1 Introduction | What is this program for? | |
2 Installation | How to install it? | |
3 Usage | How to use ‘inlines’ and ‘FD2InLine’? | |
4 Rebuilding | How to recompile it? | |
5 Internals | How do ‘inlines’ work? | |
6 Authors | Or should it be called “History”? | |
7 Index | Concept index. |
[ << ] | [ < ] | [ Up ] | [ > ] | [ >> ] | [Top] | [Contents] | [Index] | [ ? ] |
‘FD2InLine’ is useful if you want to use ‘GCC’ for Amiga-specific development and would like to make very efficient calls to Amiga shared libraries’ functions.
Format of calls to Amiga shared libraries’ functions differs substantially from the default format of C compilers (see section Background). Therefore, some tricks are necessary if you want to use these functions.
‘FD2InLine’ is a parser that converts ‘fd’ files and ‘clib’ files to ‘GCC inlines’.
‘fd’ and ‘clib’ files contain information about functions in shared libraries (see section Background).
‘FD2InLine’ reads these two files and outputs a file containing information from input files merged in a special format, suitable to use with ‘GCC’ compiler.
This output file contains so-called inlines — one for each function entry. Thanks to them ‘GCC’ can produce very efficient code for Amiga shared libraries’ functions calls.
Note: the term ‘inlines’ is misleading — ‘FD2InLine’ does not use
__inline
feature of ‘GCC’ any longer (see section New format).
[ << ] | [ < ] | [ Up ] | [ > ] | [ >> ] | [Top] | [Contents] | [Index] | [ ? ] |
I assume you have ‘fd2inline-{No value for `version'}-bin.lha’ archive.
If you use recent release of ‘GCC’, you might not need to install anything. Starting with ‘GCC’ 2.7.2, new format (see section New format) of ‘inlines’ should be available with the compiler. However, the separate ‘fd2inline-{No value for `version'}-bin.lha’ archive will always contain the latest version of ‘FD2InLine’ and ‘inlines’, which might not be the truth in case of ‘ADE’ or ‘Aminet’ distributions.
Installation is very easy, so I didn’t bother writing an Installer script :-).
If you have an older version of ‘inlines’ installed, please remove it now, or you might later have problems. Typically, you’ll have to remove the following subdirectories of ‘os-include’ directory: ‘inline’, ‘pragmas’ and ‘proto’.
Next, please change current directory to ‘ADE:’ and simply extract ‘fd2inline-{No value for `version'}-bin.lha’ archive. This should install everything in right places. More precisely, ‘inlines’ of standard system libraries and devices will go to ‘include/inline’ directory and ‘inlines’ of 3rd party libraries will go to ‘local/include/inline’ directory. ‘fd2inline’ executable will go to ‘bin’ directory and AmigaGuide documentation to ‘guide’.
[ << ] | [ < ] | [ Up ] | [ > ] | [ >> ] | [Top] | [Contents] | [Index] | [ ? ] |
This chapter describes two aspects of using ‘FD2InLine’:
3.1 Using inlines | Making efficient function calls | |
3.2 Using fd2inline | Creating ‘inlines’ |
[ << ] | [ < ] | [ Up ] | [ > ] | [ >> ] | [Top] | [Contents] | [Index] | [ ? ] |
Using ‘inlines’ is very simple. If you want to use a library called ‘foo.library’ (or a device called ‘bar.device’), just include file ‘<proto/foo.h>’ (‘<proto/bar.h>’) and that’s it. For example:
#include <proto/dos.h> int main(void) { Delay(100); /* Wait for 2 seconds */ }
Please always include ‘proto’ files, not ‘inline’ files - ‘proto’ files often fix some incompatibilities between system headers and ‘GCC’. Besides, this technique makes your code more portable across various Amiga compilers.
There are a few preprocessor symbols which alter the behaviour of ‘proto’ and ‘inlines’ files:
__NOLIBBASE__
By default, ‘proto’ files make an external declaration of library
base pointer. You can disable this behaviour by defining
__NOLIBBASE__
before including a ‘proto’ file.
__CONSTLIBBASEDECL__
The external declarations described above declare plain pointer
variables. The disadvantage of this is that library base variable has
to be reloaded every time some function is called. If you redefine
__CONSTLIBBASEDECL__
to const
, less reloading will be
necessary, so better code will be produced. However, declaring a
variable as const
makes altering it impossible, so some dirty
hacks are necessary (like defining variable as plain in one file and
altering it only there. This won’t work with base relative
code, though).
