 the tree nodes are available.
8060 If @code{SET_ASM_OP} is defined, a default definition is provided which is
8061 correct for most systems.
8064 @defmac TARGET_DEFERRED_OUTPUT_DEFS (@var{decl_of_name}, @var{decl_of_value})
8065 A C statement that evaluates to true if the assembler code which defines
8066 (equates) the symbol whose tree node is @var{decl_of_name} to have the value
8067 of the tree node @var{decl_of_value} should be emitted near the end of the
8068 current compilation unit. The default is to not defer output of defines.
8069 This macro affects defines output by @samp{ASM_OUTPUT_DEF} and
8070 @samp{ASM_OUTPUT_DEF_FROM_DECLS}.
8073 @defmac ASM_OUTPUT_WEAK_ALIAS (@var{stream}, @var{name}, @var{value})
8074 A C statement to output to the stdio stream @var{stream} assembler code
8075 which defines (equates) the weak symbol @var{name} to have the value
8076 @var{value}. If @var{value} is @code{NULL}, it defines @var{name} as
8077 an undefined weak symbol.
8079 Define this macro if the target only supports weak aliases; define
8080 @code{ASM_OUTPUT_DEF} instead if possible.
8083 @defmac OBJC_GEN_METHOD_LABEL (@var{buf}, @var{is_inst}, @var{class_name}, @var{cat_name}, @var{sel_name})
8084 Define this macro to override the default assembler names used for
8085 Objective-C methods.
8087 The default name is a unique method number followed by the name of the
8088 class (e.g.@: @samp{_1_Foo}). For methods in categories, the name of
8089 the category is also included in the assembler name (e.g.@:
8092 These names are safe on most systems, but make debugging difficult since
8093 the method's selector is not present in the name. Therefore, particular
8094 systems define other ways of computing names.
8096 @var{buf} is an expression of type @code{char *} which gives you a
8097 buffer in which to store the name; its length is as long as
8098 @var{class_name}, @var{cat_name} and @var{sel_name} put together, plus
8099 50 characters extra.
8101 The argument @var{is_inst} specifies whether the method is an instance
8102 method or a class method; @var{class_name} is the name of the class;
8103 @var{cat_name} is the name of the category (or @code{NULL} if the method is not
8104 in a category); and @var{sel_name} is the name of the selector.
8106 On systems where the assembler can handle quoted names, you can use this
8107 macro to provide more human-readable names.
8110 @defmac ASM_DECLARE_CLASS_REFERENCE (@var{stream}, @var{name})
8111 A C statement (sans semicolon) to output to the stdio stream
8112 @var{stream} commands to declare that the label @var{name} is an
8113 Objective-C class reference. This is only needed for targets whose
8114 linkers have special support for NeXT-style runtimes.
8117 @defmac ASM_DECLARE_UNRESOLVED_REFERENCE (@var{stream}, @var{name})
8118 A C statement (sans semicolon) to output to the stdio stream
8119 @var{stream} commands to declare that the label @var{name} is an
8120 unresolved Objective-C class reference. This is only needed for targets
8121 whose linkers have special support for NeXT-style runtimes.
8124 @node Initialization
8125 @subsection How Initialization Functions Are Handled
8126 @cindex initialization routines
8127 @cindex termination routines
8128 @cindex constructors, output of
8129 @cindex destructors, output of
8131 The compiled code for certain languages includes @dfn{constructors}
8132 (also called @dfn{initialization routines})---functions to initialize
8133 data in the program when the program is started. These functions need
8134 to be called before the program is ``started''---that is to say, before
8135 @code{main} is called.
8137 Compiling some languages generates @dfn{destructors} (also called
8138 @dfn{termination routines}) that should be called when the program
8141 To make the initialization and termination functions work, the compiler
8142 must output something in the assembler code to cause those functions to
8143 be called at the appropriate time. When you port the compiler to a new
8144 system, you need to specify how to do this.
8146 There are two major ways that GCC currently supports the execution of
8147 initialization and termination functions. Each way has two variants.
8148 Much of the structure is common to all four variations.
8150 @findex __CTOR_LIST__
8151 @findex __DTOR_LIST__
8152 The linker must build two lists of these functions---a list of
8153 initialization functions, called @code{__CTOR_LIST__}, and a list of
8154 termination functions, called @code{__DTOR_LIST__}.
8156 Each list always begins with an ignored function pointer (which may hold
8157 0, @minus{}1, or a count of the function pointers after it, depending on
8158 the environment). This is followed by a series of zero or more function
8159 pointers to constructors (or destructors), followed by a function
8160 pointer containing zero.
8162 Depending on the operating system and its executable file format, either
8163 @file{crtstuff.c} or @file{libgcc2.c} traverses these lists at startup
8164 time and exit time. Constructors are called in reverse order of the
8165 list; destructors in forward order.
8167 The best way to handle static constructors works only for object file
8168 formats which provide arbitrarily-named sections. A section is set
8169 aside for a list of constructors, and another for a list of destructors.
8170 Traditionally these are called @samp{.ctors} and @samp{.dtors}. Each
8171 object file that defines an initialization function also puts a word in
8172 the constructor section to point to that function. The linker
8173 accumulates all these words into one contiguous @samp{.ctors} section.
8174 Termination functions are handled similarly.
8176 This method will be chosen as the default by @file{target-def.h} if
8177 @code{TARGET_ASM_NAMED_SECTION} is defined. A target that does not
8178 support arbitrary sections, but does support special designated
8179 constructor and destructor sections may define @code{CTORS_SECTION_ASM_OP}
8180 and @code{DTORS_SECTION_ASM_OP} to achieve the same effect.
8182 When arbitrary sections are available, there are two variants, depending
8183 upon how the code in @file{crtstuff.c} is called. On systems that
8184 support a @dfn{.init} section which is executed at program startup,
8185 parts of @file{crtstuff.c} are compiled into that section. The
8186 program is linked by the @command{gcc} driver like this:
8189 ld -o @var{output_file} crti.o crtbegin.o @dots{} -lgcc crtend.o crtn.o
8192 The prologue of a function (@code{__init}) appears in the @code{.init}
8193 section of @file{crti.o}; the epilogue appears in @file{crtn.o}. Likewise
8194 for the function @code{__fini} in the @dfn{.fini} section. Normally these
8195 files are provided by the operating system or by the GNU C library, but
8196 are provided by GCC for a few targets.
8198 The objects @file{crtbegin.o} and @file{crtend.o} are (for most targets)
8199 compiled from @file{crtstuff.c}. They contain, among other things, code
8200 fragments within the @code{.init} and @code{.fini} sections that branch
8201 to routines in the @code{.text} section. The linker will pull all parts
8202 of a section together, which results in a complete @code{__init} function
8203 that invokes the routines we need at startup.
8205 To use this variant, you must define the @code{INIT_SECTION_ASM_OP}
8208 If no init section is available, when GCC compiles any function called
8209 @code{main} (or more accurately, any function designated as a program
8210 entry point by the language front end calling @code{expand_main_function}),
8211 it inserts a procedure call to @code{__main} as the first executable code
8212 after the function prologue. The @code{__main} function is defined
8213 in @file{libgcc2.c} and runs the global constructors.
8215 In file formats that don't support arbitrary sections, there are again
8216 two variants. In the simplest variant, the GNU linker (GNU @code{ld})
8217 and an `a.out' format must be used. In this case,
8218 @code{TARGET_ASM_CONSTRUCTOR} is defined to produce a @code{.stabs}
8219 entry of type @samp{N_SETT}, referencing the name @code{__CTOR_LIST__},
8220 and with the address of the void function containing the initialization
8221 code as its value. The GNU linker recognizes this as a request to add
8222 the value to a @dfn{set}; the values are accumulated, and are eventually
8223 placed in the executable as a vector in the format described above, with
8224 a leading (ignored) count and a trailing zero element.
8225 @code{TARGET_ASM_DESTRUCTOR} is handled similarly. Since no init
8226 section is available, the absence of @code{INIT_SECTION_ASM_OP} causes
8227 the compilation of @code{main} to call @code{__main} as above, starting
8228 the initialization process.
8230 The last variant uses neither arbitrary sections nor the GNU linker.
8231 This is preferable when you want to do dynamic linking and when using
8232 file formats which the GNU linker does not support, such as `ECOFF'@. In
8233 this case, @code{TARGET_HAVE_CTORS_DTORS} is false, initialization and
8234 termination functions are recognized simply by their names. This requires
8235 an extra program in the linkage step, called @command{collect2}. This program
8236 pretends to be the linker, for use with GCC; it does its job by running
8237 the ordinary linker, but also arranges to include the vectors of
8238 initialization and termination functions. These functions are called
8239 via @code{__main} as described above. In order to use this method,
8240 @code{use_collect2} must be defined in the target in @file{config.gcc}.
8243 The following section describes the specific macros that control and
8244 customize the handling of initialization and termination functions.
8247 @node Macros for Initialization
8248 @subsection Macros Controlling Initialization Routines
8250 Here are the macros that control how the compiler handles initialization
8251 and termination functions:
8253 @defmac INIT_SECTION_ASM_OP
8254 If defined, a C string constant, including spacing, for the assembler
8255 operation to identify the following data as initialization code. If not
8256 defined, GCC will assume such a section does not exist. When you are
8257 using special sections for initialization and termination functions, this
8258 macro also controls how @file{crtstuff.c} and @file{libgcc2.c} arrange to
8259 run the initialization functions.
8262 @defmac HAS_INIT_SECTION
8263 If defined, @code{main} will not call @code{__main} as described above.
8264 This macro should be defined for systems that control start-up code
8265 on a symbol-by-symbol basis, such as OSF/1, and should not
8266 be defined explicitly for systems that support @code{INIT_SECTION_ASM_OP}.
8269 @defmac LD_INIT_SWITCH
8270 If defined, a C string constant for a switch that tells the linker that
8271 the following symbol is an initialization routine.
8274 @defmac LD_FINI_SWITCH
8275 If defined, a C string constant for a switch that tells the linker that
8276 the following symbol is a finalization routine.
8279 @defmac COLLECT_SHARED_INIT_FUNC (@var{stream}, @var{func})
8280 If defined, a C statement that will write a function that can be
8281 automatically called when a shared library is loaded. The function
8282 should call @var{func}, which takes no arguments. If not defined, and
8283 the object format requires an explicit initialization function, then a
8284 function called @code{_GLOBAL__DI} will be generated.
8286 This function and the following one are used by collect2 when linking a
8287 shared library that needs constructors or destructors, or has DWARF2
8288 exception tables embedded in the code.
8291 @defmac COLLECT_SHARED_FINI_FUNC (@var{stream}, @var{func})
8292 If defined, a C statement that will write a function that can be
8293 automatically called when a shared library is unloaded. The function
8294 should call @var{func}, which takes no arguments. If not defined, and
8295 the object format requires an explicit finalization function, then a
8296 function called @code{_GLOBAL__DD} will be generated.
8299 @defmac INVOKE__main
8300 If defined, @code{main} will call @code{__main} despite the presence of
8301 @code{INIT_SECTION_ASM_OP}. This macro should be defined for systems
8302 where the init section is not actually run automatically, but is still
8303 useful for collecting the lists of constructors and destructors.
8306 @defmac SUPPORTS_INIT_PRIORITY
8307 If nonzero, the C++ @code{init_priority} attribute is supported and the
8308 compiler should emit instructions to control the order of initialization
8309 of objects. If zero, the compiler will issue an error message upon
8310 encountering an @code{init_priority} attribute.
8313 @hook TARGET_HAVE_CTORS_DTORS
8314 This value is true if the target supports some ``native'' method of
8315 collecting constructors and destructors to be run at startup and exit.
8316 It is false if we must use @command{collect2}.
8319 @hook TARGET_ASM_CONSTRUCTOR
8320 If defined, a function that outputs assembler code to arrange to call
8321 the function referenced by @var{symbol} at initialization time.
8323 Assume that @var{symbol} is a @code{SYMBOL_REF} for a function taking
8324 no arguments and with no return value. If the target supports initialization
8325 priorities, @var{priority} is a value between 0 and @code{MAX_INIT_PRIORITY};
8326 otherwise it must be @code{DEFAULT_INIT_PRIORITY}.
8328 If this macro is not defined by the target, a suitable default will
8329 be chosen if (1) the target supports arbitrary section names, (2) the
8330 target defines @code{CTORS_SECTION_ASM_OP}, or (3) @code{USE_COLLECT2}
8334 @hook TARGET_ASM_DESTRUCTOR
8335 This is like @code{TARGET_ASM_CONSTRUCTOR} but used for termination
8336 functions rather than initialization functions.
8339 If @code{TARGET_HAVE_CTORS_DTORS} is true, the initialization routine
8340 generated for the generated object file will have static linkage.
8342 If your system uses @command{collect2} as the means of processing
8343 constructors, then that program normally uses @command{nm} to scan
8344 an object file for constructor functions to be called.
8346 On certain kinds of systems, you can define this macro to make
8347 @command{collect2} work faster (and, in some cases, make it work at all):
8349 @defmac OBJECT_FORMAT_COFF
8350 Define this macro if the system uses COFF (Common Object File Format)
8351 object files, so that @command{collect2} can assume this format and scan
8352 object files directly for dynamic constructor/destructor functions.
8354 This macro is effective only in a native compiler; @command{collect2} as
8355 part of a cross compiler always uses @command{nm} for the target machine.
8358 @defmac REAL_NM_FILE_NAME
8359 Define this macro as a C string constant containing the file name to use
8360 to execute @command{nm}. The default is to search the path normally for
8365 @command{collect2} calls @command{nm} to scan object files for static
8366 constructors and destructors and LTO info. By default, @option{-n} is
8367 passed. Define @code{NM_FLAGS} to a C string constant if other options
8368 are needed to get the same output format as GNU @command{nm -n}
8372 If your system supports shared libraries and has a program to list the
8373 dynamic dependencies of a given library or executable, you can define
8374 these macros to enable support for running initialization and
8375 termination functions in shared libraries:
8378 Define this macro to a C string constant containing the name of the program
8379 which lists dynamic dependencies, like @command{ldd} under SunOS 4.
8382 @defmac PARSE_LDD_OUTPUT (@var{ptr})
8383 Define this macro to be C code that extracts filenames from the output
8384 of the program denoted by @code{LDD_SUFFIX}. @var{ptr} is a variable
8385 of type @code{char *} that points to the beginning of a line of output
8386 from @code{LDD_SUFFIX}. If the line lists a dynamic dependency, the
8387 code must advance @var{ptr} to the beginning of the filename on that
8388 line. Otherwise, it must set @var{ptr} to @code{NULL}.
8391 @defmac SHLIB_SUFFIX
8392 Define this macro to a C string constant containing the default shared
8393 library extension of the target (e.g., @samp{".so"}). @command{collect2}
8394 strips version information after this suffix when generating global
8395 constructor and destructor names. This define is only needed on targets
8396 that use @command{collect2} to process constructors and destructors.
8399 @node Instruction Output
8400 @subsection Output of Assembler Instructions
8402 @c prevent bad page break with this line
8403 This describes assembler instruction output.
8405 @defmac REGISTER_NAMES
8406 A C initializer containing the assembler's names for the machine
8407 registers, each one as a C string constant. This is what translates
8408 register numbers in the compiler into assembler language.
8411 @defmac ADDITIONAL_REGISTER_NAMES
8412 If defined, a C initializer for an array of structures containing a name
8413 and a register number. This macro defines additional names for hard
8414 registers, thus allowing the @code{asm} option in declarations to refer
8415 to registers using alternate names.
8418 @defmac OVERLAPPING_REGISTER_NAMES
8419 If defined, a C initializer for an array of structures containing a
8420 name, a register number and a count of the number of consecutive
8421 machine registers the name overlaps. This macro defines additional
8422 names for hard registers, thus allowing the @code{asm} option in
8423 declarations to refer to registers using alternate names. Unlike
8424 @code{ADDITIONAL_REGISTER_NAMES}, this macro should be used when the
8425 register name implies multiple underlying registers.
8427 This macro should be used when it is important that a clobber in an
8428 @code{asm} statement clobbers all the underlying values implied by the
8429 register name. For example, on ARM, clobbering the double-precision
8430 VFP register ``d0'' implies clobbering both single-precision registers
8434 @defmac ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr})
8435 Define this macro if you are using an unusual assembler that
8436 requires different names for the machine instructions.
8438 The definition is a C statement or statements which output an
8439 assembler instruction opcode to the stdio stream @var{stream}. The
8440 macro-operand @var{ptr} is a variable of type @code{char *} which
8441 points to the opcode name in its ``internal'' form---the form that is
8442 written in the machine description. The definition should output the
8443 opcode name to @var{stream}, performing any translation you desire, and
8444 increment the variable @var{ptr} to point at the end of the opcode
8445 so that it will not be output twice.
