2 Even faster inner loop in the DES assember for x86 and a modification
3 for IP/FP which is faster on x86. Both of these changes are
4 from Svend Olaf Mikkelsen <svolaf@inet.uni-c.dk>. His
5 changes make the assember run %40 faster on a pentium. This is just
6 a case of getting the instruction sequence 'just right'.
7 All credit to 'Svend' :-)
8 Quite a few special x86 'make' targets.
9 A libdes-l (lite) distribution.
12 After a bit of a pause, I'll up the major version number since this
13 is mostly a performace release. I've added x86 assember and
14 added more options for performance. A %28 speedup for gcc
15 on a pentium and the assember is a %50 speedup.
16 MIPS CPU's, sparc and Alpha are the main CPU's with speedups.
17 Run des_opts to work out which options should be used.
18 DES_RISC1/DES_RISC2 use alternative inner loops which use
19 more registers but should give speedups on any CPU that does
20 dual issue (pentium). DES_UNROLL unrolls the inner loop,
21 which costs in code size.
24 I've finally removed one of the shifts in D_ENCRYPT. This
25 meant I've changed the des_SPtrans table (spr.h), the set_key()
26 function and some things in des_enc.c. This has definitly
27 made things faster :-). I've known about this one for some
28 time but I've been too lazy to follow it up :-).
29 Noticed that in the D_ENCRYPT() macro, we can just do L^=(..)^(..)^..
30 instead of L^=((..)|(..)|(..).. This should save a register at
32 Assember for x86. The file to replace is des_enc.c, which is replaced
33 by one of the assember files found in asm. Look at des/asm/readme
36 /* Modification to fcrypt so it can be compiled to support
37 HPUX 10.x's long password format, define -DLONGCRYPT to use this.
38 Thanks to Jens Kupferschmidt <bt1cu@hpboot.rz.uni-leipzig.de>. */
40 SIGWINCH case put in des_read_passwd() so the function does not
41 'exit' if this function is recieved.
44 Modified read_pwd.c so that stdin can be read if not a tty.
45 Thanks to Jeff Barber <jeffb@issl.atl.hp.com> for the patches.
46 des_init_random_number_generator() shortened due to VMS linker
48 Added RSA's DESX cbc mode. It is a form of cbc encryption, with 2
49 8 byte quantites xored before and after encryption.
50 des_xcbc_encryption() - the name is funny to preserve the des_
51 prefix on all functions.
54 The DES_PTR macro option checked and used by SSLeay configuration
57 Added DES_LONG. If defined to 'unsigned int' on the DEC Alpha,
58 it gives a %20 speedup :-)
59 Fixed the problem with des.pl under perl5. The patches were
60 sent by Ed Kubaitis (ejk@uiuc.edu).
61 if fcrypt.c, changed values to handle illegal salt values the way
62 normal crypt() implementations do. Some programs apparently use
63 them :-(. The patch was sent by Bjorn Gronvall <bg@sics.se>
66 Bug in des(1), an error with the uuencoding stuff when the
67 'data' is small, thanks to Geoff Keating <keagchon@mehta.anu.edu.au>
71 After some emailing back and forth with
72 Colin Plumb <colin@nyx10.cs.du.edu>, I've tweaked a few things
73 and in a future version I will probably put in some of the
74 optimisation he suggested for use with the DES_USE_PTR option.
75 Extra routines from Mark Murray <mark@grondar.za> for use in
76 freeBSD. They mostly involve random number generation for use
77 with kerberos. They involve evil machine specific system calls
78 etc so I would normally suggest pushing this stuff into the
79 application and/or using RAND_seed()/RAND_bytes() if you are
80 using this DES library as part of SSLeay.
81 Redone the read_pw() function so that it is cleaner and
82 supports termios, thanks to Sameer Parekh <sameer@c2.org>
83 for the initial patches for this.
84 Renamed 3ecb_encrypt() to ecb3_encrypt(). This has been
85 done just to make things more consistent.
86 I have also now added triple DES versions of cfb and ofb.
89 Damn, Damn, Damn, as pointed out by Mike_Spreitzer.PARC@xerox.com,
90 my des_random_seed() function was only copying 4 bytes of the
91 passed seed into the init structure. It is now fixed to copy 8.
