1 @c Copyright (C) 1988-2015 Free Software Foundation, Inc.
2 @c This is part of the GCC manual.
3 @c For copying conditions, see the file gcc.texi.
6 @chapter Target Description Macros and Functions
7 @cindex machine description macros
8 @cindex target description macros
9 @cindex macros, target description
10 @cindex @file{tm.h} macros
12 In addition to the file @file{@var{machine}.md}, a machine description
13 includes a C header file conventionally given the name
14 @file{@var{machine}.h} and a C source file named @file{@var{machine}.c}.
15 The header file defines numerous macros that convey the information
16 about the target machine that does not fit into the scheme of the
17 @file{.md} file. The file @file{tm.h} should be a link to
18 @file{@var{machine}.h}. The header file @file{config.h} includes
19 @file{tm.h} and most compiler source files include @file{config.h}. The
20 source file defines a variable @code{targetm}, which is a structure
21 containing pointers to functions and data relating to the target
22 machine. @file{@var{machine}.c} should also contain their definitions,
23 if they are not defined elsewhere in GCC, and other functions called
24 through the macros defined in the @file{.h} file.
27 * Target Structure:: The @code{targetm} variable.
28 * Driver:: Controlling how the driver runs the compilation passes.
29 * Run-time Target:: Defining @samp{-m} options like @option{-m68000} and @option{-m68020}.
30 * Per-Function Data:: Defining data structures for per-function information.
31 * Storage Layout:: Defining sizes and alignments of data.
32 * Type Layout:: Defining sizes and properties of basic user data types.
33 * Registers:: Naming and describing the hardware registers.
34 * Register Classes:: Defining the classes of hardware registers.
35 * Stack and Calling:: Defining which way the stack grows and by how much.
36 * Varargs:: Defining the varargs macros.
37 * Trampolines:: Code set up at run time to enter a nested function.
38 * Library Calls:: Controlling how library routines are implicitly called.
39 * Addressing Modes:: Defining addressing modes valid for memory operands.
40 * Anchored Addresses:: Defining how @option{-fsection-anchors} should work.
41 * Condition Code:: Defining how insns update the condition code.
42 * Costs:: Defining relative costs of different operations.
43 * Scheduling:: Adjusting the behavior of the instruction scheduler.
44 * Sections:: Dividing storage into text, data, and other sections.
45 * PIC:: Macros for position independent code.
46 * Assembler Format:: Defining how to write insns and pseudo-ops to output.
47 * Debugging Info:: Defining the format of debugging output.
48 * Floating Point:: Handling floating point for cross-compilers.
49 * Mode Switching:: Insertion of mode-switching instructions.
50 * Target Attributes:: Defining target-specific uses of @code{__attribute__}.
51 * Emulated TLS:: Emulated TLS support.
52 * MIPS Coprocessors:: MIPS coprocessor support and how to customize it.
53 * PCH Target:: Validity checking for precompiled headers.
54 * C++ ABI:: Controlling C++ ABI changes.
55 * Named Address Spaces:: Adding support for named address spaces
56 * Misc:: Everything else.
59 @node Target Structure
60 @section The Global @code{targetm} Variable
62 @cindex target functions
64 @deftypevar {struct gcc_target} targetm
65 The target @file{.c} file must define the global @code{targetm} variable
66 which contains pointers to functions and data relating to the target
67 machine. The variable is declared in @file{target.h};
68 @file{target-def.h} defines the macro @code{TARGET_INITIALIZER} which is
69 used to initialize the variable, and macros for the default initializers
70 for elements of the structure. The @file{.c} file should override those
71 macros for which the default definition is inappropriate. For example:
74 #include "target-def.h"
76 /* @r{Initialize the GCC target structure.} */
78 #undef TARGET_COMP_TYPE_ATTRIBUTES
79 #define TARGET_COMP_TYPE_ATTRIBUTES @var{machine}_comp_type_attributes
81 struct gcc_target targetm = TARGET_INITIALIZER;
85 Where a macro should be defined in the @file{.c} file in this manner to
86 form part of the @code{targetm} structure, it is documented below as a
87 ``Target Hook'' with a prototype. Many macros will change in future
88 from being defined in the @file{.h} file to being part of the
89 @code{targetm} structure.
91 Similarly, there is a @code{targetcm} variable for hooks that are
92 specific to front ends for C-family languages, documented as ``C
93 Target Hook''. This is declared in @file{c-family/c-target.h}, the
94 initializer @code{TARGETCM_INITIALIZER} in
95 @file{c-family/c-target-def.h}. If targets initialize @code{targetcm}
96 themselves, they should set @code{target_has_targetcm=yes} in
97 @file{config.gcc}; otherwise a default definition is used.
99 Similarly, there is a @code{targetm_common} variable for hooks that
100 are shared between the compiler driver and the compilers proper,
101 documented as ``Common Target Hook''. This is declared in
102 @file{common/common-target.h}, the initializer
103 @code{TARGETM_COMMON_INITIALIZER} in
104 @file{common/common-target-def.h}. If targets initialize
105 @code{targetm_common} themselves, they should set
106 @code{target_has_targetm_common=yes} in @file{config.gcc}; otherwise a
107 default definition is used.
110 @section Controlling the Compilation Driver, @file{gcc}
112 @cindex controlling the compilation driver
114 @c prevent bad page break with this line
115 You can control the compilation driver.
117 @defmac DRIVER_SELF_SPECS
118 A list of specs for the driver itself. It should be a suitable
119 initializer for an array of strings, with no surrounding braces.
121 The driver applies these specs to its own command line between loading
122 default @file{specs} files (but not command-line specified ones) and
123 choosing the multilib directory or running any subcommands. It
124 applies them in the order given, so each spec can depend on the
125 options added by earlier ones. It is also possible to remove options
126 using @samp{%<@var{option}} in the usual way.
128 This macro can be useful when a port has several interdependent target
129 options. It provides a way of standardizing the command line so
130 that the other specs are easier to write.
132 Do not define this macro if it does not need to do anything.
135 @defmac OPTION_DEFAULT_SPECS
136 A list of specs used to support configure-time default options (i.e.@:
137 @option{--with} options) in the driver. It should be a suitable initializer
138 for an array of structures, each containing two strings, without the
139 outermost pair of surrounding braces.
141 The first item in the pair is the name of the default. This must match
142 the code in @file{config.gcc} for the target. The second item is a spec
143 to apply if a default with this name was specified. The string
144 @samp{%(VALUE)} in the spec will be replaced by the value of the default
145 everywhere it occurs.
147 The driver will apply these specs to its own command line between loading
148 default @file{specs} files and processing @code{DRIVER_SELF_SPECS}, using
149 the same mechanism as @code{DRIVER_SELF_SPECS}.
151 Do not define this macro if it does not need to do anything.
155 A C string constant that tells the GCC driver program options to
156 pass to CPP@. It can also specify how to translate options you
157 give to GCC into options for GCC to pass to the CPP@.
159 Do not define this macro if it does not need to do anything.
162 @defmac CPLUSPLUS_CPP_SPEC
163 This macro is just like @code{CPP_SPEC}, but is used for C++, rather
164 than C@. If you do not define this macro, then the value of
165 @code{CPP_SPEC} (if any) will be used instead.
169 A C string constant that tells the GCC driver program options to
170 pass to @code{cc1}, @code{cc1plus}, @code{f771}, and the other language
172 It can also specify how to translate options you give to GCC into options
173 for GCC to pass to front ends.
175 Do not define this macro if it does not need to do anything.
179 A C string constant that tells the GCC driver program options to
180 pass to @code{cc1plus}. It can also specify how to translate options you
181 give to GCC into options for GCC to pass to the @code{cc1plus}.
183 Do not define this macro if it does not need to do anything.
184 Note that everything defined in CC1_SPEC is already passed to
185 @code{cc1plus} so there is no need to duplicate the contents of
186 CC1_SPEC in CC1PLUS_SPEC@.
190 A C string constant that tells the GCC driver program options to
191 pass to the assembler. It can also specify how to translate options
192 you give to GCC into options for GCC to pass to the assembler.
193 See the file @file{sun3.h} for an example of this.
195 Do not define this macro if it does not need to do anything.
198 @defmac ASM_FINAL_SPEC
199 A C string constant that tells the GCC driver program how to
200 run any programs which cleanup after the normal assembler.
201 Normally, this is not needed. See the file @file{mips.h} for
204 Do not define this macro if it does not need to do anything.
207 @defmac AS_NEEDS_DASH_FOR_PIPED_INPUT
208 Define this macro, with no value, if the driver should give the assembler
209 an argument consisting of a single dash, @option{-}, to instruct it to
210 read from its standard input (which will be a pipe connected to the
211 output of the compiler proper). This argument is given after any
212 @option{-o} option specifying the name of the output file.
214 If you do not define this macro, the assembler is assumed to read its
215 standard input if given no non-option arguments. If your assembler
216 cannot read standard input at all, use a @samp{%@{pipe:%e@}} construct;
217 see @file{mips.h} for instance.
221 A C string constant that tells the GCC driver program options to
222 pass to the linker. It can also specify how to translate options you
223 give to GCC into options for GCC to pass to the linker.
225 Do not define this macro if it does not need to do anything.
229 Another C string constant used much like @code{LINK_SPEC}. The difference
230 between the two is that @code{LIB_SPEC} is used at the end of the
231 command given to the linker.
233 If this macro is not defined, a default is provided that
234 loads the standard C library from the usual place. See @file{gcc.c}.
238 Another C string constant that tells the GCC driver program
239 how and when to place a reference to @file{libgcc.a} into the
240 linker command line. This constant is placed both before and after
241 the value of @code{LIB_SPEC}.
243 If this macro is not defined, the GCC driver provides a default that
244 passes the string @option{-lgcc} to the linker.
247 @defmac REAL_LIBGCC_SPEC
248 By default, if @code{ENABLE_SHARED_LIBGCC} is defined, the
249 @code{LIBGCC_SPEC} is not directly used by the driver program but is
250 instead modified to refer to different versions of @file{libgcc.a}
251 depending on the values of the command line flags @option{-static},
252 @option{-shared}, @option{-static-libgcc}, and @option{-shared-libgcc}. On
253 targets where these modifications are inappropriate, define
254 @code{REAL_LIBGCC_SPEC} instead. @code{REAL_LIBGCC_SPEC} tells the
255 driver how to place a reference to @file{libgcc} on the link command
256 line, but, unlike @code{LIBGCC_SPEC}, it is used unmodified.
259 @defmac USE_LD_AS_NEEDED
260 A macro that controls the modifications to @code{LIBGCC_SPEC}
261 mentioned in @code{REAL_LIBGCC_SPEC}. If nonzero, a spec will be
262 generated that uses @option{--as-needed} or equivalent options and the
263 shared @file{libgcc} in place of the
264 static exception handler library, when linking without any of
265 @code{-static}, @code{-static-libgcc}, or @code{-shared-libgcc}.
269 If defined, this C string constant is added to @code{LINK_SPEC}.
270 When @code{USE_LD_AS_NEEDED} is zero or undefined, it also affects
271 the modifications to @code{LIBGCC_SPEC} mentioned in
272 @code{REAL_LIBGCC_SPEC}.
275 @defmac STARTFILE_SPEC
276 Another C string constant used much like @code{LINK_SPEC}. The
277 difference between the two is that @code{STARTFILE_SPEC} is used at
278 the very beginning of the command given to the linker.
280 If this macro is not defined, a default is provided that loads the
281 standard C startup file from the usual place. See @file{gcc.c}.
285 Another C string constant used much like @code{LINK_SPEC}. The
286 difference between the two is that @code{ENDFILE_SPEC} is used at
287 the very end of the command given to the linker.
289 Do not define this macro if it does not need to do anything.
292 @defmac THREAD_MODEL_SPEC
293 GCC @code{-v} will print the thread model GCC was configured to use.
294 However, this doesn't work on platforms that are multilibbed on thread
295 models, such as AIX 4.3. On such platforms, define
296 @code{THREAD_MODEL_SPEC} such that it evaluates to a string without
297 blanks that names one of the recognized thread models. @code{%*}, the
298 default value of this macro, will expand to the value of
299 @code{thread_file} set in @file{config.gcc}.
302 @defmac SYSROOT_SUFFIX_SPEC
303 Define this macro to add a suffix to the target sysroot when GCC is
304 configured with a sysroot. This will cause GCC to search for usr/lib,
305 et al, within sysroot+suffix.
308 @defmac SYSROOT_HEADERS_SUFFIX_SPEC
309 Define this macro to add a headers_suffix to the target sysroot when
310 GCC is configured with a sysroot. This will cause GCC to pass the
311 updated sysroot+headers_suffix to CPP, causing it to search for
312 usr/include, et al, within sysroot+headers_suffix.
316 Define this macro to provide additional specifications to put in the
317 @file{specs} file that can be used in various specifications like
320 The definition should be an initializer for an array of structures,
321 containing a string constant, that defines the specification name, and a
322 string constant that provides the specification.
324 Do not define this macro if it does not need to do anything.
326 @code{EXTRA_SPECS} is useful when an architecture contains several
327 related targets, which have various @code{@dots{}_SPECS} which are similar
328 to each other, and the maintainer would like one central place to keep
331 For example, the PowerPC System V.4 targets use @code{EXTRA_SPECS} to
332 define either @code{_CALL_SYSV} when the System V calling sequence is
333 used or @code{_CALL_AIX} when the older AIX-based calling sequence is
336 The @file{config/rs6000/rs6000.h} target file defines:
339 #define EXTRA_SPECS \
340 @{ "cpp_sysv_default", CPP_SYSV_DEFAULT @},
342 #define CPP_SYS_DEFAULT ""
345 The @file{config/rs6000/sysv.h} target file defines:
349 "%@{posix: -D_POSIX_SOURCE @} \
350 %@{mcall-sysv: -D_CALL_SYSV @} \
351 %@{!mcall-sysv: %(cpp_sysv_default) @} \
352 %@{msoft-float: -D_SOFT_FLOAT@} %@{mcpu=403: -D_SOFT_FLOAT@}"
354 #undef CPP_SYSV_DEFAULT
355 #define CPP_SYSV_DEFAULT "-D_CALL_SYSV"
358 while the @file{config/rs6000/eabiaix.h} target file defines
359 @code{CPP_SYSV_DEFAULT} as:
362 #undef CPP_SYSV_DEFAULT
363 #define CPP_SYSV_DEFAULT "-D_CALL_AIX"
367 @defmac LINK_LIBGCC_SPECIAL_1
368 Define this macro if the driver program should find the library
369 @file{libgcc.a}. If you do not define this macro, the driver program will pass
370 the argument @option{-lgcc} to tell the linker to do the search.
373 @defmac LINK_GCC_C_SEQUENCE_SPEC
374 The sequence in which libgcc and libc are specified to the linker.
375 By default this is @code{%G %L %G}.
378 @defmac LINK_COMMAND_SPEC
379 A C string constant giving the complete command line need to execute the
380 linker. When you do this, you will need to update your port each time a
381 change is made to the link command line within @file{gcc.c}. Therefore,
382 define this macro only if you need to completely redefine the command
383 line for invoking the linker and there is no other way to accomplish
384 the effect you need. Overriding this macro may be avoidable by overriding
385 @code{LINK_GCC_C_SEQUENCE_SPEC} instead.
388 @deftypevr {Common Target Hook} bool TARGET_ALWAYS_STRIP_DOTDOT
389 True if @file{..} components should always be removed from directory names computed relative to GCC's internal directories, false (default) if such components should be preserved and directory names containing them passed to other tools such as the linker.
392 @defmac MULTILIB_DEFAULTS
393 Define this macro as a C expression for the initializer of an array of
394 string to tell the driver program which options are defaults for this
395 target and thus do not need to be handled specially when using
396 @code{MULTILIB_OPTIONS}.
398 Do not define this macro if @code{MULTILIB_OPTIONS} is not defined in
399 the target makefile fragment or if none of the options listed in
400 @code{MULTILIB_OPTIONS} are set by default.
401 @xref{Target Fragment}.
404 @defmac RELATIVE_PREFIX_NOT_LINKDIR
405 Define this macro to tell @command{gcc} that it should only translate
406 a @option{-B} prefix into a @option{-L} linker option if the prefix
407 indicates an absolute file name.
410 @defmac MD_EXEC_PREFIX
411 If defined, this macro is an additional prefix to try after
412 @code{STANDARD_EXEC_PREFIX}. @code{MD_EXEC_PREFIX} is not searched
413 when the compiler is built as a cross
414 compiler. If you define @code{MD_EXEC_PREFIX}, then be sure to add it
415 to the list of directories used to find the assembler in @file{configure.in}.
418 @defmac STANDARD_STARTFILE_PREFIX
419 Define this macro as a C string constant if you wish to override the
420 standard choice of @code{libdir} as the default prefix to
421 try when searching for startup files such as @file{crt0.o}.
422 @code{STANDARD_STARTFILE_PREFIX} is not searched when the compiler
423 is built as a cross compiler.
426 @defmac STANDARD_STARTFILE_PREFIX_1
427 Define this macro as a C string constant if you wish to override the
428 standard choice of @code{/lib} as a prefix to try after the default prefix
429 when searching for startup files such as @file{crt0.o}.
430 @code{STANDARD_STARTFILE_PREFIX_1} is not searched when the compiler
431 is built as a cross compiler.
434 @defmac STANDARD_STARTFILE_PREFIX_2
435 Define this macro as a C string constant if you wish to override the
436 standard choice of @code{/lib} as yet another prefix to try after the
437 default prefix when searching for startup files such as @file{crt0.o}.
438 @code{STANDARD_STARTFILE_PREFIX_2} is not searched when the compiler
439 is built as a cross compiler.
442 @defmac MD_STARTFILE_PREFIX
443 If defined, this macro supplies an additional prefix to try after the
444 standard prefixes. @code{MD_EXEC_PREFIX} is not searched when the
445 compiler is built as a cross compiler.
448 @defmac MD_STARTFILE_PREFIX_1
449 If defined, this macro supplies yet another prefix to try after the
450 standard prefixes. It is not searched when the compiler is built as a
454 @defmac INIT_ENVIRONMENT
455 Define this macro as a C string constant if you wish to set environment
456 variables for programs called by the driver, such as the assembler and
457 loader. The driver passes the value of this macro to @code{putenv} to
458 initialize the necessary environment variables.
461 @defmac LOCAL_INCLUDE_DIR
462 Define this macro as a C string constant if you wish to override the
463 standard choice of @file{/usr/local/include} as the default prefix to
464 try when searching for local header files. @code{LOCAL_INCLUDE_DIR}
465 comes before @code{NATIVE_SYSTEM_HEADER_DIR} (set in
466 @file{config.gcc}, normally @file{/usr/include}) in the search order.
468 Cross compilers do not search either @file{/usr/local/include} or its
472 @defmac NATIVE_SYSTEM_HEADER_COMPONENT
473 The ``component'' corresponding to @code{NATIVE_SYSTEM_HEADER_DIR}.
474 See @code{INCLUDE_DEFAULTS}, below, for the description of components.
475 If you do not define this macro, no component is used.
478 @defmac INCLUDE_DEFAULTS
479 Define this macro if you wish to override the entire default search path
480 for include files. For a native compiler, the default search path
481 usually consists of @code{GCC_INCLUDE_DIR}, @code{LOCAL_INCLUDE_DIR},
482 @code{GPLUSPLUS_INCLUDE_DIR}, and
483 @code{NATIVE_SYSTEM_HEADER_DIR}. In addition, @code{GPLUSPLUS_INCLUDE_DIR}
484 and @code{GCC_INCLUDE_DIR} are defined automatically by @file{Makefile},
485 and specify private search areas for GCC@. The directory
486 @code{GPLUSPLUS_INCLUDE_DIR} is used only for C++ programs.
488 The definition should be an initializer for an array of structures.
489 Each array element should have four elements: the directory name (a
490 string constant), the component name (also a string constant), a flag
491 for C++-only directories,
492 and a flag showing that the includes in the directory don't need to be
493 wrapped in @code{extern @samp{C}} when compiling C++. Mark the end of
494 the array with a null element.
496 The component name denotes what GNU package the include file is part of,
497 if any, in all uppercase letters. For example, it might be @samp{GCC}
498 or @samp{BINUTILS}. If the package is part of a vendor-supplied
499 operating system, code the component name as @samp{0}.
501 For example, here is the definition used for VAX/VMS:
504 #define INCLUDE_DEFAULTS \
506 @{ "GNU_GXX_INCLUDE:", "G++", 1, 1@}, \
507 @{ "GNU_CC_INCLUDE:", "GCC", 0, 0@}, \
508 @{ "SYS$SYSROOT:[SYSLIB.]", 0, 0, 0@}, \
515 Here is the order of prefixes tried for exec files:
519 Any prefixes specified by the user with @option{-B}.
522 The environment variable @code{GCC_EXEC_PREFIX} or, if @code{GCC_EXEC_PREFIX}
523 is not set and the compiler has not been installed in the configure-time
524 @var{prefix}, the location in which the compiler has actually been installed.
527 The directories specified by the environment variable @code{COMPILER_PATH}.
530 The macro @code{STANDARD_EXEC_PREFIX}, if the compiler has been installed
531 in the configured-time @var{prefix}.
534 The location @file{/usr/libexec/gcc/}, but only if this is a native compiler.
537 The location @file{/usr/lib/gcc/}, but only if this is a native compiler.
540 The macro @code{MD_EXEC_PREFIX}, if defined, but only if this is a native
544 Here is the order of prefixes tried for startfiles:
548 Any prefixes specified by the user with @option{-B}.
551 The environment variable @code{GCC_EXEC_PREFIX} or its automatically determined
552 value based on the installed toolchain location.
555 The directories specified by the environment variable @code{LIBRARY_PATH}
556 (or port-specific name; native only, cross compilers do not use this).
559 The macro @code{STANDARD_EXEC_PREFIX}, but only if the toolchain is installed
560 in the configured @var{prefix} or this is a native compiler.
563 The location @file{/usr/lib/gcc/}, but only if this is a native compiler.
566 The macro @code{MD_EXEC_PREFIX}, if defined, but only if this is a native
570 The macro @code{MD_STARTFILE_PREFIX}, if defined, but only if this is a
571 native compiler, or we have a target system root.
574 The macro @code{MD_STARTFILE_PREFIX_1}, if defined, but only if this is a
575 native compiler, or we have a target system root.
578 The macro @code{STANDARD_STARTFILE_PREFIX}, with any sysroot modifications.
579 If this path is relative it will be prefixed by @code{GCC_EXEC_PREFIX} and
580 the machine suffix or @code{STANDARD_EXEC_PREFIX} and the machine suffix.
583 The macro @code{STANDARD_STARTFILE_PREFIX_1}, but only if this is a native
584 compiler, or we have a target system root. The default for this macro is
588 The macro @code{STANDARD_STARTFILE_PREFIX_2}, but only if this is a native
589 compiler, or we have a target system root. The default for this macro is
593 @node Run-time Target
594 @section Run-time Target Specification
595 @cindex run-time target specification
596 @cindex predefined macros
597 @cindex target specifications
599 @c prevent bad page break with this line
600 Here are run-time target specifications.
602 @defmac TARGET_CPU_CPP_BUILTINS ()
603 This function-like macro expands to a block of code that defines
604 built-in preprocessor macros and assertions for the target CPU, using
605 the functions @code{builtin_define}, @code{builtin_define_std} and
606 @code{builtin_assert}. When the front end
607 calls this macro it provides a trailing semicolon, and since it has
608 finished command line option processing your code can use those
611 @code{builtin_assert} takes a string in the form you pass to the
612 command-line option @option{-A}, such as @code{cpu=mips}, and creates
613 the assertion. @code{builtin_define} takes a string in the form
614 accepted by option @option{-D} and unconditionally defines the macro.
616 @code{builtin_define_std} takes a string representing the name of an
617 object-like macro. If it doesn't lie in the user's namespace,
618 @code{builtin_define_std} defines it unconditionally. Otherwise, it
619 defines a version with two leading underscores, and another version
620 with two leading and trailing underscores, and defines the original
621 only if an ISO standard was not requested on the command line. For
622 example, passing @code{unix} defines @code{__unix}, @code{__unix__}
623 and possibly @code{unix}; passing @code{_mips} defines @code{__mips},
624 @code{__mips__} and possibly @code{_mips}, and passing @code{_ABI64}
625 defines only @code{_ABI64}.
627 You can also test for the C dialect being compiled. The variable
628 @code{c_language} is set to one of @code{clk_c}, @code{clk_cplusplus}
629 or @code{clk_objective_c}. Note that if we are preprocessing
630 assembler, this variable will be @code{clk_c} but the function-like
631 macro @code{preprocessing_asm_p()} will return true, so you might want
632 to check for that first. If you need to check for strict ANSI, the
633 variable @code{flag_iso} can be used. The function-like macro
634 @code{preprocessing_trad_p()} can be used to check for traditional
638 @defmac TARGET_OS_CPP_BUILTINS ()
639 Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional
640 and is used for the target operating system instead.
643 @defmac TARGET_OBJFMT_CPP_BUILTINS ()
644 Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional
645 and is used for the target object format. @file{elfos.h} uses this
646 macro to define @code{__ELF__}, so you probably do not need to define
650 @deftypevar {extern int} target_flags
651 This variable is declared in @file{options.h}, which is included before
652 any target-specific headers.
655 @deftypevr {Common Target Hook} int TARGET_DEFAULT_TARGET_FLAGS
656 This variable specifies the initial value of @code{target_flags}.
657 Its default setting is 0.
660 @cindex optional hardware or system features
661 @cindex features, optional, in system conventions
663 @deftypefn {Common Target Hook} bool TARGET_HANDLE_OPTION (struct gcc_options *@var{opts}, struct gcc_options *@var{opts_set}, const struct cl_decoded_option *@var{decoded}, location_t @var{loc})
664 This hook is called whenever the user specifies one of the
665 target-specific options described by the @file{.opt} definition files
666 (@pxref{Options}). It has the opportunity to do some option-specific
667 processing and should return true if the option is valid. The default
668 definition does nothing but return true.
670 @var{decoded} specifies the option and its arguments. @var{opts} and
671 @var{opts_set} are the @code{gcc_options} structures to be used for
672 storing option state, and @var{loc} is the location at which the
673 option was passed (@code{UNKNOWN_LOCATION} except for options passed
677 @deftypefn {C Target Hook} bool TARGET_HANDLE_C_OPTION (size_t @var{code}, const char *@var{arg}, int @var{value})
678 This target hook is called whenever the user specifies one of the
679 target-specific C language family options described by the @file{.opt}
680 definition files(@pxref{Options}). It has the opportunity to do some
681 option-specific processing and should return true if the option is
682 valid. The arguments are like for @code{TARGET_HANDLE_OPTION}. The
683 default definition does nothing but return false.
685 In general, you should use @code{TARGET_HANDLE_OPTION} to handle
686 options. However, if processing an option requires routines that are
687 only available in the C (and related language) front ends, then you
688 should use @code{TARGET_HANDLE_C_OPTION} instead.
691 @deftypefn {C Target Hook} tree TARGET_OBJC_CONSTRUCT_STRING_OBJECT (tree @var{string})
692 Targets may provide a string object type that can be used within and between C, C++ and their respective Objective-C dialects. A string object might, for example, embed encoding and length information. These objects are considered opaque to the compiler and handled as references. An ideal implementation makes the composition of the string object match that of the Objective-C @code{NSString} (@code{NXString} for GNUStep), allowing efficient interworking between C-only and Objective-C code. If a target implements string objects then this hook should return a reference to such an object constructed from the normal `C' string representation provided in @var{string}. At present, the hook is used by Objective-C only, to obtain a common-format string object when the target provides one.
695 @deftypefn {C Target Hook} void TARGET_OBJC_DECLARE_UNRESOLVED_CLASS_REFERENCE (const char *@var{classname})
696 Declare that Objective C class @var{classname} is referenced by the current TU.
699 @deftypefn {C Target Hook} void TARGET_OBJC_DECLARE_CLASS_DEFINITION (const char *@var{classname})
700 Declare that Objective C class @var{classname} is defined by the current TU.
703 @deftypefn {C Target Hook} bool TARGET_STRING_OBJECT_REF_TYPE_P (const_tree @var{stringref})
704 If a target implements string objects then this hook should return @code{true} if @var{stringref} is a valid reference to such an object.
707 @deftypefn {C Target Hook} void TARGET_CHECK_STRING_OBJECT_FORMAT_ARG (tree @var{format_arg}, tree @var{args_list})
708 If a target implements string objects then this hook should should provide a facility to check the function arguments in @var{args_list} against the format specifiers in @var{format_arg} where the type of @var{format_arg} is one recognized as a valid string reference type.
711 @deftypefn {Target Hook} void TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE (void)
712 This target function is similar to the hook @code{TARGET_OPTION_OVERRIDE}
713 but is called when the optimize level is changed via an attribute or
714 pragma or when it is reset at the end of the code affected by the
715 attribute or pragma. It is not called at the beginning of compilation
716 when @code{TARGET_OPTION_OVERRIDE} is called so if you want to perform these
717 actions then, you should have @code{TARGET_OPTION_OVERRIDE} call
718 @code{TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE}.
721 @defmac C_COMMON_OVERRIDE_OPTIONS
722 This is similar to the @code{TARGET_OPTION_OVERRIDE} hook
723 but is only used in the C
724 language frontends (C, Objective-C, C++, Objective-C++) and so can be
725 used to alter option flag variables which only exist in those
729 @deftypevr {Common Target Hook} {const struct default_options *} TARGET_OPTION_OPTIMIZATION_TABLE
730 Some machines may desire to change what optimizations are performed for
731 various optimization levels. This variable, if defined, describes
732 options to enable at particular sets of optimization levels. These
733 options are processed once
734 just after the optimization level is determined and before the remainder
735 of the command options have been parsed, so may be overridden by other
736 options passed explicitly.
738 This processing is run once at program startup and when the optimization
739 options are changed via @code{#pragma GCC optimize} or by using the
740 @code{optimize} attribute.
743 @deftypefn {Common Target Hook} void TARGET_OPTION_INIT_STRUCT (struct gcc_options *@var{opts})
744 Set target-dependent initial values of fields in @var{opts}.
747 @deftypefn {Common Target Hook} void TARGET_OPTION_DEFAULT_PARAMS (void)
748 Set target-dependent default values for @option{--param} settings, using calls to @code{set_default_param_value}.
751 @defmac SWITCHABLE_TARGET
752 Some targets need to switch between substantially different subtargets
753 during compilation. For example, the MIPS target has one subtarget for
754 the traditional MIPS architecture and another for MIPS16. Source code
755 can switch between these two subarchitectures using the @code{mips16}
756 and @code{nomips16} attributes.
758 Such subtargets can differ in things like the set of available
759 registers, the set of available instructions, the costs of various
760 operations, and so on. GCC caches a lot of this type of information
761 in global variables, and recomputing them for each subtarget takes a
762 significant amount of time. The compiler therefore provides a facility
763 for maintaining several versions of the global variables and quickly
764 switching between them; see @file{target-globals.h} for details.
766 Define this macro to 1 if your target needs this facility. The default
770 @deftypefn {Target Hook} bool TARGET_FLOAT_EXCEPTIONS_ROUNDING_SUPPORTED_P (void)
771 Returns true if the target supports IEEE 754 floating-point exceptions and rounding modes, false otherwise. This is intended to relate to the @code{float} and @code{double} types, but not necessarily @code{long double}. By default, returns true if the @code{adddf3} instruction pattern is available and false otherwise, on the assumption that hardware floating point supports exceptions and rounding modes but software floating point does not.
774 @node Per-Function Data
775 @section Defining data structures for per-function information.
776 @cindex per-function data
777 @cindex data structures
779 If the target needs to store information on a per-function basis, GCC
780 provides a macro and a couple of variables to allow this. Note, just
781 using statics to store the information is a bad idea, since GCC supports
782 nested functions, so you can be halfway through encoding one function
783 when another one comes along.
785 GCC defines a data structure called @code{struct function} which
786 contains all of the data specific to an individual function. This
787 structure contains a field called @code{machine} whose type is
788 @code{struct machine_function *}, which can be used by targets to point
789 to their own specific data.
791 If a target needs per-function specific data it should define the type
792 @code{struct machine_function} and also the macro @code{INIT_EXPANDERS}.
793 This macro should be used to initialize the function pointer
794 @code{init_machine_status}. This pointer is explained below.
796 One typical use of per-function, target specific data is to create an
797 RTX to hold the register containing the function's return address. This
798 RTX can then be used to implement the @code{__builtin_return_address}
799 function, for level 0.
801 Note---earlier implementations of GCC used a single data area to hold
802 all of the per-function information. Thus when processing of a nested
803 function began the old per-function data had to be pushed onto a
804 stack, and when the processing was finished, it had to be popped off the
805 stack. GCC used to provide function pointers called
806 @code{save_machine_status} and @code{restore_machine_status} to handle
807 the saving and restoring of the target specific information. Since the
808 single data area approach is no longer used, these pointers are no
811 @defmac INIT_EXPANDERS
812 Macro called to initialize any target specific information. This macro
813 is called once per function, before generation of any RTL has begun.
814 The intention of this macro is to allow the initialization of the
815 function pointer @code{init_machine_status}.
818 @deftypevar {void (*)(struct function *)} init_machine_status
819 If this function pointer is non-@code{NULL} it will be called once per
820 function, before function compilation starts, in order to allow the
821 target to perform any target specific initialization of the
822 @code{struct function} structure. It is intended that this would be
823 used to initialize the @code{machine} of that structure.
825 @code{struct machine_function} structures are expected to be freed by GC@.
826 Generally, any memory that they reference must be allocated by using
827 GC allocation, including the structure itself.
831 @section Storage Layout
832 @cindex storage layout
834 Note that the definitions of the macros in this table which are sizes or
835 alignments measured in bits do not need to be constant. They can be C
836 expressions that refer to static variables, such as the @code{target_flags}.
837 @xref{Run-time Target}.
839 @defmac BITS_BIG_ENDIAN
840 Define this macro to have the value 1 if the most significant bit in a
841 byte has the lowest number; otherwise define it to have the value zero.
842 This means that bit-field instructions count from the most significant
843 bit. If the machine has no bit-field instructions, then this must still
844 be defined, but it doesn't matter which value it is defined to. This
845 macro need not be a constant.
847 This macro does not affect the way structure fields are packed into
848 bytes or words; that is controlled by @code{BYTES_BIG_ENDIAN}.
851 @defmac BYTES_BIG_ENDIAN
852 Define this macro to have the value 1 if the most significant byte in a
853 word has the lowest number. This macro need not be a constant.
856 @defmac WORDS_BIG_ENDIAN
857 Define this macro to have the value 1 if, in a multiword object, the
858 most significant word has the lowest number. This applies to both
859 memory locations and registers; see @code{REG_WORDS_BIG_ENDIAN} if the
860 order of words in memory is not the same as the order in registers. This
861 macro need not be a constant.
864 @defmac REG_WORDS_BIG_ENDIAN
865 On some machines, the order of words in a multiword object differs between
866 registers in memory. In such a situation, define this macro to describe
867 the order of words in a register. The macro @code{WORDS_BIG_ENDIAN} controls
868 the order of words in memory.
871 @defmac FLOAT_WORDS_BIG_ENDIAN
872 Define this macro to have the value 1 if @code{DFmode}, @code{XFmode} or
873 @code{TFmode} floating point numbers are stored in memory with the word
874 containing the sign bit at the lowest address; otherwise define it to
875 have the value 0. This macro need not be a constant.
877 You need not define this macro if the ordering is the same as for
881 @defmac BITS_PER_WORD
882 Number of bits in a word. If you do not define this macro, the default
883 is @code{BITS_PER_UNIT * UNITS_PER_WORD}.
886 @defmac MAX_BITS_PER_WORD
887 Maximum number of bits in a word. If this is undefined, the default is
888 @code{BITS_PER_WORD}. Otherwise, it is the constant value that is the
889 largest value that @code{BITS_PER_WORD} can have at run-time.
892 @defmac UNITS_PER_WORD
893 Number of storage units in a word; normally the size of a general-purpose
894 register, a power of two from 1 or 8.
897 @defmac MIN_UNITS_PER_WORD
898 Minimum number of units in a word. If this is undefined, the default is
899 @code{UNITS_PER_WORD}. Otherwise, it is the constant value that is the
900 smallest value that @code{UNITS_PER_WORD} can have at run-time.
904 Width of a pointer, in bits. You must specify a value no wider than the
905 width of @code{Pmode}. If it is not equal to the width of @code{Pmode},
906 you must define @code{POINTERS_EXTEND_UNSIGNED}. If you do not specify
907 a value the default is @code{BITS_PER_WORD}.
910 @defmac POINTERS_EXTEND_UNSIGNED
911 A C expression that determines how pointers should be extended from
912 @code{ptr_mode} to either @code{Pmode} or @code{word_mode}. It is
913 greater than zero if pointers should be zero-extended, zero if they
914 should be sign-extended, and negative if some other sort of conversion
915 is needed. In the last case, the extension is done by the target's
916 @code{ptr_extend} instruction.
918 You need not define this macro if the @code{ptr_mode}, @code{Pmode}
919 and @code{word_mode} are all the same width.
922 @defmac PROMOTE_MODE (@var{m}, @var{unsignedp}, @var{type})
923 A macro to update @var{m} and @var{unsignedp} when an object whose type
924 is @var{type} and which has the specified mode and signedness is to be
925 stored in a register. This macro is only called when @var{type} is a
928 On most RISC machines, which only have operations that operate on a full
929 register, define this macro to set @var{m} to @code{word_mode} if
930 @var{m} is an integer mode narrower than @code{BITS_PER_WORD}. In most
931 cases, only integer modes should be widened because wider-precision
932 floating-point operations are usually more expensive than their narrower
935 For most machines, the macro definition does not change @var{unsignedp}.
936 However, some machines, have instructions that preferentially handle
937 either signed or unsigned quantities of certain modes. For example, on
938 the DEC Alpha, 32-bit loads from memory and 32-bit add instructions
939 sign-extend the result to 64 bits. On such machines, set
940 @var{unsignedp} according to which kind of extension is more efficient.
942 Do not define this macro if it would never modify @var{m}.
945 @deftypefn {Target Hook} machine_mode TARGET_PROMOTE_FUNCTION_MODE (const_tree @var{type}, machine_mode @var{mode}, int *@var{punsignedp}, const_tree @var{funtype}, int @var{for_return})
946 Like @code{PROMOTE_MODE}, but it is applied to outgoing function arguments or
947 function return values. The target hook should return the new mode
948 and possibly change @code{*@var{punsignedp}} if the promotion should
949 change signedness. This function is called only for scalar @emph{or
952 @var{for_return} allows to distinguish the promotion of arguments and
953 return values. If it is @code{1}, a return value is being promoted and
954 @code{TARGET_FUNCTION_VALUE} must perform the same promotions done here.
955 If it is @code{2}, the returned mode should be that of the register in
956 which an incoming parameter is copied, or the outgoing result is computed;
957 then the hook should return the same mode as @code{promote_mode}, though
958 the signedness may be different.
960 @var{type} can be NULL when promoting function arguments of libcalls.
962 The default is to not promote arguments and return values. You can
963 also define the hook to @code{default_promote_function_mode_always_promote}
964 if you would like to apply the same rules given by @code{PROMOTE_MODE}.
967 @defmac PARM_BOUNDARY
968 Normal alignment required for function parameters on the stack, in
969 bits. All stack parameters receive at least this much alignment
970 regardless of data type. On most machines, this is the same as the
974 @defmac STACK_BOUNDARY
975 Define this macro to the minimum alignment enforced by hardware for the
976 stack pointer on this machine. The definition is a C expression for the
977 desired alignment (measured in bits). This value is used as a default
978 if @code{PREFERRED_STACK_BOUNDARY} is not defined. On most machines,
979 this should be the same as @code{PARM_BOUNDARY}.
982 @defmac PREFERRED_STACK_BOUNDARY
983 Define this macro if you wish to preserve a certain alignment for the
984 stack pointer, greater than what the hardware enforces. The definition
985 is a C expression for the desired alignment (measured in bits). This
986 macro must evaluate to a value equal to or larger than
987 @code{STACK_BOUNDARY}.
990 @defmac INCOMING_STACK_BOUNDARY
991 Define this macro if the incoming stack boundary may be different
992 from @code{PREFERRED_STACK_BOUNDARY}. This macro must evaluate
993 to a value equal to or larger than @code{STACK_BOUNDARY}.
996 @defmac FUNCTION_BOUNDARY
997 Alignment required for a function entry point, in bits.
1000 @defmac BIGGEST_ALIGNMENT
1001 Biggest alignment that any data type can require on this machine, in
1002 bits. Note that this is not the biggest alignment that is supported,
1003 just the biggest alignment that, when violated, may cause a fault.
1006 @deftypevr {Target Hook} HOST_WIDE_INT TARGET_ABSOLUTE_BIGGEST_ALIGNMENT
1007 If defined, this target hook specifies the absolute biggest alignment
1008 that a type or variable can have on this machine, otherwise,
1009 @code{BIGGEST_ALIGNMENT} is used.
1012 @defmac MALLOC_ABI_ALIGNMENT
1013 Alignment, in bits, a C conformant malloc implementation has to
1014 provide. If not defined, the default value is @code{BITS_PER_WORD}.
1017 @defmac ATTRIBUTE_ALIGNED_VALUE
1018 Alignment used by the @code{__attribute__ ((aligned))} construct. If
1019 not defined, the default value is @code{BIGGEST_ALIGNMENT}.
1022 @defmac MINIMUM_ATOMIC_ALIGNMENT
1023 If defined, the smallest alignment, in bits, that can be given to an
1024 object that can be referenced in one operation, without disturbing any
1025 nearby object. Normally, this is @code{BITS_PER_UNIT}, but may be larger
1026 on machines that don't have byte or half-word store operations.
1029 @defmac BIGGEST_FIELD_ALIGNMENT
1030 Biggest alignment that any structure or union field can require on this
1031 machine, in bits. If defined, this overrides @code{BIGGEST_ALIGNMENT} for
1032 structure and union fields only, unless the field alignment has been set
1033 by the @code{__attribute__ ((aligned (@var{n})))} construct.
1036 @defmac ADJUST_FIELD_ALIGN (@var{field}, @var{computed})
1037 An expression for the alignment of a structure field @var{field} if the
1038 alignment computed in the usual way (including applying of
1039 @code{BIGGEST_ALIGNMENT} and @code{BIGGEST_FIELD_ALIGNMENT} to the
1040 alignment) is @var{computed}. It overrides alignment only if the
1041 field alignment has not been set by the
1042 @code{__attribute__ ((aligned (@var{n})))} construct.
1045 @defmac MAX_STACK_ALIGNMENT
1046 Biggest stack alignment guaranteed by the backend. Use this macro
1047 to specify the maximum alignment of a variable on stack.
1049 If not defined, the default value is @code{STACK_BOUNDARY}.
1051 @c FIXME: The default should be @code{PREFERRED_STACK_BOUNDARY}.
1052 @c But the fix for PR 32893 indicates that we can only guarantee
1053 @c maximum stack alignment on stack up to @code{STACK_BOUNDARY}, not
1054 @c @code{PREFERRED_STACK_BOUNDARY}, if stack alignment isn't supported.
1057 @defmac MAX_OFILE_ALIGNMENT
1058 Biggest alignment supported by the object file format of this machine.
1059 Use this macro to limit the alignment which can be specified using the
1060 @code{__attribute__ ((aligned (@var{n})))} construct. If not defined,
1061 the default value is @code{BIGGEST_ALIGNMENT}.
1063 On systems that use ELF, the default (in @file{config/elfos.h}) is
1064 the largest supported 32-bit ELF section alignment representable on
1065 a 32-bit host e.g. @samp{(((uint64_t) 1 << 28) * 8)}.
1066 On 32-bit ELF the largest supported section alignment in bits is
1067 @samp{(0x80000000 * 8)}, but this is not representable on 32-bit hosts.
1070 @defmac DATA_ALIGNMENT (@var{type}, @var{basic-align})
1071 If defined, a C expression to compute the alignment for a variable in
1072 the static store. @var{type} is the data type, and @var{basic-align} is
1073 the alignment that the object would ordinarily have. The value of this
1074 macro is used instead of that alignment to align the object.
1076 If this macro is not defined, then @var{basic-align} is used.
1079 One use of this macro is to increase alignment of medium-size data to
1080 make it all fit in fewer cache lines. Another is to cause character
1081 arrays to be word-aligned so that @code{strcpy} calls that copy
1082 constants to character arrays can be done inline.
1085 @defmac DATA_ABI_ALIGNMENT (@var{type}, @var{basic-align})
1086 Similar to @code{DATA_ALIGNMENT}, but for the cases where the ABI mandates
1087 some alignment increase, instead of optimization only purposes. E.g.@
1088 AMD x86-64 psABI says that variables with array type larger than 15 bytes
1089 must be aligned to 16 byte boundaries.
1091 If this macro is not defined, then @var{basic-align} is used.
1094 @defmac CONSTANT_ALIGNMENT (@var{constant}, @var{basic-align})
1095 If defined, a C expression to compute the alignment given to a constant
1096 that is being placed in memory. @var{constant} is the constant and
1097 @var{basic-align} is the alignment that the object would ordinarily
1098 have. The value of this macro is used instead of that alignment to
1101 If this macro is not defined, then @var{basic-align} is used.
1103 The typical use of this macro is to increase alignment for string
1104 constants to be word aligned so that @code{strcpy} calls that copy
1105 constants can be done inline.
1108 @defmac LOCAL_ALIGNMENT (@var{type}, @var{basic-align})
1109 If defined, a C expression to compute the alignment for a variable in
1110 the local store. @var{type} is the data type, and @var{basic-align} is
1111 the alignment that the object would ordinarily have. The value of this
1112 macro is used instead of that alignment to align the object.
1114 If this macro is not defined, then @var{basic-align} is used.
1116 One use of this macro is to increase alignment of medium-size data to
1117 make it all fit in fewer cache lines.
1119 If the value of this macro has a type, it should be an unsigned type.
1122 @deftypefn {Target Hook} HOST_WIDE_INT TARGET_VECTOR_ALIGNMENT (const_tree @var{type})
1123 This hook can be used to define the alignment for a vector of type
1124 @var{type}, in order to comply with a platform ABI. The default is to
1125 require natural alignment for vector types. The alignment returned by
1126 this hook must be a power-of-two multiple of the default alignment of
1127 the vector element type.
1130 @defmac STACK_SLOT_ALIGNMENT (@var{type}, @var{mode}, @var{basic-align})
1131 If defined, a C expression to compute the alignment for stack slot.
1132 @var{type} is the data type, @var{mode} is the widest mode available,
1133 and @var{basic-align} is the alignment that the slot would ordinarily
1134 have. The value of this macro is used instead of that alignment to
1137 If this macro is not defined, then @var{basic-align} is used when
1138 @var{type} is @code{NULL}. Otherwise, @code{LOCAL_ALIGNMENT} will
1141 This macro is to set alignment of stack slot to the maximum alignment
1142 of all possible modes which the slot may have.
1144 If the value of this macro has a type, it should be an unsigned type.
1147 @defmac LOCAL_DECL_ALIGNMENT (@var{decl})
1148 If defined, a C expression to compute the alignment for a local
1149 variable @var{decl}.
1151 If this macro is not defined, then
1152 @code{LOCAL_ALIGNMENT (TREE_TYPE (@var{decl}), DECL_ALIGN (@var{decl}))}
1155 One use of this macro is to increase alignment of medium-size data to
1156 make it all fit in fewer cache lines.
1158 If the value of this macro has a type, it should be an unsigned type.
1161 @defmac MINIMUM_ALIGNMENT (@var{exp}, @var{mode}, @var{align})
1162 If defined, a C expression to compute the minimum required alignment
1163 for dynamic stack realignment purposes for @var{exp} (a type or decl),
1164 @var{mode}, assuming normal alignment @var{align}.
1166 If this macro is not defined, then @var{align} will be used.
1169 @defmac EMPTY_FIELD_BOUNDARY
1170 Alignment in bits to be given to a structure bit-field that follows an
1171 empty field such as @code{int : 0;}.
1173 If @code{PCC_BITFIELD_TYPE_MATTERS} is true, it overrides this macro.
1176 @defmac STRUCTURE_SIZE_BOUNDARY
1177 Number of bits which any structure or union's size must be a multiple of.
1178 Each structure or union's size is rounded up to a multiple of this.
1180 If you do not define this macro, the default is the same as
1181 @code{BITS_PER_UNIT}.
1184 @defmac STRICT_ALIGNMENT
1185 Define this macro to be the value 1 if instructions will fail to work
1186 if given data not on the nominal alignment. If instructions will merely
1187 go slower in that case, define this macro as 0.
1190 @defmac PCC_BITFIELD_TYPE_MATTERS
1191 Define this if you wish to imitate the way many other C compilers handle
1192 alignment of bit-fields and the structures that contain them.
1194 The behavior is that the type written for a named bit-field (@code{int},
1195 @code{short}, or other integer type) imposes an alignment for the entire
1196 structure, as if the structure really did contain an ordinary field of
1197 that type. In addition, the bit-field is placed within the structure so
1198 that it would fit within such a field, not crossing a boundary for it.
1200 Thus, on most machines, a named bit-field whose type is written as
1201 @code{int} would not cross a four-byte boundary, and would force
1202 four-byte alignment for the whole structure. (The alignment used may
1203 not be four bytes; it is controlled by the other alignment parameters.)
1205 An unnamed bit-field will not affect the alignment of the containing
1208 If the macro is defined, its definition should be a C expression;
1209 a nonzero value for the expression enables this behavior.
1211 Note that if this macro is not defined, or its value is zero, some
1212 bit-fields may cross more than one alignment boundary. The compiler can
1213 support such references if there are @samp{insv}, @samp{extv}, and
1214 @samp{extzv} insns that can directly reference memory.
1216 The other known way of making bit-fields work is to define
1217 @code{STRUCTURE_SIZE_BOUNDARY} as large as @code{BIGGEST_ALIGNMENT}.
1218 Then every structure can be accessed with fullwords.
1220 Unless the machine has bit-field instructions or you define
1221 @code{STRUCTURE_SIZE_BOUNDARY} that way, you must define
1222 @code{PCC_BITFIELD_TYPE_MATTERS} to have a nonzero value.
1224 If your aim is to make GCC use the same conventions for laying out
1225 bit-fields as are used by another compiler, here is how to investigate
1226 what the other compiler does. Compile and run this program:
1245 printf ("Size of foo1 is %d\n",
1246 sizeof (struct foo1));
1247 printf ("Size of foo2 is %d\n",
1248 sizeof (struct foo2));
1253 If this prints 2 and 5, then the compiler's behavior is what you would
1254 get from @code{PCC_BITFIELD_TYPE_MATTERS}.
1257 @defmac BITFIELD_NBYTES_LIMITED
1258 Like @code{PCC_BITFIELD_TYPE_MATTERS} except that its effect is limited
1259 to aligning a bit-field within the structure.
1262 @deftypefn {Target Hook} bool TARGET_ALIGN_ANON_BITFIELD (void)
1263 When @code{PCC_BITFIELD_TYPE_MATTERS} is true this hook will determine
1264 whether unnamed bitfields affect the alignment of the containing
1265 structure. The hook should return true if the structure should inherit
1266 the alignment requirements of an unnamed bitfield's type.
1269 @deftypefn {Target Hook} bool TARGET_NARROW_VOLATILE_BITFIELD (void)
1270 This target hook should return @code{true} if accesses to volatile bitfields
1271 should use the narrowest mode possible. It should return @code{false} if
1272 these accesses should use the bitfield container type.
1274 The default is @code{false}.
1277 @deftypefn {Target Hook} bool TARGET_MEMBER_TYPE_FORCES_BLK (const_tree @var{field}, machine_mode @var{mode})
1278 Return true if a structure, union or array containing @var{field} should
1279 be accessed using @code{BLKMODE}.
1281 If @var{field} is the only field in the structure, @var{mode} is its
1282 mode, otherwise @var{mode} is VOIDmode. @var{mode} is provided in the
1283 case where structures of one field would require the structure's mode to
1284 retain the field's mode.
1286 Normally, this is not needed.
1289 @defmac ROUND_TYPE_ALIGN (@var{type}, @var{computed}, @var{specified})
1290 Define this macro as an expression for the alignment of a type (given
1291 by @var{type} as a tree node) if the alignment computed in the usual
1292 way is @var{computed} and the alignment explicitly specified was
1295 The default is to use @var{specified} if it is larger; otherwise, use
1296 the smaller of @var{computed} and @code{BIGGEST_ALIGNMENT}
1299 @defmac MAX_FIXED_MODE_SIZE
1300 An integer expression for the size in bits of the largest integer
1301 machine mode that should actually be used. All integer machine modes of
1302 this size or smaller can be used for structures and unions with the
1303 appropriate sizes. If this macro is undefined, @code{GET_MODE_BITSIZE
1304 (DImode)} is assumed.
1307 @defmac STACK_SAVEAREA_MODE (@var{save_level})
1308 If defined, an expression of type @code{machine_mode} that
1309 specifies the mode of the save area operand of a
1310 @code{save_stack_@var{level}} named pattern (@pxref{Standard Names}).
1311 @var{save_level} is one of @code{SAVE_BLOCK}, @code{SAVE_FUNCTION}, or
1312 @code{SAVE_NONLOCAL} and selects which of the three named patterns is
1313 having its mode specified.
1315 You need not define this macro if it always returns @code{Pmode}. You
1316 would most commonly define this macro if the
1317 @code{save_stack_@var{level}} patterns need to support both a 32- and a
1321 @defmac STACK_SIZE_MODE
1322 If defined, an expression of type @code{machine_mode} that
1323 specifies the mode of the size increment operand of an
1324 @code{allocate_stack} named pattern (@pxref{Standard Names}).
1326 You need not define this macro if it always returns @code{word_mode}.
1327 You would most commonly define this macro if the @code{allocate_stack}
1328 pattern needs to support both a 32- and a 64-bit mode.
1331 @deftypefn {Target Hook} machine_mode TARGET_LIBGCC_CMP_RETURN_MODE (void)
1332 This target hook should return the mode to be used for the return value
1333 of compare instructions expanded to libgcc calls. If not defined
1334 @code{word_mode} is returned which is the right choice for a majority of
1338 @deftypefn {Target Hook} machine_mode TARGET_LIBGCC_SHIFT_COUNT_MODE (void)
1339 This target hook should return the mode to be used for the shift count operand
1340 of shift instructions expanded to libgcc calls. If not defined
1341 @code{word_mode} is returned which is the right choice for a majority of
1345 @deftypefn {Target Hook} machine_mode TARGET_UNWIND_WORD_MODE (void)
1346 Return machine mode to be used for @code{_Unwind_Word} type.
1347 The default is to use @code{word_mode}.
1350 @deftypefn {Target Hook} bool TARGET_MS_BITFIELD_LAYOUT_P (const_tree @var{record_type})
1351 This target hook returns @code{true} if bit-fields in the given
1352 @var{record_type} are to be laid out following the rules of Microsoft
1353 Visual C/C++, namely: (i) a bit-field won't share the same storage
1354 unit with the previous bit-field if their underlying types have
1355 different sizes, and the bit-field will be aligned to the highest
1356 alignment of the underlying types of itself and of the previous
1357 bit-field; (ii) a zero-sized bit-field will affect the alignment of
1358 the whole enclosing structure, even if it is unnamed; except that
1359 (iii) a zero-sized bit-field will be disregarded unless it follows
1360 another bit-field of nonzero size. If this hook returns @code{true},
1361 other macros that control bit-field layout are ignored.
1363 When a bit-field is inserted into a packed record, the whole size
1364 of the underlying type is used by one or more same-size adjacent
1365 bit-fields (that is, if its long:3, 32 bits is used in the record,
1366 and any additional adjacent long bit-fields are packed into the same
1367 chunk of 32 bits. However, if the size changes, a new field of that
1368 size is allocated). In an unpacked record, this is the same as using
1369 alignment, but not equivalent when packing.
1371 If both MS bit-fields and @samp{__attribute__((packed))} are used,
1372 the latter will take precedence. If @samp{__attribute__((packed))} is
1373 used on a single field when MS bit-fields are in use, it will take
1374 precedence for that field, but the alignment of the rest of the structure
1375 may affect its placement.
1378 @deftypefn {Target Hook} bool TARGET_DECIMAL_FLOAT_SUPPORTED_P (void)
1379 Returns true if the target supports decimal floating point.
1382 @deftypefn {Target Hook} bool TARGET_FIXED_POINT_SUPPORTED_P (void)
1383 Returns true if the target supports fixed-point arithmetic.
1386 @deftypefn {Target Hook} void TARGET_EXPAND_TO_RTL_HOOK (void)
1387 This hook is called just before expansion into rtl, allowing the target
1388 to perform additional initializations or analysis before the expansion.
1389 For example, the rs6000 port uses it to allocate a scratch stack slot
1390 for use in copying SDmode values between memory and floating point
1391 registers whenever the function being expanded has any SDmode
1395 @deftypefn {Target Hook} void TARGET_INSTANTIATE_DECLS (void)
1396 This hook allows the backend to perform additional instantiations on rtl
1397 that are not actually in any insns yet, but will be later.
1400 @deftypefn {Target Hook} {const char *} TARGET_MANGLE_TYPE (const_tree @var{type})
1401 If your target defines any fundamental types, or any types your target
1402 uses should be mangled differently from the default, define this hook
1403 to return the appropriate encoding for these types as part of a C++
1404 mangled name. The @var{type} argument is the tree structure representing
1405 the type to be mangled. The hook may be applied to trees which are
1406 not target-specific fundamental types; it should return @code{NULL}
1407 for all such types, as well as arguments it does not recognize. If the
1408 return value is not @code{NULL}, it must point to a statically-allocated
1411 Target-specific fundamental types might be new fundamental types or
1412 qualified versions of ordinary fundamental types. Encode new
1413 fundamental types as @samp{@w{u @var{n} @var{name}}}, where @var{name}
1414 is the name used for the type in source code, and @var{n} is the
1415 length of @var{name} in decimal. Encode qualified versions of
1416 ordinary types as @samp{@w{U @var{n} @var{name} @var{code}}}, where
1417 @var{name} is the name used for the type qualifier in source code,
1418 @var{n} is the length of @var{name} as above, and @var{code} is the
1419 code used to represent the unqualified version of this type. (See
1420 @code{write_builtin_type} in @file{cp/mangle.c} for the list of
1421 codes.) In both cases the spaces are for clarity; do not include any
1422 spaces in your string.
1424 This hook is applied to types prior to typedef resolution. If the mangled
1425 name for a particular type depends only on that type's main variant, you
1426 can perform typedef resolution yourself using @code{TYPE_MAIN_VARIANT}
1429 The default version of this hook always returns @code{NULL}, which is
1430 appropriate for a target that does not define any new fundamental
1435 @section Layout of Source Language Data Types
1437 These macros define the sizes and other characteristics of the standard
1438 basic data types used in programs being compiled. Unlike the macros in
1439 the previous section, these apply to specific features of C and related
1440 languages, rather than to fundamental aspects of storage layout.
1442 @defmac INT_TYPE_SIZE
1443 A C expression for the size in bits of the type @code{int} on the
1444 target machine. If you don't define this, the default is one word.
1447 @defmac SHORT_TYPE_SIZE
1448 A C expression for the size in bits of the type @code{short} on the
1449 target machine. If you don't define this, the default is half a word.
1450 (If this would be less than one storage unit, it is rounded up to one
1454 @defmac LONG_TYPE_SIZE
1455 A C expression for the size in bits of the type @code{long} on the
1456 target machine. If you don't define this, the default is one word.
1459 @defmac ADA_LONG_TYPE_SIZE
1460 On some machines, the size used for the Ada equivalent of the type
1461 @code{long} by a native Ada compiler differs from that used by C@. In
1462 that situation, define this macro to be a C expression to be used for
1463 the size of that type. If you don't define this, the default is the
1464 value of @code{LONG_TYPE_SIZE}.
1467 @defmac LONG_LONG_TYPE_SIZE
1468 A C expression for the size in bits of the type @code{long long} on the
1469 target machine. If you don't define this, the default is two
1470 words. If you want to support GNU Ada on your machine, the value of this
1471 macro must be at least 64.
1474 @defmac CHAR_TYPE_SIZE
1475 A C expression for the size in bits of the type @code{char} on the
1476 target machine. If you don't define this, the default is
1477 @code{BITS_PER_UNIT}.
1480 @defmac BOOL_TYPE_SIZE
1481 A C expression for the size in bits of the C++ type @code{bool} and
1482 C99 type @code{_Bool} on the target machine. If you don't define
1483 this, and you probably shouldn't, the default is @code{CHAR_TYPE_SIZE}.
1486 @defmac FLOAT_TYPE_SIZE
1487 A C expression for the size in bits of the type @code{float} on the
1488 target machine. If you don't define this, the default is one word.
1491 @defmac DOUBLE_TYPE_SIZE
1492 A C expression for the size in bits of the type @code{double} on the
1493 target machine. If you don't define this, the default is two
1497 @defmac LONG_DOUBLE_TYPE_SIZE
1498 A C expression for the size in bits of the type @code{long double} on
1499 the target machine. If you don't define this, the default is two
1503 @defmac SHORT_FRACT_TYPE_SIZE
1504 A C expression for the size in bits of the type @code{short _Fract} on
1505 the target machine. If you don't define this, the default is
1506 @code{BITS_PER_UNIT}.
1509 @defmac FRACT_TYPE_SIZE
1510 A C expression for the size in bits of the type @code{_Fract} on
1511 the target machine. If you don't define this, the default is
1512 @code{BITS_PER_UNIT * 2}.
1515 @defmac LONG_FRACT_TYPE_SIZE
1516 A C expression for the size in bits of the type @code{long _Fract} on
1517 the target machine. If you don't define this, the default is
1518 @code{BITS_PER_UNIT * 4}.
1521 @defmac LONG_LONG_FRACT_TYPE_SIZE
1522 A C expression for the size in bits of the type @code{long long _Fract} on
1523 the target machine. If you don't define this, the default is
1524 @code{BITS_PER_UNIT * 8}.
1527 @defmac SHORT_ACCUM_TYPE_SIZE
1528 A C expression for the size in bits of the type @code{short _Accum} on
1529 the target machine. If you don't define this, the default is
1530 @code{BITS_PER_UNIT * 2}.
1533 @defmac ACCUM_TYPE_SIZE
1534 A C expression for the size in bits of the type @code{_Accum} on
1535 the target machine. If you don't define this, the default is
1536 @code{BITS_PER_UNIT * 4}.
1539 @defmac LONG_ACCUM_TYPE_SIZE
1540 A C expression for the size in bits of the type @code{long _Accum} on
1541 the target machine. If you don't define this, the default is
1542 @code{BITS_PER_UNIT * 8}.
1545 @defmac LONG_LONG_ACCUM_TYPE_SIZE
1546 A C expression for the size in bits of the type @code{long long _Accum} on
1547 the target machine. If you don't define this, the default is
1548 @code{BITS_PER_UNIT * 16}.
1551 @defmac LIBGCC2_GNU_PREFIX
1552 This macro corresponds to the @code{TARGET_LIBFUNC_GNU_PREFIX} target
1553 hook and should be defined if that hook is overriden to be true. It
1554 causes function names in libgcc to be changed to use a @code{__gnu_}
1555 prefix for their name rather than the default @code{__}. A port which
1556 uses this macro should also arrange to use @file{t-gnu-prefix} in
1557 the libgcc @file{config.host}.
1560 @defmac TARGET_FLT_EVAL_METHOD
1561 A C expression for the value for @code{FLT_EVAL_METHOD} in @file{float.h},
1562 assuming, if applicable, that the floating-point control word is in its
1563 default state. If you do not define this macro the value of
1564 @code{FLT_EVAL_METHOD} will be zero.
1567 @defmac WIDEST_HARDWARE_FP_SIZE
1568 A C expression for the size in bits of the widest floating-point format
1569 supported by the hardware. If you define this macro, you must specify a
1570 value less than or equal to the value of @code{LONG_DOUBLE_TYPE_SIZE}.
