/vendor/gc/doc/README.amiga

  1===========================================================================
  2            Kjetil S. Matheussen's notes (28-11-2000)
  3===========================================================================
  4Compiles under SAS/C again. Should allso still compile under other
  5amiga compilers without big changes. I haven't checked if it still
  6works under gcc, because I don't have gcc for amiga. But I have
  7updated 'Makefile', and hope it compiles fine.
  8
  9
 10WHATS NEW:
 11
 121.
 13   Made a pretty big effort in preventing GCs allocating-functions from returning
 14   chip-mem.
 15
 16   The lower part of the new file AmigaOS.c does this in various ways, mainly by
 17   wrapping GC_malloc, GC_malloc_atomic, GC_malloc_uncollectable,
 18   GC_malloc_atomic_uncollectable, GC_malloc_stubborn, GC_malloc_ignore_off_page
 19   and GC_malloc_atomic_ignore_off_page. GC_realloc is allso wrapped, but
 20   doesn't do the same effort in preventing to return chip-mem.
 21   Other allocating-functions (f.ex. GC_*_typed_) can probably be
 22   used without any problems, but beware that the warn hook will not be called.
 23   In case of problems, don't define GC_AMIGA_FASTALLOC.
 24
 25   Programs using more time actually using the memory allocated
 26   (instead of just allocate and free rapidly) have
 27   the most to earn on this, but even gctest now normally runs twice
 28   as fast and uses less memory, on my poor 8MB machine.
 29
 30   The changes have only effect when there is no more
 31   fast-mem left. But with the way GC works, it
 32   could happen quite often. Beware that an atexit handler had to be added,
 33   so using the abort() function will make a big memory-loss.
 34   If you absolutely must call abort() instead of exit(), try calling
 35   the GC_amiga_free_all_mem function before abort().
 36
 37   New amiga-spesific compilation flags:
 38
 39   GC_AMIGA_FASTALLOC - By NOT defining this option, GC will work like before,
 40                        it will not try to force fast-mem out of the OS, and
 41                        it will use normal calloc for allocation, and the rest
 42                        of the following flags will have no effect.
 43
 44   GC_AMIGA_ONLYFAST - Makes GC never to return chip-mem. GC_AMIGA_RETRY have
 45                       no effect if this flag is set.
 46
 47   GC_AMIGA_GC - If gc returns NULL, do a GC_gcollect, and try again. This
 48                 usually is a success with the standard GC configuration. 
 49                 It is allso the most important flag to set to prevent
 50                 GC from returning chip-mem. Beware that it slows down a lot
 51                 when a program is rapidly allocating/deallocating when
 52                 theres either very little fast-memory left or verly little
 53                 chip-memory left. Its not a very common situation, but gctest
 54                 sometimes (very rare) use many minutes because of this.
 55
 56   GC_AMIGA_RETRY - If gc succeed allocating memory, but it is chip-mem,
 57                    try again and see if it is fast-mem. Most of the time,
 58                    it will actually return fast-mem for the second try.
 59                    I have set max number of retries to 9 or size/5000. You
 60                    can change this if you like. (see GC_amiga_rec_alloc())
 61
 62   GC_AMIGA_PRINTSTATS - Gather some statistics during the execution of a
 63                         program, and prints out the info when the atexit-handler
 64                         is called.
 65
 66   My reccomendation is to set all this flags, except GC_AMIGA_PRINTSTATS and
 67   GC_AMIGA_ONLYFAST.
 68
 69   If your program demands high response-time, you should
 70   not define GC_AMIGA_GC, and possible allso define GC_AMIGA_ONLYFAST.
 71   GC_AMIGA_RETRY does not seem to slow down much.
 72
 73   Allso, when compiling up programs, and GC_AMIGA_FASTALLOC was not defined when
 74   compilling gc, you can define GC_AMIGA_MAKINGLIB to avoid having these allocation-
 75   functions wrapped. (see gc.h)
 76
 77   Note that GC_realloc must not be called before any of
 78   the other above mentioned allocating-functions have been called. (shouldn't be
 79   any programs doing so either, I hope).
 80
 81   Another note. The allocation-function is wrapped when defining
 82   GC_AMIGA_FASTALLOC by letting the function go thru the new
 83   GC_amiga_allocwrapper_do function-pointer (see gc.h). Means that
 84   sending function-pointers, such as GC_malloc, GC_malloc_atomic, etc.,
 85   for later to be called like f.ex this, (*GC_malloc_functionpointer)(size),
 86   will not wrap the function. This is normally not a big problem, unless
 87   all allocation function is called like this, which will cause the
 88   atexit un-allocating function never to be called. Then you either
 89   have to manually add the atexit handler, or call the allocation-
 90   functions function-pointer functions like this;
 91   (*GC_amiga_allocwrapper_do)(size,GC_malloc_functionpointer).