92 My own suggestion is to used something like MD5 :-)
95 While looking at my code one day, I though, why do I keep on
96 calling des_encrypt(in,out,ks,enc) when every function that
97 calls it has in and out the same. So I dropped the 'out'
98 parameter, people should not be using this function.
100 Version 3.18 30/08/95
101 Fixed a few bit with the distribution and the filenames.
102 3.17 had been munged via a move to DOS and back again.
105 Version 3.17 14/07/95
106 Fixed ede3 cbc which I had broken in 3.16. I have also
107 removed some unneeded variables in 7-8 of the routines.
109 Version 3.16 26/06/95
110 Added des_encrypt2() which does not use IP/FP, used by triple
111 des routines. Tweaked things a bit elsewhere. %13 speedup on
112 sparc and %6 on a R4400 for ede3 cbc mode.
114 Version 3.15 06/06/95
115 Added des_ncbc_encrypt(), it is des_cbc mode except that it is
116 'normal' and copies the new iv value back over the top of the
118 CHANGED des_ede3_cbc_encrypt() so that it too now overwrites
119 the iv. THIS WILL BREAK EXISTING CODE, but since this function
120 only new, I feel I can change it, not so with des_cbc_encrypt :-(.
121 I need to update the documentation.
123 Version 3.14 31/05/95
124 New release upon the world, as part of my SSL implementation.
125 New copyright and usage stuff. Basically free for all to use
126 as long as you say it came from me :-)
128 Version 3.13 31/05/95
129 A fix in speed.c, if HZ is not defined, I set it to 100.0
130 which is reasonable for most unixes except SunOS 4.x.
131 I now have a #ifdef sun but timing for SunOS 4.x looked very
132 good :-(. At my last job where I used SunOS 4.x, it was
133 defined to be 60.0 (look at the old INSTALL documentation), at
134 the last release had it changed to 100.0 since I now work with
135 Solaris2 and SVR4 boxes.
136 Thanks to Rory Chisholm <rchishol@math.ethz.ch> for pointing this
139 Version 3.12 08/05/95
140 As pointed out by The Crypt Keeper <tck@bend.UCSD.EDU>,
141 my D_ENCRYPT macro in crypt() had an un-necessary variable.
144 Version 3.11 03/05/95
145 Added des_ede3_cbc_encrypt() which is cbc mode des with 3 keys
146 and one iv. It is a standard and I needed it for my SSL code.
147 It makes more sense to use this for triple DES than
148 3cbc_encrypt(). I have also added (or should I say tested :-)
149 cfb64_encrypt() which is cfb64 but it will encrypt a partial
150 number of bytes - 3 bytes in 3 bytes out. Again this is for
151 my SSL library, as a form of encryption to use with SSL
154 Version 3.10 22/03/95
155 Fixed a bug in 3cbc_encrypt() :-(. When making repeated calls
156 to cbc3_encrypt, the 2 iv values that were being returned to
157 be used in the next call were reversed :-(.
158 Many thanks to Bill Wade <wade@Stoner.COM> for pointing out
161 Version 3.09 01/02/95
162 Fixed des_random_key to far more random, it was rather feeble
163 with regards to picking the initial seed. The problem was
164 pointed out by Olaf Kirch <okir@monad.swb.de>.
166 Version 3.08 14/12/94
167 Added Makefile.PL so libdes can be built into perl5.
168 Changed des_locl.h so RAND is always defined.
170 Version 3.07 05/12/94
171 Added GNUmake and stuff so the library can be build with
174 Version 3.06 30/08/94
175 Added rpc_enc.c which contains _des_crypt. This is for use in
177 Finally fixed the cfb_enc problems.
178 Fixed a few parameter parsing bugs in des (-3 and -b), thanks
179 to Rob McMillan <R.McMillan@its.gu.edu.au>
181 Version 3.05 21/04/94
182 for unsigned long l; gcc does not produce ((l>>34) == 0)