1571 If you do not define this macro, the value of @code{LONG_DOUBLE_TYPE_SIZE}
1575 @defmac DEFAULT_SIGNED_CHAR
1576 An expression whose value is 1 or 0, according to whether the type
1577 @code{char} should be signed or unsigned by default. The user can
1578 always override this default with the options @option{-fsigned-char}
1579 and @option{-funsigned-char}.
1582 @deftypefn {Target Hook} bool TARGET_DEFAULT_SHORT_ENUMS (void)
1583 This target hook should return true if the compiler should give an
1584 @code{enum} type only as many bytes as it takes to represent the range
1585 of possible values of that type. It should return false if all
1586 @code{enum} types should be allocated like @code{int}.
1588 The default is to return false.
1592 A C expression for a string describing the name of the data type to use
1593 for size values. The typedef name @code{size_t} is defined using the
1594 contents of the string.
1596 The string can contain more than one keyword. If so, separate them with
1597 spaces, and write first any length keyword, then @code{unsigned} if
1598 appropriate, and finally @code{int}. The string must exactly match one
1599 of the data type names defined in the function
1600 @code{c_common_nodes_and_builtins} in the file @file{c-family/c-common.c}.
1601 You may not omit @code{int} or change the order---that would cause the
1602 compiler to crash on startup.
1604 If you don't define this macro, the default is @code{"long unsigned
1609 GCC defines internal types (@code{sizetype}, @code{ssizetype},
1610 @code{bitsizetype} and @code{sbitsizetype}) for expressions
1611 dealing with size. This macro is a C expression for a string describing
1612 the name of the data type from which the precision of @code{sizetype}
1615 The string has the same restrictions as @code{SIZE_TYPE} string.
1617 If you don't define this macro, the default is @code{SIZE_TYPE}.
1620 @defmac PTRDIFF_TYPE
1621 A C expression for a string describing the name of the data type to use
1622 for the result of subtracting two pointers. The typedef name
1623 @code{ptrdiff_t} is defined using the contents of the string. See
1624 @code{SIZE_TYPE} above for more information.
1626 If you don't define this macro, the default is @code{"long int"}.
1630 A C expression for a string describing the name of the data type to use
1631 for wide characters. The typedef name @code{wchar_t} is defined using
1632 the contents of the string. See @code{SIZE_TYPE} above for more
1635 If you don't define this macro, the default is @code{"int"}.
1638 @defmac WCHAR_TYPE_SIZE
1639 A C expression for the size in bits of the data type for wide
1640 characters. This is used in @code{cpp}, which cannot make use of
1645 A C expression for a string describing the name of the data type to
1646 use for wide characters passed to @code{printf} and returned from
1647 @code{getwc}. The typedef name @code{wint_t} is defined using the
1648 contents of the string. See @code{SIZE_TYPE} above for more
1651 If you don't define this macro, the default is @code{"unsigned int"}.
1655 A C expression for a string describing the name of the data type that
1656 can represent any value of any standard or extended signed integer type.
1657 The typedef name @code{intmax_t} is defined using the contents of the
1658 string. See @code{SIZE_TYPE} above for more information.
1660 If you don't define this macro, the default is the first of
1661 @code{"int"}, @code{"long int"}, or @code{"long long int"} that has as
1662 much precision as @code{long long int}.
1665 @defmac UINTMAX_TYPE
1666 A C expression for a string describing the name of the data type that
1667 can represent any value of any standard or extended unsigned integer
1668 type. The typedef name @code{uintmax_t} is defined using the contents
1669 of the string. See @code{SIZE_TYPE} above for more information.
1671 If you don't define this macro, the default is the first of
1672 @code{"unsigned int"}, @code{"long unsigned int"}, or @code{"long long
1673 unsigned int"} that has as much precision as @code{long long unsigned
1677 @defmac SIG_ATOMIC_TYPE
1683 @defmacx UINT16_TYPE
1684 @defmacx UINT32_TYPE
1685 @defmacx UINT64_TYPE
1686 @defmacx INT_LEAST8_TYPE
1687 @defmacx INT_LEAST16_TYPE
1688 @defmacx INT_LEAST32_TYPE
1689 @defmacx INT_LEAST64_TYPE
1690 @defmacx UINT_LEAST8_TYPE
1691 @defmacx UINT_LEAST16_TYPE
1692 @defmacx UINT_LEAST32_TYPE
1693 @defmacx UINT_LEAST64_TYPE
1694 @defmacx INT_FAST8_TYPE
1695 @defmacx INT_FAST16_TYPE
1696 @defmacx INT_FAST32_TYPE
1697 @defmacx INT_FAST64_TYPE
1698 @defmacx UINT_FAST8_TYPE
1699 @defmacx UINT_FAST16_TYPE
1700 @defmacx UINT_FAST32_TYPE
1701 @defmacx UINT_FAST64_TYPE
1702 @defmacx INTPTR_TYPE
1703 @defmacx UINTPTR_TYPE
1704 C expressions for the standard types @code{sig_atomic_t},
1705 @code{int8_t}, @code{int16_t}, @code{int32_t}, @code{int64_t},
1706 @code{uint8_t}, @code{uint16_t}, @code{uint32_t}, @code{uint64_t},
1707 @code{int_least8_t}, @code{int_least16_t}, @code{int_least32_t},
1708 @code{int_least64_t}, @code{uint_least8_t}, @code{uint_least16_t},
1709 @code{uint_least32_t}, @code{uint_least64_t}, @code{int_fast8_t},
1710 @code{int_fast16_t}, @code{int_fast32_t}, @code{int_fast64_t},
1711 @code{uint_fast8_t}, @code{uint_fast16_t}, @code{uint_fast32_t},
1712 @code{uint_fast64_t}, @code{intptr_t}, and @code{uintptr_t}. See
1713 @code{SIZE_TYPE} above for more information.
1715 If any of these macros evaluates to a null pointer, the corresponding
1716 type is not supported; if GCC is configured to provide
1717 @code{<stdint.h>} in such a case, the header provided may not conform
1718 to C99, depending on the type in question. The defaults for all of
1719 these macros are null pointers.
1722 @defmac TARGET_PTRMEMFUNC_VBIT_LOCATION
1723 The C++ compiler represents a pointer-to-member-function with a struct
1730 ptrdiff_t vtable_index;
1737 The C++ compiler must use one bit to indicate whether the function that
1738 will be called through a pointer-to-member-function is virtual.
1739 Normally, we assume that the low-order bit of a function pointer must
1740 always be zero. Then, by ensuring that the vtable_index is odd, we can
1741 distinguish which variant of the union is in use. But, on some
1742 platforms function pointers can be odd, and so this doesn't work. In
1743 that case, we use the low-order bit of the @code{delta} field, and shift
1744 the remainder of the @code{delta} field to the left.
1746 GCC will automatically make the right selection about where to store
1747 this bit using the @code{FUNCTION_BOUNDARY} setting for your platform.
1748 However, some platforms such as ARM/Thumb have @code{FUNCTION_BOUNDARY}
1749 set such that functions always start at even addresses, but the lowest
1750 bit of pointers to functions indicate whether the function at that
1751 address is in ARM or Thumb mode. If this is the case of your
1752 architecture, you should define this macro to
1753 @code{ptrmemfunc_vbit_in_delta}.
1755 In general, you should not have to define this macro. On architectures
1756 in which function addresses are always even, according to
1757 @code{FUNCTION_BOUNDARY}, GCC will automatically define this macro to
1758 @code{ptrmemfunc_vbit_in_pfn}.
1761 @defmac TARGET_VTABLE_USES_DESCRIPTORS
1762 Normally, the C++ compiler uses function pointers in vtables. This
1763 macro allows the target to change to use ``function descriptors''
1764 instead. Function descriptors are found on targets for whom a
1765 function pointer is actually a small data structure. Normally the
1766 data structure consists of the actual code address plus a data
1767 pointer to which the function's data is relative.
1769 If vtables are used, the value of this macro should be the number
1770 of words that the function descriptor occupies.
1773 @defmac TARGET_VTABLE_ENTRY_ALIGN
1774 By default, the vtable entries are void pointers, the so the alignment
1775 is the same as pointer alignment. The value of this macro specifies
1776 the alignment of the vtable entry in bits. It should be defined only
1777 when special alignment is necessary. */
1780 @defmac TARGET_VTABLE_DATA_ENTRY_DISTANCE
1781 There are a few non-descriptor entries in the vtable at offsets below
1782 zero. If these entries must be padded (say, to preserve the alignment
1783 specified by @code{TARGET_VTABLE_ENTRY_ALIGN}), set this to the number
1784 of words in each data entry.
1788 @section Register Usage
1789 @cindex register usage
1791 This section explains how to describe what registers the target machine
1792 has, and how (in general) they can be used.
1794 The description of which registers a specific instruction can use is
1795 done with register classes; see @ref{Register Classes}. For information
1796 on using registers to access a stack frame, see @ref{Frame Registers}.
1797 For passing values in registers, see @ref{Register Arguments}.
1798 For returning values in registers, see @ref{Scalar Return}.
1801 * Register Basics:: Number and kinds of registers.
1802 * Allocation Order:: Order in which registers are allocated.
1803 * Values in Registers:: What kinds of values each reg can hold.
1804 * Leaf Functions:: Renumbering registers for leaf functions.
1805 * Stack Registers:: Handling a register stack such as 80387.
1808 @node Register Basics
1809 @subsection Basic Characteristics of Registers
1811 @c prevent bad page break with this line
1812 Registers have various characteristics.
1814 @defmac FIRST_PSEUDO_REGISTER
1815 Number of hardware registers known to the compiler. They receive
1816 numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first
1817 pseudo register's number really is assigned the number
1818 @code{FIRST_PSEUDO_REGISTER}.
1821 @defmac FIXED_REGISTERS
1822 @cindex fixed register
1823 An initializer that says which registers are used for fixed purposes
1824 all throughout the compiled code and are therefore not available for
1825 general allocation. These would include the stack pointer, the frame
1826 pointer (except on machines where that can be used as a general
1827 register when no frame pointer is needed), the program counter on
1828 machines where that is considered one of the addressable registers,
1829 and any other numbered register with a standard use.
1831 This information is expressed as a sequence of numbers, separated by
1832 commas and surrounded by braces. The @var{n}th number is 1 if
1833 register @var{n} is fixed, 0 otherwise.
1835 The table initialized from this macro, and the table initialized by
1836 the following one, may be overridden at run time either automatically,
1837 by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by
1838 the user with the command options @option{-ffixed-@var{reg}},
1839 @option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}.
1842 @defmac CALL_USED_REGISTERS
1843 @cindex call-used register
1844 @cindex call-clobbered register
1845 @cindex call-saved register
1846 Like @code{FIXED_REGISTERS} but has 1 for each register that is
1847 clobbered (in general) by function calls as well as for fixed
1848 registers. This macro therefore identifies the registers that are not
1849 available for general allocation of values that must live across
1852 If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler
1853 automatically saves it on function entry and restores it on function
1854 exit, if the register is used within the function.
1857 @defmac CALL_REALLY_USED_REGISTERS
1858 @cindex call-used register
1859 @cindex call-clobbered register
1860 @cindex call-saved register
1861 Like @code{CALL_USED_REGISTERS} except this macro doesn't require
1862 that the entire set of @code{FIXED_REGISTERS} be included.
1863 (@code{CALL_USED_REGISTERS} must be a superset of @code{FIXED_REGISTERS}).
1864 This macro is optional. If not specified, it defaults to the value
1865 of @code{CALL_USED_REGISTERS}.
1868 @defmac HARD_REGNO_CALL_PART_CLOBBERED (@var{regno}, @var{mode})
1869 @cindex call-used register
1870 @cindex call-clobbered register
1871 @cindex call-saved register
1872 A C expression that is nonzero if it is not permissible to store a
1873 value of mode @var{mode} in hard register number @var{regno} across a
1874 call without some part of it being clobbered. For most machines this
1875 macro need not be defined. It is only required for machines that do not
1876 preserve the entire contents of a register across a call.
1880 @findex call_used_regs
1883 @findex reg_class_contents
1884 @deftypefn {Target Hook} void TARGET_CONDITIONAL_REGISTER_USAGE (void)
1885 This hook may conditionally modify five variables
1886 @code{fixed_regs}, @code{call_used_regs}, @code{global_regs},
1887 @code{reg_names}, and @code{reg_class_contents}, to take into account
1888 any dependence of these register sets on target flags. The first three
1889 of these are of type @code{char []} (interpreted as Boolean vectors).
1890 @code{global_regs} is a @code{const char *[]}, and
1891 @code{reg_class_contents} is a @code{HARD_REG_SET}. Before the macro is
1892 called, @code{fixed_regs}, @code{call_used_regs},
1893 @code{reg_class_contents}, and @code{reg_names} have been initialized
1894 from @code{FIXED_REGISTERS}, @code{CALL_USED_REGISTERS},
1895 @code{REG_CLASS_CONTENTS}, and @code{REGISTER_NAMES}, respectively.
1896 @code{global_regs} has been cleared, and any @option{-ffixed-@var{reg}},
1897 @option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}
1898 command options have been applied.
1900 @cindex disabling certain registers
1901 @cindex controlling register usage
1902 If the usage of an entire class of registers depends on the target
1903 flags, you may indicate this to GCC by using this macro to modify
1904 @code{fixed_regs} and @code{call_used_regs} to 1 for each of the
1905 registers in the classes which should not be used by GCC@. Also make
1906 @code{define_register_constraint}s return @code{NO_REGS} for constraints
1907 that shouldn't be used.
1909 (However, if this class is not included in @code{GENERAL_REGS} and all
1910 of the insn patterns whose constraints permit this class are
1911 controlled by target switches, then GCC will automatically avoid using
1912 these registers when the target switches are opposed to them.)
1915 @defmac INCOMING_REGNO (@var{out})
1916 Define this macro if the target machine has register windows. This C
1917 expression returns the register number as seen by the called function
1918 corresponding to the register number @var{out} as seen by the calling
1919 function. Return @var{out} if register number @var{out} is not an
1923 @defmac OUTGOING_REGNO (@var{in})
1924 Define this macro if the target machine has register windows. This C
1925 expression returns the register number as seen by the calling function
1926 corresponding to the register number @var{in} as seen by the called
1927 function. Return @var{in} if register number @var{in} is not an inbound
1931 @defmac LOCAL_REGNO (@var{regno})
1932 Define this macro if the target machine has register windows. This C
1933 expression returns true if the register is call-saved but is in the
1934 register window. Unlike most call-saved registers, such registers
1935 need not be explicitly restored on function exit or during non-local
1940 If the program counter has a register number, define this as that
1941 register number. Otherwise, do not define it.
1944 @node Allocation Order
1945 @subsection Order of Allocation of Registers
1946 @cindex order of register allocation
1947 @cindex register allocation order
1949 @c prevent bad page break with this line
1950 Registers are allocated in order.
1952 @defmac REG_ALLOC_ORDER
1953 If defined, an initializer for a vector of integers, containing the
1954 numbers of hard registers in the order in which GCC should prefer
1955 to use them (from most preferred to least).
1957 If this macro is not defined, registers are used lowest numbered first
1958 (all else being equal).
1960 One use of this macro is on machines where the highest numbered
1961 registers must always be saved and the save-multiple-registers
1962 instruction supports only sequences of consecutive registers. On such
1963 machines, define @code{REG_ALLOC_ORDER} to be an initializer that lists
1964 the highest numbered allocable register first.
1967 @defmac ADJUST_REG_ALLOC_ORDER
1968 A C statement (sans semicolon) to choose the order in which to allocate
1969 hard registers for pseudo-registers local to a basic block.
1971 Store the desired register order in the array @code{reg_alloc_order}.
1972 Element 0 should be the register to allocate first; element 1, the next
1973 register; and so on.
1975 The macro body should not assume anything about the contents of
1976 @code{reg_alloc_order} before execution of the macro.
1978 On most machines, it is not necessary to define this macro.
1981 @defmac HONOR_REG_ALLOC_ORDER
1982 Normally, IRA tries to estimate the costs for saving a register in the
1983 prologue and restoring it in the epilogue. This discourages it from
1984 using call-saved registers. If a machine wants to ensure that IRA
1985 allocates registers in the order given by REG_ALLOC_ORDER even if some
1986 call-saved registers appear earlier than call-used ones, then define this
1987 macro as a C expression to nonzero. Default is 0.
1990 @defmac IRA_HARD_REGNO_ADD_COST_MULTIPLIER (@var{regno})
1991 In some case register allocation order is not enough for the
1992 Integrated Register Allocator (@acronym{IRA}) to generate a good code.
1993 If this macro is defined, it should return a floating point value
1994 based on @var{regno}. The cost of using @var{regno} for a pseudo will
1995 be increased by approximately the pseudo's usage frequency times the
1996 value returned by this macro. Not defining this macro is equivalent
1997 to having it always return @code{0.0}.
1999 On most machines, it is not necessary to define this macro.
2002 @node Values in Registers
2003 @subsection How Values Fit in Registers
2005 This section discusses the macros that describe which kinds of values
2006 (specifically, which machine modes) each register can hold, and how many
2007 consecutive registers are needed for a given mode.
2009 @defmac HARD_REGNO_NREGS (@var{regno}, @var{mode})
2010 A C expression for the number of consecutive hard registers, starting
2011 at register number @var{regno}, required to hold a value of mode
2012 @var{mode}. This macro must never return zero, even if a register
2013 cannot hold the requested mode - indicate that with HARD_REGNO_MODE_OK
2014 and/or CANNOT_CHANGE_MODE_CLASS instead.
2016 On a machine where all registers are exactly one word, a suitable
2017 definition of this macro is
2020 #define HARD_REGNO_NREGS(REGNO, MODE) \
2021 ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) \
2026 @defmac HARD_REGNO_NREGS_HAS_PADDING (@var{regno}, @var{mode})
2027 A C expression that is nonzero if a value of mode @var{mode}, stored
2028 in memory, ends with padding that causes it to take up more space than
2029 in registers starting at register number @var{regno} (as determined by
2030 multiplying GCC's notion of the size of the register when containing
2031 this mode by the number of registers returned by
2032 @code{HARD_REGNO_NREGS}). By default this is zero.
2034 For example, if a floating-point value is stored in three 32-bit
2035 registers but takes up 128 bits in memory, then this would be
2038 This macros only needs to be defined if there are cases where
2039 @code{subreg_get_info}
2040 would otherwise wrongly determine that a @code{subreg} can be
2041 represented by an offset to the register number, when in fact such a
2042 @code{subreg} would contain some of the padding not stored in
2043 registers and so not be representable.
2046 @defmac HARD_REGNO_NREGS_WITH_PADDING (@var{regno}, @var{mode})
2047 For values of @var{regno} and @var{mode} for which
2048 @code{HARD_REGNO_NREGS_HAS_PADDING} returns nonzero, a C expression
2049 returning the greater number of registers required to hold the value
2050 including any padding. In the example above, the value would be four.
2053 @defmac REGMODE_NATURAL_SIZE (@var{mode})
2054 Define this macro if the natural size of registers that hold values
2055 of mode @var{mode} is not the word size. It is a C expression that
2056 should give the natural size in bytes for the specified mode. It is
2057 used by the register allocator to try to optimize its results. This
2058 happens for example on SPARC 64-bit where the natural size of
2059 floating-point registers is still 32-bit.
2062 @defmac HARD_REGNO_MODE_OK (@var{regno}, @var{mode})
2063 A C expression that is nonzero if it is permissible to store a value
2064 of mode @var{mode} in hard register number @var{regno} (or in several
2065 registers starting with that one). For a machine where all registers
2066 are equivalent, a suitable definition is
2069 #define HARD_REGNO_MODE_OK(REGNO, MODE) 1
2072 You need not include code to check for the numbers of fixed registers,
2073 because the allocation mechanism considers them to be always occupied.
2075 @cindex register pairs
2076 On some machines, double-precision values must be kept in even/odd
2077 register pairs. You can implement that by defining this macro to reject
2078 odd register numbers for such modes.
2080 The minimum requirement for a mode to be OK in a register is that the
2081 @samp{mov@var{mode}} instruction pattern support moves between the
2082 register and other hard register in the same class and that moving a
2083 value into the register and back out not alter it.
2085 Since the same instruction used to move @code{word_mode} will work for
2086 all narrower integer modes, it is not necessary on any machine for
2087 @code{HARD_REGNO_MODE_OK} to distinguish between these modes, provided
2088 you define patterns @samp{movhi}, etc., to take advantage of this. This
2089 is useful because of the interaction between @code{HARD_REGNO_MODE_OK}
2090 and @code{MODES_TIEABLE_P}; it is very desirable for all integer modes
2093 Many machines have special registers for floating point arithmetic.
2094 Often people assume that floating point machine modes are allowed only
2095 in floating point registers. This is not true. Any registers that
2096 can hold integers can safely @emph{hold} a floating point machine
2097 mode, whether or not floating arithmetic can be done on it in those
2098 registers. Integer move instructions can be used to move the values.
2100 On some machines, though, the converse is true: fixed-point machine
2101 modes may not go in floating registers. This is true if the floating
2102 registers normalize any value stored in them, because storing a
2103 non-floating value there would garble it. In this case,
2104 @code{HARD_REGNO_MODE_OK} should reject fixed-point machine modes in
2105 floating registers. But if the floating registers do not automatically
2106 normalize, if you can store any bit pattern in one and retrieve it
2107 unchanged without a trap, then any machine mode may go in a floating
2108 register, so you can define this macro to say so.
2110 The primary significance of special floating registers is rather that
2111 they are the registers acceptable in floating point arithmetic
2112 instructions. However, this is of no concern to
2113 @code{HARD_REGNO_MODE_OK}. You handle it by writing the proper
2114 constraints for those instructions.
2116 On some machines, the floating registers are especially slow to access,
2117 so that it is better to store a value in a stack frame than in such a
2118 register if floating point arithmetic is not being done. As long as the
2119 floating registers are not in class @code{GENERAL_REGS}, they will not
2120 be used unless some pattern's constraint asks for one.
2123 @defmac HARD_REGNO_RENAME_OK (@var{from}, @var{to})
2124 A C expression that is nonzero if it is OK to rename a hard register
2125 @var{from} to another hard register @var{to}.
2127 One common use of this macro is to prevent renaming of a register to
2128 another register that is not saved by a prologue in an interrupt
2131 The default is always nonzero.
2134 @defmac MODES_TIEABLE_P (@var{mode1}, @var{mode2})
2135 A C expression that is nonzero if a value of mode
2136 @var{mode1} is accessible in mode @var{mode2} without copying.
2138 If @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and
2139 @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are always the same for
2140 any @var{r}, then @code{MODES_TIEABLE_P (@var{mode1}, @var{mode2})}
2141 should be nonzero. If they differ for any @var{r}, you should define
2142 this macro to return zero unless some other mechanism ensures the
2143 accessibility of the value in a narrower mode.
2145 You should define this macro to return nonzero in as many cases as
2146 possible since doing so will allow GCC to perform better register
2150 @deftypefn {Target Hook} bool TARGET_HARD_REGNO_SCRATCH_OK (unsigned int @var{regno})
2151 This target hook should return @code{true} if it is OK to use a hard register
2152 @var{regno} as scratch reg in peephole2.
2154 One common use of this macro is to prevent using of a register that
2155 is not saved by a prologue in an interrupt handler.
2157 The default version of this hook always returns @code{true}.
2160 @defmac AVOID_CCMODE_COPIES
2161 Define this macro if the compiler should avoid copies to/from @code{CCmode}
2162 registers. You should only define this macro if support for copying to/from
2163 @code{CCmode} is incomplete.
2166 @node Leaf Functions
2167 @subsection Handling Leaf Functions
2169 @cindex leaf functions
2170 @cindex functions, leaf
2171 On some machines, a leaf function (i.e., one which makes no calls) can run
2172 more efficiently if it does not make its own register window. Often this
2173 means it is required to receive its arguments in the registers where they
2174 are passed by the caller, instead of the registers where they would
2177 The special treatment for leaf functions generally applies only when
2178 other conditions are met; for example, often they may use only those
2179 registers for its own variables and temporaries. We use the term ``leaf
2180 function'' to mean a function that is suitable for this special
2181 handling, so that functions with no calls are not necessarily ``leaf
2184 GCC assigns register numbers before it knows whether the function is
2185 suitable for leaf function treatment. So it needs to renumber the
2186 registers in order to output a leaf function. The following macros
2189 @defmac LEAF_REGISTERS
2190 Name of a char vector, indexed by hard register number, which
2191 contains 1 for a register that is allowable in a candidate for leaf
2194 If leaf function treatment involves renumbering the registers, then the
2195 registers marked here should be the ones before renumbering---those that
2196 GCC would ordinarily allocate. The registers which will actually be
2197 used in the assembler code, after renumbering, should not be marked with 1
2200 Define this macro only if the target machine offers a way to optimize
2201 the treatment of leaf functions.
2204 @defmac LEAF_REG_REMAP (@var{regno})
2205 A C expression whose value is the register number to which @var{regno}
2206 should be renumbered, when a function is treated as a leaf function.
2208 If @var{regno} is a register number which should not appear in a leaf
2209 function before renumbering, then the expression should yield @minus{}1, which
2210 will cause the compiler to abort.
2212 Define this macro only if the target machine offers a way to optimize the
2213 treatment of leaf functions, and registers need to be renumbered to do
2217 @findex current_function_is_leaf
2218 @findex current_function_uses_only_leaf_regs
2219 @code{TARGET_ASM_FUNCTION_PROLOGUE} and
2220 @code{TARGET_ASM_FUNCTION_EPILOGUE} must usually treat leaf functions
2221 specially. They can test the C variable @code{current_function_is_leaf}
2222 which is nonzero for leaf functions. @code{current_function_is_leaf} is
2223 set prior to local register allocation and is valid for the remaining
2224 compiler passes. They can also test the C variable
2225 @code{current_function_uses_only_leaf_regs} which is nonzero for leaf
2226 functions which only use leaf registers.
2227 @code{current_function_uses_only_leaf_regs} is valid after all passes
2228 that modify the instructions have been run and is only useful if
2229 @code{LEAF_REGISTERS} is defined.
2230 @c changed this to fix overfull. ALSO: why the "it" at the beginning
2231 @c of the next paragraph?! --mew 2feb93
2233 @node Stack Registers
2234 @subsection Registers That Form a Stack
2236 There are special features to handle computers where some of the
2237 ``registers'' form a stack. Stack registers are normally written by
2238 pushing onto the stack, and are numbered relative to the top of the
2241 Currently, GCC can only handle one group of stack-like registers, and
2242 they must be consecutively numbered. Furthermore, the existing
2243 support for stack-like registers is specific to the 80387 floating
2244 point coprocessor. If you have a new architecture that uses
2245 stack-like registers, you will need to do substantial work on
2246 @file{reg-stack.c} and write your machine description to cooperate
2247 with it, as well as defining these macros.
2250 Define this if the machine has any stack-like registers.
2253 @defmac STACK_REG_COVER_CLASS
2254 This is a cover class containing the stack registers. Define this if
2255 the machine has any stack-like registers.
2258 @defmac FIRST_STACK_REG
2259 The number of the first stack-like register. This one is the top
2263 @defmac LAST_STACK_REG
2264 The number of the last stack-like register. This one is the bottom of
2268 @node Register Classes
2269 @section Register Classes
2270 @cindex register class definitions
2271 @cindex class definitions, register
2273 On many machines, the numbered registers are not all equivalent.
2274 For example, certain registers may not be allowed for indexed addressing;
2275 certain registers may not be allowed in some instructions. These machine
2276 restrictions are described to the compiler using @dfn{register classes}.
2278 You define a number of register classes, giving each one a name and saying
2279 which of the registers belong to it. Then you can specify register classes
2280 that are allowed as operands to particular instruction patterns.
2284 In general, each register will belong to several classes. In fact, one
2285 class must be named @code{ALL_REGS} and contain all the registers. Another
2286 class must be named @code{NO_REGS} and contain no registers. Often the
2287 union of two classes will be another class; however, this is not required.
2289 @findex GENERAL_REGS
2290 One of the classes must be named @code{GENERAL_REGS}. There is nothing
2291 terribly special about the name, but the operand constraint letters
2292 @samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is
2293 the same as @code{ALL_REGS}, just define it as a macro which expands
2296 Order the classes so that if class @var{x} is contained in class @var{y}
2297 then @var{x} has a lower class number than @var{y}.
2299 The way classes other than @code{GENERAL_REGS} are specified in operand
2300 constraints is through machine-dependent operand constraint letters.
2301 You can define such letters to correspond to various classes, then use
2302 them in operand constraints.
2304 You must define the narrowest register classes for allocatable
2305 registers, so that each class either has no subclasses, or that for
2306 some mode, the move cost between registers within the class is
2307 cheaper than moving a register in the class to or from memory
2310 You should define a class for the union of two classes whenever some
2311 instruction allows both classes. For example, if an instruction allows
2312 either a floating point (coprocessor) register or a general register for a
2313 certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS}
2314 which includes both of them. Otherwise you will get suboptimal code,
2315 or even internal compiler errors when reload cannot find a register in the
2316 class computed via @code{reg_class_subunion}.
2318 You must also specify certain redundant information about the register
2319 classes: for each class, which classes contain it and which ones are
2320 contained in it; for each pair of classes, the largest class contained
2323 When a value occupying several consecutive registers is expected in a
2324 certain class, all the registers used must belong to that class.
2325 Therefore, register classes cannot be used to enforce a requirement for
2326 a register pair to start with an even-numbered register. The way to
2327 specify this requirement is with @code{HARD_REGNO_MODE_OK}.
2329 Register classes used for input-operands of bitwise-and or shift
2330 instructions have a special requirement: each such class must have, for
2331 each fixed-point machine mode, a subclass whose registers can transfer that
2332 mode to or from memory. For example, on some machines, the operations for
2333 single-byte values (@code{QImode}) are limited to certain registers. When
2334 this is so, each register class that is used in a bitwise-and or shift
2335 instruction must have a subclass consisting of registers from which
2336 single-byte values can be loaded or stored. This is so that
2337 @code{PREFERRED_RELOAD_CLASS} can always have a possible value to return.
2339 @deftp {Data type} {enum reg_class}
2340 An enumerated type that must be defined with all the register class names
2341 as enumerated values. @code{NO_REGS} must be first. @code{ALL_REGS}
2342 must be the last register class, followed by one more enumerated value,
2343 @code{LIM_REG_CLASSES}, which is not a register class but rather
2344 tells how many classes there are.
2346 Each register class has a number, which is the value of casting
2347 the class name to type @code{int}. The number serves as an index
2348 in many of the tables described below.
2351 @defmac N_REG_CLASSES
2352 The number of distinct register classes, defined as follows:
2355 #define N_REG_CLASSES (int) LIM_REG_CLASSES
2359 @defmac REG_CLASS_NAMES
2360 An initializer containing the names of the register classes as C string
2361 constants. These names are used in writing some of the debugging dumps.
2364 @defmac REG_CLASS_CONTENTS
2365 An initializer containing the contents of the register classes, as integers
2366 which are bit masks. The @var{n}th integer specifies the contents of class
2367 @var{n}. The way the integer @var{mask} is interpreted is that
2368 register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1.
2370 When the machine has more than 32 registers, an integer does not suffice.
2371 Then the integers are replaced by sub-initializers, braced groupings containing
2372 several integers. Each sub-initializer must be suitable as an initializer
2373 for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}.
2374 In this situation, the first integer in each sub-initializer corresponds to
2375 registers 0 through 31, the second integer to registers 32 through 63, and
2379 @defmac REGNO_REG_CLASS (@var{regno})
2380 A C expression whose value is a register class containing hard register
2381 @var{regno}. In general there is more than one such class; choose a class
2382 which is @dfn{minimal}, meaning that no smaller class also contains the
2386 @defmac BASE_REG_CLASS
2387 A macro whose definition is the name of the class to which a valid
2388 base register must belong. A base register is one used in an address
2389 which is the register value plus a displacement.
2392 @defmac MODE_BASE_REG_CLASS (@var{mode})
2393 This is a variation of the @code{BASE_REG_CLASS} macro which allows
2394 the selection of a base register in a mode dependent manner. If
2395 @var{mode} is VOIDmode then it should return the same value as
2396 @code{BASE_REG_CLASS}.
2399 @defmac MODE_BASE_REG_REG_CLASS (@var{mode})
2400 A C expression whose value is the register class to which a valid
2401 base register must belong in order to be used in a base plus index
2402 register address. You should define this macro if base plus index
2403 addresses have different requirements than other base register uses.
2406 @defmac MODE_CODE_BASE_REG_CLASS (@var{mode}, @var{address_space}, @var{outer_code}, @var{index_code})
2407 A C expression whose value is the register class to which a valid
2408 base register for a memory reference in mode @var{mode} to address
2409 space @var{address_space} must belong. @var{outer_code} and @var{index_code}
2410 define the context in which the base register occurs. @var{outer_code} is
2411 the code of the immediately enclosing expression (@code{MEM} for the top level
2412 of an 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{address_space}, @var{outer_code}, @var{index_code})
2453 A C expression which is nonzero if register number @var{num} is
2454 suitable for use as a base register in operand addresses, accessing
2455 memory in mode @var{mode} in address space @var{address_space}.
2456 This is similar to @code{REGNO_MODE_OK_FOR_BASE_P}, except
2457 that that expression may examine the context in which the register
2458 appears in the memory reference. @var{outer_code} is the code of the
2459 immediately enclosing expression (@code{MEM} if at the top level of the
2460 address, @code{ADDRESS} for something that occurs in an
2461 @code{address_operand}). @var{index_code} is the code of the
2462 corresponding index expression if @var{outer_code} is @code{PLUS};
2463 @code{SCRATCH} otherwise. The mode may be @code{VOIDmode} for addresses
2464 that appear outside a @code{MEM}, i.e., as an @code{address_operand}.
2467 @defmac REGNO_OK_FOR_INDEX_P (@var{num})
2468 A C expression which is nonzero if register number @var{num} is
2469 suitable for use as an index register in operand addresses. It may be
2470 either a suitable hard register or a pseudo register that has been
2471 allocated such a hard register.
2473 The difference between an index register and a base register is that
2474 the index register may be scaled. If an address involves the sum of
2475 two registers, neither one of them scaled, then either one may be
2476 labeled the ``base'' and the other the ``index''; but whichever
2477 labeling is used must fit the machine's constraints of which registers
2478 may serve in each capacity. The compiler will try both labelings,
2479 looking for one that is valid, and will reload one or both registers
2480 only if neither labeling works.
2483 @deftypefn {Target Hook} reg_class_t TARGET_PREFERRED_RENAME_CLASS (reg_class_t @var{rclass})
2484 A target hook that places additional preference on the register class to use when it is necessary to rename a register in class @var{rclass} to another class, or perhaps @var{NO_REGS}, if no preferred register class is found or hook @code{preferred_rename_class} is not implemented. Sometimes returning a more restrictive class makes better code. For example, on ARM, thumb-2 instructions using @code{LO_REGS} may be smaller than instructions using @code{GENERIC_REGS}. By returning @code{LO_REGS} from @code{preferred_rename_class}, code size can be reduced.
2487 @deftypefn {Target Hook} reg_class_t TARGET_PREFERRED_RELOAD_CLASS (rtx @var{x}, reg_class_t @var{rclass})
2488 A target hook that places additional restrictions on the register class
2489 to use when it is necessary to copy value @var{x} into a register in class
2490 @var{rclass}. The value is a register class; perhaps @var{rclass}, or perhaps
2491 another, smaller class.
2493 The default version of this hook always returns value of @code{rclass} argument.
2495 Sometimes returning a more restrictive class makes better code. For
2496 example, on the 68000, when @var{x} is an integer constant that is in range
2497 for a @samp{moveq} instruction, the value of this macro is always
2498 @code{DATA_REGS} as long as @var{rclass} includes the data registers.
2499 Requiring a data register guarantees that a @samp{moveq} will be used.
2501 One case where @code{TARGET_PREFERRED_RELOAD_CLASS} must not return
2502 @var{rclass} is if @var{x} is a legitimate constant which cannot be
2503 loaded into some register class. By returning @code{NO_REGS} you can
2504 force @var{x} into a memory location. For example, rs6000 can load
2505 immediate values into general-purpose registers, but does not have an
2506 instruction for loading an immediate value into a floating-point
2507 register, so @code{TARGET_PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when
2508 @var{x} is a floating-point constant. If the constant can't be loaded
2509 into any kind of register, code generation will be better if
2510 @code{TARGET_LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead
2511 of using @code{TARGET_PREFERRED_RELOAD_CLASS}.
2513 If an insn has pseudos in it after register allocation, reload will go
2514 through the alternatives and call repeatedly @code{TARGET_PREFERRED_RELOAD_CLASS}
2515 to find the best one. Returning @code{NO_REGS}, in this case, makes
2516 reload add a @code{!} in front of the constraint: the x86 back-end uses
2517 this feature to discourage usage of 387 registers when math is done in
2518 the SSE registers (and vice versa).
2521 @defmac PREFERRED_RELOAD_CLASS (@var{x}, @var{class})
2522 A C expression that places additional restrictions on the register class
2523 to use when it is necessary to copy value @var{x} into a register in class
2524 @var{class}. The value is a register class; perhaps @var{class}, or perhaps
2525 another, smaller class. On many machines, the following definition is
2529 #define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS
2532 Sometimes returning a more restrictive class makes better code. For
2533 example, on the 68000, when @var{x} is an integer constant that is in range
2534 for a @samp{moveq} instruction, the value of this macro is always
2535 @code{DATA_REGS} as long as @var{class} includes the data registers.
2536 Requiring a data register guarantees that a @samp{moveq} will be used.
2538 One case where @code{PREFERRED_RELOAD_CLASS} must not return
2539 @var{class} is if @var{x} is a legitimate constant which cannot be
2540 loaded into some register class. By returning @code{NO_REGS} you can
2541 force @var{x} into a memory location. For example, rs6000 can load
2542 immediate values into general-purpose registers, but does not have an
2543 instruction for loading an immediate value into a floating-point
2544 register, so @code{PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when
2545 @var{x} is a floating-point constant. If the constant can't be loaded
2546 into any kind of register, code generation will be better if
2547 @code{TARGET_LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead
2548 of using @code{TARGET_PREFERRED_RELOAD_CLASS}.
2550 If an insn has pseudos in it after register allocation, reload will go
2551 through the alternatives and call repeatedly @code{PREFERRED_RELOAD_CLASS}
2552 to find the best one. Returning @code{NO_REGS}, in this case, makes
2553 reload add a @code{!} in front of the constraint: the x86 back-end uses
2554 this feature to discourage usage of 387 registers when math is done in
2555 the SSE registers (and vice versa).
2558 @deftypefn {Target Hook} reg_class_t TARGET_PREFERRED_OUTPUT_RELOAD_CLASS (rtx @var{x}, reg_class_t @var{rclass})
2559 Like @code{TARGET_PREFERRED_RELOAD_CLASS}, but for output reloads instead of
2562 The default version of this hook always returns value of @code{rclass}
2565 You can also use @code{TARGET_PREFERRED_OUTPUT_RELOAD_CLASS} to discourage
2566 reload from using some alternatives, like @code{TARGET_PREFERRED_RELOAD_CLASS}.
2569 @defmac LIMIT_RELOAD_CLASS (@var{mode}, @var{class})
2570 A C expression that places additional restrictions on the register class
2571 to use when it is necessary to be able to hold a value of mode
2572 @var{mode} in a reload register for which class @var{class} would
2575 Unlike @code{PREFERRED_RELOAD_CLASS}, this macro should be used when
2576 there are certain modes that simply can't go in certain reload classes.
2578 The value is a register class; perhaps @var{class}, or perhaps another,
2581 Don't define this macro unless the target machine has limitations which
2582 require the macro to do something nontrivial.
2585 @deftypefn {Target Hook} reg_class_t TARGET_SECONDARY_RELOAD (bool @var{in_p}, rtx @var{x}, reg_class_t @var{reload_class}, machine_mode @var{reload_mode}, secondary_reload_info *@var{sri})
2586 Many machines have some registers that cannot be copied directly to or
2587 from memory or even from other types of registers. An example is the
2588 @samp{MQ} register, which on most machines, can only be copied to or
2589 from general registers, but not memory. Below, we shall be using the
2590 term 'intermediate register' when a move operation cannot be performed
2591 directly, but has to be done by copying the source into the intermediate
2592 register first, and then copying the intermediate register to the
2593 destination. An intermediate register always has the same mode as
2594 source and destination. Since it holds the actual value being copied,
2595 reload might apply optimizations to re-use an intermediate register
2596 and eliding the copy from the source when it can determine that the
2597 intermediate register still holds the required value.
2599 Another kind of secondary reload is required on some machines which
2600 allow copying all registers to and from memory, but require a scratch
2601 register for stores to some memory locations (e.g., those with symbolic
2602 address on the RT, and those with certain symbolic address on the SPARC
2603 when compiling PIC)@. Scratch registers need not have the same mode
2604 as the value being copied, and usually hold a different value than
2605 that being copied. Special patterns in the md file are needed to
2606 describe how the copy is performed with the help of the scratch register;
2607 these patterns also describe the number, register class(es) and mode(s)
2608 of the scratch register(s).
2610 In some cases, both an intermediate and a scratch register are required.
2612 For input reloads, this target hook is called with nonzero @var{in_p},
2613 and @var{x} is an rtx that needs to be copied to a register of class
2614 @var{reload_class} in @var{reload_mode}. For output reloads, this target
2615 hook is called with zero @var{in_p}, and a register of class @var{reload_class}
2616 needs to be copied to rtx @var{x} in @var{reload_mode}.
2618 If copying a register of @var{reload_class} from/to @var{x} requires
2619 an intermediate register, the hook @code{secondary_reload} should
2620 return the register class required for this intermediate register.
2621 If no intermediate register is required, it should return NO_REGS.
2622 If more than one intermediate register is required, describe the one
2623 that is closest in the copy chain to the reload register.
2625 If scratch registers are needed, you also have to describe how to
2626 perform the copy from/to the reload register to/from this
2627 closest intermediate register. Or if no intermediate register is
2628 required, but still a scratch register is needed, describe the
2629 copy from/to the reload register to/from the reload operand @var{x}.
2631 You do this by setting @code{sri->icode} to the instruction code of a pattern
2632 in the md file which performs the move. Operands 0 and 1 are the output
2633 and input of this copy, respectively. Operands from operand 2 onward are
2634 for scratch operands. These scratch operands must have a mode, and a
2635 single-register-class
2636 @c [later: or memory]
2639 When an intermediate register is used, the @code{secondary_reload}
2640 hook will be called again to determine how to copy the intermediate
2641 register to/from the reload operand @var{x}, so your hook must also
2642 have code to handle the register class of the intermediate operand.
2644 @c [For later: maybe we'll allow multi-alternative reload patterns -
2645 @c the port maintainer could name a mov<mode> pattern that has clobbers -
2646 @c and match the constraints of input and output to determine the required
2647 @c alternative. A restriction would be that constraints used to match
2648 @c against reloads registers would have to be written as register class
2649 @c constraints, or we need a new target macro / hook that tells us if an
2650 @c arbitrary constraint can match an unknown register of a given class.
2651 @c Such a macro / hook would also be useful in other places.]
2654 @var{x} might be a pseudo-register or a @code{subreg} of a
2655 pseudo-register, which could either be in a hard register or in memory.
2656 Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is
2657 in memory and the hard register number if it is in a register.
2659 Scratch operands in memory (constraint @code{"=m"} / @code{"=&m"}) are
2660 currently not supported. For the time being, you will have to continue
2661 to use @code{SECONDARY_MEMORY_NEEDED} for that purpose.
2663 @code{copy_cost} also uses this target hook to find out how values are
2664 copied. If you want it to include some extra cost for the need to allocate
2665 (a) scratch register(s), set @code{sri->extra_cost} to the additional cost.
2666 Or if two dependent moves are supposed to have a lower cost than the sum
2667 of the individual moves due to expected fortuitous scheduling and/or special
2668 forwarding logic, you can set @code{sri->extra_cost} to a negative amount.
2671 @defmac SECONDARY_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2672 @defmacx SECONDARY_INPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2673 @defmacx SECONDARY_OUTPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2674 These macros are obsolete, new ports should use the target hook
2675 @code{TARGET_SECONDARY_RELOAD} instead.
2677 These are obsolete macros, replaced by the @code{TARGET_SECONDARY_RELOAD}
2678 target hook. Older ports still define these macros to indicate to the
2679 reload phase that it may
2680 need to allocate at least one register for a reload in addition to the
2681 register to contain the data. Specifically, if copying @var{x} to a
2682 register @var{class} in @var{mode} requires an intermediate register,
2683 you were supposed to define @code{SECONDARY_INPUT_RELOAD_CLASS} to return the
2684 largest register class all of whose registers can be used as
2685 intermediate registers or scratch registers.
2687 If copying a register @var{class} in @var{mode} to @var{x} requires an
2688 intermediate or scratch register, @code{SECONDARY_OUTPUT_RELOAD_CLASS}
2689 was supposed to be defined be defined to return the largest register
2690 class required. If the
2691 requirements for input and output reloads were the same, the macro
2692 @code{SECONDARY_RELOAD_CLASS} should have been used instead of defining both
2695 The values returned by these macros are often @code{GENERAL_REGS}.
2696 Return @code{NO_REGS} if no spare register is needed; i.e., if @var{x}
2697 can be directly copied to or from a register of @var{class} in
2698 @var{mode} without requiring a scratch register. Do not define this
2699 macro if it would always return @code{NO_REGS}.
2701 If a scratch register is required (either with or without an
2702 intermediate register), you were supposed to define patterns for
2703 @samp{reload_in@var{m}} or @samp{reload_out@var{m}}, as required
2704 (@pxref{Standard Names}. These patterns, which were normally
2705 implemented with a @code{define_expand}, should be similar to the
2706 @samp{mov@var{m}} patterns, except that operand 2 is the scratch
2709 These patterns need constraints for the reload register and scratch
2711 contain a single register class. If the original reload register (whose
2712 class is @var{class}) can meet the constraint given in the pattern, the
2713 value returned by these macros is used for the class of the scratch
2714 register. Otherwise, two additional reload registers are required.
2715 Their classes are obtained from the constraints in the insn pattern.
2717 @var{x} might be a pseudo-register or a @code{subreg} of a
2718 pseudo-register, which could either be in a hard register or in memory.
2719 Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is
2720 in memory and the hard register number if it is in a register.
2722 These macros should not be used in the case where a particular class of
2723 registers can only be copied to memory and not to another class of
2724 registers. In that case, secondary reload registers are not needed and
2725 would not be helpful. Instead, a stack location must be used to perform
2726 the copy and the @code{mov@var{m}} pattern should use memory as an
2727 intermediate storage. This case often occurs between floating-point and
2731 @defmac SECONDARY_MEMORY_NEEDED (@var{class1}, @var{class2}, @var{m})
2732 Certain machines have the property that some registers cannot be copied
2733 to some other registers without using memory. Define this macro on
2734 those machines to be a C expression that is nonzero if objects of mode
2735 @var{m} in registers of @var{class1} can only be copied to registers of
2736 class @var{class2} by storing a register of @var{class1} into memory
2737 and loading that memory location into a register of @var{class2}.
2739 Do not define this macro if its value would always be zero.
2742 @defmac SECONDARY_MEMORY_NEEDED_RTX (@var{mode})
2743 Normally when @code{SECONDARY_MEMORY_NEEDED} is defined, the compiler
2744 allocates a stack slot for a memory location needed for register copies.
2745 If this macro is defined, the compiler instead uses the memory location
2746 defined by this macro.
2748 Do not define this macro if you do not define
2749 @code{SECONDARY_MEMORY_NEEDED}.
2752 @defmac SECONDARY_MEMORY_NEEDED_MODE (@var{mode})
2753 When the compiler needs a secondary memory location to copy between two
2754 registers of mode @var{mode}, it normally allocates sufficient memory to
2755 hold a quantity of @code{BITS_PER_WORD} bits and performs the store and
2756 load operations in a mode that many bits wide and whose class is the
2757 same as that of @var{mode}.
2759 This is right thing to do on most machines because it ensures that all
2760 bits of the register are copied and prevents accesses to the registers
2761 in a narrower mode, which some machines prohibit for floating-point
2764 However, this default behavior is not correct on some machines, such as
2765 the DEC Alpha, that store short integers in floating-point registers
2766 differently than in integer registers. On those machines, the default
2767 widening will not work correctly and you must define this macro to
2768 suppress that widening in some cases. See the file @file{alpha.h} for
2771 Do not define this macro if you do not define
2772 @code{SECONDARY_MEMORY_NEEDED} or if widening @var{mode} to a mode that
2773 is @code{BITS_PER_WORD} bits wide is correct for your machine.
2776 @deftypefn {Target Hook} bool TARGET_CLASS_LIKELY_SPILLED_P (reg_class_t @var{rclass})
2777 A target hook which returns @code{true} if pseudos that have been assigned
2778 to registers of class @var{rclass} would likely be spilled because
2779 registers of @var{rclass} are needed for spill registers.
2781 The default version of this target hook returns @code{true} if @var{rclass}
2782 has exactly one register and @code{false} otherwise. On most machines, this
2783 default should be used. For generally register-starved machines, such as
2784 i386, or machines with right register constraints, such as SH, this hook
2785 can be used to avoid excessive spilling.
2787 This hook is also used by some of the global intra-procedural code
2788 transformations to throtle code motion, to avoid increasing register
2792 @deftypefn {Target Hook} {unsigned char} TARGET_CLASS_MAX_NREGS (reg_class_t @var{rclass}, machine_mode @var{mode})
2793 A target hook returns the maximum number of consecutive registers
2794 of class @var{rclass} needed to hold a value of mode @var{mode}.
2796 This is closely related to the macro @code{HARD_REGNO_NREGS}. In fact,
2797 the value returned by @code{TARGET_CLASS_MAX_NREGS (@var{rclass},
2798 @var{mode})} target hook should be the maximum value of
2799 @code{HARD_REGNO_NREGS (@var{regno}, @var{mode})} for all @var{regno}
2800 values in the class @var{rclass}.
2802 This target hook helps control the handling of multiple-word values
2805 The default version of this target hook returns the size of @var{mode}
2809 @defmac CLASS_MAX_NREGS (@var{class}, @var{mode})
2810 A C expression for the maximum number of consecutive registers
2811 of class @var{class} needed to hold a value of mode @var{mode}.
2813 This is closely related to the macro @code{HARD_REGNO_NREGS}. In fact,
2814 the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})}
2815 should be the maximum value of @code{HARD_REGNO_NREGS (@var{regno},
2816 @var{mode})} for all @var{regno} values in the class @var{class}.
2818 This macro helps control the handling of multiple-word values
2822 @defmac CANNOT_CHANGE_MODE_CLASS (@var{from}, @var{to}, @var{class})
2823 If defined, a C expression that returns nonzero for a @var{class} for which
2824 a change from mode @var{from} to mode @var{to} is invalid.
2826 For example, loading 32-bit integer or floating-point objects into
2827 floating-point registers on Alpha extends them to 64 bits.
2828 Therefore loading a 64-bit object and then storing it as a 32-bit object
2829 does not store the low-order 32 bits, as would be the case for a normal
2830 register. Therefore, @file{alpha.h} defines @code{CANNOT_CHANGE_MODE_CLASS}
2834 #define CANNOT_CHANGE_MODE_CLASS(FROM, TO, CLASS) \
2835 (GET_MODE_SIZE (FROM) != GET_MODE_SIZE (TO) \
2836 ? reg_classes_intersect_p (FLOAT_REGS, (CLASS)) : 0)
2839 Even if storing from a register in mode @var{to} would be valid,
2840 if both @var{from} and @code{raw_reg_mode} for @var{class} are wider
2841 than @code{word_mode}, then we must prevent @var{to} narrowing the
2842 mode. This happens when the middle-end assumes that it can load
2843 or store pieces of an @var{N}-word pseudo, and that the pseudo will
2844 eventually be allocated to @var{N} @code{word_mode} hard registers.
2845 Failure to prevent this kind of mode change will result in the
2846 entire @code{raw_reg_mode} being modified instead of the partial
2847 value that the middle-end intended.
2851 @deftypefn {Target Hook} bool TARGET_LRA_P (void)
2852 A target hook which returns true if we use LRA instead of reload pass. It means that LRA was ported to the target. The default version of this target hook returns always false.
2855 @deftypefn {Target Hook} int TARGET_REGISTER_PRIORITY (int)
2856 A target hook which returns the register priority number to which the register @var{hard_regno} belongs to. The bigger the number, the more preferable the hard register usage (when all other conditions are the same). This hook can be used to prefer some hard register over others in LRA. For example, some x86-64 register usage needs additional prefix which makes instructions longer. The hook can return lower priority number for such registers make them less favorable and as result making the generated code smaller. The default version of this target hook returns always zero.
2859 @deftypefn {Target Hook} bool TARGET_REGISTER_USAGE_LEVELING_P (void)
2860 A target hook which returns true if we need register usage leveling. That means if a few hard registers are equally good for the assignment, we choose the least used hard register. The register usage leveling may be profitable for some targets. Don't use the usage leveling for targets with conditional execution or targets with big register files as it hurts if-conversion and cross-jumping optimizations. The default version of this target hook returns always false.
2863 @deftypefn {Target Hook} bool TARGET_DIFFERENT_ADDR_DISPLACEMENT_P (void)
2864 A target hook which returns true if an address with the same structure can have different maximal legitimate displacement. For example, the displacement can depend on memory mode or on operand combinations in the insn. The default version of this target hook returns always false.
2867 @deftypefn {Target Hook} bool TARGET_CANNOT_SUBSTITUTE_MEM_EQUIV_P (rtx @var{subst})
2868 A target hook which returns @code{true} if @var{subst} can't
2869 substitute safely pseudos with equivalent memory values during
2870 register allocation.
2871 The default version of this target hook returns @code{false}.
2872 On most machines, this default should be used. For generally
2873 machines with non orthogonal register usage for addressing, such
2874 as SH, this hook can be used to avoid excessive spilling.
2877 @deftypefn {Target Hook} bool TARGET_LEGITIMIZE_ADDRESS_DISPLACEMENT (rtx *@var{disp}, rtx *@var{offset}, machine_mode @var{mode})
2878 A target hook which returns @code{true} if *@var{disp} is
2879 legitimezed to valid address displacement with subtracting *@var{offset}
2880 at memory mode @var{mode}.
2881 The default version of this target hook returns @code{false}.
2882 This hook will benefit machines with limited base plus displacement
2886 @deftypefn {Target Hook} reg_class_t TARGET_SPILL_CLASS (reg_class_t, @var{machine_mode})
2887 This hook defines a class of registers which could be used for spilling pseudos of the given mode and class, or @code{NO_REGS} if only memory should be used. Not defining this hook is equivalent to returning @code{NO_REGS} for all inputs.
2890 @deftypefn {Target Hook} machine_mode TARGET_CSTORE_MODE (enum insn_code @var{icode})
2891 This hook defines the machine mode to use for the boolean result of conditional store patterns. The ICODE argument is the instruction code for the cstore being performed. Not definiting this hook is the same as accepting the mode encoded into operand 0 of the cstore expander patterns.
2894 @node Stack and Calling
2895 @section Stack Layout and Calling Conventions
2896 @cindex calling conventions
2898 @c prevent bad page break with this line
2899 This describes the stack layout and calling conventions.
2903 * Exception Handling::
2908 * Register Arguments::
2910 * Aggregate Return::
2915 * Stack Smashing Protection::
2916 * Miscellaneous Register Hooks::
2920 @subsection Basic Stack Layout
2921 @cindex stack frame layout
2922 @cindex frame layout
2924 @c prevent bad page break with this line
2925 Here is the basic stack layout.
2927 @defmac STACK_GROWS_DOWNWARD
2928 Define this macro if pushing a word onto the stack moves the stack
2929 pointer to a smaller address.
2931 When we say, ``define this macro if @dots{}'', it means that the
2932 compiler checks this macro only with @code{#ifdef} so the precise
2933 definition used does not matter.
2936 @defmac STACK_PUSH_CODE
2937 This macro defines the operation used when something is pushed
2938 on the stack. In RTL, a push operation will be
2939 @code{(set (mem (STACK_PUSH_CODE (reg sp))) @dots{})}
2941 The choices are @code{PRE_DEC}, @code{POST_DEC}, @code{PRE_INC},
2942 and @code{POST_INC}. Which of these is correct depends on
2943 the stack direction and on whether the stack pointer points
2944 to the last item on the stack or whether it points to the
2945 space for the next item on the stack.
2947 The default is @code{PRE_DEC} when @code{STACK_GROWS_DOWNWARD} is
2948 defined, which is almost always right, and @code{PRE_INC} otherwise,
2949 which is often wrong.
2952 @defmac FRAME_GROWS_DOWNWARD
2953 Define this macro to nonzero value if the addresses of local variable slots
2954 are at negative offsets from the frame pointer.
2957 @defmac ARGS_GROW_DOWNWARD
2958 Define this macro if successive arguments to a function occupy decreasing
2959 addresses on the stack.
2962 @defmac STARTING_FRAME_OFFSET
2963 Offset from the frame pointer to the first local variable slot to be allocated.
2965 If @code{FRAME_GROWS_DOWNWARD}, find the next slot's offset by
2966 subtracting the first slot's length from @code{STARTING_FRAME_OFFSET}.
2967 Otherwise, it is found by adding the length of the first slot to the
2968 value @code{STARTING_FRAME_OFFSET}.
2969 @c i'm not sure if the above is still correct.. had to change it to get
2970 @c rid of an overfull. --mew 2feb93
2973 @defmac STACK_ALIGNMENT_NEEDED
2974 Define to zero to disable final alignment of the stack during reload.
2975 The nonzero default for this macro is suitable for most ports.
2977 On ports where @code{STARTING_FRAME_OFFSET} is nonzero or where there
2978 is a register save block following the local block that doesn't require
2979 alignment to @code{STACK_BOUNDARY}, it may be beneficial to disable
2980 stack alignment and do it in the backend.
2983 @defmac STACK_POINTER_OFFSET
2984 Offset from the stack pointer register to the first location at which
2985 outgoing arguments are placed. If not specified, the default value of
2986 zero is used. This is the proper value for most machines.
2988 If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
2989 the first location at which outgoing arguments are placed.
2992 @defmac FIRST_PARM_OFFSET (@var{fundecl})
2993 Offset from the argument pointer register to the first argument's
2994 address. On some machines it may depend on the data type of the
2997 If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
2998 the first argument's address.
3001 @defmac STACK_DYNAMIC_OFFSET (@var{fundecl})
3002 Offset from the stack pointer register to an item dynamically allocated
3003 on the stack, e.g., by @code{alloca}.
3005 The default value for this macro is @code{STACK_POINTER_OFFSET} plus the
3006 length of the outgoing arguments. The default is correct for most
3007 machines. See @file{function.c} for details.
3010 @defmac INITIAL_FRAME_ADDRESS_RTX
3011 A C expression whose value is RTL representing the address of the initial
3012 stack frame. This address is passed to @code{RETURN_ADDR_RTX} and
3013 @code{DYNAMIC_CHAIN_ADDRESS}. If you don't define this macro, a reasonable
3014 default value will be used. Define this macro in order to make frame pointer
3015 elimination work in the presence of @code{__builtin_frame_address (count)} and
3016 @code{__builtin_return_address (count)} for @code{count} not equal to zero.
3019 @defmac DYNAMIC_CHAIN_ADDRESS (@var{frameaddr})
3020 A C expression whose value is RTL representing the address in a stack
3021 frame where the pointer to the caller's frame is stored. Assume that
3022 @var{frameaddr} is an RTL expression for the address of the stack frame
3025 If you don't define this macro, the default is to return the value
3026 of @var{frameaddr}---that is, the stack frame address is also the
3027 address of the stack word that points to the previous frame.
3030 @defmac SETUP_FRAME_ADDRESSES
3031 If defined, a C expression that produces the machine-specific code to
3032 setup the stack so that arbitrary frames can be accessed. For example,
3033 on the SPARC, we must flush all of the register windows to the stack
3034 before we can access arbitrary stack frames. You will seldom need to
3038 @deftypefn {Target Hook} rtx TARGET_BUILTIN_SETJMP_FRAME_VALUE (void)
3039 This target hook should return an rtx that is used to store
3040 the address of the current frame into the built in @code{setjmp} buffer.
3041 The default value, @code{virtual_stack_vars_rtx}, is correct for most
3042 machines. One reason you may need to define this target hook is if
3043 @code{hard_frame_pointer_rtx} is the appropriate value on your machine.
3046 @defmac FRAME_ADDR_RTX (@var{frameaddr})
3047 A C expression whose value is RTL representing the value of the frame
3048 address for the current frame. @var{frameaddr} is the frame pointer
3049 of the current frame. This is used for __builtin_frame_address.
3050 You need only define this macro if the frame address is not the same
3051 as the frame pointer. Most machines do not need to define it.
3054 @defmac RETURN_ADDR_RTX (@var{count}, @var{frameaddr})
3055 A C expression whose value is RTL representing the value of the return
3056 address for the frame @var{count} steps up from the current frame, after
3057 the prologue. @var{frameaddr} is the frame pointer of the @var{count}
3058 frame, or the frame pointer of the @var{count} @minus{} 1 frame if
3059 @code{RETURN_ADDR_IN_PREVIOUS_FRAME} is nonzero.
3061 The value of the expression must always be the correct address when
3062 @var{count} is zero, but may be @code{NULL_RTX} if there is no way to
3063 determine the return address of other frames.
3066 @defmac RETURN_ADDR_IN_PREVIOUS_FRAME
3067 Define this macro to nonzero value if the return address of a particular
3068 stack frame is accessed from the frame pointer of the previous stack
3069 frame. The zero default for this macro is suitable for most ports.
3072 @defmac INCOMING_RETURN_ADDR_RTX
3073 A C expression whose value is RTL representing the location of the
3074 incoming return address at the beginning of any function, before the
3075 prologue. This RTL is either a @code{REG}, indicating that the return
3076 value is saved in @samp{REG}, or a @code{MEM} representing a location in
3079 You only need to define this macro if you want to support call frame
3080 debugging information like that provided by DWARF 2.
3082 If this RTL is a @code{REG}, you should also define
3083 @code{DWARF_FRAME_RETURN_COLUMN} to @code{DWARF_FRAME_REGNUM (REGNO)}.
3086 @defmac DWARF_ALT_FRAME_RETURN_COLUMN
3087 A C expression whose value is an integer giving a DWARF 2 column
3088 number that may be used as an alternative return column. The column
3089 must not correspond to any gcc hard register (that is, it must not
3090 be in the range of @code{DWARF_FRAME_REGNUM}).
3092 This macro can be useful if @code{DWARF_FRAME_RETURN_COLUMN} is set to a
3093 general register, but an alternative column needs to be used for signal
3094 frames. Some targets have also used different frame return columns
3098 @defmac DWARF_ZERO_REG
3099 A C expression whose value is an integer giving a DWARF 2 register
3100 number that is considered to always have the value zero. This should
3101 only be defined if the target has an architected zero register, and
3102 someone decided it was a good idea to use that register number to
3103 terminate the stack backtrace. New ports should avoid this.
3106 @deftypefn {Target Hook} void TARGET_DWARF_HANDLE_FRAME_UNSPEC (const char *@var{label}, rtx @var{pattern}, int @var{index})
3107 This target hook allows the backend to emit frame-related insns that
3108 contain UNSPECs or UNSPEC_VOLATILEs. The DWARF 2 call frame debugging
3109 info engine will invoke it on insns of the form
3111 (set (reg) (unspec [@dots{}] UNSPEC_INDEX))
3115 (set (reg) (unspec_volatile [@dots{}] UNSPECV_INDEX)).
3117 to let the backend emit the call frame instructions. @var{label} is
3118 the CFI label attached to the insn, @var{pattern} is the pattern of
3119 the insn and @var{index} is @code{UNSPEC_INDEX} or @code{UNSPECV_INDEX}.
3122 @defmac INCOMING_FRAME_SP_OFFSET
3123 A C expression whose value is an integer giving the offset, in bytes,
3124 from the value of the stack pointer register to the top of the stack
3125 frame at the beginning of any function, before the prologue. The top of
3126 the frame is defined to be the value of the stack pointer in the
3127 previous frame, just before the call instruction.