8447 In fact, your macro definition may process less than the entire opcode
8448 name, or more than the opcode name; but if you want to process text
8449 that includes @samp{%}-sequences to substitute operands, you must take
8450 care of the substitution yourself. Just be sure to increment
8451 @var{ptr} over whatever text should not be output normally.
8453 @findex recog_data.operand
8454 If you need to look at the operand values, they can be found as the
8455 elements of @code{recog_data.operand}.
8457 If the macro definition does nothing, the instruction is output
8461 @defmac FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands})
8462 If defined, a C statement to be executed just prior to the output of
8463 assembler code for @var{insn}, to modify the extracted operands so
8464 they will be output differently.
8466 Here the argument @var{opvec} is the vector containing the operands
8467 extracted from @var{insn}, and @var{noperands} is the number of
8468 elements of the vector which contain meaningful data for this insn.
8469 The contents of this vector are what will be used to convert the insn
8470 template into assembler code, so you can change the assembler output
8471 by changing the contents of the vector.
8473 This macro is useful when various assembler syntaxes share a single
8474 file of instruction patterns; by defining this macro differently, you
8475 can cause a large class of instructions to be output differently (such
8476 as with rearranged operands). Naturally, variations in assembler
8477 syntax affecting individual insn patterns ought to be handled by
8478 writing conditional output routines in those patterns.
8480 If this macro is not defined, it is equivalent to a null statement.
8483 @hook TARGET_ASM_FINAL_POSTSCAN_INSN
8484 If defined, this target hook is a function which is executed just after the
8485 output of assembler code for @var{insn}, to change the mode of the assembler
8488 Here the argument @var{opvec} is the vector containing the operands
8489 extracted from @var{insn}, and @var{noperands} is the number of
8490 elements of the vector which contain meaningful data for this insn.
8491 The contents of this vector are what was used to convert the insn
8492 template into assembler code, so you can change the assembler mode
8493 by checking the contents of the vector.
8496 @defmac PRINT_OPERAND (@var{stream}, @var{x}, @var{code})
8497 A C compound statement to output to stdio stream @var{stream} the
8498 assembler syntax for an instruction operand @var{x}. @var{x} is an
8501 @var{code} is a value that can be used to specify one of several ways
8502 of printing the operand. It is used when identical operands must be
8503 printed differently depending on the context. @var{code} comes from
8504 the @samp{%} specification that was used to request printing of the
8505 operand. If the specification was just @samp{%@var{digit}} then
8506 @var{code} is 0; if the specification was @samp{%@var{ltr}
8507 @var{digit}} then @var{code} is the ASCII code for @var{ltr}.
8510 If @var{x} is a register, this macro should print the register's name.
8511 The names can be found in an array @code{reg_names} whose type is
8512 @code{char *[]}. @code{reg_names} is initialized from
8513 @code{REGISTER_NAMES}.
8515 When the machine description has a specification @samp{%@var{punct}}
8516 (a @samp{%} followed by a punctuation character), this macro is called
8517 with a null pointer for @var{x} and the punctuation character for
8521 @defmac PRINT_OPERAND_PUNCT_VALID_P (@var{code})
8522 A C expression which evaluates to true if @var{code} is a valid
8523 punctuation character for use in the @code{PRINT_OPERAND} macro. If
8524 @code{PRINT_OPERAND_PUNCT_VALID_P} is not defined, it means that no
8525 punctuation characters (except for the standard one, @samp{%}) are used
8529 @defmac PRINT_OPERAND_ADDRESS (@var{stream}, @var{x})
8530 A C compound statement to output to stdio stream @var{stream} the
8531 assembler syntax for an instruction operand that is a memory reference
8532 whose address is @var{x}. @var{x} is an RTL expression.
8534 @cindex @code{TARGET_ENCODE_SECTION_INFO} usage
8535 On some machines, the syntax for a symbolic address depends on the
8536 section that the address refers to. On these machines, define the hook
8537 @code{TARGET_ENCODE_SECTION_INFO} to store the information into the
8538 @code{symbol_ref}, and then check for it here. @xref{Assembler
8542 @findex dbr_sequence_length
8543 @defmac DBR_OUTPUT_SEQEND (@var{file})
8544 A C statement, to be executed after all slot-filler instructions have
8545 been output. If necessary, call @code{dbr_sequence_length} to
8546 determine the number of slots filled in a sequence (zero if not
8547 currently outputting a sequence), to decide how many no-ops to output,
8550 Don't define this macro if it has nothing to do, but it is helpful in
8551 reading assembly output if the extent of the delay sequence is made
8552 explicit (e.g.@: with white space).
8555 @findex final_sequence
8556 Note that output routines for instructions with delay slots must be
8557 prepared to deal with not being output as part of a sequence
8558 (i.e.@: when the scheduling pass is not run, or when no slot fillers could be
8559 found.) The variable @code{final_sequence} is null when not
8560 processing a sequence, otherwise it contains the @code{sequence} rtx
8564 @defmac REGISTER_PREFIX
8565 @defmacx LOCAL_LABEL_PREFIX
8566 @defmacx USER_LABEL_PREFIX
8567 @defmacx IMMEDIATE_PREFIX
8568 If defined, C string expressions to be used for the @samp{%R}, @samp{%L},
8569 @samp{%U}, and @samp{%I} options of @code{asm_fprintf} (see
8570 @file{final.c}). These are useful when a single @file{md} file must
8571 support multiple assembler formats. In that case, the various @file{tm.h}
8572 files can define these macros differently.
8575 @defmac ASM_FPRINTF_EXTENSIONS (@var{file}, @var{argptr}, @var{format})
8576 If defined this macro should expand to a series of @code{case}
8577 statements which will be parsed inside the @code{switch} statement of
8578 the @code{asm_fprintf} function. This allows targets to define extra
8579 printf formats which may useful when generating their assembler
8580 statements. Note that uppercase letters are reserved for future
8581 generic extensions to asm_fprintf, and so are not available to target
8582 specific code. The output file is given by the parameter @var{file}.
8583 The varargs input pointer is @var{argptr} and the rest of the format
8584 string, starting the character after the one that is being switched
8585 upon, is pointed to by @var{format}.
8588 @defmac ASSEMBLER_DIALECT
8589 If your target supports multiple dialects of assembler language (such as
8590 different opcodes), define this macro as a C expression that gives the
8591 numeric index of the assembler language dialect to use, with zero as the
8594 If this macro is defined, you may use constructs of the form
8596 @samp{@{option0|option1|option2@dots{}@}}
8599 in the output templates of patterns (@pxref{Output Template}) or in the
8600 first argument of @code{asm_fprintf}. This construct outputs
8601 @samp{option0}, @samp{option1}, @samp{option2}, etc., if the value of
8602 @code{ASSEMBLER_DIALECT} is zero, one, two, etc. Any special characters
8603 within these strings retain their usual meaning. If there are fewer
8604 alternatives within the braces than the value of
8605 @code{ASSEMBLER_DIALECT}, the construct outputs nothing.
8607 If you do not define this macro, the characters @samp{@{}, @samp{|} and
8608 @samp{@}} do not have any special meaning when used in templates or
8609 operands to @code{asm_fprintf}.
8611 Define the macros @code{REGISTER_PREFIX}, @code{LOCAL_LABEL_PREFIX},
8612 @code{USER_LABEL_PREFIX} and @code{IMMEDIATE_PREFIX} if you can express
8613 the variations in assembler language syntax with that mechanism. Define
8614 @code{ASSEMBLER_DIALECT} and use the @samp{@{option0|option1@}} syntax
8615 if the syntax variant are larger and involve such things as different
8616 opcodes or operand order.
8619 @defmac ASM_OUTPUT_REG_PUSH (@var{stream}, @var{regno})
8620 A C expression to output to @var{stream} some assembler code
8621 which will push hard register number @var{regno} onto the stack.
8622 The code need not be optimal, since this macro is used only when
8626 @defmac ASM_OUTPUT_REG_POP (@var{stream}, @var{regno})
8627 A C expression to output to @var{stream} some assembler code
8628 which will pop hard register number @var{regno} off of the stack.
8629 The code need not be optimal, since this macro is used only when
8633 @node Dispatch Tables
8634 @subsection Output of Dispatch Tables
8636 @c prevent bad page break with this line
8637 This concerns dispatch tables.
8639 @cindex dispatch table
8640 @defmac ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{body}, @var{value}, @var{rel})
8641 A C statement to output to the stdio stream @var{stream} an assembler
8642 pseudo-instruction to generate a difference between two labels.
8643 @var{value} and @var{rel} are the numbers of two internal labels. The
8644 definitions of these labels are output using
8645 @code{(*targetm.asm_out.internal_label)}, and they must be printed in the same
8646 way here. For example,
8649 fprintf (@var{stream}, "\t.word L%d-L%d\n",
8650 @var{value}, @var{rel})
8653 You must provide this macro on machines where the addresses in a
8654 dispatch table are relative to the table's own address. If defined, GCC
8655 will also use this macro on all machines when producing PIC@.
8656 @var{body} is the body of the @code{ADDR_DIFF_VEC}; it is provided so that the
8657 mode and flags can be read.
8660 @defmac ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value})
8661 This macro should be provided on machines where the addresses
8662 in a dispatch table are absolute.
8664 The definition should be a C statement to output to the stdio stream
8665 @var{stream} an assembler pseudo-instruction to generate a reference to
8666 a label. @var{value} is the number of an internal label whose
8667 definition is output using @code{(*targetm.asm_out.internal_label)}.
8671 fprintf (@var{stream}, "\t.word L%d\n", @var{value})
8675 @defmac ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table})
8676 Define this if the label before a jump-table needs to be output
8677 specially. The first three arguments are the same as for
8678 @code{(*targetm.asm_out.internal_label)}; the fourth argument is the
8679 jump-table which follows (a @code{jump_insn} containing an
8680 @code{addr_vec} or @code{addr_diff_vec}).
8682 This feature is used on system V to output a @code{swbeg} statement
8685 If this macro is not defined, these labels are output with
8686 @code{(*targetm.asm_out.internal_label)}.
8689 @defmac ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table})
8690 Define this if something special must be output at the end of a
8691 jump-table. The definition should be a C statement to be executed
8692 after the assembler code for the table is written. It should write
8693 the appropriate code to stdio stream @var{stream}. The argument
8694 @var{table} is the jump-table insn, and @var{num} is the label-number
8695 of the preceding label.
8697 If this macro is not defined, nothing special is output at the end of
8701 @hook TARGET_ASM_EMIT_UNWIND_LABEL
8702 This target hook emits a label at the beginning of each FDE@. It
8703 should be defined on targets where FDEs need special labels, and it
8704 should write the appropriate label, for the FDE associated with the
8705 function declaration @var{decl}, to the stdio stream @var{stream}.
8706 The third argument, @var{for_eh}, is a boolean: true if this is for an
8707 exception table. The fourth argument, @var{empty}, is a boolean:
8708 true if this is a placeholder label for an omitted FDE@.
8710 The default is that FDEs are not given nonlocal labels.
8713 @hook TARGET_ASM_EMIT_EXCEPT_TABLE_LABEL
8714 This target hook emits a label at the beginning of the exception table.
8715 It should be defined on targets where it is desirable for the table
8716 to be broken up according to function.
8718 The default is that no label is emitted.
8721 @hook TARGET_ASM_EMIT_EXCEPT_PERSONALITY
8723 @hook TARGET_ASM_UNWIND_EMIT
8724 This target hook emits assembly directives required to unwind the
8725 given instruction. This is only used when @code{TARGET_EXCEPT_UNWIND_INFO}
8726 returns @code{UI_TARGET}.
8729 @hook TARGET_ASM_UNWIND_EMIT_BEFORE_INSN
8731 @node Exception Region Output
8732 @subsection Assembler Commands for Exception Regions
8734 @c prevent bad page break with this line
8736 This describes commands marking the start and the end of an exception
8739 @defmac EH_FRAME_SECTION_NAME
8740 If defined, a C string constant for the name of the section containing
8741 exception handling frame unwind information. If not defined, GCC will
8742 provide a default definition if the target supports named sections.
8743 @file{crtstuff.c} uses this macro to switch to the appropriate section.
8745 You should define this symbol if your target supports DWARF 2 frame
8746 unwind information and the default definition does not work.
8749 @defmac EH_FRAME_IN_DATA_SECTION
8750 If defined, DWARF 2 frame unwind information will be placed in the
8751 data section even though the target supports named sections. This
8752 might be necessary, for instance, if the system linker does garbage
8753 collection and sections cannot be marked as not to be collected.
8755 Do not define this macro unless @code{TARGET_ASM_NAMED_SECTION} is
8759 @defmac EH_TABLES_CAN_BE_READ_ONLY
8760 Define this macro to 1 if your target is such that no frame unwind
8761 information encoding used with non-PIC code will ever require a
8762 runtime relocation, but the linker may not support merging read-only
8763 and read-write sections into a single read-write section.
8766 @defmac MASK_RETURN_ADDR
8767 An rtx used to mask the return address found via @code{RETURN_ADDR_RTX}, so
8768 that it does not contain any extraneous set bits in it.
8771 @defmac DWARF2_UNWIND_INFO
8772 Define this macro to 0 if your target supports DWARF 2 frame unwind
8773 information, but it does not yet work with exception handling.
8774 Otherwise, if your target supports this information (if it defines
8775 @code{INCOMING_RETURN_ADDR_RTX} and either @code{UNALIGNED_INT_ASM_OP}
8776 or @code{OBJECT_FORMAT_ELF}), GCC will provide a default definition of 1.
8779 @hook TARGET_EXCEPT_UNWIND_INFO
8780 This hook defines the mechanism that will be used for exception handling
8781 by the target. If the target has ABI specified unwind tables, the hook
8782 should return @code{UI_TARGET}. If the target is to use the
8783 @code{setjmp}/@code{longjmp}-based exception handling scheme, the hook
8784 should return @code{UI_SJLJ}. If the target supports DWARF 2 frame unwind
8785 information, the hook should return @code{UI_DWARF2}.
8787 A target may, if exceptions are disabled, choose to return @code{UI_NONE}.
8788 This may end up simplifying other parts of target-specific code. The
8789 default implementation of this hook never returns @code{UI_NONE}.
8791 Note that the value returned by this hook should be constant. It should
8792 not depend on anything except the command-line switches described by
8793 @var{opts}. In particular, the
8794 setting @code{UI_SJLJ} must be fixed at compiler start-up as C pre-processor
8795 macros and builtin functions related to exception handling are set up
8796 depending on this setting.
8798 The default implementation of the hook first honors the
8799 @option{--enable-sjlj-exceptions} configure option, then
8800 @code{DWARF2_UNWIND_INFO}, and finally defaults to @code{UI_SJLJ}. If
8801 @code{DWARF2_UNWIND_INFO} depends on command-line options, the target
8802 must define this hook so that @var{opts} is used correctly.
8805 @hook TARGET_UNWIND_TABLES_DEFAULT
8806 This variable should be set to @code{true} if the target ABI requires unwinding
8807 tables even when exceptions are not used. It must not be modified by
8808 command-line option processing.
8811 @defmac DONT_USE_BUILTIN_SETJMP
8812 Define this macro to 1 if the @code{setjmp}/@code{longjmp}-based scheme
8813 should use the @code{setjmp}/@code{longjmp} functions from the C library
8814 instead of the @code{__builtin_setjmp}/@code{__builtin_longjmp} machinery.
8817 @defmac DWARF_CIE_DATA_ALIGNMENT
8818 This macro need only be defined if the target might save registers in the
8819 function prologue at an offset to the stack pointer that is not aligned to
8820 @code{UNITS_PER_WORD}. The definition should be the negative minimum
8821 alignment if @code{STACK_GROWS_DOWNWARD} is defined, and the positive
8822 minimum alignment otherwise. @xref{SDB and DWARF}. Only applicable if
8823 the target supports DWARF 2 frame unwind information.
8826 @hook TARGET_TERMINATE_DW2_EH_FRAME_INFO
8827 Contains the value true if the target should add a zero word onto the
8828 end of a Dwarf-2 frame info section when used for exception handling.
8829 Default value is false if @code{EH_FRAME_SECTION_NAME} is defined, and
8833 @hook TARGET_DWARF_REGISTER_SPAN
8834 Given a register, this hook should return a parallel of registers to
8835 represent where to find the register pieces. Define this hook if the
8836 register and its mode are represented in Dwarf in non-contiguous
8837 locations, or if the register should be represented in more than one
8838 register in Dwarf. Otherwise, this hook should return @code{NULL_RTX}.
8839 If not defined, the default is to return @code{NULL_RTX}.
8842 @hook TARGET_INIT_DWARF_REG_SIZES_EXTRA
8843 If some registers are represented in Dwarf-2 unwind information in
8844 multiple pieces, define this hook to fill in information about the
8845 sizes of those pieces in the table used by the unwinder at runtime.
8846 It will be called by @code{expand_builtin_init_dwarf_reg_sizes} after
8847 filling in a single size corresponding to each hard register;
8848 @var{address} is the address of the table.
