This is Info file gcc.info, produced by Makeinfo-1.49 from the input
file gcc.texi.

   This file documents the use and the internals of the GNU compiler.

   Copyright (C) 1988, 1989, 1992 Free Software Foundation, Inc.

   Permission is granted to make and distribute verbatim copies of this
manual provided the copyright notice and this permission notice are
preserved on all copies.

   Permission is granted to copy and distribute modified versions of
this manual under the conditions for verbatim copying, provided also
that the sections entitled "GNU General Public License" and "Protect
Your Freedom--Fight `Look And Feel'" are included exactly as in the
original, and provided that the entire resulting derived work is
distributed under the terms of a permission notice identical to this

   Permission is granted to copy and distribute translations of this
manual into another language, under the above conditions for modified
versions, except that the sections entitled "GNU General Public
License" and "Protect Your Freedom--Fight `Look And Feel'", and this
permission notice, may be included in translations approved by the Free
Software Foundation instead of in the original English.

File: gcc.info,  Node: Sending Patches,  Prev: Bug Reporting,  Up: Bugs

Sending Patches for GNU CC

   If you would like to write bug fixes or improvements for the GNU C
compiler, that is very helpful.  When you send your changes, please
follow these guidelines to avoid causing extra work for us in studying
the patches.

   If you don't follow these guidelines, your information might still be
useful, but using it will take extra work.  Maintaining GNU C is a lot
of work in the best of circumstances, and we can't keep up unless you do
your best to help.

   * Send an explanation with your changes of what problem they fix or
     what improvement they bring about.  For a bug fix, just include a
     copy of the bug report, and explain why the change fixes the bug.

     (Referring to a bug report is not as good as including it, because
     then we will have to look it up, and we have probably already
     deleted it if we've already fixed the bug.)

   * Always include a proper bug report for the problem you think you
     have fixed.  We need to convince ourselves that the change is
     right before installing it.  Even if it is right, we might have
     trouble judging it if we don't have a way to reproduce the problem.

   * Include all the comments that are appropriate to help people
     reading the source in the future understand why this change was

   * Don't mix together changes made for different reasons. Send them

     If you make two changes for separate reasons, then we might not
     want to install them both.  We might want to install just one.  If
     you send them all jumbled together in a single set of diffs, we
     have to do extra work to disentangle them--to figure out which
     parts of the change serve which purpose.  If we don't have time
     for this, we might have to ignore your changes entirely.

     If you send each change as soon as you have written it, with its
     own explanation, then the two changes never get tangled up, and we
     can consider each one properly without any extra work to
     disentangle them.

     Ideally, each change you send should be impossible to subdivide
     into parts that we might want to consider separately, because each
     of its parts gets its motivation from the other parts.

   * Send each change as soon as that change is finished.  Sometimes
     people think they are helping us by accumulating many changes to
     send them all together.  As explained above, this is absolutely
     the worst thing you could do.

     Since you should send each change separately, you might as well
     send it right away.  That gives us the option of installing it
     immediately if it is important.

   * Use `diff -c' to make your diffs.  Diffs without context are hard
     for us to install reliably.  More than that, they make it hard for
     us to study the diffs to decide whether we want to install them. 
     Unidiff format is better than contextless diffs, but not as easy
     to read as `-c' format.

     If you have GNU diff, use `diff -cp', which shows the name of the
     function that each change occurs in.

   * Write the change log entries for your changes.  We get lots of
     changes, and we don't have time to do all the change log writing

     Read the `ChangeLog' file to see what sorts of information to put
     in, and to learn the style that we use.  The purpose of the change
     log is to show people where to find what was changed.  So you need
     to be specific about what functions you changed; in large
     functions, it's often helpful to indicate where within the
     function the change was.

     On the other hand, once you have shown people where to find the
     change, you need not explain its purpose. Thus, if you add a new
     function, all you need to say about it is that it is new.  If you
     feel that the purpose needs explaining, it probably does--but the
     explanation will be much more useful if you put it in comments in
     the code.

     If you would like your name to appear in the header line for who
     made the change, send us the header line.

   * When you write the fix, keep in mind that we can't install a
     change that would break other systems.

     People often suggest fixing a problem by changing
     machine-independent files such as `toplev.c' to do something
     special that a particular system needs.  Sometimes it is totally
     obvious that such changes would break GNU CC for almost all users.
      We can't possibly make a change like that.  At best it might tell
     us how to write another patch that would solve the problem

     Sometimes people send fixes that *might* be an improvement in
     general--but it is hard to be sure of this.  It's hard to install
     such changes because we have to study them very carefully.  Of
     course, a good explanation of the reasoning by which you concluded
     the change was correct can help convince us.

     The safest changes are changes to the configuration files for a
     particular machine.  These are safe because they can't create new
     bugs on other machines.

     Please help us keep up with the workload by designing the patch in
     a form that is good to install.