3129 You only need to define this macro if you want to support call frame
3130 debugging information like that provided by DWARF 2.
3133 @defmac ARG_POINTER_CFA_OFFSET (@var{fundecl})
3134 A C expression whose value is an integer giving the offset, in bytes,
3135 from the argument pointer to the canonical frame address (cfa). The
3136 final value should coincide with that calculated by
3137 @code{INCOMING_FRAME_SP_OFFSET}. Which is unfortunately not usable
3138 during virtual register instantiation.
3140 The default value for this macro is
3141 @code{FIRST_PARM_OFFSET (fundecl) + crtl->args.pretend_args_size},
3142 which is correct for most machines; in general, the arguments are found
3143 immediately before the stack frame. Note that this is not the case on
3144 some targets that save registers into the caller's frame, such as SPARC
3145 and rs6000, and so such targets need to define this macro.
3147 You only need to define this macro if the default is incorrect, and you
3148 want to support call frame debugging information like that provided by
3152 @defmac FRAME_POINTER_CFA_OFFSET (@var{fundecl})
3153 If defined, a C expression whose value is an integer giving the offset
3154 in bytes from the frame pointer to the canonical frame address (cfa).
3155 The final value should coincide with that calculated by
3156 @code{INCOMING_FRAME_SP_OFFSET}.
3158 Normally the CFA is calculated as an offset from the argument pointer,
3159 via @code{ARG_POINTER_CFA_OFFSET}, but if the argument pointer is
3160 variable due to the ABI, this may not be possible. If this macro is
3161 defined, it implies that the virtual register instantiation should be
3162 based on the frame pointer instead of the argument pointer. Only one
3163 of @code{FRAME_POINTER_CFA_OFFSET} and @code{ARG_POINTER_CFA_OFFSET}
3167 @defmac CFA_FRAME_BASE_OFFSET (@var{fundecl})
3168 If defined, a C expression whose value is an integer giving the offset
3169 in bytes from the canonical frame address (cfa) to the frame base used
3170 in DWARF 2 debug information. The default is zero. A different value
3171 may reduce the size of debug information on some ports.
3174 @node Exception Handling
3175 @subsection Exception Handling Support
3176 @cindex exception handling
3178 @defmac EH_RETURN_DATA_REGNO (@var{N})
3179 A C expression whose value is the @var{N}th register number used for
3180 data by exception handlers, or @code{INVALID_REGNUM} if fewer than
3181 @var{N} registers are usable.
3183 The exception handling library routines communicate with the exception
3184 handlers via a set of agreed upon registers. Ideally these registers
3185 should be call-clobbered; it is possible to use call-saved registers,
3186 but may negatively impact code size. The target must support at least
3187 2 data registers, but should define 4 if there are enough free registers.
3189 You must define this macro if you want to support call frame exception
3190 handling like that provided by DWARF 2.
3193 @defmac EH_RETURN_STACKADJ_RTX
3194 A C expression whose value is RTL representing a location in which
3195 to store a stack adjustment to be applied before function return.
3196 This is used to unwind the stack to an exception handler's call frame.
3197 It will be assigned zero on code paths that return normally.
3199 Typically this is a call-clobbered hard register that is otherwise
3200 untouched by the epilogue, but could also be a stack slot.
3202 Do not define this macro if the stack pointer is saved and restored
3203 by the regular prolog and epilog code in the call frame itself; in
3204 this case, the exception handling library routines will update the
3205 stack location to be restored in place. Otherwise, you must define
3206 this macro if you want to support call frame exception handling like
3207 that provided by DWARF 2.
3210 @defmac EH_RETURN_HANDLER_RTX
3211 A C expression whose value is RTL representing a location in which
3212 to store the address of an exception handler to which we should
3213 return. It will not be assigned on code paths that return normally.
3215 Typically this is the location in the call frame at which the normal
3216 return address is stored. For targets that return by popping an
3217 address off the stack, this might be a memory address just below
3218 the @emph{target} call frame rather than inside the current call
3219 frame. If defined, @code{EH_RETURN_STACKADJ_RTX} will have already
3220 been assigned, so it may be used to calculate the location of the
3223 Some targets have more complex requirements than storing to an
3224 address calculable during initial code generation. In that case
3225 the @code{eh_return} instruction pattern should be used instead.
3227 If you want to support call frame exception handling, you must
3228 define either this macro or the @code{eh_return} instruction pattern.
3231 @defmac RETURN_ADDR_OFFSET
3232 If defined, an integer-valued C expression for which rtl will be generated
3233 to add it to the exception handler address before it is searched in the
3234 exception handling tables, and to subtract it again from the address before
3235 using it to return to the exception handler.
3238 @defmac ASM_PREFERRED_EH_DATA_FORMAT (@var{code}, @var{global})
3239 This macro chooses the encoding of pointers embedded in the exception
3240 handling sections. If at all possible, this should be defined such
3241 that the exception handling section will not require dynamic relocations,
3242 and so may be read-only.
3244 @var{code} is 0 for data, 1 for code labels, 2 for function pointers.
3245 @var{global} is true if the symbol may be affected by dynamic relocations.
3246 The macro should return a combination of the @code{DW_EH_PE_*} defines
3247 as found in @file{dwarf2.h}.
3249 If this macro is not defined, pointers will not be encoded but
3250 represented directly.
3253 @defmac ASM_MAYBE_OUTPUT_ENCODED_ADDR_RTX (@var{file}, @var{encoding}, @var{size}, @var{addr}, @var{done})
3254 This macro allows the target to emit whatever special magic is required
3255 to represent the encoding chosen by @code{ASM_PREFERRED_EH_DATA_FORMAT}.
3256 Generic code takes care of pc-relative and indirect encodings; this must
3257 be defined if the target uses text-relative or data-relative encodings.
3259 This is a C statement that branches to @var{done} if the format was
3260 handled. @var{encoding} is the format chosen, @var{size} is the number
3261 of bytes that the format occupies, @var{addr} is the @code{SYMBOL_REF}
3265 @defmac MD_FALLBACK_FRAME_STATE_FOR (@var{context}, @var{fs})
3266 This macro allows the target to add CPU and operating system specific
3267 code to the call-frame unwinder for use when there is no unwind data
3268 available. The most common reason to implement this macro is to unwind
3269 through signal frames.
3271 This macro is called from @code{uw_frame_state_for} in
3272 @file{unwind-dw2.c}, @file{unwind-dw2-xtensa.c} and
3273 @file{unwind-ia64.c}. @var{context} is an @code{_Unwind_Context};
3274 @var{fs} is an @code{_Unwind_FrameState}. Examine @code{context->ra}
3275 for the address of the code being executed and @code{context->cfa} for
3276 the stack pointer value. If the frame can be decoded, the register
3277 save addresses should be updated in @var{fs} and the macro should
3278 evaluate to @code{_URC_NO_REASON}. If the frame cannot be decoded,
3279 the macro should evaluate to @code{_URC_END_OF_STACK}.
3281 For proper signal handling in Java this macro is accompanied by
3282 @code{MAKE_THROW_FRAME}, defined in @file{libjava/include/*-signal.h} headers.
3285 @defmac MD_HANDLE_UNWABI (@var{context}, @var{fs})
3286 This macro allows the target to add operating system specific code to the
3287 call-frame unwinder to handle the IA-64 @code{.unwabi} unwinding directive,
3288 usually used for signal or interrupt frames.
3290 This macro is called from @code{uw_update_context} in libgcc's
3291 @file{unwind-ia64.c}. @var{context} is an @code{_Unwind_Context};
3292 @var{fs} is an @code{_Unwind_FrameState}. Examine @code{fs->unwabi}
3293 for the abi and context in the @code{.unwabi} directive. If the
3294 @code{.unwabi} directive can be handled, the register save addresses should
3295 be updated in @var{fs}.
3298 @defmac TARGET_USES_WEAK_UNWIND_INFO
3299 A C expression that evaluates to true if the target requires unwind
3300 info to be given comdat linkage. Define it to be @code{1} if comdat
3301 linkage is necessary. The default is @code{0}.
3304 @node Stack Checking
3305 @subsection Specifying How Stack Checking is Done
3307 GCC will check that stack references are within the boundaries of the
3308 stack, if the option @option{-fstack-check} is specified, in one of
3313 If the value of the @code{STACK_CHECK_BUILTIN} macro is nonzero, GCC
3314 will assume that you have arranged for full stack checking to be done
3315 at appropriate places in the configuration files. GCC will not do
3316 other special processing.
3319 If @code{STACK_CHECK_BUILTIN} is zero and the value of the
3320 @code{STACK_CHECK_STATIC_BUILTIN} macro is nonzero, GCC will assume
3321 that you have arranged for static stack checking (checking of the
3322 static stack frame of functions) to be done at appropriate places
3323 in the configuration files. GCC will only emit code to do dynamic
3324 stack checking (checking on dynamic stack allocations) using the third
3328 If neither of the above are true, GCC will generate code to periodically
3329 ``probe'' the stack pointer using the values of the macros defined below.
3332 If neither STACK_CHECK_BUILTIN nor STACK_CHECK_STATIC_BUILTIN is defined,
3333 GCC will change its allocation strategy for large objects if the option
3334 @option{-fstack-check} is specified: they will always be allocated
3335 dynamically if their size exceeds @code{STACK_CHECK_MAX_VAR_SIZE} bytes.
3337 @defmac STACK_CHECK_BUILTIN
3338 A nonzero value if stack checking is done by the configuration files in a
3339 machine-dependent manner. You should define this macro if stack checking
3340 is required by the ABI of your machine or if you would like to do stack
3341 checking in some more efficient way than the generic approach. The default
3342 value of this macro is zero.
3345 @defmac STACK_CHECK_STATIC_BUILTIN
3346 A nonzero value if static stack checking is done by the configuration files
3347 in a machine-dependent manner. You should define this macro if you would
3348 like to do static stack checking in some more efficient way than the generic
3349 approach. The default value of this macro is zero.
3352 @defmac STACK_CHECK_PROBE_INTERVAL_EXP
3353 An integer specifying the interval at which GCC must generate stack probe
3354 instructions, defined as 2 raised to this integer. You will normally
3355 define this macro so that the interval be no larger than the size of
3356 the ``guard pages'' at the end of a stack area. The default value
3357 of 12 (4096-byte interval) is suitable for most systems.
3360 @defmac STACK_CHECK_MOVING_SP
3361 An integer which is nonzero if GCC should move the stack pointer page by page
3362 when doing probes. This can be necessary on systems where the stack pointer
3363 contains the bottom address of the memory area accessible to the executing
3364 thread at any point in time. In this situation an alternate signal stack
3365 is required in order to be able to recover from a stack overflow. The
3366 default value of this macro is zero.
3369 @defmac STACK_CHECK_PROTECT
3370 The number of bytes of stack needed to recover from a stack overflow, for
3371 languages where such a recovery is supported. The default value of 75 words
3372 with the @code{setjmp}/@code{longjmp}-based exception handling mechanism and
3373 8192 bytes with other exception handling mechanisms should be adequate for
3377 The following macros are relevant only if neither STACK_CHECK_BUILTIN
3378 nor STACK_CHECK_STATIC_BUILTIN is defined; you can omit them altogether
3379 in the opposite case.
3381 @defmac STACK_CHECK_MAX_FRAME_SIZE
3382 The maximum size of a stack frame, in bytes. GCC will generate probe
3383 instructions in non-leaf functions to ensure at least this many bytes of
3384 stack are available. If a stack frame is larger than this size, stack
3385 checking will not be reliable and GCC will issue a warning. The
3386 default is chosen so that GCC only generates one instruction on most
3387 systems. You should normally not change the default value of this macro.
3390 @defmac STACK_CHECK_FIXED_FRAME_SIZE
3391 GCC uses this value to generate the above warning message. It
3392 represents the amount of fixed frame used by a function, not including
3393 space for any callee-saved registers, temporaries and user variables.
3394 You need only specify an upper bound for this amount and will normally
3395 use the default of four words.
3398 @defmac STACK_CHECK_MAX_VAR_SIZE
3399 The maximum size, in bytes, of an object that GCC will place in the
3400 fixed area of the stack frame when the user specifies
3401 @option{-fstack-check}.
3402 GCC computed the default from the values of the above macros and you will
3403 normally not need to override that default.
3407 @node Frame Registers
3408 @subsection Registers That Address the Stack Frame
3410 @c prevent bad page break with this line
3411 This discusses registers that address the stack frame.
3413 @defmac STACK_POINTER_REGNUM
3414 The register number of the stack pointer register, which must also be a
3415 fixed register according to @code{FIXED_REGISTERS}. On most machines,
3416 the hardware determines which register this is.
3419 @defmac FRAME_POINTER_REGNUM
3420 The register number of the frame pointer register, which is used to
3421 access automatic variables in the stack frame. On some machines, the
3422 hardware determines which register this is. On other machines, you can
3423 choose any register you wish for this purpose.
3426 @defmac HARD_FRAME_POINTER_REGNUM
3427 On some machines the offset between the frame pointer and starting
3428 offset of the automatic variables is not known until after register
3429 allocation has been done (for example, because the saved registers are
3430 between these two locations). On those machines, define
3431 @code{FRAME_POINTER_REGNUM} the number of a special, fixed register to
3432 be used internally until the offset is known, and define
3433 @code{HARD_FRAME_POINTER_REGNUM} to be the actual hard register number
3434 used for the frame pointer.
3436 You should define this macro only in the very rare circumstances when it
3437 is not possible to calculate the offset between the frame pointer and
3438 the automatic variables until after register allocation has been
3439 completed. When this macro is defined, you must also indicate in your
3440 definition of @code{ELIMINABLE_REGS} how to eliminate
3441 @code{FRAME_POINTER_REGNUM} into either @code{HARD_FRAME_POINTER_REGNUM}
3442 or @code{STACK_POINTER_REGNUM}.
3444 Do not define this macro if it would be the same as
3445 @code{FRAME_POINTER_REGNUM}.
3448 @defmac ARG_POINTER_REGNUM
3449 The register number of the arg pointer register, which is used to access
3450 the function's argument list. On some machines, this is the same as the
3451 frame pointer register. On some machines, the hardware determines which
3452 register this is. On other machines, you can choose any register you
3453 wish for this purpose. If this is not the same register as the frame
3454 pointer register, then you must mark it as a fixed register according to
3455 @code{FIXED_REGISTERS}, or arrange to be able to eliminate it
3456 (@pxref{Elimination}).
3459 @defmac HARD_FRAME_POINTER_IS_FRAME_POINTER
3460 Define this to a preprocessor constant that is nonzero if
3461 @code{hard_frame_pointer_rtx} and @code{frame_pointer_rtx} should be
3462 the same. The default definition is @samp{(HARD_FRAME_POINTER_REGNUM
3463 == FRAME_POINTER_REGNUM)}; you only need to define this macro if that
3464 definition is not suitable for use in preprocessor conditionals.
3467 @defmac HARD_FRAME_POINTER_IS_ARG_POINTER
3468 Define this to a preprocessor constant that is nonzero if
3469 @code{hard_frame_pointer_rtx} and @code{arg_pointer_rtx} should be the
3470 same. The default definition is @samp{(HARD_FRAME_POINTER_REGNUM ==
3471 ARG_POINTER_REGNUM)}; you only need to define this macro if that
3472 definition is not suitable for use in preprocessor conditionals.
3475 @defmac RETURN_ADDRESS_POINTER_REGNUM
3476 The register number of the return address pointer register, which is used to
3477 access the current function's return address from the stack. On some
3478 machines, the return address is not at a fixed offset from the frame
3479 pointer or stack pointer or argument pointer. This register can be defined
3480 to point to the return address on the stack, and then be converted by
3481 @code{ELIMINABLE_REGS} into either the frame pointer or stack pointer.
3483 Do not define this macro unless there is no other way to get the return
3484 address from the stack.
3487 @defmac STATIC_CHAIN_REGNUM
3488 @defmacx STATIC_CHAIN_INCOMING_REGNUM
3489 Register numbers used for passing a function's static chain pointer. If
3490 register windows are used, the register number as seen by the called
3491 function is @code{STATIC_CHAIN_INCOMING_REGNUM}, while the register
3492 number as seen by the calling function is @code{STATIC_CHAIN_REGNUM}. If
3493 these registers are the same, @code{STATIC_CHAIN_INCOMING_REGNUM} need
3496 The static chain register need not be a fixed register.
3498 If the static chain is passed in memory, these macros should not be
3499 defined; instead, the @code{TARGET_STATIC_CHAIN} hook should be used.
3502 @deftypefn {Target Hook} rtx TARGET_STATIC_CHAIN (const_tree @var{fndecl_or_type}, bool @var{incoming_p})
3503 This hook replaces the use of @code{STATIC_CHAIN_REGNUM} et al for
3504 targets that may use different static chain locations for different
3505 nested functions. This may be required if the target has function
3506 attributes that affect the calling conventions of the function and
3507 those calling conventions use different static chain locations.
3509 The default version of this hook uses @code{STATIC_CHAIN_REGNUM} et al.
3511 If the static chain is passed in memory, this hook should be used to
3512 provide rtx giving @code{mem} expressions that denote where they are stored.
3513 Often the @code{mem} expression as seen by the caller will be at an offset
3514 from the stack pointer and the @code{mem} expression as seen by the callee
3515 will be at an offset from the frame pointer.
3516 @findex stack_pointer_rtx
3517 @findex frame_pointer_rtx
3518 @findex arg_pointer_rtx
3519 The variables @code{stack_pointer_rtx}, @code{frame_pointer_rtx}, and
3520 @code{arg_pointer_rtx} will have been initialized and should be used
3521 to refer to those items.
3524 @defmac DWARF_FRAME_REGISTERS
3525 This macro specifies the maximum number of hard registers that can be
3526 saved in a call frame. This is used to size data structures used in
3527 DWARF2 exception handling.
3529 Prior to GCC 3.0, this macro was needed in order to establish a stable
3530 exception handling ABI in the face of adding new hard registers for ISA
3531 extensions. In GCC 3.0 and later, the EH ABI is insulated from changes
3532 in the number of hard registers. Nevertheless, this macro can still be
3533 used to reduce the runtime memory requirements of the exception handling
3534 routines, which can be substantial if the ISA contains a lot of
3535 registers that are not call-saved.
3537 If this macro is not defined, it defaults to
3538 @code{FIRST_PSEUDO_REGISTER}.
3541 @defmac PRE_GCC3_DWARF_FRAME_REGISTERS
3543 This macro is similar to @code{DWARF_FRAME_REGISTERS}, but is provided
3544 for backward compatibility in pre GCC 3.0 compiled code.
3546 If this macro is not defined, it defaults to
3547 @code{DWARF_FRAME_REGISTERS}.
3550 @defmac DWARF_REG_TO_UNWIND_COLUMN (@var{regno})
3552 Define this macro if the target's representation for dwarf registers
3553 is different than the internal representation for unwind column.
3554 Given a dwarf register, this macro should return the internal unwind
3555 column number to use instead.
3557 See the PowerPC's SPE target for an example.
3560 @defmac DWARF_FRAME_REGNUM (@var{regno})
3562 Define this macro if the target's representation for dwarf registers
3563 used in .eh_frame or .debug_frame is different from that used in other
3564 debug info sections. Given a GCC hard register number, this macro
3565 should return the .eh_frame register number. The default is
3566 @code{DBX_REGISTER_NUMBER (@var{regno})}.
3570 @defmac DWARF2_FRAME_REG_OUT (@var{regno}, @var{for_eh})
3572 Define this macro to map register numbers held in the call frame info
3573 that GCC has collected using @code{DWARF_FRAME_REGNUM} to those that
3574 should be output in .debug_frame (@code{@var{for_eh}} is zero) and
3575 .eh_frame (@code{@var{for_eh}} is nonzero). The default is to
3576 return @code{@var{regno}}.
3580 @defmac REG_VALUE_IN_UNWIND_CONTEXT
3582 Define this macro if the target stores register values as
3583 @code{_Unwind_Word} type in unwind context. It should be defined if
3584 target register size is larger than the size of @code{void *}. The
3585 default is to store register values as @code{void *} type.
3589 @defmac ASSUME_EXTENDED_UNWIND_CONTEXT
3591 Define this macro to be 1 if the target always uses extended unwind
3592 context with version, args_size and by_value fields. If it is undefined,
3593 it will be defined to 1 when @code{REG_VALUE_IN_UNWIND_CONTEXT} is
3594 defined and 0 otherwise.
3599 @subsection Eliminating Frame Pointer and Arg Pointer
3601 @c prevent bad page break with this line
3602 This is about eliminating the frame pointer and arg pointer.
3604 @deftypefn {Target Hook} bool TARGET_FRAME_POINTER_REQUIRED (void)
3605 This target hook should return @code{true} if a function must have and use
3606 a frame pointer. This target hook is called in the reload pass. If its return
3607 value is @code{true} the function will have a frame pointer.
3609 This target hook can in principle examine the current function and decide
3610 according to the facts, but on most machines the constant @code{false} or the
3611 constant @code{true} suffices. Use @code{false} when the machine allows code
3612 to be generated with no frame pointer, and doing so saves some time or space.
3613 Use @code{true} when there is no possible advantage to avoiding a frame
3616 In certain cases, the compiler does not know how to produce valid code
3617 without a frame pointer. The compiler recognizes those cases and
3618 automatically gives the function a frame pointer regardless of what
3619 @code{TARGET_FRAME_POINTER_REQUIRED} returns. You don't need to worry about
3622 In a function that does not require a frame pointer, the frame pointer
3623 register can be allocated for ordinary usage, unless you mark it as a
3624 fixed register. See @code{FIXED_REGISTERS} for more information.
3626 Default return value is @code{false}.
3629 @findex get_frame_size
3630 @defmac INITIAL_FRAME_POINTER_OFFSET (@var{depth-var})
3631 A C statement to store in the variable @var{depth-var} the difference
3632 between the frame pointer and the stack pointer values immediately after
3633 the function prologue. The value would be computed from information
3634 such as the result of @code{get_frame_size ()} and the tables of
3635 registers @code{regs_ever_live} and @code{call_used_regs}.
3637 If @code{ELIMINABLE_REGS} is defined, this macro will be not be used and
3638 need not be defined. Otherwise, it must be defined even if
3639 @code{TARGET_FRAME_POINTER_REQUIRED} always returns true; in that
3640 case, you may set @var{depth-var} to anything.
3643 @defmac ELIMINABLE_REGS
3644 If defined, this macro specifies a table of register pairs used to
3645 eliminate unneeded registers that point into the stack frame. If it is not
3646 defined, the only elimination attempted by the compiler is to replace
3647 references to the frame pointer with references to the stack pointer.
3649 The definition of this macro is a list of structure initializations, each
3650 of which specifies an original and replacement register.
3652 On some machines, the position of the argument pointer is not known until
3653 the compilation is completed. In such a case, a separate hard register
3654 must be used for the argument pointer. This register can be eliminated by
3655 replacing it with either the frame pointer or the argument pointer,
3656 depending on whether or not the frame pointer has been eliminated.
3658 In this case, you might specify:
3660 #define ELIMINABLE_REGS \
3661 @{@{ARG_POINTER_REGNUM, STACK_POINTER_REGNUM@}, \
3662 @{ARG_POINTER_REGNUM, FRAME_POINTER_REGNUM@}, \
3663 @{FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM@}@}
3666 Note that the elimination of the argument pointer with the stack pointer is
3667 specified first since that is the preferred elimination.
3670 @deftypefn {Target Hook} bool TARGET_CAN_ELIMINATE (const int @var{from_reg}, const int @var{to_reg})
3671 This target hook should returns @code{true} if the compiler is allowed to
3672 try to replace register number @var{from_reg} with register number
3673 @var{to_reg}. This target hook need only be defined if @code{ELIMINABLE_REGS}
3674 is defined, and will usually be @code{true}, since most of the cases
3675 preventing register elimination are things that the compiler already
3678 Default return value is @code{true}.
3681 @defmac INITIAL_ELIMINATION_OFFSET (@var{from-reg}, @var{to-reg}, @var{offset-var})
3682 This macro is similar to @code{INITIAL_FRAME_POINTER_OFFSET}. It
3683 specifies the initial difference between the specified pair of
3684 registers. This macro must be defined if @code{ELIMINABLE_REGS} is
3688 @node Stack Arguments
3689 @subsection Passing Function Arguments on the Stack
3690 @cindex arguments on stack
3691 @cindex stack arguments
3693 The macros in this section control how arguments are passed
3694 on the stack. See the following section for other macros that
3695 control passing certain arguments in registers.
3697 @deftypefn {Target Hook} bool TARGET_PROMOTE_PROTOTYPES (const_tree @var{fntype})
3698 This target hook returns @code{true} if an argument declared in a
3699 prototype as an integral type smaller than @code{int} should actually be
3700 passed as an @code{int}. In addition to avoiding errors in certain
3701 cases of mismatch, it also makes for better code on certain machines.
3702 The default is to not promote prototypes.
3706 A C expression. If nonzero, push insns will be used to pass
3708 If the target machine does not have a push instruction, set it to zero.
3709 That directs GCC to use an alternate strategy: to
3710 allocate the entire argument block and then store the arguments into
3711 it. When @code{PUSH_ARGS} is nonzero, @code{PUSH_ROUNDING} must be defined too.
3714 @defmac PUSH_ARGS_REVERSED
3715 A C expression. If nonzero, function arguments will be evaluated from
3716 last to first, rather than from first to last. If this macro is not
3717 defined, it defaults to @code{PUSH_ARGS} on targets where the stack
3718 and args grow in opposite directions, and 0 otherwise.
3721 @defmac PUSH_ROUNDING (@var{npushed})
3722 A C expression that is the number of bytes actually pushed onto the
3723 stack when an instruction attempts to push @var{npushed} bytes.
3725 On some machines, the definition
3728 #define PUSH_ROUNDING(BYTES) (BYTES)
3732 will suffice. But on other machines, instructions that appear
3733 to push one byte actually push two bytes in an attempt to maintain
3734 alignment. Then the definition should be
3737 #define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1)
3740 If the value of this macro has a type, it should be an unsigned type.
3743 @findex outgoing_args_size
3744 @findex crtl->outgoing_args_size
3745 @defmac ACCUMULATE_OUTGOING_ARGS
3746 A C expression. If nonzero, the maximum amount of space required for outgoing arguments
3747 will be computed and placed into
3748 @code{crtl->outgoing_args_size}. No space will be pushed
3749 onto the stack for each call; instead, the function prologue should
3750 increase the stack frame size by this amount.
3752 Setting both @code{PUSH_ARGS} and @code{ACCUMULATE_OUTGOING_ARGS}
3756 @defmac REG_PARM_STACK_SPACE (@var{fndecl})
3757 Define this macro if functions should assume that stack space has been
3758 allocated for arguments even when their values are passed in
3761 The value of this macro is the size, in bytes, of the area reserved for
3762 arguments passed in registers for the function represented by @var{fndecl},
3763 which can be zero if GCC is calling a library function.
3764 The argument @var{fndecl} can be the FUNCTION_DECL, or the type itself
3767 This space can be allocated by the caller, or be a part of the
3768 machine-dependent stack frame: @code{OUTGOING_REG_PARM_STACK_SPACE} says
3771 @c above is overfull. not sure what to do. --mew 5feb93 did
3772 @c something, not sure if it looks good. --mew 10feb93
3774 @defmac INCOMING_REG_PARM_STACK_SPACE (@var{fndecl})
3775 Like @code{REG_PARM_STACK_SPACE}, but for incoming register arguments.
3776 Define this macro if space guaranteed when compiling a function body
3777 is different to space required when making a call, a situation that
3778 can arise with K&R style function definitions.
3781 @defmac OUTGOING_REG_PARM_STACK_SPACE (@var{fntype})
3782 Define this to a nonzero value if it is the responsibility of the
3783 caller to allocate the area reserved for arguments passed in registers
3784 when calling a function of @var{fntype}. @var{fntype} may be NULL
3785 if the function called is a library function.
3787 If @code{ACCUMULATE_OUTGOING_ARGS} is defined, this macro controls
3788 whether the space for these arguments counts in the value of
3789 @code{crtl->outgoing_args_size}.
3792 @defmac STACK_PARMS_IN_REG_PARM_AREA
3793 Define this macro if @code{REG_PARM_STACK_SPACE} is defined, but the
3794 stack parameters don't skip the area specified by it.
3795 @c i changed this, makes more sens and it should have taken care of the
3796 @c overfull.. not as specific, tho. --mew 5feb93
3798 Normally, when a parameter is not passed in registers, it is placed on the
3799 stack beyond the @code{REG_PARM_STACK_SPACE} area. Defining this macro
3800 suppresses this behavior and causes the parameter to be passed on the
3801 stack in its natural location.
3804 @deftypefn {Target Hook} int TARGET_RETURN_POPS_ARGS (tree @var{fundecl}, tree @var{funtype}, int @var{size})
3805 This target hook returns the number of bytes of its own arguments that
3806 a function pops on returning, or 0 if the function pops no arguments
3807 and the caller must therefore pop them all after the function returns.
3809 @var{fundecl} is a C variable whose value is a tree node that describes
3810 the function in question. Normally it is a node of type
3811 @code{FUNCTION_DECL} that describes the declaration of the function.
3812 From this you can obtain the @code{DECL_ATTRIBUTES} of the function.
3814 @var{funtype} is a C variable whose value is a tree node that
3815 describes the function in question. Normally it is a node of type
3816 @code{FUNCTION_TYPE} that describes the data type of the function.
3817 From this it is possible to obtain the data types of the value and
3818 arguments (if known).
3820 When a call to a library function is being considered, @var{fundecl}
3821 will contain an identifier node for the library function. Thus, if
3822 you need to distinguish among various library functions, you can do so
3823 by their names. Note that ``library function'' in this context means
3824 a function used to perform arithmetic, whose name is known specially
3825 in the compiler and was not mentioned in the C code being compiled.
3827 @var{size} is the number of bytes of arguments passed on the
3828 stack. If a variable number of bytes is passed, it is zero, and
3829 argument popping will always be the responsibility of the calling function.
3831 On the VAX, all functions always pop their arguments, so the definition
3832 of this macro is @var{size}. On the 68000, using the standard
3833 calling convention, no functions pop their arguments, so the value of
3834 the macro is always 0 in this case. But an alternative calling
3835 convention is available in which functions that take a fixed number of
3836 arguments pop them but other functions (such as @code{printf}) pop
3837 nothing (the caller pops all). When this convention is in use,
3838 @var{funtype} is examined to determine whether a function takes a fixed
3839 number of arguments.
3842 @defmac CALL_POPS_ARGS (@var{cum})
3843 A C expression that should indicate the number of bytes a call sequence
3844 pops off the stack. It is added to the value of @code{RETURN_POPS_ARGS}
3845 when compiling a function call.
3847 @var{cum} is the variable in which all arguments to the called function
3848 have been accumulated.
3850 On certain architectures, such as the SH5, a call trampoline is used
3851 that pops certain registers off the stack, depending on the arguments
3852 that have been passed to the function. Since this is a property of the
3853 call site, not of the called function, @code{RETURN_POPS_ARGS} is not
3857 @node Register Arguments
3858 @subsection Passing Arguments in Registers
3859 @cindex arguments in registers
3860 @cindex registers arguments
3862 This section describes the macros which let you control how various
3863 types of arguments are passed in registers or how they are arranged in
3866 @deftypefn {Target Hook} rtx TARGET_FUNCTION_ARG (cumulative_args_t @var{ca}, machine_mode @var{mode}, const_tree @var{type}, bool @var{named})
3867 Return an RTX indicating whether a function argument is passed in a
3868 register and if so, which register.
3870 The arguments are @var{ca}, which summarizes all the previous
3871 arguments; @var{mode}, the machine mode of the argument; @var{type},
3872 the data type of the argument as a tree node or 0 if that is not known
3873 (which happens for C support library functions); and @var{named},
3874 which is @code{true} for an ordinary argument and @code{false} for
3875 nameless arguments that correspond to @samp{@dots{}} in the called
3876 function's prototype. @var{type} can be an incomplete type if a
3877 syntax error has previously occurred.
3879 The return value is usually either a @code{reg} RTX for the hard
3880 register in which to pass the argument, or zero to pass the argument
3883 The return value can be a @code{const_int} which means argument is
3884 passed in a target specific slot with specified number. Target hooks
3885 should be used to store or load argument in such case. See
3886 @code{TARGET_STORE_BOUNDS_FOR_ARG} and @code{TARGET_LOAD_BOUNDS_FOR_ARG}
3887 for more information.
3889 The value of the expression can also be a @code{parallel} RTX@. This is
3890 used when an argument is passed in multiple locations. The mode of the
3891 @code{parallel} should be the mode of the entire argument. The
3892 @code{parallel} holds any number of @code{expr_list} pairs; each one
3893 describes where part of the argument is passed. In each
3894 @code{expr_list} the first operand must be a @code{reg} RTX for the hard
3895 register in which to pass this part of the argument, and the mode of the
3896 register RTX indicates how large this part of the argument is. The
3897 second operand of the @code{expr_list} is a @code{const_int} which gives
3898 the offset in bytes into the entire argument of where this part starts.
3899 As a special exception the first @code{expr_list} in the @code{parallel}
3900 RTX may have a first operand of zero. This indicates that the entire
3901 argument is also stored on the stack.
3903 The last time this hook is called, it is called with @code{MODE ==
3904 VOIDmode}, and its result is passed to the @code{call} or @code{call_value}
3905 pattern as operands 2 and 3 respectively.
3907 @cindex @file{stdarg.h} and register arguments
3908 The usual way to make the ISO library @file{stdarg.h} work on a
3909 machine where some arguments are usually passed in registers, is to
3910 cause nameless arguments to be passed on the stack instead. This is
3911 done by making @code{TARGET_FUNCTION_ARG} return 0 whenever
3912 @var{named} is @code{false}.
3914 @cindex @code{TARGET_MUST_PASS_IN_STACK}, and @code{TARGET_FUNCTION_ARG}
3915 @cindex @code{REG_PARM_STACK_SPACE}, and @code{TARGET_FUNCTION_ARG}
3916 You may use the hook @code{targetm.calls.must_pass_in_stack}
3917 in the definition of this macro to determine if this argument is of a
3918 type that must be passed in the stack. If @code{REG_PARM_STACK_SPACE}
3919 is not defined and @code{TARGET_FUNCTION_ARG} returns nonzero for such an
3920 argument, the compiler will abort. If @code{REG_PARM_STACK_SPACE} is
3921 defined, the argument will be computed in the stack and then loaded into
3925 @deftypefn {Target Hook} bool TARGET_MUST_PASS_IN_STACK (machine_mode @var{mode}, const_tree @var{type})
3926 This target hook should return @code{true} if we should not pass @var{type}
3927 solely in registers. The file @file{expr.h} defines a
3928 definition that is usually appropriate, refer to @file{expr.h} for additional
3932 @deftypefn {Target Hook} rtx TARGET_FUNCTION_INCOMING_ARG (cumulative_args_t @var{ca}, machine_mode @var{mode}, const_tree @var{type}, bool @var{named})
3933 Define this hook if the target machine has ``register windows'', so
3934 that the register in which a function sees an arguments is not
3935 necessarily the same as the one in which the caller passed the
3938 For such machines, @code{TARGET_FUNCTION_ARG} computes the register in
3939 which the caller passes the value, and
3940 @code{TARGET_FUNCTION_INCOMING_ARG} should be defined in a similar
3941 fashion to tell the function being called where the arguments will
3944 If @code{TARGET_FUNCTION_INCOMING_ARG} is not defined,
3945 @code{TARGET_FUNCTION_ARG} serves both purposes.
3948 @deftypefn {Target Hook} bool TARGET_USE_PSEUDO_PIC_REG (void)
3949 This hook should return 1 in case pseudo register should be created
3950 for pic_offset_table_rtx during function expand.
3953 @deftypefn {Target Hook} void TARGET_INIT_PIC_REG (void)
3954 Perform a target dependent initialization of pic_offset_table_rtx.
3955 This hook is called at the start of register allocation.
3958 @deftypefn {Target Hook} int TARGET_ARG_PARTIAL_BYTES (cumulative_args_t @var{cum}, machine_mode @var{mode}, tree @var{type}, bool @var{named})
3959 This target hook returns the number of bytes at the beginning of an
3960 argument that must be put in registers. The value must be zero for
3961 arguments that are passed entirely in registers or that are entirely
3962 pushed on the stack.
3964 On some machines, certain arguments must be passed partially in
3965 registers and partially in memory. On these machines, typically the
3966 first few words of arguments are passed in registers, and the rest
3967 on the stack. If a multi-word argument (a @code{double} or a
3968 structure) crosses that boundary, its first few words must be passed
3969 in registers and the rest must be pushed. This macro tells the
3970 compiler when this occurs, and how many bytes should go in registers.
3972 @code{TARGET_FUNCTION_ARG} for these arguments should return the first
3973 register to be used by the caller for this argument; likewise
3974 @code{TARGET_FUNCTION_INCOMING_ARG}, for the called function.
3977 @deftypefn {Target Hook} bool TARGET_PASS_BY_REFERENCE (cumulative_args_t @var{cum}, machine_mode @var{mode}, const_tree @var{type}, bool @var{named})
3978 This target hook should return @code{true} if an argument at the
3979 position indicated by @var{cum} should be passed by reference. This
3980 predicate is queried after target independent reasons for being
3981 passed by reference, such as @code{TREE_ADDRESSABLE (type)}.
3983 If the hook returns true, a copy of that argument is made in memory and a
3984 pointer to the argument is passed instead of the argument itself.
3985 The pointer is passed in whatever way is appropriate for passing a pointer
3989 @deftypefn {Target Hook} bool TARGET_CALLEE_COPIES (cumulative_args_t @var{cum}, machine_mode @var{mode}, const_tree @var{type}, bool @var{named})
3990 The function argument described by the parameters to this hook is
3991 known to be passed by reference. The hook should return true if the
3992 function argument should be copied by the callee instead of copied
3995 For any argument for which the hook returns true, if it can be
3996 determined that the argument is not modified, then a copy need
3999 The default version of this hook always returns false.
4002 @defmac CUMULATIVE_ARGS
4003 A C type for declaring a variable that is used as the first argument
4004 of @code{TARGET_FUNCTION_ARG} and other related values. For some
4005 target machines, the type @code{int} suffices and can hold the number
4006 of bytes of argument so far.
4008 There is no need to record in @code{CUMULATIVE_ARGS} anything about the
4009 arguments that have been passed on the stack. The compiler has other
4010 variables to keep track of that. For target machines on which all
4011 arguments are passed on the stack, there is no need to store anything in
4012 @code{CUMULATIVE_ARGS}; however, the data structure must exist and
4013 should not be empty, so use @code{int}.
4016 @defmac OVERRIDE_ABI_FORMAT (@var{fndecl})
4017 If defined, this macro is called before generating any code for a
4018 function, but after the @var{cfun} descriptor for the function has been
4019 created. The back end may use this macro to update @var{cfun} to
4020 reflect an ABI other than that which would normally be used by default.
4021 If the compiler is generating code for a compiler-generated function,
4022 @var{fndecl} may be @code{NULL}.
4025 @defmac INIT_CUMULATIVE_ARGS (@var{cum}, @var{fntype}, @var{libname}, @var{fndecl}, @var{n_named_args})
4026 A C statement (sans semicolon) for initializing the variable
4027 @var{cum} for the state at the beginning of the argument list. The
4028 variable has type @code{CUMULATIVE_ARGS}. The value of @var{fntype}
4029 is the tree node for the data type of the function which will receive
4030 the args, or 0 if the args are to a compiler support library function.
4031 For direct calls that are not libcalls, @var{fndecl} contain the
4032 declaration node of the function. @var{fndecl} is also set when
4033 @code{INIT_CUMULATIVE_ARGS} is used to find arguments for the function
4034 being compiled. @var{n_named_args} is set to the number of named
4035 arguments, including a structure return address if it is passed as a
4036 parameter, when making a call. When processing incoming arguments,
4037 @var{n_named_args} is set to @minus{}1.
4039 When processing a call to a compiler support library function,
4040 @var{libname} identifies which one. It is a @code{symbol_ref} rtx which
4041 contains the name of the function, as a string. @var{libname} is 0 when
4042 an ordinary C function call is being processed. Thus, each time this
4043 macro is called, either @var{libname} or @var{fntype} is nonzero, but
4044 never both of them at once.
4047 @defmac INIT_CUMULATIVE_LIBCALL_ARGS (@var{cum}, @var{mode}, @var{libname})
4048 Like @code{INIT_CUMULATIVE_ARGS} but only used for outgoing libcalls,
4049 it gets a @code{MODE} argument instead of @var{fntype}, that would be
4050 @code{NULL}. @var{indirect} would always be zero, too. If this macro
4051 is not defined, @code{INIT_CUMULATIVE_ARGS (cum, NULL_RTX, libname,
4052 0)} is used instead.
4055 @defmac INIT_CUMULATIVE_INCOMING_ARGS (@var{cum}, @var{fntype}, @var{libname})
4056 Like @code{INIT_CUMULATIVE_ARGS} but overrides it for the purposes of
4057 finding the arguments for the function being compiled. If this macro is
4058 undefined, @code{INIT_CUMULATIVE_ARGS} is used instead.
4060 The value passed for @var{libname} is always 0, since library routines
4061 with special calling conventions are never compiled with GCC@. The
4062 argument @var{libname} exists for symmetry with
4063 @code{INIT_CUMULATIVE_ARGS}.
4064 @c could use "this macro" in place of @code{INIT_CUMULATIVE_ARGS}, maybe.
4065 @c --mew 5feb93 i switched the order of the sentences. --mew 10feb93
4068 @deftypefn {Target Hook} void TARGET_FUNCTION_ARG_ADVANCE (cumulative_args_t @var{ca}, machine_mode @var{mode}, const_tree @var{type}, bool @var{named})
4069 This hook updates the summarizer variable pointed to by @var{ca} to
4070 advance past an argument in the argument list. The values @var{mode},
4071 @var{type} and @var{named} describe that argument. Once this is done,
4072 the variable @var{cum} is suitable for analyzing the @emph{following}
4073 argument with @code{TARGET_FUNCTION_ARG}, etc.
4075 This hook need not do anything if the argument in question was passed
4076 on the stack. The compiler knows how to track the amount of stack space
4077 used for arguments without any special help.
4080 @defmac FUNCTION_ARG_OFFSET (@var{mode}, @var{type})
4081 If defined, a C expression that is the number of bytes to add to the
4082 offset of the argument passed in memory. This is needed for the SPU,
4083 which passes @code{char} and @code{short} arguments in the preferred
4084 slot that is in the middle of the quad word instead of starting at the
4088 @defmac FUNCTION_ARG_PADDING (@var{mode}, @var{type})
4089 If defined, a C expression which determines whether, and in which direction,
4090 to pad out an argument with extra space. The value should be of type
4091 @code{enum direction}: either @code{upward} to pad above the argument,
4092 @code{downward} to pad below, or @code{none} to inhibit padding.
4094 The @emph{amount} of padding is not controlled by this macro, but by the
4095 target hook @code{TARGET_FUNCTION_ARG_ROUND_BOUNDARY}. It is
4096 always just enough to reach the next multiple of that boundary.
4098 This macro has a default definition which is right for most systems.
4099 For little-endian machines, the default is to pad upward. For
4100 big-endian machines, the default is to pad downward for an argument of
4101 constant size shorter than an @code{int}, and upward otherwise.
4104 @defmac PAD_VARARGS_DOWN
4105 If defined, a C expression which determines whether the default
4106 implementation of va_arg will attempt to pad down before reading the
4107 next argument, if that argument is smaller than its aligned space as
4108 controlled by @code{PARM_BOUNDARY}. If this macro is not defined, all such
4109 arguments are padded down if @code{BYTES_BIG_ENDIAN} is true.
4112 @defmac BLOCK_REG_PADDING (@var{mode}, @var{type}, @var{first})
4113 Specify padding for the last element of a block move between registers and
4114 memory. @var{first} is nonzero if this is the only element. Defining this
4115 macro allows better control of register function parameters on big-endian
4116 machines, without using @code{PARALLEL} rtl. In particular,
4117 @code{MUST_PASS_IN_STACK} need not test padding and mode of types in
4118 registers, as there is no longer a "wrong" part of a register; For example,
4119 a three byte aggregate may be passed in the high part of a register if so
4123 @deftypefn {Target Hook} {unsigned int} TARGET_FUNCTION_ARG_BOUNDARY (machine_mode @var{mode}, const_tree @var{type})
4124 This hook returns the alignment boundary, in bits, of an argument
4125 with the specified mode and type. The default hook returns
4126 @code{PARM_BOUNDARY} for all arguments.
4129 @deftypefn {Target Hook} {unsigned int} TARGET_FUNCTION_ARG_ROUND_BOUNDARY (machine_mode @var{mode}, const_tree @var{type})
4130 Normally, the size of an argument is rounded up to @code{PARM_BOUNDARY},
4131 which is the default value for this hook. You can define this hook to
4132 return a different value if an argument size must be rounded to a larger
4136 @defmac FUNCTION_ARG_REGNO_P (@var{regno})
4137 A C expression that is nonzero if @var{regno} is the number of a hard
4138 register in which function arguments are sometimes passed. This does
4139 @emph{not} include implicit arguments such as the static chain and
4140 the structure-value address. On many machines, no registers can be
4141 used for this purpose since all function arguments are pushed on the
4145 @deftypefn {Target Hook} bool TARGET_SPLIT_COMPLEX_ARG (const_tree @var{type})
4146 This hook should return true if parameter of type @var{type} are passed
4147 as two scalar parameters. By default, GCC will attempt to pack complex
4148 arguments into the target's word size. Some ABIs require complex arguments
4149 to be split and treated as their individual components. For example, on
4150 AIX64, complex floats should be passed in a pair of floating point
4151 registers, even though a complex float would fit in one 64-bit floating
4154 The default value of this hook is @code{NULL}, which is treated as always
4158 @deftypefn {Target Hook} tree TARGET_BUILD_BUILTIN_VA_LIST (void)
4159 This hook returns a type node for @code{va_list} for the target.
4160 The default version of the hook returns @code{void*}.
4163 @deftypefn {Target Hook} int TARGET_ENUM_VA_LIST_P (int @var{idx}, const char **@var{pname}, tree *@var{ptree})
4164 This target hook is used in function @code{c_common_nodes_and_builtins}
4165 to iterate through the target specific builtin types for va_list. The
4166 variable @var{idx} is used as iterator. @var{pname} has to be a pointer
4167 to a @code{const char *} and @var{ptree} a pointer to a @code{tree} typed
4169 The arguments @var{pname} and @var{ptree} are used to store the result of
4170 this macro and are set to the name of the va_list builtin type and its
4172 If the return value of this macro is zero, then there is no more element.
4173 Otherwise the @var{IDX} should be increased for the next call of this
4174 macro to iterate through all types.
4177 @deftypefn {Target Hook} tree TARGET_FN_ABI_VA_LIST (tree @var{fndecl})
4178 This hook returns the va_list type of the calling convention specified by
4180 The default version of this hook returns @code{va_list_type_node}.
4183 @deftypefn {Target Hook} tree TARGET_CANONICAL_VA_LIST_TYPE (tree @var{type})
4184 This hook returns the va_list type of the calling convention specified by the
4185 type of @var{type}. If @var{type} is not a valid va_list type, it returns
4189 @deftypefn {Target Hook} tree TARGET_GIMPLIFY_VA_ARG_EXPR (tree @var{valist}, tree @var{type}, gimple_seq *@var{pre_p}, gimple_seq *@var{post_p})
4190 This hook performs target-specific gimplification of
4191 @code{VA_ARG_EXPR}. The first two parameters correspond to the
4192 arguments to @code{va_arg}; the latter two are as in
4193 @code{gimplify.c:gimplify_expr}.
4196 @deftypefn {Target Hook} bool TARGET_VALID_POINTER_MODE (machine_mode @var{mode})
4197 Define this to return nonzero if the port can handle pointers
4198 with machine mode @var{mode}. The default version of this
4199 hook returns true for both @code{ptr_mode} and @code{Pmode}.
4202 @deftypefn {Target Hook} bool TARGET_REF_MAY_ALIAS_ERRNO (struct ao_ref *@var{ref})
4203 Define this to return nonzero if the memory reference @var{ref} may alias with the system C library errno location. The default version of this hook assumes the system C library errno location is either a declaration of type int or accessed by dereferencing a pointer to int.
4206 @deftypefn {Target Hook} bool TARGET_SCALAR_MODE_SUPPORTED_P (machine_mode @var{mode})
4207 Define this to return nonzero if the port is prepared to handle
4208 insns involving scalar mode @var{mode}. For a scalar mode to be
4209 considered supported, all the basic arithmetic and comparisons
4212 The default version of this hook returns true for any mode
4213 required to handle the basic C types (as defined by the port).
4214 Included here are the double-word arithmetic supported by the
4215 code in @file{optabs.c}.
4218 @deftypefn {Target Hook} bool TARGET_VECTOR_MODE_SUPPORTED_P (machine_mode @var{mode})
4219 Define this to return nonzero if the port is prepared to handle
4220 insns involving vector mode @var{mode}. At the very least, it
4221 must have move patterns for this mode.
4224 @deftypefn {Target Hook} bool TARGET_ARRAY_MODE_SUPPORTED_P (machine_mode @var{mode}, unsigned HOST_WIDE_INT @var{nelems})
4225 Return true if GCC should try to use a scalar mode to store an array
4226 of @var{nelems} elements, given that each element has mode @var{mode}.
4227 Returning true here overrides the usual @code{MAX_FIXED_MODE} limit
4228 and allows GCC to use any defined integer mode.
4230 One use of this hook is to support vector load and store operations
4231 that operate on several homogeneous vectors. For example, ARM NEON
4232 has operations like:
4235 int8x8x3_t vld3_s8 (const int8_t *)
4238 where the return type is defined as:
4241 typedef struct int8x8x3_t
4247 If this hook allows @code{val} to have a scalar mode, then
4248 @code{int8x8x3_t} can have the same mode. GCC can then store
4249 @code{int8x8x3_t}s in registers rather than forcing them onto the stack.
4252 @deftypefn {Target Hook} bool TARGET_LIBGCC_FLOATING_MODE_SUPPORTED_P (machine_mode @var{mode})
4253 Define this to return nonzero if libgcc provides support for the
4254 floating-point mode @var{mode}, which is known to pass
4255 @code{TARGET_SCALAR_MODE_SUPPORTED_P}. The default version of this
4256 hook returns true for all of @code{SFmode}, @code{DFmode},
4257 @code{XFmode} and @code{TFmode}, if such modes exist.
4260 @deftypefn {Target Hook} bool TARGET_SMALL_REGISTER_CLASSES_FOR_MODE_P (machine_mode @var{mode})
4261 Define this to return nonzero for machine modes for which the port has
4262 small register classes. If this target hook returns nonzero for a given
4263 @var{mode}, the compiler will try to minimize the lifetime of registers
4264 in @var{mode}. The hook may be called with @code{VOIDmode} as argument.
4265 In this case, the hook is expected to return nonzero if it returns nonzero
4268 On some machines, it is risky to let hard registers live across arbitrary
4269 insns. Typically, these machines have instructions that require values
4270 to be in specific registers (like an accumulator), and reload will fail
4271 if the required hard register is used for another purpose across such an
4274 Passes before reload do not know which hard registers will be used
4275 in an instruction, but the machine modes of the registers set or used in
4276 the instruction are already known. And for some machines, register
4277 classes are small for, say, integer registers but not for floating point
4278 registers. For example, the AMD x86-64 architecture requires specific
4279 registers for the legacy x86 integer instructions, but there are many
4280 SSE registers for floating point operations. On such targets, a good
4281 strategy may be to return nonzero from this hook for @code{INTEGRAL_MODE_P}
4282 machine modes but zero for the SSE register classes.
4284 The default version of this hook returns false for any mode. It is always
4285 safe to redefine this hook to return with a nonzero value. But if you
4286 unnecessarily define it, you will reduce the amount of optimizations
4287 that can be performed in some cases. If you do not define this hook
4288 to return a nonzero value when it is required, the compiler will run out
4289 of spill registers and print a fatal error message.
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 @deftypefn {Target Hook} rtx TARGET_FUNCTION_VALUE (const_tree @var{ret_type}, const_tree @var{fn_decl_or_type}, bool @var{outgoing})
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 @deftypefn {Target Hook} rtx TARGET_LIBCALL_VALUE (machine_mode @var{mode}, const_rtx @var{fun})
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 @deftypefn {Target Hook} bool TARGET_FUNCTION_VALUE_REGNO_P (const unsigned int @var{regno})
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 @deftypevr {Target Hook} bool TARGET_OMIT_STRUCT_RETURN_REG
4414 Normally, when a function returns a structure by memory, the address
4415 is passed as an invisible pointer argument, but the compiler also
4416 arranges to return the address from the function like it would a normal
4417 pointer return value. Define this to true if that behaviour is
4418 undesirable on your target.
4421 @deftypefn {Target Hook} bool TARGET_RETURN_IN_MSB (const_tree @var{type})
4422 This hook should return true if values of type @var{type} are returned
4423 at the most significant end of a register (in other words, if they are
4424 padded at the least significant end). You can assume that @var{type}
4425 is returned in a register; the caller is required to check this.
4427 Note that the register provided by @code{TARGET_FUNCTION_VALUE} must
4428 be able to hold the complete return value. For example, if a 1-, 2-
4429 or 3-byte structure is returned at the most significant end of a
4430 4-byte register, @code{TARGET_FUNCTION_VALUE} should provide an
4434 @node Aggregate Return
4435 @subsection How Large Values Are Returned
4436 @cindex aggregates as return values
4437 @cindex large return values
4438 @cindex returning aggregate values
4439 @cindex structure value address
4441 When a function value's mode is @code{BLKmode} (and in some other
4442 cases), the value is not returned according to
4443 @code{TARGET_FUNCTION_VALUE} (@pxref{Scalar Return}). Instead, the
4444 caller passes the address of a block of memory in which the value
4445 should be stored. This address is called the @dfn{structure value
4448 This section describes how to control returning structure values in
4451 @deftypefn {Target Hook} bool TARGET_RETURN_IN_MEMORY (const_tree @var{type}, const_tree @var{fntype})
4452 This target hook should return a nonzero value to say to return the
4453 function value in memory, just as large structures are always returned.
4454 Here @var{type} will be the data type of the value, and @var{fntype}
4455 will be the type of the function doing the returning, or @code{NULL} for
4458 Note that values of mode @code{BLKmode} must be explicitly handled
4459 by this function. Also, the option @option{-fpcc-struct-return}
4460 takes effect regardless of this macro. On most systems, it is
4461 possible to leave the hook undefined; this causes a default
4462 definition to be used, whose value is the constant 1 for @code{BLKmode}
4463 values, and 0 otherwise.
4465 Do not use this hook to indicate that structures and unions should always
4466 be returned in memory. You should instead use @code{DEFAULT_PCC_STRUCT_RETURN}
4470 @defmac DEFAULT_PCC_STRUCT_RETURN
4471 Define this macro to be 1 if all structure and union return values must be
4472 in memory. Since this results in slower code, this should be defined
4473 only if needed for compatibility with other compilers or with an ABI@.
4474 If you define this macro to be 0, then the conventions used for structure
4475 and union return values are decided by the @code{TARGET_RETURN_IN_MEMORY}
4478 If not defined, this defaults to the value 1.
4481 @deftypefn {Target Hook} rtx TARGET_STRUCT_VALUE_RTX (tree @var{fndecl}, int @var{incoming})
4482 This target hook should return the location of the structure value
4483 address (normally a @code{mem} or @code{reg}), or 0 if the address is
4484 passed as an ``invisible'' first argument. Note that @var{fndecl} may
4485 be @code{NULL}, for libcalls. You do not need to define this target
4486 hook if the address is always passed as an ``invisible'' first
4489 On some architectures the place where the structure value address
4490 is found by the called function is not the same place that the
4491 caller put it. This can be due to register windows, or it could
4492 be because the function prologue moves it to a different place.
4493 @var{incoming} is @code{1} or @code{2} when the location is needed in
4494 the context of the called function, and @code{0} in the context of
4497 If @var{incoming} is nonzero and the address is to be found on the
4498 stack, return a @code{mem} which refers to the frame pointer. If
4499 @var{incoming} is @code{2}, the result is being used to fetch the
4500 structure value address at the beginning of a function. If you need
4501 to emit adjusting code, you should do it at this point.
4504 @defmac PCC_STATIC_STRUCT_RETURN
4505 Define this macro if the usual system convention on the target machine
4506 for returning structures and unions is for the called function to return
4507 the address of a static variable containing the value.
4509 Do not define this if the usual system convention is for the caller to
4510 pass an address to the subroutine.
4512 This macro has effect in @option{-fpcc-struct-return} mode, but it does
4513 nothing when you use @option{-freg-struct-return} mode.
4516 @deftypefn {Target Hook} machine_mode TARGET_GET_RAW_RESULT_MODE (int @var{regno})
4517 This target hook returns the mode to be used when accessing raw return registers in @code{__builtin_return}. Define this macro if the value in @var{reg_raw_mode} is not correct.
4520 @deftypefn {Target Hook} machine_mode TARGET_GET_RAW_ARG_MODE (int @var{regno})
4521 This target hook returns the mode to be used when accessing raw argument registers in @code{__builtin_apply_args}. Define this macro if the value in @var{reg_raw_mode} is not correct.
4525 @subsection Caller-Saves Register Allocation
4527 If you enable it, GCC can save registers around function calls. This
4528 makes it possible to use call-clobbered registers to hold variables that
4529 must live across calls.
4531 @defmac HARD_REGNO_CALLER_SAVE_MODE (@var{regno}, @var{nregs})
4532 A C expression specifying which mode is required for saving @var{nregs}
4533 of a pseudo-register in call-clobbered hard register @var{regno}. If
4534 @var{regno} is unsuitable for caller save, @code{VOIDmode} should be
4535 returned. For most machines this macro need not be defined since GCC
4536 will select the smallest suitable mode.
4539 @node Function Entry
4540 @subsection Function Entry and Exit
4541 @cindex function entry and exit
4545 This section describes the macros that output function entry
4546 (@dfn{prologue}) and exit (@dfn{epilogue}) code.
4548 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_PROLOGUE (FILE *@var{file}, HOST_WIDE_INT @var{size})
4549 If defined, a function that outputs the assembler code for entry to a
4550 function. The prologue is responsible for setting up the stack frame,
4551 initializing the frame pointer register, saving registers that must be
4552 saved, and allocating @var{size} additional bytes of storage for the
4553 local variables. @var{size} is an integer. @var{file} is a stdio
4554 stream to which the assembler code should be output.
4556 The label for the beginning of the function need not be output by this
4557 macro. That has already been done when the macro is run.
4559 @findex regs_ever_live
4560 To determine which registers to save, the macro can refer to the array
4561 @code{regs_ever_live}: element @var{r} is nonzero if hard register
4562 @var{r} is used anywhere within the function. This implies the function
4563 prologue should save register @var{r}, provided it is not one of the
4564 call-used registers. (@code{TARGET_ASM_FUNCTION_EPILOGUE} must likewise use
4565 @code{regs_ever_live}.)
4567 On machines that have ``register windows'', the function entry code does
4568 not save on the stack the registers that are in the windows, even if
4569 they are supposed to be preserved by function calls; instead it takes
4570 appropriate steps to ``push'' the register stack, if any non-call-used
4571 registers are used in the function.
4573 @findex frame_pointer_needed
4574 On machines where functions may or may not have frame-pointers, the
4575 function entry code must vary accordingly; it must set up the frame
4576 pointer if one is wanted, and not otherwise. To determine whether a
4577 frame pointer is in wanted, the macro can refer to the variable
4578 @code{frame_pointer_needed}. The variable's value will be 1 at run
4579 time in a function that needs a frame pointer. @xref{Elimination}.
4581 The function entry code is responsible for allocating any stack space
4582 required for the function. This stack space consists of the regions
4583 listed below. In most cases, these regions are allocated in the
4584 order listed, with the last listed region closest to the top of the
4585 stack (the lowest address if @code{STACK_GROWS_DOWNWARD} is defined, and
4586 the highest address if it is not defined). You can use a different order
4587 for a machine if doing so is more convenient or required for
4588 compatibility reasons. Except in cases where required by standard
4589 or by a debugger, there is no reason why the stack layout used by GCC
4590 need agree with that used by other compilers for a machine.
4593 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_END_PROLOGUE (FILE *@var{file})
4594 If defined, a function that outputs assembler code at the end of a
4595 prologue. This should be used when the function prologue is being
4596 emitted as RTL, and you have some extra assembler that needs to be
4597 emitted. @xref{prologue instruction pattern}.
4600 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_BEGIN_EPILOGUE (FILE *@var{file})
4601 If defined, a function that outputs assembler code at the start of an
4602 epilogue. This should be used when the function epilogue is being
4603 emitted as RTL, and you have some extra assembler that needs to be
4604 emitted. @xref{epilogue instruction pattern}.
4607 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_EPILOGUE (FILE *@var{file}, HOST_WIDE_INT @var{size})
4608 If defined, a function that outputs the assembler code for exit from a
4609 function. The epilogue is responsible for restoring the saved
4610 registers and stack pointer to their values when the function was
4611 called, and returning control to the caller. This macro takes the
4612 same arguments as the macro @code{TARGET_ASM_FUNCTION_PROLOGUE}, and the
4613 registers to restore are determined from @code{regs_ever_live} and
4614 @code{CALL_USED_REGISTERS} in the same way.
4616 On some machines, there is a single instruction that does all the work
4617 of returning from the function. On these machines, give that
4618 instruction the name @samp{return} and do not define the macro
4619 @code{TARGET_ASM_FUNCTION_EPILOGUE} at all.
4621 Do not define a pattern named @samp{return} if you want the
4622 @code{TARGET_ASM_FUNCTION_EPILOGUE} to be used. If you want the target
4623 switches to control whether return instructions or epilogues are used,
4624 define a @samp{return} pattern with a validity condition that tests the
4625 target switches appropriately. If the @samp{return} pattern's validity
4626 condition is false, epilogues will be used.
4628 On machines where functions may or may not have frame-pointers, the
4629 function exit code must vary accordingly. Sometimes the code for these
4630 two cases is completely different. To determine whether a frame pointer
4631 is wanted, the macro can refer to the variable
4632 @code{frame_pointer_needed}. The variable's value will be 1 when compiling
4633 a function that needs a frame pointer.
4635 Normally, @code{TARGET_ASM_FUNCTION_PROLOGUE} and
4636 @code{TARGET_ASM_FUNCTION_EPILOGUE} must treat leaf functions specially.
4637 The C variable @code{current_function_is_leaf} is nonzero for such a
4638 function. @xref{Leaf Functions}.
4640 On some machines, some functions pop their arguments on exit while
4641 others leave that for the caller to do. For example, the 68020 when
4642 given @option{-mrtd} pops arguments in functions that take a fixed
4643 number of arguments.
4646 @findex crtl->args.pops_args
4647 Your definition of the macro @code{RETURN_POPS_ARGS} decides which
4648 functions pop their own arguments. @code{TARGET_ASM_FUNCTION_EPILOGUE}
4649 needs to know what was decided. The number of bytes of the current
4650 function's arguments that this function should pop is available in
4651 @code{crtl->args.pops_args}. @xref{Scalar Return}.
4656 @findex pretend_args_size
4657 @findex crtl->args.pretend_args_size
4658 A region of @code{crtl->args.pretend_args_size} bytes of
4659 uninitialized space just underneath the first argument arriving on the
4660 stack. (This may not be at the very start of the allocated stack region
4661 if the calling sequence has pushed anything else since pushing the stack
4662 arguments. But usually, on such machines, nothing else has been pushed
4663 yet, because the function prologue itself does all the pushing.) This
4664 region is used on machines where an argument may be passed partly in
4665 registers and partly in memory, and, in some cases to support the
4666 features in @code{<stdarg.h>}.
4669 An area of memory used to save certain registers used by the function.
4670 The size of this area, which may also include space for such things as
4671 the return address and pointers to previous stack frames, is
4672 machine-specific and usually depends on which registers have been used
4673 in the function. Machines with register windows often do not require
4677 A region of at least @var{size} bytes, possibly rounded up to an allocation
4678 boundary, to contain the local variables of the function. On some machines,
4679 this region and the save area may occur in the opposite order, with the
4680 save area closer to the top of the stack.
