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_STRING_OBJECT_REF_TYPE_P 700 701@hook TARGET_CHECK_STRING_OBJECT_FORMAT_ARG 702 703@hook TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE 704This target function is similar to the hook @code{TARGET_OPTION_OVERRIDE} 705but is called when the optimize level is changed via an attribute or 706pragma or when it is reset at the end of the code affected by the 707attribute or pragma. It is not called at the beginning of compilation 708when @code{TARGET_OPTION_OVERRIDE} is called so if you want to perform these 709actions then, you should have @code{TARGET_OPTION_OVERRIDE} call 710@code{TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE}. 711@end deftypefn 712 713@defmac C_COMMON_OVERRIDE_OPTIONS 714This is similar to the @code{TARGET_OPTION_OVERRIDE} hook 715but is only used in the C 716language frontends (C, Objective-C, C++, Objective-C++) and so can be 717used to alter option flag variables which only exist in those 718frontends. 719@end defmac 720 721@hook TARGET_OPTION_OPTIMIZATION_TABLE 722Some machines may desire to change what optimizations are performed for 723various optimization levels. This variable, if defined, describes 724options to enable at particular sets of optimization levels. These 725options are processed once 726just after the optimization level is determined and before the remainder 727of the command options have been parsed, so may be overridden by other 728options passed explicitly. 729 730This processing is run once at program startup and when the optimization 731options are changed via @code{#pragma GCC optimize} or by using the 732@code{optimize} attribute. 733@end deftypevr 734 735@hook TARGET_OPTION_INIT_STRUCT 736 737@hook TARGET_OPTION_DEFAULT_PARAMS 738 739@defmac SWITCHABLE_TARGET 740Some targets need to switch between substantially different subtargets 741during compilation. For example, the MIPS target has one subtarget for 742the traditional MIPS architecture and another for MIPS16. Source code 743can switch between these two subarchitectures using the @code{mips16} 744and @code{nomips16} attributes. 745 746Such subtargets can differ in things like the set of available 747registers, the set of available instructions, the costs of various 748operations, and so on. GCC caches a lot of this type of information 749in global variables, and recomputing them for each subtarget takes a 750significant amount of time. The compiler therefore provides a facility 751for maintaining several versions of the global variables and quickly 752switching between them; see @file{target-globals.h} for details. 753 754Define this macro to 1 if your target needs this facility. The default 755is 0. 756@end defmac 757 758@node Per-Function Data 759@section Defining data structures for per-function information. 760@cindex per-function data 761@cindex data structures 762 763If the target needs to store information on a per-function basis, GCC 764provides a macro and a couple of variables to allow this. Note, just 765using statics to store the information is a bad idea, since GCC supports 766nested functions, so you can be halfway through encoding one function 767when another one comes along. 768 769GCC defines a data structure called @code{struct function} which 770contains all of the data specific to an individual function. This 771structure contains a field called @code{machine} whose type is 772@code{struct machine_function *}, which can be used by targets to point 773to their own specific data. 774 775If a target needs per-function specific data it should define the type 776@code{struct machine_function} and also the macro @code{INIT_EXPANDERS}. 777This macro should be used to initialize the function pointer 778@code{init_machine_status}. This pointer is explained below. 779 780One typical use of per-function, target specific data is to create an 781RTX to hold the register containing the function's return address. This 782RTX can then be used to implement the @code{__builtin_return_address} 783function, for level 0. 784 785Note---earlier implementations of GCC used a single data area to hold 786all of the per-function information. Thus when processing of a nested 787function began the old per-function data had to be pushed onto a 788stack, and when the processing was finished, it had to be popped off the 789stack. GCC used to provide function pointers called 790@code{save_machine_status} and @code{restore_machine_status} to handle 791the saving and restoring of the target specific information. Since the 792single data area approach is no longer used, these pointers are no 793longer supported. 794 795@defmac INIT_EXPANDERS 796Macro called to initialize any target specific information. This macro 797is called once per function, before generation of any RTL has begun. 798The intention of this macro is to allow the initialization of the 799function pointer @code{init_machine_status}. 800@end defmac 801 802@deftypevar {void (*)(struct function *)} init_machine_status 803If this function pointer is non-@code{NULL} it will be called once per 804function, before function compilation starts, in order to allow the 805target to perform any target specific initialization of the 806@code{struct function} structure. It is intended that this would be 807used to initialize the @code{machine} of that structure. 808 809@code{struct machine_function} structures are expected to be freed by GC@. 810Generally, any memory that they reference must be allocated by using 811GC allocation, including the structure itself. 812@end deftypevar 813 814@node Storage Layout 815@section Storage Layout 816@cindex storage layout 817 818Note that the definitions of the macros in this table which are sizes or 819alignments measured in bits do not need to be constant. They can be C 820expressions that refer to static variables, such as the @code{target_flags}. 821@xref{Run-time Target}. 822 823@defmac BITS_BIG_ENDIAN 824Define this macro to have the value 1 if the most significant bit in a 825byte has the lowest number; otherwise define it to have the value zero. 826This means that bit-field instructions count from the most significant 827bit. If the machine has no bit-field instructions, then this must still 828be defined, but it doesn't matter which value it is defined to. This 829macro need not be a constant. 830 831This macro does not affect the way structure fields are packed into 832bytes or words; that is controlled by @code{BYTES_BIG_ENDIAN}. 833@end defmac 834 835@defmac BYTES_BIG_ENDIAN 836Define this macro to have the value 1 if the most significant byte in a 837word has the lowest number. This macro need not be a constant. 838@end defmac 839 840@defmac WORDS_BIG_ENDIAN 841Define this macro to have the value 1 if, in a multiword object, the 842most significant word has the lowest number. This applies to both 843memory locations and registers; see @code{REG_WORDS_BIG_ENDIAN} if the 844order of words in memory is not the same as the order in registers. This 845macro need not be a constant. 846@end defmac 847 848@defmac REG_WORDS_BIG_ENDIAN 849On some machines, the order of words in a multiword object differs between 850registers in memory. In such a situation, define this macro to describe 851the order of words in a register. The macro @code{WORDS_BIG_ENDIAN} controls 852the order of words in memory. 853@end defmac 854 855@defmac FLOAT_WORDS_BIG_ENDIAN 856Define this macro to have the value 1 if @code{DFmode}, @code{XFmode} or 857@code{TFmode} floating point numbers are stored in memory with the word 858containing the sign bit at the lowest address; otherwise define it to 859have the value 0. This macro need not be a constant. 860 861You need not define this macro if the ordering is the same as for 862multi-word integers. 863@end defmac 864 865@defmac BITS_PER_UNIT 866Define this macro to be the number of bits in an addressable storage 867unit (byte). If you do not define this macro the default is 8. 868@end defmac 869 870@defmac BITS_PER_WORD 871Number of bits in a word. If you do not define this macro, the default 872is @code{BITS_PER_UNIT * UNITS_PER_WORD}. 873@end defmac 874 875@defmac MAX_BITS_PER_WORD 876Maximum number of bits in a word. If this is undefined, the default is 877@code{BITS_PER_WORD}. Otherwise, it is the constant value that is the 878largest value that @code{BITS_PER_WORD} can have at run-time. 879@end defmac 880 881@defmac UNITS_PER_WORD 882Number of storage units in a word; normally the size of a general-purpose 883register, a power of two from 1 or 8. 884@end defmac 885 886@defmac MIN_UNITS_PER_WORD 887Minimum number of units in a word. If this is undefined, the default is 888@code{UNITS_PER_WORD}. Otherwise, it is the constant value that is the 889smallest value that @code{UNITS_PER_WORD} can have at run-time. 890@end defmac 891 892@defmac POINTER_SIZE 893Width of a pointer, in bits. You must specify a value no wider than the 894width of @code{Pmode}. If it is not equal to the width of @code{Pmode}, 895you must define @code{POINTERS_EXTEND_UNSIGNED}. If you do not specify 896a value the default is @code{BITS_PER_WORD}. 897@end defmac 898 899@defmac POINTERS_EXTEND_UNSIGNED 900A C expression that determines how pointers should be extended from 901@code{ptr_mode} to either @code{Pmode} or @code{word_mode}. It is 902greater than zero if pointers should be zero-extended, zero if they 903should be sign-extended, and negative if some other sort of conversion 904is needed. In the last case, the extension is done by the target's 905@code{ptr_extend} instruction. 906 907You need not define this macro if the @code{ptr_mode}, @code{Pmode} 908and @code{word_mode} are all the same width. 909@end defmac 910 911@defmac PROMOTE_MODE (@var{m}, @var{unsignedp}, @var{type}) 912A macro to update @var{m} and @var{unsignedp} when an object whose type 913is @var{type} and which has the specified mode and signedness is to be 914stored in a register. This macro is only called when @var{type} is a 915scalar type. 916 917On most RISC machines, which only have operations that operate on a full 918register, define this macro to set @var{m} to @code{word_mode} if 919@var{m} is an integer mode narrower than @code{BITS_PER_WORD}. In most 920cases, only integer modes should be widened because wider-precision 921floating-point operations are usually more expensive than their narrower 922counterparts. 923 924For most machines, the macro definition does not change @var{unsignedp}. 925However, some machines, have instructions that preferentially handle 926either signed or unsigned quantities of certain modes. For example, on 927the DEC Alpha, 32-bit loads from memory and 32-bit add instructions 928sign-extend the result to 64 bits. On such machines, set 929@var{unsignedp} according to which kind of extension is more efficient. 930 931Do not define this macro if it would never modify @var{m}. 932@end defmac 933 934@hook TARGET_PROMOTE_FUNCTION_MODE 935Like @code{PROMOTE_MODE}, but it is applied to outgoing function arguments or 936function return values. The target hook should return the new mode 937and possibly change @code{*@var{punsignedp}} if the promotion should 938change signedness. This function is called only for scalar @emph{or 939pointer} types. 940 941@var{for_return} allows to distinguish the promotion of arguments and 942return values. If it is @code{1}, a return value is being promoted and 943@code{TARGET_FUNCTION_VALUE} must perform the same promotions done here. 944If it is @code{2}, the returned mode should be that of the register in 945which an incoming parameter is copied, or the outgoing result is computed; 946then the hook should return the same mode as @code{promote_mode}, though 947the signedness may be different. 948 949@var{type} can be NULL when promoting function arguments of libcalls. 950 951The default is to not promote arguments and return values. You can 952also define the hook to @code{default_promote_function_mode_always_promote} 953if you would like to apply the same rules given by @code{PROMOTE_MODE}. 954@end deftypefn 955 956@defmac PARM_BOUNDARY 957Normal alignment required for function parameters on the stack, in 958bits. All stack parameters receive at least this much alignment 959regardless of data type. On most machines, this is the same as the 960size of an integer. 961@end defmac 962 963@defmac STACK_BOUNDARY 964Define this macro to the minimum alignment enforced by hardware for the 965stack pointer on this machine. The definition is a C expression for the 966desired alignment (measured in bits). This value is used as a default 967if @code{PREFERRED_STACK_BOUNDARY} is not defined. On most machines, 968this should be the same as @code{PARM_BOUNDARY}. 969@end defmac 970 971@defmac PREFERRED_STACK_BOUNDARY 972Define this macro if you wish to preserve a certain alignment for the 973stack pointer, greater than what the hardware enforces. The definition 974is a C expression for the desired alignment (measured in bits). This 975macro must evaluate to a value equal to or larger than 976@code{STACK_BOUNDARY}. 977@end defmac 978 979@defmac INCOMING_STACK_BOUNDARY 980Define this macro if the incoming stack boundary may be different 981from @code{PREFERRED_STACK_BOUNDARY}. This macro must evaluate 982to a value equal to or larger than @code{STACK_BOUNDARY}. 983@end defmac 984 985@defmac FUNCTION_BOUNDARY 986Alignment required for a function entry point, in bits. 987@end defmac 988 989@defmac BIGGEST_ALIGNMENT 990Biggest alignment that any data type can require on this machine, in 991bits. Note that this is not the biggest alignment that is supported, 992just the biggest alignment that, when violated, may cause a fault. 993@end defmac 994 995@defmac MALLOC_ABI_ALIGNMENT 996Alignment, in bits, a C conformant malloc implementation has to 997provide. If not defined, the default value is @code{BITS_PER_WORD}. 998@end defmac 999 1000@defmac ATTRIBUTE_ALIGNED_VALUE 1001Alignment used by the @code{__attribute__ ((aligned))} construct. If 1002not defined, the default value is @code{BIGGEST_ALIGNMENT}. 1003@end defmac 1004 1005@defmac MINIMUM_ATOMIC_ALIGNMENT 1006If defined, the smallest alignment, in bits, that can be given to an 1007object that can be referenced in one operation, without disturbing any 1008nearby object. Normally, this is @code{BITS_PER_UNIT}, but may be larger 1009on machines that don't have byte or half-word store operations. 1010@end defmac 1011 1012@defmac BIGGEST_FIELD_ALIGNMENT 1013Biggest alignment that any structure or union field can require on this 1014machine, in bits. If defined, this overrides @code{BIGGEST_ALIGNMENT} for 1015structure and union fields only, unless the field alignment has been set 1016by the @code{__attribute__ ((aligned (@var{n})))} construct. 1017@end defmac 1018 1019@defmac ADJUST_FIELD_ALIGN (@var{field}, @var{computed}) 1020An expression for the alignment of a structure field @var{field} if the 1021alignment computed in the usual way (including applying of 1022@code{BIGGEST_ALIGNMENT} and @code{BIGGEST_FIELD_ALIGNMENT} to the 1023alignment) is @var{computed}. It overrides alignment only if the 1024field alignment has not been set by the 1025@code{__attribute__ ((aligned (@var{n})))} construct. 1026@end defmac 1027 1028@defmac MAX_STACK_ALIGNMENT 1029Biggest stack alignment guaranteed by the backend. Use this macro 1030to specify the maximum alignment of a variable on stack. 1031 1032If not defined, the default value is @code{STACK_BOUNDARY}. 1033 1034@c FIXME: The default should be @code{PREFERRED_STACK_BOUNDARY}. 1035@c But the fix for PR 32893 indicates that we can only guarantee 1036@c maximum stack alignment on stack up to @code{STACK_BOUNDARY}, not 1037@c @code{PREFERRED_STACK_BOUNDARY}, if stack alignment isn't supported. 1038@end defmac 1039 1040@defmac MAX_OFILE_ALIGNMENT 1041Biggest alignment supported by the object file format of this machine. 1042Use this macro to limit the alignment which can be specified using the 1043@code{__attribute__ ((aligned (@var{n})))} construct. If not defined, 1044the default value is @code{BIGGEST_ALIGNMENT}. 1045 1046On systems that use ELF, the default (in @file{config/elfos.h}) is 1047the largest supported 32-bit ELF section alignment representable on 1048a 32-bit host e.g. @samp{(((unsigned HOST_WIDEST_INT) 1 << 28) * 8)}. 1049On 32-bit ELF the largest supported section alignment in bits is 1050@samp{(0x80000000 * 8)}, but this is not representable on 32-bit hosts. 1051@end defmac 1052 1053@defmac DATA_ALIGNMENT (@var{type}, @var{basic-align}) 1054If defined, a C expression to compute the alignment for a variable in 1055the static store. @var{type} is the data type, and @var{basic-align} is 1056the alignment that the object would ordinarily have. The value of this 1057macro is used instead of that alignment to align the object. 1058 1059If this macro is not defined, then @var{basic-align} is used. 1060 1061@findex strcpy 1062One use of this macro is to increase alignment of medium-size data to 1063make it all fit in fewer cache lines. Another is to cause character 1064arrays to be word-aligned so that @code{strcpy} calls that copy 1065constants to character arrays can be done inline. 1066@end defmac 1067 1068@defmac CONSTANT_ALIGNMENT (@var{constant}, @var{basic-align}) 1069If defined, a C expression to compute the alignment given to a constant 1070that is being placed in memory. @var{constant} is the constant and 1071@var{basic-align} is the alignment that the object would ordinarily 1072have. The value of this macro is used instead of that alignment to 1073align the object. 1074 1075If this macro is not defined, then @var{basic-align} is used. 1076 1077The typical use of this macro is to increase alignment for string 1078constants to be word aligned so that @code{strcpy} calls that copy 1079constants can be done inline. 1080@end defmac 1081 1082@defmac LOCAL_ALIGNMENT (@var{type}, @var{basic-align}) 1083If defined, a C expression to compute the alignment for a variable in 1084the local store. @var{type} is the data type, and @var{basic-align} is 1085the alignment that the object would ordinarily have. The value of this 1086macro is used instead of that alignment to align the object. 1087 1088If this macro is not defined, then @var{basic-align} is used. 1089 1090One use of this macro is to increase alignment of medium-size data to 1091make it all fit in fewer cache lines. 1092 1093If the value of this macro has a type, it should be an unsigned type. 1094@end defmac 1095 1096@hook TARGET_VECTOR_ALIGNMENT 1097 1098@defmac STACK_SLOT_ALIGNMENT (@var{type}, @var{mode}, @var{basic-align}) 1099If defined, a C expression to compute the alignment for stack slot. 1100@var{type} is the data type, @var{mode} is the widest mode available, 1101and @var{basic-align} is the alignment that the slot would ordinarily 1102have. The value of this macro is used instead of that alignment to 1103align the slot. 1104 1105If this macro is not defined, then @var{basic-align} is used when 1106@var{type} is @code{NULL}. Otherwise, @code{LOCAL_ALIGNMENT} will 1107be used. 1108 1109This macro is to set alignment of stack slot to the maximum alignment 1110of all possible modes which the slot may have. 1111 1112If the value of this macro has a type, it should be an unsigned type. 1113@end defmac 1114 1115@defmac LOCAL_DECL_ALIGNMENT (@var{decl}) 1116If defined, a C expression to compute the alignment for a local 1117variable @var{decl}. 1118 1119If this macro is not defined, then 1120@code{LOCAL_ALIGNMENT (TREE_TYPE (@var{decl}), DECL_ALIGN (@var{decl}))} 1121is used. 1122 1123One use of this macro is to increase alignment of medium-size data to 1124make it all fit in fewer cache lines. 1125 1126If the value of this macro has a type, it should be an unsigned type. 1127@end defmac 1128 1129@defmac MINIMUM_ALIGNMENT (@var{exp}, @var{mode}, @var{align}) 1130If defined, a C expression to compute the minimum required alignment 1131for dynamic stack realignment purposes for @var{exp} (a type or decl), 1132@var{mode}, assuming normal alignment @var{align}. 1133 1134If this macro is not defined, then @var{align} will be used. 1135@end defmac 1136 1137@defmac EMPTY_FIELD_BOUNDARY 1138Alignment in bits to be given to a structure bit-field that follows an 1139empty field such as @code{int : 0;}. 1140 1141If @code{PCC_BITFIELD_TYPE_MATTERS} is true, it overrides this macro. 1142@end defmac 1143 1144@defmac STRUCTURE_SIZE_BOUNDARY 1145Number of bits which any structure or union's size must be a multiple of. 1146Each structure or union's size is rounded up to a multiple of this. 1147 1148If you do not define this macro, the default is the same as 1149@code{BITS_PER_UNIT}. 1150@end defmac 1151 1152@defmac STRICT_ALIGNMENT 1153Define this macro to be the value 1 if instructions will fail to work 1154if given data not on the nominal alignment. If instructions will merely 1155go slower in that case, define this macro as 0. 1156@end defmac 1157 1158@defmac PCC_BITFIELD_TYPE_MATTERS 1159Define this if you wish to imitate the way many other C compilers handle 1160alignment of bit-fields and the structures that contain them. 1161 1162The behavior is that the type written for a named bit-field (@code{int}, 1163@code{short}, or other integer type) imposes an alignment for the entire 1164structure, as if the structure really did contain an ordinary field of 1165that type. In addition, the bit-field is placed within the structure so 1166that it would fit within such a field, not crossing a boundary for it. 1167 1168Thus, on most machines, a named bit-field whose type is written as 1169@code{int} would not cross a four-byte boundary, and would force 1170four-byte alignment for the whole structure. (The alignment used may 1171not be four bytes; it is controlled by the other alignment parameters.) 1172 1173An unnamed bit-field will not affect the alignment of the containing 1174structure. 1175 1176If the macro is defined, its definition should be a C expression; 1177a nonzero value for the expression enables this behavior. 1178 1179Note that if this macro is not defined, or its value is zero, some 1180bit-fields may cross more than one alignment boundary. The compiler can 1181support such references if there are @samp{insv}, @samp{extv}, and 1182@samp{extzv} insns that can directly reference memory. 1183 1184The other known way of making bit-fields work is to define 1185@code{STRUCTURE_SIZE_BOUNDARY} as large as @code{BIGGEST_ALIGNMENT}. 1186Then every structure can be accessed with fullwords. 1187 1188Unless the machine has bit-field instructions or you define 1189@code{STRUCTURE_SIZE_BOUNDARY} that way, you must define 1190@code{PCC_BITFIELD_TYPE_MATTERS} to have a nonzero value. 1191 1192If your aim is to make GCC use the same conventions for laying out 1193bit-fields as are used by another compiler, here is how to investigate 1194what the other compiler does. Compile and run this program: 1195 1196@smallexample 1197struct foo1 1198@{ 1199 char x; 1200 char :0; 1201 char y; 1202@}; 1203 1204struct foo2 1205@{ 1206 char x; 1207 int :0; 1208 char y; 1209@}; 1210 1211main () 1212@{ 1213 printf ("Size of foo1 is %d\n", 1214 sizeof (struct foo1)); 1215 printf ("Size of foo2 is %d\n", 1216 sizeof (struct foo2)); 1217 exit (0); 1218@} 1219@end smallexample 1220 1221If this prints 2 and 5, then the compiler's behavior is what you would 1222get from @code{PCC_BITFIELD_TYPE_MATTERS}. 1223@end defmac 1224 1225@defmac BITFIELD_NBYTES_LIMITED 1226Like @code{PCC_BITFIELD_TYPE_MATTERS} except that its effect is limited 1227to aligning a bit-field within the structure. 1228@end defmac 1229 1230@hook TARGET_ALIGN_ANON_BITFIELD 1231When @code{PCC_BITFIELD_TYPE_MATTERS} is true this hook will determine 1232whether unnamed bitfields affect the alignment of the containing 1233structure. The hook should return true if the structure should inherit 1234the alignment requirements of an unnamed bitfield's type. 1235@end deftypefn 1236 1237@hook TARGET_NARROW_VOLATILE_BITFIELD 1238This target hook should return @code{true} if accesses to volatile bitfields 1239should use the narrowest mode possible. It should return @code{false} if 1240these accesses should use the bitfield container type. 1241 1242The default is @code{!TARGET_STRICT_ALIGN}. 1243@end deftypefn 1244 1245@defmac MEMBER_TYPE_FORCES_BLK (@var{field}, @var{mode}) 1246Return 1 if a structure or array containing @var{field} should be accessed using 1247@code{BLKMODE}. 1248 1249If @var{field} is the only field in the structure, @var{mode} is its 1250mode, otherwise @var{mode} is VOIDmode. @var{mode} is provided in the 1251case where structures of one field would require the structure's mode to 1252retain the field's mode. 1253 1254Normally, this is not needed. 1255@end defmac 1256 1257@defmac ROUND_TYPE_ALIGN (@var{type}, @var{computed}, @var{specified}) 1258Define this macro as an expression for the alignment of a type (given 1259by @var{type} as a tree node) if the alignment computed in the usual 1260way is @var{computed} and the alignment explicitly specified was 1261@var{specified}. 1262 1263The default is to use @var{specified} if it is larger; otherwise, use 1264the smaller of @var{computed} and @code{BIGGEST_ALIGNMENT} 1265@end defmac 1266 1267@defmac MAX_FIXED_MODE_SIZE 1268An integer expression for the size in bits of the largest integer 1269machine mode that should actually be used. All integer machine modes of 1270this size or smaller can be used for structures and unions with the 1271appropriate sizes. If this macro is undefined, @code{GET_MODE_BITSIZE 1272(DImode)} is assumed. 1273@end defmac 1274 1275@defmac STACK_SAVEAREA_MODE (@var{save_level}) 1276If defined, an expression of type @code{enum machine_mode} that 1277specifies the mode of the save area operand of a 1278@code{save_stack_@var{level}} named pattern (@pxref{Standard Names}). 1279@var{save_level} is one of @code{SAVE_BLOCK}, @code{SAVE_FUNCTION}, or 1280@code{SAVE_NONLOCAL} and selects which of the three named patterns is 1281having its mode specified. 1282 1283You need not define this macro if it always returns @code{Pmode}. You 1284would most commonly define this macro if the 1285@code{save_stack_@var{level}} patterns need to support both a 32- and a 128664-bit mode. 1287@end defmac 1288 1289@defmac STACK_SIZE_MODE 1290If defined, an expression of type @code{enum machine_mode} that 1291specifies the mode of the size increment operand of an 1292@code{allocate_stack} named pattern (@pxref{Standard Names}). 1293 1294You need not define this macro if it always returns @code{word_mode}. 1295You would most commonly define this macro if the @code{allocate_stack} 1296pattern needs to support both a 32- and a 64-bit mode. 1297@end defmac 1298 1299@hook TARGET_LIBGCC_CMP_RETURN_MODE 1300This target hook should return the mode to be used for the return value 1301of compare instructions expanded to libgcc calls. If not defined 1302@code{word_mode} is returned which is the right choice for a majority of 1303targets. 1304@end deftypefn 1305 1306@hook TARGET_LIBGCC_SHIFT_COUNT_MODE 1307This target hook should return the mode to be used for the shift count operand 1308of shift instructions expanded to libgcc calls. If not defined 1309@code{word_mode} is returned which is the right choice for a majority of 1310targets. 1311@end deftypefn 1312 1313@hook TARGET_UNWIND_WORD_MODE 1314Return machine mode to be used for @code{_Unwind_Word} type. 1315The default is to use @code{word_mode}. 1316@end deftypefn 1317 1318@defmac ROUND_TOWARDS_ZERO 1319If defined, this macro should be true if the prevailing rounding 1320mode is towards zero. 1321 1322Defining this macro only affects the way @file{libgcc.a} emulates 1323floating-point arithmetic. 1324 1325Not defining this macro is equivalent to returning zero. 1326@end defmac 1327 1328@defmac LARGEST_EXPONENT_IS_NORMAL (@var{size}) 1329This macro should return true if floats with @var{size} 1330bits do not have a NaN or infinity representation, but use the largest 1331exponent for normal numbers instead. 1332 1333Defining this macro only affects the way @file{libgcc.a} emulates 1334floating-point arithmetic. 1335 1336The default definition of this macro returns false for all sizes. 1337@end defmac 1338 1339@hook TARGET_MS_BITFIELD_LAYOUT_P 1340This target hook returns @code{true} if bit-fields in the given 1341@var{record_type} are to be laid out following the rules of Microsoft 1342Visual C/C++, namely: (i) a bit-field won't share the same storage 1343unit with the previous bit-field if their underlying types have 1344different sizes, and the bit-field will be aligned to the highest 1345alignment of the underlying types of itself and of the previous 1346bit-field; (ii) a zero-sized bit-field will affect the alignment of 1347the whole enclosing structure, even if it is unnamed; except that 1348(iii) a zero-sized bit-field will be disregarded unless it follows 1349another bit-field of nonzero size. If this hook returns @code{true}, 1350other macros that control bit-field layout are ignored. 1351 1352When a bit-field is inserted into a packed record, the whole size 1353of the underlying type is used by one or more same-size adjacent 1354bit-fields (that is, if its long:3, 32 bits is used in the record, 1355and any additional adjacent long bit-fields are packed into the same 1356chunk of 32 bits. However, if the size changes, a new field of that 1357size is allocated). In an unpacked record, this is the same as using 1358alignment, but not equivalent when packing. 1359 1360If both MS bit-fields and @samp{__attribute__((packed))} are used, 1361the latter will take precedence. If @samp{__attribute__((packed))} is 1362used on a single field when MS bit-fields are in use, it will take 1363precedence for that field, but the alignment of the rest of the structure 1364may affect its placement. 1365@end deftypefn 1366 1367@hook TARGET_DECIMAL_FLOAT_SUPPORTED_P 1368Returns true if the target supports decimal floating point. 1369@end deftypefn 1370 1371@hook TARGET_FIXED_POINT_SUPPORTED_P 1372Returns true if the target supports fixed-point arithmetic. 1373@end deftypefn 1374 1375@hook TARGET_EXPAND_TO_RTL_HOOK 1376This hook is called just before expansion into rtl, allowing the target 1377to perform additional initializations or analysis before the expansion. 1378For example, the rs6000 port uses it to allocate a scratch stack slot 1379for use in copying SDmode values between memory and floating point 1380registers whenever the function being expanded has any SDmode 1381usage. 1382@end deftypefn 1383 1384@hook TARGET_INSTANTIATE_DECLS 1385This hook allows the backend to perform additional instantiations on rtl 1386that are not actually in any insns yet, but will be later. 1387@end deftypefn 1388 1389@hook TARGET_MANGLE_TYPE 1390If your target defines any fundamental types, or any types your target 1391uses should be mangled differently from the default, define this hook 1392to return the appropriate encoding for these types as part of a C++ 1393mangled name. The @var{type} argument is the tree structure representing 1394the type to be mangled. The hook may be applied to trees which are 1395not target-specific fundamental types; it should return @code{NULL} 1396for all such types, as well as arguments it does not recognize. If the 1397return value is not @code{NULL}, it must point to a statically-allocated 1398string constant. 1399 1400Target-specific fundamental types might be new fundamental types or 1401qualified versions of ordinary fundamental types. Encode new 1402fundamental types as @samp{@w{u @var{n} @var{name}}}, where @var{name} 1403is the name used for the type in source code, and @var{n} is the 1404length of @var{name} in decimal. Encode qualified versions of 1405ordinary types as @samp{@w{U @var{n} @var{name} @var{code}}}, where 1406@var{name} is the name used for the type qualifier in source code, 1407@var{n} is the length of @var{name} as above, and @var{code} is the 1408code used to represent the unqualified version of this type. (See 1409@code{write_builtin_type} in @file{cp/mangle.c} for the list of 1410codes.) In both cases the spaces are for clarity; do not include any 1411spaces in your string. 1412 1413This hook is applied to types prior to typedef resolution. If the mangled 1414name for a particular type depends only on that type's main variant, you 1415can perform typedef resolution yourself using @code{TYPE_MAIN_VARIANT} 1416before mangling. 1417 1418The default version of this hook always returns @code{NULL}, which is 1419appropriate for a target that does not define any new fundamental 1420types. 1421@end deftypefn 1422 1423@node Type Layout 1424@section Layout of Source Language Data Types 1425 1426These macros define the sizes and other characteristics of the standard 1427basic data types used in programs being compiled. Unlike the macros in 1428the previous section, these apply to specific features of C and related 1429languages, rather than to fundamental aspects of storage layout. 1430 1431@defmac INT_TYPE_SIZE 1432A C expression for the size in bits of the type @code{int} on the 1433target machine. If you don't define this, the default is one word. 1434@end defmac 1435 1436@defmac SHORT_TYPE_SIZE 1437A C expression for the size in bits of the type @code{short} on the 1438target machine. If you don't define this, the default is half a word. 1439(If this would be less than one storage unit, it is rounded up to one 1440unit.) 1441@end defmac 1442 1443@defmac LONG_TYPE_SIZE 1444A C expression for the size in bits of the type @code{long} on the 1445target machine. If you don't define this, the default is one word. 1446@end defmac 1447 1448@defmac ADA_LONG_TYPE_SIZE 1449On some machines, the size used for the Ada equivalent of the type 1450@code{long} by a native Ada compiler differs from that used by C@. In 1451that situation, define this macro to be a C expression to be used for 1452the size of that type. If you don't define this, the default is the 1453value of @code{LONG_TYPE_SIZE}. 1454@end defmac 1455 1456@defmac LONG_LONG_TYPE_SIZE 1457A C expression for the size in bits of the type @code{long long} on the 1458target machine. If you don't define this, the default is two 1459words. If you want to support GNU Ada on your machine, the value of this 1460macro must be at least 64. 1461@end defmac 1462 1463@defmac CHAR_TYPE_SIZE 1464A C expression for the size in bits of the type @code{char} on the 1465target machine. If you don't define this, the default is 1466@code{BITS_PER_UNIT}. 1467@end defmac 1468 1469@defmac BOOL_TYPE_SIZE 1470A C expression for the size in bits of the C++ type @code{bool} and 1471C99 type @code{_Bool} on the target machine. If you don't define 1472this, and you probably shouldn't, the default is @code{CHAR_TYPE_SIZE}. 1473@end defmac 1474 1475@defmac FLOAT_TYPE_SIZE 1476A C expression for the size in bits of the type @code{float} on the 1477target machine. If you don't define this, the default is one word. 1478@end defmac 1479 1480@defmac DOUBLE_TYPE_SIZE 1481A C expression for the size in bits of the type @code{double} on the 1482target machine. If you don't define this, the default is two 1483words. 1484@end defmac 1485 1486@defmac LONG_DOUBLE_TYPE_SIZE 1487A C expression for the size in bits of the type @code{long double} on 1488the target machine. If you don't define this, the default is two 1489words. 1490@end defmac 1491 1492@defmac SHORT_FRACT_TYPE_SIZE 1493A C expression for the size in bits of the type @code{short _Fract} on 1494the target machine. If you don't define this, the default is 1495@code{BITS_PER_UNIT}. 1496@end defmac 1497 1498@defmac FRACT_TYPE_SIZE 1499A C expression for the size in bits of the type @code{_Fract} on 1500the target machine. If you don't define this, the default is 1501@code{BITS_PER_UNIT * 2}. 1502@end defmac 1503 1504@defmac LONG_FRACT_TYPE_SIZE 1505A C expression for the size in bits of the type @code{long _Fract} on 1506the target machine. If you don't define this, the default is 1507@code{BITS_PER_UNIT * 4}. 1508@end defmac 1509 1510@defmac LONG_LONG_FRACT_TYPE_SIZE 1511A C expression for the size in bits of the type @code{long long _Fract} on 1512the target machine. If you don't define this, the default is 1513@code{BITS_PER_UNIT * 8}. 1514@end defmac 1515 1516@defmac SHORT_ACCUM_TYPE_SIZE 1517A C expression for the size in bits of the type @code{short _Accum} on 1518the target machine. If you don't define this, the default is 1519@code{BITS_PER_UNIT * 2}. 1520@end defmac 1521 1522@defmac ACCUM_TYPE_SIZE 1523A C expression for the size in bits of the type @code{_Accum} on 1524the target machine. If you don't define this, the default is 1525@code{BITS_PER_UNIT * 4}. 1526@end defmac 1527 1528@defmac LONG_ACCUM_TYPE_SIZE 1529A C expression for the size in bits of the type @code{long _Accum} on 1530the target machine. If you don't define this, the default is 1531@code{BITS_PER_UNIT * 8}. 1532@end defmac 1533 1534@defmac LONG_LONG_ACCUM_TYPE_SIZE 1535A C expression for the size in bits of the type @code{long long _Accum} on 1536the target machine. If you don't define this, the default is 1537@code{BITS_PER_UNIT * 16}. 1538@end defmac 1539 1540@defmac LIBGCC2_LONG_DOUBLE_TYPE_SIZE 1541Define this macro if @code{LONG_DOUBLE_TYPE_SIZE} is not constant or 1542if you want routines in @file{libgcc2.a} for a size other than 1543@code{LONG_DOUBLE_TYPE_SIZE}. If you don't define this, the 1544default is @code{LONG_DOUBLE_TYPE_SIZE}. 1545@end defmac 1546 1547@defmac LIBGCC2_HAS_DF_MODE 1548Define this macro if neither @code{DOUBLE_TYPE_SIZE} nor 1549@code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 1550@code{DFmode} but you want @code{DFmode} routines in @file{libgcc2.a} 1551anyway. If you don't define this and either @code{DOUBLE_TYPE_SIZE} 1552or @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 64 then the default is 1, 1553otherwise it is 0. 1554@end defmac 1555 1556@defmac LIBGCC2_HAS_XF_MODE 1557Define this macro if @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is not 1558@code{XFmode} but you want @code{XFmode} routines in @file{libgcc2.a} 1559anyway. If you don't define this and @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} 1560is 80 then the default is 1, otherwise it is 0. 1561@end defmac 1562 1563@defmac LIBGCC2_HAS_TF_MODE 1564Define this macro if @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is not 1565@code{TFmode} but you want @code{TFmode} routines in @file{libgcc2.a} 1566anyway. If you don't define this and @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} 1567is 128 then the default is 1, otherwise it is 0. 1568@end defmac 1569 1570@defmac LIBGCC2_GNU_PREFIX 1571This macro corresponds to the @code{TARGET_LIBFUNC_GNU_PREFIX} target 1572hook and should be defined if that hook is overriden to be true. It 1573causes function names in libgcc to be changed to use a @code{__gnu_} 1574prefix for their name rather than the default @code{__}. A port which 1575uses this macro should also arrange to use @file{t-gnu-prefix} in 1576the libgcc @file{config.host}. 1577@end defmac 1578 1579@defmac SF_SIZE 1580@defmacx DF_SIZE 1581@defmacx XF_SIZE 1582@defmacx TF_SIZE 1583Define these macros to be the size in bits of the mantissa of 1584@code{SFmode}, @code{DFmode}, @code{XFmode} and @code{TFmode} values, 1585if the defaults in @file{libgcc2.h} are inappropriate. By default, 1586@code{FLT_MANT_DIG} is used for @code{SF_SIZE}, @code{LDBL_MANT_DIG} 1587for @code{XF_SIZE} and @code{TF_SIZE}, and @code{DBL_MANT_DIG} or 1588@code{LDBL_MANT_DIG} for @code{DF_SIZE} according to whether 1589@code{DOUBLE_TYPE_SIZE} or 1590@code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 64. 1591@end defmac 1592 1593@defmac TARGET_FLT_EVAL_METHOD 1594A C expression for the value for @code{FLT_EVAL_METHOD} in @file{float.h}, 1595assuming, if applicable, that the floating-point control word is in its 1596default state. If you do not define this macro the value of 1597@code{FLT_EVAL_METHOD} will be zero. 1598@end defmac 1599 1600@defmac WIDEST_HARDWARE_FP_SIZE 1601A C expression for the size in bits of the widest floating-point format 1602supported by the hardware. If you define this macro, you must specify a 1603value less than or equal to the value of @code{LONG_DOUBLE_TYPE_SIZE}. 1604If you do not define this macro, the value of @code{LONG_DOUBLE_TYPE_SIZE} 1605is the default. 1606@end defmac 1607 1608@defmac DEFAULT_SIGNED_CHAR 1609An expression whose value is 1 or 0, according to whether the type 1610@code{char} should be signed or unsigned by default. The user can 1611always override this default with the options @option{-fsigned-char} 1612and @option{-funsigned-char}. 1613@end defmac 1614 1615@hook TARGET_DEFAULT_SHORT_ENUMS 1616This target hook should return true if the compiler should give an 1617@code{enum} type only as many bytes as it takes to represent the range 1618of possible values of that type. It should return false if all 1619@code{enum} types should be allocated like @code{int}. 1620 1621The default is to return false. 1622@end deftypefn 1623 1624@defmac SIZE_TYPE 1625A C expression for a string describing the name of the data type to use 1626for size values. The typedef name @code{size_t} is defined using the 1627contents of the string. 1628 1629The string can contain more than one keyword. If so, separate them with 1630spaces, and write first any length keyword, then @code{unsigned} if 1631appropriate, and finally @code{int}. The string must exactly match one 1632of the data type names defined in the function 1633@code{init_decl_processing} in the file @file{c-decl.c}. You may not 1634omit @code{int} or change the order---that would cause the compiler to 1635crash on startup. 1636 1637If you don't define this macro, the default is @code{"long unsigned 1638int"}. 1639@end defmac 1640 1641@defmac PTRDIFF_TYPE 1642A C expression for a string describing the name of the data type to use 1643for the result of subtracting two pointers. The typedef name 1644@code{ptrdiff_t} is defined using the contents of the string. See 1645@code{SIZE_TYPE} above for more information. 1646 1647If you don't define this macro, the default is @code{"long int"}. 1648@end defmac 1649 1650@defmac WCHAR_TYPE 1651A C expression for a string describing the name of the data type to use 1652for wide characters. The typedef name @code{wchar_t} is defined using 1653the contents of the string. See @code{SIZE_TYPE} above for more 1654information. 1655 1656If you don't define this macro, the default is @code{"int"}. 1657@end defmac 1658 1659@defmac WCHAR_TYPE_SIZE 1660A C expression for the size in bits of the data type for wide 1661characters. This is used in @code{cpp}, which cannot make use of 1662@code{WCHAR_TYPE}. 1663@end defmac 1664 1665@defmac WINT_TYPE 1666A C expression for a string describing the name of the data type to 1667use for wide characters passed to @code{printf} and returned from 1668@code{getwc}. The typedef name @code{wint_t} is defined using the 1669contents of the string. See @code{SIZE_TYPE} above for more 1670information. 1671 1672If you don't define this macro, the default is @code{"unsigned int"}. 1673@end defmac 1674 1675@defmac INTMAX_TYPE 1676A C expression for a string describing the name of the data type that 1677can represent any value of any standard or extended signed integer type. 1678The typedef name @code{intmax_t} is defined using the contents of the 1679string. See @code{SIZE_TYPE} above for more information. 1680 1681If you don't define this macro, the default is the first of 1682@code{"int"}, @code{"long int"}, or @code{"long long int"} that has as 1683much precision as @code{long long int}. 1684@end defmac 1685 1686@defmac UINTMAX_TYPE 1687A C expression for a string describing the name of the data type that 1688can represent any value of any standard or extended unsigned integer 1689type. The typedef name @code{uintmax_t} is defined using the contents 1690of the string. See @code{SIZE_TYPE} above for more information. 1691 1692If you don't define this macro, the default is the first of 1693@code{"unsigned int"}, @code{"long unsigned int"}, or @code{"long long 1694unsigned int"} that has as much precision as @code{long long unsigned 1695int}. 1696@end defmac 1697 1698@defmac SIG_ATOMIC_TYPE 1699@defmacx INT8_TYPE 1700@defmacx INT16_TYPE 1701@defmacx INT32_TYPE 1702@defmacx INT64_TYPE 1703@defmacx UINT8_TYPE 1704@defmacx UINT16_TYPE 1705@defmacx UINT32_TYPE 1706@defmacx UINT64_TYPE 1707@defmacx INT_LEAST8_TYPE 1708@defmacx INT_LEAST16_TYPE 1709@defmacx INT_LEAST32_TYPE 1710@defmacx INT_LEAST64_TYPE 1711@defmacx UINT_LEAST8_TYPE 1712@defmacx UINT_LEAST16_TYPE 1713@defmacx UINT_LEAST32_TYPE 1714@defmacx UINT_LEAST64_TYPE 1715@defmacx INT_FAST8_TYPE 1716@defmacx INT_FAST16_TYPE 1717@defmacx INT_FAST32_TYPE 1718@defmacx INT_FAST64_TYPE 1719@defmacx UINT_FAST8_TYPE 1720@defmacx UINT_FAST16_TYPE 1721@defmacx UINT_FAST32_TYPE 1722@defmacx UINT_FAST64_TYPE 1723@defmacx INTPTR_TYPE 1724@defmacx UINTPTR_TYPE 1725C expressions for the standard types @code{sig_atomic_t}, 1726@code{int8_t}, @code{int16_t}, @code{int32_t}, @code{int64_t}, 1727@code{uint8_t}, @code{uint16_t}, @code{uint32_t}, @code{uint64_t}, 1728@code{int_least8_t}, @code{int_least16_t}, @code{int_least32_t}, 1729@code{int_least64_t}, @code{uint_least8_t}, @code{uint_least16_t}, 1730@code{uint_least32_t}, @code{uint_least64_t}, @code{int_fast8_t}, 1731@code{int_fast16_t}, @code{int_fast32_t}, @code{int_fast64_t}, 1732@code{uint_fast8_t}, @code{uint_fast16_t}, @code{uint_fast32_t}, 1733@code{uint_fast64_t}, @code{intptr_t}, and @code{uintptr_t}. See 1734@code{SIZE_TYPE} above for more information. 1735 1736If any of these macros evaluates to a null pointer, the corresponding 1737type is not supported; if GCC is configured to provide 1738@code{<stdint.h>} in such a case, the header provided may not conform 1739to C99, depending on the type in question. The defaults for all of 1740these macros are null pointers. 1741@end defmac 1742 1743@defmac TARGET_PTRMEMFUNC_VBIT_LOCATION 1744The C++ compiler represents a pointer-to-member-function with a struct 1745that looks like: 1746 1747@smallexample 1748 struct @{ 1749 union @{ 1750 void (*fn)(); 1751 ptrdiff_t vtable_index; 1752 @}; 1753 ptrdiff_t delta; 1754 @}; 1755@end smallexample 1756 1757@noindent 1758The C++ compiler must use one bit to indicate whether the function that 1759will be called through a pointer-to-member-function is virtual. 1760Normally, we assume that the low-order bit of a function pointer must 1761always be zero. Then, by ensuring that the vtable_index is odd, we can 1762distinguish which variant of the union is in use. But, on some 1763platforms function pointers can be odd, and so this doesn't work. In 1764that case, we use the low-order bit of the @code{delta} field, and shift 1765the remainder of the @code{delta} field to the left. 1766 1767GCC will automatically make the right selection about where to store 1768this bit using the @code{FUNCTION_BOUNDARY} setting for your platform. 1769However, some platforms such as ARM/Thumb have @code{FUNCTION_BOUNDARY} 1770set such that functions always start at even addresses, but the lowest 1771bit of pointers to functions indicate whether the function at that 1772address is in ARM or Thumb mode. If this is the case of your 1773architecture, you should define this macro to 1774@code{ptrmemfunc_vbit_in_delta}. 1775 1776In general, you should not have to define this macro. On architectures 1777in which function addresses are always even, according to 1778@code{FUNCTION_BOUNDARY}, GCC will automatically define this macro to 1779@code{ptrmemfunc_vbit_in_pfn}. 1780@end defmac 1781 1782@defmac TARGET_VTABLE_USES_DESCRIPTORS 1783Normally, the C++ compiler uses function pointers in vtables. This 1784macro allows the target to change to use ``function descriptors'' 1785instead. Function descriptors are found on targets for whom a 1786function pointer is actually a small data structure. Normally the 1787data structure consists of the actual code address plus a data 1788pointer to which the function's data is relative. 1789 1790If vtables are used, the value of this macro should be the number 1791of words that the function descriptor occupies. 1792@end defmac 1793 1794@defmac TARGET_VTABLE_ENTRY_ALIGN 1795By default, the vtable entries are void pointers, the so the alignment 1796is the same as pointer alignment. The value of this macro specifies 1797the alignment of the vtable entry in bits. It should be defined only 1798when special alignment is necessary. */ 1799@end defmac 1800 1801@defmac TARGET_VTABLE_DATA_ENTRY_DISTANCE 1802There are a few non-descriptor entries in the vtable at offsets below 1803zero. If these entries must be padded (say, to preserve the alignment 1804specified by @code{TARGET_VTABLE_ENTRY_ALIGN}), set this to the number 1805of words in each data entry. 1806@end defmac 1807 1808@node Registers 1809@section Register Usage 1810@cindex register usage 1811 1812This section explains how to describe what registers the target machine 1813has, and how (in general) they can be used. 1814 1815The description of which registers a specific instruction can use is 1816done with register classes; see @ref{Register Classes}. For information 1817on using registers to access a stack frame, see @ref{Frame Registers}. 1818For passing values in registers, see @ref{Register Arguments}. 1819For returning values in registers, see @ref{Scalar Return}. 1820 1821@menu 1822* Register Basics:: Number and kinds of registers. 1823* Allocation Order:: Order in which registers are allocated. 1824* Values in Registers:: What kinds of values each reg can hold. 1825* Leaf Functions:: Renumbering registers for leaf functions. 1826* Stack Registers:: Handling a register stack such as 80387. 1827@end menu 1828 1829@node Register Basics 1830@subsection Basic Characteristics of Registers 1831 1832@c prevent bad page break with this line 1833Registers have various characteristics. 1834 1835@defmac FIRST_PSEUDO_REGISTER 1836Number of hardware registers known to the compiler. They receive 1837numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first 1838pseudo register's number really is assigned the number 1839@code{FIRST_PSEUDO_REGISTER}. 1840@end defmac 1841 1842@defmac FIXED_REGISTERS 1843@cindex fixed register 1844An initializer that says which registers are used for fixed purposes 1845all throughout the compiled code and are therefore not available for 1846general allocation. These would include the stack pointer, the frame 1847pointer (except on machines where that can be used as a general 1848register when no frame pointer is needed), the program counter on 1849machines where that is considered one of the addressable registers, 1850and any other numbered register with a standard use. 1851 1852This information is expressed as a sequence of numbers, separated by 1853commas and surrounded by braces. The @var{n}th number is 1 if 1854register @var{n} is fixed, 0 otherwise. 1855 1856The table initialized from this macro, and the table initialized by 1857the following one, may be overridden at run time either automatically, 1858by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by 1859the user with the command options @option{-ffixed-@var{reg}}, 1860@option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}. 1861@end defmac 1862 1863@defmac CALL_USED_REGISTERS 1864@cindex call-used register 1865@cindex call-clobbered register 1866@cindex call-saved register 1867Like @code{FIXED_REGISTERS} but has 1 for each register that is 1868clobbered (in general) by function calls as well as for fixed 1869registers. This macro therefore identifies the registers that are not 1870available for general allocation of values that must live across 1871function calls. 1872 1873If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler 1874automatically saves it on function entry and restores it on function 1875exit, if the register is used within the function. 1876@end defmac 1877 1878@defmac CALL_REALLY_USED_REGISTERS 1879@cindex call-used register 1880@cindex call-clobbered register 1881@cindex call-saved register 1882Like @code{CALL_USED_REGISTERS} except this macro doesn't require 1883that the entire set of @code{FIXED_REGISTERS} be included. 1884(@code{CALL_USED_REGISTERS} must be a superset of @code{FIXED_REGISTERS}). 1885This macro is optional. If not specified, it defaults to the value 1886of @code{CALL_USED_REGISTERS}. 1887@end defmac 1888 1889@defmac HARD_REGNO_CALL_PART_CLOBBERED (@var{regno}, @var{mode}) 1890@cindex call-used register 1891@cindex call-clobbered register 1892@cindex call-saved register 1893A C expression that is nonzero if it is not permissible to store a 1894value of mode @var{mode} in hard register number @var{regno} across a 1895call without some part of it being clobbered. For most machines this 1896macro need not be defined. It is only required for machines that do not 1897preserve the entire contents of a register across a call. 1898@end defmac 1899 1900@findex fixed_regs 1901@findex call_used_regs 1902@findex global_regs 1903@findex reg_names 1904@findex reg_class_contents 1905@hook TARGET_CONDITIONAL_REGISTER_USAGE 1906This hook may conditionally modify five variables 1907@code{fixed_regs}, @code{call_used_regs}, @code{global_regs}, 1908@code{reg_names}, and @code{reg_class_contents}, to take into account 1909any dependence of these register sets on target flags. The first three 1910of these are of type @code{char []} (interpreted as Boolean vectors). 1911@code{global_regs} is a @code{const char *[]}, and 1912@code{reg_class_contents} is a @code{HARD_REG_SET}. Before the macro is 1913called, @code{fixed_regs}, @code{call_used_regs}, 1914@code{reg_class_contents}, and @code{reg_names} have been initialized 1915from @code{FIXED_REGISTERS}, @code{CALL_USED_REGISTERS}, 1916@code{REG_CLASS_CONTENTS}, and @code{REGISTER_NAMES}, respectively. 1917@code{global_regs} has been cleared, and any @option{-ffixed-@var{reg}}, 1918@option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}} 1919command options have been applied. 1920 1921@cindex disabling certain registers 1922@cindex controlling register usage 1923If the usage of an entire class of registers depends on the target 1924flags, you may indicate this to GCC by using this macro to modify 1925@code{fixed_regs} and @code{call_used_regs} to 1 for each of the 1926registers in the classes which should not be used by GCC@. Also define 1927the macro @code{REG_CLASS_FROM_LETTER} / @code{REG_CLASS_FROM_CONSTRAINT} 1928to return @code{NO_REGS} if it 1929is called with a letter for a class that shouldn't be used. 1930 1931(However, if this class is not included in @code{GENERAL_REGS} and all 1932of the insn patterns whose constraints permit this class are 1933controlled by target switches, then GCC will automatically avoid using 1934these registers when the target switches are opposed to them.) 1935@end deftypefn 1936 1937@defmac INCOMING_REGNO (@var{out}) 1938Define this macro if the target machine has register windows. This C 1939expression returns the register number as seen by the called function 1940corresponding to the register number @var{out} as seen by the calling 1941function. Return @var{out} if register number @var{out} is not an 1942outbound register. 1943@end defmac 1944 1945@defmac OUTGOING_REGNO (@var{in}) 1946Define this macro if the target machine has register windows. This C 1947expression returns the register number as seen by the calling function 1948corresponding to the register number @var{in} as seen by the called 1949function. Return @var{in} if register number @var{in} is not an inbound 1950register. 1951@end defmac 1952 1953@defmac LOCAL_REGNO (@var{regno}) 1954Define this macro if the target machine has register windows. This C 1955expression returns true if the register is call-saved but is in the 1956register window. Unlike most call-saved registers, such registers 1957need not be explicitly restored on function exit or during non-local 1958gotos. 1959@end defmac 1960 1961@defmac PC_REGNUM 1962If the program counter has a register number, define this as that 1963register number. Otherwise, do not define it. 1964@end defmac 1965 1966@node Allocation Order 1967@subsection Order of Allocation of Registers 1968@cindex order of register allocation 1969@cindex register allocation order 1970 1971@c prevent bad page break with this line 1972Registers are allocated in order. 1973 1974@defmac REG_ALLOC_ORDER 1975If defined, an initializer for a vector of integers, containing the 1976numbers of hard registers in the order in which GCC should prefer 1977to use them (from most preferred to least). 1978 1979If this macro is not defined, registers are used lowest numbered first 1980(all else being equal). 1981 1982One use of this macro is on machines where the highest numbered 1983registers must always be saved and the save-multiple-registers 1984instruction supports only sequences of consecutive registers. On such 1985machines, define @code{REG_ALLOC_ORDER} to be an initializer that lists 1986the highest numbered allocable register first. 1987@end defmac 1988 1989@defmac ADJUST_REG_ALLOC_ORDER 1990A C statement (sans semicolon) to choose the order in which to allocate 1991hard registers for pseudo-registers local to a basic block. 1992 1993Store the desired register order in the array @code{reg_alloc_order}. 1994Element 0 should be the register to allocate first; element 1, the next 1995register; and so on. 1996 1997The macro body should not assume anything about the contents of 1998@code{reg_alloc_order} before execution of the macro. 1999 2000On most machines, it is not necessary to define this macro. 2001@end defmac 2002 2003@defmac HONOR_REG_ALLOC_ORDER 2004Normally, IRA tries to estimate the costs for saving a register in the 2005prologue and restoring it in the epilogue. This discourages it from 2006using call-saved registers. If a machine wants to ensure that IRA 2007allocates registers in the order given by REG_ALLOC_ORDER even if some 2008call-saved registers appear earlier than call-used ones, this macro 2009should be defined. 2010@end defmac 2011 2012@defmac IRA_HARD_REGNO_ADD_COST_MULTIPLIER (@var{regno}) 2013In some case register allocation order is not enough for the 2014Integrated Register Allocator (@acronym{IRA}) to generate a good code. 2015If this macro is defined, it should return a floating point value 2016based on @var{regno}. The cost of using @var{regno} for a pseudo will 2017be increased by approximately the pseudo's usage frequency times the 2018value returned by this macro. Not defining this macro is equivalent 2019to having it always return @code{0.0}. 2020 2021On most machines, it is not necessary to define this macro. 2022@end defmac 2023 2024@node Values in Registers 2025@subsection How Values Fit in Registers 2026 2027This section discusses the macros that describe which kinds of values 2028(specifically, which machine modes) each register can hold, and how many 2029consecutive registers are needed for a given mode. 2030 2031@defmac HARD_REGNO_NREGS (@var{regno}, @var{mode}) 2032A C expression for the number of consecutive hard registers, starting 2033at register number @var{regno}, required to hold a value of mode 2034@var{mode}. This macro must never return zero, even if a register 2035cannot hold the requested mode - indicate that with HARD_REGNO_MODE_OK 2036and/or CANNOT_CHANGE_MODE_CLASS instead. 2037 2038On a machine where all registers are exactly one word, a suitable 2039definition of this macro is 2040 2041@smallexample 2042#define HARD_REGNO_NREGS(REGNO, MODE) \ 2043 ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) \ 2044 / UNITS_PER_WORD) 2045@end smallexample 2046@end defmac 2047 2048@defmac HARD_REGNO_NREGS_HAS_PADDING (@var{regno}, @var{mode}) 2049A C expression that is nonzero if a value of mode @var{mode}, stored 2050in memory, ends with padding that causes it to take up more space than 2051in registers starting at register number @var{regno} (as determined by 2052multiplying GCC's notion of the size of the register when containing 2053this mode by the number of registers returned by 2054@code{HARD_REGNO_NREGS}). By default this is zero. 2055 2056For example, if a floating-point value is stored in three 32-bit 2057registers but takes up 128 bits in memory, then this would be 2058nonzero. 2059 2060This macros only needs to be defined if there are cases where 2061@code{subreg_get_info} 2062would otherwise wrongly determine that a @code{subreg} can be 2063represented by an offset to the register number, when in fact such a 2064@code{subreg} would contain some of the padding not stored in 2065registers and so not be representable. 2066@end defmac 2067 2068@defmac HARD_REGNO_NREGS_WITH_PADDING (@var{regno}, @var{mode}) 2069For values of @var{regno} and @var{mode} for which 2070@code{HARD_REGNO_NREGS_HAS_PADDING} returns nonzero, a C expression 2071returning the greater number of registers required to hold the value 2072including any padding. In the example above, the value would be four. 2073@end defmac 2074 2075@defmac REGMODE_NATURAL_SIZE (@var{mode}) 2076Define this macro if the natural size of registers that hold values 2077of mode @var{mode} is not the word size. It is a C expression that 2078should give the natural size in bytes for the specified mode. It is 2079used by the register allocator to try to optimize its results. This 2080happens for example on SPARC 64-bit where the natural size of 2081floating-point registers is still 32-bit. 2082@end defmac 2083 2084@defmac HARD_REGNO_MODE_OK (@var{regno}, @var{mode}) 2085A C expression that is nonzero if it is permissible to store a value 2086of mode @var{mode} in hard register number @var{regno} (or in several 2087registers starting with that one). For a machine where all registers 2088are equivalent, a suitable definition is 2089 2090@smallexample 2091#define HARD_REGNO_MODE_OK(REGNO, MODE) 1 2092@end smallexample 2093 2094You need not include code to check for the numbers of fixed registers, 2095because the allocation mechanism considers them to be always occupied. 2096 2097@cindex register pairs 2098On some machines, double-precision values must be kept in even/odd 2099register pairs. You can implement that by defining this macro to reject 2100odd register numbers for such modes. 2101 2102The minimum requirement for a mode to be OK in a register is that the 2103@samp{mov@var{mode}} instruction pattern support moves between the 2104register and other hard register in the same class and that moving a 2105value into the register and back out not alter it. 2106 2107Since the same instruction used to move @code{word_mode} will work for 2108all narrower integer modes, it is not necessary on any machine for 2109@code{HARD_REGNO_MODE_OK} to distinguish between these modes, provided 2110you define patterns @samp{movhi}, etc., to take advantage of this. This 2111is useful because of the interaction between @code{HARD_REGNO_MODE_OK} 2112and @code{MODES_TIEABLE_P}; it is very desirable for all integer modes 2113to be tieable. 