File: gcc.info,  Node: Service,  Next: VMS,  Prev: Bugs,  Up: Top

How To Get Help with GNU CC

   If you need help installing, using or changing GNU CC, there are two
ways to find it:

   * Send a message to a suitable network mailing list.  First try
     `bug-gcc@prep.ai.mit.edu', and if that brings no response, try

   * Look in the service directory for someone who might help you for a
     fee. The service directory is found in the file named `SERVICE' in
     the GNU CC distribution.

File: gcc.info,  Node: VMS,  Next: Portability,  Prev: Service,  Up: Top

Using GNU CC on VMS

* Menu:

* Include Files and VMS::  Where the preprocessor looks for the include files.
* Global Declarations::    How to do globaldef, globalref and globalvalue with
                           GNU CC.
* VMS Misc::		   Misc information.

File: gcc.info,  Node: Include Files and VMS,  Next: Global Declarations,  Up: VMS

Include Files and VMS

   Due to the differences between the filesystems of Unix and VMS, GNU
CC attempts to translate file names in `#include' into names that VMS
will understand.  The basic strategy is to prepend a prefix to the
specification of the include file, convert the whole filename to a VMS
filename, and then try to open the file.  GNU CC tries various prefixes
one by one until one of them succeeds:

  1. The first prefix is the `GNU_CC_INCLUDE:' logical name: this is
     where GNU C header files are traditionally stored.  If you wish to
     store header files in non-standard locations, then you can assign
     the logical `GNU_CC_INCLUDE' to be a search list, where each
     element of the list is suitable for use with a rooted logical.

  2. The next prefix tried is `SYS$SYSROOT:[SYSLIB.]'.  This is where
     VAX-C header files are traditionally stored.

  3. If the include file specification by itself is a valid VMS
     filename, the preprocessor then uses this name with no prefix in
     an attempt to open the include file.

  4. If the file specification is not a valid VMS filename (i.e. does
     not contain a device or a directory specifier, and contains a `/'
     character), the preprocessor tries to convert it from Unix syntax
     to VMS syntax.

     Conversion works like this: the first directory name becomes a
     device, and the rest of the directories are converted into
     VMS-format directory names.  For example, `X11/foobar.h' is
     translated to `X11:[000000]foobar.h' or `X11:foobar.h', whichever
     one can be opened.  This strategy allows you to assign a logical
     name to point to the actual location of the header files.

  5. If none of these strategies succeeds, the `#include' fails.

   Include directives of the form:

     #include foobar

are a common source of incompatibility between VAX-C and GNU CC.  VAX-C
treats this much like a standard `#include <foobar.h>' directive. That
is incompatible with the ANSI C behavior implemented by GNU CC: to
expand the name `foobar' as a macro.  Macro expansion should eventually
yield one of the two standard formats for `#include':

     #include "FILE"
     #include <FILE>

   If you have this problem, the best solution is to modify the source
to convert the `#include' directives to one of the two standard forms.
That will work with either compiler.  If you want a quick and dirty fix,
define the file names as macros with the proper expansion, like this:

     #define stdio <stdio.h>

This will work, as long as the name doesn't conflict with anything else
in the program.

   Another source of incompatibility is that VAX-C assumes that:

     #include "foobar"

is actually asking for the file `foobar.h'.  GNU CC does not make this
assumption, and instead takes what you ask for literally; it tries to
read the file `foobar'.  The best way to avoid this problem is to
always specify the desired file extension in your include directives.

   GNU CC for VMS is distributed with a set of include files that is
sufficient to compile most general purpose programs.  Even though the
GNU CC distribution does not contain header files to define constants
and structures for some VMS system-specific functions, there is no
reason why you cannot use GNU CC with any of these functions.  You first
may have to generate or create header files, either by using the public
domain utility `UNSDL' (which can be found on a DECUS tape), or by
extracting the relevant modules from one of the system macro libraries,
and using an editor to construct a C header file.

   A `#include' file name cannot contain a DECNET node name.  The
preprocessor reports an I/O error if you attempt to use a node name,
whether explicitly, or implicitly via a logical name.

File: gcc.info,  Node: Global Declarations,  Next: VMS Misc,  Prev: Include Files and VMS,  Up: VMS

Global Declarations and VMS

   GNU CC does not provide the `globalref', `globaldef' and
`globalvalue' keywords of VAX-C.  You can get the same effect with an
obscure feature of GAS, the GNU assembler.  (This requires GAS version
1.39 or later.)  The following macros allow you to use this feature in
a fairly natural way:

