Documentation/crypto/descore-readme.rst

1 .. SPDX-License-Identifier: GPL-2.0
13 ------------------------------------------------------------------------------
15 des - fast & portable DES encryption & decryption.
42 2.	PORTABILITY to any byte-addressable host with a 32bit unsigned C type
43 3.	Plug-compatible replacement for KERBEROS's low-level routines.
46 register-starved machines.  My discussions with Richard Outerbridge,
51 up in a parameterized fashion so it can easily be modified by speed-daemon
58 compile on a SPARCStation 1 (cc -O4, gcc -O2):
60 this code (byte-order independent):
62   - 30us per encryption (options: 64k tables, no IP/FP)
63   - 33us per encryption (options: 64k tables, FIPS standard bit ordering)
64   - 45us per encryption (options:  2k tables, no IP/FP)
65   - 48us per encryption (options:  2k tables, FIPS standard bit ordering)
66   - 275us to set a new key (uses 1k of key tables)
71 	the key setting routine. also, i have no interest in re-implementing
74 	used as drop-in replacements with mit's code or any of the mit-
80   - 53us per encryption (uses 2k of tables)
81   - 96us to set a new key (uses 2.25k of key tables)
86 	more specifically, 19-40% slower on the 68020 and 11-35% slower
93 	gcc 2.1 -O2	Sun 3/110	304  uS	369.5uS	461.8uS	 22%
94 	cc      -O1	Sun 3/110	336  uS	436.6uS	399.3uS	 19%
95 	cc      -O2	Sun 3/110	360  uS	532.4uS	505.1uS	 40%
96 	cc      -O4	Sun 3/110	365  uS	532.3uS	505.3uS	 38%
97 	gcc 2.1 -O2	Sun 4/50	 48  uS	 53.4uS	 57.5uS	 11%
98 	cc      -O2	Sun 4/50	 48  uS	 64.6uS	 64.7uS	 35%
99 	cc      -O4	Sun 4/50	 48  uS	 64.7uS	 64.9uS	 35%
106    - 68us per encryption (uses 2k of tables)
107    - 96us to set a new key (uses 2.25k of key tables)
111 	it's a bit weak on common low-level optimizations which is why
112 	it's 39%-106% slower.  because he was interested in fast crypt(3) and
113 	password-cracking applications,  he also used the same ideas to
114 	speed up the key-setting routines with impressive results.
133   - 108us per encryption (uses 34k worth of tables)
134   - 134us to set a new key (uses 32k of key tables to get this speed!)
136 	the tables used seem to be machine-independent;
151 ``highly optimized`` and tweaked Kerberos/Athena code (byte-order dependent):
153   - 165us per encryption (uses 6k worth of tables)
154   - 478us to set a new key (uses <1k of key tables)
158 	machine-independent.
161 UC Berkeley code (depends on machine-endedness):
162   -  226us per encryption
163   - 10848us to set a new key
175 i got it and looked at the code,  it really set off a lot of pet peeves -
188 the fact that this guy was computing 2 sboxes per table lookup rather
190 do the same - it was a trivial change from which i had been scared away
197 to crypt(3) in his tables - i didn't check.
201 than versions hand-written in assembly for the sparc!
211 of what i didn't want to do; all their endedness-specific ``optimizations``
219    depend on the byte order, and that bytes are 8 bits.
220    i assume word pointers can be freely cast to and from char pointers.
223 2) the typedef ``word`` means a 32 bit unsigned integral type.
224    if ``unsigned long`` is not 32 bits, change the typedef in desCore.h.
225    i assume sizeof(word) == 4 EVERYWHERE.
227 the (worst-case) cost of my NOT doing endedness-specific optimizations
230 the input and output work areas do not need to be word-aligned.
243    this will save 2 instructions and a temporary per use,
244    or about 32 to 40 instructions per en/decryption.
250 be able to get performance equal to assembly-coding, except that:
255 2) if your machine has less than 12 32-bit registers i doubt your compiler will
260    however, if you like assembly coding, the 386 does have 7 32-bit registers,
265    the 6 bits of data is the low part; it helps to exchange these.
285 	- anything more than 12 bits (bad for RISC and CISC)
286 	- greater than 127 in value (can't use movq or byte immediate on CISC)
287 	- 9-127 (may not be able to use CISC shift immediate or add/sub quick),
288 	- 1-8 were never registered, being the cheapest constants.
296 where the index is 7 bits unsigned or smaller.
302 all index scaling is done explicitly - no hidden shifts by log2(sizeof).
317 bits are manipulated in this arrangement most of the time (S7 S5 S3 S1)::
321 (the x bits are still there, i'm just emphasizing where the S boxes are).
322 bits are rotated left 4 when computing S6 S4 S2 S0::
326 the rightmost two bits are usually cleared so the lower byte can be used
327 as an index into an sbox mapping table. the next two x'd bits are set
338 	used to hold full 768-bit key.
339 	must be long-aligned.
349 	which must have odd parity (or -1 is returned) and which must
350 	not be a (semi-)weak key (or -2 is returned).
357 	m yourself, you will get a 768bit key - but then it won't
358 	be standard.  it's 768bits not 1024 because the least significant
359 	two bits of each byte are not used.  note that these two bits
365 	provide a routine that converts 128 6-bit bytes (specified in
366 	S-box mapping order or something) into the right format for you.
393 there are no machine-dependencies in the code (see porting),
397 for your compiler (MAX optimization).
403 note that i have included a kerberos-compatible interface in desUtil.c
405 to use these with kerberos or kerberos-compatible code put desCore.a
406 ahead of the kerberos-compatible library on your linker's command line.