8851 @hook TARGET_ASM_TTYPE
8852 This hook is used to output a reference from a frame unwinding table to
8853 the type_info object identified by @var{sym}. It should return @code{true}
8854 if the reference was output. Returning @code{false} will cause the
8855 reference to be output using the normal Dwarf2 routines.
8858 @hook TARGET_ARM_EABI_UNWINDER
8859 This flag should be set to @code{true} on targets that use an ARM EABI
8860 based unwinding library, and @code{false} on other targets. This effects
8861 the format of unwinding tables, and how the unwinder in entered after
8862 running a cleanup. The default is @code{false}.
8865 @node Alignment Output
8866 @subsection Assembler Commands for Alignment
8868 @c prevent bad page break with this line
8869 This describes commands for alignment.
8871 @defmac JUMP_ALIGN (@var{label})
8872 The alignment (log base 2) to put in front of @var{label}, which is
8873 a common destination of jumps and has no fallthru incoming edge.
8875 This macro need not be defined if you don't want any special alignment
8876 to be done at such a time. Most machine descriptions do not currently
8879 Unless it's necessary to inspect the @var{label} parameter, it is better
8880 to set the variable @var{align_jumps} in the target's
8881 @code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's
8882 selection in @var{align_jumps} in a @code{JUMP_ALIGN} implementation.
8885 @hook TARGET_ASM_JUMP_ALIGN_MAX_SKIP
8886 The maximum number of bytes to skip before @var{label} when applying
8887 @code{JUMP_ALIGN}. This works only if
8888 @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
8891 @defmac LABEL_ALIGN_AFTER_BARRIER (@var{label})
8892 The alignment (log base 2) to put in front of @var{label}, which follows
8895 This macro need not be defined if you don't want any special alignment
8896 to be done at such a time. Most machine descriptions do not currently
8900 @hook TARGET_ASM_LABEL_ALIGN_AFTER_BARRIER_MAX_SKIP
8901 The maximum number of bytes to skip before @var{label} when applying
8902 @code{LABEL_ALIGN_AFTER_BARRIER}. This works only if
8903 @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
8906 @defmac LOOP_ALIGN (@var{label})
8907 The alignment (log base 2) to put in front of @var{label}, which follows
8908 a @code{NOTE_INSN_LOOP_BEG} note.
8910 This macro need not be defined if you don't want any special alignment
8911 to be done at such a time. Most machine descriptions do not currently
8914 Unless it's necessary to inspect the @var{label} parameter, it is better
8915 to set the variable @code{align_loops} in the target's
8916 @code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's
8917 selection in @code{align_loops} in a @code{LOOP_ALIGN} implementation.
8920 @hook TARGET_ASM_LOOP_ALIGN_MAX_SKIP
8921 The maximum number of bytes to skip when applying @code{LOOP_ALIGN} to
8922 @var{label}. This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is
8926 @defmac LABEL_ALIGN (@var{label})
8927 The alignment (log base 2) to put in front of @var{label}.
8928 If @code{LABEL_ALIGN_AFTER_BARRIER} / @code{LOOP_ALIGN} specify a different alignment,
8929 the maximum of the specified values is used.
8931 Unless it's necessary to inspect the @var{label} parameter, it is better
8932 to set the variable @code{align_labels} in the target's
8933 @code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's
8934 selection in @code{align_labels} in a @code{LABEL_ALIGN} implementation.
8937 @hook TARGET_ASM_LABEL_ALIGN_MAX_SKIP
8938 The maximum number of bytes to skip when applying @code{LABEL_ALIGN}
8939 to @var{label}. This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN}
8943 @defmac ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes})
8944 A C statement to output to the stdio stream @var{stream} an assembler
8945 instruction to advance the location counter by @var{nbytes} bytes.
8946 Those bytes should be zero when loaded. @var{nbytes} will be a C
8947 expression of type @code{unsigned HOST_WIDE_INT}.
8950 @defmac ASM_NO_SKIP_IN_TEXT
8951 Define this macro if @code{ASM_OUTPUT_SKIP} should not be used in the
8952 text section because it fails to put zeros in the bytes that are skipped.
8953 This is true on many Unix systems, where the pseudo--op to skip bytes
8954 produces no-op instructions rather than zeros when used in the text
8958 @defmac ASM_OUTPUT_ALIGN (@var{stream}, @var{power})
8959 A C statement to output to the stdio stream @var{stream} an assembler
8960 command to advance the location counter to a multiple of 2 to the
8961 @var{power} bytes. @var{power} will be a C expression of type @code{int}.
8964 @defmac ASM_OUTPUT_ALIGN_WITH_NOP (@var{stream}, @var{power})
8965 Like @code{ASM_OUTPUT_ALIGN}, except that the ``nop'' instruction is used
8966 for padding, if necessary.
8969 @defmac ASM_OUTPUT_MAX_SKIP_ALIGN (@var{stream}, @var{power}, @var{max_skip})
8970 A C statement to output to the stdio stream @var{stream} an assembler
8971 command to advance the location counter to a multiple of 2 to the
8972 @var{power} bytes, but only if @var{max_skip} or fewer bytes are needed to
8973 satisfy the alignment request. @var{power} and @var{max_skip} will be
8974 a C expression of type @code{int}.
8978 @node Debugging Info
8979 @section Controlling Debugging Information Format
8981 @c prevent bad page break with this line
8982 This describes how to specify debugging information.
8985 * All Debuggers:: Macros that affect all debugging formats uniformly.
8986 * DBX Options:: Macros enabling specific options in DBX format.
8987 * DBX Hooks:: Hook macros for varying DBX format.
8988 * File Names and DBX:: Macros controlling output of file names in DBX format.
8989 * SDB and DWARF:: Macros for SDB (COFF) and DWARF formats.
8990 * VMS Debug:: Macros for VMS debug format.
8994 @subsection Macros Affecting All Debugging Formats
8996 @c prevent bad page break with this line
8997 These macros affect all debugging formats.
8999 @defmac DBX_REGISTER_NUMBER (@var{regno})
9000 A C expression that returns the DBX register number for the compiler
9001 register number @var{regno}. In the default macro provided, the value
9002 of this expression will be @var{regno} itself. But sometimes there are
9003 some registers that the compiler knows about and DBX does not, or vice
9004 versa. In such cases, some register may need to have one number in the
9005 compiler and another for DBX@.
9007 If two registers have consecutive numbers inside GCC, and they can be
9008 used as a pair to hold a multiword value, then they @emph{must} have
9009 consecutive numbers after renumbering with @code{DBX_REGISTER_NUMBER}.
9010 Otherwise, debuggers will be unable to access such a pair, because they
9011 expect register pairs to be consecutive in their own numbering scheme.
9013 If you find yourself defining @code{DBX_REGISTER_NUMBER} in way that
9014 does not preserve register pairs, then what you must do instead is
9015 redefine the actual register numbering scheme.
9018 @defmac DEBUGGER_AUTO_OFFSET (@var{x})
9019 A C expression that returns the integer offset value for an automatic
9020 variable having address @var{x} (an RTL expression). The default
9021 computation assumes that @var{x} is based on the frame-pointer and
9022 gives the offset from the frame-pointer. This is required for targets
9023 that produce debugging output for DBX or COFF-style debugging output
9024 for SDB and allow the frame-pointer to be eliminated when the
9025 @option{-g} options is used.
9028 @defmac DEBUGGER_ARG_OFFSET (@var{offset}, @var{x})
9029 A C expression that returns the integer offset value for an argument
9030 having address @var{x} (an RTL expression). The nominal offset is
9034 @defmac PREFERRED_DEBUGGING_TYPE
9035 A C expression that returns the type of debugging output GCC should
9036 produce when the user specifies just @option{-g}. Define
9037 this if you have arranged for GCC to support more than one format of
9038 debugging output. Currently, the allowable values are @code{DBX_DEBUG},
9039 @code{SDB_DEBUG}, @code{DWARF_DEBUG}, @code{DWARF2_DEBUG},
9040 @code{XCOFF_DEBUG}, @code{VMS_DEBUG}, and @code{VMS_AND_DWARF2_DEBUG}.
9042 When the user specifies @option{-ggdb}, GCC normally also uses the
9043 value of this macro to select the debugging output format, but with two
9044 exceptions. If @code{DWARF2_DEBUGGING_INFO} is defined, GCC uses the
9045 value @code{DWARF2_DEBUG}. Otherwise, if @code{DBX_DEBUGGING_INFO} is
9046 defined, GCC uses @code{DBX_DEBUG}.
9048 The value of this macro only affects the default debugging output; the
9049 user can always get a specific type of output by using @option{-gstabs},
9050 @option{-gcoff}, @option{-gdwarf-2}, @option{-gxcoff}, or @option{-gvms}.
9054 @subsection Specific Options for DBX Output
9056 @c prevent bad page break with this line
9057 These are specific options for DBX output.
9059 @defmac DBX_DEBUGGING_INFO
9060 Define this macro if GCC should produce debugging output for DBX
9061 in response to the @option{-g} option.
9064 @defmac XCOFF_DEBUGGING_INFO
9065 Define this macro if GCC should produce XCOFF format debugging output
9066 in response to the @option{-g} option. This is a variant of DBX format.
9069 @defmac DEFAULT_GDB_EXTENSIONS
9070 Define this macro to control whether GCC should by default generate
9071 GDB's extended version of DBX debugging information (assuming DBX-format
9072 debugging information is enabled at all). If you don't define the
9073 macro, the default is 1: always generate the extended information
9074 if there is any occasion to.
9077 @defmac DEBUG_SYMS_TEXT
9078 Define this macro if all @code{.stabs} commands should be output while
9079 in the text section.
9082 @defmac ASM_STABS_OP
9083 A C string constant, including spacing, naming the assembler pseudo op to
9084 use instead of @code{"\t.stabs\t"} to define an ordinary debugging symbol.
9085 If you don't define this macro, @code{"\t.stabs\t"} is used. This macro
9086 applies only to DBX debugging information format.
9089 @defmac ASM_STABD_OP
9090 A C string constant, including spacing, naming the assembler pseudo op to
9091 use instead of @code{"\t.stabd\t"} to define a debugging symbol whose
9092 value is the current location. If you don't define this macro,
9093 @code{"\t.stabd\t"} is used. This macro applies only to DBX debugging
9097 @defmac ASM_STABN_OP
9098 A C string constant, including spacing, naming the assembler pseudo op to
9099 use instead of @code{"\t.stabn\t"} to define a debugging symbol with no
9100 name. If you don't define this macro, @code{"\t.stabn\t"} is used. This
9101 macro applies only to DBX debugging information format.
9104 @defmac DBX_NO_XREFS
9105 Define this macro if DBX on your system does not support the construct
9106 @samp{xs@var{tagname}}. On some systems, this construct is used to
9107 describe a forward reference to a structure named @var{tagname}.
9108 On other systems, this construct is not supported at all.
9111 @defmac DBX_CONTIN_LENGTH
9112 A symbol name in DBX-format debugging information is normally
9113 continued (split into two separate @code{.stabs} directives) when it
9114 exceeds a certain length (by default, 80 characters). On some
9115 operating systems, DBX requires this splitting; on others, splitting
9116 must not be done. You can inhibit splitting by defining this macro
9117 with the value zero. You can override the default splitting-length by
9118 defining this macro as an expression for the length you desire.
9121 @defmac DBX_CONTIN_CHAR
9122 Normally continuation is indicated by adding a @samp{\} character to
9123 the end of a @code{.stabs} string when a continuation follows. To use
9124 a different character instead, define this macro as a character
9125 constant for the character you want to use. Do not define this macro
9126 if backslash is correct for your system.
9129 @defmac DBX_STATIC_STAB_DATA_SECTION
9130 Define this macro if it is necessary to go to the data section before
9131 outputting the @samp{.stabs} pseudo-op for a non-global static
9135 @defmac DBX_TYPE_DECL_STABS_CODE
9136 The value to use in the ``code'' field of the @code{.stabs} directive
9137 for a typedef. The default is @code{N_LSYM}.
9140 @defmac DBX_STATIC_CONST_VAR_CODE
9141 The value to use in the ``code'' field of the @code{.stabs} directive
9142 for a static variable located in the text section. DBX format does not
9143 provide any ``right'' way to do this. The default is @code{N_FUN}.
9146 @defmac DBX_REGPARM_STABS_CODE
9147 The value to use in the ``code'' field of the @code{.stabs} directive
9148 for a parameter passed in registers. DBX format does not provide any
9149 ``right'' way to do this. The default is @code{N_RSYM}.
9152 @defmac DBX_REGPARM_STABS_LETTER
9153 The letter to use in DBX symbol data to identify a symbol as a parameter
9154 passed in registers. DBX format does not customarily provide any way to
9155 do this. The default is @code{'P'}.
9158 @defmac DBX_FUNCTION_FIRST
9159 Define this macro if the DBX information for a function and its
9160 arguments should precede the assembler code for the function. Normally,
9161 in DBX format, the debugging information entirely follows the assembler
9165 @defmac DBX_BLOCKS_FUNCTION_RELATIVE
9166 Define this macro, with value 1, if the value of a symbol describing
9167 the scope of a block (@code{N_LBRAC} or @code{N_RBRAC}) should be
9168 relative to the start of the enclosing function. Normally, GCC uses
9169 an absolute address.
9172 @defmac DBX_LINES_FUNCTION_RELATIVE
9173 Define this macro, with value 1, if the value of a symbol indicating
9174 the current line number (@code{N_SLINE}) should be relative to the
9175 start of the enclosing function. Normally, GCC uses an absolute address.
9178 @defmac DBX_USE_BINCL
9179 Define this macro if GCC should generate @code{N_BINCL} and
9180 @code{N_EINCL} stabs for included header files, as on Sun systems. This
9181 macro also directs GCC to output a type number as a pair of a file
9182 number and a type number within the file. Normally, GCC does not
9183 generate @code{N_BINCL} or @code{N_EINCL} stabs, and it outputs a single
9184 number for a type number.
9188 @subsection Open-Ended Hooks for DBX Format
9190 @c prevent bad page break with this line
9191 These are hooks for DBX format.
9193 @defmac DBX_OUTPUT_LBRAC (@var{stream}, @var{name})
9194 Define this macro to say how to output to @var{stream} the debugging
9195 information for the start of a scope level for variable names. The
9196 argument @var{name} is the name of an assembler symbol (for use with
9197 @code{assemble_name}) whose value is the address where the scope begins.
9200 @defmac DBX_OUTPUT_RBRAC (@var{stream}, @var{name})
9201 Like @code{DBX_OUTPUT_LBRAC}, but for the end of a scope level.
9204 @defmac DBX_OUTPUT_NFUN (@var{stream}, @var{lscope_label}, @var{decl})
9205 Define this macro if the target machine requires special handling to
9206 output an @code{N_FUN} entry for the function @var{decl}.
9209 @defmac DBX_OUTPUT_SOURCE_LINE (@var{stream}, @var{line}, @var{counter})
9210 A C statement to output DBX debugging information before code for line
9211 number @var{line} of the current source file to the stdio stream
9212 @var{stream}. @var{counter} is the number of time the macro was
9213 invoked, including the current invocation; it is intended to generate
9214 unique labels in the assembly output.
9216 This macro should not be defined if the default output is correct, or
9217 if it can be made correct by defining @code{DBX_LINES_FUNCTION_RELATIVE}.
9220 @defmac NO_DBX_FUNCTION_END
9221 Some stabs encapsulation formats (in particular ECOFF), cannot handle the
9222 @code{.stabs "",N_FUN,,0,0,Lscope-function-1} gdb dbx extension construct.
9223 On those machines, define this macro to turn this feature off without
9224 disturbing the rest of the gdb extensions.
9227 @defmac NO_DBX_BNSYM_ENSYM
9228 Some assemblers cannot handle the @code{.stabd BNSYM/ENSYM,0,0} gdb dbx
9229 extension construct. On those machines, define this macro to turn this
9230 feature off without disturbing the rest of the gdb extensions.
9233 @node File Names and DBX
9234 @subsection File Names in DBX Format
9236 @c prevent bad page break with this line
9237 This describes file names in DBX format.
9239 @defmac DBX_OUTPUT_MAIN_SOURCE_FILENAME (@var{stream}, @var{name})
9240 A C statement to output DBX debugging information to the stdio stream
9241 @var{stream}, which indicates that file @var{name} is the main source
9242 file---the file specified as the input file for compilation.
9243 This macro is called only once, at the beginning of compilation.
9245 This macro need not be defined if the standard form of output
9246 for DBX debugging information is appropriate.
9248 It may be necessary to refer to a label equal to the beginning of the
9249 text section. You can use @samp{assemble_name (stream, ltext_label_name)}
9250 to do so. If you do this, you must also set the variable
9251 @var{used_ltext_label_name} to @code{true}.