2114 2115Many machines have special registers for floating point arithmetic. 2116Often people assume that floating point machine modes are allowed only 2117in floating point registers. This is not true. Any registers that 2118can hold integers can safely @emph{hold} a floating point machine 2119mode, whether or not floating arithmetic can be done on it in those 2120registers. Integer move instructions can be used to move the values. 2121 2122On some machines, though, the converse is true: fixed-point machine 2123modes may not go in floating registers. This is true if the floating 2124registers normalize any value stored in them, because storing a 2125non-floating value there would garble it. In this case, 2126@code{HARD_REGNO_MODE_OK} should reject fixed-point machine modes in 2127floating registers. But if the floating registers do not automatically 2128normalize, if you can store any bit pattern in one and retrieve it 2129unchanged without a trap, then any machine mode may go in a floating 2130register, so you can define this macro to say so. 2131 2132The primary significance of special floating registers is rather that 2133they are the registers acceptable in floating point arithmetic 2134instructions. However, this is of no concern to 2135@code{HARD_REGNO_MODE_OK}. You handle it by writing the proper 2136constraints for those instructions. 2137 2138On some machines, the floating registers are especially slow to access, 2139so that it is better to store a value in a stack frame than in such a 2140register if floating point arithmetic is not being done. As long as the 2141floating registers are not in class @code{GENERAL_REGS}, they will not 2142be used unless some pattern's constraint asks for one. 2143@end defmac 2144 2145@defmac HARD_REGNO_RENAME_OK (@var{from}, @var{to}) 2146A C expression that is nonzero if it is OK to rename a hard register 2147@var{from} to another hard register @var{to}. 2148 2149One common use of this macro is to prevent renaming of a register to 2150another register that is not saved by a prologue in an interrupt 2151handler. 2152 2153The default is always nonzero. 2154@end defmac 2155 2156@defmac MODES_TIEABLE_P (@var{mode1}, @var{mode2}) 2157A C expression that is nonzero if a value of mode 2158@var{mode1} is accessible in mode @var{mode2} without copying. 2159 2160If @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and 2161@code{HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are always the same for 2162any @var{r}, then @code{MODES_TIEABLE_P (@var{mode1}, @var{mode2})} 2163should be nonzero. If they differ for any @var{r}, you should define 2164this macro to return zero unless some other mechanism ensures the 2165accessibility of the value in a narrower mode. 2166 2167You should define this macro to return nonzero in as many cases as 2168possible since doing so will allow GCC to perform better register 2169allocation. 2170@end defmac 2171 2172@hook TARGET_HARD_REGNO_SCRATCH_OK 2173This target hook should return @code{true} if it is OK to use a hard register 2174@var{regno} as scratch reg in peephole2. 2175 2176One common use of this macro is to prevent using of a register that 2177is not saved by a prologue in an interrupt handler. 2178 2179The default version of this hook always returns @code{true}. 2180@end deftypefn 2181 2182@defmac AVOID_CCMODE_COPIES 2183Define this macro if the compiler should avoid copies to/from @code{CCmode} 2184registers. You should only define this macro if support for copying to/from 2185@code{CCmode} is incomplete. 2186@end defmac 2187 2188@node Leaf Functions 2189@subsection Handling Leaf Functions 2190 2191@cindex leaf functions 2192@cindex functions, leaf 2193On some machines, a leaf function (i.e., one which makes no calls) can run 2194more efficiently if it does not make its own register window. Often this 2195means it is required to receive its arguments in the registers where they 2196are passed by the caller, instead of the registers where they would 2197normally arrive. 2198 2199The special treatment for leaf functions generally applies only when 2200other conditions are met; for example, often they may use only those 2201registers for its own variables and temporaries. We use the term ``leaf 2202function'' to mean a function that is suitable for this special 2203handling, so that functions with no calls are not necessarily ``leaf 2204functions''. 2205 2206GCC assigns register numbers before it knows whether the function is 2207suitable for leaf function treatment. So it needs to renumber the 2208registers in order to output a leaf function. The following macros 2209accomplish this. 2210 2211@defmac LEAF_REGISTERS 2212Name of a char vector, indexed by hard register number, which 2213contains 1 for a register that is allowable in a candidate for leaf 2214function treatment. 2215 2216If leaf function treatment involves renumbering the registers, then the 2217registers marked here should be the ones before renumbering---those that 2218GCC would ordinarily allocate. The registers which will actually be 2219used in the assembler code, after renumbering, should not be marked with 1 2220in this vector. 2221 2222Define this macro only if the target machine offers a way to optimize 2223the treatment of leaf functions. 2224@end defmac 2225 2226@defmac LEAF_REG_REMAP (@var{regno}) 2227A C expression whose value is the register number to which @var{regno} 2228should be renumbered, when a function is treated as a leaf function. 2229 2230If @var{regno} is a register number which should not appear in a leaf 2231function before renumbering, then the expression should yield @minus{}1, which 2232will cause the compiler to abort. 2233 2234Define this macro only if the target machine offers a way to optimize the 2235treatment of leaf functions, and registers need to be renumbered to do 2236this. 2237@end defmac 2238 2239@findex current_function_is_leaf 2240@findex current_function_uses_only_leaf_regs 2241@code{TARGET_ASM_FUNCTION_PROLOGUE} and 2242@code{TARGET_ASM_FUNCTION_EPILOGUE} must usually treat leaf functions 2243specially. They can test the C variable @code{current_function_is_leaf} 2244which is nonzero for leaf functions. @code{current_function_is_leaf} is 2245set prior to local register allocation and is valid for the remaining 2246compiler passes. They can also test the C variable 2247@code{current_function_uses_only_leaf_regs} which is nonzero for leaf 2248functions which only use leaf registers. 2249@code{current_function_uses_only_leaf_regs} is valid after all passes 2250that modify the instructions have been run and is only useful if 2251@code{LEAF_REGISTERS} is defined. 2252@c changed this to fix overfull. ALSO: why the "it" at the beginning 2253@c of the next paragraph?! --mew 2feb93 2254 2255@node Stack Registers 2256@subsection Registers That Form a Stack 2257 2258There are special features to handle computers where some of the 2259``registers'' form a stack. Stack registers are normally written by 2260pushing onto the stack, and are numbered relative to the top of the 2261stack. 2262 2263Currently, GCC can only handle one group of stack-like registers, and 2264they must be consecutively numbered. Furthermore, the existing 2265support for stack-like registers is specific to the 80387 floating 2266point coprocessor. If you have a new architecture that uses 2267stack-like registers, you will need to do substantial work on 2268@file{reg-stack.c} and write your machine description to cooperate 2269with it, as well as defining these macros. 2270 2271@defmac STACK_REGS 2272Define this if the machine has any stack-like registers. 2273@end defmac 2274 2275@defmac STACK_REG_COVER_CLASS 2276This is a cover class containing the stack registers. Define this if 2277the machine has any stack-like registers. 2278@end defmac 2279 2280@defmac FIRST_STACK_REG 2281The number of the first stack-like register. This one is the top 2282of the stack. 2283@end defmac 2284 2285@defmac LAST_STACK_REG 2286The number of the last stack-like register. This one is the bottom of 2287the stack. 2288@end defmac 2289 2290@node Register Classes 2291@section Register Classes 2292@cindex register class definitions 2293@cindex class definitions, register 2294 2295On many machines, the numbered registers are not all equivalent. 2296For example, certain registers may not be allowed for indexed addressing; 2297certain registers may not be allowed in some instructions. These machine 2298restrictions are described to the compiler using @dfn{register classes}. 2299 2300You define a number of register classes, giving each one a name and saying 2301which of the registers belong to it. Then you can specify register classes 2302that are allowed as operands to particular instruction patterns. 2303 2304@findex ALL_REGS 2305@findex NO_REGS 2306In general, each register will belong to several classes. In fact, one 2307class must be named @code{ALL_REGS} and contain all the registers. Another 2308class must be named @code{NO_REGS} and contain no registers. Often the 2309union of two classes will be another class; however, this is not required. 2310 2311@findex GENERAL_REGS 2312One of the classes must be named @code{GENERAL_REGS}. There is nothing 2313terribly special about the name, but the operand constraint letters 2314@samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is 2315the same as @code{ALL_REGS}, just define it as a macro which expands 2316to @code{ALL_REGS}. 2317 2318Order the classes so that if class @var{x} is contained in class @var{y} 2319then @var{x} has a lower class number than @var{y}. 2320 2321The way classes other than @code{GENERAL_REGS} are specified in operand 2322constraints is through machine-dependent operand constraint letters. 2323You can define such letters to correspond to various classes, then use 2324them in operand constraints. 2325 2326You must define the narrowest register classes for allocatable 2327registers, so that each class either has no subclasses, or that for 2328some mode, the move cost between registers within the class is 2329cheaper than moving a register in the class to or from memory 2330(@pxref{Costs}). 2331 2332You should define a class for the union of two classes whenever some 2333instruction allows both classes. For example, if an instruction allows 2334either a floating point (coprocessor) register or a general register for a 2335certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS} 2336which includes both of them. Otherwise you will get suboptimal code, 2337or even internal compiler errors when reload cannot find a register in the 2338class computed via @code{reg_class_subunion}. 2339 2340You must also specify certain redundant information about the register 2341classes: for each class, which classes contain it and which ones are 2342contained in it; for each pair of classes, the largest class contained 2343in their union. 2344 2345When a value occupying several consecutive registers is expected in a 2346certain class, all the registers used must belong to that class. 2347Therefore, register classes cannot be used to enforce a requirement for 2348a register pair to start with an even-numbered register. The way to 2349specify this requirement is with @code{HARD_REGNO_MODE_OK}. 2350 2351Register classes used for input-operands of bitwise-and or shift 2352instructions have a special requirement: each such class must have, for 2353each fixed-point machine mode, a subclass whose registers can transfer that 2354mode to or from memory. For example, on some machines, the operations for 2355single-byte values (@code{QImode}) are limited to certain registers. When 2356this is so, each register class that is used in a bitwise-and or shift 2357instruction must have a subclass consisting of registers from which 2358single-byte values can be loaded or stored. This is so that 2359@code{PREFERRED_RELOAD_CLASS} can always have a possible value to return. 2360 2361@deftp {Data type} {enum reg_class} 2362An enumerated type that must be defined with all the register class names 2363as enumerated values. @code{NO_REGS} must be first. @code{ALL_REGS} 2364must be the last register class, followed by one more enumerated value, 2365@code{LIM_REG_CLASSES}, which is not a register class but rather 2366tells how many classes there are. 2367 2368Each register class has a number, which is the value of casting 2369the class name to type @code{int}. The number serves as an index 2370in many of the tables described below. 2371@end deftp 2372 2373@defmac N_REG_CLASSES 2374The number of distinct register classes, defined as follows: 2375 2376@smallexample 2377#define N_REG_CLASSES (int) LIM_REG_CLASSES 2378@end smallexample 2379@end defmac 2380 2381@defmac REG_CLASS_NAMES 2382An initializer containing the names of the register classes as C string 2383constants. These names are used in writing some of the debugging dumps. 2384@end defmac 2385 2386@defmac REG_CLASS_CONTENTS 2387An initializer containing the contents of the register classes, as integers 2388which are bit masks. The @var{n}th integer specifies the contents of class 2389@var{n}. The way the integer @var{mask} is interpreted is that 2390register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1. 2391 2392When the machine has more than 32 registers, an integer does not suffice. 2393Then the integers are replaced by sub-initializers, braced groupings containing 2394several integers. Each sub-initializer must be suitable as an initializer 2395for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}. 2396In this situation, the first integer in each sub-initializer corresponds to 2397registers 0 through 31, the second integer to registers 32 through 63, and 2398so on. 2399@end defmac 2400 2401@defmac REGNO_REG_CLASS (@var{regno}) 2402A C expression whose value is a register class containing hard register 2403@var{regno}. In general there is more than one such class; choose a class 2404which is @dfn{minimal}, meaning that no smaller class also contains the 2405register. 2406@end defmac 2407 2408@defmac BASE_REG_CLASS 2409A macro whose definition is the name of the class to which a valid 2410base register must belong. A base register is one used in an address 2411which is the register value plus a displacement. 2412@end defmac 2413 2414@defmac MODE_BASE_REG_CLASS (@var{mode}) 2415This is a variation of the @code{BASE_REG_CLASS} macro which allows 2416the selection of a base register in a mode dependent manner. If 2417@var{mode} is VOIDmode then it should return the same value as 2418@code{BASE_REG_CLASS}. 2419@end defmac 2420 2421@defmac MODE_BASE_REG_REG_CLASS (@var{mode}) 2422A C expression whose value is the register class to which a valid 2423base register must belong in order to be used in a base plus index 2424register address. You should define this macro if base plus index 2425addresses have different requirements than other base register uses. 2426@end defmac 2427 2428@defmac MODE_CODE_BASE_REG_CLASS (@var{mode}, @var{address_space}, @var{outer_code}, @var{index_code}) 2429A C expression whose value is the register class to which a valid 2430base register for a memory reference in mode @var{mode} to address 2431space @var{address_space} must belong. @var{outer_code} and @var{index_code} 2432define the context in which the base register occurs. @var{outer_code} is 2433the code of the immediately enclosing expression (@code{MEM} for the top level 2434of an address, @code{ADDRESS} for something that occurs in an 2435@code{address_operand}). @var{index_code} is the code of the corresponding 2436index expression if @var{outer_code} is @code{PLUS}; @code{SCRATCH} otherwise. 2437@end defmac 2438 2439@defmac INDEX_REG_CLASS 2440A macro whose definition is the name of the class to which a valid 2441index register must belong. An index register is one used in an 2442address where its value is either multiplied by a scale factor or 2443added to another register (as well as added to a displacement). 2444@end defmac 2445 2446@defmac REGNO_OK_FOR_BASE_P (@var{num}) 2447A C expression which is nonzero if register number @var{num} is 2448suitable for use as a base register in operand addresses. 2449@end defmac 2450 2451@defmac REGNO_MODE_OK_FOR_BASE_P (@var{num}, @var{mode}) 2452A C expression that is just like @code{REGNO_OK_FOR_BASE_P}, except that 2453that expression may examine the mode of the memory reference in 2454@var{mode}. You should define this macro if the mode of the memory 2455reference affects whether a register may be used as a base register. If 2456you define this macro, the compiler will use it instead of 2457@code{REGNO_OK_FOR_BASE_P}. The mode may be @code{VOIDmode} for 2458addresses that appear outside a @code{MEM}, i.e., as an 2459@code{address_operand}. 2460@end defmac 2461 2462@defmac REGNO_MODE_OK_FOR_REG_BASE_P (@var{num}, @var{mode}) 2463A C expression which is nonzero if register number @var{num} is suitable for 2464use as a base register in base plus index operand addresses, accessing 2465memory in mode @var{mode}. It may be either a suitable hard register or a 2466pseudo register that has been allocated such a hard register. You should 2467define this macro if base plus index addresses have different requirements 2468than other base register uses. 2469 2470Use of this macro is deprecated; please use the more general 2471@code{REGNO_MODE_CODE_OK_FOR_BASE_P}. 2472@end defmac 2473 2474@defmac REGNO_MODE_CODE_OK_FOR_BASE_P (@var{num}, @var{mode}, @var{address_space}, @var{outer_code}, @var{index_code}) 2475A C expression which is nonzero if register number @var{num} is 2476suitable for use as a base register in operand addresses, accessing 2477memory in mode @var{mode} in address space @var{address_space}. 2478This is similar to @code{REGNO_MODE_OK_FOR_BASE_P}, except 2479that that expression may examine the context in which the register 2480appears in the memory reference. @var{outer_code} is the code of the 2481immediately enclosing expression (@code{MEM} if at the top level of the 2482address, @code{ADDRESS} for something that occurs in an 2483@code{address_operand}). @var{index_code} is the code of the 2484corresponding index expression if @var{outer_code} is @code{PLUS}; 2485@code{SCRATCH} otherwise. The mode may be @code{VOIDmode} for addresses 2486that appear outside a @code{MEM}, i.e., as an @code{address_operand}. 2487@end defmac 2488 2489@defmac REGNO_OK_FOR_INDEX_P (@var{num}) 2490A C expression which is nonzero if register number @var{num} is 2491suitable for use as an index register in operand addresses. It may be 2492either a suitable hard register or a pseudo register that has been 2493allocated such a hard register. 2494 2495The difference between an index register and a base register is that 2496the index register may be scaled. If an address involves the sum of 2497two registers, neither one of them scaled, then either one may be 2498labeled the ``base'' and the other the ``index''; but whichever 2499labeling is used must fit the machine's constraints of which registers 2500may serve in each capacity. The compiler will try both labelings, 2501looking for one that is valid, and will reload one or both registers 2502only if neither labeling works. 2503@end defmac 2504 2505@hook TARGET_PREFERRED_RENAME_CLASS 2506 2507@hook TARGET_PREFERRED_RELOAD_CLASS 2508A target hook that places additional restrictions on the register class 2509to use when it is necessary to copy value @var{x} into a register in class 2510@var{rclass}. The value is a register class; perhaps @var{rclass}, or perhaps 2511another, smaller class. 2512 2513The default version of this hook always returns value of @code{rclass} argument. 2514 2515Sometimes returning a more restrictive class makes better code. For 2516example, on the 68000, when @var{x} is an integer constant that is in range 2517for a @samp{moveq} instruction, the value of this macro is always 2518@code{DATA_REGS} as long as @var{rclass} includes the data registers. 2519Requiring a data register guarantees that a @samp{moveq} will be used. 2520 2521One case where @code{TARGET_PREFERRED_RELOAD_CLASS} must not return 2522@var{rclass} is if @var{x} is a legitimate constant which cannot be 2523loaded into some register class. By returning @code{NO_REGS} you can 2524force @var{x} into a memory location. For example, rs6000 can load 2525immediate values into general-purpose registers, but does not have an 2526instruction for loading an immediate value into a floating-point 2527register, so @code{TARGET_PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when 2528@var{x} is a floating-point constant. If the constant can't be loaded 2529into any kind of register, code generation will be better if 2530@code{TARGET_LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead 2531of using @code{TARGET_PREFERRED_RELOAD_CLASS}. 2532 2533If an insn has pseudos in it after register allocation, reload will go 2534through the alternatives and call repeatedly @code{TARGET_PREFERRED_RELOAD_CLASS} 2535to find the best one. Returning @code{NO_REGS}, in this case, makes 2536reload add a @code{!} in front of the constraint: the x86 back-end uses 2537this feature to discourage usage of 387 registers when math is done in 2538the SSE registers (and vice versa). 2539@end deftypefn 2540 2541@defmac PREFERRED_RELOAD_CLASS (@var{x}, @var{class}) 2542A C expression that places additional restrictions on the register class 2543to use when it is necessary to copy value @var{x} into a register in class 2544@var{class}. The value is a register class; perhaps @var{class}, or perhaps 2545another, smaller class. On many machines, the following definition is 2546safe: 2547 2548@smallexample 2549#define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS 2550@end smallexample 2551 2552Sometimes returning a more restrictive class makes better code. For 2553example, on the 68000, when @var{x} is an integer constant that is in range 2554for a @samp{moveq} instruction, the value of this macro is always 2555@code{DATA_REGS} as long as @var{class} includes the data registers. 2556Requiring a data register guarantees that a @samp{moveq} will be used. 2557 2558One case where @code{PREFERRED_RELOAD_CLASS} must not return 2559@var{class} is if @var{x} is a legitimate constant which cannot be 2560loaded into some register class. By returning @code{NO_REGS} you can 2561force @var{x} into a memory location. For example, rs6000 can load 2562immediate values into general-purpose registers, but does not have an 2563instruction for loading an immediate value into a floating-point 2564register, so @code{PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when 2565@var{x} is a floating-point constant. If the constant can't be loaded 2566into any kind of register, code generation will be better if 2567@code{TARGET_LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead 2568of using @code{TARGET_PREFERRED_RELOAD_CLASS}. 2569 2570If an insn has pseudos in it after register allocation, reload will go 2571through the alternatives and call repeatedly @code{PREFERRED_RELOAD_CLASS} 2572to find the best one. Returning @code{NO_REGS}, in this case, makes 2573reload add a @code{!} in front of the constraint: the x86 back-end uses 2574this feature to discourage usage of 387 registers when math is done in 2575the SSE registers (and vice versa). 2576@end defmac 2577 2578@hook TARGET_PREFERRED_OUTPUT_RELOAD_CLASS 2579Like @code{TARGET_PREFERRED_RELOAD_CLASS}, but for output reloads instead of 2580input reloads. 2581 2582The default version of this hook always returns value of @code{rclass} 2583argument. 2584 2585You can also use @code{TARGET_PREFERRED_OUTPUT_RELOAD_CLASS} to discourage 2586reload from using some alternatives, like @code{TARGET_PREFERRED_RELOAD_CLASS}. 2587@end deftypefn 2588 2589@defmac LIMIT_RELOAD_CLASS (@var{mode}, @var{class}) 2590A C expression that places additional restrictions on the register class 2591to use when it is necessary to be able to hold a value of mode 2592@var{mode} in a reload register for which class @var{class} would 2593ordinarily be used. 2594 2595Unlike @code{PREFERRED_RELOAD_CLASS}, this macro should be used when 2596there are certain modes that simply can't go in certain reload classes. 2597 2598The value is a register class; perhaps @var{class}, or perhaps another, 2599smaller class. 2600 2601Don't define this macro unless the target machine has limitations which 2602require the macro to do something nontrivial. 2603@end defmac 2604 2605@hook TARGET_SECONDARY_RELOAD 2606Many machines have some registers that cannot be copied directly to or 2607from memory or even from other types of registers. An example is the 2608@samp{MQ} register, which on most machines, can only be copied to or 2609from general registers, but not memory. Below, we shall be using the 2610term 'intermediate register' when a move operation cannot be performed 2611directly, but has to be done by copying the source into the intermediate 2612register first, and then copying the intermediate register to the 2613destination. An intermediate register always has the same mode as 2614source and destination. Since it holds the actual value being copied, 2615reload might apply optimizations to re-use an intermediate register 2616and eliding the copy from the source when it can determine that the 2617intermediate register still holds the required value. 2618 2619Another kind of secondary reload is required on some machines which 2620allow copying all registers to and from memory, but require a scratch 2621register for stores to some memory locations (e.g., those with symbolic 2622address on the RT, and those with certain symbolic address on the SPARC 2623when compiling PIC)@. Scratch registers need not have the same mode 2624as the value being copied, and usually hold a different value than 2625that being copied. Special patterns in the md file are needed to 2626describe how the copy is performed with the help of the scratch register; 2627these patterns also describe the number, register class(es) and mode(s) 2628of the scratch register(s). 2629 2630In some cases, both an intermediate and a scratch register are required. 2631 2632For input reloads, this target hook is called with nonzero @var{in_p}, 2633and @var{x} is an rtx that needs to be copied to a register of class 2634@var{reload_class} in @var{reload_mode}. For output reloads, this target 2635hook is called with zero @var{in_p}, and a register of class @var{reload_class} 2636needs to be copied to rtx @var{x} in @var{reload_mode}. 2637 2638If copying a register of @var{reload_class} from/to @var{x} requires 2639an intermediate register, the hook @code{secondary_reload} should 2640return the register class required for this intermediate register. 2641If no intermediate register is required, it should return NO_REGS. 2642If more than one intermediate register is required, describe the one 2643that is closest in the copy chain to the reload register. 2644 2645If scratch registers are needed, you also have to describe how to 2646perform the copy from/to the reload register to/from this 2647closest intermediate register. Or if no intermediate register is 2648required, but still a scratch register is needed, describe the 2649copy from/to the reload register to/from the reload operand @var{x}. 2650 2651You do this by setting @code{sri->icode} to the instruction code of a pattern 2652in the md file which performs the move. Operands 0 and 1 are the output 2653and input of this copy, respectively. Operands from operand 2 onward are 2654for scratch operands. These scratch operands must have a mode, and a 2655single-register-class 2656@c [later: or memory] 2657output constraint. 2658 2659When an intermediate register is used, the @code{secondary_reload} 2660hook will be called again to determine how to copy the intermediate 2661register to/from the reload operand @var{x}, so your hook must also 2662have code to handle the register class of the intermediate operand. 2663 2664@c [For later: maybe we'll allow multi-alternative reload patterns - 2665@c the port maintainer could name a mov<mode> pattern that has clobbers - 2666@c and match the constraints of input and output to determine the required 2667@c alternative. A restriction would be that constraints used to match 2668@c against reloads registers would have to be written as register class 2669@c constraints, or we need a new target macro / hook that tells us if an 2670@c arbitrary constraint can match an unknown register of a given class. 2671@c Such a macro / hook would also be useful in other places.] 2672 2673 2674@var{x} might be a pseudo-register or a @code{subreg} of a 2675pseudo-register, which could either be in a hard register or in memory. 2676Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is 2677in memory and the hard register number if it is in a register. 2678 2679Scratch operands in memory (constraint @code{"=m"} / @code{"=&m"}) are 2680currently not supported. For the time being, you will have to continue 2681to use @code{SECONDARY_MEMORY_NEEDED} for that purpose. 2682 2683@code{copy_cost} also uses this target hook to find out how values are 2684copied. If you want it to include some extra cost for the need to allocate 2685(a) scratch register(s), set @code{sri->extra_cost} to the additional cost. 2686Or if two dependent moves are supposed to have a lower cost than the sum 2687of the individual moves due to expected fortuitous scheduling and/or special 2688forwarding logic, you can set @code{sri->extra_cost} to a negative amount. 2689@end deftypefn 2690 2691@defmac SECONDARY_RELOAD_CLASS (@var{class}, @var{mode}, @var{x}) 2692@defmacx SECONDARY_INPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x}) 2693@defmacx SECONDARY_OUTPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x}) 2694These macros are obsolete, new ports should use the target hook 2695@code{TARGET_SECONDARY_RELOAD} instead. 2696 2697These are obsolete macros, replaced by the @code{TARGET_SECONDARY_RELOAD} 2698target hook. Older ports still define these macros to indicate to the 2699reload phase that it may 2700need to allocate at least one register for a reload in addition to the 2701register to contain the data. Specifically, if copying @var{x} to a 2702register @var{class} in @var{mode} requires an intermediate register, 2703you were supposed to define @code{SECONDARY_INPUT_RELOAD_CLASS} to return the 2704largest register class all of whose registers can be used as 2705intermediate registers or scratch registers. 2706 2707If copying a register @var{class} in @var{mode} to @var{x} requires an 2708intermediate or scratch register, @code{SECONDARY_OUTPUT_RELOAD_CLASS} 2709was supposed to be defined be defined to return the largest register 2710class required. If the 2711requirements for input and output reloads were the same, the macro 2712@code{SECONDARY_RELOAD_CLASS} should have been used instead of defining both 2713macros identically. 2714 2715The values returned by these macros are often @code{GENERAL_REGS}. 2716Return @code{NO_REGS} if no spare register is needed; i.e., if @var{x} 2717can be directly copied to or from a register of @var{class} in 2718@var{mode} without requiring a scratch register. Do not define this 2719macro if it would always return @code{NO_REGS}. 2720 2721If a scratch register is required (either with or without an 2722intermediate register), you were supposed to define patterns for 2723@samp{reload_in@var{m}} or @samp{reload_out@var{m}}, as required 2724(@pxref{Standard Names}. These patterns, which were normally 2725implemented with a @code{define_expand}, should be similar to the 2726@samp{mov@var{m}} patterns, except that operand 2 is the scratch 2727register. 2728 2729These patterns need constraints for the reload register and scratch 2730register that 2731contain a single register class. If the original reload register (whose 2732class is @var{class}) can meet the constraint given in the pattern, the 2733value returned by these macros is used for the class of the scratch 2734register. Otherwise, two additional reload registers are required. 2735Their classes are obtained from the constraints in the insn pattern. 2736 2737@var{x} might be a pseudo-register or a @code{subreg} of a 2738pseudo-register, which could either be in a hard register or in memory. 2739Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is 2740in memory and the hard register number if it is in a register. 2741 2742These macros should not be used in the case where a particular class of 2743registers can only be copied to memory and not to another class of 2744registers. In that case, secondary reload registers are not needed and 2745would not be helpful. Instead, a stack location must be used to perform 2746the copy and the @code{mov@var{m}} pattern should use memory as an 2747intermediate storage. This case often occurs between floating-point and 2748general registers. 2749@end defmac 2750 2751@defmac SECONDARY_MEMORY_NEEDED (@var{class1}, @var{class2}, @var{m}) 2752Certain machines have the property that some registers cannot be copied 2753to some other registers without using memory. Define this macro on 2754those machines to be a C expression that is nonzero if objects of mode 2755@var{m} in registers of @var{class1} can only be copied to registers of 2756class @var{class2} by storing a register of @var{class1} into memory 2757and loading that memory location into a register of @var{class2}. 2758 2759Do not define this macro if its value would always be zero. 2760@end defmac 2761 2762@defmac SECONDARY_MEMORY_NEEDED_RTX (@var{mode}) 2763Normally when @code{SECONDARY_MEMORY_NEEDED} is defined, the compiler 2764allocates a stack slot for a memory location needed for register copies. 2765If this macro is defined, the compiler instead uses the memory location 2766defined by this macro. 2767 2768Do not define this macro if you do not define 2769@code{SECONDARY_MEMORY_NEEDED}. 2770@end defmac 2771 2772@defmac SECONDARY_MEMORY_NEEDED_MODE (@var{mode}) 2773When the compiler needs a secondary memory location to copy between two 2774registers of mode @var{mode}, it normally allocates sufficient memory to 2775hold a quantity of @code{BITS_PER_WORD} bits and performs the store and 2776load operations in a mode that many bits wide and whose class is the 2777same as that of @var{mode}. 2778 2779This is right thing to do on most machines because it ensures that all 2780bits of the register are copied and prevents accesses to the registers 2781in a narrower mode, which some machines prohibit for floating-point 2782registers. 2783 2784However, this default behavior is not correct on some machines, such as 2785the DEC Alpha, that store short integers in floating-point registers 2786differently than in integer registers. On those machines, the default 2787widening will not work correctly and you must define this macro to 2788suppress that widening in some cases. See the file @file{alpha.h} for 2789details. 2790 2791Do not define this macro if you do not define 2792@code{SECONDARY_MEMORY_NEEDED} or if widening @var{mode} to a mode that 2793is @code{BITS_PER_WORD} bits wide is correct for your machine. 2794@end defmac 2795 2796@hook TARGET_CLASS_LIKELY_SPILLED_P 2797A target hook which returns @code{true} if pseudos that have been assigned 2798to registers of class @var{rclass} would likely be spilled because 2799registers of @var{rclass} are needed for spill registers. 2800 2801The default version of this target hook returns @code{true} if @var{rclass} 2802has exactly one register and @code{false} otherwise. On most machines, this 2803default should be used. Only use this target hook to some other expression 2804if pseudos allocated by @file{local-alloc.c} end up in memory because their 2805hard registers were needed for spill registers. If this target hook returns 2806@code{false} for those classes, those pseudos will only be allocated by 2807@file{global.c}, which knows how to reallocate the pseudo to another 2808register. If there would not be another register available for reallocation, 2809you should not change the implementation of this target hook since 2810the only effect of such implementation would be to slow down register 2811allocation. 2812@end deftypefn 2813 2814@hook TARGET_CLASS_MAX_NREGS 2815A target hook returns the maximum number of consecutive registers 2816of class @var{rclass} needed to hold a value of mode @var{mode}. 2817 2818This is closely related to the macro @code{HARD_REGNO_NREGS}. In fact, 2819the value returned by @code{TARGET_CLASS_MAX_NREGS (@var{rclass}, 2820@var{mode})} target hook should be the maximum value of 2821@code{HARD_REGNO_NREGS (@var{regno}, @var{mode})} for all @var{regno} 2822values in the class @var{rclass}. 2823 2824This target hook helps control the handling of multiple-word values 2825in the reload pass. 2826 2827The default version of this target hook returns the size of @var{mode} 2828in words. 2829@end deftypefn 2830 2831@defmac CLASS_MAX_NREGS (@var{class}, @var{mode}) 2832A C expression for the maximum number of consecutive registers 2833of class @var{class} needed to hold a value of mode @var{mode}. 2834 2835This is closely related to the macro @code{HARD_REGNO_NREGS}. In fact, 2836the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})} 2837should be the maximum value of @code{HARD_REGNO_NREGS (@var{regno}, 2838@var{mode})} for all @var{regno} values in the class @var{class}. 2839 2840This macro helps control the handling of multiple-word values 2841in the reload pass. 2842@end defmac 2843 2844@defmac CANNOT_CHANGE_MODE_CLASS (@var{from}, @var{to}, @var{class}) 2845If defined, a C expression that returns nonzero for a @var{class} for which 2846a change from mode @var{from} to mode @var{to} is invalid. 2847 2848For the example, loading 32-bit integer or floating-point objects into 2849floating-point registers on the Alpha extends them to 64 bits. 2850Therefore loading a 64-bit object and then storing it as a 32-bit object 2851does not store the low-order 32 bits, as would be the case for a normal 2852register. Therefore, @file{alpha.h} defines @code{CANNOT_CHANGE_MODE_CLASS} 2853as below: 2854 2855@smallexample 2856#define CANNOT_CHANGE_MODE_CLASS(FROM, TO, CLASS) \ 2857 (GET_MODE_SIZE (FROM) != GET_MODE_SIZE (TO) \ 2858 ? reg_classes_intersect_p (FLOAT_REGS, (CLASS)) : 0) 2859@end smallexample 2860@end defmac 2861 2862@node Old Constraints 2863@section Obsolete Macros for Defining Constraints 2864@cindex defining constraints, obsolete method 2865@cindex constraints, defining, obsolete method 2866 2867Machine-specific constraints can be defined with these macros instead 2868of the machine description constructs described in @ref{Define 2869Constraints}. This mechanism is obsolete. New ports should not use 2870it; old ports should convert to the new mechanism. 2871 2872@defmac CONSTRAINT_LEN (@var{char}, @var{str}) 2873For the constraint at the start of @var{str}, which starts with the letter 2874@var{c}, return the length. This allows you to have register class / 2875constant / extra constraints that are longer than a single letter; 2876you don't need to define this macro if you can do with single-letter 2877constraints only. The definition of this macro should use 2878DEFAULT_CONSTRAINT_LEN for all the characters that you don't want 2879to handle specially. 2880There are some sanity checks in genoutput.c that check the constraint lengths 2881for the md file, so you can also use this macro to help you while you are 2882transitioning from a byzantine single-letter-constraint scheme: when you 2883return a negative length for a constraint you want to re-use, genoutput 2884will complain about every instance where it is used in the md file. 2885@end defmac 2886 2887@defmac REG_CLASS_FROM_LETTER (@var{char}) 2888A C expression which defines the machine-dependent operand constraint 2889letters for register classes. If @var{char} is such a letter, the 2890value should be the register class corresponding to it. Otherwise, 2891the value should be @code{NO_REGS}. The register letter @samp{r}, 2892corresponding to class @code{GENERAL_REGS}, will not be passed 2893to this macro; you do not need to handle it. 2894@end defmac 2895 2896@defmac REG_CLASS_FROM_CONSTRAINT (@var{char}, @var{str}) 2897Like @code{REG_CLASS_FROM_LETTER}, but you also get the constraint string 2898passed in @var{str}, so that you can use suffixes to distinguish between 2899different variants. 2900@end defmac 2901 2902@defmac CONST_OK_FOR_LETTER_P (@var{value}, @var{c}) 2903A C expression that defines the machine-dependent operand constraint 2904letters (@samp{I}, @samp{J}, @samp{K}, @dots{} @samp{P}) that specify 2905particular ranges of integer values. If @var{c} is one of those 2906letters, the expression should check that @var{value}, an integer, is in 2907the appropriate range and return 1 if so, 0 otherwise. If @var{c} is 2908not one of those letters, the value should be 0 regardless of 2909@var{value}. 2910@end defmac 2911 2912@defmac CONST_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str}) 2913Like @code{CONST_OK_FOR_LETTER_P}, but you also get the constraint 2914string passed in @var{str}, so that you can use suffixes to distinguish 2915between different variants. 2916@end defmac 2917 2918@defmac CONST_DOUBLE_OK_FOR_LETTER_P (@var{value}, @var{c}) 2919A C expression that defines the machine-dependent operand constraint 2920letters that specify particular ranges of @code{const_double} values 2921(@samp{G} or @samp{H}). 2922 2923If @var{c} is one of those letters, the expression should check that 2924@var{value}, an RTX of code @code{const_double}, is in the appropriate 2925range and return 1 if so, 0 otherwise. If @var{c} is not one of those 2926letters, the value should be 0 regardless of @var{value}. 2927 2928@code{const_double} is used for all floating-point constants and for 2929@code{DImode} fixed-point constants. A given letter can accept either 2930or both kinds of values. It can use @code{GET_MODE} to distinguish 2931between these kinds. 2932@end defmac 2933 2934@defmac CONST_DOUBLE_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str}) 2935Like @code{CONST_DOUBLE_OK_FOR_LETTER_P}, but you also get the constraint 2936string passed in @var{str}, so that you can use suffixes to distinguish 2937between different variants. 2938@end defmac 2939 2940@defmac EXTRA_CONSTRAINT (@var{value}, @var{c}) 2941A C expression that defines the optional machine-dependent constraint 2942letters that can be used to segregate specific types of operands, usually 2943memory references, for the target machine. Any letter that is not 2944elsewhere defined and not matched by @code{REG_CLASS_FROM_LETTER} / 2945@code{REG_CLASS_FROM_CONSTRAINT} 2946may be used. Normally this macro will not be defined. 2947 2948If it is required for a particular target machine, it should return 1 2949if @var{value} corresponds to the operand type represented by the 2950constraint letter @var{c}. If @var{c} is not defined as an extra 2951constraint, the value returned should be 0 regardless of @var{value}. 2952 2953For example, on the ROMP, load instructions cannot have their output 2954in r0 if the memory reference contains a symbolic address. Constraint 2955letter @samp{Q} is defined as representing a memory address that does 2956@emph{not} contain a symbolic address. An alternative is specified with 2957a @samp{Q} constraint on the input and @samp{r} on the output. The next 2958alternative specifies @samp{m} on the input and a register class that 2959does not include r0 on the output. 2960@end defmac 2961 2962@defmac EXTRA_CONSTRAINT_STR (@var{value}, @var{c}, @var{str}) 2963Like @code{EXTRA_CONSTRAINT}, but you also get the constraint string passed 2964in @var{str}, so that you can use suffixes to distinguish between different 2965variants. 2966@end defmac 2967 2968@defmac EXTRA_MEMORY_CONSTRAINT (@var{c}, @var{str}) 2969A C expression that defines the optional machine-dependent constraint 2970letters, amongst those accepted by @code{EXTRA_CONSTRAINT}, that should 2971be treated like memory constraints by the reload pass. 2972 2973It should return 1 if the operand type represented by the constraint 2974at the start of @var{str}, the first letter of which is the letter @var{c}, 2975comprises a subset of all memory references including 2976all those whose address is simply a base register. This allows the reload 2977pass to reload an operand, if it does not directly correspond to the operand 2978type of @var{c}, by copying its address into a base register. 2979 2980For example, on the S/390, some instructions do not accept arbitrary 2981memory references, but only those that do not make use of an index 2982register. The constraint letter @samp{Q} is defined via 2983@code{EXTRA_CONSTRAINT} as representing a memory address of this type. 2984If the letter @samp{Q} is marked as @code{EXTRA_MEMORY_CONSTRAINT}, 2985a @samp{Q} constraint can handle any memory operand, because the 2986reload pass knows it can be reloaded by copying the memory address 2987into a base register if required. This is analogous to the way 2988an @samp{o} constraint can handle any memory operand. 2989@end defmac 2990 2991@defmac EXTRA_ADDRESS_CONSTRAINT (@var{c}, @var{str}) 2992A C expression that defines the optional machine-dependent constraint 2993letters, amongst those accepted by @code{EXTRA_CONSTRAINT} / 2994@code{EXTRA_CONSTRAINT_STR}, that should 2995be treated like address constraints by the reload pass. 2996 2997It should return 1 if the operand type represented by the constraint 2998at the start of @var{str}, which starts with the letter @var{c}, comprises 2999a subset of all memory addresses including 3000all those that consist of just a base register. This allows the reload 3001pass to reload an operand, if it does not directly correspond to the operand 3002type of @var{str}, by copying it into a base register. 3003 3004Any constraint marked as @code{EXTRA_ADDRESS_CONSTRAINT} can only 3005be used with the @code{address_operand} predicate. It is treated 3006analogously to the @samp{p} constraint. 3007@end defmac 3008 3009@node Stack and Calling 3010@section Stack Layout and Calling Conventions 3011@cindex calling conventions 3012 3013@c prevent bad page break with this line 3014This describes the stack layout and calling conventions. 3015 3016@menu 3017* Frame Layout:: 3018* Exception Handling:: 3019* Stack Checking:: 3020* Frame Registers:: 3021* Elimination:: 3022* Stack Arguments:: 3023* Register Arguments:: 3024* Scalar Return:: 3025* Aggregate Return:: 3026* Caller Saves:: 3027* Function Entry:: 3028* Profiling:: 3029* Tail Calls:: 3030* Stack Smashing Protection:: 3031@end menu 3032 3033@node Frame Layout 3034@subsection Basic Stack Layout 3035@cindex stack frame layout 3036@cindex frame layout 3037 3038@c prevent bad page break with this line 3039Here is the basic stack layout. 3040 3041@defmac STACK_GROWS_DOWNWARD 3042Define this macro if pushing a word onto the stack moves the stack 3043pointer to a smaller address. 3044 3045When we say, ``define this macro if @dots{}'', it means that the 3046compiler checks this macro only with @code{#ifdef} so the precise 3047definition used does not matter. 3048@end defmac 3049 3050@defmac STACK_PUSH_CODE 3051This macro defines the operation used when something is pushed 3052on the stack. In RTL, a push operation will be 3053@code{(set (mem (STACK_PUSH_CODE (reg sp))) @dots{})} 3054 3055The choices are @code{PRE_DEC}, @code{POST_DEC}, @code{PRE_INC}, 3056and @code{POST_INC}. Which of these is correct depends on 3057the stack direction and on whether the stack pointer points 3058to the last item on the stack or whether it points to the 3059space for the next item on the stack. 3060 3061The default is @code{PRE_DEC} when @code{STACK_GROWS_DOWNWARD} is 3062defined, which is almost always right, and @code{PRE_INC} otherwise, 3063which is often wrong. 3064@end defmac 3065 3066@defmac FRAME_GROWS_DOWNWARD 3067Define this macro to nonzero value if the addresses of local variable slots 3068are at negative offsets from the frame pointer. 3069@end defmac 3070 3071@defmac ARGS_GROW_DOWNWARD 3072Define this macro if successive arguments to a function occupy decreasing 3073addresses on the stack. 3074@end defmac 3075 3076@defmac STARTING_FRAME_OFFSET 3077Offset from the frame pointer to the first local variable slot to be allocated. 3078 3079If @code{FRAME_GROWS_DOWNWARD}, find the next slot's offset by 3080subtracting the first slot's length from @code{STARTING_FRAME_OFFSET}. 3081Otherwise, it is found by adding the length of the first slot to the 3082value @code{STARTING_FRAME_OFFSET}. 3083@c i'm not sure if the above is still correct.. had to change it to get 3084@c rid of an overfull. --mew 2feb93 3085@end defmac 3086 3087@defmac STACK_ALIGNMENT_NEEDED 3088Define to zero to disable final alignment of the stack during reload. 3089The nonzero default for this macro is suitable for most ports. 3090 3091On ports where @code{STARTING_FRAME_OFFSET} is nonzero or where there 3092is a register save block following the local block that doesn't require 3093alignment to @code{STACK_BOUNDARY}, it may be beneficial to disable 3094stack alignment and do it in the backend. 3095@end defmac 3096 3097@defmac STACK_POINTER_OFFSET 3098Offset from the stack pointer register to the first location at which 3099outgoing arguments are placed. If not specified, the default value of 3100zero is used. This is the proper value for most machines. 3101 3102If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above 3103the first location at which outgoing arguments are placed. 3104@end defmac 3105 3106@defmac FIRST_PARM_OFFSET (@var{fundecl}) 3107Offset from the argument pointer register to the first argument's 3108address. On some machines it may depend on the data type of the 3109function. 3110 3111If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above 3112the first argument's address. 3113@end defmac 3114 3115@defmac STACK_DYNAMIC_OFFSET (@var{fundecl}) 3116Offset from the stack pointer register to an item dynamically allocated 3117on the stack, e.g., by @code{alloca}. 3118 3119The default value for this macro is @code{STACK_POINTER_OFFSET} plus the 3120length of the outgoing arguments. The default is correct for most 3121machines. See @file{function.c} for details. 3122@end defmac 3123 3124@defmac INITIAL_FRAME_ADDRESS_RTX 3125A C expression whose value is RTL representing the address of the initial 3126stack frame. This address is passed to @code{RETURN_ADDR_RTX} and 3127@code{DYNAMIC_CHAIN_ADDRESS}. If you don't define this macro, a reasonable 3128default value will be used. Define this macro in order to make frame pointer 3129elimination work in the presence of @code{__builtin_frame_address (count)} and 3130@code{__builtin_return_address (count)} for @code{count} not equal to zero. 3131@end defmac 3132 3133@defmac DYNAMIC_CHAIN_ADDRESS (@var{frameaddr}) 3134A C expression whose value is RTL representing the address in a stack 3135frame where the pointer to the caller's frame is stored. Assume that 3136@var{frameaddr} is an RTL expression for the address of the stack frame 3137itself. 3138 3139If you don't define this macro, the default is to return the value 3140of @var{frameaddr}---that is, the stack frame address is also the 3141address of the stack word that points to the previous frame. 3142@end defmac 3143 3144@defmac SETUP_FRAME_ADDRESSES 3145If defined, a C expression that produces the machine-specific code to 3146setup the stack so that arbitrary frames can be accessed. For example, 3147on the SPARC, we must flush all of the register windows to the stack 3148before we can access arbitrary stack frames. You will seldom need to 3149define this macro. 3150@end defmac 3151 3152@hook TARGET_BUILTIN_SETJMP_FRAME_VALUE 3153This target hook should return an rtx that is used to store 3154the address of the current frame into the built in @code{setjmp} buffer. 3155The default value, @code{virtual_stack_vars_rtx}, is correct for most 3156machines. One reason you may need to define this target hook is if 3157@code{hard_frame_pointer_rtx} is the appropriate value on your machine. 3158@end deftypefn 3159 3160@defmac FRAME_ADDR_RTX (@var{frameaddr}) 3161A C expression whose value is RTL representing the value of the frame 3162address for the current frame. @var{frameaddr} is the frame pointer 3163of the current frame. This is used for __builtin_frame_address. 3164You need only define this macro if the frame address is not the same 3165as the frame pointer. Most machines do not need to define it. 3166@end defmac 3167 3168@defmac RETURN_ADDR_RTX (@var{count}, @var{frameaddr}) 3169A C expression whose value is RTL representing the value of the return 3170address for the frame @var{count} steps up from the current frame, after 3171the prologue. @var{frameaddr} is the frame pointer of the @var{count} 3172frame, or the frame pointer of the @var{count} @minus{} 1 frame if 3173@code{RETURN_ADDR_IN_PREVIOUS_FRAME} is defined. 3174 3175The value of the expression must always be the correct address when 3176@var{count} is zero, but may be @code{NULL_RTX} if there is no way to 3177determine the return address of other frames. 3178@end defmac 3179 3180@defmac RETURN_ADDR_IN_PREVIOUS_FRAME 3181Define this if the return address of a particular stack frame is accessed 3182from the frame pointer of the previous stack frame. 3183@end defmac 3184 3185@defmac INCOMING_RETURN_ADDR_RTX 3186A C expression whose value is RTL representing the location of the 3187incoming return address at the beginning of any function, before the 3188prologue. This RTL is either a @code{REG}, indicating that the return 3189value is saved in @samp{REG}, or a @code{MEM} representing a location in 3190the stack. 3191 3192You only need to define this macro if you want to support call frame 3193debugging information like that provided by DWARF 2. 3194 3195If this RTL is a @code{REG}, you should also define 3196@code{DWARF_FRAME_RETURN_COLUMN} to @code{DWARF_FRAME_REGNUM (REGNO)}. 3197@end defmac 3198 3199@defmac DWARF_ALT_FRAME_RETURN_COLUMN 3200A C expression whose value is an integer giving a DWARF 2 column 3201number that may be used as an alternative return column. The column 3202must not correspond to any gcc hard register (that is, it must not 3203be in the range of @code{DWARF_FRAME_REGNUM}). 3204 3205This macro can be useful if @code{DWARF_FRAME_RETURN_COLUMN} is set to a 3206general register, but an alternative column needs to be used for signal 3207frames. Some targets have also used different frame return columns 3208over time. 3209@end defmac 3210 3211@defmac DWARF_ZERO_REG 3212A C expression whose value is an integer giving a DWARF 2 register 3213number that is considered to always have the value zero. This should 3214only be defined if the target has an architected zero register, and 3215someone decided it was a good idea to use that register number to 3216terminate the stack backtrace. New ports should avoid this. 3217@end defmac 3218 3219@hook TARGET_DWARF_HANDLE_FRAME_UNSPEC 3220This target hook allows the backend to emit frame-related insns that 3221contain UNSPECs or UNSPEC_VOLATILEs. The DWARF 2 call frame debugging 3222info engine will invoke it on insns of the form 3223@smallexample 3224(set (reg) (unspec [@dots{}] UNSPEC_INDEX)) 3225@end smallexample 3226and 3227@smallexample 3228(set (reg) (unspec_volatile [@dots{}] UNSPECV_INDEX)). 3229@end smallexample 3230to let the backend emit the call frame instructions. @var{label} is 3231the CFI label attached to the insn, @var{pattern} is the pattern of 3232the insn and @var{index} is @code{UNSPEC_INDEX} or @code{UNSPECV_INDEX}. 3233@end deftypefn 3234 3235@defmac INCOMING_FRAME_SP_OFFSET 3236A C expression whose value is an integer giving the offset, in bytes, 3237from the value of the stack pointer register to the top of the stack 3238frame at the beginning of any function, before the prologue. The top of 3239the frame is defined to be the value of the stack pointer in the 3240previous frame, just before the call instruction. 3241 3242You only need to define this macro if you want to support call frame 3243debugging information like that provided by DWARF 2. 3244@end defmac 3245 3246@defmac ARG_POINTER_CFA_OFFSET (@var{fundecl}) 3247A C expression whose value is an integer giving the offset, in bytes, 3248from the argument pointer to the canonical frame address (cfa). The 3249final value should coincide with that calculated by 3250@code{INCOMING_FRAME_SP_OFFSET}. Which is unfortunately not usable 3251during virtual register instantiation. 3252 3253The default value for this macro is 3254@code{FIRST_PARM_OFFSET (fundecl) + crtl->args.pretend_args_size}, 3255which is correct for most machines; in general, the arguments are found 3256immediately before the stack frame. Note that this is not the case on 3257some targets that save registers into the caller's frame, such as SPARC 3258and rs6000, and so such targets need to define this macro. 3259 3260You only need to define this macro if the default is incorrect, and you 3261want to support call frame debugging information like that provided by 3262DWARF 2. 3263@end defmac 3264 3265@defmac FRAME_POINTER_CFA_OFFSET (@var{fundecl}) 3266If defined, a C expression whose value is an integer giving the offset 3267in bytes from the frame pointer to the canonical frame address (cfa). 3268The final value should coincide with that calculated by 3269@code{INCOMING_FRAME_SP_OFFSET}. 3270 3271Normally the CFA is calculated as an offset from the argument pointer, 3272via @code{ARG_POINTER_CFA_OFFSET}, but if the argument pointer is 3273variable due to the ABI, this may not be possible. If this macro is 3274defined, it implies that the virtual register instantiation should be 3275based on the frame pointer instead of the argument pointer. Only one 3276of @code{FRAME_POINTER_CFA_OFFSET} and @code{ARG_POINTER_CFA_OFFSET} 3277should be defined. 3278@end defmac 3279 3280@defmac CFA_FRAME_BASE_OFFSET (@var{fundecl}) 3281If defined, a C expression whose value is an integer giving the offset 3282in bytes from the canonical frame address (cfa) to the frame base used 3283in DWARF 2 debug information. The default is zero. A different value 3284may reduce the size of debug information on some ports. 3285@end defmac 3286 3287@node Exception Handling 3288@subsection Exception Handling Support 3289@cindex exception handling 3290 3291@defmac EH_RETURN_DATA_REGNO (@var{N}) 3292A C expression whose value is the @var{N}th register number used for 3293data by exception handlers, or @code{INVALID_REGNUM} if fewer than 3294@var{N} registers are usable. 3295 3296The exception handling library routines communicate with the exception 3297handlers via a set of agreed upon registers. Ideally these registers 3298should be call-clobbered; it is possible to use call-saved registers, 3299but may negatively impact code size. The target must support at least 33002 data registers, but should define 4 if there are enough free registers. 3301 3302You must define this macro if you want to support call frame exception 3303handling like that provided by DWARF 2. 3304@end defmac 3305 3306@defmac EH_RETURN_STACKADJ_RTX 3307A C expression whose value is RTL representing a location in which 3308to store a stack adjustment to be applied before function return. 3309This is used to unwind the stack to an exception handler's call frame. 3310It will be assigned zero on code paths that return normally. 3311 3312Typically this is a call-clobbered hard register that is otherwise 3313untouched by the epilogue, but could also be a stack slot. 3314 3315Do not define this macro if the stack pointer is saved and restored 3316by the regular prolog and epilog code in the call frame itself; in 3317this case, the exception handling library routines will update the 3318stack location to be restored in place. Otherwise, you must define 3319this macro if you want to support call frame exception handling like 3320that provided by DWARF 2. 3321@end defmac 3322 3323@defmac EH_RETURN_HANDLER_RTX 3324A C expression whose value is RTL representing a location in which 3325to store the address of an exception handler to which we should 3326return. It will not be assigned on code paths that return normally. 3327 3328Typically this is the location in the call frame at which the normal 3329return address is stored. For targets that return by popping an 3330address off the stack, this might be a memory address just below 3331the @emph{target} call frame rather than inside the current call 3332frame. If defined, @code{EH_RETURN_STACKADJ_RTX} will have already 3333been assigned, so it may be used to calculate the location of the 3334target call frame. 3335 3336Some targets have more complex requirements than storing to an 3337address calculable during initial code generation. In that case 3338the @code{eh_return} instruction pattern should be used instead. 3339 3340If you want to support call frame exception handling, you must 3341define either this macro or the @code{eh_return} instruction pattern. 3342@end defmac 3343 3344@defmac RETURN_ADDR_OFFSET 3345If defined, an integer-valued C expression for which rtl will be generated 3346to add it to the exception handler address before it is searched in the 3347exception handling tables, and to subtract it again from the address before 3348using it to return to the exception handler. 3349@end defmac 3350 3351@defmac ASM_PREFERRED_EH_DATA_FORMAT (@var{code}, @var{global}) 3352This macro chooses the encoding of pointers embedded in the exception 3353handling sections. If at all possible, this should be defined such 3354that the exception handling section will not require dynamic relocations, 3355and so may be read-only. 3356 3357@var{code} is 0 for data, 1 for code labels, 2 for function pointers. 3358@var{global} is true if the symbol may be affected by dynamic relocations. 3359The macro should return a combination of the @code{DW_EH_PE_*} defines 3360as found in @file{dwarf2.h}. 3361 3362If this macro is not defined, pointers will not be encoded but 3363represented directly. 3364@end defmac 3365 3366@defmac ASM_MAYBE_OUTPUT_ENCODED_ADDR_RTX (@var{file}, @var{encoding}, @var{size}, @var{addr}, @var{done}) 3367This macro allows the target to emit whatever special magic is required 3368to represent the encoding chosen by @code{ASM_PREFERRED_EH_DATA_FORMAT}. 3369Generic code takes care of pc-relative and indirect encodings; this must 3370be defined if the target uses text-relative or data-relative encodings. 3371 3372This is a C statement that branches to @var{done} if the format was 3373handled. @var{encoding} is the format chosen, @var{size} is the number 3374of bytes that the format occupies, @var{addr} is the @code{SYMBOL_REF} 3375to be emitted. 3376@end defmac 3377 3378@defmac MD_FALLBACK_FRAME_STATE_FOR (@var{context}, @var{fs}) 3379This macro allows the target to add CPU and operating system specific 3380code to the call-frame unwinder for use when there is no unwind data 3381available. The most common reason to implement this macro is to unwind 3382through signal frames. 