| 1 |
Kjetil S. Matheussen's notes (28-11-2000) |
| 2 |
|
| 3 |
Compiles under SAS/C again. Should also still compile under other |
| 4 |
Amiga compilers without big changes. I haven't checked if it still |
| 5 |
works under gcc, because I don't have gcc for Amiga. But I have |
| 6 |
updated 'Makefile', and hope it compiles fine. |
| 7 |
|
| 8 |
|
| 9 |
WHATS NEW: |
| 10 |
|
| 11 |
1. |
| 12 |
Made a pretty big effort in preventing GC allocating-functions from |
| 13 |
returning chip-mem. |
| 14 |
|
| 15 |
The lower part of the new file AmigaOS.c does this in various ways, mainly by |
| 16 |
wrapping GC_malloc, GC_malloc_atomic, GC_malloc_uncollectable, |
| 17 |
GC_malloc_atomic_uncollectable, GC_malloc_ignore_off_page |
| 18 |
and GC_malloc_atomic_ignore_off_page. GC_realloc is also wrapped, but |
| 19 |
doesn't do the same effort in preventing to return chip-mem. |
| 20 |
Other allocating-functions (e.g., GC_*_typed) can probably be |
| 21 |
used without any problems, but beware that the warn hook will not be called. |
| 22 |
In case of problems, don't define GC_AMIGA_FASTALLOC. |
| 23 |
|
| 24 |
Programs using more time actually using the memory allocated |
| 25 |
(instead of just allocate and free rapidly) have |
| 26 |
the most to earn on this, but even gctest now normally runs twice |
| 27 |
as fast and uses less memory, on my poor 8 MB machine. |
| 28 |
|
| 29 |
The changes have only effect when there is no more |
| 30 |
fast-mem left. But with the way GC works, it |
| 31 |
could happen quite often. Beware that an atexit handler had to be added, |
| 32 |
so using the abort() function will make a big memory-loss. |
| 33 |
If you absolutely must call abort() instead of exit(), try calling |
| 34 |
the GC_amiga_free_all_mem function before abort(). |
| 35 |
|
| 36 |
New Amiga-specific compilation flags: |
| 37 |
|
| 38 |
GC_AMIGA_FASTALLOC - By NOT defining this option, GC will work like before, |
| 39 |
it will not try to force fast-mem out of the OS, and |
| 40 |
it will use normal calloc for allocation, and the rest |
| 41 |
of the following flags will have no effect. |
| 42 |
|
| 43 |
GC_AMIGA_ONLYFAST - Makes GC never to return chip-mem. GC_AMIGA_RETRY have |
| 44 |
no effect if this flag is set. |
| 45 |
|
| 46 |
GC_AMIGA_GC - If gc returns NULL, do a GC_gcollect, and try again. This |
| 47 |
usually is a success with the standard GC configuration. |
| 48 |
It is also the most important flag to set to prevent |
| 49 |
GC from returning chip-mem. Beware that it slows down a lot |
| 50 |
when a program is rapidly allocating/deallocating when |
| 51 |
there's either very little fast-memory left or very little |
| 52 |
chip-memory left. It's not a very common situation, but gctest |
| 53 |
sometimes (very rare) use many minutes because of this. |
| 54 |
|
| 55 |
GC_AMIGA_RETRY - If gc succeed allocating memory, but it is chip-mem, |
| 56 |
try again and see if it is fast-mem. Most of the time, |
| 57 |
it will actually return fast-mem for the second try. |
| 58 |
I have set max number of retries to 9 or size/5000. You |
| 59 |
can change this if you like. (see GC_amiga_rec_alloc()) |
| 60 |
|
| 61 |
GC_AMIGA_PRINTSTATS - Gather some statistics during the execution of a |
| 62 |
program, and prints out the info when the atexit-handler |
| 63 |
is called. |
| 64 |
|
| 65 |
My recommendation is to set all this flags, except GC_AMIGA_PRINTSTATS and |
| 66 |
GC_AMIGA_ONLYFAST. |
| 67 |
|
| 68 |
If your program demands high response-time, you should |
| 69 |