9254 @defmac NO_DBX_MAIN_SOURCE_DIRECTORY
9255 Define this macro, with value 1, if GCC should not emit an indication
9256 of the current directory for compilation and current source language at
9257 the beginning of the file.
9260 @defmac NO_DBX_GCC_MARKER
9261 Define this macro, with value 1, if GCC should not emit an indication
9262 that this object file was compiled by GCC@. The default is to emit
9263 an @code{N_OPT} stab at the beginning of every source file, with
9264 @samp{gcc2_compiled.} for the string and value 0.
9267 @defmac DBX_OUTPUT_MAIN_SOURCE_FILE_END (@var{stream}, @var{name})
9268 A C statement to output DBX debugging information at the end of
9269 compilation of the main source file @var{name}. Output should be
9270 written to the stdio stream @var{stream}.
9272 If you don't define this macro, nothing special is output at the end
9273 of compilation, which is correct for most machines.
9276 @defmac DBX_OUTPUT_NULL_N_SO_AT_MAIN_SOURCE_FILE_END
9277 Define this macro @emph{instead of} defining
9278 @code{DBX_OUTPUT_MAIN_SOURCE_FILE_END}, if what needs to be output at
9279 the end of compilation is an @code{N_SO} stab with an empty string,
9280 whose value is the highest absolute text address in the file.
9285 @subsection Macros for SDB and DWARF Output
9287 @c prevent bad page break with this line
9288 Here are macros for SDB and DWARF output.
9290 @defmac SDB_DEBUGGING_INFO
9291 Define this macro if GCC should produce COFF-style debugging output
9292 for SDB in response to the @option{-g} option.
9295 @defmac DWARF2_DEBUGGING_INFO
9296 Define this macro if GCC should produce dwarf version 2 format
9297 debugging output in response to the @option{-g} option.
9299 @hook TARGET_DWARF_CALLING_CONVENTION
9300 Define this to enable the dwarf attribute @code{DW_AT_calling_convention} to
9301 be emitted for each function. Instead of an integer return the enum
9302 value for the @code{DW_CC_} tag.
9305 To support optional call frame debugging information, you must also
9306 define @code{INCOMING_RETURN_ADDR_RTX} and either set
9307 @code{RTX_FRAME_RELATED_P} on the prologue insns if you use RTL for the
9308 prologue, or call @code{dwarf2out_def_cfa} and @code{dwarf2out_reg_save}
9309 as appropriate from @code{TARGET_ASM_FUNCTION_PROLOGUE} if you don't.
9312 @defmac DWARF2_FRAME_INFO
9313 Define this macro to a nonzero value if GCC should always output
9314 Dwarf 2 frame information. If @code{TARGET_EXCEPT_UNWIND_INFO}
9315 (@pxref{Exception Region Output}) returns @code{UI_DWARF2}, and
9316 exceptions are enabled, GCC will output this information not matter
9317 how you define @code{DWARF2_FRAME_INFO}.
9320 @hook TARGET_DEBUG_UNWIND_INFO
9321 This hook defines the mechanism that will be used for describing frame
9322 unwind information to the debugger. Normally the hook will return
9323 @code{UI_DWARF2} if DWARF 2 debug information is enabled, and
9324 return @code{UI_NONE} otherwise.
9326 A target may return @code{UI_DWARF2} even when DWARF 2 debug information
9327 is disabled in order to always output DWARF 2 frame information.
9329 A target may return @code{UI_TARGET} if it has ABI specified unwind tables.
9330 This will suppress generation of the normal debug frame unwind information.
9333 @defmac DWARF2_ASM_LINE_DEBUG_INFO
9334 Define this macro to be a nonzero value if the assembler can generate Dwarf 2
9335 line debug info sections. This will result in much more compact line number
9336 tables, and hence is desirable if it works.
9339 @hook TARGET_WANT_DEBUG_PUB_SECTIONS
9341 @defmac ASM_OUTPUT_DWARF_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2})
9342 A C statement to issue assembly directives that create a difference
9343 @var{lab1} minus @var{lab2}, using an integer of the given @var{size}.
9346 @defmac ASM_OUTPUT_DWARF_VMS_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2})
9347 A C statement to issue assembly directives that create a difference
9348 between the two given labels in system defined units, e.g. instruction
9349 slots on IA64 VMS, using an integer of the given size.
9352 @defmac ASM_OUTPUT_DWARF_OFFSET (@var{stream}, @var{size}, @var{label}, @var{section})
9353 A C statement to issue assembly directives that create a
9354 section-relative reference to the given @var{label}, using an integer of the
9355 given @var{size}. The label is known to be defined in the given @var{section}.
9358 @defmac ASM_OUTPUT_DWARF_PCREL (@var{stream}, @var{size}, @var{label})
9359 A C statement to issue assembly directives that create a self-relative
9360 reference to the given @var{label}, using an integer of the given @var{size}.
9363 @defmac ASM_OUTPUT_DWARF_TABLE_REF (@var{label})
9364 A C statement to issue assembly directives that create a reference to
9365 the DWARF table identifier @var{label} from the current section. This
9366 is used on some systems to avoid garbage collecting a DWARF table which
9367 is referenced by a function.
9370 @hook TARGET_ASM_OUTPUT_DWARF_DTPREL
9371 If defined, this target hook is a function which outputs a DTP-relative
9372 reference to the given TLS symbol of the specified size.
9375 @defmac PUT_SDB_@dots{}
9376 Define these macros to override the assembler syntax for the special
9377 SDB assembler directives. See @file{sdbout.c} for a list of these
9378 macros and their arguments. If the standard syntax is used, you need
9379 not define them yourself.
9383 Some assemblers do not support a semicolon as a delimiter, even between
9384 SDB assembler directives. In that case, define this macro to be the
9385 delimiter to use (usually @samp{\n}). It is not necessary to define
9386 a new set of @code{PUT_SDB_@var{op}} macros if this is the only change
9390 @defmac SDB_ALLOW_UNKNOWN_REFERENCES
9391 Define this macro to allow references to unknown structure,
9392 union, or enumeration tags to be emitted. Standard COFF does not
9393 allow handling of unknown references, MIPS ECOFF has support for
9397 @defmac SDB_ALLOW_FORWARD_REFERENCES
9398 Define this macro to allow references to structure, union, or
9399 enumeration tags that have not yet been seen to be handled. Some
9400 assemblers choke if forward tags are used, while some require it.
9403 @defmac SDB_OUTPUT_SOURCE_LINE (@var{stream}, @var{line})
9404 A C statement to output SDB debugging information before code for line
9405 number @var{line} of the current source file to the stdio stream
9406 @var{stream}. The default is to emit an @code{.ln} directive.
9411 @subsection Macros for VMS Debug Format
9413 @c prevent bad page break with this line
9414 Here are macros for VMS debug format.
9416 @defmac VMS_DEBUGGING_INFO
9417 Define this macro if GCC should produce debugging output for VMS
9418 in response to the @option{-g} option. The default behavior for VMS
9419 is to generate minimal debug info for a traceback in the absence of
9420 @option{-g} unless explicitly overridden with @option{-g0}. This
9421 behavior is controlled by @code{TARGET_OPTION_OPTIMIZATION} and
9422 @code{TARGET_OPTION_OVERRIDE}.
9425 @node Floating Point
9426 @section Cross Compilation and Floating Point
9427 @cindex cross compilation and floating point
9428 @cindex floating point and cross compilation
9430 While all modern machines use twos-complement representation for integers,
9431 there are a variety of representations for floating point numbers. This
9432 means that in a cross-compiler the representation of floating point numbers
9433 in the compiled program may be different from that used in the machine
9434 doing the compilation.
9436 Because different representation systems may offer different amounts of
9437 range and precision, all floating point constants must be represented in
9438 the target machine's format. Therefore, the cross compiler cannot
9439 safely use the host machine's floating point arithmetic; it must emulate
9440 the target's arithmetic. To ensure consistency, GCC always uses
9441 emulation to work with floating point values, even when the host and
9442 target floating point formats are identical.
9444 The following macros are provided by @file{real.h} for the compiler to
9445 use. All parts of the compiler which generate or optimize
9446 floating-point calculations must use these macros. They may evaluate
9447 their operands more than once, so operands must not have side effects.
9449 @defmac REAL_VALUE_TYPE
9450 The C data type to be used to hold a floating point value in the target
9451 machine's format. Typically this is a @code{struct} containing an
9452 array of @code{HOST_WIDE_INT}, but all code should treat it as an opaque
9456 @deftypefn Macro int REAL_VALUES_EQUAL (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y})
9457 Compares for equality the two values, @var{x} and @var{y}. If the target
9458 floating point format supports negative zeroes and/or NaNs,
9459 @samp{REAL_VALUES_EQUAL (-0.0, 0.0)} is true, and
9460 @samp{REAL_VALUES_EQUAL (NaN, NaN)} is false.
9463 @deftypefn Macro int REAL_VALUES_LESS (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y})
9464 Tests whether @var{x} is less than @var{y}.
9467 @deftypefn Macro HOST_WIDE_INT REAL_VALUE_FIX (REAL_VALUE_TYPE @var{x})
9468 Truncates @var{x} to a signed integer, rounding toward zero.
9471 @deftypefn Macro {unsigned HOST_WIDE_INT} REAL_VALUE_UNSIGNED_FIX (REAL_VALUE_TYPE @var{x})
9472 Truncates @var{x} to an unsigned integer, rounding toward zero. If
9473 @var{x} is negative, returns zero.
9476 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ATOF (const char *@var{string}, enum machine_mode @var{mode})
9477 Converts @var{string} into a floating point number in the target machine's
9478 representation for mode @var{mode}. This routine can handle both
9479 decimal and hexadecimal floating point constants, using the syntax
9480 defined by the C language for both.
9483 @deftypefn Macro int REAL_VALUE_NEGATIVE (REAL_VALUE_TYPE @var{x})
9484 Returns 1 if @var{x} is negative (including negative zero), 0 otherwise.
9487 @deftypefn Macro int REAL_VALUE_ISINF (REAL_VALUE_TYPE @var{x})
9488 Determines whether @var{x} represents infinity (positive or negative).
9491 @deftypefn Macro int REAL_VALUE_ISNAN (REAL_VALUE_TYPE @var{x})
9492 Determines whether @var{x} represents a ``NaN'' (not-a-number).
9495 @deftypefn Macro void REAL_ARITHMETIC (REAL_VALUE_TYPE @var{output}, enum tree_code @var{code}, REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y})
9496 Calculates an arithmetic operation on the two floating point values
9497 @var{x} and @var{y}, storing the result in @var{output} (which must be a
9500 The operation to be performed is specified by @var{code}. Only the
9501 following codes are supported: @code{PLUS_EXPR}, @code{MINUS_EXPR},
9502 @code{MULT_EXPR}, @code{RDIV_EXPR}, @code{MAX_EXPR}, @code{MIN_EXPR}.
9504 If @code{REAL_ARITHMETIC} is asked to evaluate division by zero and the
9505 target's floating point format cannot represent infinity, it will call
9506 @code{abort}. Callers should check for this situation first, using
9507 @code{MODE_HAS_INFINITIES}. @xref{Storage Layout}.
9510 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_NEGATE (REAL_VALUE_TYPE @var{x})
9511 Returns the negative of the floating point value @var{x}.
9514 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ABS (REAL_VALUE_TYPE @var{x})
9515 Returns the absolute value of @var{x}.
9518 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_TRUNCATE (REAL_VALUE_TYPE @var{mode}, enum machine_mode @var{x})
9519 Truncates the floating point value @var{x} to fit in @var{mode}. The
9520 return value is still a full-size @code{REAL_VALUE_TYPE}, but it has an
9521 appropriate bit pattern to be output as a floating constant whose
9522 precision accords with mode @var{mode}.
9525 @deftypefn Macro void REAL_VALUE_TO_INT (HOST_WIDE_INT @var{low}, HOST_WIDE_INT @var{high}, REAL_VALUE_TYPE @var{x})
9526 Converts a floating point value @var{x} into a double-precision integer
9527 which is then stored into @var{low} and @var{high}. If the value is not
9528 integral, it is truncated.
9531 @deftypefn Macro void REAL_VALUE_FROM_INT (REAL_VALUE_TYPE @var{x}, HOST_WIDE_INT @var{low}, HOST_WIDE_INT @var{high}, enum machine_mode @var{mode})
9532 Converts a double-precision integer found in @var{low} and @var{high},
9533 into a floating point value which is then stored into @var{x}. The
9534 value is truncated to fit in mode @var{mode}.
9537 @node Mode Switching
9538 @section Mode Switching Instructions
9539 @cindex mode switching
9540 The following macros control mode switching optimizations:
9542 @defmac OPTIMIZE_MODE_SWITCHING (@var{entity})
9543 Define this macro if the port needs extra instructions inserted for mode
9544 switching in an optimizing compilation.
9546 For an example, the SH4 can perform both single and double precision
9547 floating point operations, but to perform a single precision operation,
9548 the FPSCR PR bit has to be cleared, while for a double precision
9549 operation, this bit has to be set. Changing the PR bit requires a general
9550 purpose register as a scratch register, hence these FPSCR sets have to
9551 be inserted before reload, i.e.@: you can't put this into instruction emitting
9552 or @code{TARGET_MACHINE_DEPENDENT_REORG}.
9554 You can have multiple entities that are mode-switched, and select at run time
9555 which entities actually need it. @code{OPTIMIZE_MODE_SWITCHING} should
9556 return nonzero for any @var{entity} that needs mode-switching.
9557 If you define this macro, you also have to define
9558 @code{NUM_MODES_FOR_MODE_SWITCHING}, @code{MODE_NEEDED},
9559 @code{MODE_PRIORITY_TO_MODE} and @code{EMIT_MODE_SET}.
9560 @code{MODE_AFTER}, @code{MODE_ENTRY}, and @code{MODE_EXIT}
9564 @defmac NUM_MODES_FOR_MODE_SWITCHING
9565 If you define @code{OPTIMIZE_MODE_SWITCHING}, you have to define this as
9566 initializer for an array of integers. Each initializer element
9567 N refers to an entity that needs mode switching, and specifies the number
9568 of different modes that might need to be set for this entity.
9569 The position of the initializer in the initializer---starting counting at
9570 zero---determines the integer that is used to refer to the mode-switched
9572 In macros that take mode arguments / yield a mode result, modes are
9573 represented as numbers 0 @dots{} N @minus{} 1. N is used to specify that no mode
9574 switch is needed / supplied.
9577 @defmac MODE_NEEDED (@var{entity}, @var{insn})
9578 @var{entity} is an integer specifying a mode-switched entity. If
9579 @code{OPTIMIZE_MODE_SWITCHING} is defined, you must define this macro to
9580 return an integer value not larger than the corresponding element in
9581 @code{NUM_MODES_FOR_MODE_SWITCHING}, to denote the mode that @var{entity} must
9582 be switched into prior to the execution of @var{insn}.
9585 @defmac MODE_AFTER (@var{mode}, @var{insn})
9586 If this macro is defined, it is evaluated for every @var{insn} during
9587 mode switching. It determines the mode that an insn results in (if
9588 different from the incoming mode).
9591 @defmac MODE_ENTRY (@var{entity})
9592 If this macro is defined, it is evaluated for every @var{entity} that needs
9593 mode switching. It should evaluate to an integer, which is a mode that
9594 @var{entity} is assumed to be switched to at function entry. If @code{MODE_ENTRY}
9595 is defined then @code{MODE_EXIT} must be defined.
9598 @defmac MODE_EXIT (@var{entity})
9599 If this macro is defined, it is evaluated for every @var{entity} that needs
9600 mode switching. It should evaluate to an integer, which is a mode that
9601 @var{entity} is assumed to be switched to at function exit. If @code{MODE_EXIT}
9602 is defined then @code{MODE_ENTRY} must be defined.
9605 @defmac MODE_PRIORITY_TO_MODE (@var{entity}, @var{n})
9606 This macro specifies the order in which modes for @var{entity} are processed.
9607 0 is the highest priority, @code{NUM_MODES_FOR_MODE_SWITCHING[@var{entity}] - 1} the
9608 lowest. The value of the macro should be an integer designating a mode
9609 for @var{entity}. For any fixed @var{entity}, @code{mode_priority_to_mode}
9610 (@var{entity}, @var{n}) shall be a bijection in 0 @dots{}
9611 @code{num_modes_for_mode_switching[@var{entity}] - 1}.
9614 @defmac EMIT_MODE_SET (@var{entity}, @var{mode}, @var{hard_regs_live})
9615 Generate one or more insns to set @var{entity} to @var{mode}.
9616 @var{hard_reg_live} is the set of hard registers live at the point where
9617 the insn(s) are to be inserted.
9620 @node Target Attributes
9621 @section Defining target-specific uses of @code{__attribute__}
9622 @cindex target attributes
9623 @cindex machine attributes
9624 @cindex attributes, target-specific
9626 Target-specific attributes may be defined for functions, data and types.
9627 These are described using the following target hooks; they also need to
9628 be documented in @file{extend.texi}.