 92   There are probably better ways this problem could be handled, unfortunately,
 93   I didn't find any without rewriting or replacing a lot of the GC-code, which
 94   I really didn't want to. (Making new GC_malloc_* functions, and just
 95   define f.ex GC_malloc as GC_amiga_malloc should allso work).
 96
 97
 98   New amiga-spesific function:
 99
100     void GC_amiga_set_toany(void (*func)(void));
101
102   'func' is a function that will be called right before gc has to change
103   allocation-method from MEMF_FAST to MEMF_ANY. Ie. when it is likely
104   it will return chip-mem.
105
106
1072. A few small compiler-spesific additions to make it compile with SAS/C again.
108
1093. Updated and rewritten the smakefile, so that it works again and that
110   the "unnecesarry" 'SCOPTIONS' files could be removed. Allso included
111   the cord-smakefile stuff in the main smakefile, so that the cord smakefile
112   could be removed too. By writing smake -f Smakefile.smk, both gc.lib and
113   cord.lib will be made.
114
115
116
117STILL MISSING:
118
119Programs can not be started from workbench, at least not for SAS/C. (Martin
120Tauchmanns note about that it now works with workbench is definitely wrong
121when concerning SAS/C). I guess it works if you use the old "#if 0'ed"-code,
122but I haven't tested it. I think the reason for MT to replace the
123"#if 0'ed"-code was only because it was a bit to SAS/C-spesific. But I
124don't know. An iconx-script solves this problem anyway.
125
126
127BEWARE!
128
129-To run gctest, set the stack to around 200000 bytes first.
130-SAS/C-spesific: cord will crash if you compile gc.lib with
131 either parm=reg or parm=both. (missing legal prototypes for
132 function-pointers someplace is the reason I guess.).
133
134
135tested with software: Radium, http://www.stud.ifi.uio.no/~ksvalast/radium/
136
137tested with hardware: MC68060
138
139
140-ksvalast@ifi.uio.no
141
142
143===========================================================================
144			   Martin Tauchmann's notes		(1-Apr-99)
145===========================================================================
146
147Works now, also with the GNU-C compiler V2.7.2.1. <ftp://ftp.unina.it/pub/amiga/geekgadgets/amiga/m68k/snapshots/971125/amiga-bin/>
148Modify the `Makefile`
149CC=cc $(ABI_FLAG)
150to
151CC=gcc $(ABI_FLAG)
152
153TECHNICAL NOTES
154
155- `GC_get_stack_base()`, `GC_register_data_segments()` works now with every
156   C compiler; also Workbench.
157
158- Removed AMIGA_SKIP_SEG, but the Code-Segment must not be scanned by GC.
159
160
161PROBLEMS
162- When the Linker, does`t merge all Code-Segments to an single one. LD of GCC
163  do it always.
164
165- With ixemul.library V47.3, when an GC program launched from another program
166  (example: `Make` or `if_mach M68K AMIGA gctest`), `GC_register_data_segments()`
167  found the Segment-List of the caller program.
168  Can be fixed, if the run-time initialization code (for C programs, usually *crt0*)
169  support `__data` and `__bss`.
170
171- PowerPC Amiga currently not supported.
172
173- Dynamic libraries (dyn_load.c) not supported.
174
175
176TESTED WITH SOFTWARE
177
178`Optimized Oberon 2 C` (oo2c) <http://cognac.informatik.uni-kl.de/download/index.html>
179
180
181TESTED WITH HARDWARE
182
183MC68030
184
185
186CONTACT
187
188Please, contact me at <martintauchmann@bigfoot.com>, when you change the
189Amiga port. <http://martintauchmann.home.pages.de>
190 
191===========================================================================
192			   Michel Schinz's notes
193===========================================================================
194WHO DID WHAT
195
196The original Amiga port was made by Jesper Peterson. I (Michel Schinz)
197modified it slightly to reflect the changes made in the new official
198distributions, and to take advantage of the new SAS/C 6.x features. I also
199created a makefile to compile the "cord" package (see the cord
200subdirectory).