not define GC_AMIGA_GC, and possible also define GC_AMIGA_ONLYFAST. |
| 70 |
GC_AMIGA_RETRY does not seem to slow down much. |
| 71 |
|
| 72 |
Also, when compiling up programs, and GC_AMIGA_FASTALLOC was not defined when |
| 73 |
compiling gc, you can define GC_AMIGA_MAKINGLIB to avoid having these allocation- |
| 74 |
functions wrapped. (see gc.h) |
| 75 |
|
| 76 |
Note that GC_realloc must not be called before any of |
| 77 |
the other above mentioned allocating-functions have been called. (shouldn't be |
| 78 |
any programs doing so either, I hope). |
| 79 |
|
| 80 |
Another note. The allocation-function is wrapped when defining |
| 81 |
GC_AMIGA_FASTALLOC by letting the function go thru the new |
| 82 |
GC_amiga_allocwrapper_do function-pointer (see gc.h). Means that |
| 83 |
sending function-pointers, such as GC_malloc, GC_malloc_atomic, etc., |
| 84 |
for later to be called, e.g., like this, (*GC_malloc_function_pointer)(size), |
| 85 |
will not wrap the function. This is normally not a big problem, unless |
| 86 |
all allocation function is called like this, which will cause the |
| 87 |
atexit un-allocating function never to be called. Then you either |
| 88 |
have to manually add the atexit handler, or call the allocation- |
| 89 |
functions function-pointer functions like this; |
| 90 |
(*GC_amiga_allocwrapper_do)(size,GC_malloc_function_pointer). |
| 91 |
There are probably better ways this problem could be handled, unfortunately, |
| 92 |
I didn't find any without rewriting or replacing a lot of the GC-code, which |
| 93 |
I really didn't want to. (Making new GC_malloc_* functions, and just |
| 94 |
defining, e.g., GC_malloc as GC_amiga_malloc should work too). |
| 95 |
|
| 96 |
|
| 97 |
New Amiga-specific function: |
| 98 |
|
| 99 |
void GC_amiga_set_toany(void (*func)(void)); |
| 100 |
|
| 101 |
'func' is a function that will be called right before gc has to change |
| 102 |
allocation-method from MEMF_FAST to MEMF_ANY. I.e., when it is likely |
| 103 |
it will return chip-mem. |
| 104 |
|
| 105 |
|
| 106 |
2. A few small compiler-specific additions to make it compile with SAS/C again. |
| 107 |
|
| 108 |
3. Updated and rewritten the SMakefile.amiga, so that it works again and that |
| 109 |
the "unnecessary" 'SCOPTIONS' files could be removed. Also included |
| 110 |
the cord-smakefile stuff in the main smakefile, so that the cord smakefile |
| 111 |
could be removed too. By typing "smake -f SMakefile.amiga", both gc.lib and |
| 112 |
cord.lib will be made. |
| 113 |
|
| 114 |
|
| 115 |
|
| 116 |
STILL MISSING: |
| 117 |
|
| 118 |
Programs can not be started from workbench, at least not for SAS/C. (Martin |
| 119 |
Tauchmanns note about that it now works with workbench is definitely wrong |
| 120 |
when concerning SAS/C). An iconx-script solves this problem. |
| 121 |
|
| 122 |
|
| 123 |
BEWARE! |
| 124 |
|
| 125 |
-To run gctest, set the stack to around 200000 bytes first. |
| 126 |
-SAS/C-specific: cord will crash if you compile gc.lib with |
| 127 |
either parm=reg or parm=both. (missing legal prototypes for |
| 128 |
function-pointers someplace is the reason I guess.). |
| 129 |
|
| 130 |
|
| 131 |
tested with software: Radium, http://www.stud.ifi.uio.no/~ksvalast/radium/ |
| 132 |
tested with hardware: MC68060 |
| 133 |
|
| 134 |
|
| 135 |
Martin Tauchmann's notes (1-Apr-99) |
| 136 |
|
| 137 |
Works now, also with the GNU-C compiler V2.7.2.1. <ftp://ftp.unina.it/pub/amiga/geekgadgets/amiga/m68k/snapshots/971125/amiga-bin/> |
| 138 |
Modify the `Makefile` |
| 139 |