9630 @hook TARGET_ATTRIBUTE_TABLE
9631 If defined, this target hook points to an array of @samp{struct
9632 attribute_spec} (defined in @file{tree.h}) specifying the machine
9633 specific attributes for this target and some of the restrictions on the
9634 entities to which these attributes are applied and the arguments they
9638 @hook TARGET_ATTRIBUTE_TAKES_IDENTIFIER_P
9639 If defined, this target hook is a function which returns true if the
9640 machine-specific attribute named @var{name} expects an identifier
9641 given as its first argument to be passed on as a plain identifier, not
9642 subjected to name lookup. If this is not defined, the default is
9643 false for all machine-specific attributes.
9646 @hook TARGET_COMP_TYPE_ATTRIBUTES
9647 If defined, this target hook is a function which returns zero if the attributes on
9648 @var{type1} and @var{type2} are incompatible, one if they are compatible,
9649 and two if they are nearly compatible (which causes a warning to be
9650 generated). If this is not defined, machine-specific attributes are
9651 supposed always to be compatible.
9654 @hook TARGET_SET_DEFAULT_TYPE_ATTRIBUTES
9655 If defined, this target hook is a function which assigns default attributes to
9656 the newly defined @var{type}.
9659 @hook TARGET_MERGE_TYPE_ATTRIBUTES
9660 Define this target hook if the merging of type attributes needs special
9661 handling. If defined, the result is a list of the combined
9662 @code{TYPE_ATTRIBUTES} of @var{type1} and @var{type2}. It is assumed
9663 that @code{comptypes} has already been called and returned 1. This
9664 function may call @code{merge_attributes} to handle machine-independent
9668 @hook TARGET_MERGE_DECL_ATTRIBUTES
9669 Define this target hook if the merging of decl attributes needs special
9670 handling. If defined, the result is a list of the combined
9671 @code{DECL_ATTRIBUTES} of @var{olddecl} and @var{newdecl}.
9672 @var{newdecl} is a duplicate declaration of @var{olddecl}. Examples of
9673 when this is needed are when one attribute overrides another, or when an
9674 attribute is nullified by a subsequent definition. This function may
9675 call @code{merge_attributes} to handle machine-independent merging.
9677 @findex TARGET_DLLIMPORT_DECL_ATTRIBUTES
9678 If the only target-specific handling you require is @samp{dllimport}
9679 for Microsoft Windows targets, you should define the macro
9680 @code{TARGET_DLLIMPORT_DECL_ATTRIBUTES} to @code{1}. The compiler
9681 will then define a function called
9682 @code{merge_dllimport_decl_attributes} which can then be defined as
9683 the expansion of @code{TARGET_MERGE_DECL_ATTRIBUTES}. You can also
9684 add @code{handle_dll_attribute} in the attribute table for your port
9685 to perform initial processing of the @samp{dllimport} and
9686 @samp{dllexport} attributes. This is done in @file{i386/cygwin.h} and
9687 @file{i386/i386.c}, for example.
9690 @hook TARGET_VALID_DLLIMPORT_ATTRIBUTE_P
9692 @defmac TARGET_DECLSPEC
9693 Define this macro to a nonzero value if you want to treat
9694 @code{__declspec(X)} as equivalent to @code{__attribute((X))}. By
9695 default, this behavior is enabled only for targets that define
9696 @code{TARGET_DLLIMPORT_DECL_ATTRIBUTES}. The current implementation
9697 of @code{__declspec} is via a built-in macro, but you should not rely
9698 on this implementation detail.
9701 @hook TARGET_INSERT_ATTRIBUTES
9702 Define this target hook if you want to be able to add attributes to a decl
9703 when it is being created. This is normally useful for back ends which
9704 wish to implement a pragma by using the attributes which correspond to
9705 the pragma's effect. The @var{node} argument is the decl which is being
9706 created. The @var{attr_ptr} argument is a pointer to the attribute list
9707 for this decl. The list itself should not be modified, since it may be
9708 shared with other decls, but attributes may be chained on the head of
9709 the list and @code{*@var{attr_ptr}} modified to point to the new
9710 attributes, or a copy of the list may be made if further changes are
9714 @hook TARGET_FUNCTION_ATTRIBUTE_INLINABLE_P
9716 This target hook returns @code{true} if it is ok to inline @var{fndecl}
9717 into the current function, despite its having target-specific
9718 attributes, @code{false} otherwise. By default, if a function has a
9719 target specific attribute attached to it, it will not be inlined.
9722 @hook TARGET_OPTION_VALID_ATTRIBUTE_P
9723 This hook is called to parse the @code{attribute(option("..."))}, and
9724 it allows the function to set different target machine compile time
9725 options for the current function that might be different than the
9726 options specified on the command line. The hook should return
9727 @code{true} if the options are valid.
9729 The hook should set the @var{DECL_FUNCTION_SPECIFIC_TARGET} field in
9730 the function declaration to hold a pointer to a target specific
9731 @var{struct cl_target_option} structure.
9734 @hook TARGET_OPTION_SAVE
9735 This hook is called to save any additional target specific information
9736 in the @var{struct cl_target_option} structure for function specific
9738 @xref{Option file format}.
9741 @hook TARGET_OPTION_RESTORE
9742 This hook is called to restore any additional target specific
9743 information in the @var{struct cl_target_option} structure for
9744 function specific options.
9747 @hook TARGET_OPTION_PRINT
9748 This hook is called to print any additional target specific
9749 information in the @var{struct cl_target_option} structure for
9750 function specific options.
9753 @hook TARGET_OPTION_PRAGMA_PARSE
9754 This target hook parses the options for @code{#pragma GCC option} to
9755 set the machine specific options for functions that occur later in the
9756 input stream. The options should be the same as handled by the
9757 @code{TARGET_OPTION_VALID_ATTRIBUTE_P} hook.
9760 @hook TARGET_OPTION_OVERRIDE
9761 Sometimes certain combinations of command options do not make sense on
9762 a particular target machine. You can override the hook
9763 @code{TARGET_OPTION_OVERRIDE} to take account of this. This hooks is called
9764 once just after all the command options have been parsed.
9766 Don't use this hook to turn on various extra optimizations for
9767 @option{-O}. That is what @code{TARGET_OPTION_OPTIMIZATION} is for.
9769 If you need to do something whenever the optimization level is
9770 changed via the optimize attribute or pragma, see
9771 @code{TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE}
9774 @hook TARGET_CAN_INLINE_P
9775 This target hook returns @code{false} if the @var{caller} function
9776 cannot inline @var{callee}, based on target specific information. By
9777 default, inlining is not allowed if the callee function has function
9778 specific target options and the caller does not use the same options.
9782 @section Emulating TLS
9783 @cindex Emulated TLS
9785 For targets whose psABI does not provide Thread Local Storage via
9786 specific relocations and instruction sequences, an emulation layer is
9787 used. A set of target hooks allows this emulation layer to be
9788 configured for the requirements of a particular target. For instance
9789 the psABI may in fact specify TLS support in terms of an emulation
9792 The emulation layer works by creating a control object for every TLS
9793 object. To access the TLS object, a lookup function is provided
9794 which, when given the address of the control object, will return the
9795 address of the current thread's instance of the TLS object.
9797 @hook TARGET_EMUTLS_GET_ADDRESS
9798 Contains the name of the helper function that uses a TLS control
9799 object to locate a TLS instance. The default causes libgcc's
9800 emulated TLS helper function to be used.
9803 @hook TARGET_EMUTLS_REGISTER_COMMON
9804 Contains the name of the helper function that should be used at
9805 program startup to register TLS objects that are implicitly
9806 initialized to zero. If this is @code{NULL}, all TLS objects will
9807 have explicit initializers. The default causes libgcc's emulated TLS
9808 registration function to be used.
9811 @hook TARGET_EMUTLS_VAR_SECTION
9812 Contains the name of the section in which TLS control variables should
9813 be placed. The default of @code{NULL} allows these to be placed in
9817 @hook TARGET_EMUTLS_TMPL_SECTION
9818 Contains the name of the section in which TLS initializers should be
9819 placed. The default of @code{NULL} allows these to be placed in any
9823 @hook TARGET_EMUTLS_VAR_PREFIX
9824 Contains the prefix to be prepended to TLS control variable names.
9825 The default of @code{NULL} uses a target-specific prefix.
9828 @hook TARGET_EMUTLS_TMPL_PREFIX
9829 Contains the prefix to be prepended to TLS initializer objects. The
9830 default of @code{NULL} uses a target-specific prefix.
9833 @hook TARGET_EMUTLS_VAR_FIELDS
9834 Specifies a function that generates the FIELD_DECLs for a TLS control
9835 object type. @var{type} is the RECORD_TYPE the fields are for and
9836 @var{name} should be filled with the structure tag, if the default of
9837 @code{__emutls_object} is unsuitable. The default creates a type suitable
9838 for libgcc's emulated TLS function.
9841 @hook TARGET_EMUTLS_VAR_INIT
9842 Specifies a function that generates the CONSTRUCTOR to initialize a
9843 TLS control object. @var{var} is the TLS control object, @var{decl}
9844 is the TLS object and @var{tmpl_addr} is the address of the
9845 initializer. The default initializes libgcc's emulated TLS control object.
9848 @hook TARGET_EMUTLS_VAR_ALIGN_FIXED
9849 Specifies whether the alignment of TLS control variable objects is
9850 fixed and should not be increased as some backends may do to optimize
9851 single objects. The default is false.
9854 @hook TARGET_EMUTLS_DEBUG_FORM_TLS_ADDRESS
9855 Specifies whether a DWARF @code{DW_OP_form_tls_address} location descriptor
9856 may be used to describe emulated TLS control objects.
9859 @node MIPS Coprocessors
9860 @section Defining coprocessor specifics for MIPS targets.
9861 @cindex MIPS coprocessor-definition macros
9863 The MIPS specification allows MIPS implementations to have as many as 4
9864 coprocessors, each with as many as 32 private registers. GCC supports
9865 accessing these registers and transferring values between the registers
9866 and memory using asm-ized variables. For example:
9869 register unsigned int cp0count asm ("c0r1");
9875 (``c0r1'' is the default name of register 1 in coprocessor 0; alternate
9876 names may be added as described below, or the default names may be
9877 overridden entirely in @code{SUBTARGET_CONDITIONAL_REGISTER_USAGE}.)
9879 Coprocessor registers are assumed to be epilogue-used; sets to them will
9880 be preserved even if it does not appear that the register is used again
9881 later in the function.
9883 Another note: according to the MIPS spec, coprocessor 1 (if present) is
9884 the FPU@. One accesses COP1 registers through standard mips
9885 floating-point support; they are not included in this mechanism.
9887 There is one macro used in defining the MIPS coprocessor interface which
9888 you may want to override in subtargets; it is described below.
9890 @defmac ALL_COP_ADDITIONAL_REGISTER_NAMES
9891 A comma-separated list (with leading comma) of pairs describing the
9892 alternate names of coprocessor registers. The format of each entry should be
9894 @{ @var{alternatename}, @var{register_number}@}
9900 @section Parameters for Precompiled Header Validity Checking
9901 @cindex parameters, precompiled headers
9903 @hook TARGET_GET_PCH_VALIDITY
9904 This hook returns a pointer to the data needed by
9905 @code{TARGET_PCH_VALID_P} and sets
9906 @samp{*@var{sz}} to the size of the data in bytes.
9909 @hook TARGET_PCH_VALID_P
9910 This hook checks whether the options used to create a PCH file are
9911 compatible with the current settings. It returns @code{NULL}
9912 if so and a suitable error message if not. Error messages will
9913 be presented to the user and must be localized using @samp{_(@var{msg})}.
9915 @var{data} is the data that was returned by @code{TARGET_GET_PCH_VALIDITY}
9916 when the PCH file was created and @var{sz} is the size of that data in bytes.
9917 It's safe to assume that the data was created by the same version of the
9918 compiler, so no format checking is needed.
9920 The default definition of @code{default_pch_valid_p} should be
9921 suitable for most targets.
9924 @hook TARGET_CHECK_PCH_TARGET_FLAGS
9925 If this hook is nonnull, the default implementation of
9926 @code{TARGET_PCH_VALID_P} will use it to check for compatible values
9927 of @code{target_flags}. @var{pch_flags} specifies the value that
9928 @code{target_flags} had when the PCH file was created. The return
9929 value is the same as for @code{TARGET_PCH_VALID_P}.
9933 @section C++ ABI parameters
9934 @cindex parameters, c++ abi
9936 @hook TARGET_CXX_GUARD_TYPE
9937 Define this hook to override the integer type used for guard variables.
9938 These are used to implement one-time construction of static objects. The
9939 default is long_long_integer_type_node.
9942 @hook TARGET_CXX_GUARD_MASK_BIT
9943 This hook determines how guard variables are used. It should return
9944 @code{false} (the default) if the first byte should be used. A return value of
9945 @code{true} indicates that only the least significant bit should be used.
9948 @hook TARGET_CXX_GET_COOKIE_SIZE
9949 This hook returns the size of the cookie to use when allocating an array
9950 whose elements have the indicated @var{type}. Assumes that it is already
9951 known that a cookie is needed. The default is
9952 @code{max(sizeof (size_t), alignof(type))}, as defined in section 2.7 of the
9953 IA64/Generic C++ ABI@.
9956 @hook TARGET_CXX_COOKIE_HAS_SIZE
9957 This hook should return @code{true} if the element size should be stored in
9958 array cookies. The default is to return @code{false}.
9961 @hook TARGET_CXX_IMPORT_EXPORT_CLASS
9962 If defined by a backend this hook allows the decision made to export
9963 class @var{type} to be overruled. Upon entry @var{import_export}
9964 will contain 1 if the class is going to be exported, @minus{}1 if it is going
9965 to be imported and 0 otherwise. This function should return the
9966 modified value and perform any other actions necessary to support the
9967 backend's targeted operating system.
9970 @hook TARGET_CXX_CDTOR_RETURNS_THIS
9971 This hook should return @code{true} if constructors and destructors return
9972 the address of the object created/destroyed. The default is to return
9976 @hook TARGET_CXX_KEY_METHOD_MAY_BE_INLINE
9977 This hook returns true if the key method for a class (i.e., the method
9978 which, if defined in the current translation unit, causes the virtual
9979 table to be emitted) may be an inline function. Under the standard
9980 Itanium C++ ABI the key method may be an inline function so long as
9981 the function is not declared inline in the class definition. Under
9982 some variants of the ABI, an inline function can never be the key
9983 method. The default is to return @code{true}.
9986 @hook TARGET_CXX_DETERMINE_CLASS_DATA_VISIBILITY
9988 @hook TARGET_CXX_CLASS_DATA_ALWAYS_COMDAT
9989 This hook returns true (the default) if virtual tables and other
9990 similar implicit class data objects are always COMDAT if they have
9991 external linkage. If this hook returns false, then class data for
9992 classes whose virtual table will be emitted in only one translation
9993 unit will not be COMDAT.
9996 @hook TARGET_CXX_LIBRARY_RTTI_COMDAT
9997 This hook returns true (the default) if the RTTI information for
9998 the basic types which is defined in the C++ runtime should always
9999 be COMDAT, false if it should not be COMDAT.
10002 @hook TARGET_CXX_USE_AEABI_ATEXIT
10003 This hook returns true if @code{__aeabi_atexit} (as defined by the ARM EABI)
10004 should be used to register static destructors when @option{-fuse-cxa-atexit}
10005 is in effect. The default is to return false to use @code{__cxa_atexit}.
10008 @hook TARGET_CXX_USE_ATEXIT_FOR_CXA_ATEXIT
10009 This hook returns true if the target @code{atexit} function can be used
10010 in the same manner as @code{__cxa_atexit} to register C++ static
10011 destructors. This requires that @code{atexit}-registered functions in
10012 shared libraries are run in the correct order when the libraries are
10013 unloaded. The default is to return false.
10016 @hook TARGET_CXX_ADJUST_CLASS_AT_DEFINITION
10018 @node Named Address Spaces
10019 @section Adding support for named address spaces
10020 @cindex named address spaces
10022 The draft technical report of the ISO/IEC JTC1 S22 WG14 N1275
10023 standards committee, @cite{Programming Languages - C - Extensions to
10024 support embedded processors}, specifies a syntax for embedded
10025 processors to specify alternate address spaces. You can configure a
10026 GCC port to support section 5.1 of the draft report to add support for
10027 address spaces other than the default address space. These address
10028 spaces are new keywords that are similar to the @code{volatile} and
10029 @code{const} type attributes.
10031 Pointers to named address spaces can have a different size than
10032 pointers to the generic address space.