183 This causes bugs in cfb_enc.
184 Thanks to Hadmut Danisch <danisch@ira.uka.de>
186 Version 3.04 20/04/94
187 Added a version number to des.c and libdes.a
189 Version 3.03 12/01/94
190 Fixed a bug in non zero iv in 3cbc_enc.
192 Version 3.02 29/10/93
193 I now work in a place where there are 6+ architectures and 14+
195 Fixed TERMIO definition so the most sys V boxes will work :-)
197 Release upon comp.sources.misc
198 Version 3.01 08/10/93
199 Added des_3cbc_encrypt()
201 Version 3.00 07/10/93
202 Fixed up documentation.
203 quad_cksum definitely compatible with MIT's now.
205 Version 2.30 24/08/93
206 Triple DES now defaults to triple cbc but can do triple ecb
208 Fixed some MSDOS uuen/uudecoding problems, thanks to
211 Version 2.22 29/06/93
212 Fixed a bug in des_is_weak_key() which stopped it working :-(
213 thanks to engineering@MorningStar.Com.
215 Version 2.21 03/06/93
216 des(1) with no arguments gives quite a bit of help.
217 Added -c (generate ckecksum) flag to des(1).
218 Added -3 (triple DES) flag to des(1).
219 Added cfb and ofb routines to the library.
221 Version 2.20 11/03/93
222 Added -u (uuencode) flag to des(1).
223 I have been playing with byte order in quad_cksum to make it
224 compatible with MIT's version. All I can say is avid this
225 function if possible since MIT's output is endian dependent.
227 Version 2.12 14/10/92
228 Added MSDOS specific macro in ecb_encrypt which gives a %70
229 speed up when the code is compiled with turbo C.
231 Version 2.11 12/10/92
232 Speedup in set_key (recoding of PC-1)
233 I now do it in 47 simple operations, down from 60.
234 Thanks to John Fletcher (john_fletcher@lccmail.ocf.llnl.gov)
235 for motivating me to look for a faster system :-)
236 The speedup is probably less that 1% but it is still 13
237 instructions less :-).
239 Version 2.10 06/10/92
240 The code now works on the 64bit ETA10 and CRAY without modifications or
241 #defines. I believe the code should work on any machine that
242 defines long, int or short to be 8 bytes long.
243 Thanks to Shabbir J. Safdar (shabby@mentor.cc.purdue.edu)
244 for helping me fix the code to run on 64bit machines (he had
246 Thanks also to John Fletcher <john_fletcher@lccmail.ocf.llnl.gov>
247 for testing the routines on a CRAY.
248 read_password.c has been renamed to read_passwd.c
249 string_to_key.c has been renamed to string2key.c
251 Version 2.00 14/09/92
252 Made mods so that the library should work on 64bit CPU's.
253 Removed all my uchar and ulong defs. To many different
254 versions of unix define them in their header files in too many
255 different combinations :-)
256 IRIX - Sillicon Graphics mods (mostly in read_password.c).
257 Thanks to Andrew Daviel (advax@erich.triumf.ca)
259 Version 1.99 26/08/92
260 Fixed a bug or 2 in enc_read.c
261 Fixed a bug in enc_write.c
262 Fixed a pseudo bug in fcrypt.c (very obscure).
264 Version 1.98 31/07/92
265 Support for the ETA10. This is a strange machine that defines
266 longs and ints as 8 bytes and shorts as 4 bytes.
267 Since I do evil things with long * that assume that they are 4
268 bytes. Look in the Makefile for the option to compile for
269 this machine. quad_cksum appears to have problems but I
270 will don't have the time to fix it right now, and this is not
271 a function that uses DES and so will not effect the main uses
274 Version 1.97 20/05/92 eay
275 Fixed the Imakefile and made some changes to des.h to fix some
276 problems when building this package with Kerberos v 4.
278 Version 1.96 18/05/92 eay
279 Fixed a small bug in string_to_key() where problems could
280 occur if des_check_key was set to true and the string
281 generated a weak key.
