1@c Copyright (C) 1988,1989,1992,1993,1994,1995,1996,1997,1998,1999,2000,2001, 2@c 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011, 2012 3@c Free Software Foundation, Inc. 4@c This is part of the GCC manual. 5@c For copying conditions, see the file gcc.texi. 6 7@node Target Macros 8@chapter Target Description Macros and Functions 9@cindex machine description macros 10@cindex target description macros 11@cindex macros, target description 12@cindex @file{tm.h} macros 13 14In addition to the file @file{@var{machine}.md}, a machine description 15includes a C header file conventionally given the name 16@file{@var{machine}.h} and a C source file named @file{@var{machine}.c}. 17The header file defines numerous macros that convey the information 18about the target machine that does not fit into the scheme of the 19@file{.md} file. The file @file{tm.h} should be a link to 20@file{@var{machine}.h}. The header file @file{config.h} includes 21@file{tm.h} and most compiler source files include @file{config.h}. The 22source file defines a variable @code{targetm}, which is a structure 23containing pointers to functions and data relating to the target 24machine. @file{@var{machine}.c} should also contain their definitions, 25if they are not defined elsewhere in GCC, and other functions called 26through the macros defined in the @file{.h} file. 27 28@menu 29* Target Structure:: The @code{targetm} variable. 30* Driver:: Controlling how the driver runs the compilation passes. 31* Run-time Target:: Defining @samp{-m} options like @option{-m68000} and @option{-m68020}. 32* Per-Function Data:: Defining data structures for per-function information. 33* Storage Layout:: Defining sizes and alignments of data. 34* Type Layout:: Defining sizes and properties of basic user data types. 35* Registers:: Naming and describing the hardware registers. 36* Register Classes:: Defining the classes of hardware registers. 37* Old Constraints:: The old way to define machine-specific constraints. 38* Stack and Calling:: Defining which way the stack grows and by how much. 39* Varargs:: Defining the varargs macros. 40* Trampolines:: Code set up at run time to enter a nested function. 41* Library Calls:: Controlling how library routines are implicitly called. 42* Addressing Modes:: Defining addressing modes valid for memory operands. 43* Anchored Addresses:: Defining how @option{-fsection-anchors} should work. 44* Condition Code:: Defining how insns update the condition code. 45* Costs:: Defining relative costs of different operations. 46* Scheduling:: Adjusting the behavior of the instruction scheduler. 47* Sections:: Dividing storage into text, data, and other sections. 48* PIC:: Macros for position independent code. 49* Assembler Format:: Defining how to write insns and pseudo-ops to output. 50* Debugging Info:: Defining the format of debugging output. 51* Floating Point:: Handling floating point for cross-compilers. 52* Mode Switching:: Insertion of mode-switching instructions. 53* Target Attributes:: Defining target-specific uses of @code{__attribute__}. 54* Emulated TLS:: Emulated TLS support. 55* MIPS Coprocessors:: MIPS coprocessor support and how to customize it. 56* PCH Target:: Validity checking for precompiled headers. 57* C++ ABI:: Controlling C++ ABI changes. 58* Named Address Spaces:: Adding support for named address spaces 59* Misc:: Everything else. 60@end menu 61 62@node Target Structure 63@section The Global @code{targetm} Variable 64@cindex target hooks 65@cindex target functions 66 67@deftypevar {struct gcc_target} targetm 68The target @file{.c} file must define the global @code{targetm} variable 69which contains pointers to functions and data relating to the target 70machine. The variable is declared in @file{target.h}; 71@file{target-def.h} defines the macro @code{TARGET_INITIALIZER} which is 72used to initialize the variable, and macros for the default initializers 73for elements of the structure. The @file{.c} file should override those 74macros for which the default definition is inappropriate. For example: 75@smallexample 76#include "target.h" 77#include "target-def.h" 78 79/* @r{Initialize the GCC target structure.} */ 80 81#undef TARGET_COMP_TYPE_ATTRIBUTES 82#define TARGET_COMP_TYPE_ATTRIBUTES @var{machine}_comp_type_attributes 83 84struct gcc_target targetm = TARGET_INITIALIZER; 85@end smallexample 86@end deftypevar 87 88Where a macro should be defined in the @file{.c} file in this manner to 89form part of the @code{targetm} structure, it is documented below as a 90``Target Hook'' with a prototype. Many macros will change in future 91from being defined in the @file{.h} file to being part of the 92@code{targetm} structure. 93 94Similarly, there is a @code{targetcm} variable for hooks that are 95specific to front ends for C-family languages, documented as ``C 96Target Hook''. This is declared in @file{c-family/c-target.h}, the 97initializer @code{TARGETCM_INITIALIZER} in 98@file{c-family/c-target-def.h}. If targets initialize @code{targetcm} 99themselves, they should set @code{target_has_targetcm=yes} in 100@file{config.gcc}; otherwise a default definition is used. 101 102Similarly, there is a @code{targetm_common} variable for hooks that 103are shared between the compiler driver and the compilers proper, 104documented as ``Common Target Hook''. This is declared in 105@file{common/common-target.h}, the initializer 106@code{TARGETM_COMMON_INITIALIZER} in 107@file{common/common-target-def.h}. If targets initialize 108@code{targetm_common} themselves, they should set 109@code{target_has_targetm_common=yes} in @file{config.gcc}; otherwise a 110default definition is used. 111 112@node Driver 113@section Controlling the Compilation Driver, @file{gcc} 114@cindex driver 115@cindex controlling the compilation driver 116 117@c prevent bad page break with this line 118You can control the compilation driver. 119 120@defmac DRIVER_SELF_SPECS 121A list of specs for the driver itself. It should be a suitable 122initializer for an array of strings, with no surrounding braces. 123 124The driver applies these specs to its own command line between loading 125default @file{specs} files (but not command-line specified ones) and 126choosing the multilib directory or running any subcommands. It 127applies them in the order given, so each spec can depend on the 128options added by earlier ones. It is also possible to remove options 129using @samp{%<@var{option}} in the usual way. 130 131This macro can be useful when a port has several interdependent target 132options. It provides a way of standardizing the command line so 133that the other specs are easier to write. 134 135Do not define this macro if it does not need to do anything. 136@end defmac 137 138@defmac OPTION_DEFAULT_SPECS 139A list of specs used to support configure-time default options (i.e.@: 140@option{--with} options) in the driver. It should be a suitable initializer 141for an array of structures, each containing two strings, without the 142outermost pair of surrounding braces. 143 144The first item in the pair is the name of the default. This must match 145the code in @file{config.gcc} for the target. The second item is a spec 146to apply if a default with this name was specified. The string 147@samp{%(VALUE)} in the spec will be replaced by the value of the default 148everywhere it occurs. 149 150The driver will apply these specs to its own command line between loading 151default @file{specs} files and processing @code{DRIVER_SELF_SPECS}, using 152the same mechanism as @code{DRIVER_SELF_SPECS}. 153 154Do not define this macro if it does not need to do anything. 155@end defmac 156 157@defmac CPP_SPEC 158A C string constant that tells the GCC driver program options to 159pass to CPP@. It can also specify how to translate options you 160give to GCC into options for GCC to pass to the CPP@. 161 162Do not define this macro if it does not need to do anything. 163@end defmac 164 165@defmac CPLUSPLUS_CPP_SPEC 166This macro is just like @code{CPP_SPEC}, but is used for C++, rather 167than C@. If you do not define this macro, then the value of 168@code{CPP_SPEC} (if any) will be used instead. 169@end defmac 170 171@defmac CC1_SPEC 172A C string constant that tells the GCC driver program options to 173pass to @code{cc1}, @code{cc1plus}, @code{f771}, and the other language 174front ends. 175It can also specify how to translate options you give to GCC into options 176for GCC to pass to front ends. 177 178Do not define this macro if it does not need to do anything. 179@end defmac 180 181@defmac CC1PLUS_SPEC 182A C string constant that tells the GCC driver program options to 183pass to @code{cc1plus}. It can also specify how to translate options you 184give to GCC into options for GCC to pass to the @code{cc1plus}. 185 186Do not define this macro if it does not need to do anything. 187Note that everything defined in CC1_SPEC is already passed to 188@code{cc1plus} so there is no need to duplicate the contents of 189CC1_SPEC in CC1PLUS_SPEC@. 190@end defmac 191 192@defmac ASM_SPEC 193A C string constant that tells the GCC driver program options to 194pass to the assembler. It can also specify how to translate options 195you give to GCC into options for GCC to pass to the assembler. 196See the file @file{sun3.h} for an example of this. 197 198Do not define this macro if it does not need to do anything. 199@end defmac 200 201@defmac ASM_FINAL_SPEC 202A C string constant that tells the GCC driver program how to 203run any programs which cleanup after the normal assembler. 204Normally, this is not needed. See the file @file{mips.h} for 205an example of this. 206 207Do not define this macro if it does not need to do anything. 208@end defmac 209 210@defmac AS_NEEDS_DASH_FOR_PIPED_INPUT 211Define this macro, with no value, if the driver should give the assembler 212an argument consisting of a single dash, @option{-}, to instruct it to 213read from its standard input (which will be a pipe connected to the 214output of the compiler proper). This argument is given after any 215@option{-o} option specifying the name of the output file. 216 217If you do not define this macro, the assembler is assumed to read its 218standard input if given no non-option arguments. If your assembler 219cannot read standard input at all, use a @samp{%@{pipe:%e@}} construct; 220see @file{mips.h} for instance. 221@end defmac 222 223@defmac LINK_SPEC 224A C string constant that tells the GCC driver program options to 225pass to the linker. It can also specify how to translate options you 226give to GCC into options for GCC to pass to the linker. 227 228Do not define this macro if it does not need to do anything. 229@end defmac 230 231@defmac LIB_SPEC 232Another C string constant used much like @code{LINK_SPEC}. The difference 233between the two is that @code{LIB_SPEC} is used at the end of the 234command given to the linker. 235 236If this macro is not defined, a default is provided that 237loads the standard C library from the usual place. See @file{gcc.c}. 238@end defmac 239 240@defmac LIBGCC_SPEC 241Another C string constant that tells the GCC driver program 242how and when to place a reference to @file{libgcc.a} into the 243linker command line. This constant is placed both before and after 244the value of @code{LIB_SPEC}. 245 246If this macro is not defined, the GCC driver provides a default that 247passes the string @option{-lgcc} to the linker. 248@end defmac 249 250@defmac REAL_LIBGCC_SPEC 251By default, if @code{ENABLE_SHARED_LIBGCC} is defined, the 252@code{LIBGCC_SPEC} is not directly used by the driver program but is 253instead modified to refer to different versions of @file{libgcc.a} 254depending on the values of the command line flags @option{-static}, 255@option{-shared}, @option{-static-libgcc}, and @option{-shared-libgcc}. On 256targets where these modifications are inappropriate, define 257@code{REAL_LIBGCC_SPEC} instead. @code{REAL_LIBGCC_SPEC} tells the 258driver how to place a reference to @file{libgcc} on the link command 259line, but, unlike @code{LIBGCC_SPEC}, it is used unmodified. 260@end defmac 261 262@defmac USE_LD_AS_NEEDED 263A macro that controls the modifications to @code{LIBGCC_SPEC} 264mentioned in @code{REAL_LIBGCC_SPEC}. If nonzero, a spec will be 265generated that uses --as-needed and the shared libgcc in place of the 266static exception handler library, when linking without any of 267@code{-static}, @code{-static-libgcc}, or @code{-shared-libgcc}. 268@end defmac 269 270@defmac LINK_EH_SPEC 271If defined, this C string constant is added to @code{LINK_SPEC}. 272When @code{USE_LD_AS_NEEDED} is zero or undefined, it also affects 273the modifications to @code{LIBGCC_SPEC} mentioned in 274@code{REAL_LIBGCC_SPEC}. 275@end defmac 276 277@defmac STARTFILE_SPEC 278Another C string constant used much like @code{LINK_SPEC}. The 279difference between the two is that @code{STARTFILE_SPEC} is used at 280the very beginning of the command given to the linker. 281 282If this macro is not defined, a default is provided that loads the 283standard C startup file from the usual place. See @file{gcc.c}. 284@end defmac 285 286@defmac ENDFILE_SPEC 287Another C string constant used much like @code{LINK_SPEC}. The 288difference between the two is that @code{ENDFILE_SPEC} is used at 289the very end of the command given to the linker. 290 291Do not define this macro if it does not need to do anything. 292@end defmac 293 294@defmac THREAD_MODEL_SPEC 295GCC @code{-v} will print the thread model GCC was configured to use. 296However, this doesn't work on platforms that are multilibbed on thread 297models, such as AIX 4.3. On such platforms, define 298@code{THREAD_MODEL_SPEC} such that it evaluates to a string without 299blanks that names one of the recognized thread models. @code{%*}, the 300default value of this macro, will expand to the value of 301@code{thread_file} set in @file{config.gcc}. 302@end defmac 303 304@defmac SYSROOT_SUFFIX_SPEC 305Define this macro to add a suffix to the target sysroot when GCC is 306configured with a sysroot. This will cause GCC to search for usr/lib, 307et al, within sysroot+suffix. 308@end defmac 309 310@defmac SYSROOT_HEADERS_SUFFIX_SPEC 311Define this macro to add a headers_suffix to the target sysroot when 312GCC is configured with a sysroot. This will cause GCC to pass the 313updated sysroot+headers_suffix to CPP, causing it to search for 314usr/include, et al, within sysroot+headers_suffix. 315@end defmac 316 317@defmac EXTRA_SPECS 318Define this macro to provide additional specifications to put in the 319@file{specs} file that can be used in various specifications like 320@code{CC1_SPEC}. 321 322The definition should be an initializer for an array of structures, 323containing a string constant, that defines the specification name, and a 324string constant that provides the specification. 325 326Do not define this macro if it does not need to do anything. 327 328@code{EXTRA_SPECS} is useful when an architecture contains several 329related targets, which have various @code{@dots{}_SPECS} which are similar 330to each other, and the maintainer would like one central place to keep 331these definitions. 332 333For example, the PowerPC System V.4 targets use @code{EXTRA_SPECS} to 334define either @code{_CALL_SYSV} when the System V calling sequence is 335used or @code{_CALL_AIX} when the older AIX-based calling sequence is 336used. 337 338The @file{config/rs6000/rs6000.h} target file defines: 339 340@smallexample 341#define EXTRA_SPECS \ 342 @{ "cpp_sysv_default", CPP_SYSV_DEFAULT @}, 343 344#define CPP_SYS_DEFAULT "" 345@end smallexample 346 347The @file{config/rs6000/sysv.h} target file defines: 348@smallexample 349#undef CPP_SPEC 350#define CPP_SPEC \ 351"%@{posix: -D_POSIX_SOURCE @} \ 352%@{mcall-sysv: -D_CALL_SYSV @} \ 353%@{!mcall-sysv: %(cpp_sysv_default) @} \ 354%@{msoft-float: -D_SOFT_FLOAT@} %@{mcpu=403: -D_SOFT_FLOAT@}" 355 356#undef CPP_SYSV_DEFAULT 357#define CPP_SYSV_DEFAULT "-D_CALL_SYSV" 358@end smallexample 359 360while the @file{config/rs6000/eabiaix.h} target file defines 361@code{CPP_SYSV_DEFAULT} as: 362 363@smallexample 364#undef CPP_SYSV_DEFAULT 365#define CPP_SYSV_DEFAULT "-D_CALL_AIX" 366@end smallexample 367@end defmac 368 369@defmac LINK_LIBGCC_SPECIAL_1 370Define this macro if the driver program should find the library 371@file{libgcc.a}. If you do not define this macro, the driver program will pass 372the argument @option{-lgcc} to tell the linker to do the search. 373@end defmac 374 375@defmac LINK_GCC_C_SEQUENCE_SPEC 376The sequence in which libgcc and libc are specified to the linker. 377By default this is @code{%G %L %G}. 378@end defmac 379 380@defmac LINK_COMMAND_SPEC 381A C string constant giving the complete command line need to execute the 382linker. When you do this, you will need to update your port each time a 383change is made to the link command line within @file{gcc.c}. Therefore, 384define this macro only if you need to completely redefine the command 385line for invoking the linker and there is no other way to accomplish 386the effect you need. Overriding this macro may be avoidable by overriding 387@code{LINK_GCC_C_SEQUENCE_SPEC} instead. 388@end defmac 389 390@defmac LINK_ELIMINATE_DUPLICATE_LDIRECTORIES 391A nonzero value causes @command{collect2} to remove duplicate @option{-L@var{directory}} search 392directories from linking commands. Do not give it a nonzero value if 393removing duplicate search directories changes the linker's semantics. 394@end defmac 395 396@hook TARGET_ALWAYS_STRIP_DOTDOT 397 398@defmac MULTILIB_DEFAULTS 399Define this macro as a C expression for the initializer of an array of 400string to tell the driver program which options are defaults for this 401target and thus do not need to be handled specially when using 402@code{MULTILIB_OPTIONS}. 403 404Do not define this macro if @code{MULTILIB_OPTIONS} is not defined in 405the target makefile fragment or if none of the options listed in 406@code{MULTILIB_OPTIONS} are set by default. 407@xref{Target Fragment}. 408@end defmac 409 410@defmac RELATIVE_PREFIX_NOT_LINKDIR 411Define this macro to tell @command{gcc} that it should only translate 412a @option{-B} prefix into a @option{-L} linker option if the prefix 413indicates an absolute file name. 414@end defmac 415 416@defmac MD_EXEC_PREFIX 417If defined, this macro is an additional prefix to try after 418@code{STANDARD_EXEC_PREFIX}. @code{MD_EXEC_PREFIX} is not searched 419when the compiler is built as a cross 420compiler. If you define @code{MD_EXEC_PREFIX}, then be sure to add it 421to the list of directories used to find the assembler in @file{configure.in}. 422@end defmac 423 424@defmac STANDARD_STARTFILE_PREFIX 425Define this macro as a C string constant if you wish to override the 426standard choice of @code{libdir} as the default prefix to 427try when searching for startup files such as @file{crt0.o}. 428@code{STANDARD_STARTFILE_PREFIX} is not searched when the compiler 429is built as a cross compiler. 430@end defmac 431 432@defmac STANDARD_STARTFILE_PREFIX_1 433Define this macro as a C string constant if you wish to override the 434standard choice of @code{/lib} as a prefix to try after the default prefix 435when searching for startup files such as @file{crt0.o}. 436@code{STANDARD_STARTFILE_PREFIX_1} is not searched when the compiler 437is built as a cross compiler. 438@end defmac 439 440@defmac STANDARD_STARTFILE_PREFIX_2 441Define this macro as a C string constant if you wish to override the 442standard choice of @code{/lib} as yet another prefix to try after the 443default prefix when searching for startup files such as @file{crt0.o}. 444@code{STANDARD_STARTFILE_PREFIX_2} is not searched when the compiler 445is built as a cross compiler. 446@end defmac 447 448@defmac MD_STARTFILE_PREFIX 449If defined, this macro supplies an additional prefix to try after the 450standard prefixes. @code{MD_EXEC_PREFIX} is not searched when the 451compiler is built as a cross compiler. 452@end defmac 453 454@defmac MD_STARTFILE_PREFIX_1 455If defined, this macro supplies yet another prefix to try after the 456standard prefixes. It is not searched when the compiler is built as a 457cross compiler. 458@end defmac 459 460@defmac INIT_ENVIRONMENT 461Define this macro as a C string constant if you wish to set environment 462variables for programs called by the driver, such as the assembler and 463loader. The driver passes the value of this macro to @code{putenv} to 464initialize the necessary environment variables. 465@end defmac 466 467@defmac LOCAL_INCLUDE_DIR 468Define this macro as a C string constant if you wish to override the 469standard choice of @file{/usr/local/include} as the default prefix to 470try when searching for local header files. @code{LOCAL_INCLUDE_DIR} 471comes before @code{NATIVE_SYSTEM_HEADER_DIR} (set in 472@file{config.gcc}, normally @file{/usr/include}) in the search order. 473 474Cross compilers do not search either @file{/usr/local/include} or its 475replacement. 476@end defmac 477 478@defmac NATIVE_SYSTEM_HEADER_COMPONENT 479The ``component'' corresponding to @code{NATIVE_SYSTEM_HEADER_DIR}. 480See @code{INCLUDE_DEFAULTS}, below, for the description of components. 481If you do not define this macro, no component is used. 482@end defmac 483 484@defmac INCLUDE_DEFAULTS 485Define this macro if you wish to override the entire default search path 486for include files. For a native compiler, the default search path 487usually consists of @code{GCC_INCLUDE_DIR}, @code{LOCAL_INCLUDE_DIR}, 488@code{GPLUSPLUS_INCLUDE_DIR}, and 489@code{NATIVE_SYSTEM_HEADER_DIR}. In addition, @code{GPLUSPLUS_INCLUDE_DIR} 490and @code{GCC_INCLUDE_DIR} are defined automatically by @file{Makefile}, 491and specify private search areas for GCC@. The directory 492@code{GPLUSPLUS_INCLUDE_DIR} is used only for C++ programs. 493 494The definition should be an initializer for an array of structures. 495Each array element should have four elements: the directory name (a 496string constant), the component name (also a string constant), a flag 497for C++-only directories, 498and a flag showing that the includes in the directory don't need to be 499wrapped in @code{extern @samp{C}} when compiling C++. Mark the end of 500the array with a null element. 501 502The component name denotes what GNU package the include file is part of, 503if any, in all uppercase letters. For example, it might be @samp{GCC} 504or @samp{BINUTILS}. If the package is part of a vendor-supplied 505operating system, code the component name as @samp{0}. 506 507For example, here is the definition used for VAX/VMS: 508 509@smallexample 510#define INCLUDE_DEFAULTS \ 511@{ \ 512 @{ "GNU_GXX_INCLUDE:", "G++", 1, 1@}, \ 513 @{ "GNU_CC_INCLUDE:", "GCC", 0, 0@}, \ 514 @{ "SYS$SYSROOT:[SYSLIB.]", 0, 0, 0@}, \ 515 @{ ".", 0, 0, 0@}, \ 516 @{ 0, 0, 0, 0@} \ 517@} 518@end smallexample 519@end defmac 520 521Here is the order of prefixes tried for exec files: 522 523@enumerate 524@item 525Any prefixes specified by the user with @option{-B}. 526 527@item 528The environment variable @code{GCC_EXEC_PREFIX} or, if @code{GCC_EXEC_PREFIX} 529is not set and the compiler has not been installed in the configure-time 530@var{prefix}, the location in which the compiler has actually been installed. 531 532@item 533The directories specified by the environment variable @code{COMPILER_PATH}. 534 535@item 536The macro @code{STANDARD_EXEC_PREFIX}, if the compiler has been installed 537in the configured-time @var{prefix}. 538 539@item 540The location @file{/usr/libexec/gcc/}, but only if this is a native compiler. 541 542@item 543The location @file{/usr/lib/gcc/}, but only if this is a native compiler. 544 545@item 546The macro @code{MD_EXEC_PREFIX}, if defined, but only if this is a native 547compiler. 548@end enumerate 549 550Here is the order of prefixes tried for startfiles: 551 552@enumerate 553@item 554Any prefixes specified by the user with @option{-B}. 555 556@item 557The environment variable @code{GCC_EXEC_PREFIX} or its automatically determined 558value based on the installed toolchain location. 559 560@item 561The directories specified by the environment variable @code{LIBRARY_PATH} 562(or port-specific name; native only, cross compilers do not use this). 563 564@item 565The macro @code{STANDARD_EXEC_PREFIX}, but only if the toolchain is installed 566in the configured @var{prefix} or this is a native compiler. 567 568@item 569The location @file{/usr/lib/gcc/}, but only if this is a native compiler. 570 571@item 572The macro @code{MD_EXEC_PREFIX}, if defined, but only if this is a native 573compiler. 574 575@item 576The macro @code{MD_STARTFILE_PREFIX}, if defined, but only if this is a 577native compiler, or we have a target system root. 578 579@item 580The macro @code{MD_STARTFILE_PREFIX_1}, if defined, but only if this is a 581native compiler, or we have a target system root. 582 583@item 584The macro @code{STANDARD_STARTFILE_PREFIX}, with any sysroot modifications. 585If this path is relative it will be prefixed by @code{GCC_EXEC_PREFIX} and 586the machine suffix or @code{STANDARD_EXEC_PREFIX} and the machine suffix. 587 588@item 589The macro @code{STANDARD_STARTFILE_PREFIX_1}, but only if this is a native 590compiler, or we have a target system root. The default for this macro is 591@file{/lib/}. 592 593@item 594The macro @code{STANDARD_STARTFILE_PREFIX_2}, but only if this is a native 595compiler, or we have a target system root. The default for this macro is 596@file{/usr/lib/}. 597@end enumerate 598 599@node Run-time Target 600@section Run-time Target Specification 601@cindex run-time target specification 602@cindex predefined macros 603@cindex target specifications 604 605@c prevent bad page break with this line 606Here are run-time target specifications. 607 608@defmac TARGET_CPU_CPP_BUILTINS () 609This function-like macro expands to a block of code that defines 610built-in preprocessor macros and assertions for the target CPU, using 611the functions @code{builtin_define}, @code{builtin_define_std} and 612@code{builtin_assert}. When the front end 613calls this macro it provides a trailing semicolon, and since it has 614finished command line option processing your code can use those 615results freely. 616 617@code{builtin_assert} takes a string in the form you pass to the 618command-line option @option{-A}, such as @code{cpu=mips}, and creates 619the assertion. @code{builtin_define} takes a string in the form 620accepted by option @option{-D} and unconditionally defines the macro. 621 622@code{builtin_define_std} takes a string representing the name of an 623object-like macro. If it doesn't lie in the user's namespace, 624@code{builtin_define_std} defines it unconditionally. Otherwise, it 625defines a version with two leading underscores, and another version 626with two leading and trailing underscores, and defines the original 627only if an ISO standard was not requested on the command line. For 628example, passing @code{unix} defines @code{__unix}, @code{__unix__} 629and possibly @code{unix}; passing @code{_mips} defines @code{__mips}, 630@code{__mips__} and possibly @code{_mips}, and passing @code{_ABI64} 631defines only @code{_ABI64}. 632 633You can also test for the C dialect being compiled. The variable 634@code{c_language} is set to one of @code{clk_c}, @code{clk_cplusplus} 635or @code{clk_objective_c}. Note that if we are preprocessing 636assembler, this variable will be @code{clk_c} but the function-like 637macro @code{preprocessing_asm_p()} will return true, so you might want 638to check for that first. If you need to check for strict ANSI, the 639variable @code{flag_iso} can be used. The function-like macro 640@code{preprocessing_trad_p()} can be used to check for traditional 641preprocessing. 642@end defmac 643 644@defmac TARGET_OS_CPP_BUILTINS () 645Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional 646and is used for the target operating system instead. 647@end defmac 648 649@defmac TARGET_OBJFMT_CPP_BUILTINS () 650Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional 651and is used for the target object format. @file{elfos.h} uses this 652macro to define @code{__ELF__}, so you probably do not need to define 653it yourself. 654@end defmac 655 656@deftypevar {extern int} target_flags 657This variable is declared in @file{options.h}, which is included before 658any target-specific headers. 659@end deftypevar 660 661@hook TARGET_DEFAULT_TARGET_FLAGS 662This variable specifies the initial value of @code{target_flags}. 663Its default setting is 0. 664@end deftypevr 665 666@cindex optional hardware or system features 667@cindex features, optional, in system conventions 668 669@hook TARGET_HANDLE_OPTION 670This hook is called whenever the user specifies one of the 671target-specific options described by the @file{.opt} definition files 672(@pxref{Options}). It has the opportunity to do some option-specific 673processing and should return true if the option is valid. The default 674definition does nothing but return true. 675 676@var{decoded} specifies the option and its arguments. @var{opts} and 677@var{opts_set} are the @code{gcc_options} structures to be used for 678storing option state, and @var{loc} is the location at which the 679option was passed (@code{UNKNOWN_LOCATION} except for options passed 680via attributes). 681@end deftypefn 682 683@hook TARGET_HANDLE_C_OPTION 684This target hook is called whenever the user specifies one of the 685target-specific C language family options described by the @file{.opt} 686definition files(@pxref{Options}). It has the opportunity to do some 687option-specific processing and should return true if the option is 688valid. The arguments are like for @code{TARGET_HANDLE_OPTION}. The 689default definition does nothing but return false. 690 691In general, you should use @code{TARGET_HANDLE_OPTION} to handle 692options. However, if processing an option requires routines that are 693only available in the C (and related language) front ends, then you 694should use @code{TARGET_HANDLE_C_OPTION} instead. 695@end deftypefn 696 697@hook TARGET_OBJC_CONSTRUCT_STRING_OBJECT 698 699@hook TARGET_OBJC_DECLARE_UNRESOLVED_CLASS_REFERENCE 700 701@hook TARGET_OBJC_DECLARE_CLASS_DEFINITION 702 703@hook TARGET_STRING_OBJECT_REF_TYPE_P 704 705@hook TARGET_CHECK_STRING_OBJECT_FORMAT_ARG 706 707@hook TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE 708This target function is similar to the hook @code{TARGET_OPTION_OVERRIDE} 709but is called when the optimize level is changed via an attribute or 710pragma or when it is reset at the end of the code affected by the 711attribute or pragma. It is not called at the beginning of compilation 712when @code{TARGET_OPTION_OVERRIDE} is called so if you want to perform these 713actions then, you should have @code{TARGET_OPTION_OVERRIDE} call 714@code{TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE}. 715@end deftypefn 716 717@defmac C_COMMON_OVERRIDE_OPTIONS 718This is similar to the @code{TARGET_OPTION_OVERRIDE} hook 719but is only used in the C 720language frontends (C, Objective-C, C++, Objective-C++) and so can be 721used to alter option flag variables which only exist in those 722frontends. 723@end defmac 724 725@hook TARGET_OPTION_OPTIMIZATION_TABLE 726Some machines may desire to change what optimizations are performed for 727various optimization levels. This variable, if defined, describes 728options to enable at particular sets of optimization levels. These 729options are processed once 730just after the optimization level is determined and before the remainder 731of the command options have been parsed, so may be overridden by other 732options passed explicitly. 733 734This processing is run once at program startup and when the optimization 735options are changed via @code{#pragma GCC optimize} or by using the 736@code{optimize} attribute. 737@end deftypevr 738 739@hook TARGET_OPTION_INIT_STRUCT 740 741@hook TARGET_OPTION_DEFAULT_PARAMS 742 743@defmac SWITCHABLE_TARGET 744Some targets need to switch between substantially different subtargets 745during compilation. For example, the MIPS target has one subtarget for 746the traditional MIPS architecture and another for MIPS16. Source code 747can switch between these two subarchitectures using the @code{mips16} 748and @code{nomips16} attributes. 749 750Such subtargets can differ in things like the set of available 751registers, the set of available instructions, the costs of various 752operations, and so on. GCC caches a lot of this type of information 753in global variables, and recomputing them for each subtarget takes a 754significant amount of time. The compiler therefore provides a facility 755for maintaining several versions of the global variables and quickly 756switching between them; see @file{target-globals.h} for details. 757 758Define this macro to 1 if your target needs this facility. The default 759is 0. 760@end defmac 761 762@node Per-Function Data 763@section Defining data structures for per-function information. 764@cindex per-function data 765@cindex data structures 766 767If the target needs to store information on a per-function basis, GCC 768provides a macro and a couple of variables to allow this. Note, just 769using statics to store the information is a bad idea, since GCC supports 770nested functions, so you can be halfway through encoding one function 771when another one comes along. 772 773GCC defines a data structure called @code{struct function} which 774contains all of the data specific to an individual function. This 775structure contains a field called @code{machine} whose type is 776@code{struct machine_function *}, which can be used by targets to point 777to their own specific data. 778 779If a target needs per-function specific data it should define the type 780@code{struct machine_function} and also the macro @code{INIT_EXPANDERS}. 781This macro should be used to initialize the function pointer 782@code{init_machine_status}. This pointer is explained below. 783 784One typical use of per-function, target specific data is to create an 785RTX to hold the register containing the function's return address. This 786RTX can then be used to implement the @code{__builtin_return_address} 787function, for level 0. 788 789Note---earlier implementations of GCC used a single data area to hold 790all of the per-function information. Thus when processing of a nested 791function began the old per-function data had to be pushed onto a 792stack, and when the processing was finished, it had to be popped off the 793stack. GCC used to provide function pointers called 794@code{save_machine_status} and @code{restore_machine_status} to handle 795the saving and restoring of the target specific information. Since the 796single data area approach is no longer used, these pointers are no 797longer supported. 798 799@defmac INIT_EXPANDERS 800Macro called to initialize any target specific information. This macro 801is called once per function, before generation of any RTL has begun. 802The intention of this macro is to allow the initialization of the 803function pointer @code{init_machine_status}. 804@end defmac 805 806@deftypevar {void (*)(struct function *)} init_machine_status 807If this function pointer is non-@code{NULL} it will be called once per 808function, before function compilation starts, in order to allow the 809target to perform any target specific initialization of the 810@code{struct function} structure. It is intended that this would be 811used to initialize the @code{machine} of that structure. 812 813@code{struct machine_function} structures are expected to be freed by GC@. 814Generally, any memory that they reference must be allocated by using 815GC allocation, including the structure itself. 816@end deftypevar 817 818@node Storage Layout 819@section Storage Layout 820@cindex storage layout 821 822Note that the definitions of the macros in this table which are sizes or 823alignments measured in bits do not need to be constant. They can be C 824expressions that refer to static variables, such as the @code{target_flags}. 825@xref{Run-time Target}. 826 827@defmac BITS_BIG_ENDIAN 828Define this macro to have the value 1 if the most significant bit in a 829byte has the lowest number; otherwise define it to have the value zero. 830This means that bit-field instructions count from the most significant 831bit. If the machine has no bit-field instructions, then this must still 832be defined, but it doesn't matter which value it is defined to. This 833macro need not be a constant. 834 835This macro does not affect the way structure fields are packed into 836bytes or words; that is controlled by @code{BYTES_BIG_ENDIAN}. 837@end defmac 838 839@defmac BYTES_BIG_ENDIAN 840Define this macro to have the value 1 if the most significant byte in a 841word has the lowest number. This macro need not be a constant. 842@end defmac 843 844@defmac WORDS_BIG_ENDIAN 845Define this macro to have the value 1 if, in a multiword object, the 846most significant word has the lowest number. This applies to both 847memory locations and registers; see @code{REG_WORDS_BIG_ENDIAN} if the 848order of words in memory is not the same as the order in registers. This 849macro need not be a constant. 850@end defmac 851 852@defmac REG_WORDS_BIG_ENDIAN 853On some machines, the order of words in a multiword object differs between 854registers in memory. In such a situation, define this macro to describe 855the order of words in a register. The macro @code{WORDS_BIG_ENDIAN} controls 856the order of words in memory. 857@end defmac 858 859@defmac FLOAT_WORDS_BIG_ENDIAN 860Define this macro to have the value 1 if @code{DFmode}, @code{XFmode} or 861@code{TFmode} floating point numbers are stored in memory with the word 862containing the sign bit at the lowest address; otherwise define it to 863have the value 0. This macro need not be a constant. 864 865You need not define this macro if the ordering is the same as for 866multi-word integers. 867@end defmac 868 869@defmac BITS_PER_UNIT 870Define this macro to be the number of bits in an addressable storage 871unit (byte). If you do not define this macro the default is 8. 872@end defmac 873 874@defmac BITS_PER_WORD 875Number of bits in a word. If you do not define this macro, the default 876is @code{BITS_PER_UNIT * UNITS_PER_WORD}. 877@end defmac 878 879@defmac MAX_BITS_PER_WORD 880Maximum number of bits in a word. If this is undefined, the default is 881@code{BITS_PER_WORD}. Otherwise, it is the constant value that is the 882largest value that @code{BITS_PER_WORD} can have at run-time. 883@end defmac 884 885@defmac UNITS_PER_WORD 886Number of storage units in a word; normally the size of a general-purpose 887register, a power of two from 1 or 8. 888@end defmac 889 890@defmac MIN_UNITS_PER_WORD 891Minimum number of units in a word. If this is undefined, the default is 892@code{UNITS_PER_WORD}. Otherwise, it is the constant value that is the 893smallest value that @code{UNITS_PER_WORD} can have at run-time. 894@end defmac 895 896@defmac POINTER_SIZE 897Width of a pointer, in bits. You must specify a value no wider than the 898width of @code{Pmode}. If it is not equal to the width of @code{Pmode}, 899you must define @code{POINTERS_EXTEND_UNSIGNED}. If you do not specify 900a value the default is @code{BITS_PER_WORD}. 901@end defmac 902 903@defmac POINTERS_EXTEND_UNSIGNED 904A C expression that determines how pointers should be extended from 905@code{ptr_mode} to either @code{Pmode} or @code{word_mode}. It is 906greater than zero if pointers should be zero-extended, zero if they 907should be sign-extended, and negative if some other sort of conversion 908is needed. In the last case, the extension is done by the target's 909@code{ptr_extend} instruction. 910 911You need not define this macro if the @code{ptr_mode}, @code{Pmode} 912and @code{word_mode} are all the same width. 913@end defmac 914 915@defmac PROMOTE_MODE (@var{m}, @var{unsignedp}, @var{type}) 916A macro to update @var{m} and @var{unsignedp} when an object whose type 917is @var{type} and which has the specified mode and signedness is to be 918stored in a register. This macro is only called when @var{type} is a 919scalar type. 920 921On most RISC machines, which only have operations that operate on a full 922register, define this macro to set @var{m} to @code{word_mode} if 923@var{m} is an integer mode narrower than @code{BITS_PER_WORD}. In most 924cases, only integer modes should be widened because wider-precision 925floating-point operations are usually more expensive than their narrower 926counterparts. 927 928For most machines, the macro definition does not change @var{unsignedp}. 929However, some machines, have instructions that preferentially handle 930either signed or unsigned quantities of certain modes. For example, on 931the DEC Alpha, 32-bit loads from memory and 32-bit add instructions 932sign-extend the result to 64 bits. On such machines, set 933@var{unsignedp} according to which kind of extension is more efficient. 934 935Do not define this macro if it would never modify @var{m}. 936@end defmac 937 938@hook TARGET_PROMOTE_FUNCTION_MODE 939Like @code{PROMOTE_MODE}, but it is applied to outgoing function arguments or 940function return values. The target hook should return the new mode 941and possibly change @code{*@var{punsignedp}} if the promotion should 942change signedness. This function is called only for scalar @emph{or 943pointer} types. 944 945@var{for_return} allows to distinguish the promotion of arguments and 946return values. If it is @code{1}, a return value is being promoted and 947@code{TARGET_FUNCTION_VALUE} must perform the same promotions done here. 948If it is @code{2}, the returned mode should be that of the register in 949which an incoming parameter is copied, or the outgoing result is computed; 950then the hook should return the same mode as @code{promote_mode}, though 951the signedness may be different. 952 953@var{type} can be NULL when promoting function arguments of libcalls. 954 955The default is to not promote arguments and return values. You can 956also define the hook to @code{default_promote_function_mode_always_promote} 957if you would like to apply the same rules given by @code{PROMOTE_MODE}. 958@end deftypefn 959 960@defmac PARM_BOUNDARY 961Normal alignment required for function parameters on the stack, in 962bits. All stack parameters receive at least this much alignment 963regardless of data type. On most machines, this is the same as the 964size of an integer. 965@end defmac 966 967@defmac STACK_BOUNDARY 968Define this macro to the minimum alignment enforced by hardware for the 969stack pointer on this machine. The definition is a C expression for the 970desired alignment (measured in bits). This value is used as a default 971if @code{PREFERRED_STACK_BOUNDARY} is not defined. On most machines, 972this should be the same as @code{PARM_BOUNDARY}. 973@end defmac 974 975@defmac PREFERRED_STACK_BOUNDARY 976Define this macro if you wish to preserve a certain alignment for the 977stack pointer, greater than what the hardware enforces. The definition 978is a C expression for the desired alignment (measured in bits). This 979macro must evaluate to a value equal to or larger than 980@code{STACK_BOUNDARY}. 981@end defmac 982 983@defmac INCOMING_STACK_BOUNDARY 984Define this macro if the incoming stack boundary may be different 985from @code{PREFERRED_STACK_BOUNDARY}. This macro must evaluate 986to a value equal to or larger than @code{STACK_BOUNDARY}. 987@end defmac 988 989@defmac FUNCTION_BOUNDARY 990Alignment required for a function entry point, in bits. 991@end defmac 992 993@defmac BIGGEST_ALIGNMENT 994Biggest alignment that any data type can require on this machine, in 995bits. Note that this is not the biggest alignment that is supported, 996just the biggest alignment that, when violated, may cause a fault. 997@end defmac 998 999@defmac MALLOC_ABI_ALIGNMENT 1000Alignment, in bits, a C conformant malloc implementation has to 1001provide. If not defined, the default value is @code{BITS_PER_WORD}. 1002@end defmac 1003 1004@defmac ATTRIBUTE_ALIGNED_VALUE 1005Alignment used by the @code{__attribute__ ((aligned))} construct. If 1006not defined, the default value is @code{BIGGEST_ALIGNMENT}. 1007@end defmac 1008 1009@defmac MINIMUM_ATOMIC_ALIGNMENT 1010If defined, the smallest alignment, in bits, that can be given to an 1011object that can be referenced in one operation, without disturbing any 1012nearby object. Normally, this is @code{BITS_PER_UNIT}, but may be larger 1013on machines that don't have byte or half-word store operations. 1014@end defmac 1015 1016@defmac BIGGEST_FIELD_ALIGNMENT 1017Biggest alignment that any structure or union field can require on this 1018machine, in bits. If defined, this overrides @code{BIGGEST_ALIGNMENT} for 1019structure and union fields only, unless the field alignment has been set 1020by the @code{__attribute__ ((aligned (@var{n})))} construct. 1021@end defmac 1022 1023@defmac ADJUST_FIELD_ALIGN (@var{field}, @var{computed}) 1024An expression for the alignment of a structure field @var{field} if the 1025alignment computed in the usual way (including applying of 1026@code{BIGGEST_ALIGNMENT} and @code{BIGGEST_FIELD_ALIGNMENT} to the 1027alignment) is @var{computed}. It overrides alignment only if the 1028field alignment has not been set by the 1029@code{__attribute__ ((aligned (@var{n})))} construct. 1030@end defmac 1031 1032@defmac MAX_STACK_ALIGNMENT 1033Biggest stack alignment guaranteed by the backend. Use this macro 1034to specify the maximum alignment of a variable on stack. 1035 1036If not defined, the default value is @code{STACK_BOUNDARY}. 1037 1038@c FIXME: The default should be @code{PREFERRED_STACK_BOUNDARY}. 1039@c But the fix for PR 32893 indicates that we can only guarantee 1040@c maximum stack alignment on stack up to @code{STACK_BOUNDARY}, not 1041@c @code{PREFERRED_STACK_BOUNDARY}, if stack alignment isn't supported. 1042@end defmac 1043 1044@defmac MAX_OFILE_ALIGNMENT 1045Biggest alignment supported by the object file format of this machine. 1046Use this macro to limit the alignment which can be specified using the 1047@code{__attribute__ ((aligned (@var{n})))} construct. If not defined, 1048the default value is @code{BIGGEST_ALIGNMENT}. 1049 1050On systems that use ELF, the default (in @file{config/elfos.h}) is 1051the largest supported 32-bit ELF section alignment representable on 1052a 32-bit host e.g. @samp{(((unsigned HOST_WIDEST_INT) 1 << 28) * 8)}. 1053On 32-bit ELF the largest supported section alignment in bits is 1054@samp{(0x80000000 * 8)}, but this is not representable on 32-bit hosts. 1055@end defmac 1056 1057@defmac DATA_ALIGNMENT (@var{type}, @var{basic-align}) 1058If defined, a C expression to compute the alignment for a variable in 1059the static store. @var{type} is the data type, and @var{basic-align} is 1060the alignment that the object would ordinarily have. The value of this 1061macro is used instead of that alignment to align the object. 1062 1063If this macro is not defined, then @var{basic-align} is used. 1064 1065@findex strcpy 1066One use of this macro is to increase alignment of medium-size data to 1067make it all fit in fewer cache lines. Another is to cause character 1068arrays to be word-aligned so that @code{strcpy} calls that copy 1069constants to character arrays can be done inline. 1070@end defmac 1071 1072@defmac CONSTANT_ALIGNMENT (@var{constant}, @var{basic-align}) 1073If defined, a C expression to compute the alignment given to a constant 1074that is being placed in memory. @var{constant} is the constant and 1075@var{basic-align} is the alignment that the object would ordinarily 1076have. The value of this macro is used instead of that alignment to 1077align the object. 1078 1079If this macro is not defined, then @var{basic-align} is used. 1080 1081The typical use of this macro is to increase alignment for string 1082constants to be word aligned so that @code{strcpy} calls that copy 1083constants can be done inline. 1084@end defmac 1085 1086@defmac LOCAL_ALIGNMENT (@var{type}, @var{basic-align}) 1087If defined, a C expression to compute the alignment for a variable in 1088the local store. @var{type} is the data type, and @var{basic-align} is 1089the alignment that the object would ordinarily have. The value of this 1090macro is used instead of that alignment to align the object. 1091 1092If this macro is not defined, then @var{basic-align} is used. 1093 1094One use of this macro is to increase alignment of medium-size data to 1095make it all fit in fewer cache lines. 1096 1097If the value of this macro has a type, it should be an unsigned type. 1098@end defmac 1099 1100@hook TARGET_VECTOR_ALIGNMENT 1101 1102@defmac STACK_SLOT_ALIGNMENT (@var{type}, @var{mode}, @var{basic-align}) 1103If defined, a C expression to compute the alignment for stack slot. 1104@var{type} is the data type, @var{mode} is the widest mode available, 1105and @var{basic-align} is the alignment that the slot would ordinarily 1106have. The value of this macro is used instead of that alignment to 1107align the slot. 1108 1109If this macro is not defined, then @var{basic-align} is used when 1110@var{type} is @code{NULL}. Otherwise, @code{LOCAL_ALIGNMENT} will 1111be used. 1112 1113This macro is to set alignment of stack slot to the maximum alignment 1114of all possible modes which the slot may have. 1115 1116If the value of this macro has a type, it should be an unsigned type. 1117@end defmac 1118 1119@defmac LOCAL_DECL_ALIGNMENT (@var{decl}) 1120If defined, a C expression to compute the alignment for a local 1121variable @var{decl}. 1122 1123If this macro is not defined, then 1124@code{LOCAL_ALIGNMENT (TREE_TYPE (@var{decl}), DECL_ALIGN (@var{decl}))} 1125is used. 1126 1127One use of this macro is to increase alignment of medium-size data to 1128make it all fit in fewer cache lines. 1129 1130If the value of this macro has a type, it should be an unsigned type. 1131@end defmac 1132 1133@defmac MINIMUM_ALIGNMENT (@var{exp}, @var{mode}, @var{align}) 1134If defined, a C expression to compute the minimum required alignment 1135for dynamic stack realignment purposes for @var{exp} (a type or decl), 1136@var{mode}, assuming normal alignment @var{align}. 1137 1138If this macro is not defined, then @var{align} will be used. 1139@end defmac 1140 1141@defmac EMPTY_FIELD_BOUNDARY 1142Alignment in bits to be given to a structure bit-field that follows an 1143empty field such as @code{int : 0;}. 1144 1145If @code{PCC_BITFIELD_TYPE_MATTERS} is true, it overrides this macro. 1146@end defmac 1147 1148@defmac STRUCTURE_SIZE_BOUNDARY 1149Number of bits which any structure or union's size must be a multiple of. 1150Each structure or union's size is rounded up to a multiple of this. 1151 1152If you do not define this macro, the default is the same as 1153@code{BITS_PER_UNIT}. 1154@end defmac 1155 1156@defmac STRICT_ALIGNMENT 1157Define this macro to be the value 1 if instructions will fail to work 1158if given data not on the nominal alignment. If instructions will merely 1159go slower in that case, define this macro as 0. 1160@end defmac 1161 1162@defmac PCC_BITFIELD_TYPE_MATTERS 1163Define this if you wish to imitate the way many other C compilers handle 1164alignment of bit-fields and the structures that contain them. 1165 1166The behavior is that the type written for a named bit-field (@code{int}, 1167@code{short}, or other integer type) imposes an alignment for the entire 1168structure, as if the structure really did contain an ordinary field of 1169that type. In addition, the bit-field is placed within the structure so 1170that it would fit within such a field, not crossing a boundary for it. 1171 1172Thus, on most machines, a named bit-field whose type is written as 1173@code{int} would not cross a four-byte boundary, and would force 1174four-byte alignment for the whole structure. (The alignment used may 1175not be four bytes; it is controlled by the other alignment parameters.) 1176 1177An unnamed bit-field will not affect the alignment of the containing 1178structure. 1179 1180If the macro is defined, its definition should be a C expression; 1181a nonzero value for the expression enables this behavior. 1182 1183Note that if this macro is not defined, or its value is zero, some 1184bit-fields may cross more than one alignment boundary. The compiler can 1185support such references if there are @samp{insv}, @samp{extv}, and 1186@samp{extzv} insns that can directly reference memory. 1187 1188The other known way of making bit-fields work is to define 1189@code{STRUCTURE_SIZE_BOUNDARY} as large as @code{BIGGEST_ALIGNMENT}. 1190Then every structure can be accessed with fullwords. 1191 1192Unless the machine has bit-field instructions or you define 1193@code{STRUCTURE_SIZE_BOUNDARY} that way, you must define 1194@code{PCC_BITFIELD_TYPE_MATTERS} to have a nonzero value. 1195 1196If your aim is to make GCC use the same conventions for laying out 1197bit-fields as are used by another compiler, here is how to investigate 1198what the other compiler does. Compile and run this program: 1199 1200@smallexample 1201struct foo1 1202@{ 1203 char x; 1204 char :0; 1205 char y; 1206@}; 1207 1208struct foo2 1209@{ 1210 char x; 1211 int :0; 1212 char y; 1213@}; 1214 1215main () 1216@{ 1217 printf ("Size of foo1 is %d\n", 1218 sizeof (struct foo1)); 1219 printf ("Size of foo2 is %d\n", 1220 sizeof (struct foo2)); 1221 exit (0); 1222@} 1223@end smallexample 1224 1225If this prints 2 and 5, then the compiler's behavior is what you would 1226get from @code{PCC_BITFIELD_TYPE_MATTERS}. 1227@end defmac 1228 1229@defmac BITFIELD_NBYTES_LIMITED 1230Like @code{PCC_BITFIELD_TYPE_MATTERS} except that its effect is limited 1231to aligning a bit-field within the structure. 1232@end defmac 1233 1234@hook TARGET_ALIGN_ANON_BITFIELD 1235When @code{PCC_BITFIELD_TYPE_MATTERS} is true this hook will determine 1236whether unnamed bitfields affect the alignment of the containing 1237structure. The hook should return true if the structure should inherit 1238the alignment requirements of an unnamed bitfield's type. 1239@end deftypefn 1240 1241@hook TARGET_NARROW_VOLATILE_BITFIELD 1242This target hook should return @code{true} if accesses to volatile bitfields 1243should use the narrowest mode possible. It should return @code{false} if 1244these accesses should use the bitfield container type. 1245 1246The default is @code{!TARGET_STRICT_ALIGN}. 1247@end deftypefn 1248 1249@defmac MEMBER_TYPE_FORCES_BLK (@var{field}, @var{mode}) 1250Return 1 if a structure or array containing @var{field} should be accessed using 1251@code{BLKMODE}. 1252 1253If @var{field} is the only field in the structure, @var{mode} is its 1254mode, otherwise @var{mode} is VOIDmode. @var{mode} is provided in the 1255case where structures of one field would require the structure's mode to 1256retain the field's mode. 1257 1258Normally, this is not needed. 1259@end defmac 1260 1261@defmac ROUND_TYPE_ALIGN (@var{type}, @var{computed}, @var{specified}) 1262Define this macro as an expression for the alignment of a type (given 1263by @var{type} as a tree node) if the alignment computed in the usual 1264way is @var{computed} and the alignment explicitly specified was 1265@var{specified}. 1266 1267The default is to use @var{specified} if it is larger; otherwise, use 1268the smaller of @var{computed} and @code{BIGGEST_ALIGNMENT} 1269@end defmac 1270 1271@defmac MAX_FIXED_MODE_SIZE 1272An integer expression for the size in bits of the largest integer 1273machine mode that should actually be used. All integer machine modes of 1274this size or smaller can be used for structures and unions with the 1275appropriate sizes. If this macro is undefined, @code{GET_MODE_BITSIZE 1276(DImode)} is assumed. 1277@end defmac 1278 1279@defmac STACK_SAVEAREA_MODE (@var{save_level}) 1280If defined, an expression of type @code{enum machine_mode} that 1281specifies the mode of the save area operand of a 1282@code{save_stack_@var{level}} named pattern (@pxref{Standard Names}). 1283@var{save_level} is one of @code{SAVE_BLOCK}, @code{SAVE_FUNCTION}, or 1284@code{SAVE_NONLOCAL} and selects which of the three named patterns is 1285having its mode specified. 1286 1287You need not define this macro if it always returns @code{Pmode}. You 1288would most commonly define this macro if the 1289@code{save_stack_@var{level}} patterns need to support both a 32- and a 129064-bit mode. 1291@end defmac 1292 1293@defmac STACK_SIZE_MODE 1294If defined, an expression of type @code{enum machine_mode} that 1295specifies the mode of the size increment operand of an 1296@code{allocate_stack} named pattern (@pxref{Standard Names}). 1297 1298You need not define this macro if it always returns @code{word_mode}. 1299You would most commonly define this macro if the @code{allocate_stack} 1300pattern needs to support both a 32- and a 64-bit mode. 1301@end defmac 1302 1303@hook TARGET_LIBGCC_CMP_RETURN_MODE 1304This target hook should return the mode to be used for the return value 1305of compare instructions expanded to libgcc calls. If not defined 1306@code{word_mode} is returned which is the right choice for a majority of 1307targets. 1308@end deftypefn 1309 1310@hook TARGET_LIBGCC_SHIFT_COUNT_MODE 1311This target hook should return the mode to be used for the shift count operand 1312of shift instructions expanded to libgcc calls. If not defined 1313@code{word_mode} is returned which is the right choice for a majority of 1314targets. 1315@end deftypefn 1316 1317@hook TARGET_UNWIND_WORD_MODE 1318Return machine mode to be used for @code{_Unwind_Word} type. 1319The default is to use @code{word_mode}. 1320@end deftypefn 1321 1322@defmac ROUND_TOWARDS_ZERO 1323If defined, this macro should be true if the prevailing rounding 1324mode is towards zero. 1325 1326Defining this macro only affects the way @file{libgcc.a} emulates 1327floating-point arithmetic. 1328 1329Not defining this macro is equivalent to returning zero. 1330@end defmac 1331 1332@defmac LARGEST_EXPONENT_IS_NORMAL (@var{size}) 1333This macro should return true if floats with @var{size} 1334bits do not have a NaN or infinity representation, but use the largest 1335exponent for normal numbers instead. 1336 1337Defining this macro only affects the way @file{libgcc.a} emulates 1338floating-point arithmetic. 1339 1340The default definition of this macro returns false for all sizes. 1341@end defmac 1342 1343@hook TARGET_MS_BITFIELD_LAYOUT_P 1344This target hook returns @code{true} if bit-fields in the given 1345@var{record_type} are to be laid out following the rules of Microsoft 1346Visual C/C++, namely: (i) a bit-field won't share the same storage 1347unit with the previous bit-field if their underlying types have 1348different sizes, and the bit-field will be aligned to the highest 1349alignment of the underlying types of itself and of the previous 1350bit-field; (ii) a zero-sized bit-field will affect the alignment of 1351the whole enclosing structure, even if it is unnamed; except that 1352(iii) a zero-sized bit-field will be disregarded unless it follows 1353another bit-field of nonzero size. If this hook returns @code{true}, 1354other macros that control bit-field layout are ignored. 1355 1356When a bit-field is inserted into a packed record, the whole size 1357of the underlying type is used by one or more same-size adjacent 1358bit-fields (that is, if its long:3, 32 bits is used in the record, 1359and any additional adjacent long bit-fields are packed into the same 1360chunk of 32 bits. However, if the size changes, a new field of that 1361size is allocated). In an unpacked record, this is the same as using 1362alignment, but not equivalent when packing. 1363 1364If both MS bit-fields and @samp{__attribute__((packed))} are used, 1365the latter will take precedence. If @samp{__attribute__((packed))} is 1366used on a single field when MS bit-fields are in use, it will take 1367precedence for that field, but the alignment of the rest of the structure 1368may affect its placement. 1369@end deftypefn 1370 1371@hook TARGET_DECIMAL_FLOAT_SUPPORTED_P 1372Returns true if the target supports decimal floating point. 1373@end deftypefn 1374 1375@hook TARGET_FIXED_POINT_SUPPORTED_P 1376Returns true if the target supports fixed-point arithmetic. 1377@end deftypefn 1378 1379@hook TARGET_EXPAND_TO_RTL_HOOK 1380This hook is called just before expansion into rtl, allowing the target 1381to perform additional initializations or analysis before the expansion. 1382For example, the rs6000 port uses it to allocate a scratch stack slot 1383for use in copying SDmode values between memory and floating point 1384registers whenever the function being expanded has any SDmode 1385usage. 1386@end deftypefn 1387 1388@hook TARGET_INSTANTIATE_DECLS 1389This hook allows the backend to perform additional instantiations on rtl 1390that are not actually in any insns yet, but will be later. 1391@end deftypefn 1392 1393@hook TARGET_MANGLE_TYPE 1394If your target defines any fundamental types, or any types your target 1395uses should be mangled differently from the default, define this hook 1396to return the appropriate encoding for these types as part of a C++ 1397mangled name. The @var{type} argument is the tree structure representing 1398the type to be mangled. The hook may be applied to trees which are 1399not target-specific fundamental types; it should return @code{NULL} 1400for all such types, as well as arguments it does not recognize. If the 1401return value is not @code{NULL}, it must point to a statically-allocated 1402string constant. 1403 1404Target-specific fundamental types might be new fundamental types or 1405qualified versions of ordinary fundamental types. Encode new 1406fundamental types as @samp{@w{u @var{n} @var{name}}}, where @var{name} 1407is the name used for the type in source code, and @var{n} is the 1408length of @var{name} in decimal. Encode qualified versions of 1409ordinary types as @samp{@w{U @var{n} @var{name} @var{code}}}, where 1410@var{name} is the name used for the type qualifier in source code, 1411@var{n} is the length of @var{name} as above, and @var{code} is the 1412code used to represent the unqualified version of this type. (See 1413@code{write_builtin_type} in @file{cp/mangle.c} for the list of 1414codes.) In both cases the spaces are for clarity; do not include any 1415spaces in your string. 1416 1417This hook is applied to types prior to typedef resolution. If the mangled 1418name for a particular type depends only on that type's main variant, you 1419can perform typedef resolution yourself using @code{TYPE_MAIN_VARIANT} 1420before mangling. 1421 1422The default version of this hook always returns @code{NULL}, which is 1423appropriate for a target that does not define any new fundamental 1424types. 1425@end deftypefn 1426 1427@node Type Layout 1428@section Layout of Source Language Data Types 1429 1430These macros define the sizes and other characteristics of the standard 1431basic data types used in programs being compiled. Unlike the macros in 1432the previous section, these apply to specific features of C and related 1433languages, rather than to fundamental aspects of storage layout. 1434 1435@defmac INT_TYPE_SIZE 1436A C expression for the size in bits of the type @code{int} on the 1437target machine. If you don't define this, the default is one word. 1438@end defmac 1439 1440@defmac SHORT_TYPE_SIZE 1441A C expression for the size in bits of the type @code{short} on the 1442target machine. If you don't define this, the default is half a word. 1443(If this would be less than one storage unit, it is rounded up to one 1444unit.) 1445@end defmac 1446 1447@defmac LONG_TYPE_SIZE 1448A C expression for the size in bits of the type @code{long} on the 1449target machine. If you don't define this, the default is one word. 1450@end defmac 1451 1452@defmac ADA_LONG_TYPE_SIZE 1453On some machines, the size used for the Ada equivalent of the type 1454@code{long} by a native Ada compiler differs from that used by C@. In 1455that situation, define this macro to be a C expression to be used for 1456the size of that type. If you don't define this, the default is the 1457value of @code{LONG_TYPE_SIZE}. 1458@end defmac 1459 1460@defmac LONG_LONG_TYPE_SIZE 1461A C expression for the size in bits of the type @code{long long} on the 1462target machine. If you don't define this, the default is two 1463words. If you want to support GNU Ada on your machine, the value of this 1464macro must be at least 64. 1465@end defmac 1466 1467@defmac CHAR_TYPE_SIZE 1468A C expression for the size in bits of the type @code{char} on the 1469target machine. If you don't define this, the default is 1470@code{BITS_PER_UNIT}. 1471@end defmac 1472 1473@defmac BOOL_TYPE_SIZE 1474A C expression for the size in bits of the C++ type @code{bool} and 1475C99 type @code{_Bool} on the target machine. If you don't define 1476this, and you probably shouldn't, the default is @code{CHAR_TYPE_SIZE}. 1477@end defmac 1478 1479@defmac FLOAT_TYPE_SIZE 1480A C expression for the size in bits of the type @code{float} on the 1481target machine. If you don't define this, the default is one word. 1482@end defmac 1483 1484@defmac DOUBLE_TYPE_SIZE 1485A C expression for the size in bits of the type @code{double} on the 1486target machine. If you don't define this, the default is two 1487words. 1488@end defmac 1489 1490@defmac LONG_DOUBLE_TYPE_SIZE 1491A C expression for the size in bits of the type @code{long double} on 1492the target machine. If you don't define this, the default is two 1493words. 1494@end defmac 1495 1496@defmac SHORT_FRACT_TYPE_SIZE 1497A C expression for the size in bits of the type @code{short _Fract} on 1498the target machine. If you don't define this, the default is 1499@code{BITS_PER_UNIT}. 1500@end defmac 1501 1502@defmac FRACT_TYPE_SIZE 1503A C expression for the size in bits of the type @code{_Fract} on 1504the target machine. If you don't define this, the default is 1505@code{BITS_PER_UNIT * 2}. 1506@end defmac 1507 1508@defmac LONG_FRACT_TYPE_SIZE 1509A C expression for the size in bits of the type @code{long _Fract} on 1510the target machine. If you don't define this, the default is 1511@code{BITS_PER_UNIT * 4}. 1512@end defmac 1513 1514@defmac LONG_LONG_FRACT_TYPE_SIZE 1515A C expression for the size in bits of the type @code{long long _Fract} on 1516the target machine. If you don't define this, the default is 1517@code{BITS_PER_UNIT * 8}. 1518@end defmac 1519 1520@defmac SHORT_ACCUM_TYPE_SIZE 1521A C expression for the size in bits of the type @code{short _Accum} on 1522the target machine. If you don't define this, the default is 1523@code{BITS_PER_UNIT * 2}. 