3383 3384This macro is called from @code{uw_frame_state_for} in 3385@file{unwind-dw2.c}, @file{unwind-dw2-xtensa.c} and 3386@file{unwind-ia64.c}. @var{context} is an @code{_Unwind_Context}; 3387@var{fs} is an @code{_Unwind_FrameState}. Examine @code{context->ra} 3388for the address of the code being executed and @code{context->cfa} for 3389the stack pointer value. If the frame can be decoded, the register 3390save addresses should be updated in @var{fs} and the macro should 3391evaluate to @code{_URC_NO_REASON}. If the frame cannot be decoded, 3392the macro should evaluate to @code{_URC_END_OF_STACK}. 3393 3394For proper signal handling in Java this macro is accompanied by 3395@code{MAKE_THROW_FRAME}, defined in @file{libjava/include/*-signal.h} headers. 3396@end defmac 3397 3398@defmac MD_HANDLE_UNWABI (@var{context}, @var{fs}) 3399This macro allows the target to add operating system specific code to the 3400call-frame unwinder to handle the IA-64 @code{.unwabi} unwinding directive, 3401usually used for signal or interrupt frames. 3402 3403This macro is called from @code{uw_update_context} in @file{unwind-ia64.c}. 3404@var{context} is an @code{_Unwind_Context}; 3405@var{fs} is an @code{_Unwind_FrameState}. Examine @code{fs->unwabi} 3406for the abi and context in the @code{.unwabi} directive. If the 3407@code{.unwabi} directive can be handled, the register save addresses should 3408be updated in @var{fs}. 3409@end defmac 3410 3411@defmac TARGET_USES_WEAK_UNWIND_INFO 3412A C expression that evaluates to true if the target requires unwind 3413info to be given comdat linkage. Define it to be @code{1} if comdat 3414linkage is necessary. The default is @code{0}. 3415@end defmac 3416 3417@node Stack Checking 3418@subsection Specifying How Stack Checking is Done 3419 3420GCC will check that stack references are within the boundaries of the 3421stack, if the option @option{-fstack-check} is specified, in one of 3422three ways: 3423 3424@enumerate 3425@item 3426If the value of the @code{STACK_CHECK_BUILTIN} macro is nonzero, GCC 3427will assume that you have arranged for full stack checking to be done 3428at appropriate places in the configuration files. GCC will not do 3429other special processing. 3430 3431@item 3432If @code{STACK_CHECK_BUILTIN} is zero and the value of the 3433@code{STACK_CHECK_STATIC_BUILTIN} macro is nonzero, GCC will assume 3434that you have arranged for static stack checking (checking of the 3435static stack frame of functions) to be done at appropriate places 3436in the configuration files. GCC will only emit code to do dynamic 3437stack checking (checking on dynamic stack allocations) using the third 3438approach below. 3439 3440@item 3441If neither of the above are true, GCC will generate code to periodically 3442``probe'' the stack pointer using the values of the macros defined below. 3443@end enumerate 3444 3445If neither STACK_CHECK_BUILTIN nor STACK_CHECK_STATIC_BUILTIN is defined, 3446GCC will change its allocation strategy for large objects if the option 3447@option{-fstack-check} is specified: they will always be allocated 3448dynamically if their size exceeds @code{STACK_CHECK_MAX_VAR_SIZE} bytes. 3449 3450@defmac STACK_CHECK_BUILTIN 3451A nonzero value if stack checking is done by the configuration files in a 3452machine-dependent manner. You should define this macro if stack checking 3453is required by the ABI of your machine or if you would like to do stack 3454checking in some more efficient way than the generic approach. The default 3455value of this macro is zero. 3456@end defmac 3457 3458@defmac STACK_CHECK_STATIC_BUILTIN 3459A nonzero value if static stack checking is done by the configuration files 3460in a machine-dependent manner. You should define this macro if you would 3461like to do static stack checking in some more efficient way than the generic 3462approach. The default value of this macro is zero. 3463@end defmac 3464 3465@defmac STACK_CHECK_PROBE_INTERVAL_EXP 3466An integer specifying the interval at which GCC must generate stack probe 3467instructions, defined as 2 raised to this integer. You will normally 3468define this macro so that the interval be no larger than the size of 3469the ``guard pages'' at the end of a stack area. The default value 3470of 12 (4096-byte interval) is suitable for most systems. 3471@end defmac 3472 3473@defmac STACK_CHECK_MOVING_SP 3474An integer which is nonzero if GCC should move the stack pointer page by page 3475when doing probes. This can be necessary on systems where the stack pointer 3476contains the bottom address of the memory area accessible to the executing 3477thread at any point in time. In this situation an alternate signal stack 3478is required in order to be able to recover from a stack overflow. The 3479default value of this macro is zero. 3480@end defmac 3481 3482@defmac STACK_CHECK_PROTECT 3483The number of bytes of stack needed to recover from a stack overflow, for 3484languages where such a recovery is supported. The default value of 75 words 3485with the @code{setjmp}/@code{longjmp}-based exception handling mechanism and 34868192 bytes with other exception handling mechanisms should be adequate for 3487most machines. 3488@end defmac 3489 3490The following macros are relevant only if neither STACK_CHECK_BUILTIN 3491nor STACK_CHECK_STATIC_BUILTIN is defined; you can omit them altogether 3492in the opposite case. 3493 3494@defmac STACK_CHECK_MAX_FRAME_SIZE 3495The maximum size of a stack frame, in bytes. GCC will generate probe 3496instructions in non-leaf functions to ensure at least this many bytes of 3497stack are available. If a stack frame is larger than this size, stack 3498checking will not be reliable and GCC will issue a warning. The 3499default is chosen so that GCC only generates one instruction on most 3500systems. You should normally not change the default value of this macro. 3501@end defmac 3502 3503@defmac STACK_CHECK_FIXED_FRAME_SIZE 3504GCC uses this value to generate the above warning message. It 3505represents the amount of fixed frame used by a function, not including 3506space for any callee-saved registers, temporaries and user variables. 3507You need only specify an upper bound for this amount and will normally 3508use the default of four words. 3509@end defmac 3510 3511@defmac STACK_CHECK_MAX_VAR_SIZE 3512The maximum size, in bytes, of an object that GCC will place in the 3513fixed area of the stack frame when the user specifies 3514@option{-fstack-check}. 3515GCC computed the default from the values of the above macros and you will 3516normally not need to override that default. 3517@end defmac 3518 3519@need 2000 3520@node Frame Registers 3521@subsection Registers That Address the Stack Frame 3522 3523@c prevent bad page break with this line 3524This discusses registers that address the stack frame. 3525 3526@defmac STACK_POINTER_REGNUM 3527The register number of the stack pointer register, which must also be a 3528fixed register according to @code{FIXED_REGISTERS}. On most machines, 3529the hardware determines which register this is. 3530@end defmac 3531 3532@defmac FRAME_POINTER_REGNUM 3533The register number of the frame pointer register, which is used to 3534access automatic variables in the stack frame. On some machines, the 3535hardware determines which register this is. On other machines, you can 3536choose any register you wish for this purpose. 3537@end defmac 3538 3539@defmac HARD_FRAME_POINTER_REGNUM 3540On some machines the offset between the frame pointer and starting 3541offset of the automatic variables is not known until after register 3542allocation has been done (for example, because the saved registers are 3543between these two locations). On those machines, define 3544@code{FRAME_POINTER_REGNUM} the number of a special, fixed register to 3545be used internally until the offset is known, and define 3546@code{HARD_FRAME_POINTER_REGNUM} to be the actual hard register number 3547used for the frame pointer. 3548 3549You should define this macro only in the very rare circumstances when it 3550is not possible to calculate the offset between the frame pointer and 3551the automatic variables until after register allocation has been 3552completed. When this macro is defined, you must also indicate in your 3553definition of @code{ELIMINABLE_REGS} how to eliminate 3554@code{FRAME_POINTER_REGNUM} into either @code{HARD_FRAME_POINTER_REGNUM} 3555or @code{STACK_POINTER_REGNUM}. 3556 3557Do not define this macro if it would be the same as 3558@code{FRAME_POINTER_REGNUM}. 3559@end defmac 3560 3561@defmac ARG_POINTER_REGNUM 3562The register number of the arg pointer register, which is used to access 3563the function's argument list. On some machines, this is the same as the 3564frame pointer register. On some machines, the hardware determines which 3565register this is. On other machines, you can choose any register you 3566wish for this purpose. If this is not the same register as the frame 3567pointer register, then you must mark it as a fixed register according to 3568@code{FIXED_REGISTERS}, or arrange to be able to eliminate it 3569(@pxref{Elimination}). 3570@end defmac 3571 3572@defmac HARD_FRAME_POINTER_IS_FRAME_POINTER 3573Define this to a preprocessor constant that is nonzero if 3574@code{hard_frame_pointer_rtx} and @code{frame_pointer_rtx} should be 3575the same. The default definition is @samp{(HARD_FRAME_POINTER_REGNUM 3576== FRAME_POINTER_REGNUM)}; you only need to define this macro if that 3577definition is not suitable for use in preprocessor conditionals. 3578@end defmac 3579 3580@defmac HARD_FRAME_POINTER_IS_ARG_POINTER 3581Define this to a preprocessor constant that is nonzero if 3582@code{hard_frame_pointer_rtx} and @code{arg_pointer_rtx} should be the 3583same. The default definition is @samp{(HARD_FRAME_POINTER_REGNUM == 3584ARG_POINTER_REGNUM)}; you only need to define this macro if that 3585definition is not suitable for use in preprocessor conditionals. 3586@end defmac 3587 3588@defmac RETURN_ADDRESS_POINTER_REGNUM 3589The register number of the return address pointer register, which is used to 3590access the current function's return address from the stack. On some 3591machines, the return address is not at a fixed offset from the frame 3592pointer or stack pointer or argument pointer. This register can be defined 3593to point to the return address on the stack, and then be converted by 3594@code{ELIMINABLE_REGS} into either the frame pointer or stack pointer. 3595 3596Do not define this macro unless there is no other way to get the return 3597address from the stack. 3598@end defmac 3599 3600@defmac STATIC_CHAIN_REGNUM 3601@defmacx STATIC_CHAIN_INCOMING_REGNUM 3602Register numbers used for passing a function's static chain pointer. If 3603register windows are used, the register number as seen by the called 3604function is @code{STATIC_CHAIN_INCOMING_REGNUM}, while the register 3605number as seen by the calling function is @code{STATIC_CHAIN_REGNUM}. If 3606these registers are the same, @code{STATIC_CHAIN_INCOMING_REGNUM} need 3607not be defined. 3608 3609The static chain register need not be a fixed register. 3610 3611If the static chain is passed in memory, these macros should not be 3612defined; instead, the @code{TARGET_STATIC_CHAIN} hook should be used. 3613@end defmac 3614 3615@hook TARGET_STATIC_CHAIN 3616This hook replaces the use of @code{STATIC_CHAIN_REGNUM} et al for 3617targets that may use different static chain locations for different 3618nested functions. This may be required if the target has function 3619attributes that affect the calling conventions of the function and 3620those calling conventions use different static chain locations. 3621 3622The default version of this hook uses @code{STATIC_CHAIN_REGNUM} et al. 3623 3624If the static chain is passed in memory, this hook should be used to 3625provide rtx giving @code{mem} expressions that denote where they are stored. 3626Often the @code{mem} expression as seen by the caller will be at an offset 3627from the stack pointer and the @code{mem} expression as seen by the callee 3628will be at an offset from the frame pointer. 3629@findex stack_pointer_rtx 3630@findex frame_pointer_rtx 3631@findex arg_pointer_rtx 3632The variables @code{stack_pointer_rtx}, @code{frame_pointer_rtx}, and 3633@code{arg_pointer_rtx} will have been initialized and should be used 3634to refer to those items. 3635@end deftypefn 3636 3637@defmac DWARF_FRAME_REGISTERS 3638This macro specifies the maximum number of hard registers that can be 3639saved in a call frame. This is used to size data structures used in 3640DWARF2 exception handling. 3641 3642Prior to GCC 3.0, this macro was needed in order to establish a stable 3643exception handling ABI in the face of adding new hard registers for ISA 3644extensions. In GCC 3.0 and later, the EH ABI is insulated from changes 3645in the number of hard registers. Nevertheless, this macro can still be 3646used to reduce the runtime memory requirements of the exception handling 3647routines, which can be substantial if the ISA contains a lot of 3648registers that are not call-saved. 3649 3650If this macro is not defined, it defaults to 3651@code{FIRST_PSEUDO_REGISTER}. 3652@end defmac 3653 3654@defmac PRE_GCC3_DWARF_FRAME_REGISTERS 3655 3656This macro is similar to @code{DWARF_FRAME_REGISTERS}, but is provided 3657for backward compatibility in pre GCC 3.0 compiled code. 3658 3659If this macro is not defined, it defaults to 3660@code{DWARF_FRAME_REGISTERS}. 3661@end defmac 3662 3663@defmac DWARF_REG_TO_UNWIND_COLUMN (@var{regno}) 3664 3665Define this macro if the target's representation for dwarf registers 3666is different than the internal representation for unwind column. 3667Given a dwarf register, this macro should return the internal unwind 3668column number to use instead. 3669 3670See the PowerPC's SPE target for an example. 3671@end defmac 3672 3673@defmac DWARF_FRAME_REGNUM (@var{regno}) 3674 3675Define this macro if the target's representation for dwarf registers 3676used in .eh_frame or .debug_frame is different from that used in other 3677debug info sections. Given a GCC hard register number, this macro 3678should return the .eh_frame register number. The default is 3679@code{DBX_REGISTER_NUMBER (@var{regno})}. 3680 3681@end defmac 3682 3683@defmac DWARF2_FRAME_REG_OUT (@var{regno}, @var{for_eh}) 3684 3685Define this macro to map register numbers held in the call frame info 3686that GCC has collected using @code{DWARF_FRAME_REGNUM} to those that 3687should be output in .debug_frame (@code{@var{for_eh}} is zero) and 3688.eh_frame (@code{@var{for_eh}} is nonzero). The default is to 3689return @code{@var{regno}}. 3690 3691@end defmac 3692 3693@defmac REG_VALUE_IN_UNWIND_CONTEXT 3694 3695Define this macro if the target stores register values as 3696@code{_Unwind_Word} type in unwind context. It should be defined if 3697target register size is larger than the size of @code{void *}. The 3698default is to store register values as @code{void *} type. 3699 3700@end defmac 3701 3702@defmac ASSUME_EXTENDED_UNWIND_CONTEXT 3703 3704Define this macro to be 1 if the target always uses extended unwind 3705context with version, args_size and by_value fields. If it is undefined, 3706it will be defined to 1 when @code{REG_VALUE_IN_UNWIND_CONTEXT} is 3707defined and 0 otherwise. 3708 3709@end defmac 3710 3711@node Elimination 3712@subsection Eliminating Frame Pointer and Arg Pointer 3713 3714@c prevent bad page break with this line 3715This is about eliminating the frame pointer and arg pointer. 3716 3717@hook TARGET_FRAME_POINTER_REQUIRED 3718This target hook should return @code{true} if a function must have and use 3719a frame pointer. This target hook is called in the reload pass. If its return 3720value is @code{true} the function will have a frame pointer. 3721 3722This target hook can in principle examine the current function and decide 3723according to the facts, but on most machines the constant @code{false} or the 3724constant @code{true} suffices. Use @code{false} when the machine allows code 3725to be generated with no frame pointer, and doing so saves some time or space. 3726Use @code{true} when there is no possible advantage to avoiding a frame 3727pointer. 3728 3729In certain cases, the compiler does not know how to produce valid code 3730without a frame pointer. The compiler recognizes those cases and 3731automatically gives the function a frame pointer regardless of what 3732@code{TARGET_FRAME_POINTER_REQUIRED} returns. You don't need to worry about 3733them. 3734 3735In a function that does not require a frame pointer, the frame pointer 3736register can be allocated for ordinary usage, unless you mark it as a 3737fixed register. See @code{FIXED_REGISTERS} for more information. 3738 3739Default return value is @code{false}. 3740@end deftypefn 3741 3742@findex get_frame_size 3743@defmac INITIAL_FRAME_POINTER_OFFSET (@var{depth-var}) 3744A C statement to store in the variable @var{depth-var} the difference 3745between the frame pointer and the stack pointer values immediately after 3746the function prologue. The value would be computed from information 3747such as the result of @code{get_frame_size ()} and the tables of 3748registers @code{regs_ever_live} and @code{call_used_regs}. 3749 3750If @code{ELIMINABLE_REGS} is defined, this macro will be not be used and 3751need not be defined. Otherwise, it must be defined even if 3752@code{TARGET_FRAME_POINTER_REQUIRED} always returns true; in that 3753case, you may set @var{depth-var} to anything. 3754@end defmac 3755 3756@defmac ELIMINABLE_REGS 3757If defined, this macro specifies a table of register pairs used to 3758eliminate unneeded registers that point into the stack frame. If it is not 3759defined, the only elimination attempted by the compiler is to replace 3760references to the frame pointer with references to the stack pointer. 3761 3762The definition of this macro is a list of structure initializations, each 3763of which specifies an original and replacement register. 3764 3765On some machines, the position of the argument pointer is not known until 3766the compilation is completed. In such a case, a separate hard register 3767must be used for the argument pointer. This register can be eliminated by 3768replacing it with either the frame pointer or the argument pointer, 3769depending on whether or not the frame pointer has been eliminated. 3770 3771In this case, you might specify: 3772@smallexample 3773#define ELIMINABLE_REGS \ 3774@{@{ARG_POINTER_REGNUM, STACK_POINTER_REGNUM@}, \ 3775 @{ARG_POINTER_REGNUM, FRAME_POINTER_REGNUM@}, \ 3776 @{FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM@}@} 3777@end smallexample 3778 3779Note that the elimination of the argument pointer with the stack pointer is 3780specified first since that is the preferred elimination. 3781@end defmac 3782 3783@hook TARGET_CAN_ELIMINATE 3784This target hook should returns @code{true} if the compiler is allowed to 3785try to replace register number @var{from_reg} with register number 3786@var{to_reg}. This target hook need only be defined if @code{ELIMINABLE_REGS} 3787is defined, and will usually be @code{true}, since most of the cases 3788preventing register elimination are things that the compiler already 3789knows about. 3790 3791Default return value is @code{true}. 3792@end deftypefn 3793 3794@defmac INITIAL_ELIMINATION_OFFSET (@var{from-reg}, @var{to-reg}, @var{offset-var}) 3795This macro is similar to @code{INITIAL_FRAME_POINTER_OFFSET}. It 3796specifies the initial difference between the specified pair of 3797registers. This macro must be defined if @code{ELIMINABLE_REGS} is 3798defined. 3799@end defmac 3800 3801@node Stack Arguments 3802@subsection Passing Function Arguments on the Stack 3803@cindex arguments on stack 3804@cindex stack arguments 3805 3806The macros in this section control how arguments are passed 3807on the stack. See the following section for other macros that 3808control passing certain arguments in registers. 3809 3810@hook TARGET_PROMOTE_PROTOTYPES 3811This target hook returns @code{true} if an argument declared in a 3812prototype as an integral type smaller than @code{int} should actually be 3813passed as an @code{int}. In addition to avoiding errors in certain 3814cases of mismatch, it also makes for better code on certain machines. 3815The default is to not promote prototypes. 3816@end deftypefn 3817 3818@defmac PUSH_ARGS 3819A C expression. If nonzero, push insns will be used to pass 3820outgoing arguments. 3821If the target machine does not have a push instruction, set it to zero. 3822That directs GCC to use an alternate strategy: to 3823allocate the entire argument block and then store the arguments into 3824it. When @code{PUSH_ARGS} is nonzero, @code{PUSH_ROUNDING} must be defined too. 3825@end defmac 3826 3827@defmac PUSH_ARGS_REVERSED 3828A C expression. If nonzero, function arguments will be evaluated from 3829last to first, rather than from first to last. If this macro is not 3830defined, it defaults to @code{PUSH_ARGS} on targets where the stack 3831and args grow in opposite directions, and 0 otherwise. 3832@end defmac 3833 3834@defmac PUSH_ROUNDING (@var{npushed}) 3835A C expression that is the number of bytes actually pushed onto the 3836stack when an instruction attempts to push @var{npushed} bytes. 3837 3838On some machines, the definition 3839 3840@smallexample 3841#define PUSH_ROUNDING(BYTES) (BYTES) 3842@end smallexample 3843 3844@noindent 3845will suffice. But on other machines, instructions that appear 3846to push one byte actually push two bytes in an attempt to maintain 3847alignment. Then the definition should be 3848 3849@smallexample 3850#define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1) 3851@end smallexample 3852 3853If the value of this macro has a type, it should be an unsigned type. 3854@end defmac 3855 3856@findex current_function_outgoing_args_size 3857@defmac ACCUMULATE_OUTGOING_ARGS 3858A C expression. If nonzero, the maximum amount of space required for outgoing arguments 3859will be computed and placed into the variable 3860@code{current_function_outgoing_args_size}. No space will be pushed 3861onto the stack for each call; instead, the function prologue should 3862increase the stack frame size by this amount. 3863 3864Setting both @code{PUSH_ARGS} and @code{ACCUMULATE_OUTGOING_ARGS} 3865is not proper. 3866@end defmac 3867 3868@defmac REG_PARM_STACK_SPACE (@var{fndecl}) 3869Define this macro if functions should assume that stack space has been 3870allocated for arguments even when their values are passed in 3871registers. 3872 3873The value of this macro is the size, in bytes, of the area reserved for 3874arguments passed in registers for the function represented by @var{fndecl}, 3875which can be zero if GCC is calling a library function. 3876The argument @var{fndecl} can be the FUNCTION_DECL, or the type itself 3877of the function. 3878 3879This space can be allocated by the caller, or be a part of the 3880machine-dependent stack frame: @code{OUTGOING_REG_PARM_STACK_SPACE} says 3881which. 3882@end defmac 3883@c above is overfull. not sure what to do. --mew 5feb93 did 3884@c something, not sure if it looks good. --mew 10feb93 3885 3886@defmac OUTGOING_REG_PARM_STACK_SPACE (@var{fntype}) 3887Define this to a nonzero value if it is the responsibility of the 3888caller to allocate the area reserved for arguments passed in registers 3889when calling a function of @var{fntype}. @var{fntype} may be NULL 3890if the function called is a library function. 3891 3892If @code{ACCUMULATE_OUTGOING_ARGS} is defined, this macro controls 3893whether the space for these arguments counts in the value of 3894@code{current_function_outgoing_args_size}. 3895@end defmac 3896 3897@defmac STACK_PARMS_IN_REG_PARM_AREA 3898Define this macro if @code{REG_PARM_STACK_SPACE} is defined, but the 3899stack parameters don't skip the area specified by it. 3900@c i changed this, makes more sens and it should have taken care of the 3901@c overfull.. not as specific, tho. --mew 5feb93 3902 3903Normally, when a parameter is not passed in registers, it is placed on the 3904stack beyond the @code{REG_PARM_STACK_SPACE} area. Defining this macro 3905suppresses this behavior and causes the parameter to be passed on the 3906stack in its natural location. 3907@end defmac 3908 3909@hook TARGET_RETURN_POPS_ARGS 3910This target hook returns the number of bytes of its own arguments that 3911a function pops on returning, or 0 if the function pops no arguments 3912and the caller must therefore pop them all after the function returns. 3913 3914@var{fundecl} is a C variable whose value is a tree node that describes 3915the function in question. Normally it is a node of type 3916@code{FUNCTION_DECL} that describes the declaration of the function. 3917From this you can obtain the @code{DECL_ATTRIBUTES} of the function. 3918 3919@var{funtype} is a C variable whose value is a tree node that 3920describes the function in question. Normally it is a node of type 3921@code{FUNCTION_TYPE} that describes the data type of the function. 3922From this it is possible to obtain the data types of the value and 3923arguments (if known). 3924 3925When a call to a library function is being considered, @var{fundecl} 3926will contain an identifier node for the library function. Thus, if 3927you need to distinguish among various library functions, you can do so 3928by their names. Note that ``library function'' in this context means 3929a function used to perform arithmetic, whose name is known specially 3930in the compiler and was not mentioned in the C code being compiled. 3931 3932@var{size} is the number of bytes of arguments passed on the 3933stack. If a variable number of bytes is passed, it is zero, and 3934argument popping will always be the responsibility of the calling function. 3935 3936On the VAX, all functions always pop their arguments, so the definition 3937of this macro is @var{size}. On the 68000, using the standard 3938calling convention, no functions pop their arguments, so the value of 3939the macro is always 0 in this case. But an alternative calling 3940convention is available in which functions that take a fixed number of 3941arguments pop them but other functions (such as @code{printf}) pop 3942nothing (the caller pops all). When this convention is in use, 3943@var{funtype} is examined to determine whether a function takes a fixed 3944number of arguments. 3945@end deftypefn 3946 3947@defmac CALL_POPS_ARGS (@var{cum}) 3948A C expression that should indicate the number of bytes a call sequence 3949pops off the stack. It is added to the value of @code{RETURN_POPS_ARGS} 3950when compiling a function call. 3951 3952@var{cum} is the variable in which all arguments to the called function 3953have been accumulated. 3954 3955On certain architectures, such as the SH5, a call trampoline is used 3956that pops certain registers off the stack, depending on the arguments 3957that have been passed to the function. Since this is a property of the 3958call site, not of the called function, @code{RETURN_POPS_ARGS} is not 3959appropriate. 3960@end defmac 3961 3962@node Register Arguments 3963@subsection Passing Arguments in Registers 3964@cindex arguments in registers 3965@cindex registers arguments 3966 3967This section describes the macros which let you control how various 3968types of arguments are passed in registers or how they are arranged in 3969the stack. 3970 3971@hook TARGET_FUNCTION_ARG 3972Return an RTX indicating whether a function argument is passed in a 3973register and if so, which register. 3974 3975The arguments are @var{ca}, which summarizes all the previous 3976arguments; @var{mode}, the machine mode of the argument; @var{type}, 3977the data type of the argument as a tree node or 0 if that is not known 3978(which happens for C support library functions); and @var{named}, 3979which is @code{true} for an ordinary argument and @code{false} for 3980nameless arguments that correspond to @samp{@dots{}} in the called 3981function's prototype. @var{type} can be an incomplete type if a 3982syntax error has previously occurred. 3983 3984The return value is usually either a @code{reg} RTX for the hard 3985register in which to pass the argument, or zero to pass the argument 3986on the stack. 3987 3988The value of the expression can also be a @code{parallel} RTX@. This is 3989used when an argument is passed in multiple locations. The mode of the 3990@code{parallel} should be the mode of the entire argument. The 3991@code{parallel} holds any number of @code{expr_list} pairs; each one 3992describes where part of the argument is passed. In each 3993@code{expr_list} the first operand must be a @code{reg} RTX for the hard 3994register in which to pass this part of the argument, and the mode of the 3995register RTX indicates how large this part of the argument is. The 3996second operand of the @code{expr_list} is a @code{const_int} which gives 3997the offset in bytes into the entire argument of where this part starts. 3998As a special exception the first @code{expr_list} in the @code{parallel} 3999RTX may have a first operand of zero. This indicates that the entire 4000argument is also stored on the stack. 4001 4002The last time this hook is called, it is called with @code{MODE == 4003VOIDmode}, and its result is passed to the @code{call} or @code{call_value} 4004pattern as operands 2 and 3 respectively. 4005 4006@cindex @file{stdarg.h} and register arguments 4007The usual way to make the ISO library @file{stdarg.h} work on a 4008machine where some arguments are usually passed in registers, is to 4009cause nameless arguments to be passed on the stack instead. This is 4010done by making @code{TARGET_FUNCTION_ARG} return 0 whenever 4011@var{named} is @code{false}. 4012 4013@cindex @code{TARGET_MUST_PASS_IN_STACK}, and @code{TARGET_FUNCTION_ARG} 4014@cindex @code{REG_PARM_STACK_SPACE}, and @code{TARGET_FUNCTION_ARG} 4015You may use the hook @code{targetm.calls.must_pass_in_stack} 4016in the definition of this macro to determine if this argument is of a 4017type that must be passed in the stack. If @code{REG_PARM_STACK_SPACE} 4018is not defined and @code{TARGET_FUNCTION_ARG} returns nonzero for such an 4019argument, the compiler will abort. If @code{REG_PARM_STACK_SPACE} is 4020defined, the argument will be computed in the stack and then loaded into 4021a register. 4022@end deftypefn 4023 4024@hook TARGET_MUST_PASS_IN_STACK 4025This target hook should return @code{true} if we should not pass @var{type} 4026solely in registers. The file @file{expr.h} defines a 4027definition that is usually appropriate, refer to @file{expr.h} for additional 4028documentation. 4029@end deftypefn 4030 4031@hook TARGET_FUNCTION_INCOMING_ARG 4032Define this hook if the target machine has ``register windows'', so 4033that the register in which a function sees an arguments is not 4034necessarily the same as the one in which the caller passed the 4035argument. 4036 4037For such machines, @code{TARGET_FUNCTION_ARG} computes the register in 4038which the caller passes the value, and 4039@code{TARGET_FUNCTION_INCOMING_ARG} should be defined in a similar 4040fashion to tell the function being called where the arguments will 4041arrive. 4042 4043If @code{TARGET_FUNCTION_INCOMING_ARG} is not defined, 4044@code{TARGET_FUNCTION_ARG} serves both purposes. 4045@end deftypefn 4046 4047@hook TARGET_ARG_PARTIAL_BYTES 4048This target hook returns the number of bytes at the beginning of an 4049argument that must be put in registers. The value must be zero for 4050arguments that are passed entirely in registers or that are entirely 4051pushed on the stack. 4052 4053On some machines, certain arguments must be passed partially in 4054registers and partially in memory. On these machines, typically the 4055first few words of arguments are passed in registers, and the rest 4056on the stack. If a multi-word argument (a @code{double} or a 4057structure) crosses that boundary, its first few words must be passed 4058in registers and the rest must be pushed. This macro tells the 4059compiler when this occurs, and how many bytes should go in registers. 4060 4061@code{TARGET_FUNCTION_ARG} for these arguments should return the first 4062register to be used by the caller for this argument; likewise 4063@code{TARGET_FUNCTION_INCOMING_ARG}, for the called function. 4064@end deftypefn 4065 4066@hook TARGET_PASS_BY_REFERENCE 4067This target hook should return @code{true} if an argument at the 4068position indicated by @var{cum} should be passed by reference. This 4069predicate is queried after target independent reasons for being 4070passed by reference, such as @code{TREE_ADDRESSABLE (type)}. 4071 4072If the hook returns true, a copy of that argument is made in memory and a 4073pointer to the argument is passed instead of the argument itself. 4074The pointer is passed in whatever way is appropriate for passing a pointer 4075to that type. 4076@end deftypefn 4077 4078@hook TARGET_CALLEE_COPIES 4079The function argument described by the parameters to this hook is 4080known to be passed by reference. The hook should return true if the 4081function argument should be copied by the callee instead of copied 4082by the caller. 4083 4084For any argument for which the hook returns true, if it can be 4085determined that the argument is not modified, then a copy need 4086not be generated. 4087 4088The default version of this hook always returns false. 4089@end deftypefn 4090 4091@defmac CUMULATIVE_ARGS 4092A C type for declaring a variable that is used as the first argument 4093of @code{TARGET_FUNCTION_ARG} and other related values. For some 4094target machines, the type @code{int} suffices and can hold the number 4095of bytes of argument so far. 4096 4097There is no need to record in @code{CUMULATIVE_ARGS} anything about the 4098arguments that have been passed on the stack. The compiler has other 4099variables to keep track of that. For target machines on which all 4100arguments are passed on the stack, there is no need to store anything in 4101@code{CUMULATIVE_ARGS}; however, the data structure must exist and 4102should not be empty, so use @code{int}. 4103@end defmac 4104 4105@defmac OVERRIDE_ABI_FORMAT (@var{fndecl}) 4106If defined, this macro is called before generating any code for a 4107function, but after the @var{cfun} descriptor for the function has been 4108created. The back end may use this macro to update @var{cfun} to 4109reflect an ABI other than that which would normally be used by default. 4110If the compiler is generating code for a compiler-generated function, 4111@var{fndecl} may be @code{NULL}. 4112@end defmac 4113 4114@defmac INIT_CUMULATIVE_ARGS (@var{cum}, @var{fntype}, @var{libname}, @var{fndecl}, @var{n_named_args}) 4115A C statement (sans semicolon) for initializing the variable 4116@var{cum} for the state at the beginning of the argument list. The 4117variable has type @code{CUMULATIVE_ARGS}. The value of @var{fntype} 4118is the tree node for the data type of the function which will receive 4119the args, or 0 if the args are to a compiler support library function. 4120For direct calls that are not libcalls, @var{fndecl} contain the 4121declaration node of the function. @var{fndecl} is also set when 4122@code{INIT_CUMULATIVE_ARGS} is used to find arguments for the function 4123being compiled. @var{n_named_args} is set to the number of named 4124arguments, including a structure return address if it is passed as a 4125parameter, when making a call. When processing incoming arguments, 4126@var{n_named_args} is set to @minus{}1. 4127 4128When processing a call to a compiler support library function, 4129@var{libname} identifies which one. It is a @code{symbol_ref} rtx which 4130contains the name of the function, as a string. @var{libname} is 0 when 4131an ordinary C function call is being processed. Thus, each time this 4132macro is called, either @var{libname} or @var{fntype} is nonzero, but 4133never both of them at once. 4134@end defmac 4135 4136@defmac INIT_CUMULATIVE_LIBCALL_ARGS (@var{cum}, @var{mode}, @var{libname}) 4137Like @code{INIT_CUMULATIVE_ARGS} but only used for outgoing libcalls, 4138it gets a @code{MODE} argument instead of @var{fntype}, that would be 4139@code{NULL}. @var{indirect} would always be zero, too. If this macro 4140is not defined, @code{INIT_CUMULATIVE_ARGS (cum, NULL_RTX, libname, 41410)} is used instead. 4142@end defmac 4143 4144@defmac INIT_CUMULATIVE_INCOMING_ARGS (@var{cum}, @var{fntype}, @var{libname}) 4145Like @code{INIT_CUMULATIVE_ARGS} but overrides it for the purposes of 4146finding the arguments for the function being compiled. If this macro is 4147undefined, @code{INIT_CUMULATIVE_ARGS} is used instead. 4148 4149The value passed for @var{libname} is always 0, since library routines 4150with special calling conventions are never compiled with GCC@. The 4151argument @var{libname} exists for symmetry with 4152@code{INIT_CUMULATIVE_ARGS}. 4153@c could use "this macro" in place of @code{INIT_CUMULATIVE_ARGS}, maybe. 4154@c --mew 5feb93 i switched the order of the sentences. --mew 10feb93 4155@end defmac 4156 4157@hook TARGET_FUNCTION_ARG_ADVANCE 4158This hook updates the summarizer variable pointed to by @var{ca} to 4159advance past an argument in the argument list. The values @var{mode}, 4160@var{type} and @var{named} describe that argument. Once this is done, 4161the variable @var{cum} is suitable for analyzing the @emph{following} 4162argument with @code{TARGET_FUNCTION_ARG}, etc. 4163 4164This hook need not do anything if the argument in question was passed 4165on the stack. The compiler knows how to track the amount of stack space 4166used for arguments without any special help. 4167@end deftypefn 4168 4169@defmac FUNCTION_ARG_OFFSET (@var{mode}, @var{type}) 4170If defined, a C expression that is the number of bytes to add to the 4171offset of the argument passed in memory. This is needed for the SPU, 4172which passes @code{char} and @code{short} arguments in the preferred 4173slot that is in the middle of the quad word instead of starting at the 4174top. 4175@end defmac 4176 4177@defmac FUNCTION_ARG_PADDING (@var{mode}, @var{type}) 4178If defined, a C expression which determines whether, and in which direction, 4179to pad out an argument with extra space. The value should be of type 4180@code{enum direction}: either @code{upward} to pad above the argument, 4181@code{downward} to pad below, or @code{none} to inhibit padding. 4182 4183The @emph{amount} of padding is not controlled by this macro, but by the 4184target hook @code{TARGET_FUNCTION_ARG_ROUND_BOUNDARY}. It is 4185always just enough to reach the next multiple of that boundary. 4186 4187This macro has a default definition which is right for most systems. 4188For little-endian machines, the default is to pad upward. For 4189big-endian machines, the default is to pad downward for an argument of 4190constant size shorter than an @code{int}, and upward otherwise. 4191@end defmac 4192 4193@defmac PAD_VARARGS_DOWN 4194If defined, a C expression which determines whether the default 4195implementation of va_arg will attempt to pad down before reading the 4196next argument, if that argument is smaller than its aligned space as 4197controlled by @code{PARM_BOUNDARY}. If this macro is not defined, all such 4198arguments are padded down if @code{BYTES_BIG_ENDIAN} is true. 4199@end defmac 4200 4201@defmac BLOCK_REG_PADDING (@var{mode}, @var{type}, @var{first}) 4202Specify padding for the last element of a block move between registers and 4203memory. @var{first} is nonzero if this is the only element. Defining this 4204macro allows better control of register function parameters on big-endian 4205machines, without using @code{PARALLEL} rtl. In particular, 4206@code{MUST_PASS_IN_STACK} need not test padding and mode of types in 4207registers, as there is no longer a "wrong" part of a register; For example, 4208a three byte aggregate may be passed in the high part of a register if so 4209required. 4210@end defmac 4211 4212@hook TARGET_FUNCTION_ARG_BOUNDARY 4213This hook returns the alignment boundary, in bits, of an argument 4214with the specified mode and type. The default hook returns 4215@code{PARM_BOUNDARY} for all arguments. 4216@end deftypefn 4217 4218@hook TARGET_FUNCTION_ARG_ROUND_BOUNDARY 4219 4220@defmac FUNCTION_ARG_REGNO_P (@var{regno}) 4221A C expression that is nonzero if @var{regno} is the number of a hard 4222register in which function arguments are sometimes passed. This does 4223@emph{not} include implicit arguments such as the static chain and 4224the structure-value address. On many machines, no registers can be 4225used for this purpose since all function arguments are pushed on the 4226stack. 4227@end defmac 4228 4229@hook TARGET_SPLIT_COMPLEX_ARG 4230This hook should return true if parameter of type @var{type} are passed 4231as two scalar parameters. By default, GCC will attempt to pack complex 4232arguments into the target's word size. Some ABIs require complex arguments 4233to be split and treated as their individual components. For example, on 4234AIX64, complex floats should be passed in a pair of floating point 4235registers, even though a complex float would fit in one 64-bit floating 4236point register. 4237 4238The default value of this hook is @code{NULL}, which is treated as always 4239false. 4240@end deftypefn 4241 4242@hook TARGET_BUILD_BUILTIN_VA_LIST 4243This hook returns a type node for @code{va_list} for the target. 4244The default version of the hook returns @code{void*}. 4245@end deftypefn 4246 4247@hook TARGET_ENUM_VA_LIST_P 4248This target hook is used in function @code{c_common_nodes_and_builtins} 4249to iterate through the target specific builtin types for va_list. The 4250variable @var{idx} is used as iterator. @var{pname} has to be a pointer 4251to a @code{const char *} and @var{ptree} a pointer to a @code{tree} typed 4252variable. 4253The arguments @var{pname} and @var{ptree} are used to store the result of 4254this macro and are set to the name of the va_list builtin type and its 4255internal type. 4256If the return value of this macro is zero, then there is no more element. 4257Otherwise the @var{IDX} should be increased for the next call of this 4258macro to iterate through all types. 4259@end deftypefn 4260 4261@hook TARGET_FN_ABI_VA_LIST 4262This hook returns the va_list type of the calling convention specified by 4263@var{fndecl}. 4264The default version of this hook returns @code{va_list_type_node}. 4265@end deftypefn 4266 4267@hook TARGET_CANONICAL_VA_LIST_TYPE 4268This hook returns the va_list type of the calling convention specified by the 4269type of @var{type}. If @var{type} is not a valid va_list type, it returns 4270@code{NULL_TREE}. 4271@end deftypefn 4272 4273@hook TARGET_GIMPLIFY_VA_ARG_EXPR 4274This hook performs target-specific gimplification of 4275@code{VA_ARG_EXPR}. The first two parameters correspond to the 4276arguments to @code{va_arg}; the latter two are as in 4277@code{gimplify.c:gimplify_expr}. 4278@end deftypefn 4279 4280@hook TARGET_VALID_POINTER_MODE 4281Define this to return nonzero if the port can handle pointers 4282with machine mode @var{mode}. The default version of this 4283hook returns true for both @code{ptr_mode} and @code{Pmode}. 4284@end deftypefn 4285 4286@hook TARGET_REF_MAY_ALIAS_ERRNO 4287 4288@hook TARGET_SCALAR_MODE_SUPPORTED_P 4289Define this to return nonzero if the port is prepared to handle 4290insns involving scalar mode @var{mode}. For a scalar mode to be 4291considered supported, all the basic arithmetic and comparisons 4292must work. 4293 4294The default version of this hook returns true for any mode 4295required to handle the basic C types (as defined by the port). 4296Included here are the double-word arithmetic supported by the 4297code in @file{optabs.c}. 4298@end deftypefn 4299 4300@hook TARGET_VECTOR_MODE_SUPPORTED_P 4301Define this to return nonzero if the port is prepared to handle 4302insns involving vector mode @var{mode}. At the very least, it 4303must have move patterns for this mode. 4304@end deftypefn 4305 4306@hook TARGET_ARRAY_MODE_SUPPORTED_P 4307 4308@hook TARGET_SMALL_REGISTER_CLASSES_FOR_MODE_P 4309Define this to return nonzero for machine modes for which the port has 4310small register classes. If this target hook returns nonzero for a given 4311@var{mode}, the compiler will try to minimize the lifetime of registers 4312in @var{mode}. The hook may be called with @code{VOIDmode} as argument. 4313In this case, the hook is expected to return nonzero if it returns nonzero 4314for any mode. 4315 4316On some machines, it is risky to let hard registers live across arbitrary 4317insns. Typically, these machines have instructions that require values 4318to be in specific registers (like an accumulator), and reload will fail 4319if the required hard register is used for another purpose across such an 4320insn. 4321 4322Passes before reload do not know which hard registers will be used 4323in an instruction, but the machine modes of the registers set or used in 4324the instruction are already known. And for some machines, register 4325classes are small for, say, integer registers but not for floating point 4326registers. For example, the AMD x86-64 architecture requires specific 4327registers for the legacy x86 integer instructions, but there are many 4328SSE registers for floating point operations. On such targets, a good 4329strategy may be to return nonzero from this hook for @code{INTEGRAL_MODE_P} 4330machine modes but zero for the SSE register classes. 4331 4332The default version of this hook returns false for any mode. It is always 4333safe to redefine this hook to return with a nonzero value. But if you 4334unnecessarily define it, you will reduce the amount of optimizations 4335that can be performed in some cases. If you do not define this hook 4336to return a nonzero value when it is required, the compiler will run out 4337of spill registers and print a fatal error message. 4338@end deftypefn 4339 4340@hook TARGET_FLAGS_REGNUM 4341 4342@node Scalar Return 4343@subsection How Scalar Function Values Are Returned 4344@cindex return values in registers 4345@cindex values, returned by functions 4346@cindex scalars, returned as values 4347 4348This section discusses the macros that control returning scalars as 4349values---values that can fit in registers. 4350 4351@hook TARGET_FUNCTION_VALUE 4352 4353Define this to return an RTX representing the place where a function 4354returns or receives a value of data type @var{ret_type}, a tree node 4355representing a data type. @var{fn_decl_or_type} is a tree node 4356representing @code{FUNCTION_DECL} or @code{FUNCTION_TYPE} of a 4357function being called. If @var{outgoing} is false, the hook should 4358compute the register in which the caller will see the return value. 4359Otherwise, the hook should return an RTX representing the place where 4360a function returns a value. 4361 4362On many machines, only @code{TYPE_MODE (@var{ret_type})} is relevant. 4363(Actually, on most machines, scalar values are returned in the same 4364place regardless of mode.) The value of the expression is usually a 4365@code{reg} RTX for the hard register where the return value is stored. 4366The value can also be a @code{parallel} RTX, if the return value is in 4367multiple places. See @code{TARGET_FUNCTION_ARG} for an explanation of the 4368@code{parallel} form. Note that the callee will populate every 4369location specified in the @code{parallel}, but if the first element of 4370the @code{parallel} contains the whole return value, callers will use 4371that element as the canonical location and ignore the others. The m68k 4372port uses this type of @code{parallel} to return pointers in both 4373@samp{%a0} (the canonical location) and @samp{%d0}. 4374 4375If @code{TARGET_PROMOTE_FUNCTION_RETURN} returns true, you must apply 4376the same promotion rules specified in @code{PROMOTE_MODE} if 4377@var{valtype} is a scalar type. 4378 4379If the precise function being called is known, @var{func} is a tree 4380node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null 4381pointer. This makes it possible to use a different value-returning 4382convention for specific functions when all their calls are 4383known. 4384 4385Some target machines have ``register windows'' so that the register in 4386which a function returns its value is not the same as the one in which 4387the caller sees the value. For such machines, you should return 4388different RTX depending on @var{outgoing}. 4389 4390@code{TARGET_FUNCTION_VALUE} is not used for return values with 4391aggregate data types, because these are returned in another way. See 4392@code{TARGET_STRUCT_VALUE_RTX} and related macros, below. 4393@end deftypefn 4394 4395@defmac FUNCTION_VALUE (@var{valtype}, @var{func}) 4396This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE} for 4397a new target instead. 4398@end defmac 4399 4400@defmac LIBCALL_VALUE (@var{mode}) 4401A C expression to create an RTX representing the place where a library 4402function returns a value of mode @var{mode}. 4403 4404Note that ``library function'' in this context means a compiler 4405support routine, used to perform arithmetic, whose name is known 4406specially by the compiler and was not mentioned in the C code being 4407compiled. 4408@end defmac 4409 4410@hook TARGET_LIBCALL_VALUE 4411Define this hook if the back-end needs to know the name of the libcall 4412function in order to determine where the result should be returned. 4413 4414The mode of the result is given by @var{mode} and the name of the called 4415library function is given by @var{fun}. The hook should return an RTX 4416representing the place where the library function result will be returned. 4417 4418If this hook is not defined, then LIBCALL_VALUE will be used. 4419@end deftypefn 4420 4421@defmac FUNCTION_VALUE_REGNO_P (@var{regno}) 4422A C expression that is nonzero if @var{regno} is the number of a hard 4423register in which the values of called function may come back. 4424 4425A register whose use for returning values is limited to serving as the 4426second of a pair (for a value of type @code{double}, say) need not be 4427recognized by this macro. So for most machines, this definition 4428suffices: 4429 4430@smallexample 4431#define FUNCTION_VALUE_REGNO_P(N) ((N) == 0) 4432@end smallexample 4433 4434If the machine has register windows, so that the caller and the called 4435function use different registers for the return value, this macro 4436should recognize only the caller's register numbers. 4437 4438This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE_REGNO_P} 4439for a new target instead. 4440@end defmac 4441 4442@hook TARGET_FUNCTION_VALUE_REGNO_P 4443A target hook that return @code{true} if @var{regno} is the number of a hard 4444register in which the values of called function may come back. 4445 4446A register whose use for returning values is limited to serving as the 4447second of a pair (for a value of type @code{double}, say) need not be 4448recognized by this target hook. 4449 4450If the machine has register windows, so that the caller and the called 4451function use different registers for the return value, this target hook 4452should recognize only the caller's register numbers. 4453 4454If this hook is not defined, then FUNCTION_VALUE_REGNO_P will be used. 4455@end deftypefn 4456 4457@defmac APPLY_RESULT_SIZE 4458Define this macro if @samp{untyped_call} and @samp{untyped_return} 4459need more space than is implied by @code{FUNCTION_VALUE_REGNO_P} for 4460saving and restoring an arbitrary return value. 4461@end defmac 4462 4463@hook TARGET_RETURN_IN_MSB 4464This hook should return true if values of type @var{type} are returned 4465at the most significant end of a register (in other words, if they are 4466padded at the least significant end). You can assume that @var{type} 4467is returned in a register; the caller is required to check this. 4468 4469Note that the register provided by @code{TARGET_FUNCTION_VALUE} must 4470be able to hold the complete return value. For example, if a 1-, 2- 4471or 3-byte structure is returned at the most significant end of a 44724-byte register, @code{TARGET_FUNCTION_VALUE} should provide an 4473@code{SImode} rtx. 4474@end deftypefn 4475 4476@node Aggregate Return 4477@subsection How Large Values Are Returned 4478@cindex aggregates as return values 4479@cindex large return values 4480@cindex returning aggregate values 4481@cindex structure value address 4482 4483When a function value's mode is @code{BLKmode} (and in some other 4484cases), the value is not returned according to 4485@code{TARGET_FUNCTION_VALUE} (@pxref{Scalar Return}). Instead, the 4486caller passes the address of a block of memory in which the value 4487should be stored. This address is called the @dfn{structure value 4488address}. 4489 4490This section describes how to control returning structure values in 4491memory. 4492 4493@hook TARGET_RETURN_IN_MEMORY 4494This target hook should return a nonzero value to say to return the 4495function value in memory, just as large structures are always returned. 4496Here @var{type} will be the data type of the value, and @var{fntype} 4497will be the type of the function doing the returning, or @code{NULL} for 4498libcalls. 4499 4500Note that values of mode @code{BLKmode} must be explicitly handled 4501by this function. Also, the option @option{-fpcc-struct-return} 4502takes effect regardless of this macro. On most systems, it is 4503possible to leave the hook undefined; this causes a default 4504definition to be used, whose value is the constant 1 for @code{BLKmode} 4505values, and 0 otherwise. 4506 4507Do not use this hook to indicate that structures and unions should always 4508be returned in memory. You should instead use @code{DEFAULT_PCC_STRUCT_RETURN} 4509to indicate this. 4510@end deftypefn 4511 4512@defmac DEFAULT_PCC_STRUCT_RETURN 4513Define this macro to be 1 if all structure and union return values must be 4514in memory. Since this results in slower code, this should be defined 4515only if needed for compatibility with other compilers or with an ABI@. 4516If you define this macro to be 0, then the conventions used for structure 4517and union return values are decided by the @code{TARGET_RETURN_IN_MEMORY} 4518target hook. 4519 4520If not defined, this defaults to the value 1. 4521@end defmac 4522 4523@hook TARGET_STRUCT_VALUE_RTX 4524This target hook should return the location of the structure value 4525address (normally a @code{mem} or @code{reg}), or 0 if the address is 4526passed as an ``invisible'' first argument. Note that @var{fndecl} may 4527be @code{NULL}, for libcalls. You do not need to define this target 4528hook if the address is always passed as an ``invisible'' first 4529argument. 4530 4531On some architectures the place where the structure value address 4532is found by the called function is not the same place that the 4533caller put it. This can be due to register windows, or it could 4534be because the function prologue moves it to a different place. 4535@var{incoming} is @code{1} or @code{2} when the location is needed in 4536the context of the called function, and @code{0} in the context of 4537the caller. 4538 4539If @var{incoming} is nonzero and the address is to be found on the 4540stack, return a @code{mem} which refers to the frame pointer. If 4541@var{incoming} is @code{2}, the result is being used to fetch the 4542structure value address at the beginning of a function. If you need 4543to emit adjusting code, you should do it at this point. 4544@end deftypefn 4545 4546@defmac PCC_STATIC_STRUCT_RETURN 4547Define this macro if the usual system convention on the target machine 4548for returning structures and unions is for the called function to return 4549the address of a static variable containing the value. 4550 4551Do not define this if the usual system convention is for the caller to 4552pass an address to the subroutine. 4553 4554This macro has effect in @option{-fpcc-struct-return} mode, but it does 4555nothing when you use @option{-freg-struct-return} mode. 4556@end defmac 4557 4558@hook TARGET_GET_RAW_RESULT_MODE 4559 4560@hook TARGET_GET_RAW_ARG_MODE 4561 4562@node Caller Saves 4563@subsection Caller-Saves Register Allocation 4564 4565If you enable it, GCC can save registers around function calls. This 4566makes it possible to use call-clobbered registers to hold variables that 4567must live across calls. 4568 4569@defmac CALLER_SAVE_PROFITABLE (@var{refs}, @var{calls}) 4570A C expression to determine whether it is worthwhile to consider placing 4571a pseudo-register in a call-clobbered hard register and saving and 4572restoring it around each function call. The expression should be 1 when 4573this is worth doing, and 0 otherwise. 4574 4575If you don't define this macro, a default is used which is good on most 4576machines: @code{4 * @var{calls} < @var{refs}}. 4577@end defmac 4578 4579@defmac HARD_REGNO_CALLER_SAVE_MODE (@var{regno}, @var{nregs}) 4580A C expression specifying which mode is required for saving @var{nregs} 4581of a pseudo-register in call-clobbered hard register @var{regno}. If 4582@var{regno} is unsuitable for caller save, @code{VOIDmode} should be 4583returned. For most machines this macro need not be defined since GCC 4584will select the smallest suitable mode. 4585@end defmac 4586 4587@node Function Entry 4588@subsection Function Entry and Exit 4589@cindex function entry and exit 4590@cindex prologue 4591@cindex epilogue 4592 4593This section describes the macros that output function entry 4594(@dfn{prologue}) and exit (@dfn{epilogue}) code. 4595 4596@hook TARGET_ASM_FUNCTION_PROLOGUE 4597If defined, a function that outputs the assembler code for entry to a 4598function. The prologue is responsible for setting up the stack frame, 4599initializing the frame pointer register, saving registers that must be 4600saved, and allocating @var{size} additional bytes of storage for the 4601local variables. @var{size} is an integer. @var{file} is a stdio 4602stream to which the assembler code should be output. 4603 4604The label for the beginning of the function need not be output by this 4605macro. That has already been done when the macro is run. 4606 4607@findex regs_ever_live 4608To determine which registers to save, the macro can refer to the array 4609@code{regs_ever_live}: element @var{r} is nonzero if hard register 4610@var{r} is used anywhere within the function. This implies the function 4611prologue should save register @var{r}, provided it is not one of the 4612call-used registers. (@code{TARGET_ASM_FUNCTION_EPILOGUE} must likewise use 4613@code{regs_ever_live}.) 4614 4615On machines that have ``register windows'', the function entry code does 4616not save on the stack the registers that are in the windows, even if 4617they are supposed to be preserved by function calls; instead it takes 4618appropriate steps to ``push'' the register stack, if any non-call-used 4619registers are used in the function. 4620 4621@findex frame_pointer_needed 4622On machines where functions may or may not have frame-pointers, the 4623function entry code must vary accordingly; it must set up the frame 4624pointer if one is wanted, and not otherwise. To determine whether a 4625frame pointer is in wanted, the macro can refer to the variable 4626@code{frame_pointer_needed}. The variable's value will be 1 at run 4627time in a function that needs a frame pointer. @xref{Elimination}. 4628 4629The function entry code is responsible for allocating any stack space 4630required for the function. This stack space consists of the regions 4631listed below. In most cases, these regions are allocated in the 4632order listed, with the last listed region closest to the top of the 4633stack (the lowest address if @code{STACK_GROWS_DOWNWARD} is defined, and 4634the highest address if it is not defined). You can use a different order 4635for a machine if doing so is more convenient or required for 4636compatibility reasons. Except in cases where required by standard 4637or by a debugger, there is no reason why the stack layout used by GCC 4638need agree with that used by other compilers for a machine. 4639@end deftypefn 4640 4641@hook TARGET_ASM_FUNCTION_END_PROLOGUE 4642If defined, a function that outputs assembler code at the end of a 4643prologue. This should be used when the function prologue is being 4644emitted as RTL, and you have some extra assembler that needs to be 4645emitted. @xref{prologue instruction pattern}. 4646@end deftypefn 4647 4648@hook TARGET_ASM_FUNCTION_BEGIN_EPILOGUE 4649If defined, a function that outputs assembler code at the start of an 4650epilogue. This should be used when the function epilogue is being 4651emitted as RTL, and you have some extra assembler that needs to be 4652emitted. @xref{epilogue instruction pattern}. 4653@end deftypefn 4654 4655@hook TARGET_ASM_FUNCTION_EPILOGUE 4656If defined, a function that outputs the assembler code for exit from a 4657function. The epilogue is responsible for restoring the saved 4658registers and stack pointer to their values when the function was 4659called, and returning control to the caller. This macro takes the 4660same arguments as the macro @code{TARGET_ASM_FUNCTION_PROLOGUE}, and the 4661registers to restore are determined from @code{regs_ever_live} and 4662@code{CALL_USED_REGISTERS} in the same way. 4663 4664On some machines, there is a single instruction that does all the work 4665of returning from the function. On these machines, give that 4666instruction the name @samp{return} and do not define the macro 4667@code{TARGET_ASM_FUNCTION_EPILOGUE} at all. 4668 4669Do not define a pattern named @samp{return} if you want the 4670@code{TARGET_ASM_FUNCTION_EPILOGUE} to be used. If you want the target 4671switches to control whether return instructions or epilogues are used, 4672define a @samp{return} pattern with a validity condition that tests the 4673target switches appropriately. If the @samp{return} pattern's validity 4674condition is false, epilogues will be used. 4675 4676On machines where functions may or may not have frame-pointers, the 4677function exit code must vary accordingly. Sometimes the code for these 4678two cases is completely different. To determine whether a frame pointer 4679is wanted, the macro can refer to the variable 4680@code{frame_pointer_needed}. The variable's value will be 1 when compiling 4681a function that needs a frame pointer. 4682 4683Normally, @code{TARGET_ASM_FUNCTION_PROLOGUE} and 4684@code{TARGET_ASM_FUNCTION_EPILOGUE} must treat leaf functions specially. 4685The C variable @code{current_function_is_leaf} is nonzero for such a 4686function. @xref{Leaf Functions}. 4687 4688On some machines, some functions pop their arguments on exit while 4689others leave that for the caller to do. For example, the 68020 when 4690given @option{-mrtd} pops arguments in functions that take a fixed 4691number of arguments. 4692 4693@findex current_function_pops_args 4694Your definition of the macro @code{RETURN_POPS_ARGS} decides which 4695functions pop their own arguments. @code{TARGET_ASM_FUNCTION_EPILOGUE} 4696needs to know what was decided. The number of bytes of the current 4697function's arguments that this function should pop is available in 4698@code{crtl->args.pops_args}. @xref{Scalar Return}. 