     #ifdef __GNUC__
     #define GLOBALREF(TYPE,NAME)                      \
       TYPE NAME                                       \
       asm ("_$$PsectAttributes_GLOBALSYMBOL$$" #NAME)
     #define GLOBALDEF(TYPE,NAME,VALUE)                \
       TYPE NAME                                       \
       asm ("_$$PsectAttributes_GLOBALSYMBOL$$" #NAME) \
         = VALUE
     #define GLOBALVALUEREF(TYPE,NAME)                 \
       const TYPE NAME[1]                              \
       asm ("_$$PsectAttributes_GLOBALVALUE$$" #NAME)
     #define GLOBALVALUEDEF(TYPE,NAME,VALUE)           \
       const TYPE NAME[1]                              \
       asm ("_$$PsectAttributes_GLOBALVALUE$$" #NAME)  \
         = {VALUE}
     #define GLOBALREF(TYPE,NAME) \
       globalref TYPE NAME
       globaldef TYPE NAME = VALUE
       globalvalue TYPE NAME = VALUE
       globalvalue TYPE NAME

(The `_$$PsectAttributes_GLOBALSYMBOL' prefix at the start of the name
is removed by the assembler, after it has modified the attributes of
the symbol).  These macros are provided in the VMS binaries
distribution in a header file `GNU_HACKS.H'.  An example of the usage

     GLOBALREF (int, ijk);
     GLOBALDEF (int, jkl, 0);

   The macros `GLOBALREF' and `GLOBALDEF' cannot be used
straightforwardly for arrays, since there is no way to insert the array
dimension into the declaration at the right place.  However, you can
declare an array with these macros if you first define a typedef for the
array type, like this:

     typedef int intvector[10];
     GLOBALREF (intvector, foo);

   Array and structure initializers will also break the macros; you can
define the initializer to be a macro of its own, or you can expand the
`GLOBALDEF' macro by hand.  You may find a case where you wish to use
the `GLOBALDEF' macro with a large array, but you are not interested in
explicitly initializing each element of the array.  In such cases you
can use an initializer like: `{0,}', which will initialize the entire
array to `0'.

   A shortcoming of this implementation is that a variable declared with
`GLOBALVALUEREF' or `GLOBALVALUEDEF' is always an array.  For example,
the declaration:

     GLOBALVALUEREF(int, ijk);

declares the variable `ijk' as an array of type `int [1]'. This is done
because a globalvalue is actually a constant; its "value" is what the
linker would normally consider an address.  That is not how an integer
value works in C, but it is how an array works.  So treating the symbol
as an array name gives consistent results--with the exception that the
value seems to have the wrong type.  *Don't try to access an element of
the array.*  It doesn't have any elements. The array "address" may not
be the address of actual storage.

   The fact that the symbol is an array may lead to warnings where the
variable is used.  Insert type casts to avoid the warnings.  Here is an
example; it takes advantage of the ANSI C feature allowing macros that
expand to use the same name as the macro itself.

     GLOBALVALUEREF (int, ss$_normal);
     GLOBALVALUEDEF (int, xyzzy,123);
     #ifdef __GNUC__
     #define ss$_normal ((int) ss$_normal)
     #define xyzzy ((int) xyzzy)

   Don't use `globaldef' or `globalref' with a variable whose type is
an enumeration type; this is not implemented.  Instead, make the
variable an integer, and use a `globalvaluedef' for each of the
enumeration values.  An example of this would be:

     #ifdef __GNUC__
     GLOBALDEF (int, color, 0);
     GLOBALVALUEDEF (int, RED, 0);
     enum globaldef color {RED, BLUE, GREEN = 3};

File: gcc.info,  Node: VMS Misc,  Prev: Global Declarations,  Up: VMS

Other VMS Issues

   GNU CC automatically arranges for `main' to return 1 by default if
you fail to specify an explicit return value.  This will be interpreted
by VMS as a status code indicating a normal successful completion.
Version 1 of GNU CC did not provide this default.

   GNU CC on VMS works only with the GNU assembler, GAS.  You need
version 1.37 or later of GAS in order to produce value debugging
information for the VMS debugger.  Use the ordinary VMS linker with the
object files produced by GAS.

   Under previous versions of GNU CC, the generated code would
occasionally give strange results when linked to the sharable `VAXCRTL'
library. Now this should work.

   A caveat for use of `const' global variables: the `const' modifier
must be specified in every external declaration of the variable in all
of the source files that use that variable.  Otherwise the linker will
issue warnings about conflicting attributes for the variable.  Your
program will still work despite the warnings, but the variable will be
placed in writable storage.

   Although the VMS linker does distinguish between upper and lower case
letters in global symbols, most VMS compilers convert all such symbols
into upper case and most run-time library routines also have upper case
names.  To be able to reliably call such routines, GNU CC (by means of
the assembler GAS) converts global symbols into upper case like other
VMS compilers.  However, since the usual practice in C is to distinguish
case, GNU CC (via GAS) tries to preserve usual C behavior by augmenting
each name that is not all lower case.  This means truncating the name
to at most 23 characters and then adding more characters at the end
which encode the case pattern of those 23.   Names which contain at
least one dollar sign are an exception; they are converted directly into
upper case without augmentation.