1524@end defmac 1525 1526@defmac ACCUM_TYPE_SIZE 1527A C expression for the size in bits of the type @code{_Accum} on 1528the target machine. If you don't define this, the default is 1529@code{BITS_PER_UNIT * 4}. 1530@end defmac 1531 1532@defmac LONG_ACCUM_TYPE_SIZE 1533A C expression for the size in bits of the type @code{long _Accum} on 1534the target machine. If you don't define this, the default is 1535@code{BITS_PER_UNIT * 8}. 1536@end defmac 1537 1538@defmac LONG_LONG_ACCUM_TYPE_SIZE 1539A C expression for the size in bits of the type @code{long long _Accum} on 1540the target machine. If you don't define this, the default is 1541@code{BITS_PER_UNIT * 16}. 1542@end defmac 1543 1544@defmac LIBGCC2_LONG_DOUBLE_TYPE_SIZE 1545Define this macro if @code{LONG_DOUBLE_TYPE_SIZE} is not constant or 1546if you want routines in @file{libgcc2.a} for a size other than 1547@code{LONG_DOUBLE_TYPE_SIZE}. If you don't define this, the 1548default is @code{LONG_DOUBLE_TYPE_SIZE}. 1549@end defmac 1550 1551@defmac LIBGCC2_HAS_DF_MODE 1552Define this macro if neither @code{DOUBLE_TYPE_SIZE} nor 1553@code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 1554@code{DFmode} but you want @code{DFmode} routines in @file{libgcc2.a} 1555anyway. If you don't define this and either @code{DOUBLE_TYPE_SIZE} 1556or @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 64 then the default is 1, 1557otherwise it is 0. 1558@end defmac 1559 1560@defmac LIBGCC2_HAS_XF_MODE 1561Define this macro if @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is not 1562@code{XFmode} but you want @code{XFmode} routines in @file{libgcc2.a} 1563anyway. If you don't define this and @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} 1564is 80 then the default is 1, otherwise it is 0. 1565@end defmac 1566 1567@defmac LIBGCC2_HAS_TF_MODE 1568Define this macro if @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is not 1569@code{TFmode} but you want @code{TFmode} routines in @file{libgcc2.a} 1570anyway. If you don't define this and @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} 1571is 128 then the default is 1, otherwise it is 0. 1572@end defmac 1573 1574@defmac LIBGCC2_GNU_PREFIX 1575This macro corresponds to the @code{TARGET_LIBFUNC_GNU_PREFIX} target 1576hook and should be defined if that hook is overriden to be true. It 1577causes function names in libgcc to be changed to use a @code{__gnu_} 1578prefix for their name rather than the default @code{__}. A port which 1579uses this macro should also arrange to use @file{t-gnu-prefix} in 1580the libgcc @file{config.host}. 1581@end defmac 1582 1583@defmac SF_SIZE 1584@defmacx DF_SIZE 1585@defmacx XF_SIZE 1586@defmacx TF_SIZE 1587Define these macros to be the size in bits of the mantissa of 1588@code{SFmode}, @code{DFmode}, @code{XFmode} and @code{TFmode} values, 1589if the defaults in @file{libgcc2.h} are inappropriate. By default, 1590@code{FLT_MANT_DIG} is used for @code{SF_SIZE}, @code{LDBL_MANT_DIG} 1591for @code{XF_SIZE} and @code{TF_SIZE}, and @code{DBL_MANT_DIG} or 1592@code{LDBL_MANT_DIG} for @code{DF_SIZE} according to whether 1593@code{DOUBLE_TYPE_SIZE} or 1594@code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 64. 1595@end defmac 1596 1597@defmac TARGET_FLT_EVAL_METHOD 1598A C expression for the value for @code{FLT_EVAL_METHOD} in @file{float.h}, 1599assuming, if applicable, that the floating-point control word is in its 1600default state. If you do not define this macro the value of 1601@code{FLT_EVAL_METHOD} will be zero. 1602@end defmac 1603 1604@defmac WIDEST_HARDWARE_FP_SIZE 1605A C expression for the size in bits of the widest floating-point format 1606supported by the hardware. If you define this macro, you must specify a 1607value less than or equal to the value of @code{LONG_DOUBLE_TYPE_SIZE}. 1608If you do not define this macro, the value of @code{LONG_DOUBLE_TYPE_SIZE} 1609is the default. 1610@end defmac 1611 1612@defmac DEFAULT_SIGNED_CHAR 1613An expression whose value is 1 or 0, according to whether the type 1614@code{char} should be signed or unsigned by default. The user can 1615always override this default with the options @option{-fsigned-char} 1616and @option{-funsigned-char}. 1617@end defmac 1618 1619@hook TARGET_DEFAULT_SHORT_ENUMS 1620This target hook should return true if the compiler should give an 1621@code{enum} type only as many bytes as it takes to represent the range 1622of possible values of that type. It should return false if all 1623@code{enum} types should be allocated like @code{int}. 1624 1625The default is to return false. 1626@end deftypefn 1627 1628@defmac SIZE_TYPE 1629A C expression for a string describing the name of the data type to use 1630for size values. The typedef name @code{size_t} is defined using the 1631contents of the string. 1632 1633The string can contain more than one keyword. If so, separate them with 1634spaces, and write first any length keyword, then @code{unsigned} if 1635appropriate, and finally @code{int}. The string must exactly match one 1636of the data type names defined in the function 1637@code{init_decl_processing} in the file @file{c-decl.c}. You may not 1638omit @code{int} or change the order---that would cause the compiler to 1639crash on startup. 1640 1641If you don't define this macro, the default is @code{"long unsigned 1642int"}. 1643@end defmac 1644 1645@defmac PTRDIFF_TYPE 1646A C expression for a string describing the name of the data type to use 1647for the result of subtracting two pointers. The typedef name 1648@code{ptrdiff_t} is defined using the contents of the string. See 1649@code{SIZE_TYPE} above for more information. 1650 1651If you don't define this macro, the default is @code{"long int"}. 1652@end defmac 1653 1654@defmac WCHAR_TYPE 1655A C expression for a string describing the name of the data type to use 1656for wide characters. The typedef name @code{wchar_t} is defined using 1657the contents of the string. See @code{SIZE_TYPE} above for more 1658information. 1659 1660If you don't define this macro, the default is @code{"int"}. 1661@end defmac 1662 1663@defmac WCHAR_TYPE_SIZE 1664A C expression for the size in bits of the data type for wide 1665characters. This is used in @code{cpp}, which cannot make use of 1666@code{WCHAR_TYPE}. 1667@end defmac 1668 1669@defmac WINT_TYPE 1670A C expression for a string describing the name of the data type to 1671use for wide characters passed to @code{printf} and returned from 1672@code{getwc}. The typedef name @code{wint_t} is defined using the 1673contents of the string. See @code{SIZE_TYPE} above for more 1674information. 1675 1676If you don't define this macro, the default is @code{"unsigned int"}. 1677@end defmac 1678 1679@defmac INTMAX_TYPE 1680A C expression for a string describing the name of the data type that 1681can represent any value of any standard or extended signed integer type. 1682The typedef name @code{intmax_t} is defined using the contents of the 1683string. See @code{SIZE_TYPE} above for more information. 1684 1685If you don't define this macro, the default is the first of 1686@code{"int"}, @code{"long int"}, or @code{"long long int"} that has as 1687much precision as @code{long long int}. 1688@end defmac 1689 1690@defmac UINTMAX_TYPE 1691A C expression for a string describing the name of the data type that 1692can represent any value of any standard or extended unsigned integer 1693type. The typedef name @code{uintmax_t} is defined using the contents 1694of the string. See @code{SIZE_TYPE} above for more information. 1695 1696If you don't define this macro, the default is the first of 1697@code{"unsigned int"}, @code{"long unsigned int"}, or @code{"long long 1698unsigned int"} that has as much precision as @code{long long unsigned 1699int}. 1700@end defmac 1701 1702@defmac SIG_ATOMIC_TYPE 1703@defmacx INT8_TYPE 1704@defmacx INT16_TYPE 1705@defmacx INT32_TYPE 1706@defmacx INT64_TYPE 1707@defmacx UINT8_TYPE 1708@defmacx UINT16_TYPE 1709@defmacx UINT32_TYPE 1710@defmacx UINT64_TYPE 1711@defmacx INT_LEAST8_TYPE 1712@defmacx INT_LEAST16_TYPE 1713@defmacx INT_LEAST32_TYPE 1714@defmacx INT_LEAST64_TYPE 1715@defmacx UINT_LEAST8_TYPE 1716@defmacx UINT_LEAST16_TYPE 1717@defmacx UINT_LEAST32_TYPE 1718@defmacx UINT_LEAST64_TYPE 1719@defmacx INT_FAST8_TYPE 1720@defmacx INT_FAST16_TYPE 1721@defmacx INT_FAST32_TYPE 1722@defmacx INT_FAST64_TYPE 1723@defmacx UINT_FAST8_TYPE 1724@defmacx UINT_FAST16_TYPE 1725@defmacx UINT_FAST32_TYPE 1726@defmacx UINT_FAST64_TYPE 1727@defmacx INTPTR_TYPE 1728@defmacx UINTPTR_TYPE 1729C expressions for the standard types @code{sig_atomic_t}, 1730@code{int8_t}, @code{int16_t}, @code{int32_t}, @code{int64_t}, 1731@code{uint8_t}, @code{uint16_t}, @code{uint32_t}, @code{uint64_t}, 1732@code{int_least8_t}, @code{int_least16_t}, @code{int_least32_t}, 1733@code{int_least64_t}, @code{uint_least8_t}, @code{uint_least16_t}, 1734@code{uint_least32_t}, @code{uint_least64_t}, @code{int_fast8_t}, 1735@code{int_fast16_t}, @code{int_fast32_t}, @code{int_fast64_t}, 1736@code{uint_fast8_t}, @code{uint_fast16_t}, @code{uint_fast32_t}, 1737@code{uint_fast64_t}, @code{intptr_t}, and @code{uintptr_t}. See 1738@code{SIZE_TYPE} above for more information. 1739 1740If any of these macros evaluates to a null pointer, the corresponding 1741type is not supported; if GCC is configured to provide 1742@code{<stdint.h>} in such a case, the header provided may not conform 1743to C99, depending on the type in question. The defaults for all of 1744these macros are null pointers. 1745@end defmac 1746 1747@defmac TARGET_PTRMEMFUNC_VBIT_LOCATION 1748The C++ compiler represents a pointer-to-member-function with a struct 1749that looks like: 1750 1751@smallexample 1752 struct @{ 1753 union @{ 1754 void (*fn)(); 1755 ptrdiff_t vtable_index; 1756 @}; 1757 ptrdiff_t delta; 1758 @}; 1759@end smallexample 1760 1761@noindent 1762The C++ compiler must use one bit to indicate whether the function that 1763will be called through a pointer-to-member-function is virtual. 1764Normally, we assume that the low-order bit of a function pointer must 1765always be zero. Then, by ensuring that the vtable_index is odd, we can 1766distinguish which variant of the union is in use. But, on some 1767platforms function pointers can be odd, and so this doesn't work. In 1768that case, we use the low-order bit of the @code{delta} field, and shift 1769the remainder of the @code{delta} field to the left. 1770 1771GCC will automatically make the right selection about where to store 1772this bit using the @code{FUNCTION_BOUNDARY} setting for your platform. 1773However, some platforms such as ARM/Thumb have @code{FUNCTION_BOUNDARY} 1774set such that functions always start at even addresses, but the lowest 1775bit of pointers to functions indicate whether the function at that 1776address is in ARM or Thumb mode. If this is the case of your 1777architecture, you should define this macro to 1778@code{ptrmemfunc_vbit_in_delta}. 1779 1780In general, you should not have to define this macro. On architectures 1781in which function addresses are always even, according to 1782@code{FUNCTION_BOUNDARY}, GCC will automatically define this macro to 1783@code{ptrmemfunc_vbit_in_pfn}. 1784@end defmac 1785 1786@defmac TARGET_VTABLE_USES_DESCRIPTORS 1787Normally, the C++ compiler uses function pointers in vtables. This 1788macro allows the target to change to use ``function descriptors'' 1789instead. Function descriptors are found on targets for whom a 1790function pointer is actually a small data structure. Normally the 1791data structure consists of the actual code address plus a data 1792pointer to which the function's data is relative. 1793 1794If vtables are used, the value of this macro should be the number 1795of words that the function descriptor occupies. 1796@end defmac 1797 1798@defmac TARGET_VTABLE_ENTRY_ALIGN 1799By default, the vtable entries are void pointers, the so the alignment 1800is the same as pointer alignment. The value of this macro specifies 1801the alignment of the vtable entry in bits. It should be defined only 1802when special alignment is necessary. */ 1803@end defmac 1804 1805@defmac TARGET_VTABLE_DATA_ENTRY_DISTANCE 1806There are a few non-descriptor entries in the vtable at offsets below 1807zero. If these entries must be padded (say, to preserve the alignment 1808specified by @code{TARGET_VTABLE_ENTRY_ALIGN}), set this to the number 1809of words in each data entry. 1810@end defmac 1811 1812@node Registers 1813@section Register Usage 1814@cindex register usage 1815 1816This section explains how to describe what registers the target machine 1817has, and how (in general) they can be used. 1818 1819The description of which registers a specific instruction can use is 1820done with register classes; see @ref{Register Classes}. For information 1821on using registers to access a stack frame, see @ref{Frame Registers}. 1822For passing values in registers, see @ref{Register Arguments}. 1823For returning values in registers, see @ref{Scalar Return}. 1824 1825@menu 1826* Register Basics:: Number and kinds of registers. 1827* Allocation Order:: Order in which registers are allocated. 1828* Values in Registers:: What kinds of values each reg can hold. 1829* Leaf Functions:: Renumbering registers for leaf functions. 1830* Stack Registers:: Handling a register stack such as 80387. 1831@end menu 1832 1833@node Register Basics 1834@subsection Basic Characteristics of Registers 1835 1836@c prevent bad page break with this line 1837Registers have various characteristics. 1838 1839@defmac FIRST_PSEUDO_REGISTER 1840Number of hardware registers known to the compiler. They receive 1841numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first 1842pseudo register's number really is assigned the number 1843@code{FIRST_PSEUDO_REGISTER}. 1844@end defmac 1845 1846@defmac FIXED_REGISTERS 1847@cindex fixed register 1848An initializer that says which registers are used for fixed purposes 1849all throughout the compiled code and are therefore not available for 1850general allocation. These would include the stack pointer, the frame 1851pointer (except on machines where that can be used as a general 1852register when no frame pointer is needed), the program counter on 1853machines where that is considered one of the addressable registers, 1854and any other numbered register with a standard use. 1855 1856This information is expressed as a sequence of numbers, separated by 1857commas and surrounded by braces. The @var{n}th number is 1 if 1858register @var{n} is fixed, 0 otherwise. 1859 1860The table initialized from this macro, and the table initialized by 1861the following one, may be overridden at run time either automatically, 1862by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by 1863the user with the command options @option{-ffixed-@var{reg}}, 1864@option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}. 1865@end defmac 1866 1867@defmac CALL_USED_REGISTERS 1868@cindex call-used register 1869@cindex call-clobbered register 1870@cindex call-saved register 1871Like @code{FIXED_REGISTERS} but has 1 for each register that is 1872clobbered (in general) by function calls as well as for fixed 1873registers. This macro therefore identifies the registers that are not 1874available for general allocation of values that must live across 1875function calls. 1876 1877If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler 1878automatically saves it on function entry and restores it on function 1879exit, if the register is used within the function. 1880@end defmac 1881 1882@defmac CALL_REALLY_USED_REGISTERS 1883@cindex call-used register 1884@cindex call-clobbered register 1885@cindex call-saved register 1886Like @code{CALL_USED_REGISTERS} except this macro doesn't require 1887that the entire set of @code{FIXED_REGISTERS} be included. 1888(@code{CALL_USED_REGISTERS} must be a superset of @code{FIXED_REGISTERS}). 1889This macro is optional. If not specified, it defaults to the value 1890of @code{CALL_USED_REGISTERS}. 1891@end defmac 1892 1893@defmac HARD_REGNO_CALL_PART_CLOBBERED (@var{regno}, @var{mode}) 1894@cindex call-used register 1895@cindex call-clobbered register 1896@cindex call-saved register 1897A C expression that is nonzero if it is not permissible to store a 1898value of mode @var{mode} in hard register number @var{regno} across a 1899call without some part of it being clobbered. For most machines this 1900macro need not be defined. It is only required for machines that do not 1901preserve the entire contents of a register across a call. 1902@end defmac 1903 1904@findex fixed_regs 1905@findex call_used_regs 1906@findex global_regs 1907@findex reg_names 1908@findex reg_class_contents 1909@hook TARGET_CONDITIONAL_REGISTER_USAGE 1910This hook may conditionally modify five variables 1911@code{fixed_regs}, @code{call_used_regs}, @code{global_regs}, 1912@code{reg_names}, and @code{reg_class_contents}, to take into account 1913any dependence of these register sets on target flags. The first three 1914of these are of type @code{char []} (interpreted as Boolean vectors). 1915@code{global_regs} is a @code{const char *[]}, and 1916@code{reg_class_contents} is a @code{HARD_REG_SET}. Before the macro is 1917called, @code{fixed_regs}, @code{call_used_regs}, 1918@code{reg_class_contents}, and @code{reg_names} have been initialized 1919from @code{FIXED_REGISTERS}, @code{CALL_USED_REGISTERS}, 1920@code{REG_CLASS_CONTENTS}, and @code{REGISTER_NAMES}, respectively. 1921@code{global_regs} has been cleared, and any @option{-ffixed-@var{reg}}, 1922@option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}} 1923command options have been applied. 1924 1925@cindex disabling certain registers 1926@cindex controlling register usage 1927If the usage of an entire class of registers depends on the target 1928flags, you may indicate this to GCC by using this macro to modify 1929@code{fixed_regs} and @code{call_used_regs} to 1 for each of the 1930registers in the classes which should not be used by GCC@. Also define 1931the macro @code{REG_CLASS_FROM_LETTER} / @code{REG_CLASS_FROM_CONSTRAINT} 1932to return @code{NO_REGS} if it 1933is called with a letter for a class that shouldn't be used. 1934 1935(However, if this class is not included in @code{GENERAL_REGS} and all 1936of the insn patterns whose constraints permit this class are 1937controlled by target switches, then GCC will automatically avoid using 1938these registers when the target switches are opposed to them.) 1939@end deftypefn 1940 1941@defmac INCOMING_REGNO (@var{out}) 1942Define this macro if the target machine has register windows. This C 1943expression returns the register number as seen by the called function 1944corresponding to the register number @var{out} as seen by the calling 1945function. Return @var{out} if register number @var{out} is not an 1946outbound register. 1947@end defmac 1948 1949@defmac OUTGOING_REGNO (@var{in}) 1950Define this macro if the target machine has register windows. This C 1951expression returns the register number as seen by the calling function 1952corresponding to the register number @var{in} as seen by the called 1953function. Return @var{in} if register number @var{in} is not an inbound 1954register. 1955@end defmac 1956 1957@defmac LOCAL_REGNO (@var{regno}) 1958Define this macro if the target machine has register windows. This C 1959expression returns true if the register is call-saved but is in the 1960register window. Unlike most call-saved registers, such registers 1961need not be explicitly restored on function exit or during non-local 1962gotos. 1963@end defmac 1964 1965@defmac PC_REGNUM 1966If the program counter has a register number, define this as that 1967register number. Otherwise, do not define it. 1968@end defmac 1969 1970@node Allocation Order 1971@subsection Order of Allocation of Registers 1972@cindex order of register allocation 1973@cindex register allocation order 1974 1975@c prevent bad page break with this line 1976Registers are allocated in order. 1977 1978@defmac REG_ALLOC_ORDER 1979If defined, an initializer for a vector of integers, containing the 1980numbers of hard registers in the order in which GCC should prefer 1981to use them (from most preferred to least). 1982 1983If this macro is not defined, registers are used lowest numbered first 1984(all else being equal). 1985 1986One use of this macro is on machines where the highest numbered 1987registers must always be saved and the save-multiple-registers 1988instruction supports only sequences of consecutive registers. On such 1989machines, define @code{REG_ALLOC_ORDER} to be an initializer that lists 1990the highest numbered allocable register first. 1991@end defmac 1992 1993@defmac ADJUST_REG_ALLOC_ORDER 1994A C statement (sans semicolon) to choose the order in which to allocate 1995hard registers for pseudo-registers local to a basic block. 1996 1997Store the desired register order in the array @code{reg_alloc_order}. 1998Element 0 should be the register to allocate first; element 1, the next 1999register; and so on. 2000 2001The macro body should not assume anything about the contents of 2002@code{reg_alloc_order} before execution of the macro. 2003 2004On most machines, it is not necessary to define this macro. 2005@end defmac 2006 2007@defmac HONOR_REG_ALLOC_ORDER 2008Normally, IRA tries to estimate the costs for saving a register in the 2009prologue and restoring it in the epilogue. This discourages it from 2010using call-saved registers. If a machine wants to ensure that IRA 2011allocates registers in the order given by REG_ALLOC_ORDER even if some 2012call-saved registers appear earlier than call-used ones, this macro 2013should be defined. 2014@end defmac 2015 2016@defmac IRA_HARD_REGNO_ADD_COST_MULTIPLIER (@var{regno}) 2017In some case register allocation order is not enough for the 2018Integrated Register Allocator (@acronym{IRA}) to generate a good code. 2019If this macro is defined, it should return a floating point value 2020based on @var{regno}. The cost of using @var{regno} for a pseudo will 2021be increased by approximately the pseudo's usage frequency times the 2022value returned by this macro. Not defining this macro is equivalent 2023to having it always return @code{0.0}. 2024 2025On most machines, it is not necessary to define this macro. 2026@end defmac 2027 2028@node Values in Registers 2029@subsection How Values Fit in Registers 2030 2031This section discusses the macros that describe which kinds of values 2032(specifically, which machine modes) each register can hold, and how many 2033consecutive registers are needed for a given mode. 2034 2035@defmac HARD_REGNO_NREGS (@var{regno}, @var{mode}) 2036A C expression for the number of consecutive hard registers, starting 2037at register number @var{regno}, required to hold a value of mode 2038@var{mode}. This macro must never return zero, even if a register 2039cannot hold the requested mode - indicate that with HARD_REGNO_MODE_OK 2040and/or CANNOT_CHANGE_MODE_CLASS instead. 2041 2042On a machine where all registers are exactly one word, a suitable 2043definition of this macro is 2044 2045@smallexample 2046#define HARD_REGNO_NREGS(REGNO, MODE) \ 2047 ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) \ 2048 / UNITS_PER_WORD) 2049@end smallexample 2050@end defmac 2051 2052@defmac HARD_REGNO_NREGS_HAS_PADDING (@var{regno}, @var{mode}) 2053A C expression that is nonzero if a value of mode @var{mode}, stored 2054in memory, ends with padding that causes it to take up more space than 2055in registers starting at register number @var{regno} (as determined by 2056multiplying GCC's notion of the size of the register when containing 2057this mode by the number of registers returned by 2058@code{HARD_REGNO_NREGS}). By default this is zero. 2059 2060For example, if a floating-point value is stored in three 32-bit 2061registers but takes up 128 bits in memory, then this would be 2062nonzero. 2063 2064This macros only needs to be defined if there are cases where 2065@code{subreg_get_info} 2066would otherwise wrongly determine that a @code{subreg} can be 2067represented by an offset to the register number, when in fact such a 2068@code{subreg} would contain some of the padding not stored in 2069registers and so not be representable. 2070@end defmac 2071 2072@defmac HARD_REGNO_NREGS_WITH_PADDING (@var{regno}, @var{mode}) 2073For values of @var{regno} and @var{mode} for which 2074@code{HARD_REGNO_NREGS_HAS_PADDING} returns nonzero, a C expression 2075returning the greater number of registers required to hold the value 2076including any padding. In the example above, the value would be four. 2077@end defmac 2078 2079@defmac REGMODE_NATURAL_SIZE (@var{mode}) 2080Define this macro if the natural size of registers that hold values 2081of mode @var{mode} is not the word size. It is a C expression that 2082should give the natural size in bytes for the specified mode. It is 2083used by the register allocator to try to optimize its results. This 2084happens for example on SPARC 64-bit where the natural size of 2085floating-point registers is still 32-bit. 2086@end defmac 2087 2088@defmac HARD_REGNO_MODE_OK (@var{regno}, @var{mode}) 2089A C expression that is nonzero if it is permissible to store a value 2090of mode @var{mode} in hard register number @var{regno} (or in several 2091registers starting with that one). For a machine where all registers 2092are equivalent, a suitable definition is 2093 2094@smallexample 2095#define HARD_REGNO_MODE_OK(REGNO, MODE) 1 2096@end smallexample 2097 2098You need not include code to check for the numbers of fixed registers, 2099because the allocation mechanism considers them to be always occupied. 2100 2101@cindex register pairs 2102On some machines, double-precision values must be kept in even/odd 2103register pairs. You can implement that by defining this macro to reject 2104odd register numbers for such modes. 2105 2106The minimum requirement for a mode to be OK in a register is that the 2107@samp{mov@var{mode}} instruction pattern support moves between the 2108register and other hard register in the same class and that moving a 2109value into the register and back out not alter it. 2110 2111Since the same instruction used to move @code{word_mode} will work for 2112all narrower integer modes, it is not necessary on any machine for 2113@code{HARD_REGNO_MODE_OK} to distinguish between these modes, provided 2114you define patterns @samp{movhi}, etc., to take advantage of this. This 2115is useful because of the interaction between @code{HARD_REGNO_MODE_OK} 2116and @code{MODES_TIEABLE_P}; it is very desirable for all integer modes 2117to be tieable. 2118 2119Many machines have special registers for floating point arithmetic. 2120Often people assume that floating point machine modes are allowed only 2121in floating point registers. This is not true. Any registers that 2122can hold integers can safely @emph{hold} a floating point machine 2123mode, whether or not floating arithmetic can be done on it in those 2124registers. Integer move instructions can be used to move the values. 2125 2126On some machines, though, the converse is true: fixed-point machine 2127modes may not go in floating registers. This is true if the floating 2128registers normalize any value stored in them, because storing a 2129non-floating value there would garble it. In this case, 2130@code{HARD_REGNO_MODE_OK} should reject fixed-point machine modes in 2131floating registers. But if the floating registers do not automatically 2132normalize, if you can store any bit pattern in one and retrieve it 2133unchanged without a trap, then any machine mode may go in a floating 2134register, so you can define this macro to say so. 2135 2136The primary significance of special floating registers is rather that 2137they are the registers acceptable in floating point arithmetic 2138instructions. However, this is of no concern to 2139@code{HARD_REGNO_MODE_OK}. You handle it by writing the proper 2140constraints for those instructions. 2141 2142On some machines, the floating registers are especially slow to access, 2143so that it is better to store a value in a stack frame than in such a 2144register if floating point arithmetic is not being done. As long as the 2145floating registers are not in class @code{GENERAL_REGS}, they will not 2146be used unless some pattern's constraint asks for one. 2147@end defmac 2148 2149@defmac HARD_REGNO_RENAME_OK (@var{from}, @var{to}) 2150A C expression that is nonzero if it is OK to rename a hard register 2151@var{from} to another hard register @var{to}. 2152 2153One common use of this macro is to prevent renaming of a register to 2154another register that is not saved by a prologue in an interrupt 2155handler. 2156 2157The default is always nonzero. 2158@end defmac 2159 2160@defmac MODES_TIEABLE_P (@var{mode1}, @var{mode2}) 2161A C expression that is nonzero if a value of mode 2162@var{mode1} is accessible in mode @var{mode2} without copying. 2163 2164If @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and 2165@code{HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are always the same for 2166any @var{r}, then @code{MODES_TIEABLE_P (@var{mode1}, @var{mode2})} 2167should be nonzero. If they differ for any @var{r}, you should define 2168this macro to return zero unless some other mechanism ensures the 2169accessibility of the value in a narrower mode. 2170 2171You should define this macro to return nonzero in as many cases as 2172possible since doing so will allow GCC to perform better register 2173allocation. 2174@end defmac 2175 2176@hook TARGET_HARD_REGNO_SCRATCH_OK 2177This target hook should return @code{true} if it is OK to use a hard register 2178@var{regno} as scratch reg in peephole2. 2179 2180One common use of this macro is to prevent using of a register that 2181is not saved by a prologue in an interrupt handler. 2182 2183The default version of this hook always returns @code{true}. 2184@end deftypefn 2185 2186@defmac AVOID_CCMODE_COPIES 2187Define this macro if the compiler should avoid copies to/from @code{CCmode} 2188registers. You should only define this macro if support for copying to/from 2189@code{CCmode} is incomplete. 2190@end defmac 2191 2192@node Leaf Functions 2193@subsection Handling Leaf Functions 2194 2195@cindex leaf functions 2196@cindex functions, leaf 2197On some machines, a leaf function (i.e., one which makes no calls) can run 2198more efficiently if it does not make its own register window. Often this 2199means it is required to receive its arguments in the registers where they 2200are passed by the caller, instead of the registers where they would 2201normally arrive. 2202 2203The special treatment for leaf functions generally applies only when 2204other conditions are met; for example, often they may use only those 2205registers for its own variables and temporaries. We use the term ``leaf 2206function'' to mean a function that is suitable for this special 2207handling, so that functions with no calls are not necessarily ``leaf 2208functions''. 2209 2210GCC assigns register numbers before it knows whether the function is 2211suitable for leaf function treatment. So it needs to renumber the 2212registers in order to output a leaf function. The following macros 2213accomplish this. 2214 2215@defmac LEAF_REGISTERS 2216Name of a char vector, indexed by hard register number, which 2217contains 1 for a register that is allowable in a candidate for leaf 2218function treatment. 2219 2220If leaf function treatment involves renumbering the registers, then the 2221registers marked here should be the ones before renumbering---those that 2222GCC would ordinarily allocate. The registers which will actually be 2223used in the assembler code, after renumbering, should not be marked with 1 2224in this vector. 2225 2226Define this macro only if the target machine offers a way to optimize 2227the treatment of leaf functions. 2228@end defmac 2229 2230@defmac LEAF_REG_REMAP (@var{regno}) 2231A C expression whose value is the register number to which @var{regno} 2232should be renumbered, when a function is treated as a leaf function. 2233 2234If @var{regno} is a register number which should not appear in a leaf 2235function before renumbering, then the expression should yield @minus{}1, which 2236will cause the compiler to abort. 2237 2238Define this macro only if the target machine offers a way to optimize the 2239treatment of leaf functions, and registers need to be renumbered to do 2240this. 2241@end defmac 2242 2243@findex current_function_is_leaf 2244@findex current_function_uses_only_leaf_regs 2245@code{TARGET_ASM_FUNCTION_PROLOGUE} and 2246@code{TARGET_ASM_FUNCTION_EPILOGUE} must usually treat leaf functions 2247specially. They can test the C variable @code{current_function_is_leaf} 2248which is nonzero for leaf functions. @code{current_function_is_leaf} is 2249set prior to local register allocation and is valid for the remaining 2250compiler passes. They can also test the C variable 2251@code{current_function_uses_only_leaf_regs} which is nonzero for leaf 2252functions which only use leaf registers. 2253@code{current_function_uses_only_leaf_regs} is valid after all passes 2254that modify the instructions have been run and is only useful if 2255@code{LEAF_REGISTERS} is defined. 2256@c changed this to fix overfull. ALSO: why the "it" at the beginning 2257@c of the next paragraph?! --mew 2feb93 2258 2259@node Stack Registers 2260@subsection Registers That Form a Stack 2261 2262There are special features to handle computers where some of the 2263``registers'' form a stack. Stack registers are normally written by 2264pushing onto the stack, and are numbered relative to the top of the 2265stack. 2266 2267Currently, GCC can only handle one group of stack-like registers, and 2268they must be consecutively numbered. Furthermore, the existing 2269support for stack-like registers is specific to the 80387 floating 2270point coprocessor. If you have a new architecture that uses 2271stack-like registers, you will need to do substantial work on 2272@file{reg-stack.c} and write your machine description to cooperate 2273with it, as well as defining these macros. 2274 2275@defmac STACK_REGS 2276Define this if the machine has any stack-like registers. 2277@end defmac 2278 2279@defmac STACK_REG_COVER_CLASS 2280This is a cover class containing the stack registers. Define this if 2281the machine has any stack-like registers. 2282@end defmac 2283 2284@defmac FIRST_STACK_REG 2285The number of the first stack-like register. This one is the top 2286of the stack. 2287@end defmac 2288 2289@defmac LAST_STACK_REG 2290The number of the last stack-like register. This one is the bottom of 2291the stack. 2292@end defmac 2293 2294@node Register Classes 2295@section Register Classes 2296@cindex register class definitions 2297@cindex class definitions, register 2298 2299On many machines, the numbered registers are not all equivalent. 2300For example, certain registers may not be allowed for indexed addressing; 2301certain registers may not be allowed in some instructions. These machine 2302restrictions are described to the compiler using @dfn{register classes}. 2303 2304You define a number of register classes, giving each one a name and saying 2305which of the registers belong to it. Then you can specify register classes 2306that are allowed as operands to particular instruction patterns. 2307 2308@findex ALL_REGS 2309@findex NO_REGS 2310In general, each register will belong to several classes. In fact, one 2311class must be named @code{ALL_REGS} and contain all the registers. Another 2312class must be named @code{NO_REGS} and contain no registers. Often the 2313union of two classes will be another class; however, this is not required. 2314 2315@findex GENERAL_REGS 2316One of the classes must be named @code{GENERAL_REGS}. There is nothing 2317terribly special about the name, but the operand constraint letters 2318@samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is 2319the same as @code{ALL_REGS}, just define it as a macro which expands 2320to @code{ALL_REGS}. 2321 2322Order the classes so that if class @var{x} is contained in class @var{y} 2323then @var{x} has a lower class number than @var{y}. 2324 2325The way classes other than @code{GENERAL_REGS} are specified in operand 2326constraints is through machine-dependent operand constraint letters. 2327You can define such letters to correspond to various classes, then use 2328them in operand constraints. 2329 2330You must define the narrowest register classes for allocatable 2331registers, so that each class either has no subclasses, or that for 2332some mode, the move cost between registers within the class is 2333cheaper than moving a register in the class to or from memory 2334(@pxref{Costs}). 2335 2336You should define a class for the union of two classes whenever some 2337instruction allows both classes. For example, if an instruction allows 2338either a floating point (coprocessor) register or a general register for a 2339certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS} 2340which includes both of them. Otherwise you will get suboptimal code, 2341or even internal compiler errors when reload cannot find a register in the 2342class computed via @code{reg_class_subunion}. 2343 2344You must also specify certain redundant information about the register 2345classes: for each class, which classes contain it and which ones are 2346contained in it; for each pair of classes, the largest class contained 2347in their union. 2348 2349When a value occupying several consecutive registers is expected in a 2350certain class, all the registers used must belong to that class. 2351Therefore, register classes cannot be used to enforce a requirement for 2352a register pair to start with an even-numbered register. The way to 2353specify this requirement is with @code{HARD_REGNO_MODE_OK}. 2354 2355Register classes used for input-operands of bitwise-and or shift 2356instructions have a special requirement: each such class must have, for 2357each fixed-point machine mode, a subclass whose registers can transfer that 2358mode to or from memory. For example, on some machines, the operations for 2359single-byte values (@code{QImode}) are limited to certain registers. When 2360this is so, each register class that is used in a bitwise-and or shift 2361instruction must have a subclass consisting of registers from which 2362single-byte values can be loaded or stored. This is so that 2363@code{PREFERRED_RELOAD_CLASS} can always have a possible value to return. 2364 2365@deftp {Data type} {enum reg_class} 2366An enumerated type that must be defined with all the register class names 2367as enumerated values. @code{NO_REGS} must be first. @code{ALL_REGS} 2368must be the last register class, followed by one more enumerated value, 2369@code{LIM_REG_CLASSES}, which is not a register class but rather 2370tells how many classes there are. 2371 2372Each register class has a number, which is the value of casting 2373the class name to type @code{int}. The number serves as an index 2374in many of the tables described below. 2375@end deftp 2376 2377@defmac N_REG_CLASSES 2378The number of distinct register classes, defined as follows: 2379 2380@smallexample 2381#define N_REG_CLASSES (int) LIM_REG_CLASSES 2382@end smallexample 2383@end defmac 2384 2385@defmac REG_CLASS_NAMES 2386An initializer containing the names of the register classes as C string 2387constants. These names are used in writing some of the debugging dumps. 2388@end defmac 2389 2390@defmac REG_CLASS_CONTENTS 2391An initializer containing the contents of the register classes, as integers 2392which are bit masks. The @var{n}th integer specifies the contents of class 2393@var{n}. The way the integer @var{mask} is interpreted is that 2394register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1. 2395 2396When the machine has more than 32 registers, an integer does not suffice. 2397Then the integers are replaced by sub-initializers, braced groupings containing 2398several integers. Each sub-initializer must be suitable as an initializer 2399for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}. 2400In this situation, the first integer in each sub-initializer corresponds to 2401registers 0 through 31, the second integer to registers 32 through 63, and 2402so on. 2403@end defmac 2404 2405@defmac REGNO_REG_CLASS (@var{regno}) 2406A C expression whose value is a register class containing hard register 2407@var{regno}. In general there is more than one such class; choose a class 2408which is @dfn{minimal}, meaning that no smaller class also contains the 2409register. 2410@end defmac 2411 2412@defmac BASE_REG_CLASS 2413A macro whose definition is the name of the class to which a valid 2414base register must belong. A base register is one used in an address 2415which is the register value plus a displacement. 2416@end defmac 2417 2418@defmac MODE_BASE_REG_CLASS (@var{mode}) 2419This is a variation of the @code{BASE_REG_CLASS} macro which allows 2420the selection of a base register in a mode dependent manner. If 2421@var{mode} is VOIDmode then it should return the same value as 2422@code{BASE_REG_CLASS}. 2423@end defmac 2424 2425@defmac MODE_BASE_REG_REG_CLASS (@var{mode}) 2426A C expression whose value is the register class to which a valid 2427base register must belong in order to be used in a base plus index 2428register address. You should define this macro if base plus index 2429addresses have different requirements than other base register uses. 2430@end defmac 2431 2432@defmac MODE_CODE_BASE_REG_CLASS (@var{mode}, @var{address_space}, @var{outer_code}, @var{index_code}) 2433A C expression whose value is the register class to which a valid 2434base register for a memory reference in mode @var{mode} to address 2435space @var{address_space} must belong. @var{outer_code} and @var{index_code} 2436define the context in which the base register occurs. @var{outer_code} is 2437the code of the immediately enclosing expression (@code{MEM} for the top level 2438of an address, @code{ADDRESS} for something that occurs in an 2439@code{address_operand}). @var{index_code} is the code of the corresponding 2440index expression if @var{outer_code} is @code{PLUS}; @code{SCRATCH} otherwise. 2441@end defmac 2442 2443@defmac INDEX_REG_CLASS 2444A macro whose definition is the name of the class to which a valid 2445index register must belong. An index register is one used in an 2446address where its value is either multiplied by a scale factor or 2447added to another register (as well as added to a displacement). 2448@end defmac 2449 2450@defmac REGNO_OK_FOR_BASE_P (@var{num}) 2451A C expression which is nonzero if register number @var{num} is 2452suitable for use as a base register in operand addresses. 2453@end defmac 2454 2455@defmac REGNO_MODE_OK_FOR_BASE_P (@var{num}, @var{mode}) 2456A C expression that is just like @code{REGNO_OK_FOR_BASE_P}, except that 2457that expression may examine the mode of the memory reference in 2458@var{mode}. You should define this macro if the mode of the memory 2459reference affects whether a register may be used as a base register. If 2460you define this macro, the compiler will use it instead of 2461@code{REGNO_OK_FOR_BASE_P}. The mode may be @code{VOIDmode} for 2462addresses that appear outside a @code{MEM}, i.e., as an 2463@code{address_operand}. 2464@end defmac 2465 2466@defmac REGNO_MODE_OK_FOR_REG_BASE_P (@var{num}, @var{mode}) 2467A C expression which is nonzero if register number @var{num} is suitable for 2468use as a base register in base plus index operand addresses, accessing 2469memory in mode @var{mode}. It may be either a suitable hard register or a 2470pseudo register that has been allocated such a hard register. You should 2471define this macro if base plus index addresses have different requirements 2472than other base register uses. 2473 2474Use of this macro is deprecated; please use the more general 2475@code{REGNO_MODE_CODE_OK_FOR_BASE_P}. 2476@end defmac 2477 2478@defmac REGNO_MODE_CODE_OK_FOR_BASE_P (@var{num}, @var{mode}, @var{address_space}, @var{outer_code}, @var{index_code}) 2479A C expression which is nonzero if register number @var{num} is 2480suitable for use as a base register in operand addresses, accessing 2481memory in mode @var{mode} in address space @var{address_space}. 2482This is similar to @code{REGNO_MODE_OK_FOR_BASE_P}, except 2483that that expression may examine the context in which the register 2484appears in the memory reference. @var{outer_code} is the code of the 2485immediately enclosing expression (@code{MEM} if at the top level of the 2486address, @code{ADDRESS} for something that occurs in an 2487@code{address_operand}). @var{index_code} is the code of the 2488corresponding index expression if @var{outer_code} is @code{PLUS}; 2489@code{SCRATCH} otherwise. The mode may be @code{VOIDmode} for addresses 2490that appear outside a @code{MEM}, i.e., as an @code{address_operand}. 2491@end defmac 2492 2493@defmac REGNO_OK_FOR_INDEX_P (@var{num}) 2494A C expression which is nonzero if register number @var{num} is 2495suitable for use as an index register in operand addresses. It may be 2496either a suitable hard register or a pseudo register that has been 2497allocated such a hard register. 2498 2499The difference between an index register and a base register is that 2500the index register may be scaled. If an address involves the sum of 2501two registers, neither one of them scaled, then either one may be 2502labeled the ``base'' and the other the ``index''; but whichever 2503labeling is used must fit the machine's constraints of which registers 2504may serve in each capacity. The compiler will try both labelings, 2505looking for one that is valid, and will reload one or both registers 2506only if neither labeling works. 2507@end defmac 2508 2509@hook TARGET_PREFERRED_RENAME_CLASS 2510 2511@hook TARGET_PREFERRED_RELOAD_CLASS 2512A target hook that places additional restrictions on the register class 2513to use when it is necessary to copy value @var{x} into a register in class 2514@var{rclass}. The value is a register class; perhaps @var{rclass}, or perhaps 2515another, smaller class. 2516 2517The default version of this hook always returns value of @code{rclass} argument. 2518 2519Sometimes returning a more restrictive class makes better code. For 2520example, on the 68000, when @var{x} is an integer constant that is in range 2521for a @samp{moveq} instruction, the value of this macro is always 2522@code{DATA_REGS} as long as @var{rclass} includes the data registers. 2523Requiring a data register guarantees that a @samp{moveq} will be used. 2524 2525One case where @code{TARGET_PREFERRED_RELOAD_CLASS} must not return 2526@var{rclass} is if @var{x} is a legitimate constant which cannot be 2527loaded into some register class. By returning @code{NO_REGS} you can 2528force @var{x} into a memory location. For example, rs6000 can load 2529immediate values into general-purpose registers, but does not have an 2530instruction for loading an immediate value into a floating-point 2531register, so @code{TARGET_PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when 2532@var{x} is a floating-point constant. If the constant can't be loaded 2533into any kind of register, code generation will be better if 2534@code{TARGET_LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead 2535of using @code{TARGET_PREFERRED_RELOAD_CLASS}. 2536 2537If an insn has pseudos in it after register allocation, reload will go 2538through the alternatives and call repeatedly @code{TARGET_PREFERRED_RELOAD_CLASS} 2539to find the best one. Returning @code{NO_REGS}, in this case, makes 2540reload add a @code{!} in front of the constraint: the x86 back-end uses 2541this feature to discourage usage of 387 registers when math is done in 2542the SSE registers (and vice versa). 2543@end deftypefn 2544 2545@defmac PREFERRED_RELOAD_CLASS (@var{x}, @var{class}) 2546A C expression that places additional restrictions on the register class 2547to use when it is necessary to copy value @var{x} into a register in class 2548@var{class}. The value is a register class; perhaps @var{class}, or perhaps 2549another, smaller class. On many machines, the following definition is 2550safe: 2551 2552@smallexample 2553#define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS 2554@end smallexample 2555 2556Sometimes returning a more restrictive class makes better code. For 2557example, on the 68000, when @var{x} is an integer constant that is in range 2558for a @samp{moveq} instruction, the value of this macro is always 2559@code{DATA_REGS} as long as @var{class} includes the data registers. 2560Requiring a data register guarantees that a @samp{moveq} will be used. 2561 2562One case where @code{PREFERRED_RELOAD_CLASS} must not return 2563@var{class} is if @var{x} is a legitimate constant which cannot be 2564loaded into some register class. By returning @code{NO_REGS} you can 2565force @var{x} into a memory location. For example, rs6000 can load 2566immediate values into general-purpose registers, but does not have an 2567instruction for loading an immediate value into a floating-point 2568register, so @code{PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when 2569@var{x} is a floating-point constant. If the constant can't be loaded 2570into any kind of register, code generation will be better if 2571@code{TARGET_LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead 2572of using @code{TARGET_PREFERRED_RELOAD_CLASS}. 2573 2574If an insn has pseudos in it after register allocation, reload will go 2575through the alternatives and call repeatedly @code{PREFERRED_RELOAD_CLASS} 2576to find the best one. Returning @code{NO_REGS}, in this case, makes 2577reload add a @code{!} in front of the constraint: the x86 back-end uses 2578this feature to discourage usage of 387 registers when math is done in 2579the SSE registers (and vice versa). 2580@end defmac 2581 2582@hook TARGET_PREFERRED_OUTPUT_RELOAD_CLASS 2583Like @code{TARGET_PREFERRED_RELOAD_CLASS}, but for output reloads instead of 2584input reloads. 2585 2586The default version of this hook always returns value of @code{rclass} 2587argument. 2588 2589You can also use @code{TARGET_PREFERRED_OUTPUT_RELOAD_CLASS} to discourage 2590reload from using some alternatives, like @code{TARGET_PREFERRED_RELOAD_CLASS}. 2591@end deftypefn 2592 2593@defmac LIMIT_RELOAD_CLASS (@var{mode}, @var{class}) 2594A C expression that places additional restrictions on the register class 2595to use when it is necessary to be able to hold a value of mode 2596@var{mode} in a reload register for which class @var{class} would 2597ordinarily be used. 2598 2599Unlike @code{PREFERRED_RELOAD_CLASS}, this macro should be used when 2600there are certain modes that simply can't go in certain reload classes. 2601 2602The value is a register class; perhaps @var{class}, or perhaps another, 2603smaller class. 2604 2605Don't define this macro unless the target machine has limitations which 2606require the macro to do something nontrivial. 2607@end defmac 2608 2609@hook TARGET_SECONDARY_RELOAD 2610Many machines have some registers that cannot be copied directly to or 2611from memory or even from other types of registers. An example is the 2612@samp{MQ} register, which on most machines, can only be copied to or 2613from general registers, but not memory. Below, we shall be using the 2614term 'intermediate register' when a move operation cannot be performed 2615directly, but has to be done by copying the source into the intermediate 2616register first, and then copying the intermediate register to the 2617destination. An intermediate register always has the same mode as 2618source and destination. Since it holds the actual value being copied, 2619reload might apply optimizations to re-use an intermediate register 2620and eliding the copy from the source when it can determine that the 2621intermediate register still holds the required value. 2622 2623Another kind of secondary reload is required on some machines which 2624allow copying all registers to and from memory, but require a scratch 2625register for stores to some memory locations (e.g., those with symbolic 2626address on the RT, and those with certain symbolic address on the SPARC 2627when compiling PIC)@. Scratch registers need not have the same mode 2628as the value being copied, and usually hold a different value than 2629that being copied. Special patterns in the md file are needed to 2630describe how the copy is performed with the help of the scratch register; 2631these patterns also describe the number, register class(es) and mode(s) 2632of the scratch register(s). 2633 2634In some cases, both an intermediate and a scratch register are required. 2635 2636For input reloads, this target hook is called with nonzero @var{in_p}, 2637and @var{x} is an rtx that needs to be copied to a register of class 2638@var{reload_class} in @var{reload_mode}. For output reloads, this target 2639hook is called with zero @var{in_p}, and a register of class @var{reload_class} 2640needs to be copied to rtx @var{x} in @var{reload_mode}. 2641 2642If copying a register of @var{reload_class} from/to @var{x} requires 2643an intermediate register, the hook @code{secondary_reload} should 2644return the register class required for this intermediate register. 2645If no intermediate register is required, it should return NO_REGS. 2646If more than one intermediate register is required, describe the one 2647that is closest in the copy chain to the reload register. 2648 2649If scratch registers are needed, you also have to describe how to 2650perform the copy from/to the reload register to/from this 2651closest intermediate register. Or if no intermediate register is 2652required, but still a scratch register is needed, describe the 2653copy from/to the reload register to/from the reload operand @var{x}. 2654 2655You do this by setting @code{sri->icode} to the instruction code of a pattern 2656in the md file which performs the move. Operands 0 and 1 are the output 2657and input of this copy, respectively. Operands from operand 2 onward are 2658for scratch operands. These scratch operands must have a mode, and a 2659single-register-class 2660@c [later: or memory] 2661output constraint. 2662 2663When an intermediate register is used, the @code{secondary_reload} 2664hook will be called again to determine how to copy the intermediate 2665register to/from the reload operand @var{x}, so your hook must also 2666have code to handle the register class of the intermediate operand. 2667 2668@c [For later: maybe we'll allow multi-alternative reload patterns - 2669@c the port maintainer could name a mov<mode> pattern that has clobbers - 2670@c and match the constraints of input and output to determine the required 2671@c alternative. A restriction would be that constraints used to match 2672@c against reloads registers would have to be written as register class 2673@c constraints, or we need a new target macro / hook that tells us if an 2674@c arbitrary constraint can match an unknown register of a given class. 2675@c Such a macro / hook would also be useful in other places.] 2676 2677 2678@var{x} might be a pseudo-register or a @code{subreg} of a 2679pseudo-register, which could either be in a hard register or in memory. 2680Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is 2681in memory and the hard register number if it is in a register. 2682 2683Scratch operands in memory (constraint @code{"=m"} / @code{"=&m"}) are 2684currently not supported. For the time being, you will have to continue 2685to use @code{SECONDARY_MEMORY_NEEDED} for that purpose. 2686 2687@code{copy_cost} also uses this target hook to find out how values are 2688copied. If you want it to include some extra cost for the need to allocate 2689(a) scratch register(s), set @code{sri->extra_cost} to the additional cost. 2690Or if two dependent moves are supposed to have a lower cost than the sum 2691of the individual moves due to expected fortuitous scheduling and/or special 2692forwarding logic, you can set @code{sri->extra_cost} to a negative amount. 2693@end deftypefn 2694 2695@defmac SECONDARY_RELOAD_CLASS (@var{class}, @var{mode}, @var{x}) 2696@defmacx SECONDARY_INPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x}) 2697@defmacx SECONDARY_OUTPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x}) 2698These macros are obsolete, new ports should use the target hook 2699@code{TARGET_SECONDARY_RELOAD} instead. 2700 2701These are obsolete macros, replaced by the @code{TARGET_SECONDARY_RELOAD} 2702target hook. Older ports still define these macros to indicate to the 2703reload phase that it may 2704need to allocate at least one register for a reload in addition to the 2705register to contain the data. Specifically, if copying @var{x} to a 2706register @var{class} in @var{mode} requires an intermediate register, 2707you were supposed to define @code{SECONDARY_INPUT_RELOAD_CLASS} to return the 2708largest register class all of whose registers can be used as 2709intermediate registers or scratch registers. 2710 2711If copying a register @var{class} in @var{mode} to @var{x} requires an 2712intermediate or scratch register, @code{SECONDARY_OUTPUT_RELOAD_CLASS} 2713was supposed to be defined be defined to return the largest register 2714class required. If the 2715requirements for input and output reloads were the same, the macro 2716@code{SECONDARY_RELOAD_CLASS} should have been used instead of defining both 2717macros identically. 2718 2719The values returned by these macros are often @code{GENERAL_REGS}. 2720Return @code{NO_REGS} if no spare register is needed; i.e., if @var{x} 2721can be directly copied to or from a register of @var{class} in 2722@var{mode} without requiring a scratch register. Do not define this 2723macro if it would always return @code{NO_REGS}. 2724 2725If a scratch register is required (either with or without an 2726intermediate register), you were supposed to define patterns for 2727@samp{reload_in@var{m}} or @samp{reload_out@var{m}}, as required 2728(@pxref{Standard Names}. These patterns, which were normally 2729implemented with a @code{define_expand}, should be similar to the 2730@samp{mov@var{m}} patterns, except that operand 2 is the scratch 2731register. 2732 2733These patterns need constraints for the reload register and scratch 2734register that 2735contain a single register class. If the original reload register (whose 2736class is @var{class}) can meet the constraint given in the pattern, the 2737value returned by these macros is used for the class of the scratch 2738register. Otherwise, two additional reload registers are required. 2739Their classes are obtained from the constraints in the insn pattern. 2740 2741@var{x} might be a pseudo-register or a @code{subreg} of a 2742pseudo-register, which could either be in a hard register or in memory. 2743Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is 2744in memory and the hard register number if it is in a register. 2745 2746These macros should not be used in the case where a particular class of 2747registers can only be copied to memory and not to another class of 2748registers. In that case, secondary reload registers are not needed and 2749would not be helpful. Instead, a stack location must be used to perform 2750the copy and the @code{mov@var{m}} pattern should use memory as an 2751intermediate storage. This case often occurs between floating-point and 2752general registers. 2753@end defmac 2754 2755@defmac SECONDARY_MEMORY_NEEDED (@var{class1}, @var{class2}, @var{m}) 2756Certain machines have the property that some registers cannot be copied 2757to some other registers without using memory. Define this macro on 2758those machines to be a C expression that is nonzero if objects of mode 2759@var{m} in registers of @var{class1} can only be copied to registers of 2760class @var{class2} by storing a register of @var{class1} into memory 2761and loading that memory location into a register of @var{class2}. 2762 2763Do not define this macro if its value would always be zero. 2764@end defmac 2765 2766@defmac SECONDARY_MEMORY_NEEDED_RTX (@var{mode}) 2767Normally when @code{SECONDARY_MEMORY_NEEDED} is defined, the compiler 2768allocates a stack slot for a memory location needed for register copies. 2769If this macro is defined, the compiler instead uses the memory location 2770defined by this macro. 2771 2772Do not define this macro if you do not define 2773@code{SECONDARY_MEMORY_NEEDED}. 2774@end defmac 2775 2776@defmac SECONDARY_MEMORY_NEEDED_MODE (@var{mode}) 2777When the compiler needs a secondary memory location to copy between two 2778registers of mode @var{mode}, it normally allocates sufficient memory to 2779hold a quantity of @code{BITS_PER_WORD} bits and performs the store and 2780load operations in a mode that many bits wide and whose class is the 2781same as that of @var{mode}. 2782 2783This is right thing to do on most machines because it ensures that all 2784bits of the register are copied and prevents accesses to the registers 2785in a narrower mode, which some machines prohibit for floating-point 2786registers. 2787 2788However, this default behavior is not correct on some machines, such as 2789the DEC Alpha, that store short integers in floating-point registers 2790differently than in integer registers. On those machines, the default 2791widening will not work correctly and you must define this macro to 2792suppress that widening in some cases. See the file @file{alpha.h} for 2793details. 2794 2795Do not define this macro if you do not define 2796@code{SECONDARY_MEMORY_NEEDED} or if widening @var{mode} to a mode that 2797is @code{BITS_PER_WORD} bits wide is correct for your machine. 2798@end defmac 2799 2800@hook TARGET_CLASS_LIKELY_SPILLED_P 2801A target hook which returns @code{true} if pseudos that have been assigned 2802to registers of class @var{rclass} would likely be spilled because 2803registers of @var{rclass} are needed for spill registers. 2804 2805The default version of this target hook returns @code{true} if @var{rclass} 2806has exactly one register and @code{false} otherwise. On most machines, this 2807default should be used. Only use this target hook to some other expression 2808if pseudos allocated by @file{local-alloc.c} end up in memory because their 2809hard registers were needed for spill registers. If this target hook returns 2810@code{false} for those classes, those pseudos will only be allocated by 2811@file{global.c}, which knows how to reallocate the pseudo to another 2812register. If there would not be another register available for reallocation, 2813you should not change the implementation of this target hook since 2814the only effect of such implementation would be to slow down register 2815allocation. 2816@end deftypefn 2817 2818@hook TARGET_CLASS_MAX_NREGS 2819A target hook returns the maximum number of consecutive registers 2820of class @var{rclass} needed to hold a value of mode @var{mode}. 2821 2822This is closely related to the macro @code{HARD_REGNO_NREGS}. In fact, 2823the value returned by @code{TARGET_CLASS_MAX_NREGS (@var{rclass}, 2824@var{mode})} target hook should be the maximum value of 2825@code{HARD_REGNO_NREGS (@var{regno}, @var{mode})} for all @var{regno} 2826values in the class @var{rclass}. 2827 2828This target hook helps control the handling of multiple-word values 2829in the reload pass. 2830 2831The default version of this target hook returns the size of @var{mode} 2832in words. 2833@end deftypefn 2834 2835@defmac CLASS_MAX_NREGS (@var{class}, @var{mode}) 2836A C expression for the maximum number of consecutive registers 2837of class @var{class} needed to hold a value of mode @var{mode}. 2838 2839This is closely related to the macro @code{HARD_REGNO_NREGS}. In fact, 2840the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})} 2841should be the maximum value of @code{HARD_REGNO_NREGS (@var{regno}, 2842@var{mode})} for all @var{regno} values in the class @var{class}. 2843 2844This macro helps control the handling of multiple-word values 2845in the reload pass. 2846@end defmac 2847 2848@defmac CANNOT_CHANGE_MODE_CLASS (@var{from}, @var{to}, @var{class}) 2849If defined, a C expression that returns nonzero for a @var{class} for which 2850a change from mode @var{from} to mode @var{to} is invalid. 2851 2852For the example, loading 32-bit integer or floating-point objects into 2853floating-point registers on the Alpha extends them to 64 bits. 2854Therefore loading a 64-bit object and then storing it as a 32-bit object 2855does not store the low-order 32 bits, as would be the case for a normal 2856register. Therefore, @file{alpha.h} defines @code{CANNOT_CHANGE_MODE_CLASS} 2857as below: 2858 2859@smallexample 2860#define CANNOT_CHANGE_MODE_CLASS(FROM, TO, CLASS) \ 2861 (GET_MODE_SIZE (FROM) != GET_MODE_SIZE (TO) \ 2862 ? reg_classes_intersect_p (FLOAT_REGS, (CLASS)) : 0) 2863@end smallexample 2864@end defmac 2865 2866@node Old Constraints 2867@section Obsolete Macros for Defining Constraints 2868@cindex defining constraints, obsolete method 2869@cindex constraints, defining, obsolete method 2870 2871Machine-specific constraints can be defined with these macros instead 2872of the machine description constructs described in @ref{Define 2873Constraints}. This mechanism is obsolete. New ports should not use 2874it; old ports should convert to the new mechanism. 2875 2876@defmac CONSTRAINT_LEN (@var{char}, @var{str}) 2877For the constraint at the start of @var{str}, which starts with the letter 2878@var{c}, return the length. This allows you to have register class / 2879constant / extra constraints that are longer than a single letter; 2880you don't need to define this macro if you can do with single-letter 2881constraints only. The definition of this macro should use 2882DEFAULT_CONSTRAINT_LEN for all the characters that you don't want 2883to handle specially. 2884There are some sanity checks in genoutput.c that check the constraint lengths 2885for the md file, so you can also use this macro to help you while you are 2886transitioning from a byzantine single-letter-constraint scheme: when you 2887return a negative length for a constraint you want to re-use, genoutput 2888will complain about every instance where it is used in the md file. 2889@end defmac 2890 2891@defmac REG_CLASS_FROM_LETTER (@var{char}) 2892A C expression which defines the machine-dependent operand constraint 2893letters for register classes. If @var{char} is such a letter, the 2894value should be the register class corresponding to it. Otherwise, 2895the value should be @code{NO_REGS}. The register letter @samp{r}, 2896corresponding to class @code{GENERAL_REGS}, will not be passed 2897to this macro; you do not need to handle it. 2898@end defmac 2899 2900@defmac REG_CLASS_FROM_CONSTRAINT (@var{char}, @var{str}) 2901Like @code{REG_CLASS_FROM_LETTER}, but you also get the constraint string 2902passed in @var{str}, so that you can use suffixes to distinguish between 2903different variants. 