4699@end deftypefn 4700 4701@itemize @bullet 4702@item 4703@findex current_function_pretend_args_size 4704A region of @code{current_function_pretend_args_size} bytes of 4705uninitialized space just underneath the first argument arriving on the 4706stack. (This may not be at the very start of the allocated stack region 4707if the calling sequence has pushed anything else since pushing the stack 4708arguments. But usually, on such machines, nothing else has been pushed 4709yet, because the function prologue itself does all the pushing.) This 4710region is used on machines where an argument may be passed partly in 4711registers and partly in memory, and, in some cases to support the 4712features in @code{<stdarg.h>}. 4713 4714@item 4715An area of memory used to save certain registers used by the function. 4716The size of this area, which may also include space for such things as 4717the return address and pointers to previous stack frames, is 4718machine-specific and usually depends on which registers have been used 4719in the function. Machines with register windows often do not require 4720a save area. 4721 4722@item 4723A region of at least @var{size} bytes, possibly rounded up to an allocation 4724boundary, to contain the local variables of the function. On some machines, 4725this region and the save area may occur in the opposite order, with the 4726save area closer to the top of the stack. 4727 4728@item 4729@cindex @code{ACCUMULATE_OUTGOING_ARGS} and stack frames 4730Optionally, when @code{ACCUMULATE_OUTGOING_ARGS} is defined, a region of 4731@code{current_function_outgoing_args_size} bytes to be used for outgoing 4732argument lists of the function. @xref{Stack Arguments}. 4733@end itemize 4734 4735@defmac EXIT_IGNORE_STACK 4736Define this macro as a C expression that is nonzero if the return 4737instruction or the function epilogue ignores the value of the stack 4738pointer; in other words, if it is safe to delete an instruction to 4739adjust the stack pointer before a return from the function. The 4740default is 0. 4741 4742Note that this macro's value is relevant only for functions for which 4743frame pointers are maintained. It is never safe to delete a final 4744stack adjustment in a function that has no frame pointer, and the 4745compiler knows this regardless of @code{EXIT_IGNORE_STACK}. 4746@end defmac 4747 4748@defmac EPILOGUE_USES (@var{regno}) 4749Define this macro as a C expression that is nonzero for registers that are 4750used by the epilogue or the @samp{return} pattern. The stack and frame 4751pointer registers are already assumed to be used as needed. 4752@end defmac 4753 4754@defmac EH_USES (@var{regno}) 4755Define this macro as a C expression that is nonzero for registers that are 4756used by the exception handling mechanism, and so should be considered live 4757on entry to an exception edge. 4758@end defmac 4759 4760@defmac DELAY_SLOTS_FOR_EPILOGUE 4761Define this macro if the function epilogue contains delay slots to which 4762instructions from the rest of the function can be ``moved''. The 4763definition should be a C expression whose value is an integer 4764representing the number of delay slots there. 4765@end defmac 4766 4767@defmac ELIGIBLE_FOR_EPILOGUE_DELAY (@var{insn}, @var{n}) 4768A C expression that returns 1 if @var{insn} can be placed in delay 4769slot number @var{n} of the epilogue. 4770 4771The argument @var{n} is an integer which identifies the delay slot now 4772being considered (since different slots may have different rules of 4773eligibility). It is never negative and is always less than the number 4774of epilogue delay slots (what @code{DELAY_SLOTS_FOR_EPILOGUE} returns). 4775If you reject a particular insn for a given delay slot, in principle, it 4776may be reconsidered for a subsequent delay slot. Also, other insns may 4777(at least in principle) be considered for the so far unfilled delay 4778slot. 4779 4780@findex current_function_epilogue_delay_list 4781@findex final_scan_insn 4782The insns accepted to fill the epilogue delay slots are put in an RTL 4783list made with @code{insn_list} objects, stored in the variable 4784@code{current_function_epilogue_delay_list}. The insn for the first 4785delay slot comes first in the list. Your definition of the macro 4786@code{TARGET_ASM_FUNCTION_EPILOGUE} should fill the delay slots by 4787outputting the insns in this list, usually by calling 4788@code{final_scan_insn}. 4789 4790You need not define this macro if you did not define 4791@code{DELAY_SLOTS_FOR_EPILOGUE}. 4792@end defmac 4793 4794@hook TARGET_ASM_OUTPUT_MI_THUNK 4795A function that outputs the assembler code for a thunk 4796function, used to implement C++ virtual function calls with multiple 4797inheritance. The thunk acts as a wrapper around a virtual function, 4798adjusting the implicit object parameter before handing control off to 4799the real function. 4800 4801First, emit code to add the integer @var{delta} to the location that 4802contains the incoming first argument. Assume that this argument 4803contains a pointer, and is the one used to pass the @code{this} pointer 4804in C++. This is the incoming argument @emph{before} the function prologue, 4805e.g.@: @samp{%o0} on a sparc. The addition must preserve the values of 4806all other incoming arguments. 4807 4808Then, if @var{vcall_offset} is nonzero, an additional adjustment should be 4809made after adding @code{delta}. In particular, if @var{p} is the 4810adjusted pointer, the following adjustment should be made: 4811 4812@smallexample 4813p += (*((ptrdiff_t **)p))[vcall_offset/sizeof(ptrdiff_t)] 4814@end smallexample 4815 4816After the additions, emit code to jump to @var{function}, which is a 4817@code{FUNCTION_DECL}. This is a direct pure jump, not a call, and does 4818not touch the return address. Hence returning from @var{FUNCTION} will 4819return to whoever called the current @samp{thunk}. 4820 4821The effect must be as if @var{function} had been called directly with 4822the adjusted first argument. This macro is responsible for emitting all 4823of the code for a thunk function; @code{TARGET_ASM_FUNCTION_PROLOGUE} 4824and @code{TARGET_ASM_FUNCTION_EPILOGUE} are not invoked. 4825 4826The @var{thunk_fndecl} is redundant. (@var{delta} and @var{function} 4827have already been extracted from it.) It might possibly be useful on 4828some targets, but probably not. 4829 4830If you do not define this macro, the target-independent code in the C++ 4831front end will generate a less efficient heavyweight thunk that calls 4832@var{function} instead of jumping to it. The generic approach does 4833not support varargs. 4834@end deftypefn 4835 4836@hook TARGET_ASM_CAN_OUTPUT_MI_THUNK 4837A function that returns true if TARGET_ASM_OUTPUT_MI_THUNK would be able 4838to output the assembler code for the thunk function specified by the 4839arguments it is passed, and false otherwise. In the latter case, the 4840generic approach will be used by the C++ front end, with the limitations 4841previously exposed. 4842@end deftypefn 4843 4844@node Profiling 4845@subsection Generating Code for Profiling 4846@cindex profiling, code generation 4847 4848These macros will help you generate code for profiling. 4849 4850@defmac FUNCTION_PROFILER (@var{file}, @var{labelno}) 4851A C statement or compound statement to output to @var{file} some 4852assembler code to call the profiling subroutine @code{mcount}. 4853 4854@findex mcount 4855The details of how @code{mcount} expects to be called are determined by 4856your operating system environment, not by GCC@. To figure them out, 4857compile a small program for profiling using the system's installed C 4858compiler and look at the assembler code that results. 4859 4860Older implementations of @code{mcount} expect the address of a counter 4861variable to be loaded into some register. The name of this variable is 4862@samp{LP} followed by the number @var{labelno}, so you would generate 4863the name using @samp{LP%d} in a @code{fprintf}. 4864@end defmac 4865 4866@defmac PROFILE_HOOK 4867A C statement or compound statement to output to @var{file} some assembly 4868code to call the profiling subroutine @code{mcount} even the target does 4869not support profiling. 4870@end defmac 4871 4872@defmac NO_PROFILE_COUNTERS 4873Define this macro to be an expression with a nonzero value if the 4874@code{mcount} subroutine on your system does not need a counter variable 4875allocated for each function. This is true for almost all modern 4876implementations. If you define this macro, you must not use the 4877@var{labelno} argument to @code{FUNCTION_PROFILER}. 4878@end defmac 4879 4880@defmac PROFILE_BEFORE_PROLOGUE 4881Define this macro if the code for function profiling should come before 4882the function prologue. Normally, the profiling code comes after. 4883@end defmac 4884 4885@node Tail Calls 4886@subsection Permitting tail calls 4887@cindex tail calls 4888 4889@hook TARGET_FUNCTION_OK_FOR_SIBCALL 4890True if it is ok to do sibling call optimization for the specified 4891call expression @var{exp}. @var{decl} will be the called function, 4892or @code{NULL} if this is an indirect call. 4893 4894It is not uncommon for limitations of calling conventions to prevent 4895tail calls to functions outside the current unit of translation, or 4896during PIC compilation. The hook is used to enforce these restrictions, 4897as the @code{sibcall} md pattern can not fail, or fall over to a 4898``normal'' call. The criteria for successful sibling call optimization 4899may vary greatly between different architectures. 4900@end deftypefn 4901 4902@hook TARGET_EXTRA_LIVE_ON_ENTRY 4903Add any hard registers to @var{regs} that are live on entry to the 4904function. This hook only needs to be defined to provide registers that 4905cannot be found by examination of FUNCTION_ARG_REGNO_P, the callee saved 4906registers, STATIC_CHAIN_INCOMING_REGNUM, STATIC_CHAIN_REGNUM, 4907TARGET_STRUCT_VALUE_RTX, FRAME_POINTER_REGNUM, EH_USES, 4908FRAME_POINTER_REGNUM, ARG_POINTER_REGNUM, and the PIC_OFFSET_TABLE_REGNUM. 4909@end deftypefn 4910 4911@hook TARGET_SET_UP_BY_PROLOGUE 4912 4913@node Stack Smashing Protection 4914@subsection Stack smashing protection 4915@cindex stack smashing protection 4916 4917@hook TARGET_STACK_PROTECT_GUARD 4918This hook returns a @code{DECL} node for the external variable to use 4919for the stack protection guard. This variable is initialized by the 4920runtime to some random value and is used to initialize the guard value 4921that is placed at the top of the local stack frame. The type of this 4922variable must be @code{ptr_type_node}. 4923 4924The default version of this hook creates a variable called 4925@samp{__stack_chk_guard}, which is normally defined in @file{libgcc2.c}. 4926@end deftypefn 4927 4928@hook TARGET_STACK_PROTECT_FAIL 4929This hook returns a tree expression that alerts the runtime that the 4930stack protect guard variable has been modified. This expression should 4931involve a call to a @code{noreturn} function. 4932 4933The default version of this hook invokes a function called 4934@samp{__stack_chk_fail}, taking no arguments. This function is 4935normally defined in @file{libgcc2.c}. 4936@end deftypefn 4937 4938@hook TARGET_SUPPORTS_SPLIT_STACK 4939 4940@node Varargs 4941@section Implementing the Varargs Macros 4942@cindex varargs implementation 4943 4944GCC comes with an implementation of @code{<varargs.h>} and 4945@code{<stdarg.h>} that work without change on machines that pass arguments 4946on the stack. Other machines require their own implementations of 4947varargs, and the two machine independent header files must have 4948conditionals to include it. 4949 4950ISO @code{<stdarg.h>} differs from traditional @code{<varargs.h>} mainly in 4951the calling convention for @code{va_start}. The traditional 4952implementation takes just one argument, which is the variable in which 4953to store the argument pointer. The ISO implementation of 4954@code{va_start} takes an additional second argument. The user is 4955supposed to write the last named argument of the function here. 4956 4957However, @code{va_start} should not use this argument. The way to find 4958the end of the named arguments is with the built-in functions described 4959below. 4960 4961@defmac __builtin_saveregs () 4962Use this built-in function to save the argument registers in memory so 4963that the varargs mechanism can access them. Both ISO and traditional 4964versions of @code{va_start} must use @code{__builtin_saveregs}, unless 4965you use @code{TARGET_SETUP_INCOMING_VARARGS} (see below) instead. 4966 4967On some machines, @code{__builtin_saveregs} is open-coded under the 4968control of the target hook @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. On 4969other machines, it calls a routine written in assembler language, 4970found in @file{libgcc2.c}. 4971 4972Code generated for the call to @code{__builtin_saveregs} appears at the 4973beginning of the function, as opposed to where the call to 4974@code{__builtin_saveregs} is written, regardless of what the code is. 4975This is because the registers must be saved before the function starts 4976to use them for its own purposes. 4977@c i rewrote the first sentence above to fix an overfull hbox. --mew 4978@c 10feb93 4979@end defmac 4980 4981@defmac __builtin_next_arg (@var{lastarg}) 4982This builtin returns the address of the first anonymous stack 4983argument, as type @code{void *}. If @code{ARGS_GROW_DOWNWARD}, it 4984returns the address of the location above the first anonymous stack 4985argument. Use it in @code{va_start} to initialize the pointer for 4986fetching arguments from the stack. Also use it in @code{va_start} to 4987verify that the second parameter @var{lastarg} is the last named argument 4988of the current function. 4989@end defmac 4990 4991@defmac __builtin_classify_type (@var{object}) 4992Since each machine has its own conventions for which data types are 4993passed in which kind of register, your implementation of @code{va_arg} 4994has to embody these conventions. The easiest way to categorize the 4995specified data type is to use @code{__builtin_classify_type} together 4996with @code{sizeof} and @code{__alignof__}. 4997 4998@code{__builtin_classify_type} ignores the value of @var{object}, 4999considering only its data type. It returns an integer describing what 5000kind of type that is---integer, floating, pointer, structure, and so on. 5001 5002The file @file{typeclass.h} defines an enumeration that you can use to 5003interpret the values of @code{__builtin_classify_type}. 5004@end defmac 5005 5006These machine description macros help implement varargs: 5007 5008@hook TARGET_EXPAND_BUILTIN_SAVEREGS 5009If defined, this hook produces the machine-specific code for a call to 5010@code{__builtin_saveregs}. This code will be moved to the very 5011beginning of the function, before any parameter access are made. The 5012return value of this function should be an RTX that contains the value 5013to use as the return of @code{__builtin_saveregs}. 5014@end deftypefn 5015 5016@hook TARGET_SETUP_INCOMING_VARARGS 5017This target hook offers an alternative to using 5018@code{__builtin_saveregs} and defining the hook 5019@code{TARGET_EXPAND_BUILTIN_SAVEREGS}. Use it to store the anonymous 5020register arguments into the stack so that all the arguments appear to 5021have been passed consecutively on the stack. Once this is done, you can 5022use the standard implementation of varargs that works for machines that 5023pass all their arguments on the stack. 5024 5025The argument @var{args_so_far} points to the @code{CUMULATIVE_ARGS} data 5026structure, containing the values that are obtained after processing the 5027named arguments. The arguments @var{mode} and @var{type} describe the 5028last named argument---its machine mode and its data type as a tree node. 5029 5030The target hook should do two things: first, push onto the stack all the 5031argument registers @emph{not} used for the named arguments, and second, 5032store the size of the data thus pushed into the @code{int}-valued 5033variable pointed to by @var{pretend_args_size}. The value that you 5034store here will serve as additional offset for setting up the stack 5035frame. 5036 5037Because you must generate code to push the anonymous arguments at 5038compile time without knowing their data types, 5039@code{TARGET_SETUP_INCOMING_VARARGS} is only useful on machines that 5040have just a single category of argument register and use it uniformly 5041for all data types. 5042 5043If the argument @var{second_time} is nonzero, it means that the 5044arguments of the function are being analyzed for the second time. This 5045happens for an inline function, which is not actually compiled until the 5046end of the source file. The hook @code{TARGET_SETUP_INCOMING_VARARGS} should 5047not generate any instructions in this case. 5048@end deftypefn 5049 5050@hook TARGET_STRICT_ARGUMENT_NAMING 5051Define this hook to return @code{true} if the location where a function 5052argument is passed depends on whether or not it is a named argument. 5053 5054This hook controls how the @var{named} argument to @code{TARGET_FUNCTION_ARG} 5055is set for varargs and stdarg functions. If this hook returns 5056@code{true}, the @var{named} argument is always true for named 5057arguments, and false for unnamed arguments. If it returns @code{false}, 5058but @code{TARGET_PRETEND_OUTGOING_VARARGS_NAMED} returns @code{true}, 5059then all arguments are treated as named. Otherwise, all named arguments 5060except the last are treated as named. 5061 5062You need not define this hook if it always returns @code{false}. 5063@end deftypefn 5064 5065@hook TARGET_PRETEND_OUTGOING_VARARGS_NAMED 5066If you need to conditionally change ABIs so that one works with 5067@code{TARGET_SETUP_INCOMING_VARARGS}, but the other works like neither 5068@code{TARGET_SETUP_INCOMING_VARARGS} nor @code{TARGET_STRICT_ARGUMENT_NAMING} was 5069defined, then define this hook to return @code{true} if 5070@code{TARGET_SETUP_INCOMING_VARARGS} is used, @code{false} otherwise. 5071Otherwise, you should not define this hook. 5072@end deftypefn 5073 5074@node Trampolines 5075@section Trampolines for Nested Functions 5076@cindex trampolines for nested functions 5077@cindex nested functions, trampolines for 5078 5079A @dfn{trampoline} is a small piece of code that is created at run time 5080when the address of a nested function is taken. It normally resides on 5081the stack, in the stack frame of the containing function. These macros 5082tell GCC how to generate code to allocate and initialize a 5083trampoline. 5084 5085The instructions in the trampoline must do two things: load a constant 5086address into the static chain register, and jump to the real address of 5087the nested function. On CISC machines such as the m68k, this requires 5088two instructions, a move immediate and a jump. Then the two addresses 5089exist in the trampoline as word-long immediate operands. On RISC 5090machines, it is often necessary to load each address into a register in 5091two parts. Then pieces of each address form separate immediate 5092operands. 5093 5094The code generated to initialize the trampoline must store the variable 5095parts---the static chain value and the function address---into the 5096immediate operands of the instructions. On a CISC machine, this is 5097simply a matter of copying each address to a memory reference at the 5098proper offset from the start of the trampoline. On a RISC machine, it 5099may be necessary to take out pieces of the address and store them 5100separately. 5101 5102@hook TARGET_ASM_TRAMPOLINE_TEMPLATE 5103This hook is called by @code{assemble_trampoline_template} to output, 5104on the stream @var{f}, assembler code for a block of data that contains 5105the constant parts of a trampoline. This code should not include a 5106label---the label is taken care of automatically. 5107 5108If you do not define this hook, it means no template is needed 5109for the target. Do not define this hook on systems where the block move 5110code to copy the trampoline into place would be larger than the code 5111to generate it on the spot. 5112@end deftypefn 5113 5114@defmac TRAMPOLINE_SECTION 5115Return the section into which the trampoline template is to be placed 5116(@pxref{Sections}). The default value is @code{readonly_data_section}. 5117@end defmac 5118 5119@defmac TRAMPOLINE_SIZE 5120A C expression for the size in bytes of the trampoline, as an integer. 5121@end defmac 5122 5123@defmac TRAMPOLINE_ALIGNMENT 5124Alignment required for trampolines, in bits. 5125 5126If you don't define this macro, the value of @code{FUNCTION_ALIGNMENT} 5127is used for aligning trampolines. 5128@end defmac 5129 5130@hook TARGET_TRAMPOLINE_INIT 5131This hook is called to initialize a trampoline. 5132@var{m_tramp} is an RTX for the memory block for the trampoline; @var{fndecl} 5133is the @code{FUNCTION_DECL} for the nested function; @var{static_chain} is an 5134RTX for the static chain value that should be passed to the function 5135when it is called. 5136 5137If the target defines @code{TARGET_ASM_TRAMPOLINE_TEMPLATE}, then the 5138first thing this hook should do is emit a block move into @var{m_tramp} 5139from the memory block returned by @code{assemble_trampoline_template}. 5140Note that the block move need only cover the constant parts of the 5141trampoline. If the target isolates the variable parts of the trampoline 5142to the end, not all @code{TRAMPOLINE_SIZE} bytes need be copied. 5143 5144If the target requires any other actions, such as flushing caches or 5145enabling stack execution, these actions should be performed after 5146initializing the trampoline proper. 5147@end deftypefn 5148 5149@hook TARGET_TRAMPOLINE_ADJUST_ADDRESS 5150This hook should perform any machine-specific adjustment in 5151the address of the trampoline. Its argument contains the address of the 5152memory block that was passed to @code{TARGET_TRAMPOLINE_INIT}. In case 5153the address to be used for a function call should be different from the 5154address at which the template was stored, the different address should 5155be returned; otherwise @var{addr} should be returned unchanged. 5156If this hook is not defined, @var{addr} will be used for function calls. 5157@end deftypefn 5158 5159Implementing trampolines is difficult on many machines because they have 5160separate instruction and data caches. Writing into a stack location 5161fails to clear the memory in the instruction cache, so when the program 5162jumps to that location, it executes the old contents. 5163 5164Here are two possible solutions. One is to clear the relevant parts of 5165the instruction cache whenever a trampoline is set up. The other is to 5166make all trampolines identical, by having them jump to a standard 5167subroutine. The former technique makes trampoline execution faster; the 5168latter makes initialization faster. 5169 5170To clear the instruction cache when a trampoline is initialized, define 5171the following macro. 5172 5173@defmac CLEAR_INSN_CACHE (@var{beg}, @var{end}) 5174If defined, expands to a C expression clearing the @emph{instruction 5175cache} in the specified interval. The definition of this macro would 5176typically be a series of @code{asm} statements. Both @var{beg} and 5177@var{end} are both pointer expressions. 5178@end defmac 5179 5180To use a standard subroutine, define the following macro. In addition, 5181you must make sure that the instructions in a trampoline fill an entire 5182cache line with identical instructions, or else ensure that the 5183beginning of the trampoline code is always aligned at the same point in 5184its cache line. Look in @file{m68k.h} as a guide. 5185 5186@defmac TRANSFER_FROM_TRAMPOLINE 5187Define this macro if trampolines need a special subroutine to do their 5188work. The macro should expand to a series of @code{asm} statements 5189which will be compiled with GCC@. They go in a library function named 5190@code{__transfer_from_trampoline}. 5191 5192If you need to avoid executing the ordinary prologue code of a compiled 5193C function when you jump to the subroutine, you can do so by placing a 5194special label of your own in the assembler code. Use one @code{asm} 5195statement to generate an assembler label, and another to make the label 5196global. Then trampolines can use that label to jump directly to your 5197special assembler code. 5198@end defmac 5199 5200@node Library Calls 5201@section Implicit Calls to Library Routines 5202@cindex library subroutine names 5203@cindex @file{libgcc.a} 5204 5205@c prevent bad page break with this line 5206Here is an explanation of implicit calls to library routines. 5207 5208@defmac DECLARE_LIBRARY_RENAMES 5209This macro, if defined, should expand to a piece of C code that will get 5210expanded when compiling functions for libgcc.a. It can be used to 5211provide alternate names for GCC's internal library functions if there 5212are ABI-mandated names that the compiler should provide. 5213@end defmac 5214 5215@findex set_optab_libfunc 5216@findex init_one_libfunc 5217@hook TARGET_INIT_LIBFUNCS 5218This hook should declare additional library routines or rename 5219existing ones, using the functions @code{set_optab_libfunc} and 5220@code{init_one_libfunc} defined in @file{optabs.c}. 5221@code{init_optabs} calls this macro after initializing all the normal 5222library routines. 5223 5224The default is to do nothing. Most ports don't need to define this hook. 5225@end deftypefn 5226 5227@hook TARGET_LIBFUNC_GNU_PREFIX 5228 5229@defmac FLOAT_LIB_COMPARE_RETURNS_BOOL (@var{mode}, @var{comparison}) 5230This macro should return @code{true} if the library routine that 5231implements the floating point comparison operator @var{comparison} in 5232mode @var{mode} will return a boolean, and @var{false} if it will 5233return a tristate. 5234 5235GCC's own floating point libraries return tristates from the 5236comparison operators, so the default returns false always. Most ports 5237don't need to define this macro. 5238@end defmac 5239 5240@defmac TARGET_LIB_INT_CMP_BIASED 5241This macro should evaluate to @code{true} if the integer comparison 5242functions (like @code{__cmpdi2}) return 0 to indicate that the first 5243operand is smaller than the second, 1 to indicate that they are equal, 5244and 2 to indicate that the first operand is greater than the second. 5245If this macro evaluates to @code{false} the comparison functions return 5246@minus{}1, 0, and 1 instead of 0, 1, and 2. If the target uses the routines 5247in @file{libgcc.a}, you do not need to define this macro. 5248@end defmac 5249 5250@cindex @code{EDOM}, implicit usage 5251@findex matherr 5252@defmac TARGET_EDOM 5253The value of @code{EDOM} on the target machine, as a C integer constant 5254expression. If you don't define this macro, GCC does not attempt to 5255deposit the value of @code{EDOM} into @code{errno} directly. Look in 5256@file{/usr/include/errno.h} to find the value of @code{EDOM} on your 5257system. 5258 5259If you do not define @code{TARGET_EDOM}, then compiled code reports 5260domain errors by calling the library function and letting it report the 5261error. If mathematical functions on your system use @code{matherr} when 5262there is an error, then you should leave @code{TARGET_EDOM} undefined so 5263that @code{matherr} is used normally. 5264@end defmac 5265 5266@cindex @code{errno}, implicit usage 5267@defmac GEN_ERRNO_RTX 5268Define this macro as a C expression to create an rtl expression that 5269refers to the global ``variable'' @code{errno}. (On certain systems, 5270@code{errno} may not actually be a variable.) If you don't define this 5271macro, a reasonable default is used. 5272@end defmac 5273 5274@cindex C99 math functions, implicit usage 5275@defmac TARGET_C99_FUNCTIONS 5276When this macro is nonzero, GCC will implicitly optimize @code{sin} calls into 5277@code{sinf} and similarly for other functions defined by C99 standard. The 5278default is zero because a number of existing systems lack support for these 5279functions in their runtime so this macro needs to be redefined to one on 5280systems that do support the C99 runtime. 5281@end defmac 5282 5283@cindex sincos math function, implicit usage 5284@defmac TARGET_HAS_SINCOS 5285When this macro is nonzero, GCC will implicitly optimize calls to @code{sin} 5286and @code{cos} with the same argument to a call to @code{sincos}. The 5287default is zero. The target has to provide the following functions: 5288@smallexample 5289void sincos(double x, double *sin, double *cos); 5290void sincosf(float x, float *sin, float *cos); 5291void sincosl(long double x, long double *sin, long double *cos); 5292@end smallexample 5293@end defmac 5294 5295@defmac NEXT_OBJC_RUNTIME 5296Set this macro to 1 to use the "NeXT" Objective-C message sending conventions 5297by default. This calling convention involves passing the object, the selector 5298and the method arguments all at once to the method-lookup library function. 5299This is the usual setting when targeting Darwin/Mac OS X systems, which have 5300the NeXT runtime installed. 5301 5302If the macro is set to 0, the "GNU" Objective-C message sending convention 5303will be used by default. This convention passes just the object and the 5304selector to the method-lookup function, which returns a pointer to the method. 5305 5306In either case, it remains possible to select code-generation for the alternate 5307scheme, by means of compiler command line switches. 5308@end defmac 5309 5310@node Addressing Modes 5311@section Addressing Modes 5312@cindex addressing modes 5313 5314@c prevent bad page break with this line 5315This is about addressing modes. 5316 5317@defmac HAVE_PRE_INCREMENT 5318@defmacx HAVE_PRE_DECREMENT 5319@defmacx HAVE_POST_INCREMENT 5320@defmacx HAVE_POST_DECREMENT 5321A C expression that is nonzero if the machine supports pre-increment, 5322pre-decrement, post-increment, or post-decrement addressing respectively. 5323@end defmac 5324 5325@defmac HAVE_PRE_MODIFY_DISP 5326@defmacx HAVE_POST_MODIFY_DISP 5327A C expression that is nonzero if the machine supports pre- or 5328post-address side-effect generation involving constants other than 5329the size of the memory operand. 5330@end defmac 5331 5332@defmac HAVE_PRE_MODIFY_REG 5333@defmacx HAVE_POST_MODIFY_REG 5334A C expression that is nonzero if the machine supports pre- or 5335post-address side-effect generation involving a register displacement. 5336@end defmac 5337 5338@defmac CONSTANT_ADDRESS_P (@var{x}) 5339A C expression that is 1 if the RTX @var{x} is a constant which 5340is a valid address. On most machines the default definition of 5341@code{(CONSTANT_P (@var{x}) && GET_CODE (@var{x}) != CONST_DOUBLE)} 5342is acceptable, but a few machines are more restrictive as to which 5343constant addresses are supported. 5344@end defmac 5345 5346@defmac CONSTANT_P (@var{x}) 5347@code{CONSTANT_P}, which is defined by target-independent code, 5348accepts integer-values expressions whose values are not explicitly 5349known, such as @code{symbol_ref}, @code{label_ref}, and @code{high} 5350expressions and @code{const} arithmetic expressions, in addition to 5351@code{const_int} and @code{const_double} expressions. 5352@end defmac 5353 5354@defmac MAX_REGS_PER_ADDRESS 5355A number, the maximum number of registers that can appear in a valid 5356memory address. Note that it is up to you to specify a value equal to 5357the maximum number that @code{TARGET_LEGITIMATE_ADDRESS_P} would ever 5358accept. 5359@end defmac 5360 5361@hook TARGET_LEGITIMATE_ADDRESS_P 5362A function that returns whether @var{x} (an RTX) is a legitimate memory 5363address on the target machine for a memory operand of mode @var{mode}. 5364 5365Legitimate addresses are defined in two variants: a strict variant and a 5366non-strict one. The @var{strict} parameter chooses which variant is 5367desired by the caller. 5368 5369The strict variant is used in the reload pass. It must be defined so 5370that any pseudo-register that has not been allocated a hard register is 5371considered a memory reference. This is because in contexts where some 5372kind of register is required, a pseudo-register with no hard register 5373must be rejected. For non-hard registers, the strict variant should look 5374up the @code{reg_renumber} array; it should then proceed using the hard 5375register number in the array, or treat the pseudo as a memory reference 5376if the array holds @code{-1}. 5377 5378The non-strict variant is used in other passes. It must be defined to 5379accept all pseudo-registers in every context where some kind of 5380register is required. 5381 5382Normally, constant addresses which are the sum of a @code{symbol_ref} 5383and an integer are stored inside a @code{const} RTX to mark them as 5384constant. Therefore, there is no need to recognize such sums 5385specifically as legitimate addresses. Normally you would simply 5386recognize any @code{const} as legitimate. 5387 5388Usually @code{PRINT_OPERAND_ADDRESS} is not prepared to handle constant 5389sums that are not marked with @code{const}. It assumes that a naked 5390@code{plus} indicates indexing. If so, then you @emph{must} reject such 5391naked constant sums as illegitimate addresses, so that none of them will 5392be given to @code{PRINT_OPERAND_ADDRESS}. 5393 5394@cindex @code{TARGET_ENCODE_SECTION_INFO} and address validation 5395On some machines, whether a symbolic address is legitimate depends on 5396the section that the address refers to. On these machines, define the 5397target hook @code{TARGET_ENCODE_SECTION_INFO} to store the information 5398into the @code{symbol_ref}, and then check for it here. When you see a 5399@code{const}, you will have to look inside it to find the 5400@code{symbol_ref} in order to determine the section. @xref{Assembler 5401Format}. 5402 5403@cindex @code{GO_IF_LEGITIMATE_ADDRESS} 5404Some ports are still using a deprecated legacy substitute for 5405this hook, the @code{GO_IF_LEGITIMATE_ADDRESS} macro. This macro 5406has this syntax: 5407 5408@example 5409#define GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label}) 5410@end example 5411 5412@noindent 5413and should @code{goto @var{label}} if the address @var{x} is a valid 5414address on the target machine for a memory operand of mode @var{mode}. 5415 5416@findex REG_OK_STRICT 5417Compiler source files that want to use the strict variant of this 5418macro define the macro @code{REG_OK_STRICT}. You should use an 5419@code{#ifdef REG_OK_STRICT} conditional to define the strict variant in 5420that case and the non-strict variant otherwise. 5421 5422Using the hook is usually simpler because it limits the number of 5423files that are recompiled when changes are made. 5424@end deftypefn 5425 5426@defmac TARGET_MEM_CONSTRAINT 5427A single character to be used instead of the default @code{'m'} 5428character for general memory addresses. This defines the constraint 5429letter which matches the memory addresses accepted by 5430@code{TARGET_LEGITIMATE_ADDRESS_P}. Define this macro if you want to 5431support new address formats in your back end without changing the 5432semantics of the @code{'m'} constraint. This is necessary in order to 5433preserve functionality of inline assembly constructs using the 5434@code{'m'} constraint. 5435@end defmac 5436 5437@defmac FIND_BASE_TERM (@var{x}) 5438A C expression to determine the base term of address @var{x}, 5439or to provide a simplified version of @var{x} from which @file{alias.c} 5440can easily find the base term. This macro is used in only two places: 5441@code{find_base_value} and @code{find_base_term} in @file{alias.c}. 5442 5443It is always safe for this macro to not be defined. It exists so 5444that alias analysis can understand machine-dependent addresses. 5445 5446The typical use of this macro is to handle addresses containing 5447a label_ref or symbol_ref within an UNSPEC@. 5448@end defmac 5449 5450@hook TARGET_LEGITIMIZE_ADDRESS 5451This hook is given an invalid memory address @var{x} for an 5452operand of mode @var{mode} and should try to return a valid memory 5453address. 5454 5455@findex break_out_memory_refs 5456@var{x} will always be the result of a call to @code{break_out_memory_refs}, 5457and @var{oldx} will be the operand that was given to that function to produce 5458@var{x}. 5459 5460The code of the hook should not alter the substructure of 5461@var{x}. If it transforms @var{x} into a more legitimate form, it 5462should return the new @var{x}. 5463 5464It is not necessary for this hook to come up with a legitimate address. 5465The compiler has standard ways of doing so in all cases. In fact, it 5466is safe to omit this hook or make it return @var{x} if it cannot find 5467a valid way to legitimize the address. But often a machine-dependent 5468strategy can generate better code. 5469@end deftypefn 5470 5471@defmac LEGITIMIZE_RELOAD_ADDRESS (@var{x}, @var{mode}, @var{opnum}, @var{type}, @var{ind_levels}, @var{win}) 5472A C compound statement that attempts to replace @var{x}, which is an address 5473that needs reloading, with a valid memory address for an operand of mode 5474@var{mode}. @var{win} will be a C statement label elsewhere in the code. 5475It is not necessary to define this macro, but it might be useful for 5476performance reasons. 5477 5478For example, on the i386, it is sometimes possible to use a single 5479reload register instead of two by reloading a sum of two pseudo 5480registers into a register. On the other hand, for number of RISC 5481processors offsets are limited so that often an intermediate address 5482needs to be generated in order to address a stack slot. By defining 5483@code{LEGITIMIZE_RELOAD_ADDRESS} appropriately, the intermediate addresses 5484generated for adjacent some stack slots can be made identical, and thus 5485be shared. 5486 5487@emph{Note}: This macro should be used with caution. It is necessary 5488to know something of how reload works in order to effectively use this, 5489and it is quite easy to produce macros that build in too much knowledge 5490of reload internals. 5491 5492@emph{Note}: This macro must be able to reload an address created by a 5493previous invocation of this macro. If it fails to handle such addresses 5494then the compiler may generate incorrect code or abort. 5495 5496@findex push_reload 5497The macro definition should use @code{push_reload} to indicate parts that 5498need reloading; @var{opnum}, @var{type} and @var{ind_levels} are usually 5499suitable to be passed unaltered to @code{push_reload}. 5500 5501The code generated by this macro must not alter the substructure of 5502@var{x}. If it transforms @var{x} into a more legitimate form, it 5503should assign @var{x} (which will always be a C variable) a new value. 5504This also applies to parts that you change indirectly by calling 5505@code{push_reload}. 5506 5507@findex strict_memory_address_p 5508The macro definition may use @code{strict_memory_address_p} to test if 5509the address has become legitimate. 5510 5511@findex copy_rtx 5512If you want to change only a part of @var{x}, one standard way of doing 5513this is to use @code{copy_rtx}. Note, however, that it unshares only a 5514single level of rtl. Thus, if the part to be changed is not at the 5515top level, you'll need to replace first the top level. 5516It is not necessary for this macro to come up with a legitimate 5517address; but often a machine-dependent strategy can generate better code. 5518@end defmac 5519 5520@hook TARGET_MODE_DEPENDENT_ADDRESS_P 5521This hook returns @code{true} if memory address @var{addr} can have 5522different meanings depending on the machine mode of the memory 5523reference it is used for or if the address is valid for some modes 5524but not others. 5525 5526Autoincrement and autodecrement addresses typically have mode-dependent 5527effects because the amount of the increment or decrement is the size 5528of the operand being addressed. Some machines have other mode-dependent 5529addresses. Many RISC machines have no mode-dependent addresses. 5530 5531You may assume that @var{addr} is a valid address for the machine. 5532 5533The default version of this hook returns @code{false}. 5534@end deftypefn 5535 5536@defmac GO_IF_MODE_DEPENDENT_ADDRESS (@var{addr}, @var{label}) 5537A C statement or compound statement with a conditional @code{goto 5538@var{label};} executed if memory address @var{x} (an RTX) can have 5539different meanings depending on the machine mode of the memory 5540reference it is used for or if the address is valid for some modes 5541but not others. 5542 5543Autoincrement and autodecrement addresses typically have mode-dependent 5544effects because the amount of the increment or decrement is the size 5545of the operand being addressed. Some machines have other mode-dependent 5546addresses. Many RISC machines have no mode-dependent addresses. 5547 5548You may assume that @var{addr} is a valid address for the machine. 5549 5550These are obsolete macros, replaced by the 5551@code{TARGET_MODE_DEPENDENT_ADDRESS_P} target hook. 5552@end defmac 5553 5554@hook TARGET_LEGITIMATE_CONSTANT_P 5555This hook returns true if @var{x} is a legitimate constant for a 5556@var{mode}-mode immediate operand on the target machine. You can assume that 5557@var{x} satisfies @code{CONSTANT_P}, so you need not check this. 5558 5559The default definition returns true. 5560@end deftypefn 5561 5562@hook TARGET_DELEGITIMIZE_ADDRESS 5563This hook is used to undo the possibly obfuscating effects of the 5564@code{LEGITIMIZE_ADDRESS} and @code{LEGITIMIZE_RELOAD_ADDRESS} target 5565macros. Some backend implementations of these macros wrap symbol 5566references inside an @code{UNSPEC} rtx to represent PIC or similar 5567addressing modes. This target hook allows GCC's optimizers to understand 5568the semantics of these opaque @code{UNSPEC}s by converting them back 5569into their original form. 5570@end deftypefn 5571 5572@hook TARGET_CONST_NOT_OK_FOR_DEBUG_P 5573This hook should return true if @var{x} should not be emitted into 5574debug sections. 5575@end deftypefn 5576 5577@hook TARGET_CANNOT_FORCE_CONST_MEM 5578This hook should return true if @var{x} is of a form that cannot (or 5579should not) be spilled to the constant pool. @var{mode} is the mode 5580of @var{x}. 5581 5582The default version of this hook returns false. 5583 5584The primary reason to define this hook is to prevent reload from 5585deciding that a non-legitimate constant would be better reloaded 5586from the constant pool instead of spilling and reloading a register 5587holding the constant. This restriction is often true of addresses 5588of TLS symbols for various targets. 5589@end deftypefn 5590 5591@hook TARGET_USE_BLOCKS_FOR_CONSTANT_P 5592This hook should return true if pool entries for constant @var{x} can 5593be placed in an @code{object_block} structure. @var{mode} is the mode 5594of @var{x}. 5595 5596The default version returns false for all constants. 5597@end deftypefn 5598 5599@hook TARGET_BUILTIN_RECIPROCAL 5600This hook should return the DECL of a function that implements reciprocal of 5601the builtin function with builtin function code @var{fn}, or 5602@code{NULL_TREE} if such a function is not available. @var{md_fn} is true 5603when @var{fn} is a code of a machine-dependent builtin function. When 5604@var{sqrt} is true, additional optimizations that apply only to the reciprocal 5605of a square root function are performed, and only reciprocals of @code{sqrt} 5606function are valid. 5607@end deftypefn 5608 5609@hook TARGET_VECTORIZE_BUILTIN_MASK_FOR_LOAD 5610This hook should return the DECL of a function @var{f} that given an 5611address @var{addr} as an argument returns a mask @var{m} that can be 5612used to extract from two vectors the relevant data that resides in 5613@var{addr} in case @var{addr} is not properly aligned. 5614 5615The autovectorizer, when vectorizing a load operation from an address 5616@var{addr} that may be unaligned, will generate two vector loads from 5617the two aligned addresses around @var{addr}. It then generates a 5618@code{REALIGN_LOAD} operation to extract the relevant data from the 5619two loaded vectors. The first two arguments to @code{REALIGN_LOAD}, 5620@var{v1} and @var{v2}, are the two vectors, each of size @var{VS}, and 5621the third argument, @var{OFF}, defines how the data will be extracted 5622from these two vectors: if @var{OFF} is 0, then the returned vector is 5623@var{v2}; otherwise, the returned vector is composed from the last 5624@var{VS}-@var{OFF} elements of @var{v1} concatenated to the first 5625@var{OFF} elements of @var{v2}. 5626 5627If this hook is defined, the autovectorizer will generate a call 5628to @var{f} (using the DECL tree that this hook returns) and will 5629use the return value of @var{f} as the argument @var{OFF} to 5630@code{REALIGN_LOAD}. Therefore, the mask @var{m} returned by @var{f} 5631should comply with the semantics expected by @code{REALIGN_LOAD} 5632described above. 5633If this hook is not defined, then @var{addr} will be used as 5634the argument @var{OFF} to @code{REALIGN_LOAD}, in which case the low 5635log2(@var{VS}) @minus{} 1 bits of @var{addr} will be considered. 5636@end deftypefn 5637 5638@hook TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_EVEN 5639This hook should return the DECL of a function @var{f} that implements 5640widening multiplication of the even elements of two input vectors of type @var{x}. 5641 5642If this hook is defined, the autovectorizer will use it along with the 5643@code{TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_ODD} target hook when vectorizing 5644widening multiplication in cases that the order of the results does not have to be 5645preserved (e.g.@: used only by a reduction computation). Otherwise, the 5646@code{widen_mult_hi/lo} idioms will be used. 5647@end deftypefn 5648 5649@hook TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_ODD 5650This hook should return the DECL of a function @var{f} that implements 5651widening multiplication of the odd elements of two input vectors of type @var{x}. 5652 5653If this hook is defined, the autovectorizer will use it along with the 5654@code{TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_EVEN} target hook when vectorizing 5655widening multiplication in cases that the order of the results does not have to be 5656preserved (e.g.@: used only by a reduction computation). Otherwise, the 5657@code{widen_mult_hi/lo} idioms will be used. 5658@end deftypefn 5659 5660@hook TARGET_VECTORIZE_BUILTIN_VECTORIZATION_COST 5661Returns cost of different scalar or vector statements for vectorization cost model. 5662For vector memory operations the cost may depend on type (@var{vectype}) and 5663misalignment value (@var{misalign}). 5664@end deftypefn 5665 5666@hook TARGET_VECTORIZE_VECTOR_ALIGNMENT_REACHABLE 5667Return true if vector alignment is reachable (by peeling N iterations) for the given type. 5668@end deftypefn 5669 5670@hook TARGET_VECTORIZE_VEC_PERM_CONST_OK 5671Return true if a vector created for @code{vec_perm_const} is valid. 5672@end deftypefn 5673 5674@hook TARGET_VECTORIZE_BUILTIN_CONVERSION 5675This hook should return the DECL of a function that implements conversion of the 5676input vector of type @var{src_type} to type @var{dest_type}. 5677The value of @var{code} is one of the enumerators in @code{enum tree_code} and 5678specifies how the conversion is to be applied 5679(truncation, rounding, etc.). 5680 5681If this hook is defined, the autovectorizer will use the 5682@code{TARGET_VECTORIZE_BUILTIN_CONVERSION} target hook when vectorizing 5683conversion. Otherwise, it will return @code{NULL_TREE}. 5684@end deftypefn 5685 5686@hook TARGET_VECTORIZE_BUILTIN_VECTORIZED_FUNCTION 5687This hook should return the decl of a function that implements the 5688vectorized variant of the builtin function with builtin function code 5689@var{code} or @code{NULL_TREE} if such a function is not available. 5690The value of @var{fndecl} is the builtin function declaration. The 5691return type of the vectorized function shall be of vector type 5692@var{vec_type_out} and the argument types should be @var{vec_type_in}. 5693@end deftypefn 5694 5695@hook TARGET_VECTORIZE_SUPPORT_VECTOR_MISALIGNMENT 5696This hook should return true if the target supports misaligned vector 5697store/load of a specific factor denoted in the @var{misalignment} 5698parameter. The vector store/load should be of machine mode @var{mode} and 5699the elements in the vectors should be of type @var{type}. @var{is_packed} 5700parameter is true if the memory access is defined in a packed struct. 5701@end deftypefn 5702 5703@hook TARGET_VECTORIZE_PREFERRED_SIMD_MODE 5704This hook should return the preferred mode for vectorizing scalar 5705mode @var{mode}. The default is 5706equal to @code{word_mode}, because the vectorizer can do some 5707transformations even in absence of specialized @acronym{SIMD} hardware. 5708@end deftypefn 5709 5710@hook TARGET_VECTORIZE_AUTOVECTORIZE_VECTOR_SIZES 5711This hook should return a mask of sizes that should be iterated over 5712after trying to autovectorize using the vector size derived from the 5713mode returned by @code{TARGET_VECTORIZE_PREFERRED_SIMD_MODE}. 5714The default is zero which means to not iterate over other vector sizes. 5715@end deftypefn 5716 5717@hook TARGET_VECTORIZE_BUILTIN_TM_LOAD 5718 5719@hook TARGET_VECTORIZE_BUILTIN_TM_STORE 5720 5721@hook TARGET_VECTORIZE_BUILTIN_GATHER 5722Target builtin that implements vector gather operation. @var{mem_vectype} 5723is the vector type of the load and @var{index_type} is scalar type of 5724the index, scaled by @var{scale}. 5725The default is @code{NULL_TREE} which means to not vectorize gather 5726loads. 5727@end deftypefn 5728 5729@node Anchored Addresses 5730@section Anchored Addresses 5731@cindex anchored addresses 5732@cindex @option{-fsection-anchors} 5733 5734GCC usually addresses every static object as a separate entity. 5735For example, if we have: 5736 5737@smallexample 5738static int a, b, c; 5739int foo (void) @{ return a + b + c; @} 5740@end smallexample 5741 5742the code for @code{foo} will usually calculate three separate symbolic 5743addresses: those of @code{a}, @code{b} and @code{c}. On some targets, 5744it would be better to calculate just one symbolic address and access 5745the three variables relative to it. The equivalent pseudocode would 5746be something like: 5747 5748@smallexample 5749int foo (void) 5750@{ 5751 register int *xr = &x; 5752 return xr[&a - &x] + xr[&b - &x] + xr[&c - &x]; 5753@} 5754@end smallexample 5755 5756(which isn't valid C). We refer to shared addresses like @code{x} as 5757``section anchors''. Their use is controlled by @option{-fsection-anchors}. 5758 5759The hooks below describe the target properties that GCC needs to know 5760in order to make effective use of section anchors. It won't use 5761section anchors at all unless either @code{TARGET_MIN_ANCHOR_OFFSET} 5762or @code{TARGET_MAX_ANCHOR_OFFSET} is set to a nonzero value. 5763 5764@hook TARGET_MIN_ANCHOR_OFFSET 5765The minimum offset that should be applied to a section anchor. 5766On most targets, it should be the smallest offset that can be 5767applied to a base register while still giving a legitimate address 5768for every mode. The default value is 0. 5769@end deftypevr 5770 5771@hook TARGET_MAX_ANCHOR_OFFSET 5772Like @code{TARGET_MIN_ANCHOR_OFFSET}, but the maximum (inclusive) 5773offset that should be applied to section anchors. The default 5774value is 0. 5775@end deftypevr 5776 5777@hook TARGET_ASM_OUTPUT_ANCHOR 5778Write the assembly code to define section anchor @var{x}, which is a 5779@code{SYMBOL_REF} for which @samp{SYMBOL_REF_ANCHOR_P (@var{x})} is true. 5780The hook is called with the assembly output position set to the beginning 5781of @code{SYMBOL_REF_BLOCK (@var{x})}. 5782 5783If @code{ASM_OUTPUT_DEF} is available, the hook's default definition uses 5784it to define the symbol as @samp{. + SYMBOL_REF_BLOCK_OFFSET (@var{x})}. 5785If @code{ASM_OUTPUT_DEF} is not available, the hook's default definition 5786is @code{NULL}, which disables the use of section anchors altogether. 5787@end deftypefn 5788 5789@hook TARGET_USE_ANCHORS_FOR_SYMBOL_P 5790Return true if GCC should attempt to use anchors to access @code{SYMBOL_REF} 5791@var{x}. You can assume @samp{SYMBOL_REF_HAS_BLOCK_INFO_P (@var{x})} and 5792@samp{!SYMBOL_REF_ANCHOR_P (@var{x})}. 5793 5794The default version is correct for most targets, but you might need to 5795intercept this hook to handle things like target-specific attributes 5796or target-specific sections. 5797@end deftypefn 5798 5799@node Condition Code 5800@section Condition Code Status 5801@cindex condition code status 5802 5803The macros in this section can be split in two families, according to the 5804two ways of representing condition codes in GCC. 5805 5806The first representation is the so called @code{(cc0)} representation 5807(@pxref{Jump Patterns}), where all instructions can have an implicit 5808clobber of the condition codes. The second is the condition code 5809register representation, which provides better schedulability for 5810architectures that do have a condition code register, but on which 5811most instructions do not affect it. The latter category includes 5812most RISC machines. 5813 5814The implicit clobbering poses a strong restriction on the placement of 5815the definition and use of the condition code, which need to be in adjacent 5816insns for machines using @code{(cc0)}. This can prevent important 5817optimizations on some machines. For example, on the IBM RS/6000, there 5818is a delay for taken branches unless the condition code register is set 5819three instructions earlier than the conditional branch. The instruction 5820scheduler cannot perform this optimization if it is not permitted to 5821separate the definition and use of the condition code register. 5822 5823For this reason, it is possible and suggested to use a register to 5824represent the condition code for new ports. If there is a specific 5825condition code register in the machine, use a hard register. If the 5826condition code or comparison result can be placed in any general register, 5827or if there are multiple condition registers, use a pseudo register. 5828Registers used to store the condition code value will usually have a mode 5829that is in class @code{MODE_CC}. 5830 5831Alternatively, you can use @code{BImode} if the comparison operator is 5832specified already in the compare instruction. In this case, you are not 5833interested in most macros in this section. 5834 5835@menu 5836* CC0 Condition Codes:: Old style representation of condition codes. 5837* MODE_CC Condition Codes:: Modern representation of condition codes. 5838* Cond Exec Macros:: Macros to control conditional execution. 5839@end menu 5840 5841@node CC0 Condition Codes 5842@subsection Representation of condition codes using @code{(cc0)} 5843@findex cc0 5844 5845@findex cc_status 5846The file @file{conditions.h} defines a variable @code{cc_status} to 5847describe how the condition code was computed (in case the interpretation of 5848the condition code depends on the instruction that it was set by). This 5849variable contains the RTL expressions on which the condition code is 5850currently based, and several standard flags. 5851 5852Sometimes additional machine-specific flags must be defined in the machine 5853description header file. It can also add additional machine-specific 5854information by defining @code{CC_STATUS_MDEP}. 5855 5856@defmac CC_STATUS_MDEP 5857C code for a data type which is used for declaring the @code{mdep} 5858component of @code{cc_status}. It defaults to @code{int}. 5859 5860This macro is not used on machines that do not use @code{cc0}. 5861@end defmac 5862 5863@defmac CC_STATUS_MDEP_INIT 5864A C expression to initialize the @code{mdep} field to ``empty''. 5865The default definition does nothing, since most machines don't use 5866the field anyway. If you want to use the field, you should probably 5867define this macro to initialize it. 5868 5869This macro is not used on machines that do not use @code{cc0}. 5870@end defmac 5871 5872@defmac NOTICE_UPDATE_CC (@var{exp}, @var{insn}) 5873A C compound statement to set the components of @code{cc_status} 5874appropriately for an insn @var{insn} whose body is @var{exp}. It is 5875this macro's responsibility to recognize insns that set the condition 5876code as a byproduct of other activity as well as those that explicitly 5877set @code{(cc0)}. 5878 5879This macro is not used on machines that do not use @code{cc0}. 5880 5881If there are insns that do not set the condition code but do alter 5882other machine registers, this macro must check to see whether they 5883invalidate the expressions that the condition code is recorded as 5884reflecting. For example, on the 68000, insns that store in address 5885registers do not set the condition code, which means that usually 5886@code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such 5887insns. But suppose that the previous insn set the condition code 5888based on location @samp{a4@@(102)} and the current insn stores a new 5889value in @samp{a4}. Although the condition code is not changed by 5890this, it will no longer be true that it reflects the contents of 5891@samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter 5892@code{cc_status} in this case to say that nothing is known about the 5893condition code value. 5894 5895The definition of @code{NOTICE_UPDATE_CC} must be prepared to deal 5896with the results of peephole optimization: insns whose patterns are 5897@code{parallel} RTXs containing various @code{reg}, @code{mem} or 5898constants which are just the operands. The RTL structure of these 5899insns is not sufficient to indicate what the insns actually do. What 5900@code{NOTICE_UPDATE_CC} should do when it sees one is just to run 5901@code{CC_STATUS_INIT}. 5902 5903A possible definition of @code{NOTICE_UPDATE_CC} is to call a function 5904that looks at an attribute (@pxref{Insn Attributes}) named, for example, 5905@samp{cc}. This avoids having detailed information about patterns in 5906two places, the @file{md} file and in @code{NOTICE_UPDATE_CC}. 5907@end defmac 5908 5909@node MODE_CC Condition Codes 5910@subsection Representation of condition codes using registers 5911@findex CCmode 5912@findex MODE_CC 5913 5914@defmac SELECT_CC_MODE (@var{op}, @var{x}, @var{y}) 5915On many machines, the condition code may be produced by other instructions 5916than compares, for example the branch can use directly the condition 5917code set by a subtract instruction. However, on some machines 5918when the condition code is set this way some bits (such as the overflow 5919bit) are not set in the same way as a test instruction, so that a different 5920branch instruction must be used for some conditional branches. When 5921this happens, use the machine mode of the condition code register to 5922record different formats of the condition code register. Modes can 5923also be used to record which compare instruction (e.g. a signed or an 5924unsigned comparison) produced the condition codes. 5925 5926If other modes than @code{CCmode} are required, add them to 5927@file{@var{machine}-modes.def} and define @code{SELECT_CC_MODE} to choose 5928a mode given an operand of a compare. This is needed because the modes 5929have to be chosen not only during RTL generation but also, for example, 5930by instruction combination. The result of @code{SELECT_CC_MODE} should 5931be consistent with the mode used in the patterns; for example to support 5932the case of the add on the SPARC discussed above, we have the pattern 5933 5934@smallexample 5935(define_insn "" 5936 [(set (reg:CC_NOOV 0) 5937 (compare:CC_NOOV 5938 (plus:SI (match_operand:SI 0 "register_operand" "%r") 5939 (match_operand:SI 1 "arith_operand" "rI")) 5940 (const_int 0)))] 5941 "" 5942 "@dots{}") 5943@end smallexample 5944 5945@noindent 5946together with a @code{SELECT_CC_MODE} that returns @code{CC_NOOVmode} 5947for comparisons whose argument is a @code{plus}: 5948 5949@smallexample 5950#define SELECT_CC_MODE(OP,X,Y) \ 5951 (GET_MODE_CLASS (GET_MODE (X)) == MODE_FLOAT \ 5952 ? ((OP == EQ || OP == NE) ? CCFPmode : CCFPEmode) \ 5953 : ((GET_CODE (X) == PLUS || GET_CODE (X) == MINUS \ 5954 || GET_CODE (X) == NEG) \ 5955 ? CC_NOOVmode : CCmode)) 5956@end smallexample 5957 5958Another reason to use modes is to retain information on which operands 5959were used by the comparison; see @code{REVERSIBLE_CC_MODE} later in 5960this section. 5961 5962You should define this macro if and only if you define extra CC modes 5963in @file{@var{machine}-modes.def}. 5964@end defmac 5965 5966@defmac CANONICALIZE_COMPARISON (@var{code}, @var{op0}, @var{op1}) 5967On some machines not all possible comparisons are defined, but you can 5968convert an invalid comparison into a valid one. For example, the Alpha 5969does not have a @code{GT} comparison, but you can use an @code{LT} 5970comparison instead and swap the order of the operands. 5971 5972On such machines, define this macro to be a C statement to do any 5973required conversions. @var{code} is the initial comparison code 5974and @var{op0} and @var{op1} are the left and right operands of the 5975comparison, respectively. You should modify @var{code}, @var{op0}, and 5976@var{op1} as required. 5977 5978GCC will not assume that the comparison resulting from this macro is 5979valid but will see if the resulting insn matches a pattern in the 5980@file{md} file. 5981 5982You need not define this macro if it would never change the comparison 5983code or operands. 5984@end defmac 5985 5986@defmac REVERSIBLE_CC_MODE (@var{mode}) 5987A C expression whose value is one if it is always safe to reverse a 5988comparison whose mode is @var{mode}. If @code{SELECT_CC_MODE} 5989can ever return @var{mode} for a floating-point inequality comparison, 5990then @code{REVERSIBLE_CC_MODE (@var{mode})} must be zero. 5991 5992You need not define this macro if it would always returns zero or if the 5993floating-point format is anything other than @code{IEEE_FLOAT_FORMAT}. 5994For example, here is the definition used on the SPARC, where floating-point 5995inequality comparisons are always given @code{CCFPEmode}: 5996 5997@smallexample 5998#define REVERSIBLE_CC_MODE(MODE) ((MODE) != CCFPEmode) 5999@end smallexample 6000@end defmac 6001 6002@defmac REVERSE_CONDITION (@var{code}, @var{mode}) 6003A C expression whose value is reversed condition code of the @var{code} for 6004comparison done in CC_MODE @var{mode}. The macro is used only in case 6005@code{REVERSIBLE_CC_MODE (@var{mode})} is nonzero. Define this macro in case 6006machine has some non-standard way how to reverse certain conditionals. For 6007instance in case all floating point conditions are non-trapping, compiler may 6008freely convert unordered compares to ordered one. Then definition may look 6009like: 6010 6011@smallexample 6012#define REVERSE_CONDITION(CODE, MODE) \ 6013 ((MODE) != CCFPmode ? reverse_condition (CODE) \ 6014 : reverse_condition_maybe_unordered (CODE)) 6015@end smallexample 6016@end defmac 6017 6018@hook TARGET_FIXED_CONDITION_CODE_REGS 6019On targets which do not use @code{(cc0)}, and which use a hard 6020register rather than a pseudo-register to hold condition codes, the 6021regular CSE passes are often not able to identify cases in which the 6022hard register is set to a common value. Use this hook to enable a 6023small pass which optimizes such cases. This hook should return true 6024to enable this pass, and it should set the integers to which its 6025arguments point to the hard register numbers used for condition codes. 