4683 @cindex @code{ACCUMULATE_OUTGOING_ARGS} and stack frames
4684 Optionally, when @code{ACCUMULATE_OUTGOING_ARGS} is defined, a region of
4685 @code{crtl->outgoing_args_size} bytes to be used for outgoing
4686 argument lists of the function. @xref{Stack Arguments}.
4689 @defmac EXIT_IGNORE_STACK
4690 Define this macro as a C expression that is nonzero if the return
4691 instruction or the function epilogue ignores the value of the stack
4692 pointer; in other words, if it is safe to delete an instruction to
4693 adjust the stack pointer before a return from the function. The
4696 Note that this macro's value is relevant only for functions for which
4697 frame pointers are maintained. It is never safe to delete a final
4698 stack adjustment in a function that has no frame pointer, and the
4699 compiler knows this regardless of @code{EXIT_IGNORE_STACK}.
4702 @defmac EPILOGUE_USES (@var{regno})
4703 Define this macro as a C expression that is nonzero for registers that are
4704 used by the epilogue or the @samp{return} pattern. The stack and frame
4705 pointer registers are already assumed to be used as needed.
4708 @defmac EH_USES (@var{regno})
4709 Define this macro as a C expression that is nonzero for registers that are
4710 used by the exception handling mechanism, and so should be considered live
4711 on entry to an exception edge.
4714 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_MI_THUNK (FILE *@var{file}, tree @var{thunk_fndecl}, HOST_WIDE_INT @var{delta}, HOST_WIDE_INT @var{vcall_offset}, tree @var{function})
4715 A function that outputs the assembler code for a thunk
4716 function, used to implement C++ virtual function calls with multiple
4717 inheritance. The thunk acts as a wrapper around a virtual function,
4718 adjusting the implicit object parameter before handing control off to
4721 First, emit code to add the integer @var{delta} to the location that
4722 contains the incoming first argument. Assume that this argument
4723 contains a pointer, and is the one used to pass the @code{this} pointer
4724 in C++. This is the incoming argument @emph{before} the function prologue,
4725 e.g.@: @samp{%o0} on a sparc. The addition must preserve the values of
4726 all other incoming arguments.
4728 Then, if @var{vcall_offset} is nonzero, an additional adjustment should be
4729 made after adding @code{delta}. In particular, if @var{p} is the
4730 adjusted pointer, the following adjustment should be made:
4733 p += (*((ptrdiff_t **)p))[vcall_offset/sizeof(ptrdiff_t)]
4736 After the additions, emit code to jump to @var{function}, which is a
4737 @code{FUNCTION_DECL}. This is a direct pure jump, not a call, and does
4738 not touch the return address. Hence returning from @var{FUNCTION} will
4739 return to whoever called the current @samp{thunk}.
4741 The effect must be as if @var{function} had been called directly with
4742 the adjusted first argument. This macro is responsible for emitting all
4743 of the code for a thunk function; @code{TARGET_ASM_FUNCTION_PROLOGUE}
4744 and @code{TARGET_ASM_FUNCTION_EPILOGUE} are not invoked.
4746 The @var{thunk_fndecl} is redundant. (@var{delta} and @var{function}
4747 have already been extracted from it.) It might possibly be useful on
4748 some targets, but probably not.
4750 If you do not define this macro, the target-independent code in the C++
4751 front end will generate a less efficient heavyweight thunk that calls
4752 @var{function} instead of jumping to it. The generic approach does
4753 not support varargs.
4756 @deftypefn {Target Hook} bool TARGET_ASM_CAN_OUTPUT_MI_THUNK (const_tree @var{thunk_fndecl}, HOST_WIDE_INT @var{delta}, HOST_WIDE_INT @var{vcall_offset}, const_tree @var{function})
4757 A function that returns true if TARGET_ASM_OUTPUT_MI_THUNK would be able
4758 to output the assembler code for the thunk function specified by the
4759 arguments it is passed, and false otherwise. In the latter case, the
4760 generic approach will be used by the C++ front end, with the limitations
4765 @subsection Generating Code for Profiling
4766 @cindex profiling, code generation
4768 These macros will help you generate code for profiling.
4770 @defmac FUNCTION_PROFILER (@var{file}, @var{labelno})
4771 A C statement or compound statement to output to @var{file} some
4772 assembler code to call the profiling subroutine @code{mcount}.
4775 The details of how @code{mcount} expects to be called are determined by
4776 your operating system environment, not by GCC@. To figure them out,
4777 compile a small program for profiling using the system's installed C
4778 compiler and look at the assembler code that results.
4780 Older implementations of @code{mcount} expect the address of a counter
4781 variable to be loaded into some register. The name of this variable is
4782 @samp{LP} followed by the number @var{labelno}, so you would generate
4783 the name using @samp{LP%d} in a @code{fprintf}.
4786 @defmac PROFILE_HOOK
4787 A C statement or compound statement to output to @var{file} some assembly
4788 code to call the profiling subroutine @code{mcount} even the target does
4789 not support profiling.
4792 @defmac NO_PROFILE_COUNTERS
4793 Define this macro to be an expression with a nonzero value if the
4794 @code{mcount} subroutine on your system does not need a counter variable
4795 allocated for each function. This is true for almost all modern
4796 implementations. If you define this macro, you must not use the
4797 @var{labelno} argument to @code{FUNCTION_PROFILER}.
4800 @defmac PROFILE_BEFORE_PROLOGUE
4801 Define this macro if the code for function profiling should come before
4802 the function prologue. Normally, the profiling code comes after.
4805 @deftypefn {Target Hook} bool TARGET_KEEP_LEAF_WHEN_PROFILED (void)
4806 This target hook returns true if the target wants the leaf flag for the current function to stay true even if it calls mcount. This might make sense for targets using the leaf flag only to determine whether a stack frame needs to be generated or not and for which the call to mcount is generated before the function prologue.
4810 @subsection Permitting tail calls
4813 @deftypefn {Target Hook} bool TARGET_FUNCTION_OK_FOR_SIBCALL (tree @var{decl}, tree @var{exp})
4814 True if it is OK to do sibling call optimization for the specified
4815 call expression @var{exp}. @var{decl} will be the called function,
4816 or @code{NULL} if this is an indirect call.
4818 It is not uncommon for limitations of calling conventions to prevent
4819 tail calls to functions outside the current unit of translation, or
4820 during PIC compilation. The hook is used to enforce these restrictions,
4821 as the @code{sibcall} md pattern can not fail, or fall over to a
4822 ``normal'' call. The criteria for successful sibling call optimization
4823 may vary greatly between different architectures.
4826 @deftypefn {Target Hook} void TARGET_EXTRA_LIVE_ON_ENTRY (bitmap @var{regs})
4827 Add any hard registers to @var{regs} that are live on entry to the
4828 function. This hook only needs to be defined to provide registers that
4829 cannot be found by examination of FUNCTION_ARG_REGNO_P, the callee saved
4830 registers, STATIC_CHAIN_INCOMING_REGNUM, STATIC_CHAIN_REGNUM,
4831 TARGET_STRUCT_VALUE_RTX, FRAME_POINTER_REGNUM, EH_USES,
4832 FRAME_POINTER_REGNUM, ARG_POINTER_REGNUM, and the PIC_OFFSET_TABLE_REGNUM.
4835 @deftypefn {Target Hook} void TARGET_SET_UP_BY_PROLOGUE (struct hard_reg_set_container *@var{})
4836 This hook should add additional registers that are computed by the prologue to the hard regset for shrink-wrapping optimization purposes.
4839 @deftypefn {Target Hook} bool TARGET_WARN_FUNC_RETURN (tree)
4840 True if a function's return statements should be checked for matching the function's return type. This includes checking for falling off the end of a non-void function. Return false if no such check should be made.
4843 @node Stack Smashing Protection
4844 @subsection Stack smashing protection
4845 @cindex stack smashing protection
4847 @deftypefn {Target Hook} tree TARGET_STACK_PROTECT_GUARD (void)
4848 This hook returns a @code{DECL} node for the external variable to use
4849 for the stack protection guard. This variable is initialized by the
4850 runtime to some random value and is used to initialize the guard value
4851 that is placed at the top of the local stack frame. The type of this
4852 variable must be @code{ptr_type_node}.
4854 The default version of this hook creates a variable called
4855 @samp{__stack_chk_guard}, which is normally defined in @file{libgcc2.c}.
4858 @deftypefn {Target Hook} tree TARGET_STACK_PROTECT_FAIL (void)
4859 This hook returns a @code{CALL_EXPR} that alerts the runtime that the
4860 stack protect guard variable has been modified. This expression should
4861 involve a call to a @code{noreturn} function.
4863 The default version of this hook invokes a function called
4864 @samp{__stack_chk_fail}, taking no arguments. This function is
4865 normally defined in @file{libgcc2.c}.
4868 @deftypefn {Common Target Hook} bool TARGET_SUPPORTS_SPLIT_STACK (bool @var{report}, struct gcc_options *@var{opts})
4869 Whether this target supports splitting the stack when the options described in @var{opts} have been passed. This is called after options have been parsed, so the target may reject splitting the stack in some configurations. The default version of this hook returns false. If @var{report} is true, this function may issue a warning or error; if @var{report} is false, it must simply return a value
4872 @node Miscellaneous Register Hooks
4873 @subsection Miscellaneous register hooks
4874 @cindex miscellaneous register hooks
4876 @deftypevr {Target Hook} bool TARGET_CALL_FUSAGE_CONTAINS_NON_CALLEE_CLOBBERS
4877 Set to true if each call that binds to a local definition explicitly
4878 clobbers or sets all non-fixed registers modified by performing the call.
4879 That is, by the call pattern itself, or by code that might be inserted by the
4880 linker (e.g. stubs, veneers, branch islands), but not including those
4881 modifiable by the callee. The affected registers may be mentioned explicitly
4882 in the call pattern, or included as clobbers in CALL_INSN_FUNCTION_USAGE.
4883 The default version of this hook is set to false. The purpose of this hook
4884 is to enable the fipa-ra optimization.
4888 @section Implementing the Varargs Macros
4889 @cindex varargs implementation
4891 GCC comes with an implementation of @code{<varargs.h>} and
4892 @code{<stdarg.h>} that work without change on machines that pass arguments
4893 on the stack. Other machines require their own implementations of
4894 varargs, and the two machine independent header files must have
4895 conditionals to include it.
4897 ISO @code{<stdarg.h>} differs from traditional @code{<varargs.h>} mainly in
4898 the calling convention for @code{va_start}. The traditional
4899 implementation takes just one argument, which is the variable in which
4900 to store the argument pointer. The ISO implementation of
4901 @code{va_start} takes an additional second argument. The user is
4902 supposed to write the last named argument of the function here.
4904 However, @code{va_start} should not use this argument. The way to find
4905 the end of the named arguments is with the built-in functions described
4908 @defmac __builtin_saveregs ()
4909 Use this built-in function to save the argument registers in memory so
4910 that the varargs mechanism can access them. Both ISO and traditional
4911 versions of @code{va_start} must use @code{__builtin_saveregs}, unless
4912 you use @code{TARGET_SETUP_INCOMING_VARARGS} (see below) instead.
4914 On some machines, @code{__builtin_saveregs} is open-coded under the
4915 control of the target hook @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. On
4916 other machines, it calls a routine written in assembler language,
4917 found in @file{libgcc2.c}.
4919 Code generated for the call to @code{__builtin_saveregs} appears at the
4920 beginning of the function, as opposed to where the call to
4921 @code{__builtin_saveregs} is written, regardless of what the code is.
4922 This is because the registers must be saved before the function starts
4923 to use them for its own purposes.
4924 @c i rewrote the first sentence above to fix an overfull hbox. --mew
4928 @defmac __builtin_next_arg (@var{lastarg})
4929 This builtin returns the address of the first anonymous stack
4930 argument, as type @code{void *}. If @code{ARGS_GROW_DOWNWARD}, it
4931 returns the address of the location above the first anonymous stack
4932 argument. Use it in @code{va_start} to initialize the pointer for
4933 fetching arguments from the stack. Also use it in @code{va_start} to
4934 verify that the second parameter @var{lastarg} is the last named argument
4935 of the current function.
4938 @defmac __builtin_classify_type (@var{object})
4939 Since each machine has its own conventions for which data types are
4940 passed in which kind of register, your implementation of @code{va_arg}
4941 has to embody these conventions. The easiest way to categorize the
4942 specified data type is to use @code{__builtin_classify_type} together
4943 with @code{sizeof} and @code{__alignof__}.
4945 @code{__builtin_classify_type} ignores the value of @var{object},
4946 considering only its data type. It returns an integer describing what
4947 kind of type that is---integer, floating, pointer, structure, and so on.
4949 The file @file{typeclass.h} defines an enumeration that you can use to
4950 interpret the values of @code{__builtin_classify_type}.
4953 These machine description macros help implement varargs:
4955 @deftypefn {Target Hook} rtx TARGET_EXPAND_BUILTIN_SAVEREGS (void)
4956 If defined, this hook produces the machine-specific code for a call to
4957 @code{__builtin_saveregs}. This code will be moved to the very
4958 beginning of the function, before any parameter access are made. The
4959 return value of this function should be an RTX that contains the value
4960 to use as the return of @code{__builtin_saveregs}.
4963 @deftypefn {Target Hook} void TARGET_SETUP_INCOMING_VARARGS (cumulative_args_t @var{args_so_far}, machine_mode @var{mode}, tree @var{type}, int *@var{pretend_args_size}, int @var{second_time})
4964 This target hook offers an alternative to using
4965 @code{__builtin_saveregs} and defining the hook
4966 @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. Use it to store the anonymous
4967 register arguments into the stack so that all the arguments appear to
4968 have been passed consecutively on the stack. Once this is done, you can
4969 use the standard implementation of varargs that works for machines that
4970 pass all their arguments on the stack.
4972 The argument @var{args_so_far} points to the @code{CUMULATIVE_ARGS} data
4973 structure, containing the values that are obtained after processing the
4974 named arguments. The arguments @var{mode} and @var{type} describe the
4975 last named argument---its machine mode and its data type as a tree node.
4977 The target hook should do two things: first, push onto the stack all the
4978 argument registers @emph{not} used for the named arguments, and second,
4979 store the size of the data thus pushed into the @code{int}-valued
4980 variable pointed to by @var{pretend_args_size}. The value that you
4981 store here will serve as additional offset for setting up the stack
4984 Because you must generate code to push the anonymous arguments at
4985 compile time without knowing their data types,
4986 @code{TARGET_SETUP_INCOMING_VARARGS} is only useful on machines that
4987 have just a single category of argument register and use it uniformly
4990 If the argument @var{second_time} is nonzero, it means that the
4991 arguments of the function are being analyzed for the second time. This
4992 happens for an inline function, which is not actually compiled until the
4993 end of the source file. The hook @code{TARGET_SETUP_INCOMING_VARARGS} should
4994 not generate any instructions in this case.
4997 @deftypefn {Target Hook} bool TARGET_STRICT_ARGUMENT_NAMING (cumulative_args_t @var{ca})
4998 Define this hook to return @code{true} if the location where a function
4999 argument is passed depends on whether or not it is a named argument.
5001 This hook controls how the @var{named} argument to @code{TARGET_FUNCTION_ARG}
5002 is set for varargs and stdarg functions. If this hook returns
5003 @code{true}, the @var{named} argument is always true for named
5004 arguments, and false for unnamed arguments. If it returns @code{false},
5005 but @code{TARGET_PRETEND_OUTGOING_VARARGS_NAMED} returns @code{true},
5006 then all arguments are treated as named. Otherwise, all named arguments
5007 except the last are treated as named.
5009 You need not define this hook if it always returns @code{false}.
5012 @deftypefn {Target Hook} void TARGET_CALL_ARGS (rtx, @var{tree})
5013 While generating RTL for a function call, this target hook is invoked once
5014 for each argument passed to the function, either a register returned by
5015 @code{TARGET_FUNCTION_ARG} or a memory location. It is called just
5016 before the point where argument registers are stored. The type of the
5017 function to be called is also passed as the second argument; it is
5018 @code{NULL_TREE} for libcalls. The @code{TARGET_END_CALL_ARGS} hook is
5019 invoked just after the code to copy the return reg has been emitted.
5020 This functionality can be used to perform special setup of call argument
5021 registers if a target needs it.
5022 For functions without arguments, the hook is called once with @code{pc_rtx}
5023 passed instead of an argument register.
5024 Most ports do not need to implement anything for this hook.
5027 @deftypefn {Target Hook} void TARGET_END_CALL_ARGS (void)
5028 This target hook is invoked while generating RTL for a function call,
5029 just after the point where the return reg is copied into a pseudo. It
5030 signals that all the call argument and return registers for the just
5031 emitted call are now no longer in use.
5032 Most ports do not need to implement anything for this hook.
5035 @deftypefn {Target Hook} bool TARGET_PRETEND_OUTGOING_VARARGS_NAMED (cumulative_args_t @var{ca})
5036 If you need to conditionally change ABIs so that one works with
5037 @code{TARGET_SETUP_INCOMING_VARARGS}, but the other works like neither
5038 @code{TARGET_SETUP_INCOMING_VARARGS} nor @code{TARGET_STRICT_ARGUMENT_NAMING} was
5039 defined, then define this hook to return @code{true} if
5040 @code{TARGET_SETUP_INCOMING_VARARGS} is used, @code{false} otherwise.
5041 Otherwise, you should not define this hook.
5044 @deftypefn {Target Hook} rtx TARGET_LOAD_BOUNDS_FOR_ARG (rtx @var{slot}, rtx @var{arg}, rtx @var{slot_no})
5045 This hook is used by expand pass to emit insn to load bounds of
5046 @var{arg} passed in @var{slot}. Expand pass uses this hook in case
5047 bounds of @var{arg} are not passed in register. If @var{slot} is a
5048 memory, then bounds are loaded as for regular pointer loaded from
5049 memory. If @var{slot} is not a memory then @var{slot_no} is an integer
5050 constant holding number of the target dependent special slot which
5051 should be used to obtain bounds. Hook returns RTX holding loaded bounds.
5054 @deftypefn {Target Hook} void TARGET_STORE_BOUNDS_FOR_ARG (rtx @var{arg}, rtx @var{slot}, rtx @var{bounds}, rtx @var{slot_no})
5055 This hook is used by expand pass to emit insns to store @var{bounds} of
5056 @var{arg} passed in @var{slot}. Expand pass uses this hook in case
5057 @var{bounds} of @var{arg} are not passed in register. If @var{slot} is a
5058 memory, then @var{bounds} are stored as for regular pointer stored in
5059 memory. If @var{slot} is not a memory then @var{slot_no} is an integer
5060 constant holding number of the target dependent special slot which
5061 should be used to store @var{bounds}.
5064 @deftypefn {Target Hook} rtx TARGET_LOAD_RETURNED_BOUNDS (rtx @var{slot})
5065 This hook is used by expand pass to emit insn to load bounds
5066 returned by function call in @var{slot}. Hook returns RTX holding
5070 @deftypefn {Target Hook} void TARGET_STORE_RETURNED_BOUNDS (rtx @var{slot}, rtx @var{bounds})
5071 This hook is used by expand pass to emit insn to store @var{bounds}
5072 returned by function call into @var{slot}.
5075 @deftypefn {Target Hook} rtx TARGET_CHKP_FUNCTION_VALUE_BOUNDS (const_tree @var{ret_type}, const_tree @var{fn_decl_or_type}, bool @var{outgoing})
5076 Define this to return an RTX representing the place where a function
5077 returns bounds for returned pointers. Arguments meaning is similar to
5078 @code{TARGET_FUNCTION_VALUE}.
5081 @deftypefn {Target Hook} void TARGET_SETUP_INCOMING_VARARG_BOUNDS (cumulative_args_t @var{args_so_far}, enum machine_mode @var{mode}, tree @var{type}, int *@var{pretend_args_size}, int @var{second_time})
5082 Use it to store bounds for anonymous register arguments stored
5083 into the stack. Arguments meaning is similar to
5084 @code{TARGET_SETUP_INCOMING_VARARGS}.
5088 @section Trampolines for Nested Functions
5089 @cindex trampolines for nested functions
5090 @cindex nested functions, trampolines for
5092 A @dfn{trampoline} is a small piece of code that is created at run time
5093 when the address of a nested function is taken. It normally resides on
5094 the stack, in the stack frame of the containing function. These macros
5095 tell GCC how to generate code to allocate and initialize a
5098 The instructions in the trampoline must do two things: load a constant
5099 address into the static chain register, and jump to the real address of
5100 the nested function. On CISC machines such as the m68k, this requires
5101 two instructions, a move immediate and a jump. Then the two addresses
5102 exist in the trampoline as word-long immediate operands. On RISC
5103 machines, it is often necessary to load each address into a register in
5104 two parts. Then pieces of each address form separate immediate
5107 The code generated to initialize the trampoline must store the variable
5108 parts---the static chain value and the function address---into the
5109 immediate operands of the instructions. On a CISC machine, this is
5110 simply a matter of copying each address to a memory reference at the
5111 proper offset from the start of the trampoline. On a RISC machine, it
5112 may be necessary to take out pieces of the address and store them
5115 @deftypefn {Target Hook} void TARGET_ASM_TRAMPOLINE_TEMPLATE (FILE *@var{f})
5116 This hook is called by @code{assemble_trampoline_template} to output,
5117 on the stream @var{f}, assembler code for a block of data that contains
5118 the constant parts of a trampoline. This code should not include a
5119 label---the label is taken care of automatically.
5121 If you do not define this hook, it means no template is needed
5122 for the target. Do not define this hook on systems where the block move
5123 code to copy the trampoline into place would be larger than the code
5124 to generate it on the spot.
5127 @defmac TRAMPOLINE_SECTION
5128 Return the section into which the trampoline template is to be placed
5129 (@pxref{Sections}). The default value is @code{readonly_data_section}.
5132 @defmac TRAMPOLINE_SIZE
5133 A C expression for the size in bytes of the trampoline, as an integer.
5136 @defmac TRAMPOLINE_ALIGNMENT
5137 Alignment required for trampolines, in bits.
5139 If you don't define this macro, the value of @code{FUNCTION_ALIGNMENT}
5140 is used for aligning trampolines.
5143 @deftypefn {Target Hook} void TARGET_TRAMPOLINE_INIT (rtx @var{m_tramp}, tree @var{fndecl}, rtx @var{static_chain})
5144 This hook is called to initialize a trampoline.
5145 @var{m_tramp} is an RTX for the memory block for the trampoline; @var{fndecl}
5146 is the @code{FUNCTION_DECL} for the nested function; @var{static_chain} is an
5147 RTX for the static chain value that should be passed to the function
5150 If the target defines @code{TARGET_ASM_TRAMPOLINE_TEMPLATE}, then the
5151 first thing this hook should do is emit a block move into @var{m_tramp}
5152 from the memory block returned by @code{assemble_trampoline_template}.
5153 Note that the block move need only cover the constant parts of the
5154 trampoline. If the target isolates the variable parts of the trampoline
5155 to the end, not all @code{TRAMPOLINE_SIZE} bytes need be copied.
5157 If the target requires any other actions, such as flushing caches or
5158 enabling stack execution, these actions should be performed after
5159 initializing the trampoline proper.
5162 @deftypefn {Target Hook} rtx TARGET_TRAMPOLINE_ADJUST_ADDRESS (rtx @var{addr})
5163 This hook should perform any machine-specific adjustment in
5164 the address of the trampoline. Its argument contains the address of the
5165 memory block that was passed to @code{TARGET_TRAMPOLINE_INIT}. In case
5166 the address to be used for a function call should be different from the
5167 address at which the template was stored, the different address should
5168 be returned; otherwise @var{addr} should be returned unchanged.
5169 If this hook is not defined, @var{addr} will be used for function calls.
5172 Implementing trampolines is difficult on many machines because they have
5173 separate instruction and data caches. Writing into a stack location
5174 fails to clear the memory in the instruction cache, so when the program
5175 jumps to that location, it executes the old contents.
5177 Here are two possible solutions. One is to clear the relevant parts of
5178 the instruction cache whenever a trampoline is set up. The other is to
5179 make all trampolines identical, by having them jump to a standard
5180 subroutine. The former technique makes trampoline execution faster; the
5181 latter makes initialization faster.
5183 To clear the instruction cache when a trampoline is initialized, define
5184 the following macro.
5186 @defmac CLEAR_INSN_CACHE (@var{beg}, @var{end})
5187 If defined, expands to a C expression clearing the @emph{instruction
5188 cache} in the specified interval. The definition of this macro would
5189 typically be a series of @code{asm} statements. Both @var{beg} and
5190 @var{end} are both pointer expressions.
5193 To use a standard subroutine, define the following macro. In addition,
5194 you must make sure that the instructions in a trampoline fill an entire
5195 cache line with identical instructions, or else ensure that the
5196 beginning of the trampoline code is always aligned at the same point in
5197 its cache line. Look in @file{m68k.h} as a guide.
5199 @defmac TRANSFER_FROM_TRAMPOLINE
5200 Define this macro if trampolines need a special subroutine to do their
5201 work. The macro should expand to a series of @code{asm} statements
5202 which will be compiled with GCC@. They go in a library function named
5203 @code{__transfer_from_trampoline}.
5205 If you need to avoid executing the ordinary prologue code of a compiled
5206 C function when you jump to the subroutine, you can do so by placing a
5207 special label of your own in the assembler code. Use one @code{asm}
5208 statement to generate an assembler label, and another to make the label
5209 global. Then trampolines can use that label to jump directly to your
5210 special assembler code.
5214 @section Implicit Calls to Library Routines
5215 @cindex library subroutine names
5216 @cindex @file{libgcc.a}
5218 @c prevent bad page break with this line
5219 Here is an explanation of implicit calls to library routines.
5221 @defmac DECLARE_LIBRARY_RENAMES
5222 This macro, if defined, should expand to a piece of C code that will get
5223 expanded when compiling functions for libgcc.a. It can be used to
5224 provide alternate names for GCC's internal library functions if there
5225 are ABI-mandated names that the compiler should provide.
5228 @findex set_optab_libfunc
5229 @findex init_one_libfunc
5230 @deftypefn {Target Hook} void TARGET_INIT_LIBFUNCS (void)
5231 This hook should declare additional library routines or rename
5232 existing ones, using the functions @code{set_optab_libfunc} and
5233 @code{init_one_libfunc} defined in @file{optabs.c}.
5234 @code{init_optabs} calls this macro after initializing all the normal
5237 The default is to do nothing. Most ports don't need to define this hook.
5240 @deftypevr {Target Hook} bool TARGET_LIBFUNC_GNU_PREFIX
5241 If false (the default), internal library routines start with two
5242 underscores. If set to true, these routines start with @code{__gnu_}
5243 instead. E.g., @code{__muldi3} changes to @code{__gnu_muldi3}. This
5244 currently only affects functions defined in @file{libgcc2.c}. If this
5245 is set to true, the @file{tm.h} file must also
5246 @code{#define LIBGCC2_GNU_PREFIX}.
5249 @defmac FLOAT_LIB_COMPARE_RETURNS_BOOL (@var{mode}, @var{comparison})
5250 This macro should return @code{true} if the library routine that
5251 implements the floating point comparison operator @var{comparison} in
5252 mode @var{mode} will return a boolean, and @var{false} if it will
5255 GCC's own floating point libraries return tristates from the
5256 comparison operators, so the default returns false always. Most ports
5257 don't need to define this macro.
5260 @defmac TARGET_LIB_INT_CMP_BIASED
5261 This macro should evaluate to @code{true} if the integer comparison
5262 functions (like @code{__cmpdi2}) return 0 to indicate that the first
5263 operand is smaller than the second, 1 to indicate that they are equal,
5264 and 2 to indicate that the first operand is greater than the second.
5265 If this macro evaluates to @code{false} the comparison functions return
5266 @minus{}1, 0, and 1 instead of 0, 1, and 2. If the target uses the routines
5267 in @file{libgcc.a}, you do not need to define this macro.
5270 @defmac TARGET_HAS_NO_HW_DIVIDE
5271 This macro should be defined if the target has no hardware divide
5272 instructions. If this macro is defined, GCC will use an algorithm which
5273 make use of simple logical and arithmetic operations for 64-bit
5274 division. If the macro is not defined, GCC will use an algorithm which
5275 make use of a 64-bit by 32-bit divide primitive.
5278 @cindex @code{EDOM}, implicit usage
5281 The value of @code{EDOM} on the target machine, as a C integer constant
5282 expression. If you don't define this macro, GCC does not attempt to
5283 deposit the value of @code{EDOM} into @code{errno} directly. Look in
5284 @file{/usr/include/errno.h} to find the value of @code{EDOM} on your
5287 If you do not define @code{TARGET_EDOM}, then compiled code reports
5288 domain errors by calling the library function and letting it report the
5289 error. If mathematical functions on your system use @code{matherr} when
5290 there is an error, then you should leave @code{TARGET_EDOM} undefined so
5291 that @code{matherr} is used normally.
5294 @cindex @code{errno}, implicit usage
5295 @defmac GEN_ERRNO_RTX
5296 Define this macro as a C expression to create an rtl expression that
5297 refers to the global ``variable'' @code{errno}. (On certain systems,
5298 @code{errno} may not actually be a variable.) If you don't define this
5299 macro, a reasonable default is used.
5302 @deftypefn {Target Hook} bool TARGET_LIBC_HAS_FUNCTION (enum function_class @var{fn_class})
5303 This hook determines whether a function from a class of functions
5304 @var{fn_class} is present at the runtime.
5307 @defmac NEXT_OBJC_RUNTIME
5308 Set this macro to 1 to use the "NeXT" Objective-C message sending conventions
5309 by default. This calling convention involves passing the object, the selector
5310 and the method arguments all at once to the method-lookup library function.
5311 This is the usual setting when targeting Darwin/Mac OS X systems, which have
5312 the NeXT runtime installed.
5314 If the macro is set to 0, the "GNU" Objective-C message sending convention
5315 will be used by default. This convention passes just the object and the
5316 selector to the method-lookup function, which returns a pointer to the method.
5318 In either case, it remains possible to select code-generation for the alternate
5319 scheme, by means of compiler command line switches.
5322 @node Addressing Modes
5323 @section Addressing Modes
5324 @cindex addressing modes
5326 @c prevent bad page break with this line
5327 This is about addressing modes.
5329 @defmac HAVE_PRE_INCREMENT
5330 @defmacx HAVE_PRE_DECREMENT
5331 @defmacx HAVE_POST_INCREMENT
5332 @defmacx HAVE_POST_DECREMENT
5333 A C expression that is nonzero if the machine supports pre-increment,
5334 pre-decrement, post-increment, or post-decrement addressing respectively.
5337 @defmac HAVE_PRE_MODIFY_DISP
5338 @defmacx HAVE_POST_MODIFY_DISP
5339 A C expression that is nonzero if the machine supports pre- or
5340 post-address side-effect generation involving constants other than
5341 the size of the memory operand.
5344 @defmac HAVE_PRE_MODIFY_REG
5345 @defmacx HAVE_POST_MODIFY_REG
5346 A C expression that is nonzero if the machine supports pre- or
5347 post-address side-effect generation involving a register displacement.
5350 @defmac CONSTANT_ADDRESS_P (@var{x})
5351 A C expression that is 1 if the RTX @var{x} is a constant which
5352 is a valid address. On most machines the default definition of
5353 @code{(CONSTANT_P (@var{x}) && GET_CODE (@var{x}) != CONST_DOUBLE)}
5354 is acceptable, but a few machines are more restrictive as to which
5355 constant addresses are supported.
5358 @defmac CONSTANT_P (@var{x})
5359 @code{CONSTANT_P}, which is defined by target-independent code,
5360 accepts integer-values expressions whose values are not explicitly
5361 known, such as @code{symbol_ref}, @code{label_ref}, and @code{high}
5362 expressions and @code{const} arithmetic expressions, in addition to
5363 @code{const_int} and @code{const_double} expressions.
5366 @defmac MAX_REGS_PER_ADDRESS
5367 A number, the maximum number of registers that can appear in a valid
5368 memory address. Note that it is up to you to specify a value equal to
5369 the maximum number that @code{TARGET_LEGITIMATE_ADDRESS_P} would ever
5373 @deftypefn {Target Hook} bool TARGET_LEGITIMATE_ADDRESS_P (machine_mode @var{mode}, rtx @var{x}, bool @var{strict})
5374 A function that returns whether @var{x} (an RTX) is a legitimate memory
5375 address on the target machine for a memory operand of mode @var{mode}.
5377 Legitimate addresses are defined in two variants: a strict variant and a
5378 non-strict one. The @var{strict} parameter chooses which variant is
5379 desired by the caller.
5381 The strict variant is used in the reload pass. It must be defined so
5382 that any pseudo-register that has not been allocated a hard register is
5383 considered a memory reference. This is because in contexts where some
5384 kind of register is required, a pseudo-register with no hard register
5385 must be rejected. For non-hard registers, the strict variant should look
5386 up the @code{reg_renumber} array; it should then proceed using the hard
5387 register number in the array, or treat the pseudo as a memory reference
5388 if the array holds @code{-1}.
5390 The non-strict variant is used in other passes. It must be defined to
5391 accept all pseudo-registers in every context where some kind of
5392 register is required.
5394 Normally, constant addresses which are the sum of a @code{symbol_ref}
5395 and an integer are stored inside a @code{const} RTX to mark them as
5396 constant. Therefore, there is no need to recognize such sums
5397 specifically as legitimate addresses. Normally you would simply
5398 recognize any @code{const} as legitimate.
5400 Usually @code{PRINT_OPERAND_ADDRESS} is not prepared to handle constant
5401 sums that are not marked with @code{const}. It assumes that a naked
5402 @code{plus} indicates indexing. If so, then you @emph{must} reject such
5403 naked constant sums as illegitimate addresses, so that none of them will
5404 be given to @code{PRINT_OPERAND_ADDRESS}.
5406 @cindex @code{TARGET_ENCODE_SECTION_INFO} and address validation
5407 On some machines, whether a symbolic address is legitimate depends on
5408 the section that the address refers to. On these machines, define the
5409 target hook @code{TARGET_ENCODE_SECTION_INFO} to store the information
5410 into the @code{symbol_ref}, and then check for it here. When you see a
5411 @code{const}, you will have to look inside it to find the
5412 @code{symbol_ref} in order to determine the section. @xref{Assembler
5415 @cindex @code{GO_IF_LEGITIMATE_ADDRESS}
5416 Some ports are still using a deprecated legacy substitute for
5417 this hook, the @code{GO_IF_LEGITIMATE_ADDRESS} macro. This macro
5421 #define GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label})
5425 and should @code{goto @var{label}} if the address @var{x} is a valid
5426 address on the target machine for a memory operand of mode @var{mode}.
5428 @findex REG_OK_STRICT
5429 Compiler source files that want to use the strict variant of this
5430 macro define the macro @code{REG_OK_STRICT}. You should use an
5431 @code{#ifdef REG_OK_STRICT} conditional to define the strict variant in
5432 that case and the non-strict variant otherwise.
5434 Using the hook is usually simpler because it limits the number of
5435 files that are recompiled when changes are made.
5438 @defmac TARGET_MEM_CONSTRAINT
5439 A single character to be used instead of the default @code{'m'}
5440 character for general memory addresses. This defines the constraint
5441 letter which matches the memory addresses accepted by
5442 @code{TARGET_LEGITIMATE_ADDRESS_P}. Define this macro if you want to
5443 support new address formats in your back end without changing the
5444 semantics of the @code{'m'} constraint. This is necessary in order to
5445 preserve functionality of inline assembly constructs using the
5446 @code{'m'} constraint.
5449 @defmac FIND_BASE_TERM (@var{x})
5450 A C expression to determine the base term of address @var{x},
5451 or to provide a simplified version of @var{x} from which @file{alias.c}
5452 can easily find the base term. This macro is used in only two places:
5453 @code{find_base_value} and @code{find_base_term} in @file{alias.c}.
5455 It is always safe for this macro to not be defined. It exists so
5456 that alias analysis can understand machine-dependent addresses.
5458 The typical use of this macro is to handle addresses containing
5459 a label_ref or symbol_ref within an UNSPEC@.
5462 @deftypefn {Target Hook} rtx TARGET_LEGITIMIZE_ADDRESS (rtx @var{x}, rtx @var{oldx}, machine_mode @var{mode})
5463 This hook is given an invalid memory address @var{x} for an
5464 operand of mode @var{mode} and should try to return a valid memory
5467 @findex break_out_memory_refs
5468 @var{x} will always be the result of a call to @code{break_out_memory_refs},
5469 and @var{oldx} will be the operand that was given to that function to produce
5472 The code of the hook should not alter the substructure of
5473 @var{x}. If it transforms @var{x} into a more legitimate form, it
5474 should return the new @var{x}.
5476 It is not necessary for this hook to come up with a legitimate address,
5477 with the exception of native TLS addresses (@pxref{Emulated TLS}).
5478 The compiler has standard ways of doing so in all cases. In fact, if
5479 the target supports only emulated TLS, it
5480 is safe to omit this hook or make it return @var{x} if it cannot find
5481 a valid way to legitimize the address. But often a machine-dependent
5482 strategy can generate better code.
5485 @defmac LEGITIMIZE_RELOAD_ADDRESS (@var{x}, @var{mode}, @var{opnum}, @var{type}, @var{ind_levels}, @var{win})
5486 A C compound statement that attempts to replace @var{x}, which is an address
5487 that needs reloading, with a valid memory address for an operand of mode
5488 @var{mode}. @var{win} will be a C statement label elsewhere in the code.
5489 It is not necessary to define this macro, but it might be useful for
5490 performance reasons.
5492 For example, on the i386, it is sometimes possible to use a single
5493 reload register instead of two by reloading a sum of two pseudo
5494 registers into a register. On the other hand, for number of RISC
5495 processors offsets are limited so that often an intermediate address
5496 needs to be generated in order to address a stack slot. By defining
5497 @code{LEGITIMIZE_RELOAD_ADDRESS} appropriately, the intermediate addresses
5498 generated for adjacent some stack slots can be made identical, and thus
5501 @emph{Note}: This macro should be used with caution. It is necessary
5502 to know something of how reload works in order to effectively use this,
5503 and it is quite easy to produce macros that build in too much knowledge
5504 of reload internals.
5506 @emph{Note}: This macro must be able to reload an address created by a
5507 previous invocation of this macro. If it fails to handle such addresses
5508 then the compiler may generate incorrect code or abort.
5511 The macro definition should use @code{push_reload} to indicate parts that
5512 need reloading; @var{opnum}, @var{type} and @var{ind_levels} are usually
5513 suitable to be passed unaltered to @code{push_reload}.
5515 The code generated by this macro must not alter the substructure of
5516 @var{x}. If it transforms @var{x} into a more legitimate form, it
5517 should assign @var{x} (which will always be a C variable) a new value.
5518 This also applies to parts that you change indirectly by calling
5521 @findex strict_memory_address_p
5522 The macro definition may use @code{strict_memory_address_p} to test if
5523 the address has become legitimate.
5526 If you want to change only a part of @var{x}, one standard way of doing
5527 this is to use @code{copy_rtx}. Note, however, that it unshares only a
5528 single level of rtl. Thus, if the part to be changed is not at the
5529 top level, you'll need to replace first the top level.
5530 It is not necessary for this macro to come up with a legitimate
5531 address; but often a machine-dependent strategy can generate better code.
5534 @deftypefn {Target Hook} bool TARGET_MODE_DEPENDENT_ADDRESS_P (const_rtx @var{addr}, addr_space_t @var{addrspace})
5535 This hook returns @code{true} if memory address @var{addr} in address
5536 space @var{addrspace} can have
5537 different meanings depending on the machine mode of the memory
5538 reference it is used for or if the address is valid for some modes
5541 Autoincrement and autodecrement addresses typically have mode-dependent
5542 effects because the amount of the increment or decrement is the size
5543 of the operand being addressed. Some machines have other mode-dependent
5544 addresses. Many RISC machines have no mode-dependent addresses.
5546 You may assume that @var{addr} is a valid address for the machine.
5548 The default version of this hook returns @code{false}.
5551 @deftypefn {Target Hook} bool TARGET_LEGITIMATE_CONSTANT_P (machine_mode @var{mode}, rtx @var{x})
5552 This hook returns true if @var{x} is a legitimate constant for a
5553 @var{mode}-mode immediate operand on the target machine. You can assume that
5554 @var{x} satisfies @code{CONSTANT_P}, so you need not check this.
5556 The default definition returns true.
5559 @deftypefn {Target Hook} rtx TARGET_DELEGITIMIZE_ADDRESS (rtx @var{x})
5560 This hook is used to undo the possibly obfuscating effects of the
5561 @code{LEGITIMIZE_ADDRESS} and @code{LEGITIMIZE_RELOAD_ADDRESS} target
5562 macros. Some backend implementations of these macros wrap symbol
5563 references inside an @code{UNSPEC} rtx to represent PIC or similar
5564 addressing modes. This target hook allows GCC's optimizers to understand
5565 the semantics of these opaque @code{UNSPEC}s by converting them back
5566 into their original form.
5569 @deftypefn {Target Hook} bool TARGET_CONST_NOT_OK_FOR_DEBUG_P (rtx @var{x})
5570 This hook should return true if @var{x} should not be emitted into
5574 @deftypefn {Target Hook} bool TARGET_CANNOT_FORCE_CONST_MEM (machine_mode @var{mode}, rtx @var{x})
5575 This hook should return true if @var{x} is of a form that cannot (or
5576 should not) be spilled to the constant pool. @var{mode} is the mode
5579 The default version of this hook returns false.
5581 The primary reason to define this hook is to prevent reload from
5582 deciding that a non-legitimate constant would be better reloaded
5583 from the constant pool instead of spilling and reloading a register
5584 holding the constant. This restriction is often true of addresses
5585 of TLS symbols for various targets.
5588 @deftypefn {Target Hook} bool TARGET_USE_BLOCKS_FOR_CONSTANT_P (machine_mode @var{mode}, const_rtx @var{x})
5589 This hook should return true if pool entries for constant @var{x} can
5590 be placed in an @code{object_block} structure. @var{mode} is the mode
5593 The default version returns false for all constants.
5596 @deftypefn {Target Hook} bool TARGET_USE_BLOCKS_FOR_DECL_P (const_tree @var{decl})
5597 This hook should return true if pool entries for @var{decl} should
5598 be placed in an @code{object_block} structure.
5600 The default version returns true for all decls.
5603 @deftypefn {Target Hook} tree TARGET_BUILTIN_RECIPROCAL (unsigned @var{fn}, bool @var{md_fn}, bool @var{sqrt})
5604 This hook should return the DECL of a function that implements reciprocal of
5605 the builtin function with builtin function code @var{fn}, or
5606 @code{NULL_TREE} if such a function is not available. @var{md_fn} is true
5607 when @var{fn} is a code of a machine-dependent builtin function. When
5608 @var{sqrt} is true, additional optimizations that apply only to the reciprocal
5609 of a square root function are performed, and only reciprocals of @code{sqrt}
5613 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_MASK_FOR_LOAD (void)
5614 This hook should return the DECL of a function @var{f} that given an
5615 address @var{addr} as an argument returns a mask @var{m} that can be
5616 used to extract from two vectors the relevant data that resides in
5617 @var{addr} in case @var{addr} is not properly aligned.
5619 The autovectorizer, when vectorizing a load operation from an address
5620 @var{addr} that may be unaligned, will generate two vector loads from
5621 the two aligned addresses around @var{addr}. It then generates a
5622 @code{REALIGN_LOAD} operation to extract the relevant data from the
5623 two loaded vectors. The first two arguments to @code{REALIGN_LOAD},
5624 @var{v1} and @var{v2}, are the two vectors, each of size @var{VS}, and
5625 the third argument, @var{OFF}, defines how the data will be extracted
5626 from these two vectors: if @var{OFF} is 0, then the returned vector is
5627 @var{v2}; otherwise, the returned vector is composed from the last
5628 @var{VS}-@var{OFF} elements of @var{v1} concatenated to the first
5629 @var{OFF} elements of @var{v2}.
5631 If this hook is defined, the autovectorizer will generate a call
5632 to @var{f} (using the DECL tree that this hook returns) and will
5633 use the return value of @var{f} as the argument @var{OFF} to
5634 @code{REALIGN_LOAD}. Therefore, the mask @var{m} returned by @var{f}
5635 should comply with the semantics expected by @code{REALIGN_LOAD}
5637 If this hook is not defined, then @var{addr} will be used as
5638 the argument @var{OFF} to @code{REALIGN_LOAD}, in which case the low
5639 log2(@var{VS}) @minus{} 1 bits of @var{addr} will be considered.
5642 @deftypefn {Target Hook} int TARGET_VECTORIZE_BUILTIN_VECTORIZATION_COST (enum vect_cost_for_stmt @var{type_of_cost}, tree @var{vectype}, int @var{misalign})
5643 Returns cost of different scalar or vector statements for vectorization cost model.
5644 For vector memory operations the cost may depend on type (@var{vectype}) and
5645 misalignment value (@var{misalign}).
5648 @deftypefn {Target Hook} bool TARGET_VECTORIZE_VECTOR_ALIGNMENT_REACHABLE (const_tree @var{type}, bool @var{is_packed})
5649 Return true if vector alignment is reachable (by peeling N iterations) for the given type.
5652 @deftypefn {Target Hook} bool TARGET_VECTORIZE_VEC_PERM_CONST_OK (machine_mode, const unsigned char *@var{sel})
5653 Return true if a vector created for @code{vec_perm_const} is valid.
5656 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_CONVERSION (unsigned @var{code}, tree @var{dest_type}, tree @var{src_type})
5657 This hook should return the DECL of a function that implements conversion of the
5658 input vector of type @var{src_type} to type @var{dest_type}.
5659 The value of @var{code} is one of the enumerators in @code{enum tree_code} and
5660 specifies how the conversion is to be applied
5661 (truncation, rounding, etc.).
5663 If this hook is defined, the autovectorizer will use the
5664 @code{TARGET_VECTORIZE_BUILTIN_CONVERSION} target hook when vectorizing
5665 conversion. Otherwise, it will return @code{NULL_TREE}.
5668 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_VECTORIZED_FUNCTION (tree @var{fndecl}, tree @var{vec_type_out}, tree @var{vec_type_in})
5669 This hook should return the decl of a function that implements the
5670 vectorized variant of the builtin function with builtin function code
5671 @var{code} or @code{NULL_TREE} if such a function is not available.
5672 The value of @var{fndecl} is the builtin function declaration. The
5673 return type of the vectorized function shall be of vector type
5674 @var{vec_type_out} and the argument types should be @var{vec_type_in}.
5677 @deftypefn {Target Hook} bool TARGET_VECTORIZE_SUPPORT_VECTOR_MISALIGNMENT (machine_mode @var{mode}, const_tree @var{type}, int @var{misalignment}, bool @var{is_packed})
5678 This hook should return true if the target supports misaligned vector
5679 store/load of a specific factor denoted in the @var{misalignment}
5680 parameter. The vector store/load should be of machine mode @var{mode} and
5681 the elements in the vectors should be of type @var{type}. @var{is_packed}
5682 parameter is true if the memory access is defined in a packed struct.
5685 @deftypefn {Target Hook} machine_mode TARGET_VECTORIZE_PREFERRED_SIMD_MODE (machine_mode @var{mode})
5686 This hook should return the preferred mode for vectorizing scalar
5687 mode @var{mode}. The default is
5688 equal to @code{word_mode}, because the vectorizer can do some
5689 transformations even in absence of specialized @acronym{SIMD} hardware.
5692 @deftypefn {Target Hook} {unsigned int} TARGET_VECTORIZE_AUTOVECTORIZE_VECTOR_SIZES (void)
5693 This hook should return a mask of sizes that should be iterated over
5694 after trying to autovectorize using the vector size derived from the
5695 mode returned by @code{TARGET_VECTORIZE_PREFERRED_SIMD_MODE}.
5696 The default is zero which means to not iterate over other vector sizes.
5699 @deftypefn {Target Hook} {void *} TARGET_VECTORIZE_INIT_COST (struct loop *@var{loop_info})
5700 This hook should initialize target-specific data structures in preparation for modeling the costs of vectorizing a loop or basic block. The default allocates three unsigned integers for accumulating costs for the prologue, body, and epilogue of the loop or basic block. If @var{loop_info} is non-NULL, it identifies the loop being vectorized; otherwise a single block is being vectorized.
5703 @deftypefn {Target Hook} unsigned TARGET_VECTORIZE_ADD_STMT_COST (void *@var{data}, int @var{count}, enum vect_cost_for_stmt @var{kind}, struct _stmt_vec_info *@var{stmt_info}, int @var{misalign}, enum vect_cost_model_location @var{where})
5704 This hook should update the target-specific @var{data} in response to adding @var{count} copies of the given @var{kind} of statement to a loop or basic block. The default adds the builtin vectorizer cost for the copies of the statement to the accumulator specified by @var{where}, (the prologue, body, or epilogue) and returns the amount added. The return value should be viewed as a tentative cost that may later be revised.
5707 @deftypefn {Target Hook} void TARGET_VECTORIZE_FINISH_COST (void *@var{data}, unsigned *@var{prologue_cost}, unsigned *@var{body_cost}, unsigned *@var{epilogue_cost})
5708 This hook should complete calculations of the cost of vectorizing a loop or basic block based on @var{data}, and return the prologue, body, and epilogue costs as unsigned integers. The default returns the value of the three accumulators.
5711 @deftypefn {Target Hook} void TARGET_VECTORIZE_DESTROY_COST_DATA (void *@var{data})
5712 This hook should release @var{data} and any related data structures allocated by TARGET_VECTORIZE_INIT_COST. The default releases the accumulator.
5715 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_TM_LOAD (tree)
5716 This hook should return the built-in decl needed to load a vector of the given type within a transaction.
5719 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_TM_STORE (tree)
5720 This hook should return the built-in decl needed to store a vector of the given type within a transaction.
5723 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_GATHER (const_tree @var{mem_vectype}, const_tree @var{index_type}, int @var{scale})
5724 Target builtin that implements vector gather operation. @var{mem_vectype}
5725 is the vector type of the load and @var{index_type} is scalar type of
5726 the index, scaled by @var{scale}.
5727 The default is @code{NULL_TREE} which means to not vectorize gather
5731 @deftypefn {Target Hook} int TARGET_SIMD_CLONE_COMPUTE_VECSIZE_AND_SIMDLEN (struct cgraph_node *@var{}, struct cgraph_simd_clone *@var{}, @var{tree}, @var{int})
5732 This hook should set @var{vecsize_mangle}, @var{vecsize_int}, @var{vecsize_float}
5733 fields in @var{simd_clone} structure pointed by @var{clone_info} argument and also
5734 @var{simdlen} field if it was previously 0.
5735 The hook should return 0 if SIMD clones shouldn't be emitted,
5736 or number of @var{vecsize_mangle} variants that should be emitted.
5739 @deftypefn {Target Hook} void TARGET_SIMD_CLONE_ADJUST (struct cgraph_node *@var{})
5740 This hook should add implicit @code{attribute(target("..."))} attribute
5741 to SIMD clone @var{node} if needed.
5744 @deftypefn {Target Hook} int TARGET_SIMD_CLONE_USABLE (struct cgraph_node *@var{})
5745 This hook should return -1 if SIMD clone @var{node} shouldn't be used
5746 in vectorized loops in current function, or non-negative number if it is
5747 usable. In that case, the smaller the number is, the more desirable it is
5751 @node Anchored Addresses
5752 @section Anchored Addresses
5753 @cindex anchored addresses
5754 @cindex @option{-fsection-anchors}
5756 GCC usually addresses every static object as a separate entity.
5757 For example, if we have:
5761 int foo (void) @{ return a + b + c; @}
5764 the code for @code{foo} will usually calculate three separate symbolic
5765 addresses: those of @code{a}, @code{b} and @code{c}. On some targets,
5766 it would be better to calculate just one symbolic address and access
5767 the three variables relative to it. The equivalent pseudocode would
5773 register int *xr = &x;
5774 return xr[&a - &x] + xr[&b - &x] + xr[&c - &x];
5778 (which isn't valid C). We refer to shared addresses like @code{x} as
5779 ``section anchors''. Their use is controlled by @option{-fsection-anchors}.
5781 The hooks below describe the target properties that GCC needs to know
5782 in order to make effective use of section anchors. It won't use
5783 section anchors at all unless either @code{TARGET_MIN_ANCHOR_OFFSET}
5784 or @code{TARGET_MAX_ANCHOR_OFFSET} is set to a nonzero value.
5786 @deftypevr {Target Hook} HOST_WIDE_INT TARGET_MIN_ANCHOR_OFFSET
5787 The minimum offset that should be applied to a section anchor.
5788 On most targets, it should be the smallest offset that can be
5789 applied to a base register while still giving a legitimate address
5790 for every mode. The default value is 0.
5793 @deftypevr {Target Hook} HOST_WIDE_INT TARGET_MAX_ANCHOR_OFFSET
5794 Like @code{TARGET_MIN_ANCHOR_OFFSET}, but the maximum (inclusive)
5795 offset that should be applied to section anchors. The default
5799 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_ANCHOR (rtx @var{x})
5800 Write the assembly code to define section anchor @var{x}, which is a
5801 @code{SYMBOL_REF} for which @samp{SYMBOL_REF_ANCHOR_P (@var{x})} is true.
5802 The hook is called with the assembly output position set to the beginning
5803 of @code{SYMBOL_REF_BLOCK (@var{x})}.
5805 If @code{ASM_OUTPUT_DEF} is available, the hook's default definition uses
5806 it to define the symbol as @samp{. + SYMBOL_REF_BLOCK_OFFSET (@var{x})}.
5807 If @code{ASM_OUTPUT_DEF} is not available, the hook's default definition
5808 is @code{NULL}, which disables the use of section anchors altogether.
5811 @deftypefn {Target Hook} bool TARGET_USE_ANCHORS_FOR_SYMBOL_P (const_rtx @var{x})
5812 Return true if GCC should attempt to use anchors to access @code{SYMBOL_REF}
5813 @var{x}. You can assume @samp{SYMBOL_REF_HAS_BLOCK_INFO_P (@var{x})} and
5814 @samp{!SYMBOL_REF_ANCHOR_P (@var{x})}.
5816 The default version is correct for most targets, but you might need to
5817 intercept this hook to handle things like target-specific attributes
5818 or target-specific sections.
5821 @node Condition Code
5822 @section Condition Code Status
5823 @cindex condition code status
5825 The macros in this section can be split in two families, according to the
5826 two ways of representing condition codes in GCC.
5828 The first representation is the so called @code{(cc0)} representation
5829 (@pxref{Jump Patterns}), where all instructions can have an implicit
5830 clobber of the condition codes. The second is the condition code
5831 register representation, which provides better schedulability for
5832 architectures that do have a condition code register, but on which
5833 most instructions do not affect it. The latter category includes
5836 The implicit clobbering poses a strong restriction on the placement of
5837 the definition and use of the condition code. In the past the definition
5838 and use were always adjacent. However, recent changes to support trapping
5839 arithmatic may result in the definition and user being in different blocks.
5840 Thus, there may be a @code{NOTE_INSN_BASIC_BLOCK} between them. Additionally,
5841 the definition may be the source of exception handling edges.
5843 These restrictions can prevent important
5844 optimizations on some machines. For example, on the IBM RS/6000, there
5845 is a delay for taken branches unless the condition code register is set
5846 three instructions earlier than the conditional branch. The instruction
5847 scheduler cannot perform this optimization if it is not permitted to
5848 separate the definition and use of the condition code register.
5850 For this reason, it is possible and suggested to use a register to
5851 represent the condition code for new ports. If there is a specific
5852 condition code register in the machine, use a hard register. If the
5853 condition code or comparison result can be placed in any general register,
5854 or if there are multiple condition registers, use a pseudo register.
5855 Registers used to store the condition code value will usually have a mode
5856 that is in class @code{MODE_CC}.
5858 Alternatively, you can use @code{BImode} if the comparison operator is
5859 specified already in the compare instruction. In this case, you are not
5860 interested in most macros in this section.
5863 * CC0 Condition Codes:: Old style representation of condition codes.
5864 * MODE_CC Condition Codes:: Modern representation of condition codes.
5867 @node CC0 Condition Codes
5868 @subsection Representation of condition codes using @code{(cc0)}
5872 The file @file{conditions.h} defines a variable @code{cc_status} to
5873 describe how the condition code was computed (in case the interpretation of
5874 the condition code depends on the instruction that it was set by). This
5875 variable contains the RTL expressions on which the condition code is
5876 currently based, and several standard flags.
5878 Sometimes additional machine-specific flags must be defined in the machine
5879 description header file. It can also add additional machine-specific
5880 information by defining @code{CC_STATUS_MDEP}.
5882 @defmac CC_STATUS_MDEP
5883 C code for a data type which is used for declaring the @code{mdep}
5884 component of @code{cc_status}. It defaults to @code{int}.
5886 This macro is not used on machines that do not use @code{cc0}.
5889 @defmac CC_STATUS_MDEP_INIT
5890 A C expression to initialize the @code{mdep} field to ``empty''.
5891 The default definition does nothing, since most machines don't use
5892 the field anyway. If you want to use the field, you should probably
5893 define this macro to initialize it.
5895 This macro is not used on machines that do not use @code{cc0}.
5898 @defmac NOTICE_UPDATE_CC (@var{exp}, @var{insn})
5899 A C compound statement to set the components of @code{cc_status}
5900 appropriately for an insn @var{insn} whose body is @var{exp}. It is
5901 this macro's responsibility to recognize insns that set the condition
5902 code as a byproduct of other activity as well as those that explicitly
5905 This macro is not used on machines that do not use @code{cc0}.
5907 If there are insns that do not set the condition code but do alter
5908 other machine registers, this macro must check to see whether they
5909 invalidate the expressions that the condition code is recorded as
5910 reflecting. For example, on the 68000, insns that store in address
5911 registers do not set the condition code, which means that usually
5912 @code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such
5913 insns. But suppose that the previous insn set the condition code
5914 based on location @samp{a4@@(102)} and the current insn stores a new
5915 value in @samp{a4}. Although the condition code is not changed by
5916 this, it will no longer be true that it reflects the contents of
5917 @samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter
5918 @code{cc_status} in this case to say that nothing is known about the
5919 condition code value.
5921 The definition of @code{NOTICE_UPDATE_CC} must be prepared to deal
5922 with the results of peephole optimization: insns whose patterns are
5923 @code{parallel} RTXs containing various @code{reg}, @code{mem} or
5924 constants which are just the operands. The RTL structure of these
5925 insns is not sufficient to indicate what the insns actually do. What
5926 @code{NOTICE_UPDATE_CC} should do when it sees one is just to run
5927 @code{CC_STATUS_INIT}.
5929 A possible definition of @code{NOTICE_UPDATE_CC} is to call a function
5930 that looks at an attribute (@pxref{Insn Attributes}) named, for example,
5931 @samp{cc}. This avoids having detailed information about patterns in
5932 two places, the @file{md} file and in @code{NOTICE_UPDATE_CC}.
5935 @node MODE_CC Condition Codes
5936 @subsection Representation of condition codes using registers
5940 @defmac SELECT_CC_MODE (@var{op}, @var{x}, @var{y})
5941 On many machines, the condition code may be produced by other instructions
5942 than compares, for example the branch can use directly the condition
5943 code set by a subtract instruction. However, on some machines
5944 when the condition code is set this way some bits (such as the overflow
5945 bit) are not set in the same way as a test instruction, so that a different
5946 branch instruction must be used for some conditional branches. When
5947 this happens, use the machine mode of the condition code register to
5948 record different formats of the condition code register. Modes can
5949 also be used to record which compare instruction (e.g. a signed or an
5950 unsigned comparison) produced the condition codes.
5952 If other modes than @code{CCmode} are required, add them to
5953 @file{@var{machine}-modes.def} and define @code{SELECT_CC_MODE} to choose
5954 a mode given an operand of a compare. This is needed because the modes
5955 have to be chosen not only during RTL generation but also, for example,
5956 by instruction combination. The result of @code{SELECT_CC_MODE} should
5957 be consistent with the mode used in the patterns; for example to support
5958 the case of the add on the SPARC discussed above, we have the pattern
5962 [(set (reg:CC_NOOV 0)
5964 (plus:SI (match_operand:SI 0 "register_operand" "%r")
5965 (match_operand:SI 1 "arith_operand" "rI"))
5972 together with a @code{SELECT_CC_MODE} that returns @code{CC_NOOVmode}
5973 for comparisons whose argument is a @code{plus}:
5976 #define SELECT_CC_MODE(OP,X,Y) \
5977 (GET_MODE_CLASS (GET_MODE (X)) == MODE_FLOAT \
5978 ? ((OP == LT || OP == LE || OP == GT || OP == GE) \
5979 ? CCFPEmode : CCFPmode) \
5980 : ((GET_CODE (X) == PLUS || GET_CODE (X) == MINUS \
5981 || GET_CODE (X) == NEG || GET_CODE (x) == ASHIFT) \
5982 ? CC_NOOVmode : CCmode))
5985 Another reason to use modes is to retain information on which operands
5986 were used by the comparison; see @code{REVERSIBLE_CC_MODE} later in
5989 You should define this macro if and only if you define extra CC modes
5990 in @file{@var{machine}-modes.def}.
5993 @deftypefn {Target Hook} void TARGET_CANONICALIZE_COMPARISON (int *@var{code}, rtx *@var{op0}, rtx *@var{op1}, bool @var{op0_preserve_value})
5994 On some machines not all possible comparisons are defined, but you can
5995 convert an invalid comparison into a valid one. For example, the Alpha
5996 does not have a @code{GT} comparison, but you can use an @code{LT}
5997 comparison instead and swap the order of the operands.
5999 On such machines, implement this hook to do any required conversions.
6000 @var{code} is the initial comparison code and @var{op0} and @var{op1}
6001 are the left and right operands of the comparison, respectively. If
6002 @var{op0_preserve_value} is @code{true} the implementation is not
6003 allowed to change the value of @var{op0} since the value might be used
6004 in RTXs which aren't comparisons. E.g. the implementation is not
6005 allowed to swap operands in that case.
6007 GCC will not assume that the comparison resulting from this macro is
6008 valid but will see if the resulting insn matches a pattern in the
6011 You need not to implement this hook if it would never change the
6012 comparison code or operands.
6015 @defmac REVERSIBLE_CC_MODE (@var{mode})
6016 A C expression whose value is one if it is always safe to reverse a
6017 comparison whose mode is @var{mode}. If @code{SELECT_CC_MODE}
6018 can ever return @var{mode} for a floating-point inequality comparison,
6019 then @code{REVERSIBLE_CC_MODE (@var{mode})} must be zero.
6021 You need not define this macro if it would always returns zero or if the
6022 floating-point format is anything other than @code{IEEE_FLOAT_FORMAT}.
6023 For example, here is the definition used on the SPARC, where floating-point
6024 inequality comparisons are given either @code{CCFPEmode} or @code{CCFPmode}:
6027 #define REVERSIBLE_CC_MODE(MODE) \
6028 ((MODE) != CCFPEmode && (MODE) != CCFPmode)
6032 @defmac REVERSE_CONDITION (@var{code}, @var{mode})
6033 A C expression whose value is reversed condition code of the @var{code} for
6034 comparison done in CC_MODE @var{mode}. The macro is used only in case
6035 @code{REVERSIBLE_CC_MODE (@var{mode})} is nonzero. Define this macro in case
6036 machine has some non-standard way how to reverse certain conditionals. For
6037 instance in case all floating point conditions are non-trapping, compiler may
6038 freely convert unordered compares to ordered ones. Then definition may look
6042 #define REVERSE_CONDITION(CODE, MODE) \
6043 ((MODE) != CCFPmode ? reverse_condition (CODE) \
6044 : reverse_condition_maybe_unordered (CODE))
6048 @deftypefn {Target Hook} bool TARGET_FIXED_CONDITION_CODE_REGS (unsigned int *@var{p1}, unsigned int *@var{p2})
6049 On targets which do not use @code{(cc0)}, and which use a hard
6050 register rather than a pseudo-register to hold condition codes, the
6051 regular CSE passes are often not able to identify cases in which the
6052 hard register is set to a common value. Use this hook to enable a
6053 small pass which optimizes such cases. This hook should return true
6054 to enable this pass, and it should set the integers to which its
6055 arguments point to the hard register numbers used for condition codes.