2904@end defmac 2905 2906@defmac CONST_OK_FOR_LETTER_P (@var{value}, @var{c}) 2907A C expression that defines the machine-dependent operand constraint 2908letters (@samp{I}, @samp{J}, @samp{K}, @dots{} @samp{P}) that specify 2909particular ranges of integer values. If @var{c} is one of those 2910letters, the expression should check that @var{value}, an integer, is in 2911the appropriate range and return 1 if so, 0 otherwise. If @var{c} is 2912not one of those letters, the value should be 0 regardless of 2913@var{value}. 2914@end defmac 2915 2916@defmac CONST_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str}) 2917Like @code{CONST_OK_FOR_LETTER_P}, but you also get the constraint 2918string passed in @var{str}, so that you can use suffixes to distinguish 2919between different variants. 2920@end defmac 2921 2922@defmac CONST_DOUBLE_OK_FOR_LETTER_P (@var{value}, @var{c}) 2923A C expression that defines the machine-dependent operand constraint 2924letters that specify particular ranges of @code{const_double} values 2925(@samp{G} or @samp{H}). 2926 2927If @var{c} is one of those letters, the expression should check that 2928@var{value}, an RTX of code @code{const_double}, is in the appropriate 2929range and return 1 if so, 0 otherwise. If @var{c} is not one of those 2930letters, the value should be 0 regardless of @var{value}. 2931 2932@code{const_double} is used for all floating-point constants and for 2933@code{DImode} fixed-point constants. A given letter can accept either 2934or both kinds of values. It can use @code{GET_MODE} to distinguish 2935between these kinds. 2936@end defmac 2937 2938@defmac CONST_DOUBLE_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str}) 2939Like @code{CONST_DOUBLE_OK_FOR_LETTER_P}, but you also get the constraint 2940string passed in @var{str}, so that you can use suffixes to distinguish 2941between different variants. 2942@end defmac 2943 2944@defmac EXTRA_CONSTRAINT (@var{value}, @var{c}) 2945A C expression that defines the optional machine-dependent constraint 2946letters that can be used to segregate specific types of operands, usually 2947memory references, for the target machine. Any letter that is not 2948elsewhere defined and not matched by @code{REG_CLASS_FROM_LETTER} / 2949@code{REG_CLASS_FROM_CONSTRAINT} 2950may be used. Normally this macro will not be defined. 2951 2952If it is required for a particular target machine, it should return 1 2953if @var{value} corresponds to the operand type represented by the 2954constraint letter @var{c}. If @var{c} is not defined as an extra 2955constraint, the value returned should be 0 regardless of @var{value}. 2956 2957For example, on the ROMP, load instructions cannot have their output 2958in r0 if the memory reference contains a symbolic address. Constraint 2959letter @samp{Q} is defined as representing a memory address that does 2960@emph{not} contain a symbolic address. An alternative is specified with 2961a @samp{Q} constraint on the input and @samp{r} on the output. The next 2962alternative specifies @samp{m} on the input and a register class that 2963does not include r0 on the output. 2964@end defmac 2965 2966@defmac EXTRA_CONSTRAINT_STR (@var{value}, @var{c}, @var{str}) 2967Like @code{EXTRA_CONSTRAINT}, but you also get the constraint string passed 2968in @var{str}, so that you can use suffixes to distinguish between different 2969variants. 2970@end defmac 2971 2972@defmac EXTRA_MEMORY_CONSTRAINT (@var{c}, @var{str}) 2973A C expression that defines the optional machine-dependent constraint 2974letters, amongst those accepted by @code{EXTRA_CONSTRAINT}, that should 2975be treated like memory constraints by the reload pass. 2976 2977It should return 1 if the operand type represented by the constraint 2978at the start of @var{str}, the first letter of which is the letter @var{c}, 2979comprises a subset of all memory references including 2980all those whose address is simply a base register. This allows the reload 2981pass to reload an operand, if it does not directly correspond to the operand 2982type of @var{c}, by copying its address into a base register. 2983 2984For example, on the S/390, some instructions do not accept arbitrary 2985memory references, but only those that do not make use of an index 2986register. The constraint letter @samp{Q} is defined via 2987@code{EXTRA_CONSTRAINT} as representing a memory address of this type. 2988If the letter @samp{Q} is marked as @code{EXTRA_MEMORY_CONSTRAINT}, 2989a @samp{Q} constraint can handle any memory operand, because the 2990reload pass knows it can be reloaded by copying the memory address 2991into a base register if required. This is analogous to the way 2992an @samp{o} constraint can handle any memory operand. 2993@end defmac 2994 2995@defmac EXTRA_ADDRESS_CONSTRAINT (@var{c}, @var{str}) 2996A C expression that defines the optional machine-dependent constraint 2997letters, amongst those accepted by @code{EXTRA_CONSTRAINT} / 2998@code{EXTRA_CONSTRAINT_STR}, that should 2999be treated like address constraints by the reload pass. 3000 3001It should return 1 if the operand type represented by the constraint 3002at the start of @var{str}, which starts with the letter @var{c}, comprises 3003a subset of all memory addresses including 3004all those that consist of just a base register. This allows the reload 3005pass to reload an operand, if it does not directly correspond to the operand 3006type of @var{str}, by copying it into a base register. 3007 3008Any constraint marked as @code{EXTRA_ADDRESS_CONSTRAINT} can only 3009be used with the @code{address_operand} predicate. It is treated 3010analogously to the @samp{p} constraint. 3011@end defmac 3012 3013@node Stack and Calling 3014@section Stack Layout and Calling Conventions 3015@cindex calling conventions 3016 3017@c prevent bad page break with this line 3018This describes the stack layout and calling conventions. 3019 3020@menu 3021* Frame Layout:: 3022* Exception Handling:: 3023* Stack Checking:: 3024* Frame Registers:: 3025* Elimination:: 3026* Stack Arguments:: 3027* Register Arguments:: 3028* Scalar Return:: 3029* Aggregate Return:: 3030* Caller Saves:: 3031* Function Entry:: 3032* Profiling:: 3033* Tail Calls:: 3034* Stack Smashing Protection:: 3035@end menu 3036 3037@node Frame Layout 3038@subsection Basic Stack Layout 3039@cindex stack frame layout 3040@cindex frame layout 3041 3042@c prevent bad page break with this line 3043Here is the basic stack layout. 3044 3045@defmac STACK_GROWS_DOWNWARD 3046Define this macro if pushing a word onto the stack moves the stack 3047pointer to a smaller address. 3048 3049When we say, ``define this macro if @dots{}'', it means that the 3050compiler checks this macro only with @code{#ifdef} so the precise 3051definition used does not matter. 3052@end defmac 3053 3054@defmac STACK_PUSH_CODE 3055This macro defines the operation used when something is pushed 3056on the stack. In RTL, a push operation will be 3057@code{(set (mem (STACK_PUSH_CODE (reg sp))) @dots{})} 3058 3059The choices are @code{PRE_DEC}, @code{POST_DEC}, @code{PRE_INC}, 3060and @code{POST_INC}. Which of these is correct depends on 3061the stack direction and on whether the stack pointer points 3062to the last item on the stack or whether it points to the 3063space for the next item on the stack. 3064 3065The default is @code{PRE_DEC} when @code{STACK_GROWS_DOWNWARD} is 3066defined, which is almost always right, and @code{PRE_INC} otherwise, 3067which is often wrong. 3068@end defmac 3069 3070@defmac FRAME_GROWS_DOWNWARD 3071Define this macro to nonzero value if the addresses of local variable slots 3072are at negative offsets from the frame pointer. 3073@end defmac 3074 3075@defmac ARGS_GROW_DOWNWARD 3076Define this macro if successive arguments to a function occupy decreasing 3077addresses on the stack. 3078@end defmac 3079 3080@defmac STARTING_FRAME_OFFSET 3081Offset from the frame pointer to the first local variable slot to be allocated. 3082 3083If @code{FRAME_GROWS_DOWNWARD}, find the next slot's offset by 3084subtracting the first slot's length from @code{STARTING_FRAME_OFFSET}. 3085Otherwise, it is found by adding the length of the first slot to the 3086value @code{STARTING_FRAME_OFFSET}. 3087@c i'm not sure if the above is still correct.. had to change it to get 3088@c rid of an overfull. --mew 2feb93 3089@end defmac 3090 3091@defmac STACK_ALIGNMENT_NEEDED 3092Define to zero to disable final alignment of the stack during reload. 3093The nonzero default for this macro is suitable for most ports. 3094 3095On ports where @code{STARTING_FRAME_OFFSET} is nonzero or where there 3096is a register save block following the local block that doesn't require 3097alignment to @code{STACK_BOUNDARY}, it may be beneficial to disable 3098stack alignment and do it in the backend. 3099@end defmac 3100 3101@defmac STACK_POINTER_OFFSET 3102Offset from the stack pointer register to the first location at which 3103outgoing arguments are placed. If not specified, the default value of 3104zero is used. This is the proper value for most machines. 3105 3106If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above 3107the first location at which outgoing arguments are placed. 3108@end defmac 3109 3110@defmac FIRST_PARM_OFFSET (@var{fundecl}) 3111Offset from the argument pointer register to the first argument's 3112address. On some machines it may depend on the data type of the 3113function. 3114 3115If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above 3116the first argument's address. 3117@end defmac 3118 3119@defmac STACK_DYNAMIC_OFFSET (@var{fundecl}) 3120Offset from the stack pointer register to an item dynamically allocated 3121on the stack, e.g., by @code{alloca}. 3122 3123The default value for this macro is @code{STACK_POINTER_OFFSET} plus the 3124length of the outgoing arguments. The default is correct for most 3125machines. See @file{function.c} for details. 3126@end defmac 3127 3128@defmac INITIAL_FRAME_ADDRESS_RTX 3129A C expression whose value is RTL representing the address of the initial 3130stack frame. This address is passed to @code{RETURN_ADDR_RTX} and 3131@code{DYNAMIC_CHAIN_ADDRESS}. If you don't define this macro, a reasonable 3132default value will be used. Define this macro in order to make frame pointer 3133elimination work in the presence of @code{__builtin_frame_address (count)} and 3134@code{__builtin_return_address (count)} for @code{count} not equal to zero. 3135@end defmac 3136 3137@defmac DYNAMIC_CHAIN_ADDRESS (@var{frameaddr}) 3138A C expression whose value is RTL representing the address in a stack 3139frame where the pointer to the caller's frame is stored. Assume that 3140@var{frameaddr} is an RTL expression for the address of the stack frame 3141itself. 3142 3143If you don't define this macro, the default is to return the value 3144of @var{frameaddr}---that is, the stack frame address is also the 3145address of the stack word that points to the previous frame. 3146@end defmac 3147 3148@defmac SETUP_FRAME_ADDRESSES 3149If defined, a C expression that produces the machine-specific code to 3150setup the stack so that arbitrary frames can be accessed. For example, 3151on the SPARC, we must flush all of the register windows to the stack 3152before we can access arbitrary stack frames. You will seldom need to 3153define this macro. 3154@end defmac 3155 3156@hook TARGET_BUILTIN_SETJMP_FRAME_VALUE 3157This target hook should return an rtx that is used to store 3158the address of the current frame into the built in @code{setjmp} buffer. 3159The default value, @code{virtual_stack_vars_rtx}, is correct for most 3160machines. One reason you may need to define this target hook is if 3161@code{hard_frame_pointer_rtx} is the appropriate value on your machine. 3162@end deftypefn 3163 3164@defmac FRAME_ADDR_RTX (@var{frameaddr}) 3165A C expression whose value is RTL representing the value of the frame 3166address for the current frame. @var{frameaddr} is the frame pointer 3167of the current frame. This is used for __builtin_frame_address. 3168You need only define this macro if the frame address is not the same 3169as the frame pointer. Most machines do not need to define it. 3170@end defmac 3171 3172@defmac RETURN_ADDR_RTX (@var{count}, @var{frameaddr}) 3173A C expression whose value is RTL representing the value of the return 3174address for the frame @var{count} steps up from the current frame, after 3175the prologue. @var{frameaddr} is the frame pointer of the @var{count} 3176frame, or the frame pointer of the @var{count} @minus{} 1 frame if 3177@code{RETURN_ADDR_IN_PREVIOUS_FRAME} is defined. 3178 3179The value of the expression must always be the correct address when 3180@var{count} is zero, but may be @code{NULL_RTX} if there is no way to 3181determine the return address of other frames. 3182@end defmac 3183 3184@defmac RETURN_ADDR_IN_PREVIOUS_FRAME 3185Define this if the return address of a particular stack frame is accessed 3186from the frame pointer of the previous stack frame. 3187@end defmac 3188 3189@defmac INCOMING_RETURN_ADDR_RTX 3190A C expression whose value is RTL representing the location of the 3191incoming return address at the beginning of any function, before the 3192prologue. This RTL is either a @code{REG}, indicating that the return 3193value is saved in @samp{REG}, or a @code{MEM} representing a location in 3194the stack. 3195 3196You only need to define this macro if you want to support call frame 3197debugging information like that provided by DWARF 2. 3198 3199If this RTL is a @code{REG}, you should also define 3200@code{DWARF_FRAME_RETURN_COLUMN} to @code{DWARF_FRAME_REGNUM (REGNO)}. 3201@end defmac 3202 3203@defmac DWARF_ALT_FRAME_RETURN_COLUMN 3204A C expression whose value is an integer giving a DWARF 2 column 3205number that may be used as an alternative return column. The column 3206must not correspond to any gcc hard register (that is, it must not 3207be in the range of @code{DWARF_FRAME_REGNUM}). 3208 3209This macro can be useful if @code{DWARF_FRAME_RETURN_COLUMN} is set to a 3210general register, but an alternative column needs to be used for signal 3211frames. Some targets have also used different frame return columns 3212over time. 3213@end defmac 3214 3215@defmac DWARF_ZERO_REG 3216A C expression whose value is an integer giving a DWARF 2 register 3217number that is considered to always have the value zero. This should 3218only be defined if the target has an architected zero register, and 3219someone decided it was a good idea to use that register number to 3220terminate the stack backtrace. New ports should avoid this. 3221@end defmac 3222 3223@hook TARGET_DWARF_HANDLE_FRAME_UNSPEC 3224This target hook allows the backend to emit frame-related insns that 3225contain UNSPECs or UNSPEC_VOLATILEs. The DWARF 2 call frame debugging 3226info engine will invoke it on insns of the form 3227@smallexample 3228(set (reg) (unspec [@dots{}] UNSPEC_INDEX)) 3229@end smallexample 3230and 3231@smallexample 3232(set (reg) (unspec_volatile [@dots{}] UNSPECV_INDEX)). 3233@end smallexample 3234to let the backend emit the call frame instructions. @var{label} is 3235the CFI label attached to the insn, @var{pattern} is the pattern of 3236the insn and @var{index} is @code{UNSPEC_INDEX} or @code{UNSPECV_INDEX}. 3237@end deftypefn 3238 3239@defmac INCOMING_FRAME_SP_OFFSET 3240A C expression whose value is an integer giving the offset, in bytes, 3241from the value of the stack pointer register to the top of the stack 3242frame at the beginning of any function, before the prologue. The top of 3243the frame is defined to be the value of the stack pointer in the 3244previous frame, just before the call instruction. 3245 3246You only need to define this macro if you want to support call frame 3247debugging information like that provided by DWARF 2. 3248@end defmac 3249 3250@defmac ARG_POINTER_CFA_OFFSET (@var{fundecl}) 3251A C expression whose value is an integer giving the offset, in bytes, 3252from the argument pointer to the canonical frame address (cfa). The 3253final value should coincide with that calculated by 3254@code{INCOMING_FRAME_SP_OFFSET}. Which is unfortunately not usable 3255during virtual register instantiation. 3256 3257The default value for this macro is 3258@code{FIRST_PARM_OFFSET (fundecl) + crtl->args.pretend_args_size}, 3259which is correct for most machines; in general, the arguments are found 3260immediately before the stack frame. Note that this is not the case on 3261some targets that save registers into the caller's frame, such as SPARC 3262and rs6000, and so such targets need to define this macro. 3263 3264You only need to define this macro if the default is incorrect, and you 3265want to support call frame debugging information like that provided by 3266DWARF 2. 3267@end defmac 3268 3269@defmac FRAME_POINTER_CFA_OFFSET (@var{fundecl}) 3270If defined, a C expression whose value is an integer giving the offset 3271in bytes from the frame pointer to the canonical frame address (cfa). 3272The final value should coincide with that calculated by 3273@code{INCOMING_FRAME_SP_OFFSET}. 3274 3275Normally the CFA is calculated as an offset from the argument pointer, 3276via @code{ARG_POINTER_CFA_OFFSET}, but if the argument pointer is 3277variable due to the ABI, this may not be possible. If this macro is 3278defined, it implies that the virtual register instantiation should be 3279based on the frame pointer instead of the argument pointer. Only one 3280of @code{FRAME_POINTER_CFA_OFFSET} and @code{ARG_POINTER_CFA_OFFSET} 3281should be defined. 3282@end defmac 3283 3284@defmac CFA_FRAME_BASE_OFFSET (@var{fundecl}) 3285If defined, a C expression whose value is an integer giving the offset 3286in bytes from the canonical frame address (cfa) to the frame base used 3287in DWARF 2 debug information. The default is zero. A different value 3288may reduce the size of debug information on some ports. 3289@end defmac 3290 3291@node Exception Handling 3292@subsection Exception Handling Support 3293@cindex exception handling 3294 3295@defmac EH_RETURN_DATA_REGNO (@var{N}) 3296A C expression whose value is the @var{N}th register number used for 3297data by exception handlers, or @code{INVALID_REGNUM} if fewer than 3298@var{N} registers are usable. 3299 3300The exception handling library routines communicate with the exception 3301handlers via a set of agreed upon registers. Ideally these registers 3302should be call-clobbered; it is possible to use call-saved registers, 3303but may negatively impact code size. The target must support at least 33042 data registers, but should define 4 if there are enough free registers. 3305 3306You must define this macro if you want to support call frame exception 3307handling like that provided by DWARF 2. 3308@end defmac 3309 3310@defmac EH_RETURN_STACKADJ_RTX 3311A C expression whose value is RTL representing a location in which 3312to store a stack adjustment to be applied before function return. 3313This is used to unwind the stack to an exception handler's call frame. 3314It will be assigned zero on code paths that return normally. 3315 3316Typically this is a call-clobbered hard register that is otherwise 3317untouched by the epilogue, but could also be a stack slot. 3318 3319Do not define this macro if the stack pointer is saved and restored 3320by the regular prolog and epilog code in the call frame itself; in 3321this case, the exception handling library routines will update the 3322stack location to be restored in place. Otherwise, you must define 3323this macro if you want to support call frame exception handling like 3324that provided by DWARF 2. 3325@end defmac 3326 3327@defmac EH_RETURN_HANDLER_RTX 3328A C expression whose value is RTL representing a location in which 3329to store the address of an exception handler to which we should 3330return. It will not be assigned on code paths that return normally. 3331 3332Typically this is the location in the call frame at which the normal 3333return address is stored. For targets that return by popping an 3334address off the stack, this might be a memory address just below 3335the @emph{target} call frame rather than inside the current call 3336frame. If defined, @code{EH_RETURN_STACKADJ_RTX} will have already 3337been assigned, so it may be used to calculate the location of the 3338target call frame. 3339 3340Some targets have more complex requirements than storing to an 3341address calculable during initial code generation. In that case 3342the @code{eh_return} instruction pattern should be used instead. 3343 3344If you want to support call frame exception handling, you must 3345define either this macro or the @code{eh_return} instruction pattern. 3346@end defmac 3347 3348@defmac RETURN_ADDR_OFFSET 3349If defined, an integer-valued C expression for which rtl will be generated 3350to add it to the exception handler address before it is searched in the 3351exception handling tables, and to subtract it again from the address before 3352using it to return to the exception handler. 3353@end defmac 3354 3355@defmac ASM_PREFERRED_EH_DATA_FORMAT (@var{code}, @var{global}) 3356This macro chooses the encoding of pointers embedded in the exception 3357handling sections. If at all possible, this should be defined such 3358that the exception handling section will not require dynamic relocations, 3359and so may be read-only. 3360 3361@var{code} is 0 for data, 1 for code labels, 2 for function pointers. 3362@var{global} is true if the symbol may be affected by dynamic relocations. 3363The macro should return a combination of the @code{DW_EH_PE_*} defines 3364as found in @file{dwarf2.h}. 3365 3366If this macro is not defined, pointers will not be encoded but 3367represented directly. 3368@end defmac 3369 3370@defmac ASM_MAYBE_OUTPUT_ENCODED_ADDR_RTX (@var{file}, @var{encoding}, @var{size}, @var{addr}, @var{done}) 3371This macro allows the target to emit whatever special magic is required 3372to represent the encoding chosen by @code{ASM_PREFERRED_EH_DATA_FORMAT}. 3373Generic code takes care of pc-relative and indirect encodings; this must 3374be defined if the target uses text-relative or data-relative encodings. 3375 3376This is a C statement that branches to @var{done} if the format was 3377handled. @var{encoding} is the format chosen, @var{size} is the number 3378of bytes that the format occupies, @var{addr} is the @code{SYMBOL_REF} 3379to be emitted. 3380@end defmac 3381 3382@defmac MD_FALLBACK_FRAME_STATE_FOR (@var{context}, @var{fs}) 3383This macro allows the target to add CPU and operating system specific 3384code to the call-frame unwinder for use when there is no unwind data 3385available. The most common reason to implement this macro is to unwind 3386through signal frames. 3387 3388This macro is called from @code{uw_frame_state_for} in 3389@file{unwind-dw2.c}, @file{unwind-dw2-xtensa.c} and 3390@file{unwind-ia64.c}. @var{context} is an @code{_Unwind_Context}; 3391@var{fs} is an @code{_Unwind_FrameState}. Examine @code{context->ra} 3392for the address of the code being executed and @code{context->cfa} for 3393the stack pointer value. If the frame can be decoded, the register 3394save addresses should be updated in @var{fs} and the macro should 3395evaluate to @code{_URC_NO_REASON}. If the frame cannot be decoded, 3396the macro should evaluate to @code{_URC_END_OF_STACK}. 3397 3398For proper signal handling in Java this macro is accompanied by 3399@code{MAKE_THROW_FRAME}, defined in @file{libjava/include/*-signal.h} headers. 3400@end defmac 3401 3402@defmac MD_HANDLE_UNWABI (@var{context}, @var{fs}) 3403This macro allows the target to add operating system specific code to the 3404call-frame unwinder to handle the IA-64 @code{.unwabi} unwinding directive, 3405usually used for signal or interrupt frames. 3406 3407This macro is called from @code{uw_update_context} in @file{unwind-ia64.c}. 3408@var{context} is an @code{_Unwind_Context}; 3409@var{fs} is an @code{_Unwind_FrameState}. Examine @code{fs->unwabi} 3410for the abi and context in the @code{.unwabi} directive. If the 3411@code{.unwabi} directive can be handled, the register save addresses should 3412be updated in @var{fs}. 3413@end defmac 3414 3415@defmac TARGET_USES_WEAK_UNWIND_INFO 3416A C expression that evaluates to true if the target requires unwind 3417info to be given comdat linkage. Define it to be @code{1} if comdat 3418linkage is necessary. The default is @code{0}. 3419@end defmac 3420 3421@node Stack Checking 3422@subsection Specifying How Stack Checking is Done 3423 3424GCC will check that stack references are within the boundaries of the 3425stack, if the option @option{-fstack-check} is specified, in one of 3426three ways: 3427 3428@enumerate 3429@item 3430If the value of the @code{STACK_CHECK_BUILTIN} macro is nonzero, GCC 3431will assume that you have arranged for full stack checking to be done 3432at appropriate places in the configuration files. GCC will not do 3433other special processing. 3434 3435@item 3436If @code{STACK_CHECK_BUILTIN} is zero and the value of the 3437@code{STACK_CHECK_STATIC_BUILTIN} macro is nonzero, GCC will assume 3438that you have arranged for static stack checking (checking of the 3439static stack frame of functions) to be done at appropriate places 3440in the configuration files. GCC will only emit code to do dynamic 3441stack checking (checking on dynamic stack allocations) using the third 3442approach below. 3443 3444@item 3445If neither of the above are true, GCC will generate code to periodically 3446``probe'' the stack pointer using the values of the macros defined below. 3447@end enumerate 3448 3449If neither STACK_CHECK_BUILTIN nor STACK_CHECK_STATIC_BUILTIN is defined, 3450GCC will change its allocation strategy for large objects if the option 3451@option{-fstack-check} is specified: they will always be allocated 3452dynamically if their size exceeds @code{STACK_CHECK_MAX_VAR_SIZE} bytes. 3453 3454@defmac STACK_CHECK_BUILTIN 3455A nonzero value if stack checking is done by the configuration files in a 3456machine-dependent manner. You should define this macro if stack checking 3457is required by the ABI of your machine or if you would like to do stack 3458checking in some more efficient way than the generic approach. The default 3459value of this macro is zero. 3460@end defmac 3461 3462@defmac STACK_CHECK_STATIC_BUILTIN 3463A nonzero value if static stack checking is done by the configuration files 3464in a machine-dependent manner. You should define this macro if you would 3465like to do static stack checking in some more efficient way than the generic 3466approach. The default value of this macro is zero. 3467@end defmac 3468 3469@defmac STACK_CHECK_PROBE_INTERVAL_EXP 3470An integer specifying the interval at which GCC must generate stack probe 3471instructions, defined as 2 raised to this integer. You will normally 3472define this macro so that the interval be no larger than the size of 3473the ``guard pages'' at the end of a stack area. The default value 3474of 12 (4096-byte interval) is suitable for most systems. 3475@end defmac 3476 3477@defmac STACK_CHECK_MOVING_SP 3478An integer which is nonzero if GCC should move the stack pointer page by page 3479when doing probes. This can be necessary on systems where the stack pointer 3480contains the bottom address of the memory area accessible to the executing 3481thread at any point in time. In this situation an alternate signal stack 3482is required in order to be able to recover from a stack overflow. The 3483default value of this macro is zero. 3484@end defmac 3485 3486@defmac STACK_CHECK_PROTECT 3487The number of bytes of stack needed to recover from a stack overflow, for 3488languages where such a recovery is supported. The default value of 75 words 3489with the @code{setjmp}/@code{longjmp}-based exception handling mechanism and 34908192 bytes with other exception handling mechanisms should be adequate for 3491most machines. 3492@end defmac 3493 3494The following macros are relevant only if neither STACK_CHECK_BUILTIN 3495nor STACK_CHECK_STATIC_BUILTIN is defined; you can omit them altogether 3496in the opposite case. 3497 3498@defmac STACK_CHECK_MAX_FRAME_SIZE 3499The maximum size of a stack frame, in bytes. GCC will generate probe 3500instructions in non-leaf functions to ensure at least this many bytes of 3501stack are available. If a stack frame is larger than this size, stack 3502checking will not be reliable and GCC will issue a warning. The 3503default is chosen so that GCC only generates one instruction on most 3504systems. You should normally not change the default value of this macro. 3505@end defmac 3506 3507@defmac STACK_CHECK_FIXED_FRAME_SIZE 3508GCC uses this value to generate the above warning message. It 3509represents the amount of fixed frame used by a function, not including 3510space for any callee-saved registers, temporaries and user variables. 3511You need only specify an upper bound for this amount and will normally 3512use the default of four words. 3513@end defmac 3514 3515@defmac STACK_CHECK_MAX_VAR_SIZE 3516The maximum size, in bytes, of an object that GCC will place in the 3517fixed area of the stack frame when the user specifies 3518@option{-fstack-check}. 3519GCC computed the default from the values of the above macros and you will 3520normally not need to override that default. 3521@end defmac 3522 3523@need 2000 3524@node Frame Registers 3525@subsection Registers That Address the Stack Frame 3526 3527@c prevent bad page break with this line 3528This discusses registers that address the stack frame. 3529 3530@defmac STACK_POINTER_REGNUM 3531The register number of the stack pointer register, which must also be a 3532fixed register according to @code{FIXED_REGISTERS}. On most machines, 3533the hardware determines which register this is. 3534@end defmac 3535 3536@defmac FRAME_POINTER_REGNUM 3537The register number of the frame pointer register, which is used to 3538access automatic variables in the stack frame. On some machines, the 3539hardware determines which register this is. On other machines, you can 3540choose any register you wish for this purpose. 3541@end defmac 3542 3543@defmac HARD_FRAME_POINTER_REGNUM 3544On some machines the offset between the frame pointer and starting 3545offset of the automatic variables is not known until after register 3546allocation has been done (for example, because the saved registers are 3547between these two locations). On those machines, define 3548@code{FRAME_POINTER_REGNUM} the number of a special, fixed register to 3549be used internally until the offset is known, and define 3550@code{HARD_FRAME_POINTER_REGNUM} to be the actual hard register number 3551used for the frame pointer. 3552 3553You should define this macro only in the very rare circumstances when it 3554is not possible to calculate the offset between the frame pointer and 3555the automatic variables until after register allocation has been 3556completed. When this macro is defined, you must also indicate in your 3557definition of @code{ELIMINABLE_REGS} how to eliminate 3558@code{FRAME_POINTER_REGNUM} into either @code{HARD_FRAME_POINTER_REGNUM} 3559or @code{STACK_POINTER_REGNUM}. 3560 3561Do not define this macro if it would be the same as 3562@code{FRAME_POINTER_REGNUM}. 3563@end defmac 3564 3565@defmac ARG_POINTER_REGNUM 3566The register number of the arg pointer register, which is used to access 3567the function's argument list. On some machines, this is the same as the 3568frame pointer register. On some machines, the hardware determines which 3569register this is. On other machines, you can choose any register you 3570wish for this purpose. If this is not the same register as the frame 3571pointer register, then you must mark it as a fixed register according to 3572@code{FIXED_REGISTERS}, or arrange to be able to eliminate it 3573(@pxref{Elimination}). 3574@end defmac 3575 3576@defmac HARD_FRAME_POINTER_IS_FRAME_POINTER 3577Define this to a preprocessor constant that is nonzero if 3578@code{hard_frame_pointer_rtx} and @code{frame_pointer_rtx} should be 3579the same. The default definition is @samp{(HARD_FRAME_POINTER_REGNUM 3580== FRAME_POINTER_REGNUM)}; you only need to define this macro if that 3581definition is not suitable for use in preprocessor conditionals. 3582@end defmac 3583 3584@defmac HARD_FRAME_POINTER_IS_ARG_POINTER 3585Define this to a preprocessor constant that is nonzero if 3586@code{hard_frame_pointer_rtx} and @code{arg_pointer_rtx} should be the 3587same. The default definition is @samp{(HARD_FRAME_POINTER_REGNUM == 3588ARG_POINTER_REGNUM)}; you only need to define this macro if that 3589definition is not suitable for use in preprocessor conditionals. 3590@end defmac 3591 3592@defmac RETURN_ADDRESS_POINTER_REGNUM 3593The register number of the return address pointer register, which is used to 3594access the current function's return address from the stack. On some 3595machines, the return address is not at a fixed offset from the frame 3596pointer or stack pointer or argument pointer. This register can be defined 3597to point to the return address on the stack, and then be converted by 3598@code{ELIMINABLE_REGS} into either the frame pointer or stack pointer. 3599 3600Do not define this macro unless there is no other way to get the return 3601address from the stack. 3602@end defmac 3603 3604@defmac STATIC_CHAIN_REGNUM 3605@defmacx STATIC_CHAIN_INCOMING_REGNUM 3606Register numbers used for passing a function's static chain pointer. If 3607register windows are used, the register number as seen by the called 3608function is @code{STATIC_CHAIN_INCOMING_REGNUM}, while the register 3609number as seen by the calling function is @code{STATIC_CHAIN_REGNUM}. If 3610these registers are the same, @code{STATIC_CHAIN_INCOMING_REGNUM} need 3611not be defined. 3612 3613The static chain register need not be a fixed register. 3614 3615If the static chain is passed in memory, these macros should not be 3616defined; instead, the @code{TARGET_STATIC_CHAIN} hook should be used. 3617@end defmac 3618 3619@hook TARGET_STATIC_CHAIN 3620This hook replaces the use of @code{STATIC_CHAIN_REGNUM} et al for 3621targets that may use different static chain locations for different 3622nested functions. This may be required if the target has function 3623attributes that affect the calling conventions of the function and 3624those calling conventions use different static chain locations. 3625 3626The default version of this hook uses @code{STATIC_CHAIN_REGNUM} et al. 3627 3628If the static chain is passed in memory, this hook should be used to 3629provide rtx giving @code{mem} expressions that denote where they are stored. 3630Often the @code{mem} expression as seen by the caller will be at an offset 3631from the stack pointer and the @code{mem} expression as seen by the callee 3632will be at an offset from the frame pointer. 3633@findex stack_pointer_rtx 3634@findex frame_pointer_rtx 3635@findex arg_pointer_rtx 3636The variables @code{stack_pointer_rtx}, @code{frame_pointer_rtx}, and 3637@code{arg_pointer_rtx} will have been initialized and should be used 3638to refer to those items. 3639@end deftypefn 3640 3641@defmac DWARF_FRAME_REGISTERS 3642This macro specifies the maximum number of hard registers that can be 3643saved in a call frame. This is used to size data structures used in 3644DWARF2 exception handling. 3645 3646Prior to GCC 3.0, this macro was needed in order to establish a stable 3647exception handling ABI in the face of adding new hard registers for ISA 3648extensions. In GCC 3.0 and later, the EH ABI is insulated from changes 3649in the number of hard registers. Nevertheless, this macro can still be 3650used to reduce the runtime memory requirements of the exception handling 3651routines, which can be substantial if the ISA contains a lot of 3652registers that are not call-saved. 3653 3654If this macro is not defined, it defaults to 3655@code{FIRST_PSEUDO_REGISTER}. 3656@end defmac 3657 3658@defmac PRE_GCC3_DWARF_FRAME_REGISTERS 3659 3660This macro is similar to @code{DWARF_FRAME_REGISTERS}, but is provided 3661for backward compatibility in pre GCC 3.0 compiled code. 3662 3663If this macro is not defined, it defaults to 3664@code{DWARF_FRAME_REGISTERS}. 3665@end defmac 3666 3667@defmac DWARF_REG_TO_UNWIND_COLUMN (@var{regno}) 3668 3669Define this macro if the target's representation for dwarf registers 3670is different than the internal representation for unwind column. 3671Given a dwarf register, this macro should return the internal unwind 3672column number to use instead. 3673 3674See the PowerPC's SPE target for an example. 3675@end defmac 3676 3677@defmac DWARF_FRAME_REGNUM (@var{regno}) 3678 3679Define this macro if the target's representation for dwarf registers 3680used in .eh_frame or .debug_frame is different from that used in other 3681debug info sections. Given a GCC hard register number, this macro 3682should return the .eh_frame register number. The default is 3683@code{DBX_REGISTER_NUMBER (@var{regno})}. 3684 3685@end defmac 3686 3687@defmac DWARF2_FRAME_REG_OUT (@var{regno}, @var{for_eh}) 3688 3689Define this macro to map register numbers held in the call frame info 3690that GCC has collected using @code{DWARF_FRAME_REGNUM} to those that 3691should be output in .debug_frame (@code{@var{for_eh}} is zero) and 3692.eh_frame (@code{@var{for_eh}} is nonzero). The default is to 3693return @code{@var{regno}}. 3694 3695@end defmac 3696 3697@defmac REG_VALUE_IN_UNWIND_CONTEXT 3698 3699Define this macro if the target stores register values as 3700@code{_Unwind_Word} type in unwind context. It should be defined if 3701target register size is larger than the size of @code{void *}. The 3702default is to store register values as @code{void *} type. 3703 3704@end defmac 3705 3706@defmac ASSUME_EXTENDED_UNWIND_CONTEXT 3707 3708Define this macro to be 1 if the target always uses extended unwind 3709context with version, args_size and by_value fields. If it is undefined, 3710it will be defined to 1 when @code{REG_VALUE_IN_UNWIND_CONTEXT} is 3711defined and 0 otherwise. 3712 3713@end defmac 3714 3715@node Elimination 3716@subsection Eliminating Frame Pointer and Arg Pointer 3717 3718@c prevent bad page break with this line 3719This is about eliminating the frame pointer and arg pointer. 3720 3721@hook TARGET_FRAME_POINTER_REQUIRED 3722This target hook should return @code{true} if a function must have and use 3723a frame pointer. This target hook is called in the reload pass. If its return 3724value is @code{true} the function will have a frame pointer. 3725 3726This target hook can in principle examine the current function and decide 3727according to the facts, but on most machines the constant @code{false} or the 3728constant @code{true} suffices. Use @code{false} when the machine allows code 3729to be generated with no frame pointer, and doing so saves some time or space. 3730Use @code{true} when there is no possible advantage to avoiding a frame 3731pointer. 3732 3733In certain cases, the compiler does not know how to produce valid code 3734without a frame pointer. The compiler recognizes those cases and 3735automatically gives the function a frame pointer regardless of what 3736@code{TARGET_FRAME_POINTER_REQUIRED} returns. You don't need to worry about 3737them. 3738 3739In a function that does not require a frame pointer, the frame pointer 3740register can be allocated for ordinary usage, unless you mark it as a 3741fixed register. See @code{FIXED_REGISTERS} for more information. 3742 3743Default return value is @code{false}. 3744@end deftypefn 3745 3746@findex get_frame_size 3747@defmac INITIAL_FRAME_POINTER_OFFSET (@var{depth-var}) 3748A C statement to store in the variable @var{depth-var} the difference 3749between the frame pointer and the stack pointer values immediately after 3750the function prologue. The value would be computed from information 3751such as the result of @code{get_frame_size ()} and the tables of 3752registers @code{regs_ever_live} and @code{call_used_regs}. 3753 3754If @code{ELIMINABLE_REGS} is defined, this macro will be not be used and 3755need not be defined. Otherwise, it must be defined even if 3756@code{TARGET_FRAME_POINTER_REQUIRED} always returns true; in that 3757case, you may set @var{depth-var} to anything. 3758@end defmac 3759 3760@defmac ELIMINABLE_REGS 3761If defined, this macro specifies a table of register pairs used to 3762eliminate unneeded registers that point into the stack frame. If it is not 3763defined, the only elimination attempted by the compiler is to replace 3764references to the frame pointer with references to the stack pointer. 3765 3766The definition of this macro is a list of structure initializations, each 3767of which specifies an original and replacement register. 3768 3769On some machines, the position of the argument pointer is not known until 3770the compilation is completed. In such a case, a separate hard register 3771must be used for the argument pointer. This register can be eliminated by 3772replacing it with either the frame pointer or the argument pointer, 3773depending on whether or not the frame pointer has been eliminated. 3774 3775In this case, you might specify: 3776@smallexample 3777#define ELIMINABLE_REGS \ 3778@{@{ARG_POINTER_REGNUM, STACK_POINTER_REGNUM@}, \ 3779 @{ARG_POINTER_REGNUM, FRAME_POINTER_REGNUM@}, \ 3780 @{FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM@}@} 3781@end smallexample 3782 3783Note that the elimination of the argument pointer with the stack pointer is 3784specified first since that is the preferred elimination. 3785@end defmac 3786 3787@hook TARGET_CAN_ELIMINATE 3788This target hook should returns @code{true} if the compiler is allowed to 3789try to replace register number @var{from_reg} with register number 3790@var{to_reg}. This target hook need only be defined if @code{ELIMINABLE_REGS} 3791is defined, and will usually be @code{true}, since most of the cases 3792preventing register elimination are things that the compiler already 3793knows about. 3794 3795Default return value is @code{true}. 3796@end deftypefn 3797 3798@defmac INITIAL_ELIMINATION_OFFSET (@var{from-reg}, @var{to-reg}, @var{offset-var}) 3799This macro is similar to @code{INITIAL_FRAME_POINTER_OFFSET}. It 3800specifies the initial difference between the specified pair of 3801registers. This macro must be defined if @code{ELIMINABLE_REGS} is 3802defined. 3803@end defmac 3804 3805@node Stack Arguments 3806@subsection Passing Function Arguments on the Stack 3807@cindex arguments on stack 3808@cindex stack arguments 3809 3810The macros in this section control how arguments are passed 3811on the stack. See the following section for other macros that 3812control passing certain arguments in registers. 3813 3814@hook TARGET_PROMOTE_PROTOTYPES 3815This target hook returns @code{true} if an argument declared in a 3816prototype as an integral type smaller than @code{int} should actually be 3817passed as an @code{int}. In addition to avoiding errors in certain 3818cases of mismatch, it also makes for better code on certain machines. 3819The default is to not promote prototypes. 3820@end deftypefn 3821 3822@defmac PUSH_ARGS 3823A C expression. If nonzero, push insns will be used to pass 3824outgoing arguments. 3825If the target machine does not have a push instruction, set it to zero. 3826That directs GCC to use an alternate strategy: to 3827allocate the entire argument block and then store the arguments into 3828it. When @code{PUSH_ARGS} is nonzero, @code{PUSH_ROUNDING} must be defined too. 3829@end defmac 3830 3831@defmac PUSH_ARGS_REVERSED 3832A C expression. If nonzero, function arguments will be evaluated from 3833last to first, rather than from first to last. If this macro is not 3834defined, it defaults to @code{PUSH_ARGS} on targets where the stack 3835and args grow in opposite directions, and 0 otherwise. 3836@end defmac 3837 3838@defmac PUSH_ROUNDING (@var{npushed}) 3839A C expression that is the number of bytes actually pushed onto the 3840stack when an instruction attempts to push @var{npushed} bytes. 3841 3842On some machines, the definition 3843 3844@smallexample 3845#define PUSH_ROUNDING(BYTES) (BYTES) 3846@end smallexample 3847 3848@noindent 3849will suffice. But on other machines, instructions that appear 3850to push one byte actually push two bytes in an attempt to maintain 3851alignment. Then the definition should be 3852 3853@smallexample 3854#define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1) 3855@end smallexample 3856 3857If the value of this macro has a type, it should be an unsigned type. 3858@end defmac 3859 3860@findex current_function_outgoing_args_size 3861@defmac ACCUMULATE_OUTGOING_ARGS 3862A C expression. If nonzero, the maximum amount of space required for outgoing arguments 3863will be computed and placed into the variable 3864@code{current_function_outgoing_args_size}. No space will be pushed 3865onto the stack for each call; instead, the function prologue should 3866increase the stack frame size by this amount. 3867 3868Setting both @code{PUSH_ARGS} and @code{ACCUMULATE_OUTGOING_ARGS} 3869is not proper. 3870@end defmac 3871 3872@defmac REG_PARM_STACK_SPACE (@var{fndecl}) 3873Define this macro if functions should assume that stack space has been 3874allocated for arguments even when their values are passed in 3875registers. 3876 3877The value of this macro is the size, in bytes, of the area reserved for 3878arguments passed in registers for the function represented by @var{fndecl}, 3879which can be zero if GCC is calling a library function. 3880The argument @var{fndecl} can be the FUNCTION_DECL, or the type itself 3881of the function. 3882 3883This space can be allocated by the caller, or be a part of the 3884machine-dependent stack frame: @code{OUTGOING_REG_PARM_STACK_SPACE} says 3885which. 3886@end defmac 3887@c above is overfull. not sure what to do. --mew 5feb93 did 3888@c something, not sure if it looks good. --mew 10feb93 3889 3890@defmac OUTGOING_REG_PARM_STACK_SPACE (@var{fntype}) 3891Define this to a nonzero value if it is the responsibility of the 3892caller to allocate the area reserved for arguments passed in registers 3893when calling a function of @var{fntype}. @var{fntype} may be NULL 3894if the function called is a library function. 3895 3896If @code{ACCUMULATE_OUTGOING_ARGS} is defined, this macro controls 3897whether the space for these arguments counts in the value of 3898@code{current_function_outgoing_args_size}. 3899@end defmac 3900 3901@defmac STACK_PARMS_IN_REG_PARM_AREA 3902Define this macro if @code{REG_PARM_STACK_SPACE} is defined, but the 3903stack parameters don't skip the area specified by it. 3904@c i changed this, makes more sens and it should have taken care of the 3905@c overfull.. not as specific, tho. --mew 5feb93 3906 3907Normally, when a parameter is not passed in registers, it is placed on the 3908stack beyond the @code{REG_PARM_STACK_SPACE} area. Defining this macro 3909suppresses this behavior and causes the parameter to be passed on the 3910stack in its natural location. 3911@end defmac 3912 3913@hook TARGET_RETURN_POPS_ARGS 3914This target hook returns the number of bytes of its own arguments that 3915a function pops on returning, or 0 if the function pops no arguments 3916and the caller must therefore pop them all after the function returns. 3917 3918@var{fundecl} is a C variable whose value is a tree node that describes 3919the function in question. Normally it is a node of type 3920@code{FUNCTION_DECL} that describes the declaration of the function. 3921From this you can obtain the @code{DECL_ATTRIBUTES} of the function. 3922 3923@var{funtype} is a C variable whose value is a tree node that 3924describes the function in question. Normally it is a node of type 3925@code{FUNCTION_TYPE} that describes the data type of the function. 3926From this it is possible to obtain the data types of the value and 3927arguments (if known). 3928 3929When a call to a library function is being considered, @var{fundecl} 3930will contain an identifier node for the library function. Thus, if 3931you need to distinguish among various library functions, you can do so 3932by their names. Note that ``library function'' in this context means 3933a function used to perform arithmetic, whose name is known specially 3934in the compiler and was not mentioned in the C code being compiled. 3935 3936@var{size} is the number of bytes of arguments passed on the 3937stack. If a variable number of bytes is passed, it is zero, and 3938argument popping will always be the responsibility of the calling function. 3939 3940On the VAX, all functions always pop their arguments, so the definition 3941of this macro is @var{size}. On the 68000, using the standard 3942calling convention, no functions pop their arguments, so the value of 3943the macro is always 0 in this case. But an alternative calling 3944convention is available in which functions that take a fixed number of 3945arguments pop them but other functions (such as @code{printf}) pop 3946nothing (the caller pops all). When this convention is in use, 3947@var{funtype} is examined to determine whether a function takes a fixed 3948number of arguments. 3949@end deftypefn 3950 3951@defmac CALL_POPS_ARGS (@var{cum}) 3952A C expression that should indicate the number of bytes a call sequence 3953pops off the stack. It is added to the value of @code{RETURN_POPS_ARGS} 3954when compiling a function call. 3955 3956@var{cum} is the variable in which all arguments to the called function 3957have been accumulated. 3958 3959On certain architectures, such as the SH5, a call trampoline is used 3960that pops certain registers off the stack, depending on the arguments 3961that have been passed to the function. Since this is a property of the 3962call site, not of the called function, @code{RETURN_POPS_ARGS} is not 3963appropriate. 3964@end defmac 3965 3966@node Register Arguments 3967@subsection Passing Arguments in Registers 3968@cindex arguments in registers 3969@cindex registers arguments 3970 3971This section describes the macros which let you control how various 3972types of arguments are passed in registers or how they are arranged in 3973the stack. 3974 3975@hook TARGET_FUNCTION_ARG 3976Return an RTX indicating whether a function argument is passed in a 3977register and if so, which register. 3978 3979The arguments are @var{ca}, which summarizes all the previous 3980arguments; @var{mode}, the machine mode of the argument; @var{type}, 3981the data type of the argument as a tree node or 0 if that is not known 3982(which happens for C support library functions); and @var{named}, 3983which is @code{true} for an ordinary argument and @code{false} for 3984nameless arguments that correspond to @samp{@dots{}} in the called 3985function's prototype. @var{type} can be an incomplete type if a 3986syntax error has previously occurred. 3987 3988The return value is usually either a @code{reg} RTX for the hard 3989register in which to pass the argument, or zero to pass the argument 3990on the stack. 3991 3992The value of the expression can also be a @code{parallel} RTX@. This is 3993used when an argument is passed in multiple locations. The mode of the 3994@code{parallel} should be the mode of the entire argument. The 3995@code{parallel} holds any number of @code{expr_list} pairs; each one 3996describes where part of the argument is passed. In each 3997@code{expr_list} the first operand must be a @code{reg} RTX for the hard 3998register in which to pass this part of the argument, and the mode of the 3999register RTX indicates how large this part of the argument is. The 4000second operand of the @code{expr_list} is a @code{const_int} which gives 4001the offset in bytes into the entire argument of where this part starts. 4002As a special exception the first @code{expr_list} in the @code{parallel} 4003RTX may have a first operand of zero. This indicates that the entire 4004argument is also stored on the stack. 4005 4006The last time this hook is called, it is called with @code{MODE == 4007VOIDmode}, and its result is passed to the @code{call} or @code{call_value} 4008pattern as operands 2 and 3 respectively. 4009 4010@cindex @file{stdarg.h} and register arguments 4011The usual way to make the ISO library @file{stdarg.h} work on a 4012machine where some arguments are usually passed in registers, is to 4013cause nameless arguments to be passed on the stack instead. This is 4014done by making @code{TARGET_FUNCTION_ARG} return 0 whenever 4015@var{named} is @code{false}. 4016 4017@cindex @code{TARGET_MUST_PASS_IN_STACK}, and @code{TARGET_FUNCTION_ARG} 4018@cindex @code{REG_PARM_STACK_SPACE}, and @code{TARGET_FUNCTION_ARG} 4019You may use the hook @code{targetm.calls.must_pass_in_stack} 4020in the definition of this macro to determine if this argument is of a 4021type that must be passed in the stack. If @code{REG_PARM_STACK_SPACE} 4022is not defined and @code{TARGET_FUNCTION_ARG} returns nonzero for such an 4023argument, the compiler will abort. If @code{REG_PARM_STACK_SPACE} is 4024defined, the argument will be computed in the stack and then loaded into 4025a register. 4026@end deftypefn 4027 4028@hook TARGET_MUST_PASS_IN_STACK 4029This target hook should return @code{true} if we should not pass @var{type} 4030solely in registers. The file @file{expr.h} defines a 4031definition that is usually appropriate, refer to @file{expr.h} for additional 4032documentation. 4033@end deftypefn 4034 4035@hook TARGET_FUNCTION_INCOMING_ARG 4036Define this hook if the target machine has ``register windows'', so 4037that the register in which a function sees an arguments is not 4038necessarily the same as the one in which the caller passed the 4039argument. 4040 4041For such machines, @code{TARGET_FUNCTION_ARG} computes the register in 4042which the caller passes the value, and 4043@code{TARGET_FUNCTION_INCOMING_ARG} should be defined in a similar 4044fashion to tell the function being called where the arguments will 4045arrive. 4046 4047If @code{TARGET_FUNCTION_INCOMING_ARG} is not defined, 4048@code{TARGET_FUNCTION_ARG} serves both purposes. 4049@end deftypefn 4050 4051@hook TARGET_ARG_PARTIAL_BYTES 4052This target hook returns the number of bytes at the beginning of an 4053argument that must be put in registers. The value must be zero for 4054arguments that are passed entirely in registers or that are entirely 4055pushed on the stack. 4056 4057On some machines, certain arguments must be passed partially in 4058registers and partially in memory. On these machines, typically the 4059first few words of arguments are passed in registers, and the rest 4060on the stack. If a multi-word argument (a @code{double} or a 4061structure) crosses that boundary, its first few words must be passed 4062in registers and the rest must be pushed. This macro tells the 4063compiler when this occurs, and how many bytes should go in registers. 4064 4065@code{TARGET_FUNCTION_ARG} for these arguments should return the first 4066register to be used by the caller for this argument; likewise 4067@code{TARGET_FUNCTION_INCOMING_ARG}, for the called function. 4068@end deftypefn 4069 4070@hook TARGET_PASS_BY_REFERENCE 4071This target hook should return @code{true} if an argument at the 4072position indicated by @var{cum} should be passed by reference. This 4073predicate is queried after target independent reasons for being 4074passed by reference, such as @code{TREE_ADDRESSABLE (type)}. 4075 4076If the hook returns true, a copy of that argument is made in memory and a 4077pointer to the argument is passed instead of the argument itself. 4078The pointer is passed in whatever way is appropriate for passing a pointer 4079to that type. 4080@end deftypefn 4081 4082@hook TARGET_CALLEE_COPIES 4083The function argument described by the parameters to this hook is 4084known to be passed by reference. The hook should return true if the 4085function argument should be copied by the callee instead of copied 4086by the caller. 4087 4088For any argument for which the hook returns true, if it can be 4089determined that the argument is not modified, then a copy need 4090not be generated. 4091 4092The default version of this hook always returns false. 4093@end deftypefn 4094 4095@defmac CUMULATIVE_ARGS 4096A C type for declaring a variable that is used as the first argument 4097of @code{TARGET_FUNCTION_ARG} and other related values. For some 4098target machines, the type @code{int} suffices and can hold the number 4099of bytes of argument so far. 4100 4101There is no need to record in @code{CUMULATIVE_ARGS} anything about the 4102arguments that have been passed on the stack. The compiler has other 4103variables to keep track of that. For target machines on which all 4104arguments are passed on the stack, there is no need to store anything in 4105@code{CUMULATIVE_ARGS}; however, the data structure must exist and 4106should not be empty, so use @code{int}. 4107@end defmac 4108 4109@defmac OVERRIDE_ABI_FORMAT (@var{fndecl}) 4110If defined, this macro is called before generating any code for a 4111function, but after the @var{cfun} descriptor for the function has been 4112created. The back end may use this macro to update @var{cfun} to 4113reflect an ABI other than that which would normally be used by default. 4114If the compiler is generating code for a compiler-generated function, 4115@var{fndecl} may be @code{NULL}. 4116@end defmac 4117 4118@defmac INIT_CUMULATIVE_ARGS (@var{cum}, @var{fntype}, @var{libname}, @var{fndecl}, @var{n_named_args}) 4119A C statement (sans semicolon) for initializing the variable 4120@var{cum} for the state at the beginning of the argument list. The 4121variable has type @code{CUMULATIVE_ARGS}. The value of @var{fntype} 4122is the tree node for the data type of the function which will receive 4123the args, or 0 if the args are to a compiler support library function. 4124For direct calls that are not libcalls, @var{fndecl} contain the 4125declaration node of the function. @var{fndecl} is also set when 4126@code{INIT_CUMULATIVE_ARGS} is used to find arguments for the function 4127being compiled. @var{n_named_args} is set to the number of named 4128arguments, including a structure return address if it is passed as a 4129parameter, when making a call. When processing incoming arguments, 4130@var{n_named_args} is set to @minus{}1. 4131 4132When processing a call to a compiler support library function, 4133@var{libname} identifies which one. It is a @code{symbol_ref} rtx which 4134contains the name of the function, as a string. @var{libname} is 0 when 4135an ordinary C function call is being processed. Thus, each time this 4136macro is called, either @var{libname} or @var{fntype} is nonzero, but 4137never both of them at once. 4138@end defmac 4139 4140@defmac INIT_CUMULATIVE_LIBCALL_ARGS (@var{cum}, @var{mode}, @var{libname}) 4141Like @code{INIT_CUMULATIVE_ARGS} but only used for outgoing libcalls, 4142it gets a @code{MODE} argument instead of @var{fntype}, that would be 4143@code{NULL}. @var{indirect} would always be zero, too. If this macro 4144is not defined, @code{INIT_CUMULATIVE_ARGS (cum, NULL_RTX, libname, 41450)} is used instead. 4146@end defmac 4147 4148@defmac INIT_CUMULATIVE_INCOMING_ARGS (@var{cum}, @var{fntype}, @var{libname}) 4149Like @code{INIT_CUMULATIVE_ARGS} but overrides it for the purposes of 4150finding the arguments for the function being compiled. If this macro is 4151undefined, @code{INIT_CUMULATIVE_ARGS} is used instead. 4152 4153The value passed for @var{libname} is always 0, since library routines 4154with special calling conventions are never compiled with GCC@. The 4155argument @var{libname} exists for symmetry with 4156@code{INIT_CUMULATIVE_ARGS}. 4157@c could use "this macro" in place of @code{INIT_CUMULATIVE_ARGS}, maybe. 4158@c --mew 5feb93 i switched the order of the sentences. --mew 10feb93 4159@end defmac 4160 4161@hook TARGET_FUNCTION_ARG_ADVANCE 4162This hook updates the summarizer variable pointed to by @var{ca} to 4163advance past an argument in the argument list. The values @var{mode}, 4164@var{type} and @var{named} describe that argument. Once this is done, 4165the variable @var{cum} is suitable for analyzing the @emph{following} 4166argument with @code{TARGET_FUNCTION_ARG}, etc. 4167 4168This hook need not do anything if the argument in question was passed 4169on the stack. The compiler knows how to track the amount of stack space 4170used for arguments without any special help. 4171@end deftypefn 4172 4173@defmac FUNCTION_ARG_OFFSET (@var{mode}, @var{type}) 4174If defined, a C expression that is the number of bytes to add to the 4175offset of the argument passed in memory. This is needed for the SPU, 4176which passes @code{char} and @code{short} arguments in the preferred 4177slot that is in the middle of the quad word instead of starting at the 4178top. 4179@end defmac 4180 4181@defmac FUNCTION_ARG_PADDING (@var{mode}, @var{type}) 4182If defined, a C expression which determines whether, and in which direction, 4183to pad out an argument with extra space. The value should be of type 4184@code{enum direction}: either @code{upward} to pad above the argument, 4185@code{downward} to pad below, or @code{none} to inhibit padding. 4186 4187The @emph{amount} of padding is not controlled by this macro, but by the 4188target hook @code{TARGET_FUNCTION_ARG_ROUND_BOUNDARY}. It is 4189always just enough to reach the next multiple of that boundary. 4190 4191This macro has a default definition which is right for most systems. 4192For little-endian machines, the default is to pad upward. For 4193big-endian machines, the default is to pad downward for an argument of 4194constant size shorter than an @code{int}, and upward otherwise. 4195@end defmac 4196 4197@defmac PAD_VARARGS_DOWN 4198If defined, a C expression which determines whether the default 4199implementation of va_arg will attempt to pad down before reading the 4200next argument, if that argument is smaller than its aligned space as 4201controlled by @code{PARM_BOUNDARY}. If this macro is not defined, all such 4202arguments are padded down if @code{BYTES_BIG_ENDIAN} is true. 4203@end defmac 4204 4205@defmac BLOCK_REG_PADDING (@var{mode}, @var{type}, @var{first}) 4206Specify padding for the last element of a block move between registers and 4207memory. @var{first} is nonzero if this is the only element. Defining this 4208macro allows better control of register function parameters on big-endian 4209machines, without using @code{PARALLEL} rtl. In particular, 4210@code{MUST_PASS_IN_STACK} need not test padding and mode of types in 4211registers, as there is no longer a "wrong" part of a register; For example, 4212a three byte aggregate may be passed in the high part of a register if so 4213required. 4214@end defmac 4215 4216@hook TARGET_FUNCTION_ARG_BOUNDARY 4217This hook returns the alignment boundary, in bits, of an argument 4218with the specified mode and type. The default hook returns 4219@code{PARM_BOUNDARY} for all arguments. 4220@end deftypefn 4221 4222@hook TARGET_FUNCTION_ARG_ROUND_BOUNDARY 4223 4224@defmac FUNCTION_ARG_REGNO_P (@var{regno}) 4225A C expression that is nonzero if @var{regno} is the number of a hard 4226register in which function arguments are sometimes passed. This does 4227@emph{not} include implicit arguments such as the static chain and 4228the structure-value address. On many machines, no registers can be 4229used for this purpose since all function arguments are pushed on the 4230stack. 4231@end defmac 4232 4233@hook TARGET_SPLIT_COMPLEX_ARG 4234This hook should return true if parameter of type @var{type} are passed 4235as two scalar parameters. By default, GCC will attempt to pack complex 4236arguments into the target's word size. Some ABIs require complex arguments 4237to be split and treated as their individual components. For example, on 4238AIX64, complex floats should be passed in a pair of floating point 4239registers, even though a complex float would fit in one 64-bit floating 4240point register. 4241 4242The default value of this hook is @code{NULL}, which is treated as always 4243false. 4244@end deftypefn 4245 4246@hook TARGET_BUILD_BUILTIN_VA_LIST 4247This hook returns a type node for @code{va_list} for the target. 4248The default version of the hook returns @code{void*}. 4249@end deftypefn 4250 4251@hook TARGET_ENUM_VA_LIST_P 4252This target hook is used in function @code{c_common_nodes_and_builtins} 4253to iterate through the target specific builtin types for va_list. The 4254variable @var{idx} is used as iterator. @var{pname} has to be a pointer 4255to a @code{const char *} and @var{ptree} a pointer to a @code{tree} typed 4256variable. 4257The arguments @var{pname} and @var{ptree} are used to store the result of 4258this macro and are set to the name of the va_list builtin type and its 4259internal type. 4260If the return value of this macro is zero, then there is no more element. 4261Otherwise the @var{IDX} should be increased for the next call of this 4262macro to iterate through all types. 4263@end deftypefn 4264 4265@hook TARGET_FN_ABI_VA_LIST 4266This hook returns the va_list type of the calling convention specified by 4267@var{fndecl}. 4268The default version of this hook returns @code{va_list_type_node}. 4269@end deftypefn 4270 4271@hook TARGET_CANONICAL_VA_LIST_TYPE 4272This hook returns the va_list type of the calling convention specified by the 4273type of @var{type}. If @var{type} is not a valid va_list type, it returns 4274@code{NULL_TREE}. 4275@end deftypefn 4276 4277@hook TARGET_GIMPLIFY_VA_ARG_EXPR 4278This hook performs target-specific gimplification of 4279@code{VA_ARG_EXPR}. The first two parameters correspond to the 4280arguments to @code{va_arg}; the latter two are as in 4281@code{gimplify.c:gimplify_expr}. 4282@end deftypefn 4283 4284@hook TARGET_VALID_POINTER_MODE 4285Define this to return nonzero if the port can handle pointers 4286with machine mode @var{mode}. The default version of this 4287hook returns true for both @code{ptr_mode} and @code{Pmode}. 4288@end deftypefn 4289 4290@hook TARGET_REF_MAY_ALIAS_ERRNO 4291 4292@hook TARGET_SCALAR_MODE_SUPPORTED_P 4293Define this to return nonzero if the port is prepared to handle 4294insns involving scalar mode @var{mode}. For a scalar mode to be 4295considered supported, all the basic arithmetic and comparisons 4296must work. 4297 4298The default version of this hook returns true for any mode 4299required to handle the basic C types (as defined by the port). 4300Included here are the double-word arithmetic supported by the 4301code in @file{optabs.c}. 