10034 For example, the SPU port uses the @code{__ea} address space to refer
10035 to memory in the host processor, rather than memory local to the SPU
10036 processor. Access to memory in the @code{__ea} address space involves
10037 issuing DMA operations to move data between the host processor and the
10038 local processor memory address space. Pointers in the @code{__ea}
10039 address space are either 32 bits or 64 bits based on the
10040 @option{-mea32} or @option{-mea64} switches (native SPU pointers are
10043 Internally, address spaces are represented as a small integer in the
10044 range 0 to 15 with address space 0 being reserved for the generic
10047 To register a named address space qualifier keyword with the C front end,
10048 the target may call the @code{c_register_addr_space} routine. For example,
10049 the SPU port uses the following to declare @code{__ea} as the keyword for
10050 named address space #1:
10052 #define ADDR_SPACE_EA 1
10053 c_register_addr_space ("__ea", ADDR_SPACE_EA);
10056 @hook TARGET_ADDR_SPACE_POINTER_MODE
10057 Define this to return the machine mode to use for pointers to
10058 @var{address_space} if the target supports named address spaces.
10059 The default version of this hook returns @code{ptr_mode} for the
10060 generic address space only.
10063 @hook TARGET_ADDR_SPACE_ADDRESS_MODE
10064 Define this to return the machine mode to use for addresses in
10065 @var{address_space} if the target supports named address spaces.
10066 The default version of this hook returns @code{Pmode} for the
10067 generic address space only.
10070 @hook TARGET_ADDR_SPACE_VALID_POINTER_MODE
10071 Define this to return nonzero if the port can handle pointers
10072 with machine mode @var{mode} to address space @var{as}. This target
10073 hook is the same as the @code{TARGET_VALID_POINTER_MODE} target hook,
10074 except that it includes explicit named address space support. The default
10075 version of this hook returns true for the modes returned by either the
10076 @code{TARGET_ADDR_SPACE_POINTER_MODE} or @code{TARGET_ADDR_SPACE_ADDRESS_MODE}
10077 target hooks for the given address space.
10080 @hook TARGET_ADDR_SPACE_LEGITIMATE_ADDRESS_P
10081 Define this to return true if @var{exp} is a valid address for mode
10082 @var{mode} in the named address space @var{as}. The @var{strict}
10083 parameter says whether strict addressing is in effect after reload has
10084 finished. This target hook is the same as the
10085 @code{TARGET_LEGITIMATE_ADDRESS_P} target hook, except that it includes
10086 explicit named address space support.
10089 @hook TARGET_ADDR_SPACE_LEGITIMIZE_ADDRESS
10090 Define this to modify an invalid address @var{x} to be a valid address
10091 with mode @var{mode} in the named address space @var{as}. This target
10092 hook is the same as the @code{TARGET_LEGITIMIZE_ADDRESS} target hook,
10093 except that it includes explicit named address space support.
10096 @hook TARGET_ADDR_SPACE_SUBSET_P
10097 Define this to return whether the @var{subset} named address space is
10098 contained within the @var{superset} named address space. Pointers to
10099 a named address space that is a subset of another named address space
10100 will be converted automatically without a cast if used together in
10101 arithmetic operations. Pointers to a superset address space can be
10102 converted to pointers to a subset address space via explicit casts.
10105 @hook TARGET_ADDR_SPACE_CONVERT
10106 Define this to convert the pointer expression represented by the RTL
10107 @var{op} with type @var{from_type} that points to a named address
10108 space to a new pointer expression with type @var{to_type} that points
10109 to a different named address space. When this hook it called, it is
10110 guaranteed that one of the two address spaces is a subset of the other,
10111 as determined by the @code{TARGET_ADDR_SPACE_SUBSET_P} target hook.
10115 @section Miscellaneous Parameters
10116 @cindex parameters, miscellaneous
10118 @c prevent bad page break with this line
10119 Here are several miscellaneous parameters.
10121 @defmac HAS_LONG_COND_BRANCH
10122 Define this boolean macro to indicate whether or not your architecture
10123 has conditional branches that can span all of memory. It is used in
10124 conjunction with an optimization that partitions hot and cold basic
10125 blocks into separate sections of the executable. If this macro is
10126 set to false, gcc will convert any conditional branches that attempt
10127 to cross between sections into unconditional branches or indirect jumps.
10130 @defmac HAS_LONG_UNCOND_BRANCH
10131 Define this boolean macro to indicate whether or not your architecture
10132 has unconditional branches that can span all of memory. It is used in
10133 conjunction with an optimization that partitions hot and cold basic
10134 blocks into separate sections of the executable. If this macro is
10135 set to false, gcc will convert any unconditional branches that attempt
10136 to cross between sections into indirect jumps.
10139 @defmac CASE_VECTOR_MODE
10140 An alias for a machine mode name. This is the machine mode that
10141 elements of a jump-table should have.
10144 @defmac CASE_VECTOR_SHORTEN_MODE (@var{min_offset}, @var{max_offset}, @var{body})
10145 Optional: return the preferred mode for an @code{addr_diff_vec}
10146 when the minimum and maximum offset are known. If you define this,
10147 it enables extra code in branch shortening to deal with @code{addr_diff_vec}.
10148 To make this work, you also have to define @code{INSN_ALIGN} and
10149 make the alignment for @code{addr_diff_vec} explicit.
10150 The @var{body} argument is provided so that the offset_unsigned and scale
10151 flags can be updated.
10154 @defmac CASE_VECTOR_PC_RELATIVE
10155 Define this macro to be a C expression to indicate when jump-tables
10156 should contain relative addresses. You need not define this macro if
10157 jump-tables never contain relative addresses, or jump-tables should
10158 contain relative addresses only when @option{-fPIC} or @option{-fPIC}
10162 @hook TARGET_CASE_VALUES_THRESHOLD
10163 This function return the smallest number of different values for which it
10164 is best to use a jump-table instead of a tree of conditional branches.
10165 The default is four for machines with a @code{casesi} instruction and
10166 five otherwise. This is best for most machines.
10169 @defmac CASE_USE_BIT_TESTS
10170 Define this macro to be a C expression to indicate whether C switch
10171 statements may be implemented by a sequence of bit tests. This is
10172 advantageous on processors that can efficiently implement left shift
10173 of 1 by the number of bits held in a register, but inappropriate on
10174 targets that would require a loop. By default, this macro returns
10175 @code{true} if the target defines an @code{ashlsi3} pattern, and
10176 @code{false} otherwise.
10179 @defmac WORD_REGISTER_OPERATIONS
10180 Define this macro if operations between registers with integral mode
10181 smaller than a word are always performed on the entire register.
10182 Most RISC machines have this property and most CISC machines do not.
10185 @defmac LOAD_EXTEND_OP (@var{mem_mode})
10186 Define this macro to be a C expression indicating when insns that read
10187 memory in @var{mem_mode}, an integral mode narrower than a word, set the
10188 bits outside of @var{mem_mode} to be either the sign-extension or the
10189 zero-extension of the data read. Return @code{SIGN_EXTEND} for values
10190 of @var{mem_mode} for which the
10191 insn sign-extends, @code{ZERO_EXTEND} for which it zero-extends, and
10192 @code{UNKNOWN} for other modes.
10194 This macro is not called with @var{mem_mode} non-integral or with a width
10195 greater than or equal to @code{BITS_PER_WORD}, so you may return any
10196 value in this case. Do not define this macro if it would always return
10197 @code{UNKNOWN}. On machines where this macro is defined, you will normally
10198 define it as the constant @code{SIGN_EXTEND} or @code{ZERO_EXTEND}.
10200 You may return a non-@code{UNKNOWN} value even if for some hard registers
10201 the sign extension is not performed, if for the @code{REGNO_REG_CLASS}
10202 of these hard registers @code{CANNOT_CHANGE_MODE_CLASS} returns nonzero
10203 when the @var{from} mode is @var{mem_mode} and the @var{to} mode is any
10204 integral mode larger than this but not larger than @code{word_mode}.
10206 You must return @code{UNKNOWN} if for some hard registers that allow this
10207 mode, @code{CANNOT_CHANGE_MODE_CLASS} says that they cannot change to
10208 @code{word_mode}, but that they can change to another integral mode that
10209 is larger then @var{mem_mode} but still smaller than @code{word_mode}.
10212 @defmac SHORT_IMMEDIATES_SIGN_EXTEND
10213 Define this macro if loading short immediate values into registers sign
10217 @defmac FIXUNS_TRUNC_LIKE_FIX_TRUNC
10218 Define this macro if the same instructions that convert a floating
10219 point number to a signed fixed point number also convert validly to an
10223 @hook TARGET_MIN_DIVISIONS_FOR_RECIP_MUL
10224 When @option{-ffast-math} is in effect, GCC tries to optimize
10225 divisions by the same divisor, by turning them into multiplications by
10226 the reciprocal. This target hook specifies the minimum number of divisions
10227 that should be there for GCC to perform the optimization for a variable
10228 of mode @var{mode}. The default implementation returns 3 if the machine
10229 has an instruction for the division, and 2 if it does not.
10233 The maximum number of bytes that a single instruction can move quickly
10234 between memory and registers or between two memory locations.
10237 @defmac MAX_MOVE_MAX
10238 The maximum number of bytes that a single instruction can move quickly
10239 between memory and registers or between two memory locations. If this
10240 is undefined, the default is @code{MOVE_MAX}. Otherwise, it is the
10241 constant value that is the largest value that @code{MOVE_MAX} can have
10245 @defmac SHIFT_COUNT_TRUNCATED
10246 A C expression that is nonzero if on this machine the number of bits
10247 actually used for the count of a shift operation is equal to the number
10248 of bits needed to represent the size of the object being shifted. When
10249 this macro is nonzero, the compiler will assume that it is safe to omit
10250 a sign-extend, zero-extend, and certain bitwise `and' instructions that
10251 truncates the count of a shift operation. On machines that have
10252 instructions that act on bit-fields at variable positions, which may
10253 include `bit test' instructions, a nonzero @code{SHIFT_COUNT_TRUNCATED}
10254 also enables deletion of truncations of the values that serve as
10255 arguments to bit-field instructions.
10257 If both types of instructions truncate the count (for shifts) and
10258 position (for bit-field operations), or if no variable-position bit-field
10259 instructions exist, you should define this macro.
10261 However, on some machines, such as the 80386 and the 680x0, truncation
10262 only applies to shift operations and not the (real or pretended)
10263 bit-field operations. Define @code{SHIFT_COUNT_TRUNCATED} to be zero on
10264 such machines. Instead, add patterns to the @file{md} file that include
10265 the implied truncation of the shift instructions.
10267 You need not define this macro if it would always have the value of zero.
10270 @anchor{TARGET_SHIFT_TRUNCATION_MASK}
10271 @hook TARGET_SHIFT_TRUNCATION_MASK
10272 This function describes how the standard shift patterns for @var{mode}
10273 deal with shifts by negative amounts or by more than the width of the mode.
10274 @xref{shift patterns}.
10276 On many machines, the shift patterns will apply a mask @var{m} to the
10277 shift count, meaning that a fixed-width shift of @var{x} by @var{y} is
10278 equivalent to an arbitrary-width shift of @var{x} by @var{y & m}. If
10279 this is true for mode @var{mode}, the function should return @var{m},
10280 otherwise it should return 0. A return value of 0 indicates that no
10281 particular behavior is guaranteed.
10283 Note that, unlike @code{SHIFT_COUNT_TRUNCATED}, this function does
10284 @emph{not} apply to general shift rtxes; it applies only to instructions
10285 that are generated by the named shift patterns.
10287 The default implementation of this function returns
10288 @code{GET_MODE_BITSIZE (@var{mode}) - 1} if @code{SHIFT_COUNT_TRUNCATED}
10289 and 0 otherwise. This definition is always safe, but if
10290 @code{SHIFT_COUNT_TRUNCATED} is false, and some shift patterns
10291 nevertheless truncate the shift count, you may get better code
10295 @defmac TRULY_NOOP_TRUNCATION (@var{outprec}, @var{inprec})
10296 A C expression which is nonzero if on this machine it is safe to
10297 ``convert'' an integer of @var{inprec} bits to one of @var{outprec}
10298 bits (where @var{outprec} is smaller than @var{inprec}) by merely
10299 operating on it as if it had only @var{outprec} bits.
10301 On many machines, this expression can be 1.
10303 @c rearranged this, removed the phrase "it is reported that". this was
10304 @c to fix an overfull hbox. --mew 10feb93
10305 When @code{TRULY_NOOP_TRUNCATION} returns 1 for a pair of sizes for
10306 modes for which @code{MODES_TIEABLE_P} is 0, suboptimal code can result.
10307 If this is the case, making @code{TRULY_NOOP_TRUNCATION} return 0 in
10308 such cases may improve things.
10311 @hook TARGET_MODE_REP_EXTENDED
10312 The representation of an integral mode can be such that the values
10313 are always extended to a wider integral mode. Return
10314 @code{SIGN_EXTEND} if values of @var{mode} are represented in
10315 sign-extended form to @var{rep_mode}. Return @code{UNKNOWN}
10316 otherwise. (Currently, none of the targets use zero-extended
10317 representation this way so unlike @code{LOAD_EXTEND_OP},
10318 @code{TARGET_MODE_REP_EXTENDED} is expected to return either
10319 @code{SIGN_EXTEND} or @code{UNKNOWN}. Also no target extends
10320 @var{mode} to @var{rep_mode} so that @var{rep_mode} is not the next
10321 widest integral mode and currently we take advantage of this fact.)
10323 Similarly to @code{LOAD_EXTEND_OP} you may return a non-@code{UNKNOWN}
10324 value even if the extension is not performed on certain hard registers
10325 as long as for the @code{REGNO_REG_CLASS} of these hard registers
10326 @code{CANNOT_CHANGE_MODE_CLASS} returns nonzero.
10328 Note that @code{TARGET_MODE_REP_EXTENDED} and @code{LOAD_EXTEND_OP}
10329 describe two related properties. If you define
10330 @code{TARGET_MODE_REP_EXTENDED (mode, word_mode)} you probably also want
10331 to define @code{LOAD_EXTEND_OP (mode)} to return the same type of
10334 In order to enforce the representation of @code{mode},
10335 @code{TRULY_NOOP_TRUNCATION} should return false when truncating to
10339 @defmac STORE_FLAG_VALUE
10340 A C expression describing the value returned by a comparison operator
10341 with an integral mode and stored by a store-flag instruction
10342 (@samp{cstore@var{mode}4}) when the condition is true. This description must
10343 apply to @emph{all} the @samp{cstore@var{mode}4} patterns and all the
10344 comparison operators whose results have a @code{MODE_INT} mode.
10346 A value of 1 or @minus{}1 means that the instruction implementing the
10347 comparison operator returns exactly 1 or @minus{}1 when the comparison is true
10348 and 0 when the comparison is false. Otherwise, the value indicates
10349 which bits of the result are guaranteed to be 1 when the comparison is
10350 true. This value is interpreted in the mode of the comparison
10351 operation, which is given by the mode of the first operand in the
10352 @samp{cstore@var{mode}4} pattern. Either the low bit or the sign bit of
10353 @code{STORE_FLAG_VALUE} be on. Presently, only those bits are used by
10356 If @code{STORE_FLAG_VALUE} is neither 1 or @minus{}1, the compiler will
10357 generate code that depends only on the specified bits. It can also
10358 replace comparison operators with equivalent operations if they cause
10359 the required bits to be set, even if the remaining bits are undefined.
10360 For example, on a machine whose comparison operators return an
10361 @code{SImode} value and where @code{STORE_FLAG_VALUE} is defined as
10362 @samp{0x80000000}, saying that just the sign bit is relevant, the
10366 (ne:SI (and:SI @var{x} (const_int @var{power-of-2})) (const_int 0))
10370 can be converted to
10373 (ashift:SI @var{x} (const_int @var{n}))
10377 where @var{n} is the appropriate shift count to move the bit being
10378 tested into the sign bit.
10380 There is no way to describe a machine that always sets the low-order bit
10381 for a true value, but does not guarantee the value of any other bits,
10382 but we do not know of any machine that has such an instruction. If you
10383 are trying to port GCC to such a machine, include an instruction to
10384 perform a logical-and of the result with 1 in the pattern for the
10385 comparison operators and let us know at @email{gcc@@gcc.gnu.org}.
10387 Often, a machine will have multiple instructions that obtain a value
10388 from a comparison (or the condition codes). Here are rules to guide the
10389 choice of value for @code{STORE_FLAG_VALUE}, and hence the instructions
10394 Use the shortest sequence that yields a valid definition for
10395 @code{STORE_FLAG_VALUE}. It is more efficient for the compiler to
10396 ``normalize'' the value (convert it to, e.g., 1 or 0) than for the
10397 comparison operators to do so because there may be opportunities to
10398 combine the normalization with other operations.