6026When there is only one such register, as is true on most systems, the 6027integer pointed to by @var{p2} should be set to 6028@code{INVALID_REGNUM}. 6029 6030The default version of this hook returns false. 6031@end deftypefn 6032 6033@hook TARGET_CC_MODES_COMPATIBLE 6034On targets which use multiple condition code modes in class 6035@code{MODE_CC}, it is sometimes the case that a comparison can be 6036validly done in more than one mode. On such a system, define this 6037target hook to take two mode arguments and to return a mode in which 6038both comparisons may be validly done. If there is no such mode, 6039return @code{VOIDmode}. 6040 6041The default version of this hook checks whether the modes are the 6042same. If they are, it returns that mode. If they are different, it 6043returns @code{VOIDmode}. 6044@end deftypefn 6045 6046@node Cond Exec Macros 6047@subsection Macros to control conditional execution 6048@findex conditional execution 6049@findex predication 6050 6051There is one macro that may need to be defined for targets 6052supporting conditional execution, independent of how they 6053represent conditional branches. 6054 6055@defmac REVERSE_CONDEXEC_PREDICATES_P (@var{op1}, @var{op2}) 6056A C expression that returns true if the conditional execution predicate 6057@var{op1}, a comparison operation, is the inverse of @var{op2} and vice 6058versa. Define this to return 0 if the target has conditional execution 6059predicates that cannot be reversed safely. There is no need to validate 6060that the arguments of op1 and op2 are the same, this is done separately. 6061If no expansion is specified, this macro is defined as follows: 6062 6063@smallexample 6064#define REVERSE_CONDEXEC_PREDICATES_P (x, y) \ 6065 (GET_CODE ((x)) == reversed_comparison_code ((y), NULL)) 6066@end smallexample 6067@end defmac 6068 6069@node Costs 6070@section Describing Relative Costs of Operations 6071@cindex costs of instructions 6072@cindex relative costs 6073@cindex speed of instructions 6074 6075These macros let you describe the relative speed of various operations 6076on the target machine. 6077 6078@defmac REGISTER_MOVE_COST (@var{mode}, @var{from}, @var{to}) 6079A C expression for the cost of moving data of mode @var{mode} from a 6080register in class @var{from} to one in class @var{to}. The classes are 6081expressed using the enumeration values such as @code{GENERAL_REGS}. A 6082value of 2 is the default; other values are interpreted relative to 6083that. 6084 6085It is not required that the cost always equal 2 when @var{from} is the 6086same as @var{to}; on some machines it is expensive to move between 6087registers if they are not general registers. 6088 6089If reload sees an insn consisting of a single @code{set} between two 6090hard registers, and if @code{REGISTER_MOVE_COST} applied to their 6091classes returns a value of 2, reload does not check to ensure that the 6092constraints of the insn are met. Setting a cost of other than 2 will 6093allow reload to verify that the constraints are met. You should do this 6094if the @samp{mov@var{m}} pattern's constraints do not allow such copying. 6095 6096These macros are obsolete, new ports should use the target hook 6097@code{TARGET_REGISTER_MOVE_COST} instead. 6098@end defmac 6099 6100@hook TARGET_REGISTER_MOVE_COST 6101This target hook should return the cost of moving data of mode @var{mode} 6102from a register in class @var{from} to one in class @var{to}. The classes 6103are expressed using the enumeration values such as @code{GENERAL_REGS}. 6104A value of 2 is the default; other values are interpreted relative to 6105that. 6106 6107It is not required that the cost always equal 2 when @var{from} is the 6108same as @var{to}; on some machines it is expensive to move between 6109registers if they are not general registers. 6110 6111If reload sees an insn consisting of a single @code{set} between two 6112hard registers, and if @code{TARGET_REGISTER_MOVE_COST} applied to their 6113classes returns a value of 2, reload does not check to ensure that the 6114constraints of the insn are met. Setting a cost of other than 2 will 6115allow reload to verify that the constraints are met. You should do this 6116if the @samp{mov@var{m}} pattern's constraints do not allow such copying. 6117 6118The default version of this function returns 2. 6119@end deftypefn 6120 6121@defmac MEMORY_MOVE_COST (@var{mode}, @var{class}, @var{in}) 6122A C expression for the cost of moving data of mode @var{mode} between a 6123register of class @var{class} and memory; @var{in} is zero if the value 6124is to be written to memory, nonzero if it is to be read in. This cost 6125is relative to those in @code{REGISTER_MOVE_COST}. If moving between 6126registers and memory is more expensive than between two registers, you 6127should define this macro to express the relative cost. 6128 6129If you do not define this macro, GCC uses a default cost of 4 plus 6130the cost of copying via a secondary reload register, if one is 6131needed. If your machine requires a secondary reload register to copy 6132between memory and a register of @var{class} but the reload mechanism is 6133more complex than copying via an intermediate, define this macro to 6134reflect the actual cost of the move. 6135 6136GCC defines the function @code{memory_move_secondary_cost} if 6137secondary reloads are needed. It computes the costs due to copying via 6138a secondary register. If your machine copies from memory using a 6139secondary register in the conventional way but the default base value of 61404 is not correct for your machine, define this macro to add some other 6141value to the result of that function. The arguments to that function 6142are the same as to this macro. 6143 6144These macros are obsolete, new ports should use the target hook 6145@code{TARGET_MEMORY_MOVE_COST} instead. 6146@end defmac 6147 6148@hook TARGET_MEMORY_MOVE_COST 6149This target hook should return the cost of moving data of mode @var{mode} 6150between a register of class @var{rclass} and memory; @var{in} is @code{false} 6151if the value is to be written to memory, @code{true} if it is to be read in. 6152This cost is relative to those in @code{TARGET_REGISTER_MOVE_COST}. 6153If moving between registers and memory is more expensive than between two 6154registers, you should add this target hook to express the relative cost. 6155 6156If you do not add this target hook, GCC uses a default cost of 4 plus 6157the cost of copying via a secondary reload register, if one is 6158needed. If your machine requires a secondary reload register to copy 6159between memory and a register of @var{rclass} but the reload mechanism is 6160more complex than copying via an intermediate, use this target hook to 6161reflect the actual cost of the move. 6162 6163GCC defines the function @code{memory_move_secondary_cost} if 6164secondary reloads are needed. It computes the costs due to copying via 6165a secondary register. If your machine copies from memory using a 6166secondary register in the conventional way but the default base value of 61674 is not correct for your machine, use this target hook to add some other 6168value to the result of that function. The arguments to that function 6169are the same as to this target hook. 6170@end deftypefn 6171 6172@defmac BRANCH_COST (@var{speed_p}, @var{predictable_p}) 6173A C expression for the cost of a branch instruction. A value of 1 is 6174the default; other values are interpreted relative to that. Parameter 6175@var{speed_p} is true when the branch in question should be optimized 6176for speed. When it is false, @code{BRANCH_COST} should return a value 6177optimal for code size rather than performance. @var{predictable_p} is 6178true for well-predicted branches. On many architectures the 6179@code{BRANCH_COST} can be reduced then. 6180@end defmac 6181 6182Here are additional macros which do not specify precise relative costs, 6183but only that certain actions are more expensive than GCC would 6184ordinarily expect. 6185 6186@defmac SLOW_BYTE_ACCESS 6187Define this macro as a C expression which is nonzero if accessing less 6188than a word of memory (i.e.@: a @code{char} or a @code{short}) is no 6189faster than accessing a word of memory, i.e., if such access 6190require more than one instruction or if there is no difference in cost 6191between byte and (aligned) word loads. 6192 6193When this macro is not defined, the compiler will access a field by 6194finding the smallest containing object; when it is defined, a fullword 6195load will be used if alignment permits. Unless bytes accesses are 6196faster than word accesses, using word accesses is preferable since it 6197may eliminate subsequent memory access if subsequent accesses occur to 6198other fields in the same word of the structure, but to different bytes. 6199@end defmac 6200 6201@defmac SLOW_UNALIGNED_ACCESS (@var{mode}, @var{alignment}) 6202Define this macro to be the value 1 if memory accesses described by the 6203@var{mode} and @var{alignment} parameters have a cost many times greater 6204than aligned accesses, for example if they are emulated in a trap 6205handler. 6206 6207When this macro is nonzero, the compiler will act as if 6208@code{STRICT_ALIGNMENT} were nonzero when generating code for block 6209moves. This can cause significantly more instructions to be produced. 6210Therefore, do not set this macro nonzero if unaligned accesses only add a 6211cycle or two to the time for a memory access. 6212 6213If the value of this macro is always zero, it need not be defined. If 6214this macro is defined, it should produce a nonzero value when 6215@code{STRICT_ALIGNMENT} is nonzero. 6216@end defmac 6217 6218@defmac MOVE_RATIO (@var{speed}) 6219The threshold of number of scalar memory-to-memory move insns, @emph{below} 6220which a sequence of insns should be generated instead of a 6221string move insn or a library call. Increasing the value will always 6222make code faster, but eventually incurs high cost in increased code size. 6223 6224Note that on machines where the corresponding move insn is a 6225@code{define_expand} that emits a sequence of insns, this macro counts 6226the number of such sequences. 6227 6228The parameter @var{speed} is true if the code is currently being 6229optimized for speed rather than size. 6230 6231If you don't define this, a reasonable default is used. 6232@end defmac 6233 6234@defmac MOVE_BY_PIECES_P (@var{size}, @var{alignment}) 6235A C expression used to determine whether @code{move_by_pieces} will be used to 6236copy a chunk of memory, or whether some other block move mechanism 6237will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less 6238than @code{MOVE_RATIO}. 6239@end defmac 6240 6241@defmac MOVE_MAX_PIECES 6242A C expression used by @code{move_by_pieces} to determine the largest unit 6243a load or store used to copy memory is. Defaults to @code{MOVE_MAX}. 6244@end defmac 6245 6246@defmac CLEAR_RATIO (@var{speed}) 6247The threshold of number of scalar move insns, @emph{below} which a sequence 6248of insns should be generated to clear memory instead of a string clear insn 6249or a library call. Increasing the value will always make code faster, but 6250eventually incurs high cost in increased code size. 6251 6252The parameter @var{speed} is true if the code is currently being 6253optimized for speed rather than size. 6254 6255If you don't define this, a reasonable default is used. 6256@end defmac 6257 6258@defmac CLEAR_BY_PIECES_P (@var{size}, @var{alignment}) 6259A C expression used to determine whether @code{clear_by_pieces} will be used 6260to clear a chunk of memory, or whether some other block clear mechanism 6261will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less 6262than @code{CLEAR_RATIO}. 6263@end defmac 6264 6265@defmac SET_RATIO (@var{speed}) 6266The threshold of number of scalar move insns, @emph{below} which a sequence 6267of insns should be generated to set memory to a constant value, instead of 6268a block set insn or a library call. 6269Increasing the value will always make code faster, but 6270eventually incurs high cost in increased code size. 6271 6272The parameter @var{speed} is true if the code is currently being 6273optimized for speed rather than size. 6274 6275If you don't define this, it defaults to the value of @code{MOVE_RATIO}. 6276@end defmac 6277 6278@defmac SET_BY_PIECES_P (@var{size}, @var{alignment}) 6279A C expression used to determine whether @code{store_by_pieces} will be 6280used to set a chunk of memory to a constant value, or whether some 6281other mechanism will be used. Used by @code{__builtin_memset} when 6282storing values other than constant zero. 6283Defaults to 1 if @code{move_by_pieces_ninsns} returns less 6284than @code{SET_RATIO}. 6285@end defmac 6286 6287@defmac STORE_BY_PIECES_P (@var{size}, @var{alignment}) 6288A C expression used to determine whether @code{store_by_pieces} will be 6289used to set a chunk of memory to a constant string value, or whether some 6290other mechanism will be used. Used by @code{__builtin_strcpy} when 6291called with a constant source string. 6292Defaults to 1 if @code{move_by_pieces_ninsns} returns less 6293than @code{MOVE_RATIO}. 6294@end defmac 6295 6296@defmac USE_LOAD_POST_INCREMENT (@var{mode}) 6297A C expression used to determine whether a load postincrement is a good 6298thing to use for a given mode. Defaults to the value of 6299@code{HAVE_POST_INCREMENT}. 6300@end defmac 6301 6302@defmac USE_LOAD_POST_DECREMENT (@var{mode}) 6303A C expression used to determine whether a load postdecrement is a good 6304thing to use for a given mode. Defaults to the value of 6305@code{HAVE_POST_DECREMENT}. 6306@end defmac 6307 6308@defmac USE_LOAD_PRE_INCREMENT (@var{mode}) 6309A C expression used to determine whether a load preincrement is a good 6310thing to use for a given mode. Defaults to the value of 6311@code{HAVE_PRE_INCREMENT}. 6312@end defmac 6313 6314@defmac USE_LOAD_PRE_DECREMENT (@var{mode}) 6315A C expression used to determine whether a load predecrement is a good 6316thing to use for a given mode. Defaults to the value of 6317@code{HAVE_PRE_DECREMENT}. 6318@end defmac 6319 6320@defmac USE_STORE_POST_INCREMENT (@var{mode}) 6321A C expression used to determine whether a store postincrement is a good 6322thing to use for a given mode. Defaults to the value of 6323@code{HAVE_POST_INCREMENT}. 6324@end defmac 6325 6326@defmac USE_STORE_POST_DECREMENT (@var{mode}) 6327A C expression used to determine whether a store postdecrement is a good 6328thing to use for a given mode. Defaults to the value of 6329@code{HAVE_POST_DECREMENT}. 6330@end defmac 6331 6332@defmac USE_STORE_PRE_INCREMENT (@var{mode}) 6333This macro is used to determine whether a store preincrement is a good 6334thing to use for a given mode. Defaults to the value of 6335@code{HAVE_PRE_INCREMENT}. 6336@end defmac 6337 6338@defmac USE_STORE_PRE_DECREMENT (@var{mode}) 6339This macro is used to determine whether a store predecrement is a good 6340thing to use for a given mode. Defaults to the value of 6341@code{HAVE_PRE_DECREMENT}. 6342@end defmac 6343 6344@defmac NO_FUNCTION_CSE 6345Define this macro if it is as good or better to call a constant 6346function address than to call an address kept in a register. 6347@end defmac 6348 6349@defmac RANGE_TEST_NON_SHORT_CIRCUIT 6350Define this macro if a non-short-circuit operation produced by 6351@samp{fold_range_test ()} is optimal. This macro defaults to true if 6352@code{BRANCH_COST} is greater than or equal to the value 2. 6353@end defmac 6354 6355@hook TARGET_RTX_COSTS 6356This target hook describes the relative costs of RTL expressions. 6357 6358The cost may depend on the precise form of the expression, which is 6359available for examination in @var{x}, and the fact that @var{x} appears 6360as operand @var{opno} of an expression with rtx code @var{outer_code}. 6361That is, the hook can assume that there is some rtx @var{y} such 6362that @samp{GET_CODE (@var{y}) == @var{outer_code}} and such that 6363either (a) @samp{XEXP (@var{y}, @var{opno}) == @var{x}} or 6364(b) @samp{XVEC (@var{y}, @var{opno})} contains @var{x}. 6365 6366@var{code} is @var{x}'s expression code---redundant, since it can be 6367obtained with @code{GET_CODE (@var{x})}. 6368 6369In implementing this hook, you can use the construct 6370@code{COSTS_N_INSNS (@var{n})} to specify a cost equal to @var{n} fast 6371instructions. 6372 6373On entry to the hook, @code{*@var{total}} contains a default estimate 6374for the cost of the expression. The hook should modify this value as 6375necessary. Traditionally, the default costs are @code{COSTS_N_INSNS (5)} 6376for multiplications, @code{COSTS_N_INSNS (7)} for division and modulus 6377operations, and @code{COSTS_N_INSNS (1)} for all other operations. 6378 6379When optimizing for code size, i.e.@: when @code{speed} is 6380false, this target hook should be used to estimate the relative 6381size cost of an expression, again relative to @code{COSTS_N_INSNS}. 6382 6383The hook returns true when all subexpressions of @var{x} have been 6384processed, and false when @code{rtx_cost} should recurse. 6385@end deftypefn 6386 6387@hook TARGET_ADDRESS_COST 6388This hook computes the cost of an addressing mode that contains 6389@var{address}. If not defined, the cost is computed from 6390the @var{address} expression and the @code{TARGET_RTX_COST} hook. 6391 6392For most CISC machines, the default cost is a good approximation of the 6393true cost of the addressing mode. However, on RISC machines, all 6394instructions normally have the same length and execution time. Hence 6395all addresses will have equal costs. 6396 6397In cases where more than one form of an address is known, the form with 6398the lowest cost will be used. If multiple forms have the same, lowest, 6399cost, the one that is the most complex will be used. 6400 6401For example, suppose an address that is equal to the sum of a register 6402and a constant is used twice in the same basic block. When this macro 6403is not defined, the address will be computed in a register and memory 6404references will be indirect through that register. On machines where 6405the cost of the addressing mode containing the sum is no higher than 6406that of a simple indirect reference, this will produce an additional 6407instruction and possibly require an additional register. Proper 6408specification of this macro eliminates this overhead for such machines. 6409 6410This hook is never called with an invalid address. 6411 6412On machines where an address involving more than one register is as 6413cheap as an address computation involving only one register, defining 6414@code{TARGET_ADDRESS_COST} to reflect this can cause two registers to 6415be live over a region of code where only one would have been if 6416@code{TARGET_ADDRESS_COST} were not defined in that manner. This effect 6417should be considered in the definition of this macro. Equivalent costs 6418should probably only be given to addresses with different numbers of 6419registers on machines with lots of registers. 6420@end deftypefn 6421 6422@node Scheduling 6423@section Adjusting the Instruction Scheduler 6424 6425The instruction scheduler may need a fair amount of machine-specific 6426adjustment in order to produce good code. GCC provides several target 6427hooks for this purpose. It is usually enough to define just a few of 6428them: try the first ones in this list first. 6429 6430@hook TARGET_SCHED_ISSUE_RATE 6431This hook returns the maximum number of instructions that can ever 6432issue at the same time on the target machine. The default is one. 6433Although the insn scheduler can define itself the possibility of issue 6434an insn on the same cycle, the value can serve as an additional 6435constraint to issue insns on the same simulated processor cycle (see 6436hooks @samp{TARGET_SCHED_REORDER} and @samp{TARGET_SCHED_REORDER2}). 6437This value must be constant over the entire compilation. If you need 6438it to vary depending on what the instructions are, you must use 6439@samp{TARGET_SCHED_VARIABLE_ISSUE}. 6440@end deftypefn 6441 6442@hook TARGET_SCHED_VARIABLE_ISSUE 6443This hook is executed by the scheduler after it has scheduled an insn 6444from the ready list. It should return the number of insns which can 6445still be issued in the current cycle. The default is 6446@samp{@w{@var{more} - 1}} for insns other than @code{CLOBBER} and 6447@code{USE}, which normally are not counted against the issue rate. 6448You should define this hook if some insns take more machine resources 6449than others, so that fewer insns can follow them in the same cycle. 6450@var{file} is either a null pointer, or a stdio stream to write any 6451debug output to. @var{verbose} is the verbose level provided by 6452@option{-fsched-verbose-@var{n}}. @var{insn} is the instruction that 6453was scheduled. 6454@end deftypefn 6455 6456@hook TARGET_SCHED_ADJUST_COST 6457This function corrects the value of @var{cost} based on the 6458relationship between @var{insn} and @var{dep_insn} through the 6459dependence @var{link}. It should return the new value. The default 6460is to make no adjustment to @var{cost}. This can be used for example 6461to specify to the scheduler using the traditional pipeline description 6462that an output- or anti-dependence does not incur the same cost as a 6463data-dependence. If the scheduler using the automaton based pipeline 6464description, the cost of anti-dependence is zero and the cost of 6465output-dependence is maximum of one and the difference of latency 6466times of the first and the second insns. If these values are not 6467acceptable, you could use the hook to modify them too. See also 6468@pxref{Processor pipeline description}. 6469@end deftypefn 6470 6471@hook TARGET_SCHED_ADJUST_PRIORITY 6472This hook adjusts the integer scheduling priority @var{priority} of 6473@var{insn}. It should return the new priority. Increase the priority to 6474execute @var{insn} earlier, reduce the priority to execute @var{insn} 6475later. Do not define this hook if you do not need to adjust the 6476scheduling priorities of insns. 6477@end deftypefn 6478 6479@hook TARGET_SCHED_REORDER 6480This hook is executed by the scheduler after it has scheduled the ready 6481list, to allow the machine description to reorder it (for example to 6482combine two small instructions together on @samp{VLIW} machines). 6483@var{file} is either a null pointer, or a stdio stream to write any 6484debug output to. @var{verbose} is the verbose level provided by 6485@option{-fsched-verbose-@var{n}}. @var{ready} is a pointer to the ready 6486list of instructions that are ready to be scheduled. @var{n_readyp} is 6487a pointer to the number of elements in the ready list. The scheduler 6488reads the ready list in reverse order, starting with 6489@var{ready}[@var{*n_readyp} @minus{} 1] and going to @var{ready}[0]. @var{clock} 6490is the timer tick of the scheduler. You may modify the ready list and 6491the number of ready insns. The return value is the number of insns that 6492can issue this cycle; normally this is just @code{issue_rate}. See also 6493@samp{TARGET_SCHED_REORDER2}. 6494@end deftypefn 6495 6496@hook TARGET_SCHED_REORDER2 6497Like @samp{TARGET_SCHED_REORDER}, but called at a different time. That 6498function is called whenever the scheduler starts a new cycle. This one 6499is called once per iteration over a cycle, immediately after 6500@samp{TARGET_SCHED_VARIABLE_ISSUE}; it can reorder the ready list and 6501return the number of insns to be scheduled in the same cycle. Defining 6502this hook can be useful if there are frequent situations where 6503scheduling one insn causes other insns to become ready in the same 6504cycle. These other insns can then be taken into account properly. 6505@end deftypefn 6506 6507@hook TARGET_SCHED_DEPENDENCIES_EVALUATION_HOOK 6508This hook is called after evaluation forward dependencies of insns in 6509chain given by two parameter values (@var{head} and @var{tail} 6510correspondingly) but before insns scheduling of the insn chain. For 6511example, it can be used for better insn classification if it requires 6512analysis of dependencies. This hook can use backward and forward 6513dependencies of the insn scheduler because they are already 6514calculated. 6515@end deftypefn 6516 6517@hook TARGET_SCHED_INIT 6518This hook is executed by the scheduler at the beginning of each block of 6519instructions that are to be scheduled. @var{file} is either a null 6520pointer, or a stdio stream to write any debug output to. @var{verbose} 6521is the verbose level provided by @option{-fsched-verbose-@var{n}}. 6522@var{max_ready} is the maximum number of insns in the current scheduling 6523region that can be live at the same time. This can be used to allocate 6524scratch space if it is needed, e.g.@: by @samp{TARGET_SCHED_REORDER}. 6525@end deftypefn 6526 6527@hook TARGET_SCHED_FINISH 6528This hook is executed by the scheduler at the end of each block of 6529instructions that are to be scheduled. It can be used to perform 6530cleanup of any actions done by the other scheduling hooks. @var{file} 6531is either a null pointer, or a stdio stream to write any debug output 6532to. @var{verbose} is the verbose level provided by 6533@option{-fsched-verbose-@var{n}}. 6534@end deftypefn 6535 6536@hook TARGET_SCHED_INIT_GLOBAL 6537This hook is executed by the scheduler after function level initializations. 6538@var{file} is either a null pointer, or a stdio stream to write any debug output to. 6539@var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}. 6540@var{old_max_uid} is the maximum insn uid when scheduling begins. 6541@end deftypefn 6542 6543@hook TARGET_SCHED_FINISH_GLOBAL 6544This is the cleanup hook corresponding to @code{TARGET_SCHED_INIT_GLOBAL}. 6545@var{file} is either a null pointer, or a stdio stream to write any debug output to. 6546@var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}. 6547@end deftypefn 6548 6549@hook TARGET_SCHED_DFA_PRE_CYCLE_INSN 6550The hook returns an RTL insn. The automaton state used in the 6551pipeline hazard recognizer is changed as if the insn were scheduled 6552when the new simulated processor cycle starts. Usage of the hook may 6553simplify the automaton pipeline description for some @acronym{VLIW} 6554processors. If the hook is defined, it is used only for the automaton 6555based pipeline description. The default is not to change the state 6556when the new simulated processor cycle starts. 6557@end deftypefn 6558 6559@hook TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN 6560The hook can be used to initialize data used by the previous hook. 6561@end deftypefn 6562 6563@hook TARGET_SCHED_DFA_POST_CYCLE_INSN 6564The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used 6565to changed the state as if the insn were scheduled when the new 6566simulated processor cycle finishes. 6567@end deftypefn 6568 6569@hook TARGET_SCHED_INIT_DFA_POST_CYCLE_INSN 6570The hook is analogous to @samp{TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN} but 6571used to initialize data used by the previous hook. 6572@end deftypefn 6573 6574@hook TARGET_SCHED_DFA_PRE_ADVANCE_CYCLE 6575The hook to notify target that the current simulated cycle is about to finish. 6576The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used 6577to change the state in more complicated situations - e.g., when advancing 6578state on a single insn is not enough. 6579@end deftypefn 6580 6581@hook TARGET_SCHED_DFA_POST_ADVANCE_CYCLE 6582The hook to notify target that new simulated cycle has just started. 6583The hook is analogous to @samp{TARGET_SCHED_DFA_POST_CYCLE_INSN} but used 6584to change the state in more complicated situations - e.g., when advancing 6585state on a single insn is not enough. 6586@end deftypefn 6587 6588@hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD 6589This hook controls better choosing an insn from the ready insn queue 6590for the @acronym{DFA}-based insn scheduler. Usually the scheduler 6591chooses the first insn from the queue. If the hook returns a positive 6592value, an additional scheduler code tries all permutations of 6593@samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD ()} 6594subsequent ready insns to choose an insn whose issue will result in 6595maximal number of issued insns on the same cycle. For the 6596@acronym{VLIW} processor, the code could actually solve the problem of 6597packing simple insns into the @acronym{VLIW} insn. Of course, if the 6598rules of @acronym{VLIW} packing are described in the automaton. 6599 6600This code also could be used for superscalar @acronym{RISC} 6601processors. Let us consider a superscalar @acronym{RISC} processor 6602with 3 pipelines. Some insns can be executed in pipelines @var{A} or 6603@var{B}, some insns can be executed only in pipelines @var{B} or 6604@var{C}, and one insn can be executed in pipeline @var{B}. The 6605processor may issue the 1st insn into @var{A} and the 2nd one into 6606@var{B}. In this case, the 3rd insn will wait for freeing @var{B} 6607until the next cycle. If the scheduler issues the 3rd insn the first, 6608the processor could issue all 3 insns per cycle. 6609 6610Actually this code demonstrates advantages of the automaton based 6611pipeline hazard recognizer. We try quickly and easy many insn 6612schedules to choose the best one. 6613 6614The default is no multipass scheduling. 6615@end deftypefn 6616 6617@hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD 6618 6619This hook controls what insns from the ready insn queue will be 6620considered for the multipass insn scheduling. If the hook returns 6621zero for @var{insn}, the insn will be not chosen to 6622be issued. 6623 6624The default is that any ready insns can be chosen to be issued. 6625@end deftypefn 6626 6627@hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_BEGIN 6628This hook prepares the target backend for a new round of multipass 6629scheduling. 6630@end deftypefn 6631 6632@hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_ISSUE 6633This hook is called when multipass scheduling evaluates instruction INSN. 6634@end deftypefn 6635 6636@hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_BACKTRACK 6637This is called when multipass scheduling backtracks from evaluation of 6638an instruction. 6639@end deftypefn 6640 6641@hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_END 6642This hook notifies the target about the result of the concluded current 6643round of multipass scheduling. 6644@end deftypefn 6645 6646@hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_INIT 6647This hook initializes target-specific data used in multipass scheduling. 6648@end deftypefn 6649 6650@hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_FINI 6651This hook finalizes target-specific data used in multipass scheduling. 6652@end deftypefn 6653 6654@hook TARGET_SCHED_DFA_NEW_CYCLE 6655This hook is called by the insn scheduler before issuing @var{insn} 6656on cycle @var{clock}. If the hook returns nonzero, 6657@var{insn} is not issued on this processor cycle. Instead, 6658the processor cycle is advanced. If *@var{sort_p} 6659is zero, the insn ready queue is not sorted on the new cycle 6660start as usually. @var{dump} and @var{verbose} specify the file and 6661verbosity level to use for debugging output. 6662@var{last_clock} and @var{clock} are, respectively, the 6663processor cycle on which the previous insn has been issued, 6664and the current processor cycle. 6665@end deftypefn 6666 6667@hook TARGET_SCHED_IS_COSTLY_DEPENDENCE 6668This hook is used to define which dependences are considered costly by 6669the target, so costly that it is not advisable to schedule the insns that 6670are involved in the dependence too close to one another. The parameters 6671to this hook are as follows: The first parameter @var{_dep} is the dependence 6672being evaluated. The second parameter @var{cost} is the cost of the 6673dependence as estimated by the scheduler, and the third 6674parameter @var{distance} is the distance in cycles between the two insns. 6675The hook returns @code{true} if considering the distance between the two 6676insns the dependence between them is considered costly by the target, 6677and @code{false} otherwise. 6678 6679Defining this hook can be useful in multiple-issue out-of-order machines, 6680where (a) it's practically hopeless to predict the actual data/resource 6681delays, however: (b) there's a better chance to predict the actual grouping 6682that will be formed, and (c) correctly emulating the grouping can be very 6683important. In such targets one may want to allow issuing dependent insns 6684closer to one another---i.e., closer than the dependence distance; however, 6685not in cases of ``costly dependences'', which this hooks allows to define. 6686@end deftypefn 6687 6688@hook TARGET_SCHED_H_I_D_EXTENDED 6689This hook is called by the insn scheduler after emitting a new instruction to 6690the instruction stream. The hook notifies a target backend to extend its 6691per instruction data structures. 6692@end deftypefn 6693 6694@hook TARGET_SCHED_ALLOC_SCHED_CONTEXT 6695Return a pointer to a store large enough to hold target scheduling context. 6696@end deftypefn 6697 6698@hook TARGET_SCHED_INIT_SCHED_CONTEXT 6699Initialize store pointed to by @var{tc} to hold target scheduling context. 6700It @var{clean_p} is true then initialize @var{tc} as if scheduler is at the 6701beginning of the block. Otherwise, copy the current context into @var{tc}. 6702@end deftypefn 6703 6704@hook TARGET_SCHED_SET_SCHED_CONTEXT 6705Copy target scheduling context pointed to by @var{tc} to the current context. 6706@end deftypefn 6707 6708@hook TARGET_SCHED_CLEAR_SCHED_CONTEXT 6709Deallocate internal data in target scheduling context pointed to by @var{tc}. 6710@end deftypefn 6711 6712@hook TARGET_SCHED_FREE_SCHED_CONTEXT 6713Deallocate a store for target scheduling context pointed to by @var{tc}. 6714@end deftypefn 6715 6716@hook TARGET_SCHED_SPECULATE_INSN 6717This hook is called by the insn scheduler when @var{insn} has only 6718speculative dependencies and therefore can be scheduled speculatively. 6719The hook is used to check if the pattern of @var{insn} has a speculative 6720version and, in case of successful check, to generate that speculative 6721pattern. The hook should return 1, if the instruction has a speculative form, 6722or @minus{}1, if it doesn't. @var{request} describes the type of requested 6723speculation. If the return value equals 1 then @var{new_pat} is assigned 6724the generated speculative pattern. 6725@end deftypefn 6726 6727@hook TARGET_SCHED_NEEDS_BLOCK_P 6728This hook is called by the insn scheduler during generation of recovery code 6729for @var{insn}. It should return @code{true}, if the corresponding check 6730instruction should branch to recovery code, or @code{false} otherwise. 6731@end deftypefn 6732 6733@hook TARGET_SCHED_GEN_SPEC_CHECK 6734This hook is called by the insn scheduler to generate a pattern for recovery 6735check instruction. If @var{mutate_p} is zero, then @var{insn} is a 6736speculative instruction for which the check should be generated. 6737@var{label} is either a label of a basic block, where recovery code should 6738be emitted, or a null pointer, when requested check doesn't branch to 6739recovery code (a simple check). If @var{mutate_p} is nonzero, then 6740a pattern for a branchy check corresponding to a simple check denoted by 6741@var{insn} should be generated. In this case @var{label} can't be null. 6742@end deftypefn 6743 6744@hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD_SPEC 6745This hook is used as a workaround for 6746@samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD} not being 6747called on the first instruction of the ready list. The hook is used to 6748discard speculative instructions that stand first in the ready list from 6749being scheduled on the current cycle. If the hook returns @code{false}, 6750@var{insn} will not be chosen to be issued. 6751For non-speculative instructions, 6752the hook should always return @code{true}. For example, in the ia64 backend 6753the hook is used to cancel data speculative insns when the ALAT table 6754is nearly full. 6755@end deftypefn 6756 6757@hook TARGET_SCHED_SET_SCHED_FLAGS 6758This hook is used by the insn scheduler to find out what features should be 6759enabled/used. 6760The structure *@var{spec_info} should be filled in by the target. 6761The structure describes speculation types that can be used in the scheduler. 6762@end deftypefn 6763 6764@hook TARGET_SCHED_SMS_RES_MII 6765This hook is called by the swing modulo scheduler to calculate a 6766resource-based lower bound which is based on the resources available in 6767the machine and the resources required by each instruction. The target 6768backend can use @var{g} to calculate such bound. A very simple lower 6769bound will be used in case this hook is not implemented: the total number 6770of instructions divided by the issue rate. 6771@end deftypefn 6772 6773@hook TARGET_SCHED_DISPATCH 6774This hook is called by Haifa Scheduler. It returns true if dispatch scheduling 6775is supported in hardware and the condition specified in the parameter is true. 6776@end deftypefn 6777 6778@hook TARGET_SCHED_DISPATCH_DO 6779This hook is called by Haifa Scheduler. It performs the operation specified 6780in its second parameter. 6781@end deftypefn 6782 6783@hook TARGET_SCHED_EXPOSED_PIPELINE 6784 6785@hook TARGET_SCHED_REASSOCIATION_WIDTH 6786 6787@node Sections 6788@section Dividing the Output into Sections (Texts, Data, @dots{}) 6789@c the above section title is WAY too long. maybe cut the part between 6790@c the (...)? --mew 10feb93 6791 6792An object file is divided into sections containing different types of 6793data. In the most common case, there are three sections: the @dfn{text 6794section}, which holds instructions and read-only data; the @dfn{data 6795section}, which holds initialized writable data; and the @dfn{bss 6796section}, which holds uninitialized data. Some systems have other kinds 6797of sections. 6798 6799@file{varasm.c} provides several well-known sections, such as 6800@code{text_section}, @code{data_section} and @code{bss_section}. 6801The normal way of controlling a @code{@var{foo}_section} variable 6802is to define the associated @code{@var{FOO}_SECTION_ASM_OP} macro, 6803as described below. The macros are only read once, when @file{varasm.c} 6804initializes itself, so their values must be run-time constants. 6805They may however depend on command-line flags. 6806 6807@emph{Note:} Some run-time files, such @file{crtstuff.c}, also make 6808use of the @code{@var{FOO}_SECTION_ASM_OP} macros, and expect them 6809to be string literals. 6810 6811Some assemblers require a different string to be written every time a 6812section is selected. If your assembler falls into this category, you 6813should define the @code{TARGET_ASM_INIT_SECTIONS} hook and use 6814@code{get_unnamed_section} to set up the sections. 6815 6816You must always create a @code{text_section}, either by defining 6817@code{TEXT_SECTION_ASM_OP} or by initializing @code{text_section} 6818in @code{TARGET_ASM_INIT_SECTIONS}. The same is true of 6819@code{data_section} and @code{DATA_SECTION_ASM_OP}. If you do not 6820create a distinct @code{readonly_data_section}, the default is to 6821reuse @code{text_section}. 6822 6823All the other @file{varasm.c} sections are optional, and are null 6824if the target does not provide them. 6825 6826@defmac TEXT_SECTION_ASM_OP 6827A C expression whose value is a string, including spacing, containing the 6828assembler operation that should precede instructions and read-only data. 6829Normally @code{"\t.text"} is right. 6830@end defmac 6831 6832@defmac HOT_TEXT_SECTION_NAME 6833If defined, a C string constant for the name of the section containing most 6834frequently executed functions of the program. If not defined, GCC will provide 6835a default definition if the target supports named sections. 6836@end defmac 6837 6838@defmac UNLIKELY_EXECUTED_TEXT_SECTION_NAME 6839If defined, a C string constant for the name of the section containing unlikely 6840executed functions in the program. 6841@end defmac 6842 6843@defmac DATA_SECTION_ASM_OP 6844A C expression whose value is a string, including spacing, containing the 6845assembler operation to identify the following data as writable initialized 6846data. Normally @code{"\t.data"} is right. 6847@end defmac 6848 6849@defmac SDATA_SECTION_ASM_OP 6850If defined, a C expression whose value is a string, including spacing, 6851containing the assembler operation to identify the following data as 6852initialized, writable small data. 6853@end defmac 6854 6855@defmac READONLY_DATA_SECTION_ASM_OP 6856A C expression whose value is a string, including spacing, containing the 6857assembler operation to identify the following data as read-only initialized 6858data. 6859@end defmac 6860 6861@defmac BSS_SECTION_ASM_OP 6862If defined, a C expression whose value is a string, including spacing, 6863containing the assembler operation to identify the following data as 6864uninitialized global data. If not defined, and 6865@code{ASM_OUTPUT_ALIGNED_BSS} not defined, 6866uninitialized global data will be output in the data section if 6867@option{-fno-common} is passed, otherwise @code{ASM_OUTPUT_COMMON} will be 6868used. 6869@end defmac 6870 6871@defmac SBSS_SECTION_ASM_OP 6872If defined, a C expression whose value is a string, including spacing, 6873containing the assembler operation to identify the following data as 6874uninitialized, writable small data. 6875@end defmac 6876 6877@defmac TLS_COMMON_ASM_OP 6878If defined, a C expression whose value is a string containing the 6879assembler operation to identify the following data as thread-local 6880common data. The default is @code{".tls_common"}. 6881@end defmac 6882 6883@defmac TLS_SECTION_ASM_FLAG 6884If defined, a C expression whose value is a character constant 6885containing the flag used to mark a section as a TLS section. The 6886default is @code{'T'}. 6887@end defmac 6888 6889@defmac INIT_SECTION_ASM_OP 6890If defined, a C expression whose value is a string, including spacing, 6891containing the assembler operation to identify the following data as 6892initialization code. If not defined, GCC will assume such a section does 6893not exist. This section has no corresponding @code{init_section} 6894variable; it is used entirely in runtime code. 6895@end defmac 6896 6897@defmac FINI_SECTION_ASM_OP 6898If defined, a C expression whose value is a string, including spacing, 6899containing the assembler operation to identify the following data as 6900finalization code. If not defined, GCC will assume such a section does 6901not exist. This section has no corresponding @code{fini_section} 6902variable; it is used entirely in runtime code. 6903@end defmac 6904 6905@defmac INIT_ARRAY_SECTION_ASM_OP 6906If defined, a C expression whose value is a string, including spacing, 6907containing the assembler operation to identify the following data as 6908part of the @code{.init_array} (or equivalent) section. If not 6909defined, GCC will assume such a section does not exist. Do not define 6910both this macro and @code{INIT_SECTION_ASM_OP}. 6911@end defmac 6912 6913@defmac FINI_ARRAY_SECTION_ASM_OP 6914If defined, a C expression whose value is a string, including spacing, 6915containing the assembler operation to identify the following data as 6916part of the @code{.fini_array} (or equivalent) section. If not 6917defined, GCC will assume such a section does not exist. Do not define 6918both this macro and @code{FINI_SECTION_ASM_OP}. 6919@end defmac 6920 6921@defmac CRT_CALL_STATIC_FUNCTION (@var{section_op}, @var{function}) 6922If defined, an ASM statement that switches to a different section 6923via @var{section_op}, calls @var{function}, and switches back to 6924the text section. This is used in @file{crtstuff.c} if 6925@code{INIT_SECTION_ASM_OP} or @code{FINI_SECTION_ASM_OP} to calls 6926to initialization and finalization functions from the init and fini 6927sections. By default, this macro uses a simple function call. Some 6928ports need hand-crafted assembly code to avoid dependencies on 6929registers initialized in the function prologue or to ensure that 6930constant pools don't end up too far way in the text section. 6931@end defmac 6932 6933@defmac TARGET_LIBGCC_SDATA_SECTION 6934If defined, a string which names the section into which small 6935variables defined in crtstuff and libgcc should go. This is useful 6936when the target has options for optimizing access to small data, and 6937you want the crtstuff and libgcc routines to be conservative in what 6938they expect of your application yet liberal in what your application 6939expects. For example, for targets with a @code{.sdata} section (like 6940MIPS), you could compile crtstuff with @code{-G 0} so that it doesn't 6941require small data support from your application, but use this macro 6942to put small data into @code{.sdata} so that your application can 6943access these variables whether it uses small data or not. 6944@end defmac 6945 6946@defmac FORCE_CODE_SECTION_ALIGN 6947If defined, an ASM statement that aligns a code section to some 6948arbitrary boundary. This is used to force all fragments of the 6949@code{.init} and @code{.fini} sections to have to same alignment 6950and thus prevent the linker from having to add any padding. 6951@end defmac 6952 6953@defmac JUMP_TABLES_IN_TEXT_SECTION 6954Define this macro to be an expression with a nonzero value if jump 6955tables (for @code{tablejump} insns) should be output in the text 6956section, along with the assembler instructions. Otherwise, the 6957readonly data section is used. 6958 6959This macro is irrelevant if there is no separate readonly data section. 6960@end defmac 6961 6962@hook TARGET_ASM_INIT_SECTIONS 6963Define this hook if you need to do something special to set up the 6964@file{varasm.c} sections, or if your target has some special sections 6965of its own that you need to create. 6966 6967GCC calls this hook after processing the command line, but before writing 6968any assembly code, and before calling any of the section-returning hooks 6969described below. 6970@end deftypefn 6971 6972@hook TARGET_ASM_RELOC_RW_MASK 6973Return a mask describing how relocations should be treated when 6974selecting sections. Bit 1 should be set if global relocations 6975should be placed in a read-write section; bit 0 should be set if 6976local relocations should be placed in a read-write section. 6977 6978The default version of this function returns 3 when @option{-fpic} 6979is in effect, and 0 otherwise. The hook is typically redefined 6980when the target cannot support (some kinds of) dynamic relocations 6981in read-only sections even in executables. 6982@end deftypefn 6983 6984@hook TARGET_ASM_SELECT_SECTION 6985Return the section into which @var{exp} should be placed. You can 6986assume that @var{exp} is either a @code{VAR_DECL} node or a constant of 6987some sort. @var{reloc} indicates whether the initial value of @var{exp} 6988requires link-time relocations. Bit 0 is set when variable contains 6989local relocations only, while bit 1 is set for global relocations. 6990@var{align} is the constant alignment in bits. 6991 6992The default version of this function takes care of putting read-only 6993variables in @code{readonly_data_section}. 6994 6995See also @var{USE_SELECT_SECTION_FOR_FUNCTIONS}. 6996@end deftypefn 6997 6998@defmac USE_SELECT_SECTION_FOR_FUNCTIONS 6999Define this macro if you wish TARGET_ASM_SELECT_SECTION to be called 7000for @code{FUNCTION_DECL}s as well as for variables and constants. 7001 7002In the case of a @code{FUNCTION_DECL}, @var{reloc} will be zero if the 7003function has been determined to be likely to be called, and nonzero if 7004it is unlikely to be called. 7005@end defmac 7006 7007@hook TARGET_ASM_UNIQUE_SECTION 7008Build up a unique section name, expressed as a @code{STRING_CST} node, 7009and assign it to @samp{DECL_SECTION_NAME (@var{decl})}. 7010As with @code{TARGET_ASM_SELECT_SECTION}, @var{reloc} indicates whether 7011the initial value of @var{exp} requires link-time relocations. 7012 7013The default version of this function appends the symbol name to the 7014ELF section name that would normally be used for the symbol. For 7015example, the function @code{foo} would be placed in @code{.text.foo}. 7016Whatever the actual target object format, this is often good enough. 7017@end deftypefn 7018 7019@hook TARGET_ASM_FUNCTION_RODATA_SECTION 7020Return the readonly data section associated with 7021@samp{DECL_SECTION_NAME (@var{decl})}. 7022The default version of this function selects @code{.gnu.linkonce.r.name} if 7023the function's section is @code{.gnu.linkonce.t.name}, @code{.rodata.name} 7024if function is in @code{.text.name}, and the normal readonly-data section 7025otherwise. 7026@end deftypefn 7027 7028@hook TARGET_ASM_MERGEABLE_RODATA_PREFIX 7029 7030@hook TARGET_ASM_TM_CLONE_TABLE_SECTION 7031 7032@hook TARGET_ASM_SELECT_RTX_SECTION 7033Return the section into which a constant @var{x}, of mode @var{mode}, 7034should be placed. You can assume that @var{x} is some kind of 7035constant in RTL@. The argument @var{mode} is redundant except in the 7036case of a @code{const_int} rtx. @var{align} is the constant alignment 7037in bits. 7038 7039The default version of this function takes care of putting symbolic 7040constants in @code{flag_pic} mode in @code{data_section} and everything 7041else in @code{readonly_data_section}. 7042@end deftypefn 7043 7044@hook TARGET_MANGLE_DECL_ASSEMBLER_NAME 7045Define this hook if you need to postprocess the assembler name generated 7046by target-independent code. The @var{id} provided to this hook will be 7047the computed name (e.g., the macro @code{DECL_NAME} of the @var{decl} in C, 7048or the mangled name of the @var{decl} in C++). The return value of the 7049hook is an @code{IDENTIFIER_NODE} for the appropriate mangled name on 7050your target system. The default implementation of this hook just 7051returns the @var{id} provided. 7052@end deftypefn 7053 7054@hook TARGET_ENCODE_SECTION_INFO 7055Define this hook if references to a symbol or a constant must be 7056treated differently depending on something about the variable or 7057function named by the symbol (such as what section it is in). 7058 7059The hook is executed immediately after rtl has been created for 7060@var{decl}, which may be a variable or function declaration or 7061an entry in the constant pool. In either case, @var{rtl} is the 7062rtl in question. Do @emph{not} use @code{DECL_RTL (@var{decl})} 7063in this hook; that field may not have been initialized yet. 7064 7065In the case of a constant, it is safe to assume that the rtl is 7066a @code{mem} whose address is a @code{symbol_ref}. Most decls 7067will also have this form, but that is not guaranteed. Global 7068register variables, for instance, will have a @code{reg} for their 7069rtl. (Normally the right thing to do with such unusual rtl is 7070leave it alone.) 7071 7072The @var{new_decl_p} argument will be true if this is the first time 7073that @code{TARGET_ENCODE_SECTION_INFO} has been invoked on this decl. It will 7074be false for subsequent invocations, which will happen for duplicate 7075declarations. Whether or not anything must be done for the duplicate 7076declaration depends on whether the hook examines @code{DECL_ATTRIBUTES}. 7077@var{new_decl_p} is always true when the hook is called for a constant. 7078 7079@cindex @code{SYMBOL_REF_FLAG}, in @code{TARGET_ENCODE_SECTION_INFO} 7080The usual thing for this hook to do is to record flags in the 7081@code{symbol_ref}, using @code{SYMBOL_REF_FLAG} or @code{SYMBOL_REF_FLAGS}. 7082Historically, the name string was modified if it was necessary to 7083encode more than one bit of information, but this practice is now 7084discouraged; use @code{SYMBOL_REF_FLAGS}. 7085 7086The default definition of this hook, @code{default_encode_section_info} 7087in @file{varasm.c}, sets a number of commonly-useful bits in 7088@code{SYMBOL_REF_FLAGS}. Check whether the default does what you need 7089before overriding it. 7090@end deftypefn 7091 7092@hook TARGET_STRIP_NAME_ENCODING 7093Decode @var{name} and return the real name part, sans 7094the characters that @code{TARGET_ENCODE_SECTION_INFO} 7095may have added. 7096@end deftypefn 7097 7098@hook TARGET_IN_SMALL_DATA_P 7099Returns true if @var{exp} should be placed into a ``small data'' section. 7100The default version of this hook always returns false. 7101@end deftypefn 7102 7103@hook TARGET_HAVE_SRODATA_SECTION 7104Contains the value true if the target places read-only 7105``small data'' into a separate section. The default value is false. 7106@end deftypevr 7107 7108@hook TARGET_PROFILE_BEFORE_PROLOGUE 7109 7110@hook TARGET_BINDS_LOCAL_P 7111Returns true if @var{exp} names an object for which name resolution 7112rules must resolve to the current ``module'' (dynamic shared library 7113or executable image). 7114 7115The default version of this hook implements the name resolution rules 7116for ELF, which has a looser model of global name binding than other 7117currently supported object file formats. 7118@end deftypefn 7119 7120@hook TARGET_HAVE_TLS 7121Contains the value true if the target supports thread-local storage. 7122The default value is false. 7123@end deftypevr 7124 7125 7126@node PIC 7127@section Position Independent Code 7128@cindex position independent code 7129@cindex PIC 7130 7131This section describes macros that help implement generation of position 7132independent code. Simply defining these macros is not enough to 7133generate valid PIC; you must also add support to the hook 7134@code{TARGET_LEGITIMATE_ADDRESS_P} and to the macro 7135@code{PRINT_OPERAND_ADDRESS}, as well as @code{LEGITIMIZE_ADDRESS}. You 7136must modify the definition of @samp{movsi} to do something appropriate 7137when the source operand contains a symbolic address. You may also 7138need to alter the handling of switch statements so that they use 7139relative addresses. 7140@c i rearranged the order of the macros above to try to force one of 7141@c them to the next line, to eliminate an overfull hbox. --mew 10feb93 7142 7143@defmac PIC_OFFSET_TABLE_REGNUM 7144The register number of the register used to address a table of static 7145data addresses in memory. In some cases this register is defined by a 7146processor's ``application binary interface'' (ABI)@. When this macro 7147is defined, RTL is generated for this register once, as with the stack 7148pointer and frame pointer registers. If this macro is not defined, it 7149is up to the machine-dependent files to allocate such a register (if 7150necessary). Note that this register must be fixed when in use (e.g.@: 7151when @code{flag_pic} is true). 7152@end defmac 7153 7154@defmac PIC_OFFSET_TABLE_REG_CALL_CLOBBERED 7155A C expression that is nonzero if the register defined by 7156@code{PIC_OFFSET_TABLE_REGNUM} is clobbered by calls. If not defined, 7157the default is zero. Do not define 7158this macro if @code{PIC_OFFSET_TABLE_REGNUM} is not defined. 7159@end defmac 7160 7161@defmac LEGITIMATE_PIC_OPERAND_P (@var{x}) 7162A C expression that is nonzero if @var{x} is a legitimate immediate 7163operand on the target machine when generating position independent code. 7164You can assume that @var{x} satisfies @code{CONSTANT_P}, so you need not 7165check this. You can also assume @var{flag_pic} is true, so you need not 7166check it either. You need not define this macro if all constants 7167(including @code{SYMBOL_REF}) can be immediate operands when generating 7168position independent code. 7169@end defmac 7170 7171@node Assembler Format 7172@section Defining the Output Assembler Language 7173 7174This section describes macros whose principal purpose is to describe how 7175to write instructions in assembler language---rather than what the 7176instructions do. 7177 7178@menu 7179* File Framework:: Structural information for the assembler file. 7180* Data Output:: Output of constants (numbers, strings, addresses). 7181* Uninitialized Data:: Output of uninitialized variables. 7182* Label Output:: Output and generation of labels. 7183* Initialization:: General principles of initialization 7184 and termination routines. 7185* Macros for Initialization:: 7186 Specific macros that control the handling of 7187 initialization and termination routines. 7188* Instruction Output:: Output of actual instructions. 7189* Dispatch Tables:: Output of jump tables. 7190* Exception Region Output:: Output of exception region code. 7191* Alignment Output:: Pseudo ops for alignment and skipping data. 7192@end menu 7193 7194@node File Framework 7195@subsection The Overall Framework of an Assembler File 7196@cindex assembler format 7197@cindex output of assembler code 7198 7199@c prevent bad page break with this line 7200This describes the overall framework of an assembly file. 7201 7202@findex default_file_start 7203@hook TARGET_ASM_FILE_START 7204Output to @code{asm_out_file} any text which the assembler expects to 7205find at the beginning of a file. The default behavior is controlled 7206by two flags, documented below. Unless your target's assembler is 7207quite unusual, if you override the default, you should call 7208@code{default_file_start} at some point in your target hook. This 7209lets other target files rely on these variables. 7210@end deftypefn 7211 7212@hook TARGET_ASM_FILE_START_APP_OFF 7213If this flag is true, the text of the macro @code{ASM_APP_OFF} will be 7214printed as the very first line in the assembly file, unless 7215@option{-fverbose-asm} is in effect. (If that macro has been defined 7216to the empty string, this variable has no effect.) With the normal 7217definition of @code{ASM_APP_OFF}, the effect is to notify the GNU 7218assembler that it need not bother stripping comments or extra 7219whitespace from its input. This allows it to work a bit faster. 7220 7221The default is false. You should not set it to true unless you have 7222verified that your port does not generate any extra whitespace or 7223comments that will cause GAS to issue errors in NO_APP mode. 7224@end deftypevr 7225 7226@hook TARGET_ASM_FILE_START_FILE_DIRECTIVE 7227If this flag is true, @code{output_file_directive} will be called 7228for the primary source file, immediately after printing 7229@code{ASM_APP_OFF} (if that is enabled). Most ELF assemblers expect 7230this to be done. The default is false. 7231@end deftypevr 7232 7233@hook TARGET_ASM_FILE_END 7234Output to @code{asm_out_file} any text which the assembler expects 7235to find at the end of a file. The default is to output nothing. 7236@end deftypefn 7237 7238@deftypefun void file_end_indicate_exec_stack () 7239Some systems use a common convention, the @samp{.note.GNU-stack} 7240special section, to indicate whether or not an object file relies on 7241the stack being executable. If your system uses this convention, you 7242should define @code{TARGET_ASM_FILE_END} to this function. If you 7243need to do other things in that hook, have your hook function call 7244this function. 7245@end deftypefun 7246 7247@hook TARGET_ASM_LTO_START 7248Output to @code{asm_out_file} any text which the assembler expects 7249to find at the start of an LTO section. The default is to output 7250nothing. 7251@end deftypefn 7252 7253@hook TARGET_ASM_LTO_END 7254Output to @code{asm_out_file} any text which the assembler expects 7255to find at the end of an LTO section. The default is to output 7256nothing. 7257@end deftypefn 7258 7259@hook TARGET_ASM_CODE_END 7260Output to @code{asm_out_file} any text which is needed before emitting 7261unwind info and debug info at the end of a file. Some targets emit 7262here PIC setup thunks that cannot be emitted at the end of file, 7263because they couldn't have unwind info then. The default is to output 7264nothing. 7265@end deftypefn 7266 7267@defmac ASM_COMMENT_START 7268A C string constant describing how to begin a comment in the target 7269assembler language. The compiler assumes that the comment will end at 7270the end of the line. 7271@end defmac 7272 7273@defmac ASM_APP_ON 7274A C string constant for text to be output before each @code{asm} 7275statement or group of consecutive ones. Normally this is 7276@code{"#APP"}, which is a comment that has no effect on most 7277assemblers but tells the GNU assembler that it must check the lines 7278that follow for all valid assembler constructs. 7279@end defmac 7280 7281@defmac ASM_APP_OFF 7282A C string constant for text to be output after each @code{asm} 7283statement or group of consecutive ones. Normally this is 7284@code{"#NO_APP"}, which tells the GNU assembler to resume making the 7285time-saving assumptions that are valid for ordinary compiler output. 7286@end defmac 7287 7288@defmac ASM_OUTPUT_SOURCE_FILENAME (@var{stream}, @var{name}) 7289A C statement to output COFF information or DWARF debugging information 7290which indicates that filename @var{name} is the current source file to 7291the stdio stream @var{stream}. 7292 7293This macro need not be defined if the standard form of output 7294for the file format in use is appropriate. 7295@end defmac 7296 7297@hook TARGET_ASM_OUTPUT_SOURCE_FILENAME 7298 7299@defmac OUTPUT_QUOTED_STRING (@var{stream}, @var{string}) 7300A C statement to output the string @var{string} to the stdio stream 7301@var{stream}. If you do not call the function @code{output_quoted_string} 7302in your config files, GCC will only call it to output filenames to 7303the assembler source. So you can use it to canonicalize the format 7304of the filename using this macro. 7305@end defmac 7306 7307@defmac ASM_OUTPUT_IDENT (@var{stream}, @var{string}) 7308A C statement to output something to the assembler file to handle a 7309@samp{#ident} directive containing the text @var{string}. If this 7310macro is not defined, nothing is output for a @samp{#ident} directive. 7311@end defmac 7312 7313@hook TARGET_ASM_NAMED_SECTION 7314Output assembly directives to switch to section @var{name}. The section 7315should have attributes as specified by @var{flags}, which is a bit mask 7316of the @code{SECTION_*} flags defined in @file{output.h}. If @var{decl} 7317is non-NULL, it is the @code{VAR_DECL} or @code{FUNCTION_DECL} with which 7318this section is associated. 7319@end deftypefn 7320 7321@hook TARGET_ASM_FUNCTION_SECTION 7322Return preferred text (sub)section for function @var{decl}. 