   Name augmentation yields bad results for programs that use
precompiled libraries (such as Xlib) which were generated by another
compiler.  You can use the compiler option `/NOCASE_HACK' to inhibit
augmentation; it makes external C functions and variables
case-independent as is usual on VMS.  Alternatively, you could write
all references to the functions and variables in such libraries using
lower case; this will work on VMS, but is not portable to other
systems.  The compiler option `/NAMES' also provides control over
global name handling.

   Function and variable names are handled somewhat differently with GNU
C++.  The GNU C++ compiler performs "name mangling" on function names,
which means that it adds information to the function name to describe
the data types of the arguments that the function takes. One result of
this is that the name of a function can become very long. Since the VMS
linker only recognizes the first 31 characters in a name, special
action is taken to ensure that each function and variable has a unique
name that can be represented in 31 characters.

   If the name (plus a name augmentation, if required) is less than 32
characters in length, then no special action is performed. If the name
is longer than 31 characters, the assembler (GAS) will generate a hash
string based upon the function name, truncate the function name to 23
characters, and append the hash string to the truncated name.  If the
`/VERBOSE' compiler option is used, the assembler will print both the
full and truncated names of each symbol that is truncated.

   The `/NOCASE_HACK' compiler option should not be used when you are
compiling programs that use libg++. libg++ has several instances of
objects (i.e.  `Filebuf' and `filebuf') which become indistinguishable
in a case-insensitive environment.  This leads to cases where you need
to inhibit augmentation selectively (if you were using libg++ and Xlib
in the same program, for example).  There is no special feature for
doing this, but you can get the result by defining a macro for each
mixed case symbol for which you wish to inhibit augmentation.  The
macro should expand into the lower case equivalent of itself.  For

     #define StuDlyCapS studlycaps

   These macro definitions can be placed in a header file to minimize
the number of changes to your source code.

File: gcc.info,  Node: Portability,  Next: Interface,  Prev: VMS,  Up: Top

GNU CC and Portability

   The main goal of GNU CC was to make a good, fast compiler for
machines in the class that the GNU system aims to run on: 32-bit
machines that address 8-bit bytes and have several general registers. 
Elegance, theoretical power and simplicity are only secondary.

   GNU CC gets most of the information about the target machine from a
machine description which gives an algebraic formula for each of the
machine's instructions.  This is a very clean way to describe the
target.  But when the compiler needs information that is difficult to
express in this fashion, I have not hesitated to define an ad-hoc
parameter to the machine description.  The purpose of portability is to
reduce the total work needed on the compiler; it was not of interest
for its own sake.

   GNU CC does not contain machine dependent code, but it does contain
code that depends on machine parameters such as endianness (whether the
most significant byte has the highest or lowest address of the bytes in
a word) and the availability of autoincrement addressing.  In the
RTL-generation pass, it is often necessary to have multiple strategies
for generating code for a particular kind of syntax tree, strategies
that are usable for different combinations of parameters.  Often I have
not tried to address all possible cases, but only the common ones or
only the ones that I have encountered. As a result, a new target may
require additional strategies.  You will know if this happens because
the compiler will call `abort'.  Fortunately, the new strategies can be
added in a machine-independent fashion, and will affect only the target
machines that need them.

File: gcc.info,  Node: Interface,  Next: Passes,  Prev: Portability,  Up: Top

Interfacing to GNU CC Output

   GNU CC is normally configured to use the same function calling
convention normally in use on the target system.  This is done with the
machine-description macros described (*note Target Macros::.).

   However, returning of structure and union values is done differently
on some target machines.  As a result, functions compiled with PCC
returning such types cannot be called from code compiled with GNU CC,
and vice versa.  This does not cause trouble often because few Unix
library routines return structures or unions.

   GNU CC code returns structures and unions that are 1, 2, 4 or 8 bytes
long in the same registers used for `int' or `double' return values. 
(GNU CC typically allocates variables of such types in registers also.)
 Structures and unions of other sizes are returned by storing them into
an address passed by the caller (usually in a register).  The
machine-description macros `STRUCT_VALUE' and `STRUCT_INCOMING_VALUE'
tell GNU CC where to pass this address.

   By contrast, PCC on most target machines returns structures and
unions of any size by copying the data into an area of static storage,
and then returning the address of that storage as if it were a pointer
value. The caller must copy the data from that memory area to the place
where the value is wanted.  This is slower than the method used by GNU
CC, and fails to be reentrant.

   On some target machines, such as RISC machines and the 80386, the
standard system convention is to pass to the subroutine the address of
where to return the value.  On these machines, GNU CC has been
configured to be compatible with the standard compiler, when this method
is used.  It may not be compatible for structures of 1, 2, 4 or 8 bytes.

   GNU CC uses the system's standard convention for passing arguments. 
On some machines, the first few arguments are passed in registers; in
others, all are passed on the stack.  It would be possible to use
registers for argument passing on any machine, and this would probably
result in a significant speedup.  But the result would be complete
incompatibility with code that follows the standard convention.  So this
change is practical only if you are switching to GNU CC as the sole C
compiler for the system.  We may implement register argument passing on
certain machines once we have a complete GNU system so that we can
compile the libraries with GNU CC.