CC=cc $(ABI_FLAG) |
| 140 |
to |
| 141 |
CC=gcc $(ABI_FLAG) |
| 142 |
|
| 143 |
TECHNICAL NOTES |
| 144 |
|
| 145 |
- `GC_get_stack_base()`, `GC_register_data_segments()` works now with every |
| 146 |
C compiler; also Workbench. |
| 147 |
|
| 148 |
- Removed AMIGA_SKIP_SEG, but the Code-Segment must not be scanned by GC. |
| 149 |
|
| 150 |
|
| 151 |
PROBLEMS |
| 152 |
- When the Linker, does`t merge all Code-Segments to an single one. LD of GCC |
| 153 |
do it always. |
| 154 |
|
| 155 |
- With ixemul.library V47.3, when an GC program launched from another program |
| 156 |
(example: `Make` or `if_mach M68K AMIGA gctest`), `GC_register_data_segments()` |
| 157 |
found the Segment-List of the caller program. |
| 158 |
Can be fixed, if the run-time initialization code (for C programs, usually *crt0*) |
| 159 |
support `__data` and `__bss`. |
| 160 |
|
| 161 |
- PowerPC Amiga currently not supported. |
| 162 |
|
| 163 |
- Dynamic libraries (dyn_load.c) not supported. |
| 164 |
|
| 165 |
|
| 166 |
TESTED WITH SOFTWARE |
| 167 |
|
| 168 |
`Optimized Oberon 2 C` (oo2c) <http://cognac.informatik.uni-kl.de/download/index.html> |
| 169 |
|
| 170 |
|
| 171 |
TESTED WITH HARDWARE |
| 172 |
|
| 173 |
MC68030 |
| 174 |
|
| 175 |
|
| 176 |
Michel Schinz's notes |
| 177 |
|
| 178 |
WHO DID WHAT |
| 179 |
|
| 180 |
The original Amiga port was made by Jesper Peterson. I (Michel Schinz) |
| 181 |
modified it slightly to reflect the changes made in the new official |
| 182 |
distributions, and to take advantage of the new SAS/C 6.x features. I also |
| 183 |
created a makefile to compile the "cord" package (see the cord |
| 184 |
subdirectory). |
| 185 |
|
| 186 |
TECHNICAL NOTES |
| 187 |
|
| 188 |
In addition to Jesper's notes, I have the following to say: |
| 189 |
|
| 190 |
- gctest checks to see if the code segment is added to the root set or not, |
| 191 |
and complains if it is. The only problem is that, as far as I know, it is |
| 192 |
impossible to know which segments are code segments and which are data |
| 193 |
segments (there are indeed solutions to this problem, like scanning the |
| 194 |
program on disk or patch the LoadSeg functions, but they are rather |
| 195 |
complicated). The solution I have chosen (see os_dep.c) is to test whether |
| 196 |
the program counter is in the segment we are about to add to the root set, |
| 197 |
and if it is, to skip the segment. The problems are that this solution is |
| 198 |
rather awkward and that it works only for one code segment. This means that |
| 199 |
if your program has more than one code segment, all of them but one will be |
| 200 |
added to the root set. This isn't a big problem in fact, since the |
| 201 |
collector will continue to work correctly, but it may be slower. |
| 202 |
|
| 203 |
Anyway, the code which decides whether to skip a segment or not can be |
| 204 |
removed simply by not defining AMIGA_SKIP_SEG. But notice that if you do |
| 205 |
so, gctest will complain (it will say that "GC_is_visible produced wrong |
| 206 |
failure indication"). However, it may be useful if you happen to have |
| 207 |
pointers stored in a code segment (you really shouldn't). |
| 208 |
|
| 209 |
If anyone has a good solution to the problem of finding, when a program |
| 210 |
is loaded in memory, whether a segment is a code or a data segment, |
| 211 |
please let me know. |
| 212 |
|
| 213 |
|
| 214 |
Jesper Peterson's notes |
| 215 |
|
| 216 |
ADDITIONAL NOTES FOR AMIGA PORT |
| 217 |
|
| 218 |
These notes assume some familiarity with Amiga internals. |