7323Main purpose of this function is to separate cold, normal and hot 7324functions. @var{startup} is true when function is known to be used only 7325at startup (from static constructors or it is @code{main()}). 7326@var{exit} is true when function is known to be used only at exit 7327(from static destructors). 7328Return NULL if function should go to default text section. 7329@end deftypefn 7330 7331@hook TARGET_ASM_FUNCTION_SWITCHED_TEXT_SECTIONS 7332 7333@hook TARGET_HAVE_NAMED_SECTIONS 7334This flag is true if the target supports @code{TARGET_ASM_NAMED_SECTION}. 7335It must not be modified by command-line option processing. 7336@end deftypevr 7337 7338@anchor{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS} 7339@hook TARGET_HAVE_SWITCHABLE_BSS_SECTIONS 7340This flag is true if we can create zeroed data by switching to a BSS 7341section and then using @code{ASM_OUTPUT_SKIP} to allocate the space. 7342This is true on most ELF targets. 7343@end deftypevr 7344 7345@hook TARGET_SECTION_TYPE_FLAGS 7346Choose a set of section attributes for use by @code{TARGET_ASM_NAMED_SECTION} 7347based on a variable or function decl, a section name, and whether or not the 7348declaration's initializer may contain runtime relocations. @var{decl} may be 7349null, in which case read-write data should be assumed. 7350 7351The default version of this function handles choosing code vs data, 7352read-only vs read-write data, and @code{flag_pic}. You should only 7353need to override this if your target has special flags that might be 7354set via @code{__attribute__}. 7355@end deftypefn 7356 7357@hook TARGET_ASM_RECORD_GCC_SWITCHES 7358Provides the target with the ability to record the gcc command line 7359switches that have been passed to the compiler, and options that are 7360enabled. The @var{type} argument specifies what is being recorded. 7361It can take the following values: 7362 7363@table @gcctabopt 7364@item SWITCH_TYPE_PASSED 7365@var{text} is a command line switch that has been set by the user. 7366 7367@item SWITCH_TYPE_ENABLED 7368@var{text} is an option which has been enabled. This might be as a 7369direct result of a command line switch, or because it is enabled by 7370default or because it has been enabled as a side effect of a different 7371command line switch. For example, the @option{-O2} switch enables 7372various different individual optimization passes. 7373 7374@item SWITCH_TYPE_DESCRIPTIVE 7375@var{text} is either NULL or some descriptive text which should be 7376ignored. If @var{text} is NULL then it is being used to warn the 7377target hook that either recording is starting or ending. The first 7378time @var{type} is SWITCH_TYPE_DESCRIPTIVE and @var{text} is NULL, the 7379warning is for start up and the second time the warning is for 7380wind down. This feature is to allow the target hook to make any 7381necessary preparations before it starts to record switches and to 7382perform any necessary tidying up after it has finished recording 7383switches. 7384 7385@item SWITCH_TYPE_LINE_START 7386This option can be ignored by this target hook. 7387 7388@item SWITCH_TYPE_LINE_END 7389This option can be ignored by this target hook. 7390@end table 7391 7392The hook's return value must be zero. Other return values may be 7393supported in the future. 7394 7395By default this hook is set to NULL, but an example implementation is 7396provided for ELF based targets. Called @var{elf_record_gcc_switches}, 7397it records the switches as ASCII text inside a new, string mergeable 7398section in the assembler output file. The name of the new section is 7399provided by the @code{TARGET_ASM_RECORD_GCC_SWITCHES_SECTION} target 7400hook. 7401@end deftypefn 7402 7403@hook TARGET_ASM_RECORD_GCC_SWITCHES_SECTION 7404This is the name of the section that will be created by the example 7405ELF implementation of the @code{TARGET_ASM_RECORD_GCC_SWITCHES} target 7406hook. 7407@end deftypevr 7408 7409@need 2000 7410@node Data Output 7411@subsection Output of Data 7412 7413 7414@hook TARGET_ASM_BYTE_OP 7415@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_HI_OP 7416@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_SI_OP 7417@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_DI_OP 7418@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_TI_OP 7419@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_HI_OP 7420@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_SI_OP 7421@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_DI_OP 7422@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_TI_OP 7423These hooks specify assembly directives for creating certain kinds 7424of integer object. The @code{TARGET_ASM_BYTE_OP} directive creates a 7425byte-sized object, the @code{TARGET_ASM_ALIGNED_HI_OP} one creates an 7426aligned two-byte object, and so on. Any of the hooks may be 7427@code{NULL}, indicating that no suitable directive is available. 7428 7429The compiler will print these strings at the start of a new line, 7430followed immediately by the object's initial value. In most cases, 7431the string should contain a tab, a pseudo-op, and then another tab. 7432@end deftypevr 7433 7434@hook TARGET_ASM_INTEGER 7435The @code{assemble_integer} function uses this hook to output an 7436integer object. @var{x} is the object's value, @var{size} is its size 7437in bytes and @var{aligned_p} indicates whether it is aligned. The 7438function should return @code{true} if it was able to output the 7439object. If it returns false, @code{assemble_integer} will try to 7440split the object into smaller parts. 7441 7442The default implementation of this hook will use the 7443@code{TARGET_ASM_BYTE_OP} family of strings, returning @code{false} 7444when the relevant string is @code{NULL}. 7445@end deftypefn 7446 7447@hook TARGET_ASM_OUTPUT_ADDR_CONST_EXTRA 7448A target hook to recognize @var{rtx} patterns that @code{output_addr_const} 7449can't deal with, and output assembly code to @var{file} corresponding to 7450the pattern @var{x}. This may be used to allow machine-dependent 7451@code{UNSPEC}s to appear within constants. 7452 7453If target hook fails to recognize a pattern, it must return @code{false}, 7454so that a standard error message is printed. If it prints an error message 7455itself, by calling, for example, @code{output_operand_lossage}, it may just 7456return @code{true}. 7457@end deftypefn 7458 7459@defmac ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len}) 7460A C statement to output to the stdio stream @var{stream} an assembler 7461instruction to assemble a string constant containing the @var{len} 7462bytes at @var{ptr}. @var{ptr} will be a C expression of type 7463@code{char *} and @var{len} a C expression of type @code{int}. 7464 7465If the assembler has a @code{.ascii} pseudo-op as found in the 7466Berkeley Unix assembler, do not define the macro 7467@code{ASM_OUTPUT_ASCII}. 7468@end defmac 7469 7470@defmac ASM_OUTPUT_FDESC (@var{stream}, @var{decl}, @var{n}) 7471A C statement to output word @var{n} of a function descriptor for 7472@var{decl}. This must be defined if @code{TARGET_VTABLE_USES_DESCRIPTORS} 7473is defined, and is otherwise unused. 7474@end defmac 7475 7476@defmac CONSTANT_POOL_BEFORE_FUNCTION 7477You may define this macro as a C expression. You should define the 7478expression to have a nonzero value if GCC should output the constant 7479pool for a function before the code for the function, or a zero value if 7480GCC should output the constant pool after the function. If you do 7481not define this macro, the usual case, GCC will output the constant 7482pool before the function. 7483@end defmac 7484 7485@defmac ASM_OUTPUT_POOL_PROLOGUE (@var{file}, @var{funname}, @var{fundecl}, @var{size}) 7486A C statement to output assembler commands to define the start of the 7487constant pool for a function. @var{funname} is a string giving 7488the name of the function. Should the return type of the function 7489be required, it can be obtained via @var{fundecl}. @var{size} 7490is the size, in bytes, of the constant pool that will be written 7491immediately after this call. 7492 7493If no constant-pool prefix is required, the usual case, this macro need 7494not be defined. 7495@end defmac 7496 7497@defmac ASM_OUTPUT_SPECIAL_POOL_ENTRY (@var{file}, @var{x}, @var{mode}, @var{align}, @var{labelno}, @var{jumpto}) 7498A C statement (with or without semicolon) to output a constant in the 7499constant pool, if it needs special treatment. (This macro need not do 7500anything for RTL expressions that can be output normally.) 7501 7502The argument @var{file} is the standard I/O stream to output the 7503assembler code on. @var{x} is the RTL expression for the constant to 7504output, and @var{mode} is the machine mode (in case @var{x} is a 7505@samp{const_int}). @var{align} is the required alignment for the value 7506@var{x}; you should output an assembler directive to force this much 7507alignment. 7508 7509The argument @var{labelno} is a number to use in an internal label for 7510the address of this pool entry. The definition of this macro is 7511responsible for outputting the label definition at the proper place. 7512Here is how to do this: 7513 7514@smallexample 7515@code{(*targetm.asm_out.internal_label)} (@var{file}, "LC", @var{labelno}); 7516@end smallexample 7517 7518When you output a pool entry specially, you should end with a 7519@code{goto} to the label @var{jumpto}. This will prevent the same pool 7520entry from being output a second time in the usual manner. 7521 7522You need not define this macro if it would do nothing. 7523@end defmac 7524 7525@defmac ASM_OUTPUT_POOL_EPILOGUE (@var{file} @var{funname} @var{fundecl} @var{size}) 7526A C statement to output assembler commands to at the end of the constant 7527pool for a function. @var{funname} is a string giving the name of the 7528function. Should the return type of the function be required, you can 7529obtain it via @var{fundecl}. @var{size} is the size, in bytes, of the 7530constant pool that GCC wrote immediately before this call. 7531 7532If no constant-pool epilogue is required, the usual case, you need not 7533define this macro. 7534@end defmac 7535 7536@defmac IS_ASM_LOGICAL_LINE_SEPARATOR (@var{C}, @var{STR}) 7537Define this macro as a C expression which is nonzero if @var{C} is 7538used as a logical line separator by the assembler. @var{STR} points 7539to the position in the string where @var{C} was found; this can be used if 7540a line separator uses multiple characters. 7541 7542If you do not define this macro, the default is that only 7543the character @samp{;} is treated as a logical line separator. 7544@end defmac 7545 7546@hook TARGET_ASM_OPEN_PAREN 7547These target hooks are C string constants, describing the syntax in the 7548assembler for grouping arithmetic expressions. If not overridden, they 7549default to normal parentheses, which is correct for most assemblers. 7550@end deftypevr 7551 7552These macros are provided by @file{real.h} for writing the definitions 7553of @code{ASM_OUTPUT_DOUBLE} and the like: 7554 7555@defmac REAL_VALUE_TO_TARGET_SINGLE (@var{x}, @var{l}) 7556@defmacx REAL_VALUE_TO_TARGET_DOUBLE (@var{x}, @var{l}) 7557@defmacx REAL_VALUE_TO_TARGET_LONG_DOUBLE (@var{x}, @var{l}) 7558@defmacx REAL_VALUE_TO_TARGET_DECIMAL32 (@var{x}, @var{l}) 7559@defmacx REAL_VALUE_TO_TARGET_DECIMAL64 (@var{x}, @var{l}) 7560@defmacx REAL_VALUE_TO_TARGET_DECIMAL128 (@var{x}, @var{l}) 7561These translate @var{x}, of type @code{REAL_VALUE_TYPE}, to the 7562target's floating point representation, and store its bit pattern in 7563the variable @var{l}. For @code{REAL_VALUE_TO_TARGET_SINGLE} and 7564@code{REAL_VALUE_TO_TARGET_DECIMAL32}, this variable should be a 7565simple @code{long int}. For the others, it should be an array of 7566@code{long int}. The number of elements in this array is determined 7567by the size of the desired target floating point data type: 32 bits of 7568it go in each @code{long int} array element. Each array element holds 756932 bits of the result, even if @code{long int} is wider than 32 bits 7570on the host machine. 7571 7572The array element values are designed so that you can print them out 7573using @code{fprintf} in the order they should appear in the target 7574machine's memory. 7575@end defmac 7576 7577@node Uninitialized Data 7578@subsection Output of Uninitialized Variables 7579 7580Each of the macros in this section is used to do the whole job of 7581outputting a single uninitialized variable. 7582 7583@defmac ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded}) 7584A C statement (sans semicolon) to output to the stdio stream 7585@var{stream} the assembler definition of a common-label named 7586@var{name} whose size is @var{size} bytes. The variable @var{rounded} 7587is the size rounded up to whatever alignment the caller wants. It is 7588possible that @var{size} may be zero, for instance if a struct with no 7589other member than a zero-length array is defined. In this case, the 7590backend must output a symbol definition that allocates at least one 7591byte, both so that the address of the resulting object does not compare 7592equal to any other, and because some object formats cannot even express 7593the concept of a zero-sized common symbol, as that is how they represent 7594an ordinary undefined external. 7595 7596Use the expression @code{assemble_name (@var{stream}, @var{name})} to 7597output the name itself; before and after that, output the additional 7598assembler syntax for defining the name, and a newline. 7599 7600This macro controls how the assembler definitions of uninitialized 7601common global variables are output. 7602@end defmac 7603 7604@defmac ASM_OUTPUT_ALIGNED_COMMON (@var{stream}, @var{name}, @var{size}, @var{alignment}) 7605Like @code{ASM_OUTPUT_COMMON} except takes the required alignment as a 7606separate, explicit argument. If you define this macro, it is used in 7607place of @code{ASM_OUTPUT_COMMON}, and gives you more flexibility in 7608handling the required alignment of the variable. The alignment is specified 7609as the number of bits. 7610@end defmac 7611 7612@defmac ASM_OUTPUT_ALIGNED_DECL_COMMON (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment}) 7613Like @code{ASM_OUTPUT_ALIGNED_COMMON} except that @var{decl} of the 7614variable to be output, if there is one, or @code{NULL_TREE} if there 7615is no corresponding variable. If you define this macro, GCC will use it 7616in place of both @code{ASM_OUTPUT_COMMON} and 7617@code{ASM_OUTPUT_ALIGNED_COMMON}. Define this macro when you need to see 7618the variable's decl in order to chose what to output. 7619@end defmac 7620 7621@defmac ASM_OUTPUT_ALIGNED_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment}) 7622A C statement (sans semicolon) to output to the stdio stream 7623@var{stream} the assembler definition of uninitialized global @var{decl} named 7624@var{name} whose size is @var{size} bytes. The variable @var{alignment} 7625is the alignment specified as the number of bits. 7626 7627Try to use function @code{asm_output_aligned_bss} defined in file 7628@file{varasm.c} when defining this macro. If unable, use the expression 7629@code{assemble_name (@var{stream}, @var{name})} to output the name itself; 7630before and after that, output the additional assembler syntax for defining 7631the name, and a newline. 7632 7633There are two ways of handling global BSS@. One is to define this macro. 7634The other is to have @code{TARGET_ASM_SELECT_SECTION} return a 7635switchable BSS section (@pxref{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS}). 7636You do not need to do both. 7637 7638Some languages do not have @code{common} data, and require a 7639non-common form of global BSS in order to handle uninitialized globals 7640efficiently. C++ is one example of this. However, if the target does 7641not support global BSS, the front end may choose to make globals 7642common in order to save space in the object file. 7643@end defmac 7644 7645@defmac ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded}) 7646A C statement (sans semicolon) to output to the stdio stream 7647@var{stream} the assembler definition of a local-common-label named 7648@var{name} whose size is @var{size} bytes. The variable @var{rounded} 7649is the size rounded up to whatever alignment the caller wants. 7650 7651Use the expression @code{assemble_name (@var{stream}, @var{name})} to 7652output the name itself; before and after that, output the additional 7653assembler syntax for defining the name, and a newline. 7654 7655This macro controls how the assembler definitions of uninitialized 7656static variables are output. 7657@end defmac 7658 7659@defmac ASM_OUTPUT_ALIGNED_LOCAL (@var{stream}, @var{name}, @var{size}, @var{alignment}) 7660Like @code{ASM_OUTPUT_LOCAL} except takes the required alignment as a 7661separate, explicit argument. If you define this macro, it is used in 7662place of @code{ASM_OUTPUT_LOCAL}, and gives you more flexibility in 7663handling the required alignment of the variable. The alignment is specified 7664as the number of bits. 7665@end defmac 7666 7667@defmac ASM_OUTPUT_ALIGNED_DECL_LOCAL (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment}) 7668Like @code{ASM_OUTPUT_ALIGNED_DECL} except that @var{decl} of the 7669variable to be output, if there is one, or @code{NULL_TREE} if there 7670is no corresponding variable. If you define this macro, GCC will use it 7671in place of both @code{ASM_OUTPUT_DECL} and 7672@code{ASM_OUTPUT_ALIGNED_DECL}. Define this macro when you need to see 7673the variable's decl in order to chose what to output. 7674@end defmac 7675 7676@node Label Output 7677@subsection Output and Generation of Labels 7678 7679@c prevent bad page break with this line 7680This is about outputting labels. 7681 7682@findex assemble_name 7683@defmac ASM_OUTPUT_LABEL (@var{stream}, @var{name}) 7684A C statement (sans semicolon) to output to the stdio stream 7685@var{stream} the assembler definition of a label named @var{name}. 7686Use the expression @code{assemble_name (@var{stream}, @var{name})} to 7687output the name itself; before and after that, output the additional 7688assembler syntax for defining the name, and a newline. A default 7689definition of this macro is provided which is correct for most systems. 7690@end defmac 7691 7692@defmac ASM_OUTPUT_FUNCTION_LABEL (@var{stream}, @var{name}, @var{decl}) 7693A C statement (sans semicolon) to output to the stdio stream 7694@var{stream} the assembler definition of a label named @var{name} of 7695a function. 7696Use the expression @code{assemble_name (@var{stream}, @var{name})} to 7697output the name itself; before and after that, output the additional 7698assembler syntax for defining the name, and a newline. A default 7699definition of this macro is provided which is correct for most systems. 7700 7701If this macro is not defined, then the function name is defined in the 7702usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}). 7703@end defmac 7704 7705@findex assemble_name_raw 7706@defmac ASM_OUTPUT_INTERNAL_LABEL (@var{stream}, @var{name}) 7707Identical to @code{ASM_OUTPUT_LABEL}, except that @var{name} is known 7708to refer to a compiler-generated label. The default definition uses 7709@code{assemble_name_raw}, which is like @code{assemble_name} except 7710that it is more efficient. 7711@end defmac 7712 7713@defmac SIZE_ASM_OP 7714A C string containing the appropriate assembler directive to specify the 7715size of a symbol, without any arguments. On systems that use ELF, the 7716default (in @file{config/elfos.h}) is @samp{"\t.size\t"}; on other 7717systems, the default is not to define this macro. 7718 7719Define this macro only if it is correct to use the default definitions 7720of @code{ASM_OUTPUT_SIZE_DIRECTIVE} and @code{ASM_OUTPUT_MEASURED_SIZE} 7721for your system. If you need your own custom definitions of those 7722macros, or if you do not need explicit symbol sizes at all, do not 7723define this macro. 7724@end defmac 7725 7726@defmac ASM_OUTPUT_SIZE_DIRECTIVE (@var{stream}, @var{name}, @var{size}) 7727A C statement (sans semicolon) to output to the stdio stream 7728@var{stream} a directive telling the assembler that the size of the 7729symbol @var{name} is @var{size}. @var{size} is a @code{HOST_WIDE_INT}. 7730If you define @code{SIZE_ASM_OP}, a default definition of this macro is 7731provided. 7732@end defmac 7733 7734@defmac ASM_OUTPUT_MEASURED_SIZE (@var{stream}, @var{name}) 7735A C statement (sans semicolon) to output to the stdio stream 7736@var{stream} a directive telling the assembler to calculate the size of 7737the symbol @var{name} by subtracting its address from the current 7738address. 7739 7740If you define @code{SIZE_ASM_OP}, a default definition of this macro is 7741provided. The default assumes that the assembler recognizes a special 7742@samp{.} symbol as referring to the current address, and can calculate 7743the difference between this and another symbol. If your assembler does 7744not recognize @samp{.} or cannot do calculations with it, you will need 7745to redefine @code{ASM_OUTPUT_MEASURED_SIZE} to use some other technique. 7746@end defmac 7747 7748@defmac TYPE_ASM_OP 7749A C string containing the appropriate assembler directive to specify the 7750type of a symbol, without any arguments. On systems that use ELF, the 7751default (in @file{config/elfos.h}) is @samp{"\t.type\t"}; on other 7752systems, the default is not to define this macro. 7753 7754Define this macro only if it is correct to use the default definition of 7755@code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own 7756custom definition of this macro, or if you do not need explicit symbol 7757types at all, do not define this macro. 7758@end defmac 7759 7760@defmac TYPE_OPERAND_FMT 7761A C string which specifies (using @code{printf} syntax) the format of 7762the second operand to @code{TYPE_ASM_OP}. On systems that use ELF, the 7763default (in @file{config/elfos.h}) is @samp{"@@%s"}; on other systems, 7764the default is not to define this macro. 7765 7766Define this macro only if it is correct to use the default definition of 7767@code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own 7768custom definition of this macro, or if you do not need explicit symbol 7769types at all, do not define this macro. 7770@end defmac 7771 7772@defmac ASM_OUTPUT_TYPE_DIRECTIVE (@var{stream}, @var{type}) 7773A C statement (sans semicolon) to output to the stdio stream 7774@var{stream} a directive telling the assembler that the type of the 7775symbol @var{name} is @var{type}. @var{type} is a C string; currently, 7776that string is always either @samp{"function"} or @samp{"object"}, but 7777you should not count on this. 7778 7779If you define @code{TYPE_ASM_OP} and @code{TYPE_OPERAND_FMT}, a default 7780definition of this macro is provided. 7781@end defmac 7782 7783@defmac ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl}) 7784A C statement (sans semicolon) to output to the stdio stream 7785@var{stream} any text necessary for declaring the name @var{name} of a 7786function which is being defined. This macro is responsible for 7787outputting the label definition (perhaps using 7788@code{ASM_OUTPUT_FUNCTION_LABEL}). The argument @var{decl} is the 7789@code{FUNCTION_DECL} tree node representing the function. 7790 7791If this macro is not defined, then the function name is defined in the 7792usual manner as a label (by means of @code{ASM_OUTPUT_FUNCTION_LABEL}). 7793 7794You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition 7795of this macro. 7796@end defmac 7797 7798@defmac ASM_DECLARE_FUNCTION_SIZE (@var{stream}, @var{name}, @var{decl}) 7799A C statement (sans semicolon) to output to the stdio stream 7800@var{stream} any text necessary for declaring the size of a function 7801which is being defined. The argument @var{name} is the name of the 7802function. The argument @var{decl} is the @code{FUNCTION_DECL} tree node 7803representing the function. 7804 7805If this macro is not defined, then the function size is not defined. 7806 7807You may wish to use @code{ASM_OUTPUT_MEASURED_SIZE} in the definition 7808of this macro. 7809@end defmac 7810 7811@defmac ASM_DECLARE_OBJECT_NAME (@var{stream}, @var{name}, @var{decl}) 7812A C statement (sans semicolon) to output to the stdio stream 7813@var{stream} any text necessary for declaring the name @var{name} of an 7814initialized variable which is being defined. This macro must output the 7815label definition (perhaps using @code{ASM_OUTPUT_LABEL}). The argument 7816@var{decl} is the @code{VAR_DECL} tree node representing the variable. 7817 7818If this macro is not defined, then the variable name is defined in the 7819usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}). 7820 7821You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} and/or 7822@code{ASM_OUTPUT_SIZE_DIRECTIVE} in the definition of this macro. 7823@end defmac 7824 7825@hook TARGET_ASM_DECLARE_CONSTANT_NAME 7826A target hook to output to the stdio stream @var{file} any text necessary 7827for declaring the name @var{name} of a constant which is being defined. This 7828target hook is responsible for outputting the label definition (perhaps using 7829@code{assemble_label}). The argument @var{exp} is the value of the constant, 7830and @var{size} is the size of the constant in bytes. The @var{name} 7831will be an internal label. 7832 7833The default version of this target hook, define the @var{name} in the 7834usual manner as a label (by means of @code{assemble_label}). 7835 7836You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in this target hook. 7837@end deftypefn 7838 7839@defmac ASM_DECLARE_REGISTER_GLOBAL (@var{stream}, @var{decl}, @var{regno}, @var{name}) 7840A C statement (sans semicolon) to output to the stdio stream 7841@var{stream} any text necessary for claiming a register @var{regno} 7842for a global variable @var{decl} with name @var{name}. 7843 7844If you don't define this macro, that is equivalent to defining it to do 7845nothing. 7846@end defmac 7847 7848@defmac ASM_FINISH_DECLARE_OBJECT (@var{stream}, @var{decl}, @var{toplevel}, @var{atend}) 7849A C statement (sans semicolon) to finish up declaring a variable name 7850once the compiler has processed its initializer fully and thus has had a 7851chance to determine the size of an array when controlled by an 7852initializer. This is used on systems where it's necessary to declare 7853something about the size of the object. 7854 7855If you don't define this macro, that is equivalent to defining it to do 7856nothing. 7857 7858You may wish to use @code{ASM_OUTPUT_SIZE_DIRECTIVE} and/or 7859@code{ASM_OUTPUT_MEASURED_SIZE} in the definition of this macro. 7860@end defmac 7861 7862@hook TARGET_ASM_GLOBALIZE_LABEL 7863This target hook is a function to output to the stdio stream 7864@var{stream} some commands that will make the label @var{name} global; 7865that is, available for reference from other files. 7866 7867The default implementation relies on a proper definition of 7868@code{GLOBAL_ASM_OP}. 7869@end deftypefn 7870 7871@hook TARGET_ASM_GLOBALIZE_DECL_NAME 7872This target hook is a function to output to the stdio stream 7873@var{stream} some commands that will make the name associated with @var{decl} 7874global; that is, available for reference from other files. 7875 7876The default implementation uses the TARGET_ASM_GLOBALIZE_LABEL target hook. 7877@end deftypefn 7878 7879@defmac ASM_WEAKEN_LABEL (@var{stream}, @var{name}) 7880A C statement (sans semicolon) to output to the stdio stream 7881@var{stream} some commands that will make the label @var{name} weak; 7882that is, available for reference from other files but only used if 7883no other definition is available. Use the expression 7884@code{assemble_name (@var{stream}, @var{name})} to output the name 7885itself; before and after that, output the additional assembler syntax 7886for making that name weak, and a newline. 7887 7888If you don't define this macro or @code{ASM_WEAKEN_DECL}, GCC will not 7889support weak symbols and you should not define the @code{SUPPORTS_WEAK} 7890macro. 7891@end defmac 7892 7893@defmac ASM_WEAKEN_DECL (@var{stream}, @var{decl}, @var{name}, @var{value}) 7894Combines (and replaces) the function of @code{ASM_WEAKEN_LABEL} and 7895@code{ASM_OUTPUT_WEAK_ALIAS}, allowing access to the associated function 7896or variable decl. If @var{value} is not @code{NULL}, this C statement 7897should output to the stdio stream @var{stream} assembler code which 7898defines (equates) the weak symbol @var{name} to have the value 7899@var{value}. If @var{value} is @code{NULL}, it should output commands 7900to make @var{name} weak. 7901@end defmac 7902 7903@defmac ASM_OUTPUT_WEAKREF (@var{stream}, @var{decl}, @var{name}, @var{value}) 7904Outputs a directive that enables @var{name} to be used to refer to 7905symbol @var{value} with weak-symbol semantics. @code{decl} is the 7906declaration of @code{name}. 7907@end defmac 7908 7909@defmac SUPPORTS_WEAK 7910A preprocessor constant expression which evaluates to true if the target 7911supports weak symbols. 7912 7913If you don't define this macro, @file{defaults.h} provides a default 7914definition. If either @code{ASM_WEAKEN_LABEL} or @code{ASM_WEAKEN_DECL} 7915is defined, the default definition is @samp{1}; otherwise, it is @samp{0}. 7916@end defmac 7917 7918@defmac TARGET_SUPPORTS_WEAK 7919A C expression which evaluates to true if the target supports weak symbols. 7920 7921If you don't define this macro, @file{defaults.h} provides a default 7922definition. The default definition is @samp{(SUPPORTS_WEAK)}. Define 7923this macro if you want to control weak symbol support with a compiler 7924flag such as @option{-melf}. 7925@end defmac 7926 7927@defmac MAKE_DECL_ONE_ONLY (@var{decl}) 7928A C statement (sans semicolon) to mark @var{decl} to be emitted as a 7929public symbol such that extra copies in multiple translation units will 7930be discarded by the linker. Define this macro if your object file 7931format provides support for this concept, such as the @samp{COMDAT} 7932section flags in the Microsoft Windows PE/COFF format, and this support 7933requires changes to @var{decl}, such as putting it in a separate section. 7934@end defmac 7935 7936@defmac SUPPORTS_ONE_ONLY 7937A C expression which evaluates to true if the target supports one-only 7938semantics. 7939 7940If you don't define this macro, @file{varasm.c} provides a default 7941definition. If @code{MAKE_DECL_ONE_ONLY} is defined, the default 7942definition is @samp{1}; otherwise, it is @samp{0}. Define this macro if 7943you want to control one-only symbol support with a compiler flag, or if 7944setting the @code{DECL_ONE_ONLY} flag is enough to mark a declaration to 7945be emitted as one-only. 7946@end defmac 7947 7948@hook TARGET_ASM_ASSEMBLE_VISIBILITY 7949This target hook is a function to output to @var{asm_out_file} some 7950commands that will make the symbol(s) associated with @var{decl} have 7951hidden, protected or internal visibility as specified by @var{visibility}. 7952@end deftypefn 7953 7954@defmac TARGET_WEAK_NOT_IN_ARCHIVE_TOC 7955A C expression that evaluates to true if the target's linker expects 7956that weak symbols do not appear in a static archive's table of contents. 7957The default is @code{0}. 7958 7959Leaving weak symbols out of an archive's table of contents means that, 7960if a symbol will only have a definition in one translation unit and 7961will have undefined references from other translation units, that 7962symbol should not be weak. Defining this macro to be nonzero will 7963thus have the effect that certain symbols that would normally be weak 7964(explicit template instantiations, and vtables for polymorphic classes 7965with noninline key methods) will instead be nonweak. 7966 7967The C++ ABI requires this macro to be zero. Define this macro for 7968targets where full C++ ABI compliance is impossible and where linker 7969restrictions require weak symbols to be left out of a static archive's 7970table of contents. 7971@end defmac 7972 7973@defmac ASM_OUTPUT_EXTERNAL (@var{stream}, @var{decl}, @var{name}) 7974A C statement (sans semicolon) to output to the stdio stream 7975@var{stream} any text necessary for declaring the name of an external 7976symbol named @var{name} which is referenced in this compilation but 7977not defined. The value of @var{decl} is the tree node for the 7978declaration. 7979 7980This macro need not be defined if it does not need to output anything. 7981The GNU assembler and most Unix assemblers don't require anything. 7982@end defmac 7983 7984@hook TARGET_ASM_EXTERNAL_LIBCALL 7985This target hook is a function to output to @var{asm_out_file} an assembler 7986pseudo-op to declare a library function name external. The name of the 7987library function is given by @var{symref}, which is a @code{symbol_ref}. 7988@end deftypefn 7989 7990@hook TARGET_ASM_MARK_DECL_PRESERVED 7991This target hook is a function to output to @var{asm_out_file} an assembler 7992directive to annotate @var{symbol} as used. The Darwin target uses the 7993.no_dead_code_strip directive. 7994@end deftypefn 7995 7996@defmac ASM_OUTPUT_LABELREF (@var{stream}, @var{name}) 7997A C statement (sans semicolon) to output to the stdio stream 7998@var{stream} a reference in assembler syntax to a label named 7999@var{name}. This should add @samp{_} to the front of the name, if that 8000is customary on your operating system, as it is in most Berkeley Unix 8001systems. This macro is used in @code{assemble_name}. 8002@end defmac 8003 8004@hook TARGET_MANGLE_ASSEMBLER_NAME 8005 8006@defmac ASM_OUTPUT_SYMBOL_REF (@var{stream}, @var{sym}) 8007A C statement (sans semicolon) to output a reference to 8008@code{SYMBOL_REF} @var{sym}. If not defined, @code{assemble_name} 8009will be used to output the name of the symbol. This macro may be used 8010to modify the way a symbol is referenced depending on information 8011encoded by @code{TARGET_ENCODE_SECTION_INFO}. 8012@end defmac 8013 8014@defmac ASM_OUTPUT_LABEL_REF (@var{stream}, @var{buf}) 8015A C statement (sans semicolon) to output a reference to @var{buf}, the 8016result of @code{ASM_GENERATE_INTERNAL_LABEL}. If not defined, 8017@code{assemble_name} will be used to output the name of the symbol. 8018This macro is not used by @code{output_asm_label}, or the @code{%l} 8019specifier that calls it; the intention is that this macro should be set 8020when it is necessary to output a label differently when its address is 8021being taken. 8022@end defmac 8023 8024@hook TARGET_ASM_INTERNAL_LABEL 8025A function to output to the stdio stream @var{stream} a label whose 8026name is made from the string @var{prefix} and the number @var{labelno}. 8027 8028It is absolutely essential that these labels be distinct from the labels 8029used for user-level functions and variables. Otherwise, certain programs 8030will have name conflicts with internal labels. 8031 8032It is desirable to exclude internal labels from the symbol table of the 8033object file. Most assemblers have a naming convention for labels that 8034should be excluded; on many systems, the letter @samp{L} at the 8035beginning of a label has this effect. You should find out what 8036convention your system uses, and follow it. 8037 8038The default version of this function utilizes @code{ASM_GENERATE_INTERNAL_LABEL}. 8039@end deftypefn 8040 8041@defmac ASM_OUTPUT_DEBUG_LABEL (@var{stream}, @var{prefix}, @var{num}) 8042A C statement to output to the stdio stream @var{stream} a debug info 8043label whose name is made from the string @var{prefix} and the number 8044@var{num}. This is useful for VLIW targets, where debug info labels 8045may need to be treated differently than branch target labels. On some 8046systems, branch target labels must be at the beginning of instruction 8047bundles, but debug info labels can occur in the middle of instruction 8048bundles. 8049 8050If this macro is not defined, then @code{(*targetm.asm_out.internal_label)} will be 8051used. 8052@end defmac 8053 8054@defmac ASM_GENERATE_INTERNAL_LABEL (@var{string}, @var{prefix}, @var{num}) 8055A C statement to store into the string @var{string} a label whose name 8056is made from the string @var{prefix} and the number @var{num}. 8057 8058This string, when output subsequently by @code{assemble_name}, should 8059produce the output that @code{(*targetm.asm_out.internal_label)} would produce 8060with the same @var{prefix} and @var{num}. 8061 8062If the string begins with @samp{*}, then @code{assemble_name} will 8063output the rest of the string unchanged. It is often convenient for 8064@code{ASM_GENERATE_INTERNAL_LABEL} to use @samp{*} in this way. If the 8065string doesn't start with @samp{*}, then @code{ASM_OUTPUT_LABELREF} gets 8066to output the string, and may change it. (Of course, 8067@code{ASM_OUTPUT_LABELREF} is also part of your machine description, so 8068you should know what it does on your machine.) 8069@end defmac 8070 8071@defmac ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number}) 8072A C expression to assign to @var{outvar} (which is a variable of type 8073@code{char *}) a newly allocated string made from the string 8074@var{name} and the number @var{number}, with some suitable punctuation 8075added. Use @code{alloca} to get space for the string. 8076 8077The string will be used as an argument to @code{ASM_OUTPUT_LABELREF} to 8078produce an assembler label for an internal static variable whose name is 8079@var{name}. Therefore, the string must be such as to result in valid 8080assembler code. The argument @var{number} is different each time this 8081macro is executed; it prevents conflicts between similarly-named 8082internal static variables in different scopes. 8083 8084Ideally this string should not be a valid C identifier, to prevent any 8085conflict with the user's own symbols. Most assemblers allow periods 8086or percent signs in assembler symbols; putting at least one of these 8087between the name and the number will suffice. 8088 8089If this macro is not defined, a default definition will be provided 8090which is correct for most systems. 8091@end defmac 8092 8093@defmac ASM_OUTPUT_DEF (@var{stream}, @var{name}, @var{value}) 8094A C statement to output to the stdio stream @var{stream} assembler code 8095which defines (equates) the symbol @var{name} to have the value @var{value}. 8096 8097@findex SET_ASM_OP 8098If @code{SET_ASM_OP} is defined, a default definition is provided which is 8099correct for most systems. 8100@end defmac 8101 8102@defmac ASM_OUTPUT_DEF_FROM_DECLS (@var{stream}, @var{decl_of_name}, @var{decl_of_value}) 8103A C statement to output to the stdio stream @var{stream} assembler code 8104which defines (equates) the symbol whose tree node is @var{decl_of_name} 8105to have the value of the tree node @var{decl_of_value}. This macro will 8106be used in preference to @samp{ASM_OUTPUT_DEF} if it is defined and if 8107the tree nodes are available. 8108 8109@findex SET_ASM_OP 8110If @code{SET_ASM_OP} is defined, a default definition is provided which is 8111correct for most systems. 8112@end defmac 8113 8114@defmac TARGET_DEFERRED_OUTPUT_DEFS (@var{decl_of_name}, @var{decl_of_value}) 8115A C statement that evaluates to true if the assembler code which defines 8116(equates) the symbol whose tree node is @var{decl_of_name} to have the value 8117of the tree node @var{decl_of_value} should be emitted near the end of the 8118current compilation unit. The default is to not defer output of defines. 8119This macro affects defines output by @samp{ASM_OUTPUT_DEF} and 8120@samp{ASM_OUTPUT_DEF_FROM_DECLS}. 8121@end defmac 8122 8123@defmac ASM_OUTPUT_WEAK_ALIAS (@var{stream}, @var{name}, @var{value}) 8124A C statement to output to the stdio stream @var{stream} assembler code 8125which defines (equates) the weak symbol @var{name} to have the value 8126@var{value}. If @var{value} is @code{NULL}, it defines @var{name} as 8127an undefined weak symbol. 8128 8129Define this macro if the target only supports weak aliases; define 8130@code{ASM_OUTPUT_DEF} instead if possible. 8131@end defmac 8132 8133@defmac OBJC_GEN_METHOD_LABEL (@var{buf}, @var{is_inst}, @var{class_name}, @var{cat_name}, @var{sel_name}) 8134Define this macro to override the default assembler names used for 8135Objective-C methods. 8136 8137The default name is a unique method number followed by the name of the 8138class (e.g.@: @samp{_1_Foo}). For methods in categories, the name of 8139the category is also included in the assembler name (e.g.@: 8140@samp{_1_Foo_Bar}). 8141 8142These names are safe on most systems, but make debugging difficult since 8143the method's selector is not present in the name. Therefore, particular 8144systems define other ways of computing names. 8145 8146@var{buf} is an expression of type @code{char *} which gives you a 8147buffer in which to store the name; its length is as long as 8148@var{class_name}, @var{cat_name} and @var{sel_name} put together, plus 814950 characters extra. 8150 8151The argument @var{is_inst} specifies whether the method is an instance 8152method or a class method; @var{class_name} is the name of the class; 8153@var{cat_name} is the name of the category (or @code{NULL} if the method is not 8154in a category); and @var{sel_name} is the name of the selector. 8155 8156On systems where the assembler can handle quoted names, you can use this 8157macro to provide more human-readable names. 8158@end defmac 8159 8160@defmac ASM_DECLARE_CLASS_REFERENCE (@var{stream}, @var{name}) 8161A C statement (sans semicolon) to output to the stdio stream 8162@var{stream} commands to declare that the label @var{name} is an 8163Objective-C class reference. This is only needed for targets whose 8164linkers have special support for NeXT-style runtimes. 8165@end defmac 8166 8167@defmac ASM_DECLARE_UNRESOLVED_REFERENCE (@var{stream}, @var{name}) 8168A C statement (sans semicolon) to output to the stdio stream 8169@var{stream} commands to declare that the label @var{name} is an 8170unresolved Objective-C class reference. This is only needed for targets 8171whose linkers have special support for NeXT-style runtimes. 8172@end defmac 8173 8174@node Initialization 8175@subsection How Initialization Functions Are Handled 8176@cindex initialization routines 8177@cindex termination routines 8178@cindex constructors, output of 8179@cindex destructors, output of 8180 8181The compiled code for certain languages includes @dfn{constructors} 8182(also called @dfn{initialization routines})---functions to initialize 8183data in the program when the program is started. These functions need 8184to be called before the program is ``started''---that is to say, before 8185@code{main} is called. 8186 8187Compiling some languages generates @dfn{destructors} (also called 8188@dfn{termination routines}) that should be called when the program 8189terminates. 8190 8191To make the initialization and termination functions work, the compiler 8192must output something in the assembler code to cause those functions to 8193be called at the appropriate time. When you port the compiler to a new 8194system, you need to specify how to do this. 8195 8196There are two major ways that GCC currently supports the execution of 8197initialization and termination functions. Each way has two variants. 8198Much of the structure is common to all four variations. 8199 8200@findex __CTOR_LIST__ 8201@findex __DTOR_LIST__ 8202The linker must build two lists of these functions---a list of 8203initialization functions, called @code{__CTOR_LIST__}, and a list of 8204termination functions, called @code{__DTOR_LIST__}. 8205 8206Each list always begins with an ignored function pointer (which may hold 82070, @minus{}1, or a count of the function pointers after it, depending on 8208the environment). This is followed by a series of zero or more function 8209pointers to constructors (or destructors), followed by a function 8210pointer containing zero. 8211 8212Depending on the operating system and its executable file format, either 8213@file{crtstuff.c} or @file{libgcc2.c} traverses these lists at startup 8214time and exit time. Constructors are called in reverse order of the 8215list; destructors in forward order. 8216 8217The best way to handle static constructors works only for object file 8218formats which provide arbitrarily-named sections. A section is set 8219aside for a list of constructors, and another for a list of destructors. 8220Traditionally these are called @samp{.ctors} and @samp{.dtors}. Each 8221object file that defines an initialization function also puts a word in 8222the constructor section to point to that function. The linker 8223accumulates all these words into one contiguous @samp{.ctors} section. 8224Termination functions are handled similarly. 8225 8226This method will be chosen as the default by @file{target-def.h} if 8227@code{TARGET_ASM_NAMED_SECTION} is defined. A target that does not 8228support arbitrary sections, but does support special designated 8229constructor and destructor sections may define @code{CTORS_SECTION_ASM_OP} 8230and @code{DTORS_SECTION_ASM_OP} to achieve the same effect. 8231 8232When arbitrary sections are available, there are two variants, depending 8233upon how the code in @file{crtstuff.c} is called. On systems that 8234support a @dfn{.init} section which is executed at program startup, 8235parts of @file{crtstuff.c} are compiled into that section. The 8236program is linked by the @command{gcc} driver like this: 8237 8238@smallexample 8239ld -o @var{output_file} crti.o crtbegin.o @dots{} -lgcc crtend.o crtn.o 8240@end smallexample 8241 8242The prologue of a function (@code{__init}) appears in the @code{.init} 8243section of @file{crti.o}; the epilogue appears in @file{crtn.o}. Likewise 8244for the function @code{__fini} in the @dfn{.fini} section. Normally these 8245files are provided by the operating system or by the GNU C library, but 8246are provided by GCC for a few targets. 8247 8248The objects @file{crtbegin.o} and @file{crtend.o} are (for most targets) 8249compiled from @file{crtstuff.c}. They contain, among other things, code 8250fragments within the @code{.init} and @code{.fini} sections that branch 8251to routines in the @code{.text} section. The linker will pull all parts 8252of a section together, which results in a complete @code{__init} function 8253that invokes the routines we need at startup. 8254 8255To use this variant, you must define the @code{INIT_SECTION_ASM_OP} 8256macro properly. 8257 8258If no init section is available, when GCC compiles any function called 8259@code{main} (or more accurately, any function designated as a program 8260entry point by the language front end calling @code{expand_main_function}), 8261it inserts a procedure call to @code{__main} as the first executable code 8262after the function prologue. The @code{__main} function is defined 8263in @file{libgcc2.c} and runs the global constructors. 8264 8265In file formats that don't support arbitrary sections, there are again 8266two variants. In the simplest variant, the GNU linker (GNU @code{ld}) 8267and an `a.out' format must be used. In this case, 8268@code{TARGET_ASM_CONSTRUCTOR} is defined to produce a @code{.stabs} 8269entry of type @samp{N_SETT}, referencing the name @code{__CTOR_LIST__}, 8270and with the address of the void function containing the initialization 8271code as its value. The GNU linker recognizes this as a request to add 8272the value to a @dfn{set}; the values are accumulated, and are eventually 8273placed in the executable as a vector in the format described above, with 8274a leading (ignored) count and a trailing zero element. 8275@code{TARGET_ASM_DESTRUCTOR} is handled similarly. Since no init 8276section is available, the absence of @code{INIT_SECTION_ASM_OP} causes 8277the compilation of @code{main} to call @code{__main} as above, starting 8278the initialization process. 8279 8280The last variant uses neither arbitrary sections nor the GNU linker. 8281This is preferable when you want to do dynamic linking and when using 8282file formats which the GNU linker does not support, such as `ECOFF'@. In 8283this case, @code{TARGET_HAVE_CTORS_DTORS} is false, initialization and 8284termination functions are recognized simply by their names. This requires 8285an extra program in the linkage step, called @command{collect2}. This program 8286pretends to be the linker, for use with GCC; it does its job by running 8287the ordinary linker, but also arranges to include the vectors of 8288initialization and termination functions. These functions are called 8289via @code{__main} as described above. In order to use this method, 8290@code{use_collect2} must be defined in the target in @file{config.gcc}. 8291 8292@ifinfo 8293The following section describes the specific macros that control and 8294customize the handling of initialization and termination functions. 8295@end ifinfo 8296 8297@node Macros for Initialization 8298@subsection Macros Controlling Initialization Routines 8299 8300Here are the macros that control how the compiler handles initialization 8301and termination functions: 8302 8303@defmac INIT_SECTION_ASM_OP 8304If defined, a C string constant, including spacing, for the assembler 8305operation to identify the following data as initialization code. If not 8306defined, GCC will assume such a section does not exist. When you are 8307using special sections for initialization and termination functions, this 8308macro also controls how @file{crtstuff.c} and @file{libgcc2.c} arrange to 8309run the initialization functions. 8310@end defmac 8311 8312@defmac HAS_INIT_SECTION 8313If defined, @code{main} will not call @code{__main} as described above. 8314This macro should be defined for systems that control start-up code 8315on a symbol-by-symbol basis, such as OSF/1, and should not 8316be defined explicitly for systems that support @code{INIT_SECTION_ASM_OP}. 8317@end defmac 8318 8319@defmac LD_INIT_SWITCH 8320If defined, a C string constant for a switch that tells the linker that 8321the following symbol is an initialization routine. 8322@end defmac 8323 8324@defmac LD_FINI_SWITCH 8325If defined, a C string constant for a switch that tells the linker that 8326the following symbol is a finalization routine. 8327@end defmac 8328 8329@defmac COLLECT_SHARED_INIT_FUNC (@var{stream}, @var{func}) 8330If defined, a C statement that will write a function that can be 8331automatically called when a shared library is loaded. The function 8332should call @var{func}, which takes no arguments. If not defined, and 8333the object format requires an explicit initialization function, then a 8334function called @code{_GLOBAL__DI} will be generated. 8335 8336This function and the following one are used by collect2 when linking a 8337shared library that needs constructors or destructors, or has DWARF2 8338exception tables embedded in the code. 8339@end defmac 8340 8341@defmac COLLECT_SHARED_FINI_FUNC (@var{stream}, @var{func}) 8342If defined, a C statement that will write a function that can be 8343automatically called when a shared library is unloaded. The function 8344should call @var{func}, which takes no arguments. If not defined, and 8345the object format requires an explicit finalization function, then a 8346function called @code{_GLOBAL__DD} will be generated. 8347@end defmac 8348 8349@defmac INVOKE__main 8350If defined, @code{main} will call @code{__main} despite the presence of 8351@code{INIT_SECTION_ASM_OP}. This macro should be defined for systems 8352where the init section is not actually run automatically, but is still 8353useful for collecting the lists of constructors and destructors. 8354@end defmac 8355 8356@defmac SUPPORTS_INIT_PRIORITY 8357If nonzero, the C++ @code{init_priority} attribute is supported and the 8358compiler should emit instructions to control the order of initialization 8359of objects. If zero, the compiler will issue an error message upon 8360encountering an @code{init_priority} attribute. 8361@end defmac 8362 8363@hook TARGET_HAVE_CTORS_DTORS 8364This value is true if the target supports some ``native'' method of 8365collecting constructors and destructors to be run at startup and exit. 8366It is false if we must use @command{collect2}. 8367@end deftypevr 8368 8369@hook TARGET_ASM_CONSTRUCTOR 8370If defined, a function that outputs assembler code to arrange to call 8371the function referenced by @var{symbol} at initialization time. 8372 8373Assume that @var{symbol} is a @code{SYMBOL_REF} for a function taking 8374no arguments and with no return value. If the target supports initialization 8375priorities, @var{priority} is a value between 0 and @code{MAX_INIT_PRIORITY}; 8376otherwise it must be @code{DEFAULT_INIT_PRIORITY}. 8377 8378If this macro is not defined by the target, a suitable default will 8379be chosen if (1) the target supports arbitrary section names, (2) the 8380target defines @code{CTORS_SECTION_ASM_OP}, or (3) @code{USE_COLLECT2} 8381is not defined. 8382@end deftypefn 8383 8384@hook TARGET_ASM_DESTRUCTOR 8385This is like @code{TARGET_ASM_CONSTRUCTOR} but used for termination 8386functions rather than initialization functions. 8387@end deftypefn 8388 8389If @code{TARGET_HAVE_CTORS_DTORS} is true, the initialization routine 8390generated for the generated object file will have static linkage. 8391 8392If your system uses @command{collect2} as the means of processing 8393constructors, then that program normally uses @command{nm} to scan 8394an object file for constructor functions to be called. 8395 8396On certain kinds of systems, you can define this macro to make 8397@command{collect2} work faster (and, in some cases, make it work at all): 8398 8399@defmac OBJECT_FORMAT_COFF 8400Define this macro if the system uses COFF (Common Object File Format) 8401object files, so that @command{collect2} can assume this format and scan 8402object files directly for dynamic constructor/destructor functions. 8403 8404This macro is effective only in a native compiler; @command{collect2} as 8405part of a cross compiler always uses @command{nm} for the target machine. 8406@end defmac 8407 8408@defmac REAL_NM_FILE_NAME 8409Define this macro as a C string constant containing the file name to use 8410to execute @command{nm}. The default is to search the path normally for 8411@command{nm}. 8412@end defmac 8413 8414@defmac NM_FLAGS 8415@command{collect2} calls @command{nm} to scan object files for static 8416constructors and destructors and LTO info. By default, @option{-n} is 8417passed. Define @code{NM_FLAGS} to a C string constant if other options 8418are needed to get the same output format as GNU @command{nm -n} 8419produces. 8420@end defmac 8421 8422If your system supports shared libraries and has a program to list the 8423dynamic dependencies of a given library or executable, you can define 8424these macros to enable support for running initialization and 8425termination functions in shared libraries: 8426 8427@defmac LDD_SUFFIX 8428Define this macro to a C string constant containing the name of the program 8429which lists dynamic dependencies, like @command{ldd} under SunOS 4. 8430@end defmac 8431 8432@defmac PARSE_LDD_OUTPUT (@var{ptr}) 8433Define this macro to be C code that extracts filenames from the output 8434of the program denoted by @code{LDD_SUFFIX}. @var{ptr} is a variable 8435of type @code{char *} that points to the beginning of a line of output 8436from @code{LDD_SUFFIX}. If the line lists a dynamic dependency, the 8437code must advance @var{ptr} to the beginning of the filename on that 8438line. Otherwise, it must set @var{ptr} to @code{NULL}. 8439@end defmac 8440 8441@defmac SHLIB_SUFFIX 8442Define this macro to a C string constant containing the default shared 8443library extension of the target (e.g., @samp{".so"}). @command{collect2} 8444strips version information after this suffix when generating global 8445constructor and destructor names. This define is only needed on targets 8446that use @command{collect2} to process constructors and destructors. 8447@end defmac 8448 8449@node Instruction Output 8450@subsection Output of Assembler Instructions 8451 8452@c prevent bad page break with this line 8453This describes assembler instruction output. 8454 8455@defmac REGISTER_NAMES 8456A C initializer containing the assembler's names for the machine 8457registers, each one as a C string constant. This is what translates 8458register numbers in the compiler into assembler language. 8459@end defmac 8460 8461@defmac ADDITIONAL_REGISTER_NAMES 8462If defined, a C initializer for an array of structures containing a name 8463and a register number. This macro defines additional names for hard 8464registers, thus allowing the @code{asm} option in declarations to refer 8465to registers using alternate names. 8466@end defmac 8467 8468@defmac OVERLAPPING_REGISTER_NAMES 8469If defined, a C initializer for an array of structures containing a 8470name, a register number and a count of the number of consecutive 8471machine registers the name overlaps. This macro defines additional 8472names for hard registers, thus allowing the @code{asm} option in 8473declarations to refer to registers using alternate names. Unlike 8474@code{ADDITIONAL_REGISTER_NAMES}, this macro should be used when the 8475register name implies multiple underlying registers. 8476 8477This macro should be used when it is important that a clobber in an 8478@code{asm} statement clobbers all the underlying values implied by the 8479register name. For example, on ARM, clobbering the double-precision 8480VFP register ``d0'' implies clobbering both single-precision registers 8481``s0'' and ``s1''. 8482@end defmac 8483 8484@defmac ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr}) 8485Define this macro if you are using an unusual assembler that 8486requires different names for the machine instructions. 8487 8488The definition is a C statement or statements which output an 8489assembler instruction opcode to the stdio stream @var{stream}. The 8490macro-operand @var{ptr} is a variable of type @code{char *} which 8491points to the opcode name in its ``internal'' form---the form that is 8492written in the machine description. The definition should output the 8493opcode name to @var{stream}, performing any translation you desire, and 8494increment the variable @var{ptr} to point at the end of the opcode 8495so that it will not be output twice. 8496 8497In fact, your macro definition may process less than the entire opcode 8498name, or more than the opcode name; but if you want to process text 8499that includes @samp{%}-sequences to substitute operands, you must take 8500care of the substitution yourself. Just be sure to increment 8501@var{ptr} over whatever text should not be output normally. 8502 8503@findex recog_data.operand 8504If you need to look at the operand values, they can be found as the 8505elements of @code{recog_data.operand}. 8506 8507If the macro definition does nothing, the instruction is output 8508in the usual way. 8509@end defmac 8510 8511@defmac FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands}) 8512If defined, a C statement to be executed just prior to the output of 8513assembler code for @var{insn}, to modify the extracted operands so 8514they will be output differently. 8515 8516Here the argument @var{opvec} is the vector containing the operands 8517extracted from @var{insn}, and @var{noperands} is the number of 8518elements of the vector which contain meaningful data for this insn. 8519The contents of this vector are what will be used to convert the insn 8520template into assembler code, so you can change the assembler output 8521by changing the contents of the vector. 8522 8523This macro is useful when various assembler syntaxes share a single 8524file of instruction patterns; by defining this macro differently, you 8525can cause a large class of instructions to be output differently (such 8526as with rearranged operands). Naturally, variations in assembler 8527syntax affecting individual insn patterns ought to be handled by 8528writing conditional output routines in those patterns. 8529 8530If this macro is not defined, it is equivalent to a null statement. 8531@end defmac 8532 8533@hook TARGET_ASM_FINAL_POSTSCAN_INSN 8534If defined, this target hook is a function which is executed just after the 8535output of assembler code for @var{insn}, to change the mode of the assembler 8536if necessary. 8537 8538Here the argument @var{opvec} is the vector containing the operands 8539extracted from @var{insn}, and @var{noperands} is the number of 8540elements of the vector which contain meaningful data for this insn. 8541The contents of this vector are what was used to convert the insn 8542template into assembler code, so you can change the assembler mode 8543by checking the contents of the vector. 8544@end deftypefn 8545 8546@defmac PRINT_OPERAND (@var{stream}, @var{x}, @var{code}) 8547A C compound statement to output to stdio stream @var{stream} the 8548assembler syntax for an instruction operand @var{x}. @var{x} is an 8549RTL expression. 8550 8551@var{code} is a value that can be used to specify one of several ways 8552of printing the operand. It is used when identical operands must be 8553printed differently depending on the context. @var{code} comes from 8554the @samp{%} specification that was used to request printing of the 8555operand. If the specification was just @samp{%@var{digit}} then 8556@var{code} is 0; if the specification was @samp{%@var{ltr} 8557@var{digit}} then @var{code} is the ASCII code for @var{ltr}. 8558 8559@findex reg_names 8560If @var{x} is a register, this macro should print the register's name. 8561The names can be found in an array @code{reg_names} whose type is 8562@code{char *[]}. @code{reg_names} is initialized from 8563@code{REGISTER_NAMES}. 8564 8565When the machine description has a specification @samp{%@var{punct}} 8566(a @samp{%} followed by a punctuation character), this macro is called 8567with a null pointer for @var{x} and the punctuation character for 8568@var{code}. 8569@end defmac 8570 8571@defmac PRINT_OPERAND_PUNCT_VALID_P (@var{code}) 8572A C expression which evaluates to true if @var{code} is a valid 8573punctuation character for use in the @code{PRINT_OPERAND} macro. If 8574@code{PRINT_OPERAND_PUNCT_VALID_P} is not defined, it means that no 8575punctuation characters (except for the standard one, @samp{%}) are used 8576in this way. 8577@end defmac 8578 8579@defmac PRINT_OPERAND_ADDRESS (@var{stream}, @var{x}) 8580A C compound statement to output to stdio stream @var{stream} the 8581assembler syntax for an instruction operand that is a memory reference 8582whose address is @var{x}. @var{x} is an RTL expression. 8583 8584@cindex @code{TARGET_ENCODE_SECTION_INFO} usage 8585On some machines, the syntax for a symbolic address depends on the 8586section that the address refers to. On these machines, define the hook 8587@code{TARGET_ENCODE_SECTION_INFO} to store the information into the 8588@code{symbol_ref}, and then check for it here. @xref{Assembler 8589Format}. 