6056 When there is only one such register, as is true on most systems, the
6057 integer pointed to by @var{p2} should be set to
6058 @code{INVALID_REGNUM}.
6060 The default version of this hook returns false.
6063 @deftypefn {Target Hook} machine_mode TARGET_CC_MODES_COMPATIBLE (machine_mode @var{m1}, machine_mode @var{m2})
6064 On targets which use multiple condition code modes in class
6065 @code{MODE_CC}, it is sometimes the case that a comparison can be
6066 validly done in more than one mode. On such a system, define this
6067 target hook to take two mode arguments and to return a mode in which
6068 both comparisons may be validly done. If there is no such mode,
6069 return @code{VOIDmode}.
6071 The default version of this hook checks whether the modes are the
6072 same. If they are, it returns that mode. If they are different, it
6073 returns @code{VOIDmode}.
6076 @deftypevr {Target Hook} {unsigned int} TARGET_FLAGS_REGNUM
6077 If the target has a dedicated flags register, and it needs to use the post-reload comparison elimination pass, then this value should be set appropriately.
6081 @section Describing Relative Costs of Operations
6082 @cindex costs of instructions
6083 @cindex relative costs
6084 @cindex speed of instructions
6086 These macros let you describe the relative speed of various operations
6087 on the target machine.
6089 @defmac REGISTER_MOVE_COST (@var{mode}, @var{from}, @var{to})
6090 A C expression for the cost of moving data of mode @var{mode} from a
6091 register in class @var{from} to one in class @var{to}. The classes are
6092 expressed using the enumeration values such as @code{GENERAL_REGS}. A
6093 value of 2 is the default; other values are interpreted relative to
6096 It is not required that the cost always equal 2 when @var{from} is the
6097 same as @var{to}; on some machines it is expensive to move between
6098 registers if they are not general registers.
6100 If reload sees an insn consisting of a single @code{set} between two
6101 hard registers, and if @code{REGISTER_MOVE_COST} applied to their
6102 classes returns a value of 2, reload does not check to ensure that the
6103 constraints of the insn are met. Setting a cost of other than 2 will
6104 allow reload to verify that the constraints are met. You should do this
6105 if the @samp{mov@var{m}} pattern's constraints do not allow such copying.
6107 These macros are obsolete, new ports should use the target hook
6108 @code{TARGET_REGISTER_MOVE_COST} instead.
6111 @deftypefn {Target Hook} int TARGET_REGISTER_MOVE_COST (machine_mode @var{mode}, reg_class_t @var{from}, reg_class_t @var{to})
6112 This target hook should return the cost of moving data of mode @var{mode}
6113 from a register in class @var{from} to one in class @var{to}. The classes
6114 are expressed using the enumeration values such as @code{GENERAL_REGS}.
6115 A value of 2 is the default; other values are interpreted relative to
6118 It is not required that the cost always equal 2 when @var{from} is the
6119 same as @var{to}; on some machines it is expensive to move between
6120 registers if they are not general registers.
6122 If reload sees an insn consisting of a single @code{set} between two
6123 hard registers, and if @code{TARGET_REGISTER_MOVE_COST} applied to their
6124 classes returns a value of 2, reload does not check to ensure that the
6125 constraints of the insn are met. Setting a cost of other than 2 will
6126 allow reload to verify that the constraints are met. You should do this
6127 if the @samp{mov@var{m}} pattern's constraints do not allow such copying.
6129 The default version of this function returns 2.
6132 @defmac MEMORY_MOVE_COST (@var{mode}, @var{class}, @var{in})
6133 A C expression for the cost of moving data of mode @var{mode} between a
6134 register of class @var{class} and memory; @var{in} is zero if the value
6135 is to be written to memory, nonzero if it is to be read in. This cost
6136 is relative to those in @code{REGISTER_MOVE_COST}. If moving between
6137 registers and memory is more expensive than between two registers, you
6138 should define this macro to express the relative cost.
6140 If you do not define this macro, GCC uses a default cost of 4 plus
6141 the cost of copying via a secondary reload register, if one is
6142 needed. If your machine requires a secondary reload register to copy
6143 between memory and a register of @var{class} but the reload mechanism is
6144 more complex than copying via an intermediate, define this macro to
6145 reflect the actual cost of the move.
6147 GCC defines the function @code{memory_move_secondary_cost} if
6148 secondary reloads are needed. It computes the costs due to copying via
6149 a secondary register. If your machine copies from memory using a
6150 secondary register in the conventional way but the default base value of
6151 4 is not correct for your machine, define this macro to add some other
6152 value to the result of that function. The arguments to that function
6153 are the same as to this macro.
6155 These macros are obsolete, new ports should use the target hook
6156 @code{TARGET_MEMORY_MOVE_COST} instead.
6159 @deftypefn {Target Hook} int TARGET_MEMORY_MOVE_COST (machine_mode @var{mode}, reg_class_t @var{rclass}, bool @var{in})
6160 This target hook should return the cost of moving data of mode @var{mode}
6161 between a register of class @var{rclass} and memory; @var{in} is @code{false}
6162 if the value is to be written to memory, @code{true} if it is to be read in.
6163 This cost is relative to those in @code{TARGET_REGISTER_MOVE_COST}.
6164 If moving between registers and memory is more expensive than between two
6165 registers, you should add this target hook to express the relative cost.
6167 If you do not add this target hook, GCC uses a default cost of 4 plus
6168 the cost of copying via a secondary reload register, if one is
6169 needed. If your machine requires a secondary reload register to copy
6170 between memory and a register of @var{rclass} but the reload mechanism is
6171 more complex than copying via an intermediate, use this target hook to
6172 reflect the actual cost of the move.
6174 GCC defines the function @code{memory_move_secondary_cost} if
6175 secondary reloads are needed. It computes the costs due to copying via
6176 a secondary register. If your machine copies from memory using a
6177 secondary register in the conventional way but the default base value of
6178 4 is not correct for your machine, use this target hook to add some other
6179 value to the result of that function. The arguments to that function
6180 are the same as to this target hook.
6183 @defmac BRANCH_COST (@var{speed_p}, @var{predictable_p})
6184 A C expression for the cost of a branch instruction. A value of 1 is
6185 the default; other values are interpreted relative to that. Parameter
6186 @var{speed_p} is true when the branch in question should be optimized
6187 for speed. When it is false, @code{BRANCH_COST} should return a value
6188 optimal for code size rather than performance. @var{predictable_p} is
6189 true for well-predicted branches. On many architectures the
6190 @code{BRANCH_COST} can be reduced then.
6193 Here are additional macros which do not specify precise relative costs,
6194 but only that certain actions are more expensive than GCC would
6197 @defmac SLOW_BYTE_ACCESS
6198 Define this macro as a C expression which is nonzero if accessing less
6199 than a word of memory (i.e.@: a @code{char} or a @code{short}) is no
6200 faster than accessing a word of memory, i.e., if such access
6201 require more than one instruction or if there is no difference in cost
6202 between byte and (aligned) word loads.
6204 When this macro is not defined, the compiler will access a field by
6205 finding the smallest containing object; when it is defined, a fullword
6206 load will be used if alignment permits. Unless bytes accesses are
6207 faster than word accesses, using word accesses is preferable since it
6208 may eliminate subsequent memory access if subsequent accesses occur to
6209 other fields in the same word of the structure, but to different bytes.
6212 @defmac SLOW_UNALIGNED_ACCESS (@var{mode}, @var{alignment})
6213 Define this macro to be the value 1 if memory accesses described by the
6214 @var{mode} and @var{alignment} parameters have a cost many times greater
6215 than aligned accesses, for example if they are emulated in a trap
6218 When this macro is nonzero, the compiler will act as if
6219 @code{STRICT_ALIGNMENT} were nonzero when generating code for block
6220 moves. This can cause significantly more instructions to be produced.
6221 Therefore, do not set this macro nonzero if unaligned accesses only add a
6222 cycle or two to the time for a memory access.
6224 If the value of this macro is always zero, it need not be defined. If
6225 this macro is defined, it should produce a nonzero value when
6226 @code{STRICT_ALIGNMENT} is nonzero.
6229 @defmac MOVE_RATIO (@var{speed})
6230 The threshold of number of scalar memory-to-memory move insns, @emph{below}
6231 which a sequence of insns should be generated instead of a
6232 string move insn or a library call. Increasing the value will always
6233 make code faster, but eventually incurs high cost in increased code size.
6235 Note that on machines where the corresponding move insn is a
6236 @code{define_expand} that emits a sequence of insns, this macro counts
6237 the number of such sequences.
6239 The parameter @var{speed} is true if the code is currently being
6240 optimized for speed rather than size.
6242 If you don't define this, a reasonable default is used.
6245 @deftypefn {Target Hook} bool TARGET_USE_BY_PIECES_INFRASTRUCTURE_P (unsigned HOST_WIDE_INT @var{size}, unsigned int @var{alignment}, enum by_pieces_operation @var{op}, bool @var{speed_p})
6246 GCC will attempt several strategies when asked to copy between
6247 two areas of memory, or to set, clear or store to memory, for example
6248 when copying a @code{struct}. The @code{by_pieces} infrastructure
6249 implements such memory operations as a sequence of load, store or move
6250 insns. Alternate strategies are to expand the
6251 @code{movmem} or @code{setmem} optabs, to emit a library call, or to emit
6252 unit-by-unit, loop-based operations.
6254 This target hook should return true if, for a memory operation with a
6255 given @var{size} and @var{alignment}, using the @code{by_pieces}
6256 infrastructure is expected to result in better code generation.
6257 Both @var{size} and @var{alignment} are measured in terms of storage
6260 The parameter @var{op} is one of: @code{CLEAR_BY_PIECES},
6261 @code{MOVE_BY_PIECES}, @code{SET_BY_PIECES}, @code{STORE_BY_PIECES}.
6262 These describe the type of memory operation under consideration.
6264 The parameter @var{speed_p} is true if the code is currently being
6265 optimized for speed rather than size.
6267 Returning true for higher values of @var{size} can improve code generation
6268 for speed if the target does not provide an implementation of the
6269 @code{movmem} or @code{setmem} standard names, if the @code{movmem} or
6270 @code{setmem} implementation would be more expensive than a sequence of
6271 insns, or if the overhead of a library call would dominate that of
6272 the body of the memory operation.
6274 Returning true for higher values of @code{size} may also cause an increase
6275 in code size, for example where the number of insns emitted to perform a
6276 move would be greater than that of a library call.
6279 @defmac MOVE_MAX_PIECES
6280 A C expression used by @code{move_by_pieces} to determine the largest unit
6281 a load or store used to copy memory is. Defaults to @code{MOVE_MAX}.
6284 @defmac CLEAR_RATIO (@var{speed})
6285 The threshold of number of scalar move insns, @emph{below} which a sequence
6286 of insns should be generated to clear memory instead of a string clear insn
6287 or a library call. Increasing the value will always make code faster, but
6288 eventually incurs high cost in increased code size.
6290 The parameter @var{speed} is true if the code is currently being
6291 optimized for speed rather than size.
6293 If you don't define this, a reasonable default is used.
6296 @defmac SET_RATIO (@var{speed})
6297 The threshold of number of scalar move insns, @emph{below} which a sequence
6298 of insns should be generated to set memory to a constant value, instead of
6299 a block set insn or a library call.
6300 Increasing the value will always make code faster, but
6301 eventually incurs high cost in increased code size.
6303 The parameter @var{speed} is true if the code is currently being
6304 optimized for speed rather than size.
6306 If you don't define this, it defaults to the value of @code{MOVE_RATIO}.
6309 @defmac USE_LOAD_POST_INCREMENT (@var{mode})
6310 A C expression used to determine whether a load postincrement is a good
6311 thing to use for a given mode. Defaults to the value of
6312 @code{HAVE_POST_INCREMENT}.
6315 @defmac USE_LOAD_POST_DECREMENT (@var{mode})
6316 A C expression used to determine whether a load postdecrement is a good
6317 thing to use for a given mode. Defaults to the value of
6318 @code{HAVE_POST_DECREMENT}.
6321 @defmac USE_LOAD_PRE_INCREMENT (@var{mode})
6322 A C expression used to determine whether a load preincrement is a good
6323 thing to use for a given mode. Defaults to the value of
6324 @code{HAVE_PRE_INCREMENT}.
6327 @defmac USE_LOAD_PRE_DECREMENT (@var{mode})
6328 A C expression used to determine whether a load predecrement is a good
6329 thing to use for a given mode. Defaults to the value of
6330 @code{HAVE_PRE_DECREMENT}.
6333 @defmac USE_STORE_POST_INCREMENT (@var{mode})
6334 A C expression used to determine whether a store postincrement is a good
6335 thing to use for a given mode. Defaults to the value of
6336 @code{HAVE_POST_INCREMENT}.
6339 @defmac USE_STORE_POST_DECREMENT (@var{mode})
6340 A C expression used to determine whether a store postdecrement is a good
6341 thing to use for a given mode. Defaults to the value of
6342 @code{HAVE_POST_DECREMENT}.
6345 @defmac USE_STORE_PRE_INCREMENT (@var{mode})
6346 This macro is used to determine whether a store preincrement is a good
6347 thing to use for a given mode. Defaults to the value of
6348 @code{HAVE_PRE_INCREMENT}.
6351 @defmac USE_STORE_PRE_DECREMENT (@var{mode})
6352 This macro is used to determine whether a store predecrement is a good
6353 thing to use for a given mode. Defaults to the value of
6354 @code{HAVE_PRE_DECREMENT}.
6357 @defmac NO_FUNCTION_CSE
6358 Define this macro if it is as good or better to call a constant
6359 function address than to call an address kept in a register.
6362 @defmac LOGICAL_OP_NON_SHORT_CIRCUIT
6363 Define this macro if a non-short-circuit operation produced by
6364 @samp{fold_range_test ()} is optimal. This macro defaults to true if
6365 @code{BRANCH_COST} is greater than or equal to the value 2.
6368 @deftypefn {Target Hook} bool TARGET_RTX_COSTS (rtx @var{x}, int @var{code}, int @var{outer_code}, int @var{opno}, int *@var{total}, bool @var{speed})
6369 This target hook describes the relative costs of RTL expressions.
6371 The cost may depend on the precise form of the expression, which is
6372 available for examination in @var{x}, and the fact that @var{x} appears
6373 as operand @var{opno} of an expression with rtx code @var{outer_code}.
6374 That is, the hook can assume that there is some rtx @var{y} such
6375 that @samp{GET_CODE (@var{y}) == @var{outer_code}} and such that
6376 either (a) @samp{XEXP (@var{y}, @var{opno}) == @var{x}} or
6377 (b) @samp{XVEC (@var{y}, @var{opno})} contains @var{x}.
6379 @var{code} is @var{x}'s expression code---redundant, since it can be
6380 obtained with @code{GET_CODE (@var{x})}.
6382 In implementing this hook, you can use the construct
6383 @code{COSTS_N_INSNS (@var{n})} to specify a cost equal to @var{n} fast
6386 On entry to the hook, @code{*@var{total}} contains a default estimate
6387 for the cost of the expression. The hook should modify this value as
6388 necessary. Traditionally, the default costs are @code{COSTS_N_INSNS (5)}
6389 for multiplications, @code{COSTS_N_INSNS (7)} for division and modulus
6390 operations, and @code{COSTS_N_INSNS (1)} for all other operations.
6392 When optimizing for code size, i.e.@: when @code{speed} is
6393 false, this target hook should be used to estimate the relative
6394 size cost of an expression, again relative to @code{COSTS_N_INSNS}.
6396 The hook returns true when all subexpressions of @var{x} have been
6397 processed, and false when @code{rtx_cost} should recurse.
6400 @deftypefn {Target Hook} int TARGET_ADDRESS_COST (rtx @var{address}, machine_mode @var{mode}, addr_space_t @var{as}, bool @var{speed})
6401 This hook computes the cost of an addressing mode that contains
6402 @var{address}. If not defined, the cost is computed from
6403 the @var{address} expression and the @code{TARGET_RTX_COST} hook.
6405 For most CISC machines, the default cost is a good approximation of the
6406 true cost of the addressing mode. However, on RISC machines, all
6407 instructions normally have the same length and execution time. Hence
6408 all addresses will have equal costs.
6410 In cases where more than one form of an address is known, the form with
6411 the lowest cost will be used. If multiple forms have the same, lowest,
6412 cost, the one that is the most complex will be used.
6414 For example, suppose an address that is equal to the sum of a register
6415 and a constant is used twice in the same basic block. When this macro
6416 is not defined, the address will be computed in a register and memory
6417 references will be indirect through that register. On machines where
6418 the cost of the addressing mode containing the sum is no higher than
6419 that of a simple indirect reference, this will produce an additional
6420 instruction and possibly require an additional register. Proper
6421 specification of this macro eliminates this overhead for such machines.
6423 This hook is never called with an invalid address.
6425 On machines where an address involving more than one register is as
6426 cheap as an address computation involving only one register, defining
6427 @code{TARGET_ADDRESS_COST} to reflect this can cause two registers to
6428 be live over a region of code where only one would have been if
6429 @code{TARGET_ADDRESS_COST} were not defined in that manner. This effect
6430 should be considered in the definition of this macro. Equivalent costs
6431 should probably only be given to addresses with different numbers of
6432 registers on machines with lots of registers.
6436 @section Adjusting the Instruction Scheduler
6438 The instruction scheduler may need a fair amount of machine-specific
6439 adjustment in order to produce good code. GCC provides several target
6440 hooks for this purpose. It is usually enough to define just a few of
6441 them: try the first ones in this list first.
6443 @deftypefn {Target Hook} int TARGET_SCHED_ISSUE_RATE (void)
6444 This hook returns the maximum number of instructions that can ever
6445 issue at the same time on the target machine. The default is one.
6446 Although the insn scheduler can define itself the possibility of issue
6447 an insn on the same cycle, the value can serve as an additional
6448 constraint to issue insns on the same simulated processor cycle (see
6449 hooks @samp{TARGET_SCHED_REORDER} and @samp{TARGET_SCHED_REORDER2}).
6450 This value must be constant over the entire compilation. If you need
6451 it to vary depending on what the instructions are, you must use
6452 @samp{TARGET_SCHED_VARIABLE_ISSUE}.
6455 @deftypefn {Target Hook} int TARGET_SCHED_VARIABLE_ISSUE (FILE *@var{file}, int @var{verbose}, rtx_insn *@var{insn}, int @var{more})
6456 This hook is executed by the scheduler after it has scheduled an insn
6457 from the ready list. It should return the number of insns which can
6458 still be issued in the current cycle. The default is
6459 @samp{@w{@var{more} - 1}} for insns other than @code{CLOBBER} and
6460 @code{USE}, which normally are not counted against the issue rate.
6461 You should define this hook if some insns take more machine resources
6462 than others, so that fewer insns can follow them in the same cycle.
6463 @var{file} is either a null pointer, or a stdio stream to write any
6464 debug output to. @var{verbose} is the verbose level provided by
6465 @option{-fsched-verbose-@var{n}}. @var{insn} is the instruction that
6469 @deftypefn {Target Hook} int TARGET_SCHED_ADJUST_COST (rtx_insn *@var{insn}, rtx @var{link}, rtx_insn *@var{dep_insn}, int @var{cost})
6470 This function corrects the value of @var{cost} based on the
6471 relationship between @var{insn} and @var{dep_insn} through the
6472 dependence @var{link}. It should return the new value. The default
6473 is to make no adjustment to @var{cost}. This can be used for example
6474 to specify to the scheduler using the traditional pipeline description
6475 that an output- or anti-dependence does not incur the same cost as a
6476 data-dependence. If the scheduler using the automaton based pipeline
6477 description, the cost of anti-dependence is zero and the cost of
6478 output-dependence is maximum of one and the difference of latency
6479 times of the first and the second insns. If these values are not
6480 acceptable, you could use the hook to modify them too. See also
6481 @pxref{Processor pipeline description}.
6484 @deftypefn {Target Hook} int TARGET_SCHED_ADJUST_PRIORITY (rtx_insn *@var{insn}, int @var{priority})
6485 This hook adjusts the integer scheduling priority @var{priority} of
6486 @var{insn}. It should return the new priority. Increase the priority to
6487 execute @var{insn} earlier, reduce the priority to execute @var{insn}
6488 later. Do not define this hook if you do not need to adjust the
6489 scheduling priorities of insns.
6492 @deftypefn {Target Hook} int TARGET_SCHED_REORDER (FILE *@var{file}, int @var{verbose}, rtx_insn **@var{ready}, int *@var{n_readyp}, int @var{clock})
6493 This hook is executed by the scheduler after it has scheduled the ready
6494 list, to allow the machine description to reorder it (for example to
6495 combine two small instructions together on @samp{VLIW} machines).
6496 @var{file} is either a null pointer, or a stdio stream to write any
6497 debug output to. @var{verbose} is the verbose level provided by
6498 @option{-fsched-verbose-@var{n}}. @var{ready} is a pointer to the ready
6499 list of instructions that are ready to be scheduled. @var{n_readyp} is
6500 a pointer to the number of elements in the ready list. The scheduler
6501 reads the ready list in reverse order, starting with
6502 @var{ready}[@var{*n_readyp} @minus{} 1] and going to @var{ready}[0]. @var{clock}
6503 is the timer tick of the scheduler. You may modify the ready list and
6504 the number of ready insns. The return value is the number of insns that
6505 can issue this cycle; normally this is just @code{issue_rate}. See also
6506 @samp{TARGET_SCHED_REORDER2}.
6509 @deftypefn {Target Hook} int TARGET_SCHED_REORDER2 (FILE *@var{file}, int @var{verbose}, rtx_insn **@var{ready}, int *@var{n_readyp}, int @var{clock})
6510 Like @samp{TARGET_SCHED_REORDER}, but called at a different time. That
6511 function is called whenever the scheduler starts a new cycle. This one
6512 is called once per iteration over a cycle, immediately after
6513 @samp{TARGET_SCHED_VARIABLE_ISSUE}; it can reorder the ready list and
6514 return the number of insns to be scheduled in the same cycle. Defining
6515 this hook can be useful if there are frequent situations where
6516 scheduling one insn causes other insns to become ready in the same
6517 cycle. These other insns can then be taken into account properly.
6520 @deftypefn {Target Hook} bool TARGET_SCHED_MACRO_FUSION_P (void)
6521 This hook is used to check whether target platform supports macro fusion.
6524 @deftypefn {Target Hook} bool TARGET_SCHED_MACRO_FUSION_PAIR_P (rtx_insn *@var{prev}, rtx_insn *@var{curr})
6525 This hook is used to check whether two insns should be macro fused for
6526 a target microarchitecture. If this hook returns true for the given insn pair
6527 (@var{prev} and @var{curr}), the scheduler will put them into a sched
6528 group, and they will not be scheduled apart. The two insns will be either
6529 two SET insns or a compare and a conditional jump and this hook should
6530 validate any dependencies needed to fuse the two insns together.
6533 @deftypefn {Target Hook} void TARGET_SCHED_DEPENDENCIES_EVALUATION_HOOK (rtx_insn *@var{head}, rtx_insn *@var{tail})
6534 This hook is called after evaluation forward dependencies of insns in
6535 chain given by two parameter values (@var{head} and @var{tail}
6536 correspondingly) but before insns scheduling of the insn chain. For
6537 example, it can be used for better insn classification if it requires
6538 analysis of dependencies. This hook can use backward and forward
6539 dependencies of the insn scheduler because they are already
6543 @deftypefn {Target Hook} void TARGET_SCHED_INIT (FILE *@var{file}, int @var{verbose}, int @var{max_ready})
6544 This hook is executed by the scheduler at the beginning of each block of
6545 instructions that are to be scheduled. @var{file} is either a null
6546 pointer, or a stdio stream to write any debug output to. @var{verbose}
6547 is the verbose level provided by @option{-fsched-verbose-@var{n}}.
6548 @var{max_ready} is the maximum number of insns in the current scheduling
6549 region that can be live at the same time. This can be used to allocate
6550 scratch space if it is needed, e.g.@: by @samp{TARGET_SCHED_REORDER}.
6553 @deftypefn {Target Hook} void TARGET_SCHED_FINISH (FILE *@var{file}, int @var{verbose})
6554 This hook is executed by the scheduler at the end of each block of
6555 instructions that are to be scheduled. It can be used to perform
6556 cleanup of any actions done by the other scheduling hooks. @var{file}
6557 is either a null pointer, or a stdio stream to write any debug output
6558 to. @var{verbose} is the verbose level provided by
6559 @option{-fsched-verbose-@var{n}}.
6562 @deftypefn {Target Hook} void TARGET_SCHED_INIT_GLOBAL (FILE *@var{file}, int @var{verbose}, int @var{old_max_uid})
6563 This hook is executed by the scheduler after function level initializations.
6564 @var{file} is either a null pointer, or a stdio stream to write any debug output to.
6565 @var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}.
6566 @var{old_max_uid} is the maximum insn uid when scheduling begins.
6569 @deftypefn {Target Hook} void TARGET_SCHED_FINISH_GLOBAL (FILE *@var{file}, int @var{verbose})
6570 This is the cleanup hook corresponding to @code{TARGET_SCHED_INIT_GLOBAL}.
6571 @var{file} is either a null pointer, or a stdio stream to write any debug output to.
6572 @var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}.
6575 @deftypefn {Target Hook} rtx TARGET_SCHED_DFA_PRE_CYCLE_INSN (void)
6576 The hook returns an RTL insn. The automaton state used in the
6577 pipeline hazard recognizer is changed as if the insn were scheduled
6578 when the new simulated processor cycle starts. Usage of the hook may
6579 simplify the automaton pipeline description for some @acronym{VLIW}
6580 processors. If the hook is defined, it is used only for the automaton
6581 based pipeline description. The default is not to change the state
6582 when the new simulated processor cycle starts.
6585 @deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN (void)
6586 The hook can be used to initialize data used by the previous hook.
6589 @deftypefn {Target Hook} {rtx_insn *} TARGET_SCHED_DFA_POST_CYCLE_INSN (void)
6590 The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used
6591 to changed the state as if the insn were scheduled when the new
6592 simulated processor cycle finishes.
6595 @deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_POST_CYCLE_INSN (void)
6596 The hook is analogous to @samp{TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN} but
6597 used to initialize data used by the previous hook.
6600 @deftypefn {Target Hook} void TARGET_SCHED_DFA_PRE_ADVANCE_CYCLE (void)
6601 The hook to notify target that the current simulated cycle is about to finish.
6602 The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used
6603 to change the state in more complicated situations - e.g., when advancing
6604 state on a single insn is not enough.
6607 @deftypefn {Target Hook} void TARGET_SCHED_DFA_POST_ADVANCE_CYCLE (void)
6608 The hook to notify target that new simulated cycle has just started.
6609 The hook is analogous to @samp{TARGET_SCHED_DFA_POST_CYCLE_INSN} but used
6610 to change the state in more complicated situations - e.g., when advancing
6611 state on a single insn is not enough.
6614 @deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD (void)
6615 This hook controls better choosing an insn from the ready insn queue
6616 for the @acronym{DFA}-based insn scheduler. Usually the scheduler
6617 chooses the first insn from the queue. If the hook returns a positive
6618 value, an additional scheduler code tries all permutations of
6619 @samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD ()}
6620 subsequent ready insns to choose an insn whose issue will result in
6621 maximal number of issued insns on the same cycle. For the
6622 @acronym{VLIW} processor, the code could actually solve the problem of
6623 packing simple insns into the @acronym{VLIW} insn. Of course, if the
6624 rules of @acronym{VLIW} packing are described in the automaton.
6626 This code also could be used for superscalar @acronym{RISC}
6627 processors. Let us consider a superscalar @acronym{RISC} processor
6628 with 3 pipelines. Some insns can be executed in pipelines @var{A} or
6629 @var{B}, some insns can be executed only in pipelines @var{B} or
6630 @var{C}, and one insn can be executed in pipeline @var{B}. The
6631 processor may issue the 1st insn into @var{A} and the 2nd one into
6632 @var{B}. In this case, the 3rd insn will wait for freeing @var{B}
6633 until the next cycle. If the scheduler issues the 3rd insn the first,
6634 the processor could issue all 3 insns per cycle.
6636 Actually this code demonstrates advantages of the automaton based
6637 pipeline hazard recognizer. We try quickly and easy many insn
6638 schedules to choose the best one.
6640 The default is no multipass scheduling.
6643 @deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD (rtx_insn *@var{insn}, int @var{ready_index})
6645 This hook controls what insns from the ready insn queue will be
6646 considered for the multipass insn scheduling. If the hook returns
6647 zero for @var{insn}, the insn will be considered in multipass scheduling.
6648 Positive return values will remove @var{insn} from consideration on
6649 the current round of multipass scheduling.
6650 Negative return values will remove @var{insn} from consideration for given
6652 Backends should be careful about returning non-zero for highest priority
6653 instruction at position 0 in the ready list. @var{ready_index} is passed
6654 to allow backends make correct judgements.
6656 The default is that any ready insns can be chosen to be issued.
6659 @deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_BEGIN (void *@var{data}, signed char *@var{ready_try}, int @var{n_ready}, bool @var{first_cycle_insn_p})
6660 This hook prepares the target backend for a new round of multipass
6664 @deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_ISSUE (void *@var{data}, signed char *@var{ready_try}, int @var{n_ready}, rtx_insn *@var{insn}, const void *@var{prev_data})
6665 This hook is called when multipass scheduling evaluates instruction INSN.
6668 @deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_BACKTRACK (const void *@var{data}, signed char *@var{ready_try}, int @var{n_ready})
6669 This is called when multipass scheduling backtracks from evaluation of
6673 @deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_END (const void *@var{data})
6674 This hook notifies the target about the result of the concluded current
6675 round of multipass scheduling.
6678 @deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_INIT (void *@var{data})
6679 This hook initializes target-specific data used in multipass scheduling.
6682 @deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_FINI (void *@var{data})
6683 This hook finalizes target-specific data used in multipass scheduling.
6686 @deftypefn {Target Hook} int TARGET_SCHED_DFA_NEW_CYCLE (FILE *@var{dump}, int @var{verbose}, rtx_insn *@var{insn}, int @var{last_clock}, int @var{clock}, int *@var{sort_p})
6687 This hook is called by the insn scheduler before issuing @var{insn}
6688 on cycle @var{clock}. If the hook returns nonzero,
6689 @var{insn} is not issued on this processor cycle. Instead,
6690 the processor cycle is advanced. If *@var{sort_p}
6691 is zero, the insn ready queue is not sorted on the new cycle
6692 start as usually. @var{dump} and @var{verbose} specify the file and
6693 verbosity level to use for debugging output.
6694 @var{last_clock} and @var{clock} are, respectively, the
6695 processor cycle on which the previous insn has been issued,
6696 and the current processor cycle.
6699 @deftypefn {Target Hook} bool TARGET_SCHED_IS_COSTLY_DEPENDENCE (struct _dep *@var{_dep}, int @var{cost}, int @var{distance})
6700 This hook is used to define which dependences are considered costly by
6701 the target, so costly that it is not advisable to schedule the insns that
6702 are involved in the dependence too close to one another. The parameters
6703 to this hook are as follows: The first parameter @var{_dep} is the dependence
6704 being evaluated. The second parameter @var{cost} is the cost of the
6705 dependence as estimated by the scheduler, and the third
6706 parameter @var{distance} is the distance in cycles between the two insns.
6707 The hook returns @code{true} if considering the distance between the two
6708 insns the dependence between them is considered costly by the target,
6709 and @code{false} otherwise.
6711 Defining this hook can be useful in multiple-issue out-of-order machines,
6712 where (a) it's practically hopeless to predict the actual data/resource
6713 delays, however: (b) there's a better chance to predict the actual grouping
6714 that will be formed, and (c) correctly emulating the grouping can be very
6715 important. In such targets one may want to allow issuing dependent insns
6716 closer to one another---i.e., closer than the dependence distance; however,
6717 not in cases of ``costly dependences'', which this hooks allows to define.
6720 @deftypefn {Target Hook} void TARGET_SCHED_H_I_D_EXTENDED (void)
6721 This hook is called by the insn scheduler after emitting a new instruction to
6722 the instruction stream. The hook notifies a target backend to extend its
6723 per instruction data structures.
6726 @deftypefn {Target Hook} {void *} TARGET_SCHED_ALLOC_SCHED_CONTEXT (void)
6727 Return a pointer to a store large enough to hold target scheduling context.
6730 @deftypefn {Target Hook} void TARGET_SCHED_INIT_SCHED_CONTEXT (void *@var{tc}, bool @var{clean_p})
6731 Initialize store pointed to by @var{tc} to hold target scheduling context.
6732 It @var{clean_p} is true then initialize @var{tc} as if scheduler is at the
6733 beginning of the block. Otherwise, copy the current context into @var{tc}.
6736 @deftypefn {Target Hook} void TARGET_SCHED_SET_SCHED_CONTEXT (void *@var{tc})
6737 Copy target scheduling context pointed to by @var{tc} to the current context.
6740 @deftypefn {Target Hook} void TARGET_SCHED_CLEAR_SCHED_CONTEXT (void *@var{tc})
6741 Deallocate internal data in target scheduling context pointed to by @var{tc}.
6744 @deftypefn {Target Hook} void TARGET_SCHED_FREE_SCHED_CONTEXT (void *@var{tc})
6745 Deallocate a store for target scheduling context pointed to by @var{tc}.
6748 @deftypefn {Target Hook} int TARGET_SCHED_SPECULATE_INSN (rtx_insn *@var{insn}, unsigned int @var{dep_status}, rtx *@var{new_pat})
6749 This hook is called by the insn scheduler when @var{insn} has only
6750 speculative dependencies and therefore can be scheduled speculatively.
6751 The hook is used to check if the pattern of @var{insn} has a speculative
6752 version and, in case of successful check, to generate that speculative
6753 pattern. The hook should return 1, if the instruction has a speculative form,
6754 or @minus{}1, if it doesn't. @var{request} describes the type of requested
6755 speculation. If the return value equals 1 then @var{new_pat} is assigned
6756 the generated speculative pattern.
6759 @deftypefn {Target Hook} bool TARGET_SCHED_NEEDS_BLOCK_P (unsigned int @var{dep_status})
6760 This hook is called by the insn scheduler during generation of recovery code
6761 for @var{insn}. It should return @code{true}, if the corresponding check
6762 instruction should branch to recovery code, or @code{false} otherwise.
6765 @deftypefn {Target Hook} rtx TARGET_SCHED_GEN_SPEC_CHECK (rtx_insn *@var{insn}, rtx_insn *@var{label}, unsigned int @var{ds})
6766 This hook is called by the insn scheduler to generate a pattern for recovery
6767 check instruction. If @var{mutate_p} is zero, then @var{insn} is a
6768 speculative instruction for which the check should be generated.
6769 @var{label} is either a label of a basic block, where recovery code should
6770 be emitted, or a null pointer, when requested check doesn't branch to
6771 recovery code (a simple check). If @var{mutate_p} is nonzero, then
6772 a pattern for a branchy check corresponding to a simple check denoted by
6773 @var{insn} should be generated. In this case @var{label} can't be null.
6776 @deftypefn {Target Hook} void TARGET_SCHED_SET_SCHED_FLAGS (struct spec_info_def *@var{spec_info})
6777 This hook is used by the insn scheduler to find out what features should be
6779 The structure *@var{spec_info} should be filled in by the target.
6780 The structure describes speculation types that can be used in the scheduler.
6783 @deftypefn {Target Hook} int TARGET_SCHED_SMS_RES_MII (struct ddg *@var{g})
6784 This hook is called by the swing modulo scheduler to calculate a
6785 resource-based lower bound which is based on the resources available in
6786 the machine and the resources required by each instruction. The target
6787 backend can use @var{g} to calculate such bound. A very simple lower
6788 bound will be used in case this hook is not implemented: the total number
6789 of instructions divided by the issue rate.
6792 @deftypefn {Target Hook} bool TARGET_SCHED_DISPATCH (rtx_insn *@var{insn}, int @var{x})
6793 This hook is called by Haifa Scheduler. It returns true if dispatch scheduling
6794 is supported in hardware and the condition specified in the parameter is true.
6797 @deftypefn {Target Hook} void TARGET_SCHED_DISPATCH_DO (rtx_insn *@var{insn}, int @var{x})
6798 This hook is called by Haifa Scheduler. It performs the operation specified
6799 in its second parameter.
6802 @deftypevr {Target Hook} bool TARGET_SCHED_EXPOSED_PIPELINE
6803 True if the processor has an exposed pipeline, which means that not just
6804 the order of instructions is important for correctness when scheduling, but
6805 also the latencies of operations.
6808 @deftypefn {Target Hook} int TARGET_SCHED_REASSOCIATION_WIDTH (unsigned int @var{opc}, machine_mode @var{mode})
6809 This hook is called by tree reassociator to determine a level of
6810 parallelism required in output calculations chain.
6813 @deftypefn {Target Hook} void TARGET_SCHED_FUSION_PRIORITY (rtx_insn *@var{insn}, int @var{max_pri}, int *@var{fusion_pri}, int *@var{pri})
6814 This hook is called by scheduling fusion pass. It calculates fusion
6815 priorities for each instruction passed in by parameter. The priorities
6816 are returned via pointer parameters.
6818 @var{insn} is the instruction whose priorities need to be calculated.
6819 @var{max_pri} is the maximum priority can be returned in any cases.
6820 @var{fusion_pri} is the pointer parameter through which @var{insn}'s
6821 fusion priority should be calculated and returned.
6822 @var{pri} is the pointer parameter through which @var{insn}'s priority
6823 should be calculated and returned.
6825 Same @var{fusion_pri} should be returned for instructions which should
6826 be scheduled together. Different @var{pri} should be returned for
6827 instructions with same @var{fusion_pri}. @var{fusion_pri} is the major
6828 sort key, @var{pri} is the minor sort key. All instructions will be
6829 scheduled according to the two priorities. All priorities calculated
6830 should be between 0 (exclusive) and @var{max_pri} (inclusive). To avoid
6831 false dependencies, @var{fusion_pri} of instructions which need to be
6832 scheduled together should be smaller than @var{fusion_pri} of irrelevant
6835 Given below example:
6848 On targets like ARM/AArch64, the two pairs of consecutive loads should be
6849 merged. Since peephole2 pass can't help in this case unless consecutive
6850 loads are actually next to each other in instruction flow. That's where
6851 this scheduling fusion pass works. This hook calculates priority for each
6852 instruction based on its fustion type, like:
6855 ldr r10, [r1, 4] ; fusion_pri=99, pri=96
6856 add r4, r4, r10 ; fusion_pri=100, pri=100
6857 ldr r15, [r2, 8] ; fusion_pri=98, pri=92
6858 sub r5, r5, r15 ; fusion_pri=100, pri=100
6859 ldr r11, [r1, 0] ; fusion_pri=99, pri=100
6860 add r4, r4, r11 ; fusion_pri=100, pri=100
6861 ldr r16, [r2, 12] ; fusion_pri=98, pri=88
6862 sub r5, r5, r16 ; fusion_pri=100, pri=100
6865 Scheduling fusion pass then sorts all ready to issue instructions according
6866 to the priorities. As a result, instructions of same fusion type will be
6867 pushed together in instruction flow, like:
6880 Now peephole2 pass can simply merge the two pairs of loads.
6882 Since scheduling fusion pass relies on peephole2 to do real fusion
6883 work, it is only enabled by default when peephole2 is in effect.
6885 This is firstly introduced on ARM/AArch64 targets, please refer to
6886 the hook implementation for how different fusion types are supported.
6890 @section Dividing the Output into Sections (Texts, Data, @dots{})
6891 @c the above section title is WAY too long. maybe cut the part between
6892 @c the (...)? --mew 10feb93
6894 An object file is divided into sections containing different types of
6895 data. In the most common case, there are three sections: the @dfn{text
6896 section}, which holds instructions and read-only data; the @dfn{data
6897 section}, which holds initialized writable data; and the @dfn{bss
6898 section}, which holds uninitialized data. Some systems have other kinds
6901 @file{varasm.c} provides several well-known sections, such as
6902 @code{text_section}, @code{data_section} and @code{bss_section}.
6903 The normal way of controlling a @code{@var{foo}_section} variable
6904 is to define the associated @code{@var{FOO}_SECTION_ASM_OP} macro,
6905 as described below. The macros are only read once, when @file{varasm.c}
6906 initializes itself, so their values must be run-time constants.
6907 They may however depend on command-line flags.
6909 @emph{Note:} Some run-time files, such @file{crtstuff.c}, also make
6910 use of the @code{@var{FOO}_SECTION_ASM_OP} macros, and expect them
6911 to be string literals.
6913 Some assemblers require a different string to be written every time a
6914 section is selected. If your assembler falls into this category, you
6915 should define the @code{TARGET_ASM_INIT_SECTIONS} hook and use
6916 @code{get_unnamed_section} to set up the sections.
6918 You must always create a @code{text_section}, either by defining
6919 @code{TEXT_SECTION_ASM_OP} or by initializing @code{text_section}
6920 in @code{TARGET_ASM_INIT_SECTIONS}. The same is true of
6921 @code{data_section} and @code{DATA_SECTION_ASM_OP}. If you do not
6922 create a distinct @code{readonly_data_section}, the default is to
6923 reuse @code{text_section}.
6925 All the other @file{varasm.c} sections are optional, and are null
6926 if the target does not provide them.
6928 @defmac TEXT_SECTION_ASM_OP
6929 A C expression whose value is a string, including spacing, containing the
6930 assembler operation that should precede instructions and read-only data.
6931 Normally @code{"\t.text"} is right.
6934 @defmac HOT_TEXT_SECTION_NAME
6935 If defined, a C string constant for the name of the section containing most
6936 frequently executed functions of the program. If not defined, GCC will provide
6937 a default definition if the target supports named sections.
6940 @defmac UNLIKELY_EXECUTED_TEXT_SECTION_NAME
6941 If defined, a C string constant for the name of the section containing unlikely
6942 executed functions in the program.
6945 @defmac DATA_SECTION_ASM_OP
6946 A C expression whose value is a string, including spacing, containing the
6947 assembler operation to identify the following data as writable initialized
6948 data. Normally @code{"\t.data"} is right.
6951 @defmac SDATA_SECTION_ASM_OP
6952 If defined, a C expression whose value is a string, including spacing,
6953 containing the assembler operation to identify the following data as
6954 initialized, writable small data.
6957 @defmac READONLY_DATA_SECTION_ASM_OP
6958 A C expression whose value is a string, including spacing, containing the
6959 assembler operation to identify the following data as read-only initialized
6963 @defmac BSS_SECTION_ASM_OP
6964 If defined, a C expression whose value is a string, including spacing,
6965 containing the assembler operation to identify the following data as
6966 uninitialized global data. If not defined, and
6967 @code{ASM_OUTPUT_ALIGNED_BSS} not defined,
6968 uninitialized global data will be output in the data section if
6969 @option{-fno-common} is passed, otherwise @code{ASM_OUTPUT_COMMON} will be
6973 @defmac SBSS_SECTION_ASM_OP
6974 If defined, a C expression whose value is a string, including spacing,
6975 containing the assembler operation to identify the following data as
6976 uninitialized, writable small data.
6979 @defmac TLS_COMMON_ASM_OP
6980 If defined, a C expression whose value is a string containing the
6981 assembler operation to identify the following data as thread-local
6982 common data. The default is @code{".tls_common"}.
6985 @defmac TLS_SECTION_ASM_FLAG
6986 If defined, a C expression whose value is a character constant
6987 containing the flag used to mark a section as a TLS section. The
6988 default is @code{'T'}.
6991 @defmac INIT_SECTION_ASM_OP
6992 If defined, a C expression whose value is a string, including spacing,
6993 containing the assembler operation to identify the following data as
6994 initialization code. If not defined, GCC will assume such a section does
6995 not exist. This section has no corresponding @code{init_section}
6996 variable; it is used entirely in runtime code.
6999 @defmac FINI_SECTION_ASM_OP
7000 If defined, a C expression whose value is a string, including spacing,
7001 containing the assembler operation to identify the following data as
7002 finalization code. If not defined, GCC will assume such a section does
7003 not exist. This section has no corresponding @code{fini_section}
7004 variable; it is used entirely in runtime code.
7007 @defmac INIT_ARRAY_SECTION_ASM_OP
7008 If defined, a C expression whose value is a string, including spacing,
7009 containing the assembler operation to identify the following data as
7010 part of the @code{.init_array} (or equivalent) section. If not
7011 defined, GCC will assume such a section does not exist. Do not define
7012 both this macro and @code{INIT_SECTION_ASM_OP}.
7015 @defmac FINI_ARRAY_SECTION_ASM_OP
7016 If defined, a C expression whose value is a string, including spacing,
7017 containing the assembler operation to identify the following data as
7018 part of the @code{.fini_array} (or equivalent) section. If not
7019 defined, GCC will assume such a section does not exist. Do not define
7020 both this macro and @code{FINI_SECTION_ASM_OP}.
7023 @defmac CRT_CALL_STATIC_FUNCTION (@var{section_op}, @var{function})
7024 If defined, an ASM statement that switches to a different section
7025 via @var{section_op}, calls @var{function}, and switches back to
7026 the text section. This is used in @file{crtstuff.c} if
7027 @code{INIT_SECTION_ASM_OP} or @code{FINI_SECTION_ASM_OP} to calls
7028 to initialization and finalization functions from the init and fini
7029 sections. By default, this macro uses a simple function call. Some
7030 ports need hand-crafted assembly code to avoid dependencies on
7031 registers initialized in the function prologue or to ensure that
7032 constant pools don't end up too far way in the text section.
7035 @defmac TARGET_LIBGCC_SDATA_SECTION
7036 If defined, a string which names the section into which small
7037 variables defined in crtstuff and libgcc should go. This is useful
7038 when the target has options for optimizing access to small data, and
7039 you want the crtstuff and libgcc routines to be conservative in what
7040 they expect of your application yet liberal in what your application
7041 expects. For example, for targets with a @code{.sdata} section (like
7042 MIPS), you could compile crtstuff with @code{-G 0} so that it doesn't
7043 require small data support from your application, but use this macro
7044 to put small data into @code{.sdata} so that your application can
7045 access these variables whether it uses small data or not.
7048 @defmac FORCE_CODE_SECTION_ALIGN
7049 If defined, an ASM statement that aligns a code section to some
7050 arbitrary boundary. This is used to force all fragments of the
7051 @code{.init} and @code{.fini} sections to have to same alignment
7052 and thus prevent the linker from having to add any padding.
7055 @defmac JUMP_TABLES_IN_TEXT_SECTION
7056 Define this macro to be an expression with a nonzero value if jump
7057 tables (for @code{tablejump} insns) should be output in the text
7058 section, along with the assembler instructions. Otherwise, the
7059 readonly data section is used.
7061 This macro is irrelevant if there is no separate readonly data section.
7064 @deftypefn {Target Hook} void TARGET_ASM_INIT_SECTIONS (void)
7065 Define this hook if you need to do something special to set up the
7066 @file{varasm.c} sections, or if your target has some special sections
7067 of its own that you need to create.
7069 GCC calls this hook after processing the command line, but before writing
7070 any assembly code, and before calling any of the section-returning hooks
7074 @deftypefn {Target Hook} int TARGET_ASM_RELOC_RW_MASK (void)
7075 Return a mask describing how relocations should be treated when
7076 selecting sections. Bit 1 should be set if global relocations
7077 should be placed in a read-write section; bit 0 should be set if
7078 local relocations should be placed in a read-write section.
7080 The default version of this function returns 3 when @option{-fpic}
7081 is in effect, and 0 otherwise. The hook is typically redefined
7082 when the target cannot support (some kinds of) dynamic relocations
7083 in read-only sections even in executables.
7086 @deftypefn {Target Hook} {section *} TARGET_ASM_SELECT_SECTION (tree @var{exp}, int @var{reloc}, unsigned HOST_WIDE_INT @var{align})
7087 Return the section into which @var{exp} should be placed. You can
7088 assume that @var{exp} is either a @code{VAR_DECL} node or a constant of
7089 some sort. @var{reloc} indicates whether the initial value of @var{exp}
7090 requires link-time relocations. Bit 0 is set when variable contains
7091 local relocations only, while bit 1 is set for global relocations.
7092 @var{align} is the constant alignment in bits.
7094 The default version of this function takes care of putting read-only
7095 variables in @code{readonly_data_section}.
7097 See also @var{USE_SELECT_SECTION_FOR_FUNCTIONS}.
7100 @defmac USE_SELECT_SECTION_FOR_FUNCTIONS
7101 Define this macro if you wish TARGET_ASM_SELECT_SECTION to be called
7102 for @code{FUNCTION_DECL}s as well as for variables and constants.
7104 In the case of a @code{FUNCTION_DECL}, @var{reloc} will be zero if the
7105 function has been determined to be likely to be called, and nonzero if
7106 it is unlikely to be called.
7109 @deftypefn {Target Hook} void TARGET_ASM_UNIQUE_SECTION (tree @var{decl}, int @var{reloc})
7110 Build up a unique section name, expressed as a @code{STRING_CST} node,
7111 and assign it to @samp{DECL_SECTION_NAME (@var{decl})}.
7112 As with @code{TARGET_ASM_SELECT_SECTION}, @var{reloc} indicates whether
7113 the initial value of @var{exp} requires link-time relocations.
7115 The default version of this function appends the symbol name to the
7116 ELF section name that would normally be used for the symbol. For
7117 example, the function @code{foo} would be placed in @code{.text.foo}.
7118 Whatever the actual target object format, this is often good enough.
7121 @deftypefn {Target Hook} {section *} TARGET_ASM_FUNCTION_RODATA_SECTION (tree @var{decl})
7122 Return the readonly data section associated with
7123 @samp{DECL_SECTION_NAME (@var{decl})}.
7124 The default version of this function selects @code{.gnu.linkonce.r.name} if
7125 the function's section is @code{.gnu.linkonce.t.name}, @code{.rodata.name}
7126 if function is in @code{.text.name}, and the normal readonly-data section
7130 @deftypevr {Target Hook} {const char *} TARGET_ASM_MERGEABLE_RODATA_PREFIX
7131 Usually, the compiler uses the prefix @code{".rodata"} to construct
7132 section names for mergeable constant data. Define this macro to override
7133 the string if a different section name should be used.
7136 @deftypefn {Target Hook} {section *} TARGET_ASM_TM_CLONE_TABLE_SECTION (void)
7137 Return the section that should be used for transactional memory clone tables.
7140 @deftypefn {Target Hook} {section *} TARGET_ASM_SELECT_RTX_SECTION (machine_mode @var{mode}, rtx @var{x}, unsigned HOST_WIDE_INT @var{align})
7141 Return the section into which a constant @var{x}, of mode @var{mode},
7142 should be placed. You can assume that @var{x} is some kind of
7143 constant in RTL@. The argument @var{mode} is redundant except in the
7144 case of a @code{const_int} rtx. @var{align} is the constant alignment
7147 The default version of this function takes care of putting symbolic
7148 constants in @code{flag_pic} mode in @code{data_section} and everything
7149 else in @code{readonly_data_section}.
7152 @deftypefn {Target Hook} tree TARGET_MANGLE_DECL_ASSEMBLER_NAME (tree @var{decl}, tree @var{id})
7153 Define this hook if you need to postprocess the assembler name generated
7154 by target-independent code. The @var{id} provided to this hook will be
7155 the computed name (e.g., the macro @code{DECL_NAME} of the @var{decl} in C,
7156 or the mangled name of the @var{decl} in C++). The return value of the
7157 hook is an @code{IDENTIFIER_NODE} for the appropriate mangled name on
7158 your target system. The default implementation of this hook just
7159 returns the @var{id} provided.
7162 @deftypefn {Target Hook} void TARGET_ENCODE_SECTION_INFO (tree @var{decl}, rtx @var{rtl}, int @var{new_decl_p})
7163 Define this hook if references to a symbol or a constant must be
7164 treated differently depending on something about the variable or
7165 function named by the symbol (such as what section it is in).
7167 The hook is executed immediately after rtl has been created for
7168 @var{decl}, which may be a variable or function declaration or
7169 an entry in the constant pool. In either case, @var{rtl} is the
7170 rtl in question. Do @emph{not} use @code{DECL_RTL (@var{decl})}
7171 in this hook; that field may not have been initialized yet.
7173 In the case of a constant, it is safe to assume that the rtl is
7174 a @code{mem} whose address is a @code{symbol_ref}. Most decls
7175 will also have this form, but that is not guaranteed. Global
7176 register variables, for instance, will have a @code{reg} for their
7177 rtl. (Normally the right thing to do with such unusual rtl is
7180 The @var{new_decl_p} argument will be true if this is the first time
7181 that @code{TARGET_ENCODE_SECTION_INFO} has been invoked on this decl. It will
7182 be false for subsequent invocations, which will happen for duplicate
7183 declarations. Whether or not anything must be done for the duplicate
7184 declaration depends on whether the hook examines @code{DECL_ATTRIBUTES}.
7185 @var{new_decl_p} is always true when the hook is called for a constant.
7187 @cindex @code{SYMBOL_REF_FLAG}, in @code{TARGET_ENCODE_SECTION_INFO}
7188 The usual thing for this hook to do is to record flags in the
7189 @code{symbol_ref}, using @code{SYMBOL_REF_FLAG} or @code{SYMBOL_REF_FLAGS}.
7190 Historically, the name string was modified if it was necessary to
7191 encode more than one bit of information, but this practice is now
7192 discouraged; use @code{SYMBOL_REF_FLAGS}.
7194 The default definition of this hook, @code{default_encode_section_info}
7195 in @file{varasm.c}, sets a number of commonly-useful bits in
7196 @code{SYMBOL_REF_FLAGS}. Check whether the default does what you need
7197 before overriding it.
7200 @deftypefn {Target Hook} {const char *} TARGET_STRIP_NAME_ENCODING (const char *@var{name})
7201 Decode @var{name} and return the real name part, sans
7202 the characters that @code{TARGET_ENCODE_SECTION_INFO}
7206 @deftypefn {Target Hook} bool TARGET_IN_SMALL_DATA_P (const_tree @var{exp})
7207 Returns true if @var{exp} should be placed into a ``small data'' section.
7208 The default version of this hook always returns false.
7211 @deftypevr {Target Hook} bool TARGET_HAVE_SRODATA_SECTION
7212 Contains the value true if the target places read-only
7213 ``small data'' into a separate section. The default value is false.
7216 @deftypefn {Target Hook} bool TARGET_PROFILE_BEFORE_PROLOGUE (void)
7217 It returns true if target wants profile code emitted before prologue.
7219 The default version of this hook use the target macro
7220 @code{PROFILE_BEFORE_PROLOGUE}.
7223 @deftypefn {Target Hook} bool TARGET_BINDS_LOCAL_P (const_tree @var{exp})
7224 Returns true if @var{exp} names an object for which name resolution
7225 rules must resolve to the current ``module'' (dynamic shared library
7226 or executable image).
7228 The default version of this hook implements the name resolution rules
7229 for ELF, which has a looser model of global name binding than other
7230 currently supported object file formats.
7233 @deftypevr {Target Hook} bool TARGET_HAVE_TLS
7234 Contains the value true if the target supports thread-local storage.
7235 The default value is false.
7240 @section Position Independent Code
7241 @cindex position independent code
7244 This section describes macros that help implement generation of position
7245 independent code. Simply defining these macros is not enough to
7246 generate valid PIC; you must also add support to the hook
7247 @code{TARGET_LEGITIMATE_ADDRESS_P} and to the macro
7248 @code{PRINT_OPERAND_ADDRESS}, as well as @code{LEGITIMIZE_ADDRESS}. You
7249 must modify the definition of @samp{movsi} to do something appropriate
7250 when the source operand contains a symbolic address. You may also
7251 need to alter the handling of switch statements so that they use
7253 @c i rearranged the order of the macros above to try to force one of
7254 @c them to the next line, to eliminate an overfull hbox. --mew 10feb93
7256 @defmac PIC_OFFSET_TABLE_REGNUM
7257 The register number of the register used to address a table of static
7258 data addresses in memory. In some cases this register is defined by a
7259 processor's ``application binary interface'' (ABI)@. When this macro
7260 is defined, RTL is generated for this register once, as with the stack
7261 pointer and frame pointer registers. If this macro is not defined, it
7262 is up to the machine-dependent files to allocate such a register (if
7263 necessary). Note that this register must be fixed when in use (e.g.@:
7264 when @code{flag_pic} is true).
7267 @defmac PIC_OFFSET_TABLE_REG_CALL_CLOBBERED
7268 A C expression that is nonzero if the register defined by
7269 @code{PIC_OFFSET_TABLE_REGNUM} is clobbered by calls. If not defined,
7270 the default is zero. Do not define
7271 this macro if @code{PIC_OFFSET_TABLE_REGNUM} is not defined.
7274 @defmac LEGITIMATE_PIC_OPERAND_P (@var{x})
7275 A C expression that is nonzero if @var{x} is a legitimate immediate
7276 operand on the target machine when generating position independent code.
7277 You can assume that @var{x} satisfies @code{CONSTANT_P}, so you need not
7278 check this. You can also assume @var{flag_pic} is true, so you need not
7279 check it either. You need not define this macro if all constants
7280 (including @code{SYMBOL_REF}) can be immediate operands when generating
7281 position independent code.
7284 @node Assembler Format
7285 @section Defining the Output Assembler Language
7287 This section describes macros whose principal purpose is to describe how
7288 to write instructions in assembler language---rather than what the
7292 * File Framework:: Structural information for the assembler file.
7293 * Data Output:: Output of constants (numbers, strings, addresses).
7294 * Uninitialized Data:: Output of uninitialized variables.
7295 * Label Output:: Output and generation of labels.
7296 * Initialization:: General principles of initialization
7297 and termination routines.
7298 * Macros for Initialization::
7299 Specific macros that control the handling of
7300 initialization and termination routines.
7301 * Instruction Output:: Output of actual instructions.
7302 * Dispatch Tables:: Output of jump tables.
7303 * Exception Region Output:: Output of exception region code.
7304 * Alignment Output:: Pseudo ops for alignment and skipping data.
7307 @node File Framework
7308 @subsection The Overall Framework of an Assembler File
7309 @cindex assembler format
7310 @cindex output of assembler code
7312 @c prevent bad page break with this line
7313 This describes the overall framework of an assembly file.
7315 @findex default_file_start
7316 @deftypefn {Target Hook} void TARGET_ASM_FILE_START (void)
7317 Output to @code{asm_out_file} any text which the assembler expects to
7318 find at the beginning of a file. The default behavior is controlled
7319 by two flags, documented below. Unless your target's assembler is
7320 quite unusual, if you override the default, you should call
7321 @code{default_file_start} at some point in your target hook. This
7322 lets other target files rely on these variables.
7325 @deftypevr {Target Hook} bool TARGET_ASM_FILE_START_APP_OFF
7326 If this flag is true, the text of the macro @code{ASM_APP_OFF} will be
7327 printed as the very first line in the assembly file, unless
7328 @option{-fverbose-asm} is in effect. (If that macro has been defined
7329 to the empty string, this variable has no effect.) With the normal
7330 definition of @code{ASM_APP_OFF}, the effect is to notify the GNU
7331 assembler that it need not bother stripping comments or extra
7332 whitespace from its input. This allows it to work a bit faster.
7334 The default is false. You should not set it to true unless you have
7335 verified that your port does not generate any extra whitespace or
7336 comments that will cause GAS to issue errors in NO_APP mode.
7339 @deftypevr {Target Hook} bool TARGET_ASM_FILE_START_FILE_DIRECTIVE
7340 If this flag is true, @code{output_file_directive} will be called
7341 for the primary source file, immediately after printing
7342 @code{ASM_APP_OFF} (if that is enabled). Most ELF assemblers expect
7343 this to be done. The default is false.
7346 @deftypefn {Target Hook} void TARGET_ASM_FILE_END (void)
7347 Output to @code{asm_out_file} any text which the assembler expects
7348 to find at the end of a file. The default is to output nothing.
7351 @deftypefun void file_end_indicate_exec_stack ()
7352 Some systems use a common convention, the @samp{.note.GNU-stack}
7353 special section, to indicate whether or not an object file relies on
7354 the stack being executable. If your system uses this convention, you
7355 should define @code{TARGET_ASM_FILE_END} to this function. If you
7356 need to do other things in that hook, have your hook function call
7360 @deftypefn {Target Hook} void TARGET_ASM_LTO_START (void)
7361 Output to @code{asm_out_file} any text which the assembler expects
7362 to find at the start of an LTO section. The default is to output
7366 @deftypefn {Target Hook} void TARGET_ASM_LTO_END (void)
7367 Output to @code{asm_out_file} any text which the assembler expects
7368 to find at the end of an LTO section. The default is to output
7372 @deftypefn {Target Hook} void TARGET_ASM_CODE_END (void)
7373 Output to @code{asm_out_file} any text which is needed before emitting
7374 unwind info and debug info at the end of a file. Some targets emit
7375 here PIC setup thunks that cannot be emitted at the end of file,
7376 because they couldn't have unwind info then. The default is to output
7380 @defmac ASM_COMMENT_START
7381 A C string constant describing how to begin a comment in the target
7382 assembler language. The compiler assumes that the comment will end at
7383 the end of the line.
7387 A C string constant for text to be output before each @code{asm}
7388 statement or group of consecutive ones. Normally this is
7389 @code{"#APP"}, which is a comment that has no effect on most
7390 assemblers but tells the GNU assembler that it must check the lines
7391 that follow for all valid assembler constructs.
7395 A C string constant for text to be output after each @code{asm}
7396 statement or group of consecutive ones. Normally this is
7397 @code{"#NO_APP"}, which tells the GNU assembler to resume making the
7398 time-saving assumptions that are valid for ordinary compiler output.
7401 @defmac ASM_OUTPUT_SOURCE_FILENAME (@var{stream}, @var{name})
7402 A C statement to output COFF information or DWARF debugging information
7403 which indicates that filename @var{name} is the current source file to
7404 the stdio stream @var{stream}.
7406 This macro need not be defined if the standard form of output
7407 for the file format in use is appropriate.
7410 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_SOURCE_FILENAME (FILE *@var{file}, const char *@var{name})
7411 Output COFF information or DWARF debugging information which indicates that filename @var{name} is the current source file to the stdio stream @var{file}.
7413 This target hook need not be defined if the standard form of output for the file format in use is appropriate.
7416 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_IDENT (const char *@var{name})
7417 Output a string based on @var{name}, suitable for the @samp{#ident} directive, or the equivalent directive or pragma in non-C-family languages. If this hook is not defined, nothing is output for the @samp{#ident} directive.
7420 @defmac OUTPUT_QUOTED_STRING (@var{stream}, @var{string})
7421 A C statement to output the string @var{string} to the stdio stream
7422 @var{stream}. If you do not call the function @code{output_quoted_string}
7423 in your config files, GCC will only call it to output filenames to
7424 the assembler source. So you can use it to canonicalize the format
7425 of the filename using this macro.
7428 @deftypefn {Target Hook} void TARGET_ASM_NAMED_SECTION (const char *@var{name}, unsigned int @var{flags}, tree @var{decl})
7429 Output assembly directives to switch to section @var{name}. The section
7430 should have attributes as specified by @var{flags}, which is a bit mask
7431 of the @code{SECTION_*} flags defined in @file{output.h}. If @var{decl}
7432 is non-NULL, it is the @code{VAR_DECL} or @code{FUNCTION_DECL} with which
7433 this section is associated.
7436 @deftypefn {Target Hook} {section *} TARGET_ASM_FUNCTION_SECTION (tree @var{decl}, enum node_frequency @var{freq}, bool @var{startup}, bool @var{exit})
7437 Return preferred text (sub)section for function @var{decl}.
7438 Main purpose of this function is to separate cold, normal and hot
7439 functions. @var{startup} is true when function is known to be used only
7440 at startup (from static constructors or it is @code{main()}).