4302@end deftypefn 4303 4304@hook TARGET_VECTOR_MODE_SUPPORTED_P 4305Define this to return nonzero if the port is prepared to handle 4306insns involving vector mode @var{mode}. At the very least, it 4307must have move patterns for this mode. 4308@end deftypefn 4309 4310@hook TARGET_ARRAY_MODE_SUPPORTED_P 4311 4312@hook TARGET_SMALL_REGISTER_CLASSES_FOR_MODE_P 4313Define this to return nonzero for machine modes for which the port has 4314small register classes. If this target hook returns nonzero for a given 4315@var{mode}, the compiler will try to minimize the lifetime of registers 4316in @var{mode}. The hook may be called with @code{VOIDmode} as argument. 4317In this case, the hook is expected to return nonzero if it returns nonzero 4318for any mode. 4319 4320On some machines, it is risky to let hard registers live across arbitrary 4321insns. Typically, these machines have instructions that require values 4322to be in specific registers (like an accumulator), and reload will fail 4323if the required hard register is used for another purpose across such an 4324insn. 4325 4326Passes before reload do not know which hard registers will be used 4327in an instruction, but the machine modes of the registers set or used in 4328the instruction are already known. And for some machines, register 4329classes are small for, say, integer registers but not for floating point 4330registers. For example, the AMD x86-64 architecture requires specific 4331registers for the legacy x86 integer instructions, but there are many 4332SSE registers for floating point operations. On such targets, a good 4333strategy may be to return nonzero from this hook for @code{INTEGRAL_MODE_P} 4334machine modes but zero for the SSE register classes. 4335 4336The default version of this hook returns false for any mode. It is always 4337safe to redefine this hook to return with a nonzero value. But if you 4338unnecessarily define it, you will reduce the amount of optimizations 4339that can be performed in some cases. If you do not define this hook 4340to return a nonzero value when it is required, the compiler will run out 4341of spill registers and print a fatal error message. 4342@end deftypefn 4343 4344@hook TARGET_FLAGS_REGNUM 4345 4346@node Scalar Return 4347@subsection How Scalar Function Values Are Returned 4348@cindex return values in registers 4349@cindex values, returned by functions 4350@cindex scalars, returned as values 4351 4352This section discusses the macros that control returning scalars as 4353values---values that can fit in registers. 4354 4355@hook TARGET_FUNCTION_VALUE 4356 4357Define this to return an RTX representing the place where a function 4358returns or receives a value of data type @var{ret_type}, a tree node 4359representing a data type. @var{fn_decl_or_type} is a tree node 4360representing @code{FUNCTION_DECL} or @code{FUNCTION_TYPE} of a 4361function being called. If @var{outgoing} is false, the hook should 4362compute the register in which the caller will see the return value. 4363Otherwise, the hook should return an RTX representing the place where 4364a function returns a value. 4365 4366On many machines, only @code{TYPE_MODE (@var{ret_type})} is relevant. 4367(Actually, on most machines, scalar values are returned in the same 4368place regardless of mode.) The value of the expression is usually a 4369@code{reg} RTX for the hard register where the return value is stored. 4370The value can also be a @code{parallel} RTX, if the return value is in 4371multiple places. See @code{TARGET_FUNCTION_ARG} for an explanation of the 4372@code{parallel} form. Note that the callee will populate every 4373location specified in the @code{parallel}, but if the first element of 4374the @code{parallel} contains the whole return value, callers will use 4375that element as the canonical location and ignore the others. The m68k 4376port uses this type of @code{parallel} to return pointers in both 4377@samp{%a0} (the canonical location) and @samp{%d0}. 4378 4379If @code{TARGET_PROMOTE_FUNCTION_RETURN} returns true, you must apply 4380the same promotion rules specified in @code{PROMOTE_MODE} if 4381@var{valtype} is a scalar type. 4382 4383If the precise function being called is known, @var{func} is a tree 4384node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null 4385pointer. This makes it possible to use a different value-returning 4386convention for specific functions when all their calls are 4387known. 4388 4389Some target machines have ``register windows'' so that the register in 4390which a function returns its value is not the same as the one in which 4391the caller sees the value. For such machines, you should return 4392different RTX depending on @var{outgoing}. 4393 4394@code{TARGET_FUNCTION_VALUE} is not used for return values with 4395aggregate data types, because these are returned in another way. See 4396@code{TARGET_STRUCT_VALUE_RTX} and related macros, below. 4397@end deftypefn 4398 4399@defmac FUNCTION_VALUE (@var{valtype}, @var{func}) 4400This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE} for 4401a new target instead. 4402@end defmac 4403 4404@defmac LIBCALL_VALUE (@var{mode}) 4405A C expression to create an RTX representing the place where a library 4406function returns a value of mode @var{mode}. 4407 4408Note that ``library function'' in this context means a compiler 4409support routine, used to perform arithmetic, whose name is known 4410specially by the compiler and was not mentioned in the C code being 4411compiled. 4412@end defmac 4413 4414@hook TARGET_LIBCALL_VALUE 4415Define this hook if the back-end needs to know the name of the libcall 4416function in order to determine where the result should be returned. 4417 4418The mode of the result is given by @var{mode} and the name of the called 4419library function is given by @var{fun}. The hook should return an RTX 4420representing the place where the library function result will be returned. 4421 4422If this hook is not defined, then LIBCALL_VALUE will be used. 4423@end deftypefn 4424 4425@defmac FUNCTION_VALUE_REGNO_P (@var{regno}) 4426A C expression that is nonzero if @var{regno} is the number of a hard 4427register in which the values of called function may come back. 4428 4429A register whose use for returning values is limited to serving as the 4430second of a pair (for a value of type @code{double}, say) need not be 4431recognized by this macro. So for most machines, this definition 4432suffices: 4433 4434@smallexample 4435#define FUNCTION_VALUE_REGNO_P(N) ((N) == 0) 4436@end smallexample 4437 4438If the machine has register windows, so that the caller and the called 4439function use different registers for the return value, this macro 4440should recognize only the caller's register numbers. 4441 4442This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE_REGNO_P} 4443for a new target instead. 4444@end defmac 4445 4446@hook TARGET_FUNCTION_VALUE_REGNO_P 4447A target hook that return @code{true} if @var{regno} is the number of a hard 4448register in which the values of called function may come back. 4449 4450A register whose use for returning values is limited to serving as the 4451second of a pair (for a value of type @code{double}, say) need not be 4452recognized by this target hook. 4453 4454If the machine has register windows, so that the caller and the called 4455function use different registers for the return value, this target hook 4456should recognize only the caller's register numbers. 4457 4458If this hook is not defined, then FUNCTION_VALUE_REGNO_P will be used. 4459@end deftypefn 4460 4461@defmac APPLY_RESULT_SIZE 4462Define this macro if @samp{untyped_call} and @samp{untyped_return} 4463need more space than is implied by @code{FUNCTION_VALUE_REGNO_P} for 4464saving and restoring an arbitrary return value. 4465@end defmac 4466 4467@hook TARGET_RETURN_IN_MSB 4468This hook should return true if values of type @var{type} are returned 4469at the most significant end of a register (in other words, if they are 4470padded at the least significant end). You can assume that @var{type} 4471is returned in a register; the caller is required to check this. 4472 4473Note that the register provided by @code{TARGET_FUNCTION_VALUE} must 4474be able to hold the complete return value. For example, if a 1-, 2- 4475or 3-byte structure is returned at the most significant end of a 44764-byte register, @code{TARGET_FUNCTION_VALUE} should provide an 4477@code{SImode} rtx. 4478@end deftypefn 4479 4480@node Aggregate Return 4481@subsection How Large Values Are Returned 4482@cindex aggregates as return values 4483@cindex large return values 4484@cindex returning aggregate values 4485@cindex structure value address 4486 4487When a function value's mode is @code{BLKmode} (and in some other 4488cases), the value is not returned according to 4489@code{TARGET_FUNCTION_VALUE} (@pxref{Scalar Return}). Instead, the 4490caller passes the address of a block of memory in which the value 4491should be stored. This address is called the @dfn{structure value 4492address}. 4493 4494This section describes how to control returning structure values in 4495memory. 4496 4497@hook TARGET_RETURN_IN_MEMORY 4498This target hook should return a nonzero value to say to return the 4499function value in memory, just as large structures are always returned. 4500Here @var{type} will be the data type of the value, and @var{fntype} 4501will be the type of the function doing the returning, or @code{NULL} for 4502libcalls. 4503 4504Note that values of mode @code{BLKmode} must be explicitly handled 4505by this function. Also, the option @option{-fpcc-struct-return} 4506takes effect regardless of this macro. On most systems, it is 4507possible to leave the hook undefined; this causes a default 4508definition to be used, whose value is the constant 1 for @code{BLKmode} 4509values, and 0 otherwise. 4510 4511Do not use this hook to indicate that structures and unions should always 4512be returned in memory. You should instead use @code{DEFAULT_PCC_STRUCT_RETURN} 4513to indicate this. 4514@end deftypefn 4515 4516@defmac DEFAULT_PCC_STRUCT_RETURN 4517Define this macro to be 1 if all structure and union return values must be 4518in memory. Since this results in slower code, this should be defined 4519only if needed for compatibility with other compilers or with an ABI@. 4520If you define this macro to be 0, then the conventions used for structure 4521and union return values are decided by the @code{TARGET_RETURN_IN_MEMORY} 4522target hook. 4523 4524If not defined, this defaults to the value 1. 4525@end defmac 4526 4527@hook TARGET_STRUCT_VALUE_RTX 4528This target hook should return the location of the structure value 4529address (normally a @code{mem} or @code{reg}), or 0 if the address is 4530passed as an ``invisible'' first argument. Note that @var{fndecl} may 4531be @code{NULL}, for libcalls. You do not need to define this target 4532hook if the address is always passed as an ``invisible'' first 4533argument. 4534 4535On some architectures the place where the structure value address 4536is found by the called function is not the same place that the 4537caller put it. This can be due to register windows, or it could 4538be because the function prologue moves it to a different place. 4539@var{incoming} is @code{1} or @code{2} when the location is needed in 4540the context of the called function, and @code{0} in the context of 4541the caller. 4542 4543If @var{incoming} is nonzero and the address is to be found on the 4544stack, return a @code{mem} which refers to the frame pointer. If 4545@var{incoming} is @code{2}, the result is being used to fetch the 4546structure value address at the beginning of a function. If you need 4547to emit adjusting code, you should do it at this point. 4548@end deftypefn 4549 4550@defmac PCC_STATIC_STRUCT_RETURN 4551Define this macro if the usual system convention on the target machine 4552for returning structures and unions is for the called function to return 4553the address of a static variable containing the value. 4554 4555Do not define this if the usual system convention is for the caller to 4556pass an address to the subroutine. 4557 4558This macro has effect in @option{-fpcc-struct-return} mode, but it does 4559nothing when you use @option{-freg-struct-return} mode. 4560@end defmac 4561 4562@hook TARGET_GET_RAW_RESULT_MODE 4563 4564@hook TARGET_GET_RAW_ARG_MODE 4565 4566@node Caller Saves 4567@subsection Caller-Saves Register Allocation 4568 4569If you enable it, GCC can save registers around function calls. This 4570makes it possible to use call-clobbered registers to hold variables that 4571must live across calls. 4572 4573@defmac CALLER_SAVE_PROFITABLE (@var{refs}, @var{calls}) 4574A C expression to determine whether it is worthwhile to consider placing 4575a pseudo-register in a call-clobbered hard register and saving and 4576restoring it around each function call. The expression should be 1 when 4577this is worth doing, and 0 otherwise. 4578 4579If you don't define this macro, a default is used which is good on most 4580machines: @code{4 * @var{calls} < @var{refs}}. 4581@end defmac 4582 4583@defmac HARD_REGNO_CALLER_SAVE_MODE (@var{regno}, @var{nregs}) 4584A C expression specifying which mode is required for saving @var{nregs} 4585of a pseudo-register in call-clobbered hard register @var{regno}. If 4586@var{regno} is unsuitable for caller save, @code{VOIDmode} should be 4587returned. For most machines this macro need not be defined since GCC 4588will select the smallest suitable mode. 4589@end defmac 4590 4591@node Function Entry 4592@subsection Function Entry and Exit 4593@cindex function entry and exit 4594@cindex prologue 4595@cindex epilogue 4596 4597This section describes the macros that output function entry 4598(@dfn{prologue}) and exit (@dfn{epilogue}) code. 4599 4600@hook TARGET_ASM_FUNCTION_PROLOGUE 4601If defined, a function that outputs the assembler code for entry to a 4602function. The prologue is responsible for setting up the stack frame, 4603initializing the frame pointer register, saving registers that must be 4604saved, and allocating @var{size} additional bytes of storage for the 4605local variables. @var{size} is an integer. @var{file} is a stdio 4606stream to which the assembler code should be output. 4607 4608The label for the beginning of the function need not be output by this 4609macro. That has already been done when the macro is run. 4610 4611@findex regs_ever_live 4612To determine which registers to save, the macro can refer to the array 4613@code{regs_ever_live}: element @var{r} is nonzero if hard register 4614@var{r} is used anywhere within the function. This implies the function 4615prologue should save register @var{r}, provided it is not one of the 4616call-used registers. (@code{TARGET_ASM_FUNCTION_EPILOGUE} must likewise use 4617@code{regs_ever_live}.) 4618 4619On machines that have ``register windows'', the function entry code does 4620not save on the stack the registers that are in the windows, even if 4621they are supposed to be preserved by function calls; instead it takes 4622appropriate steps to ``push'' the register stack, if any non-call-used 4623registers are used in the function. 4624 4625@findex frame_pointer_needed 4626On machines where functions may or may not have frame-pointers, the 4627function entry code must vary accordingly; it must set up the frame 4628pointer if one is wanted, and not otherwise. To determine whether a 4629frame pointer is in wanted, the macro can refer to the variable 4630@code{frame_pointer_needed}. The variable's value will be 1 at run 4631time in a function that needs a frame pointer. @xref{Elimination}. 4632 4633The function entry code is responsible for allocating any stack space 4634required for the function. This stack space consists of the regions 4635listed below. In most cases, these regions are allocated in the 4636order listed, with the last listed region closest to the top of the 4637stack (the lowest address if @code{STACK_GROWS_DOWNWARD} is defined, and 4638the highest address if it is not defined). You can use a different order 4639for a machine if doing so is more convenient or required for 4640compatibility reasons. Except in cases where required by standard 4641or by a debugger, there is no reason why the stack layout used by GCC 4642need agree with that used by other compilers for a machine. 4643@end deftypefn 4644 4645@hook TARGET_ASM_FUNCTION_END_PROLOGUE 4646If defined, a function that outputs assembler code at the end of a 4647prologue. This should be used when the function prologue is being 4648emitted as RTL, and you have some extra assembler that needs to be 4649emitted. @xref{prologue instruction pattern}. 4650@end deftypefn 4651 4652@hook TARGET_ASM_FUNCTION_BEGIN_EPILOGUE 4653If defined, a function that outputs assembler code at the start of an 4654epilogue. This should be used when the function epilogue is being 4655emitted as RTL, and you have some extra assembler that needs to be 4656emitted. @xref{epilogue instruction pattern}. 4657@end deftypefn 4658 4659@hook TARGET_ASM_FUNCTION_EPILOGUE 4660If defined, a function that outputs the assembler code for exit from a 4661function. The epilogue is responsible for restoring the saved 4662registers and stack pointer to their values when the function was 4663called, and returning control to the caller. This macro takes the 4664same arguments as the macro @code{TARGET_ASM_FUNCTION_PROLOGUE}, and the 4665registers to restore are determined from @code{regs_ever_live} and 4666@code{CALL_USED_REGISTERS} in the same way. 4667 4668On some machines, there is a single instruction that does all the work 4669of returning from the function. On these machines, give that 4670instruction the name @samp{return} and do not define the macro 4671@code{TARGET_ASM_FUNCTION_EPILOGUE} at all. 4672 4673Do not define a pattern named @samp{return} if you want the 4674@code{TARGET_ASM_FUNCTION_EPILOGUE} to be used. If you want the target 4675switches to control whether return instructions or epilogues are used, 4676define a @samp{return} pattern with a validity condition that tests the 4677target switches appropriately. If the @samp{return} pattern's validity 4678condition is false, epilogues will be used. 4679 4680On machines where functions may or may not have frame-pointers, the 4681function exit code must vary accordingly. Sometimes the code for these 4682two cases is completely different. To determine whether a frame pointer 4683is wanted, the macro can refer to the variable 4684@code{frame_pointer_needed}. The variable's value will be 1 when compiling 4685a function that needs a frame pointer. 4686 4687Normally, @code{TARGET_ASM_FUNCTION_PROLOGUE} and 4688@code{TARGET_ASM_FUNCTION_EPILOGUE} must treat leaf functions specially. 4689The C variable @code{current_function_is_leaf} is nonzero for such a 4690function. @xref{Leaf Functions}. 4691 4692On some machines, some functions pop their arguments on exit while 4693others leave that for the caller to do. For example, the 68020 when 4694given @option{-mrtd} pops arguments in functions that take a fixed 4695number of arguments. 4696 4697@findex current_function_pops_args 4698Your definition of the macro @code{RETURN_POPS_ARGS} decides which 4699functions pop their own arguments. @code{TARGET_ASM_FUNCTION_EPILOGUE} 4700needs to know what was decided. The number of bytes of the current 4701function's arguments that this function should pop is available in 4702@code{crtl->args.pops_args}. @xref{Scalar Return}. 4703@end deftypefn 4704 4705@itemize @bullet 4706@item 4707@findex current_function_pretend_args_size 4708A region of @code{current_function_pretend_args_size} bytes of 4709uninitialized space just underneath the first argument arriving on the 4710stack. (This may not be at the very start of the allocated stack region 4711if the calling sequence has pushed anything else since pushing the stack 4712arguments. But usually, on such machines, nothing else has been pushed 4713yet, because the function prologue itself does all the pushing.) This 4714region is used on machines where an argument may be passed partly in 4715registers and partly in memory, and, in some cases to support the 4716features in @code{<stdarg.h>}. 4717 4718@item 4719An area of memory used to save certain registers used by the function. 4720The size of this area, which may also include space for such things as 4721the return address and pointers to previous stack frames, is 4722machine-specific and usually depends on which registers have been used 4723in the function. Machines with register windows often do not require 4724a save area. 4725 4726@item 4727A region of at least @var{size} bytes, possibly rounded up to an allocation 4728boundary, to contain the local variables of the function. On some machines, 4729this region and the save area may occur in the opposite order, with the 4730save area closer to the top of the stack. 4731 4732@item 4733@cindex @code{ACCUMULATE_OUTGOING_ARGS} and stack frames 4734Optionally, when @code{ACCUMULATE_OUTGOING_ARGS} is defined, a region of 4735@code{current_function_outgoing_args_size} bytes to be used for outgoing 4736argument lists of the function. @xref{Stack Arguments}. 4737@end itemize 4738 4739@defmac EXIT_IGNORE_STACK 4740Define this macro as a C expression that is nonzero if the return 4741instruction or the function epilogue ignores the value of the stack 4742pointer; in other words, if it is safe to delete an instruction to 4743adjust the stack pointer before a return from the function. The 4744default is 0. 4745 4746Note that this macro's value is relevant only for functions for which 4747frame pointers are maintained. It is never safe to delete a final 4748stack adjustment in a function that has no frame pointer, and the 4749compiler knows this regardless of @code{EXIT_IGNORE_STACK}. 4750@end defmac 4751 4752@defmac EPILOGUE_USES (@var{regno}) 4753Define this macro as a C expression that is nonzero for registers that are 4754used by the epilogue or the @samp{return} pattern. The stack and frame 4755pointer registers are already assumed to be used as needed. 4756@end defmac 4757 4758@defmac EH_USES (@var{regno}) 4759Define this macro as a C expression that is nonzero for registers that are 4760used by the exception handling mechanism, and so should be considered live 4761on entry to an exception edge. 4762@end defmac 4763 4764@defmac DELAY_SLOTS_FOR_EPILOGUE 4765Define this macro if the function epilogue contains delay slots to which 4766instructions from the rest of the function can be ``moved''. The 4767definition should be a C expression whose value is an integer 4768representing the number of delay slots there. 4769@end defmac 4770 4771@defmac ELIGIBLE_FOR_EPILOGUE_DELAY (@var{insn}, @var{n}) 4772A C expression that returns 1 if @var{insn} can be placed in delay 4773slot number @var{n} of the epilogue. 4774 4775The argument @var{n} is an integer which identifies the delay slot now 4776being considered (since different slots may have different rules of 4777eligibility). It is never negative and is always less than the number 4778of epilogue delay slots (what @code{DELAY_SLOTS_FOR_EPILOGUE} returns). 4779If you reject a particular insn for a given delay slot, in principle, it 4780may be reconsidered for a subsequent delay slot. Also, other insns may 4781(at least in principle) be considered for the so far unfilled delay 4782slot. 4783 4784@findex current_function_epilogue_delay_list 4785@findex final_scan_insn 4786The insns accepted to fill the epilogue delay slots are put in an RTL 4787list made with @code{insn_list} objects, stored in the variable 4788@code{current_function_epilogue_delay_list}. The insn for the first 4789delay slot comes first in the list. Your definition of the macro 4790@code{TARGET_ASM_FUNCTION_EPILOGUE} should fill the delay slots by 4791outputting the insns in this list, usually by calling 4792@code{final_scan_insn}. 4793 4794You need not define this macro if you did not define 4795@code{DELAY_SLOTS_FOR_EPILOGUE}. 4796@end defmac 4797 4798@hook TARGET_ASM_OUTPUT_MI_THUNK 4799A function that outputs the assembler code for a thunk 4800function, used to implement C++ virtual function calls with multiple 4801inheritance. The thunk acts as a wrapper around a virtual function, 4802adjusting the implicit object parameter before handing control off to 4803the real function. 4804 4805First, emit code to add the integer @var{delta} to the location that 4806contains the incoming first argument. Assume that this argument 4807contains a pointer, and is the one used to pass the @code{this} pointer 4808in C++. This is the incoming argument @emph{before} the function prologue, 4809e.g.@: @samp{%o0} on a sparc. The addition must preserve the values of 4810all other incoming arguments. 4811 4812Then, if @var{vcall_offset} is nonzero, an additional adjustment should be 4813made after adding @code{delta}. In particular, if @var{p} is the 4814adjusted pointer, the following adjustment should be made: 4815 4816@smallexample 4817p += (*((ptrdiff_t **)p))[vcall_offset/sizeof(ptrdiff_t)] 4818@end smallexample 4819 4820After the additions, emit code to jump to @var{function}, which is a 4821@code{FUNCTION_DECL}. This is a direct pure jump, not a call, and does 4822not touch the return address. Hence returning from @var{FUNCTION} will 4823return to whoever called the current @samp{thunk}. 4824 4825The effect must be as if @var{function} had been called directly with 4826the adjusted first argument. This macro is responsible for emitting all 4827of the code for a thunk function; @code{TARGET_ASM_FUNCTION_PROLOGUE} 4828and @code{TARGET_ASM_FUNCTION_EPILOGUE} are not invoked. 4829 4830The @var{thunk_fndecl} is redundant. (@var{delta} and @var{function} 4831have already been extracted from it.) It might possibly be useful on 4832some targets, but probably not. 4833 4834If you do not define this macro, the target-independent code in the C++ 4835front end will generate a less efficient heavyweight thunk that calls 4836@var{function} instead of jumping to it. The generic approach does 4837not support varargs. 4838@end deftypefn 4839 4840@hook TARGET_ASM_CAN_OUTPUT_MI_THUNK 4841A function that returns true if TARGET_ASM_OUTPUT_MI_THUNK would be able 4842to output the assembler code for the thunk function specified by the 4843arguments it is passed, and false otherwise. In the latter case, the 4844generic approach will be used by the C++ front end, with the limitations 4845previously exposed. 4846@end deftypefn 4847 4848@node Profiling 4849@subsection Generating Code for Profiling 4850@cindex profiling, code generation 4851 4852These macros will help you generate code for profiling. 4853 4854@defmac FUNCTION_PROFILER (@var{file}, @var{labelno}) 4855A C statement or compound statement to output to @var{file} some 4856assembler code to call the profiling subroutine @code{mcount}. 4857 4858@findex mcount 4859The details of how @code{mcount} expects to be called are determined by 4860your operating system environment, not by GCC@. To figure them out, 4861compile a small program for profiling using the system's installed C 4862compiler and look at the assembler code that results. 4863 4864Older implementations of @code{mcount} expect the address of a counter 4865variable to be loaded into some register. The name of this variable is 4866@samp{LP} followed by the number @var{labelno}, so you would generate 4867the name using @samp{LP%d} in a @code{fprintf}. 4868@end defmac 4869 4870@defmac PROFILE_HOOK 4871A C statement or compound statement to output to @var{file} some assembly 4872code to call the profiling subroutine @code{mcount} even the target does 4873not support profiling. 4874@end defmac 4875 4876@defmac NO_PROFILE_COUNTERS 4877Define this macro to be an expression with a nonzero value if the 4878@code{mcount} subroutine on your system does not need a counter variable 4879allocated for each function. This is true for almost all modern 4880implementations. If you define this macro, you must not use the 4881@var{labelno} argument to @code{FUNCTION_PROFILER}. 4882@end defmac 4883 4884@defmac PROFILE_BEFORE_PROLOGUE 4885Define this macro if the code for function profiling should come before 4886the function prologue. Normally, the profiling code comes after. 4887@end defmac 4888 4889@node Tail Calls 4890@subsection Permitting tail calls 4891@cindex tail calls 4892 4893@hook TARGET_FUNCTION_OK_FOR_SIBCALL 4894True if it is ok to do sibling call optimization for the specified 4895call expression @var{exp}. @var{decl} will be the called function, 4896or @code{NULL} if this is an indirect call. 4897 4898It is not uncommon for limitations of calling conventions to prevent 4899tail calls to functions outside the current unit of translation, or 4900during PIC compilation. The hook is used to enforce these restrictions, 4901as the @code{sibcall} md pattern can not fail, or fall over to a 4902``normal'' call. The criteria for successful sibling call optimization 4903may vary greatly between different architectures. 4904@end deftypefn 4905 4906@hook TARGET_EXTRA_LIVE_ON_ENTRY 4907Add any hard registers to @var{regs} that are live on entry to the 4908function. This hook only needs to be defined to provide registers that 4909cannot be found by examination of FUNCTION_ARG_REGNO_P, the callee saved 4910registers, STATIC_CHAIN_INCOMING_REGNUM, STATIC_CHAIN_REGNUM, 4911TARGET_STRUCT_VALUE_RTX, FRAME_POINTER_REGNUM, EH_USES, 4912FRAME_POINTER_REGNUM, ARG_POINTER_REGNUM, and the PIC_OFFSET_TABLE_REGNUM. 4913@end deftypefn 4914 4915@hook TARGET_SET_UP_BY_PROLOGUE 4916 4917@node Stack Smashing Protection 4918@subsection Stack smashing protection 4919@cindex stack smashing protection 4920 4921@hook TARGET_STACK_PROTECT_GUARD 4922This hook returns a @code{DECL} node for the external variable to use 4923for the stack protection guard. This variable is initialized by the 4924runtime to some random value and is used to initialize the guard value 4925that is placed at the top of the local stack frame. The type of this 4926variable must be @code{ptr_type_node}. 4927 4928The default version of this hook creates a variable called 4929@samp{__stack_chk_guard}, which is normally defined in @file{libgcc2.c}. 4930@end deftypefn 4931 4932@hook TARGET_STACK_PROTECT_FAIL 4933This hook returns a tree expression that alerts the runtime that the 4934stack protect guard variable has been modified. This expression should 4935involve a call to a @code{noreturn} function. 4936 4937The default version of this hook invokes a function called 4938@samp{__stack_chk_fail}, taking no arguments. This function is 4939normally defined in @file{libgcc2.c}. 4940@end deftypefn 4941 4942@hook TARGET_SUPPORTS_SPLIT_STACK 4943 4944@node Varargs 4945@section Implementing the Varargs Macros 4946@cindex varargs implementation 4947 4948GCC comes with an implementation of @code{<varargs.h>} and 4949@code{<stdarg.h>} that work without change on machines that pass arguments 4950on the stack. Other machines require their own implementations of 4951varargs, and the two machine independent header files must have 4952conditionals to include it. 4953 4954ISO @code{<stdarg.h>} differs from traditional @code{<varargs.h>} mainly in 4955the calling convention for @code{va_start}. The traditional 4956implementation takes just one argument, which is the variable in which 4957to store the argument pointer. The ISO implementation of 4958@code{va_start} takes an additional second argument. The user is 4959supposed to write the last named argument of the function here. 4960 4961However, @code{va_start} should not use this argument. The way to find 4962the end of the named arguments is with the built-in functions described 4963below. 4964 4965@defmac __builtin_saveregs () 4966Use this built-in function to save the argument registers in memory so 4967that the varargs mechanism can access them. Both ISO and traditional 4968versions of @code{va_start} must use @code{__builtin_saveregs}, unless 4969you use @code{TARGET_SETUP_INCOMING_VARARGS} (see below) instead. 4970 4971On some machines, @code{__builtin_saveregs} is open-coded under the 4972control of the target hook @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. On 4973other machines, it calls a routine written in assembler language, 4974found in @file{libgcc2.c}. 4975 4976Code generated for the call to @code{__builtin_saveregs} appears at the 4977beginning of the function, as opposed to where the call to 4978@code{__builtin_saveregs} is written, regardless of what the code is. 4979This is because the registers must be saved before the function starts 4980to use them for its own purposes. 4981@c i rewrote the first sentence above to fix an overfull hbox. --mew 4982@c 10feb93 4983@end defmac 4984 4985@defmac __builtin_next_arg (@var{lastarg}) 4986This builtin returns the address of the first anonymous stack 4987argument, as type @code{void *}. If @code{ARGS_GROW_DOWNWARD}, it 4988returns the address of the location above the first anonymous stack 4989argument. Use it in @code{va_start} to initialize the pointer for 4990fetching arguments from the stack. Also use it in @code{va_start} to 4991verify that the second parameter @var{lastarg} is the last named argument 4992of the current function. 4993@end defmac 4994 4995@defmac __builtin_classify_type (@var{object}) 4996Since each machine has its own conventions for which data types are 4997passed in which kind of register, your implementation of @code{va_arg} 4998has to embody these conventions. The easiest way to categorize the 4999specified data type is to use @code{__builtin_classify_type} together 5000with @code{sizeof} and @code{__alignof__}. 5001 5002@code{__builtin_classify_type} ignores the value of @var{object}, 5003considering only its data type. It returns an integer describing what 5004kind of type that is---integer, floating, pointer, structure, and so on. 5005 5006The file @file{typeclass.h} defines an enumeration that you can use to 5007interpret the values of @code{__builtin_classify_type}. 5008@end defmac 5009 5010These machine description macros help implement varargs: 5011 5012@hook TARGET_EXPAND_BUILTIN_SAVEREGS 5013If defined, this hook produces the machine-specific code for a call to 5014@code{__builtin_saveregs}. This code will be moved to the very 5015beginning of the function, before any parameter access are made. The 5016return value of this function should be an RTX that contains the value 5017to use as the return of @code{__builtin_saveregs}. 5018@end deftypefn 5019 5020@hook TARGET_SETUP_INCOMING_VARARGS 5021This target hook offers an alternative to using 5022@code{__builtin_saveregs} and defining the hook 5023@code{TARGET_EXPAND_BUILTIN_SAVEREGS}. Use it to store the anonymous 5024register arguments into the stack so that all the arguments appear to 5025have been passed consecutively on the stack. Once this is done, you can 5026use the standard implementation of varargs that works for machines that 5027pass all their arguments on the stack. 5028 5029The argument @var{args_so_far} points to the @code{CUMULATIVE_ARGS} data 5030structure, containing the values that are obtained after processing the 5031named arguments. The arguments @var{mode} and @var{type} describe the 5032last named argument---its machine mode and its data type as a tree node. 5033 5034The target hook should do two things: first, push onto the stack all the 5035argument registers @emph{not} used for the named arguments, and second, 5036store the size of the data thus pushed into the @code{int}-valued 5037variable pointed to by @var{pretend_args_size}. The value that you 5038store here will serve as additional offset for setting up the stack 5039frame. 5040 5041Because you must generate code to push the anonymous arguments at 5042compile time without knowing their data types, 5043@code{TARGET_SETUP_INCOMING_VARARGS} is only useful on machines that 5044have just a single category of argument register and use it uniformly 5045for all data types. 5046 5047If the argument @var{second_time} is nonzero, it means that the 5048arguments of the function are being analyzed for the second time. This 5049happens for an inline function, which is not actually compiled until the 5050end of the source file. The hook @code{TARGET_SETUP_INCOMING_VARARGS} should 5051not generate any instructions in this case. 5052@end deftypefn 5053 5054@hook TARGET_STRICT_ARGUMENT_NAMING 5055Define this hook to return @code{true} if the location where a function 5056argument is passed depends on whether or not it is a named argument. 5057 5058This hook controls how the @var{named} argument to @code{TARGET_FUNCTION_ARG} 5059is set for varargs and stdarg functions. If this hook returns 5060@code{true}, the @var{named} argument is always true for named 5061arguments, and false for unnamed arguments. If it returns @code{false}, 5062but @code{TARGET_PRETEND_OUTGOING_VARARGS_NAMED} returns @code{true}, 5063then all arguments are treated as named. Otherwise, all named arguments 5064except the last are treated as named. 5065 5066You need not define this hook if it always returns @code{false}. 5067@end deftypefn 5068 5069@hook TARGET_PRETEND_OUTGOING_VARARGS_NAMED 5070If you need to conditionally change ABIs so that one works with 5071@code{TARGET_SETUP_INCOMING_VARARGS}, but the other works like neither 5072@code{TARGET_SETUP_INCOMING_VARARGS} nor @code{TARGET_STRICT_ARGUMENT_NAMING} was 5073defined, then define this hook to return @code{true} if 5074@code{TARGET_SETUP_INCOMING_VARARGS} is used, @code{false} otherwise. 5075Otherwise, you should not define this hook. 5076@end deftypefn 5077 5078@node Trampolines 5079@section Trampolines for Nested Functions 5080@cindex trampolines for nested functions 5081@cindex nested functions, trampolines for 5082 5083A @dfn{trampoline} is a small piece of code that is created at run time 5084when the address of a nested function is taken. It normally resides on 5085the stack, in the stack frame of the containing function. These macros 5086tell GCC how to generate code to allocate and initialize a 5087trampoline. 5088 5089The instructions in the trampoline must do two things: load a constant 5090address into the static chain register, and jump to the real address of 5091the nested function. On CISC machines such as the m68k, this requires 5092two instructions, a move immediate and a jump. Then the two addresses 5093exist in the trampoline as word-long immediate operands. On RISC 5094machines, it is often necessary to load each address into a register in 5095two parts. Then pieces of each address form separate immediate 5096operands. 5097 5098The code generated to initialize the trampoline must store the variable 5099parts---the static chain value and the function address---into the 5100immediate operands of the instructions. On a CISC machine, this is 5101simply a matter of copying each address to a memory reference at the 5102proper offset from the start of the trampoline. On a RISC machine, it 5103may be necessary to take out pieces of the address and store them 5104separately. 5105 5106@hook TARGET_ASM_TRAMPOLINE_TEMPLATE 5107This hook is called by @code{assemble_trampoline_template} to output, 5108on the stream @var{f}, assembler code for a block of data that contains 5109the constant parts of a trampoline. This code should not include a 5110label---the label is taken care of automatically. 5111 5112If you do not define this hook, it means no template is needed 5113for the target. Do not define this hook on systems where the block move 5114code to copy the trampoline into place would be larger than the code 5115to generate it on the spot. 5116@end deftypefn 5117 5118@defmac TRAMPOLINE_SECTION 5119Return the section into which the trampoline template is to be placed 5120(@pxref{Sections}). The default value is @code{readonly_data_section}. 5121@end defmac 5122 5123@defmac TRAMPOLINE_SIZE 5124A C expression for the size in bytes of the trampoline, as an integer. 5125@end defmac 5126 5127@defmac TRAMPOLINE_ALIGNMENT 5128Alignment required for trampolines, in bits. 5129 5130If you don't define this macro, the value of @code{FUNCTION_ALIGNMENT} 5131is used for aligning trampolines. 5132@end defmac 5133 5134@hook TARGET_TRAMPOLINE_INIT 5135This hook is called to initialize a trampoline. 5136@var{m_tramp} is an RTX for the memory block for the trampoline; @var{fndecl} 5137is the @code{FUNCTION_DECL} for the nested function; @var{static_chain} is an 5138RTX for the static chain value that should be passed to the function 5139when it is called. 5140 5141If the target defines @code{TARGET_ASM_TRAMPOLINE_TEMPLATE}, then the 5142first thing this hook should do is emit a block move into @var{m_tramp} 5143from the memory block returned by @code{assemble_trampoline_template}. 5144Note that the block move need only cover the constant parts of the 5145trampoline. If the target isolates the variable parts of the trampoline 5146to the end, not all @code{TRAMPOLINE_SIZE} bytes need be copied. 5147 5148If the target requires any other actions, such as flushing caches or 5149enabling stack execution, these actions should be performed after 5150initializing the trampoline proper. 5151@end deftypefn 5152 5153@hook TARGET_TRAMPOLINE_ADJUST_ADDRESS 5154This hook should perform any machine-specific adjustment in 5155the address of the trampoline. Its argument contains the address of the 5156memory block that was passed to @code{TARGET_TRAMPOLINE_INIT}. In case 5157the address to be used for a function call should be different from the 5158address at which the template was stored, the different address should 5159be returned; otherwise @var{addr} should be returned unchanged. 5160If this hook is not defined, @var{addr} will be used for function calls. 5161@end deftypefn 5162 5163Implementing trampolines is difficult on many machines because they have 5164separate instruction and data caches. Writing into a stack location 5165fails to clear the memory in the instruction cache, so when the program 5166jumps to that location, it executes the old contents. 5167 5168Here are two possible solutions. One is to clear the relevant parts of 5169the instruction cache whenever a trampoline is set up. The other is to 5170make all trampolines identical, by having them jump to a standard 5171subroutine. The former technique makes trampoline execution faster; the 5172latter makes initialization faster. 5173 5174To clear the instruction cache when a trampoline is initialized, define 5175the following macro. 5176 5177@defmac CLEAR_INSN_CACHE (@var{beg}, @var{end}) 5178If defined, expands to a C expression clearing the @emph{instruction 5179cache} in the specified interval. The definition of this macro would 5180typically be a series of @code{asm} statements. Both @var{beg} and 5181@var{end} are both pointer expressions. 5182@end defmac 5183 5184To use a standard subroutine, define the following macro. In addition, 5185you must make sure that the instructions in a trampoline fill an entire 5186cache line with identical instructions, or else ensure that the 5187beginning of the trampoline code is always aligned at the same point in 5188its cache line. Look in @file{m68k.h} as a guide. 5189 5190@defmac TRANSFER_FROM_TRAMPOLINE 5191Define this macro if trampolines need a special subroutine to do their 5192work. The macro should expand to a series of @code{asm} statements 5193which will be compiled with GCC@. They go in a library function named 5194@code{__transfer_from_trampoline}. 5195 5196If you need to avoid executing the ordinary prologue code of a compiled 5197C function when you jump to the subroutine, you can do so by placing a 5198special label of your own in the assembler code. Use one @code{asm} 5199statement to generate an assembler label, and another to make the label 5200global. Then trampolines can use that label to jump directly to your 5201special assembler code. 5202@end defmac 5203 5204@node Library Calls 5205@section Implicit Calls to Library Routines 5206@cindex library subroutine names 5207@cindex @file{libgcc.a} 5208 5209@c prevent bad page break with this line 5210Here is an explanation of implicit calls to library routines. 5211 5212@defmac DECLARE_LIBRARY_RENAMES 5213This macro, if defined, should expand to a piece of C code that will get 5214expanded when compiling functions for libgcc.a. It can be used to 5215provide alternate names for GCC's internal library functions if there 5216are ABI-mandated names that the compiler should provide. 5217@end defmac 5218 5219@findex set_optab_libfunc 5220@findex init_one_libfunc 5221@hook TARGET_INIT_LIBFUNCS 5222This hook should declare additional library routines or rename 5223existing ones, using the functions @code{set_optab_libfunc} and 5224@code{init_one_libfunc} defined in @file{optabs.c}. 5225@code{init_optabs} calls this macro after initializing all the normal 5226library routines. 5227 5228The default is to do nothing. Most ports don't need to define this hook. 5229@end deftypefn 5230 5231@hook TARGET_LIBFUNC_GNU_PREFIX 5232 5233@defmac FLOAT_LIB_COMPARE_RETURNS_BOOL (@var{mode}, @var{comparison}) 5234This macro should return @code{true} if the library routine that 5235implements the floating point comparison operator @var{comparison} in 5236mode @var{mode} will return a boolean, and @var{false} if it will 5237return a tristate. 5238 5239GCC's own floating point libraries return tristates from the 5240comparison operators, so the default returns false always. Most ports 5241don't need to define this macro. 5242@end defmac 5243 5244@defmac TARGET_LIB_INT_CMP_BIASED 5245This macro should evaluate to @code{true} if the integer comparison 5246functions (like @code{__cmpdi2}) return 0 to indicate that the first 5247operand is smaller than the second, 1 to indicate that they are equal, 5248and 2 to indicate that the first operand is greater than the second. 5249If this macro evaluates to @code{false} the comparison functions return 5250@minus{}1, 0, and 1 instead of 0, 1, and 2. If the target uses the routines 5251in @file{libgcc.a}, you do not need to define this macro. 5252@end defmac 5253 5254@cindex @code{EDOM}, implicit usage 5255@findex matherr 5256@defmac TARGET_EDOM 5257The value of @code{EDOM} on the target machine, as a C integer constant 5258expression. If you don't define this macro, GCC does not attempt to 5259deposit the value of @code{EDOM} into @code{errno} directly. Look in 5260@file{/usr/include/errno.h} to find the value of @code{EDOM} on your 5261system. 5262 5263If you do not define @code{TARGET_EDOM}, then compiled code reports 5264domain errors by calling the library function and letting it report the 5265error. If mathematical functions on your system use @code{matherr} when 5266there is an error, then you should leave @code{TARGET_EDOM} undefined so 5267that @code{matherr} is used normally. 5268@end defmac 5269 5270@cindex @code{errno}, implicit usage 5271@defmac GEN_ERRNO_RTX 5272Define this macro as a C expression to create an rtl expression that 5273refers to the global ``variable'' @code{errno}. (On certain systems, 5274@code{errno} may not actually be a variable.) If you don't define this 5275macro, a reasonable default is used. 5276@end defmac 5277 5278@cindex C99 math functions, implicit usage 5279@defmac TARGET_C99_FUNCTIONS 5280When this macro is nonzero, GCC will implicitly optimize @code{sin} calls into 5281@code{sinf} and similarly for other functions defined by C99 standard. The 5282default is zero because a number of existing systems lack support for these 5283functions in their runtime so this macro needs to be redefined to one on 5284systems that do support the C99 runtime. 5285@end defmac 5286 5287@cindex sincos math function, implicit usage 5288@defmac TARGET_HAS_SINCOS 5289When this macro is nonzero, GCC will implicitly optimize calls to @code{sin} 5290and @code{cos} with the same argument to a call to @code{sincos}. The 5291default is zero. The target has to provide the following functions: 5292@smallexample 5293void sincos(double x, double *sin, double *cos); 5294void sincosf(float x, float *sin, float *cos); 5295void sincosl(long double x, long double *sin, long double *cos); 5296@end smallexample 5297@end defmac 5298 5299@defmac NEXT_OBJC_RUNTIME 5300Set this macro to 1 to use the "NeXT" Objective-C message sending conventions 5301by default. This calling convention involves passing the object, the selector 5302and the method arguments all at once to the method-lookup library function. 5303This is the usual setting when targeting Darwin/Mac OS X systems, which have 5304the NeXT runtime installed. 5305 5306If the macro is set to 0, the "GNU" Objective-C message sending convention 5307will be used by default. This convention passes just the object and the 5308selector to the method-lookup function, which returns a pointer to the method. 5309 5310In either case, it remains possible to select code-generation for the alternate 5311scheme, by means of compiler command line switches. 5312@end defmac 5313 5314@node Addressing Modes 5315@section Addressing Modes 5316@cindex addressing modes 5317 5318@c prevent bad page break with this line 5319This is about addressing modes. 5320 5321@defmac HAVE_PRE_INCREMENT 5322@defmacx HAVE_PRE_DECREMENT 5323@defmacx HAVE_POST_INCREMENT 5324@defmacx HAVE_POST_DECREMENT 5325A C expression that is nonzero if the machine supports pre-increment, 5326pre-decrement, post-increment, or post-decrement addressing respectively. 5327@end defmac 5328 5329@defmac HAVE_PRE_MODIFY_DISP 5330@defmacx HAVE_POST_MODIFY_DISP 5331A C expression that is nonzero if the machine supports pre- or 5332post-address side-effect generation involving constants other than 5333the size of the memory operand. 5334@end defmac 5335 5336@defmac HAVE_PRE_MODIFY_REG 5337@defmacx HAVE_POST_MODIFY_REG 5338A C expression that is nonzero if the machine supports pre- or 5339post-address side-effect generation involving a register displacement. 5340@end defmac 5341 5342@defmac CONSTANT_ADDRESS_P (@var{x}) 5343A C expression that is 1 if the RTX @var{x} is a constant which 5344is a valid address. On most machines the default definition of 5345@code{(CONSTANT_P (@var{x}) && GET_CODE (@var{x}) != CONST_DOUBLE)} 5346is acceptable, but a few machines are more restrictive as to which 5347constant addresses are supported. 5348@end defmac 5349 5350@defmac CONSTANT_P (@var{x}) 5351@code{CONSTANT_P}, which is defined by target-independent code, 5352accepts integer-values expressions whose values are not explicitly 5353known, such as @code{symbol_ref}, @code{label_ref}, and @code{high} 5354expressions and @code{const} arithmetic expressions, in addition to 5355@code{const_int} and @code{const_double} expressions. 5356@end defmac 5357 5358@defmac MAX_REGS_PER_ADDRESS 5359A number, the maximum number of registers that can appear in a valid 5360memory address. Note that it is up to you to specify a value equal to 5361the maximum number that @code{TARGET_LEGITIMATE_ADDRESS_P} would ever 5362accept. 5363@end defmac 5364 5365@hook TARGET_LEGITIMATE_ADDRESS_P 5366A function that returns whether @var{x} (an RTX) is a legitimate memory 5367address on the target machine for a memory operand of mode @var{mode}. 5368 5369Legitimate addresses are defined in two variants: a strict variant and a 5370non-strict one. The @var{strict} parameter chooses which variant is 5371desired by the caller. 5372 5373The strict variant is used in the reload pass. It must be defined so 5374that any pseudo-register that has not been allocated a hard register is 5375considered a memory reference. This is because in contexts where some 5376kind of register is required, a pseudo-register with no hard register 5377must be rejected. For non-hard registers, the strict variant should look 5378up the @code{reg_renumber} array; it should then proceed using the hard 5379register number in the array, or treat the pseudo as a memory reference 5380if the array holds @code{-1}. 5381 5382The non-strict variant is used in other passes. It must be defined to 5383accept all pseudo-registers in every context where some kind of 5384register is required. 5385 5386Normally, constant addresses which are the sum of a @code{symbol_ref} 5387and an integer are stored inside a @code{const} RTX to mark them as 5388constant. Therefore, there is no need to recognize such sums 5389specifically as legitimate addresses. Normally you would simply 5390recognize any @code{const} as legitimate. 5391 5392Usually @code{PRINT_OPERAND_ADDRESS} is not prepared to handle constant 5393sums that are not marked with @code{const}. It assumes that a naked 5394@code{plus} indicates indexing. If so, then you @emph{must} reject such 5395naked constant sums as illegitimate addresses, so that none of them will 5396be given to @code{PRINT_OPERAND_ADDRESS}. 5397 5398@cindex @code{TARGET_ENCODE_SECTION_INFO} and address validation 5399On some machines, whether a symbolic address is legitimate depends on 5400the section that the address refers to. On these machines, define the 5401target hook @code{TARGET_ENCODE_SECTION_INFO} to store the information 5402into the @code{symbol_ref}, and then check for it here. When you see a 5403@code{const}, you will have to look inside it to find the 5404@code{symbol_ref} in order to determine the section. @xref{Assembler 5405Format}. 5406 5407@cindex @code{GO_IF_LEGITIMATE_ADDRESS} 5408Some ports are still using a deprecated legacy substitute for 5409this hook, the @code{GO_IF_LEGITIMATE_ADDRESS} macro. This macro 5410has this syntax: 5411 5412@example 5413#define GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label}) 5414@end example 5415 5416@noindent 5417and should @code{goto @var{label}} if the address @var{x} is a valid 5418address on the target machine for a memory operand of mode @var{mode}. 5419 5420@findex REG_OK_STRICT 5421Compiler source files that want to use the strict variant of this 5422macro define the macro @code{REG_OK_STRICT}. You should use an 5423@code{#ifdef REG_OK_STRICT} conditional to define the strict variant in 5424that case and the non-strict variant otherwise. 5425 5426Using the hook is usually simpler because it limits the number of 5427files that are recompiled when changes are made. 5428@end deftypefn 5429 5430@defmac TARGET_MEM_CONSTRAINT 5431A single character to be used instead of the default @code{'m'} 5432character for general memory addresses. This defines the constraint 5433letter which matches the memory addresses accepted by 5434@code{TARGET_LEGITIMATE_ADDRESS_P}. Define this macro if you want to 5435support new address formats in your back end without changing the 5436semantics of the @code{'m'} constraint. This is necessary in order to 5437preserve functionality of inline assembly constructs using the 5438@code{'m'} constraint. 5439@end defmac 5440 5441@defmac FIND_BASE_TERM (@var{x}) 5442A C expression to determine the base term of address @var{x}, 5443or to provide a simplified version of @var{x} from which @file{alias.c} 5444can easily find the base term. This macro is used in only two places: 5445@code{find_base_value} and @code{find_base_term} in @file{alias.c}. 5446 5447It is always safe for this macro to not be defined. It exists so 5448that alias analysis can understand machine-dependent addresses. 5449 5450The typical use of this macro is to handle addresses containing 5451a label_ref or symbol_ref within an UNSPEC@. 5452@end defmac 5453 5454@hook TARGET_LEGITIMIZE_ADDRESS 5455This hook is given an invalid memory address @var{x} for an 5456operand of mode @var{mode} and should try to return a valid memory 5457address. 5458 5459@findex break_out_memory_refs 5460@var{x} will always be the result of a call to @code{break_out_memory_refs}, 5461and @var{oldx} will be the operand that was given to that function to produce 5462@var{x}. 5463 5464The code of the hook should not alter the substructure of 5465@var{x}. If it transforms @var{x} into a more legitimate form, it 5466should return the new @var{x}. 5467 5468It is not necessary for this hook to come up with a legitimate address. 5469The compiler has standard ways of doing so in all cases. In fact, it 5470is safe to omit this hook or make it return @var{x} if it cannot find 5471a valid way to legitimize the address. But often a machine-dependent 5472strategy can generate better code. 5473@end deftypefn 5474 5475@defmac LEGITIMIZE_RELOAD_ADDRESS (@var{x}, @var{mode}, @var{opnum}, @var{type}, @var{ind_levels}, @var{win}) 5476A C compound statement that attempts to replace @var{x}, which is an address 5477that needs reloading, with a valid memory address for an operand of mode 5478@var{mode}. @var{win} will be a C statement label elsewhere in the code. 5479It is not necessary to define this macro, but it might be useful for 5480performance reasons. 5481 5482For example, on the i386, it is sometimes possible to use a single 5483reload register instead of two by reloading a sum of two pseudo 5484registers into a register. On the other hand, for number of RISC 5485processors offsets are limited so that often an intermediate address 5486needs to be generated in order to address a stack slot. By defining 5487@code{LEGITIMIZE_RELOAD_ADDRESS} appropriately, the intermediate addresses 5488generated for adjacent some stack slots can be made identical, and thus 5489be shared. 5490 5491@emph{Note}: This macro should be used with caution. It is necessary 5492to know something of how reload works in order to effectively use this, 5493and it is quite easy to produce macros that build in too much knowledge 5494of reload internals. 5495 5496@emph{Note}: This macro must be able to reload an address created by a 5497previous invocation of this macro. If it fails to handle such addresses 5498then the compiler may generate incorrect code or abort. 5499 5500@findex push_reload 5501The macro definition should use @code{push_reload} to indicate parts that 5502need reloading; @var{opnum}, @var{type} and @var{ind_levels} are usually 5503suitable to be passed unaltered to @code{push_reload}. 5504 5505The code generated by this macro must not alter the substructure of 5506@var{x}. If it transforms @var{x} into a more legitimate form, it 5507should assign @var{x} (which will always be a C variable) a new value. 5508This also applies to parts that you change indirectly by calling 5509@code{push_reload}. 5510 5511@findex strict_memory_address_p 5512The macro definition may use @code{strict_memory_address_p} to test if 5513the address has become legitimate. 5514 5515@findex copy_rtx 5516If you want to change only a part of @var{x}, one standard way of doing 5517this is to use @code{copy_rtx}. Note, however, that it unshares only a 5518single level of rtl. Thus, if the part to be changed is not at the 5519top level, you'll need to replace first the top level. 5520It is not necessary for this macro to come up with a legitimate 5521address; but often a machine-dependent strategy can generate better code. 5522@end defmac 5523 5524@hook TARGET_MODE_DEPENDENT_ADDRESS_P 5525This hook returns @code{true} if memory address @var{addr} can have 5526different meanings depending on the machine mode of the memory 5527reference it is used for or if the address is valid for some modes 5528but not others. 5529 5530Autoincrement and autodecrement addresses typically have mode-dependent 5531effects because the amount of the increment or decrement is the size 5532of the operand being addressed. Some machines have other mode-dependent 5533addresses. Many RISC machines have no mode-dependent addresses. 5534 5535You may assume that @var{addr} is a valid address for the machine. 5536 5537The default version of this hook returns @code{false}. 5538@end deftypefn 5539 5540@defmac GO_IF_MODE_DEPENDENT_ADDRESS (@var{addr}, @var{label}) 5541A C statement or compound statement with a conditional @code{goto 5542@var{label};} executed if memory address @var{x} (an RTX) can have 5543different meanings depending on the machine mode of the memory 5544reference it is used for or if the address is valid for some modes 5545but not others. 5546 5547Autoincrement and autodecrement addresses typically have mode-dependent 5548effects because the amount of the increment or decrement is the size 5549of the operand being addressed. Some machines have other mode-dependent 5550addresses. Many RISC machines have no mode-dependent addresses. 5551 5552You may assume that @var{addr} is a valid address for the machine. 5553 5554These are obsolete macros, replaced by the 5555@code{TARGET_MODE_DEPENDENT_ADDRESS_P} target hook. 5556@end defmac 5557 5558@hook TARGET_LEGITIMATE_CONSTANT_P 5559This hook returns true if @var{x} is a legitimate constant for a 5560@var{mode}-mode immediate operand on the target machine. You can assume that 5561@var{x} satisfies @code{CONSTANT_P}, so you need not check this. 5562 5563The default definition returns true. 5564@end deftypefn 5565 5566@hook TARGET_DELEGITIMIZE_ADDRESS 5567This hook is used to undo the possibly obfuscating effects of the 5568@code{LEGITIMIZE_ADDRESS} and @code{LEGITIMIZE_RELOAD_ADDRESS} target 5569macros. Some backend implementations of these macros wrap symbol 5570references inside an @code{UNSPEC} rtx to represent PIC or similar 5571addressing modes. This target hook allows GCC's optimizers to understand 5572the semantics of these opaque @code{UNSPEC}s by converting them back 5573into their original form. 5574@end deftypefn 5575 5576@hook TARGET_CONST_NOT_OK_FOR_DEBUG_P 5577This hook should return true if @var{x} should not be emitted into 5578debug sections. 5579@end deftypefn 5580 5581@hook TARGET_CANNOT_FORCE_CONST_MEM 5582This hook should return true if @var{x} is of a form that cannot (or 5583should not) be spilled to the constant pool. @var{mode} is the mode 5584of @var{x}. 5585 5586The default version of this hook returns false. 5587 5588The primary reason to define this hook is to prevent reload from 5589deciding that a non-legitimate constant would be better reloaded 5590from the constant pool instead of spilling and reloading a register 5591holding the constant. This restriction is often true of addresses 5592of TLS symbols for various targets. 5593@end deftypefn 5594 5595@hook TARGET_USE_BLOCKS_FOR_CONSTANT_P 5596This hook should return true if pool entries for constant @var{x} can 5597be placed in an @code{object_block} structure. @var{mode} is the mode 5598of @var{x}. 5599 5600The default version returns false for all constants. 5601@end deftypefn 5602 5603@hook TARGET_BUILTIN_RECIPROCAL 5604This hook should return the DECL of a function that implements reciprocal of 5605the 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 5607when @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 5609of a square root function are performed, and only reciprocals of @code{sqrt} 5610function are valid. 5611@end deftypefn 5612 5613@hook TARGET_VECTORIZE_BUILTIN_MASK_FOR_LOAD 5614This hook should return the DECL of a function @var{f} that given an 5615address @var{addr} as an argument returns a mask @var{m} that can be 5616used to extract from two vectors the relevant data that resides in 5617@var{addr} in case @var{addr} is not properly aligned. 5618 5619The autovectorizer, when vectorizing a load operation from an address 5620@var{addr} that may be unaligned, will generate two vector loads from 5621the two aligned addresses around @var{addr}. It then generates a 5622@code{REALIGN_LOAD} operation to extract the relevant data from the 5623two 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 5625the third argument, @var{OFF}, defines how the data will be extracted 5626from 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}. 5630 5631If this hook is defined, the autovectorizer will generate a call 5632to @var{f} (using the DECL tree that this hook returns) and will 5633use 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} 5635should comply with the semantics expected by @code{REALIGN_LOAD} 5636described above. 5637If this hook is not defined, then @var{addr} will be used as 5638the argument @var{OFF} to @code{REALIGN_LOAD}, in which case the low 5639log2(@var{VS}) @minus{} 1 bits of @var{addr} will be considered. 5640@end deftypefn 5641 5642@hook TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_EVEN 5643This hook should return the DECL of a function @var{f} that implements 5644widening multiplication of the even elements of two input vectors of type @var{x}. 5645 5646If this hook is defined, the autovectorizer will use it along with the 5647@code{TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_ODD} target hook when vectorizing 5648widening multiplication in cases that the order of the results does not have to be 5649preserved (e.g.@: used only by a reduction computation). Otherwise, the 5650@code{widen_mult_hi/lo} idioms will be used. 5651@end deftypefn 5652 5653@hook TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_ODD 5654This hook should return the DECL of a function @var{f} that implements 5655widening multiplication of the odd elements of two input vectors of type @var{x}. 5656 5657If this hook is defined, the autovectorizer will use it along with the 5658@code{TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_EVEN} target hook when vectorizing 5659widening multiplication in cases that the order of the results does not have to be 5660preserved (e.g.@: used only by a reduction computation). Otherwise, the 5661@code{widen_mult_hi/lo} idioms will be used. 5662@end deftypefn 5663 5664@hook TARGET_VECTORIZE_BUILTIN_VECTORIZATION_COST 5665Returns cost of different scalar or vector statements for vectorization cost model. 5666For vector memory operations the cost may depend on type (@var{vectype}) and 5667misalignment value (@var{misalign}). 5668@end deftypefn 5669 5670@hook TARGET_VECTORIZE_VECTOR_ALIGNMENT_REACHABLE 5671Return true if vector alignment is reachable (by peeling N iterations) for the given type. 5672@end deftypefn 5673 5674@hook TARGET_VECTORIZE_VEC_PERM_CONST_OK 5675Return true if a vector created for @code{vec_perm_const} is valid. 5676@end deftypefn 5677 5678@hook TARGET_VECTORIZE_BUILTIN_CONVERSION 5679This hook should return the DECL of a function that implements conversion of the 5680input vector of type @var{src_type} to type @var{dest_type}. 5681The value of @var{code} is one of the enumerators in @code{enum tree_code} and 5682specifies how the conversion is to be applied 5683(truncation, rounding, etc.). 5684 5685If this hook is defined, the autovectorizer will use the 5686@code{TARGET_VECTORIZE_BUILTIN_CONVERSION} target hook when vectorizing 5687conversion. Otherwise, it will return @code{NULL_TREE}. 5688@end deftypefn 5689 5690@hook TARGET_VECTORIZE_BUILTIN_VECTORIZED_FUNCTION 5691This hook should return the decl of a function that implements the 5692vectorized variant of the builtin function with builtin function code 5693@var{code} or @code{NULL_TREE} if such a function is not available. 