   On some machines (particularly the Sparc), certain types of arguments
are passed "by invisible reference".  This means that the value is
stored in memory, and the address of the memory location is passed to
the subroutine.

   If you use `longjmp', beware of automatic variables.  ANSI C says
that automatic variables that are not declared `volatile' have undefined
values after a `longjmp'.  And this is all GNU CC promises to do,
because it is very difficult to restore register variables correctly,
and one of GNU CC's features is that it can put variables in registers
without your asking it to.

   If you want a variable to be unaltered by `longjmp', and you don't
want to write `volatile' because old C compilers don't accept it, just
take the address of the variable.  If a variable's address is ever
taken, even if just to compute it and ignore it, then the variable
cannot go in a register:

       int careful;

   Code compiled with GNU CC may call certain library routines.  Most of
them handle arithmetic for which there are no instructions.  This
includes multiply and divide on some machines, and floating point
operations on any machine for which floating point support is disabled
with `-msoft-float'.  Some standard parts of the C library, such as
`bcopy' or `memcpy', are also called automatically.  The usual function
call interface is used for calling the library routines.

   These library routines should be defined in the library `libgcc.a',
which GNU CC automatically searches whenever it links a program.  On
machines that have multiply and divide instructions, if hardware
floating point is in use, normally `libgcc.a' is not needed, but it is
searched just in case.

   Each arithmetic function is defined in `libgcc1.c' to use the
corresponding C arithmetic operator.  As long as the file is compiled
with another C compiler, which supports all the C arithmetic operators,
this file will work portably.  However, `libgcc1.c' does not work if
compiled with GNU CC, because each arithmetic function would compile
into a call to itself!

File: gcc.info,  Node: Passes,  Next: RTL,  Prev: Interface,  Up: Top

Passes and Files of the Compiler

   The overall control structure of the compiler is in `toplev.c'.  This
file is responsible for initialization, decoding arguments, opening and
closing files, and sequencing the passes.

   The parsing pass is invoked only once, to parse the entire input. 
The RTL intermediate code for a function is generated as the function
is parsed, a statement at a time.  Each statement is read in as a
syntax tree and then converted to RTL; then the storage for the tree
for the statement is reclaimed.  Storage for types (and the expressions
for their sizes), declarations, and a representation of the binding
contours and how they nest, remain until the function is finished being
compiled; these are all needed to output the debugging information.

   Each time the parsing pass reads a complete function definition or
top-level declaration, it calls the function `rest_of_compilation' or
`rest_of_decl_compilation' in `toplev.c', which are responsible for all
further processing necessary, ending with output of the assembler
language.  All other compiler passes run, in sequence, within
`rest_of_compilation'. When that function returns from compiling a
function definition, the storage used for that function definition's
compilation is entirely freed, unless it is an inline function (*note

   Here is a list of all the passes of the compiler and their source
files. Also included is a description of where debugging dumps can be
requested with `-d' options.

   * Parsing.  This pass reads the entire text of a function definition,
     constructing partial syntax trees.  This and RTL generation are no
     longer truly separate passes (formerly they were), but it is
     easier to think of them as separate.

     The tree representation does not entirely follow C syntax, because
     it is intended to support other languages as well.

     Language-specific data type analysis is also done in this pass,
     and every tree node that represents an expression has a data type
     attached. Variables are represented as declaration nodes.

     Constant folding and some arithmetic simplifications are also done
     during this pass.

     The language-independent source files for parsing are
     `stor-layout.c', `fold-const.c', and `tree.c'. There are also
     header files `tree.h' and `tree.def' which define the format of
     the tree representation.

     The source files for parsing C are `c-parse.y', `c-decl.c',
     `c-typeck.c', `c-convert.c', `c-lang.c', and `c-aux-info.c' along
     with header files `c-lex.h', and `c-tree.h'.

     The source files for parsing C++ are `cp-parse.y', `cp-class.c',
     `cp-cvt.c', `cp-decl.c', `cp-decl2.c', `cp-dem.c', `cp-except.c',
     `cp-expr.c', `cp-init.c', `cp-lex.c', `cp-method.c', `cp-ptree.c',
     `cp-search.c', `cp-tree.c', `cp-type2.c', and `cp-typeck.c', along
     with header files `cp-tree.def', `cp-tree.h', and `cp-decl.h'.

     The special source files for parsing Objective C are
     `objc-parse.y', `objc-actions.c', `objc-tree.def', and
     `objc-actions.h'.  Certain C-specific files are used for this as

     The file `c-common.c' is also used for all of the above languages.

   * RTL generation.  This is the conversion of syntax tree into RTL
     code. It is actually done statement-by-statement during parsing,
     but for most purposes it can be thought of as a separate pass.