<library>_BASE_NAME
Function definitions in ‘inline’ files refer to library base
variable through <library>_BASE_NAME
symbol
(AMIGAGUIDE_BASE_NAME
for ‘amigaguide.library’, for
example). On top of ‘inlines’ file, this symbol is redefined to
appropriate library base variable name (AmigaGuideBase
, for
example), unless it’s already defined. This way, you can make
inlines use a field of a structure as a library base, for example.
NO_INLINE_STDARG
This symbol prevents defining of inline macros for varargs functions (see section Old format).
_USEOLDEXEC_
This one is used in ‘proto/exec.h’ only. Unlike in ‘SAS/C’,
‘proto/exec.h’ uses SysBase
variable as ‘Exec’
library base by default. This is usually faster than direct
dereferencing of 0x00000004
(see section Background), since it
doesn’t require reading from ‘CHIP’ memory (things might be even
worse if you use ‘Enforcer’ or ‘CyberGuard’, which protect
low memory region). However, in some low-level cases (like startup
code) you might prefer dereferencing 0x00000004
. To do this,
define _USEOLDEXEC_
before including ‘proto/exec.h’.
[ << ] | [ < ] | [ Up ] | [ > ] | [ >> ] | [Top] | [Contents] | [Index] | [ ? ] |
You invoke ‘FD2InLine’ by writing:
‘fd2inline’ [options] fd-file clib-file [[-o] output-file]
Variables have the following meaning:
A file name of an input ‘fd’ file.
A file name of an input ‘clib’ file.
A file name of an output ‘inlines’ file. If it is not specified (or if ‘-’ is specified), standard output will be used instead. The file name can be preceded with a ‘-o’, for compatibility with most UN*X software.
The following options can be specified (anywhere on the command line):
Produce new format ‘inlines’.
Produce old format ‘inlines’.
Produce library stubs.
See section Internals, for more information.
Example:
fd2inline FD:exec_lib.fd ADE:os-include/clib/exec_protos.h -o ADE:include/inline/exec.h
This will build file ‘exec.h’ containing new format ‘inlines’ of ‘exec.library’ in ‘ADE:include/inline’ directory.
If you want to add support for ‘GCC’ to a library, creating ‘inlines’ is not enough — you should also create ‘proto’ file.
Use one of existing files as a template. Some libraries (like for example ‘dos.library’) have broken header files and ‘GCC’ generates warning messages. In order to avoid them, you have to include various headers in ‘proto’ file before including ‘clib’ file.
You might also want to create ‘pragmas’ file, which might be necessary for badly written ‘SAS/C’ sources. ‘pragmas’ are generated automatically during building of ‘FD2InLine’ by an ‘AWK’ script, so you might either have a look at ‘fd2inline-{No value for `version'}-src.lha’ archive or simply create ‘pragmas’ by hand.
Creating a linker library with stubs might also be useful, in case somebody didn’t want to, or couldn’t, use inline headers.
‘fd2inline-{No value for `version'}-src.lha’ contains necessary support for this. For example, to generate ‘libexec.a’ library with ‘exec.library’ stubs, you’d have to type:
make alllib INCBASE=exec LIBBASE=exec
This would create two ‘libexec.a’ libraries: plain and base relative one. Of course, this particular example doesn’t make much sense since ‘libamiga.a’ already contains this stubs.
INCBASE
and LIBBASE
specify basenames of (input)
‘proto’ and ‘fd’ files and (output) library. This will often
be the same, but not always. For example, in case of ‘MUI’
INCBASE
has to be set to muimaster
, but LIBBASE
might be set to mui
.
[ << ] | [ < ] | [ Up ] | [ > ] | [ >> ] | [Top] | [Contents] | [Index] | [ ? ] |
First, you have to download ‘fd2inline-{No value for `version'}-src.lha’ archive.
Unarchive it. You might either build ‘FD2InLine’ in source archive or in a separate, build archive. I suggest the latter. Type:
lha -mraxe x fd2inline-{No value for `version'}-src.lha mkdir fd2inline-bin cd fd2inline-bin sh ../fd2inline-{No value for `version'}/configure --prefix=/ade make
This should build ‘FD2InLine’ executable and ‘inlines’.
Please note that ‘fd’ files should be available in directory
‘ADE:os-lib/fd’. If you store them in some other place, you’ll
have to edit ‘Makefile’ before invoking make
(variable
FD_DIR
).
You can then type:
make install
This will install new ‘fd2inline’, ‘inlines’ and documentation in appropriate subdirectories of ‘ADE:’.