5694The value of @var{fndecl} is the builtin function declaration. The 5695return type of the vectorized function shall be of vector type 5696@var{vec_type_out} and the argument types should be @var{vec_type_in}. 5697@end deftypefn 5698 5699@hook TARGET_VECTORIZE_SUPPORT_VECTOR_MISALIGNMENT 5700This hook should return true if the target supports misaligned vector 5701store/load of a specific factor denoted in the @var{misalignment} 5702parameter. The vector store/load should be of machine mode @var{mode} and 5703the elements in the vectors should be of type @var{type}. @var{is_packed} 5704parameter is true if the memory access is defined in a packed struct. 5705@end deftypefn 5706 5707@hook TARGET_VECTORIZE_PREFERRED_SIMD_MODE 5708This hook should return the preferred mode for vectorizing scalar 5709mode @var{mode}. The default is 5710equal to @code{word_mode}, because the vectorizer can do some 5711transformations even in absence of specialized @acronym{SIMD} hardware. 5712@end deftypefn 5713 5714@hook TARGET_VECTORIZE_AUTOVECTORIZE_VECTOR_SIZES 5715This hook should return a mask of sizes that should be iterated over 5716after trying to autovectorize using the vector size derived from the 5717mode returned by @code{TARGET_VECTORIZE_PREFERRED_SIMD_MODE}. 5718The default is zero which means to not iterate over other vector sizes. 5719@end deftypefn 5720 5721@hook TARGET_VECTORIZE_BUILTIN_TM_LOAD 5722 5723@hook TARGET_VECTORIZE_BUILTIN_TM_STORE 5724 5725@hook TARGET_VECTORIZE_BUILTIN_GATHER 5726Target builtin that implements vector gather operation. @var{mem_vectype} 5727is the vector type of the load and @var{index_type} is scalar type of 5728the index, scaled by @var{scale}. 5729The default is @code{NULL_TREE} which means to not vectorize gather 5730loads. 5731@end deftypefn 5732 5733@node Anchored Addresses 5734@section Anchored Addresses 5735@cindex anchored addresses 5736@cindex @option{-fsection-anchors} 5737 5738GCC usually addresses every static object as a separate entity. 5739For example, if we have: 5740 5741@smallexample 5742static int a, b, c; 5743int foo (void) @{ return a + b + c; @} 5744@end smallexample 5745 5746the code for @code{foo} will usually calculate three separate symbolic 5747addresses: those of @code{a}, @code{b} and @code{c}. On some targets, 5748it would be better to calculate just one symbolic address and access 5749the three variables relative to it. The equivalent pseudocode would 5750be something like: 5751 5752@smallexample 5753int foo (void) 5754@{ 5755 register int *xr = &x; 5756 return xr[&a - &x] + xr[&b - &x] + xr[&c - &x]; 5757@} 5758@end smallexample 5759 5760(which isn't valid C). We refer to shared addresses like @code{x} as 5761``section anchors''. Their use is controlled by @option{-fsection-anchors}. 5762 5763The hooks below describe the target properties that GCC needs to know 5764in order to make effective use of section anchors. It won't use 5765section anchors at all unless either @code{TARGET_MIN_ANCHOR_OFFSET} 5766or @code{TARGET_MAX_ANCHOR_OFFSET} is set to a nonzero value. 5767 5768@hook TARGET_MIN_ANCHOR_OFFSET 5769The minimum offset that should be applied to a section anchor. 5770On most targets, it should be the smallest offset that can be 5771applied to a base register while still giving a legitimate address 5772for every mode. The default value is 0. 5773@end deftypevr 5774 5775@hook TARGET_MAX_ANCHOR_OFFSET 5776Like @code{TARGET_MIN_ANCHOR_OFFSET}, but the maximum (inclusive) 5777offset that should be applied to section anchors. The default 5778value is 0. 5779@end deftypevr 5780 5781@hook TARGET_ASM_OUTPUT_ANCHOR 5782Write the assembly code to define section anchor @var{x}, which is a 5783@code{SYMBOL_REF} for which @samp{SYMBOL_REF_ANCHOR_P (@var{x})} is true. 5784The hook is called with the assembly output position set to the beginning 5785of @code{SYMBOL_REF_BLOCK (@var{x})}. 5786 5787If @code{ASM_OUTPUT_DEF} is available, the hook's default definition uses 5788it to define the symbol as @samp{. + SYMBOL_REF_BLOCK_OFFSET (@var{x})}. 5789If @code{ASM_OUTPUT_DEF} is not available, the hook's default definition 5790is @code{NULL}, which disables the use of section anchors altogether. 5791@end deftypefn 5792 5793@hook TARGET_USE_ANCHORS_FOR_SYMBOL_P 5794Return true if GCC should attempt to use anchors to access @code{SYMBOL_REF} 5795@var{x}. You can assume @samp{SYMBOL_REF_HAS_BLOCK_INFO_P (@var{x})} and 5796@samp{!SYMBOL_REF_ANCHOR_P (@var{x})}. 5797 5798The default version is correct for most targets, but you might need to 5799intercept this hook to handle things like target-specific attributes 5800or target-specific sections. 5801@end deftypefn 5802 5803@node Condition Code 5804@section Condition Code Status 5805@cindex condition code status 5806 5807The macros in this section can be split in two families, according to the 5808two ways of representing condition codes in GCC. 5809 5810The first representation is the so called @code{(cc0)} representation 5811(@pxref{Jump Patterns}), where all instructions can have an implicit 5812clobber of the condition codes. The second is the condition code 5813register representation, which provides better schedulability for 5814architectures that do have a condition code register, but on which 5815most instructions do not affect it. The latter category includes 5816most RISC machines. 5817 5818The implicit clobbering poses a strong restriction on the placement of 5819the definition and use of the condition code, which need to be in adjacent 5820insns for machines using @code{(cc0)}. This can prevent important 5821optimizations on some machines. For example, on the IBM RS/6000, there 5822is a delay for taken branches unless the condition code register is set 5823three instructions earlier than the conditional branch. The instruction 5824scheduler cannot perform this optimization if it is not permitted to 5825separate the definition and use of the condition code register. 5826 5827For this reason, it is possible and suggested to use a register to 5828represent the condition code for new ports. If there is a specific 5829condition code register in the machine, use a hard register. If the 5830condition code or comparison result can be placed in any general register, 5831or if there are multiple condition registers, use a pseudo register. 5832Registers used to store the condition code value will usually have a mode 5833that is in class @code{MODE_CC}. 5834 5835Alternatively, you can use @code{BImode} if the comparison operator is 5836specified already in the compare instruction. In this case, you are not 5837interested in most macros in this section. 5838 5839@menu 5840* CC0 Condition Codes:: Old style representation of condition codes. 5841* MODE_CC Condition Codes:: Modern representation of condition codes. 5842* Cond Exec Macros:: Macros to control conditional execution. 5843@end menu 5844 5845@node CC0 Condition Codes 5846@subsection Representation of condition codes using @code{(cc0)} 5847@findex cc0 5848 5849@findex cc_status 5850The file @file{conditions.h} defines a variable @code{cc_status} to 5851describe how the condition code was computed (in case the interpretation of 5852the condition code depends on the instruction that it was set by). This 5853variable contains the RTL expressions on which the condition code is 5854currently based, and several standard flags. 5855 5856Sometimes additional machine-specific flags must be defined in the machine 5857description header file. It can also add additional machine-specific 5858information by defining @code{CC_STATUS_MDEP}. 5859 5860@defmac CC_STATUS_MDEP 5861C code for a data type which is used for declaring the @code{mdep} 5862component of @code{cc_status}. It defaults to @code{int}. 5863 5864This macro is not used on machines that do not use @code{cc0}. 5865@end defmac 5866 5867@defmac CC_STATUS_MDEP_INIT 5868A C expression to initialize the @code{mdep} field to ``empty''. 5869The default definition does nothing, since most machines don't use 5870the field anyway. If you want to use the field, you should probably 5871define this macro to initialize it. 5872 5873This macro is not used on machines that do not use @code{cc0}. 5874@end defmac 5875 5876@defmac NOTICE_UPDATE_CC (@var{exp}, @var{insn}) 5877A C compound statement to set the components of @code{cc_status} 5878appropriately for an insn @var{insn} whose body is @var{exp}. It is 5879this macro's responsibility to recognize insns that set the condition 5880code as a byproduct of other activity as well as those that explicitly 5881set @code{(cc0)}. 5882 5883This macro is not used on machines that do not use @code{cc0}. 5884 5885If there are insns that do not set the condition code but do alter 5886other machine registers, this macro must check to see whether they 5887invalidate the expressions that the condition code is recorded as 5888reflecting. For example, on the 68000, insns that store in address 5889registers do not set the condition code, which means that usually 5890@code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such 5891insns. But suppose that the previous insn set the condition code 5892based on location @samp{a4@@(102)} and the current insn stores a new 5893value in @samp{a4}. Although the condition code is not changed by 5894this, it will no longer be true that it reflects the contents of 5895@samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter 5896@code{cc_status} in this case to say that nothing is known about the 5897condition code value. 5898 5899The definition of @code{NOTICE_UPDATE_CC} must be prepared to deal 5900with the results of peephole optimization: insns whose patterns are 5901@code{parallel} RTXs containing various @code{reg}, @code{mem} or 5902constants which are just the operands. The RTL structure of these 5903insns is not sufficient to indicate what the insns actually do. What 5904@code{NOTICE_UPDATE_CC} should do when it sees one is just to run 5905@code{CC_STATUS_INIT}. 5906 5907A possible definition of @code{NOTICE_UPDATE_CC} is to call a function 5908that looks at an attribute (@pxref{Insn Attributes}) named, for example, 5909@samp{cc}. This avoids having detailed information about patterns in 5910two places, the @file{md} file and in @code{NOTICE_UPDATE_CC}. 5911@end defmac 5912 5913@node MODE_CC Condition Codes 5914@subsection Representation of condition codes using registers 5915@findex CCmode 5916@findex MODE_CC 5917 5918@defmac SELECT_CC_MODE (@var{op}, @var{x}, @var{y}) 5919On many machines, the condition code may be produced by other instructions 5920than compares, for example the branch can use directly the condition 5921code set by a subtract instruction. However, on some machines 5922when the condition code is set this way some bits (such as the overflow 5923bit) are not set in the same way as a test instruction, so that a different 5924branch instruction must be used for some conditional branches. When 5925this happens, use the machine mode of the condition code register to 5926record different formats of the condition code register. Modes can 5927also be used to record which compare instruction (e.g. a signed or an 5928unsigned comparison) produced the condition codes. 5929 5930If other modes than @code{CCmode} are required, add them to 5931@file{@var{machine}-modes.def} and define @code{SELECT_CC_MODE} to choose 5932a mode given an operand of a compare. This is needed because the modes 5933have to be chosen not only during RTL generation but also, for example, 5934by instruction combination. The result of @code{SELECT_CC_MODE} should 5935be consistent with the mode used in the patterns; for example to support 5936the case of the add on the SPARC discussed above, we have the pattern 5937 5938@smallexample 5939(define_insn "" 5940 [(set (reg:CC_NOOV 0) 5941 (compare:CC_NOOV 5942 (plus:SI (match_operand:SI 0 "register_operand" "%r") 5943 (match_operand:SI 1 "arith_operand" "rI")) 5944 (const_int 0)))] 5945 "" 5946 "@dots{}") 5947@end smallexample 5948 5949@noindent 5950together with a @code{SELECT_CC_MODE} that returns @code{CC_NOOVmode} 5951for comparisons whose argument is a @code{plus}: 5952 5953@smallexample 5954#define SELECT_CC_MODE(OP,X,Y) \ 5955 (GET_MODE_CLASS (GET_MODE (X)) == MODE_FLOAT \ 5956 ? ((OP == EQ || OP == NE) ? CCFPmode : CCFPEmode) \ 5957 : ((GET_CODE (X) == PLUS || GET_CODE (X) == MINUS \ 5958 || GET_CODE (X) == NEG) \ 5959 ? CC_NOOVmode : CCmode)) 5960@end smallexample 5961 5962Another reason to use modes is to retain information on which operands 5963were used by the comparison; see @code{REVERSIBLE_CC_MODE} later in 5964this section. 5965 5966You should define this macro if and only if you define extra CC modes 5967in @file{@var{machine}-modes.def}. 5968@end defmac 5969 5970@defmac CANONICALIZE_COMPARISON (@var{code}, @var{op0}, @var{op1}) 5971On some machines not all possible comparisons are defined, but you can 5972convert an invalid comparison into a valid one. For example, the Alpha 5973does not have a @code{GT} comparison, but you can use an @code{LT} 5974comparison instead and swap the order of the operands. 5975 5976On such machines, define this macro to be a C statement to do any 5977required conversions. @var{code} is the initial comparison code 5978and @var{op0} and @var{op1} are the left and right operands of the 5979comparison, respectively. You should modify @var{code}, @var{op0}, and 5980@var{op1} as required. 5981 5982GCC will not assume that the comparison resulting from this macro is 5983valid but will see if the resulting insn matches a pattern in the 5984@file{md} file. 5985 5986You need not define this macro if it would never change the comparison 5987code or operands. 5988@end defmac 5989 5990@defmac REVERSIBLE_CC_MODE (@var{mode}) 5991A C expression whose value is one if it is always safe to reverse a 5992comparison whose mode is @var{mode}. If @code{SELECT_CC_MODE} 5993can ever return @var{mode} for a floating-point inequality comparison, 5994then @code{REVERSIBLE_CC_MODE (@var{mode})} must be zero. 5995 5996You need not define this macro if it would always returns zero or if the 5997floating-point format is anything other than @code{IEEE_FLOAT_FORMAT}. 5998For example, here is the definition used on the SPARC, where floating-point 5999inequality comparisons are always given @code{CCFPEmode}: 6000 6001@smallexample 6002#define REVERSIBLE_CC_MODE(MODE) ((MODE) != CCFPEmode) 6003@end smallexample 6004@end defmac 6005 6006@defmac REVERSE_CONDITION (@var{code}, @var{mode}) 6007A C expression whose value is reversed condition code of the @var{code} for 6008comparison done in CC_MODE @var{mode}. The macro is used only in case 6009@code{REVERSIBLE_CC_MODE (@var{mode})} is nonzero. Define this macro in case 6010machine has some non-standard way how to reverse certain conditionals. For 6011instance in case all floating point conditions are non-trapping, compiler may 6012freely convert unordered compares to ordered one. Then definition may look 6013like: 6014 6015@smallexample 6016#define REVERSE_CONDITION(CODE, MODE) \ 6017 ((MODE) != CCFPmode ? reverse_condition (CODE) \ 6018 : reverse_condition_maybe_unordered (CODE)) 6019@end smallexample 6020@end defmac 6021 6022@hook TARGET_FIXED_CONDITION_CODE_REGS 6023On targets which do not use @code{(cc0)}, and which use a hard 6024register rather than a pseudo-register to hold condition codes, the 6025regular CSE passes are often not able to identify cases in which the 6026hard register is set to a common value. Use this hook to enable a 6027small pass which optimizes such cases. This hook should return true 6028to enable this pass, and it should set the integers to which its 6029arguments point to the hard register numbers used for condition codes. 6030When there is only one such register, as is true on most systems, the 6031integer pointed to by @var{p2} should be set to 6032@code{INVALID_REGNUM}. 6033 6034The default version of this hook returns false. 6035@end deftypefn 6036 6037@hook TARGET_CC_MODES_COMPATIBLE 6038On targets which use multiple condition code modes in class 6039@code{MODE_CC}, it is sometimes the case that a comparison can be 6040validly done in more than one mode. On such a system, define this 6041target hook to take two mode arguments and to return a mode in which 6042both comparisons may be validly done. If there is no such mode, 6043return @code{VOIDmode}. 6044 6045The default version of this hook checks whether the modes are the 6046same. If they are, it returns that mode. If they are different, it 6047returns @code{VOIDmode}. 6048@end deftypefn 6049 6050@node Cond Exec Macros 6051@subsection Macros to control conditional execution 6052@findex conditional execution 6053@findex predication 6054 6055There is one macro that may need to be defined for targets 6056supporting conditional execution, independent of how they 6057represent conditional branches. 6058 6059@defmac REVERSE_CONDEXEC_PREDICATES_P (@var{op1}, @var{op2}) 6060A C expression that returns true if the conditional execution predicate 6061@var{op1}, a comparison operation, is the inverse of @var{op2} and vice 6062versa. Define this to return 0 if the target has conditional execution 6063predicates that cannot be reversed safely. There is no need to validate 6064that the arguments of op1 and op2 are the same, this is done separately. 6065If no expansion is specified, this macro is defined as follows: 6066 6067@smallexample 6068#define REVERSE_CONDEXEC_PREDICATES_P (x, y) \ 6069 (GET_CODE ((x)) == reversed_comparison_code ((y), NULL)) 6070@end smallexample 6071@end defmac 6072 6073@node Costs 6074@section Describing Relative Costs of Operations 6075@cindex costs of instructions 6076@cindex relative costs 6077@cindex speed of instructions 6078 6079These macros let you describe the relative speed of various operations 6080on the target machine. 6081 6082@defmac REGISTER_MOVE_COST (@var{mode}, @var{from}, @var{to}) 6083A C expression for the cost of moving data of mode @var{mode} from a 6084register in class @var{from} to one in class @var{to}. The classes are 6085expressed using the enumeration values such as @code{GENERAL_REGS}. A 6086value of 2 is the default; other values are interpreted relative to 6087that. 6088 6089It is not required that the cost always equal 2 when @var{from} is the 6090same as @var{to}; on some machines it is expensive to move between 6091registers if they are not general registers. 6092 6093If reload sees an insn consisting of a single @code{set} between two 6094hard registers, and if @code{REGISTER_MOVE_COST} applied to their 6095classes returns a value of 2, reload does not check to ensure that the 6096constraints of the insn are met. Setting a cost of other than 2 will 6097allow reload to verify that the constraints are met. You should do this 6098if the @samp{mov@var{m}} pattern's constraints do not allow such copying. 6099 6100These macros are obsolete, new ports should use the target hook 6101@code{TARGET_REGISTER_MOVE_COST} instead. 6102@end defmac 6103 6104@hook TARGET_REGISTER_MOVE_COST 6105This target hook should return the cost of moving data of mode @var{mode} 6106from a register in class @var{from} to one in class @var{to}. The classes 6107are expressed using the enumeration values such as @code{GENERAL_REGS}. 6108A value of 2 is the default; other values are interpreted relative to 6109that. 6110 6111It is not required that the cost always equal 2 when @var{from} is the 6112same as @var{to}; on some machines it is expensive to move between 6113registers if they are not general registers. 6114 6115If reload sees an insn consisting of a single @code{set} between two 6116hard registers, and if @code{TARGET_REGISTER_MOVE_COST} applied to their 6117classes returns a value of 2, reload does not check to ensure that the 6118constraints of the insn are met. Setting a cost of other than 2 will 6119allow reload to verify that the constraints are met. You should do this 6120if the @samp{mov@var{m}} pattern's constraints do not allow such copying. 6121 6122The default version of this function returns 2. 6123@end deftypefn 6124 6125@defmac MEMORY_MOVE_COST (@var{mode}, @var{class}, @var{in}) 6126A C expression for the cost of moving data of mode @var{mode} between a 6127register of class @var{class} and memory; @var{in} is zero if the value 6128is to be written to memory, nonzero if it is to be read in. This cost 6129is relative to those in @code{REGISTER_MOVE_COST}. If moving between 6130registers and memory is more expensive than between two registers, you 6131should define this macro to express the relative cost. 6132 6133If you do not define this macro, GCC uses a default cost of 4 plus 6134the cost of copying via a secondary reload register, if one is 6135needed. If your machine requires a secondary reload register to copy 6136between memory and a register of @var{class} but the reload mechanism is 6137more complex than copying via an intermediate, define this macro to 6138reflect the actual cost of the move. 6139 6140GCC defines the function @code{memory_move_secondary_cost} if 6141secondary reloads are needed. It computes the costs due to copying via 6142a secondary register. If your machine copies from memory using a 6143secondary register in the conventional way but the default base value of 61444 is not correct for your machine, define this macro to add some other 6145value to the result of that function. The arguments to that function 6146are the same as to this macro. 6147 6148These macros are obsolete, new ports should use the target hook 6149@code{TARGET_MEMORY_MOVE_COST} instead. 6150@end defmac 6151 6152@hook TARGET_MEMORY_MOVE_COST 6153This target hook should return the cost of moving data of mode @var{mode} 6154between a register of class @var{rclass} and memory; @var{in} is @code{false} 6155if the value is to be written to memory, @code{true} if it is to be read in. 6156This cost is relative to those in @code{TARGET_REGISTER_MOVE_COST}. 6157If moving between registers and memory is more expensive than between two 6158registers, you should add this target hook to express the relative cost. 6159 6160If you do not add this target hook, GCC uses a default cost of 4 plus 6161the cost of copying via a secondary reload register, if one is 6162needed. If your machine requires a secondary reload register to copy 6163between memory and a register of @var{rclass} but the reload mechanism is 6164more complex than copying via an intermediate, use this target hook to 6165reflect the actual cost of the move. 6166 6167GCC defines the function @code{memory_move_secondary_cost} if 6168secondary reloads are needed. It computes the costs due to copying via 6169a secondary register. If your machine copies from memory using a 6170secondary register in the conventional way but the default base value of 61714 is not correct for your machine, use this target hook to add some other 6172value to the result of that function. The arguments to that function 6173are the same as to this target hook. 6174@end deftypefn 6175 6176@defmac BRANCH_COST (@var{speed_p}, @var{predictable_p}) 6177A C expression for the cost of a branch instruction. A value of 1 is 6178the default; other values are interpreted relative to that. Parameter 6179@var{speed_p} is true when the branch in question should be optimized 6180for speed. When it is false, @code{BRANCH_COST} should return a value 6181optimal for code size rather than performance. @var{predictable_p} is 6182true for well-predicted branches. On many architectures the 6183@code{BRANCH_COST} can be reduced then. 6184@end defmac 6185 6186Here are additional macros which do not specify precise relative costs, 6187but only that certain actions are more expensive than GCC would 6188ordinarily expect. 6189 6190@defmac SLOW_BYTE_ACCESS 6191Define this macro as a C expression which is nonzero if accessing less 6192than a word of memory (i.e.@: a @code{char} or a @code{short}) is no 6193faster than accessing a word of memory, i.e., if such access 6194require more than one instruction or if there is no difference in cost 6195between byte and (aligned) word loads. 6196 6197When this macro is not defined, the compiler will access a field by 6198finding the smallest containing object; when it is defined, a fullword 6199load will be used if alignment permits. Unless bytes accesses are 6200faster than word accesses, using word accesses is preferable since it 6201may eliminate subsequent memory access if subsequent accesses occur to 6202other fields in the same word of the structure, but to different bytes. 6203@end defmac 6204 6205@defmac SLOW_UNALIGNED_ACCESS (@var{mode}, @var{alignment}) 6206Define this macro to be the value 1 if memory accesses described by the 6207@var{mode} and @var{alignment} parameters have a cost many times greater 6208than aligned accesses, for example if they are emulated in a trap 6209handler. 6210 6211When this macro is nonzero, the compiler will act as if 6212@code{STRICT_ALIGNMENT} were nonzero when generating code for block 6213moves. This can cause significantly more instructions to be produced. 6214Therefore, do not set this macro nonzero if unaligned accesses only add a 6215cycle or two to the time for a memory access. 6216 6217If the value of this macro is always zero, it need not be defined. If 6218this macro is defined, it should produce a nonzero value when 6219@code{STRICT_ALIGNMENT} is nonzero. 6220@end defmac 6221 6222@defmac MOVE_RATIO (@var{speed}) 6223The threshold of number of scalar memory-to-memory move insns, @emph{below} 6224which a sequence of insns should be generated instead of a 6225string move insn or a library call. Increasing the value will always 6226make code faster, but eventually incurs high cost in increased code size. 6227 6228Note that on machines where the corresponding move insn is a 6229@code{define_expand} that emits a sequence of insns, this macro counts 6230the number of such sequences. 6231 6232The parameter @var{speed} is true if the code is currently being 6233optimized for speed rather than size. 6234 6235If you don't define this, a reasonable default is used. 6236@end defmac 6237 6238@defmac MOVE_BY_PIECES_P (@var{size}, @var{alignment}) 6239A C expression used to determine whether @code{move_by_pieces} will be used to 6240copy a chunk of memory, or whether some other block move mechanism 6241will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less 6242than @code{MOVE_RATIO}. 6243@end defmac 6244 6245@defmac MOVE_MAX_PIECES 6246A C expression used by @code{move_by_pieces} to determine the largest unit 6247a load or store used to copy memory is. Defaults to @code{MOVE_MAX}. 6248@end defmac 6249 6250@defmac CLEAR_RATIO (@var{speed}) 6251The threshold of number of scalar move insns, @emph{below} which a sequence 6252of insns should be generated to clear memory instead of a string clear insn 6253or a library call. Increasing the value will always make code faster, but 6254eventually incurs high cost in increased code size. 6255 6256The parameter @var{speed} is true if the code is currently being 6257optimized for speed rather than size. 6258 6259If you don't define this, a reasonable default is used. 6260@end defmac 6261 6262@defmac CLEAR_BY_PIECES_P (@var{size}, @var{alignment}) 6263A C expression used to determine whether @code{clear_by_pieces} will be used 6264to clear a chunk of memory, or whether some other block clear mechanism 6265will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less 6266than @code{CLEAR_RATIO}. 6267@end defmac 6268 6269@defmac SET_RATIO (@var{speed}) 6270The threshold of number of scalar move insns, @emph{below} which a sequence 6271of insns should be generated to set memory to a constant value, instead of 6272a block set insn or a library call. 6273Increasing the value will always make code faster, but 6274eventually incurs high cost in increased code size. 6275 6276The parameter @var{speed} is true if the code is currently being 6277optimized for speed rather than size. 6278 6279If you don't define this, it defaults to the value of @code{MOVE_RATIO}. 6280@end defmac 6281 6282@defmac SET_BY_PIECES_P (@var{size}, @var{alignment}) 6283A C expression used to determine whether @code{store_by_pieces} will be 6284used to set a chunk of memory to a constant value, or whether some 6285other mechanism will be used. Used by @code{__builtin_memset} when 6286storing values other than constant zero. 6287Defaults to 1 if @code{move_by_pieces_ninsns} returns less 6288than @code{SET_RATIO}. 6289@end defmac 6290 6291@defmac STORE_BY_PIECES_P (@var{size}, @var{alignment}) 6292A C expression used to determine whether @code{store_by_pieces} will be 6293used to set a chunk of memory to a constant string value, or whether some 6294other mechanism will be used. Used by @code{__builtin_strcpy} when 6295called with a constant source string. 6296Defaults to 1 if @code{move_by_pieces_ninsns} returns less 6297than @code{MOVE_RATIO}. 6298@end defmac 6299 6300@defmac USE_LOAD_POST_INCREMENT (@var{mode}) 6301A C expression used to determine whether a load postincrement is a good 6302thing to use for a given mode. Defaults to the value of 6303@code{HAVE_POST_INCREMENT}. 6304@end defmac 6305 6306@defmac USE_LOAD_POST_DECREMENT (@var{mode}) 6307A C expression used to determine whether a load postdecrement is a good 6308thing to use for a given mode. Defaults to the value of 6309@code{HAVE_POST_DECREMENT}. 6310@end defmac 6311 6312@defmac USE_LOAD_PRE_INCREMENT (@var{mode}) 6313A C expression used to determine whether a load preincrement is a good 6314thing to use for a given mode. Defaults to the value of 6315@code{HAVE_PRE_INCREMENT}. 6316@end defmac 6317 6318@defmac USE_LOAD_PRE_DECREMENT (@var{mode}) 6319A C expression used to determine whether a load predecrement is a good 6320thing to use for a given mode. Defaults to the value of 6321@code{HAVE_PRE_DECREMENT}. 6322@end defmac 6323 6324@defmac USE_STORE_POST_INCREMENT (@var{mode}) 6325A C expression used to determine whether a store postincrement is a good 6326thing to use for a given mode. Defaults to the value of 6327@code{HAVE_POST_INCREMENT}. 6328@end defmac 6329 6330@defmac USE_STORE_POST_DECREMENT (@var{mode}) 6331A C expression used to determine whether a store postdecrement is a good 6332thing to use for a given mode. Defaults to the value of 6333@code{HAVE_POST_DECREMENT}. 6334@end defmac 6335 6336@defmac USE_STORE_PRE_INCREMENT (@var{mode}) 6337This macro is used to determine whether a store preincrement is a good 6338thing to use for a given mode. Defaults to the value of 6339@code{HAVE_PRE_INCREMENT}. 6340@end defmac 6341 6342@defmac USE_STORE_PRE_DECREMENT (@var{mode}) 6343This macro is used to determine whether a store predecrement is a good 6344thing to use for a given mode. Defaults to the value of 6345@code{HAVE_PRE_DECREMENT}. 6346@end defmac 6347 6348@defmac NO_FUNCTION_CSE 6349Define this macro if it is as good or better to call a constant 6350function address than to call an address kept in a register. 6351@end defmac 6352 6353@defmac RANGE_TEST_NON_SHORT_CIRCUIT 6354Define this macro if a non-short-circuit operation produced by 6355@samp{fold_range_test ()} is optimal. This macro defaults to true if 6356@code{BRANCH_COST} is greater than or equal to the value 2. 6357@end defmac 6358 6359@hook TARGET_RTX_COSTS 6360This target hook describes the relative costs of RTL expressions. 6361 6362The cost may depend on the precise form of the expression, which is 6363available for examination in @var{x}, and the fact that @var{x} appears 6364as operand @var{opno} of an expression with rtx code @var{outer_code}. 6365That is, the hook can assume that there is some rtx @var{y} such 6366that @samp{GET_CODE (@var{y}) == @var{outer_code}} and such that 6367either (a) @samp{XEXP (@var{y}, @var{opno}) == @var{x}} or 6368(b) @samp{XVEC (@var{y}, @var{opno})} contains @var{x}. 6369 6370@var{code} is @var{x}'s expression code---redundant, since it can be 6371obtained with @code{GET_CODE (@var{x})}. 6372 6373In implementing this hook, you can use the construct 6374@code{COSTS_N_INSNS (@var{n})} to specify a cost equal to @var{n} fast 6375instructions. 6376 6377On entry to the hook, @code{*@var{total}} contains a default estimate 6378for the cost of the expression. The hook should modify this value as 6379necessary. Traditionally, the default costs are @code{COSTS_N_INSNS (5)} 6380for multiplications, @code{COSTS_N_INSNS (7)} for division and modulus 6381operations, and @code{COSTS_N_INSNS (1)} for all other operations. 6382 6383When optimizing for code size, i.e.@: when @code{speed} is 6384false, this target hook should be used to estimate the relative 6385size cost of an expression, again relative to @code{COSTS_N_INSNS}. 6386 6387The hook returns true when all subexpressions of @var{x} have been 6388processed, and false when @code{rtx_cost} should recurse. 6389@end deftypefn 6390 6391@hook TARGET_ADDRESS_COST 6392This hook computes the cost of an addressing mode that contains 6393@var{address}. If not defined, the cost is computed from 6394the @var{address} expression and the @code{TARGET_RTX_COST} hook. 6395 6396For most CISC machines, the default cost is a good approximation of the 6397true cost of the addressing mode. However, on RISC machines, all 6398instructions normally have the same length and execution time. Hence 6399all addresses will have equal costs. 6400 6401In cases where more than one form of an address is known, the form with 6402the lowest cost will be used. If multiple forms have the same, lowest, 6403cost, the one that is the most complex will be used. 6404 6405For example, suppose an address that is equal to the sum of a register 6406and a constant is used twice in the same basic block. When this macro 6407is not defined, the address will be computed in a register and memory 6408references will be indirect through that register. On machines where 6409the cost of the addressing mode containing the sum is no higher than 6410that of a simple indirect reference, this will produce an additional 6411instruction and possibly require an additional register. Proper 6412specification of this macro eliminates this overhead for such machines. 6413 6414This hook is never called with an invalid address. 6415 6416On machines where an address involving more than one register is as 6417cheap as an address computation involving only one register, defining 6418@code{TARGET_ADDRESS_COST} to reflect this can cause two registers to 6419be live over a region of code where only one would have been if 6420@code{TARGET_ADDRESS_COST} were not defined in that manner. This effect 6421should be considered in the definition of this macro. Equivalent costs 6422should probably only be given to addresses with different numbers of 6423registers on machines with lots of registers. 6424@end deftypefn 6425 6426@node Scheduling 6427@section Adjusting the Instruction Scheduler 6428 6429The instruction scheduler may need a fair amount of machine-specific 6430adjustment in order to produce good code. GCC provides several target 6431hooks for this purpose. It is usually enough to define just a few of 6432them: try the first ones in this list first. 6433 6434@hook TARGET_SCHED_ISSUE_RATE 6435This hook returns the maximum number of instructions that can ever 6436issue at the same time on the target machine. The default is one. 6437Although the insn scheduler can define itself the possibility of issue 6438an insn on the same cycle, the value can serve as an additional 6439constraint to issue insns on the same simulated processor cycle (see 6440hooks @samp{TARGET_SCHED_REORDER} and @samp{TARGET_SCHED_REORDER2}). 6441This value must be constant over the entire compilation. If you need 6442it to vary depending on what the instructions are, you must use 6443@samp{TARGET_SCHED_VARIABLE_ISSUE}. 6444@end deftypefn 6445 6446@hook TARGET_SCHED_VARIABLE_ISSUE 6447This hook is executed by the scheduler after it has scheduled an insn 6448from the ready list. It should return the number of insns which can 6449still be issued in the current cycle. The default is 6450@samp{@w{@var{more} - 1}} for insns other than @code{CLOBBER} and 6451@code{USE}, which normally are not counted against the issue rate. 6452You should define this hook if some insns take more machine resources 6453than others, so that fewer insns can follow them in the same cycle. 6454@var{file} is either a null pointer, or a stdio stream to write any 6455debug output to. @var{verbose} is the verbose level provided by 6456@option{-fsched-verbose-@var{n}}. @var{insn} is the instruction that 6457was scheduled. 6458@end deftypefn 6459 6460@hook TARGET_SCHED_ADJUST_COST 6461This function corrects the value of @var{cost} based on the 6462relationship between @var{insn} and @var{dep_insn} through the 6463dependence @var{link}. It should return the new value. The default 6464is to make no adjustment to @var{cost}. This can be used for example 6465to specify to the scheduler using the traditional pipeline description 6466that an output- or anti-dependence does not incur the same cost as a 6467data-dependence. If the scheduler using the automaton based pipeline 6468description, the cost of anti-dependence is zero and the cost of 6469output-dependence is maximum of one and the difference of latency 6470times of the first and the second insns. If these values are not 6471acceptable, you could use the hook to modify them too. See also 6472@pxref{Processor pipeline description}. 6473@end deftypefn 6474 6475@hook TARGET_SCHED_ADJUST_PRIORITY 6476This hook adjusts the integer scheduling priority @var{priority} of 6477@var{insn}. It should return the new priority. Increase the priority to 6478execute @var{insn} earlier, reduce the priority to execute @var{insn} 6479later. Do not define this hook if you do not need to adjust the 6480scheduling priorities of insns. 6481@end deftypefn 6482 6483@hook TARGET_SCHED_REORDER 6484This hook is executed by the scheduler after it has scheduled the ready 6485list, to allow the machine description to reorder it (for example to 6486combine two small instructions together on @samp{VLIW} machines). 6487@var{file} is either a null pointer, or a stdio stream to write any 6488debug output to. @var{verbose} is the verbose level provided by 6489@option{-fsched-verbose-@var{n}}. @var{ready} is a pointer to the ready 6490list of instructions that are ready to be scheduled. @var{n_readyp} is 6491a pointer to the number of elements in the ready list. The scheduler 6492reads the ready list in reverse order, starting with 6493@var{ready}[@var{*n_readyp} @minus{} 1] and going to @var{ready}[0]. @var{clock} 6494is the timer tick of the scheduler. You may modify the ready list and 6495the number of ready insns. The return value is the number of insns that 6496can issue this cycle; normally this is just @code{issue_rate}. See also 6497@samp{TARGET_SCHED_REORDER2}. 6498@end deftypefn 6499 6500@hook TARGET_SCHED_REORDER2 6501Like @samp{TARGET_SCHED_REORDER}, but called at a different time. That 6502function is called whenever the scheduler starts a new cycle. This one 6503is called once per iteration over a cycle, immediately after 6504@samp{TARGET_SCHED_VARIABLE_ISSUE}; it can reorder the ready list and 6505return the number of insns to be scheduled in the same cycle. Defining 6506this hook can be useful if there are frequent situations where 6507scheduling one insn causes other insns to become ready in the same 6508cycle. These other insns can then be taken into account properly. 6509@end deftypefn 6510 6511@hook TARGET_SCHED_DEPENDENCIES_EVALUATION_HOOK 6512This hook is called after evaluation forward dependencies of insns in 6513chain given by two parameter values (@var{head} and @var{tail} 6514correspondingly) but before insns scheduling of the insn chain. For 6515example, it can be used for better insn classification if it requires 6516analysis of dependencies. This hook can use backward and forward 6517dependencies of the insn scheduler because they are already 6518calculated. 6519@end deftypefn 6520 6521@hook TARGET_SCHED_INIT 6522This hook is executed by the scheduler at the beginning of each block of 6523instructions that are to be scheduled. @var{file} is either a null 6524pointer, or a stdio stream to write any debug output to. @var{verbose} 6525is the verbose level provided by @option{-fsched-verbose-@var{n}}. 6526@var{max_ready} is the maximum number of insns in the current scheduling 6527region that can be live at the same time. This can be used to allocate 6528scratch space if it is needed, e.g.@: by @samp{TARGET_SCHED_REORDER}. 6529@end deftypefn 6530 6531@hook TARGET_SCHED_FINISH 6532This hook is executed by the scheduler at the end of each block of 6533instructions that are to be scheduled. It can be used to perform 6534cleanup of any actions done by the other scheduling hooks. @var{file} 6535is either a null pointer, or a stdio stream to write any debug output 6536to. @var{verbose} is the verbose level provided by 6537@option{-fsched-verbose-@var{n}}. 6538@end deftypefn 6539 6540@hook TARGET_SCHED_INIT_GLOBAL 6541This hook is executed by the scheduler after function level initializations. 6542@var{file} is either a null pointer, or a stdio stream to write any debug output to. 6543@var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}. 6544@var{old_max_uid} is the maximum insn uid when scheduling begins. 6545@end deftypefn 6546 6547@hook TARGET_SCHED_FINISH_GLOBAL 6548This is the cleanup hook corresponding to @code{TARGET_SCHED_INIT_GLOBAL}. 6549@var{file} is either a null pointer, or a stdio stream to write any debug output to. 6550@var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}. 6551@end deftypefn 6552 6553@hook TARGET_SCHED_DFA_PRE_CYCLE_INSN 6554The hook returns an RTL insn. The automaton state used in the 6555pipeline hazard recognizer is changed as if the insn were scheduled 6556when the new simulated processor cycle starts. Usage of the hook may 6557simplify the automaton pipeline description for some @acronym{VLIW} 6558processors. If the hook is defined, it is used only for the automaton 6559based pipeline description. The default is not to change the state 6560when the new simulated processor cycle starts. 6561@end deftypefn 6562 6563@hook TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN 6564The hook can be used to initialize data used by the previous hook. 6565@end deftypefn 6566 6567@hook TARGET_SCHED_DFA_POST_CYCLE_INSN 6568The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used 6569to changed the state as if the insn were scheduled when the new 6570simulated processor cycle finishes. 6571@end deftypefn 6572 6573@hook TARGET_SCHED_INIT_DFA_POST_CYCLE_INSN 6574The hook is analogous to @samp{TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN} but 6575used to initialize data used by the previous hook. 6576@end deftypefn 6577 6578@hook TARGET_SCHED_DFA_PRE_ADVANCE_CYCLE 6579The hook to notify target that the current simulated cycle is about to finish. 6580The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used 6581to change the state in more complicated situations - e.g., when advancing 6582state on a single insn is not enough. 6583@end deftypefn 6584 6585@hook TARGET_SCHED_DFA_POST_ADVANCE_CYCLE 6586The hook to notify target that new simulated cycle has just started. 6587The hook is analogous to @samp{TARGET_SCHED_DFA_POST_CYCLE_INSN} but used 6588to change the state in more complicated situations - e.g., when advancing 6589state on a single insn is not enough. 6590@end deftypefn 6591 6592@hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD 6593This hook controls better choosing an insn from the ready insn queue 6594for the @acronym{DFA}-based insn scheduler. Usually the scheduler 6595chooses the first insn from the queue. If the hook returns a positive 6596value, an additional scheduler code tries all permutations of 6597@samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD ()} 6598subsequent ready insns to choose an insn whose issue will result in 6599maximal number of issued insns on the same cycle. For the 6600@acronym{VLIW} processor, the code could actually solve the problem of 6601packing simple insns into the @acronym{VLIW} insn. Of course, if the 6602rules of @acronym{VLIW} packing are described in the automaton. 6603 6604This code also could be used for superscalar @acronym{RISC} 6605processors. Let us consider a superscalar @acronym{RISC} processor 6606with 3 pipelines. Some insns can be executed in pipelines @var{A} or 6607@var{B}, some insns can be executed only in pipelines @var{B} or 6608@var{C}, and one insn can be executed in pipeline @var{B}. The 6609processor may issue the 1st insn into @var{A} and the 2nd one into 6610@var{B}. In this case, the 3rd insn will wait for freeing @var{B} 6611until the next cycle. If the scheduler issues the 3rd insn the first, 6612the processor could issue all 3 insns per cycle. 6613 6614Actually this code demonstrates advantages of the automaton based 6615pipeline hazard recognizer. We try quickly and easy many insn 6616schedules to choose the best one. 6617 6618The default is no multipass scheduling. 6619@end deftypefn 6620 6621@hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD 6622 6623This hook controls what insns from the ready insn queue will be 6624considered for the multipass insn scheduling. If the hook returns 6625zero for @var{insn}, the insn will be not chosen to 6626be issued. 6627 6628The default is that any ready insns can be chosen to be issued. 6629@end deftypefn 6630 6631@hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_BEGIN 6632This hook prepares the target backend for a new round of multipass 6633scheduling. 6634@end deftypefn 6635 6636@hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_ISSUE 6637This hook is called when multipass scheduling evaluates instruction INSN. 6638@end deftypefn 6639 6640@hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_BACKTRACK 6641This is called when multipass scheduling backtracks from evaluation of 6642an instruction. 6643@end deftypefn 6644 6645@hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_END 6646This hook notifies the target about the result of the concluded current 6647round of multipass scheduling. 6648@end deftypefn 6649 6650@hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_INIT 6651This hook initializes target-specific data used in multipass scheduling. 6652@end deftypefn 6653 6654@hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_FINI 6655This hook finalizes target-specific data used in multipass scheduling. 6656@end deftypefn 6657 6658@hook TARGET_SCHED_DFA_NEW_CYCLE 6659This hook is called by the insn scheduler before issuing @var{insn} 6660on cycle @var{clock}. If the hook returns nonzero, 6661@var{insn} is not issued on this processor cycle. Instead, 6662the processor cycle is advanced. If *@var{sort_p} 6663is zero, the insn ready queue is not sorted on the new cycle 6664start as usually. @var{dump} and @var{verbose} specify the file and 6665verbosity level to use for debugging output. 6666@var{last_clock} and @var{clock} are, respectively, the 6667processor cycle on which the previous insn has been issued, 6668and the current processor cycle. 6669@end deftypefn 6670 6671@hook TARGET_SCHED_IS_COSTLY_DEPENDENCE 6672This hook is used to define which dependences are considered costly by 6673the target, so costly that it is not advisable to schedule the insns that 6674are involved in the dependence too close to one another. The parameters 6675to this hook are as follows: The first parameter @var{_dep} is the dependence 6676being evaluated. The second parameter @var{cost} is the cost of the 6677dependence as estimated by the scheduler, and the third 6678parameter @var{distance} is the distance in cycles between the two insns. 6679The hook returns @code{true} if considering the distance between the two 6680insns the dependence between them is considered costly by the target, 6681and @code{false} otherwise. 6682 6683Defining this hook can be useful in multiple-issue out-of-order machines, 6684where (a) it's practically hopeless to predict the actual data/resource 6685delays, however: (b) there's a better chance to predict the actual grouping 6686that will be formed, and (c) correctly emulating the grouping can be very 6687important. In such targets one may want to allow issuing dependent insns 6688closer to one another---i.e., closer than the dependence distance; however, 6689not in cases of ``costly dependences'', which this hooks allows to define. 6690@end deftypefn 6691 6692@hook TARGET_SCHED_H_I_D_EXTENDED 6693This hook is called by the insn scheduler after emitting a new instruction to 6694the instruction stream. The hook notifies a target backend to extend its 6695per instruction data structures. 6696@end deftypefn 6697 6698@hook TARGET_SCHED_ALLOC_SCHED_CONTEXT 6699Return a pointer to a store large enough to hold target scheduling context. 6700@end deftypefn 6701 6702@hook TARGET_SCHED_INIT_SCHED_CONTEXT 6703Initialize store pointed to by @var{tc} to hold target scheduling context. 6704It @var{clean_p} is true then initialize @var{tc} as if scheduler is at the 6705beginning of the block. Otherwise, copy the current context into @var{tc}. 6706@end deftypefn 6707 6708@hook TARGET_SCHED_SET_SCHED_CONTEXT 6709Copy target scheduling context pointed to by @var{tc} to the current context. 6710@end deftypefn 6711 6712@hook TARGET_SCHED_CLEAR_SCHED_CONTEXT 6713Deallocate internal data in target scheduling context pointed to by @var{tc}. 6714@end deftypefn 6715 6716@hook TARGET_SCHED_FREE_SCHED_CONTEXT 6717Deallocate a store for target scheduling context pointed to by @var{tc}. 6718@end deftypefn 6719 6720@hook TARGET_SCHED_SPECULATE_INSN 6721This hook is called by the insn scheduler when @var{insn} has only 6722speculative dependencies and therefore can be scheduled speculatively. 6723The hook is used to check if the pattern of @var{insn} has a speculative 6724version and, in case of successful check, to generate that speculative 6725pattern. The hook should return 1, if the instruction has a speculative form, 6726or @minus{}1, if it doesn't. @var{request} describes the type of requested 6727speculation. If the return value equals 1 then @var{new_pat} is assigned 6728the generated speculative pattern. 6729@end deftypefn 6730 6731@hook TARGET_SCHED_NEEDS_BLOCK_P 6732This hook is called by the insn scheduler during generation of recovery code 6733for @var{insn}. It should return @code{true}, if the corresponding check 6734instruction should branch to recovery code, or @code{false} otherwise. 6735@end deftypefn 6736 6737@hook TARGET_SCHED_GEN_SPEC_CHECK 6738This hook is called by the insn scheduler to generate a pattern for recovery 6739check instruction. If @var{mutate_p} is zero, then @var{insn} is a 6740speculative instruction for which the check should be generated. 6741@var{label} is either a label of a basic block, where recovery code should 6742be emitted, or a null pointer, when requested check doesn't branch to 6743recovery code (a simple check). If @var{mutate_p} is nonzero, then 6744a pattern for a branchy check corresponding to a simple check denoted by 6745@var{insn} should be generated. In this case @var{label} can't be null. 6746@end deftypefn 6747 6748@hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD_SPEC 6749This hook is used as a workaround for 6750@samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD} not being 6751called on the first instruction of the ready list. The hook is used to 6752discard speculative instructions that stand first in the ready list from 6753being scheduled on the current cycle. If the hook returns @code{false}, 6754@var{insn} will not be chosen to be issued. 6755For non-speculative instructions, 6756the hook should always return @code{true}. For example, in the ia64 backend 6757the hook is used to cancel data speculative insns when the ALAT table 6758is nearly full. 6759@end deftypefn 6760 6761@hook TARGET_SCHED_SET_SCHED_FLAGS 6762This hook is used by the insn scheduler to find out what features should be 6763enabled/used. 6764The structure *@var{spec_info} should be filled in by the target. 6765The structure describes speculation types that can be used in the scheduler. 6766@end deftypefn 6767 6768@hook TARGET_SCHED_SMS_RES_MII 6769This hook is called by the swing modulo scheduler to calculate a 6770resource-based lower bound which is based on the resources available in 6771the machine and the resources required by each instruction. The target 6772backend can use @var{g} to calculate such bound. A very simple lower 6773bound will be used in case this hook is not implemented: the total number 6774of instructions divided by the issue rate. 6775@end deftypefn 6776 6777@hook TARGET_SCHED_DISPATCH 6778This hook is called by Haifa Scheduler. It returns true if dispatch scheduling 6779is supported in hardware and the condition specified in the parameter is true. 6780@end deftypefn 6781 6782@hook TARGET_SCHED_DISPATCH_DO 6783This hook is called by Haifa Scheduler. It performs the operation specified 6784in its second parameter. 6785@end deftypefn 6786 6787@hook TARGET_SCHED_EXPOSED_PIPELINE 6788 6789@hook TARGET_SCHED_REASSOCIATION_WIDTH 6790 6791@node Sections 6792@section Dividing the Output into Sections (Texts, Data, @dots{}) 6793@c the above section title is WAY too long. maybe cut the part between 6794@c the (...)? --mew 10feb93 6795 6796An object file is divided into sections containing different types of 6797data. In the most common case, there are three sections: the @dfn{text 6798section}, which holds instructions and read-only data; the @dfn{data 6799section}, which holds initialized writable data; and the @dfn{bss 6800section}, which holds uninitialized data. Some systems have other kinds 6801of sections. 6802 6803@file{varasm.c} provides several well-known sections, such as 6804@code{text_section}, @code{data_section} and @code{bss_section}. 6805The normal way of controlling a @code{@var{foo}_section} variable 6806is to define the associated @code{@var{FOO}_SECTION_ASM_OP} macro, 6807as described below. The macros are only read once, when @file{varasm.c} 6808initializes itself, so their values must be run-time constants. 6809They may however depend on command-line flags. 6810 6811@emph{Note:} Some run-time files, such @file{crtstuff.c}, also make 6812use of the @code{@var{FOO}_SECTION_ASM_OP} macros, and expect them 6813to be string literals. 6814 6815Some assemblers require a different string to be written every time a 6816section is selected. If your assembler falls into this category, you 6817should define the @code{TARGET_ASM_INIT_SECTIONS} hook and use 6818@code{get_unnamed_section} to set up the sections. 6819 6820You must always create a @code{text_section}, either by defining 6821@code{TEXT_SECTION_ASM_OP} or by initializing @code{text_section} 6822in @code{TARGET_ASM_INIT_SECTIONS}. The same is true of 6823@code{data_section} and @code{DATA_SECTION_ASM_OP}. If you do not 6824create a distinct @code{readonly_data_section}, the default is to 6825reuse @code{text_section}. 6826 6827All the other @file{varasm.c} sections are optional, and are null 6828if the target does not provide them. 6829 6830@defmac TEXT_SECTION_ASM_OP 6831A C expression whose value is a string, including spacing, containing the 6832assembler operation that should precede instructions and read-only data. 6833Normally @code{"\t.text"} is right. 6834@end defmac 6835 6836@defmac HOT_TEXT_SECTION_NAME 6837If defined, a C string constant for the name of the section containing most 6838frequently executed functions of the program. If not defined, GCC will provide 6839a default definition if the target supports named sections. 6840@end defmac 6841 6842@defmac UNLIKELY_EXECUTED_TEXT_SECTION_NAME 6843If defined, a C string constant for the name of the section containing unlikely 6844executed functions in the program. 6845@end defmac 6846 6847@defmac DATA_SECTION_ASM_OP 6848A C expression whose value is a string, including spacing, containing the 6849assembler operation to identify the following data as writable initialized 6850data. Normally @code{"\t.data"} is right. 6851@end defmac 6852 6853@defmac SDATA_SECTION_ASM_OP 6854If defined, a C expression whose value is a string, including spacing, 6855containing the assembler operation to identify the following data as 6856initialized, writable small data. 6857@end defmac 6858 6859@defmac READONLY_DATA_SECTION_ASM_OP 6860A C expression whose value is a string, including spacing, containing the 6861assembler operation to identify the following data as read-only initialized 6862data. 6863@end defmac 6864 6865@defmac BSS_SECTION_ASM_OP 6866If defined, a C expression whose value is a string, including spacing, 6867containing the assembler operation to identify the following data as 6868uninitialized global data. If not defined, and 6869@code{ASM_OUTPUT_ALIGNED_BSS} not defined, 6870uninitialized global data will be output in the data section if 6871@option{-fno-common} is passed, otherwise @code{ASM_OUTPUT_COMMON} will be 6872used. 6873@end defmac 6874 6875@defmac SBSS_SECTION_ASM_OP 6876If defined, a C expression whose value is a string, including spacing, 6877containing the assembler operation to identify the following data as 6878uninitialized, writable small data. 6879@end defmac 6880 6881@defmac TLS_COMMON_ASM_OP 6882If defined, a C expression whose value is a string containing the 6883assembler operation to identify the following data as thread-local 6884common data. The default is @code{".tls_common"}. 6885@end defmac 6886 6887@defmac TLS_SECTION_ASM_FLAG 6888If defined, a C expression whose value is a character constant 6889containing the flag used to mark a section as a TLS section. The 6890default is @code{'T'}. 6891@end defmac 6892 6893@defmac INIT_SECTION_ASM_OP 6894If defined, a C expression whose value is a string, including spacing, 6895containing the assembler operation to identify the following data as 6896initialization code. If not defined, GCC will assume such a section does 6897not exist. This section has no corresponding @code{init_section} 6898variable; it is used entirely in runtime code. 6899@end defmac 6900 6901@defmac FINI_SECTION_ASM_OP 6902If defined, a C expression whose value is a string, including spacing, 6903containing the assembler operation to identify the following data as 6904finalization code. If not defined, GCC will assume such a section does 6905not exist. This section has no corresponding @code{fini_section} 6906variable; it is used entirely in runtime code. 6907@end defmac 6908 6909@defmac INIT_ARRAY_SECTION_ASM_OP 6910If defined, a C expression whose value is a string, including spacing, 6911containing the assembler operation to identify the following data as 6912part of the @code{.init_array} (or equivalent) section. If not 6913defined, GCC will assume such a section does not exist. Do not define 6914both this macro and @code{INIT_SECTION_ASM_OP}. 6915@end defmac 6916 6917@defmac FINI_ARRAY_SECTION_ASM_OP 6918If defined, a C expression whose value is a string, including spacing, 6919containing the assembler operation to identify the following data as 6920part of the @code{.fini_array} (or equivalent) section. If not 6921defined, GCC will assume such a section does not exist. Do not define 6922both this macro and @code{FINI_SECTION_ASM_OP}. 6923@end defmac 6924 6925@defmac CRT_CALL_STATIC_FUNCTION (@var{section_op}, @var{function}) 6926If defined, an ASM statement that switches to a different section 6927via @var{section_op}, calls @var{function}, and switches back to 6928the text section. This is used in @file{crtstuff.c} if 6929@code{INIT_SECTION_ASM_OP} or @code{FINI_SECTION_ASM_OP} to calls 6930to initialization and finalization functions from the init and fini 6931sections. By default, this macro uses a simple function call. Some 6932ports need hand-crafted assembly code to avoid dependencies on 6933registers initialized in the function prologue or to ensure that 6934constant pools don't end up too far way in the text section. 6935@end defmac 6936 6937@defmac TARGET_LIBGCC_SDATA_SECTION 6938If defined, a string which names the section into which small 6939variables defined in crtstuff and libgcc should go. This is useful 6940when the target has options for optimizing access to small data, and 6941you want the crtstuff and libgcc routines to be conservative in what 6942they expect of your application yet liberal in what your application 6943expects. For example, for targets with a @code{.sdata} section (like 6944MIPS), you could compile crtstuff with @code{-G 0} so that it doesn't 6945require small data support from your application, but use this macro 6946to put small data into @code{.sdata} so that your application can 6947access these variables whether it uses small data or not. 6948@end defmac 6949 6950@defmac FORCE_CODE_SECTION_ALIGN 6951If defined, an ASM statement that aligns a code section to some 6952arbitrary boundary. This is used to force all fragments of the 6953@code{.init} and @code{.fini} sections to have to same alignment 6954and thus prevent the linker from having to add any padding. 6955@end defmac 6956 6957@defmac JUMP_TABLES_IN_TEXT_SECTION 6958Define this macro to be an expression with a nonzero value if jump 6959tables (for @code{tablejump} insns) should be output in the text 6960section, along with the assembler instructions. Otherwise, the 6961readonly data section is used. 6962 6963This macro is irrelevant if there is no separate readonly data section. 6964@end defmac 6965 6966@hook TARGET_ASM_INIT_SECTIONS 6967Define this hook if you need to do something special to set up the 6968@file{varasm.c} sections, or if your target has some special sections 6969of its own that you need to create. 6970 6971GCC calls this hook after processing the command line, but before writing 6972any assembly code, and before calling any of the section-returning hooks 6973described below. 6974@end deftypefn 6975 6976@hook TARGET_ASM_RELOC_RW_MASK 6977Return a mask describing how relocations should be treated when 6978selecting sections. Bit 1 should be set if global relocations 6979should be placed in a read-write section; bit 0 should be set if 6980local relocations should be placed in a read-write section. 6981 6982The default version of this function returns 3 when @option{-fpic} 6983is in effect, and 0 otherwise. The hook is typically redefined 6984when the target cannot support (some kinds of) dynamic relocations 6985in read-only sections even in executables. 6986@end deftypefn 6987 6988@hook TARGET_ASM_SELECT_SECTION 6989Return the section into which @var{exp} should be placed. You can 6990assume that @var{exp} is either a @code{VAR_DECL} node or a constant of 6991some sort. @var{reloc} indicates whether the initial value of @var{exp} 6992requires link-time relocations. Bit 0 is set when variable contains 6993local relocations only, while bit 1 is set for global relocations. 6994@var{align} is the constant alignment in bits. 6995 6996The default version of this function takes care of putting read-only 6997variables in @code{readonly_data_section}. 6998 6999See also @var{USE_SELECT_SECTION_FOR_FUNCTIONS}. 7000@end deftypefn 7001 7002@defmac USE_SELECT_SECTION_FOR_FUNCTIONS 7003Define this macro if you wish TARGET_ASM_SELECT_SECTION to be called 7004for @code{FUNCTION_DECL}s as well as for variables and constants. 7005 7006In the case of a @code{FUNCTION_DECL}, @var{reloc} will be zero if the 7007function has been determined to be likely to be called, and nonzero if 7008it is unlikely to be called. 7009@end defmac 7010 7011@hook TARGET_ASM_UNIQUE_SECTION 7012Build up a unique section name, expressed as a @code{STRING_CST} node, 7013and assign it to @samp{DECL_SECTION_NAME (@var{decl})}. 7014As with @code{TARGET_ASM_SELECT_SECTION}, @var{reloc} indicates whether 7015the initial value of @var{exp} requires link-time relocations. 7016 7017The default version of this function appends the symbol name to the 7018ELF section name that would normally be used for the symbol. For 7019example, the function @code{foo} would be placed in @code{.text.foo}. 7020Whatever the actual target object format, this is often good enough. 7021@end deftypefn 7022 7023@hook TARGET_ASM_FUNCTION_RODATA_SECTION 7024Return the readonly data section associated with 7025@samp{DECL_SECTION_NAME (@var{decl})}. 7026The default version of this function selects @code{.gnu.linkonce.r.name} if 7027the function's section is @code{.gnu.linkonce.t.name}, @code{.rodata.name} 7028if function is in @code{.text.name}, and the normal readonly-data section 7029otherwise. 7030@end deftypefn 7031 7032@hook TARGET_ASM_MERGEABLE_RODATA_PREFIX 7033 7034@hook TARGET_ASM_TM_CLONE_TABLE_SECTION 7035 7036@hook TARGET_ASM_SELECT_RTX_SECTION 7037Return the section into which a constant @var{x}, of mode @var{mode}, 7038should be placed. You can assume that @var{x} is some kind of 7039constant in RTL@. The argument @var{mode} is redundant except in the 7040case of a @code{const_int} rtx. @var{align} is the constant alignment 7041in bits. 7042 7043The default version of this function takes care of putting symbolic 7044constants in @code{flag_pic} mode in @code{data_section} and everything 7045else in @code{readonly_data_section}. 7046@end deftypefn 7047 7048@hook TARGET_MANGLE_DECL_ASSEMBLER_NAME 7049Define this hook if you need to postprocess the assembler name generated 7050by target-independent code. The @var{id} provided to this hook will be 7051the computed name (e.g., the macro @code{DECL_NAME} of the @var{decl} in C, 7052or the mangled name of the @var{decl} in C++). The return value of the 7053hook is an @code{IDENTIFIER_NODE} for the appropriate mangled name on 7054your target system. The default implementation of this hook just 7055returns the @var{id} provided. 7056@end deftypefn 7057 7058@hook TARGET_ENCODE_SECTION_INFO 7059Define this hook if references to a symbol or a constant must be 7060treated differently depending on something about the variable or 7061function named by the symbol (such as what section it is in). 7062 7063The hook is executed immediately after rtl has been created for 7064@var{decl}, which may be a variable or function declaration or 7065an entry in the constant pool. In either case, @var{rtl} is the 7066rtl in question. Do @emph{not} use @code{DECL_RTL (@var{decl})} 7067in this hook; that field may not have been initialized yet. 7068 7069In the case of a constant, it is safe to assume that the rtl is 7070a @code{mem} whose address is a @code{symbol_ref}. Most decls 7071will also have this form, but that is not guaranteed. Global 7072register variables, for instance, will have a @code{reg} for their 7073rtl. (Normally the right thing to do with such unusual rtl is 7074leave it alone.) 