     This is where the bulk of target-parameter-dependent code is found,
     since often it is necessary for strategies to apply only when
     certain standard kinds of instructions are available.  The purpose
     of named instruction patterns is to provide this information to
     the RTL generation pass.

     Optimization is done in this pass for `if'-conditions that are
     comparisons, boolean operations or conditional expressions.  Tail
     recursion is detected at this time also.  Decisions are made about
     how best to arrange loops and how to output `switch' statements.

     The source files for RTL generation include `stmt.c',
     `function.c', `expr.c', `calls.c', `explow.c', `expmed.c',
     `optabs.c' and `emit-rtl.c'.  Also, the file `insn-emit.c',
     generated from the machine description by the program `genemit',
     is used in this pass.  The header file `expr.h' is used for
     communication within this pass.

     The header files `insn-flags.h' and `insn-codes.h', generated from
     the machine description by the programs `genflags' and `gencodes',
     tell this pass which standard names are available for use and
     which patterns correspond to them.

     Aside from debugging information output, none of the following
     passes refers to the tree structure representation of the function
     (only part of which is saved).

     The decision of whether the function can and should be expanded
     inline in its subsequent callers is made at the end of rtl
     generation.  The function must meet certain criteria, currently
     related to the size of the function and the types and number of
     parameters it has.  Note that this function may contain loops,
     recursive calls to itself (tail-recursive functions can be
     inlined!), gotos, in short, all constructs supported by GNU CC. 
     The file `integrate.c' contains the code to save a function's rtl
     for later inlining and to inline that rtl when the function is
     called.  The header file `integrate.h' is also used for this

     The option `-dr' causes a debugging dump of the RTL code after
     this pass.  This dump file's name is made by appending `.rtl' to
     the input file name.

   * Jump optimization.  This pass simplifies jumps to the following
     instruction, jumps across jumps, and jumps to jumps.  It deletes
     unreferenced labels and unreachable code, except that unreachable
     code that contains a loop is not recognized as unreachable in this
     pass. (Such loops are deleted later in the basic block analysis.) 
     It also converts some code originally written with jumps into
     sequences of instructions that directly set values from the
     results of comparisons, if the machine has such instructions.

     Jump optimization is performed two or three times.  The first time
     is immediately following RTL generation.  The second time is after
     CSE, but only if CSE says repeated jump optimization is needed. 
     The last time is right before the final pass.  That time,
     cross-jumping and deletion of no-op move instructions are done
     together with the optimizations described above.

     The source file of this pass is `jump.c'.

     The option `-dj' causes a debugging dump of the RTL code after
     this pass is run for the first time.  This dump file's name is
     made by appending `.jump' to the input file name.

   * Register scan.  This pass finds the first and last use of each
     register, as a guide for common subexpression elimination.  Its
     source is in `regclass.c'.

   * Jump threading.  This pass detects a condition jump that branches
     to an identical or inverse test.  Such jumps can be `threaded'
     through the second conditional test.  The source code for this
     pass is in `jump.c'.  This optimization is only performed if
     `-fthread-jumps' is enabled.

   * Common subexpression elimination.  This pass also does constant
     propagation.  Its source file is `cse.c'.  If constant propagation
     causes conditional jumps to become unconditional or to become
     no-ops, jump optimization is run again when CSE is finished.

     The option `-ds' causes a debugging dump of the RTL code after
     this pass.  This dump file's name is made by appending `.cse' to
     the input file name.

   * Loop optimization.  This pass moves constant expressions out of
     loops, and optionally does strength-reduction and loop unrolling
     as well. Its source files are `loop.c' and `unroll.c', plus the
     header `loop.h' used for communication between them.  Loop
     unrolling uses some functions in `integrate.c' and the header

     The option `-dL' causes a debugging dump of the RTL code after
     this pass.  This dump file's name is made by appending `.loop' to
     the input file name.

   * If `-frerun-cse-after-loop' was enabled, a second common
     subexpression elimination pass is performed after the loop
     optimization pass.  Jump threading is also done again at this time
     if it was specified.

     The option `-dt' causes a debugging dump of the RTL code after
     this pass.  This dump file's name is made by appending `.cse2' to
     the input file name.

   * Stupid register allocation is performed at this point in a
     nonoptimizing compilation.  It does a little data flow analysis as
     well.  When stupid register allocation is in use, the next pass
     executed is the reloading pass; the others in between are skipped.
     The source file is `stupid.c'.

   * Data flow analysis (`flow.c').  This pass divides the program into
     basic blocks (and in the process deletes unreachable loops); then
     it computes which pseudo-registers are live at each point in the
     program, and makes the first instruction that uses a value point at
     the instruction that computed the value.

     This pass also deletes computations whose results are never used,
     and combines memory references with add or subtract instructions
     to make autoincrement or autodecrement addressing.

     The option `-df' causes a debugging dump of the RTL code after
     this pass.  This dump file's name is made by appending `.flow' to
     the input file name.  If stupid register allocation is in use, this
     dump file reflects the full results of such allocation.