7441 @var{exit} is true when function is known to be used only at exit
7442 (from static destructors).
7443 Return NULL if function should go to default text section.
7446 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_SWITCHED_TEXT_SECTIONS (FILE *@var{file}, tree @var{decl}, bool @var{new_is_cold})
7447 Used by the target to emit any assembler directives or additional labels needed when a function is partitioned between different sections. Output should be written to @var{file}. The function decl is available as @var{decl} and the new section is `cold' if @var{new_is_cold} is @code{true}.
7450 @deftypevr {Common Target Hook} bool TARGET_HAVE_NAMED_SECTIONS
7451 This flag is true if the target supports @code{TARGET_ASM_NAMED_SECTION}.
7452 It must not be modified by command-line option processing.
7455 @anchor{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS}
7456 @deftypevr {Target Hook} bool TARGET_HAVE_SWITCHABLE_BSS_SECTIONS
7457 This flag is true if we can create zeroed data by switching to a BSS
7458 section and then using @code{ASM_OUTPUT_SKIP} to allocate the space.
7459 This is true on most ELF targets.
7462 @deftypefn {Target Hook} {unsigned int} TARGET_SECTION_TYPE_FLAGS (tree @var{decl}, const char *@var{name}, int @var{reloc})
7463 Choose a set of section attributes for use by @code{TARGET_ASM_NAMED_SECTION}
7464 based on a variable or function decl, a section name, and whether or not the
7465 declaration's initializer may contain runtime relocations. @var{decl} may be
7466 null, in which case read-write data should be assumed.
7468 The default version of this function handles choosing code vs data,
7469 read-only vs read-write data, and @code{flag_pic}. You should only
7470 need to override this if your target has special flags that might be
7471 set via @code{__attribute__}.
7474 @deftypefn {Target Hook} int TARGET_ASM_RECORD_GCC_SWITCHES (print_switch_type @var{type}, const char *@var{text})
7475 Provides the target with the ability to record the gcc command line
7476 switches that have been passed to the compiler, and options that are
7477 enabled. The @var{type} argument specifies what is being recorded.
7478 It can take the following values:
7481 @item SWITCH_TYPE_PASSED
7482 @var{text} is a command line switch that has been set by the user.
7484 @item SWITCH_TYPE_ENABLED
7485 @var{text} is an option which has been enabled. This might be as a
7486 direct result of a command line switch, or because it is enabled by
7487 default or because it has been enabled as a side effect of a different
7488 command line switch. For example, the @option{-O2} switch enables
7489 various different individual optimization passes.
7491 @item SWITCH_TYPE_DESCRIPTIVE
7492 @var{text} is either NULL or some descriptive text which should be
7493 ignored. If @var{text} is NULL then it is being used to warn the
7494 target hook that either recording is starting or ending. The first
7495 time @var{type} is SWITCH_TYPE_DESCRIPTIVE and @var{text} is NULL, the
7496 warning is for start up and the second time the warning is for
7497 wind down. This feature is to allow the target hook to make any
7498 necessary preparations before it starts to record switches and to
7499 perform any necessary tidying up after it has finished recording
7502 @item SWITCH_TYPE_LINE_START
7503 This option can be ignored by this target hook.
7505 @item SWITCH_TYPE_LINE_END
7506 This option can be ignored by this target hook.
7509 The hook's return value must be zero. Other return values may be
7510 supported in the future.
7512 By default this hook is set to NULL, but an example implementation is
7513 provided for ELF based targets. Called @var{elf_record_gcc_switches},
7514 it records the switches as ASCII text inside a new, string mergeable
7515 section in the assembler output file. The name of the new section is
7516 provided by the @code{TARGET_ASM_RECORD_GCC_SWITCHES_SECTION} target
7520 @deftypevr {Target Hook} {const char *} TARGET_ASM_RECORD_GCC_SWITCHES_SECTION
7521 This is the name of the section that will be created by the example
7522 ELF implementation of the @code{TARGET_ASM_RECORD_GCC_SWITCHES} target
7528 @subsection Output of Data
7531 @deftypevr {Target Hook} {const char *} TARGET_ASM_BYTE_OP
7532 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_HI_OP
7533 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_SI_OP
7534 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_DI_OP
7535 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_TI_OP
7536 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_HI_OP
7537 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_SI_OP
7538 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_DI_OP
7539 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_TI_OP
7540 These hooks specify assembly directives for creating certain kinds
7541 of integer object. The @code{TARGET_ASM_BYTE_OP} directive creates a
7542 byte-sized object, the @code{TARGET_ASM_ALIGNED_HI_OP} one creates an
7543 aligned two-byte object, and so on. Any of the hooks may be
7544 @code{NULL}, indicating that no suitable directive is available.
7546 The compiler will print these strings at the start of a new line,
7547 followed immediately by the object's initial value. In most cases,
7548 the string should contain a tab, a pseudo-op, and then another tab.
7551 @deftypefn {Target Hook} bool TARGET_ASM_INTEGER (rtx @var{x}, unsigned int @var{size}, int @var{aligned_p})
7552 The @code{assemble_integer} function uses this hook to output an
7553 integer object. @var{x} is the object's value, @var{size} is its size
7554 in bytes and @var{aligned_p} indicates whether it is aligned. The
7555 function should return @code{true} if it was able to output the
7556 object. If it returns false, @code{assemble_integer} will try to
7557 split the object into smaller parts.
7559 The default implementation of this hook will use the
7560 @code{TARGET_ASM_BYTE_OP} family of strings, returning @code{false}
7561 when the relevant string is @code{NULL}.
7564 @deftypefn {Target Hook} void TARGET_ASM_DECL_END (void)
7565 Define this hook if the target assembler requires a special marker to
7566 terminate an initialized variable declaration.
7569 @deftypefn {Target Hook} bool TARGET_ASM_OUTPUT_ADDR_CONST_EXTRA (FILE *@var{file}, rtx @var{x})
7570 A target hook to recognize @var{rtx} patterns that @code{output_addr_const}
7571 can't deal with, and output assembly code to @var{file} corresponding to
7572 the pattern @var{x}. This may be used to allow machine-dependent
7573 @code{UNSPEC}s to appear within constants.
7575 If target hook fails to recognize a pattern, it must return @code{false},
7576 so that a standard error message is printed. If it prints an error message
7577 itself, by calling, for example, @code{output_operand_lossage}, it may just
7581 @defmac ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len})
7582 A C statement to output to the stdio stream @var{stream} an assembler
7583 instruction to assemble a string constant containing the @var{len}
7584 bytes at @var{ptr}. @var{ptr} will be a C expression of type
7585 @code{char *} and @var{len} a C expression of type @code{int}.
7587 If the assembler has a @code{.ascii} pseudo-op as found in the
7588 Berkeley Unix assembler, do not define the macro
7589 @code{ASM_OUTPUT_ASCII}.
7592 @defmac ASM_OUTPUT_FDESC (@var{stream}, @var{decl}, @var{n})
7593 A C statement to output word @var{n} of a function descriptor for
7594 @var{decl}. This must be defined if @code{TARGET_VTABLE_USES_DESCRIPTORS}
7595 is defined, and is otherwise unused.
7598 @defmac CONSTANT_POOL_BEFORE_FUNCTION
7599 You may define this macro as a C expression. You should define the
7600 expression to have a nonzero value if GCC should output the constant
7601 pool for a function before the code for the function, or a zero value if
7602 GCC should output the constant pool after the function. If you do
7603 not define this macro, the usual case, GCC will output the constant
7604 pool before the function.
7607 @defmac ASM_OUTPUT_POOL_PROLOGUE (@var{file}, @var{funname}, @var{fundecl}, @var{size})
7608 A C statement to output assembler commands to define the start of the
7609 constant pool for a function. @var{funname} is a string giving
7610 the name of the function. Should the return type of the function
7611 be required, it can be obtained via @var{fundecl}. @var{size}
7612 is the size, in bytes, of the constant pool that will be written
7613 immediately after this call.
7615 If no constant-pool prefix is required, the usual case, this macro need
7619 @defmac ASM_OUTPUT_SPECIAL_POOL_ENTRY (@var{file}, @var{x}, @var{mode}, @var{align}, @var{labelno}, @var{jumpto})
7620 A C statement (with or without semicolon) to output a constant in the
7621 constant pool, if it needs special treatment. (This macro need not do
7622 anything for RTL expressions that can be output normally.)
7624 The argument @var{file} is the standard I/O stream to output the
7625 assembler code on. @var{x} is the RTL expression for the constant to
7626 output, and @var{mode} is the machine mode (in case @var{x} is a
7627 @samp{const_int}). @var{align} is the required alignment for the value
7628 @var{x}; you should output an assembler directive to force this much
7631 The argument @var{labelno} is a number to use in an internal label for
7632 the address of this pool entry. The definition of this macro is
7633 responsible for outputting the label definition at the proper place.
7634 Here is how to do this:
7637 @code{(*targetm.asm_out.internal_label)} (@var{file}, "LC", @var{labelno});
7640 When you output a pool entry specially, you should end with a
7641 @code{goto} to the label @var{jumpto}. This will prevent the same pool
7642 entry from being output a second time in the usual manner.
7644 You need not define this macro if it would do nothing.
7647 @defmac ASM_OUTPUT_POOL_EPILOGUE (@var{file} @var{funname} @var{fundecl} @var{size})
7648 A C statement to output assembler commands to at the end of the constant
7649 pool for a function. @var{funname} is a string giving the name of the
7650 function. Should the return type of the function be required, you can
7651 obtain it via @var{fundecl}. @var{size} is the size, in bytes, of the
7652 constant pool that GCC wrote immediately before this call.
7654 If no constant-pool epilogue is required, the usual case, you need not
7658 @defmac IS_ASM_LOGICAL_LINE_SEPARATOR (@var{C}, @var{STR})
7659 Define this macro as a C expression which is nonzero if @var{C} is
7660 used as a logical line separator by the assembler. @var{STR} points
7661 to the position in the string where @var{C} was found; this can be used if
7662 a line separator uses multiple characters.
7664 If you do not define this macro, the default is that only
7665 the character @samp{;} is treated as a logical line separator.
7668 @deftypevr {Target Hook} {const char *} TARGET_ASM_OPEN_PAREN
7669 @deftypevrx {Target Hook} {const char *} TARGET_ASM_CLOSE_PAREN
7670 These target hooks are C string constants, describing the syntax in the
7671 assembler for grouping arithmetic expressions. If not overridden, they
7672 default to normal parentheses, which is correct for most assemblers.
7675 These macros are provided by @file{real.h} for writing the definitions
7676 of @code{ASM_OUTPUT_DOUBLE} and the like:
7678 @defmac REAL_VALUE_TO_TARGET_SINGLE (@var{x}, @var{l})
7679 @defmacx REAL_VALUE_TO_TARGET_DOUBLE (@var{x}, @var{l})
7680 @defmacx REAL_VALUE_TO_TARGET_LONG_DOUBLE (@var{x}, @var{l})
7681 @defmacx REAL_VALUE_TO_TARGET_DECIMAL32 (@var{x}, @var{l})
7682 @defmacx REAL_VALUE_TO_TARGET_DECIMAL64 (@var{x}, @var{l})
7683 @defmacx REAL_VALUE_TO_TARGET_DECIMAL128 (@var{x}, @var{l})
7684 These translate @var{x}, of type @code{REAL_VALUE_TYPE}, to the
7685 target's floating point representation, and store its bit pattern in
7686 the variable @var{l}. For @code{REAL_VALUE_TO_TARGET_SINGLE} and
7687 @code{REAL_VALUE_TO_TARGET_DECIMAL32}, this variable should be a
7688 simple @code{long int}. For the others, it should be an array of
7689 @code{long int}. The number of elements in this array is determined
7690 by the size of the desired target floating point data type: 32 bits of
7691 it go in each @code{long int} array element. Each array element holds
7692 32 bits of the result, even if @code{long int} is wider than 32 bits
7693 on the host machine.
7695 The array element values are designed so that you can print them out
7696 using @code{fprintf} in the order they should appear in the target
7700 @node Uninitialized Data
7701 @subsection Output of Uninitialized Variables
7703 Each of the macros in this section is used to do the whole job of
7704 outputting a single uninitialized variable.
7706 @defmac ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded})
7707 A C statement (sans semicolon) to output to the stdio stream
7708 @var{stream} the assembler definition of a common-label named
7709 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
7710 is the size rounded up to whatever alignment the caller wants. It is
7711 possible that @var{size} may be zero, for instance if a struct with no
7712 other member than a zero-length array is defined. In this case, the
7713 backend must output a symbol definition that allocates at least one
7714 byte, both so that the address of the resulting object does not compare
7715 equal to any other, and because some object formats cannot even express
7716 the concept of a zero-sized common symbol, as that is how they represent
7717 an ordinary undefined external.
7719 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7720 output the name itself; before and after that, output the additional
7721 assembler syntax for defining the name, and a newline.
7723 This macro controls how the assembler definitions of uninitialized
7724 common global variables are output.
7727 @defmac ASM_OUTPUT_ALIGNED_COMMON (@var{stream}, @var{name}, @var{size}, @var{alignment})
7728 Like @code{ASM_OUTPUT_COMMON} except takes the required alignment as a
7729 separate, explicit argument. If you define this macro, it is used in
7730 place of @code{ASM_OUTPUT_COMMON}, and gives you more flexibility in
7731 handling the required alignment of the variable. The alignment is specified
7732 as the number of bits.
7735 @defmac ASM_OUTPUT_ALIGNED_DECL_COMMON (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
7736 Like @code{ASM_OUTPUT_ALIGNED_COMMON} except that @var{decl} of the
7737 variable to be output, if there is one, or @code{NULL_TREE} if there
7738 is no corresponding variable. If you define this macro, GCC will use it
7739 in place of both @code{ASM_OUTPUT_COMMON} and
7740 @code{ASM_OUTPUT_ALIGNED_COMMON}. Define this macro when you need to see
7741 the variable's decl in order to chose what to output.
7744 @defmac ASM_OUTPUT_ALIGNED_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
7745 A C statement (sans semicolon) to output to the stdio stream
7746 @var{stream} the assembler definition of uninitialized global @var{decl} named
7747 @var{name} whose size is @var{size} bytes. The variable @var{alignment}
7748 is the alignment specified as the number of bits.
7750 Try to use function @code{asm_output_aligned_bss} defined in file
7751 @file{varasm.c} when defining this macro. If unable, use the expression
7752 @code{assemble_name (@var{stream}, @var{name})} to output the name itself;
7753 before and after that, output the additional assembler syntax for defining
7754 the name, and a newline.
7756 There are two ways of handling global BSS@. One is to define this macro.
7757 The other is to have @code{TARGET_ASM_SELECT_SECTION} return a
7758 switchable BSS section (@pxref{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS}).
7759 You do not need to do both.
7761 Some languages do not have @code{common} data, and require a
7762 non-common form of global BSS in order to handle uninitialized globals
7763 efficiently. C++ is one example of this. However, if the target does
7764 not support global BSS, the front end may choose to make globals
7765 common in order to save space in the object file.
7768 @defmac ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded})
7769 A C statement (sans semicolon) to output to the stdio stream
7770 @var{stream} the assembler definition of a local-common-label named
7771 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
7772 is the size rounded up to whatever alignment the caller wants.
7774 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7775 output the name itself; before and after that, output the additional
7776 assembler syntax for defining the name, and a newline.
7778 This macro controls how the assembler definitions of uninitialized
7779 static variables are output.
7782 @defmac ASM_OUTPUT_ALIGNED_LOCAL (@var{stream}, @var{name}, @var{size}, @var{alignment})
7783 Like @code{ASM_OUTPUT_LOCAL} except takes the required alignment as a
7784 separate, explicit argument. If you define this macro, it is used in
7785 place of @code{ASM_OUTPUT_LOCAL}, and gives you more flexibility in
7786 handling the required alignment of the variable. The alignment is specified
7787 as the number of bits.
7790 @defmac ASM_OUTPUT_ALIGNED_DECL_LOCAL (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
7791 Like @code{ASM_OUTPUT_ALIGNED_DECL} except that @var{decl} of the
7792 variable to be output, if there is one, or @code{NULL_TREE} if there
7793 is no corresponding variable. If you define this macro, GCC will use it
7794 in place of both @code{ASM_OUTPUT_DECL} and
7795 @code{ASM_OUTPUT_ALIGNED_DECL}. Define this macro when you need to see
7796 the variable's decl in order to chose what to output.
7800 @subsection Output and Generation of Labels
7802 @c prevent bad page break with this line
7803 This is about outputting labels.
7805 @findex assemble_name
7806 @defmac ASM_OUTPUT_LABEL (@var{stream}, @var{name})
7807 A C statement (sans semicolon) to output to the stdio stream
7808 @var{stream} the assembler definition of a label named @var{name}.
7809 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7810 output the name itself; before and after that, output the additional
7811 assembler syntax for defining the name, and a newline. A default
7812 definition of this macro is provided which is correct for most systems.
7815 @defmac ASM_OUTPUT_FUNCTION_LABEL (@var{stream}, @var{name}, @var{decl})
7816 A C statement (sans semicolon) to output to the stdio stream
7817 @var{stream} the assembler definition of a label named @var{name} of
7819 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7820 output the name itself; before and after that, output the additional
7821 assembler syntax for defining the name, and a newline. A default
7822 definition of this macro is provided which is correct for most systems.
7824 If this macro is not defined, then the function name is defined in the
7825 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
7828 @findex assemble_name_raw
7829 @defmac ASM_OUTPUT_INTERNAL_LABEL (@var{stream}, @var{name})
7830 Identical to @code{ASM_OUTPUT_LABEL}, except that @var{name} is known
7831 to refer to a compiler-generated label. The default definition uses
7832 @code{assemble_name_raw}, which is like @code{assemble_name} except
7833 that it is more efficient.
7837 A C string containing the appropriate assembler directive to specify the
7838 size of a symbol, without any arguments. On systems that use ELF, the
7839 default (in @file{config/elfos.h}) is @samp{"\t.size\t"}; on other
7840 systems, the default is not to define this macro.
7842 Define this macro only if it is correct to use the default definitions
7843 of @code{ASM_OUTPUT_SIZE_DIRECTIVE} and @code{ASM_OUTPUT_MEASURED_SIZE}
7844 for your system. If you need your own custom definitions of those
7845 macros, or if you do not need explicit symbol sizes at all, do not
7849 @defmac ASM_OUTPUT_SIZE_DIRECTIVE (@var{stream}, @var{name}, @var{size})
7850 A C statement (sans semicolon) to output to the stdio stream
7851 @var{stream} a directive telling the assembler that the size of the
7852 symbol @var{name} is @var{size}. @var{size} is a @code{HOST_WIDE_INT}.
7853 If you define @code{SIZE_ASM_OP}, a default definition of this macro is
7857 @defmac ASM_OUTPUT_MEASURED_SIZE (@var{stream}, @var{name})
7858 A C statement (sans semicolon) to output to the stdio stream
7859 @var{stream} a directive telling the assembler to calculate the size of
7860 the symbol @var{name} by subtracting its address from the current
7863 If you define @code{SIZE_ASM_OP}, a default definition of this macro is
7864 provided. The default assumes that the assembler recognizes a special
7865 @samp{.} symbol as referring to the current address, and can calculate
7866 the difference between this and another symbol. If your assembler does
7867 not recognize @samp{.} or cannot do calculations with it, you will need
7868 to redefine @code{ASM_OUTPUT_MEASURED_SIZE} to use some other technique.
7871 @defmac NO_DOLLAR_IN_LABEL
7872 Define this macro if the assembler does not accept the character
7873 @samp{$} in label names. By default constructors and destructors in
7874 G++ have @samp{$} in the identifiers. If this macro is defined,
7875 @samp{.} is used instead.
7878 @defmac NO_DOT_IN_LABEL
7879 Define this macro if the assembler does not accept the character
7880 @samp{.} in label names. By default constructors and destructors in G++
7881 have names that use @samp{.}. If this macro is defined, these names
7882 are rewritten to avoid @samp{.}.
7886 A C string containing the appropriate assembler directive to specify the
7887 type of a symbol, without any arguments. On systems that use ELF, the
7888 default (in @file{config/elfos.h}) is @samp{"\t.type\t"}; on other
7889 systems, the default is not to define this macro.
7891 Define this macro only if it is correct to use the default definition of
7892 @code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own
7893 custom definition of this macro, or if you do not need explicit symbol
7894 types at all, do not define this macro.
7897 @defmac TYPE_OPERAND_FMT
7898 A C string which specifies (using @code{printf} syntax) the format of
7899 the second operand to @code{TYPE_ASM_OP}. On systems that use ELF, the
7900 default (in @file{config/elfos.h}) is @samp{"@@%s"}; on other systems,
7901 the default is not to define this macro.
7903 Define this macro only if it is correct to use the default definition of
7904 @code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own
7905 custom definition of this macro, or if you do not need explicit symbol
7906 types at all, do not define this macro.
7909 @defmac ASM_OUTPUT_TYPE_DIRECTIVE (@var{stream}, @var{type})
7910 A C statement (sans semicolon) to output to the stdio stream
7911 @var{stream} a directive telling the assembler that the type of the
7912 symbol @var{name} is @var{type}. @var{type} is a C string; currently,
7913 that string is always either @samp{"function"} or @samp{"object"}, but
7914 you should not count on this.
7916 If you define @code{TYPE_ASM_OP} and @code{TYPE_OPERAND_FMT}, a default
7917 definition of this macro is provided.
7920 @defmac ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl})
7921 A C statement (sans semicolon) to output to the stdio stream
7922 @var{stream} any text necessary for declaring the name @var{name} of a
7923 function which is being defined. This macro is responsible for
7924 outputting the label definition (perhaps using
7925 @code{ASM_OUTPUT_FUNCTION_LABEL}). The argument @var{decl} is the
7926 @code{FUNCTION_DECL} tree node representing the function.
7928 If this macro is not defined, then the function name is defined in the
7929 usual manner as a label (by means of @code{ASM_OUTPUT_FUNCTION_LABEL}).
7931 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition
7935 @defmac ASM_DECLARE_FUNCTION_SIZE (@var{stream}, @var{name}, @var{decl})
7936 A C statement (sans semicolon) to output to the stdio stream
7937 @var{stream} any text necessary for declaring the size of a function
7938 which is being defined. The argument @var{name} is the name of the
7939 function. The argument @var{decl} is the @code{FUNCTION_DECL} tree node
7940 representing the function.
7942 If this macro is not defined, then the function size is not defined.
7944 You may wish to use @code{ASM_OUTPUT_MEASURED_SIZE} in the definition
7948 @defmac ASM_DECLARE_OBJECT_NAME (@var{stream}, @var{name}, @var{decl})
7949 A C statement (sans semicolon) to output to the stdio stream
7950 @var{stream} any text necessary for declaring the name @var{name} of an
7951 initialized variable which is being defined. This macro must output the
7952 label definition (perhaps using @code{ASM_OUTPUT_LABEL}). The argument
7953 @var{decl} is the @code{VAR_DECL} tree node representing the variable.
7955 If this macro is not defined, then the variable name is defined in the
7956 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
7958 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} and/or
7959 @code{ASM_OUTPUT_SIZE_DIRECTIVE} in the definition of this macro.
7962 @deftypefn {Target Hook} void TARGET_ASM_DECLARE_CONSTANT_NAME (FILE *@var{file}, const char *@var{name}, const_tree @var{expr}, HOST_WIDE_INT @var{size})
7963 A target hook to output to the stdio stream @var{file} any text necessary
7964 for declaring the name @var{name} of a constant which is being defined. This
7965 target hook is responsible for outputting the label definition (perhaps using
7966 @code{assemble_label}). The argument @var{exp} is the value of the constant,
7967 and @var{size} is the size of the constant in bytes. The @var{name}
7968 will be an internal label.
7970 The default version of this target hook, define the @var{name} in the
7971 usual manner as a label (by means of @code{assemble_label}).
7973 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in this target hook.
7976 @defmac ASM_DECLARE_REGISTER_GLOBAL (@var{stream}, @var{decl}, @var{regno}, @var{name})
7977 A C statement (sans semicolon) to output to the stdio stream
7978 @var{stream} any text necessary for claiming a register @var{regno}
7979 for a global variable @var{decl} with name @var{name}.
7981 If you don't define this macro, that is equivalent to defining it to do
7985 @defmac ASM_FINISH_DECLARE_OBJECT (@var{stream}, @var{decl}, @var{toplevel}, @var{atend})
7986 A C statement (sans semicolon) to finish up declaring a variable name
7987 once the compiler has processed its initializer fully and thus has had a
7988 chance to determine the size of an array when controlled by an
7989 initializer. This is used on systems where it's necessary to declare
7990 something about the size of the object.
7992 If you don't define this macro, that is equivalent to defining it to do
7995 You may wish to use @code{ASM_OUTPUT_SIZE_DIRECTIVE} and/or
7996 @code{ASM_OUTPUT_MEASURED_SIZE} in the definition of this macro.
7999 @deftypefn {Target Hook} void TARGET_ASM_GLOBALIZE_LABEL (FILE *@var{stream}, const char *@var{name})
8000 This target hook is a function to output to the stdio stream
8001 @var{stream} some commands that will make the label @var{name} global;
8002 that is, available for reference from other files.
8004 The default implementation relies on a proper definition of
8005 @code{GLOBAL_ASM_OP}.
8008 @deftypefn {Target Hook} void TARGET_ASM_GLOBALIZE_DECL_NAME (FILE *@var{stream}, tree @var{decl})
8009 This target hook is a function to output to the stdio stream
8010 @var{stream} some commands that will make the name associated with @var{decl}
8011 global; that is, available for reference from other files.
8013 The default implementation uses the TARGET_ASM_GLOBALIZE_LABEL target hook.
8016 @deftypefn {Target Hook} void TARGET_ASM_ASSEMBLE_UNDEFINED_DECL (FILE *@var{stream}, const char *@var{name}, const_tree @var{decl})
8017 This target hook is a function to output to the stdio stream
8018 @var{stream} some commands that will declare the name associated with
8019 @var{decl} which is not defined in the current translation unit. Most
8020 assemblers do not require anything to be output in this case.
8023 @defmac ASM_WEAKEN_LABEL (@var{stream}, @var{name})
8024 A C statement (sans semicolon) to output to the stdio stream
8025 @var{stream} some commands that will make the label @var{name} weak;
8026 that is, available for reference from other files but only used if
8027 no other definition is available. Use the expression
8028 @code{assemble_name (@var{stream}, @var{name})} to output the name
8029 itself; before and after that, output the additional assembler syntax
8030 for making that name weak, and a newline.
8032 If you don't define this macro or @code{ASM_WEAKEN_DECL}, GCC will not
8033 support weak symbols and you should not define the @code{SUPPORTS_WEAK}
8037 @defmac ASM_WEAKEN_DECL (@var{stream}, @var{decl}, @var{name}, @var{value})
8038 Combines (and replaces) the function of @code{ASM_WEAKEN_LABEL} and
8039 @code{ASM_OUTPUT_WEAK_ALIAS}, allowing access to the associated function
8040 or variable decl. If @var{value} is not @code{NULL}, this C statement
8041 should output to the stdio stream @var{stream} assembler code which
8042 defines (equates) the weak symbol @var{name} to have the value
8043 @var{value}. If @var{value} is @code{NULL}, it should output commands
8044 to make @var{name} weak.
8047 @defmac ASM_OUTPUT_WEAKREF (@var{stream}, @var{decl}, @var{name}, @var{value})
8048 Outputs a directive that enables @var{name} to be used to refer to
8049 symbol @var{value} with weak-symbol semantics. @code{decl} is the
8050 declaration of @code{name}.
8053 @defmac SUPPORTS_WEAK
8054 A preprocessor constant expression which evaluates to true if the target
8055 supports weak symbols.
8057 If you don't define this macro, @file{defaults.h} provides a default
8058 definition. If either @code{ASM_WEAKEN_LABEL} or @code{ASM_WEAKEN_DECL}
8059 is defined, the default definition is @samp{1}; otherwise, it is @samp{0}.
8062 @defmac TARGET_SUPPORTS_WEAK
8063 A C expression which evaluates to true if the target supports weak symbols.
8065 If you don't define this macro, @file{defaults.h} provides a default
8066 definition. The default definition is @samp{(SUPPORTS_WEAK)}. Define
8067 this macro if you want to control weak symbol support with a compiler
8068 flag such as @option{-melf}.
8071 @defmac MAKE_DECL_ONE_ONLY (@var{decl})
8072 A C statement (sans semicolon) to mark @var{decl} to be emitted as a
8073 public symbol such that extra copies in multiple translation units will
8074 be discarded by the linker. Define this macro if your object file
8075 format provides support for this concept, such as the @samp{COMDAT}
8076 section flags in the Microsoft Windows PE/COFF format, and this support
8077 requires changes to @var{decl}, such as putting it in a separate section.
8080 @defmac SUPPORTS_ONE_ONLY
8081 A C expression which evaluates to true if the target supports one-only
8084 If you don't define this macro, @file{varasm.c} provides a default
8085 definition. If @code{MAKE_DECL_ONE_ONLY} is defined, the default
8086 definition is @samp{1}; otherwise, it is @samp{0}. Define this macro if
8087 you want to control one-only symbol support with a compiler flag, or if
8088 setting the @code{DECL_ONE_ONLY} flag is enough to mark a declaration to
8089 be emitted as one-only.
8092 @deftypefn {Target Hook} void TARGET_ASM_ASSEMBLE_VISIBILITY (tree @var{decl}, int @var{visibility})
8093 This target hook is a function to output to @var{asm_out_file} some
8094 commands that will make the symbol(s) associated with @var{decl} have
8095 hidden, protected or internal visibility as specified by @var{visibility}.
8098 @defmac TARGET_WEAK_NOT_IN_ARCHIVE_TOC
8099 A C expression that evaluates to true if the target's linker expects
8100 that weak symbols do not appear in a static archive's table of contents.
8101 The default is @code{0}.
8103 Leaving weak symbols out of an archive's table of contents means that,
8104 if a symbol will only have a definition in one translation unit and
8105 will have undefined references from other translation units, that
8106 symbol should not be weak. Defining this macro to be nonzero will
8107 thus have the effect that certain symbols that would normally be weak
8108 (explicit template instantiations, and vtables for polymorphic classes
8109 with noninline key methods) will instead be nonweak.
8111 The C++ ABI requires this macro to be zero. Define this macro for
8112 targets where full C++ ABI compliance is impossible and where linker
8113 restrictions require weak symbols to be left out of a static archive's
8117 @defmac ASM_OUTPUT_EXTERNAL (@var{stream}, @var{decl}, @var{name})
8118 A C statement (sans semicolon) to output to the stdio stream
8119 @var{stream} any text necessary for declaring the name of an external
8120 symbol named @var{name} which is referenced in this compilation but
8121 not defined. The value of @var{decl} is the tree node for the
8124 This macro need not be defined if it does not need to output anything.
8125 The GNU assembler and most Unix assemblers don't require anything.
8128 @deftypefn {Target Hook} void TARGET_ASM_EXTERNAL_LIBCALL (rtx @var{symref})
8129 This target hook is a function to output to @var{asm_out_file} an assembler
8130 pseudo-op to declare a library function name external. The name of the
8131 library function is given by @var{symref}, which is a @code{symbol_ref}.
8134 @deftypefn {Target Hook} void TARGET_ASM_MARK_DECL_PRESERVED (const char *@var{symbol})
8135 This target hook is a function to output to @var{asm_out_file} an assembler
8136 directive to annotate @var{symbol} as used. The Darwin target uses the
8137 .no_dead_code_strip directive.
8140 @defmac ASM_OUTPUT_LABELREF (@var{stream}, @var{name})
8141 A C statement (sans semicolon) to output to the stdio stream
8142 @var{stream} a reference in assembler syntax to a label named
8143 @var{name}. This should add @samp{_} to the front of the name, if that
8144 is customary on your operating system, as it is in most Berkeley Unix
8145 systems. This macro is used in @code{assemble_name}.
8148 @deftypefn {Target Hook} tree TARGET_MANGLE_ASSEMBLER_NAME (const char *@var{name})
8149 Given a symbol @var{name}, perform same mangling as @code{varasm.c}'s @code{assemble_name}, but in memory rather than to a file stream, returning result as an @code{IDENTIFIER_NODE}. Required for correct LTO symtabs. The default implementation calls the @code{TARGET_STRIP_NAME_ENCODING} hook and then prepends the @code{USER_LABEL_PREFIX}, if any.
8152 @defmac ASM_OUTPUT_SYMBOL_REF (@var{stream}, @var{sym})
8153 A C statement (sans semicolon) to output a reference to
8154 @code{SYMBOL_REF} @var{sym}. If not defined, @code{assemble_name}
8155 will be used to output the name of the symbol. This macro may be used
8156 to modify the way a symbol is referenced depending on information
8157 encoded by @code{TARGET_ENCODE_SECTION_INFO}.
8160 @defmac ASM_OUTPUT_LABEL_REF (@var{stream}, @var{buf})
8161 A C statement (sans semicolon) to output a reference to @var{buf}, the
8162 result of @code{ASM_GENERATE_INTERNAL_LABEL}. If not defined,
8163 @code{assemble_name} will be used to output the name of the symbol.
8164 This macro is not used by @code{output_asm_label}, or the @code{%l}
8165 specifier that calls it; the intention is that this macro should be set
8166 when it is necessary to output a label differently when its address is
8170 @deftypefn {Target Hook} void TARGET_ASM_INTERNAL_LABEL (FILE *@var{stream}, const char *@var{prefix}, unsigned long @var{labelno})
8171 A function to output to the stdio stream @var{stream} a label whose
8172 name is made from the string @var{prefix} and the number @var{labelno}.
8174 It is absolutely essential that these labels be distinct from the labels
8175 used for user-level functions and variables. Otherwise, certain programs
8176 will have name conflicts with internal labels.
8178 It is desirable to exclude internal labels from the symbol table of the
8179 object file. Most assemblers have a naming convention for labels that
8180 should be excluded; on many systems, the letter @samp{L} at the
8181 beginning of a label has this effect. You should find out what
8182 convention your system uses, and follow it.
8184 The default version of this function utilizes @code{ASM_GENERATE_INTERNAL_LABEL}.
8187 @defmac ASM_OUTPUT_DEBUG_LABEL (@var{stream}, @var{prefix}, @var{num})
8188 A C statement to output to the stdio stream @var{stream} a debug info
8189 label whose name is made from the string @var{prefix} and the number
8190 @var{num}. This is useful for VLIW targets, where debug info labels
8191 may need to be treated differently than branch target labels. On some
8192 systems, branch target labels must be at the beginning of instruction
8193 bundles, but debug info labels can occur in the middle of instruction
8196 If this macro is not defined, then @code{(*targetm.asm_out.internal_label)} will be
8200 @defmac ASM_GENERATE_INTERNAL_LABEL (@var{string}, @var{prefix}, @var{num})
8201 A C statement to store into the string @var{string} a label whose name
8202 is made from the string @var{prefix} and the number @var{num}.
8204 This string, when output subsequently by @code{assemble_name}, should
8205 produce the output that @code{(*targetm.asm_out.internal_label)} would produce
8206 with the same @var{prefix} and @var{num}.
8208 If the string begins with @samp{*}, then @code{assemble_name} will
8209 output the rest of the string unchanged. It is often convenient for
8210 @code{ASM_GENERATE_INTERNAL_LABEL} to use @samp{*} in this way. If the
8211 string doesn't start with @samp{*}, then @code{ASM_OUTPUT_LABELREF} gets
8212 to output the string, and may change it. (Of course,
8213 @code{ASM_OUTPUT_LABELREF} is also part of your machine description, so
8214 you should know what it does on your machine.)
8217 @defmac ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number})
8218 A C expression to assign to @var{outvar} (which is a variable of type
8219 @code{char *}) a newly allocated string made from the string
8220 @var{name} and the number @var{number}, with some suitable punctuation
8221 added. Use @code{alloca} to get space for the string.
8223 The string will be used as an argument to @code{ASM_OUTPUT_LABELREF} to
8224 produce an assembler label for an internal static variable whose name is
8225 @var{name}. Therefore, the string must be such as to result in valid
8226 assembler code. The argument @var{number} is different each time this
8227 macro is executed; it prevents conflicts between similarly-named
8228 internal static variables in different scopes.
8230 Ideally this string should not be a valid C identifier, to prevent any
8231 conflict with the user's own symbols. Most assemblers allow periods
8232 or percent signs in assembler symbols; putting at least one of these
8233 between the name and the number will suffice.
8235 If this macro is not defined, a default definition will be provided
8236 which is correct for most systems.
8239 @defmac ASM_OUTPUT_DEF (@var{stream}, @var{name}, @var{value})
8240 A C statement to output to the stdio stream @var{stream} assembler code
8241 which defines (equates) the symbol @var{name} to have the value @var{value}.
8244 If @code{SET_ASM_OP} is defined, a default definition is provided which is
8245 correct for most systems.
8248 @defmac ASM_OUTPUT_DEF_FROM_DECLS (@var{stream}, @var{decl_of_name}, @var{decl_of_value})
8249 A C statement to output to the stdio stream @var{stream} assembler code
8250 which defines (equates) the symbol whose tree node is @var{decl_of_name}
8251 to have the value of the tree node @var{decl_of_value}. This macro will
8252 be used in preference to @samp{ASM_OUTPUT_DEF} if it is defined and if
8253 the tree nodes are available.
8256 If @code{SET_ASM_OP} is defined, a default definition is provided which is
8257 correct for most systems.
8260 @defmac TARGET_DEFERRED_OUTPUT_DEFS (@var{decl_of_name}, @var{decl_of_value})
8261 A C statement that evaluates to true if the assembler code which defines
8262 (equates) the symbol whose tree node is @var{decl_of_name} to have the value
8263 of the tree node @var{decl_of_value} should be emitted near the end of the
8264 current compilation unit. The default is to not defer output of defines.
8265 This macro affects defines output by @samp{ASM_OUTPUT_DEF} and
8266 @samp{ASM_OUTPUT_DEF_FROM_DECLS}.
8269 @defmac ASM_OUTPUT_WEAK_ALIAS (@var{stream}, @var{name}, @var{value})
8270 A C statement to output to the stdio stream @var{stream} assembler code
8271 which defines (equates) the weak symbol @var{name} to have the value
8272 @var{value}. If @var{value} is @code{NULL}, it defines @var{name} as
8273 an undefined weak symbol.
8275 Define this macro if the target only supports weak aliases; define
8276 @code{ASM_OUTPUT_DEF} instead if possible.
8279 @defmac OBJC_GEN_METHOD_LABEL (@var{buf}, @var{is_inst}, @var{class_name}, @var{cat_name}, @var{sel_name})
8280 Define this macro to override the default assembler names used for
8281 Objective-C methods.
8283 The default name is a unique method number followed by the name of the
8284 class (e.g.@: @samp{_1_Foo}). For methods in categories, the name of
8285 the category is also included in the assembler name (e.g.@:
8288 These names are safe on most systems, but make debugging difficult since
8289 the method's selector is not present in the name. Therefore, particular
8290 systems define other ways of computing names.
8292 @var{buf} is an expression of type @code{char *} which gives you a
8293 buffer in which to store the name; its length is as long as
8294 @var{class_name}, @var{cat_name} and @var{sel_name} put together, plus
8295 50 characters extra.
8297 The argument @var{is_inst} specifies whether the method is an instance
8298 method or a class method; @var{class_name} is the name of the class;
8299 @var{cat_name} is the name of the category (or @code{NULL} if the method is not
8300 in a category); and @var{sel_name} is the name of the selector.
8302 On systems where the assembler can handle quoted names, you can use this
8303 macro to provide more human-readable names.
8306 @node Initialization
8307 @subsection How Initialization Functions Are Handled
8308 @cindex initialization routines
8309 @cindex termination routines
8310 @cindex constructors, output of
8311 @cindex destructors, output of
8313 The compiled code for certain languages includes @dfn{constructors}
8314 (also called @dfn{initialization routines})---functions to initialize
8315 data in the program when the program is started. These functions need
8316 to be called before the program is ``started''---that is to say, before
8317 @code{main} is called.
8319 Compiling some languages generates @dfn{destructors} (also called
8320 @dfn{termination routines}) that should be called when the program
8323 To make the initialization and termination functions work, the compiler
8324 must output something in the assembler code to cause those functions to
8325 be called at the appropriate time. When you port the compiler to a new
8326 system, you need to specify how to do this.
8328 There are two major ways that GCC currently supports the execution of
8329 initialization and termination functions. Each way has two variants.
8330 Much of the structure is common to all four variations.
8332 @findex __CTOR_LIST__
8333 @findex __DTOR_LIST__
8334 The linker must build two lists of these functions---a list of
8335 initialization functions, called @code{__CTOR_LIST__}, and a list of
8336 termination functions, called @code{__DTOR_LIST__}.
8338 Each list always begins with an ignored function pointer (which may hold
8339 0, @minus{}1, or a count of the function pointers after it, depending on
8340 the environment). This is followed by a series of zero or more function
8341 pointers to constructors (or destructors), followed by a function
8342 pointer containing zero.
8344 Depending on the operating system and its executable file format, either
8345 @file{crtstuff.c} or @file{libgcc2.c} traverses these lists at startup
8346 time and exit time. Constructors are called in reverse order of the
8347 list; destructors in forward order.
8349 The best way to handle static constructors works only for object file
8350 formats which provide arbitrarily-named sections. A section is set
8351 aside for a list of constructors, and another for a list of destructors.
8352 Traditionally these are called @samp{.ctors} and @samp{.dtors}. Each
8353 object file that defines an initialization function also puts a word in
8354 the constructor section to point to that function. The linker
8355 accumulates all these words into one contiguous @samp{.ctors} section.
8356 Termination functions are handled similarly.
8358 This method will be chosen as the default by @file{target-def.h} if
8359 @code{TARGET_ASM_NAMED_SECTION} is defined. A target that does not
8360 support arbitrary sections, but does support special designated
8361 constructor and destructor sections may define @code{CTORS_SECTION_ASM_OP}
8362 and @code{DTORS_SECTION_ASM_OP} to achieve the same effect.
8364 When arbitrary sections are available, there are two variants, depending
8365 upon how the code in @file{crtstuff.c} is called. On systems that
8366 support a @dfn{.init} section which is executed at program startup,
8367 parts of @file{crtstuff.c} are compiled into that section. The
8368 program is linked by the @command{gcc} driver like this:
8371 ld -o @var{output_file} crti.o crtbegin.o @dots{} -lgcc crtend.o crtn.o
8374 The prologue of a function (@code{__init}) appears in the @code{.init}
8375 section of @file{crti.o}; the epilogue appears in @file{crtn.o}. Likewise
8376 for the function @code{__fini} in the @dfn{.fini} section. Normally these
8377 files are provided by the operating system or by the GNU C library, but
8378 are provided by GCC for a few targets.
8380 The objects @file{crtbegin.o} and @file{crtend.o} are (for most targets)
8381 compiled from @file{crtstuff.c}. They contain, among other things, code
8382 fragments within the @code{.init} and @code{.fini} sections that branch
8383 to routines in the @code{.text} section. The linker will pull all parts
8384 of a section together, which results in a complete @code{__init} function
8385 that invokes the routines we need at startup.
8387 To use this variant, you must define the @code{INIT_SECTION_ASM_OP}
8390 If no init section is available, when GCC compiles any function called
8391 @code{main} (or more accurately, any function designated as a program
8392 entry point by the language front end calling @code{expand_main_function}),
8393 it inserts a procedure call to @code{__main} as the first executable code
8394 after the function prologue. The @code{__main} function is defined
8395 in @file{libgcc2.c} and runs the global constructors.
8397 In file formats that don't support arbitrary sections, there are again
8398 two variants. In the simplest variant, the GNU linker (GNU @code{ld})
8399 and an `a.out' format must be used. In this case,
8400 @code{TARGET_ASM_CONSTRUCTOR} is defined to produce a @code{.stabs}
8401 entry of type @samp{N_SETT}, referencing the name @code{__CTOR_LIST__},
8402 and with the address of the void function containing the initialization
8403 code as its value. The GNU linker recognizes this as a request to add
8404 the value to a @dfn{set}; the values are accumulated, and are eventually
8405 placed in the executable as a vector in the format described above, with
8406 a leading (ignored) count and a trailing zero element.
8407 @code{TARGET_ASM_DESTRUCTOR} is handled similarly. Since no init
8408 section is available, the absence of @code{INIT_SECTION_ASM_OP} causes
8409 the compilation of @code{main} to call @code{__main} as above, starting
8410 the initialization process.
8412 The last variant uses neither arbitrary sections nor the GNU linker.
8413 This is preferable when you want to do dynamic linking and when using
8414 file formats which the GNU linker does not support, such as `ECOFF'@. In
8415 this case, @code{TARGET_HAVE_CTORS_DTORS} is false, initialization and
8416 termination functions are recognized simply by their names. This requires
8417 an extra program in the linkage step, called @command{collect2}. This program
8418 pretends to be the linker, for use with GCC; it does its job by running
8419 the ordinary linker, but also arranges to include the vectors of
8420 initialization and termination functions. These functions are called
8421 via @code{__main} as described above. In order to use this method,
8422 @code{use_collect2} must be defined in the target in @file{config.gcc}.
8425 The following section describes the specific macros that control and
8426 customize the handling of initialization and termination functions.
8429 @node Macros for Initialization
8430 @subsection Macros Controlling Initialization Routines
8432 Here are the macros that control how the compiler handles initialization
8433 and termination functions:
8435 @defmac INIT_SECTION_ASM_OP
8436 If defined, a C string constant, including spacing, for the assembler
8437 operation to identify the following data as initialization code. If not
8438 defined, GCC will assume such a section does not exist. When you are
8439 using special sections for initialization and termination functions, this
8440 macro also controls how @file{crtstuff.c} and @file{libgcc2.c} arrange to
8441 run the initialization functions.
8444 @defmac HAS_INIT_SECTION
8445 If defined, @code{main} will not call @code{__main} as described above.
8446 This macro should be defined for systems that control start-up code
8447 on a symbol-by-symbol basis, such as OSF/1, and should not
8448 be defined explicitly for systems that support @code{INIT_SECTION_ASM_OP}.
8451 @defmac LD_INIT_SWITCH
8452 If defined, a C string constant for a switch that tells the linker that
8453 the following symbol is an initialization routine.
8456 @defmac LD_FINI_SWITCH
8457 If defined, a C string constant for a switch that tells the linker that
8458 the following symbol is a finalization routine.
8461 @defmac COLLECT_SHARED_INIT_FUNC (@var{stream}, @var{func})
8462 If defined, a C statement that will write a function that can be
8463 automatically called when a shared library is loaded. The function
8464 should call @var{func}, which takes no arguments. If not defined, and
8465 the object format requires an explicit initialization function, then a
8466 function called @code{_GLOBAL__DI} will be generated.
8468 This function and the following one are used by collect2 when linking a
8469 shared library that needs constructors or destructors, or has DWARF2
8470 exception tables embedded in the code.
8473 @defmac COLLECT_SHARED_FINI_FUNC (@var{stream}, @var{func})
8474 If defined, a C statement that will write a function that can be
8475 automatically called when a shared library is unloaded. The function
8476 should call @var{func}, which takes no arguments. If not defined, and
8477 the object format requires an explicit finalization function, then a
8478 function called @code{_GLOBAL__DD} will be generated.
8481 @defmac INVOKE__main
8482 If defined, @code{main} will call @code{__main} despite the presence of
8483 @code{INIT_SECTION_ASM_OP}. This macro should be defined for systems
8484 where the init section is not actually run automatically, but is still
8485 useful for collecting the lists of constructors and destructors.
8488 @defmac SUPPORTS_INIT_PRIORITY
8489 If nonzero, the C++ @code{init_priority} attribute is supported and the
8490 compiler should emit instructions to control the order of initialization
8491 of objects. If zero, the compiler will issue an error message upon
8492 encountering an @code{init_priority} attribute.
8495 @deftypevr {Target Hook} bool TARGET_HAVE_CTORS_DTORS
8496 This value is true if the target supports some ``native'' method of
8497 collecting constructors and destructors to be run at startup and exit.
8498 It is false if we must use @command{collect2}.
8501 @deftypefn {Target Hook} void TARGET_ASM_CONSTRUCTOR (rtx @var{symbol}, int @var{priority})
8502 If defined, a function that outputs assembler code to arrange to call
8503 the function referenced by @var{symbol} at initialization time.
8505 Assume that @var{symbol} is a @code{SYMBOL_REF} for a function taking
8506 no arguments and with no return value. If the target supports initialization
8507 priorities, @var{priority} is a value between 0 and @code{MAX_INIT_PRIORITY};
8508 otherwise it must be @code{DEFAULT_INIT_PRIORITY}.
8510 If this macro is not defined by the target, a suitable default will
8511 be chosen if (1) the target supports arbitrary section names, (2) the
8512 target defines @code{CTORS_SECTION_ASM_OP}, or (3) @code{USE_COLLECT2}
8516 @deftypefn {Target Hook} void TARGET_ASM_DESTRUCTOR (rtx @var{symbol}, int @var{priority})
8517 This is like @code{TARGET_ASM_CONSTRUCTOR} but used for termination
8518 functions rather than initialization functions.
8521 If @code{TARGET_HAVE_CTORS_DTORS} is true, the initialization routine
8522 generated for the generated object file will have static linkage.
8524 If your system uses @command{collect2} as the means of processing
8525 constructors, then that program normally uses @command{nm} to scan
8526 an object file for constructor functions to be called.
8528 On certain kinds of systems, you can define this macro to make
8529 @command{collect2} work faster (and, in some cases, make it work at all):
8531 @defmac OBJECT_FORMAT_COFF
8532 Define this macro if the system uses COFF (Common Object File Format)
8533 object files, so that @command{collect2} can assume this format and scan
8534 object files directly for dynamic constructor/destructor functions.
8536 This macro is effective only in a native compiler; @command{collect2} as
8537 part of a cross compiler always uses @command{nm} for the target machine.
8540 @defmac REAL_NM_FILE_NAME
8541 Define this macro as a C string constant containing the file name to use
8542 to execute @command{nm}. The default is to search the path normally for
8547 @command{collect2} calls @command{nm} to scan object files for static
8548 constructors and destructors and LTO info. By default, @option{-n} is
8549 passed. Define @code{NM_FLAGS} to a C string constant if other options
8550 are needed to get the same output format as GNU @command{nm -n}
8554 If your system supports shared libraries and has a program to list the
8555 dynamic dependencies of a given library or executable, you can define
8556 these macros to enable support for running initialization and
8557 termination functions in shared libraries:
8560 Define this macro to a C string constant containing the name of the program
8561 which lists dynamic dependencies, like @command{ldd} under SunOS 4.
8564 @defmac PARSE_LDD_OUTPUT (@var{ptr})
8565 Define this macro to be C code that extracts filenames from the output
8566 of the program denoted by @code{LDD_SUFFIX}. @var{ptr} is a variable
8567 of type @code{char *} that points to the beginning of a line of output
8568 from @code{LDD_SUFFIX}. If the line lists a dynamic dependency, the
8569 code must advance @var{ptr} to the beginning of the filename on that
8570 line. Otherwise, it must set @var{ptr} to @code{NULL}.
8573 @defmac SHLIB_SUFFIX
8574 Define this macro to a C string constant containing the default shared
8575 library extension of the target (e.g., @samp{".so"}). @command{collect2}
8576 strips version information after this suffix when generating global
8577 constructor and destructor names. This define is only needed on targets
8578 that use @command{collect2} to process constructors and destructors.
8581 @node Instruction Output
8582 @subsection Output of Assembler Instructions
8584 @c prevent bad page break with this line
8585 This describes assembler instruction output.
8587 @defmac REGISTER_NAMES
8588 A C initializer containing the assembler's names for the machine
8589 registers, each one as a C string constant. This is what translates
8590 register numbers in the compiler into assembler language.
8593 @defmac ADDITIONAL_REGISTER_NAMES
8594 If defined, a C initializer for an array of structures containing a name
8595 and a register number. This macro defines additional names for hard
8596 registers, thus allowing the @code{asm} option in declarations to refer
8597 to registers using alternate names.
8600 @defmac OVERLAPPING_REGISTER_NAMES
8601 If defined, a C initializer for an array of structures containing a
8602 name, a register number and a count of the number of consecutive
8603 machine registers the name overlaps. This macro defines additional
8604 names for hard registers, thus allowing the @code{asm} option in
8605 declarations to refer to registers using alternate names. Unlike
8606 @code{ADDITIONAL_REGISTER_NAMES}, this macro should be used when the
8607 register name implies multiple underlying registers.
8609 This macro should be used when it is important that a clobber in an
8610 @code{asm} statement clobbers all the underlying values implied by the
8611 register name. For example, on ARM, clobbering the double-precision
8612 VFP register ``d0'' implies clobbering both single-precision registers
8616 @defmac ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr})
8617 Define this macro if you are using an unusual assembler that
8618 requires different names for the machine instructions.
8620 The definition is a C statement or statements which output an
8621 assembler instruction opcode to the stdio stream @var{stream}. The
8622 macro-operand @var{ptr} is a variable of type @code{char *} which
8623 points to the opcode name in its ``internal'' form---the form that is
8624 written in the machine description. The definition should output the
8625 opcode name to @var{stream}, performing any translation you desire, and
8626 increment the variable @var{ptr} to point at the end of the opcode
8627 so that it will not be output twice.
8629 In fact, your macro definition may process less than the entire opcode
8630 name, or more than the opcode name; but if you want to process text
8631 that includes @samp{%}-sequences to substitute operands, you must take
8632 care of the substitution yourself. Just be sure to increment
8633 @var{ptr} over whatever text should not be output normally.
8635 @findex recog_data.operand
8636 If you need to look at the operand values, they can be found as the
8637 elements of @code{recog_data.operand}.
8639 If the macro definition does nothing, the instruction is output
8643 @defmac FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands})
8644 If defined, a C statement to be executed just prior to the output of
8645 assembler code for @var{insn}, to modify the extracted operands so
8646 they will be output differently.
8648 Here the argument @var{opvec} is the vector containing the operands
8649 extracted from @var{insn}, and @var{noperands} is the number of
8650 elements of the vector which contain meaningful data for this insn.
8651 The contents of this vector are what will be used to convert the insn
8652 template into assembler code, so you can change the assembler output
8653 by changing the contents of the vector.
8655 This macro is useful when various assembler syntaxes share a single
8656 file of instruction patterns; by defining this macro differently, you
8657 can cause a large class of instructions to be output differently (such
8658 as with rearranged operands). Naturally, variations in assembler
8659 syntax affecting individual insn patterns ought to be handled by
8660 writing conditional output routines in those patterns.
8662 If this macro is not defined, it is equivalent to a null statement.
8665 @deftypefn {Target Hook} void TARGET_ASM_FINAL_POSTSCAN_INSN (FILE *@var{file}, rtx_insn *@var{insn}, rtx *@var{opvec}, int @var{noperands})
8666 If defined, this target hook is a function which is executed just after the
8667 output of assembler code for @var{insn}, to change the mode of the assembler
8670 Here the argument @var{opvec} is the vector containing the operands
8671 extracted from @var{insn}, and @var{noperands} is the number of
8672 elements of the vector which contain meaningful data for this insn.
8673 The contents of this vector are what was used to convert the insn
8674 template into assembler code, so you can change the assembler mode
8675 by checking the contents of the vector.
8678 @defmac PRINT_OPERAND (@var{stream}, @var{x}, @var{code})
8679 A C compound statement to output to stdio stream @var{stream} the
8680 assembler syntax for an instruction operand @var{x}. @var{x} is an
8683 @var{code} is a value that can be used to specify one of several ways
8684 of printing the operand. It is used when identical operands must be
8685 printed differently depending on the context. @var{code} comes from
8686 the @samp{%} specification that was used to request printing of the
8687 operand. If the specification was just @samp{%@var{digit}} then
8688 @var{code} is 0; if the specification was @samp{%@var{ltr}
8689 @var{digit}} then @var{code} is the ASCII code for @var{ltr}.
8692 If @var{x} is a register, this macro should print the register's name.
8693 The names can be found in an array @code{reg_names} whose type is
8694 @code{char *[]}. @code{reg_names} is initialized from
8695 @code{REGISTER_NAMES}.
8697 When the machine description has a specification @samp{%@var{punct}}
8698 (a @samp{%} followed by a punctuation character), this macro is called
8699 with a null pointer for @var{x} and the punctuation character for
8703 @defmac PRINT_OPERAND_PUNCT_VALID_P (@var{code})
8704 A C expression which evaluates to true if @var{code} is a valid
8705 punctuation character for use in the @code{PRINT_OPERAND} macro. If
8706 @code{PRINT_OPERAND_PUNCT_VALID_P} is not defined, it means that no
8707 punctuation characters (except for the standard one, @samp{%}) are used
8711 @defmac PRINT_OPERAND_ADDRESS (@var{stream}, @var{x})
8712 A C compound statement to output to stdio stream @var{stream} the
8713 assembler syntax for an instruction operand that is a memory reference
8714 whose address is @var{x}. @var{x} is an RTL expression.
8716 @cindex @code{TARGET_ENCODE_SECTION_INFO} usage
8717 On some machines, the syntax for a symbolic address depends on the
8718 section that the address refers to. On these machines, define the hook
8719 @code{TARGET_ENCODE_SECTION_INFO} to store the information into the
8720 @code{symbol_ref}, and then check for it here. @xref{Assembler
8724 @findex dbr_sequence_length
8725 @defmac DBR_OUTPUT_SEQEND (@var{file})
8726 A C statement, to be executed after all slot-filler instructions have
8727 been output. If necessary, call @code{dbr_sequence_length} to
8728 determine the number of slots filled in a sequence (zero if not
8729 currently outputting a sequence), to decide how many no-ops to output,
8732 Don't define this macro if it has nothing to do, but it is helpful in
8733 reading assembly output if the extent of the delay sequence is made
8734 explicit (e.g.@: with white space).
8737 @findex final_sequence
8738 Note that output routines for instructions with delay slots must be
8739 prepared to deal with not being output as part of a sequence
8740 (i.e.@: when the scheduling pass is not run, or when no slot fillers could be
8741 found.) The variable @code{final_sequence} is null when not
8742 processing a sequence, otherwise it contains the @code{sequence} rtx
8746 @defmac REGISTER_PREFIX
8747 @defmacx LOCAL_LABEL_PREFIX
8748 @defmacx USER_LABEL_PREFIX
8749 @defmacx IMMEDIATE_PREFIX
8750 If defined, C string expressions to be used for the @samp{%R}, @samp{%L},
8751 @samp{%U}, and @samp{%I} options of @code{asm_fprintf} (see
8752 @file{final.c}). These are useful when a single @file{md} file must
8753 support multiple assembler formats. In that case, the various @file{tm.h}
8754 files can define these macros differently.
8757 @defmac ASM_FPRINTF_EXTENSIONS (@var{file}, @var{argptr}, @var{format})
8758 If defined this macro should expand to a series of @code{case}
8759 statements which will be parsed inside the @code{switch} statement of
8760 the @code{asm_fprintf} function. This allows targets to define extra
8761 printf formats which may useful when generating their assembler
8762 statements. Note that uppercase letters are reserved for future
8763 generic extensions to asm_fprintf, and so are not available to target
8764 specific code. The output file is given by the parameter @var{file}.
8765 The varargs input pointer is @var{argptr} and the rest of the format
8766 string, starting the character after the one that is being switched
8767 upon, is pointed to by @var{format}.
8770 @defmac ASSEMBLER_DIALECT
8771 If your target supports multiple dialects of assembler language (such as
8772 different opcodes), define this macro as a C expression that gives the
8773 numeric index of the assembler language dialect to use, with zero as the
8776 If this macro is defined, you may use constructs of the form
8778 @samp{@{option0|option1|option2@dots{}@}}
8781 in the output templates of patterns (@pxref{Output Template}) or in the
8782 first argument of @code{asm_fprintf}. This construct outputs
8783 @samp{option0}, @samp{option1}, @samp{option2}, etc., if the value of
8784 @code{ASSEMBLER_DIALECT} is zero, one, two, etc. Any special characters
8785 within these strings retain their usual meaning. If there are fewer
8786 alternatives within the braces than the value of
8787 @code{ASSEMBLER_DIALECT}, the construct outputs nothing. If it's needed
8788 to print curly braces or @samp{|} character in assembler output directly,
8789 @samp{%@{}, @samp{%@}} and @samp{%|} can be used.
8791 If you do not define this macro, the characters @samp{@{}, @samp{|} and
8792 @samp{@}} do not have any special meaning when used in templates or
8793 operands to @code{asm_fprintf}.
8795 Define the macros @code{REGISTER_PREFIX}, @code{LOCAL_LABEL_PREFIX},
8796 @code{USER_LABEL_PREFIX} and @code{IMMEDIATE_PREFIX} if you can express
8797 the variations in assembler language syntax with that mechanism. Define
8798 @code{ASSEMBLER_DIALECT} and use the @samp{@{option0|option1@}} syntax
8799 if the syntax variant are larger and involve such things as different
8800 opcodes or operand order.
8803 @defmac ASM_OUTPUT_REG_PUSH (@var{stream}, @var{regno})
8804 A C expression to output to @var{stream} some assembler code
8805 which will push hard register number @var{regno} onto the stack.
8806 The code need not be optimal, since this macro is used only when
8810 @defmac ASM_OUTPUT_REG_POP (@var{stream}, @var{regno})
8811 A C expression to output to @var{stream} some assembler code
8812 which will pop hard register number @var{regno} off of the stack.
8813 The code need not be optimal, since this macro is used only when
8817 @node Dispatch Tables
8818 @subsection Output of Dispatch Tables
8820 @c prevent bad page break with this line
8821 This concerns dispatch tables.
8823 @cindex dispatch table
8824 @defmac ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{body}, @var{value}, @var{rel})
8825 A C statement to output to the stdio stream @var{stream} an assembler
8826 pseudo-instruction to generate a difference between two labels.
8827 @var{value} and @var{rel} are the numbers of two internal labels. The
8828 definitions of these labels are output using
8829 @code{(*targetm.asm_out.internal_label)}, and they must be printed in the same
8830 way here. For example,
8833 fprintf (@var{stream}, "\t.word L%d-L%d\n",
8834 @var{value}, @var{rel})
8837 You must provide this macro on machines where the addresses in a
8838 dispatch table are relative to the table's own address. If defined, GCC
8839 will also use this macro on all machines when producing PIC@.
8840 @var{body} is the body of the @code{ADDR_DIFF_VEC}; it is provided so that the
8841 mode and flags can be read.
8844 @defmac ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value})
8845 This macro should be provided on machines where the addresses
8846 in a dispatch table are absolute.
8848 The definition should be a C statement to output to the stdio stream
8849 @var{stream} an assembler pseudo-instruction to generate a reference to
8850 a label. @var{value} is the number of an internal label whose
8851 definition is output using @code{(*targetm.asm_out.internal_label)}.
8855 fprintf (@var{stream}, "\t.word L%d\n", @var{value})
8859 @defmac ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table})
8860 Define this if the label before a jump-table needs to be output
8861 specially. The first three arguments are the same as for
8862 @code{(*targetm.asm_out.internal_label)}; the fourth argument is the
8863 jump-table which follows (a @code{jump_table_data} containing an
8864 @code{addr_vec} or @code{addr_diff_vec}).
8866 This feature is used on system V to output a @code{swbeg} statement
8869 If this macro is not defined, these labels are output with
8870 @code{(*targetm.asm_out.internal_label)}.
8873 @defmac ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table})
8874 Define this if something special must be output at the end of a
8875 jump-table. The definition should be a C statement to be executed
8876 after the assembler code for the table is written. It should write
8877 the appropriate code to stdio stream @var{stream}. The argument
8878 @var{table} is the jump-table insn, and @var{num} is the label-number
8879 of the preceding label.
8881 If this macro is not defined, nothing special is output at the end of
8885 @deftypefn {Target Hook} void TARGET_ASM_EMIT_UNWIND_LABEL (FILE *@var{stream}, tree @var{decl}, int @var{for_eh}, int @var{empty})
8886 This target hook emits a label at the beginning of each FDE@. It
8887 should be defined on targets where FDEs need special labels, and it
8888 should write the appropriate label, for the FDE associated with the
8889 function declaration @var{decl}, to the stdio stream @var{stream}.