8590@end defmac 8591 8592@findex dbr_sequence_length 8593@defmac DBR_OUTPUT_SEQEND (@var{file}) 8594A C statement, to be executed after all slot-filler instructions have 8595been output. If necessary, call @code{dbr_sequence_length} to 8596determine the number of slots filled in a sequence (zero if not 8597currently outputting a sequence), to decide how many no-ops to output, 8598or whatever. 8599 8600Don't define this macro if it has nothing to do, but it is helpful in 8601reading assembly output if the extent of the delay sequence is made 8602explicit (e.g.@: with white space). 8603@end defmac 8604 8605@findex final_sequence 8606Note that output routines for instructions with delay slots must be 8607prepared to deal with not being output as part of a sequence 8608(i.e.@: when the scheduling pass is not run, or when no slot fillers could be 8609found.) The variable @code{final_sequence} is null when not 8610processing a sequence, otherwise it contains the @code{sequence} rtx 8611being output. 8612 8613@findex asm_fprintf 8614@defmac REGISTER_PREFIX 8615@defmacx LOCAL_LABEL_PREFIX 8616@defmacx USER_LABEL_PREFIX 8617@defmacx IMMEDIATE_PREFIX 8618If defined, C string expressions to be used for the @samp{%R}, @samp{%L}, 8619@samp{%U}, and @samp{%I} options of @code{asm_fprintf} (see 8620@file{final.c}). These are useful when a single @file{md} file must 8621support multiple assembler formats. In that case, the various @file{tm.h} 8622files can define these macros differently. 8623@end defmac 8624 8625@defmac ASM_FPRINTF_EXTENSIONS (@var{file}, @var{argptr}, @var{format}) 8626If defined this macro should expand to a series of @code{case} 8627statements which will be parsed inside the @code{switch} statement of 8628the @code{asm_fprintf} function. This allows targets to define extra 8629printf formats which may useful when generating their assembler 8630statements. Note that uppercase letters are reserved for future 8631generic extensions to asm_fprintf, and so are not available to target 8632specific code. The output file is given by the parameter @var{file}. 8633The varargs input pointer is @var{argptr} and the rest of the format 8634string, starting the character after the one that is being switched 8635upon, is pointed to by @var{format}. 8636@end defmac 8637 8638@defmac ASSEMBLER_DIALECT 8639If your target supports multiple dialects of assembler language (such as 8640different opcodes), define this macro as a C expression that gives the 8641numeric index of the assembler language dialect to use, with zero as the 8642first variant. 8643 8644If this macro is defined, you may use constructs of the form 8645@smallexample 8646@samp{@{option0|option1|option2@dots{}@}} 8647@end smallexample 8648@noindent 8649in the output templates of patterns (@pxref{Output Template}) or in the 8650first argument of @code{asm_fprintf}. This construct outputs 8651@samp{option0}, @samp{option1}, @samp{option2}, etc., if the value of 8652@code{ASSEMBLER_DIALECT} is zero, one, two, etc. Any special characters 8653within these strings retain their usual meaning. If there are fewer 8654alternatives within the braces than the value of 8655@code{ASSEMBLER_DIALECT}, the construct outputs nothing. 8656 8657If you do not define this macro, the characters @samp{@{}, @samp{|} and 8658@samp{@}} do not have any special meaning when used in templates or 8659operands to @code{asm_fprintf}. 8660 8661Define the macros @code{REGISTER_PREFIX}, @code{LOCAL_LABEL_PREFIX}, 8662@code{USER_LABEL_PREFIX} and @code{IMMEDIATE_PREFIX} if you can express 8663the variations in assembler language syntax with that mechanism. Define 8664@code{ASSEMBLER_DIALECT} and use the @samp{@{option0|option1@}} syntax 8665if the syntax variant are larger and involve such things as different 8666opcodes or operand order. 8667@end defmac 8668 8669@defmac ASM_OUTPUT_REG_PUSH (@var{stream}, @var{regno}) 8670A C expression to output to @var{stream} some assembler code 8671which will push hard register number @var{regno} onto the stack. 8672The code need not be optimal, since this macro is used only when 8673profiling. 8674@end defmac 8675 8676@defmac ASM_OUTPUT_REG_POP (@var{stream}, @var{regno}) 8677A C expression to output to @var{stream} some assembler code 8678which will pop hard register number @var{regno} off of the stack. 8679The code need not be optimal, since this macro is used only when 8680profiling. 8681@end defmac 8682 8683@node Dispatch Tables 8684@subsection Output of Dispatch Tables 8685 8686@c prevent bad page break with this line 8687This concerns dispatch tables. 8688 8689@cindex dispatch table 8690@defmac ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{body}, @var{value}, @var{rel}) 8691A C statement to output to the stdio stream @var{stream} an assembler 8692pseudo-instruction to generate a difference between two labels. 8693@var{value} and @var{rel} are the numbers of two internal labels. The 8694definitions of these labels are output using 8695@code{(*targetm.asm_out.internal_label)}, and they must be printed in the same 8696way here. For example, 8697 8698@smallexample 8699fprintf (@var{stream}, "\t.word L%d-L%d\n", 8700 @var{value}, @var{rel}) 8701@end smallexample 8702 8703You must provide this macro on machines where the addresses in a 8704dispatch table are relative to the table's own address. If defined, GCC 8705will also use this macro on all machines when producing PIC@. 8706@var{body} is the body of the @code{ADDR_DIFF_VEC}; it is provided so that the 8707mode and flags can be read. 8708@end defmac 8709 8710@defmac ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value}) 8711This macro should be provided on machines where the addresses 8712in a dispatch table are absolute. 8713 8714The definition should be a C statement to output to the stdio stream 8715@var{stream} an assembler pseudo-instruction to generate a reference to 8716a label. @var{value} is the number of an internal label whose 8717definition is output using @code{(*targetm.asm_out.internal_label)}. 8718For example, 8719 8720@smallexample 8721fprintf (@var{stream}, "\t.word L%d\n", @var{value}) 8722@end smallexample 8723@end defmac 8724 8725@defmac ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table}) 8726Define this if the label before a jump-table needs to be output 8727specially. The first three arguments are the same as for 8728@code{(*targetm.asm_out.internal_label)}; the fourth argument is the 8729jump-table which follows (a @code{jump_insn} containing an 8730@code{addr_vec} or @code{addr_diff_vec}). 8731 8732This feature is used on system V to output a @code{swbeg} statement 8733for the table. 8734 8735If this macro is not defined, these labels are output with 8736@code{(*targetm.asm_out.internal_label)}. 8737@end defmac 8738 8739@defmac ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table}) 8740Define this if something special must be output at the end of a 8741jump-table. The definition should be a C statement to be executed 8742after the assembler code for the table is written. It should write 8743the appropriate code to stdio stream @var{stream}. The argument 8744@var{table} is the jump-table insn, and @var{num} is the label-number 8745of the preceding label. 8746 8747If this macro is not defined, nothing special is output at the end of 8748the jump-table. 8749@end defmac 8750 8751@hook TARGET_ASM_EMIT_UNWIND_LABEL 8752This target hook emits a label at the beginning of each FDE@. It 8753should be defined on targets where FDEs need special labels, and it 8754should write the appropriate label, for the FDE associated with the 8755function declaration @var{decl}, to the stdio stream @var{stream}. 8756The third argument, @var{for_eh}, is a boolean: true if this is for an 8757exception table. The fourth argument, @var{empty}, is a boolean: 8758true if this is a placeholder label for an omitted FDE@. 8759 8760The default is that FDEs are not given nonlocal labels. 8761@end deftypefn 8762 8763@hook TARGET_ASM_EMIT_EXCEPT_TABLE_LABEL 8764This target hook emits a label at the beginning of the exception table. 8765It should be defined on targets where it is desirable for the table 8766to be broken up according to function. 8767 8768The default is that no label is emitted. 8769@end deftypefn 8770 8771@hook TARGET_ASM_EMIT_EXCEPT_PERSONALITY 8772 8773@hook TARGET_ASM_UNWIND_EMIT 8774This target hook emits assembly directives required to unwind the 8775given instruction. This is only used when @code{TARGET_EXCEPT_UNWIND_INFO} 8776returns @code{UI_TARGET}. 8777@end deftypefn 8778 8779@hook TARGET_ASM_UNWIND_EMIT_BEFORE_INSN 8780 8781@node Exception Region Output 8782@subsection Assembler Commands for Exception Regions 8783 8784@c prevent bad page break with this line 8785 8786This describes commands marking the start and the end of an exception 8787region. 8788 8789@defmac EH_FRAME_SECTION_NAME 8790If defined, a C string constant for the name of the section containing 8791exception handling frame unwind information. If not defined, GCC will 8792provide a default definition if the target supports named sections. 8793@file{crtstuff.c} uses this macro to switch to the appropriate section. 8794 8795You should define this symbol if your target supports DWARF 2 frame 8796unwind information and the default definition does not work. 8797@end defmac 8798 8799@defmac EH_FRAME_IN_DATA_SECTION 8800If defined, DWARF 2 frame unwind information will be placed in the 8801data section even though the target supports named sections. This 8802might be necessary, for instance, if the system linker does garbage 8803collection and sections cannot be marked as not to be collected. 8804 8805Do not define this macro unless @code{TARGET_ASM_NAMED_SECTION} is 8806also defined. 8807@end defmac 8808 8809@defmac EH_TABLES_CAN_BE_READ_ONLY 8810Define this macro to 1 if your target is such that no frame unwind 8811information encoding used with non-PIC code will ever require a 8812runtime relocation, but the linker may not support merging read-only 8813and read-write sections into a single read-write section. 8814@end defmac 8815 8816@defmac MASK_RETURN_ADDR 8817An rtx used to mask the return address found via @code{RETURN_ADDR_RTX}, so 8818that it does not contain any extraneous set bits in it. 8819@end defmac 8820 8821@defmac DWARF2_UNWIND_INFO 8822Define this macro to 0 if your target supports DWARF 2 frame unwind 8823information, but it does not yet work with exception handling. 8824Otherwise, if your target supports this information (if it defines 8825@code{INCOMING_RETURN_ADDR_RTX} and either @code{UNALIGNED_INT_ASM_OP} 8826or @code{OBJECT_FORMAT_ELF}), GCC will provide a default definition of 1. 8827@end defmac 8828 8829@hook TARGET_EXCEPT_UNWIND_INFO 8830This hook defines the mechanism that will be used for exception handling 8831by the target. If the target has ABI specified unwind tables, the hook 8832should return @code{UI_TARGET}. If the target is to use the 8833@code{setjmp}/@code{longjmp}-based exception handling scheme, the hook 8834should return @code{UI_SJLJ}. If the target supports DWARF 2 frame unwind 8835information, the hook should return @code{UI_DWARF2}. 8836 8837A target may, if exceptions are disabled, choose to return @code{UI_NONE}. 8838This may end up simplifying other parts of target-specific code. The 8839default implementation of this hook never returns @code{UI_NONE}. 8840 8841Note that the value returned by this hook should be constant. It should 8842not depend on anything except the command-line switches described by 8843@var{opts}. In particular, the 8844setting @code{UI_SJLJ} must be fixed at compiler start-up as C pre-processor 8845macros and builtin functions related to exception handling are set up 8846depending on this setting. 8847 8848The default implementation of the hook first honors the 8849@option{--enable-sjlj-exceptions} configure option, then 8850@code{DWARF2_UNWIND_INFO}, and finally defaults to @code{UI_SJLJ}. If 8851@code{DWARF2_UNWIND_INFO} depends on command-line options, the target 8852must define this hook so that @var{opts} is used correctly. 8853@end deftypefn 8854 8855@hook TARGET_UNWIND_TABLES_DEFAULT 8856This variable should be set to @code{true} if the target ABI requires unwinding 8857tables even when exceptions are not used. It must not be modified by 8858command-line option processing. 8859@end deftypevr 8860 8861@defmac DONT_USE_BUILTIN_SETJMP 8862Define this macro to 1 if the @code{setjmp}/@code{longjmp}-based scheme 8863should use the @code{setjmp}/@code{longjmp} functions from the C library 8864instead of the @code{__builtin_setjmp}/@code{__builtin_longjmp} machinery. 8865@end defmac 8866 8867@defmac DWARF_CIE_DATA_ALIGNMENT 8868This macro need only be defined if the target might save registers in the 8869function prologue at an offset to the stack pointer that is not aligned to 8870@code{UNITS_PER_WORD}. The definition should be the negative minimum 8871alignment if @code{STACK_GROWS_DOWNWARD} is defined, and the positive 8872minimum alignment otherwise. @xref{SDB and DWARF}. Only applicable if 8873the target supports DWARF 2 frame unwind information. 8874@end defmac 8875 8876@hook TARGET_TERMINATE_DW2_EH_FRAME_INFO 8877Contains the value true if the target should add a zero word onto the 8878end of a Dwarf-2 frame info section when used for exception handling. 8879Default value is false if @code{EH_FRAME_SECTION_NAME} is defined, and 8880true otherwise. 8881@end deftypevr 8882 8883@hook TARGET_DWARF_REGISTER_SPAN 8884Given a register, this hook should return a parallel of registers to 8885represent where to find the register pieces. Define this hook if the 8886register and its mode are represented in Dwarf in non-contiguous 8887locations, or if the register should be represented in more than one 8888register in Dwarf. Otherwise, this hook should return @code{NULL_RTX}. 8889If not defined, the default is to return @code{NULL_RTX}. 8890@end deftypefn 8891 8892@hook TARGET_INIT_DWARF_REG_SIZES_EXTRA 8893If some registers are represented in Dwarf-2 unwind information in 8894multiple pieces, define this hook to fill in information about the 8895sizes of those pieces in the table used by the unwinder at runtime. 8896It will be called by @code{expand_builtin_init_dwarf_reg_sizes} after 8897filling in a single size corresponding to each hard register; 8898@var{address} is the address of the table. 8899@end deftypefn 8900 8901@hook TARGET_ASM_TTYPE 8902This hook is used to output a reference from a frame unwinding table to 8903the type_info object identified by @var{sym}. It should return @code{true} 8904if the reference was output. Returning @code{false} will cause the 8905reference to be output using the normal Dwarf2 routines. 8906@end deftypefn 8907 8908@hook TARGET_ARM_EABI_UNWINDER 8909This flag should be set to @code{true} on targets that use an ARM EABI 8910based unwinding library, and @code{false} on other targets. This effects 8911the format of unwinding tables, and how the unwinder in entered after 8912running a cleanup. The default is @code{false}. 8913@end deftypevr 8914 8915@node Alignment Output 8916@subsection Assembler Commands for Alignment 8917 8918@c prevent bad page break with this line 8919This describes commands for alignment. 8920 8921@defmac JUMP_ALIGN (@var{label}) 8922The alignment (log base 2) to put in front of @var{label}, which is 8923a common destination of jumps and has no fallthru incoming edge. 8924 8925This macro need not be defined if you don't want any special alignment 8926to be done at such a time. Most machine descriptions do not currently 8927define the macro. 8928 8929Unless it's necessary to inspect the @var{label} parameter, it is better 8930to set the variable @var{align_jumps} in the target's 8931@code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's 8932selection in @var{align_jumps} in a @code{JUMP_ALIGN} implementation. 8933@end defmac 8934 8935@hook TARGET_ASM_JUMP_ALIGN_MAX_SKIP 8936The maximum number of bytes to skip before @var{label} when applying 8937@code{JUMP_ALIGN}. This works only if 8938@code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined. 8939@end deftypefn 8940 8941@defmac LABEL_ALIGN_AFTER_BARRIER (@var{label}) 8942The alignment (log base 2) to put in front of @var{label}, which follows 8943a @code{BARRIER}. 8944 8945This macro need not be defined if you don't want any special alignment 8946to be done at such a time. Most machine descriptions do not currently 8947define the macro. 8948@end defmac 8949 8950@hook TARGET_ASM_LABEL_ALIGN_AFTER_BARRIER_MAX_SKIP 8951The maximum number of bytes to skip before @var{label} when applying 8952@code{LABEL_ALIGN_AFTER_BARRIER}. This works only if 8953@code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined. 8954@end deftypefn 8955 8956@defmac LOOP_ALIGN (@var{label}) 8957The alignment (log base 2) to put in front of @var{label}, which follows 8958a @code{NOTE_INSN_LOOP_BEG} note. 8959 8960This macro need not be defined if you don't want any special alignment 8961to be done at such a time. Most machine descriptions do not currently 8962define the macro. 8963 8964Unless it's necessary to inspect the @var{label} parameter, it is better 8965to set the variable @code{align_loops} in the target's 8966@code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's 8967selection in @code{align_loops} in a @code{LOOP_ALIGN} implementation. 8968@end defmac 8969 8970@hook TARGET_ASM_LOOP_ALIGN_MAX_SKIP 8971The maximum number of bytes to skip when applying @code{LOOP_ALIGN} to 8972@var{label}. This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is 8973defined. 8974@end deftypefn 8975 8976@defmac LABEL_ALIGN (@var{label}) 8977The alignment (log base 2) to put in front of @var{label}. 8978If @code{LABEL_ALIGN_AFTER_BARRIER} / @code{LOOP_ALIGN} specify a different alignment, 8979the maximum of the specified values is used. 8980 8981Unless it's necessary to inspect the @var{label} parameter, it is better 8982to set the variable @code{align_labels} in the target's 8983@code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's 8984selection in @code{align_labels} in a @code{LABEL_ALIGN} implementation. 8985@end defmac 8986 8987@hook TARGET_ASM_LABEL_ALIGN_MAX_SKIP 8988The maximum number of bytes to skip when applying @code{LABEL_ALIGN} 8989to @var{label}. This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} 8990is defined. 8991@end deftypefn 8992 8993@defmac ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes}) 8994A C statement to output to the stdio stream @var{stream} an assembler 8995instruction to advance the location counter by @var{nbytes} bytes. 8996Those bytes should be zero when loaded. @var{nbytes} will be a C 8997expression of type @code{unsigned HOST_WIDE_INT}. 8998@end defmac 8999 9000@defmac ASM_NO_SKIP_IN_TEXT 9001Define this macro if @code{ASM_OUTPUT_SKIP} should not be used in the 9002text section because it fails to put zeros in the bytes that are skipped. 9003This is true on many Unix systems, where the pseudo--op to skip bytes 9004produces no-op instructions rather than zeros when used in the text 9005section. 9006@end defmac 9007 9008@defmac ASM_OUTPUT_ALIGN (@var{stream}, @var{power}) 9009A C statement to output to the stdio stream @var{stream} an assembler 9010command to advance the location counter to a multiple of 2 to the 9011@var{power} bytes. @var{power} will be a C expression of type @code{int}. 9012@end defmac 9013 9014@defmac ASM_OUTPUT_ALIGN_WITH_NOP (@var{stream}, @var{power}) 9015Like @code{ASM_OUTPUT_ALIGN}, except that the ``nop'' instruction is used 9016for padding, if necessary. 9017@end defmac 9018 9019@defmac ASM_OUTPUT_MAX_SKIP_ALIGN (@var{stream}, @var{power}, @var{max_skip}) 9020A C statement to output to the stdio stream @var{stream} an assembler 9021command to advance the location counter to a multiple of 2 to the 9022@var{power} bytes, but only if @var{max_skip} or fewer bytes are needed to 9023satisfy the alignment request. @var{power} and @var{max_skip} will be 9024a C expression of type @code{int}. 9025@end defmac 9026 9027@need 3000 9028@node Debugging Info 9029@section Controlling Debugging Information Format 9030 9031@c prevent bad page break with this line 9032This describes how to specify debugging information. 9033 9034@menu 9035* All Debuggers:: Macros that affect all debugging formats uniformly. 9036* DBX Options:: Macros enabling specific options in DBX format. 9037* DBX Hooks:: Hook macros for varying DBX format. 9038* File Names and DBX:: Macros controlling output of file names in DBX format. 9039* SDB and DWARF:: Macros for SDB (COFF) and DWARF formats. 9040* VMS Debug:: Macros for VMS debug format. 9041@end menu 9042 9043@node All Debuggers 9044@subsection Macros Affecting All Debugging Formats 9045 9046@c prevent bad page break with this line 9047These macros affect all debugging formats. 9048 9049@defmac DBX_REGISTER_NUMBER (@var{regno}) 9050A C expression that returns the DBX register number for the compiler 9051register number @var{regno}. In the default macro provided, the value 9052of this expression will be @var{regno} itself. But sometimes there are 9053some registers that the compiler knows about and DBX does not, or vice 9054versa. In such cases, some register may need to have one number in the 9055compiler and another for DBX@. 9056 9057If two registers have consecutive numbers inside GCC, and they can be 9058used as a pair to hold a multiword value, then they @emph{must} have 9059consecutive numbers after renumbering with @code{DBX_REGISTER_NUMBER}. 9060Otherwise, debuggers will be unable to access such a pair, because they 9061expect register pairs to be consecutive in their own numbering scheme. 9062 9063If you find yourself defining @code{DBX_REGISTER_NUMBER} in way that 9064does not preserve register pairs, then what you must do instead is 9065redefine the actual register numbering scheme. 9066@end defmac 9067 9068@defmac DEBUGGER_AUTO_OFFSET (@var{x}) 9069A C expression that returns the integer offset value for an automatic 9070variable having address @var{x} (an RTL expression). The default 9071computation assumes that @var{x} is based on the frame-pointer and 9072gives the offset from the frame-pointer. This is required for targets 9073that produce debugging output for DBX or COFF-style debugging output 9074for SDB and allow the frame-pointer to be eliminated when the 9075@option{-g} options is used. 9076@end defmac 9077 9078@defmac DEBUGGER_ARG_OFFSET (@var{offset}, @var{x}) 9079A C expression that returns the integer offset value for an argument 9080having address @var{x} (an RTL expression). The nominal offset is 9081@var{offset}. 9082@end defmac 9083 9084@defmac PREFERRED_DEBUGGING_TYPE 9085A C expression that returns the type of debugging output GCC should 9086produce when the user specifies just @option{-g}. Define 9087this if you have arranged for GCC to support more than one format of 9088debugging output. Currently, the allowable values are @code{DBX_DEBUG}, 9089@code{SDB_DEBUG}, @code{DWARF_DEBUG}, @code{DWARF2_DEBUG}, 9090@code{XCOFF_DEBUG}, @code{VMS_DEBUG}, and @code{VMS_AND_DWARF2_DEBUG}. 9091 9092When the user specifies @option{-ggdb}, GCC normally also uses the 9093value of this macro to select the debugging output format, but with two 9094exceptions. If @code{DWARF2_DEBUGGING_INFO} is defined, GCC uses the 9095value @code{DWARF2_DEBUG}. Otherwise, if @code{DBX_DEBUGGING_INFO} is 9096defined, GCC uses @code{DBX_DEBUG}. 9097 9098The value of this macro only affects the default debugging output; the 9099user can always get a specific type of output by using @option{-gstabs}, 9100@option{-gcoff}, @option{-gdwarf-2}, @option{-gxcoff}, or @option{-gvms}. 9101@end defmac 9102 9103@node DBX Options 9104@subsection Specific Options for DBX Output 9105 9106@c prevent bad page break with this line 9107These are specific options for DBX output. 9108 9109@defmac DBX_DEBUGGING_INFO 9110Define this macro if GCC should produce debugging output for DBX 9111in response to the @option{-g} option. 9112@end defmac 9113 9114@defmac XCOFF_DEBUGGING_INFO 9115Define this macro if GCC should produce XCOFF format debugging output 9116in response to the @option{-g} option. This is a variant of DBX format. 9117@end defmac 9118 9119@defmac DEFAULT_GDB_EXTENSIONS 9120Define this macro to control whether GCC should by default generate 9121GDB's extended version of DBX debugging information (assuming DBX-format 9122debugging information is enabled at all). If you don't define the 9123macro, the default is 1: always generate the extended information 9124if there is any occasion to. 9125@end defmac 9126 9127@defmac DEBUG_SYMS_TEXT 9128Define this macro if all @code{.stabs} commands should be output while 9129in the text section. 9130@end defmac 9131 9132@defmac ASM_STABS_OP 9133A C string constant, including spacing, naming the assembler pseudo op to 9134use instead of @code{"\t.stabs\t"} to define an ordinary debugging symbol. 9135If you don't define this macro, @code{"\t.stabs\t"} is used. This macro 9136applies only to DBX debugging information format. 9137@end defmac 9138 9139@defmac ASM_STABD_OP 9140A C string constant, including spacing, naming the assembler pseudo op to 9141use instead of @code{"\t.stabd\t"} to define a debugging symbol whose 9142value is the current location. If you don't define this macro, 9143@code{"\t.stabd\t"} is used. This macro applies only to DBX debugging 9144information format. 9145@end defmac 9146 9147@defmac ASM_STABN_OP 9148A C string constant, including spacing, naming the assembler pseudo op to 9149use instead of @code{"\t.stabn\t"} to define a debugging symbol with no 9150name. If you don't define this macro, @code{"\t.stabn\t"} is used. This 9151macro applies only to DBX debugging information format. 9152@end defmac 9153 9154@defmac DBX_NO_XREFS 9155Define this macro if DBX on your system does not support the construct 9156@samp{xs@var{tagname}}. On some systems, this construct is used to 9157describe a forward reference to a structure named @var{tagname}. 9158On other systems, this construct is not supported at all. 9159@end defmac 9160 9161@defmac DBX_CONTIN_LENGTH 9162A symbol name in DBX-format debugging information is normally 9163continued (split into two separate @code{.stabs} directives) when it 9164exceeds a certain length (by default, 80 characters). On some 9165operating systems, DBX requires this splitting; on others, splitting 9166must not be done. You can inhibit splitting by defining this macro 9167with the value zero. You can override the default splitting-length by 9168defining this macro as an expression for the length you desire. 9169@end defmac 9170 9171@defmac DBX_CONTIN_CHAR 9172Normally continuation is indicated by adding a @samp{\} character to 9173the end of a @code{.stabs} string when a continuation follows. To use 9174a different character instead, define this macro as a character 9175constant for the character you want to use. Do not define this macro 9176if backslash is correct for your system. 9177@end defmac 9178 9179@defmac DBX_STATIC_STAB_DATA_SECTION 9180Define this macro if it is necessary to go to the data section before 9181outputting the @samp{.stabs} pseudo-op for a non-global static 9182variable. 9183@end defmac 9184 9185@defmac DBX_TYPE_DECL_STABS_CODE 9186The value to use in the ``code'' field of the @code{.stabs} directive 9187for a typedef. The default is @code{N_LSYM}. 9188@end defmac 9189 9190@defmac DBX_STATIC_CONST_VAR_CODE 9191The value to use in the ``code'' field of the @code{.stabs} directive 9192for a static variable located in the text section. DBX format does not 9193provide any ``right'' way to do this. The default is @code{N_FUN}. 9194@end defmac 9195 9196@defmac DBX_REGPARM_STABS_CODE 9197The value to use in the ``code'' field of the @code{.stabs} directive 9198for a parameter passed in registers. DBX format does not provide any 9199``right'' way to do this. The default is @code{N_RSYM}. 9200@end defmac 9201 9202@defmac DBX_REGPARM_STABS_LETTER 9203The letter to use in DBX symbol data to identify a symbol as a parameter 9204passed in registers. DBX format does not customarily provide any way to 9205do this. The default is @code{'P'}. 9206@end defmac 9207 9208@defmac DBX_FUNCTION_FIRST 9209Define this macro if the DBX information for a function and its 9210arguments should precede the assembler code for the function. Normally, 9211in DBX format, the debugging information entirely follows the assembler 9212code. 9213@end defmac 9214 9215@defmac DBX_BLOCKS_FUNCTION_RELATIVE 9216Define this macro, with value 1, if the value of a symbol describing 9217the scope of a block (@code{N_LBRAC} or @code{N_RBRAC}) should be 9218relative to the start of the enclosing function. Normally, GCC uses 9219an absolute address. 9220@end defmac 9221 9222@defmac DBX_LINES_FUNCTION_RELATIVE 9223Define this macro, with value 1, if the value of a symbol indicating 9224the current line number (@code{N_SLINE}) should be relative to the 9225start of the enclosing function. Normally, GCC uses an absolute address. 9226@end defmac 9227 9228@defmac DBX_USE_BINCL 9229Define this macro if GCC should generate @code{N_BINCL} and 9230@code{N_EINCL} stabs for included header files, as on Sun systems. This 9231macro also directs GCC to output a type number as a pair of a file 9232number and a type number within the file. Normally, GCC does not 9233generate @code{N_BINCL} or @code{N_EINCL} stabs, and it outputs a single 9234number for a type number. 9235@end defmac 9236 9237@node DBX Hooks 9238@subsection Open-Ended Hooks for DBX Format 9239 9240@c prevent bad page break with this line 9241These are hooks for DBX format. 9242 9243@defmac DBX_OUTPUT_LBRAC (@var{stream}, @var{name}) 9244Define this macro to say how to output to @var{stream} the debugging 9245information for the start of a scope level for variable names. The 9246argument @var{name} is the name of an assembler symbol (for use with 9247@code{assemble_name}) whose value is the address where the scope begins. 9248@end defmac 9249 9250@defmac DBX_OUTPUT_RBRAC (@var{stream}, @var{name}) 9251Like @code{DBX_OUTPUT_LBRAC}, but for the end of a scope level. 9252@end defmac 9253 9254@defmac DBX_OUTPUT_NFUN (@var{stream}, @var{lscope_label}, @var{decl}) 9255Define this macro if the target machine requires special handling to 9256output an @code{N_FUN} entry for the function @var{decl}. 9257@end defmac 9258 9259@defmac DBX_OUTPUT_SOURCE_LINE (@var{stream}, @var{line}, @var{counter}) 9260A C statement to output DBX debugging information before code for line 9261number @var{line} of the current source file to the stdio stream 9262@var{stream}. @var{counter} is the number of time the macro was 9263invoked, including the current invocation; it is intended to generate 9264unique labels in the assembly output. 9265 9266This macro should not be defined if the default output is correct, or 9267if it can be made correct by defining @code{DBX_LINES_FUNCTION_RELATIVE}. 9268@end defmac 9269 9270@defmac NO_DBX_FUNCTION_END 9271Some stabs encapsulation formats (in particular ECOFF), cannot handle the 9272@code{.stabs "",N_FUN,,0,0,Lscope-function-1} gdb dbx extension construct. 9273On those machines, define this macro to turn this feature off without 9274disturbing the rest of the gdb extensions. 9275@end defmac 9276 9277@defmac NO_DBX_BNSYM_ENSYM 9278Some assemblers cannot handle the @code{.stabd BNSYM/ENSYM,0,0} gdb dbx 9279extension construct. On those machines, define this macro to turn this 9280feature off without disturbing the rest of the gdb extensions. 9281@end defmac 9282 9283@node File Names and DBX 9284@subsection File Names in DBX Format 9285 9286@c prevent bad page break with this line 9287This describes file names in DBX format. 9288 9289@defmac DBX_OUTPUT_MAIN_SOURCE_FILENAME (@var{stream}, @var{name}) 9290A C statement to output DBX debugging information to the stdio stream 9291@var{stream}, which indicates that file @var{name} is the main source 9292file---the file specified as the input file for compilation. 9293This macro is called only once, at the beginning of compilation. 9294 9295This macro need not be defined if the standard form of output 9296for DBX debugging information is appropriate. 9297 9298It may be necessary to refer to a label equal to the beginning of the 9299text section. You can use @samp{assemble_name (stream, ltext_label_name)} 9300to do so. If you do this, you must also set the variable 9301@var{used_ltext_label_name} to @code{true}. 9302@end defmac 9303 9304@defmac NO_DBX_MAIN_SOURCE_DIRECTORY 9305Define this macro, with value 1, if GCC should not emit an indication 9306of the current directory for compilation and current source language at 9307the beginning of the file. 9308@end defmac 9309 9310@defmac NO_DBX_GCC_MARKER 9311Define this macro, with value 1, if GCC should not emit an indication 9312that this object file was compiled by GCC@. The default is to emit 9313an @code{N_OPT} stab at the beginning of every source file, with 9314@samp{gcc2_compiled.} for the string and value 0. 9315@end defmac 9316 9317@defmac DBX_OUTPUT_MAIN_SOURCE_FILE_END (@var{stream}, @var{name}) 9318A C statement to output DBX debugging information at the end of 9319compilation of the main source file @var{name}. Output should be 9320written to the stdio stream @var{stream}. 9321 9322If you don't define this macro, nothing special is output at the end 9323of compilation, which is correct for most machines. 9324@end defmac 9325 9326@defmac DBX_OUTPUT_NULL_N_SO_AT_MAIN_SOURCE_FILE_END 9327Define this macro @emph{instead of} defining 9328@code{DBX_OUTPUT_MAIN_SOURCE_FILE_END}, if what needs to be output at 9329the end of compilation is an @code{N_SO} stab with an empty string, 9330whose value is the highest absolute text address in the file. 9331@end defmac 9332 9333@need 2000 9334@node SDB and DWARF 9335@subsection Macros for SDB and DWARF Output 9336 9337@c prevent bad page break with this line 9338Here are macros for SDB and DWARF output. 9339 9340@defmac SDB_DEBUGGING_INFO 9341Define this macro if GCC should produce COFF-style debugging output 9342for SDB in response to the @option{-g} option. 9343@end defmac 9344 9345@defmac DWARF2_DEBUGGING_INFO 9346Define this macro if GCC should produce dwarf version 2 format 9347debugging output in response to the @option{-g} option. 9348 9349@hook TARGET_DWARF_CALLING_CONVENTION 9350Define this to enable the dwarf attribute @code{DW_AT_calling_convention} to 9351be emitted for each function. Instead of an integer return the enum 9352value for the @code{DW_CC_} tag. 9353@end deftypefn 9354 9355To support optional call frame debugging information, you must also 9356define @code{INCOMING_RETURN_ADDR_RTX} and either set 9357@code{RTX_FRAME_RELATED_P} on the prologue insns if you use RTL for the 9358prologue, or call @code{dwarf2out_def_cfa} and @code{dwarf2out_reg_save} 9359as appropriate from @code{TARGET_ASM_FUNCTION_PROLOGUE} if you don't. 9360@end defmac 9361 9362@defmac DWARF2_FRAME_INFO 9363Define this macro to a nonzero value if GCC should always output 9364Dwarf 2 frame information. If @code{TARGET_EXCEPT_UNWIND_INFO} 9365(@pxref{Exception Region Output}) returns @code{UI_DWARF2}, and 9366exceptions are enabled, GCC will output this information not matter 9367how you define @code{DWARF2_FRAME_INFO}. 9368@end defmac 9369 9370@hook TARGET_DEBUG_UNWIND_INFO 9371This hook defines the mechanism that will be used for describing frame 9372unwind information to the debugger. Normally the hook will return 9373@code{UI_DWARF2} if DWARF 2 debug information is enabled, and 9374return @code{UI_NONE} otherwise. 9375 9376A target may return @code{UI_DWARF2} even when DWARF 2 debug information 9377is disabled in order to always output DWARF 2 frame information. 9378 9379A target may return @code{UI_TARGET} if it has ABI specified unwind tables. 9380This will suppress generation of the normal debug frame unwind information. 9381@end deftypefn 9382 9383@defmac DWARF2_ASM_LINE_DEBUG_INFO 9384Define this macro to be a nonzero value if the assembler can generate Dwarf 2 9385line debug info sections. This will result in much more compact line number 9386tables, and hence is desirable if it works. 9387@end defmac 9388 9389@hook TARGET_WANT_DEBUG_PUB_SECTIONS 9390 9391@hook TARGET_DELAY_SCHED2 9392 9393@hook TARGET_DELAY_VARTRACK 9394 9395@defmac ASM_OUTPUT_DWARF_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2}) 9396A C statement to issue assembly directives that create a difference 9397@var{lab1} minus @var{lab2}, using an integer of the given @var{size}. 9398@end defmac 9399 9400@defmac ASM_OUTPUT_DWARF_VMS_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2}) 9401A C statement to issue assembly directives that create a difference 9402between the two given labels in system defined units, e.g. instruction 9403slots on IA64 VMS, using an integer of the given size. 9404@end defmac 9405 9406@defmac ASM_OUTPUT_DWARF_OFFSET (@var{stream}, @var{size}, @var{label}, @var{section}) 9407A C statement to issue assembly directives that create a 9408section-relative reference to the given @var{label}, using an integer of the 9409given @var{size}. The label is known to be defined in the given @var{section}. 9410@end defmac 9411 9412@defmac ASM_OUTPUT_DWARF_PCREL (@var{stream}, @var{size}, @var{label}) 9413A C statement to issue assembly directives that create a self-relative 9414reference to the given @var{label}, using an integer of the given @var{size}. 9415@end defmac 9416 9417@defmac ASM_OUTPUT_DWARF_TABLE_REF (@var{label}) 9418A C statement to issue assembly directives that create a reference to 9419the DWARF table identifier @var{label} from the current section. This 9420is used on some systems to avoid garbage collecting a DWARF table which 9421is referenced by a function. 9422@end defmac 9423 9424@hook TARGET_ASM_OUTPUT_DWARF_DTPREL 9425If defined, this target hook is a function which outputs a DTP-relative 9426reference to the given TLS symbol of the specified size. 9427@end deftypefn 9428 9429@defmac PUT_SDB_@dots{} 9430Define these macros to override the assembler syntax for the special 9431SDB assembler directives. See @file{sdbout.c} for a list of these 9432macros and their arguments. If the standard syntax is used, you need 9433not define them yourself. 9434@end defmac 9435 9436@defmac SDB_DELIM 9437Some assemblers do not support a semicolon as a delimiter, even between 9438SDB assembler directives. In that case, define this macro to be the 9439delimiter to use (usually @samp{\n}). It is not necessary to define 9440a new set of @code{PUT_SDB_@var{op}} macros if this is the only change 9441required. 9442@end defmac 9443 9444@defmac SDB_ALLOW_UNKNOWN_REFERENCES 9445Define this macro to allow references to unknown structure, 9446union, or enumeration tags to be emitted. Standard COFF does not 9447allow handling of unknown references, MIPS ECOFF has support for 9448it. 9449@end defmac 9450 9451@defmac SDB_ALLOW_FORWARD_REFERENCES 9452Define this macro to allow references to structure, union, or 9453enumeration tags that have not yet been seen to be handled. Some 9454assemblers choke if forward tags are used, while some require it. 9455@end defmac 9456 9457@defmac SDB_OUTPUT_SOURCE_LINE (@var{stream}, @var{line}) 9458A C statement to output SDB debugging information before code for line 9459number @var{line} of the current source file to the stdio stream 9460@var{stream}. The default is to emit an @code{.ln} directive. 9461@end defmac 9462 9463@need 2000 9464@node VMS Debug 9465@subsection Macros for VMS Debug Format 9466 9467@c prevent bad page break with this line 9468Here are macros for VMS debug format. 9469 9470@defmac VMS_DEBUGGING_INFO 9471Define this macro if GCC should produce debugging output for VMS 9472in response to the @option{-g} option. The default behavior for VMS 9473is to generate minimal debug info for a traceback in the absence of 9474@option{-g} unless explicitly overridden with @option{-g0}. This 9475behavior is controlled by @code{TARGET_OPTION_OPTIMIZATION} and 9476@code{TARGET_OPTION_OVERRIDE}. 9477@end defmac 9478 9479@node Floating Point 9480@section Cross Compilation and Floating Point 9481@cindex cross compilation and floating point 9482@cindex floating point and cross compilation 9483 9484While all modern machines use twos-complement representation for integers, 9485there are a variety of representations for floating point numbers. This 9486means that in a cross-compiler the representation of floating point numbers 9487in the compiled program may be different from that used in the machine 9488doing the compilation. 9489 9490Because different representation systems may offer different amounts of 9491range and precision, all floating point constants must be represented in 9492the target machine's format. Therefore, the cross compiler cannot 9493safely use the host machine's floating point arithmetic; it must emulate 9494the target's arithmetic. To ensure consistency, GCC always uses 9495emulation to work with floating point values, even when the host and 9496target floating point formats are identical. 9497 9498The following macros are provided by @file{real.h} for the compiler to 9499use. All parts of the compiler which generate or optimize 9500floating-point calculations must use these macros. They may evaluate 9501their operands more than once, so operands must not have side effects. 9502 9503@defmac REAL_VALUE_TYPE 9504The C data type to be used to hold a floating point value in the target 9505machine's format. Typically this is a @code{struct} containing an 9506array of @code{HOST_WIDE_INT}, but all code should treat it as an opaque 9507quantity. 9508@end defmac 9509 9510@deftypefn Macro int REAL_VALUES_EQUAL (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y}) 9511Compares for equality the two values, @var{x} and @var{y}. If the target 9512floating point format supports negative zeroes and/or NaNs, 9513@samp{REAL_VALUES_EQUAL (-0.0, 0.0)} is true, and 9514@samp{REAL_VALUES_EQUAL (NaN, NaN)} is false. 9515@end deftypefn 9516 9517@deftypefn Macro int REAL_VALUES_LESS (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y}) 9518Tests whether @var{x} is less than @var{y}. 9519@end deftypefn 9520 9521@deftypefn Macro HOST_WIDE_INT REAL_VALUE_FIX (REAL_VALUE_TYPE @var{x}) 9522Truncates @var{x} to a signed integer, rounding toward zero. 9523@end deftypefn 9524 9525@deftypefn Macro {unsigned HOST_WIDE_INT} REAL_VALUE_UNSIGNED_FIX (REAL_VALUE_TYPE @var{x}) 9526Truncates @var{x} to an unsigned integer, rounding toward zero. If 9527@var{x} is negative, returns zero. 9528@end deftypefn 9529 9530@deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ATOF (const char *@var{string}, enum machine_mode @var{mode}) 9531Converts @var{string} into a floating point number in the target machine's 9532representation for mode @var{mode}. This routine can handle both 9533decimal and hexadecimal floating point constants, using the syntax 9534defined by the C language for both. 9535@end deftypefn 9536 9537@deftypefn Macro int REAL_VALUE_NEGATIVE (REAL_VALUE_TYPE @var{x}) 9538Returns 1 if @var{x} is negative (including negative zero), 0 otherwise. 9539@end deftypefn 9540 9541@deftypefn Macro int REAL_VALUE_ISINF (REAL_VALUE_TYPE @var{x}) 9542Determines whether @var{x} represents infinity (positive or negative). 9543@end deftypefn 9544 9545@deftypefn Macro int REAL_VALUE_ISNAN (REAL_VALUE_TYPE @var{x}) 9546Determines whether @var{x} represents a ``NaN'' (not-a-number). 9547@end deftypefn 9548 9549@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}) 9550Calculates an arithmetic operation on the two floating point values 9551@var{x} and @var{y}, storing the result in @var{output} (which must be a 9552variable). 9553 9554The operation to be performed is specified by @var{code}. Only the 9555following codes are supported: @code{PLUS_EXPR}, @code{MINUS_EXPR}, 9556@code{MULT_EXPR}, @code{RDIV_EXPR}, @code{MAX_EXPR}, @code{MIN_EXPR}. 9557 9558If @code{REAL_ARITHMETIC} is asked to evaluate division by zero and the 9559target's floating point format cannot represent infinity, it will call 9560@code{abort}. Callers should check for this situation first, using 9561@code{MODE_HAS_INFINITIES}. @xref{Storage Layout}. 9562@end deftypefn 9563 9564@deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_NEGATE (REAL_VALUE_TYPE @var{x}) 9565Returns the negative of the floating point value @var{x}. 9566@end deftypefn 9567 9568@deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ABS (REAL_VALUE_TYPE @var{x}) 9569Returns the absolute value of @var{x}. 9570@end deftypefn 9571 9572@deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_TRUNCATE (REAL_VALUE_TYPE @var{mode}, enum machine_mode @var{x}) 9573Truncates the floating point value @var{x} to fit in @var{mode}. The 9574return value is still a full-size @code{REAL_VALUE_TYPE}, but it has an 9575appropriate bit pattern to be output as a floating constant whose 9576precision accords with mode @var{mode}. 9577@end deftypefn 9578 9579@deftypefn Macro void REAL_VALUE_TO_INT (HOST_WIDE_INT @var{low}, HOST_WIDE_INT @var{high}, REAL_VALUE_TYPE @var{x}) 9580Converts a floating point value @var{x} into a double-precision integer 9581which is then stored into @var{low} and @var{high}. If the value is not 9582integral, it is truncated. 9583@end deftypefn 9584 9585@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}) 9586Converts a double-precision integer found in @var{low} and @var{high}, 9587into a floating point value which is then stored into @var{x}. The 9588value is truncated to fit in mode @var{mode}. 9589@end deftypefn 9590 9591@node Mode Switching 9592@section Mode Switching Instructions 9593@cindex mode switching 9594The following macros control mode switching optimizations: 9595 9596@defmac OPTIMIZE_MODE_SWITCHING (@var{entity}) 9597Define this macro if the port needs extra instructions inserted for mode 9598switching in an optimizing compilation. 9599 9600For an example, the SH4 can perform both single and double precision 9601floating point operations, but to perform a single precision operation, 9602the FPSCR PR bit has to be cleared, while for a double precision 9603operation, this bit has to be set. Changing the PR bit requires a general 9604purpose register as a scratch register, hence these FPSCR sets have to 9605be inserted before reload, i.e.@: you can't put this into instruction emitting 9606or @code{TARGET_MACHINE_DEPENDENT_REORG}. 9607 9608You can have multiple entities that are mode-switched, and select at run time 9609which entities actually need it. @code{OPTIMIZE_MODE_SWITCHING} should 9610return nonzero for any @var{entity} that needs mode-switching. 9611If you define this macro, you also have to define 9612@code{NUM_MODES_FOR_MODE_SWITCHING}, @code{MODE_NEEDED}, 9613@code{MODE_PRIORITY_TO_MODE} and @code{EMIT_MODE_SET}. 9614@code{MODE_AFTER}, @code{MODE_ENTRY}, and @code{MODE_EXIT} 9615are optional. 9616@end defmac 9617 9618@defmac NUM_MODES_FOR_MODE_SWITCHING 9619If you define @code{OPTIMIZE_MODE_SWITCHING}, you have to define this as 9620initializer for an array of integers. Each initializer element 9621N refers to an entity that needs mode switching, and specifies the number 9622of different modes that might need to be set for this entity. 9623The position of the initializer in the initializer---starting counting at 9624zero---determines the integer that is used to refer to the mode-switched 9625entity in question. 9626In macros that take mode arguments / yield a mode result, modes are 9627represented as numbers 0 @dots{} N @minus{} 1. N is used to specify that no mode 9628switch is needed / supplied. 9629@end defmac 9630 9631@defmac MODE_NEEDED (@var{entity}, @var{insn}) 9632@var{entity} is an integer specifying a mode-switched entity. If 9633@code{OPTIMIZE_MODE_SWITCHING} is defined, you must define this macro to 9634return an integer value not larger than the corresponding element in 9635@code{NUM_MODES_FOR_MODE_SWITCHING}, to denote the mode that @var{entity} must 9636be switched into prior to the execution of @var{insn}. 9637@end defmac 9638 9639@defmac MODE_AFTER (@var{mode}, @var{insn}) 9640If this macro is defined, it is evaluated for every @var{insn} during 9641mode switching. It determines the mode that an insn results in (if 9642different from the incoming mode). 9643@end defmac 9644 9645@defmac MODE_ENTRY (@var{entity}) 9646If this macro is defined, it is evaluated for every @var{entity} that needs 9647mode switching. It should evaluate to an integer, which is a mode that 9648@var{entity} is assumed to be switched to at function entry. If @code{MODE_ENTRY} 9649is defined then @code{MODE_EXIT} must be defined. 9650@end defmac 9651 9652@defmac MODE_EXIT (@var{entity}) 9653If this macro is defined, it is evaluated for every @var{entity} that needs 9654mode switching. It should evaluate to an integer, which is a mode that 9655@var{entity} is assumed to be switched to at function exit. If @code{MODE_EXIT} 9656is defined then @code{MODE_ENTRY} must be defined. 9657@end defmac 9658 9659@defmac MODE_PRIORITY_TO_MODE (@var{entity}, @var{n}) 9660This macro specifies the order in which modes for @var{entity} are processed. 96610 is the highest priority, @code{NUM_MODES_FOR_MODE_SWITCHING[@var{entity}] - 1} the 9662lowest. The value of the macro should be an integer designating a mode 9663for @var{entity}. For any fixed @var{entity}, @code{mode_priority_to_mode} 9664(@var{entity}, @var{n}) shall be a bijection in 0 @dots{} 9665@code{num_modes_for_mode_switching[@var{entity}] - 1}. 9666@end defmac 9667 9668@defmac EMIT_MODE_SET (@var{entity}, @var{mode}, @var{hard_regs_live}) 9669Generate one or more insns to set @var{entity} to @var{mode}. 9670@var{hard_reg_live} is the set of hard registers live at the point where 9671the insn(s) are to be inserted. 9672@end defmac 9673 9674@node Target Attributes 9675@section Defining target-specific uses of @code{__attribute__} 9676@cindex target attributes 9677@cindex machine attributes 9678@cindex attributes, target-specific 9679 9680Target-specific attributes may be defined for functions, data and types. 9681These are described using the following target hooks; they also need to 9682be documented in @file{extend.texi}. 9683 9684@hook TARGET_ATTRIBUTE_TABLE 9685If defined, this target hook points to an array of @samp{struct 9686attribute_spec} (defined in @file{tree.h}) specifying the machine 9687specific attributes for this target and some of the restrictions on the 9688entities to which these attributes are applied and the arguments they 9689take. 9690@end deftypevr 9691 9692@hook TARGET_ATTRIBUTE_TAKES_IDENTIFIER_P 9693If defined, this target hook is a function which returns true if the 9694machine-specific attribute named @var{name} expects an identifier 9695given as its first argument to be passed on as a plain identifier, not 9696subjected to name lookup. If this is not defined, the default is 9697false for all machine-specific attributes. 9698@end deftypefn 9699 9700@hook TARGET_COMP_TYPE_ATTRIBUTES 9701If defined, this target hook is a function which returns zero if the attributes on 9702@var{type1} and @var{type2} are incompatible, one if they are compatible, 9703and two if they are nearly compatible (which causes a warning to be 9704generated). If this is not defined, machine-specific attributes are 9705supposed always to be compatible. 9706@end deftypefn 9707 9708@hook TARGET_SET_DEFAULT_TYPE_ATTRIBUTES 9709If defined, this target hook is a function which assigns default attributes to 9710the newly defined @var{type}. 9711@end deftypefn 9712 9713@hook TARGET_MERGE_TYPE_ATTRIBUTES 9714Define this target hook if the merging of type attributes needs special 9715handling. If defined, the result is a list of the combined 9716@code{TYPE_ATTRIBUTES} of @var{type1} and @var{type2}. It is assumed 9717that @code{comptypes} has already been called and returned 1. This 9718function may call @code{merge_attributes} to handle machine-independent 9719merging. 9720@end deftypefn 9721 9722@hook TARGET_MERGE_DECL_ATTRIBUTES 9723Define this target hook if the merging of decl attributes needs special 9724handling. If defined, the result is a list of the combined 9725@code{DECL_ATTRIBUTES} of @var{olddecl} and @var{newdecl}. 9726@var{newdecl} is a duplicate declaration of @var{olddecl}. Examples of 9727when this is needed are when one attribute overrides another, or when an 9728attribute is nullified by a subsequent definition. This function may 9729call @code{merge_attributes} to handle machine-independent merging. 9730 9731@findex TARGET_DLLIMPORT_DECL_ATTRIBUTES 9732If the only target-specific handling you require is @samp{dllimport} 9733for Microsoft Windows targets, you should define the macro 9734@code{TARGET_DLLIMPORT_DECL_ATTRIBUTES} to @code{1}. The compiler 9735will then define a function called 9736@code{merge_dllimport_decl_attributes} which can then be defined as 9737the expansion of @code{TARGET_MERGE_DECL_ATTRIBUTES}. You can also 9738add @code{handle_dll_attribute} in the attribute table for your port 9739to perform initial processing of the @samp{dllimport} and 9740@samp{dllexport} attributes. This is done in @file{i386/cygwin.h} and 9741@file{i386/i386.c}, for example. 9742@end deftypefn 9743 9744@hook TARGET_VALID_DLLIMPORT_ATTRIBUTE_P 9745 9746@defmac TARGET_DECLSPEC 9747Define this macro to a nonzero value if you want to treat 9748@code{__declspec(X)} as equivalent to @code{__attribute((X))}. By 9749default, this behavior is enabled only for targets that define 9750@code{TARGET_DLLIMPORT_DECL_ATTRIBUTES}. The current implementation 9751of @code{__declspec} is via a built-in macro, but you should not rely 9752on this implementation detail. 9753@end defmac 9754 9755@hook TARGET_INSERT_ATTRIBUTES 9756Define this target hook if you want to be able to add attributes to a decl 9757when it is being created. This is normally useful for back ends which 9758wish to implement a pragma by using the attributes which correspond to 9759the pragma's effect. The @var{node} argument is the decl which is being 9760created. The @var{attr_ptr} argument is a pointer to the attribute list 9761for this decl. The list itself should not be modified, since it may be 9762shared with other decls, but attributes may be chained on the head of 9763the list and @code{*@var{attr_ptr}} modified to point to the new 9764attributes, or a copy of the list may be made if further changes are 9765needed. 9766@end deftypefn 9767 9768@hook TARGET_FUNCTION_ATTRIBUTE_INLINABLE_P 9769@cindex inlining 9770This target hook returns @code{true} if it is ok to inline @var{fndecl} 9771into the current function, despite its having target-specific 9772attributes, @code{false} otherwise. By default, if a function has a 9773target specific attribute attached to it, it will not be inlined. 9774@end deftypefn 9775 9776@hook TARGET_OPTION_VALID_ATTRIBUTE_P 9777This hook is called to parse the @code{attribute(option("..."))}, and 9778it allows the function to set different target machine compile time 9779options for the current function that might be different than the 9780options specified on the command line. The hook should return 9781@code{true} if the options are valid. 9782 9783The hook should set the @var{DECL_FUNCTION_SPECIFIC_TARGET} field in 9784the function declaration to hold a pointer to a target specific 9785@var{struct cl_target_option} structure. 9786@end deftypefn 9787 9788@hook TARGET_OPTION_SAVE 9789This hook is called to save any additional target specific information 9790in the @var{struct cl_target_option} structure for function specific 9791options. 9792@xref{Option file format}. 9793@end deftypefn 9794 9795@hook TARGET_OPTION_RESTORE 9796This hook is called to restore any additional target specific 9797information in the @var{struct cl_target_option} structure for 9798function specific options. 9799@end deftypefn 9800 9801@hook TARGET_OPTION_PRINT 9802This hook is called to print any additional target specific 9803information in the @var{struct cl_target_option} structure for 9804function specific options. 9805@end deftypefn 9806 9807@hook TARGET_OPTION_PRAGMA_PARSE 9808This target hook parses the options for @code{#pragma GCC option} to 9809set the machine specific options for functions that occur later in the 9810input stream. The options should be the same as handled by the 9811@code{TARGET_OPTION_VALID_ATTRIBUTE_P} hook. 9812@end deftypefn 9813 9814@hook TARGET_OPTION_OVERRIDE 9815Sometimes certain combinations of command options do not make sense on 9816a particular target machine. You can override the hook 9817@code{TARGET_OPTION_OVERRIDE} to take account of this. This hooks is called 9818once just after all the command options have been parsed. 9819 9820Don't use this hook to turn on various extra optimizations for 9821@option{-O}. That is what @code{TARGET_OPTION_OPTIMIZATION} is for. 9822 9823If you need to do something whenever the optimization level is 9824changed via the optimize attribute or pragma, see 9825@code{TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE} 9826@end deftypefn 9827 9828@hook TARGET_CAN_INLINE_P 9829This target hook returns @code{false} if the @var{caller} function 9830cannot inline @var{callee}, based on target specific information. By 9831default, inlining is not allowed if the callee function has function 9832specific target options and the caller does not use the same options. 9833@end deftypefn 9834 9835@node Emulated TLS 9836@section Emulating TLS 9837@cindex Emulated TLS 9838 9839For targets whose psABI does not provide Thread Local Storage via 9840specific relocations and instruction sequences, an emulation layer is 9841used. A set of target hooks allows this emulation layer to be 9842configured for the requirements of a particular target. For instance 9843the psABI may in fact specify TLS support in terms of an emulation 9844layer. 9845 9846The emulation layer works by creating a control object for every TLS 9847object. To access the TLS object, a lookup function is provided 9848which, when given the address of the control object, will return the 9849address of the current thread's instance of the TLS object. 9850 9851@hook TARGET_EMUTLS_GET_ADDRESS 9852Contains the name of the helper function that uses a TLS control 9853object to locate a TLS instance. The default causes libgcc's 9854emulated TLS helper function to be used. 9855@end deftypevr 9856 9857@hook TARGET_EMUTLS_REGISTER_COMMON 9858Contains the name of the helper function that should be used at 9859program startup to register TLS objects that are implicitly 9860initialized to zero. If this is @code{NULL}, all TLS objects will 9861have explicit initializers. The default causes libgcc's emulated TLS 9862registration function to be used. 9863@end deftypevr 9864 9865@hook TARGET_EMUTLS_VAR_SECTION 9866Contains the name of the section in which TLS control variables should 9867be placed. The default of @code{NULL} allows these to be placed in 9868any section. 9869@end deftypevr 9870 9871@hook TARGET_EMUTLS_TMPL_SECTION 9872Contains the name of the section in which TLS initializers should be 9873placed. The default of @code{NULL} allows these to be placed in any 9874section. 9875@end deftypevr 9876 9877@hook TARGET_EMUTLS_VAR_PREFIX 9878Contains the prefix to be prepended to TLS control variable names. 9879The default of @code{NULL} uses a target-specific prefix. 9880@end deftypevr 9881 9882@hook TARGET_EMUTLS_TMPL_PREFIX 9883Contains the prefix to be prepended to TLS initializer objects. The 9884default of @code{NULL} uses a target-specific prefix. 9885@end deftypevr 9886 9887@hook TARGET_EMUTLS_VAR_FIELDS 9888Specifies a function that generates the FIELD_DECLs for a TLS control 9889object type. @var{type} is the RECORD_TYPE the fields are for and 9890@var{name} should be filled with the structure tag, if the default of 9891@code{__emutls_object} is unsuitable. The default creates a type suitable 9892for libgcc's emulated TLS function. 9893@end deftypefn 9894 9895@hook TARGET_EMUTLS_VAR_INIT 9896Specifies a function that generates the CONSTRUCTOR to initialize a 9897TLS control object. @var{var} is the TLS control object, @var{decl} 9898is the TLS object and @var{tmpl_addr} is the address of the 9899initializer. The default initializes libgcc's emulated TLS control object. 9900@end deftypefn 9901 9902@hook TARGET_EMUTLS_VAR_ALIGN_FIXED 9903Specifies whether the alignment of TLS control variable objects is 9904fixed and should not be increased as some backends may do to optimize 9905single objects. The default is false. 9906@end deftypevr 9907 9908@hook TARGET_EMUTLS_DEBUG_FORM_TLS_ADDRESS 9909Specifies whether a DWARF @code{DW_OP_form_tls_address} location descriptor 9910may be used to describe emulated TLS control objects. 9911@end deftypevr 9912 9913@node MIPS Coprocessors 9914@section Defining coprocessor specifics for MIPS targets. 9915@cindex MIPS coprocessor-definition macros 9916 9917The MIPS specification allows MIPS implementations to have as many as 4 9918coprocessors, each with as many as 32 private registers. GCC supports 9919accessing these registers and transferring values between the registers 9920and memory using asm-ized variables. For example: 9921 9922@smallexample 9923 register unsigned int cp0count asm ("c0r1"); 9924 unsigned int d; 9925 9926 d = cp0count + 3; 9927@end smallexample 9928 9929(``c0r1'' is the default name of register 1 in coprocessor 0; alternate 9930names may be added as described below, or the default names may be 9931overridden entirely in @code{SUBTARGET_CONDITIONAL_REGISTER_USAGE}.) 9932 9933Coprocessor registers are assumed to be epilogue-used; sets to them will 9934be preserved even if it does not appear that the register is used again 9935later in the function. 9936 9937Another note: according to the MIPS spec, coprocessor 1 (if present) is 9938the FPU@. One accesses COP1 registers through standard mips 9939floating-point support; they are not included in this mechanism. 9940 9941There is one macro used in defining the MIPS coprocessor interface which 9942you may want to override in subtargets; it is described below. 9943 9944@defmac ALL_COP_ADDITIONAL_REGISTER_NAMES 9945A comma-separated list (with leading comma) of pairs describing the 9946alternate names of coprocessor registers. The format of each entry should be 9947@smallexample 9948@{ @var{alternatename}, @var{register_number}@} 9949@end smallexample 9950Default: empty. 9951@end defmac 9952 9953@node PCH Target 9954@section Parameters for Precompiled Header Validity Checking 9955@cindex parameters, precompiled headers 9956 9957@hook TARGET_GET_PCH_VALIDITY 9958This hook returns a pointer to the data needed by 9959@code{TARGET_PCH_VALID_P} and sets 9960@samp{*@var{sz}} to the size of the data in bytes. 9961@end deftypefn 9962 9963@hook TARGET_PCH_VALID_P 9964This hook checks whether the options used to create a PCH file are 9965compatible with the current settings. It returns @code{NULL} 9966if so and a suitable error message if not. Error messages will 9967be presented to the user and must be localized using @samp{_(@var{msg})}. 9968 9969@var{data} is the data that was returned by @code{TARGET_GET_PCH_VALIDITY} 9970when the PCH file was created and @var{sz} is the size of that data in bytes. 