   * Instruction combination (`combine.c').  This pass attempts to
     combine groups of two or three instructions that are related by
     data flow into single instructions.  It combines the RTL
     expressions for the instructions by substitution, simplifies the
     result using algebra, and then attempts to match the result
     against the machine description.

     The option `-dc' causes a debugging dump of the RTL code after
     this pass.  This dump file's name is made by appending `.combine'
     to the input file name.

   * Instruction scheduling (`sched.c').  This pass looks for
     instructions whose output will not be available by the time that
     it is used in subsequent instructions.  (Memory loads and floating
     point instructions often have this behavior on RISC machines).  It
     re-orders instructions within a basic block to try to separate the
     definition and use of items that otherwise would cause pipeline

     Instruction scheduling is performed twice.  The first time is
     immediately after instruction combination and the second is
     immediately after reload.

     The option `-dS' causes a debugging dump of the RTL code after this
     pass is run for the first time.  The dump file's name is made by
     appending `.sched' to the input file name.

   * Register class preferencing.  The RTL code is scanned to find out
     which register class is best for each pseudo register.  The source
     file is `regclass.c'.

   * Local register allocation (`local-alloc.c').  This pass allocates
     hard registers to pseudo registers that are used only within one
     basic block.  Because the basic block is linear, it can use fast
     and powerful techniques to do a very good job.

     The option `-dl' causes a debugging dump of the RTL code after
     this pass.  This dump file's name is made by appending `.lreg' to
     the input file name.

   * Global register allocation (`global.c').  This pass allocates hard
     registers for the remaining pseudo registers (those whose life
     spans are not contained in one basic block).

   * Reloading.  This pass renumbers pseudo registers with the hardware
     registers numbers they were allocated.  Pseudo registers that did
     not get hard registers are replaced with stack slots.  Then it
     finds instructions that are invalid because a value has failed to
     end up in a register, or has ended up in a register of the wrong
     kind.  It fixes up these instructions by reloading the
     problematical values temporarily into registers.  Additional
     instructions are generated to do the copying.

     The reload pass also optionally eliminates the frame pointer and
     inserts instructions to save and restore call-clobbered registers
     around calls.

     Source files are `reload.c' and `reload1.c', plus the header
     `reload.h' used for communication between them.

     The option `-dg' causes a debugging dump of the RTL code after
     this pass.  This dump file's name is made by appending `.greg' to
     the input file name.

   * Instruction scheduling is repeated here to try to avoid pipeline
     stalls due to memory loads generated for spilled pseudo registers.

     The option `-dR' causes a debugging dump of the RTL code after
     this pass.  This dump file's name is made by appending `.sched2'
     to the input file name.

   * Jump optimization is repeated, this time including cross-jumping
     and deletion of no-op move instructions.

     The option `-dJ' causes a debugging dump of the RTL code after
     this pass.  This dump file's name is made by appending `.jump2' to
     the input file name.

   * Delayed branch scheduling.  This optional pass attempts to find
     instructions that can go into the delay slots of other
     instructions, usually jumps and calls.  The source file name is

     The option `-dd' causes a debugging dump of the RTL code after
     this pass.  This dump file's name is made by appending `.dbr' to
     the input file name.

   * Conversion from usage of some hard registers to usage of a register
     stack may be done at this point.  Currently, this is supported only
     for the floating-point registers of the Intel 80387 coprocessor.  
     The source file name is `reg-stack.c'.

     The options `-dk' causes a debugging dump of the RTL code after
     this pass.  This dump file's name is made by appending `.stack' to
     the input file name.

   * Final.  This pass outputs the assembler code for the function.  It
     is also responsible for identifying spurious test and compare
     instructions.  Machine-specific peephole optimizations are
     performed at the same time.  The function entry and exit sequences
     are generated directly as assembler code in this pass; they never
     exist as RTL.

     The source files are `final.c' plus `insn-output.c'; the latter is
     generated automatically from the machine description by the tool
     `genoutput'.  The header file `conditions.h' is used for
     communication between these files.

   * Debugging information output.  This is run after final because it
     must output the stack slot offsets for pseudo registers that did
     not get hard registers.  Source files are `dbxout.c' for DBX
     symbol table format, `sdbout.c' for SDB symbol table format, and
     `dwarfout.c' for DWARF symbol table format.

   Some additional files are used by all or many passes:

   * Every pass uses `machmode.def' and `machmode.h' which define the
     machine modes.

   * Several passes use `real.h', which defines the default
     representation of floating point constants and how to operate on

   * All the passes that work with RTL use the header files `rtl.h' and
     `rtl.def', and subroutines in file `rtl.c'.  The tools `gen*' also
     use these files to read and work with the machine description RTL.

   * Several passes refer to the header file `insn-config.h' which
     contains a few parameters (C macro definitions) generated
     automatically from the machine description RTL by the tool

   * Several passes use the instruction recognizer, which consists of
     `recog.c' and `recog.h', plus the files `insn-recog.c' and
     `insn-extract.c' that are generated automatically from the machine
     description by the tools `genrecog' and `genextract'.