8890 The third argument, @var{for_eh}, is a boolean: true if this is for an
8891 exception table. The fourth argument, @var{empty}, is a boolean:
8892 true if this is a placeholder label for an omitted FDE@.
8894 The default is that FDEs are not given nonlocal labels.
8897 @deftypefn {Target Hook} void TARGET_ASM_EMIT_EXCEPT_TABLE_LABEL (FILE *@var{stream})
8898 This target hook emits a label at the beginning of the exception table.
8899 It should be defined on targets where it is desirable for the table
8900 to be broken up according to function.
8902 The default is that no label is emitted.
8905 @deftypefn {Target Hook} void TARGET_ASM_EMIT_EXCEPT_PERSONALITY (rtx @var{personality})
8906 If the target implements @code{TARGET_ASM_UNWIND_EMIT}, this hook may be used to emit a directive to install a personality hook into the unwind info. This hook should not be used if dwarf2 unwind info is used.
8909 @deftypefn {Target Hook} void TARGET_ASM_UNWIND_EMIT (FILE *@var{stream}, rtx_insn *@var{insn})
8910 This target hook emits assembly directives required to unwind the
8911 given instruction. This is only used when @code{TARGET_EXCEPT_UNWIND_INFO}
8912 returns @code{UI_TARGET}.
8915 @deftypevr {Target Hook} bool TARGET_ASM_UNWIND_EMIT_BEFORE_INSN
8916 True if the @code{TARGET_ASM_UNWIND_EMIT} hook should be called before the assembly for @var{insn} has been emitted, false if the hook should be called afterward.
8919 @node Exception Region Output
8920 @subsection Assembler Commands for Exception Regions
8922 @c prevent bad page break with this line
8924 This describes commands marking the start and the end of an exception
8927 @defmac EH_FRAME_SECTION_NAME
8928 If defined, a C string constant for the name of the section containing
8929 exception handling frame unwind information. If not defined, GCC will
8930 provide a default definition if the target supports named sections.
8931 @file{crtstuff.c} uses this macro to switch to the appropriate section.
8933 You should define this symbol if your target supports DWARF 2 frame
8934 unwind information and the default definition does not work.
8937 @defmac EH_FRAME_IN_DATA_SECTION
8938 If defined, DWARF 2 frame unwind information will be placed in the
8939 data section even though the target supports named sections. This
8940 might be necessary, for instance, if the system linker does garbage
8941 collection and sections cannot be marked as not to be collected.
8943 Do not define this macro unless @code{TARGET_ASM_NAMED_SECTION} is
8947 @defmac EH_TABLES_CAN_BE_READ_ONLY
8948 Define this macro to 1 if your target is such that no frame unwind
8949 information encoding used with non-PIC code will ever require a
8950 runtime relocation, but the linker may not support merging read-only
8951 and read-write sections into a single read-write section.
8954 @defmac MASK_RETURN_ADDR
8955 An rtx used to mask the return address found via @code{RETURN_ADDR_RTX}, so
8956 that it does not contain any extraneous set bits in it.
8959 @defmac DWARF2_UNWIND_INFO
8960 Define this macro to 0 if your target supports DWARF 2 frame unwind
8961 information, but it does not yet work with exception handling.
8962 Otherwise, if your target supports this information (if it defines
8963 @code{INCOMING_RETURN_ADDR_RTX} and @code{OBJECT_FORMAT_ELF}),
8964 GCC will provide a default definition of 1.
8967 @deftypefn {Common Target Hook} {enum unwind_info_type} TARGET_EXCEPT_UNWIND_INFO (struct gcc_options *@var{opts})
8968 This hook defines the mechanism that will be used for exception handling
8969 by the target. If the target has ABI specified unwind tables, the hook
8970 should return @code{UI_TARGET}. If the target is to use the
8971 @code{setjmp}/@code{longjmp}-based exception handling scheme, the hook
8972 should return @code{UI_SJLJ}. If the target supports DWARF 2 frame unwind
8973 information, the hook should return @code{UI_DWARF2}.
8975 A target may, if exceptions are disabled, choose to return @code{UI_NONE}.
8976 This may end up simplifying other parts of target-specific code. The
8977 default implementation of this hook never returns @code{UI_NONE}.
8979 Note that the value returned by this hook should be constant. It should
8980 not depend on anything except the command-line switches described by
8981 @var{opts}. In particular, the
8982 setting @code{UI_SJLJ} must be fixed at compiler start-up as C pre-processor
8983 macros and builtin functions related to exception handling are set up
8984 depending on this setting.
8986 The default implementation of the hook first honors the
8987 @option{--enable-sjlj-exceptions} configure option, then
8988 @code{DWARF2_UNWIND_INFO}, and finally defaults to @code{UI_SJLJ}. If
8989 @code{DWARF2_UNWIND_INFO} depends on command-line options, the target
8990 must define this hook so that @var{opts} is used correctly.
8993 @deftypevr {Common Target Hook} bool TARGET_UNWIND_TABLES_DEFAULT
8994 This variable should be set to @code{true} if the target ABI requires unwinding
8995 tables even when exceptions are not used. It must not be modified by
8996 command-line option processing.
8999 @defmac DONT_USE_BUILTIN_SETJMP
9000 Define this macro to 1 if the @code{setjmp}/@code{longjmp}-based scheme
9001 should use the @code{setjmp}/@code{longjmp} functions from the C library
9002 instead of the @code{__builtin_setjmp}/@code{__builtin_longjmp} machinery.
9005 @defmac JMP_BUF_SIZE
9006 This macro has no effect unless @code{DONT_USE_BUILTIN_SETJMP} is also
9007 defined. Define this macro if the default size of @code{jmp_buf} buffer
9008 for the @code{setjmp}/@code{longjmp}-based exception handling mechanism
9009 is not large enough, or if it is much too large.
9010 The default size is @code{FIRST_PSEUDO_REGISTER * sizeof(void *)}.
9013 @defmac DWARF_CIE_DATA_ALIGNMENT
9014 This macro need only be defined if the target might save registers in the
9015 function prologue at an offset to the stack pointer that is not aligned to
9016 @code{UNITS_PER_WORD}. The definition should be the negative minimum
9017 alignment if @code{STACK_GROWS_DOWNWARD} is defined, and the positive
9018 minimum alignment otherwise. @xref{SDB and DWARF}. Only applicable if
9019 the target supports DWARF 2 frame unwind information.
9022 @deftypevr {Target Hook} bool TARGET_TERMINATE_DW2_EH_FRAME_INFO
9023 Contains the value true if the target should add a zero word onto the
9024 end of a Dwarf-2 frame info section when used for exception handling.
9025 Default value is false if @code{EH_FRAME_SECTION_NAME} is defined, and
9029 @deftypefn {Target Hook} rtx TARGET_DWARF_REGISTER_SPAN (rtx @var{reg})
9030 Given a register, this hook should return a parallel of registers to
9031 represent where to find the register pieces. Define this hook if the
9032 register and its mode are represented in Dwarf in non-contiguous
9033 locations, or if the register should be represented in more than one
9034 register in Dwarf. Otherwise, this hook should return @code{NULL_RTX}.
9035 If not defined, the default is to return @code{NULL_RTX}.
9038 @deftypefn {Target Hook} machine_mode TARGET_DWARF_FRAME_REG_MODE (int @var{regno})
9039 Given a register, this hook should return the mode which the
9040 corresponding Dwarf frame register should have. This is normally
9041 used to return a smaller mode than the raw mode to prevent call
9042 clobbered parts of a register altering the frame register size
9045 @deftypefn {Target Hook} void TARGET_INIT_DWARF_REG_SIZES_EXTRA (tree @var{address})
9046 If some registers are represented in Dwarf-2 unwind information in
9047 multiple pieces, define this hook to fill in information about the
9048 sizes of those pieces in the table used by the unwinder at runtime.
9049 It will be called by @code{expand_builtin_init_dwarf_reg_sizes} after
9050 filling in a single size corresponding to each hard register;
9051 @var{address} is the address of the table.
9054 @deftypefn {Target Hook} bool TARGET_ASM_TTYPE (rtx @var{sym})
9055 This hook is used to output a reference from a frame unwinding table to
9056 the type_info object identified by @var{sym}. It should return @code{true}
9057 if the reference was output. Returning @code{false} will cause the
9058 reference to be output using the normal Dwarf2 routines.
9061 @deftypevr {Target Hook} bool TARGET_ARM_EABI_UNWINDER
9062 This flag should be set to @code{true} on targets that use an ARM EABI
9063 based unwinding library, and @code{false} on other targets. This effects
9064 the format of unwinding tables, and how the unwinder in entered after
9065 running a cleanup. The default is @code{false}.
9068 @node Alignment Output
9069 @subsection Assembler Commands for Alignment
9071 @c prevent bad page break with this line
9072 This describes commands for alignment.
9074 @defmac JUMP_ALIGN (@var{label})
9075 The alignment (log base 2) to put in front of @var{label}, which is
9076 a common destination of jumps and has no fallthru incoming edge.
9078 This macro need not be defined if you don't want any special alignment
9079 to be done at such a time. Most machine descriptions do not currently
9082 Unless it's necessary to inspect the @var{label} parameter, it is better
9083 to set the variable @var{align_jumps} in the target's
9084 @code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's
9085 selection in @var{align_jumps} in a @code{JUMP_ALIGN} implementation.
9088 @deftypefn {Target Hook} int TARGET_ASM_JUMP_ALIGN_MAX_SKIP (rtx_insn *@var{label})
9089 The maximum number of bytes to skip before @var{label} when applying
9090 @code{JUMP_ALIGN}. This works only if
9091 @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
9094 @defmac LABEL_ALIGN_AFTER_BARRIER (@var{label})
9095 The alignment (log base 2) to put in front of @var{label}, which follows
9098 This macro need not be defined if you don't want any special alignment
9099 to be done at such a time. Most machine descriptions do not currently
9103 @deftypefn {Target Hook} int TARGET_ASM_LABEL_ALIGN_AFTER_BARRIER_MAX_SKIP (rtx_insn *@var{label})
9104 The maximum number of bytes to skip before @var{label} when applying
9105 @code{LABEL_ALIGN_AFTER_BARRIER}. This works only if
9106 @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
9109 @defmac LOOP_ALIGN (@var{label})
9110 The alignment (log base 2) to put in front of @var{label} that heads
9111 a frequently executed basic block (usually the header of a loop).
9113 This macro need not be defined if you don't want any special alignment
9114 to be done at such a time. Most machine descriptions do not currently
9117 Unless it's necessary to inspect the @var{label} parameter, it is better
9118 to set the variable @code{align_loops} in the target's
9119 @code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's
9120 selection in @code{align_loops} in a @code{LOOP_ALIGN} implementation.
9123 @deftypefn {Target Hook} int TARGET_ASM_LOOP_ALIGN_MAX_SKIP (rtx_insn *@var{label})
9124 The maximum number of bytes to skip when applying @code{LOOP_ALIGN} to
9125 @var{label}. This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is
9129 @defmac LABEL_ALIGN (@var{label})
9130 The alignment (log base 2) to put in front of @var{label}.
9131 If @code{LABEL_ALIGN_AFTER_BARRIER} / @code{LOOP_ALIGN} specify a different alignment,
9132 the maximum of the specified values is used.
9134 Unless it's necessary to inspect the @var{label} parameter, it is better
9135 to set the variable @code{align_labels} in the target's
9136 @code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's
9137 selection in @code{align_labels} in a @code{LABEL_ALIGN} implementation.
9140 @deftypefn {Target Hook} int TARGET_ASM_LABEL_ALIGN_MAX_SKIP (rtx_insn *@var{label})
9141 The maximum number of bytes to skip when applying @code{LABEL_ALIGN}
9142 to @var{label}. This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN}
9146 @defmac ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes})
9147 A C statement to output to the stdio stream @var{stream} an assembler
9148 instruction to advance the location counter by @var{nbytes} bytes.
9149 Those bytes should be zero when loaded. @var{nbytes} will be a C
9150 expression of type @code{unsigned HOST_WIDE_INT}.
9153 @defmac ASM_NO_SKIP_IN_TEXT
9154 Define this macro if @code{ASM_OUTPUT_SKIP} should not be used in the
9155 text section because it fails to put zeros in the bytes that are skipped.
9156 This is true on many Unix systems, where the pseudo--op to skip bytes
9157 produces no-op instructions rather than zeros when used in the text
9161 @defmac ASM_OUTPUT_ALIGN (@var{stream}, @var{power})
9162 A C statement to output to the stdio stream @var{stream} an assembler
9163 command to advance the location counter to a multiple of 2 to the
9164 @var{power} bytes. @var{power} will be a C expression of type @code{int}.
9167 @defmac ASM_OUTPUT_ALIGN_WITH_NOP (@var{stream}, @var{power})
9168 Like @code{ASM_OUTPUT_ALIGN}, except that the ``nop'' instruction is used
9169 for padding, if necessary.
9172 @defmac ASM_OUTPUT_MAX_SKIP_ALIGN (@var{stream}, @var{power}, @var{max_skip})
9173 A C statement to output to the stdio stream @var{stream} an assembler
9174 command to advance the location counter to a multiple of 2 to the
9175 @var{power} bytes, but only if @var{max_skip} or fewer bytes are needed to
9176 satisfy the alignment request. @var{power} and @var{max_skip} will be
9177 a C expression of type @code{int}.
9181 @node Debugging Info
9182 @section Controlling Debugging Information Format
9184 @c prevent bad page break with this line
9185 This describes how to specify debugging information.
9188 * All Debuggers:: Macros that affect all debugging formats uniformly.
9189 * DBX Options:: Macros enabling specific options in DBX format.
9190 * DBX Hooks:: Hook macros for varying DBX format.
9191 * File Names and DBX:: Macros controlling output of file names in DBX format.
9192 * SDB and DWARF:: Macros for SDB (COFF) and DWARF formats.
9193 * VMS Debug:: Macros for VMS debug format.
9197 @subsection Macros Affecting All Debugging Formats
9199 @c prevent bad page break with this line
9200 These macros affect all debugging formats.
9202 @defmac DBX_REGISTER_NUMBER (@var{regno})
9203 A C expression that returns the DBX register number for the compiler
9204 register number @var{regno}. In the default macro provided, the value
9205 of this expression will be @var{regno} itself. But sometimes there are
9206 some registers that the compiler knows about and DBX does not, or vice
9207 versa. In such cases, some register may need to have one number in the
9208 compiler and another for DBX@.
9210 If two registers have consecutive numbers inside GCC, and they can be
9211 used as a pair to hold a multiword value, then they @emph{must} have
9212 consecutive numbers after renumbering with @code{DBX_REGISTER_NUMBER}.
9213 Otherwise, debuggers will be unable to access such a pair, because they
9214 expect register pairs to be consecutive in their own numbering scheme.
9216 If you find yourself defining @code{DBX_REGISTER_NUMBER} in way that
9217 does not preserve register pairs, then what you must do instead is
9218 redefine the actual register numbering scheme.
9221 @defmac DEBUGGER_AUTO_OFFSET (@var{x})
9222 A C expression that returns the integer offset value for an automatic
9223 variable having address @var{x} (an RTL expression). The default
9224 computation assumes that @var{x} is based on the frame-pointer and
9225 gives the offset from the frame-pointer. This is required for targets
9226 that produce debugging output for DBX or COFF-style debugging output
9227 for SDB and allow the frame-pointer to be eliminated when the
9228 @option{-g} options is used.
9231 @defmac DEBUGGER_ARG_OFFSET (@var{offset}, @var{x})
9232 A C expression that returns the integer offset value for an argument
9233 having address @var{x} (an RTL expression). The nominal offset is
9237 @defmac PREFERRED_DEBUGGING_TYPE
9238 A C expression that returns the type of debugging output GCC should
9239 produce when the user specifies just @option{-g}. Define
9240 this if you have arranged for GCC to support more than one format of
9241 debugging output. Currently, the allowable values are @code{DBX_DEBUG},
9242 @code{SDB_DEBUG}, @code{DWARF_DEBUG}, @code{DWARF2_DEBUG},
9243 @code{XCOFF_DEBUG}, @code{VMS_DEBUG}, and @code{VMS_AND_DWARF2_DEBUG}.
9245 When the user specifies @option{-ggdb}, GCC normally also uses the
9246 value of this macro to select the debugging output format, but with two
9247 exceptions. If @code{DWARF2_DEBUGGING_INFO} is defined, GCC uses the
9248 value @code{DWARF2_DEBUG}. Otherwise, if @code{DBX_DEBUGGING_INFO} is
9249 defined, GCC uses @code{DBX_DEBUG}.
9251 The value of this macro only affects the default debugging output; the
9252 user can always get a specific type of output by using @option{-gstabs},
9253 @option{-gcoff}, @option{-gdwarf-2}, @option{-gxcoff}, or @option{-gvms}.
9257 @subsection Specific Options for DBX Output
9259 @c prevent bad page break with this line
9260 These are specific options for DBX output.
9262 @defmac DBX_DEBUGGING_INFO
9263 Define this macro if GCC should produce debugging output for DBX
9264 in response to the @option{-g} option.
9267 @defmac XCOFF_DEBUGGING_INFO
9268 Define this macro if GCC should produce XCOFF format debugging output
9269 in response to the @option{-g} option. This is a variant of DBX format.
9272 @defmac DEFAULT_GDB_EXTENSIONS
9273 Define this macro to control whether GCC should by default generate
9274 GDB's extended version of DBX debugging information (assuming DBX-format
9275 debugging information is enabled at all). If you don't define the
9276 macro, the default is 1: always generate the extended information
9277 if there is any occasion to.
9280 @defmac DEBUG_SYMS_TEXT
9281 Define this macro if all @code{.stabs} commands should be output while
9282 in the text section.
9285 @defmac ASM_STABS_OP
9286 A C string constant, including spacing, naming the assembler pseudo op to
9287 use instead of @code{"\t.stabs\t"} to define an ordinary debugging symbol.
9288 If you don't define this macro, @code{"\t.stabs\t"} is used. This macro
9289 applies only to DBX debugging information format.
9292 @defmac ASM_STABD_OP
9293 A C string constant, including spacing, naming the assembler pseudo op to
9294 use instead of @code{"\t.stabd\t"} to define a debugging symbol whose
9295 value is the current location. If you don't define this macro,
9296 @code{"\t.stabd\t"} is used. This macro applies only to DBX debugging
9300 @defmac ASM_STABN_OP
9301 A C string constant, including spacing, naming the assembler pseudo op to
9302 use instead of @code{"\t.stabn\t"} to define a debugging symbol with no
9303 name. If you don't define this macro, @code{"\t.stabn\t"} is used. This
9304 macro applies only to DBX debugging information format.
9307 @defmac DBX_NO_XREFS
9308 Define this macro if DBX on your system does not support the construct
9309 @samp{xs@var{tagname}}. On some systems, this construct is used to
9310 describe a forward reference to a structure named @var{tagname}.
9311 On other systems, this construct is not supported at all.
9314 @defmac DBX_CONTIN_LENGTH
9315 A symbol name in DBX-format debugging information is normally
9316 continued (split into two separate @code{.stabs} directives) when it
9317 exceeds a certain length (by default, 80 characters). On some
9318 operating systems, DBX requires this splitting; on others, splitting
9319 must not be done. You can inhibit splitting by defining this macro
9320 with the value zero. You can override the default splitting-length by
9321 defining this macro as an expression for the length you desire.
9324 @defmac DBX_CONTIN_CHAR
9325 Normally continuation is indicated by adding a @samp{\} character to
9326 the end of a @code{.stabs} string when a continuation follows. To use
9327 a different character instead, define this macro as a character
9328 constant for the character you want to use. Do not define this macro
9329 if backslash is correct for your system.
9332 @defmac DBX_STATIC_STAB_DATA_SECTION
9333 Define this macro if it is necessary to go to the data section before
9334 outputting the @samp{.stabs} pseudo-op for a non-global static
9338 @defmac DBX_TYPE_DECL_STABS_CODE
9339 The value to use in the ``code'' field of the @code{.stabs} directive
9340 for a typedef. The default is @code{N_LSYM}.
9343 @defmac DBX_STATIC_CONST_VAR_CODE
9344 The value to use in the ``code'' field of the @code{.stabs} directive
9345 for a static variable located in the text section. DBX format does not
9346 provide any ``right'' way to do this. The default is @code{N_FUN}.
9349 @defmac DBX_REGPARM_STABS_CODE
9350 The value to use in the ``code'' field of the @code{.stabs} directive
9351 for a parameter passed in registers. DBX format does not provide any
9352 ``right'' way to do this. The default is @code{N_RSYM}.
9355 @defmac DBX_REGPARM_STABS_LETTER
9356 The letter to use in DBX symbol data to identify a symbol as a parameter
9357 passed in registers. DBX format does not customarily provide any way to
9358 do this. The default is @code{'P'}.
9361 @defmac DBX_FUNCTION_FIRST
9362 Define this macro if the DBX information for a function and its
9363 arguments should precede the assembler code for the function. Normally,
9364 in DBX format, the debugging information entirely follows the assembler
9368 @defmac DBX_BLOCKS_FUNCTION_RELATIVE
9369 Define this macro, with value 1, if the value of a symbol describing
9370 the scope of a block (@code{N_LBRAC} or @code{N_RBRAC}) should be
9371 relative to the start of the enclosing function. Normally, GCC uses
9372 an absolute address.
9375 @defmac DBX_LINES_FUNCTION_RELATIVE
9376 Define this macro, with value 1, if the value of a symbol indicating
9377 the current line number (@code{N_SLINE}) should be relative to the
9378 start of the enclosing function. Normally, GCC uses an absolute address.
9381 @defmac DBX_USE_BINCL
9382 Define this macro if GCC should generate @code{N_BINCL} and
9383 @code{N_EINCL} stabs for included header files, as on Sun systems. This
9384 macro also directs GCC to output a type number as a pair of a file
9385 number and a type number within the file. Normally, GCC does not
9386 generate @code{N_BINCL} or @code{N_EINCL} stabs, and it outputs a single
9387 number for a type number.
9391 @subsection Open-Ended Hooks for DBX Format
9393 @c prevent bad page break with this line
9394 These are hooks for DBX format.
9396 @defmac DBX_OUTPUT_SOURCE_LINE (@var{stream}, @var{line}, @var{counter})
9397 A C statement to output DBX debugging information before code for line
9398 number @var{line} of the current source file to the stdio stream
9399 @var{stream}. @var{counter} is the number of time the macro was
9400 invoked, including the current invocation; it is intended to generate
9401 unique labels in the assembly output.
9403 This macro should not be defined if the default output is correct, or
9404 if it can be made correct by defining @code{DBX_LINES_FUNCTION_RELATIVE}.
9407 @defmac NO_DBX_FUNCTION_END
9408 Some stabs encapsulation formats (in particular ECOFF), cannot handle the
9409 @code{.stabs "",N_FUN,,0,0,Lscope-function-1} gdb dbx extension construct.
9410 On those machines, define this macro to turn this feature off without
9411 disturbing the rest of the gdb extensions.
9414 @defmac NO_DBX_BNSYM_ENSYM
9415 Some assemblers cannot handle the @code{.stabd BNSYM/ENSYM,0,0} gdb dbx
9416 extension construct. On those machines, define this macro to turn this
9417 feature off without disturbing the rest of the gdb extensions.
9420 @node File Names and DBX
9421 @subsection File Names in DBX Format
9423 @c prevent bad page break with this line
9424 This describes file names in DBX format.
9426 @defmac DBX_OUTPUT_MAIN_SOURCE_FILENAME (@var{stream}, @var{name})
9427 A C statement to output DBX debugging information to the stdio stream
9428 @var{stream}, which indicates that file @var{name} is the main source
9429 file---the file specified as the input file for compilation.
9430 This macro is called only once, at the beginning of compilation.
9432 This macro need not be defined if the standard form of output
9433 for DBX debugging information is appropriate.
9435 It may be necessary to refer to a label equal to the beginning of the
9436 text section. You can use @samp{assemble_name (stream, ltext_label_name)}
9437 to do so. If you do this, you must also set the variable
9438 @var{used_ltext_label_name} to @code{true}.
9441 @defmac NO_DBX_MAIN_SOURCE_DIRECTORY
9442 Define this macro, with value 1, if GCC should not emit an indication
9443 of the current directory for compilation and current source language at
9444 the beginning of the file.
9447 @defmac NO_DBX_GCC_MARKER
9448 Define this macro, with value 1, if GCC should not emit an indication
9449 that this object file was compiled by GCC@. The default is to emit
9450 an @code{N_OPT} stab at the beginning of every source file, with
9451 @samp{gcc2_compiled.} for the string and value 0.
9454 @defmac DBX_OUTPUT_MAIN_SOURCE_FILE_END (@var{stream}, @var{name})
9455 A C statement to output DBX debugging information at the end of
9456 compilation of the main source file @var{name}. Output should be
9457 written to the stdio stream @var{stream}.
9459 If you don't define this macro, nothing special is output at the end
9460 of compilation, which is correct for most machines.
9463 @defmac DBX_OUTPUT_NULL_N_SO_AT_MAIN_SOURCE_FILE_END
9464 Define this macro @emph{instead of} defining
9465 @code{DBX_OUTPUT_MAIN_SOURCE_FILE_END}, if what needs to be output at
9466 the end of compilation is an @code{N_SO} stab with an empty string,
9467 whose value is the highest absolute text address in the file.
9472 @subsection Macros for SDB and DWARF Output
9474 @c prevent bad page break with this line
9475 Here are macros for SDB and DWARF output.
9477 @defmac SDB_DEBUGGING_INFO
9478 Define this macro if GCC should produce COFF-style debugging output
9479 for SDB in response to the @option{-g} option.
9482 @defmac DWARF2_DEBUGGING_INFO
9483 Define this macro if GCC should produce dwarf version 2 format
9484 debugging output in response to the @option{-g} option.
9486 @deftypefn {Target Hook} int TARGET_DWARF_CALLING_CONVENTION (const_tree @var{function})
9487 Define this to enable the dwarf attribute @code{DW_AT_calling_convention} to
9488 be emitted for each function. Instead of an integer return the enum
9489 value for the @code{DW_CC_} tag.
9492 To support optional call frame debugging information, you must also
9493 define @code{INCOMING_RETURN_ADDR_RTX} and either set
9494 @code{RTX_FRAME_RELATED_P} on the prologue insns if you use RTL for the
9495 prologue, or call @code{dwarf2out_def_cfa} and @code{dwarf2out_reg_save}
9496 as appropriate from @code{TARGET_ASM_FUNCTION_PROLOGUE} if you don't.
9499 @defmac DWARF2_FRAME_INFO
9500 Define this macro to a nonzero value if GCC should always output
9501 Dwarf 2 frame information. If @code{TARGET_EXCEPT_UNWIND_INFO}
9502 (@pxref{Exception Region Output}) returns @code{UI_DWARF2}, and
9503 exceptions are enabled, GCC will output this information not matter
9504 how you define @code{DWARF2_FRAME_INFO}.
9507 @deftypefn {Target Hook} {enum unwind_info_type} TARGET_DEBUG_UNWIND_INFO (void)
9508 This hook defines the mechanism that will be used for describing frame
9509 unwind information to the debugger. Normally the hook will return
9510 @code{UI_DWARF2} if DWARF 2 debug information is enabled, and
9511 return @code{UI_NONE} otherwise.
9513 A target may return @code{UI_DWARF2} even when DWARF 2 debug information
9514 is disabled in order to always output DWARF 2 frame information.
9516 A target may return @code{UI_TARGET} if it has ABI specified unwind tables.
9517 This will suppress generation of the normal debug frame unwind information.
9520 @defmac DWARF2_ASM_LINE_DEBUG_INFO
9521 Define this macro to be a nonzero value if the assembler can generate Dwarf 2
9522 line debug info sections. This will result in much more compact line number
9523 tables, and hence is desirable if it works.
9526 @deftypevr {Target Hook} bool TARGET_WANT_DEBUG_PUB_SECTIONS
9527 True if the @code{.debug_pubtypes} and @code{.debug_pubnames} sections should be emitted. These sections are not used on most platforms, and in particular GDB does not use them.
9530 @deftypevr {Target Hook} bool TARGET_FORCE_AT_COMP_DIR
9531 True if the @code{DW_AT_comp_dir} attribute should be emitted for each compilation unit. This attribute is required for the darwin linker to emit debug information.
9534 @deftypevr {Target Hook} bool TARGET_DELAY_SCHED2
9535 True if sched2 is not to be run at its normal place.
9536 This usually means it will be run as part of machine-specific reorg.
9539 @deftypevr {Target Hook} bool TARGET_DELAY_VARTRACK
9540 True if vartrack is not to be run at its normal place.
9541 This usually means it will be run as part of machine-specific reorg.
9544 @deftypevr {Target Hook} bool TARGET_NO_REGISTER_ALLOCATION
9545 True if register allocation and the passes
9546 following it should not be run. Usually true only for virtual assembler
9550 @defmac ASM_OUTPUT_DWARF_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2})
9551 A C statement to issue assembly directives that create a difference
9552 @var{lab1} minus @var{lab2}, using an integer of the given @var{size}.
9555 @defmac ASM_OUTPUT_DWARF_VMS_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2})
9556 A C statement to issue assembly directives that create a difference
9557 between the two given labels in system defined units, e.g. instruction
9558 slots on IA64 VMS, using an integer of the given size.
9561 @defmac ASM_OUTPUT_DWARF_OFFSET (@var{stream}, @var{size}, @var{label}, @var{section})
9562 A C statement to issue assembly directives that create a
9563 section-relative reference to the given @var{label}, using an integer of the
9564 given @var{size}. The label is known to be defined in the given @var{section}.
9567 @defmac ASM_OUTPUT_DWARF_PCREL (@var{stream}, @var{size}, @var{label})
9568 A C statement to issue assembly directives that create a self-relative
9569 reference to the given @var{label}, using an integer of the given @var{size}.
9572 @defmac ASM_OUTPUT_DWARF_TABLE_REF (@var{label})
9573 A C statement to issue assembly directives that create a reference to
9574 the DWARF table identifier @var{label} from the current section. This
9575 is used on some systems to avoid garbage collecting a DWARF table which
9576 is referenced by a function.
9579 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_DWARF_DTPREL (FILE *@var{file}, int @var{size}, rtx @var{x})
9580 If defined, this target hook is a function which outputs a DTP-relative
9581 reference to the given TLS symbol of the specified size.
9584 @defmac PUT_SDB_@dots{}
9585 Define these macros to override the assembler syntax for the special
9586 SDB assembler directives. See @file{sdbout.c} for a list of these
9587 macros and their arguments. If the standard syntax is used, you need
9588 not define them yourself.
9592 Some assemblers do not support a semicolon as a delimiter, even between
9593 SDB assembler directives. In that case, define this macro to be the
9594 delimiter to use (usually @samp{\n}). It is not necessary to define
9595 a new set of @code{PUT_SDB_@var{op}} macros if this is the only change
9599 @defmac SDB_ALLOW_UNKNOWN_REFERENCES
9600 Define this macro to allow references to unknown structure,
9601 union, or enumeration tags to be emitted. Standard COFF does not
9602 allow handling of unknown references, MIPS ECOFF has support for
9606 @defmac SDB_ALLOW_FORWARD_REFERENCES
9607 Define this macro to allow references to structure, union, or
9608 enumeration tags that have not yet been seen to be handled. Some
9609 assemblers choke if forward tags are used, while some require it.
9612 @defmac SDB_OUTPUT_SOURCE_LINE (@var{stream}, @var{line})
9613 A C statement to output SDB debugging information before code for line
9614 number @var{line} of the current source file to the stdio stream
9615 @var{stream}. The default is to emit an @code{.ln} directive.
9620 @subsection Macros for VMS Debug Format
9622 @c prevent bad page break with this line
9623 Here are macros for VMS debug format.
9625 @defmac VMS_DEBUGGING_INFO
9626 Define this macro if GCC should produce debugging output for VMS
9627 in response to the @option{-g} option. The default behavior for VMS
9628 is to generate minimal debug info for a traceback in the absence of
9629 @option{-g} unless explicitly overridden with @option{-g0}. This
9630 behavior is controlled by @code{TARGET_OPTION_OPTIMIZATION} and
9631 @code{TARGET_OPTION_OVERRIDE}.
9634 @node Floating Point
9635 @section Cross Compilation and Floating Point
9636 @cindex cross compilation and floating point
9637 @cindex floating point and cross compilation
9639 While all modern machines use twos-complement representation for integers,
9640 there are a variety of representations for floating point numbers. This
9641 means that in a cross-compiler the representation of floating point numbers
9642 in the compiled program may be different from that used in the machine
9643 doing the compilation.
9645 Because different representation systems may offer different amounts of
9646 range and precision, all floating point constants must be represented in
9647 the target machine's format. Therefore, the cross compiler cannot
9648 safely use the host machine's floating point arithmetic; it must emulate
9649 the target's arithmetic. To ensure consistency, GCC always uses
9650 emulation to work with floating point values, even when the host and
9651 target floating point formats are identical.
9653 The following macros are provided by @file{real.h} for the compiler to
9654 use. All parts of the compiler which generate or optimize
9655 floating-point calculations must use these macros. They may evaluate
9656 their operands more than once, so operands must not have side effects.
9658 @defmac REAL_VALUE_TYPE
9659 The C data type to be used to hold a floating point value in the target
9660 machine's format. Typically this is a @code{struct} containing an
9661 array of @code{HOST_WIDE_INT}, but all code should treat it as an opaque
9665 @deftypefn Macro int REAL_VALUES_EQUAL (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y})
9666 Compares for equality the two values, @var{x} and @var{y}. If the target
9667 floating point format supports negative zeroes and/or NaNs,
9668 @samp{REAL_VALUES_EQUAL (-0.0, 0.0)} is true, and
9669 @samp{REAL_VALUES_EQUAL (NaN, NaN)} is false.
9672 @deftypefn Macro int REAL_VALUES_LESS (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y})
9673 Tests whether @var{x} is less than @var{y}.
9676 @deftypefn Macro HOST_WIDE_INT REAL_VALUE_FIX (REAL_VALUE_TYPE @var{x})
9677 Truncates @var{x} to a signed integer, rounding toward zero.
9680 @deftypefn Macro {unsigned HOST_WIDE_INT} REAL_VALUE_UNSIGNED_FIX (REAL_VALUE_TYPE @var{x})
9681 Truncates @var{x} to an unsigned integer, rounding toward zero. If
9682 @var{x} is negative, returns zero.
9685 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ATOF (const char *@var{string}, machine_mode @var{mode})
9686 Converts @var{string} into a floating point number in the target machine's
9687 representation for mode @var{mode}. This routine can handle both
9688 decimal and hexadecimal floating point constants, using the syntax
9689 defined by the C language for both.
9692 @deftypefn Macro int REAL_VALUE_NEGATIVE (REAL_VALUE_TYPE @var{x})
9693 Returns 1 if @var{x} is negative (including negative zero), 0 otherwise.
9696 @deftypefn Macro int REAL_VALUE_ISINF (REAL_VALUE_TYPE @var{x})
9697 Determines whether @var{x} represents infinity (positive or negative).
9700 @deftypefn Macro int REAL_VALUE_ISNAN (REAL_VALUE_TYPE @var{x})
9701 Determines whether @var{x} represents a ``NaN'' (not-a-number).
9704 @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})
9705 Calculates an arithmetic operation on the two floating point values
9706 @var{x} and @var{y}, storing the result in @var{output} (which must be a
9709 The operation to be performed is specified by @var{code}. Only the
9710 following codes are supported: @code{PLUS_EXPR}, @code{MINUS_EXPR},
9711 @code{MULT_EXPR}, @code{RDIV_EXPR}, @code{MAX_EXPR}, @code{MIN_EXPR}.
9713 If @code{REAL_ARITHMETIC} is asked to evaluate division by zero and the
9714 target's floating point format cannot represent infinity, it will call
9715 @code{abort}. Callers should check for this situation first, using
9716 @code{MODE_HAS_INFINITIES}. @xref{Storage Layout}.
9719 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_NEGATE (REAL_VALUE_TYPE @var{x})
9720 Returns the negative of the floating point value @var{x}.
9723 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ABS (REAL_VALUE_TYPE @var{x})
9724 Returns the absolute value of @var{x}.
9727 @node Mode Switching
9728 @section Mode Switching Instructions
9729 @cindex mode switching
9730 The following macros control mode switching optimizations:
9732 @defmac OPTIMIZE_MODE_SWITCHING (@var{entity})
9733 Define this macro if the port needs extra instructions inserted for mode
9734 switching in an optimizing compilation.
9736 For an example, the SH4 can perform both single and double precision
9737 floating point operations, but to perform a single precision operation,
9738 the FPSCR PR bit has to be cleared, while for a double precision
9739 operation, this bit has to be set. Changing the PR bit requires a general
9740 purpose register as a scratch register, hence these FPSCR sets have to
9741 be inserted before reload, i.e.@: you can't put this into instruction emitting
9742 or @code{TARGET_MACHINE_DEPENDENT_REORG}.
9744 You can have multiple entities that are mode-switched, and select at run time
9745 which entities actually need it. @code{OPTIMIZE_MODE_SWITCHING} should
9746 return nonzero for any @var{entity} that needs mode-switching.
9747 If you define this macro, you also have to define
9748 @code{NUM_MODES_FOR_MODE_SWITCHING}, @code{TARGET_MODE_NEEDED},
9749 @code{TARGET_MODE_PRIORITY} and @code{TARGET_MODE_EMIT}.
9750 @code{TARGET_MODE_AFTER}, @code{TARGET_MODE_ENTRY}, and @code{TARGET_MODE_EXIT}
9754 @defmac NUM_MODES_FOR_MODE_SWITCHING
9755 If you define @code{OPTIMIZE_MODE_SWITCHING}, you have to define this as
9756 initializer for an array of integers. Each initializer element
9757 N refers to an entity that needs mode switching, and specifies the number
9758 of different modes that might need to be set for this entity.
9759 The position of the initializer in the initializer---starting counting at
9760 zero---determines the integer that is used to refer to the mode-switched
9762 In macros that take mode arguments / yield a mode result, modes are
9763 represented as numbers 0 @dots{} N @minus{} 1. N is used to specify that no mode
9764 switch is needed / supplied.
9767 @deftypefn {Target Hook} void TARGET_MODE_EMIT (int @var{entity}, int @var{mode}, int @var{prev_mode}, HARD_REG_SET @var{regs_live})
9768 Generate one or more insns to set @var{entity} to @var{mode}. @var{hard_reg_live} is the set of hard registers live at the point where the insn(s) are to be inserted. @var{prev_moxde} indicates the mode to switch from. Sets of a lower numbered entity will be emitted before sets of a higher numbered entity to a mode of the same or lower priority.
9771 @deftypefn {Target Hook} int TARGET_MODE_NEEDED (int @var{entity}, rtx_insn *@var{insn})
9772 @var{entity} is an integer specifying a mode-switched entity. If @code{OPTIMIZE_MODE_SWITCHING} is defined, you must define this macro to return an integer value not larger than the corresponding element in @code{NUM_MODES_FOR_MODE_SWITCHING}, to denote the mode that @var{entity} must be switched into prior to the execution of @var{insn}.
9775 @deftypefn {Target Hook} int TARGET_MODE_AFTER (int @var{entity}, int @var{mode}, rtx_insn *@var{insn})
9776 @var{entity} is an integer specifying a mode-switched entity. If this macro is defined, it is evaluated for every @var{insn} during mode switching. It determines the mode that an insn results in (if different from the incoming mode).
9779 @deftypefn {Target Hook} int TARGET_MODE_ENTRY (int @var{entity})
9780 If this macro is defined, it is evaluated for every @var{entity} that needs mode switching. It should evaluate to an integer, which is a mode that @var{entity} is assumed to be switched to at function entry. If @code{TARGET_MODE_ENTRY} is defined then @code{TARGET_MODE_EXIT} must be defined.
9783 @deftypefn {Target Hook} int TARGET_MODE_EXIT (int @var{entity})
9784 If this macro is defined, it is evaluated for every @var{entity} that needs mode switching. It should evaluate to an integer, which is a mode that @var{entity} is assumed to be switched to at function exit. If @code{TARGET_MODE_EXIT} is defined then @code{TARGET_MODE_ENTRY} must be defined.
9787 @deftypefn {Target Hook} int TARGET_MODE_PRIORITY (int @var{entity}, int @var{n})
9788 This macro specifies the order in which modes for @var{entity} are processed. 0 is the highest priority, @code{NUM_MODES_FOR_MODE_SWITCHING[@var{entity}] - 1} the lowest. The value of the macro should be an integer designating a mode for @var{entity}. For any fixed @var{entity}, @code{mode_priority} (@var{entity}, @var{n}) shall be a bijection in 0 @dots{} @code{num_modes_for_mode_switching[@var{entity}] - 1}.
9791 @node Target Attributes
9792 @section Defining target-specific uses of @code{__attribute__}
9793 @cindex target attributes
9794 @cindex machine attributes
9795 @cindex attributes, target-specific
9797 Target-specific attributes may be defined for functions, data and types.
9798 These are described using the following target hooks; they also need to
9799 be documented in @file{extend.texi}.
9801 @deftypevr {Target Hook} {const struct attribute_spec *} TARGET_ATTRIBUTE_TABLE
9802 If defined, this target hook points to an array of @samp{struct
9803 attribute_spec} (defined in @file{tree.h}) specifying the machine
9804 specific attributes for this target and some of the restrictions on the
9805 entities to which these attributes are applied and the arguments they
9809 @deftypefn {Target Hook} bool TARGET_ATTRIBUTE_TAKES_IDENTIFIER_P (const_tree @var{name})
9810 If defined, this target hook is a function which returns true if the
9811 machine-specific attribute named @var{name} expects an identifier
9812 given as its first argument to be passed on as a plain identifier, not
9813 subjected to name lookup. If this is not defined, the default is
9814 false for all machine-specific attributes.
9817 @deftypefn {Target Hook} int TARGET_COMP_TYPE_ATTRIBUTES (const_tree @var{type1}, const_tree @var{type2})
9818 If defined, this target hook is a function which returns zero if the attributes on
9819 @var{type1} and @var{type2} are incompatible, one if they are compatible,
9820 and two if they are nearly compatible (which causes a warning to be
9821 generated). If this is not defined, machine-specific attributes are
9822 supposed always to be compatible.
9825 @deftypefn {Target Hook} void TARGET_SET_DEFAULT_TYPE_ATTRIBUTES (tree @var{type})
9826 If defined, this target hook is a function which assigns default attributes to
9827 the newly defined @var{type}.
9830 @deftypefn {Target Hook} tree TARGET_MERGE_TYPE_ATTRIBUTES (tree @var{type1}, tree @var{type2})
9831 Define this target hook if the merging of type attributes needs special
9832 handling. If defined, the result is a list of the combined
9833 @code{TYPE_ATTRIBUTES} of @var{type1} and @var{type2}. It is assumed
9834 that @code{comptypes} has already been called and returned 1. This
9835 function may call @code{merge_attributes} to handle machine-independent
9839 @deftypefn {Target Hook} tree TARGET_MERGE_DECL_ATTRIBUTES (tree @var{olddecl}, tree @var{newdecl})
9840 Define this target hook if the merging of decl attributes needs special
9841 handling. If defined, the result is a list of the combined
9842 @code{DECL_ATTRIBUTES} of @var{olddecl} and @var{newdecl}.
9843 @var{newdecl} is a duplicate declaration of @var{olddecl}. Examples of
9844 when this is needed are when one attribute overrides another, or when an
9845 attribute is nullified by a subsequent definition. This function may
9846 call @code{merge_attributes} to handle machine-independent merging.
9848 @findex TARGET_DLLIMPORT_DECL_ATTRIBUTES
9849 If the only target-specific handling you require is @samp{dllimport}
9850 for Microsoft Windows targets, you should define the macro
9851 @code{TARGET_DLLIMPORT_DECL_ATTRIBUTES} to @code{1}. The compiler
9852 will then define a function called
9853 @code{merge_dllimport_decl_attributes} which can then be defined as
9854 the expansion of @code{TARGET_MERGE_DECL_ATTRIBUTES}. You can also
9855 add @code{handle_dll_attribute} in the attribute table for your port
9856 to perform initial processing of the @samp{dllimport} and
9857 @samp{dllexport} attributes. This is done in @file{i386/cygwin.h} and
9858 @file{i386/i386.c}, for example.
9861 @deftypefn {Target Hook} bool TARGET_VALID_DLLIMPORT_ATTRIBUTE_P (const_tree @var{decl})
9862 @var{decl} is a variable or function with @code{__attribute__((dllimport))} specified. Use this hook if the target needs to add extra validation checks to @code{handle_dll_attribute}.
9865 @defmac TARGET_DECLSPEC
9866 Define this macro to a nonzero value if you want to treat
9867 @code{__declspec(X)} as equivalent to @code{__attribute((X))}. By
9868 default, this behavior is enabled only for targets that define
9869 @code{TARGET_DLLIMPORT_DECL_ATTRIBUTES}. The current implementation
9870 of @code{__declspec} is via a built-in macro, but you should not rely
9871 on this implementation detail.
9874 @deftypefn {Target Hook} void TARGET_INSERT_ATTRIBUTES (tree @var{node}, tree *@var{attr_ptr})
9875 Define this target hook if you want to be able to add attributes to a decl
9876 when it is being created. This is normally useful for back ends which
9877 wish to implement a pragma by using the attributes which correspond to
9878 the pragma's effect. The @var{node} argument is the decl which is being
9879 created. The @var{attr_ptr} argument is a pointer to the attribute list
9880 for this decl. The list itself should not be modified, since it may be
9881 shared with other decls, but attributes may be chained on the head of
9882 the list and @code{*@var{attr_ptr}} modified to point to the new
9883 attributes, or a copy of the list may be made if further changes are
9887 @deftypefn {Target Hook} bool TARGET_FUNCTION_ATTRIBUTE_INLINABLE_P (const_tree @var{fndecl})
9889 This target hook returns @code{true} if it is OK to inline @var{fndecl}
9890 into the current function, despite its having target-specific
9891 attributes, @code{false} otherwise. By default, if a function has a
9892 target specific attribute attached to it, it will not be inlined.
9895 @deftypefn {Target Hook} bool TARGET_OPTION_VALID_ATTRIBUTE_P (tree @var{fndecl}, tree @var{name}, tree @var{args}, int @var{flags})
9896 This hook is called to parse @code{attribute(target("..."))}, which
9897 allows setting target-specific options on individual functions.
9898 These function-specific options may differ
9899 from the options specified on the command line. The hook should return
9900 @code{true} if the options are valid.
9902 The hook should set the @code{DECL_FUNCTION_SPECIFIC_TARGET} field in
9903 the function declaration to hold a pointer to a target-specific
9904 @code{struct cl_target_option} structure.
9907 @deftypefn {Target Hook} void TARGET_OPTION_SAVE (struct cl_target_option *@var{ptr}, struct gcc_options *@var{opts})
9908 This hook is called to save any additional target-specific information
9909 in the @code{struct cl_target_option} structure for function-specific
9910 options from the @code{struct gcc_options} structure.
9911 @xref{Option file format}.
9914 @deftypefn {Target Hook} void TARGET_OPTION_RESTORE (struct gcc_options *@var{opts}, struct cl_target_option *@var{ptr})
9915 This hook is called to restore any additional target-specific
9916 information in the @code{struct cl_target_option} structure for
9917 function-specific options to the @code{struct gcc_options} structure.
9920 @deftypefn {Target Hook} void TARGET_OPTION_POST_STREAM_IN (struct cl_target_option *@var{ptr})
9921 This hook is called to update target-specific information in the
9922 @code{struct cl_target_option} structure after it is streamed in from
9926 @deftypefn {Target Hook} void TARGET_OPTION_PRINT (FILE *@var{file}, int @var{indent}, struct cl_target_option *@var{ptr})
9927 This hook is called to print any additional target-specific
9928 information in the @code{struct cl_target_option} structure for
9929 function-specific options.
9932 @deftypefn {Target Hook} bool TARGET_OPTION_PRAGMA_PARSE (tree @var{args}, tree @var{pop_target})
9933 This target hook parses the options for @code{#pragma GCC target}, which
9934 sets the target-specific options for functions that occur later in the
9935 input stream. The options accepted should be the same as those handled by the
9936 @code{TARGET_OPTION_VALID_ATTRIBUTE_P} hook.
9939 @deftypefn {Target Hook} void TARGET_OPTION_OVERRIDE (void)
9940 Sometimes certain combinations of command options do not make sense on
9941 a particular target machine. You can override the hook
9942 @code{TARGET_OPTION_OVERRIDE} to take account of this. This hooks is called
9943 once just after all the command options have been parsed.
9945 Don't use this hook to turn on various extra optimizations for
9946 @option{-O}. That is what @code{TARGET_OPTION_OPTIMIZATION} is for.
9948 If you need to do something whenever the optimization level is
9949 changed via the optimize attribute or pragma, see
9950 @code{TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE}
9953 @deftypefn {Target Hook} bool TARGET_OPTION_FUNCTION_VERSIONS (tree @var{decl1}, tree @var{decl2})
9954 This target hook returns @code{true} if @var{DECL1} and @var{DECL2} are
9955 versions of the same function. @var{DECL1} and @var{DECL2} are function
9956 versions if and only if they have the same function signature and
9957 different target specific attributes, that is, they are compiled for
9958 different target machines.
9961 @deftypefn {Target Hook} bool TARGET_CAN_INLINE_P (tree @var{caller}, tree @var{callee})
9962 This target hook returns @code{false} if the @var{caller} function
9963 cannot inline @var{callee}, based on target specific information. By
9964 default, inlining is not allowed if the callee function has function
9965 specific target options and the caller does not use the same options.
9969 @section Emulating TLS
9970 @cindex Emulated TLS
9972 For targets whose psABI does not provide Thread Local Storage via
9973 specific relocations and instruction sequences, an emulation layer is
9974 used. A set of target hooks allows this emulation layer to be
9975 configured for the requirements of a particular target. For instance
9976 the psABI may in fact specify TLS support in terms of an emulation
9979 The emulation layer works by creating a control object for every TLS
9980 object. To access the TLS object, a lookup function is provided
9981 which, when given the address of the control object, will return the
9982 address of the current thread's instance of the TLS object.
9984 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_GET_ADDRESS
9985 Contains the name of the helper function that uses a TLS control
9986 object to locate a TLS instance. The default causes libgcc's
9987 emulated TLS helper function to be used.
9990 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_REGISTER_COMMON
9991 Contains the name of the helper function that should be used at
9992 program startup to register TLS objects that are implicitly
9993 initialized to zero. If this is @code{NULL}, all TLS objects will
9994 have explicit initializers. The default causes libgcc's emulated TLS
9995 registration function to be used.
9998 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_VAR_SECTION
9999 Contains the name of the section in which TLS control variables should
10000 be placed. The default of @code{NULL} allows these to be placed in
10004 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_TMPL_SECTION
10005 Contains the name of the section in which TLS initializers should be
10006 placed. The default of @code{NULL} allows these to be placed in any
10010 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_VAR_PREFIX
10011 Contains the prefix to be prepended to TLS control variable names.
10012 The default of @code{NULL} uses a target-specific prefix.
10015 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_TMPL_PREFIX
10016 Contains the prefix to be prepended to TLS initializer objects. The
10017 default of @code{NULL} uses a target-specific prefix.
10020 @deftypefn {Target Hook} tree TARGET_EMUTLS_VAR_FIELDS (tree @var{type}, tree *@var{name})
10021 Specifies a function that generates the FIELD_DECLs for a TLS control
10022 object type. @var{type} is the RECORD_TYPE the fields are for and
10023 @var{name} should be filled with the structure tag, if the default of
10024 @code{__emutls_object} is unsuitable. The default creates a type suitable
10025 for libgcc's emulated TLS function.
10028 @deftypefn {Target Hook} tree TARGET_EMUTLS_VAR_INIT (tree @var{var}, tree @var{decl}, tree @var{tmpl_addr})
10029 Specifies a function that generates the CONSTRUCTOR to initialize a
10030 TLS control object. @var{var} is the TLS control object, @var{decl}
10031 is the TLS object and @var{tmpl_addr} is the address of the
10032 initializer. The default initializes libgcc's emulated TLS control object.
10035 @deftypevr {Target Hook} bool TARGET_EMUTLS_VAR_ALIGN_FIXED
10036 Specifies whether the alignment of TLS control variable objects is
10037 fixed and should not be increased as some backends may do to optimize
10038 single objects. The default is false.
10041 @deftypevr {Target Hook} bool TARGET_EMUTLS_DEBUG_FORM_TLS_ADDRESS
10042 Specifies whether a DWARF @code{DW_OP_form_tls_address} location descriptor
10043 may be used to describe emulated TLS control objects.
10046 @node MIPS Coprocessors
10047 @section Defining coprocessor specifics for MIPS targets.
10048 @cindex MIPS coprocessor-definition macros
10050 The MIPS specification allows MIPS implementations to have as many as 4
10051 coprocessors, each with as many as 32 private registers. GCC supports
10052 accessing these registers and transferring values between the registers
10053 and memory using asm-ized variables. For example:
10056 register unsigned int cp0count asm ("c0r1");
10062 (``c0r1'' is the default name of register 1 in coprocessor 0; alternate
10063 names may be added as described below, or the default names may be
10064 overridden entirely in @code{SUBTARGET_CONDITIONAL_REGISTER_USAGE}.)
10066 Coprocessor registers are assumed to be epilogue-used; sets to them will
10067 be preserved even if it does not appear that the register is used again
10068 later in the function.
10070 Another note: according to the MIPS spec, coprocessor 1 (if present) is
10071 the FPU@. One accesses COP1 registers through standard mips
10072 floating-point support; they are not included in this mechanism.
10075 @section Parameters for Precompiled Header Validity Checking
10076 @cindex parameters, precompiled headers
10078 @deftypefn {Target Hook} {void *} TARGET_GET_PCH_VALIDITY (size_t *@var{sz})
10079 This hook returns a pointer to the data needed by
10080 @code{TARGET_PCH_VALID_P} and sets
10081 @samp{*@var{sz}} to the size of the data in bytes.
10084 @deftypefn {Target Hook} {const char *} TARGET_PCH_VALID_P (const void *@var{data}, size_t @var{sz})
10085 This hook checks whether the options used to create a PCH file are
10086 compatible with the current settings. It returns @code{NULL}
10087 if so and a suitable error message if not. Error messages will
10088 be presented to the user and must be localized using @samp{_(@var{msg})}.
10090 @var{data} is the data that was returned by @code{TARGET_GET_PCH_VALIDITY}
10091 when the PCH file was created and @var{sz} is the size of that data in bytes.
10092 It's safe to assume that the data was created by the same version of the
10093 compiler, so no format checking is needed.
10095 The default definition of @code{default_pch_valid_p} should be
10096 suitable for most targets.
10099 @deftypefn {Target Hook} {const char *} TARGET_CHECK_PCH_TARGET_FLAGS (int @var{pch_flags})
10100 If this hook is nonnull, the default implementation of
10101 @code{TARGET_PCH_VALID_P} will use it to check for compatible values
10102 of @code{target_flags}. @var{pch_flags} specifies the value that
10103 @code{target_flags} had when the PCH file was created. The return
10104 value is the same as for @code{TARGET_PCH_VALID_P}.
10107 @deftypefn {Target Hook} void TARGET_PREPARE_PCH_SAVE (void)
10108 Called before writing out a PCH file. If the target has some
10109 garbage-collected data that needs to be in a particular state on PCH loads,
10110 it can use this hook to enforce that state. Very few targets need
10111 to do anything here.
10115 @section C++ ABI parameters
10116 @cindex parameters, c++ abi
10118 @deftypefn {Target Hook} tree TARGET_CXX_GUARD_TYPE (void)
10119 Define this hook to override the integer type used for guard variables.
10120 These are used to implement one-time construction of static objects. The
10121 default is long_long_integer_type_node.
10124 @deftypefn {Target Hook} bool TARGET_CXX_GUARD_MASK_BIT (void)
10125 This hook determines how guard variables are used. It should return
10126 @code{false} (the default) if the first byte should be used. A return value of
10127 @code{true} indicates that only the least significant bit should be used.
10130 @deftypefn {Target Hook} tree TARGET_CXX_GET_COOKIE_SIZE (tree @var{type})
10131 This hook returns the size of the cookie to use when allocating an array
10132 whose elements have the indicated @var{type}. Assumes that it is already
10133 known that a cookie is needed. The default is
10134 @code{max(sizeof (size_t), alignof(type))}, as defined in section 2.7 of the
10135 IA64/Generic C++ ABI@.
10138 @deftypefn {Target Hook} bool TARGET_CXX_COOKIE_HAS_SIZE (void)
10139 This hook should return @code{true} if the element size should be stored in
10140 array cookies. The default is to return @code{false}.
10143 @deftypefn {Target Hook} int TARGET_CXX_IMPORT_EXPORT_CLASS (tree @var{type}, int @var{import_export})
10144 If defined by a backend this hook allows the decision made to export
10145 class @var{type} to be overruled. Upon entry @var{import_export}
10146 will contain 1 if the class is going to be exported, @minus{}1 if it is going
10147 to be imported and 0 otherwise. This function should return the
10148 modified value and perform any other actions necessary to support the
10149 backend's targeted operating system.
10152 @deftypefn {Target Hook} bool TARGET_CXX_CDTOR_RETURNS_THIS (void)
10153 This hook should return @code{true} if constructors and destructors return
10154 the address of the object created/destroyed. The default is to return
10158 @deftypefn {Target Hook} bool TARGET_CXX_KEY_METHOD_MAY_BE_INLINE (void)
10159 This hook returns true if the key method for a class (i.e., the method
10160 which, if defined in the current translation unit, causes the virtual
10161 table to be emitted) may be an inline function. Under the standard
10162 Itanium C++ ABI the key method may be an inline function so long as
10163 the function is not declared inline in the class definition. Under
10164 some variants of the ABI, an inline function can never be the key
10165 method. The default is to return @code{true}.
10168 @deftypefn {Target Hook} void TARGET_CXX_DETERMINE_CLASS_DATA_VISIBILITY (tree @var{decl})
10169 @var{decl} is a virtual table, virtual table table, typeinfo object, or other similar implicit class data object that will be emitted with external linkage in this translation unit. No ELF visibility has been explicitly specified. If the target needs to specify a visibility other than that of the containing class, use this hook to set @code{DECL_VISIBILITY} and @code{DECL_VISIBILITY_SPECIFIED}.
10172 @deftypefn {Target Hook} bool TARGET_CXX_CLASS_DATA_ALWAYS_COMDAT (void)
10173 This hook returns true (the default) if virtual tables and other
10174 similar implicit class data objects are always COMDAT if they have
10175 external linkage. If this hook returns false, then class data for
10176 classes whose virtual table will be emitted in only one translation
10177 unit will not be COMDAT.
10180 @deftypefn {Target Hook} bool TARGET_CXX_LIBRARY_RTTI_COMDAT (void)
10181 This hook returns true (the default) if the RTTI information for
10182 the basic types which is defined in the C++ runtime should always
10183 be COMDAT, false if it should not be COMDAT.
10186 @deftypefn {Target Hook} bool TARGET_CXX_USE_AEABI_ATEXIT (void)
10187 This hook returns true if @code{__aeabi_atexit} (as defined by the ARM EABI)
10188 should be used to register static destructors when @option{-fuse-cxa-atexit}
10189 is in effect. The default is to return false to use @code{__cxa_atexit}.
10192 @deftypefn {Target Hook} bool TARGET_CXX_USE_ATEXIT_FOR_CXA_ATEXIT (void)
10193 This hook returns true if the target @code{atexit} function can be used
10194 in the same manner as @code{__cxa_atexit} to register C++ static
10195 destructors. This requires that @code{atexit}-registered functions in
10196 shared libraries are run in the correct order when the libraries are
10197 unloaded. The default is to return false.
10200 @deftypefn {Target Hook} void TARGET_CXX_ADJUST_CLASS_AT_DEFINITION (tree @var{type})
10201 @var{type} is a C++ class (i.e., RECORD_TYPE or UNION_TYPE) that has just been defined. Use this hook to make adjustments to the class (eg, tweak visibility or perform any other required target modifications).
10204 @deftypefn {Target Hook} tree TARGET_CXX_DECL_MANGLING_CONTEXT (const_tree @var{decl})
10205 Return target-specific mangling context of @var{decl} or @code{NULL_TREE}.
10208 @node Named Address Spaces
10209 @section Adding support for named address spaces
10210 @cindex named address spaces
10212 The draft technical report of the ISO/IEC JTC1 S22 WG14 N1275
10213 standards committee, @cite{Programming Languages - C - Extensions to
10214 support embedded processors}, specifies a syntax for embedded
10215 processors to specify alternate address spaces. You can configure a
10216 GCC port to support section 5.1 of the draft report to add support for
10217 address spaces other than the default address space. These address
10218 spaces are new keywords that are similar to the @code{volatile} and
10219 @code{const} type attributes.
10221 Pointers to named address spaces can have a different size than
10222 pointers to the generic address space.
10224 For example, the SPU port uses the @code{__ea} address space to refer
10225 to memory in the host processor, rather than memory local to the SPU
10226 processor. Access to memory in the @code{__ea} address space involves
10227 issuing DMA operations to move data between the host processor and the
10228 local processor memory address space. Pointers in the @code{__ea}
10229 address space are either 32 bits or 64 bits based on the
10230 @option{-mea32} or @option{-mea64} switches (native SPU pointers are
10233 Internally, address spaces are represented as a small integer in the
10234 range 0 to 15 with address space 0 being reserved for the generic
10237 To register a named address space qualifier keyword with the C front end,
10238 the target may call the @code{c_register_addr_space} routine. For example,
10239 the SPU port uses the following to declare @code{__ea} as the keyword for
10240 named address space #1:
10242 #define ADDR_SPACE_EA 1
10243 c_register_addr_space ("__ea", ADDR_SPACE_EA);
10246 @deftypefn {Target Hook} machine_mode TARGET_ADDR_SPACE_POINTER_MODE (addr_space_t @var{address_space})
10247 Define this to return the machine mode to use for pointers to
10248 @var{address_space} if the target supports named address spaces.
10249 The default version of this hook returns @code{ptr_mode} for the
10250 generic address space only.
10253 @deftypefn {Target Hook} machine_mode TARGET_ADDR_SPACE_ADDRESS_MODE (addr_space_t @var{address_space})
10254 Define this to return the machine mode to use for addresses in
10255 @var{address_space} if the target supports named address spaces.
10256 The default version of this hook returns @code{Pmode} for the
10257 generic address space only.
10260 @deftypefn {Target Hook} bool TARGET_ADDR_SPACE_VALID_POINTER_MODE (machine_mode @var{mode}, addr_space_t @var{as})
10261 Define this to return nonzero if the port can handle pointers
10262 with machine mode @var{mode} to address space @var{as}. This target
10263 hook is the same as the @code{TARGET_VALID_POINTER_MODE} target hook,
10264 except that it includes explicit named address space support. The default
10265 version of this hook returns true for the modes returned by either the
10266 @code{TARGET_ADDR_SPACE_POINTER_MODE} or @code{TARGET_ADDR_SPACE_ADDRESS_MODE}
10267 target hooks for the given address space.
10270 @deftypefn {Target Hook} bool TARGET_ADDR_SPACE_LEGITIMATE_ADDRESS_P (machine_mode @var{mode}, rtx @var{exp}, bool @var{strict}, addr_space_t @var{as})
10271 Define this to return true if @var{exp} is a valid address for mode
10272 @var{mode} in the named address space @var{as}. The @var{strict}
10273 parameter says whether strict addressing is in effect after reload has
10274 finished. This target hook is the same as the
10275 @code{TARGET_LEGITIMATE_ADDRESS_P} target hook, except that it includes
10276 explicit named address space support.
10279 @deftypefn {Target Hook} rtx TARGET_ADDR_SPACE_LEGITIMIZE_ADDRESS (rtx @var{x}, rtx @var{oldx}, machine_mode @var{mode}, addr_space_t @var{as})
10280 Define this to modify an invalid address @var{x} to be a valid address
10281 with mode @var{mode} in the named address space @var{as}. This target
10282 hook is the same as the @code{TARGET_LEGITIMIZE_ADDRESS} target hook,
10283 except that it includes explicit named address space support.