9971It's safe to assume that the data was created by the same version of the 9972compiler, so no format checking is needed. 9973 9974The default definition of @code{default_pch_valid_p} should be 9975suitable for most targets. 9976@end deftypefn 9977 9978@hook TARGET_CHECK_PCH_TARGET_FLAGS 9979If this hook is nonnull, the default implementation of 9980@code{TARGET_PCH_VALID_P} will use it to check for compatible values 9981of @code{target_flags}. @var{pch_flags} specifies the value that 9982@code{target_flags} had when the PCH file was created. The return 9983value is the same as for @code{TARGET_PCH_VALID_P}. 9984@end deftypefn 9985 9986@hook TARGET_PREPARE_PCH_SAVE 9987 9988@node C++ ABI 9989@section C++ ABI parameters 9990@cindex parameters, c++ abi 9991 9992@hook TARGET_CXX_GUARD_TYPE 9993Define this hook to override the integer type used for guard variables. 9994These are used to implement one-time construction of static objects. The 9995default is long_long_integer_type_node. 9996@end deftypefn 9997 9998@hook TARGET_CXX_GUARD_MASK_BIT 9999This hook determines how guard variables are used. It should return 10000@code{false} (the default) if the first byte should be used. A return value of 10001@code{true} indicates that only the least significant bit should be used. 10002@end deftypefn 10003 10004@hook TARGET_CXX_GET_COOKIE_SIZE 10005This hook returns the size of the cookie to use when allocating an array 10006whose elements have the indicated @var{type}. Assumes that it is already 10007known that a cookie is needed. The default is 10008@code{max(sizeof (size_t), alignof(type))}, as defined in section 2.7 of the 10009IA64/Generic C++ ABI@. 10010@end deftypefn 10011 10012@hook TARGET_CXX_COOKIE_HAS_SIZE 10013This hook should return @code{true} if the element size should be stored in 10014array cookies. The default is to return @code{false}. 10015@end deftypefn 10016 10017@hook TARGET_CXX_IMPORT_EXPORT_CLASS 10018If defined by a backend this hook allows the decision made to export 10019class @var{type} to be overruled. Upon entry @var{import_export} 10020will contain 1 if the class is going to be exported, @minus{}1 if it is going 10021to be imported and 0 otherwise. This function should return the 10022modified value and perform any other actions necessary to support the 10023backend's targeted operating system. 10024@end deftypefn 10025 10026@hook TARGET_CXX_CDTOR_RETURNS_THIS 10027This hook should return @code{true} if constructors and destructors return 10028the address of the object created/destroyed. The default is to return 10029@code{false}. 10030@end deftypefn 10031 10032@hook TARGET_CXX_KEY_METHOD_MAY_BE_INLINE 10033This hook returns true if the key method for a class (i.e., the method 10034which, if defined in the current translation unit, causes the virtual 10035table to be emitted) may be an inline function. Under the standard 10036Itanium C++ ABI the key method may be an inline function so long as 10037the function is not declared inline in the class definition. Under 10038some variants of the ABI, an inline function can never be the key 10039method. The default is to return @code{true}. 10040@end deftypefn 10041 10042@hook TARGET_CXX_DETERMINE_CLASS_DATA_VISIBILITY 10043 10044@hook TARGET_CXX_CLASS_DATA_ALWAYS_COMDAT 10045This hook returns true (the default) if virtual tables and other 10046similar implicit class data objects are always COMDAT if they have 10047external linkage. If this hook returns false, then class data for 10048classes whose virtual table will be emitted in only one translation 10049unit will not be COMDAT. 10050@end deftypefn 10051 10052@hook TARGET_CXX_LIBRARY_RTTI_COMDAT 10053This hook returns true (the default) if the RTTI information for 10054the basic types which is defined in the C++ runtime should always 10055be COMDAT, false if it should not be COMDAT. 10056@end deftypefn 10057 10058@hook TARGET_CXX_USE_AEABI_ATEXIT 10059This hook returns true if @code{__aeabi_atexit} (as defined by the ARM EABI) 10060should be used to register static destructors when @option{-fuse-cxa-atexit} 10061is in effect. The default is to return false to use @code{__cxa_atexit}. 10062@end deftypefn 10063 10064@hook TARGET_CXX_USE_ATEXIT_FOR_CXA_ATEXIT 10065This hook returns true if the target @code{atexit} function can be used 10066in the same manner as @code{__cxa_atexit} to register C++ static 10067destructors. This requires that @code{atexit}-registered functions in 10068shared libraries are run in the correct order when the libraries are 10069unloaded. The default is to return false. 10070@end deftypefn 10071 10072@hook TARGET_CXX_ADJUST_CLASS_AT_DEFINITION 10073 10074@node Named Address Spaces 10075@section Adding support for named address spaces 10076@cindex named address spaces 10077 10078The draft technical report of the ISO/IEC JTC1 S22 WG14 N1275 10079standards committee, @cite{Programming Languages - C - Extensions to 10080support embedded processors}, specifies a syntax for embedded 10081processors to specify alternate address spaces. You can configure a 10082GCC port to support section 5.1 of the draft report to add support for 10083address spaces other than the default address space. These address 10084spaces are new keywords that are similar to the @code{volatile} and 10085@code{const} type attributes. 10086 10087Pointers to named address spaces can have a different size than 10088pointers to the generic address space. 10089 10090For example, the SPU port uses the @code{__ea} address space to refer 10091to memory in the host processor, rather than memory local to the SPU 10092processor. Access to memory in the @code{__ea} address space involves 10093issuing DMA operations to move data between the host processor and the 10094local processor memory address space. Pointers in the @code{__ea} 10095address space are either 32 bits or 64 bits based on the 10096@option{-mea32} or @option{-mea64} switches (native SPU pointers are 10097always 32 bits). 10098 10099Internally, address spaces are represented as a small integer in the 10100range 0 to 15 with address space 0 being reserved for the generic 10101address space. 10102 10103To register a named address space qualifier keyword with the C front end, 10104the target may call the @code{c_register_addr_space} routine. For example, 10105the SPU port uses the following to declare @code{__ea} as the keyword for 10106named address space #1: 10107@smallexample 10108#define ADDR_SPACE_EA 1 10109c_register_addr_space ("__ea", ADDR_SPACE_EA); 10110@end smallexample 10111 10112@hook TARGET_ADDR_SPACE_POINTER_MODE 10113Define this to return the machine mode to use for pointers to 10114@var{address_space} if the target supports named address spaces. 10115The default version of this hook returns @code{ptr_mode} for the 10116generic address space only. 10117@end deftypefn 10118 10119@hook TARGET_ADDR_SPACE_ADDRESS_MODE 10120Define this to return the machine mode to use for addresses in 10121@var{address_space} if the target supports named address spaces. 10122The default version of this hook returns @code{Pmode} for the 10123generic address space only. 10124@end deftypefn 10125 10126@hook TARGET_ADDR_SPACE_VALID_POINTER_MODE 10127Define this to return nonzero if the port can handle pointers 10128with machine mode @var{mode} to address space @var{as}. This target 10129hook is the same as the @code{TARGET_VALID_POINTER_MODE} target hook, 10130except that it includes explicit named address space support. The default 10131version of this hook returns true for the modes returned by either the 10132@code{TARGET_ADDR_SPACE_POINTER_MODE} or @code{TARGET_ADDR_SPACE_ADDRESS_MODE} 10133target hooks for the given address space. 10134@end deftypefn 10135 10136@hook TARGET_ADDR_SPACE_LEGITIMATE_ADDRESS_P 10137Define this to return true if @var{exp} is a valid address for mode 10138@var{mode} in the named address space @var{as}. The @var{strict} 10139parameter says whether strict addressing is in effect after reload has 10140finished. This target hook is the same as the 10141@code{TARGET_LEGITIMATE_ADDRESS_P} target hook, except that it includes 10142explicit named address space support. 10143@end deftypefn 10144 10145@hook TARGET_ADDR_SPACE_LEGITIMIZE_ADDRESS 10146Define this to modify an invalid address @var{x} to be a valid address 10147with mode @var{mode} in the named address space @var{as}. This target 10148hook is the same as the @code{TARGET_LEGITIMIZE_ADDRESS} target hook, 10149except that it includes explicit named address space support. 10150@end deftypefn 10151 10152@hook TARGET_ADDR_SPACE_SUBSET_P 10153Define this to return whether the @var{subset} named address space is 10154contained within the @var{superset} named address space. Pointers to 10155a named address space that is a subset of another named address space 10156will be converted automatically without a cast if used together in 10157arithmetic operations. Pointers to a superset address space can be 10158converted to pointers to a subset address space via explicit casts. 10159@end deftypefn 10160 10161@hook TARGET_ADDR_SPACE_CONVERT 10162Define this to convert the pointer expression represented by the RTL 10163@var{op} with type @var{from_type} that points to a named address 10164space to a new pointer expression with type @var{to_type} that points 10165to a different named address space. When this hook it called, it is 10166guaranteed that one of the two address spaces is a subset of the other, 10167as determined by the @code{TARGET_ADDR_SPACE_SUBSET_P} target hook. 10168@end deftypefn 10169 10170@node Misc 10171@section Miscellaneous Parameters 10172@cindex parameters, miscellaneous 10173 10174@c prevent bad page break with this line 10175Here are several miscellaneous parameters. 10176 10177@defmac HAS_LONG_COND_BRANCH 10178Define this boolean macro to indicate whether or not your architecture 10179has conditional branches that can span all of memory. It is used in 10180conjunction with an optimization that partitions hot and cold basic 10181blocks into separate sections of the executable. If this macro is 10182set to false, gcc will convert any conditional branches that attempt 10183to cross between sections into unconditional branches or indirect jumps. 10184@end defmac 10185 10186@defmac HAS_LONG_UNCOND_BRANCH 10187Define this boolean macro to indicate whether or not your architecture 10188has unconditional branches that can span all of memory. It is used in 10189conjunction with an optimization that partitions hot and cold basic 10190blocks into separate sections of the executable. If this macro is 10191set to false, gcc will convert any unconditional branches that attempt 10192to cross between sections into indirect jumps. 10193@end defmac 10194 10195@defmac CASE_VECTOR_MODE 10196An alias for a machine mode name. This is the machine mode that 10197elements of a jump-table should have. 10198@end defmac 10199 10200@defmac CASE_VECTOR_SHORTEN_MODE (@var{min_offset}, @var{max_offset}, @var{body}) 10201Optional: return the preferred mode for an @code{addr_diff_vec} 10202when the minimum and maximum offset are known. If you define this, 10203it enables extra code in branch shortening to deal with @code{addr_diff_vec}. 10204To make this work, you also have to define @code{INSN_ALIGN} and 10205make the alignment for @code{addr_diff_vec} explicit. 10206The @var{body} argument is provided so that the offset_unsigned and scale 10207flags can be updated. 10208@end defmac 10209 10210@defmac CASE_VECTOR_PC_RELATIVE 10211Define this macro to be a C expression to indicate when jump-tables 10212should contain relative addresses. You need not define this macro if 10213jump-tables never contain relative addresses, or jump-tables should 10214contain relative addresses only when @option{-fPIC} or @option{-fPIC} 10215is in effect. 10216@end defmac 10217 10218@hook TARGET_CASE_VALUES_THRESHOLD 10219This function return the smallest number of different values for which it 10220is best to use a jump-table instead of a tree of conditional branches. 10221The default is four for machines with a @code{casesi} instruction and 10222five otherwise. This is best for most machines. 10223@end deftypefn 10224 10225@defmac CASE_USE_BIT_TESTS 10226Define this macro to be a C expression to indicate whether C switch 10227statements may be implemented by a sequence of bit tests. This is 10228advantageous on processors that can efficiently implement left shift 10229of 1 by the number of bits held in a register, but inappropriate on 10230targets that would require a loop. By default, this macro returns 10231@code{true} if the target defines an @code{ashlsi3} pattern, and 10232@code{false} otherwise. 10233@end defmac 10234 10235@defmac WORD_REGISTER_OPERATIONS 10236Define this macro if operations between registers with integral mode 10237smaller than a word are always performed on the entire register. 10238Most RISC machines have this property and most CISC machines do not. 10239@end defmac 10240 10241@defmac LOAD_EXTEND_OP (@var{mem_mode}) 10242Define this macro to be a C expression indicating when insns that read 10243memory in @var{mem_mode}, an integral mode narrower than a word, set the 10244bits outside of @var{mem_mode} to be either the sign-extension or the 10245zero-extension of the data read. Return @code{SIGN_EXTEND} for values 10246of @var{mem_mode} for which the 10247insn sign-extends, @code{ZERO_EXTEND} for which it zero-extends, and 10248@code{UNKNOWN} for other modes. 10249 10250This macro is not called with @var{mem_mode} non-integral or with a width 10251greater than or equal to @code{BITS_PER_WORD}, so you may return any 10252value in this case. Do not define this macro if it would always return 10253@code{UNKNOWN}. On machines where this macro is defined, you will normally 10254define it as the constant @code{SIGN_EXTEND} or @code{ZERO_EXTEND}. 10255 10256You may return a non-@code{UNKNOWN} value even if for some hard registers 10257the sign extension is not performed, if for the @code{REGNO_REG_CLASS} 10258of these hard registers @code{CANNOT_CHANGE_MODE_CLASS} returns nonzero 10259when the @var{from} mode is @var{mem_mode} and the @var{to} mode is any 10260integral mode larger than this but not larger than @code{word_mode}. 10261 10262You must return @code{UNKNOWN} if for some hard registers that allow this 10263mode, @code{CANNOT_CHANGE_MODE_CLASS} says that they cannot change to 10264@code{word_mode}, but that they can change to another integral mode that 10265is larger then @var{mem_mode} but still smaller than @code{word_mode}. 10266@end defmac 10267 10268@defmac SHORT_IMMEDIATES_SIGN_EXTEND 10269Define this macro if loading short immediate values into registers sign 10270extends. 10271@end defmac 10272 10273@defmac FIXUNS_TRUNC_LIKE_FIX_TRUNC 10274Define this macro if the same instructions that convert a floating 10275point number to a signed fixed point number also convert validly to an 10276unsigned one. 10277@end defmac 10278 10279@hook TARGET_MIN_DIVISIONS_FOR_RECIP_MUL 10280When @option{-ffast-math} is in effect, GCC tries to optimize 10281divisions by the same divisor, by turning them into multiplications by 10282the reciprocal. This target hook specifies the minimum number of divisions 10283that should be there for GCC to perform the optimization for a variable 10284of mode @var{mode}. The default implementation returns 3 if the machine 10285has an instruction for the division, and 2 if it does not. 10286@end deftypefn 10287 10288@defmac MOVE_MAX 10289The maximum number of bytes that a single instruction can move quickly 10290between memory and registers or between two memory locations. 10291@end defmac 10292 10293@defmac MAX_MOVE_MAX 10294The maximum number of bytes that a single instruction can move quickly 10295between memory and registers or between two memory locations. If this 10296is undefined, the default is @code{MOVE_MAX}. Otherwise, it is the 10297constant value that is the largest value that @code{MOVE_MAX} can have 10298at run-time. 10299@end defmac 10300 10301@defmac SHIFT_COUNT_TRUNCATED 10302A C expression that is nonzero if on this machine the number of bits 10303actually used for the count of a shift operation is equal to the number 10304of bits needed to represent the size of the object being shifted. When 10305this macro is nonzero, the compiler will assume that it is safe to omit 10306a sign-extend, zero-extend, and certain bitwise `and' instructions that 10307truncates the count of a shift operation. On machines that have 10308instructions that act on bit-fields at variable positions, which may 10309include `bit test' instructions, a nonzero @code{SHIFT_COUNT_TRUNCATED} 10310also enables deletion of truncations of the values that serve as 10311arguments to bit-field instructions. 10312 10313If both types of instructions truncate the count (for shifts) and 10314position (for bit-field operations), or if no variable-position bit-field 10315instructions exist, you should define this macro. 10316 10317However, on some machines, such as the 80386 and the 680x0, truncation 10318only applies to shift operations and not the (real or pretended) 10319bit-field operations. Define @code{SHIFT_COUNT_TRUNCATED} to be zero on 10320such machines. Instead, add patterns to the @file{md} file that include 10321the implied truncation of the shift instructions. 10322 10323You need not define this macro if it would always have the value of zero. 10324@end defmac 10325 10326@anchor{TARGET_SHIFT_TRUNCATION_MASK} 10327@hook TARGET_SHIFT_TRUNCATION_MASK 10328This function describes how the standard shift patterns for @var{mode} 10329deal with shifts by negative amounts or by more than the width of the mode. 10330@xref{shift patterns}. 10331 10332On many machines, the shift patterns will apply a mask @var{m} to the 10333shift count, meaning that a fixed-width shift of @var{x} by @var{y} is 10334equivalent to an arbitrary-width shift of @var{x} by @var{y & m}. If 10335this is true for mode @var{mode}, the function should return @var{m}, 10336otherwise it should return 0. A return value of 0 indicates that no 10337particular behavior is guaranteed. 10338 10339Note that, unlike @code{SHIFT_COUNT_TRUNCATED}, this function does 10340@emph{not} apply to general shift rtxes; it applies only to instructions 10341that are generated by the named shift patterns. 10342 10343The default implementation of this function returns 10344@code{GET_MODE_BITSIZE (@var{mode}) - 1} if @code{SHIFT_COUNT_TRUNCATED} 10345and 0 otherwise. This definition is always safe, but if 10346@code{SHIFT_COUNT_TRUNCATED} is false, and some shift patterns 10347nevertheless truncate the shift count, you may get better code 10348by overriding it. 10349@end deftypefn 10350 10351@defmac TRULY_NOOP_TRUNCATION (@var{outprec}, @var{inprec}) 10352A C expression which is nonzero if on this machine it is safe to 10353``convert'' an integer of @var{inprec} bits to one of @var{outprec} 10354bits (where @var{outprec} is smaller than @var{inprec}) by merely 10355operating on it as if it had only @var{outprec} bits. 10356 10357On many machines, this expression can be 1. 10358 10359@c rearranged this, removed the phrase "it is reported that". this was 10360@c to fix an overfull hbox. --mew 10feb93 10361When @code{TRULY_NOOP_TRUNCATION} returns 1 for a pair of sizes for 10362modes for which @code{MODES_TIEABLE_P} is 0, suboptimal code can result. 10363If this is the case, making @code{TRULY_NOOP_TRUNCATION} return 0 in 10364such cases may improve things. 10365@end defmac 10366 10367@hook TARGET_MODE_REP_EXTENDED 10368The representation of an integral mode can be such that the values 10369are always extended to a wider integral mode. Return 10370@code{SIGN_EXTEND} if values of @var{mode} are represented in 10371sign-extended form to @var{rep_mode}. Return @code{UNKNOWN} 10372otherwise. (Currently, none of the targets use zero-extended 10373representation this way so unlike @code{LOAD_EXTEND_OP}, 10374@code{TARGET_MODE_REP_EXTENDED} is expected to return either 10375@code{SIGN_EXTEND} or @code{UNKNOWN}. Also no target extends 10376@var{mode} to @var{rep_mode} so that @var{rep_mode} is not the next 10377widest integral mode and currently we take advantage of this fact.) 10378 10379Similarly to @code{LOAD_EXTEND_OP} you may return a non-@code{UNKNOWN} 10380value even if the extension is not performed on certain hard registers 10381as long as for the @code{REGNO_REG_CLASS} of these hard registers 10382@code{CANNOT_CHANGE_MODE_CLASS} returns nonzero. 10383 10384Note that @code{TARGET_MODE_REP_EXTENDED} and @code{LOAD_EXTEND_OP} 10385describe two related properties. If you define 10386@code{TARGET_MODE_REP_EXTENDED (mode, word_mode)} you probably also want 10387to define @code{LOAD_EXTEND_OP (mode)} to return the same type of 10388extension. 10389 10390In order to enforce the representation of @code{mode}, 10391@code{TRULY_NOOP_TRUNCATION} should return false when truncating to 10392@code{mode}. 10393@end deftypefn 10394 10395@defmac STORE_FLAG_VALUE 10396A C expression describing the value returned by a comparison operator 10397with an integral mode and stored by a store-flag instruction 10398(@samp{cstore@var{mode}4}) when the condition is true. This description must 10399apply to @emph{all} the @samp{cstore@var{mode}4} patterns and all the 10400comparison operators whose results have a @code{MODE_INT} mode. 10401 10402A value of 1 or @minus{}1 means that the instruction implementing the 10403comparison operator returns exactly 1 or @minus{}1 when the comparison is true 10404and 0 when the comparison is false. Otherwise, the value indicates 10405which bits of the result are guaranteed to be 1 when the comparison is 10406true. This value is interpreted in the mode of the comparison 10407operation, which is given by the mode of the first operand in the 10408@samp{cstore@var{mode}4} pattern. Either the low bit or the sign bit of 10409@code{STORE_FLAG_VALUE} be on. Presently, only those bits are used by 10410the compiler. 10411 10412If @code{STORE_FLAG_VALUE} is neither 1 or @minus{}1, the compiler will 10413generate code that depends only on the specified bits. It can also 10414replace comparison operators with equivalent operations if they cause 10415the required bits to be set, even if the remaining bits are undefined. 10416For example, on a machine whose comparison operators return an 10417@code{SImode} value and where @code{STORE_FLAG_VALUE} is defined as 10418@samp{0x80000000}, saying that just the sign bit is relevant, the 10419expression 10420 10421@smallexample 10422(ne:SI (and:SI @var{x} (const_int @var{power-of-2})) (const_int 0)) 10423@end smallexample 10424 10425@noindent 10426can be converted to 10427 10428@smallexample 10429(ashift:SI @var{x} (const_int @var{n})) 10430@end smallexample 10431 10432@noindent 10433where @var{n} is the appropriate shift count to move the bit being 10434tested into the sign bit. 10435 10436There is no way to describe a machine that always sets the low-order bit 10437for a true value, but does not guarantee the value of any other bits, 10438but we do not know of any machine that has such an instruction. If you 10439are trying to port GCC to such a machine, include an instruction to 10440perform a logical-and of the result with 1 in the pattern for the 10441comparison operators and let us know at @email{gcc@@gcc.gnu.org}. 10442 10443Often, a machine will have multiple instructions that obtain a value 10444from a comparison (or the condition codes). Here are rules to guide the 10445choice of value for @code{STORE_FLAG_VALUE}, and hence the instructions 10446to be used: 10447 10448@itemize @bullet 10449@item 10450Use the shortest sequence that yields a valid definition for 10451@code{STORE_FLAG_VALUE}. It is more efficient for the compiler to 10452``normalize'' the value (convert it to, e.g., 1 or 0) than for the 10453comparison operators to do so because there may be opportunities to 10454combine the normalization with other operations. 10455 10456@item 10457For equal-length sequences, use a value of 1 or @minus{}1, with @minus{}1 being 10458slightly preferred on machines with expensive jumps and 1 preferred on 10459other machines. 10460 10461@item 10462As a second choice, choose a value of @samp{0x80000001} if instructions 10463exist that set both the sign and low-order bits but do not define the 10464others. 10465 10466@item 10467Otherwise, use a value of @samp{0x80000000}. 10468@end itemize 10469 10470Many machines can produce both the value chosen for 10471@code{STORE_FLAG_VALUE} and its negation in the same number of 10472instructions. On those machines, you should also define a pattern for 10473those cases, e.g., one matching 10474 10475@smallexample 10476(set @var{A} (neg:@var{m} (ne:@var{m} @var{B} @var{C}))) 10477@end smallexample 10478 10479Some machines can also perform @code{and} or @code{plus} operations on 10480condition code values with less instructions than the corresponding 10481@samp{cstore@var{mode}4} insn followed by @code{and} or @code{plus}. On those 10482machines, define the appropriate patterns. Use the names @code{incscc} 10483and @code{decscc}, respectively, for the patterns which perform 10484@code{plus} or @code{minus} operations on condition code values. See 10485@file{rs6000.md} for some examples. The GNU Superoptimizer can be used to 10486find such instruction sequences on other machines. 10487 10488If this macro is not defined, the default value, 1, is used. You need 10489not define @code{STORE_FLAG_VALUE} if the machine has no store-flag 10490instructions, or if the value generated by these instructions is 1. 10491@end defmac 10492 10493@defmac FLOAT_STORE_FLAG_VALUE (@var{mode}) 10494A C expression that gives a nonzero @code{REAL_VALUE_TYPE} value that is 10495returned when comparison operators with floating-point results are true. 10496Define this macro on machines that have comparison operations that return 10497floating-point values. If there are no such operations, do not define 10498this macro. 10499@end defmac 10500 10501@defmac VECTOR_STORE_FLAG_VALUE (@var{mode}) 10502A C expression that gives a rtx representing the nonzero true element 10503for vector comparisons. The returned rtx should be valid for the inner 10504mode of @var{mode} which is guaranteed to be a vector mode. Define 10505this macro on machines that have vector comparison operations that 10506return a vector result. If there are no such operations, do not define 10507this macro. Typically, this macro is defined as @code{const1_rtx} or 10508@code{constm1_rtx}. This macro may return @code{NULL_RTX} to prevent 10509the compiler optimizing such vector comparison operations for the 10510given mode. 10511@end defmac 10512 10513@defmac CLZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value}) 10514@defmacx CTZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value}) 10515A C expression that indicates whether the architecture defines a value 10516for @code{clz} or @code{ctz} with a zero operand. 10517A result of @code{0} indicates the value is undefined. 10518If the value is defined for only the RTL expression, the macro should 10519evaluate to @code{1}; if the value applies also to the corresponding optab 10520entry (which is normally the case if it expands directly into 10521the corresponding RTL), then the macro should evaluate to @code{2}. 10522In the cases where the value is defined, @var{value} should be set to 10523this value. 10524 10525If this macro is not defined, the value of @code{clz} or 10526@code{ctz} at zero is assumed to be undefined. 10527 10528This macro must be defined if the target's expansion for @code{ffs} 10529relies on a particular value to get correct results. Otherwise it 10530is not necessary, though it may be used to optimize some corner cases, and 10531to provide a default expansion for the @code{ffs} optab. 10532 10533Note that regardless of this macro the ``definedness'' of @code{clz} 10534and @code{ctz} at zero do @emph{not} extend to the builtin functions 10535visible to the user. Thus one may be free to adjust the value at will 10536to match the target expansion of these operations without fear of 10537breaking the API@. 10538@end defmac 10539 10540@defmac Pmode 10541An alias for the machine mode for pointers. On most machines, define 10542this to be the integer mode corresponding to the width of a hardware 10543pointer; @code{SImode} on 32-bit machine or @code{DImode} on 64-bit machines. 10544On some machines you must define this to be one of the partial integer 10545modes, such as @code{PSImode}. 10546 10547The width of @code{Pmode} must be at least as large as the value of 10548@code{POINTER_SIZE}. If it is not equal, you must define the macro 10549@code{POINTERS_EXTEND_UNSIGNED} to specify how pointers are extended 10550to @code{Pmode}. 10551@end defmac 10552 10553@defmac FUNCTION_MODE 10554An alias for the machine mode used for memory references to functions 10555being called, in @code{call} RTL expressions. On most CISC machines, 10556where an instruction can begin at any byte address, this should be 10557@code{QImode}. On most RISC machines, where all instructions have fixed 10558size and alignment, this should be a mode with the same size and alignment 10559as the machine instruction words - typically @code{SImode} or @code{HImode}. 10560@end defmac 10561 10562@defmac STDC_0_IN_SYSTEM_HEADERS 10563In normal operation, the preprocessor expands @code{__STDC__} to the 10564constant 1, to signify that GCC conforms to ISO Standard C@. On some 10565hosts, like Solaris, the system compiler uses a different convention, 10566where @code{__STDC__} is normally 0, but is 1 if the user specifies 10567strict conformance to the C Standard. 10568 10569Defining @code{STDC_0_IN_SYSTEM_HEADERS} makes GNU CPP follows the host 10570convention when processing system header files, but when processing user 10571files @code{__STDC__} will always expand to 1. 10572@end defmac 10573 10574@defmac NO_IMPLICIT_EXTERN_C 10575Define this macro if the system header files support C++ as well as C@. 10576This macro inhibits the usual method of using system header files in 10577C++, which is to pretend that the file's contents are enclosed in 10578@samp{extern "C" @{@dots{}@}}. 10579@end defmac 10580 10581@findex #pragma 10582@findex pragma 10583@defmac REGISTER_TARGET_PRAGMAS () 10584Define this macro if you want to implement any target-specific pragmas. 10585If defined, it is a C expression which makes a series of calls to 10586@code{c_register_pragma} or @code{c_register_pragma_with_expansion} 10587for each pragma. The macro may also do any 10588setup required for the pragmas. 10589 10590The primary reason to define this macro is to provide compatibility with 10591other compilers for the same target. In general, we discourage 10592definition of target-specific pragmas for GCC@. 10593 10594If the pragma can be implemented by attributes then you should consider 10595defining the target hook @samp{TARGET_INSERT_ATTRIBUTES} as well. 10596 10597Preprocessor macros that appear on pragma lines are not expanded. All 10598@samp{#pragma} directives that do not match any registered pragma are 10599silently ignored, unless the user specifies @option{-Wunknown-pragmas}. 10600@end defmac 10601 10602@deftypefun void c_register_pragma (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *)) 10603@deftypefunx void c_register_pragma_with_expansion (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *)) 10604 10605Each call to @code{c_register_pragma} or 10606@code{c_register_pragma_with_expansion} establishes one pragma. The 10607@var{callback} routine will be called when the preprocessor encounters a 10608pragma of the form 10609 10610@smallexample 10611#pragma [@var{space}] @var{name} @dots{} 10612@end smallexample 10613 10614@var{space} is the case-sensitive namespace of the pragma, or 10615@code{NULL} to put the pragma in the global namespace. The callback 10616routine receives @var{pfile} as its first argument, which can be passed 10617on to cpplib's functions if necessary. You can lex tokens after the 10618@var{name} by calling @code{pragma_lex}. Tokens that are not read by the 10619callback will be silently ignored. The end of the line is indicated by 10620a token of type @code{CPP_EOF}. Macro expansion occurs on the 10621arguments of pragmas registered with 10622@code{c_register_pragma_with_expansion} but not on the arguments of 10623pragmas registered with @code{c_register_pragma}. 10624 10625Note that the use of @code{pragma_lex} is specific to the C and C++ 10626compilers. It will not work in the Java or Fortran compilers, or any 10627other language compilers for that matter. Thus if @code{pragma_lex} is going 10628to be called from target-specific code, it must only be done so when 10629building the C and C++ compilers. This can be done by defining the 10630variables @code{c_target_objs} and @code{cxx_target_objs} in the 10631target entry in the @file{config.gcc} file. These variables should name 10632the target-specific, language-specific object file which contains the 10633code that uses @code{pragma_lex}. Note it will also be necessary to add a 10634rule to the makefile fragment pointed to by @code{tmake_file} that shows 10635how to build this object file. 10636@end deftypefun 10637 10638@defmac HANDLE_PRAGMA_PACK_WITH_EXPANSION 10639Define this macro if macros should be expanded in the 10640arguments of @samp{#pragma pack}. 10641@end defmac 10642 10643@hook TARGET_HANDLE_PRAGMA_EXTERN_PREFIX 10644 10645@defmac TARGET_DEFAULT_PACK_STRUCT 10646If your target requires a structure packing default other than 0 (meaning 10647the machine default), define this macro to the necessary value (in bytes). 10648This must be a value that would also be valid to use with 10649@samp{#pragma pack()} (that is, a small power of two). 10650@end defmac 10651 10652@defmac DOLLARS_IN_IDENTIFIERS 10653Define this macro to control use of the character @samp{$} in 10654identifier names for the C family of languages. 0 means @samp{$} is 10655not allowed by default; 1 means it is allowed. 1 is the default; 10656there is no need to define this macro in that case. 10657@end defmac 10658 10659@defmac NO_DOLLAR_IN_LABEL 10660Define this macro if the assembler does not accept the character 10661@samp{$} in label names. By default constructors and destructors in 10662G++ have @samp{$} in the identifiers. If this macro is defined, 10663@samp{.} is used instead. 10664@end defmac 10665 10666@defmac NO_DOT_IN_LABEL 10667Define this macro if the assembler does not accept the character 10668@samp{.} in label names. By default constructors and destructors in G++ 10669have names that use @samp{.}. If this macro is defined, these names 10670are rewritten to avoid @samp{.}. 10671@end defmac 10672 10673@defmac INSN_SETS_ARE_DELAYED (@var{insn}) 10674Define this macro as a C expression that is nonzero if it is safe for the 10675delay slot scheduler to place instructions in the delay slot of @var{insn}, 10676even if they appear to use a resource set or clobbered in @var{insn}. 10677@var{insn} is always a @code{jump_insn} or an @code{insn}; GCC knows that 10678every @code{call_insn} has this behavior. On machines where some @code{insn} 10679or @code{jump_insn} is really a function call and hence has this behavior, 10680you should define this macro. 10681 10682You need not define this macro if it would always return zero. 10683@end defmac 10684 10685@defmac INSN_REFERENCES_ARE_DELAYED (@var{insn}) 10686Define this macro as a C expression that is nonzero if it is safe for the 10687delay slot scheduler to place instructions in the delay slot of @var{insn}, 10688even if they appear to set or clobber a resource referenced in @var{insn}. 10689@var{insn} is always a @code{jump_insn} or an @code{insn}. On machines where 10690some @code{insn} or @code{jump_insn} is really a function call and its operands 10691are registers whose use is actually in the subroutine it calls, you should 10692define this macro. Doing so allows the delay slot scheduler to move 10693instructions which copy arguments into the argument registers into the delay 10694slot of @var{insn}. 10695 10696You need not define this macro if it would always return zero. 10697@end defmac 10698 10699@defmac MULTIPLE_SYMBOL_SPACES 10700Define this macro as a C expression that is nonzero if, in some cases, 10701global symbols from one translation unit may not be bound to undefined 10702symbols in another translation unit without user intervention. For 10703instance, under Microsoft Windows symbols must be explicitly imported 10704from shared libraries (DLLs). 10705 10706You need not define this macro if it would always evaluate to zero. 10707@end defmac 10708 10709@hook TARGET_MD_ASM_CLOBBERS 10710This target hook should add to @var{clobbers} @code{STRING_CST} trees for 10711any hard regs the port wishes to automatically clobber for an asm. 10712It should return the result of the last @code{tree_cons} used to add a 10713clobber. The @var{outputs}, @var{inputs} and @var{clobber} lists are the 10714corresponding parameters to the asm and may be inspected to avoid 10715clobbering a register that is an input or output of the asm. You can use 10716@code{tree_overlaps_hard_reg_set}, declared in @file{tree.h}, to test 10717for overlap with regards to asm-declared registers. 10718@end deftypefn 10719 10720@defmac MATH_LIBRARY 10721Define this macro as a C string constant for the linker argument to link 10722in the system math library, minus the initial @samp{"-l"}, or 10723@samp{""} if the target does not have a 10724separate math library. 10725 10726You need only define this macro if the default of @samp{"m"} is wrong. 10727@end defmac 10728 10729@defmac LIBRARY_PATH_ENV 10730Define this macro as a C string constant for the environment variable that 10731specifies where the linker should look for libraries. 10732 10733You need only define this macro if the default of @samp{"LIBRARY_PATH"} 10734is wrong. 10735@end defmac 10736 10737@defmac TARGET_POSIX_IO 10738Define this macro if the target supports the following POSIX@ file 10739functions, access, mkdir and file locking with fcntl / F_SETLKW@. 10740Defining @code{TARGET_POSIX_IO} will enable the test coverage code 10741to use file locking when exiting a program, which avoids race conditions 10742if the program has forked. It will also create directories at run-time 10743for cross-profiling. 10744@end defmac 10745 10746@defmac MAX_CONDITIONAL_EXECUTE 10747 10748A C expression for the maximum number of instructions to execute via 10749conditional execution instructions instead of a branch. A value of 10750@code{BRANCH_COST}+1 is the default if the machine does not use cc0, and 107511 if it does use cc0. 10752@end defmac 10753 10754@defmac IFCVT_MODIFY_TESTS (@var{ce_info}, @var{true_expr}, @var{false_expr}) 10755Used if the target needs to perform machine-dependent modifications on the 10756conditionals used for turning basic blocks into conditionally executed code. 10757@var{ce_info} points to a data structure, @code{struct ce_if_block}, which 10758contains information about the currently processed blocks. @var{true_expr} 10759and @var{false_expr} are the tests that are used for converting the 10760then-block and the else-block, respectively. Set either @var{true_expr} or 10761@var{false_expr} to a null pointer if the tests cannot be converted. 10762@end defmac 10763 10764@defmac IFCVT_MODIFY_MULTIPLE_TESTS (@var{ce_info}, @var{bb}, @var{true_expr}, @var{false_expr}) 10765Like @code{IFCVT_MODIFY_TESTS}, but used when converting more complicated 10766if-statements into conditions combined by @code{and} and @code{or} operations. 10767@var{bb} contains the basic block that contains the test that is currently 10768being processed and about to be turned into a condition. 10769@end defmac 10770 10771@defmac IFCVT_MODIFY_INSN (@var{ce_info}, @var{pattern}, @var{insn}) 10772A C expression to modify the @var{PATTERN} of an @var{INSN} that is to 10773be converted to conditional execution format. @var{ce_info} points to 10774a data structure, @code{struct ce_if_block}, which contains information 10775about the currently processed blocks. 10776@end defmac 10777 10778@defmac IFCVT_MODIFY_FINAL (@var{ce_info}) 10779A C expression to perform any final machine dependent modifications in 10780converting code to conditional execution. The involved basic blocks 10781can be found in the @code{struct ce_if_block} structure that is pointed 10782to by @var{ce_info}. 10783@end defmac 10784 10785@defmac IFCVT_MODIFY_CANCEL (@var{ce_info}) 10786A C expression to cancel any machine dependent modifications in 10787converting code to conditional execution. The involved basic blocks 10788can be found in the @code{struct ce_if_block} structure that is pointed 10789to by @var{ce_info}. 10790@end defmac 10791 10792@defmac IFCVT_INIT_EXTRA_FIELDS (@var{ce_info}) 10793A C expression to initialize any extra fields in a @code{struct ce_if_block} 10794structure, which are defined by the @code{IFCVT_EXTRA_FIELDS} macro. 10795@end defmac 10796 10797@defmac IFCVT_EXTRA_FIELDS 10798If defined, it should expand to a set of field declarations that will be 10799added to the @code{struct ce_if_block} structure. These should be initialized 10800by the @code{IFCVT_INIT_EXTRA_FIELDS} macro. 10801@end defmac 10802 10803@hook TARGET_MACHINE_DEPENDENT_REORG 10804If non-null, this hook performs a target-specific pass over the 10805instruction stream. The compiler will run it at all optimization levels, 10806just before the point at which it normally does delayed-branch scheduling. 10807 10808The exact purpose of the hook varies from target to target. Some use 10809it to do transformations that are necessary for correctness, such as 10810laying out in-function constant pools or avoiding hardware hazards. 10811Others use it as an opportunity to do some machine-dependent optimizations. 10812 10813You need not implement the hook if it has nothing to do. The default 10814definition is null. 10815@end deftypefn 10816 10817@hook TARGET_INIT_BUILTINS 10818Define this hook if you have any machine-specific built-in functions 10819that need to be defined. It should be a function that performs the 10820necessary setup. 10821 10822Machine specific built-in functions can be useful to expand special machine 10823instructions that would otherwise not normally be generated because 10824they have no equivalent in the source language (for example, SIMD vector 10825instructions or prefetch instructions). 10826 10827To create a built-in function, call the function 10828@code{lang_hooks.builtin_function} 10829which is defined by the language front end. You can use any type nodes set 10830up by @code{build_common_tree_nodes}; 10831only language front ends that use those two functions will call 10832@samp{TARGET_INIT_BUILTINS}. 10833@end deftypefn 10834 10835@hook TARGET_BUILTIN_DECL 10836Define this hook if you have any machine-specific built-in functions 10837that need to be defined. It should be a function that returns the 10838builtin function declaration for the builtin function code @var{code}. 10839If there is no such builtin and it cannot be initialized at this time 10840if @var{initialize_p} is true the function should return @code{NULL_TREE}. 10841If @var{code} is out of range the function should return 10842@code{error_mark_node}. 10843@end deftypefn 10844 10845@hook TARGET_EXPAND_BUILTIN 10846 10847Expand a call to a machine specific built-in function that was set up by 10848@samp{TARGET_INIT_BUILTINS}. @var{exp} is the expression for the 10849function call; the result should go to @var{target} if that is 10850convenient, and have mode @var{mode} if that is convenient. 10851@var{subtarget} may be used as the target for computing one of 10852@var{exp}'s operands. @var{ignore} is nonzero if the value is to be 10853ignored. This function should return the result of the call to the 10854built-in function. 10855@end deftypefn 10856 10857@hook TARGET_RESOLVE_OVERLOADED_BUILTIN 10858Select a replacement for a machine specific built-in function that 10859was set up by @samp{TARGET_INIT_BUILTINS}. This is done 10860@emph{before} regular type checking, and so allows the target to 10861implement a crude form of function overloading. @var{fndecl} is the 10862declaration of the built-in function. @var{arglist} is the list of 10863arguments passed to the built-in function. The result is a 10864complete expression that implements the operation, usually 10865another @code{CALL_EXPR}. 10866@var{arglist} really has type @samp{VEC(tree,gc)*} 10867@end deftypefn 10868 10869@hook TARGET_FOLD_BUILTIN 10870Fold a call to a machine specific built-in function that was set up by 10871@samp{TARGET_INIT_BUILTINS}. @var{fndecl} is the declaration of the 10872built-in function. @var{n_args} is the number of arguments passed to 10873the function; the arguments themselves are pointed to by @var{argp}. 10874The result is another tree containing a simplified expression for the 10875call's result. If @var{ignore} is true the value will be ignored. 10876@end deftypefn 10877 10878@hook TARGET_INVALID_WITHIN_DOLOOP 10879 10880Take an instruction in @var{insn} and return NULL if it is valid within a 10881low-overhead loop, otherwise return a string explaining why doloop 10882could not be applied. 10883 10884Many targets use special registers for low-overhead looping. For any 10885instruction that clobbers these this function should return a string indicating 10886the reason why the doloop could not be applied. 10887By default, the RTL loop optimizer does not use a present doloop pattern for 10888loops containing function calls or branch on table instructions. 10889@end deftypefn 10890 10891@defmac MD_CAN_REDIRECT_BRANCH (@var{branch1}, @var{branch2}) 10892 10893Take a branch insn in @var{branch1} and another in @var{branch2}. 10894Return true if redirecting @var{branch1} to the destination of 10895@var{branch2} is possible. 10896 10897On some targets, branches may have a limited range. Optimizing the 10898filling of delay slots can result in branches being redirected, and this 10899may in turn cause a branch offset to overflow. 10900@end defmac 10901 10902@hook TARGET_COMMUTATIVE_P 10903This target hook returns @code{true} if @var{x} is considered to be commutative. 10904Usually, this is just COMMUTATIVE_P (@var{x}), but the HP PA doesn't consider 10905PLUS to be commutative inside a MEM@. @var{outer_code} is the rtx code 10906of the enclosing rtl, if known, otherwise it is UNKNOWN. 10907@end deftypefn 10908 10909@hook TARGET_ALLOCATE_INITIAL_VALUE 10910 10911When the initial value of a hard register has been copied in a pseudo 10912register, it is often not necessary to actually allocate another register 10913to this pseudo register, because the original hard register or a stack slot 10914it has been saved into can be used. @code{TARGET_ALLOCATE_INITIAL_VALUE} 10915is called at the start of register allocation once for each hard register 10916that had its initial value copied by using 10917@code{get_func_hard_reg_initial_val} or @code{get_hard_reg_initial_val}. 10918Possible values are @code{NULL_RTX}, if you don't want 10919to do any special allocation, a @code{REG} rtx---that would typically be 10920the hard register itself, if it is known not to be clobbered---or a 10921@code{MEM}. 10922If you are returning a @code{MEM}, this is only a hint for the allocator; 10923it might decide to use another register anyways. 10924You may use @code{current_function_leaf_function} in the hook, functions 10925that use @code{REG_N_SETS}, to determine if the hard 10926register in question will not be clobbered. 10927The default value of this hook is @code{NULL}, which disables any special 10928allocation. 10929@end deftypefn 10930 10931@hook TARGET_UNSPEC_MAY_TRAP_P 10932This target hook returns nonzero if @var{x}, an @code{unspec} or 10933@code{unspec_volatile} operation, might cause a trap. Targets can use 10934this hook to enhance precision of analysis for @code{unspec} and 10935@code{unspec_volatile} operations. You may call @code{may_trap_p_1} 10936to analyze inner elements of @var{x} in which case @var{flags} should be 10937passed along. 10938@end deftypefn 10939 10940@hook TARGET_SET_CURRENT_FUNCTION 10941The compiler invokes this hook whenever it changes its current function 10942context (@code{cfun}). You can define this function if 10943the back end needs to perform any initialization or reset actions on a 10944per-function basis. For example, it may be used to implement function 10945attributes that affect register usage or code generation patterns. 10946The argument @var{decl} is the declaration for the new function context, 10947and may be null to indicate that the compiler has left a function context 10948and is returning to processing at the top level. 10949The default hook function does nothing. 10950 10951GCC sets @code{cfun} to a dummy function context during initialization of 10952some parts of the back end. The hook function is not invoked in this 10953situation; you need not worry about the hook being invoked recursively, 10954or when the back end is in a partially-initialized state. 10955@code{cfun} might be @code{NULL} to indicate processing at top level, 10956outside of any function scope. 10957@end deftypefn 10958 10959@defmac TARGET_OBJECT_SUFFIX 10960Define this macro to be a C string representing the suffix for object 10961files on your target machine. If you do not define this macro, GCC will 10962use @samp{.o} as the suffix for object files. 10963@end defmac 10964 10965@defmac TARGET_EXECUTABLE_SUFFIX 10966Define this macro to be a C string representing the suffix to be 10967automatically added to executable files on your target machine. If you 10968do not define this macro, GCC will use the null string as the suffix for 10969executable files. 10970@end defmac 10971 10972@defmac COLLECT_EXPORT_LIST 10973If defined, @code{collect2} will scan the individual object files 10974specified on its command line and create an export list for the linker. 10975Define this macro for systems like AIX, where the linker discards 10976object files that are not referenced from @code{main} and uses export 10977lists. 10978@end defmac 10979 10980@defmac MODIFY_JNI_METHOD_CALL (@var{mdecl}) 10981Define this macro to a C expression representing a variant of the 10982method call @var{mdecl}, if Java Native Interface (JNI) methods 10983must be invoked differently from other methods on your target. 10984For example, on 32-bit Microsoft Windows, JNI methods must be invoked using 10985the @code{stdcall} calling convention and this macro is then 10986defined as this expression: 10987 10988@smallexample 10989build_type_attribute_variant (@var{mdecl}, 10990 build_tree_list 10991 (get_identifier ("stdcall"), 10992 NULL)) 10993@end smallexample 10994@end defmac 10995 10996@hook TARGET_CANNOT_MODIFY_JUMPS_P 10997This target hook returns @code{true} past the point in which new jump 10998instructions could be created. On machines that require a register for 10999every jump such as the SHmedia ISA of SH5, this point would typically be 11000reload, so this target hook should be defined to a function such as: 11001 11002@smallexample 11003static bool 11004cannot_modify_jumps_past_reload_p () 11005@{ 11006 return (reload_completed || reload_in_progress); 11007@} 11008@end smallexample 11009@end deftypefn 11010 11011@hook TARGET_BRANCH_TARGET_REGISTER_CLASS 11012This target hook returns a register class for which branch target register 11013optimizations should be applied. All registers in this class should be 11014usable interchangeably. After reload, registers in this class will be 11015re-allocated and loads will be hoisted out of loops and be subjected 11016to inter-block scheduling. 11017@end deftypefn 11018 11019@hook TARGET_BRANCH_TARGET_REGISTER_CALLEE_SAVED 11020Branch target register optimization will by default exclude callee-saved 11021registers 11022that are not already live during the current function; if this target hook 11023returns true, they will be included. The target code must than make sure 11024that all target registers in the class returned by 11025@samp{TARGET_BRANCH_TARGET_REGISTER_CLASS} that might need saving are 11026saved. @var{after_prologue_epilogue_gen} indicates if prologues and 11027epilogues have already been generated. Note, even if you only return 11028true when @var{after_prologue_epilogue_gen} is false, you still are likely 11029to have to make special provisions in @code{INITIAL_ELIMINATION_OFFSET} 11030to reserve space for caller-saved target registers. 11031@end deftypefn 11032 11033@hook TARGET_HAVE_CONDITIONAL_EXECUTION 11034This target hook returns true if the target supports conditional execution. 11035This target hook is required only when the target has several different 11036modes and they have different conditional execution capability, such as ARM. 11037@end deftypefn 11038 11039@hook TARGET_LOOP_UNROLL_ADJUST 11040This target hook returns a new value for the number of times @var{loop} 11041should be unrolled. The parameter @var{nunroll} is the number of times 11042the loop is to be unrolled. The parameter @var{loop} is a pointer to 11043the loop, which is going to be checked for unrolling. This target hook 11044is required only when the target has special constraints like maximum 11045number of memory accesses. 11046@end deftypefn 11047 11048@defmac POWI_MAX_MULTS 11049If defined, this macro is interpreted as a signed integer C expression 11050that specifies the maximum number of floating point multiplications 11051that should be emitted when expanding exponentiation by an integer 11052constant inline. When this value is defined, exponentiation requiring 11053more than this number of multiplications is implemented by calling the 11054system library's @code{pow}, @code{powf} or @code{powl} routines. 11055The default value places no upper bound on the multiplication count. 11056@end defmac 11057 11058@deftypefn Macro void TARGET_EXTRA_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc}) 11059This target hook should register any extra include files for the 11060target. The parameter @var{stdinc} indicates if normal include files 11061are present. The parameter @var{sysroot} is the system root directory. 11062The parameter @var{iprefix} is the prefix for the gcc directory. 11063@end deftypefn 11064 11065@deftypefn Macro void TARGET_EXTRA_PRE_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc}) 11066This target hook should register any extra include files for the 11067target before any standard headers. The parameter @var{stdinc} 11068indicates if normal include files are present. The parameter 11069@var{sysroot} is the system root directory. The parameter 11070@var{iprefix} is the prefix for the gcc directory. 11071@end deftypefn 11072 11073@deftypefn Macro void TARGET_OPTF (char *@var{path}) 11074This target hook should register special include paths for the target. 11075The parameter @var{path} is the include to register. On Darwin 11076systems, this is used for Framework includes, which have semantics 11077that are different from @option{-I}. 11078@end deftypefn 11079 11080@defmac bool TARGET_USE_LOCAL_THUNK_ALIAS_P (tree @var{fndecl}) 11081This target macro returns @code{true} if it is safe to use a local alias 11082for a virtual function @var{fndecl} when constructing thunks, 11083@code{false} otherwise. By default, the macro returns @code{true} for all 11084functions, if a target supports aliases (i.e.@: defines 11085@code{ASM_OUTPUT_DEF}), @code{false} otherwise, 11086@end defmac 11087 11088@defmac TARGET_FORMAT_TYPES 11089If defined, this macro is the name of a global variable containing 11090target-specific format checking information for the @option{-Wformat} 11091option. The default is to have no target-specific format checks. 11092@end defmac 11093 11094@defmac TARGET_N_FORMAT_TYPES 11095If defined, this macro is the number of entries in 11096@code{TARGET_FORMAT_TYPES}. 11097@end defmac 11098 11099@defmac TARGET_OVERRIDES_FORMAT_ATTRIBUTES 11100If defined, this macro is the name of a global variable containing 11101target-specific format overrides for the @option{-Wformat} option. The 11102default is to have no target-specific format overrides. If defined, 11103@code{TARGET_FORMAT_TYPES} must be defined, too. 11104@end defmac 11105 11106@defmac TARGET_OVERRIDES_FORMAT_ATTRIBUTES_COUNT 11107If defined, this macro specifies the number of entries in 11108@code{TARGET_OVERRIDES_FORMAT_ATTRIBUTES}. 11109@end defmac 11110 11111@defmac TARGET_OVERRIDES_FORMAT_INIT 11112If defined, this macro specifies the optional initialization 11113routine for target specific customizations of the system printf 11114and scanf formatter settings. 11115@end defmac 11116 11117@hook TARGET_RELAXED_ORDERING 11118If set to @code{true}, means that the target's memory model does not 11119guarantee that loads which do not depend on one another will access 11120main memory in the order of the instruction stream; if ordering is 11121important, an explicit memory barrier must be used. This is true of 11122many recent processors which implement a policy of ``relaxed,'' 11123``weak,'' or ``release'' memory consistency, such as Alpha, PowerPC, 11124and ia64. The default is @code{false}. 11125@end deftypevr 11126 11127@hook TARGET_INVALID_ARG_FOR_UNPROTOTYPED_FN 11128If defined, this macro returns the diagnostic message when it is 11129illegal to pass argument @var{val} to function @var{funcdecl} 11130with prototype @var{typelist}. 11131@end deftypefn 11132 11133@hook TARGET_INVALID_CONVERSION 11134If defined, this macro returns the diagnostic message when it is 11135invalid to convert from @var{fromtype} to @var{totype}, or @code{NULL} 11136if validity should be determined by the front end. 11137@end deftypefn 11138 11139@hook TARGET_INVALID_UNARY_OP 11140If defined, this macro returns the diagnostic message when it is 11141invalid to apply operation @var{op} (where unary plus is denoted by 11142@code{CONVERT_EXPR}) to an operand of type @var{type}, or @code{NULL} 11143if validity should be determined by the front end. 11144@end deftypefn 11145 11146@hook TARGET_INVALID_BINARY_OP 11147If defined, this macro returns the diagnostic message when it is 11148invalid to apply operation @var{op} to operands of types @var{type1} 11149and @var{type2}, or @code{NULL} if validity should be determined by 11150the front end. 11151@end deftypefn 11152 11153@hook TARGET_INVALID_PARAMETER_TYPE 11154If defined, this macro returns the diagnostic message when it is 11155invalid for functions to include parameters of type @var{type}, 11156or @code{NULL} if validity should be determined by 11157the front end. This is currently used only by the C and C++ front ends. 11158@end deftypefn 11159 11160@hook TARGET_INVALID_RETURN_TYPE 11161If defined, this macro returns the diagnostic message when it is 11162invalid for functions to have return type @var{type}, 11163or @code{NULL} if validity should be determined by 11164the front end. This is currently used only by the C and C++ front ends. 11165@end deftypefn 11166 11167@hook TARGET_PROMOTED_TYPE 11168If defined, this target hook returns the type to which values of 11169@var{type} should be promoted when they appear in expressions, 11170analogous to the integer promotions, or @code{NULL_TREE} to use the 11171front end's normal promotion rules. This hook is useful when there are 11172target-specific types with special promotion rules. 11173This is currently used only by the C and C++ front ends. 11174@end deftypefn 11175 11176@hook TARGET_CONVERT_TO_TYPE 11177If defined, this hook returns the result of converting @var{expr} to 11178@var{type}. It should return the converted expression, 11179or @code{NULL_TREE} to apply the front end's normal conversion rules. 11180This hook is useful when there are target-specific types with special 11181conversion rules. 11182This is currently used only by the C and C++ front ends. 11183@end deftypefn 11184 11185@defmac TARGET_USE_JCR_SECTION 11186This macro determines whether to use the JCR section to register Java 11187classes. By default, TARGET_USE_JCR_SECTION is defined to 1 if both 11188SUPPORTS_WEAK and TARGET_HAVE_NAMED_SECTIONS are true, else 0. 11189@end defmac 11190 11191@defmac OBJC_JBLEN 11192This macro determines the size of the objective C jump buffer for the 11193NeXT runtime. By default, OBJC_JBLEN is defined to an innocuous value. 11194@end defmac 11195 11196@defmac LIBGCC2_UNWIND_ATTRIBUTE 11197Define this macro if any target-specific attributes need to be attached 11198to the functions in @file{libgcc} that provide low-level support for 11199call stack unwinding. It is used in declarations in @file{unwind-generic.h} 11200and the associated definitions of those functions. 11201@end defmac 11202 11203@hook TARGET_UPDATE_STACK_BOUNDARY 11204Define this macro to update the current function stack boundary if 11205necessary. 11206@end deftypefn 11207 11208@hook TARGET_GET_DRAP_RTX 11209This hook should return an rtx for Dynamic Realign Argument Pointer (DRAP) if a 11210different argument pointer register is needed to access the function's 11211argument list due to stack realignment. Return @code{NULL} if no DRAP 11212is needed. 11213@end deftypefn 11214 11215@hook TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS 11216When optimization is disabled, this hook indicates whether or not 11217arguments should be allocated to stack slots. Normally, GCC allocates 11218stacks slots for arguments when not optimizing in order to make 11219debugging easier. However, when a function is declared with 11220@code{__attribute__((naked))}, there is no stack frame, and the compiler 11221cannot safely move arguments from the registers in which they are passed 11222to the stack. Therefore, this hook should return true in general, but 11223false for naked functions. The default implementation always returns true. 11224@end deftypefn 11225 11226@hook TARGET_CONST_ANCHOR 11227On some architectures it can take multiple instructions to synthesize 11228a constant. If there is another constant already in a register that 11229is close enough in value then it is preferable that the new constant 11230is computed from this register using immediate addition or 11231subtraction. We accomplish this through CSE. Besides the value of 11232the constant we also add a lower and an upper constant anchor to the 11233available expressions. These are then queried when encountering new 11234constants. The anchors are computed by rounding the constant up and 11235down to a multiple of the value of @code{TARGET_CONST_ANCHOR}. 11236@code{TARGET_CONST_ANCHOR} should be the maximum positive value 11237accepted by immediate-add plus one. We currently assume that the 11238value of @code{TARGET_CONST_ANCHOR} is a power of 2. For example, on 11239MIPS, where add-immediate takes a 16-bit signed value, 11240@code{TARGET_CONST_ANCHOR} is set to @samp{0x8000}. The default value 11241is zero, which disables this optimization. @end deftypevr 11242 11243@hook TARGET_ATOMIC_TEST_AND_SET_TRUEVAL 11244