‘fd2inline-{No value for `version'}-src.lha’ archive contains three
patches in context diff format. They fix bugs in OS 3.1 headers and
‘fd’ files. Without applying ‘amigaguide_lib.fd.diff’ you
won’t be able to build ‘inlines’ for ‘amigaguide.library’.
Two other patches rename an argument true
to tf
, since
true
is a reserved word in ‘C++’. Use patch
to
apply the patches, for example:
cd ADE:os-lib/fd patch <amigaguide_lib.fd.diff
A few words about the source code:
I know, it’s not state-of-the-art ‘C’ programming example.
However, believe me, it was in much worse condition when I took
it over. In its current state it is at least readable (if you use tab
size 3, as I do :-). I think that rewriting it in ‘C++’ would
clean it up considerably (it’s already written in ‘OO’ fashion,
so this should be quite easy). Using flex
and bison
to
create the parser would also be a nice thing, I guess. However, I
don’t think it’s worth the effort. But, if somebody wanted to do it:
feel free, I’ll be more than happy.
[ << ] | [ < ] | [ Up ] | [ > ] | [ >> ] | [Top] | [Contents] | [Index] | [ ? ] |
This chapter describes the implementation details of ‘inlines’.
5.1 Background | Function calls in shared libraries. | |
5.2 Old format | Inlines that used __inline
| |
5.3 New format | Inlines that use preprocessor. | |
5.4 Stubs format | Not really inlines, but... |
[ << ] | [ < ] | [ Up ] | [ > ] | [ >> ] | [Top] | [Contents] | [Index] | [ ? ] |
This section describes calling conventions used in Amiga shared libraries.
User-callable functions in AmigaOS are organized in libraries.
From our point of view, most important part of a library is a
library base. It always resides in RAM and contains library
variables and a jump table. The location of library base varies.
You can obtain the library base location of main system library —
exec.library
— by dereferencing 0x00000004
. Locations
of other libraries’ bases can be obtained using OpenLibrary
function of ‘exec.library’.
Without providing unnecessary details, every function has a fixed place in library’s jump table. In order to call a function, one has to jump into this place.
Most functions require some arguments. In ‘C’, these are usually passed on CPU stack. However, for some obscure reason, AmigaOS system designers decided that arguments to shared libraries should be passed in CPU registers.
All the information required to make function calls are provided in fd files. Every shared library should have such a file. It provides information about the name a library base variable should have and, for every function separately, about offset in jump table where the function resides and about registers in which arguments should be passed.
In order to check if arguments passed to function have correct type, C compiler needs function prototypes. These are provided in clib files — every library should have such a file.
Starting with AmigaOS release 2.0, some functions have been provided which accept variable number of arguments (so-called varargs functions). Actually, these are only ‘C’ language stubs. Internally, all optional arguments have to be put into an array of ‘long ints’ and address of this array is passed to fixed args library function.
To implement calls to shared libraries’ functions, compiler vendors have to either use some compiler-dependent tricks to make this calls directly (so-called in line), or provide linker libraries with functions stubs, usually written in assembler. In the latter case, function call from user’s code is compiled as usual — arguments are passed on stack. Then, in linking stage, a library stub gets linked in, and when it’s called, it moves arguments from stack to appropriate registers and jumps to library jump table. Needless to say, this is slower than making a call in line.
[ << ] | [ < ] | [ Up ] | [ > ] | [ >> ] | [Top] | [Contents] | [Index] | [ ? ] |
extern __inline APTR OpenAmigaGuideA(BASE_PAR_DECL struct NewAmigaGuide *nag, struct TagItem *attrs) { BASE_EXT_DECL register APTR res __asm("d0"); register struct Library *a6 __asm("a6") = BASE_NAME; register struct NewAmigaGuide *a0 __asm("a0") = nag; register struct TagItem *a1 __asm("a1") = attrs; __asm volatile ("jsr a6@(-0x36:W)" : "=r" (res) : "r" (a6), "r" (a0), "r" (a1) : "d0", "d1", "a0", "a1", "cc", "memory"); return res; }
In this implementation, Amiga shared libraries’ function stubs are
external functions. They are defined as __inline
, what makes
‘GCC’ insert them at every place of call. Mysterious
BASE_PAR_DECL
and BASE_EXT_DECL
defines are hacks
necessary for local library base support (which is quite hard to
achieve, so I won’t describe it here). The biggest disadvantage of
this ‘inlines’ is the fact that ‘GCC’ becomes very slow and
requires huge amount of memory when compiling them. Besides, inlining
works with optimization enabled only.