10286 @deftypefn {Target Hook} bool TARGET_ADDR_SPACE_SUBSET_P (addr_space_t @var{subset}, addr_space_t @var{superset})
10287 Define this to return whether the @var{subset} named address space is
10288 contained within the @var{superset} named address space. Pointers to
10289 a named address space that is a subset of another named address space
10290 will be converted automatically without a cast if used together in
10291 arithmetic operations. Pointers to a superset address space can be
10292 converted to pointers to a subset address space via explicit casts.
10295 @deftypefn {Target Hook} rtx TARGET_ADDR_SPACE_CONVERT (rtx @var{op}, tree @var{from_type}, tree @var{to_type})
10296 Define this to convert the pointer expression represented by the RTL
10297 @var{op} with type @var{from_type} that points to a named address
10298 space to a new pointer expression with type @var{to_type} that points
10299 to a different named address space. When this hook it called, it is
10300 guaranteed that one of the two address spaces is a subset of the other,
10301 as determined by the @code{TARGET_ADDR_SPACE_SUBSET_P} target hook.
10305 @section Miscellaneous Parameters
10306 @cindex parameters, miscellaneous
10308 @c prevent bad page break with this line
10309 Here are several miscellaneous parameters.
10311 @defmac HAS_LONG_COND_BRANCH
10312 Define this boolean macro to indicate whether or not your architecture
10313 has conditional branches that can span all of memory. It is used in
10314 conjunction with an optimization that partitions hot and cold basic
10315 blocks into separate sections of the executable. If this macro is
10316 set to false, gcc will convert any conditional branches that attempt
10317 to cross between sections into unconditional branches or indirect jumps.
10320 @defmac HAS_LONG_UNCOND_BRANCH
10321 Define this boolean macro to indicate whether or not your architecture
10322 has unconditional branches that can span all of memory. It is used in
10323 conjunction with an optimization that partitions hot and cold basic
10324 blocks into separate sections of the executable. If this macro is
10325 set to false, gcc will convert any unconditional branches that attempt
10326 to cross between sections into indirect jumps.
10329 @defmac CASE_VECTOR_MODE
10330 An alias for a machine mode name. This is the machine mode that
10331 elements of a jump-table should have.
10334 @defmac CASE_VECTOR_SHORTEN_MODE (@var{min_offset}, @var{max_offset}, @var{body})
10335 Optional: return the preferred mode for an @code{addr_diff_vec}
10336 when the minimum and maximum offset are known. If you define this,
10337 it enables extra code in branch shortening to deal with @code{addr_diff_vec}.
10338 To make this work, you also have to define @code{INSN_ALIGN} and
10339 make the alignment for @code{addr_diff_vec} explicit.
10340 The @var{body} argument is provided so that the offset_unsigned and scale
10341 flags can be updated.
10344 @defmac CASE_VECTOR_PC_RELATIVE
10345 Define this macro to be a C expression to indicate when jump-tables
10346 should contain relative addresses. You need not define this macro if
10347 jump-tables never contain relative addresses, or jump-tables should
10348 contain relative addresses only when @option{-fPIC} or @option{-fPIC}
10352 @deftypefn {Target Hook} {unsigned int} TARGET_CASE_VALUES_THRESHOLD (void)
10353 This function return the smallest number of different values for which it
10354 is best to use a jump-table instead of a tree of conditional branches.
10355 The default is four for machines with a @code{casesi} instruction and
10356 five otherwise. This is best for most machines.
10359 @defmac WORD_REGISTER_OPERATIONS
10360 Define this macro if operations between registers with integral mode
10361 smaller than a word are always performed on the entire register.
10362 Most RISC machines have this property and most CISC machines do not.
10365 @defmac LOAD_EXTEND_OP (@var{mem_mode})
10366 Define this macro to be a C expression indicating when insns that read
10367 memory in @var{mem_mode}, an integral mode narrower than a word, set the
10368 bits outside of @var{mem_mode} to be either the sign-extension or the
10369 zero-extension of the data read. Return @code{SIGN_EXTEND} for values
10370 of @var{mem_mode} for which the
10371 insn sign-extends, @code{ZERO_EXTEND} for which it zero-extends, and
10372 @code{UNKNOWN} for other modes.
10374 This macro is not called with @var{mem_mode} non-integral or with a width
10375 greater than or equal to @code{BITS_PER_WORD}, so you may return any
10376 value in this case. Do not define this macro if it would always return
10377 @code{UNKNOWN}. On machines where this macro is defined, you will normally
10378 define it as the constant @code{SIGN_EXTEND} or @code{ZERO_EXTEND}.
10380 You may return a non-@code{UNKNOWN} value even if for some hard registers
10381 the sign extension is not performed, if for the @code{REGNO_REG_CLASS}
10382 of these hard registers @code{CANNOT_CHANGE_MODE_CLASS} returns nonzero
10383 when the @var{from} mode is @var{mem_mode} and the @var{to} mode is any
10384 integral mode larger than this but not larger than @code{word_mode}.
10386 You must return @code{UNKNOWN} if for some hard registers that allow this
10387 mode, @code{CANNOT_CHANGE_MODE_CLASS} says that they cannot change to
10388 @code{word_mode}, but that they can change to another integral mode that
10389 is larger then @var{mem_mode} but still smaller than @code{word_mode}.
10392 @defmac SHORT_IMMEDIATES_SIGN_EXTEND
10393 Define this macro if loading short immediate values into registers sign
10397 @deftypefn {Target Hook} {unsigned int} TARGET_MIN_DIVISIONS_FOR_RECIP_MUL (machine_mode @var{mode})
10398 When @option{-ffast-math} is in effect, GCC tries to optimize
10399 divisions by the same divisor, by turning them into multiplications by
10400 the reciprocal. This target hook specifies the minimum number of divisions
10401 that should be there for GCC to perform the optimization for a variable
10402 of mode @var{mode}. The default implementation returns 3 if the machine
10403 has an instruction for the division, and 2 if it does not.
10407 The maximum number of bytes that a single instruction can move quickly
10408 between memory and registers or between two memory locations.
10411 @defmac MAX_MOVE_MAX
10412 The maximum number of bytes that a single instruction can move quickly
10413 between memory and registers or between two memory locations. If this
10414 is undefined, the default is @code{MOVE_MAX}. Otherwise, it is the
10415 constant value that is the largest value that @code{MOVE_MAX} can have
10419 @defmac SHIFT_COUNT_TRUNCATED
10420 A C expression that is nonzero if on this machine the number of bits
10421 actually used for the count of a shift operation is equal to the number
10422 of bits needed to represent the size of the object being shifted. When
10423 this macro is nonzero, the compiler will assume that it is safe to omit
10424 a sign-extend, zero-extend, and certain bitwise `and' instructions that
10425 truncates the count of a shift operation. On machines that have
10426 instructions that act on bit-fields at variable positions, which may
10427 include `bit test' instructions, a nonzero @code{SHIFT_COUNT_TRUNCATED}
10428 also enables deletion of truncations of the values that serve as
10429 arguments to bit-field instructions.
10431 If both types of instructions truncate the count (for shifts) and
10432 position (for bit-field operations), or if no variable-position bit-field
10433 instructions exist, you should define this macro.
10435 However, on some machines, such as the 80386 and the 680x0, truncation
10436 only applies to shift operations and not the (real or pretended)
10437 bit-field operations. Define @code{SHIFT_COUNT_TRUNCATED} to be zero on
10438 such machines. Instead, add patterns to the @file{md} file that include
10439 the implied truncation of the shift instructions.
10441 You need not define this macro if it would always have the value of zero.
10444 @anchor{TARGET_SHIFT_TRUNCATION_MASK}
10445 @deftypefn {Target Hook} {unsigned HOST_WIDE_INT} TARGET_SHIFT_TRUNCATION_MASK (machine_mode @var{mode})
10446 This function describes how the standard shift patterns for @var{mode}
10447 deal with shifts by negative amounts or by more than the width of the mode.
10448 @xref{shift patterns}.
10450 On many machines, the shift patterns will apply a mask @var{m} to the
10451 shift count, meaning that a fixed-width shift of @var{x} by @var{y} is
10452 equivalent to an arbitrary-width shift of @var{x} by @var{y & m}. If
10453 this is true for mode @var{mode}, the function should return @var{m},
10454 otherwise it should return 0. A return value of 0 indicates that no
10455 particular behavior is guaranteed.
10457 Note that, unlike @code{SHIFT_COUNT_TRUNCATED}, this function does
10458 @emph{not} apply to general shift rtxes; it applies only to instructions
10459 that are generated by the named shift patterns.
10461 The default implementation of this function returns
10462 @code{GET_MODE_BITSIZE (@var{mode}) - 1} if @code{SHIFT_COUNT_TRUNCATED}
10463 and 0 otherwise. This definition is always safe, but if
10464 @code{SHIFT_COUNT_TRUNCATED} is false, and some shift patterns
10465 nevertheless truncate the shift count, you may get better code
10469 @defmac TRULY_NOOP_TRUNCATION (@var{outprec}, @var{inprec})
10470 A C expression which is nonzero if on this machine it is safe to
10471 ``convert'' an integer of @var{inprec} bits to one of @var{outprec}
10472 bits (where @var{outprec} is smaller than @var{inprec}) by merely
10473 operating on it as if it had only @var{outprec} bits.
10475 On many machines, this expression can be 1.
10477 @c rearranged this, removed the phrase "it is reported that". this was
10478 @c to fix an overfull hbox. --mew 10feb93
10479 When @code{TRULY_NOOP_TRUNCATION} returns 1 for a pair of sizes for
10480 modes for which @code{MODES_TIEABLE_P} is 0, suboptimal code can result.
10481 If this is the case, making @code{TRULY_NOOP_TRUNCATION} return 0 in
10482 such cases may improve things.
10485 @deftypefn {Target Hook} int TARGET_MODE_REP_EXTENDED (machine_mode @var{mode}, machine_mode @var{rep_mode})
10486 The representation of an integral mode can be such that the values
10487 are always extended to a wider integral mode. Return
10488 @code{SIGN_EXTEND} if values of @var{mode} are represented in
10489 sign-extended form to @var{rep_mode}. Return @code{UNKNOWN}
10490 otherwise. (Currently, none of the targets use zero-extended
10491 representation this way so unlike @code{LOAD_EXTEND_OP},
10492 @code{TARGET_MODE_REP_EXTENDED} is expected to return either
10493 @code{SIGN_EXTEND} or @code{UNKNOWN}. Also no target extends
10494 @var{mode} to @var{rep_mode} so that @var{rep_mode} is not the next
10495 widest integral mode and currently we take advantage of this fact.)
10497 Similarly to @code{LOAD_EXTEND_OP} you may return a non-@code{UNKNOWN}
10498 value even if the extension is not performed on certain hard registers
10499 as long as for the @code{REGNO_REG_CLASS} of these hard registers
10500 @code{CANNOT_CHANGE_MODE_CLASS} returns nonzero.
10502 Note that @code{TARGET_MODE_REP_EXTENDED} and @code{LOAD_EXTEND_OP}
10503 describe two related properties. If you define
10504 @code{TARGET_MODE_REP_EXTENDED (mode, word_mode)} you probably also want
10505 to define @code{LOAD_EXTEND_OP (mode)} to return the same type of
10508 In order to enforce the representation of @code{mode},
10509 @code{TRULY_NOOP_TRUNCATION} should return false when truncating to
10513 @defmac STORE_FLAG_VALUE
10514 A C expression describing the value returned by a comparison operator
10515 with an integral mode and stored by a store-flag instruction
10516 (@samp{cstore@var{mode}4}) when the condition is true. This description must
10517 apply to @emph{all} the @samp{cstore@var{mode}4} patterns and all the
10518 comparison operators whose results have a @code{MODE_INT} mode.
10520 A value of 1 or @minus{}1 means that the instruction implementing the
10521 comparison operator returns exactly 1 or @minus{}1 when the comparison is true
10522 and 0 when the comparison is false. Otherwise, the value indicates
10523 which bits of the result are guaranteed to be 1 when the comparison is
10524 true. This value is interpreted in the mode of the comparison
10525 operation, which is given by the mode of the first operand in the
10526 @samp{cstore@var{mode}4} pattern. Either the low bit or the sign bit of
10527 @code{STORE_FLAG_VALUE} be on. Presently, only those bits are used by
10530 If @code{STORE_FLAG_VALUE} is neither 1 or @minus{}1, the compiler will
10531 generate code that depends only on the specified bits. It can also
10532 replace comparison operators with equivalent operations if they cause
10533 the required bits to be set, even if the remaining bits are undefined.
10534 For example, on a machine whose comparison operators return an
10535 @code{SImode} value and where @code{STORE_FLAG_VALUE} is defined as
10536 @samp{0x80000000}, saying that just the sign bit is relevant, the
10540 (ne:SI (and:SI @var{x} (const_int @var{power-of-2})) (const_int 0))
10544 can be converted to
10547 (ashift:SI @var{x} (const_int @var{n}))
10551 where @var{n} is the appropriate shift count to move the bit being
10552 tested into the sign bit.
10554 There is no way to describe a machine that always sets the low-order bit
10555 for a true value, but does not guarantee the value of any other bits,
10556 but we do not know of any machine that has such an instruction. If you
10557 are trying to port GCC to such a machine, include an instruction to
10558 perform a logical-and of the result with 1 in the pattern for the
10559 comparison operators and let us know at @email{gcc@@gcc.gnu.org}.
10561 Often, a machine will have multiple instructions that obtain a value
10562 from a comparison (or the condition codes). Here are rules to guide the
10563 choice of value for @code{STORE_FLAG_VALUE}, and hence the instructions
10568 Use the shortest sequence that yields a valid definition for
10569 @code{STORE_FLAG_VALUE}. It is more efficient for the compiler to
10570 ``normalize'' the value (convert it to, e.g., 1 or 0) than for the
10571 comparison operators to do so because there may be opportunities to
10572 combine the normalization with other operations.
10575 For equal-length sequences, use a value of 1 or @minus{}1, with @minus{}1 being
10576 slightly preferred on machines with expensive jumps and 1 preferred on
10580 As a second choice, choose a value of @samp{0x80000001} if instructions
10581 exist that set both the sign and low-order bits but do not define the
10585 Otherwise, use a value of @samp{0x80000000}.
10588 Many machines can produce both the value chosen for
10589 @code{STORE_FLAG_VALUE} and its negation in the same number of
10590 instructions. On those machines, you should also define a pattern for
10591 those cases, e.g., one matching
10594 (set @var{A} (neg:@var{m} (ne:@var{m} @var{B} @var{C})))
10597 Some machines can also perform @code{and} or @code{plus} operations on
10598 condition code values with less instructions than the corresponding
10599 @samp{cstore@var{mode}4} insn followed by @code{and} or @code{plus}. On those
10600 machines, define the appropriate patterns. Use the names @code{incscc}
10601 and @code{decscc}, respectively, for the patterns which perform
10602 @code{plus} or @code{minus} operations on condition code values. See
10603 @file{rs6000.md} for some examples. The GNU Superoptimizer can be used to
10604 find such instruction sequences on other machines.
10606 If this macro is not defined, the default value, 1, is used. You need
10607 not define @code{STORE_FLAG_VALUE} if the machine has no store-flag
10608 instructions, or if the value generated by these instructions is 1.
10611 @defmac FLOAT_STORE_FLAG_VALUE (@var{mode})
10612 A C expression that gives a nonzero @code{REAL_VALUE_TYPE} value that is
10613 returned when comparison operators with floating-point results are true.
10614 Define this macro on machines that have comparison operations that return
10615 floating-point values. If there are no such operations, do not define
10619 @defmac VECTOR_STORE_FLAG_VALUE (@var{mode})
10620 A C expression that gives a rtx representing the nonzero true element
10621 for vector comparisons. The returned rtx should be valid for the inner
10622 mode of @var{mode} which is guaranteed to be a vector mode. Define
10623 this macro on machines that have vector comparison operations that
10624 return a vector result. If there are no such operations, do not define
10625 this macro. Typically, this macro is defined as @code{const1_rtx} or
10626 @code{constm1_rtx}. This macro may return @code{NULL_RTX} to prevent
10627 the compiler optimizing such vector comparison operations for the
10631 @defmac CLZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value})
10632 @defmacx CTZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value})
10633 A C expression that indicates whether the architecture defines a value
10634 for @code{clz} or @code{ctz} with a zero operand.
10635 A result of @code{0} indicates the value is undefined.
10636 If the value is defined for only the RTL expression, the macro should
10637 evaluate to @code{1}; if the value applies also to the corresponding optab
10638 entry (which is normally the case if it expands directly into
10639 the corresponding RTL), then the macro should evaluate to @code{2}.
10640 In the cases where the value is defined, @var{value} should be set to
10643 If this macro is not defined, the value of @code{clz} or
10644 @code{ctz} at zero is assumed to be undefined.
10646 This macro must be defined if the target's expansion for @code{ffs}
10647 relies on a particular value to get correct results. Otherwise it
10648 is not necessary, though it may be used to optimize some corner cases, and
10649 to provide a default expansion for the @code{ffs} optab.
10651 Note that regardless of this macro the ``definedness'' of @code{clz}
10652 and @code{ctz} at zero do @emph{not} extend to the builtin functions
10653 visible to the user. Thus one may be free to adjust the value at will
10654 to match the target expansion of these operations without fear of
10659 An alias for the machine mode for pointers. On most machines, define
10660 this to be the integer mode corresponding to the width of a hardware
10661 pointer; @code{SImode} on 32-bit machine or @code{DImode} on 64-bit machines.
10662 On some machines you must define this to be one of the partial integer
10663 modes, such as @code{PSImode}.
10665 The width of @code{Pmode} must be at least as large as the value of
10666 @code{POINTER_SIZE}. If it is not equal, you must define the macro
10667 @code{POINTERS_EXTEND_UNSIGNED} to specify how pointers are extended
10671 @defmac FUNCTION_MODE
10672 An alias for the machine mode used for memory references to functions
10673 being called, in @code{call} RTL expressions. On most CISC machines,
10674 where an instruction can begin at any byte address, this should be
10675 @code{QImode}. On most RISC machines, where all instructions have fixed
10676 size and alignment, this should be a mode with the same size and alignment
10677 as the machine instruction words - typically @code{SImode} or @code{HImode}.
10680 @defmac STDC_0_IN_SYSTEM_HEADERS
10681 In normal operation, the preprocessor expands @code{__STDC__} to the
10682 constant 1, to signify that GCC conforms to ISO Standard C@. On some
10683 hosts, like Solaris, the system compiler uses a different convention,
10684 where @code{__STDC__} is normally 0, but is 1 if the user specifies
10685 strict conformance to the C Standard.
10687 Defining @code{STDC_0_IN_SYSTEM_HEADERS} makes GNU CPP follows the host
10688 convention when processing system header files, but when processing user
10689 files @code{__STDC__} will always expand to 1.
10692 @deftypefn {C Target Hook} {const char *} TARGET_C_PREINCLUDE (void)
10693 Define this hook to return the name of a header file to be included at the start of all compilations, as if it had been included with @code{#include <@var{file}>}. If this hook returns @code{NULL}, or is not defined, or the header is not found, or if the user specifies @option{-ffreestanding} or @option{-nostdinc}, no header is included.
10695 This hook can be used together with a header provided by the system C library to implement ISO C requirements for certain macros to be predefined that describe properties of the whole implementation rather than just the compiler.
10698 @deftypefn {C Target Hook} bool TARGET_CXX_IMPLICIT_EXTERN_C (const char*@var{})
10699 Define this hook to add target-specific C++ implicit extern C functions. If this function returns true for the name of a file-scope function, that function implicitly gets extern "C" linkage rather than whatever language linkage the declaration would normally have. An example of such function is WinMain on Win32 targets.
10702 @defmac NO_IMPLICIT_EXTERN_C
10703 Define this macro if the system header files support C++ as well as C@.
10704 This macro inhibits the usual method of using system header files in
10705 C++, which is to pretend that the file's contents are enclosed in
10706 @samp{extern "C" @{@dots{}@}}.
10711 @defmac REGISTER_TARGET_PRAGMAS ()
10712 Define this macro if you want to implement any target-specific pragmas.
10713 If defined, it is a C expression which makes a series of calls to
10714 @code{c_register_pragma} or @code{c_register_pragma_with_expansion}
10715 for each pragma. The macro may also do any
10716 setup required for the pragmas.
10718 The primary reason to define this macro is to provide compatibility with
10719 other compilers for the same target. In general, we discourage
10720 definition of target-specific pragmas for GCC@.
10722 If the pragma can be implemented by attributes then you should consider
10723 defining the target hook @samp{TARGET_INSERT_ATTRIBUTES} as well.
10725 Preprocessor macros that appear on pragma lines are not expanded. All
10726 @samp{#pragma} directives that do not match any registered pragma are
10727 silently ignored, unless the user specifies @option{-Wunknown-pragmas}.
10730 @deftypefun void c_register_pragma (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *))
10731 @deftypefunx void c_register_pragma_with_expansion (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *))
10733 Each call to @code{c_register_pragma} or
10734 @code{c_register_pragma_with_expansion} establishes one pragma. The
10735 @var{callback} routine will be called when the preprocessor encounters a
10739 #pragma [@var{space}] @var{name} @dots{}
10742 @var{space} is the case-sensitive namespace of the pragma, or
10743 @code{NULL} to put the pragma in the global namespace. The callback
10744 routine receives @var{pfile} as its first argument, which can be passed
10745 on to cpplib's functions if necessary. You can lex tokens after the
10746 @var{name} by calling @code{pragma_lex}. Tokens that are not read by the
10747 callback will be silently ignored. The end of the line is indicated by
10748 a token of type @code{CPP_EOF}. Macro expansion occurs on the
10749 arguments of pragmas registered with
10750 @code{c_register_pragma_with_expansion} but not on the arguments of
10751 pragmas registered with @code{c_register_pragma}.
10753 Note that the use of @code{pragma_lex} is specific to the C and C++
10754 compilers. It will not work in the Java or Fortran compilers, or any
10755 other language compilers for that matter. Thus if @code{pragma_lex} is going
10756 to be called from target-specific code, it must only be done so when
10757 building the C and C++ compilers. This can be done by defining the
10758 variables @code{c_target_objs} and @code{cxx_target_objs} in the
10759 target entry in the @file{config.gcc} file. These variables should name
10760 the target-specific, language-specific object file which contains the
10761 code that uses @code{pragma_lex}. Note it will also be necessary to add a
10762 rule to the makefile fragment pointed to by @code{tmake_file} that shows
10763 how to build this object file.
10766 @defmac HANDLE_PRAGMA_PACK_WITH_EXPANSION
10767 Define this macro if macros should be expanded in the
10768 arguments of @samp{#pragma pack}.
10771 @defmac TARGET_DEFAULT_PACK_STRUCT
10772 If your target requires a structure packing default other than 0 (meaning
10773 the machine default), define this macro to the necessary value (in bytes).
10774 This must be a value that would also be valid to use with
10775 @samp{#pragma pack()} (that is, a small power of two).
10778 @defmac DOLLARS_IN_IDENTIFIERS
10779 Define this macro to control use of the character @samp{$} in
10780 identifier names for the C family of languages. 0 means @samp{$} is
10781 not allowed by default; 1 means it is allowed. 1 is the default;
10782 there is no need to define this macro in that case.
10785 @defmac INSN_SETS_ARE_DELAYED (@var{insn})
10786 Define this macro as a C expression that is nonzero if it is safe for the
10787 delay slot scheduler to place instructions in the delay slot of @var{insn},
10788 even if they appear to use a resource set or clobbered in @var{insn}.
10789 @var{insn} is always a @code{jump_insn} or an @code{insn}; GCC knows that
10790 every @code{call_insn} has this behavior. On machines where some @code{insn}
10791 or @code{jump_insn} is really a function call and hence has this behavior,
10792 you should define this macro.
10794 You need not define this macro if it would always return zero.
10797 @defmac INSN_REFERENCES_ARE_DELAYED (@var{insn})
10798 Define this macro as a C expression that is nonzero if it is safe for the
10799 delay slot scheduler to place instructions in the delay slot of @var{insn},
10800 even if they appear to set or clobber a resource referenced in @var{insn}.
10801 @var{insn} is always a @code{jump_insn} or an @code{insn}. On machines where
10802 some @code{insn} or @code{jump_insn} is really a function call and its operands
10803 are registers whose use is actually in the subroutine it calls, you should
10804 define this macro. Doing so allows the delay slot scheduler to move
10805 instructions which copy arguments into the argument registers into the delay
10806 slot of @var{insn}.
10808 You need not define this macro if it would always return zero.
10811 @defmac MULTIPLE_SYMBOL_SPACES
10812 Define this macro as a C expression that is nonzero if, in some cases,
10813 global symbols from one translation unit may not be bound to undefined
10814 symbols in another translation unit without user intervention. For
10815 instance, under Microsoft Windows symbols must be explicitly imported
10816 from shared libraries (DLLs).
10818 You need not define this macro if it would always evaluate to zero.
10821 @deftypefn {Target Hook} tree TARGET_MD_ASM_CLOBBERS (tree @var{outputs}, tree @var{inputs}, tree @var{clobbers})
10822 This target hook should add to @var{clobbers} @code{STRING_CST} trees for
10823 any hard regs the port wishes to automatically clobber for an asm.
10824 It should return the result of the last @code{tree_cons} used to add a
10825 clobber. The @var{outputs}, @var{inputs} and @var{clobber} lists are the
10826 corresponding parameters to the asm and may be inspected to avoid
10827 clobbering a register that is an input or output of the asm. You can use
10828 @code{tree_overlaps_hard_reg_set}, declared in @file{tree.h}, to test
10829 for overlap with regards to asm-declared registers.
10832 @defmac MATH_LIBRARY
10833 Define this macro as a C string constant for the linker argument to link
10834 in the system math library, minus the initial @samp{"-l"}, or
10835 @samp{""} if the target does not have a
10836 separate math library.
10838 You need only define this macro if the default of @samp{"m"} is wrong.
10841 @defmac LIBRARY_PATH_ENV
10842 Define this macro as a C string constant for the environment variable that
10843 specifies where the linker should look for libraries.
10845 You need only define this macro if the default of @samp{"LIBRARY_PATH"}
10849 @defmac TARGET_POSIX_IO
10850 Define this macro if the target supports the following POSIX@ file
10851 functions, access, mkdir and file locking with fcntl / F_SETLKW@.
10852 Defining @code{TARGET_POSIX_IO} will enable the test coverage code
10853 to use file locking when exiting a program, which avoids race conditions
10854 if the program has forked. It will also create directories at run-time
10855 for cross-profiling.
10858 @defmac MAX_CONDITIONAL_EXECUTE
10860 A C expression for the maximum number of instructions to execute via
10861 conditional execution instructions instead of a branch. A value of
10862 @code{BRANCH_COST}+1 is the default if the machine does not use cc0, and
10863 1 if it does use cc0.
10866 @defmac IFCVT_MODIFY_TESTS (@var{ce_info}, @var{true_expr}, @var{false_expr})
10867 Used if the target needs to perform machine-dependent modifications on the
10868 conditionals used for turning basic blocks into conditionally executed code.
10869 @var{ce_info} points to a data structure, @code{struct ce_if_block}, which
10870 contains information about the currently processed blocks. @var{true_expr}
10871 and @var{false_expr} are the tests that are used for converting the
10872 then-block and the else-block, respectively. Set either @var{true_expr} or
10873 @var{false_expr} to a null pointer if the tests cannot be converted.
10876 @defmac IFCVT_MODIFY_MULTIPLE_TESTS (@var{ce_info}, @var{bb}, @var{true_expr}, @var{false_expr})
10877 Like @code{IFCVT_MODIFY_TESTS}, but used when converting more complicated
10878 if-statements into conditions combined by @code{and} and @code{or} operations.
10879 @var{bb} contains the basic block that contains the test that is currently
10880 being processed and about to be turned into a condition.
10883 @defmac IFCVT_MODIFY_INSN (@var{ce_info}, @var{pattern}, @var{insn})
10884 A C expression to modify the @var{PATTERN} of an @var{INSN} that is to
10885 be converted to conditional execution format. @var{ce_info} points to
10886 a data structure, @code{struct ce_if_block}, which contains information
10887 about the currently processed blocks.
10890 @defmac IFCVT_MODIFY_FINAL (@var{ce_info})
10891 A C expression to perform any final machine dependent modifications in
10892 converting code to conditional execution. The involved basic blocks
10893 can be found in the @code{struct ce_if_block} structure that is pointed
10894 to by @var{ce_info}.
10897 @defmac IFCVT_MODIFY_CANCEL (@var{ce_info})
10898 A C expression to cancel any machine dependent modifications in
10899 converting code to conditional execution. The involved basic blocks
10900 can be found in the @code{struct ce_if_block} structure that is pointed
10901 to by @var{ce_info}.
10904 @defmac IFCVT_MACHDEP_INIT (@var{ce_info})
10905 A C expression to initialize any machine specific data for if-conversion
10906 of the if-block in the @code{struct ce_if_block} structure that is pointed
10907 to by @var{ce_info}.
10910 @deftypefn {Target Hook} void TARGET_MACHINE_DEPENDENT_REORG (void)
10911 If non-null, this hook performs a target-specific pass over the
10912 instruction stream. The compiler will run it at all optimization levels,
10913 just before the point at which it normally does delayed-branch scheduling.
10915 The exact purpose of the hook varies from target to target. Some use
10916 it to do transformations that are necessary for correctness, such as
10917 laying out in-function constant pools or avoiding hardware hazards.
10918 Others use it as an opportunity to do some machine-dependent optimizations.
10920 You need not implement the hook if it has nothing to do. The default
10921 definition is null.
10924 @deftypefn {Target Hook} void TARGET_INIT_BUILTINS (void)
10925 Define this hook if you have any machine-specific built-in functions
10926 that need to be defined. It should be a function that performs the
10929 Machine specific built-in functions can be useful to expand special machine
10930 instructions that would otherwise not normally be generated because
10931 they have no equivalent in the source language (for example, SIMD vector
10932 instructions or prefetch instructions).
10934 To create a built-in function, call the function
10935 @code{lang_hooks.builtin_function}
10936 which is defined by the language front end. You can use any type nodes set
10937 up by @code{build_common_tree_nodes};
10938 only language front ends that use those two functions will call
10939 @samp{TARGET_INIT_BUILTINS}.
10942 @deftypefn {Target Hook} tree TARGET_BUILTIN_DECL (unsigned @var{code}, bool @var{initialize_p})
10943 Define this hook if you have any machine-specific built-in functions
10944 that need to be defined. It should be a function that returns the
10945 builtin function declaration for the builtin function code @var{code}.
10946 If there is no such builtin and it cannot be initialized at this time
10947 if @var{initialize_p} is true the function should return @code{NULL_TREE}.
10948 If @var{code} is out of range the function should return
10949 @code{error_mark_node}.
10952 @deftypefn {Target Hook} rtx TARGET_EXPAND_BUILTIN (tree @var{exp}, rtx @var{target}, rtx @var{subtarget}, machine_mode @var{mode}, int @var{ignore})
10954 Expand a call to a machine specific built-in function that was set up by
10955 @samp{TARGET_INIT_BUILTINS}. @var{exp} is the expression for the
10956 function call; the result should go to @var{target} if that is
10957 convenient, and have mode @var{mode} if that is convenient.
10958 @var{subtarget} may be used as the target for computing one of
10959 @var{exp}'s operands. @var{ignore} is nonzero if the value is to be
10960 ignored. This function should return the result of the call to the
10964 @deftypefn {Target Hook} tree TARGET_BUILTIN_CHKP_FUNCTION (unsigned @var{fcode})
10965 This hook allows target to redefine built-in functions used by
10966 Pointer Bounds Checker for code instrumentation. Hook should return
10967 fndecl of function implementing generic builtin whose code is
10968 passed in @var{fcode}. Currently following built-in functions are
10969 obtained using this hook:
10970 @deftypefn {Built-in Function} __bounds_type __chkp_bndmk (const void *@var{lb}, size_t @var{size})
10971 Function code - BUILT_IN_CHKP_BNDMK. This built-in function is used
10972 by Pointer Bounds Checker to create bound values. @var{lb} holds low
10973 bound of the resulting bounds. @var{size} holds size of created bounds.
10976 @deftypefn {Built-in Function} void __chkp_bndstx (const void *@var{ptr}, __bounds_type @var{b}, const void **@var{loc})
10977 Function code - @code{BUILT_IN_CHKP_BNDSTX}. This built-in function is used
10978 by Pointer Bounds Checker to store bounds @var{b} for pointer @var{ptr}
10979 when @var{ptr} is stored by address @var{loc}.
10982 @deftypefn {Built-in Function} __bounds_type __chkp_bndldx (const void **@var{loc}, const void *@var{ptr})
10983 Function code - @code{BUILT_IN_CHKP_BNDLDX}. This built-in function is used
10984 by Pointer Bounds Checker to get bounds of pointer @var{ptr} loaded by
10988 @deftypefn {Built-in Function} void __chkp_bndcl (const void *@var{ptr}, __bounds_type @var{b})
10989 Function code - @code{BUILT_IN_CHKP_BNDCL}. This built-in function is used
10990 by Pointer Bounds Checker to perform check for pointer @var{ptr} against
10991 lower bound of bounds @var{b}.
10994 @deftypefn {Built-in Function} void __chkp_bndcu (const void *@var{ptr}, __bounds_type @var{b})
10995 Function code - @code{BUILT_IN_CHKP_BNDCU}. This built-in function is used
10996 by Pointer Bounds Checker to perform check for pointer @var{ptr} against
10997 upper bound of bounds @var{b}.
11000 @deftypefn {Built-in Function} __bounds_type __chkp_bndret (void *@var{ptr})
11001 Function code - @code{BUILT_IN_CHKP_BNDRET}. This built-in function is used
11002 by Pointer Bounds Checker to obtain bounds returned by a call statement.
11003 @var{ptr} passed to built-in is @code{SSA_NAME} returned by the call.
11006 @deftypefn {Built-in Function} __bounds_type __chkp_intersect (__bounds_type @var{b1}, __bounds_type @var{b2})
11007 Function code - @code{BUILT_IN_CHKP_INTERSECT}. This built-in function
11008 returns intersection of bounds @var{b1} and @var{b2}.
11011 @deftypefn {Built-in Function} __bounds_type __chkp_narrow (const void *@var{ptr}, __bounds_type @var{b}, size_t @var{s})
11012 Function code - @code{BUILT_IN_CHKP_NARROW}. This built-in function
11013 returns intersection of bounds @var{b} and
11014 [@var{ptr}, @var{ptr} + @var{s} - @code{1}].
11017 @deftypefn {Built-in Function} size_t __chkp_sizeof (const void *@var{ptr})
11018 Function code - @code{BUILT_IN_CHKP_SIZEOF}. This built-in function
11019 returns size of object referenced by @var{ptr}. @var{ptr} is always
11020 @code{ADDR_EXPR} of @code{VAR_DECL}. This built-in is used by
11021 Pointer Bounds Checker when bounds of object cannot be computed statically
11022 (e.g. object has incomplete type).
11025 @deftypefn {Built-in Function} const void *__chkp_extract_lower (__bounds_type @var{b})
11026 Function code - @code{BUILT_IN_CHKP_EXTRACT_LOWER}. This built-in function
11027 returns lower bound of bounds @var{b}.
11030 @deftypefn {Built-in Function} const void *__chkp_extract_upper (__bounds_type @var{b})
11031 Function code - @code{BUILT_IN_CHKP_EXTRACT_UPPER}. This built-in function
11032 returns upper bound of bounds @var{b}.
11035 @deftypefn {Target Hook} tree TARGET_CHKP_BOUND_TYPE (void)
11036 Return type to be used for bounds
11038 @deftypefn {Target Hook} {enum machine_mode} TARGET_CHKP_BOUND_MODE (void)
11039 Return mode to be used for bounds.
11041 @deftypefn {Target Hook} tree TARGET_CHKP_MAKE_BOUNDS_CONSTANT (HOST_WIDE_INT @var{lb}, HOST_WIDE_INT @var{ub})
11042 Return constant used to statically initialize constant bounds
11043 with specified lower bound @var{lb} and upper bounds @var{ub}.
11045 @deftypefn {Target Hook} int TARGET_CHKP_INITIALIZE_BOUNDS (tree @var{var}, tree @var{lb}, tree @var{ub}, tree *@var{stmts})
11046 Generate a list of statements @var{stmts} to initialize pointer
11047 bounds variable @var{var} with bounds @var{lb} and @var{ub}. Return
11048 the number of generated statements.
11051 @deftypefn {Target Hook} tree TARGET_RESOLVE_OVERLOADED_BUILTIN (unsigned int @var{loc}, tree @var{fndecl}, void *@var{arglist})
11052 Select a replacement for a machine specific built-in function that
11053 was set up by @samp{TARGET_INIT_BUILTINS}. This is done
11054 @emph{before} regular type checking, and so allows the target to
11055 implement a crude form of function overloading. @var{fndecl} is the
11056 declaration of the built-in function. @var{arglist} is the list of
11057 arguments passed to the built-in function. The result is a
11058 complete expression that implements the operation, usually
11059 another @code{CALL_EXPR}.
11060 @var{arglist} really has type @samp{VEC(tree,gc)*}
11063 @deftypefn {Target Hook} tree TARGET_FOLD_BUILTIN (tree @var{fndecl}, int @var{n_args}, tree *@var{argp}, bool @var{ignore})
11064 Fold a call to a machine specific built-in function that was set up by
11065 @samp{TARGET_INIT_BUILTINS}. @var{fndecl} is the declaration of the
11066 built-in function. @var{n_args} is the number of arguments passed to
11067 the function; the arguments themselves are pointed to by @var{argp}.
11068 The result is another tree, valid for both GIMPLE and GENERIC,
11069 containing a simplified expression for the call's result. If
11070 @var{ignore} is true the value will be ignored.
11073 @deftypefn {Target Hook} bool TARGET_GIMPLE_FOLD_BUILTIN (gimple_stmt_iterator *@var{gsi})
11074 Fold a call to a machine specific built-in function that was set up
11075 by @samp{TARGET_INIT_BUILTINS}. @var{gsi} points to the gimple
11076 statement holding the function call. Returns true if any change
11077 was made to the GIMPLE stream.
11080 @deftypefn {Target Hook} int TARGET_COMPARE_VERSION_PRIORITY (tree @var{decl1}, tree @var{decl2})
11081 This hook is used to compare the target attributes in two functions to
11082 determine which function's features get higher priority. This is used
11083 during function multi-versioning to figure out the order in which two
11084 versions must be dispatched. A function version with a higher priority
11085 is checked for dispatching earlier. @var{decl1} and @var{decl2} are
11086 the two function decls that will be compared.
11089 @deftypefn {Target Hook} tree TARGET_GET_FUNCTION_VERSIONS_DISPATCHER (void *@var{decl})
11090 This hook is used to get the dispatcher function for a set of function
11091 versions. The dispatcher function is called to invoke the right function
11092 version at run-time. @var{decl} is one version from a set of semantically
11093 identical versions.
11096 @deftypefn {Target Hook} tree TARGET_GENERATE_VERSION_DISPATCHER_BODY (void *@var{arg})
11097 This hook is used to generate the dispatcher logic to invoke the right
11098 function version at run-time for a given set of function versions.
11099 @var{arg} points to the callgraph node of the dispatcher function whose
11100 body must be generated.
11103 @deftypefn {Target Hook} bool TARGET_CAN_USE_DOLOOP_P (const widest_int @var{&iterations}, const widest_int @var{&iterations_max}, unsigned int @var{loop_depth}, bool @var{entered_at_top})
11104 Return true if it is possible to use low-overhead loops (@code{doloop_end}
11105 and @code{doloop_begin}) for a particular loop. @var{iterations} gives the
11106 exact number of iterations, or 0 if not known. @var{iterations_max} gives
11107 the maximum number of iterations, or 0 if not known. @var{loop_depth} is
11108 the nesting depth of the loop, with 1 for innermost loops, 2 for loops that
11109 contain innermost loops, and so on. @var{entered_at_top} is true if the
11110 loop is only entered from the top.
11112 This hook is only used if @code{doloop_end} is available. The default
11113 implementation returns true. You can use @code{can_use_doloop_if_innermost}
11114 if the loop must be the innermost, and if there are no other restrictions.
11117 @deftypefn {Target Hook} {const char *} TARGET_INVALID_WITHIN_DOLOOP (const rtx_insn *@var{insn})
11119 Take an instruction in @var{insn} and return NULL if it is valid within a
11120 low-overhead loop, otherwise return a string explaining why doloop
11121 could not be applied.
11123 Many targets use special registers for low-overhead looping. For any
11124 instruction that clobbers these this function should return a string indicating
11125 the reason why the doloop could not be applied.
11126 By default, the RTL loop optimizer does not use a present doloop pattern for
11127 loops containing function calls or branch on table instructions.
11130 @deftypefn {Target Hook} bool TARGET_LEGITIMATE_COMBINED_INSN (rtx_insn *@var{insn})
11131 Take an instruction in @var{insn} and return @code{false} if the instruction is not appropriate as a combination of two or more instructions. The default is to accept all instructions.
11134 @deftypefn {Target Hook} bool TARGET_CAN_FOLLOW_JUMP (const rtx_insn *@var{follower}, const rtx_insn *@var{followee})
11135 FOLLOWER and FOLLOWEE are JUMP_INSN instructions; return true if FOLLOWER may be modified to follow FOLLOWEE; false, if it can't. For example, on some targets, certain kinds of branches can't be made to follow through a hot/cold partitioning.
11138 @deftypefn {Target Hook} bool TARGET_COMMUTATIVE_P (const_rtx @var{x}, int @var{outer_code})
11139 This target hook returns @code{true} if @var{x} is considered to be commutative.
11140 Usually, this is just COMMUTATIVE_P (@var{x}), but the HP PA doesn't consider
11141 PLUS to be commutative inside a MEM@. @var{outer_code} is the rtx code
11142 of the enclosing rtl, if known, otherwise it is UNKNOWN.
11145 @deftypefn {Target Hook} rtx TARGET_ALLOCATE_INITIAL_VALUE (rtx @var{hard_reg})
11147 When the initial value of a hard register has been copied in a pseudo
11148 register, it is often not necessary to actually allocate another register
11149 to this pseudo register, because the original hard register or a stack slot
11150 it has been saved into can be used. @code{TARGET_ALLOCATE_INITIAL_VALUE}
11151 is called at the start of register allocation once for each hard register
11152 that had its initial value copied by using
11153 @code{get_func_hard_reg_initial_val} or @code{get_hard_reg_initial_val}.
11154 Possible values are @code{NULL_RTX}, if you don't want
11155 to do any special allocation, a @code{REG} rtx---that would typically be
11156 the hard register itself, if it is known not to be clobbered---or a
11158 If you are returning a @code{MEM}, this is only a hint for the allocator;
11159 it might decide to use another register anyways.
11160 You may use @code{current_function_is_leaf} or
11161 @code{REG_N_SETS} in the hook to determine if the hard
11162 register in question will not be clobbered.
11163 The default value of this hook is @code{NULL}, which disables any special
11167 @deftypefn {Target Hook} int TARGET_UNSPEC_MAY_TRAP_P (const_rtx @var{x}, unsigned @var{flags})
11168 This target hook returns nonzero if @var{x}, an @code{unspec} or
11169 @code{unspec_volatile} operation, might cause a trap. Targets can use
11170 this hook to enhance precision of analysis for @code{unspec} and
11171 @code{unspec_volatile} operations. You may call @code{may_trap_p_1}
11172 to analyze inner elements of @var{x} in which case @var{flags} should be
11176 @deftypefn {Target Hook} void TARGET_SET_CURRENT_FUNCTION (tree @var{decl})
11177 The compiler invokes this hook whenever it changes its current function
11178 context (@code{cfun}). You can define this function if
11179 the back end needs to perform any initialization or reset actions on a
11180 per-function basis. For example, it may be used to implement function
11181 attributes that affect register usage or code generation patterns.
11182 The argument @var{decl} is the declaration for the new function context,
11183 and may be null to indicate that the compiler has left a function context
11184 and is returning to processing at the top level.
11185 The default hook function does nothing.
11187 GCC sets @code{cfun} to a dummy function context during initialization of
11188 some parts of the back end. The hook function is not invoked in this
11189 situation; you need not worry about the hook being invoked recursively,
11190 or when the back end is in a partially-initialized state.
11191 @code{cfun} might be @code{NULL} to indicate processing at top level,
11192 outside of any function scope.
11195 @defmac TARGET_OBJECT_SUFFIX
11196 Define this macro to be a C string representing the suffix for object
11197 files on your target machine. If you do not define this macro, GCC will
11198 use @samp{.o} as the suffix for object files.
11201 @defmac TARGET_EXECUTABLE_SUFFIX
11202 Define this macro to be a C string representing the suffix to be
11203 automatically added to executable files on your target machine. If you
11204 do not define this macro, GCC will use the null string as the suffix for
11208 @defmac COLLECT_EXPORT_LIST
11209 If defined, @code{collect2} will scan the individual object files
11210 specified on its command line and create an export list for the linker.
11211 Define this macro for systems like AIX, where the linker discards
11212 object files that are not referenced from @code{main} and uses export
11216 @defmac MODIFY_JNI_METHOD_CALL (@var{mdecl})
11217 Define this macro to a C expression representing a variant of the
11218 method call @var{mdecl}, if Java Native Interface (JNI) methods
11219 must be invoked differently from other methods on your target.
11220 For example, on 32-bit Microsoft Windows, JNI methods must be invoked using
11221 the @code{stdcall} calling convention and this macro is then
11222 defined as this expression:
11225 build_type_attribute_variant (@var{mdecl},
11227 (get_identifier ("stdcall"),
11232 @deftypefn {Target Hook} bool TARGET_CANNOT_MODIFY_JUMPS_P (void)
11233 This target hook returns @code{true} past the point in which new jump
11234 instructions could be created. On machines that require a register for
11235 every jump such as the SHmedia ISA of SH5, this point would typically be
11236 reload, so this target hook should be defined to a function such as:
11240 cannot_modify_jumps_past_reload_p ()
11242 return (reload_completed || reload_in_progress);
11247 @deftypefn {Target Hook} reg_class_t TARGET_BRANCH_TARGET_REGISTER_CLASS (void)
11248 This target hook returns a register class for which branch target register
11249 optimizations should be applied. All registers in this class should be
11250 usable interchangeably. After reload, registers in this class will be
11251 re-allocated and loads will be hoisted out of loops and be subjected
11252 to inter-block scheduling.
11255 @deftypefn {Target Hook} bool TARGET_BRANCH_TARGET_REGISTER_CALLEE_SAVED (bool @var{after_prologue_epilogue_gen})
11256 Branch target register optimization will by default exclude callee-saved
11258 that are not already live during the current function; if this target hook
11259 returns true, they will be included. The target code must than make sure
11260 that all target registers in the class returned by
11261 @samp{TARGET_BRANCH_TARGET_REGISTER_CLASS} that might need saving are
11262 saved. @var{after_prologue_epilogue_gen} indicates if prologues and
11263 epilogues have already been generated. Note, even if you only return
11264 true when @var{after_prologue_epilogue_gen} is false, you still are likely
11265 to have to make special provisions in @code{INITIAL_ELIMINATION_OFFSET}
11266 to reserve space for caller-saved target registers.
11269 @deftypefn {Target Hook} bool TARGET_HAVE_CONDITIONAL_EXECUTION (void)
11270 This target hook returns true if the target supports conditional execution.
11271 This target hook is required only when the target has several different
11272 modes and they have different conditional execution capability, such as ARM.
11275 @deftypefn {Target Hook} rtx TARGET_GEN_CCMP_FIRST (rtx *@var{prep_seq}, rtx *@var{gen_seq}, int @var{code}, tree @var{op0}, tree @var{op1})
11276 This function prepares to emit a comparison insn for the first compare in a
11277 sequence of conditional comparisions. It returns a appropriate @code{CC}
11278 for passing to @code{gen_ccmp_next} or @code{cbranch_optab}. The insns to
11279 prepare the compare are saved in @var{prep_seq} and the compare insns are
11280 saved in @var{gen_seq}. They will be emitted when all the compares in the
11281 the conditional comparision are generated without error. @var{code} is
11282 the @code{rtx_code} of the compare for @var{op0} and @var{op1}.
11285 @deftypefn {Target Hook} rtx TARGET_GEN_CCMP_NEXT (rtx *@var{prep_seq}, rtx *@var{gen_seq}, rtx @var{prev}, int @var{cmp_code}, tree @var{op0}, tree @var{op1}, int @var{bit_code})
11286 This function prepare to emit a conditional comparison within a sequence of
11287 conditional comparisons. It returns a appropriate @code{CC} for passing to
11288 @code{gen_ccmp_next} or @code{cbranch_optab}. The insns to prepare the
11289 compare are saved in @var{prep_seq} and the compare insns are saved in
11290 @var{gen_seq}. They will be emitted when all the compares in the conditional
11291 comparision are generated without error. The @var{prev} expression is the
11292 result of a prior call to @code{gen_ccmp_first} or @code{gen_ccmp_next}. It
11293 may return @code{NULL} if the combination of @var{prev} and this comparison is
11294 not supported, otherwise the result must be appropriate for passing to
11295 @code{gen_ccmp_next} or @code{cbranch_optab}. @var{code} is the
11296 @code{rtx_code} of the compare for @var{op0} and @var{op1}. @var{bit_code}
11297 is @code{AND} or @code{IOR}, which is the op on the two compares.
11300 @deftypefn {Target Hook} unsigned TARGET_LOOP_UNROLL_ADJUST (unsigned @var{nunroll}, struct loop *@var{loop})
11301 This target hook returns a new value for the number of times @var{loop}
11302 should be unrolled. The parameter @var{nunroll} is the number of times
11303 the loop is to be unrolled. The parameter @var{loop} is a pointer to
11304 the loop, which is going to be checked for unrolling. This target hook
11305 is required only when the target has special constraints like maximum
11306 number of memory accesses.
11309 @defmac POWI_MAX_MULTS
11310 If defined, this macro is interpreted as a signed integer C expression
11311 that specifies the maximum number of floating point multiplications
11312 that should be emitted when expanding exponentiation by an integer
11313 constant inline. When this value is defined, exponentiation requiring
11314 more than this number of multiplications is implemented by calling the
11315 system library's @code{pow}, @code{powf} or @code{powl} routines.
11316 The default value places no upper bound on the multiplication count.
11319 @deftypefn Macro void TARGET_EXTRA_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc})
11320 This target hook should register any extra include files for the
11321 target. The parameter @var{stdinc} indicates if normal include files
11322 are present. The parameter @var{sysroot} is the system root directory.
11323 The parameter @var{iprefix} is the prefix for the gcc directory.
11326 @deftypefn Macro void TARGET_EXTRA_PRE_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc})
11327 This target hook should register any extra include files for the
11328 target before any standard headers. The parameter @var{stdinc}
11329 indicates if normal include files are present. The parameter
11330 @var{sysroot} is the system root directory. The parameter
11331 @var{iprefix} is the prefix for the gcc directory.
11334 @deftypefn Macro void TARGET_OPTF (char *@var{path})
11335 This target hook should register special include paths for the target.
11336 The parameter @var{path} is the include to register. On Darwin
11337 systems, this is used for Framework includes, which have semantics
11338 that are different from @option{-I}.
11341 @defmac bool TARGET_USE_LOCAL_THUNK_ALIAS_P (tree @var{fndecl})
11342 This target macro returns @code{true} if it is safe to use a local alias
11343 for a virtual function @var{fndecl} when constructing thunks,
11344 @code{false} otherwise. By default, the macro returns @code{true} for all
11345 functions, if a target supports aliases (i.e.@: defines
11346 @code{ASM_OUTPUT_DEF}), @code{false} otherwise,
11349 @defmac TARGET_FORMAT_TYPES
11350 If defined, this macro is the name of a global variable containing
11351 target-specific format checking information for the @option{-Wformat}
11352 option. The default is to have no target-specific format checks.
11355 @defmac TARGET_N_FORMAT_TYPES
11356 If defined, this macro is the number of entries in
11357 @code{TARGET_FORMAT_TYPES}.
11360 @defmac TARGET_OVERRIDES_FORMAT_ATTRIBUTES
11361 If defined, this macro is the name of a global variable containing
11362 target-specific format overrides for the @option{-Wformat} option. The
11363 default is to have no target-specific format overrides. If defined,
11364 @code{TARGET_FORMAT_TYPES} must be defined, too.
11367 @defmac TARGET_OVERRIDES_FORMAT_ATTRIBUTES_COUNT
11368 If defined, this macro specifies the number of entries in
11369 @code{TARGET_OVERRIDES_FORMAT_ATTRIBUTES}.
11372 @defmac TARGET_OVERRIDES_FORMAT_INIT
11373 If defined, this macro specifies the optional initialization
11374 routine for target specific customizations of the system printf
11375 and scanf formatter settings.
11378 @deftypevr {Target Hook} bool TARGET_RELAXED_ORDERING
11379 If set to @code{true}, means that the target's memory model does not
11380 guarantee that loads which do not depend on one another will access
11381 main memory in the order of the instruction stream; if ordering is
11382 important, an explicit memory barrier must be used. This is true of
11383 many recent processors which implement a policy of ``relaxed,''
11384 ``weak,'' or ``release'' memory consistency, such as Alpha, PowerPC,
11385 and ia64. The default is @code{false}.
11388 @deftypefn {Target Hook} {const char *} TARGET_INVALID_ARG_FOR_UNPROTOTYPED_FN (const_tree @var{typelist}, const_tree @var{funcdecl}, const_tree @var{val})
11389 If defined, this macro returns the diagnostic message when it is
11390 illegal to pass argument @var{val} to function @var{funcdecl}
11391 with prototype @var{typelist}.
11394 @deftypefn {Target Hook} {const char *} TARGET_INVALID_CONVERSION (const_tree @var{fromtype}, const_tree @var{totype})
11395 If defined, this macro returns the diagnostic message when it is
11396 invalid to convert from @var{fromtype} to @var{totype}, or @code{NULL}
11397 if validity should be determined by the front end.
11400 @deftypefn {Target Hook} {const char *} TARGET_INVALID_UNARY_OP (int @var{op}, const_tree @var{type})
11401 If defined, this macro returns the diagnostic message when it is
11402 invalid to apply operation @var{op} (where unary plus is denoted by
11403 @code{CONVERT_EXPR}) to an operand of type @var{type}, or @code{NULL}
11404 if validity should be determined by the front end.
11407 @deftypefn {Target Hook} {const char *} TARGET_INVALID_BINARY_OP (int @var{op}, const_tree @var{type1}, const_tree @var{type2})
11408 If defined, this macro returns the diagnostic message when it is
11409 invalid to apply operation @var{op} to operands of types @var{type1}
11410 and @var{type2}, or @code{NULL} if validity should be determined by
11414 @deftypefn {Target Hook} {const char *} TARGET_INVALID_PARAMETER_TYPE (const_tree @var{type})
11415 If defined, this macro returns the diagnostic message when it is
11416 invalid for functions to include parameters of type @var{type},
11417 or @code{NULL} if validity should be determined by
11418 the front end. This is currently used only by the C and C++ front ends.
11421 @deftypefn {Target Hook} {const char *} TARGET_INVALID_RETURN_TYPE (const_tree @var{type})
11422 If defined, this macro returns the diagnostic message when it is
11423 invalid for functions to have return type @var{type},
11424 or @code{NULL} if validity should be determined by
11425 the front end. This is currently used only by the C and C++ front ends.
11428 @deftypefn {Target Hook} tree TARGET_PROMOTED_TYPE (const_tree @var{type})
11429 If defined, this target hook returns the type to which values of
11430 @var{type} should be promoted when they appear in expressions,
11431 analogous to the integer promotions, or @code{NULL_TREE} to use the
11432 front end's normal promotion rules. This hook is useful when there are
11433 target-specific types with special promotion rules.
11434 This is currently used only by the C and C++ front ends.
11437 @deftypefn {Target Hook} tree TARGET_CONVERT_TO_TYPE (tree @var{type}, tree @var{expr})
11438 If defined, this hook returns the result of converting @var{expr} to
11439 @var{type}. It should return the converted expression,
11440 or @code{NULL_TREE} to apply the front end's normal conversion rules.
11441 This hook is useful when there are target-specific types with special
11443 This is currently used only by the C and C++ front ends.
11446 @defmac TARGET_USE_JCR_SECTION
11447 This macro determines whether to use the JCR section to register Java
11448 classes. By default, TARGET_USE_JCR_SECTION is defined to 1 if both
11449 SUPPORTS_WEAK and TARGET_HAVE_NAMED_SECTIONS are true, else 0.
11453 This macro determines the size of the objective C jump buffer for the
11454 NeXT runtime. By default, OBJC_JBLEN is defined to an innocuous value.
11457 @defmac LIBGCC2_UNWIND_ATTRIBUTE
11458 Define this macro if any target-specific attributes need to be attached
11459 to the functions in @file{libgcc} that provide low-level support for
11460 call stack unwinding. It is used in declarations in @file{unwind-generic.h}
11461 and the associated definitions of those functions.
11464 @deftypefn {Target Hook} void TARGET_UPDATE_STACK_BOUNDARY (void)
11465 Define this macro to update the current function stack boundary if
11469 @deftypefn {Target Hook} rtx TARGET_GET_DRAP_RTX (void)
11470 This hook should return an rtx for Dynamic Realign Argument Pointer (DRAP) if a
11471 different argument pointer register is needed to access the function's
11472 argument list due to stack realignment. Return @code{NULL} if no DRAP
11476 @deftypefn {Target Hook} bool TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS (void)
11477 When optimization is disabled, this hook indicates whether or not
11478 arguments should be allocated to stack slots. Normally, GCC allocates
11479 stacks slots for arguments when not optimizing in order to make
11480 debugging easier. However, when a function is declared with
11481 @code{__attribute__((naked))}, there is no stack frame, and the compiler
11482 cannot safely move arguments from the registers in which they are passed
11483 to the stack. Therefore, this hook should return true in general, but
11484 false for naked functions. The default implementation always returns true.
11487 @deftypevr {Target Hook} {unsigned HOST_WIDE_INT} TARGET_CONST_ANCHOR
11488 On some architectures it can take multiple instructions to synthesize
11489 a constant. If there is another constant already in a register that
11490 is close enough in value then it is preferable that the new constant
11491 is computed from this register using immediate addition or
11492 subtraction. We accomplish this through CSE. Besides the value of
11493 the constant we also add a lower and an upper constant anchor to the
11494 available expressions. These are then queried when encountering new
11495 constants. The anchors are computed by rounding the constant up and
11496 down to a multiple of the value of @code{TARGET_CONST_ANCHOR}.
11497 @code{TARGET_CONST_ANCHOR} should be the maximum positive value
11498 accepted by immediate-add plus one. We currently assume that the
11499 value of @code{TARGET_CONST_ANCHOR} is a power of 2. For example, on
11500 MIPS, where add-immediate takes a 16-bit signed value,
11501 @code{TARGET_CONST_ANCHOR} is set to @samp{0x8000}. The default value
11502 is zero, which disables this optimization.
11505 @deftypefn {Target Hook} {unsigned HOST_WIDE_INT} TARGET_ASAN_SHADOW_OFFSET (void)
11506 Return the offset bitwise ored into shifted address to get corresponding
11507 Address Sanitizer shadow memory address. NULL if Address Sanitizer is not
11508 supported by the target.
11511 @deftypefn {Target Hook} {unsigned HOST_WIDE_INT} TARGET_MEMMODEL_CHECK (unsigned HOST_WIDE_INT @var{val})
11512 Validate target specific memory model mask bits. When NULL no target specific
11513 memory model bits are allowed.
11516 @deftypevr {Target Hook} {unsigned char} TARGET_ATOMIC_TEST_AND_SET_TRUEVAL
11517 This value should be set if the result written by @code{atomic_test_and_set} is not exactly 1, i.e. the @code{bool} @code{true}.
11520 @deftypefn {Target Hook} bool TARGET_HAS_IFUNC_P (void)
11521 It returns true if the target supports GNU indirect functions.
11522 The support includes the assembler, linker and dynamic linker.
11523 The default value of this hook is based on target's libc.
11526 @deftypefn {Target Hook} {unsigned int} TARGET_ATOMIC_ALIGN_FOR_MODE (machine_mode @var{mode})
11527 If defined, this function returns an appropriate alignment in bits for an atomic object of machine_mode @var{mode}. If 0 is returned then the default alignment for the specified mode is used.
11530 @deftypefn {Target Hook} void TARGET_ATOMIC_ASSIGN_EXPAND_FENV (tree *@var{hold}, tree *@var{clear}, tree *@var{update})
11531 ISO C11 requires atomic compound assignments that may raise floating-point exceptions to raise exceptions corresponding to the arithmetic operation whose result was successfully stored in a compare-and-exchange sequence. This requires code equivalent to calls to @code{feholdexcept}, @code{feclearexcept} and @code{feupdateenv} to be generated at appropriate points in the compare-and-exchange sequence. This hook should set @code{*@var{hold}} to an expression equivalent to the call to @code{feholdexcept}, @code{*@var{clear}} to an expression equivalent to the call to @code{feclearexcept} and @code{*@var{update}} to an expression equivalent to the call to @code{feupdateenv}. The three expressions are @code{NULL_TREE} on entry to the hook and may be left as @code{NULL_TREE} if no code is required in a particular place. The default implementation leaves all three expressions as @code{NULL_TREE}. The @code{__atomic_feraiseexcept} function from @code{libatomic} may be of use as part of the code generated in @code{*@var{update}}.
11534 @deftypefn {Target Hook} void TARGET_RECORD_OFFLOAD_SYMBOL (tree)
11535 Used when offloaded functions are seen in the compilation unit and no named
11536 sections are available. It is called once for each symbol that must be
11537 recorded in the offload function and variable table.
11540 @deftypefn {Target Hook} {char *} TARGET_OFFLOAD_OPTIONS (void)
11541 Used when writing out the list of options into an LTO file. It should
11542 translate any relevant target-specific options (such as the ABI in use)
11543 into one of the @option{-foffload} options that exist as a common interface
11544 to express such options. It should return a string containing these options,
11545 separated by spaces, which the caller will free.
11549 @defmac TARGET_SUPPORTS_WIDE_INT
11551 On older ports, large integers are stored in @code{CONST_DOUBLE} rtl
11552 objects. Newer ports define @code{TARGET_SUPPORTS_WIDE_INT} to be nonzero
11553 to indicate that large integers are stored in
11554 @code{CONST_WIDE_INT} rtl objects. The @code{CONST_WIDE_INT} allows
11555 very large integer constants to be represented. @code{CONST_DOUBLE}
11556 is limited to twice the size of the host's @code{HOST_WIDE_INT}
11559 Converting a port mostly requires looking for the places where
11560 @code{CONST_DOUBLE}s are used with @code{VOIDmode} and replacing that
11561 code with code that accesses @code{CONST_WIDE_INT}s. @samp{"grep -i
11562 const_double"} at the port level gets you to 95% of the changes that
11563 need to be made. There are a few places that require a deeper look.
11567 There is no equivalent to @code{hval} and @code{lval} for
11568 @code{CONST_WIDE_INT}s. This would be difficult to express in the md
11569 language since there are a variable number of elements.
11571 Most ports only check that @code{hval} is either 0 or -1 to see if the
11572 value is small. As mentioned above, this will no longer be necessary
11573 since small constants are always @code{CONST_INT}. Of course there
11574 are still a few exceptions, the alpha's constraint used by the zap
11575 instruction certainly requires careful examination by C code.
11576 However, all the current code does is pass the hval and lval to C
11577 code, so evolving the c code to look at the @code{CONST_WIDE_INT} is
11578 not really a large change.
11581 Because there is no standard template that ports use to materialize
11582 constants, there is likely to be some futzing that is unique to each
11586 The rtx costs may have to be adjusted to properly account for larger
11587 constants that are represented as @code{CONST_WIDE_INT}.
11590 All and all it does not take long to convert ports that the
11591 maintainer is familiar with.