   * Several passes use the header files `regs.h' which defines the
     information recorded about pseudo register usage, and
     `basic-block.h' which defines the information recorded about basic

   * `hard-reg-set.h' defines the type `HARD_REG_SET', a bit-vector
     with a bit for each hard register, and some macros to manipulate
     it. This type is just `int' if the machine has few enough hard
     registers; otherwise it is an array of `int' and some of the
     macros expand into loops.

   * Several passes use instruction attributes.  A definition of the
     attributes defined for a particular machine is in file
     `insn-attr.h', which is generated from the machine description by
     the program `genattr'.  The file `insn-attrtab.c' contains
     subroutines to obtain the attribute values for insns.  It is
     generated from the machine description by the program `genattrtab'.

File: gcc.info,  Node: RTL,  Next: Machine Desc,  Prev: Passes,  Up: Top

RTL Representation

   Most of the work of the compiler is done on an intermediate
representation called register transfer language.  In this language,
the instructions to be output are described, pretty much one by one, in
an algebraic form that describes what the instruction does.

   RTL is inspired by Lisp lists.  It has both an internal form, made
up of structures that point at other structures, and a textual form
that is used in the machine description and in printed debugging dumps.
 The textual form uses nested parentheses to indicate the pointers in
the internal form.

* Menu:

* RTL Objects::       Expressions vs vectors vs strings vs integers.
* Accessors::         Macros to access expression operands or vector elts.
* Flags::             Other flags in an RTL expression.
* Machine Modes::     Describing the size and format of a datum.
* Constants::         Expressions with constant values.
* Regs and Memory::   Expressions representing register contents or memory.
* Arithmetic::        Expressions representing arithmetic on other expressions.
* Comparisons::       Expressions representing comparison of expressions.
* Bit Fields::        Expressions representing bit-fields in memory or reg.
* Conversions::       Extending, truncating, floating or fixing.
* RTL Declarations::  Declaring volatility, constancy, etc.
* Side Effects::      Expressions for storing in registers, etc.
* Incdec::            Embedded side-effects for autoincrement addressing.
* Assembler::         Representing `asm' with operands.
* Insns::             Expression types for entire insns.
* Calls::             RTL representation of function call insns.
* Sharing::           Some expressions are unique; others *must* be copied.

File: gcc.info,  Node: RTL Objects,  Next: Accessors,  Prev: RTL,  Up: RTL

RTL Object Types

   RTL uses five kinds of objects: expressions, integers, wide integers,
strings and vectors.  Expressions are the most important ones.  An RTL
expression ("RTX", for short) is a C structure, but it is usually
referred to with a pointer; a type that is given the typedef name `rtx'.

   An integer is simply an `int'; their written form uses decimal
digits. A wide integer is an integral object whose type is
`HOST_WIDE_INT' (*note Config::.); their written form used decimal

   A string is a sequence of characters.  In core it is represented as a
`char *' in usual C fashion, and it is written in C syntax as well.
However, strings in RTL may never be null.  If you write an empty
string in a machine description, it is represented in core as a null
pointer rather than as a pointer to a null character.  In certain
contexts, these null pointers instead of strings are valid.  Within RTL
code, strings are most commonly found inside `symbol_ref' expressions,
but they appear in other contexts in the RTL expressions that make up
machine descriptions.

   A vector contains an arbitrary number of pointers to expressions. 
The number of elements in the vector is explicitly present in the
vector. The written form of a vector consists of square brackets
(`[...]') surrounding the elements, in sequence and with whitespace
separating them.  Vectors of length zero are not created; null pointers
are used instead.

   Expressions are classified by "expression codes" (also called RTX
codes).  The expression code is a name defined in `rtl.def', which is
also (in upper case) a C enumeration constant.  The possible expression
codes and their meanings are machine-independent.  The code of an RTX
can be extracted with the macro `GET_CODE (X)' and altered with

   The expression code determines how many operands the expression
contains, and what kinds of objects they are.  In RTL, unlike Lisp, you
cannot tell by looking at an operand what kind of object it is. 
Instead, you must know from its context--from the expression code of
the containing expression. For example, in an expression of code
`subreg', the first operand is to be regarded as an expression and the
second operand as an integer.  In an expression of code `plus', there
are two operands, both of which are to be regarded as expressions.  In
a `symbol_ref' expression, there is one operand, which is to be
regarded as a string.

   Expressions are written as parentheses containing the name of the
expression type, its flags and machine mode if any, and then the
operands of the expression (separated by spaces).

   Expression code names in the `md' file are written in lower case,
but when they appear in C code they are written in upper case.  In this
manual, they are shown as follows: `const_int'.

   In a few contexts a null pointer is valid where an expression is
normally wanted.  The written form of this is `(nil)'.