283 Patch2 posted to comp.sources.misc
284 Version 1.95 13/05/92 eay
285 Added an alternative version of the D_ENCRYPT macro in
286 ecb_encrypt and fcrypt. Depending on the compiler, one version or the
287 other will be faster. This was inspired by
288 Dana How <how@isl.stanford.edu>, and her pointers about doing the
289 *(ulong *)((uchar *)ptr+(value&0xfc))
292 to stop the C compiler doing a <<2 to convert the long array index.
294 Version 1.94 05/05/92 eay
295 Fixed an incompatibility between my string_to_key and the MIT
296 version. When the key is longer than 8 chars, I was wrapping
297 with a different method. To use the old version, define
298 OLD_STR_TO_KEY in the makefile. Thanks to
299 viktor@newsu.shearson.com (Viktor Dukhovni).
301 Version 1.93 28/04/92 eay
302 Fixed the VMS mods so that echo is now turned off in
303 read_password. Thanks again to brennan@coco.cchs.su.oz.AU.
304 MSDOS support added. The routines can be compiled with
305 Turbo C (v2.0) and MSC (v5.1). Make sure MSDOS is defined.
307 Patch1 posted to comp.sources.misc
308 Version 1.92 13/04/92 eay
309 Changed D_ENCRYPT so that the rotation of R occurs outside of
310 the loop. This required rotating all the longs in sp.h (now
311 called spr.h). Thanks to Richard Outerbridge <71755.204@CompuServe.COM>
312 speed.c has been changed so it will work without SIGALRM. If
313 times(3) is not present it will try to use ftime() instead.
315 Version 1.91 08/04/92 eay
316 Added -E/-D options to des(1) so it can use string_to_key.
317 Added SVR4 mods suggested by witr@rwwa.COM
318 Added VMS mods suggested by brennan@coco.cchs.su.oz.AU. If
319 anyone knows how to turn of tty echo in VMS please tell me or
320 implement it yourself :-).
321 Changed FILE *IN/*OUT to *DES_IN/*DES_OUT since it appears VMS
322 does not like IN/OUT being used.
324 Libdes posted to comp.sources.misc
325 Version 1.9 24/03/92 eay
326 Now contains a fast small crypt replacement.
327 Added des(1) command.
328 Added des_rw_mode so people can use cbc encryption with
329 enc_read and enc_write.
331 Version 1.8 15/10/91 eay
333 Many thanks to Keith Reynolds (keithr@sco.COM) for pointing this
336 Version 1.7 24/09/91 eay
338 set_key is 4 times faster and takes less space.
339 There are a few minor changes that could be made.
341 Version 1.6 19/09/1991 eay
342 Finally go IP and FP finished.
343 Now I need to fix set_key.
344 This version is quite a bit faster that 1.51
346 Version 1.52 15/06/1991 eay
347 20% speedup in ecb_encrypt by changing the E bit selection
348 to use 2 32bit words. This also required modification of the
349 sp table. There is still a way to speedup the IP and IP-1
350 (hints from outer@sq.com) still working on this one :-(.
352 Version 1.51 07/06/1991 eay
353 Faster des_encrypt by loop unrolling
354 Fixed bug in quad_cksum.c (thanks to hughes@logos.ucs.indiana.edu)
356 Version 1.50 28/05/1991 eay
357 Optimised the code a bit more for the sparc. I have improved the
358 speed of the inner des_encrypt by speeding up the initial and
361 Version 1.40 23/10/1990 eay
362 Fixed des_random_key, it did not produce a random key :-(
364 Version 1.30 2/10/1990 eay
365 Have made des_quad_cksum the same as MIT's, the full package
366 should be compatible with MIT's
367 Have tested on a DECstation 3100
368 Still need to fix des_set_key (make it faster).
369 Does des_cbc_encrypts at 70.5k/sec on a 3100.
371 Version 1.20 18/09/1990 eay
372 Fixed byte order dependencies.
373 Fixed (I hope) all the word alignment problems.
374 Speedup in des_ecb_encrypt.
376 Version 1.10 11/09/1990 eay
377 Added des_enc_read and des_enc_write.
378 Still need to fix des_quad_cksum.
379 Still need to document des_enc_read and des_enc_write.
381 Version 1.00 27/08/1990 eay