7075 7076The @var{new_decl_p} argument will be true if this is the first time 7077that @code{TARGET_ENCODE_SECTION_INFO} has been invoked on this decl. It will 7078be false for subsequent invocations, which will happen for duplicate 7079declarations. Whether or not anything must be done for the duplicate 7080declaration depends on whether the hook examines @code{DECL_ATTRIBUTES}. 7081@var{new_decl_p} is always true when the hook is called for a constant. 7082 7083@cindex @code{SYMBOL_REF_FLAG}, in @code{TARGET_ENCODE_SECTION_INFO} 7084The usual thing for this hook to do is to record flags in the 7085@code{symbol_ref}, using @code{SYMBOL_REF_FLAG} or @code{SYMBOL_REF_FLAGS}. 7086Historically, the name string was modified if it was necessary to 7087encode more than one bit of information, but this practice is now 7088discouraged; use @code{SYMBOL_REF_FLAGS}. 7089 7090The default definition of this hook, @code{default_encode_section_info} 7091in @file{varasm.c}, sets a number of commonly-useful bits in 7092@code{SYMBOL_REF_FLAGS}. Check whether the default does what you need 7093before overriding it. 7094@end deftypefn 7095 7096@hook TARGET_STRIP_NAME_ENCODING 7097Decode @var{name} and return the real name part, sans 7098the characters that @code{TARGET_ENCODE_SECTION_INFO} 7099may have added. 7100@end deftypefn 7101 7102@hook TARGET_IN_SMALL_DATA_P 7103Returns true if @var{exp} should be placed into a ``small data'' section. 7104The default version of this hook always returns false. 7105@end deftypefn 7106 7107@hook TARGET_HAVE_SRODATA_SECTION 7108Contains the value true if the target places read-only 7109``small data'' into a separate section. The default value is false. 7110@end deftypevr 7111 7112@hook TARGET_PROFILE_BEFORE_PROLOGUE 7113 7114@hook TARGET_BINDS_LOCAL_P 7115Returns true if @var{exp} names an object for which name resolution 7116rules must resolve to the current ``module'' (dynamic shared library 7117or executable image). 7118 7119The default version of this hook implements the name resolution rules 7120for ELF, which has a looser model of global name binding than other 7121currently supported object file formats. 7122@end deftypefn 7123 7124@hook TARGET_HAVE_TLS 7125Contains the value true if the target supports thread-local storage. 7126The default value is false. 7127@end deftypevr 7128 7129 7130@node PIC 7131@section Position Independent Code 7132@cindex position independent code 7133@cindex PIC 7134 7135This section describes macros that help implement generation of position 7136independent code. Simply defining these macros is not enough to 7137generate valid PIC; you must also add support to the hook 7138@code{TARGET_LEGITIMATE_ADDRESS_P} and to the macro 7139@code{PRINT_OPERAND_ADDRESS}, as well as @code{LEGITIMIZE_ADDRESS}. You 7140must modify the definition of @samp{movsi} to do something appropriate 7141when the source operand contains a symbolic address. You may also 7142need to alter the handling of switch statements so that they use 7143relative addresses. 7144@c i rearranged the order of the macros above to try to force one of 7145@c them to the next line, to eliminate an overfull hbox. --mew 10feb93 7146 7147@defmac PIC_OFFSET_TABLE_REGNUM 7148The register number of the register used to address a table of static 7149data addresses in memory. In some cases this register is defined by a 7150processor's ``application binary interface'' (ABI)@. When this macro 7151is defined, RTL is generated for this register once, as with the stack 7152pointer and frame pointer registers. If this macro is not defined, it 7153is up to the machine-dependent files to allocate such a register (if 7154necessary). Note that this register must be fixed when in use (e.g.@: 7155when @code{flag_pic} is true). 7156@end defmac 7157 7158@defmac PIC_OFFSET_TABLE_REG_CALL_CLOBBERED 7159A C expression that is nonzero if the register defined by 7160@code{PIC_OFFSET_TABLE_REGNUM} is clobbered by calls. If not defined, 7161the default is zero. Do not define 7162this macro if @code{PIC_OFFSET_TABLE_REGNUM} is not defined. 7163@end defmac 7164 7165@defmac LEGITIMATE_PIC_OPERAND_P (@var{x}) 7166A C expression that is nonzero if @var{x} is a legitimate immediate 7167operand on the target machine when generating position independent code. 7168You can assume that @var{x} satisfies @code{CONSTANT_P}, so you need not 7169check this. You can also assume @var{flag_pic} is true, so you need not 7170check it either. You need not define this macro if all constants 7171(including @code{SYMBOL_REF}) can be immediate operands when generating 7172position independent code. 7173@end defmac 7174 7175@node Assembler Format 7176@section Defining the Output Assembler Language 7177 7178This section describes macros whose principal purpose is to describe how 7179to write instructions in assembler language---rather than what the 7180instructions do. 7181 7182@menu 7183* File Framework:: Structural information for the assembler file. 7184* Data Output:: Output of constants (numbers, strings, addresses). 7185* Uninitialized Data:: Output of uninitialized variables. 7186* Label Output:: Output and generation of labels. 7187* Initialization:: General principles of initialization 7188 and termination routines. 7189* Macros for Initialization:: 7190 Specific macros that control the handling of 7191 initialization and termination routines. 7192* Instruction Output:: Output of actual instructions. 7193* Dispatch Tables:: Output of jump tables. 7194* Exception Region Output:: Output of exception region code. 7195* Alignment Output:: Pseudo ops for alignment and skipping data. 7196@end menu 7197 7198@node File Framework 7199@subsection The Overall Framework of an Assembler File 7200@cindex assembler format 7201@cindex output of assembler code 7202 7203@c prevent bad page break with this line 7204This describes the overall framework of an assembly file. 7205 7206@findex default_file_start 7207@hook TARGET_ASM_FILE_START 7208Output to @code{asm_out_file} any text which the assembler expects to 7209find at the beginning of a file. The default behavior is controlled 7210by two flags, documented below. Unless your target's assembler is 7211quite unusual, if you override the default, you should call 7212@code{default_file_start} at some point in your target hook. This 7213lets other target files rely on these variables. 7214@end deftypefn 7215 7216@hook TARGET_ASM_FILE_START_APP_OFF 7217If this flag is true, the text of the macro @code{ASM_APP_OFF} will be 7218printed as the very first line in the assembly file, unless 7219@option{-fverbose-asm} is in effect. (If that macro has been defined 7220to the empty string, this variable has no effect.) With the normal 7221definition of @code{ASM_APP_OFF}, the effect is to notify the GNU 7222assembler that it need not bother stripping comments or extra 7223whitespace from its input. This allows it to work a bit faster. 7224 7225The default is false. You should not set it to true unless you have 7226verified that your port does not generate any extra whitespace or 7227comments that will cause GAS to issue errors in NO_APP mode. 7228@end deftypevr 7229 7230@hook TARGET_ASM_FILE_START_FILE_DIRECTIVE 7231If this flag is true, @code{output_file_directive} will be called 7232for the primary source file, immediately after printing 7233@code{ASM_APP_OFF} (if that is enabled). Most ELF assemblers expect 7234this to be done. The default is false. 7235@end deftypevr 7236 7237@hook TARGET_ASM_FILE_END 7238Output to @code{asm_out_file} any text which the assembler expects 7239to find at the end of a file. The default is to output nothing. 7240@end deftypefn 7241 7242@deftypefun void file_end_indicate_exec_stack () 7243Some systems use a common convention, the @samp{.note.GNU-stack} 7244special section, to indicate whether or not an object file relies on 7245the stack being executable. If your system uses this convention, you 7246should define @code{TARGET_ASM_FILE_END} to this function. If you 7247need to do other things in that hook, have your hook function call 7248this function. 7249@end deftypefun 7250 7251@hook TARGET_ASM_LTO_START 7252Output to @code{asm_out_file} any text which the assembler expects 7253to find at the start of an LTO section. The default is to output 7254nothing. 7255@end deftypefn 7256 7257@hook TARGET_ASM_LTO_END 7258Output to @code{asm_out_file} any text which the assembler expects 7259to find at the end of an LTO section. The default is to output 7260nothing. 7261@end deftypefn 7262 7263@hook TARGET_ASM_CODE_END 7264Output to @code{asm_out_file} any text which is needed before emitting 7265unwind info and debug info at the end of a file. Some targets emit 7266here PIC setup thunks that cannot be emitted at the end of file, 7267because they couldn't have unwind info then. The default is to output 7268nothing. 7269@end deftypefn 7270 7271@defmac ASM_COMMENT_START 7272A C string constant describing how to begin a comment in the target 7273assembler language. The compiler assumes that the comment will end at 7274the end of the line. 7275@end defmac 7276 7277@defmac ASM_APP_ON 7278A C string constant for text to be output before each @code{asm} 7279statement or group of consecutive ones. Normally this is 7280@code{"#APP"}, which is a comment that has no effect on most 7281assemblers but tells the GNU assembler that it must check the lines 7282that follow for all valid assembler constructs. 7283@end defmac 7284 7285@defmac ASM_APP_OFF 7286A C string constant for text to be output after each @code{asm} 7287statement or group of consecutive ones. Normally this is 7288@code{"#NO_APP"}, which tells the GNU assembler to resume making the 7289time-saving assumptions that are valid for ordinary compiler output. 7290@end defmac 7291 7292@defmac ASM_OUTPUT_SOURCE_FILENAME (@var{stream}, @var{name}) 7293A C statement to output COFF information or DWARF debugging information 7294which indicates that filename @var{name} is the current source file to 7295the stdio stream @var{stream}. 7296 7297This macro need not be defined if the standard form of output 7298for the file format in use is appropriate. 7299@end defmac 7300 7301@hook TARGET_ASM_OUTPUT_SOURCE_FILENAME 7302 7303@defmac OUTPUT_QUOTED_STRING (@var{stream}, @var{string}) 7304A C statement to output the string @var{string} to the stdio stream 7305@var{stream}. If you do not call the function @code{output_quoted_string} 7306in your config files, GCC will only call it to output filenames to 7307the assembler source. So you can use it to canonicalize the format 7308of the filename using this macro. 7309@end defmac 7310 7311@defmac ASM_OUTPUT_IDENT (@var{stream}, @var{string}) 7312A C statement to output something to the assembler file to handle a 7313@samp{#ident} directive containing the text @var{string}. If this 7314macro is not defined, nothing is output for a @samp{#ident} directive. 7315@end defmac 7316 7317@hook TARGET_ASM_NAMED_SECTION 7318Output assembly directives to switch to section @var{name}. The section 7319should have attributes as specified by @var{flags}, which is a bit mask 7320of the @code{SECTION_*} flags defined in @file{output.h}. If @var{decl} 7321is non-NULL, it is the @code{VAR_DECL} or @code{FUNCTION_DECL} with which 7322this section is associated. 7323@end deftypefn 7324 7325@hook TARGET_ASM_FUNCTION_SECTION 7326Return preferred text (sub)section for function @var{decl}. 7327Main purpose of this function is to separate cold, normal and hot 7328functions. @var{startup} is true when function is known to be used only 7329at startup (from static constructors or it is @code{main()}). 7330@var{exit} is true when function is known to be used only at exit 7331(from static destructors). 7332Return NULL if function should go to default text section. 7333@end deftypefn 7334 7335@hook TARGET_ASM_FUNCTION_SWITCHED_TEXT_SECTIONS 7336 7337@hook TARGET_HAVE_NAMED_SECTIONS 7338This flag is true if the target supports @code{TARGET_ASM_NAMED_SECTION}. 7339It must not be modified by command-line option processing. 7340@end deftypevr 7341 7342@anchor{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS} 7343@hook TARGET_HAVE_SWITCHABLE_BSS_SECTIONS 7344This flag is true if we can create zeroed data by switching to a BSS 7345section and then using @code{ASM_OUTPUT_SKIP} to allocate the space. 7346This is true on most ELF targets. 7347@end deftypevr 7348 7349@hook TARGET_SECTION_TYPE_FLAGS 7350Choose a set of section attributes for use by @code{TARGET_ASM_NAMED_SECTION} 7351based on a variable or function decl, a section name, and whether or not the 7352declaration's initializer may contain runtime relocations. @var{decl} may be 7353null, in which case read-write data should be assumed. 7354 7355The default version of this function handles choosing code vs data, 7356read-only vs read-write data, and @code{flag_pic}. You should only 7357need to override this if your target has special flags that might be 7358set via @code{__attribute__}. 7359@end deftypefn 7360 7361@hook TARGET_ASM_RECORD_GCC_SWITCHES 7362Provides the target with the ability to record the gcc command line 7363switches that have been passed to the compiler, and options that are 7364enabled. The @var{type} argument specifies what is being recorded. 7365It can take the following values: 7366 7367@table @gcctabopt 7368@item SWITCH_TYPE_PASSED 7369@var{text} is a command line switch that has been set by the user. 7370 7371@item SWITCH_TYPE_ENABLED 7372@var{text} is an option which has been enabled. This might be as a 7373direct result of a command line switch, or because it is enabled by 7374default or because it has been enabled as a side effect of a different 7375command line switch. For example, the @option{-O2} switch enables 7376various different individual optimization passes. 7377 7378@item SWITCH_TYPE_DESCRIPTIVE 7379@var{text} is either NULL or some descriptive text which should be 7380ignored. If @var{text} is NULL then it is being used to warn the 7381target hook that either recording is starting or ending. The first 7382time @var{type} is SWITCH_TYPE_DESCRIPTIVE and @var{text} is NULL, the 7383warning is for start up and the second time the warning is for 7384wind down. This feature is to allow the target hook to make any 7385necessary preparations before it starts to record switches and to 7386perform any necessary tidying up after it has finished recording 7387switches. 7388 7389@item SWITCH_TYPE_LINE_START 7390This option can be ignored by this target hook. 7391 7392@item SWITCH_TYPE_LINE_END 7393This option can be ignored by this target hook. 7394@end table 7395 7396The hook's return value must be zero. Other return values may be 7397supported in the future. 7398 7399By default this hook is set to NULL, but an example implementation is 7400provided for ELF based targets. Called @var{elf_record_gcc_switches}, 7401it records the switches as ASCII text inside a new, string mergeable 7402section in the assembler output file. The name of the new section is 7403provided by the @code{TARGET_ASM_RECORD_GCC_SWITCHES_SECTION} target 7404hook. 7405@end deftypefn 7406 7407@hook TARGET_ASM_RECORD_GCC_SWITCHES_SECTION 7408This is the name of the section that will be created by the example 7409ELF implementation of the @code{TARGET_ASM_RECORD_GCC_SWITCHES} target 7410hook. 7411@end deftypevr 7412 7413@need 2000 7414@node Data Output 7415@subsection Output of Data 7416 7417 7418@hook TARGET_ASM_BYTE_OP 7419@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_HI_OP 7420@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_SI_OP 7421@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_DI_OP 7422@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_TI_OP 7423@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_HI_OP 7424@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_SI_OP 7425@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_DI_OP 7426@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_TI_OP 7427These hooks specify assembly directives for creating certain kinds 7428of integer object. The @code{TARGET_ASM_BYTE_OP} directive creates a 7429byte-sized object, the @code{TARGET_ASM_ALIGNED_HI_OP} one creates an 7430aligned two-byte object, and so on. Any of the hooks may be 7431@code{NULL}, indicating that no suitable directive is available. 7432 7433The compiler will print these strings at the start of a new line, 7434followed immediately by the object's initial value. In most cases, 7435the string should contain a tab, a pseudo-op, and then another tab. 7436@end deftypevr 7437 7438@hook TARGET_ASM_INTEGER 7439The @code{assemble_integer} function uses this hook to output an 7440integer object. @var{x} is the object's value, @var{size} is its size 7441in bytes and @var{aligned_p} indicates whether it is aligned. The 7442function should return @code{true} if it was able to output the 7443object. If it returns false, @code{assemble_integer} will try to 7444split the object into smaller parts. 7445 7446The default implementation of this hook will use the 7447@code{TARGET_ASM_BYTE_OP} family of strings, returning @code{false} 7448when the relevant string is @code{NULL}. 7449@end deftypefn 7450 7451@hook TARGET_ASM_OUTPUT_ADDR_CONST_EXTRA 7452A target hook to recognize @var{rtx} patterns that @code{output_addr_const} 7453can't deal with, and output assembly code to @var{file} corresponding to 7454the pattern @var{x}. This may be used to allow machine-dependent 7455@code{UNSPEC}s to appear within constants. 7456 7457If target hook fails to recognize a pattern, it must return @code{false}, 7458so that a standard error message is printed. If it prints an error message 7459itself, by calling, for example, @code{output_operand_lossage}, it may just 7460return @code{true}. 7461@end deftypefn 7462 7463@defmac ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len}) 7464A C statement to output to the stdio stream @var{stream} an assembler 7465instruction to assemble a string constant containing the @var{len} 7466bytes at @var{ptr}. @var{ptr} will be a C expression of type 7467@code{char *} and @var{len} a C expression of type @code{int}. 7468 7469If the assembler has a @code{.ascii} pseudo-op as found in the 7470Berkeley Unix assembler, do not define the macro 7471@code{ASM_OUTPUT_ASCII}. 7472@end defmac 7473 7474@defmac ASM_OUTPUT_FDESC (@var{stream}, @var{decl}, @var{n}) 7475A C statement to output word @var{n} of a function descriptor for 7476@var{decl}. This must be defined if @code{TARGET_VTABLE_USES_DESCRIPTORS} 7477is defined, and is otherwise unused. 7478@end defmac 7479 7480@defmac CONSTANT_POOL_BEFORE_FUNCTION 7481You may define this macro as a C expression. You should define the 7482expression to have a nonzero value if GCC should output the constant 7483pool for a function before the code for the function, or a zero value if 7484GCC should output the constant pool after the function. If you do 7485not define this macro, the usual case, GCC will output the constant 7486pool before the function. 7487@end defmac 7488 7489@defmac ASM_OUTPUT_POOL_PROLOGUE (@var{file}, @var{funname}, @var{fundecl}, @var{size}) 7490A C statement to output assembler commands to define the start of the 7491constant pool for a function. @var{funname} is a string giving 7492the name of the function. Should the return type of the function 7493be required, it can be obtained via @var{fundecl}. @var{size} 7494is the size, in bytes, of the constant pool that will be written 7495immediately after this call. 7496 7497If no constant-pool prefix is required, the usual case, this macro need 7498not be defined. 7499@end defmac 7500 7501@defmac ASM_OUTPUT_SPECIAL_POOL_ENTRY (@var{file}, @var{x}, @var{mode}, @var{align}, @var{labelno}, @var{jumpto}) 7502A C statement (with or without semicolon) to output a constant in the 7503constant pool, if it needs special treatment. (This macro need not do 7504anything for RTL expressions that can be output normally.) 7505 7506The argument @var{file} is the standard I/O stream to output the 7507assembler code on. @var{x} is the RTL expression for the constant to 7508output, and @var{mode} is the machine mode (in case @var{x} is a 7509@samp{const_int}). @var{align} is the required alignment for the value 7510@var{x}; you should output an assembler directive to force this much 7511alignment. 7512 7513The argument @var{labelno} is a number to use in an internal label for 7514the address of this pool entry. The definition of this macro is 7515responsible for outputting the label definition at the proper place. 7516Here is how to do this: 7517 7518@smallexample 7519@code{(*targetm.asm_out.internal_label)} (@var{file}, "LC", @var{labelno}); 7520@end smallexample 7521 7522When you output a pool entry specially, you should end with a 7523@code{goto} to the label @var{jumpto}. This will prevent the same pool 7524entry from being output a second time in the usual manner. 7525 7526You need not define this macro if it would do nothing. 7527@end defmac 7528 7529@defmac ASM_OUTPUT_POOL_EPILOGUE (@var{file} @var{funname} @var{fundecl} @var{size}) 7530A C statement to output assembler commands to at the end of the constant 7531pool for a function. @var{funname} is a string giving the name of the 7532function. Should the return type of the function be required, you can 7533obtain it via @var{fundecl}. @var{size} is the size, in bytes, of the 7534constant pool that GCC wrote immediately before this call. 7535 7536If no constant-pool epilogue is required, the usual case, you need not 7537define this macro. 7538@end defmac 7539 7540@defmac IS_ASM_LOGICAL_LINE_SEPARATOR (@var{C}, @var{STR}) 7541Define this macro as a C expression which is nonzero if @var{C} is 7542used as a logical line separator by the assembler. @var{STR} points 7543to the position in the string where @var{C} was found; this can be used if 7544a line separator uses multiple characters. 7545 7546If you do not define this macro, the default is that only 7547the character @samp{;} is treated as a logical line separator. 7548@end defmac 7549 7550@hook TARGET_ASM_OPEN_PAREN 7551These target hooks are C string constants, describing the syntax in the 7552assembler for grouping arithmetic expressions. If not overridden, they 7553default to normal parentheses, which is correct for most assemblers. 7554@end deftypevr 7555 7556These macros are provided by @file{real.h} for writing the definitions 7557of @code{ASM_OUTPUT_DOUBLE} and the like: 7558 7559@defmac REAL_VALUE_TO_TARGET_SINGLE (@var{x}, @var{l}) 7560@defmacx REAL_VALUE_TO_TARGET_DOUBLE (@var{x}, @var{l}) 7561@defmacx REAL_VALUE_TO_TARGET_LONG_DOUBLE (@var{x}, @var{l}) 7562@defmacx REAL_VALUE_TO_TARGET_DECIMAL32 (@var{x}, @var{l}) 7563@defmacx REAL_VALUE_TO_TARGET_DECIMAL64 (@var{x}, @var{l}) 7564@defmacx REAL_VALUE_TO_TARGET_DECIMAL128 (@var{x}, @var{l}) 7565These translate @var{x}, of type @code{REAL_VALUE_TYPE}, to the 7566target's floating point representation, and store its bit pattern in 7567the variable @var{l}. For @code{REAL_VALUE_TO_TARGET_SINGLE} and 7568@code{REAL_VALUE_TO_TARGET_DECIMAL32}, this variable should be a 7569simple @code{long int}. For the others, it should be an array of 7570@code{long int}. The number of elements in this array is determined 7571by the size of the desired target floating point data type: 32 bits of 7572it go in each @code{long int} array element. Each array element holds 757332 bits of the result, even if @code{long int} is wider than 32 bits 7574on the host machine. 7575 7576The array element values are designed so that you can print them out 7577using @code{fprintf} in the order they should appear in the target 7578machine's memory. 7579@end defmac 7580 7581@node Uninitialized Data 7582@subsection Output of Uninitialized Variables 7583 7584Each of the macros in this section is used to do the whole job of 7585outputting a single uninitialized variable. 7586 7587@defmac ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded}) 7588A C statement (sans semicolon) to output to the stdio stream 7589@var{stream} the assembler definition of a common-label named 7590@var{name} whose size is @var{size} bytes. The variable @var{rounded} 7591is the size rounded up to whatever alignment the caller wants. It is 7592possible that @var{size} may be zero, for instance if a struct with no 7593other member than a zero-length array is defined. In this case, the 7594backend must output a symbol definition that allocates at least one 7595byte, both so that the address of the resulting object does not compare 7596equal to any other, and because some object formats cannot even express 7597the concept of a zero-sized common symbol, as that is how they represent 7598an ordinary undefined external. 7599 7600Use the expression @code{assemble_name (@var{stream}, @var{name})} to 7601output the name itself; before and after that, output the additional 7602assembler syntax for defining the name, and a newline. 7603 7604This macro controls how the assembler definitions of uninitialized 7605common global variables are output. 7606@end defmac 7607 7608@defmac ASM_OUTPUT_ALIGNED_COMMON (@var{stream}, @var{name}, @var{size}, @var{alignment}) 7609Like @code{ASM_OUTPUT_COMMON} except takes the required alignment as a 7610separate, explicit argument. If you define this macro, it is used in 7611place of @code{ASM_OUTPUT_COMMON}, and gives you more flexibility in 7612handling the required alignment of the variable. The alignment is specified 7613as the number of bits. 7614@end defmac 7615 7616@defmac ASM_OUTPUT_ALIGNED_DECL_COMMON (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment}) 7617Like @code{ASM_OUTPUT_ALIGNED_COMMON} except that @var{decl} of the 7618variable to be output, if there is one, or @code{NULL_TREE} if there 7619is no corresponding variable. If you define this macro, GCC will use it 7620in place of both @code{ASM_OUTPUT_COMMON} and 7621@code{ASM_OUTPUT_ALIGNED_COMMON}. Define this macro when you need to see 7622the variable's decl in order to chose what to output. 7623@end defmac 7624 7625@defmac ASM_OUTPUT_ALIGNED_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment}) 7626A C statement (sans semicolon) to output to the stdio stream 7627@var{stream} the assembler definition of uninitialized global @var{decl} named 7628@var{name} whose size is @var{size} bytes. The variable @var{alignment} 7629is the alignment specified as the number of bits. 7630 7631Try to use function @code{asm_output_aligned_bss} defined in file 7632@file{varasm.c} when defining this macro. If unable, use the expression 7633@code{assemble_name (@var{stream}, @var{name})} to output the name itself; 7634before and after that, output the additional assembler syntax for defining 7635the name, and a newline. 7636 7637There are two ways of handling global BSS@. One is to define this macro. 7638The other is to have @code{TARGET_ASM_SELECT_SECTION} return a 7639switchable BSS section (@pxref{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS}). 7640You do not need to do both. 7641 7642Some languages do not have @code{common} data, and require a 7643non-common form of global BSS in order to handle uninitialized globals 7644efficiently. C++ is one example of this. However, if the target does 7645not support global BSS, the front end may choose to make globals 7646common in order to save space in the object file. 7647@end defmac 7648 7649@defmac ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded}) 7650A C statement (sans semicolon) to output to the stdio stream 7651@var{stream} the assembler definition of a local-common-label named 7652@var{name} whose size is @var{size} bytes. The variable @var{rounded} 7653is the size rounded up to whatever alignment the caller wants. 7654 7655Use the expression @code{assemble_name (@var{stream}, @var{name})} to 7656output the name itself; before and after that, output the additional 7657assembler syntax for defining the name, and a newline. 7658 7659This macro controls how the assembler definitions of uninitialized 7660static variables are output. 7661@end defmac 7662 7663@defmac ASM_OUTPUT_ALIGNED_LOCAL (@var{stream}, @var{name}, @var{size}, @var{alignment}) 7664Like @code{ASM_OUTPUT_LOCAL} except takes the required alignment as a 7665separate, explicit argument. If you define this macro, it is used in 7666place of @code{ASM_OUTPUT_LOCAL}, and gives you more flexibility in 7667handling the required alignment of the variable. The alignment is specified 7668as the number of bits. 7669@end defmac 7670 7671@defmac ASM_OUTPUT_ALIGNED_DECL_LOCAL (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment}) 7672Like @code{ASM_OUTPUT_ALIGNED_DECL} except that @var{decl} of the 7673variable to be output, if there is one, or @code{NULL_TREE} if there 7674is no corresponding variable. If you define this macro, GCC will use it 7675in place of both @code{ASM_OUTPUT_DECL} and 7676@code{ASM_OUTPUT_ALIGNED_DECL}. Define this macro when you need to see 7677the variable's decl in order to chose what to output. 7678@end defmac 7679 7680@node Label Output 7681@subsection Output and Generation of Labels 7682 7683@c prevent bad page break with this line 7684This is about outputting labels. 7685 7686@findex assemble_name 7687@defmac ASM_OUTPUT_LABEL (@var{stream}, @var{name}) 7688A C statement (sans semicolon) to output to the stdio stream 7689@var{stream} the assembler definition of a label named @var{name}. 7690Use the expression @code{assemble_name (@var{stream}, @var{name})} to 7691output the name itself; before and after that, output the additional 7692assembler syntax for defining the name, and a newline. A default 7693definition of this macro is provided which is correct for most systems. 7694@end defmac 7695 7696@defmac ASM_OUTPUT_FUNCTION_LABEL (@var{stream}, @var{name}, @var{decl}) 7697A C statement (sans semicolon) to output to the stdio stream 7698@var{stream} the assembler definition of a label named @var{name} of 7699a function. 7700Use the expression @code{assemble_name (@var{stream}, @var{name})} to 7701output the name itself; before and after that, output the additional 7702assembler syntax for defining the name, and a newline. A default 7703definition of this macro is provided which is correct for most systems. 7704 7705If this macro is not defined, then the function name is defined in the 7706usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}). 7707@end defmac 7708 7709@findex assemble_name_raw 7710@defmac ASM_OUTPUT_INTERNAL_LABEL (@var{stream}, @var{name}) 7711Identical to @code{ASM_OUTPUT_LABEL}, except that @var{name} is known 7712to refer to a compiler-generated label. The default definition uses 7713@code{assemble_name_raw}, which is like @code{assemble_name} except 7714that it is more efficient. 7715@end defmac 7716 7717@defmac SIZE_ASM_OP 7718A C string containing the appropriate assembler directive to specify the 7719size of a symbol, without any arguments. On systems that use ELF, the 7720default (in @file{config/elfos.h}) is @samp{"\t.size\t"}; on other 7721systems, the default is not to define this macro. 7722 7723Define this macro only if it is correct to use the default definitions 7724of @code{ASM_OUTPUT_SIZE_DIRECTIVE} and @code{ASM_OUTPUT_MEASURED_SIZE} 7725for your system. If you need your own custom definitions of those 7726macros, or if you do not need explicit symbol sizes at all, do not 7727define this macro. 7728@end defmac 7729 7730@defmac ASM_OUTPUT_SIZE_DIRECTIVE (@var{stream}, @var{name}, @var{size}) 7731A C statement (sans semicolon) to output to the stdio stream 7732@var{stream} a directive telling the assembler that the size of the 7733symbol @var{name} is @var{size}. @var{size} is a @code{HOST_WIDE_INT}. 7734If you define @code{SIZE_ASM_OP}, a default definition of this macro is 7735provided. 7736@end defmac 7737 7738@defmac ASM_OUTPUT_MEASURED_SIZE (@var{stream}, @var{name}) 7739A C statement (sans semicolon) to output to the stdio stream 7740@var{stream} a directive telling the assembler to calculate the size of 7741the symbol @var{name} by subtracting its address from the current 7742address. 7743 7744If you define @code{SIZE_ASM_OP}, a default definition of this macro is 7745provided. The default assumes that the assembler recognizes a special 7746@samp{.} symbol as referring to the current address, and can calculate 7747the difference between this and another symbol. If your assembler does 7748not recognize @samp{.} or cannot do calculations with it, you will need 7749to redefine @code{ASM_OUTPUT_MEASURED_SIZE} to use some other technique. 7750@end defmac 7751 7752@defmac TYPE_ASM_OP 7753A C string containing the appropriate assembler directive to specify the 7754type of a symbol, without any arguments. On systems that use ELF, the 7755default (in @file{config/elfos.h}) is @samp{"\t.type\t"}; on other 7756systems, the default is not to define this macro. 7757 7758Define this macro only if it is correct to use the default definition of 7759@code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own 7760custom definition of this macro, or if you do not need explicit symbol 7761types at all, do not define this macro. 7762@end defmac 7763 7764@defmac TYPE_OPERAND_FMT 7765A C string which specifies (using @code{printf} syntax) the format of 7766the second operand to @code{TYPE_ASM_OP}. On systems that use ELF, the 7767default (in @file{config/elfos.h}) is @samp{"@@%s"}; on other systems, 7768the default is not to define this macro. 7769 7770Define this macro only if it is correct to use the default definition of 7771@code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own 7772custom definition of this macro, or if you do not need explicit symbol 7773types at all, do not define this macro. 7774@end defmac 7775 7776@defmac ASM_OUTPUT_TYPE_DIRECTIVE (@var{stream}, @var{type}) 7777A C statement (sans semicolon) to output to the stdio stream 7778@var{stream} a directive telling the assembler that the type of the 7779symbol @var{name} is @var{type}. @var{type} is a C string; currently, 7780that string is always either @samp{"function"} or @samp{"object"}, but 7781you should not count on this. 7782 7783If you define @code{TYPE_ASM_OP} and @code{TYPE_OPERAND_FMT}, a default 7784definition of this macro is provided. 7785@end defmac 7786 7787@defmac ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl}) 7788A C statement (sans semicolon) to output to the stdio stream 7789@var{stream} any text necessary for declaring the name @var{name} of a 7790function which is being defined. This macro is responsible for 7791outputting the label definition (perhaps using 7792@code{ASM_OUTPUT_FUNCTION_LABEL}). The argument @var{decl} is the 7793@code{FUNCTION_DECL} tree node representing the function. 7794 7795If this macro is not defined, then the function name is defined in the 7796usual manner as a label (by means of @code{ASM_OUTPUT_FUNCTION_LABEL}). 7797 7798You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition 7799of this macro. 7800@end defmac 7801 7802@defmac ASM_DECLARE_FUNCTION_SIZE (@var{stream}, @var{name}, @var{decl}) 7803A C statement (sans semicolon) to output to the stdio stream 7804@var{stream} any text necessary for declaring the size of a function 7805which is being defined. The argument @var{name} is the name of the 7806function. The argument @var{decl} is the @code{FUNCTION_DECL} tree node 7807representing the function. 7808 7809If this macro is not defined, then the function size is not defined. 7810 7811You may wish to use @code{ASM_OUTPUT_MEASURED_SIZE} in the definition 7812of this macro. 7813@end defmac 7814 7815@defmac ASM_DECLARE_OBJECT_NAME (@var{stream}, @var{name}, @var{decl}) 7816A C statement (sans semicolon) to output to the stdio stream 7817@var{stream} any text necessary for declaring the name @var{name} of an 7818initialized variable which is being defined. This macro must output the 7819label definition (perhaps using @code{ASM_OUTPUT_LABEL}). The argument 7820@var{decl} is the @code{VAR_DECL} tree node representing the variable. 7821 7822If this macro is not defined, then the variable name is defined in the 7823usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}). 7824 7825You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} and/or 7826@code{ASM_OUTPUT_SIZE_DIRECTIVE} in the definition of this macro. 7827@end defmac 7828 7829@hook TARGET_ASM_DECLARE_CONSTANT_NAME 7830A target hook to output to the stdio stream @var{file} any text necessary 7831for declaring the name @var{name} of a constant which is being defined. This 7832target hook is responsible for outputting the label definition (perhaps using 7833@code{assemble_label}). The argument @var{exp} is the value of the constant, 7834and @var{size} is the size of the constant in bytes. The @var{name} 7835will be an internal label. 7836 7837The default version of this target hook, define the @var{name} in the 7838usual manner as a label (by means of @code{assemble_label}). 7839 7840You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in this target hook. 7841@end deftypefn 7842 7843@defmac ASM_DECLARE_REGISTER_GLOBAL (@var{stream}, @var{decl}, @var{regno}, @var{name}) 7844A C statement (sans semicolon) to output to the stdio stream 7845@var{stream} any text necessary for claiming a register @var{regno} 7846for a global variable @var{decl} with name @var{name}. 7847 7848If you don't define this macro, that is equivalent to defining it to do 7849nothing. 7850@end defmac 7851 7852@defmac ASM_FINISH_DECLARE_OBJECT (@var{stream}, @var{decl}, @var{toplevel}, @var{atend}) 7853A C statement (sans semicolon) to finish up declaring a variable name 7854once the compiler has processed its initializer fully and thus has had a 7855chance to determine the size of an array when controlled by an 7856initializer. This is used on systems where it's necessary to declare 7857something about the size of the object. 7858 7859If you don't define this macro, that is equivalent to defining it to do 7860nothing. 7861 7862You may wish to use @code{ASM_OUTPUT_SIZE_DIRECTIVE} and/or 7863@code{ASM_OUTPUT_MEASURED_SIZE} in the definition of this macro. 7864@end defmac 7865 7866@hook TARGET_ASM_GLOBALIZE_LABEL 7867This target hook is a function to output to the stdio stream 7868@var{stream} some commands that will make the label @var{name} global; 7869that is, available for reference from other files. 7870 7871The default implementation relies on a proper definition of 7872@code{GLOBAL_ASM_OP}. 7873@end deftypefn 7874 7875@hook TARGET_ASM_GLOBALIZE_DECL_NAME 7876This target hook is a function to output to the stdio stream 7877@var{stream} some commands that will make the name associated with @var{decl} 7878global; that is, available for reference from other files. 7879 7880The default implementation uses the TARGET_ASM_GLOBALIZE_LABEL target hook. 7881@end deftypefn 7882 7883@defmac ASM_WEAKEN_LABEL (@var{stream}, @var{name}) 7884A C statement (sans semicolon) to output to the stdio stream 7885@var{stream} some commands that will make the label @var{name} weak; 7886that is, available for reference from other files but only used if 7887no other definition is available. Use the expression 7888@code{assemble_name (@var{stream}, @var{name})} to output the name 7889itself; before and after that, output the additional assembler syntax 7890for making that name weak, and a newline. 7891 7892If you don't define this macro or @code{ASM_WEAKEN_DECL}, GCC will not 7893support weak symbols and you should not define the @code{SUPPORTS_WEAK} 7894macro. 7895@end defmac 7896 7897@defmac ASM_WEAKEN_DECL (@var{stream}, @var{decl}, @var{name}, @var{value}) 7898Combines (and replaces) the function of @code{ASM_WEAKEN_LABEL} and 7899@code{ASM_OUTPUT_WEAK_ALIAS}, allowing access to the associated function 7900or variable decl. If @var{value} is not @code{NULL}, this C statement 7901should output to the stdio stream @var{stream} assembler code which 7902defines (equates) the weak symbol @var{name} to have the value 7903@var{value}. If @var{value} is @code{NULL}, it should output commands 7904to make @var{name} weak. 7905@end defmac 7906 7907@defmac ASM_OUTPUT_WEAKREF (@var{stream}, @var{decl}, @var{name}, @var{value}) 7908Outputs a directive that enables @var{name} to be used to refer to 7909symbol @var{value} with weak-symbol semantics. @code{decl} is the 7910declaration of @code{name}. 7911@end defmac 7912 7913@defmac SUPPORTS_WEAK 7914A preprocessor constant expression which evaluates to true if the target 7915supports weak symbols. 7916 7917If you don't define this macro, @file{defaults.h} provides a default 7918definition. If either @code{ASM_WEAKEN_LABEL} or @code{ASM_WEAKEN_DECL} 7919is defined, the default definition is @samp{1}; otherwise, it is @samp{0}. 7920@end defmac 7921 7922@defmac TARGET_SUPPORTS_WEAK 7923A C expression which evaluates to true if the target supports weak symbols. 7924 7925If you don't define this macro, @file{defaults.h} provides a default 7926definition. The default definition is @samp{(SUPPORTS_WEAK)}. Define 7927this macro if you want to control weak symbol support with a compiler 7928flag such as @option{-melf}. 7929@end defmac 7930 7931@defmac MAKE_DECL_ONE_ONLY (@var{decl}) 7932A C statement (sans semicolon) to mark @var{decl} to be emitted as a 7933public symbol such that extra copies in multiple translation units will 7934be discarded by the linker. Define this macro if your object file 7935format provides support for this concept, such as the @samp{COMDAT} 7936section flags in the Microsoft Windows PE/COFF format, and this support 7937requires changes to @var{decl}, such as putting it in a separate section. 7938@end defmac 7939 7940@defmac SUPPORTS_ONE_ONLY 7941A C expression which evaluates to true if the target supports one-only 7942semantics. 7943 7944If you don't define this macro, @file{varasm.c} provides a default 7945definition. If @code{MAKE_DECL_ONE_ONLY} is defined, the default 7946definition is @samp{1}; otherwise, it is @samp{0}. Define this macro if 7947you want to control one-only symbol support with a compiler flag, or if 7948setting the @code{DECL_ONE_ONLY} flag is enough to mark a declaration to 7949be emitted as one-only. 7950@end defmac 7951 7952@hook TARGET_ASM_ASSEMBLE_VISIBILITY 7953This target hook is a function to output to @var{asm_out_file} some 7954commands that will make the symbol(s) associated with @var{decl} have 7955hidden, protected or internal visibility as specified by @var{visibility}. 7956@end deftypefn 7957 7958@defmac TARGET_WEAK_NOT_IN_ARCHIVE_TOC 7959A C expression that evaluates to true if the target's linker expects 7960that weak symbols do not appear in a static archive's table of contents. 7961The default is @code{0}. 7962 7963Leaving weak symbols out of an archive's table of contents means that, 7964if a symbol will only have a definition in one translation unit and 7965will have undefined references from other translation units, that 7966symbol should not be weak. Defining this macro to be nonzero will 7967thus have the effect that certain symbols that would normally be weak 7968(explicit template instantiations, and vtables for polymorphic classes 7969with noninline key methods) will instead be nonweak. 7970 7971The C++ ABI requires this macro to be zero. Define this macro for 7972targets where full C++ ABI compliance is impossible and where linker 7973restrictions require weak symbols to be left out of a static archive's 7974table of contents. 7975@end defmac 7976 7977@defmac ASM_OUTPUT_EXTERNAL (@var{stream}, @var{decl}, @var{name}) 7978A C statement (sans semicolon) to output to the stdio stream 7979@var{stream} any text necessary for declaring the name of an external 7980symbol named @var{name} which is referenced in this compilation but 7981not defined. The value of @var{decl} is the tree node for the 7982declaration. 7983 7984This macro need not be defined if it does not need to output anything. 7985The GNU assembler and most Unix assemblers don't require anything. 7986@end defmac 7987 7988@hook TARGET_ASM_EXTERNAL_LIBCALL 7989This target hook is a function to output to @var{asm_out_file} an assembler 7990pseudo-op to declare a library function name external. The name of the 7991library function is given by @var{symref}, which is a @code{symbol_ref}. 7992@end deftypefn 7993 7994@hook TARGET_ASM_MARK_DECL_PRESERVED 7995This target hook is a function to output to @var{asm_out_file} an assembler 7996directive to annotate @var{symbol} as used. The Darwin target uses the 7997.no_dead_code_strip directive. 7998@end deftypefn 7999 8000@defmac ASM_OUTPUT_LABELREF (@var{stream}, @var{name}) 8001A C statement (sans semicolon) to output to the stdio stream 8002@var{stream} a reference in assembler syntax to a label named 8003@var{name}. This should add @samp{_} to the front of the name, if that 8004is customary on your operating system, as it is in most Berkeley Unix 8005systems. This macro is used in @code{assemble_name}. 8006@end defmac 8007 8008@hook TARGET_MANGLE_ASSEMBLER_NAME 8009 8010@defmac ASM_OUTPUT_SYMBOL_REF (@var{stream}, @var{sym}) 8011A C statement (sans semicolon) to output a reference to 8012@code{SYMBOL_REF} @var{sym}. If not defined, @code{assemble_name} 8013will be used to output the name of the symbol. This macro may be used 8014to modify the way a symbol is referenced depending on information 8015encoded by @code{TARGET_ENCODE_SECTION_INFO}. 8016@end defmac 8017 8018@defmac ASM_OUTPUT_LABEL_REF (@var{stream}, @var{buf}) 8019A C statement (sans semicolon) to output a reference to @var{buf}, the 8020result of @code{ASM_GENERATE_INTERNAL_LABEL}. If not defined, 8021@code{assemble_name} will be used to output the name of the symbol. 8022This macro is not used by @code{output_asm_label}, or the @code{%l} 8023specifier that calls it; the intention is that this macro should be set 8024when it is necessary to output a label differently when its address is 8025being taken. 8026@end defmac 8027 8028@hook TARGET_ASM_INTERNAL_LABEL 8029A function to output to the stdio stream @var{stream} a label whose 8030name is made from the string @var{prefix} and the number @var{labelno}. 8031 8032It is absolutely essential that these labels be distinct from the labels 8033used for user-level functions and variables. Otherwise, certain programs 8034will have name conflicts with internal labels. 8035 8036It is desirable to exclude internal labels from the symbol table of the 8037object file. Most assemblers have a naming convention for labels that 8038should be excluded; on many systems, the letter @samp{L} at the 8039beginning of a label has this effect. You should find out what 8040convention your system uses, and follow it. 8041 8042The default version of this function utilizes @code{ASM_GENERATE_INTERNAL_LABEL}. 8043@end deftypefn 8044 8045@defmac ASM_OUTPUT_DEBUG_LABEL (@var{stream}, @var{prefix}, @var{num}) 8046A C statement to output to the stdio stream @var{stream} a debug info 8047label whose name is made from the string @var{prefix} and the number 8048@var{num}. This is useful for VLIW targets, where debug info labels 8049may need to be treated differently than branch target labels. On some 8050systems, branch target labels must be at the beginning of instruction 8051bundles, but debug info labels can occur in the middle of instruction 8052bundles. 8053 8054If this macro is not defined, then @code{(*targetm.asm_out.internal_label)} will be 8055used. 8056@end defmac 8057 8058@defmac ASM_GENERATE_INTERNAL_LABEL (@var{string}, @var{prefix}, @var{num}) 8059A C statement to store into the string @var{string} a label whose name 8060is made from the string @var{prefix} and the number @var{num}. 8061 8062This string, when output subsequently by @code{assemble_name}, should 8063produce the output that @code{(*targetm.asm_out.internal_label)} would produce 8064with the same @var{prefix} and @var{num}. 8065 8066If the string begins with @samp{*}, then @code{assemble_name} will 8067output the rest of the string unchanged. It is often convenient for 8068@code{ASM_GENERATE_INTERNAL_LABEL} to use @samp{*} in this way. If the 8069string doesn't start with @samp{*}, then @code{ASM_OUTPUT_LABELREF} gets 8070to output the string, and may change it. (Of course, 8071@code{ASM_OUTPUT_LABELREF} is also part of your machine description, so 8072you should know what it does on your machine.) 8073@end defmac 8074 8075@defmac ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number}) 8076A C expression to assign to @var{outvar} (which is a variable of type 8077@code{char *}) a newly allocated string made from the string 8078@var{name} and the number @var{number}, with some suitable punctuation 8079added. Use @code{alloca} to get space for the string. 8080 8081The string will be used as an argument to @code{ASM_OUTPUT_LABELREF} to 8082produce an assembler label for an internal static variable whose name is 8083@var{name}. Therefore, the string must be such as to result in valid 8084assembler code. The argument @var{number} is different each time this 8085macro is executed; it prevents conflicts between similarly-named 8086internal static variables in different scopes. 8087 8088Ideally this string should not be a valid C identifier, to prevent any 8089conflict with the user's own symbols. Most assemblers allow periods 8090or percent signs in assembler symbols; putting at least one of these 8091between the name and the number will suffice. 8092 8093If this macro is not defined, a default definition will be provided 8094which is correct for most systems. 8095@end defmac 8096 8097@defmac ASM_OUTPUT_DEF (@var{stream}, @var{name}, @var{value}) 8098A C statement to output to the stdio stream @var{stream} assembler code 8099which defines (equates) the symbol @var{name} to have the value @var{value}. 8100 8101@findex SET_ASM_OP 8102If @code{SET_ASM_OP} is defined, a default definition is provided which is 8103correct for most systems. 8104@end defmac 8105 8106@defmac ASM_OUTPUT_DEF_FROM_DECLS (@var{stream}, @var{decl_of_name}, @var{decl_of_value}) 8107A C statement to output to the stdio stream @var{stream} assembler code 8108which defines (equates) the symbol whose tree node is @var{decl_of_name} 8109to have the value of the tree node @var{decl_of_value}. This macro will 8110be used in preference to @samp{ASM_OUTPUT_DEF} if it is defined and if 8111the tree nodes are available. 8112 8113@findex SET_ASM_OP 8114If @code{SET_ASM_OP} is defined, a default definition is provided which is 8115correct for most systems. 8116@end defmac 8117 8118@defmac TARGET_DEFERRED_OUTPUT_DEFS (@var{decl_of_name}, @var{decl_of_value}) 8119A C statement that evaluates to true if the assembler code which defines 8120(equates) the symbol whose tree node is @var{decl_of_name} to have the value 8121of the tree node @var{decl_of_value} should be emitted near the end of the 8122current compilation unit. The default is to not defer output of defines. 8123This macro affects defines output by @samp{ASM_OUTPUT_DEF} and 8124@samp{ASM_OUTPUT_DEF_FROM_DECLS}. 8125@end defmac 8126 8127@defmac ASM_OUTPUT_WEAK_ALIAS (@var{stream}, @var{name}, @var{value}) 8128A C statement to output to the stdio stream @var{stream} assembler code 8129which defines (equates) the weak symbol @var{name} to have the value 8130@var{value}. If @var{value} is @code{NULL}, it defines @var{name} as 8131an undefined weak symbol. 8132 8133Define this macro if the target only supports weak aliases; define 8134@code{ASM_OUTPUT_DEF} instead if possible. 8135@end defmac 8136 8137@defmac OBJC_GEN_METHOD_LABEL (@var{buf}, @var{is_inst}, @var{class_name}, @var{cat_name}, @var{sel_name}) 8138Define this macro to override the default assembler names used for 8139Objective-C methods. 8140 8141The default name is a unique method number followed by the name of the 8142class (e.g.@: @samp{_1_Foo}). For methods in categories, the name of 8143the category is also included in the assembler name (e.g.@: 8144@samp{_1_Foo_Bar}). 8145 8146These names are safe on most systems, but make debugging difficult since 8147the method's selector is not present in the name. Therefore, particular 8148systems define other ways of computing names. 8149 8150@var{buf} is an expression of type @code{char *} which gives you a 8151buffer in which to store the name; its length is as long as 8152@var{class_name}, @var{cat_name} and @var{sel_name} put together, plus 815350 characters extra. 8154 8155The argument @var{is_inst} specifies whether the method is an instance 8156method or a class method; @var{class_name} is the name of the class; 8157@var{cat_name} is the name of the category (or @code{NULL} if the method is not 8158in a category); and @var{sel_name} is the name of the selector. 8159 8160On systems where the assembler can handle quoted names, you can use this 8161macro to provide more human-readable names. 8162@end defmac 8163 8164@node Initialization 8165@subsection How Initialization Functions Are Handled 8166@cindex initialization routines 8167@cindex termination routines 8168@cindex constructors, output of 8169@cindex destructors, output of 8170 8171The compiled code for certain languages includes @dfn{constructors} 8172(also called @dfn{initialization routines})---functions to initialize 8173data in the program when the program is started. These functions need 8174to be called before the program is ``started''---that is to say, before 8175@code{main} is called. 8176 8177Compiling some languages generates @dfn{destructors} (also called 8178@dfn{termination routines}) that should be called when the program 8179terminates. 8180 8181To make the initialization and termination functions work, the compiler 8182must output something in the assembler code to cause those functions to 8183be called at the appropriate time. When you port the compiler to a new 8184system, you need to specify how to do this. 8185 8186There are two major ways that GCC currently supports the execution of 8187initialization and termination functions. Each way has two variants. 8188Much of the structure is common to all four variations. 8189 8190@findex __CTOR_LIST__ 8191@findex __DTOR_LIST__ 8192The linker must build two lists of these functions---a list of 8193initialization functions, called @code{__CTOR_LIST__}, and a list of 8194termination functions, called @code{__DTOR_LIST__}. 8195 8196Each list always begins with an ignored function pointer (which may hold 81970, @minus{}1, or a count of the function pointers after it, depending on 8198the environment). This is followed by a series of zero or more function 8199pointers to constructors (or destructors), followed by a function 8200pointer containing zero. 8201 8202Depending on the operating system and its executable file format, either 8203@file{crtstuff.c} or @file{libgcc2.c} traverses these lists at startup 8204time and exit time. Constructors are called in reverse order of the 8205list; destructors in forward order. 8206 8207The best way to handle static constructors works only for object file 8208formats which provide arbitrarily-named sections. A section is set 8209aside for a list of constructors, and another for a list of destructors. 8210Traditionally these are called @samp{.ctors} and @samp{.dtors}. Each 8211object file that defines an initialization function also puts a word in 8212the constructor section to point to that function. The linker 8213accumulates all these words into one contiguous @samp{.ctors} section. 8214Termination functions are handled similarly. 8215 8216This method will be chosen as the default by @file{target-def.h} if 8217@code{TARGET_ASM_NAMED_SECTION} is defined. A target that does not 8218support arbitrary sections, but does support special designated 8219constructor and destructor sections may define @code{CTORS_SECTION_ASM_OP} 8220and @code{DTORS_SECTION_ASM_OP} to achieve the same effect. 8221 8222When arbitrary sections are available, there are two variants, depending 8223upon how the code in @file{crtstuff.c} is called. On systems that 8224support a @dfn{.init} section which is executed at program startup, 8225parts of @file{crtstuff.c} are compiled into that section. The 8226program is linked by the @command{gcc} driver like this: 8227 8228@smallexample 8229ld -o @var{output_file} crti.o crtbegin.o @dots{} -lgcc crtend.o crtn.o 8230@end smallexample 8231 8232The prologue of a function (@code{__init}) appears in the @code{.init} 8233section of @file{crti.o}; the epilogue appears in @file{crtn.o}. Likewise 8234for the function @code{__fini} in the @dfn{.fini} section. Normally these 8235files are provided by the operating system or by the GNU C library, but 8236are provided by GCC for a few targets. 8237 8238The objects @file{crtbegin.o} and @file{crtend.o} are (for most targets) 8239compiled from @file{crtstuff.c}. They contain, among other things, code 8240fragments within the @code{.init} and @code{.fini} sections that branch 8241to routines in the @code{.text} section. The linker will pull all parts 8242of a section together, which results in a complete @code{__init} function 8243that invokes the routines we need at startup. 8244 8245To use this variant, you must define the @code{INIT_SECTION_ASM_OP} 8246macro properly. 8247 8248If no init section is available, when GCC compiles any function called 8249@code{main} (or more accurately, any function designated as a program 8250entry point by the language front end calling @code{expand_main_function}), 8251it inserts a procedure call to @code{__main} as the first executable code 8252after the function prologue. The @code{__main} function is defined 8253in @file{libgcc2.c} and runs the global constructors. 8254 8255In file formats that don't support arbitrary sections, there are again 8256two variants. In the simplest variant, the GNU linker (GNU @code{ld}) 8257and an `a.out' format must be used. In this case, 8258@code{TARGET_ASM_CONSTRUCTOR} is defined to produce a @code{.stabs} 8259entry of type @samp{N_SETT}, referencing the name @code{__CTOR_LIST__}, 8260and with the address of the void function containing the initialization 8261code as its value. The GNU linker recognizes this as a request to add 8262the value to a @dfn{set}; the values are accumulated, and are eventually 8263placed in the executable as a vector in the format described above, with 8264a leading (ignored) count and a trailing zero element. 8265@code{TARGET_ASM_DESTRUCTOR} is handled similarly. Since no init 8266section is available, the absence of @code{INIT_SECTION_ASM_OP} causes 8267the compilation of @code{main} to call @code{__main} as above, starting 8268the initialization process. 8269 8270The last variant uses neither arbitrary sections nor the GNU linker. 8271This is preferable when you want to do dynamic linking and when using 8272file formats which the GNU linker does not support, such as `ECOFF'@. In 8273this case, @code{TARGET_HAVE_CTORS_DTORS} is false, initialization and 8274termination functions are recognized simply by their names. This requires 8275an extra program in the linkage step, called @command{collect2}. This program 8276pretends to be the linker, for use with GCC; it does its job by running 8277the ordinary linker, but also arranges to include the vectors of 8278initialization and termination functions. These functions are called 8279via @code{__main} as described above. In order to use this method, 8280@code{use_collect2} must be defined in the target in @file{config.gcc}. 8281 8282@ifinfo 8283The following section describes the specific macros that control and 8284customize the handling of initialization and termination functions. 8285@end ifinfo 8286 8287@node Macros for Initialization 8288@subsection Macros Controlling Initialization Routines 8289 8290Here are the macros that control how the compiler handles initialization 8291and termination functions: 8292 8293@defmac INIT_SECTION_ASM_OP 8294If defined, a C string constant, including spacing, for the assembler 8295operation to identify the following data as initialization code. If not 8296defined, GCC will assume such a section does not exist. When you are 8297using special sections for initialization and termination functions, this 8298macro also controls how @file{crtstuff.c} and @file{libgcc2.c} arrange to 8299run the initialization functions. 8300@end defmac 8301 8302@defmac HAS_INIT_SECTION 8303If defined, @code{main} will not call @code{__main} as described above. 8304This macro should be defined for systems that control start-up code 8305on a symbol-by-symbol basis, such as OSF/1, and should not 8306be defined explicitly for systems that support @code{INIT_SECTION_ASM_OP}. 8307@end defmac 8308 8309@defmac LD_INIT_SWITCH 8310If defined, a C string constant for a switch that tells the linker that 8311the following symbol is an initialization routine. 8312@end defmac 8313 8314@defmac LD_FINI_SWITCH 8315If defined, a C string constant for a switch that tells the linker that 8316the following symbol is a finalization routine. 8317@end defmac 8318 8319@defmac COLLECT_SHARED_INIT_FUNC (@var{stream}, @var{func}) 8320If defined, a C statement that will write a function that can be 8321automatically called when a shared library is loaded. The function 8322should call @var{func}, which takes no arguments. If not defined, and 8323the object format requires an explicit initialization function, then a 8324function called @code{_GLOBAL__DI} will be generated. 8325 8326This function and the following one are used by collect2 when linking a 8327shared library that needs constructors or destructors, or has DWARF2 8328exception tables embedded in the code. 8329@end defmac 8330 8331@defmac COLLECT_SHARED_FINI_FUNC (@var{stream}, @var{func}) 8332If defined, a C statement that will write a function that can be 8333automatically called when a shared library is unloaded. The function 8334should call @var{func}, which takes no arguments. If not defined, and 8335the object format requires an explicit finalization function, then a 8336function called @code{_GLOBAL__DD} will be generated. 8337@end defmac 8338 8339@defmac INVOKE__main 8340If defined, @code{main} will call @code{__main} despite the presence of 8341@code{INIT_SECTION_ASM_OP}. This macro should be defined for systems 8342where the init section is not actually run automatically, but is still 8343useful for collecting the lists of constructors and destructors. 8344@end defmac 8345 8346@defmac SUPPORTS_INIT_PRIORITY 8347If nonzero, the C++ @code{init_priority} attribute is supported and the 8348compiler should emit instructions to control the order of initialization 8349of objects. If zero, the compiler will issue an error message upon 8350encountering an @code{init_priority} attribute. 8351@end defmac 8352 8353@hook TARGET_HAVE_CTORS_DTORS 8354This value is true if the target supports some ``native'' method of 8355collecting constructors and destructors to be run at startup and exit. 8356It is false if we must use @command{collect2}. 8357@end deftypevr 8358 8359@hook TARGET_ASM_CONSTRUCTOR 8360If defined, a function that outputs assembler code to arrange to call 8361the function referenced by @var{symbol} at initialization time. 8362 8363Assume that @var{symbol} is a @code{SYMBOL_REF} for a function taking 8364no arguments and with no return value. If the target supports initialization 8365priorities, @var{priority} is a value between 0 and @code{MAX_INIT_PRIORITY}; 8366otherwise it must be @code{DEFAULT_INIT_PRIORITY}. 8367 8368If this macro is not defined by the target, a suitable default will 8369be chosen if (1) the target supports arbitrary section names, (2) the 8370target defines @code{CTORS_SECTION_ASM_OP}, or (3) @code{USE_COLLECT2} 8371is not defined. 8372@end deftypefn 8373 8374@hook TARGET_ASM_DESTRUCTOR 8375This is like @code{TARGET_ASM_CONSTRUCTOR} but used for termination 8376functions rather than initialization functions. 8377@end deftypefn 8378 8379If @code{TARGET_HAVE_CTORS_DTORS} is true, the initialization routine 8380generated for the generated object file will have static linkage. 8381 8382If your system uses @command{collect2} as the means of processing 8383constructors, then that program normally uses @command{nm} to scan 8384an object file for constructor functions to be called. 8385 8386On certain kinds of systems, you can define this macro to make 8387@command{collect2} work faster (and, in some cases, make it work at all): 8388 8389@defmac OBJECT_FORMAT_COFF 8390Define this macro if the system uses COFF (Common Object File Format) 8391object files, so that @command{collect2} can assume this format and scan 8392object files directly for dynamic constructor/destructor functions. 8393 8394This macro is effective only in a native compiler; @command{collect2} as 8395part of a cross compiler always uses @command{nm} for the target machine. 8396@end defmac 8397 8398@defmac REAL_NM_FILE_NAME 8399Define this macro as a C string constant containing the file name to use 8400to execute @command{nm}. The default is to search the path normally for 8401@command{nm}. 8402@end defmac 8403 8404@defmac NM_FLAGS 8405@command{collect2} calls @command{nm} to scan object files for static 8406constructors and destructors and LTO info. By default, @option{-n} is 8407passed. Define @code{NM_FLAGS} to a C string constant if other options 8408are needed to get the same output format as GNU @command{nm -n} 8409produces. 8410@end defmac 8411 8412If your system supports shared libraries and has a program to list the 8413dynamic dependencies of a given library or executable, you can define 8414these macros to enable support for running initialization and 8415termination functions in shared libraries: 8416 8417@defmac LDD_SUFFIX 8418Define this macro to a C string constant containing the name of the program 8419which lists dynamic dependencies, like @command{ldd} under SunOS 4. 8420@end defmac 8421 8422@defmac PARSE_LDD_OUTPUT (@var{ptr}) 8423Define this macro to be C code that extracts filenames from the output 8424of the program denoted by @code{LDD_SUFFIX}. @var{ptr} is a variable 8425of type @code{char *} that points to the beginning of a line of output 8426from @code{LDD_SUFFIX}. If the line lists a dynamic dependency, the 8427code must advance @var{ptr} to the beginning of the filename on that 8428line. Otherwise, it must set @var{ptr} to @code{NULL}. 8429@end defmac 8430 8431@defmac SHLIB_SUFFIX 8432Define this macro to a C string constant containing the default shared 8433library extension of the target (e.g., @samp{".so"}). @command{collect2} 8434strips version information after this suffix when generating global 8435constructor and destructor names. This define is only needed on targets 8436that use @command{collect2} to process constructors and destructors. 