10401 For equal-length sequences, use a value of 1 or @minus{}1, with @minus{}1 being
10402 slightly preferred on machines with expensive jumps and 1 preferred on
10406 As a second choice, choose a value of @samp{0x80000001} if instructions
10407 exist that set both the sign and low-order bits but do not define the
10411 Otherwise, use a value of @samp{0x80000000}.
10414 Many machines can produce both the value chosen for
10415 @code{STORE_FLAG_VALUE} and its negation in the same number of
10416 instructions. On those machines, you should also define a pattern for
10417 those cases, e.g., one matching
10420 (set @var{A} (neg:@var{m} (ne:@var{m} @var{B} @var{C})))
10423 Some machines can also perform @code{and} or @code{plus} operations on
10424 condition code values with less instructions than the corresponding
10425 @samp{cstore@var{mode}4} insn followed by @code{and} or @code{plus}. On those
10426 machines, define the appropriate patterns. Use the names @code{incscc}
10427 and @code{decscc}, respectively, for the patterns which perform
10428 @code{plus} or @code{minus} operations on condition code values. See
10429 @file{rs6000.md} for some examples. The GNU Superoptimizer can be used to
10430 find such instruction sequences on other machines.
10432 If this macro is not defined, the default value, 1, is used. You need
10433 not define @code{STORE_FLAG_VALUE} if the machine has no store-flag
10434 instructions, or if the value generated by these instructions is 1.
10437 @defmac FLOAT_STORE_FLAG_VALUE (@var{mode})
10438 A C expression that gives a nonzero @code{REAL_VALUE_TYPE} value that is
10439 returned when comparison operators with floating-point results are true.
10440 Define this macro on machines that have comparison operations that return
10441 floating-point values. If there are no such operations, do not define
10445 @defmac VECTOR_STORE_FLAG_VALUE (@var{mode})
10446 A C expression that gives a rtx representing the nonzero true element
10447 for vector comparisons. The returned rtx should be valid for the inner
10448 mode of @var{mode} which is guaranteed to be a vector mode. Define
10449 this macro on machines that have vector comparison operations that
10450 return a vector result. If there are no such operations, do not define
10451 this macro. Typically, this macro is defined as @code{const1_rtx} or
10452 @code{constm1_rtx}. This macro may return @code{NULL_RTX} to prevent
10453 the compiler optimizing such vector comparison operations for the
10457 @defmac CLZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value})
10458 @defmacx CTZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value})
10459 A C expression that indicates whether the architecture defines a value
10460 for @code{clz} or @code{ctz} with a zero operand.
10461 A result of @code{0} indicates the value is undefined.
10462 If the value is defined for only the RTL expression, the macro should
10463 evaluate to @code{1}; if the value applies also to the corresponding optab
10464 entry (which is normally the case if it expands directly into
10465 the corresponding RTL), then the macro should evaluate to @code{2}.
10466 In the cases where the value is defined, @var{value} should be set to
10469 If this macro is not defined, the value of @code{clz} or
10470 @code{ctz} at zero is assumed to be undefined.
10472 This macro must be defined if the target's expansion for @code{ffs}
10473 relies on a particular value to get correct results. Otherwise it
10474 is not necessary, though it may be used to optimize some corner cases, and
10475 to provide a default expansion for the @code{ffs} optab.
10477 Note that regardless of this macro the ``definedness'' of @code{clz}
10478 and @code{ctz} at zero do @emph{not} extend to the builtin functions
10479 visible to the user. Thus one may be free to adjust the value at will
10480 to match the target expansion of these operations without fear of
10485 An alias for the machine mode for pointers. On most machines, define
10486 this to be the integer mode corresponding to the width of a hardware
10487 pointer; @code{SImode} on 32-bit machine or @code{DImode} on 64-bit machines.
10488 On some machines you must define this to be one of the partial integer
10489 modes, such as @code{PSImode}.
10491 The width of @code{Pmode} must be at least as large as the value of
10492 @code{POINTER_SIZE}. If it is not equal, you must define the macro
10493 @code{POINTERS_EXTEND_UNSIGNED} to specify how pointers are extended
10497 @defmac FUNCTION_MODE
10498 An alias for the machine mode used for memory references to functions
10499 being called, in @code{call} RTL expressions. On most CISC machines,
10500 where an instruction can begin at any byte address, this should be
10501 @code{QImode}. On most RISC machines, where all instructions have fixed
10502 size and alignment, this should be a mode with the same size and alignment
10503 as the machine instruction words - typically @code{SImode} or @code{HImode}.
10506 @defmac STDC_0_IN_SYSTEM_HEADERS
10507 In normal operation, the preprocessor expands @code{__STDC__} to the
10508 constant 1, to signify that GCC conforms to ISO Standard C@. On some
10509 hosts, like Solaris, the system compiler uses a different convention,
10510 where @code{__STDC__} is normally 0, but is 1 if the user specifies
10511 strict conformance to the C Standard.
10513 Defining @code{STDC_0_IN_SYSTEM_HEADERS} makes GNU CPP follows the host
10514 convention when processing system header files, but when processing user
10515 files @code{__STDC__} will always expand to 1.
10518 @defmac NO_IMPLICIT_EXTERN_C
10519 Define this macro if the system header files support C++ as well as C@.
10520 This macro inhibits the usual method of using system header files in
10521 C++, which is to pretend that the file's contents are enclosed in
10522 @samp{extern "C" @{@dots{}@}}.
10527 @defmac REGISTER_TARGET_PRAGMAS ()
10528 Define this macro if you want to implement any target-specific pragmas.
10529 If defined, it is a C expression which makes a series of calls to
10530 @code{c_register_pragma} or @code{c_register_pragma_with_expansion}
10531 for each pragma. The macro may also do any
10532 setup required for the pragmas.
10534 The primary reason to define this macro is to provide compatibility with
10535 other compilers for the same target. In general, we discourage
10536 definition of target-specific pragmas for GCC@.
10538 If the pragma can be implemented by attributes then you should consider
10539 defining the target hook @samp{TARGET_INSERT_ATTRIBUTES} as well.
10541 Preprocessor macros that appear on pragma lines are not expanded. All
10542 @samp{#pragma} directives that do not match any registered pragma are
10543 silently ignored, unless the user specifies @option{-Wunknown-pragmas}.
10546 @deftypefun void c_register_pragma (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *))
10547 @deftypefunx void c_register_pragma_with_expansion (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *))
10549 Each call to @code{c_register_pragma} or
10550 @code{c_register_pragma_with_expansion} establishes one pragma. The
10551 @var{callback} routine will be called when the preprocessor encounters a
10555 #pragma [@var{space}] @var{name} @dots{}
10558 @var{space} is the case-sensitive namespace of the pragma, or
10559 @code{NULL} to put the pragma in the global namespace. The callback
10560 routine receives @var{pfile} as its first argument, which can be passed
10561 on to cpplib's functions if necessary. You can lex tokens after the
10562 @var{name} by calling @code{pragma_lex}. Tokens that are not read by the
10563 callback will be silently ignored. The end of the line is indicated by
10564 a token of type @code{CPP_EOF}. Macro expansion occurs on the
10565 arguments of pragmas registered with
10566 @code{c_register_pragma_with_expansion} but not on the arguments of
10567 pragmas registered with @code{c_register_pragma}.
10569 Note that the use of @code{pragma_lex} is specific to the C and C++
10570 compilers. It will not work in the Java or Fortran compilers, or any
10571 other language compilers for that matter. Thus if @code{pragma_lex} is going
10572 to be called from target-specific code, it must only be done so when
10573 building the C and C++ compilers. This can be done by defining the
10574 variables @code{c_target_objs} and @code{cxx_target_objs} in the
10575 target entry in the @file{config.gcc} file. These variables should name
10576 the target-specific, language-specific object file which contains the
10577 code that uses @code{pragma_lex}. Note it will also be necessary to add a
10578 rule to the makefile fragment pointed to by @code{tmake_file} that shows
10579 how to build this object file.
10582 @defmac HANDLE_PRAGMA_PACK_WITH_EXPANSION
10583 Define this macro if macros should be expanded in the
10584 arguments of @samp{#pragma pack}.
10587 @hook TARGET_HANDLE_PRAGMA_EXTERN_PREFIX
10589 @defmac TARGET_DEFAULT_PACK_STRUCT
10590 If your target requires a structure packing default other than 0 (meaning
10591 the machine default), define this macro to the necessary value (in bytes).
10592 This must be a value that would also be valid to use with
10593 @samp{#pragma pack()} (that is, a small power of two).
10596 @defmac DOLLARS_IN_IDENTIFIERS
10597 Define this macro to control use of the character @samp{$} in
10598 identifier names for the C family of languages. 0 means @samp{$} is
10599 not allowed by default; 1 means it is allowed. 1 is the default;
10600 there is no need to define this macro in that case.
10603 @defmac NO_DOLLAR_IN_LABEL
10604 Define this macro if the assembler does not accept the character
10605 @samp{$} in label names. By default constructors and destructors in
10606 G++ have @samp{$} in the identifiers. If this macro is defined,
10607 @samp{.} is used instead.
10610 @defmac NO_DOT_IN_LABEL
10611 Define this macro if the assembler does not accept the character
10612 @samp{.} in label names. By default constructors and destructors in G++
10613 have names that use @samp{.}. If this macro is defined, these names
10614 are rewritten to avoid @samp{.}.
10617 @defmac INSN_SETS_ARE_DELAYED (@var{insn})
10618 Define this macro as a C expression that is nonzero if it is safe for the
10619 delay slot scheduler to place instructions in the delay slot of @var{insn},
10620 even if they appear to use a resource set or clobbered in @var{insn}.
10621 @var{insn} is always a @code{jump_insn} or an @code{insn}; GCC knows that
10622 every @code{call_insn} has this behavior. On machines where some @code{insn}
10623 or @code{jump_insn} is really a function call and hence has this behavior,
10624 you should define this macro.
10626 You need not define this macro if it would always return zero.
10629 @defmac INSN_REFERENCES_ARE_DELAYED (@var{insn})
10630 Define this macro as a C expression that is nonzero if it is safe for the
10631 delay slot scheduler to place instructions in the delay slot of @var{insn},
10632 even if they appear to set or clobber a resource referenced in @var{insn}.
10633 @var{insn} is always a @code{jump_insn} or an @code{insn}. On machines where
10634 some @code{insn} or @code{jump_insn} is really a function call and its operands
10635 are registers whose use is actually in the subroutine it calls, you should
10636 define this macro. Doing so allows the delay slot scheduler to move
10637 instructions which copy arguments into the argument registers into the delay
10638 slot of @var{insn}.
10640 You need not define this macro if it would always return zero.
10643 @defmac MULTIPLE_SYMBOL_SPACES
10644 Define this macro as a C expression that is nonzero if, in some cases,
10645 global symbols from one translation unit may not be bound to undefined
10646 symbols in another translation unit without user intervention. For
10647 instance, under Microsoft Windows symbols must be explicitly imported
10648 from shared libraries (DLLs).
10650 You need not define this macro if it would always evaluate to zero.
10653 @hook TARGET_MD_ASM_CLOBBERS
10654 This target hook should add to @var{clobbers} @code{STRING_CST} trees for
10655 any hard regs the port wishes to automatically clobber for an asm.
10656 It should return the result of the last @code{tree_cons} used to add a
10657 clobber. The @var{outputs}, @var{inputs} and @var{clobber} lists are the
10658 corresponding parameters to the asm and may be inspected to avoid
10659 clobbering a register that is an input or output of the asm. You can use
10660 @code{tree_overlaps_hard_reg_set}, declared in @file{tree.h}, to test
10661 for overlap with regards to asm-declared registers.
10664 @defmac MATH_LIBRARY
10665 Define this macro as a C string constant for the linker argument to link
10666 in the system math library, minus the initial @samp{"-l"}, or
10667 @samp{""} if the target does not have a
10668 separate math library.
10670 You need only define this macro if the default of @samp{"m"} is wrong.
10673 @defmac LIBRARY_PATH_ENV
10674 Define this macro as a C string constant for the environment variable that
10675 specifies where the linker should look for libraries.
10677 You need only define this macro if the default of @samp{"LIBRARY_PATH"}
10681 @defmac TARGET_POSIX_IO
10682 Define this macro if the target supports the following POSIX@ file
10683 functions, access, mkdir and file locking with fcntl / F_SETLKW@.
10684 Defining @code{TARGET_POSIX_IO} will enable the test coverage code
10685 to use file locking when exiting a program, which avoids race conditions
10686 if the program has forked. It will also create directories at run-time
10687 for cross-profiling.
10690 @defmac MAX_CONDITIONAL_EXECUTE
10692 A C expression for the maximum number of instructions to execute via
10693 conditional execution instructions instead of a branch. A value of
10694 @code{BRANCH_COST}+1 is the default if the machine does not use cc0, and
10695 1 if it does use cc0.
10698 @defmac IFCVT_MODIFY_TESTS (@var{ce_info}, @var{true_expr}, @var{false_expr})
10699 Used if the target needs to perform machine-dependent modifications on the
10700 conditionals used for turning basic blocks into conditionally executed code.
10701 @var{ce_info} points to a data structure, @code{struct ce_if_block}, which
10702 contains information about the currently processed blocks. @var{true_expr}
10703 and @var{false_expr} are the tests that are used for converting the
10704 then-block and the else-block, respectively. Set either @var{true_expr} or
10705 @var{false_expr} to a null pointer if the tests cannot be converted.
10708 @defmac IFCVT_MODIFY_MULTIPLE_TESTS (@var{ce_info}, @var{bb}, @var{true_expr}, @var{false_expr})
10709 Like @code{IFCVT_MODIFY_TESTS}, but used when converting more complicated
10710 if-statements into conditions combined by @code{and} and @code{or} operations.
10711 @var{bb} contains the basic block that contains the test that is currently
10712 being processed and about to be turned into a condition.
10715 @defmac IFCVT_MODIFY_INSN (@var{ce_info}, @var{pattern}, @var{insn})
10716 A C expression to modify the @var{PATTERN} of an @var{INSN} that is to
10717 be converted to conditional execution format. @var{ce_info} points to
10718 a data structure, @code{struct ce_if_block}, which contains information
10719 about the currently processed blocks.
10722 @defmac IFCVT_MODIFY_FINAL (@var{ce_info})
10723 A C expression to perform any final machine dependent modifications in
10724 converting code to conditional execution. The involved basic blocks
10725 can be found in the @code{struct ce_if_block} structure that is pointed
10726 to by @var{ce_info}.
10729 @defmac IFCVT_MODIFY_CANCEL (@var{ce_info})
10730 A C expression to cancel any machine dependent modifications in
10731 converting code to conditional execution. The involved basic blocks
10732 can be found in the @code{struct ce_if_block} structure that is pointed
10733 to by @var{ce_info}.
10736 @defmac IFCVT_INIT_EXTRA_FIELDS (@var{ce_info})
10737 A C expression to initialize any extra fields in a @code{struct ce_if_block}
10738 structure, which are defined by the @code{IFCVT_EXTRA_FIELDS} macro.
10741 @defmac IFCVT_EXTRA_FIELDS
10742 If defined, it should expand to a set of field declarations that will be
10743 added to the @code{struct ce_if_block} structure. These should be initialized
10744 by the @code{IFCVT_INIT_EXTRA_FIELDS} macro.
10747 @hook TARGET_MACHINE_DEPENDENT_REORG
10748 If non-null, this hook performs a target-specific pass over the
10749 instruction stream. The compiler will run it at all optimization levels,
10750 just before the point at which it normally does delayed-branch scheduling.
10752 The exact purpose of the hook varies from target to target. Some use
10753 it to do transformations that are necessary for correctness, such as
10754 laying out in-function constant pools or avoiding hardware hazards.
10755 Others use it as an opportunity to do some machine-dependent optimizations.
10757 You need not implement the hook if it has nothing to do. The default
10758 definition is null.
10761 @hook TARGET_INIT_BUILTINS
10762 Define this hook if you have any machine-specific built-in functions
10763 that need to be defined. It should be a function that performs the
10766 Machine specific built-in functions can be useful to expand special machine
10767 instructions that would otherwise not normally be generated because
10768 they have no equivalent in the source language (for example, SIMD vector
10769 instructions or prefetch instructions).
10771 To create a built-in function, call the function
10772 @code{lang_hooks.builtin_function}
10773 which is defined by the language front end. You can use any type nodes set
10774 up by @code{build_common_tree_nodes} and @code{build_common_tree_nodes_2};
10775 only language front ends that use those two functions will call
10776 @samp{TARGET_INIT_BUILTINS}.
10779 @hook TARGET_BUILTIN_DECL
10780 Define this hook if you have any machine-specific built-in functions
10781 that need to be defined. It should be a function that returns the
10782 builtin function declaration for the builtin function code @var{code}.
10783 If there is no such builtin and it cannot be initialized at this time
10784 if @var{initialize_p} is true the function should return @code{NULL_TREE}.
10785 If @var{code} is out of range the function should return
10786 @code{error_mark_node}.