#ifndef NO_INLINE_STDARG #define OpenAmigaGuide(a0, tags...) \ ({ULONG _tags[] = { tags }; OpenAmigaGuideA((a0), (struct TagItem *)_tags);}) #endif /* !NO_INLINE_STDARG */
This is how ‘varargs’ functions are implemented. Handling them
cannot be made using __inline
functions, since __inline
functions require fixed number of arguments. Therefore, unique
features of GCC preprocessor (such as ‘varargs macros’) have to
be used, instead. This has some drawbacks, unfortunately. Since this
are actually preprocessor macros and not function calls, you can’t
make tricky things that involve preprocessor inside them. For example:
#include <proto/amigaguide.h> #define OPENAG_BEG OpenAmigaGuide( #define OPENAG_END , TAG_DONE) void f(void) { OPENAG_BEG "a_file.guide" OPENAG_END; OpenAmigaGuide( #ifdef ABC "abc.guide", #else "def.guide", #endif TAG_DONE); }
Neither of OpenAmigaGuide()
calls above is handled correctly.
In case of the first call, you get an error:
unterminated macro call
By the time preprocessor attempts to unwind OpenAmigaGuide
macro, OPENAG_END
is not yet unwound, so preprocessor can’t
find closing bracket. This code might look artificial, but ‘MUI’
for example defines such macros to make code look more pretty.
In case of the second call, you’ll see:
warning: preprocessing directive not recognized within macro arg
A workaround would be to either surround whole function calls with
conditions, or to conditionally define a preprocessor symbol
GUIDE
somewhere above and simply put GUIDE
as a function
argument:
#ifdef ABC #define GUIDE "abc.guide" #else #define GUIDE "def.guide" #endif void f(void) { #ifdef ABC OpenAmigaGuide("abc.guide", TAG_DONE); #else OpenAmigaGuide("def.guide", TAG_DONE); #endif OpenAmigaGuide(GUIDE, TAG_DONE); }
Because of this drawbacks, ‘varargs inlines’ can be disabled by
defining NO_INLINE_STDARG
before including ‘proto’ file.
You’ll need a library with function stubs in such case.
[ << ] | [ < ] | [ Up ] | [ > ] | [ >> ] | [Top] | [Contents] | [Index] | [ ? ] |
#define OpenAmigaGuideA(nag, attrs) \ LP2(0x36, APTR, OpenAmigaGuideA, struct NewAmigaGuide *, nag, a0, struct TagItem *, attrs, a1, \ , AMIGAGUIDE_BASE_NAME)
As you can see, this implementation is much more compact. LP2
macro (and others) are defined in ‘inline/macros.h’, which is
being included at the beginning of every ‘inline’ file.
#define LP2(offs, rt, name, t1, v1, r1, t2, v2, r2, bt, bn) \ ({ \ t1 _##name##_v1 = (v1); \ t2 _##name##_v2 = (v2); \ { \ register rt _##name##_re __asm("d0"); \ register struct Library *const _##name##_bn __asm("a6") = (struct Library*)(bn);\ register t1 _n1 __asm(#r1) = _##name##_v1; \ register t2 _n2 __asm(#r2) = _##name##_v2; \ __asm volatile ("jsr a6@(-"#offs":W)" \ : "=r" (_##name##_re) \ : "r" (_##name##_bn), "r"(_n1), "r"(_n2) \ : "d0", "d1", "a0", "a1", "cc", "memory"); \ _##name##_re; \ } \ })
If you compare this with old inlines (see section Old format) you’ll notice many similarities. Indeed, both implementations use the same tricks.
However, with new inlines, inlining is performed very early, at preprocessing stage. This makes compilation much faster, less memory hungry and independent of optimization options used. This also makes it very easy to use local library bases — all that’s needed is defining a local variable with the same name as library base.
Unfortunately, using preprocessor instead of compiler for making function calls has its drawback, as described earlier (see section Old format). There is nothing you can do about it but modify your code, I’m afraid.
Depending on the type of function, ‘fd2inline’ generates calls to different ‘LP’ macros.
Macros are distinguished by one or more of qualifiers described below:
As you might have already guessed, digit indicates the number of arguments a function accepts. Therefore, it’s mandatory.
This indicates “no return” (void
) function.
These two are used when one of the arguments has to be in either ‘a4’ or ‘a5’ register. In certain situations, these registers have special meaning and have to be handled more carefully.
This indicates “user base” — library base pointer has to be specified explicitly by user. Currently, this is used for ‘cia.resource’ only. Since there are two ‘CIA’ chips, programmer has to specify which one [s]he wants to use.
This means that one of the arguments has a “pointer to function”
type. To overcome strange ‘C’ syntax rules in this case, inside
‘FP’ macros a typedef
to __fpt
is performed.