201
202TECHNICAL NOTES
203
204In addition to Jesper's notes, I have the following to say:
205
206- Starting with version 4.3, gctest checks to see if the code segment is
207  added to the root set or not, and complains if it is. Previous versions
208  of this Amiga port added the code segment to the root set, so I tried to
209  fix that. The only problem is that, as far as I know, it is impossible to
210  know which segments are code segments and which are data segments (there
211  are indeed solutions to this problem, like scanning the program on disk
212  or patch the LoadSeg functions, but they are rather complicated). The
213  solution I have chosen (see os_dep.c) is to test whether the program
214  counter is in the segment we are about to add to the root set, and if it
215  is, to skip the segment. The problems are that this solution is rather
216  awkward and that it works only for one code segment. This means that if
217  your program has more than one code segment, all of them but one will be
218  added to the root set. This isn't a big problem in fact, since the
219  collector will continue to work correctly, but it may be slower.
220
221  Anyway, the code which decides whether to skip a segment or not can be
222  removed simply by not defining AMIGA_SKIP_SEG. But notice that if you do
223  so, gctest will complain (it will say that "GC_is_visible produced wrong
224  failure indication"). However, it may be useful if you happen to have
225  pointers stored in a code segment (you really shouldn't).
226
227  If anyone has a good solution to the problem of finding, when a program
228  is loaded in memory, whether a segment is a code or a data segment,
229  please let me know.
230
231PROBLEMS
232
233If you have any problem with this version, please contact me at
234schinz@alphanet.ch (but do *not* send long files, since we pay for
235every mail!).
236
237===========================================================================
238			  Jesper Peterson's notes
239===========================================================================
240
241ADDITIONAL NOTES FOR AMIGA PORT
242
243These notes assume some familiarity with Amiga internals.
244
245WHY I PORTED TO THE AMIGA
246
247The sole reason why I made this port was as a first step in getting
248the Sather(*) language on the Amiga. A port of this language will
249be done as soon as the Sather 1.0 sources are made available to me.
250Given this motivation, the garbage collection (GC) port is rather
251minimal.
252
253(*) For information on Sather read the comp.lang.sather newsgroup.
254
255LIMITATIONS
256
257This port assumes that the startup code linked with target programs
258is that supplied with SAS/C versions 6.0 or later. This allows
259assumptions to be made about where to find the stack base pointer
260and data segments when programs are run from WorkBench, as opposed
261to running from the CLI. The compiler dependent code is all in the
262GC_get_stack_base() and GC_register_data_segments() functions, but
263may spread as I add Amiga specific features.
264
265Given that SAS/C was assumed, the port is set up to be built with
266"smake" using the "SMakefile". Compiler options in "SCoptions" can
267be set with "scopts" program. Both "smake" and "scopts" are part of
268the SAS/C commercial development system.
269
270In keeping with the porting philosophy outlined above, this port
271will not behave well with Amiga specific code. Especially not inter-
272process comms via messages, and setting up public structures like
273Intuition objects or anything else in the system lists. For the
274time being the use of this library is limited to single threaded
275ANSI/POSIX  compliant or near-complient code. (ie. Stick to stdio
276for now). Given this limitation there is currently no mechanism for
277allocating "CHIP" or "PUBLIC" memory under the garbage collector.
278I'll add this after giving it considerable thought. The major
279problem is the entire physical address space may have to me scanned,
280since there is no telling who we may have passed memory to.
281
282If you allocate your own stack in client code, you will have to
283assign the pointer plus stack size to GC_stackbottom.
284
285The initial stack size of the target program can be compiled in by
286setting the __stack symbol (see SAS documentaion). It can be over-
287ridden from the CLI by running the AmigaDOS "stack" program, or from
288the WorkBench by setting the stack size in the tool types window.
289
290SAS/C COMPILER OPTIONS (SCoptions)
291
292You may wish to check the "CPU" code option is appropriate for your
293intended target system.
294
295Under no circumstances set the "StackExtend" code option in either
296compiling the library or *ANY* client code.
297
298All benign compiler warnings have been suppressed. These mainly
299involve lack of prototypes in the code, and dead assignments
300detected by the optimizer.
301
302THE GOOD NEWS
303
304The library as it stands is compatible with the GigaMem commercial
305virtual memory software, and probably similar PD software.
306
307The performance of "gctest" on an Amiga 2630 (68030 @ 25Mhz)
308compares favourably with an HP9000 with similar architecture (a 325
309with a 68030 I think).
310
311-----------------------------------------------------------------------
312
313The Amiga port has been brought to you by:
314
315Jesper Peterson.
316
317jep@mtiame.mtia.oz.au		(preferred, but 1 week turnaround)
318jep@orca1.vic.design.telecom.au (that's orca<one>, 1 day turnaround)
319
320At least one of these addresses should be around for a while, even
321though I don't work for either of the companies involved.
322