10789 @hook TARGET_EXPAND_BUILTIN
10791 Expand a call to a machine specific built-in function that was set up by
10792 @samp{TARGET_INIT_BUILTINS}. @var{exp} is the expression for the
10793 function call; the result should go to @var{target} if that is
10794 convenient, and have mode @var{mode} if that is convenient.
10795 @var{subtarget} may be used as the target for computing one of
10796 @var{exp}'s operands. @var{ignore} is nonzero if the value is to be
10797 ignored. This function should return the result of the call to the
10801 @hook TARGET_RESOLVE_OVERLOADED_BUILTIN
10802 Select a replacement for a machine specific built-in function that
10803 was set up by @samp{TARGET_INIT_BUILTINS}. This is done
10804 @emph{before} regular type checking, and so allows the target to
10805 implement a crude form of function overloading. @var{fndecl} is the
10806 declaration of the built-in function. @var{arglist} is the list of
10807 arguments passed to the built-in function. The result is a
10808 complete expression that implements the operation, usually
10809 another @code{CALL_EXPR}.
10810 @var{arglist} really has type @samp{VEC(tree,gc)*}
10813 @hook TARGET_FOLD_BUILTIN
10814 Fold a call to a machine specific built-in function that was set up by
10815 @samp{TARGET_INIT_BUILTINS}. @var{fndecl} is the declaration of the
10816 built-in function. @var{n_args} is the number of arguments passed to
10817 the function; the arguments themselves are pointed to by @var{argp}.
10818 The result is another tree containing a simplified expression for the
10819 call's result. If @var{ignore} is true the value will be ignored.
10822 @hook TARGET_INVALID_WITHIN_DOLOOP
10824 Take an instruction in @var{insn} and return NULL if it is valid within a
10825 low-overhead loop, otherwise return a string explaining why doloop
10826 could not be applied.
10828 Many targets use special registers for low-overhead looping. For any
10829 instruction that clobbers these this function should return a string indicating
10830 the reason why the doloop could not be applied.
10831 By default, the RTL loop optimizer does not use a present doloop pattern for
10832 loops containing function calls or branch on table instructions.
10835 @defmac MD_CAN_REDIRECT_BRANCH (@var{branch1}, @var{branch2})
10837 Take a branch insn in @var{branch1} and another in @var{branch2}.
10838 Return true if redirecting @var{branch1} to the destination of
10839 @var{branch2} is possible.
10841 On some targets, branches may have a limited range. Optimizing the
10842 filling of delay slots can result in branches being redirected, and this
10843 may in turn cause a branch offset to overflow.
10846 @hook TARGET_COMMUTATIVE_P
10847 This target hook returns @code{true} if @var{x} is considered to be commutative.
10848 Usually, this is just COMMUTATIVE_P (@var{x}), but the HP PA doesn't consider
10849 PLUS to be commutative inside a MEM@. @var{outer_code} is the rtx code
10850 of the enclosing rtl, if known, otherwise it is UNKNOWN.
10853 @hook TARGET_ALLOCATE_INITIAL_VALUE
10855 When the initial value of a hard register has been copied in a pseudo
10856 register, it is often not necessary to actually allocate another register
10857 to this pseudo register, because the original hard register or a stack slot
10858 it has been saved into can be used. @code{TARGET_ALLOCATE_INITIAL_VALUE}
10859 is called at the start of register allocation once for each hard register
10860 that had its initial value copied by using
10861 @code{get_func_hard_reg_initial_val} or @code{get_hard_reg_initial_val}.
10862 Possible values are @code{NULL_RTX}, if you don't want
10863 to do any special allocation, a @code{REG} rtx---that would typically be
10864 the hard register itself, if it is known not to be clobbered---or a
10866 If you are returning a @code{MEM}, this is only a hint for the allocator;
10867 it might decide to use another register anyways.
10868 You may use @code{current_function_leaf_function} in the hook, functions
10869 that use @code{REG_N_SETS}, to determine if the hard
10870 register in question will not be clobbered.
10871 The default value of this hook is @code{NULL}, which disables any special
10875 @hook TARGET_UNSPEC_MAY_TRAP_P
10876 This target hook returns nonzero if @var{x}, an @code{unspec} or
10877 @code{unspec_volatile} operation, might cause a trap. Targets can use
10878 this hook to enhance precision of analysis for @code{unspec} and
10879 @code{unspec_volatile} operations. You may call @code{may_trap_p_1}
10880 to analyze inner elements of @var{x} in which case @var{flags} should be
10884 @hook TARGET_SET_CURRENT_FUNCTION
10885 The compiler invokes this hook whenever it changes its current function
10886 context (@code{cfun}). You can define this function if
10887 the back end needs to perform any initialization or reset actions on a
10888 per-function basis. For example, it may be used to implement function
10889 attributes that affect register usage or code generation patterns.
10890 The argument @var{decl} is the declaration for the new function context,
10891 and may be null to indicate that the compiler has left a function context
10892 and is returning to processing at the top level.
10893 The default hook function does nothing.
10895 GCC sets @code{cfun} to a dummy function context during initialization of
10896 some parts of the back end. The hook function is not invoked in this
10897 situation; you need not worry about the hook being invoked recursively,
10898 or when the back end is in a partially-initialized state.
10899 @code{cfun} might be @code{NULL} to indicate processing at top level,
10900 outside of any function scope.
10903 @defmac TARGET_OBJECT_SUFFIX
10904 Define this macro to be a C string representing the suffix for object
10905 files on your target machine. If you do not define this macro, GCC will
10906 use @samp{.o} as the suffix for object files.
10909 @defmac TARGET_EXECUTABLE_SUFFIX
10910 Define this macro to be a C string representing the suffix to be
10911 automatically added to executable files on your target machine. If you
10912 do not define this macro, GCC will use the null string as the suffix for
10916 @defmac COLLECT_EXPORT_LIST
10917 If defined, @code{collect2} will scan the individual object files
10918 specified on its command line and create an export list for the linker.
10919 Define this macro for systems like AIX, where the linker discards
10920 object files that are not referenced from @code{main} and uses export
10924 @defmac MODIFY_JNI_METHOD_CALL (@var{mdecl})
10925 Define this macro to a C expression representing a variant of the
10926 method call @var{mdecl}, if Java Native Interface (JNI) methods
10927 must be invoked differently from other methods on your target.
10928 For example, on 32-bit Microsoft Windows, JNI methods must be invoked using
10929 the @code{stdcall} calling convention and this macro is then
10930 defined as this expression:
10933 build_type_attribute_variant (@var{mdecl},
10935 (get_identifier ("stdcall"),
10940 @hook TARGET_CANNOT_MODIFY_JUMPS_P
10941 This target hook returns @code{true} past the point in which new jump
10942 instructions could be created. On machines that require a register for
10943 every jump such as the SHmedia ISA of SH5, this point would typically be
10944 reload, so this target hook should be defined to a function such as:
10948 cannot_modify_jumps_past_reload_p ()
10950 return (reload_completed || reload_in_progress);
10955 @hook TARGET_BRANCH_TARGET_REGISTER_CLASS
10956 This target hook returns a register class for which branch target register
10957 optimizations should be applied. All registers in this class should be
10958 usable interchangeably. After reload, registers in this class will be
10959 re-allocated and loads will be hoisted out of loops and be subjected
10960 to inter-block scheduling.
10963 @hook TARGET_BRANCH_TARGET_REGISTER_CALLEE_SAVED
10964 Branch target register optimization will by default exclude callee-saved
10966 that are not already live during the current function; if this target hook
10967 returns true, they will be included. The target code must than make sure
10968 that all target registers in the class returned by
10969 @samp{TARGET_BRANCH_TARGET_REGISTER_CLASS} that might need saving are
10970 saved. @var{after_prologue_epilogue_gen} indicates if prologues and
10971 epilogues have already been generated. Note, even if you only return
10972 true when @var{after_prologue_epilogue_gen} is false, you still are likely
10973 to have to make special provisions in @code{INITIAL_ELIMINATION_OFFSET}
10974 to reserve space for caller-saved target registers.
10977 @hook TARGET_HAVE_CONDITIONAL_EXECUTION
10978 This target hook returns true if the target supports conditional execution.
10979 This target hook is required only when the target has several different
10980 modes and they have different conditional execution capability, such as ARM.
10983 @hook TARGET_LOOP_UNROLL_ADJUST
10984 This target hook returns a new value for the number of times @var{loop}
10985 should be unrolled. The parameter @var{nunroll} is the number of times
10986 the loop is to be unrolled. The parameter @var{loop} is a pointer to
10987 the loop, which is going to be checked for unrolling. This target hook
10988 is required only when the target has special constraints like maximum
10989 number of memory accesses.
10992 @defmac POWI_MAX_MULTS
10993 If defined, this macro is interpreted as a signed integer C expression
10994 that specifies the maximum number of floating point multiplications
10995 that should be emitted when expanding exponentiation by an integer
10996 constant inline. When this value is defined, exponentiation requiring
10997 more than this number of multiplications is implemented by calling the
10998 system library's @code{pow}, @code{powf} or @code{powl} routines.
10999 The default value places no upper bound on the multiplication count.
11002 @deftypefn Macro void TARGET_EXTRA_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc})
11003 This target hook should register any extra include files for the
11004 target. The parameter @var{stdinc} indicates if normal include files
11005 are present. The parameter @var{sysroot} is the system root directory.
11006 The parameter @var{iprefix} is the prefix for the gcc directory.
11009 @deftypefn Macro void TARGET_EXTRA_PRE_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc})
11010 This target hook should register any extra include files for the
11011 target before any standard headers. The parameter @var{stdinc}
11012 indicates if normal include files are present. The parameter
11013 @var{sysroot} is the system root directory. The parameter
11014 @var{iprefix} is the prefix for the gcc directory.
11017 @deftypefn Macro void TARGET_OPTF (char *@var{path})
11018 This target hook should register special include paths for the target.
11019 The parameter @var{path} is the include to register. On Darwin
11020 systems, this is used for Framework includes, which have semantics
11021 that are different from @option{-I}.
11024 @defmac bool TARGET_USE_LOCAL_THUNK_ALIAS_P (tree @var{fndecl})
11025 This target macro returns @code{true} if it is safe to use a local alias
11026 for a virtual function @var{fndecl} when constructing thunks,
11027 @code{false} otherwise. By default, the macro returns @code{true} for all
11028 functions, if a target supports aliases (i.e.@: defines
11029 @code{ASM_OUTPUT_DEF}), @code{false} otherwise,
11032 @defmac TARGET_FORMAT_TYPES
11033 If defined, this macro is the name of a global variable containing
11034 target-specific format checking information for the @option{-Wformat}
11035 option. The default is to have no target-specific format checks.
11038 @defmac TARGET_N_FORMAT_TYPES
11039 If defined, this macro is the number of entries in
11040 @code{TARGET_FORMAT_TYPES}.
11043 @defmac TARGET_OVERRIDES_FORMAT_ATTRIBUTES
11044 If defined, this macro is the name of a global variable containing
11045 target-specific format overrides for the @option{-Wformat} option. The
11046 default is to have no target-specific format overrides. If defined,
11047 @code{TARGET_FORMAT_TYPES} must be defined, too.
11050 @defmac TARGET_OVERRIDES_FORMAT_ATTRIBUTES_COUNT
11051 If defined, this macro specifies the number of entries in
11052 @code{TARGET_OVERRIDES_FORMAT_ATTRIBUTES}.
11055 @defmac TARGET_OVERRIDES_FORMAT_INIT
11056 If defined, this macro specifies the optional initialization
11057 routine for target specific customizations of the system printf
11058 and scanf formatter settings.
11061 @hook TARGET_RELAXED_ORDERING
11062 If set to @code{true}, means that the target's memory model does not
11063 guarantee that loads which do not depend on one another will access
11064 main memory in the order of the instruction stream; if ordering is
11065 important, an explicit memory barrier must be used. This is true of
11066 many recent processors which implement a policy of ``relaxed,''
11067 ``weak,'' or ``release'' memory consistency, such as Alpha, PowerPC,
11068 and ia64. The default is @code{false}.
11071 @hook TARGET_INVALID_ARG_FOR_UNPROTOTYPED_FN
11072 If defined, this macro returns the diagnostic message when it is
11073 illegal to pass argument @var{val} to function @var{funcdecl}
11074 with prototype @var{typelist}.
11077 @hook TARGET_INVALID_CONVERSION
11078 If defined, this macro returns the diagnostic message when it is
11079 invalid to convert from @var{fromtype} to @var{totype}, or @code{NULL}
11080 if validity should be determined by the front end.
11083 @hook TARGET_INVALID_UNARY_OP
11084 If defined, this macro returns the diagnostic message when it is
11085 invalid to apply operation @var{op} (where unary plus is denoted by
11086 @code{CONVERT_EXPR}) to an operand of type @var{type}, or @code{NULL}
11087 if validity should be determined by the front end.
11090 @hook TARGET_INVALID_BINARY_OP
11091 If defined, this macro returns the diagnostic message when it is
11092 invalid to apply operation @var{op} to operands of types @var{type1}
11093 and @var{type2}, or @code{NULL} if validity should be determined by
11097 @hook TARGET_INVALID_PARAMETER_TYPE
11098 If defined, this macro returns the diagnostic message when it is
11099 invalid for functions to include parameters of type @var{type},
11100 or @code{NULL} if validity should be determined by
11101 the front end. This is currently used only by the C and C++ front ends.
11104 @hook TARGET_INVALID_RETURN_TYPE
11105 If defined, this macro returns the diagnostic message when it is
11106 invalid for functions to have return type @var{type},
11107 or @code{NULL} if validity should be determined by
11108 the front end. This is currently used only by the C and C++ front ends.
11111 @hook TARGET_PROMOTED_TYPE
11112 If defined, this target hook returns the type to which values of
11113 @var{type} should be promoted when they appear in expressions,
11114 analogous to the integer promotions, or @code{NULL_TREE} to use the
11115 front end's normal promotion rules. This hook is useful when there are
11116 target-specific types with special promotion rules.
11117 This is currently used only by the C and C++ front ends.
11120 @hook TARGET_CONVERT_TO_TYPE
11121 If defined, this hook returns the result of converting @var{expr} to
11122 @var{type}. It should return the converted expression,
11123 or @code{NULL_TREE} to apply the front end's normal conversion rules.
11124 This hook is useful when there are target-specific types with special
11126 This is currently used only by the C and C++ front ends.
11129 @defmac TARGET_USE_JCR_SECTION
11130 This macro determines whether to use the JCR section to register Java
11131 classes. By default, TARGET_USE_JCR_SECTION is defined to 1 if both
11132 SUPPORTS_WEAK and TARGET_HAVE_NAMED_SECTIONS are true, else 0.
11136 This macro determines the size of the objective C jump buffer for the
11137 NeXT runtime. By default, OBJC_JBLEN is defined to an innocuous value.
11140 @defmac LIBGCC2_UNWIND_ATTRIBUTE
11141 Define this macro if any target-specific attributes need to be attached
11142 to the functions in @file{libgcc} that provide low-level support for
11143 call stack unwinding. It is used in declarations in @file{unwind-generic.h}
11144 and the associated definitions of those functions.
11147 @hook TARGET_UPDATE_STACK_BOUNDARY
11148 Define this macro to update the current function stack boundary if
11152 @hook TARGET_GET_DRAP_RTX
11153 This hook should return an rtx for Dynamic Realign Argument Pointer (DRAP) if a
11154 different argument pointer register is needed to access the function's
11155 argument list due to stack realignment. Return @code{NULL} if no DRAP
11159 @hook TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS
11160 When optimization is disabled, this hook indicates whether or not
11161 arguments should be allocated to stack slots. Normally, GCC allocates
11162 stacks slots for arguments when not optimizing in order to make
11163 debugging easier. However, when a function is declared with
11164 @code{__attribute__((naked))}, there is no stack frame, and the compiler
11165 cannot safely move arguments from the registers in which they are passed
11166 to the stack. Therefore, this hook should return true in general, but
11167 false for naked functions. The default implementation always returns true.
11170 @hook TARGET_CONST_ANCHOR
11171 On some architectures it can take multiple instructions to synthesize
11172 a constant. If there is another constant already in a register that
11173 is close enough in value then it is preferable that the new constant
11174 is computed from this register using immediate addition or
11175 subtraction. We accomplish this through CSE. Besides the value of
11176 the constant we also add a lower and an upper constant anchor to the
11177 available expressions. These are then queried when encountering new
11178 constants. The anchors are computed by rounding the constant up and
11179 down to a multiple of the value of @code{TARGET_CONST_ANCHOR}.
11180 @code{TARGET_CONST_ANCHOR} should be the maximum positive value
11181 accepted by immediate-add plus one. We currently assume that the
11182 value of @code{TARGET_CONST_ANCHOR} is a power of 2. For example, on
11183 MIPS, where add-immediate takes a 16-bit signed value,
11184 @code{TARGET_CONST_ANCHOR} is set to @samp{0x8000}. The default value
11185 is zero, which disables this optimization. @end deftypevr