‘inline’ file passes __fpt
as argument’s type to ‘LP’
macro. The actual type of the argument, in a form suitable for a
typedef
, is passed as an additional, last argument.
As you can see, there could be more than a hundred different variations of ‘LP’ macros. ‘inline/macros.h’ contains only 34, which are used in current OS version and supported 3rd party libraries. More macros will be added in future, if needed.
If you had a careful look at the definition of OpenAmigaGuideA
on the beginning of this section, you might have noticed that next to
last argument to ‘LP’ macro is not used. New inlines were not
implemented in one evening and they came through many modifications.
This unused argument (which was once a type of library base pointer)
is provided for backwards compatibility. Actually, there are more
unnecessary arguments, like function and argument names, but we
decided to left them in peace.
[ << ] | [ < ] | [ Up ] | [ > ] | [ >> ] | [Top] | [Contents] | [Index] | [ ? ] |
Stubs format is very similar to old format (see section Old format). The
functions are not defined as extern
, however.
The main difference is format of ‘varargs’ functions — they are actual functions, not preprocessor macros.
APTR OpenAmigaGuide(struct NewAmigaGuide *nag, int tag, ...) { return OpenAmigaGuideA(nag, (struct TagItem *)&tag); }
This format is not suitable for inlining, and it is not provided for this purpose. It is provided for building of linker libraries with stubs (see section Using fd2inline).
[ << ] | [ < ] | [ Up ] | [ > ] | [ >> ] | [Top] | [Contents] | [Index] | [ ? ] |
First parser for ‘GCC inlines’ was written in ‘Perl’ by Markus Wild.
It had several limitations, which were apparently hard to fix in ‘Perl’. That’s why Wolfgang Baron decided to write a new parser in ‘C’.
However, for some reason he has never finished it. In early 1995 Rainer F. Trunz took over its development and “improved, updated, simply made it workable” (it’s a quote from changes log). It still contained quite a lot of bugs, though.
In more-or-less the same time I started a discussion on ‘amiga-gcc-port’ mailing list about improving quality of ‘inlines’. The most important idea came from Matthias Fleischer — he introduced new format of ‘inlines’ (see section New format). Since I started the discussion, I volunteered to make improvements to ‘inlines’ parser. Having no idea about programming in ‘Perl’, I decided to modify the parser written in ‘C’. I fixed all the bugs known to me, added some new features and wrote this terribly long documentation :-).
Not all of the files distributed in ‘FD2InLine’ archives were created by me or ‘FD2InLine’. Most of the files in ‘include/proto’, as well as a few files in ‘include/inline’ (‘alib.h’, ‘strsub.h’ and ‘stubs.h’) were written by Gunther Nikl (with some modifications by Joerg Hoehle).
If you have any comments concerning this work, please write to:
ade-gcc@ninemoons.com
This is a list where most of ADE GCC developers and activists are subscribed, so you are practically guaranteed to get a reply.
However, if you for some reason wanted to contact me personally, do it in one of the following ways:
iskra@student.uci.agh.edu.pl
Should be valid until October 1999 (at least I hope so :-).
Kamil Iskra Luzycka 51/258 30-658 Krakow Poland
Latest version of this package should always be available on my WWW page:
http://student.uci.agh.edu.pl/~iskra
[ << ] | [ < ] | [ Up ] | [ > ] | [ >> ] | [Top] | [Contents] | [Index] | [ ? ] |
Jump to: | <
_
A B C F H I J L M N O P R S U V W |
---|
Jump to: | <
_
A B C F H I J L M N O P R S U V W |
---|
[Top] | [Contents] | [Index] | [ ? ] |
[Top] | [Contents] | [Index] | [ ? ] |
This document was generated on January 15, 2023 using texi2html 5.0.
The buttons in the navigation panels have the following meaning:
Button | Name | Go to | From 1.2.3 go to |
---|---|---|---|
[ << ] | FastBack | Beginning of this chapter or previous chapter | 1 |
[ < ] | Back | Previous section in reading order | 1.2.2 |
[ Up ] | Up | Up section | 1.2 |
[ > ] | Forward | Next section in reading order | 1.2.4 |
[ >> ] | FastForward | Next chapter | 2 |
[Top] | Top | Cover (top) of document | |
[Contents] | Contents | Table of contents | |
[Index] | Index | Index | |
[ ? ] | About | About (help) |
where the Example assumes that the current position is at Subsubsection One-Two-Three of a document of the following structure:
This document was generated on January 15, 2023 using texi2html 5.0.