8437@end defmac 8438 8439@node Instruction Output 8440@subsection Output of Assembler Instructions 8441 8442@c prevent bad page break with this line 8443This describes assembler instruction output. 8444 8445@defmac REGISTER_NAMES 8446A C initializer containing the assembler's names for the machine 8447registers, each one as a C string constant. This is what translates 8448register numbers in the compiler into assembler language. 8449@end defmac 8450 8451@defmac ADDITIONAL_REGISTER_NAMES 8452If defined, a C initializer for an array of structures containing a name 8453and a register number. This macro defines additional names for hard 8454registers, thus allowing the @code{asm} option in declarations to refer 8455to registers using alternate names. 8456@end defmac 8457 8458@defmac OVERLAPPING_REGISTER_NAMES 8459If defined, a C initializer for an array of structures containing a 8460name, a register number and a count of the number of consecutive 8461machine registers the name overlaps. This macro defines additional 8462names for hard registers, thus allowing the @code{asm} option in 8463declarations to refer to registers using alternate names. Unlike 8464@code{ADDITIONAL_REGISTER_NAMES}, this macro should be used when the 8465register name implies multiple underlying registers. 8466 8467This macro should be used when it is important that a clobber in an 8468@code{asm} statement clobbers all the underlying values implied by the 8469register name. For example, on ARM, clobbering the double-precision 8470VFP register ``d0'' implies clobbering both single-precision registers 8471``s0'' and ``s1''. 8472@end defmac 8473 8474@defmac ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr}) 8475Define this macro if you are using an unusual assembler that 8476requires different names for the machine instructions. 8477 8478The definition is a C statement or statements which output an 8479assembler instruction opcode to the stdio stream @var{stream}. The 8480macro-operand @var{ptr} is a variable of type @code{char *} which 8481points to the opcode name in its ``internal'' form---the form that is 8482written in the machine description. The definition should output the 8483opcode name to @var{stream}, performing any translation you desire, and 8484increment the variable @var{ptr} to point at the end of the opcode 8485so that it will not be output twice. 8486 8487In fact, your macro definition may process less than the entire opcode 8488name, or more than the opcode name; but if you want to process text 8489that includes @samp{%}-sequences to substitute operands, you must take 8490care of the substitution yourself. Just be sure to increment 8491@var{ptr} over whatever text should not be output normally. 8492 8493@findex recog_data.operand 8494If you need to look at the operand values, they can be found as the 8495elements of @code{recog_data.operand}. 8496 8497If the macro definition does nothing, the instruction is output 8498in the usual way. 8499@end defmac 8500 8501@defmac FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands}) 8502If defined, a C statement to be executed just prior to the output of 8503assembler code for @var{insn}, to modify the extracted operands so 8504they will be output differently. 8505 8506Here the argument @var{opvec} is the vector containing the operands 8507extracted from @var{insn}, and @var{noperands} is the number of 8508elements of the vector which contain meaningful data for this insn. 8509The contents of this vector are what will be used to convert the insn 8510template into assembler code, so you can change the assembler output 8511by changing the contents of the vector. 8512 8513This macro is useful when various assembler syntaxes share a single 8514file of instruction patterns; by defining this macro differently, you 8515can cause a large class of instructions to be output differently (such 8516as with rearranged operands). Naturally, variations in assembler 8517syntax affecting individual insn patterns ought to be handled by 8518writing conditional output routines in those patterns. 8519 8520If this macro is not defined, it is equivalent to a null statement. 8521@end defmac 8522 8523@hook TARGET_ASM_FINAL_POSTSCAN_INSN 8524If defined, this target hook is a function which is executed just after the 8525output of assembler code for @var{insn}, to change the mode of the assembler 8526if necessary. 8527 8528Here the argument @var{opvec} is the vector containing the operands 8529extracted from @var{insn}, and @var{noperands} is the number of 8530elements of the vector which contain meaningful data for this insn. 8531The contents of this vector are what was used to convert the insn 8532template into assembler code, so you can change the assembler mode 8533by checking the contents of the vector. 8534@end deftypefn 8535 8536@defmac PRINT_OPERAND (@var{stream}, @var{x}, @var{code}) 8537A C compound statement to output to stdio stream @var{stream} the 8538assembler syntax for an instruction operand @var{x}. @var{x} is an 8539RTL expression. 8540 8541@var{code} is a value that can be used to specify one of several ways 8542of printing the operand. It is used when identical operands must be 8543printed differently depending on the context. @var{code} comes from 8544the @samp{%} specification that was used to request printing of the 8545operand. If the specification was just @samp{%@var{digit}} then 8546@var{code} is 0; if the specification was @samp{%@var{ltr} 8547@var{digit}} then @var{code} is the ASCII code for @var{ltr}. 8548 8549@findex reg_names 8550If @var{x} is a register, this macro should print the register's name. 8551The names can be found in an array @code{reg_names} whose type is 8552@code{char *[]}. @code{reg_names} is initialized from 8553@code{REGISTER_NAMES}. 8554 8555When the machine description has a specification @samp{%@var{punct}} 8556(a @samp{%} followed by a punctuation character), this macro is called 8557with a null pointer for @var{x} and the punctuation character for 8558@var{code}. 8559@end defmac 8560 8561@defmac PRINT_OPERAND_PUNCT_VALID_P (@var{code}) 8562A C expression which evaluates to true if @var{code} is a valid 8563punctuation character for use in the @code{PRINT_OPERAND} macro. If 8564@code{PRINT_OPERAND_PUNCT_VALID_P} is not defined, it means that no 8565punctuation characters (except for the standard one, @samp{%}) are used 8566in this way. 8567@end defmac 8568 8569@defmac PRINT_OPERAND_ADDRESS (@var{stream}, @var{x}) 8570A C compound statement to output to stdio stream @var{stream} the 8571assembler syntax for an instruction operand that is a memory reference 8572whose address is @var{x}. @var{x} is an RTL expression. 8573 8574@cindex @code{TARGET_ENCODE_SECTION_INFO} usage 8575On some machines, the syntax for a symbolic address depends on the 8576section that the address refers to. On these machines, define the hook 8577@code{TARGET_ENCODE_SECTION_INFO} to store the information into the 8578@code{symbol_ref}, and then check for it here. @xref{Assembler 8579Format}. 8580@end defmac 8581 8582@findex dbr_sequence_length 8583@defmac DBR_OUTPUT_SEQEND (@var{file}) 8584A C statement, to be executed after all slot-filler instructions have 8585been output. If necessary, call @code{dbr_sequence_length} to 8586determine the number of slots filled in a sequence (zero if not 8587currently outputting a sequence), to decide how many no-ops to output, 8588or whatever. 8589 8590Don't define this macro if it has nothing to do, but it is helpful in 8591reading assembly output if the extent of the delay sequence is made 8592explicit (e.g.@: with white space). 8593@end defmac 8594 8595@findex final_sequence 8596Note that output routines for instructions with delay slots must be 8597prepared to deal with not being output as part of a sequence 8598(i.e.@: when the scheduling pass is not run, or when no slot fillers could be 8599found.) The variable @code{final_sequence} is null when not 8600processing a sequence, otherwise it contains the @code{sequence} rtx 8601being output. 8602 8603@findex asm_fprintf 8604@defmac REGISTER_PREFIX 8605@defmacx LOCAL_LABEL_PREFIX 8606@defmacx USER_LABEL_PREFIX 8607@defmacx IMMEDIATE_PREFIX 8608If defined, C string expressions to be used for the @samp{%R}, @samp{%L}, 8609@samp{%U}, and @samp{%I} options of @code{asm_fprintf} (see 8610@file{final.c}). These are useful when a single @file{md} file must 8611support multiple assembler formats. In that case, the various @file{tm.h} 8612files can define these macros differently. 8613@end defmac 8614 8615@defmac ASM_FPRINTF_EXTENSIONS (@var{file}, @var{argptr}, @var{format}) 8616If defined this macro should expand to a series of @code{case} 8617statements which will be parsed inside the @code{switch} statement of 8618the @code{asm_fprintf} function. This allows targets to define extra 8619printf formats which may useful when generating their assembler 8620statements. Note that uppercase letters are reserved for future 8621generic extensions to asm_fprintf, and so are not available to target 8622specific code. The output file is given by the parameter @var{file}. 8623The varargs input pointer is @var{argptr} and the rest of the format 8624string, starting the character after the one that is being switched 8625upon, is pointed to by @var{format}. 8626@end defmac 8627 8628@defmac ASSEMBLER_DIALECT 8629If your target supports multiple dialects of assembler language (such as 8630different opcodes), define this macro as a C expression that gives the 8631numeric index of the assembler language dialect to use, with zero as the 8632first variant. 8633 8634If this macro is defined, you may use constructs of the form 8635@smallexample 8636@samp{@{option0|option1|option2@dots{}@}} 8637@end smallexample 8638@noindent 8639in the output templates of patterns (@pxref{Output Template}) or in the 8640first argument of @code{asm_fprintf}. This construct outputs 8641@samp{option0}, @samp{option1}, @samp{option2}, etc., if the value of 8642@code{ASSEMBLER_DIALECT} is zero, one, two, etc. Any special characters 8643within these strings retain their usual meaning. If there are fewer 8644alternatives within the braces than the value of 8645@code{ASSEMBLER_DIALECT}, the construct outputs nothing. 8646 8647If you do not define this macro, the characters @samp{@{}, @samp{|} and 8648@samp{@}} do not have any special meaning when used in templates or 8649operands to @code{asm_fprintf}. 8650 8651Define the macros @code{REGISTER_PREFIX}, @code{LOCAL_LABEL_PREFIX}, 8652@code{USER_LABEL_PREFIX} and @code{IMMEDIATE_PREFIX} if you can express 8653the variations in assembler language syntax with that mechanism. Define 8654@code{ASSEMBLER_DIALECT} and use the @samp{@{option0|option1@}} syntax 8655if the syntax variant are larger and involve such things as different 8656opcodes or operand order. 8657@end defmac 8658 8659@defmac ASM_OUTPUT_REG_PUSH (@var{stream}, @var{regno}) 8660A C expression to output to @var{stream} some assembler code 8661which will push hard register number @var{regno} onto the stack. 8662The code need not be optimal, since this macro is used only when 8663profiling. 8664@end defmac 8665 8666@defmac ASM_OUTPUT_REG_POP (@var{stream}, @var{regno}) 8667A C expression to output to @var{stream} some assembler code 8668which will pop hard register number @var{regno} off of the stack. 8669The code need not be optimal, since this macro is used only when 8670profiling. 8671@end defmac 8672 8673@node Dispatch Tables 8674@subsection Output of Dispatch Tables 8675 8676@c prevent bad page break with this line 8677This concerns dispatch tables. 8678 8679@cindex dispatch table 8680@defmac ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{body}, @var{value}, @var{rel}) 8681A C statement to output to the stdio stream @var{stream} an assembler 8682pseudo-instruction to generate a difference between two labels. 8683@var{value} and @var{rel} are the numbers of two internal labels. The 8684definitions of these labels are output using 8685@code{(*targetm.asm_out.internal_label)}, and they must be printed in the same 8686way here. For example, 8687 8688@smallexample 8689fprintf (@var{stream}, "\t.word L%d-L%d\n", 8690 @var{value}, @var{rel}) 8691@end smallexample 8692 8693You must provide this macro on machines where the addresses in a 8694dispatch table are relative to the table's own address. If defined, GCC 8695will also use this macro on all machines when producing PIC@. 8696@var{body} is the body of the @code{ADDR_DIFF_VEC}; it is provided so that the 8697mode and flags can be read. 8698@end defmac 8699 8700@defmac ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value}) 8701This macro should be provided on machines where the addresses 8702in a dispatch table are absolute. 8703 8704The definition should be a C statement to output to the stdio stream 8705@var{stream} an assembler pseudo-instruction to generate a reference to 8706a label. @var{value} is the number of an internal label whose 8707definition is output using @code{(*targetm.asm_out.internal_label)}. 8708For example, 8709 8710@smallexample 8711fprintf (@var{stream}, "\t.word L%d\n", @var{value}) 8712@end smallexample 8713@end defmac 8714 8715@defmac ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table}) 8716Define this if the label before a jump-table needs to be output 8717specially. The first three arguments are the same as for 8718@code{(*targetm.asm_out.internal_label)}; the fourth argument is the 8719jump-table which follows (a @code{jump_insn} containing an 8720@code{addr_vec} or @code{addr_diff_vec}). 8721 8722This feature is used on system V to output a @code{swbeg} statement 8723for the table. 8724 8725If this macro is not defined, these labels are output with 8726@code{(*targetm.asm_out.internal_label)}. 8727@end defmac 8728 8729@defmac ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table}) 8730Define this if something special must be output at the end of a 8731jump-table. The definition should be a C statement to be executed 8732after the assembler code for the table is written. It should write 8733the appropriate code to stdio stream @var{stream}. The argument 8734@var{table} is the jump-table insn, and @var{num} is the label-number 8735of the preceding label. 8736 8737If this macro is not defined, nothing special is output at the end of 8738the jump-table. 8739@end defmac 8740 8741@hook TARGET_ASM_EMIT_UNWIND_LABEL 8742This target hook emits a label at the beginning of each FDE@. It 8743should be defined on targets where FDEs need special labels, and it 8744should write the appropriate label, for the FDE associated with the 8745function declaration @var{decl}, to the stdio stream @var{stream}. 8746The third argument, @var{for_eh}, is a boolean: true if this is for an 8747exception table. The fourth argument, @var{empty}, is a boolean: 8748true if this is a placeholder label for an omitted FDE@. 8749 8750The default is that FDEs are not given nonlocal labels. 8751@end deftypefn 8752 8753@hook TARGET_ASM_EMIT_EXCEPT_TABLE_LABEL 8754This target hook emits a label at the beginning of the exception table. 8755It should be defined on targets where it is desirable for the table 8756to be broken up according to function. 8757 8758The default is that no label is emitted. 8759@end deftypefn 8760 8761@hook TARGET_ASM_EMIT_EXCEPT_PERSONALITY 8762 8763@hook TARGET_ASM_UNWIND_EMIT 8764This target hook emits assembly directives required to unwind the 8765given instruction. This is only used when @code{TARGET_EXCEPT_UNWIND_INFO} 8766returns @code{UI_TARGET}. 8767@end deftypefn 8768 8769@hook TARGET_ASM_UNWIND_EMIT_BEFORE_INSN 8770 8771@node Exception Region Output 8772@subsection Assembler Commands for Exception Regions 8773 8774@c prevent bad page break with this line 8775 8776This describes commands marking the start and the end of an exception 8777region. 8778 8779@defmac EH_FRAME_SECTION_NAME 8780If defined, a C string constant for the name of the section containing 8781exception handling frame unwind information. If not defined, GCC will 8782provide a default definition if the target supports named sections. 8783@file{crtstuff.c} uses this macro to switch to the appropriate section. 8784 8785You should define this symbol if your target supports DWARF 2 frame 8786unwind information and the default definition does not work. 8787@end defmac 8788 8789@defmac EH_FRAME_IN_DATA_SECTION 8790If defined, DWARF 2 frame unwind information will be placed in the 8791data section even though the target supports named sections. This 8792might be necessary, for instance, if the system linker does garbage 8793collection and sections cannot be marked as not to be collected. 8794 8795Do not define this macro unless @code{TARGET_ASM_NAMED_SECTION} is 8796also defined. 8797@end defmac 8798 8799@defmac EH_TABLES_CAN_BE_READ_ONLY 8800Define this macro to 1 if your target is such that no frame unwind 8801information encoding used with non-PIC code will ever require a 8802runtime relocation, but the linker may not support merging read-only 8803and read-write sections into a single read-write section. 8804@end defmac 8805 8806@defmac MASK_RETURN_ADDR 8807An rtx used to mask the return address found via @code{RETURN_ADDR_RTX}, so 8808that it does not contain any extraneous set bits in it. 8809@end defmac 8810 8811@defmac DWARF2_UNWIND_INFO 8812Define this macro to 0 if your target supports DWARF 2 frame unwind 8813information, but it does not yet work with exception handling. 8814Otherwise, if your target supports this information (if it defines 8815@code{INCOMING_RETURN_ADDR_RTX} and either @code{UNALIGNED_INT_ASM_OP} 8816or @code{OBJECT_FORMAT_ELF}), GCC will provide a default definition of 1. 8817@end defmac 8818 8819@hook TARGET_EXCEPT_UNWIND_INFO 8820This hook defines the mechanism that will be used for exception handling 8821by the target. If the target has ABI specified unwind tables, the hook 8822should return @code{UI_TARGET}. If the target is to use the 8823@code{setjmp}/@code{longjmp}-based exception handling scheme, the hook 8824should return @code{UI_SJLJ}. If the target supports DWARF 2 frame unwind 8825information, the hook should return @code{UI_DWARF2}. 8826 8827A target may, if exceptions are disabled, choose to return @code{UI_NONE}. 8828This may end up simplifying other parts of target-specific code. The 8829default implementation of this hook never returns @code{UI_NONE}. 8830 8831Note that the value returned by this hook should be constant. It should 8832not depend on anything except the command-line switches described by 8833@var{opts}. In particular, the 8834setting @code{UI_SJLJ} must be fixed at compiler start-up as C pre-processor 8835macros and builtin functions related to exception handling are set up 8836depending on this setting. 8837 8838The default implementation of the hook first honors the 8839@option{--enable-sjlj-exceptions} configure option, then 8840@code{DWARF2_UNWIND_INFO}, and finally defaults to @code{UI_SJLJ}. If 8841@code{DWARF2_UNWIND_INFO} depends on command-line options, the target 8842must define this hook so that @var{opts} is used correctly. 8843@end deftypefn 8844 8845@hook TARGET_UNWIND_TABLES_DEFAULT 8846This variable should be set to @code{true} if the target ABI requires unwinding 8847tables even when exceptions are not used. It must not be modified by 8848command-line option processing. 8849@end deftypevr 8850 8851@defmac DONT_USE_BUILTIN_SETJMP 8852Define this macro to 1 if the @code{setjmp}/@code{longjmp}-based scheme 8853should use the @code{setjmp}/@code{longjmp} functions from the C library 8854instead of the @code{__builtin_setjmp}/@code{__builtin_longjmp} machinery. 8855@end defmac 8856 8857@defmac DWARF_CIE_DATA_ALIGNMENT 8858This macro need only be defined if the target might save registers in the 8859function prologue at an offset to the stack pointer that is not aligned to 8860@code{UNITS_PER_WORD}. The definition should be the negative minimum 8861alignment if @code{STACK_GROWS_DOWNWARD} is defined, and the positive 8862minimum alignment otherwise. @xref{SDB and DWARF}. Only applicable if 8863the target supports DWARF 2 frame unwind information. 8864@end defmac 8865 8866@hook TARGET_TERMINATE_DW2_EH_FRAME_INFO 8867Contains the value true if the target should add a zero word onto the 8868end of a Dwarf-2 frame info section when used for exception handling. 8869Default value is false if @code{EH_FRAME_SECTION_NAME} is defined, and 8870true otherwise. 8871@end deftypevr 8872 8873@hook TARGET_DWARF_REGISTER_SPAN 8874Given a register, this hook should return a parallel of registers to 8875represent where to find the register pieces. Define this hook if the 8876register and its mode are represented in Dwarf in non-contiguous 8877locations, or if the register should be represented in more than one 8878register in Dwarf. Otherwise, this hook should return @code{NULL_RTX}. 8879If not defined, the default is to return @code{NULL_RTX}. 8880@end deftypefn 8881 8882@hook TARGET_INIT_DWARF_REG_SIZES_EXTRA 8883If some registers are represented in Dwarf-2 unwind information in 8884multiple pieces, define this hook to fill in information about the 8885sizes of those pieces in the table used by the unwinder at runtime. 8886It will be called by @code{expand_builtin_init_dwarf_reg_sizes} after 8887filling in a single size corresponding to each hard register; 8888@var{address} is the address of the table. 8889@end deftypefn 8890 8891@hook TARGET_ASM_TTYPE 8892This hook is used to output a reference from a frame unwinding table to 8893the type_info object identified by @var{sym}. It should return @code{true} 8894if the reference was output. Returning @code{false} will cause the 8895reference to be output using the normal Dwarf2 routines. 8896@end deftypefn 8897 8898@hook TARGET_ARM_EABI_UNWINDER 8899This flag should be set to @code{true} on targets that use an ARM EABI 8900based unwinding library, and @code{false} on other targets. This effects 8901the format of unwinding tables, and how the unwinder in entered after 8902running a cleanup. The default is @code{false}. 8903@end deftypevr 8904 8905@node Alignment Output 8906@subsection Assembler Commands for Alignment 8907 8908@c prevent bad page break with this line 8909This describes commands for alignment. 8910 8911@defmac JUMP_ALIGN (@var{label}) 8912The alignment (log base 2) to put in front of @var{label}, which is 8913a common destination of jumps and has no fallthru incoming edge. 8914 8915This macro need not be defined if you don't want any special alignment 8916to be done at such a time. Most machine descriptions do not currently 8917define the macro. 8918 8919Unless it's necessary to inspect the @var{label} parameter, it is better 8920to set the variable @var{align_jumps} in the target's 8921@code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's 8922selection in @var{align_jumps} in a @code{JUMP_ALIGN} implementation. 8923@end defmac 8924 8925@hook TARGET_ASM_JUMP_ALIGN_MAX_SKIP 8926The maximum number of bytes to skip before @var{label} when applying 8927@code{JUMP_ALIGN}. This works only if 8928@code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined. 8929@end deftypefn 8930 8931@defmac LABEL_ALIGN_AFTER_BARRIER (@var{label}) 8932The alignment (log base 2) to put in front of @var{label}, which follows 8933a @code{BARRIER}. 8934 8935This macro need not be defined if you don't want any special alignment 8936to be done at such a time. Most machine descriptions do not currently 8937define the macro. 8938@end defmac 8939 8940@hook TARGET_ASM_LABEL_ALIGN_AFTER_BARRIER_MAX_SKIP 8941The maximum number of bytes to skip before @var{label} when applying 8942@code{LABEL_ALIGN_AFTER_BARRIER}. This works only if 8943@code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined. 8944@end deftypefn 8945 8946@defmac LOOP_ALIGN (@var{label}) 8947The alignment (log base 2) to put in front of @var{label}, which follows 8948a @code{NOTE_INSN_LOOP_BEG} note. 8949 8950This macro need not be defined if you don't want any special alignment 8951to be done at such a time. Most machine descriptions do not currently 8952define the macro. 8953 8954Unless it's necessary to inspect the @var{label} parameter, it is better 8955to set the variable @code{align_loops} in the target's 8956@code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's 8957selection in @code{align_loops} in a @code{LOOP_ALIGN} implementation. 8958@end defmac 8959 8960@hook TARGET_ASM_LOOP_ALIGN_MAX_SKIP 8961The maximum number of bytes to skip when applying @code{LOOP_ALIGN} to 8962@var{label}. This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is 8963defined. 8964@end deftypefn 8965 8966@defmac LABEL_ALIGN (@var{label}) 8967The alignment (log base 2) to put in front of @var{label}. 8968If @code{LABEL_ALIGN_AFTER_BARRIER} / @code{LOOP_ALIGN} specify a different alignment, 8969the maximum of the specified values is used. 8970 8971Unless it's necessary to inspect the @var{label} parameter, it is better 8972to set the variable @code{align_labels} in the target's 8973@code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's 8974selection in @code{align_labels} in a @code{LABEL_ALIGN} implementation. 8975@end defmac 8976 8977@hook TARGET_ASM_LABEL_ALIGN_MAX_SKIP 8978The maximum number of bytes to skip when applying @code{LABEL_ALIGN} 8979to @var{label}. This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} 8980is defined. 8981@end deftypefn 8982 8983@defmac ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes}) 8984A C statement to output to the stdio stream @var{stream} an assembler 8985instruction to advance the location counter by @var{nbytes} bytes. 8986Those bytes should be zero when loaded. @var{nbytes} will be a C 8987expression of type @code{unsigned HOST_WIDE_INT}. 8988@end defmac 8989 8990@defmac ASM_NO_SKIP_IN_TEXT 8991Define this macro if @code{ASM_OUTPUT_SKIP} should not be used in the 8992text section because it fails to put zeros in the bytes that are skipped. 8993This is true on many Unix systems, where the pseudo--op to skip bytes 8994produces no-op instructions rather than zeros when used in the text 8995section. 8996@end defmac 8997 8998@defmac ASM_OUTPUT_ALIGN (@var{stream}, @var{power}) 8999A C statement to output to the stdio stream @var{stream} an assembler 9000command to advance the location counter to a multiple of 2 to the 9001@var{power} bytes. @var{power} will be a C expression of type @code{int}. 9002@end defmac 9003 9004@defmac ASM_OUTPUT_ALIGN_WITH_NOP (@var{stream}, @var{power}) 9005Like @code{ASM_OUTPUT_ALIGN}, except that the ``nop'' instruction is used 9006for padding, if necessary. 9007@end defmac 9008 9009@defmac ASM_OUTPUT_MAX_SKIP_ALIGN (@var{stream}, @var{power}, @var{max_skip}) 9010A C statement to output to the stdio stream @var{stream} an assembler 9011command to advance the location counter to a multiple of 2 to the 9012@var{power} bytes, but only if @var{max_skip} or fewer bytes are needed to 9013satisfy the alignment request. @var{power} and @var{max_skip} will be 9014a C expression of type @code{int}. 9015@end defmac 9016 9017@need 3000 9018@node Debugging Info 9019@section Controlling Debugging Information Format 9020 9021@c prevent bad page break with this line 9022This describes how to specify debugging information. 9023 9024@menu 9025* All Debuggers:: Macros that affect all debugging formats uniformly. 9026* DBX Options:: Macros enabling specific options in DBX format. 9027* DBX Hooks:: Hook macros for varying DBX format. 9028* File Names and DBX:: Macros controlling output of file names in DBX format. 9029* SDB and DWARF:: Macros for SDB (COFF) and DWARF formats. 9030* VMS Debug:: Macros for VMS debug format. 9031@end menu 9032 9033@node All Debuggers 9034@subsection Macros Affecting All Debugging Formats 9035 9036@c prevent bad page break with this line 9037These macros affect all debugging formats. 9038 9039@defmac DBX_REGISTER_NUMBER (@var{regno}) 9040A C expression that returns the DBX register number for the compiler 9041register number @var{regno}. In the default macro provided, the value 9042of this expression will be @var{regno} itself. But sometimes there are 9043some registers that the compiler knows about and DBX does not, or vice 9044versa. In such cases, some register may need to have one number in the 9045compiler and another for DBX@. 9046 9047If two registers have consecutive numbers inside GCC, and they can be 9048used as a pair to hold a multiword value, then they @emph{must} have 9049consecutive numbers after renumbering with @code{DBX_REGISTER_NUMBER}. 9050Otherwise, debuggers will be unable to access such a pair, because they 9051expect register pairs to be consecutive in their own numbering scheme. 9052 9053If you find yourself defining @code{DBX_REGISTER_NUMBER} in way that 9054does not preserve register pairs, then what you must do instead is 9055redefine the actual register numbering scheme. 9056@end defmac 9057 9058@defmac DEBUGGER_AUTO_OFFSET (@var{x}) 9059A C expression that returns the integer offset value for an automatic 9060variable having address @var{x} (an RTL expression). The default 9061computation assumes that @var{x} is based on the frame-pointer and 9062gives the offset from the frame-pointer. This is required for targets 9063that produce debugging output for DBX or COFF-style debugging output 9064for SDB and allow the frame-pointer to be eliminated when the 9065@option{-g} options is used. 9066@end defmac 9067 9068@defmac DEBUGGER_ARG_OFFSET (@var{offset}, @var{x}) 9069A C expression that returns the integer offset value for an argument 9070having address @var{x} (an RTL expression). The nominal offset is 9071@var{offset}. 9072@end defmac 9073 9074@defmac PREFERRED_DEBUGGING_TYPE 9075A C expression that returns the type of debugging output GCC should 9076produce when the user specifies just @option{-g}. Define 9077this if you have arranged for GCC to support more than one format of 9078debugging output. Currently, the allowable values are @code{DBX_DEBUG}, 9079@code{SDB_DEBUG}, @code{DWARF_DEBUG}, @code{DWARF2_DEBUG}, 9080@code{XCOFF_DEBUG}, @code{VMS_DEBUG}, and @code{VMS_AND_DWARF2_DEBUG}. 9081 9082When the user specifies @option{-ggdb}, GCC normally also uses the 9083value of this macro to select the debugging output format, but with two 9084exceptions. If @code{DWARF2_DEBUGGING_INFO} is defined, GCC uses the 9085value @code{DWARF2_DEBUG}. Otherwise, if @code{DBX_DEBUGGING_INFO} is 9086defined, GCC uses @code{DBX_DEBUG}. 9087 9088The value of this macro only affects the default debugging output; the 9089user can always get a specific type of output by using @option{-gstabs}, 9090@option{-gcoff}, @option{-gdwarf-2}, @option{-gxcoff}, or @option{-gvms}. 9091@end defmac 9092 9093@node DBX Options 9094@subsection Specific Options for DBX Output 9095 9096@c prevent bad page break with this line 9097These are specific options for DBX output. 9098 9099@defmac DBX_DEBUGGING_INFO 9100Define this macro if GCC should produce debugging output for DBX 9101in response to the @option{-g} option. 9102@end defmac 9103 9104@defmac XCOFF_DEBUGGING_INFO 9105Define this macro if GCC should produce XCOFF format debugging output 9106in response to the @option{-g} option. This is a variant of DBX format. 9107@end defmac 9108 9109@defmac DEFAULT_GDB_EXTENSIONS 9110Define this macro to control whether GCC should by default generate 9111GDB's extended version of DBX debugging information (assuming DBX-format 9112debugging information is enabled at all). If you don't define the 9113macro, the default is 1: always generate the extended information 9114if there is any occasion to. 9115@end defmac 9116 9117@defmac DEBUG_SYMS_TEXT 9118Define this macro if all @code{.stabs} commands should be output while 9119in the text section. 9120@end defmac 9121 9122@defmac ASM_STABS_OP 9123A C string constant, including spacing, naming the assembler pseudo op to 9124use instead of @code{"\t.stabs\t"} to define an ordinary debugging symbol. 9125If you don't define this macro, @code{"\t.stabs\t"} is used. This macro 9126applies only to DBX debugging information format. 9127@end defmac 9128 9129@defmac ASM_STABD_OP 9130A C string constant, including spacing, naming the assembler pseudo op to 9131use instead of @code{"\t.stabd\t"} to define a debugging symbol whose 9132value is the current location. If you don't define this macro, 9133@code{"\t.stabd\t"} is used. This macro applies only to DBX debugging 9134information format. 9135@end defmac 9136 9137@defmac ASM_STABN_OP 9138A C string constant, including spacing, naming the assembler pseudo op to 9139use instead of @code{"\t.stabn\t"} to define a debugging symbol with no 9140name. If you don't define this macro, @code{"\t.stabn\t"} is used. This 9141macro applies only to DBX debugging information format. 9142@end defmac 9143 9144@defmac DBX_NO_XREFS 9145Define this macro if DBX on your system does not support the construct 9146@samp{xs@var{tagname}}. On some systems, this construct is used to 9147describe a forward reference to a structure named @var{tagname}. 9148On other systems, this construct is not supported at all. 9149@end defmac 9150 9151@defmac DBX_CONTIN_LENGTH 9152A symbol name in DBX-format debugging information is normally 9153continued (split into two separate @code{.stabs} directives) when it 9154exceeds a certain length (by default, 80 characters). On some 9155operating systems, DBX requires this splitting; on others, splitting 9156must not be done. You can inhibit splitting by defining this macro 9157with the value zero. You can override the default splitting-length by 9158defining this macro as an expression for the length you desire. 9159@end defmac 9160 9161@defmac DBX_CONTIN_CHAR 9162Normally continuation is indicated by adding a @samp{\} character to 9163the end of a @code{.stabs} string when a continuation follows. To use 9164a different character instead, define this macro as a character 9165constant for the character you want to use. Do not define this macro 9166if backslash is correct for your system. 9167@end defmac 9168 9169@defmac DBX_STATIC_STAB_DATA_SECTION 9170Define this macro if it is necessary to go to the data section before 9171outputting the @samp{.stabs} pseudo-op for a non-global static 9172variable. 9173@end defmac 9174 9175@defmac DBX_TYPE_DECL_STABS_CODE 9176The value to use in the ``code'' field of the @code{.stabs} directive 9177for a typedef. The default is @code{N_LSYM}. 9178@end defmac 9179 9180@defmac DBX_STATIC_CONST_VAR_CODE 9181The value to use in the ``code'' field of the @code{.stabs} directive 9182for a static variable located in the text section. DBX format does not 9183provide any ``right'' way to do this. The default is @code{N_FUN}. 9184@end defmac 9185 9186@defmac DBX_REGPARM_STABS_CODE 9187The value to use in the ``code'' field of the @code{.stabs} directive 9188for a parameter passed in registers. DBX format does not provide any 9189``right'' way to do this. The default is @code{N_RSYM}. 9190@end defmac 9191 9192@defmac DBX_REGPARM_STABS_LETTER 9193The letter to use in DBX symbol data to identify a symbol as a parameter 9194passed in registers. DBX format does not customarily provide any way to 9195do this. The default is @code{'P'}. 9196@end defmac 9197 9198@defmac DBX_FUNCTION_FIRST 9199Define this macro if the DBX information for a function and its 9200arguments should precede the assembler code for the function. Normally, 9201in DBX format, the debugging information entirely follows the assembler 9202code. 9203@end defmac 9204 9205@defmac DBX_BLOCKS_FUNCTION_RELATIVE 9206Define this macro, with value 1, if the value of a symbol describing 9207the scope of a block (@code{N_LBRAC} or @code{N_RBRAC}) should be 9208relative to the start of the enclosing function. Normally, GCC uses 9209an absolute address. 9210@end defmac 9211 9212@defmac DBX_LINES_FUNCTION_RELATIVE 9213Define this macro, with value 1, if the value of a symbol indicating 9214the current line number (@code{N_SLINE}) should be relative to the 9215start of the enclosing function. Normally, GCC uses an absolute address. 9216@end defmac 9217 9218@defmac DBX_USE_BINCL 9219Define this macro if GCC should generate @code{N_BINCL} and 9220@code{N_EINCL} stabs for included header files, as on Sun systems. This 9221macro also directs GCC to output a type number as a pair of a file 9222number and a type number within the file. Normally, GCC does not 9223generate @code{N_BINCL} or @code{N_EINCL} stabs, and it outputs a single 9224number for a type number. 9225@end defmac 9226 9227@node DBX Hooks 9228@subsection Open-Ended Hooks for DBX Format 9229 9230@c prevent bad page break with this line 9231These are hooks for DBX format. 9232 9233@defmac DBX_OUTPUT_LBRAC (@var{stream}, @var{name}) 9234Define this macro to say how to output to @var{stream} the debugging 9235information for the start of a scope level for variable names. The 9236argument @var{name} is the name of an assembler symbol (for use with 9237@code{assemble_name}) whose value is the address where the scope begins. 9238@end defmac 9239 9240@defmac DBX_OUTPUT_RBRAC (@var{stream}, @var{name}) 9241Like @code{DBX_OUTPUT_LBRAC}, but for the end of a scope level. 9242@end defmac 9243 9244@defmac DBX_OUTPUT_NFUN (@var{stream}, @var{lscope_label}, @var{decl}) 9245Define this macro if the target machine requires special handling to 9246output an @code{N_FUN} entry for the function @var{decl}. 9247@end defmac 9248 9249@defmac DBX_OUTPUT_SOURCE_LINE (@var{stream}, @var{line}, @var{counter}) 9250A C statement to output DBX debugging information before code for line 9251number @var{line} of the current source file to the stdio stream 9252@var{stream}. @var{counter} is the number of time the macro was 9253invoked, including the current invocation; it is intended to generate 9254unique labels in the assembly output. 9255 9256This macro should not be defined if the default output is correct, or 9257if it can be made correct by defining @code{DBX_LINES_FUNCTION_RELATIVE}. 9258@end defmac 9259 9260@defmac NO_DBX_FUNCTION_END 9261Some stabs encapsulation formats (in particular ECOFF), cannot handle the 9262@code{.stabs "",N_FUN,,0,0,Lscope-function-1} gdb dbx extension construct. 9263On those machines, define this macro to turn this feature off without 9264disturbing the rest of the gdb extensions. 9265@end defmac 9266 9267@defmac NO_DBX_BNSYM_ENSYM 9268Some assemblers cannot handle the @code{.stabd BNSYM/ENSYM,0,0} gdb dbx 9269extension construct. On those machines, define this macro to turn this 9270feature off without disturbing the rest of the gdb extensions. 9271@end defmac 9272 9273@node File Names and DBX 9274@subsection File Names in DBX Format 9275 9276@c prevent bad page break with this line 9277This describes file names in DBX format. 9278 9279@defmac DBX_OUTPUT_MAIN_SOURCE_FILENAME (@var{stream}, @var{name}) 9280A C statement to output DBX debugging information to the stdio stream 9281@var{stream}, which indicates that file @var{name} is the main source 9282file---the file specified as the input file for compilation. 9283This macro is called only once, at the beginning of compilation. 9284 9285This macro need not be defined if the standard form of output 9286for DBX debugging information is appropriate. 9287 9288It may be necessary to refer to a label equal to the beginning of the 9289text section. You can use @samp{assemble_name (stream, ltext_label_name)} 9290to do so. If you do this, you must also set the variable 9291@var{used_ltext_label_name} to @code{true}. 9292@end defmac 9293 9294@defmac NO_DBX_MAIN_SOURCE_DIRECTORY 9295Define this macro, with value 1, if GCC should not emit an indication 9296of the current directory for compilation and current source language at 9297the beginning of the file. 9298@end defmac 9299 9300@defmac NO_DBX_GCC_MARKER 9301Define this macro, with value 1, if GCC should not emit an indication 9302that this object file was compiled by GCC@. The default is to emit 9303an @code{N_OPT} stab at the beginning of every source file, with 9304@samp{gcc2_compiled.} for the string and value 0. 9305@end defmac 9306 9307@defmac DBX_OUTPUT_MAIN_SOURCE_FILE_END (@var{stream}, @var{name}) 9308A C statement to output DBX debugging information at the end of 9309compilation of the main source file @var{name}. Output should be 9310written to the stdio stream @var{stream}. 9311 9312If you don't define this macro, nothing special is output at the end 9313of compilation, which is correct for most machines. 9314@end defmac 9315 9316@defmac DBX_OUTPUT_NULL_N_SO_AT_MAIN_SOURCE_FILE_END 9317Define this macro @emph{instead of} defining 9318@code{DBX_OUTPUT_MAIN_SOURCE_FILE_END}, if what needs to be output at 9319the end of compilation is an @code{N_SO} stab with an empty string, 9320whose value is the highest absolute text address in the file. 9321@end defmac 9322 9323@need 2000 9324@node SDB and DWARF 9325@subsection Macros for SDB and DWARF Output 9326 9327@c prevent bad page break with this line 9328Here are macros for SDB and DWARF output. 9329 9330@defmac SDB_DEBUGGING_INFO 9331Define this macro if GCC should produce COFF-style debugging output 9332for SDB in response to the @option{-g} option. 9333@end defmac 9334 9335@defmac DWARF2_DEBUGGING_INFO 9336Define this macro if GCC should produce dwarf version 2 format 9337debugging output in response to the @option{-g} option. 9338 9339@hook TARGET_DWARF_CALLING_CONVENTION 9340Define this to enable the dwarf attribute @code{DW_AT_calling_convention} to 9341be emitted for each function. Instead of an integer return the enum 9342value for the @code{DW_CC_} tag. 9343@end deftypefn 9344 9345To support optional call frame debugging information, you must also 9346define @code{INCOMING_RETURN_ADDR_RTX} and either set 9347@code{RTX_FRAME_RELATED_P} on the prologue insns if you use RTL for the 9348prologue, or call @code{dwarf2out_def_cfa} and @code{dwarf2out_reg_save} 9349as appropriate from @code{TARGET_ASM_FUNCTION_PROLOGUE} if you don't. 9350@end defmac 9351 9352@defmac DWARF2_FRAME_INFO 9353Define this macro to a nonzero value if GCC should always output 9354Dwarf 2 frame information. If @code{TARGET_EXCEPT_UNWIND_INFO} 9355(@pxref{Exception Region Output}) returns @code{UI_DWARF2}, and 9356exceptions are enabled, GCC will output this information not matter 9357how you define @code{DWARF2_FRAME_INFO}. 9358@end defmac 9359 9360@hook TARGET_DEBUG_UNWIND_INFO 9361This hook defines the mechanism that will be used for describing frame 9362unwind information to the debugger. Normally the hook will return 9363@code{UI_DWARF2} if DWARF 2 debug information is enabled, and 9364return @code{UI_NONE} otherwise. 9365 9366A target may return @code{UI_DWARF2} even when DWARF 2 debug information 9367is disabled in order to always output DWARF 2 frame information. 9368 9369A target may return @code{UI_TARGET} if it has ABI specified unwind tables. 9370This will suppress generation of the normal debug frame unwind information. 9371@end deftypefn 9372 9373@defmac DWARF2_ASM_LINE_DEBUG_INFO 9374Define this macro to be a nonzero value if the assembler can generate Dwarf 2 9375line debug info sections. This will result in much more compact line number 9376tables, and hence is desirable if it works. 9377@end defmac 9378 9379@hook TARGET_WANT_DEBUG_PUB_SECTIONS 9380 9381@hook TARGET_FORCE_AT_COMP_DIR 9382 9383@hook TARGET_DELAY_SCHED2 9384 9385@hook TARGET_DELAY_VARTRACK 9386 9387@defmac ASM_OUTPUT_DWARF_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2}) 9388A C statement to issue assembly directives that create a difference 9389@var{lab1} minus @var{lab2}, using an integer of the given @var{size}. 9390@end defmac 9391 9392@defmac ASM_OUTPUT_DWARF_VMS_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2}) 9393A C statement to issue assembly directives that create a difference 9394between the two given labels in system defined units, e.g. instruction 9395slots on IA64 VMS, using an integer of the given size. 9396@end defmac 9397 9398@defmac ASM_OUTPUT_DWARF_OFFSET (@var{stream}, @var{size}, @var{label}, @var{section}) 9399A C statement to issue assembly directives that create a 9400section-relative reference to the given @var{label}, using an integer of the 9401given @var{size}. The label is known to be defined in the given @var{section}. 9402@end defmac 9403 9404@defmac ASM_OUTPUT_DWARF_PCREL (@var{stream}, @var{size}, @var{label}) 9405A C statement to issue assembly directives that create a self-relative 9406reference to the given @var{label}, using an integer of the given @var{size}. 9407@end defmac 9408 9409@defmac ASM_OUTPUT_DWARF_TABLE_REF (@var{label}) 9410A C statement to issue assembly directives that create a reference to 9411the DWARF table identifier @var{label} from the current section. This 9412is used on some systems to avoid garbage collecting a DWARF table which 9413is referenced by a function. 9414@end defmac 9415 9416@hook TARGET_ASM_OUTPUT_DWARF_DTPREL 9417If defined, this target hook is a function which outputs a DTP-relative 9418reference to the given TLS symbol of the specified size. 9419@end deftypefn 9420 9421@defmac PUT_SDB_@dots{} 9422Define these macros to override the assembler syntax for the special 9423SDB assembler directives. See @file{sdbout.c} for a list of these 9424macros and their arguments. If the standard syntax is used, you need 9425not define them yourself. 9426@end defmac 9427 9428@defmac SDB_DELIM 9429Some assemblers do not support a semicolon as a delimiter, even between 9430SDB assembler directives. In that case, define this macro to be the 9431delimiter to use (usually @samp{\n}). It is not necessary to define 9432a new set of @code{PUT_SDB_@var{op}} macros if this is the only change 9433required. 9434@end defmac 9435 9436@defmac SDB_ALLOW_UNKNOWN_REFERENCES 9437Define this macro to allow references to unknown structure, 9438union, or enumeration tags to be emitted. Standard COFF does not 9439allow handling of unknown references, MIPS ECOFF has support for 9440it. 9441@end defmac 9442 9443@defmac SDB_ALLOW_FORWARD_REFERENCES 9444Define this macro to allow references to structure, union, or 9445enumeration tags that have not yet been seen to be handled. Some 9446assemblers choke if forward tags are used, while some require it. 9447@end defmac 9448 9449@defmac SDB_OUTPUT_SOURCE_LINE (@var{stream}, @var{line}) 9450A C statement to output SDB debugging information before code for line 9451number @var{line} of the current source file to the stdio stream 9452@var{stream}. The default is to emit an @code{.ln} directive. 9453@end defmac 9454 9455@need 2000 9456@node VMS Debug 9457@subsection Macros for VMS Debug Format 9458 9459@c prevent bad page break with this line 9460Here are macros for VMS debug format. 9461 9462@defmac VMS_DEBUGGING_INFO 9463Define this macro if GCC should produce debugging output for VMS 9464in response to the @option{-g} option. The default behavior for VMS 9465is to generate minimal debug info for a traceback in the absence of 9466@option{-g} unless explicitly overridden with @option{-g0}. This 9467behavior is controlled by @code{TARGET_OPTION_OPTIMIZATION} and 9468@code{TARGET_OPTION_OVERRIDE}. 9469@end defmac 9470 9471@node Floating Point 9472@section Cross Compilation and Floating Point 9473@cindex cross compilation and floating point 9474@cindex floating point and cross compilation 9475 9476While all modern machines use twos-complement representation for integers, 9477there are a variety of representations for floating point numbers. This 9478means that in a cross-compiler the representation of floating point numbers 9479in the compiled program may be different from that used in the machine 9480doing the compilation. 9481 9482Because different representation systems may offer different amounts of 9483range and precision, all floating point constants must be represented in 9484the target machine's format. Therefore, the cross compiler cannot 9485safely use the host machine's floating point arithmetic; it must emulate 9486the target's arithmetic. To ensure consistency, GCC always uses 9487emulation to work with floating point values, even when the host and 9488target floating point formats are identical. 9489 9490The following macros are provided by @file{real.h} for the compiler to 9491use. All parts of the compiler which generate or optimize 9492floating-point calculations must use these macros. They may evaluate 9493their operands more than once, so operands must not have side effects. 9494 9495@defmac REAL_VALUE_TYPE 9496The C data type to be used to hold a floating point value in the target 9497machine's format. Typically this is a @code{struct} containing an 9498array of @code{HOST_WIDE_INT}, but all code should treat it as an opaque 9499quantity. 9500@end defmac 9501 9502@deftypefn Macro int REAL_VALUES_EQUAL (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y}) 9503Compares for equality the two values, @var{x} and @var{y}. If the target 9504floating point format supports negative zeroes and/or NaNs, 9505@samp{REAL_VALUES_EQUAL (-0.0, 0.0)} is true, and 9506@samp{REAL_VALUES_EQUAL (NaN, NaN)} is false. 9507@end deftypefn 9508 9509@deftypefn Macro int REAL_VALUES_LESS (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y}) 9510Tests whether @var{x} is less than @var{y}. 9511@end deftypefn 9512 9513@deftypefn Macro HOST_WIDE_INT REAL_VALUE_FIX (REAL_VALUE_TYPE @var{x}) 9514Truncates @var{x} to a signed integer, rounding toward zero. 9515@end deftypefn 9516 9517@deftypefn Macro {unsigned HOST_WIDE_INT} REAL_VALUE_UNSIGNED_FIX (REAL_VALUE_TYPE @var{x}) 9518Truncates @var{x} to an unsigned integer, rounding toward zero. If 9519@var{x} is negative, returns zero. 9520@end deftypefn 9521 9522@deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ATOF (const char *@var{string}, enum machine_mode @var{mode}) 9523Converts @var{string} into a floating point number in the target machine's 9524representation for mode @var{mode}. This routine can handle both 9525decimal and hexadecimal floating point constants, using the syntax 9526defined by the C language for both. 9527@end deftypefn 9528 9529@deftypefn Macro int REAL_VALUE_NEGATIVE (REAL_VALUE_TYPE @var{x}) 9530Returns 1 if @var{x} is negative (including negative zero), 0 otherwise. 9531@end deftypefn 9532 9533@deftypefn Macro int REAL_VALUE_ISINF (REAL_VALUE_TYPE @var{x}) 9534Determines whether @var{x} represents infinity (positive or negative). 9535@end deftypefn 9536 9537@deftypefn Macro int REAL_VALUE_ISNAN (REAL_VALUE_TYPE @var{x}) 9538Determines whether @var{x} represents a ``NaN'' (not-a-number). 9539@end deftypefn 9540 9541@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}) 9542Calculates an arithmetic operation on the two floating point values 9543@var{x} and @var{y}, storing the result in @var{output} (which must be a 9544variable). 9545 9546The operation to be performed is specified by @var{code}. Only the 9547following codes are supported: @code{PLUS_EXPR}, @code{MINUS_EXPR}, 9548@code{MULT_EXPR}, @code{RDIV_EXPR}, @code{MAX_EXPR}, @code{MIN_EXPR}. 9549 9550If @code{REAL_ARITHMETIC} is asked to evaluate division by zero and the 9551target's floating point format cannot represent infinity, it will call 9552@code{abort}. Callers should check for this situation first, using 9553@code{MODE_HAS_INFINITIES}. @xref{Storage Layout}. 9554@end deftypefn 9555 9556@deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_NEGATE (REAL_VALUE_TYPE @var{x}) 9557Returns the negative of the floating point value @var{x}. 9558@end deftypefn 9559 9560@deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ABS (REAL_VALUE_TYPE @var{x}) 9561Returns the absolute value of @var{x}. 9562@end deftypefn 9563 9564@deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_TRUNCATE (REAL_VALUE_TYPE @var{mode}, enum machine_mode @var{x}) 9565Truncates the floating point value @var{x} to fit in @var{mode}. The 9566return value is still a full-size @code{REAL_VALUE_TYPE}, but it has an 9567appropriate bit pattern to be output as a floating constant whose 9568precision accords with mode @var{mode}. 9569@end deftypefn 9570 9571@deftypefn Macro void REAL_VALUE_TO_INT (HOST_WIDE_INT @var{low}, HOST_WIDE_INT @var{high}, REAL_VALUE_TYPE @var{x}) 9572Converts a floating point value @var{x} into a double-precision integer 9573which is then stored into @var{low} and @var{high}. If the value is not 9574integral, it is truncated. 9575@end deftypefn 9576 9577@deftypefn Macro void REAL_VALUE_FROM_INT (REAL_VALUE_TYPE @var{x}, HOST_WIDE_INT @var{low}, HOST_WIDE_INT @var{high}, enum machine_mode @var{mode}) 9578Converts a double-precision integer found in @var{low} and @var{high}, 9579into a floating point value which is then stored into @var{x}. The 9580value is truncated to fit in mode @var{mode}. 9581@end deftypefn 9582 9583@node Mode Switching 9584@section Mode Switching Instructions 9585@cindex mode switching 9586The following macros control mode switching optimizations: 9587 9588@defmac OPTIMIZE_MODE_SWITCHING (@var{entity}) 9589Define this macro if the port needs extra instructions inserted for mode 9590switching in an optimizing compilation. 9591 9592For an example, the SH4 can perform both single and double precision 9593floating point operations, but to perform a single precision operation, 9594the FPSCR PR bit has to be cleared, while for a double precision 9595operation, this bit has to be set. Changing the PR bit requires a general 9596purpose register as a scratch register, hence these FPSCR sets have to 9597be inserted before reload, i.e.@: you can't put this into instruction emitting 9598or @code{TARGET_MACHINE_DEPENDENT_REORG}. 9599 9600You can have multiple entities that are mode-switched, and select at run time 9601which entities actually need it. @code{OPTIMIZE_MODE_SWITCHING} should 9602return nonzero for any @var{entity} that needs mode-switching. 9603If you define this macro, you also have to define 9604@code{NUM_MODES_FOR_MODE_SWITCHING}, @code{MODE_NEEDED}, 9605@code{MODE_PRIORITY_TO_MODE} and @code{EMIT_MODE_SET}. 9606@code{MODE_AFTER}, @code{MODE_ENTRY}, and @code{MODE_EXIT} 9607are optional. 9608@end defmac 9609 9610@defmac NUM_MODES_FOR_MODE_SWITCHING 9611If you define @code{OPTIMIZE_MODE_SWITCHING}, you have to define this as 9612initializer for an array of integers. Each initializer element 9613N refers to an entity that needs mode switching, and specifies the number 9614of different modes that might need to be set for this entity. 9615The position of the initializer in the initializer---starting counting at 9616zero---determines the integer that is used to refer to the mode-switched 9617entity in question. 9618In macros that take mode arguments / yield a mode result, modes are 9619represented as numbers 0 @dots{} N @minus{} 1. N is used to specify that no mode 9620switch is needed / supplied. 9621@end defmac 9622 9623@defmac MODE_NEEDED (@var{entity}, @var{insn}) 9624@var{entity} is an integer specifying a mode-switched entity. If 9625@code{OPTIMIZE_MODE_SWITCHING} is defined, you must define this macro to 9626return an integer value not larger than the corresponding element in 9627@code{NUM_MODES_FOR_MODE_SWITCHING}, to denote the mode that @var{entity} must 9628be switched into prior to the execution of @var{insn}. 9629@end defmac 9630 9631@defmac MODE_AFTER (@var{mode}, @var{insn}) 9632If this macro is defined, it is evaluated for every @var{insn} during 9633mode switching. It determines the mode that an insn results in (if 9634different from the incoming mode). 9635@end defmac 9636 9637@defmac MODE_ENTRY (@var{entity}) 9638If this macro is defined, it is evaluated for every @var{entity} that needs 9639mode switching. It should evaluate to an integer, which is a mode that 9640@var{entity} is assumed to be switched to at function entry. If @code{MODE_ENTRY} 9641is defined then @code{MODE_EXIT} must be defined. 9642@end defmac 9643 9644@defmac MODE_EXIT (@var{entity}) 9645If this macro is defined, it is evaluated for every @var{entity} that needs 9646mode switching. It should evaluate to an integer, which is a mode that 9647@var{entity} is assumed to be switched to at function exit. If @code{MODE_EXIT} 9648is defined then @code{MODE_ENTRY} must be defined. 9649@end defmac 9650 9651@defmac MODE_PRIORITY_TO_MODE (@var{entity}, @var{n}) 9652This macro specifies the order in which modes for @var{entity} are processed. 96530 is the highest priority, @code{NUM_MODES_FOR_MODE_SWITCHING[@var{entity}] - 1} the 9654lowest. The value of the macro should be an integer designating a mode 9655for @var{entity}. For any fixed @var{entity}, @code{mode_priority_to_mode} 9656(@var{entity}, @var{n}) shall be a bijection in 0 @dots{} 9657@code{num_modes_for_mode_switching[@var{entity}] - 1}. 9658@end defmac 9659 9660@defmac EMIT_MODE_SET (@var{entity}, @var{mode}, @var{hard_regs_live}) 9661Generate one or more insns to set @var{entity} to @var{mode}. 9662@var{hard_reg_live} is the set of hard registers live at the point where 9663the insn(s) are to be inserted. 9664@end defmac 9665 9666@node Target Attributes 9667@section Defining target-specific uses of @code{__attribute__} 9668@cindex target attributes 9669@cindex machine attributes 9670@cindex attributes, target-specific 9671 9672Target-specific attributes may be defined for functions, data and types. 9673These are described using the following target hooks; they also need to 9674be documented in @file{extend.texi}. 9675 9676@hook TARGET_ATTRIBUTE_TABLE 9677If defined, this target hook points to an array of @samp{struct 9678attribute_spec} (defined in @file{tree.h}) specifying the machine 9679specific attributes for this target and some of the restrictions on the 9680entities to which these attributes are applied and the arguments they 9681take. 9682@end deftypevr 9683 9684@hook TARGET_ATTRIBUTE_TAKES_IDENTIFIER_P 9685If defined, this target hook is a function which returns true if the 9686machine-specific attribute named @var{name} expects an identifier 9687given as its first argument to be passed on as a plain identifier, not 9688subjected to name lookup. If this is not defined, the default is 9689false for all machine-specific attributes. 9690@end deftypefn 9691 9692@hook TARGET_COMP_TYPE_ATTRIBUTES 9693If defined, this target hook is a function which returns zero if the attributes on 9694@var{type1} and @var{type2} are incompatible, one if they are compatible, 9695and two if they are nearly compatible (which causes a warning to be 9696generated). If this is not defined, machine-specific attributes are 9697supposed always to be compatible. 9698@end deftypefn 9699 9700@hook TARGET_SET_DEFAULT_TYPE_ATTRIBUTES 9701If defined, this target hook is a function which assigns default attributes to 9702the newly defined @var{type}. 9703@end deftypefn 9704 9705@hook TARGET_MERGE_TYPE_ATTRIBUTES 9706Define this target hook if the merging of type attributes needs special 9707handling. If defined, the result is a list of the combined 9708@code{TYPE_ATTRIBUTES} of @var{type1} and @var{type2}. It is assumed 9709that @code{comptypes} has already been called and returned 1. This 9710function may call @code{merge_attributes} to handle machine-independent 9711merging. 9712@end deftypefn 9713 9714@hook TARGET_MERGE_DECL_ATTRIBUTES 9715Define this target hook if the merging of decl attributes needs special 9716handling. If defined, the result is a list of the combined 9717@code{DECL_ATTRIBUTES} of @var{olddecl} and @var{newdecl}. 9718@var{newdecl} is a duplicate declaration of @var{olddecl}. Examples of 9719when this is needed are when one attribute overrides another, or when an 9720attribute is nullified by a subsequent definition. This function may 9721call @code{merge_attributes} to handle machine-independent merging. 9722 9723@findex TARGET_DLLIMPORT_DECL_ATTRIBUTES 9724If the only target-specific handling you require is @samp{dllimport} 9725for Microsoft Windows targets, you should define the macro 9726@code{TARGET_DLLIMPORT_DECL_ATTRIBUTES} to @code{1}. The compiler 9727will then define a function called 9728@code{merge_dllimport_decl_attributes} which can then be defined as 9729the expansion of @code{TARGET_MERGE_DECL_ATTRIBUTES}. You can also 9730add @code{handle_dll_attribute} in the attribute table for your port 9731to perform initial processing of the @samp{dllimport} and 9732@samp{dllexport} attributes. This is done in @file{i386/cygwin.h} and 9733@file{i386/i386.c}, for example. 9734@end deftypefn 9735 9736@hook TARGET_VALID_DLLIMPORT_ATTRIBUTE_P 9737 9738@defmac TARGET_DECLSPEC 9739Define this macro to a nonzero value if you want to treat 9740@code{__declspec(X)} as equivalent to @code{__attribute((X))}. By 9741default, this behavior is enabled only for targets that define 9742@code{TARGET_DLLIMPORT_DECL_ATTRIBUTES}. The current implementation 9743of @code{__declspec} is via a built-in macro, but you should not rely 9744on this implementation detail. 9745@end defmac 9746 9747@hook TARGET_INSERT_ATTRIBUTES 9748Define this target hook if you want to be able to add attributes to a decl 9749when it is being created. This is normally useful for back ends which 9750wish to implement a pragma by using the attributes which correspond to 9751the pragma's effect. The @var{node} argument is the decl which is being 9752created. The @var{attr_ptr} argument is a pointer to the attribute list 9753for this decl. The list itself should not be modified, since it may be 9754shared with other decls, but attributes may be chained on the head of 9755the list and @code{*@var{attr_ptr}} modified to point to the new 9756attributes, or a copy of the list may be made if further changes are 9757needed. 9758@end deftypefn 9759 9760@hook TARGET_FUNCTION_ATTRIBUTE_INLINABLE_P 9761@cindex inlining 9762This target hook returns @code{true} if it is ok to inline @var{fndecl} 9763into the current function, despite its having target-specific 9764attributes, @code{false} otherwise. By default, if a function has a 9765target specific attribute attached to it, it will not be inlined. 9766@end deftypefn 9767 9768@hook TARGET_OPTION_VALID_ATTRIBUTE_P 9769This hook is called to parse the @code{attribute(option("..."))}, and 9770it allows the function to set different target machine compile time 9771options for the current function that might be different than the 9772options specified on the command line. The hook should return 9773@code{true} if the options are valid. 9774 9775The hook should set the @var{DECL_FUNCTION_SPECIFIC_TARGET} field in 9776the function declaration to hold a pointer to a target specific 9777@var{struct cl_target_option} structure. 9778@end deftypefn 9779 9780@hook TARGET_OPTION_SAVE 9781This hook is called to save any additional target specific information 9782in the @var{struct cl_target_option} structure for function specific 9783options. 9784@xref{Option file format}. 9785@end deftypefn 9786 9787@hook TARGET_OPTION_RESTORE 9788This hook is called to restore any additional target specific 9789information in the @var{struct cl_target_option} structure for 9790function specific options. 9791@end deftypefn 9792 9793@hook TARGET_OPTION_PRINT 9794This hook is called to print any additional target specific 9795information in the @var{struct cl_target_option} structure for 9796function specific options. 9797@end deftypefn 9798 9799@hook TARGET_OPTION_PRAGMA_PARSE 9800This target hook parses the options for @code{#pragma GCC option} to 9801set the machine specific options for functions that occur later in the 9802input stream. The options should be the same as handled by the 9803@code{TARGET_OPTION_VALID_ATTRIBUTE_P} hook. 9804@end deftypefn 9805 9806@hook TARGET_OPTION_OVERRIDE 9807Sometimes certain combinations of command options do not make sense on 9808a particular target machine. You can override the hook 9809@code{TARGET_OPTION_OVERRIDE} to take account of this. This hooks is called 9810once just after all the command options have been parsed. 9811 9812Don't use this hook to turn on various extra optimizations for 9813@option{-O}. That is what @code{TARGET_OPTION_OPTIMIZATION} is for. 9814 9815If you need to do something whenever the optimization level is 9816changed via the optimize attribute or pragma, see 9817@code{TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE} 9818@end deftypefn 9819 9820@hook TARGET_CAN_INLINE_P 9821This target hook returns @code{false} if the @var{caller} function 9822cannot inline @var{callee}, based on target specific information. By 9823default, inlining is not allowed if the callee function has function 9824specific target options and the caller does not use the same options. 9825@end deftypefn 9826 9827@node Emulated TLS 9828@section Emulating TLS 9829@cindex Emulated TLS 9830 9831For targets whose psABI does not provide Thread Local Storage via 9832specific relocations and instruction sequences, an emulation layer is 9833used. A set of target hooks allows this emulation layer to be 9834configured for the requirements of a particular target. For instance 9835the psABI may in fact specify TLS support in terms of an emulation 9836layer. 9837 9838The emulation layer works by creating a control object for every TLS 9839object. To access the TLS object, a lookup function is provided 9840which, when given the address of the control object, will return the 9841address of the current thread's instance of the TLS object. 9842 9843@hook TARGET_EMUTLS_GET_ADDRESS 9844Contains the name of the helper function that uses a TLS control 9845object to locate a TLS instance. The default causes libgcc's 9846emulated TLS helper function to be used. 9847@end deftypevr 9848 9849@hook TARGET_EMUTLS_REGISTER_COMMON 9850Contains the name of the helper function that should be used at 9851program startup to register TLS objects that are implicitly 9852initialized to zero. If this is @code{NULL}, all TLS objects will 9853have explicit initializers. The default causes libgcc's emulated TLS 9854registration function to be used. 9855@end deftypevr 9856 9857@hook TARGET_EMUTLS_VAR_SECTION 9858Contains the name of the section in which TLS control variables should 9859be placed. The default of @code{NULL} allows these to be placed in 9860any section. 9861@end deftypevr 9862 9863@hook TARGET_EMUTLS_TMPL_SECTION 9864Contains the name of the section in which TLS initializers should be 9865placed. The default of @code{NULL} allows these to be placed in any 9866section. 9867@end deftypevr 9868 9869@hook TARGET_EMUTLS_VAR_PREFIX 9870Contains the prefix to be prepended to TLS control variable names. 9871The default of @code{NULL} uses a target-specific prefix. 9872@end deftypevr 9873 9874@hook TARGET_EMUTLS_TMPL_PREFIX 9875Contains the prefix to be prepended to TLS initializer objects. The 9876default of @code{NULL} uses a target-specific prefix. 