| 219 |
|
| 220 |
WHY I PORTED TO THE AMIGA |
| 221 |
|
| 222 |
The sole reason why I made this port was as a first step in getting |
| 223 |
the Sather(*) language on the Amiga. A port of this language will |
| 224 |
be done as soon as the Sather 1.0 sources are made available to me. |
| 225 |
Given this motivation, the garbage collection (GC) port is rather |
| 226 |
minimal. |
| 227 |
|
| 228 |
(*) For information on Sather read the comp.lang.sather newsgroup. |
| 229 |
|
| 230 |
LIMITATIONS |
| 231 |
|
| 232 |
This port assumes that the startup code linked with target programs |
| 233 |
is that supplied with SAS/C versions 6.0 or later. This allows |
| 234 |
assumptions to be made about where to find the stack base pointer |
| 235 |
and data segments when programs are run from WorkBench, as opposed |
| 236 |
to running from the CLI. The compiler dependent code is all in the |
| 237 |
GC_get_stack_base() and GC_register_data_segments() functions, but |
| 238 |
may spread as I add Amiga specific features. |
| 239 |
|
| 240 |
Given that SAS/C was assumed, the port is set up to be built with |
| 241 |
"smake" using the "SMakefile". Compiler options in "SCoptions" can |
| 242 |
be set with "scopts" program. Both "smake" and "scopts" are part of |
| 243 |
the SAS/C commercial development system. |
| 244 |
|
| 245 |
In keeping with the porting philosophy outlined above, this port |
| 246 |
will not behave well with Amiga specific code. Especially not inter- |
| 247 |
process communications via messages, and setting up public structures like |
| 248 |
Intuition objects or anything else in the system lists. For the |
| 249 |
time being the use of this library is limited to single threaded |
| 250 |
ANSI/POSIX compliant or near-compliant code. (i.e., stick to stdio |
| 251 |
for now). Given this limitation there is currently no mechanism for |
| 252 |
allocating "CHIP" or "PUBLIC" memory under the garbage collector. |
| 253 |
I'll add this after giving it considerable thought. The major |
| 254 |
problem is the entire physical address space may have to me scanned, |
| 255 |
since there is no telling who we may have passed memory to. |
| 256 |
|
| 257 |
If you allocate your own stack in client code, you will have to |
| 258 |
assign the pointer plus stack size to GC_stackbottom. |
| 259 |
|
| 260 |
The initial stack size of the target program can be compiled in by |
| 261 |
setting the __stack symbol (see SAS documentation). It can be over- |
| 262 |
ridden from the CLI by running the AmigaDOS "stack" program, or from |
| 263 |
the WorkBench by setting the stack size in the tool types window. |
| 264 |
|
| 265 |
SAS/C COMPILER OPTIONS (SCoptions) |
| 266 |
|
| 267 |
You may wish to check the "CPU" code option is appropriate for your |
| 268 |
intended target system. |
| 269 |
|
| 270 |
Under no circumstances set the "StackExtend" code option in either |
| 271 |
compiling the library or *ANY* client code. |
| 272 |
|
| 273 |
All benign compiler warnings have been suppressed. These mainly |
| 274 |
involve lack of prototypes in the code, and dead assignments |
| 275 |
detected by the optimizer. |
| 276 |
|
| 277 |
THE GOOD NEWS |
| 278 |
|
| 279 |
The library as it stands is compatible with the GigaMem commercial |
| 280 |
virtual memory software, and probably similar PD software. |
| 281 |
|
| 282 |
The performance of "gctest" on an Amiga 2630 (68030 @ 25 MHz) |
| 283 |
compares favorably with an HP9000 with similar architecture (a 325 |
| 284 |
with a 68030 I think). |