Short: Malloc() replacement using Exec pools Author: jochen.wiedmann@uni-tuebingen.de Uploader: jochen.wiedmann@uni-tuebingen.de Type: dev/c MemPools - malloc() replacement using standard Exec memory pools ================================================================ Many people don't like using malloc() and similar functions because they think it's too slow and gives too much overhead. Well, as far as I can say they're alright, at least for users of Dice and SAS/C: the Exec pool functions seem to be at least 5 times faster and it can even be faster than the malloc() of GNU-c! (See "Timings" below.) Other people like malloc(): It's portable and rather easy to use. I am one of these, however I don't want to loose the speed of pool functions. The mempools package is a link library with the functions malloc(), calloc(), realloc(), strdup() and free(). It replaces the compilers own functions by just linking against this library. No additional work is required as the library takes use of the compilers auto- initialization facility. Ths autoinitialization restricts use of the mempools library to users of Dice, SAS/C and gcc: I don't know how to handle other compilers. Note, that pools are available not only for users of OS 3.x! The current amiga.lib offers replacements which call exec.library, if the OS is 3.x and emulate pools otherwise. The MemPools library uses this replacements. 1.) Disclaimer: Copyrights, (No) Warranty ----------------------------------------- This program is Copyright (C) 1994 Jochen Wiedmann Am Eisteich 9 72555 Metzingen Germany Phone: (0049) +7123 / 14881 Mail: jochen.wiedmann@uni-tuebingen.de Permission is granted to make and distribute either verbatim and modified copies of this documentation and the library MemPools provided the copyright notice and this permission notice are preserved on all copies and the "GNU General Public License" (in the file COPYING) is distributed as well. The author gives ABSOLUTELY NO warranty that the software described in this documentation and the results produced by them are correct. The author cannot be held responsible for ANY damage resulting from the use of this software. 2.) Installation ---------------- There are four libraries included in the distribution, it depends on your compiler which one you need: mempoolss.lib and mempoolssr.lib are for dice, they should be copied to DLIB: or a similar place. mempools.lib is for SAS/C, its destination is sc:lib. Finally libmempools.a is the GNU-c version which should go to GNU:lib. If you want to recreate them, just type smake (SAS/C) or make (GNU-c). Dice, however, requires different libraries, depending on the data model, the type of argument handling and so on. To create a certain library just type lbmake mempools s (for mempoolss.lib) lbmake mempools s r (for mempoolssr.lib) lbmake mempools l (for mempoolsl.lib) . . . 3.) Usage --------- All you have to do is to link against the mempools library. a) Dice Add the option -lmempools to your compiler statement. By adding it to the environment variable "DCCOPTS" you get this automatically. b) SAS/C Add the option LIB mempools to your compiler statement. By using the "scopts" program you can get this automatically, too. c) gcc Add the option -lmempools to your compiler statement. I don't know if gcc can handle this automatically. Probably someone can enlighten me? However, you might be interested in controlling the pools attributes. This can be done by creating global variables __MemPoolPuddleSize, __MemPoolThreshSize and __MemPoolFlags which correspond to the arguments of the CreatePool() function. For example, if you want to modify the puddle size (this can increase speed, see "Timings" below), just add the following line somewhere to your program: ULONG __MemPoolPuddleSize = 16384; /* Default: 4096 */ Or if you want malloc() to obtain chip memory, just write #include ULONG __MemPoolFlags = MEMF_CHIP; /* Default: MEMF_ANY */ 4.) Timings ----------- I wrote a little timing program. (I don't claim that it is very good, probably someone can write a better one.) I was rather surprised about the results: Compiler Time (Mins:Secs) Notes Dice 2:46,74 Dice maps malloc() to AllocMem() directly. This is extremely poor! Especially the exit() function needs about 20% of the programs time. SAS/C 1:17,66 SAS/C uses an own pool system. Thus the exit() function is rather fast. malloc(), however, seems to be slow. gcc 0:12,84 (ixemul) gcc uses a pool system which is 0:14,34 (libnix) rather close to Exec pools. This gives good results. MemPools 0:15,10 (4096) By increasing the puddle size 0:12,14 (8192) (see "Usage") you get better 0:10,80 (16384) results! It seems worth to ry a little bit. ============================= Archive contents ============================= Original Packed Ratio Date Time Name -------- ------- ----- --------- -------- ------------- 1693 904 46.6% 04-Dec-94 21:16:28 mempools/calloc.c 17982 6993 61.1% 01-Sep-93 18:40:30 mempools/COPYING 3051 1224 59.8% 05-Dec-94 04:27:36 mempools/DMakefile 1519 836 44.9% 04-Dec-94 18:22:16 mempools/free.c 2727 1229 54.9% 04-Dec-94 18:39:32 mempools/init.c 1337 572 57.2% 05-Dec-94 03:27:22 mempools/lib.def 3294 1059 67.8% 06-Dec-94 00:47:26 mempools/libmempools.a 2893 1163 59.7% 05-Dec-94 03:28:12 mempools/Makefile 1596 868 45.6% 04-Dec-94 21:24:48 mempools/malloc.c 59 59 0.0% 04-Dec-94 18:42:08 mempools/MemPoolFlags.c 60 60 0.0% 04-Dec-94 18:40:54 mempools/MemPoolPuddleSize.c 1916 828 56.7% 06-Dec-94 00:43:16 mempools/mempools.lib 5016 2246 55.2% 06-Dec-94 00:41:30 mempools/MemPools.readme 2688 985 63.3% 06-Dec-94 00:42:00 mempools/mempoolss.lib 2628 958 63.5% 06-Dec-94 00:42:40 mempools/mempoolssr.lib 59 59 0.0% 04-Dec-94 20:09:24 mempools/MemPoolThreshSize.c 1689 872 48.3% 04-Dec-94 21:15:56 mempools/realloc.c 2862 1164 59.3% 05-Dec-94 03:27:46 mempools/SMakefile 1491 819 45.0% 05-Dec-94 04:13:30 mempools/strdup.c 4480 1653 63.1% 05-Dec-94 03:50:44 mempools/TimeMem.c 2861 1308 54.2% 05-Dec-94 01:32:44 mempools/TimeProg.c -------- ------- ----- --------- -------- 61901 25859 58.2% 07-Dec-94 18:33:10 21 files