9877@end deftypevr 9878 9879@hook TARGET_EMUTLS_VAR_FIELDS 9880Specifies a function that generates the FIELD_DECLs for a TLS control 9881object type. @var{type} is the RECORD_TYPE the fields are for and 9882@var{name} should be filled with the structure tag, if the default of 9883@code{__emutls_object} is unsuitable. The default creates a type suitable 9884for libgcc's emulated TLS function. 9885@end deftypefn 9886 9887@hook TARGET_EMUTLS_VAR_INIT 9888Specifies a function that generates the CONSTRUCTOR to initialize a 9889TLS control object. @var{var} is the TLS control object, @var{decl} 9890is the TLS object and @var{tmpl_addr} is the address of the 9891initializer. The default initializes libgcc's emulated TLS control object. 9892@end deftypefn 9893 9894@hook TARGET_EMUTLS_VAR_ALIGN_FIXED 9895Specifies whether the alignment of TLS control variable objects is 9896fixed and should not be increased as some backends may do to optimize 9897single objects. The default is false. 9898@end deftypevr 9899 9900@hook TARGET_EMUTLS_DEBUG_FORM_TLS_ADDRESS 9901Specifies whether a DWARF @code{DW_OP_form_tls_address} location descriptor 9902may be used to describe emulated TLS control objects. 9903@end deftypevr 9904 9905@node MIPS Coprocessors 9906@section Defining coprocessor specifics for MIPS targets. 9907@cindex MIPS coprocessor-definition macros 9908 9909The MIPS specification allows MIPS implementations to have as many as 4 9910coprocessors, each with as many as 32 private registers. GCC supports 9911accessing these registers and transferring values between the registers 9912and memory using asm-ized variables. For example: 9913 9914@smallexample 9915 register unsigned int cp0count asm ("c0r1"); 9916 unsigned int d; 9917 9918 d = cp0count + 3; 9919@end smallexample 9920 9921(``c0r1'' is the default name of register 1 in coprocessor 0; alternate 9922names may be added as described below, or the default names may be 9923overridden entirely in @code{SUBTARGET_CONDITIONAL_REGISTER_USAGE}.) 9924 9925Coprocessor registers are assumed to be epilogue-used; sets to them will 9926be preserved even if it does not appear that the register is used again 9927later in the function. 9928 9929Another note: according to the MIPS spec, coprocessor 1 (if present) is 9930the FPU@. One accesses COP1 registers through standard mips 9931floating-point support; they are not included in this mechanism. 9932 9933There is one macro used in defining the MIPS coprocessor interface which 9934you may want to override in subtargets; it is described below. 9935 9936@defmac ALL_COP_ADDITIONAL_REGISTER_NAMES 9937A comma-separated list (with leading comma) of pairs describing the 9938alternate names of coprocessor registers. The format of each entry should be 9939@smallexample 9940@{ @var{alternatename}, @var{register_number}@} 9941@end smallexample 9942Default: empty. 9943@end defmac 9944 9945@node PCH Target 9946@section Parameters for Precompiled Header Validity Checking 9947@cindex parameters, precompiled headers 9948 9949@hook TARGET_GET_PCH_VALIDITY 9950This hook returns a pointer to the data needed by 9951@code{TARGET_PCH_VALID_P} and sets 9952@samp{*@var{sz}} to the size of the data in bytes. 9953@end deftypefn 9954 9955@hook TARGET_PCH_VALID_P 9956This hook checks whether the options used to create a PCH file are 9957compatible with the current settings. It returns @code{NULL} 9958if so and a suitable error message if not. Error messages will 9959be presented to the user and must be localized using @samp{_(@var{msg})}. 9960 9961@var{data} is the data that was returned by @code{TARGET_GET_PCH_VALIDITY} 9962when the PCH file was created and @var{sz} is the size of that data in bytes. 9963It's safe to assume that the data was created by the same version of the 9964compiler, so no format checking is needed. 9965 9966The default definition of @code{default_pch_valid_p} should be 9967suitable for most targets. 9968@end deftypefn 9969 9970@hook TARGET_CHECK_PCH_TARGET_FLAGS 9971If this hook is nonnull, the default implementation of 9972@code{TARGET_PCH_VALID_P} will use it to check for compatible values 9973of @code{target_flags}. @var{pch_flags} specifies the value that 9974@code{target_flags} had when the PCH file was created. The return 9975value is the same as for @code{TARGET_PCH_VALID_P}. 9976@end deftypefn 9977 9978@hook TARGET_PREPARE_PCH_SAVE 9979 9980@node C++ ABI 9981@section C++ ABI parameters 9982@cindex parameters, c++ abi 9983 9984@hook TARGET_CXX_GUARD_TYPE 9985Define this hook to override the integer type used for guard variables. 9986These are used to implement one-time construction of static objects. The 9987default is long_long_integer_type_node. 9988@end deftypefn 9989 9990@hook TARGET_CXX_GUARD_MASK_BIT 9991This hook determines how guard variables are used. It should return 9992@code{false} (the default) if the first byte should be used. A return value of 9993@code{true} indicates that only the least significant bit should be used. 9994@end deftypefn 9995 9996@hook TARGET_CXX_GET_COOKIE_SIZE 9997This hook returns the size of the cookie to use when allocating an array 9998whose elements have the indicated @var{type}. Assumes that it is already 9999known that a cookie is needed. The default is 10000@code{max(sizeof (size_t), alignof(type))}, as defined in section 2.7 of the 10001IA64/Generic C++ ABI@. 10002@end deftypefn 10003 10004@hook TARGET_CXX_COOKIE_HAS_SIZE 10005This hook should return @code{true} if the element size should be stored in 10006array cookies. The default is to return @code{false}. 10007@end deftypefn 10008 10009@hook TARGET_CXX_IMPORT_EXPORT_CLASS 10010If defined by a backend this hook allows the decision made to export 10011class @var{type} to be overruled. Upon entry @var{import_export} 10012will contain 1 if the class is going to be exported, @minus{}1 if it is going 10013to be imported and 0 otherwise. This function should return the 10014modified value and perform any other actions necessary to support the 10015backend's targeted operating system. 10016@end deftypefn 10017 10018@hook TARGET_CXX_CDTOR_RETURNS_THIS 10019This hook should return @code{true} if constructors and destructors return 10020the address of the object created/destroyed. The default is to return 10021@code{false}. 10022@end deftypefn 10023 10024@hook TARGET_CXX_KEY_METHOD_MAY_BE_INLINE 10025This hook returns true if the key method for a class (i.e., the method 10026which, if defined in the current translation unit, causes the virtual 10027table to be emitted) may be an inline function. Under the standard 10028Itanium C++ ABI the key method may be an inline function so long as 10029the function is not declared inline in the class definition. Under 10030some variants of the ABI, an inline function can never be the key 10031method. The default is to return @code{true}. 10032@end deftypefn 10033 10034@hook TARGET_CXX_DETERMINE_CLASS_DATA_VISIBILITY 10035 10036@hook TARGET_CXX_CLASS_DATA_ALWAYS_COMDAT 10037This hook returns true (the default) if virtual tables and other 10038similar implicit class data objects are always COMDAT if they have 10039external linkage. If this hook returns false, then class data for 10040classes whose virtual table will be emitted in only one translation 10041unit will not be COMDAT. 10042@end deftypefn 10043 10044@hook TARGET_CXX_LIBRARY_RTTI_COMDAT 10045This hook returns true (the default) if the RTTI information for 10046the basic types which is defined in the C++ runtime should always 10047be COMDAT, false if it should not be COMDAT. 10048@end deftypefn 10049 10050@hook TARGET_CXX_USE_AEABI_ATEXIT 10051This hook returns true if @code{__aeabi_atexit} (as defined by the ARM EABI) 10052should be used to register static destructors when @option{-fuse-cxa-atexit} 10053is in effect. The default is to return false to use @code{__cxa_atexit}. 10054@end deftypefn 10055 10056@hook TARGET_CXX_USE_ATEXIT_FOR_CXA_ATEXIT 10057This hook returns true if the target @code{atexit} function can be used 10058in the same manner as @code{__cxa_atexit} to register C++ static 10059destructors. This requires that @code{atexit}-registered functions in 10060shared libraries are run in the correct order when the libraries are 10061unloaded. The default is to return false. 10062@end deftypefn 10063 10064@hook TARGET_CXX_ADJUST_CLASS_AT_DEFINITION 10065 10066@node Named Address Spaces 10067@section Adding support for named address spaces 10068@cindex named address spaces 10069 10070The draft technical report of the ISO/IEC JTC1 S22 WG14 N1275 10071standards committee, @cite{Programming Languages - C - Extensions to 10072support embedded processors}, specifies a syntax for embedded 10073processors to specify alternate address spaces. You can configure a 10074GCC port to support section 5.1 of the draft report to add support for 10075address spaces other than the default address space. These address 10076spaces are new keywords that are similar to the @code{volatile} and 10077@code{const} type attributes. 10078 10079Pointers to named address spaces can have a different size than 10080pointers to the generic address space. 10081 10082For example, the SPU port uses the @code{__ea} address space to refer 10083to memory in the host processor, rather than memory local to the SPU 10084processor. Access to memory in the @code{__ea} address space involves 10085issuing DMA operations to move data between the host processor and the 10086local processor memory address space. Pointers in the @code{__ea} 10087address space are either 32 bits or 64 bits based on the 10088@option{-mea32} or @option{-mea64} switches (native SPU pointers are 10089always 32 bits). 10090 10091Internally, address spaces are represented as a small integer in the 10092range 0 to 15 with address space 0 being reserved for the generic 10093address space. 10094 10095To register a named address space qualifier keyword with the C front end, 10096the target may call the @code{c_register_addr_space} routine. For example, 10097the SPU port uses the following to declare @code{__ea} as the keyword for 10098named address space #1: 10099@smallexample 10100#define ADDR_SPACE_EA 1 10101c_register_addr_space ("__ea", ADDR_SPACE_EA); 10102@end smallexample 10103 10104@hook TARGET_ADDR_SPACE_POINTER_MODE 10105Define this to return the machine mode to use for pointers to 10106@var{address_space} if the target supports named address spaces. 10107The default version of this hook returns @code{ptr_mode} for the 10108generic address space only. 10109@end deftypefn 10110 10111@hook TARGET_ADDR_SPACE_ADDRESS_MODE 10112Define this to return the machine mode to use for addresses in 10113@var{address_space} if the target supports named address spaces. 10114The default version of this hook returns @code{Pmode} for the 10115generic address space only. 10116@end deftypefn 10117 10118@hook TARGET_ADDR_SPACE_VALID_POINTER_MODE 10119Define this to return nonzero if the port can handle pointers 10120with machine mode @var{mode} to address space @var{as}. This target 10121hook is the same as the @code{TARGET_VALID_POINTER_MODE} target hook, 10122except that it includes explicit named address space support. The default 10123version of this hook returns true for the modes returned by either the 10124@code{TARGET_ADDR_SPACE_POINTER_MODE} or @code{TARGET_ADDR_SPACE_ADDRESS_MODE} 10125target hooks for the given address space. 10126@end deftypefn 10127 10128@hook TARGET_ADDR_SPACE_LEGITIMATE_ADDRESS_P 10129Define this to return true if @var{exp} is a valid address for mode 10130@var{mode} in the named address space @var{as}. The @var{strict} 10131parameter says whether strict addressing is in effect after reload has 10132finished. This target hook is the same as the 10133@code{TARGET_LEGITIMATE_ADDRESS_P} target hook, except that it includes 10134explicit named address space support. 10135@end deftypefn 10136 10137@hook TARGET_ADDR_SPACE_LEGITIMIZE_ADDRESS 10138Define this to modify an invalid address @var{x} to be a valid address 10139with mode @var{mode} in the named address space @var{as}. This target 10140hook is the same as the @code{TARGET_LEGITIMIZE_ADDRESS} target hook, 10141except that it includes explicit named address space support. 10142@end deftypefn 10143 10144@hook TARGET_ADDR_SPACE_SUBSET_P 10145Define this to return whether the @var{subset} named address space is 10146contained within the @var{superset} named address space. Pointers to 10147a named address space that is a subset of another named address space 10148will be converted automatically without a cast if used together in 10149arithmetic operations. Pointers to a superset address space can be 10150converted to pointers to a subset address space via explicit casts. 10151@end deftypefn 10152 10153@hook TARGET_ADDR_SPACE_CONVERT 10154Define this to convert the pointer expression represented by the RTL 10155@var{op} with type @var{from_type} that points to a named address 10156space to a new pointer expression with type @var{to_type} that points 10157to a different named address space. When this hook it called, it is 10158guaranteed that one of the two address spaces is a subset of the other, 10159as determined by the @code{TARGET_ADDR_SPACE_SUBSET_P} target hook. 10160@end deftypefn 10161 10162@node Misc 10163@section Miscellaneous Parameters 10164@cindex parameters, miscellaneous 10165 10166@c prevent bad page break with this line 10167Here are several miscellaneous parameters. 10168 10169@defmac HAS_LONG_COND_BRANCH 10170Define this boolean macro to indicate whether or not your architecture 10171has conditional branches that can span all of memory. It is used in 10172conjunction with an optimization that partitions hot and cold basic 10173blocks into separate sections of the executable. If this macro is 10174set to false, gcc will convert any conditional branches that attempt 10175to cross between sections into unconditional branches or indirect jumps. 10176@end defmac 10177 10178@defmac HAS_LONG_UNCOND_BRANCH 10179Define this boolean macro to indicate whether or not your architecture 10180has unconditional branches that can span all of memory. It is used in 10181conjunction with an optimization that partitions hot and cold basic 10182blocks into separate sections of the executable. If this macro is 10183set to false, gcc will convert any unconditional branches that attempt 10184to cross between sections into indirect jumps. 10185@end defmac 10186 10187@defmac CASE_VECTOR_MODE 10188An alias for a machine mode name. This is the machine mode that 10189elements of a jump-table should have. 10190@end defmac 10191 10192@defmac CASE_VECTOR_SHORTEN_MODE (@var{min_offset}, @var{max_offset}, @var{body}) 10193Optional: return the preferred mode for an @code{addr_diff_vec} 10194when the minimum and maximum offset are known. If you define this, 10195it enables extra code in branch shortening to deal with @code{addr_diff_vec}. 10196To make this work, you also have to define @code{INSN_ALIGN} and 10197make the alignment for @code{addr_diff_vec} explicit. 10198The @var{body} argument is provided so that the offset_unsigned and scale 10199flags can be updated. 10200@end defmac 10201 10202@defmac CASE_VECTOR_PC_RELATIVE 10203Define this macro to be a C expression to indicate when jump-tables 10204should contain relative addresses. You need not define this macro if 10205jump-tables never contain relative addresses, or jump-tables should 10206contain relative addresses only when @option{-fPIC} or @option{-fPIC} 10207is in effect. 10208@end defmac 10209 10210@hook TARGET_CASE_VALUES_THRESHOLD 10211This function return the smallest number of different values for which it 10212is best to use a jump-table instead of a tree of conditional branches. 10213The default is four for machines with a @code{casesi} instruction and 10214five otherwise. This is best for most machines. 10215@end deftypefn 10216 10217@defmac CASE_USE_BIT_TESTS 10218Define this macro to be a C expression to indicate whether C switch 10219statements may be implemented by a sequence of bit tests. This is 10220advantageous on processors that can efficiently implement left shift 10221of 1 by the number of bits held in a register, but inappropriate on 10222targets that would require a loop. By default, this macro returns 10223@code{true} if the target defines an @code{ashlsi3} pattern, and 10224@code{false} otherwise. 10225@end defmac 10226 10227@defmac WORD_REGISTER_OPERATIONS 10228Define this macro if operations between registers with integral mode 10229smaller than a word are always performed on the entire register. 10230Most RISC machines have this property and most CISC machines do not. 10231@end defmac 10232 10233@defmac LOAD_EXTEND_OP (@var{mem_mode}) 10234Define this macro to be a C expression indicating when insns that read 10235memory in @var{mem_mode}, an integral mode narrower than a word, set the 10236bits outside of @var{mem_mode} to be either the sign-extension or the 10237zero-extension of the data read. Return @code{SIGN_EXTEND} for values 10238of @var{mem_mode} for which the 10239insn sign-extends, @code{ZERO_EXTEND} for which it zero-extends, and 10240@code{UNKNOWN} for other modes. 10241 10242This macro is not called with @var{mem_mode} non-integral or with a width 10243greater than or equal to @code{BITS_PER_WORD}, so you may return any 10244value in this case. Do not define this macro if it would always return 10245@code{UNKNOWN}. On machines where this macro is defined, you will normally 10246define it as the constant @code{SIGN_EXTEND} or @code{ZERO_EXTEND}. 10247 10248You may return a non-@code{UNKNOWN} value even if for some hard registers 10249the sign extension is not performed, if for the @code{REGNO_REG_CLASS} 10250of these hard registers @code{CANNOT_CHANGE_MODE_CLASS} returns nonzero 10251when the @var{from} mode is @var{mem_mode} and the @var{to} mode is any 10252integral mode larger than this but not larger than @code{word_mode}. 10253 10254You must return @code{UNKNOWN} if for some hard registers that allow this 10255mode, @code{CANNOT_CHANGE_MODE_CLASS} says that they cannot change to 10256@code{word_mode}, but that they can change to another integral mode that 10257is larger then @var{mem_mode} but still smaller than @code{word_mode}. 10258@end defmac 10259 10260@defmac SHORT_IMMEDIATES_SIGN_EXTEND 10261Define this macro if loading short immediate values into registers sign 10262extends. 10263@end defmac 10264 10265@defmac FIXUNS_TRUNC_LIKE_FIX_TRUNC 10266Define this macro if the same instructions that convert a floating 10267point number to a signed fixed point number also convert validly to an 10268unsigned one. 10269@end defmac 10270 10271@hook TARGET_MIN_DIVISIONS_FOR_RECIP_MUL 10272When @option{-ffast-math} is in effect, GCC tries to optimize 10273divisions by the same divisor, by turning them into multiplications by 10274the reciprocal. This target hook specifies the minimum number of divisions 10275that should be there for GCC to perform the optimization for a variable 10276of mode @var{mode}. The default implementation returns 3 if the machine 10277has an instruction for the division, and 2 if it does not. 10278@end deftypefn 10279 10280@defmac MOVE_MAX 10281The maximum number of bytes that a single instruction can move quickly 10282between memory and registers or between two memory locations. 10283@end defmac 10284 10285@defmac MAX_MOVE_MAX 10286The maximum number of bytes that a single instruction can move quickly 10287between memory and registers or between two memory locations. If this 10288is undefined, the default is @code{MOVE_MAX}. Otherwise, it is the 10289constant value that is the largest value that @code{MOVE_MAX} can have 10290at run-time. 10291@end defmac 10292 10293@defmac SHIFT_COUNT_TRUNCATED 10294A C expression that is nonzero if on this machine the number of bits 10295actually used for the count of a shift operation is equal to the number 10296of bits needed to represent the size of the object being shifted. When 10297this macro is nonzero, the compiler will assume that it is safe to omit 10298a sign-extend, zero-extend, and certain bitwise `and' instructions that 10299truncates the count of a shift operation. On machines that have 10300instructions that act on bit-fields at variable positions, which may 10301include `bit test' instructions, a nonzero @code{SHIFT_COUNT_TRUNCATED} 10302also enables deletion of truncations of the values that serve as 10303arguments to bit-field instructions. 10304 10305If both types of instructions truncate the count (for shifts) and 10306position (for bit-field operations), or if no variable-position bit-field 10307instructions exist, you should define this macro. 10308 10309However, on some machines, such as the 80386 and the 680x0, truncation 10310only applies to shift operations and not the (real or pretended) 10311bit-field operations. Define @code{SHIFT_COUNT_TRUNCATED} to be zero on 10312such machines. Instead, add patterns to the @file{md} file that include 10313the implied truncation of the shift instructions. 10314 10315You need not define this macro if it would always have the value of zero. 10316@end defmac 10317 10318@anchor{TARGET_SHIFT_TRUNCATION_MASK} 10319@hook TARGET_SHIFT_TRUNCATION_MASK 10320This function describes how the standard shift patterns for @var{mode} 10321deal with shifts by negative amounts or by more than the width of the mode. 10322@xref{shift patterns}. 10323 10324On many machines, the shift patterns will apply a mask @var{m} to the 10325shift count, meaning that a fixed-width shift of @var{x} by @var{y} is 10326equivalent to an arbitrary-width shift of @var{x} by @var{y & m}. If 10327this is true for mode @var{mode}, the function should return @var{m}, 10328otherwise it should return 0. A return value of 0 indicates that no 10329particular behavior is guaranteed. 10330 10331Note that, unlike @code{SHIFT_COUNT_TRUNCATED}, this function does 10332@emph{not} apply to general shift rtxes; it applies only to instructions 10333that are generated by the named shift patterns. 10334 10335The default implementation of this function returns 10336@code{GET_MODE_BITSIZE (@var{mode}) - 1} if @code{SHIFT_COUNT_TRUNCATED} 10337and 0 otherwise. This definition is always safe, but if 10338@code{SHIFT_COUNT_TRUNCATED} is false, and some shift patterns 10339nevertheless truncate the shift count, you may get better code 10340by overriding it. 10341@end deftypefn 10342 10343@defmac TRULY_NOOP_TRUNCATION (@var{outprec}, @var{inprec}) 10344A C expression which is nonzero if on this machine it is safe to 10345``convert'' an integer of @var{inprec} bits to one of @var{outprec} 10346bits (where @var{outprec} is smaller than @var{inprec}) by merely 10347operating on it as if it had only @var{outprec} bits. 10348 10349On many machines, this expression can be 1. 10350 10351@c rearranged this, removed the phrase "it is reported that". this was 10352@c to fix an overfull hbox. --mew 10feb93 10353When @code{TRULY_NOOP_TRUNCATION} returns 1 for a pair of sizes for 10354modes for which @code{MODES_TIEABLE_P} is 0, suboptimal code can result. 10355If this is the case, making @code{TRULY_NOOP_TRUNCATION} return 0 in 10356such cases may improve things. 10357@end defmac 10358 10359@hook TARGET_MODE_REP_EXTENDED 10360The representation of an integral mode can be such that the values 10361are always extended to a wider integral mode. Return 10362@code{SIGN_EXTEND} if values of @var{mode} are represented in 10363sign-extended form to @var{rep_mode}. Return @code{UNKNOWN} 10364otherwise. (Currently, none of the targets use zero-extended 10365representation this way so unlike @code{LOAD_EXTEND_OP}, 10366@code{TARGET_MODE_REP_EXTENDED} is expected to return either 10367@code{SIGN_EXTEND} or @code{UNKNOWN}. Also no target extends 10368@var{mode} to @var{rep_mode} so that @var{rep_mode} is not the next 10369widest integral mode and currently we take advantage of this fact.) 10370 10371Similarly to @code{LOAD_EXTEND_OP} you may return a non-@code{UNKNOWN} 10372value even if the extension is not performed on certain hard registers 10373as long as for the @code{REGNO_REG_CLASS} of these hard registers 10374@code{CANNOT_CHANGE_MODE_CLASS} returns nonzero. 10375 10376Note that @code{TARGET_MODE_REP_EXTENDED} and @code{LOAD_EXTEND_OP} 10377describe two related properties. If you define 10378@code{TARGET_MODE_REP_EXTENDED (mode, word_mode)} you probably also want 10379to define @code{LOAD_EXTEND_OP (mode)} to return the same type of 10380extension. 10381 10382In order to enforce the representation of @code{mode}, 10383@code{TRULY_NOOP_TRUNCATION} should return false when truncating to 10384@code{mode}. 10385@end deftypefn 10386 10387@defmac STORE_FLAG_VALUE 10388A C expression describing the value returned by a comparison operator 10389with an integral mode and stored by a store-flag instruction 10390(@samp{cstore@var{mode}4}) when the condition is true. This description must 10391apply to @emph{all} the @samp{cstore@var{mode}4} patterns and all the 10392comparison operators whose results have a @code{MODE_INT} mode. 10393 10394A value of 1 or @minus{}1 means that the instruction implementing the 10395comparison operator returns exactly 1 or @minus{}1 when the comparison is true 10396and 0 when the comparison is false. Otherwise, the value indicates 10397which bits of the result are guaranteed to be 1 when the comparison is 10398true. This value is interpreted in the mode of the comparison 10399operation, which is given by the mode of the first operand in the 10400@samp{cstore@var{mode}4} pattern. Either the low bit or the sign bit of 10401@code{STORE_FLAG_VALUE} be on. Presently, only those bits are used by 10402the compiler. 10403 10404If @code{STORE_FLAG_VALUE} is neither 1 or @minus{}1, the compiler will 10405generate code that depends only on the specified bits. It can also 10406replace comparison operators with equivalent operations if they cause 10407the required bits to be set, even if the remaining bits are undefined. 10408For example, on a machine whose comparison operators return an 10409@code{SImode} value and where @code{STORE_FLAG_VALUE} is defined as 10410@samp{0x80000000}, saying that just the sign bit is relevant, the 10411expression 10412 10413@smallexample 10414(ne:SI (and:SI @var{x} (const_int @var{power-of-2})) (const_int 0)) 10415@end smallexample 10416 10417@noindent 10418can be converted to 10419 10420@smallexample 10421(ashift:SI @var{x} (const_int @var{n})) 10422@end smallexample 10423 10424@noindent 10425where @var{n} is the appropriate shift count to move the bit being 10426tested into the sign bit. 10427 10428There is no way to describe a machine that always sets the low-order bit 10429for a true value, but does not guarantee the value of any other bits, 10430but we do not know of any machine that has such an instruction. If you 10431are trying to port GCC to such a machine, include an instruction to 10432perform a logical-and of the result with 1 in the pattern for the 10433comparison operators and let us know at @email{gcc@@gcc.gnu.org}. 10434 10435Often, a machine will have multiple instructions that obtain a value 10436from a comparison (or the condition codes). Here are rules to guide the 10437choice of value for @code{STORE_FLAG_VALUE}, and hence the instructions 10438to be used: 10439 10440@itemize @bullet 10441@item 10442Use the shortest sequence that yields a valid definition for 10443@code{STORE_FLAG_VALUE}. It is more efficient for the compiler to 10444``normalize'' the value (convert it to, e.g., 1 or 0) than for the 10445comparison operators to do so because there may be opportunities to 10446combine the normalization with other operations. 10447 10448@item 10449For equal-length sequences, use a value of 1 or @minus{}1, with @minus{}1 being 10450slightly preferred on machines with expensive jumps and 1 preferred on 10451other machines. 10452 10453@item 10454As a second choice, choose a value of @samp{0x80000001} if instructions 10455exist that set both the sign and low-order bits but do not define the 10456others. 10457 10458@item 10459Otherwise, use a value of @samp{0x80000000}. 10460@end itemize 10461 10462Many machines can produce both the value chosen for 10463@code{STORE_FLAG_VALUE} and its negation in the same number of 10464instructions. On those machines, you should also define a pattern for 10465those cases, e.g., one matching 10466 10467@smallexample 10468(set @var{A} (neg:@var{m} (ne:@var{m} @var{B} @var{C}))) 10469@end smallexample 10470 10471Some machines can also perform @code{and} or @code{plus} operations on 10472condition code values with less instructions than the corresponding 10473@samp{cstore@var{mode}4} insn followed by @code{and} or @code{plus}. On those 10474machines, define the appropriate patterns. Use the names @code{incscc} 10475and @code{decscc}, respectively, for the patterns which perform 10476@code{plus} or @code{minus} operations on condition code values. See 10477@file{rs6000.md} for some examples. The GNU Superoptimizer can be used to 10478find such instruction sequences on other machines. 10479 10480If this macro is not defined, the default value, 1, is used. You need 10481not define @code{STORE_FLAG_VALUE} if the machine has no store-flag 10482instructions, or if the value generated by these instructions is 1. 10483@end defmac 10484 10485@defmac FLOAT_STORE_FLAG_VALUE (@var{mode}) 10486A C expression that gives a nonzero @code{REAL_VALUE_TYPE} value that is 10487returned when comparison operators with floating-point results are true. 10488Define this macro on machines that have comparison operations that return 10489floating-point values. If there are no such operations, do not define 10490this macro. 10491@end defmac 10492 10493@defmac VECTOR_STORE_FLAG_VALUE (@var{mode}) 10494A C expression that gives a rtx representing the nonzero true element 10495for vector comparisons. The returned rtx should be valid for the inner 10496mode of @var{mode} which is guaranteed to be a vector mode. Define 10497this macro on machines that have vector comparison operations that 10498return a vector result. If there are no such operations, do not define 10499this macro. Typically, this macro is defined as @code{const1_rtx} or 10500@code{constm1_rtx}. This macro may return @code{NULL_RTX} to prevent 10501the compiler optimizing such vector comparison operations for the 10502given mode. 10503@end defmac 10504 10505@defmac CLZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value}) 10506@defmacx CTZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value}) 10507A C expression that indicates whether the architecture defines a value 10508for @code{clz} or @code{ctz} with a zero operand. 10509A result of @code{0} indicates the value is undefined. 10510If the value is defined for only the RTL expression, the macro should 10511evaluate to @code{1}; if the value applies also to the corresponding optab 10512entry (which is normally the case if it expands directly into 10513the corresponding RTL), then the macro should evaluate to @code{2}. 10514In the cases where the value is defined, @var{value} should be set to 10515this value. 10516 10517If this macro is not defined, the value of @code{clz} or 10518@code{ctz} at zero is assumed to be undefined. 10519 10520This macro must be defined if the target's expansion for @code{ffs} 10521relies on a particular value to get correct results. Otherwise it 10522is not necessary, though it may be used to optimize some corner cases, and 10523to provide a default expansion for the @code{ffs} optab. 10524 10525Note that regardless of this macro the ``definedness'' of @code{clz} 10526and @code{ctz} at zero do @emph{not} extend to the builtin functions 10527visible to the user. Thus one may be free to adjust the value at will 10528to match the target expansion of these operations without fear of 10529breaking the API@. 10530@end defmac 10531 10532@defmac Pmode 10533An alias for the machine mode for pointers. On most machines, define 10534this to be the integer mode corresponding to the width of a hardware 10535pointer; @code{SImode} on 32-bit machine or @code{DImode} on 64-bit machines. 10536On some machines you must define this to be one of the partial integer 10537modes, such as @code{PSImode}. 10538 10539The width of @code{Pmode} must be at least as large as the value of 10540@code{POINTER_SIZE}. If it is not equal, you must define the macro 10541@code{POINTERS_EXTEND_UNSIGNED} to specify how pointers are extended 10542to @code{Pmode}. 10543@end defmac 10544 10545@defmac FUNCTION_MODE 10546An alias for the machine mode used for memory references to functions 10547being called, in @code{call} RTL expressions. On most CISC machines, 10548where an instruction can begin at any byte address, this should be 10549@code{QImode}. On most RISC machines, where all instructions have fixed 10550size and alignment, this should be a mode with the same size and alignment 10551as the machine instruction words - typically @code{SImode} or @code{HImode}. 10552@end defmac 10553 10554@defmac STDC_0_IN_SYSTEM_HEADERS 10555In normal operation, the preprocessor expands @code{__STDC__} to the 10556constant 1, to signify that GCC conforms to ISO Standard C@. On some 10557hosts, like Solaris, the system compiler uses a different convention, 10558where @code{__STDC__} is normally 0, but is 1 if the user specifies 10559strict conformance to the C Standard. 10560 10561Defining @code{STDC_0_IN_SYSTEM_HEADERS} makes GNU CPP follows the host 10562convention when processing system header files, but when processing user 10563files @code{__STDC__} will always expand to 1. 10564@end defmac 10565 10566@defmac NO_IMPLICIT_EXTERN_C 10567Define this macro if the system header files support C++ as well as C@. 10568This macro inhibits the usual method of using system header files in 10569C++, which is to pretend that the file's contents are enclosed in 10570@samp{extern "C" @{@dots{}@}}. 10571@end defmac 10572 10573@findex #pragma 10574@findex pragma 10575@defmac REGISTER_TARGET_PRAGMAS () 10576Define this macro if you want to implement any target-specific pragmas. 10577If defined, it is a C expression which makes a series of calls to 10578@code{c_register_pragma} or @code{c_register_pragma_with_expansion} 10579for each pragma. The macro may also do any 10580setup required for the pragmas. 10581 10582The primary reason to define this macro is to provide compatibility with 10583other compilers for the same target. In general, we discourage 10584definition of target-specific pragmas for GCC@. 10585 10586If the pragma can be implemented by attributes then you should consider 10587defining the target hook @samp{TARGET_INSERT_ATTRIBUTES} as well. 10588 10589Preprocessor macros that appear on pragma lines are not expanded. All 10590@samp{#pragma} directives that do not match any registered pragma are 10591silently ignored, unless the user specifies @option{-Wunknown-pragmas}. 10592@end defmac 10593 10594@deftypefun void c_register_pragma (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *)) 10595@deftypefunx void c_register_pragma_with_expansion (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *)) 10596 10597Each call to @code{c_register_pragma} or 10598@code{c_register_pragma_with_expansion} establishes one pragma. The 10599@var{callback} routine will be called when the preprocessor encounters a 10600pragma of the form 10601 10602@smallexample 10603#pragma [@var{space}] @var{name} @dots{} 10604@end smallexample 10605 10606@var{space} is the case-sensitive namespace of the pragma, or 10607@code{NULL} to put the pragma in the global namespace. The callback 10608routine receives @var{pfile} as its first argument, which can be passed 10609on to cpplib's functions if necessary. You can lex tokens after the 10610@var{name} by calling @code{pragma_lex}. Tokens that are not read by the 10611callback will be silently ignored. The end of the line is indicated by 10612a token of type @code{CPP_EOF}. Macro expansion occurs on the 10613arguments of pragmas registered with 10614@code{c_register_pragma_with_expansion} but not on the arguments of 10615pragmas registered with @code{c_register_pragma}. 10616 10617Note that the use of @code{pragma_lex} is specific to the C and C++ 10618compilers. It will not work in the Java or Fortran compilers, or any 10619other language compilers for that matter. Thus if @code{pragma_lex} is going 10620to be called from target-specific code, it must only be done so when 10621building the C and C++ compilers. This can be done by defining the 10622variables @code{c_target_objs} and @code{cxx_target_objs} in the 10623target entry in the @file{config.gcc} file. These variables should name 10624the target-specific, language-specific object file which contains the 10625code that uses @code{pragma_lex}. Note it will also be necessary to add a 10626rule to the makefile fragment pointed to by @code{tmake_file} that shows 10627how to build this object file. 10628@end deftypefun 10629 10630@defmac HANDLE_PRAGMA_PACK_WITH_EXPANSION 10631Define this macro if macros should be expanded in the 10632arguments of @samp{#pragma pack}. 10633@end defmac 10634 10635@hook TARGET_HANDLE_PRAGMA_EXTERN_PREFIX 10636 10637@defmac TARGET_DEFAULT_PACK_STRUCT 10638If your target requires a structure packing default other than 0 (meaning 10639the machine default), define this macro to the necessary value (in bytes). 10640This must be a value that would also be valid to use with 10641@samp{#pragma pack()} (that is, a small power of two). 10642@end defmac 10643 10644@defmac DOLLARS_IN_IDENTIFIERS 10645Define this macro to control use of the character @samp{$} in 10646identifier names for the C family of languages. 0 means @samp{$} is 10647not allowed by default; 1 means it is allowed. 1 is the default; 10648there is no need to define this macro in that case. 10649@end defmac 10650 10651@defmac NO_DOLLAR_IN_LABEL 10652Define this macro if the assembler does not accept the character 10653@samp{$} in label names. By default constructors and destructors in 10654G++ have @samp{$} in the identifiers. If this macro is defined, 10655@samp{.} is used instead. 10656@end defmac 10657 10658@defmac NO_DOT_IN_LABEL 10659Define this macro if the assembler does not accept the character 10660@samp{.} in label names. By default constructors and destructors in G++ 10661have names that use @samp{.}. If this macro is defined, these names 10662are rewritten to avoid @samp{.}. 10663@end defmac 10664 10665@defmac INSN_SETS_ARE_DELAYED (@var{insn}) 10666Define this macro as a C expression that is nonzero if it is safe for the 10667delay slot scheduler to place instructions in the delay slot of @var{insn}, 10668even if they appear to use a resource set or clobbered in @var{insn}. 10669@var{insn} is always a @code{jump_insn} or an @code{insn}; GCC knows that 10670every @code{call_insn} has this behavior. On machines where some @code{insn} 10671or @code{jump_insn} is really a function call and hence has this behavior, 10672you should define this macro. 10673 10674You need not define this macro if it would always return zero. 10675@end defmac 10676 10677@defmac INSN_REFERENCES_ARE_DELAYED (@var{insn}) 10678Define this macro as a C expression that is nonzero if it is safe for the 10679delay slot scheduler to place instructions in the delay slot of @var{insn}, 10680even if they appear to set or clobber a resource referenced in @var{insn}. 10681@var{insn} is always a @code{jump_insn} or an @code{insn}. On machines where 10682some @code{insn} or @code{jump_insn} is really a function call and its operands 10683are registers whose use is actually in the subroutine it calls, you should 10684define this macro. Doing so allows the delay slot scheduler to move 10685instructions which copy arguments into the argument registers into the delay 10686slot of @var{insn}. 10687 10688You need not define this macro if it would always return zero. 10689@end defmac 10690 10691@defmac MULTIPLE_SYMBOL_SPACES 10692Define this macro as a C expression that is nonzero if, in some cases, 10693global symbols from one translation unit may not be bound to undefined 10694symbols in another translation unit without user intervention. For 10695instance, under Microsoft Windows symbols must be explicitly imported 10696from shared libraries (DLLs). 10697 10698You need not define this macro if it would always evaluate to zero. 10699@end defmac 10700 10701@hook TARGET_MD_ASM_CLOBBERS 10702This target hook should add to @var{clobbers} @code{STRING_CST} trees for 10703any hard regs the port wishes to automatically clobber for an asm. 10704It should return the result of the last @code{tree_cons} used to add a 10705clobber. The @var{outputs}, @var{inputs} and @var{clobber} lists are the 10706corresponding parameters to the asm and may be inspected to avoid 10707clobbering a register that is an input or output of the asm. You can use 10708@code{tree_overlaps_hard_reg_set}, declared in @file{tree.h}, to test 10709for overlap with regards to asm-declared registers. 10710@end deftypefn 10711 10712@defmac MATH_LIBRARY 10713Define this macro as a C string constant for the linker argument to link 10714in the system math library, minus the initial @samp{"-l"}, or 10715@samp{""} if the target does not have a 10716separate math library. 10717 10718You need only define this macro if the default of @samp{"m"} is wrong. 10719@end defmac 10720 10721@defmac LIBRARY_PATH_ENV 10722Define this macro as a C string constant for the environment variable that 10723specifies where the linker should look for libraries. 10724 10725You need only define this macro if the default of @samp{"LIBRARY_PATH"} 10726is wrong. 10727@end defmac 10728 10729@defmac TARGET_POSIX_IO 10730Define this macro if the target supports the following POSIX@ file 10731functions, access, mkdir and file locking with fcntl / F_SETLKW@. 10732Defining @code{TARGET_POSIX_IO} will enable the test coverage code 10733to use file locking when exiting a program, which avoids race conditions 10734if the program has forked. It will also create directories at run-time 10735for cross-profiling. 10736@end defmac 10737 10738@defmac MAX_CONDITIONAL_EXECUTE 10739 10740A C expression for the maximum number of instructions to execute via 10741conditional execution instructions instead of a branch. A value of 10742@code{BRANCH_COST}+1 is the default if the machine does not use cc0, and 107431 if it does use cc0. 10744@end defmac 10745 10746@defmac IFCVT_MODIFY_TESTS (@var{ce_info}, @var{true_expr}, @var{false_expr}) 10747Used if the target needs to perform machine-dependent modifications on the 10748conditionals used for turning basic blocks into conditionally executed code. 10749@var{ce_info} points to a data structure, @code{struct ce_if_block}, which 10750contains information about the currently processed blocks. @var{true_expr} 10751and @var{false_expr} are the tests that are used for converting the 10752then-block and the else-block, respectively. Set either @var{true_expr} or 10753@var{false_expr} to a null pointer if the tests cannot be converted. 10754@end defmac 10755 10756@defmac IFCVT_MODIFY_MULTIPLE_TESTS (@var{ce_info}, @var{bb}, @var{true_expr}, @var{false_expr}) 10757Like @code{IFCVT_MODIFY_TESTS}, but used when converting more complicated 10758if-statements into conditions combined by @code{and} and @code{or} operations. 10759@var{bb} contains the basic block that contains the test that is currently 10760being processed and about to be turned into a condition. 10761@end defmac 10762 10763@defmac IFCVT_MODIFY_INSN (@var{ce_info}, @var{pattern}, @var{insn}) 10764A C expression to modify the @var{PATTERN} of an @var{INSN} that is to 10765be converted to conditional execution format. @var{ce_info} points to 10766a data structure, @code{struct ce_if_block}, which contains information 10767about the currently processed blocks. 10768@end defmac 10769 10770@defmac IFCVT_MODIFY_FINAL (@var{ce_info}) 10771A C expression to perform any final machine dependent modifications in 10772converting code to conditional execution. The involved basic blocks 10773can be found in the @code{struct ce_if_block} structure that is pointed 10774to by @var{ce_info}. 10775@end defmac 10776 10777@defmac IFCVT_MODIFY_CANCEL (@var{ce_info}) 10778A C expression to cancel any machine dependent modifications in 10779converting code to conditional execution. The involved basic blocks 10780can be found in the @code{struct ce_if_block} structure that is pointed 10781to by @var{ce_info}. 10782@end defmac 10783 10784@defmac IFCVT_INIT_EXTRA_FIELDS (@var{ce_info}) 10785A C expression to initialize any extra fields in a @code{struct ce_if_block} 10786structure, which are defined by the @code{IFCVT_EXTRA_FIELDS} macro. 10787@end defmac 10788 10789@defmac IFCVT_EXTRA_FIELDS 10790If defined, it should expand to a set of field declarations that will be 10791added to the @code{struct ce_if_block} structure. These should be initialized 10792by the @code{IFCVT_INIT_EXTRA_FIELDS} macro. 10793@end defmac 10794 10795@hook TARGET_MACHINE_DEPENDENT_REORG 10796If non-null, this hook performs a target-specific pass over the 10797instruction stream. The compiler will run it at all optimization levels, 10798just before the point at which it normally does delayed-branch scheduling. 10799 10800The exact purpose of the hook varies from target to target. Some use 10801it to do transformations that are necessary for correctness, such as 10802laying out in-function constant pools or avoiding hardware hazards. 10803Others use it as an opportunity to do some machine-dependent optimizations. 10804 10805You need not implement the hook if it has nothing to do. The default 10806definition is null. 10807@end deftypefn 10808 10809@hook TARGET_INIT_BUILTINS 10810Define this hook if you have any machine-specific built-in functions 10811that need to be defined. It should be a function that performs the 10812necessary setup. 10813 10814Machine specific built-in functions can be useful to expand special machine 10815instructions that would otherwise not normally be generated because 10816they have no equivalent in the source language (for example, SIMD vector 10817instructions or prefetch instructions). 10818 10819To create a built-in function, call the function 10820@code{lang_hooks.builtin_function} 10821which is defined by the language front end. You can use any type nodes set 10822up by @code{build_common_tree_nodes}; 10823only language front ends that use those two functions will call 10824@samp{TARGET_INIT_BUILTINS}. 10825@end deftypefn 10826 10827@hook TARGET_BUILTIN_DECL 10828Define this hook if you have any machine-specific built-in functions 10829that need to be defined. It should be a function that returns the 10830builtin function declaration for the builtin function code @var{code}. 10831If there is no such builtin and it cannot be initialized at this time 10832if @var{initialize_p} is true the function should return @code{NULL_TREE}. 10833If @var{code} is out of range the function should return 10834@code{error_mark_node}. 10835@end deftypefn 10836 10837@hook TARGET_EXPAND_BUILTIN 10838 10839Expand a call to a machine specific built-in function that was set up by 10840@samp{TARGET_INIT_BUILTINS}. @var{exp} is the expression for the 10841function call; the result should go to @var{target} if that is 10842convenient, and have mode @var{mode} if that is convenient. 10843@var{subtarget} may be used as the target for computing one of 10844@var{exp}'s operands. @var{ignore} is nonzero if the value is to be 10845ignored. This function should return the result of the call to the 10846built-in function. 10847@end deftypefn 10848 10849@hook TARGET_RESOLVE_OVERLOADED_BUILTIN 10850Select a replacement for a machine specific built-in function that 10851was set up by @samp{TARGET_INIT_BUILTINS}. This is done 10852@emph{before} regular type checking, and so allows the target to 10853implement a crude form of function overloading. @var{fndecl} is the 10854declaration of the built-in function. @var{arglist} is the list of 10855arguments passed to the built-in function. The result is a 10856complete expression that implements the operation, usually 10857another @code{CALL_EXPR}. 10858@var{arglist} really has type @samp{VEC(tree,gc)*} 10859@end deftypefn 10860 10861@hook TARGET_FOLD_BUILTIN 10862Fold a call to a machine specific built-in function that was set up by 10863@samp{TARGET_INIT_BUILTINS}. @var{fndecl} is the declaration of the 10864built-in function. @var{n_args} is the number of arguments passed to 10865the function; the arguments themselves are pointed to by @var{argp}. 10866The result is another tree containing a simplified expression for the 10867call's result. If @var{ignore} is true the value will be ignored. 10868@end deftypefn 10869 10870@hook TARGET_INVALID_WITHIN_DOLOOP 10871 10872Take an instruction in @var{insn} and return NULL if it is valid within a 10873low-overhead loop, otherwise return a string explaining why doloop 10874could not be applied. 10875 10876Many targets use special registers for low-overhead looping. For any 10877instruction that clobbers these this function should return a string indicating 10878the reason why the doloop could not be applied. 10879By default, the RTL loop optimizer does not use a present doloop pattern for 10880loops containing function calls or branch on table instructions. 10881@end deftypefn 10882 10883@defmac MD_CAN_REDIRECT_BRANCH (@var{branch1}, @var{branch2}) 10884 10885Take a branch insn in @var{branch1} and another in @var{branch2}. 10886Return true if redirecting @var{branch1} to the destination of 10887@var{branch2} is possible. 10888 10889On some targets, branches may have a limited range. Optimizing the 10890filling of delay slots can result in branches being redirected, and this 10891may in turn cause a branch offset to overflow. 10892@end defmac 10893 10894@hook TARGET_COMMUTATIVE_P 10895This target hook returns @code{true} if @var{x} is considered to be commutative. 10896Usually, this is just COMMUTATIVE_P (@var{x}), but the HP PA doesn't consider 10897PLUS to be commutative inside a MEM@. @var{outer_code} is the rtx code 10898of the enclosing rtl, if known, otherwise it is UNKNOWN. 10899@end deftypefn 10900 10901@hook TARGET_ALLOCATE_INITIAL_VALUE 10902 10903When the initial value of a hard register has been copied in a pseudo 10904register, it is often not necessary to actually allocate another register 10905to this pseudo register, because the original hard register or a stack slot 10906it has been saved into can be used. @code{TARGET_ALLOCATE_INITIAL_VALUE} 10907is called at the start of register allocation once for each hard register 10908that had its initial value copied by using 10909@code{get_func_hard_reg_initial_val} or @code{get_hard_reg_initial_val}. 10910Possible values are @code{NULL_RTX}, if you don't want 10911to do any special allocation, a @code{REG} rtx---that would typically be 10912the hard register itself, if it is known not to be clobbered---or a 10913@code{MEM}. 10914If you are returning a @code{MEM}, this is only a hint for the allocator; 10915it might decide to use another register anyways. 10916You may use @code{current_function_leaf_function} in the hook, functions 10917that use @code{REG_N_SETS}, to determine if the hard 10918register in question will not be clobbered. 10919The default value of this hook is @code{NULL}, which disables any special 10920allocation. 10921@end deftypefn 10922 10923@hook TARGET_UNSPEC_MAY_TRAP_P 10924This target hook returns nonzero if @var{x}, an @code{unspec} or 10925@code{unspec_volatile} operation, might cause a trap. Targets can use 10926this hook to enhance precision of analysis for @code{unspec} and 10927@code{unspec_volatile} operations. You may call @code{may_trap_p_1} 10928to analyze inner elements of @var{x} in which case @var{flags} should be 10929passed along. 10930@end deftypefn 10931 10932@hook TARGET_SET_CURRENT_FUNCTION 10933The compiler invokes this hook whenever it changes its current function 10934context (@code{cfun}). You can define this function if 10935the back end needs to perform any initialization or reset actions on a 10936per-function basis. For example, it may be used to implement function 10937attributes that affect register usage or code generation patterns. 10938The argument @var{decl} is the declaration for the new function context, 10939and may be null to indicate that the compiler has left a function context 10940and is returning to processing at the top level. 10941The default hook function does nothing. 10942 10943GCC sets @code{cfun} to a dummy function context during initialization of 10944some parts of the back end. The hook function is not invoked in this 10945situation; you need not worry about the hook being invoked recursively, 10946or when the back end is in a partially-initialized state. 10947@code{cfun} might be @code{NULL} to indicate processing at top level, 10948outside of any function scope. 10949@end deftypefn 10950 10951@defmac TARGET_OBJECT_SUFFIX 10952Define this macro to be a C string representing the suffix for object 10953files on your target machine. If you do not define this macro, GCC will 10954use @samp{.o} as the suffix for object files. 10955@end defmac 10956 10957@defmac TARGET_EXECUTABLE_SUFFIX 10958Define this macro to be a C string representing the suffix to be 10959automatically added to executable files on your target machine. If you 10960do not define this macro, GCC will use the null string as the suffix for 10961executable files. 10962@end defmac 10963 10964@defmac COLLECT_EXPORT_LIST 10965If defined, @code{collect2} will scan the individual object files 10966specified on its command line and create an export list for the linker. 10967Define this macro for systems like AIX, where the linker discards 10968object files that are not referenced from @code{main} and uses export 10969lists. 10970@end defmac 10971 10972@defmac MODIFY_JNI_METHOD_CALL (@var{mdecl}) 10973Define this macro to a C expression representing a variant of the 10974method call @var{mdecl}, if Java Native Interface (JNI) methods 10975must be invoked differently from other methods on your target. 10976For example, on 32-bit Microsoft Windows, JNI methods must be invoked using 10977the @code{stdcall} calling convention and this macro is then 10978defined as this expression: 10979 10980@smallexample 10981build_type_attribute_variant (@var{mdecl}, 10982 build_tree_list 10983 (get_identifier ("stdcall"), 10984 NULL)) 10985@end smallexample 10986@end defmac 10987 10988@hook TARGET_CANNOT_MODIFY_JUMPS_P 10989This target hook returns @code{true} past the point in which new jump 10990instructions could be created. On machines that require a register for 10991every jump such as the SHmedia ISA of SH5, this point would typically be 10992reload, so this target hook should be defined to a function such as: 10993 10994@smallexample 10995static bool 10996cannot_modify_jumps_past_reload_p () 10997@{ 10998 return (reload_completed || reload_in_progress); 10999@} 11000@end smallexample 11001@end deftypefn 11002 11003@hook TARGET_BRANCH_TARGET_REGISTER_CLASS 11004This target hook returns a register class for which branch target register 11005optimizations should be applied. All registers in this class should be 11006usable interchangeably. After reload, registers in this class will be 11007re-allocated and loads will be hoisted out of loops and be subjected 11008to inter-block scheduling. 11009@end deftypefn 11010 11011@hook TARGET_BRANCH_TARGET_REGISTER_CALLEE_SAVED 11012Branch target register optimization will by default exclude callee-saved 11013registers 11014that are not already live during the current function; if this target hook 11015returns true, they will be included. The target code must than make sure 11016that all target registers in the class returned by 11017@samp{TARGET_BRANCH_TARGET_REGISTER_CLASS} that might need saving are 11018saved. @var{after_prologue_epilogue_gen} indicates if prologues and 11019epilogues have already been generated. Note, even if you only return 11020true when @var{after_prologue_epilogue_gen} is false, you still are likely 11021to have to make special provisions in @code{INITIAL_ELIMINATION_OFFSET} 11022to reserve space for caller-saved target registers. 11023@end deftypefn 11024 11025@hook TARGET_HAVE_CONDITIONAL_EXECUTION 11026This target hook returns true if the target supports conditional execution. 11027This target hook is required only when the target has several different 11028modes and they have different conditional execution capability, such as ARM. 11029@end deftypefn 11030 11031@hook TARGET_LOOP_UNROLL_ADJUST 11032This target hook returns a new value for the number of times @var{loop} 11033should be unrolled. The parameter @var{nunroll} is the number of times 11034the loop is to be unrolled. The parameter @var{loop} is a pointer to 11035the loop, which is going to be checked for unrolling. This target hook 11036is required only when the target has special constraints like maximum 11037number of memory accesses. 11038@end deftypefn 11039 11040@defmac POWI_MAX_MULTS 11041If defined, this macro is interpreted as a signed integer C expression 11042that specifies the maximum number of floating point multiplications 11043that should be emitted when expanding exponentiation by an integer 11044constant inline. When this value is defined, exponentiation requiring 11045more than this number of multiplications is implemented by calling the 11046system library's @code{pow}, @code{powf} or @code{powl} routines. 11047The default value places no upper bound on the multiplication count. 11048@end defmac 11049 11050@deftypefn Macro void TARGET_EXTRA_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc}) 11051This target hook should register any extra include files for the 11052target. The parameter @var{stdinc} indicates if normal include files 11053are present. The parameter @var{sysroot} is the system root directory. 11054The parameter @var{iprefix} is the prefix for the gcc directory. 11055@end deftypefn 11056 11057@deftypefn Macro void TARGET_EXTRA_PRE_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc}) 11058This target hook should register any extra include files for the 11059target before any standard headers. The parameter @var{stdinc} 11060indicates if normal include files are present. The parameter 11061@var{sysroot} is the system root directory. The parameter 11062@var{iprefix} is the prefix for the gcc directory. 11063@end deftypefn 11064 11065@deftypefn Macro void TARGET_OPTF (char *@var{path}) 11066This target hook should register special include paths for the target. 11067The parameter @var{path} is the include to register. On Darwin 11068systems, this is used for Framework includes, which have semantics 11069that are different from @option{-I}. 11070@end deftypefn 11071 11072@defmac bool TARGET_USE_LOCAL_THUNK_ALIAS_P (tree @var{fndecl}) 11073This target macro returns @code{true} if it is safe to use a local alias 11074for a virtual function @var{fndecl} when constructing thunks, 11075@code{false} otherwise. By default, the macro returns @code{true} for all 11076functions, if a target supports aliases (i.e.@: defines 11077@code{ASM_OUTPUT_DEF}), @code{false} otherwise, 11078@end defmac 11079 11080@defmac TARGET_FORMAT_TYPES 11081If defined, this macro is the name of a global variable containing 11082target-specific format checking information for the @option{-Wformat} 11083option. The default is to have no target-specific format checks. 11084@end defmac 11085 11086@defmac TARGET_N_FORMAT_TYPES 11087If defined, this macro is the number of entries in 11088@code{TARGET_FORMAT_TYPES}. 11089@end defmac 11090 11091@defmac TARGET_OVERRIDES_FORMAT_ATTRIBUTES 11092If defined, this macro is the name of a global variable containing 11093target-specific format overrides for the @option{-Wformat} option. The 11094default is to have no target-specific format overrides. If defined, 11095@code{TARGET_FORMAT_TYPES} must be defined, too. 11096@end defmac 11097 11098@defmac TARGET_OVERRIDES_FORMAT_ATTRIBUTES_COUNT 11099If defined, this macro specifies the number of entries in 11100@code{TARGET_OVERRIDES_FORMAT_ATTRIBUTES}. 11101@end defmac 11102 11103@defmac TARGET_OVERRIDES_FORMAT_INIT 11104If defined, this macro specifies the optional initialization 11105routine for target specific customizations of the system printf 11106and scanf formatter settings. 11107@end defmac 11108 11109@hook TARGET_RELAXED_ORDERING 11110If set to @code{true}, means that the target's memory model does not 11111guarantee that loads which do not depend on one another will access 11112main memory in the order of the instruction stream; if ordering is 11113important, an explicit memory barrier must be used. This is true of 11114many recent processors which implement a policy of ``relaxed,'' 11115``weak,'' or ``release'' memory consistency, such as Alpha, PowerPC, 11116and ia64. The default is @code{false}. 11117@end deftypevr 11118 11119@hook TARGET_INVALID_ARG_FOR_UNPROTOTYPED_FN 11120If defined, this macro returns the diagnostic message when it is 11121illegal to pass argument @var{val} to function @var{funcdecl} 11122with prototype @var{typelist}. 11123@end deftypefn 11124 11125@hook TARGET_INVALID_CONVERSION 11126If defined, this macro returns the diagnostic message when it is 11127invalid to convert from @var{fromtype} to @var{totype}, or @code{NULL} 11128if validity should be determined by the front end. 11129@end deftypefn 11130 11131@hook TARGET_INVALID_UNARY_OP 11132If defined, this macro returns the diagnostic message when it is 11133invalid to apply operation @var{op} (where unary plus is denoted by 11134@code{CONVERT_EXPR}) to an operand of type @var{type}, or @code{NULL} 11135if validity should be determined by the front end. 11136@end deftypefn 11137 11138@hook TARGET_INVALID_BINARY_OP 11139If defined, this macro returns the diagnostic message when it is 11140invalid to apply operation @var{op} to operands of types @var{type1} 11141and @var{type2}, or @code{NULL} if validity should be determined by 11142the front end. 11143@end deftypefn 11144 11145@hook TARGET_INVALID_PARAMETER_TYPE 11146If defined, this macro returns the diagnostic message when it is 11147invalid for functions to include parameters of type @var{type}, 11148or @code{NULL} if validity should be determined by 11149the front end. This is currently used only by the C and C++ front ends. 11150@end deftypefn 11151 11152@hook TARGET_INVALID_RETURN_TYPE 11153If defined, this macro returns the diagnostic message when it is 11154invalid for functions to have return type @var{type}, 11155or @code{NULL} if validity should be determined by 11156the front end. This is currently used only by the C and C++ front ends. 11157@end deftypefn 11158 11159@hook TARGET_PROMOTED_TYPE 11160If defined, this target hook returns the type to which values of 11161@var{type} should be promoted when they appear in expressions, 11162analogous to the integer promotions, or @code{NULL_TREE} to use the 11163front end's normal promotion rules. This hook is useful when there are 11164target-specific types with special promotion rules. 11165This is currently used only by the C and C++ front ends. 11166@end deftypefn 11167 11168@hook TARGET_CONVERT_TO_TYPE 11169If defined, this hook returns the result of converting @var{expr} to 11170@var{type}. It should return the converted expression, 11171or @code{NULL_TREE} to apply the front end's normal conversion rules. 11172This hook is useful when there are target-specific types with special 11173conversion rules. 11174This is currently used only by the C and C++ front ends. 11175@end deftypefn 11176 11177@defmac TARGET_USE_JCR_SECTION 11178This macro determines whether to use the JCR section to register Java 11179classes. By default, TARGET_USE_JCR_SECTION is defined to 1 if both 11180SUPPORTS_WEAK and TARGET_HAVE_NAMED_SECTIONS are true, else 0. 11181@end defmac 11182 11183@defmac OBJC_JBLEN 11184This macro determines the size of the objective C jump buffer for the 11185NeXT runtime. By default, OBJC_JBLEN is defined to an innocuous value. 11186@end defmac 11187 11188@defmac LIBGCC2_UNWIND_ATTRIBUTE 11189Define this macro if any target-specific attributes need to be attached 11190to the functions in @file{libgcc} that provide low-level support for 11191call stack unwinding. It is used in declarations in @file{unwind-generic.h} 11192and the associated definitions of those functions. 11193@end defmac 11194 11195@hook TARGET_UPDATE_STACK_BOUNDARY 11196Define this macro to update the current function stack boundary if 11197necessary. 11198@end deftypefn 11199 11200@hook TARGET_GET_DRAP_RTX 11201This hook should return an rtx for Dynamic Realign Argument Pointer (DRAP) if a 11202different argument pointer register is needed to access the function's 11203argument list due to stack realignment. Return @code{NULL} if no DRAP 11204is needed. 11205@end deftypefn 11206 11207@hook TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS 11208When optimization is disabled, this hook indicates whether or not 11209arguments should be allocated to stack slots. Normally, GCC allocates 11210stacks slots for arguments when not optimizing in order to make 11211debugging easier. However, when a function is declared with 11212@code{__attribute__((naked))}, there is no stack frame, and the compiler 11213cannot safely move arguments from the registers in which they are passed 11214to the stack. Therefore, this hook should return true in general, but 11215false for naked functions. The default implementation always returns true. 11216@end deftypefn 11217 11218@hook TARGET_CONST_ANCHOR 11219On some architectures it can take multiple instructions to synthesize 11220a constant. If there is another constant already in a register that 11221is close enough in value then it is preferable that the new constant 11222is computed from this register using immediate addition or 11223subtraction. We accomplish this through CSE. Besides the value of 11224the constant we also add a lower and an upper constant anchor to the 11225available expressions. These are then queried when encountering new 11226constants. The anchors are computed by rounding the constant up and 11227down to a multiple of the value of @code{TARGET_CONST_ANCHOR}. 11228@code{TARGET_CONST_ANCHOR} should be the maximum positive value 11229accepted by immediate-add plus one. We currently assume that the 11230value of @code{TARGET_CONST_ANCHOR} is a power of 2. For example, on 11231MIPS, where add-immediate takes a 16-bit signed value, 11232@code{TARGET_CONST_ANCHOR} is set to @samp{0x8000}. The default value 11233is zero, which disables this optimization. @end deftypevr 11234 11235@hook TARGET_ATOMIC_TEST_AND_SET_TRUEVAL 11236