drygdryg/calc
1
0
Fork 0
mirror of https://github.com/lcn2/calc.git synced 2025-08-19 01:13:27 +03:00
Code Issues Packages Projects Releases Wiki Activity
Files
b4952bd44ff90de39942cf6e39d85acc419ad3d1
calc/zrand.c
Landon Curt Noll 8aedcf801a Release calc version 2.11.6.2
2017-05-21 15:38:43 -07:00

2111 lines
60 KiB
C
Raw Blame History

/*
* zrand - subtractive 100 shuffle generator
*
* Copyright (C) 1999 Landon Curt Noll
*
* Calc is open software; you can redistribute it and/or modify it under
* the terms of the version 2.1 of the GNU Lesser General Public License
* as published by the Free Software Foundation.
*
* Calc is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
* or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General
* Public License for more details.
*
* A copy of version 2.1 of the GNU Lesser General Public License is
* distributed with calc under the filename COPYING-LGPL. You should have
* received a copy with calc; if not, write to Free Software Foundation, Inc.
* 59 Temple Place, Suite 330, Boston, MA 02111-1307, USA.
*
* @(#) $Revision: 29.5 $
* @(#) $Id: zrand.c,v 29.5 2003/01/14 00:54:24 chongo Exp $
* @(#) $Source: /usr/local/src/cmd/calc/RCS/zrand.c,v $
*
* Under source code control: 1995/01/07 09:45:25
* File existed as early as: 1994
*
* chongo <was here> /\oo/\ http://www.isthe.com/chongo/
* Share and enjoy! :-) http://www.isthe.com/chongo/tech/comp/calc/
*/
/*
* AN OVERVIEW OF THE FUNCTIONS:
*
* This module contains an Subtractive 100 shuffle generator wrapped inside
* of a shuffle generator.
*
* We refer to this generator as the s100 generator.
*
* rand - s100 shuffle generator
* srand - seed the s100 shuffle generator
*
* This generator has two distinct parts, the s100 generator
* and the shuffle generator.
*
* The subtractive 100 generator is described in Knuth's "The Art of
* Computer Programming - Seminumerical Algorithms", Vol 2, 3rd edition
* (1998), Section 3.6, page 186, formula (2).
*
* The "use only the first 100 our of every 1009" is described in
* Knuth's "The Art of Computer Programming - Seminumerical Algorithms",
* Vol 2, 3rd edition (1998), Section 3.6, page 188".
*
* The period and other properties of this generator make it very
* useful to 'seed' other generators.
*
* The shuffle generator is described in Knuth's "The Art of Computer
* Programming - Seminumerical Algorithms", Vol 2, 3rd edition (1998),
* Section 3.2.2, page 34, Algorithm B.
*
* The shuffle generator is fast and serves as a fairly good standard
* pseudo-random generator. If you need a fast generator and do not
* need a cryptographically strong one, this generator is likely to do
* the job.
*
* The shuffle generator is feed values by the subtractive 100 process.
*
******************************************************************************
*
* GOALS:
*
* The goals of this package are:
*
* all magic numbers are explained
*
* I distrust systems with constants (magic numbers) and tables
* that have no justification (e.g., DES). I believe that I have
* done my best to justify all of the magic numbers used.
*
* full documentation
*
* You have this source file, plus background publications,
* what more could you ask?
*
* large selection of seeds
*
* Seeds are not limited to a small number of bits. A seed
* may be of any size.
*
* the strength of the generators may be tuned to meet the need
*
* By using the appropriate seed and other arguments, one may
* increase the strength of the generator to suit the need of
* the application. One does not have just a few levels.
*
* Even though I have done my best to implement a good system, you still
* must use these routines your own risk.
*
* Share and enjoy! :-)
*/
/*
* ON THE GENERATORS:
*
* The subtractive 100 generator has a good period, and is fast. It is
* reasonable as generators go, though there are better ones available.
* The shuffle generator has a very good period, and is fast. It is
* fairly good as generators go, particularly when it is feed reasonably
* random numbers. Because of this, we use feed values from the subtractive
* 100 process into the shuffle generator.
*
* The s100 generator uses 2 tables:
*
* subtractive table - 100 entries of 64 bits used by the subtractive 100
* part of the s100 generator
*
* shuffle table - 256 entries of 64 bits used by the shuffle
* part of the s100 generator and feed by the
* subtractive table.
*
* Casual direct use of the shuffle generator may be acceptable. If one
* desires cryptographically strong random numbers, or if one is paranoid,
* one should use the Blum generator instead (see zrandom.c).
*
* The s100 generator as the following calc interfaces:
*
* rand(min,max) (where min < max)
*
* Print an s100 generator random value over interval [a,b).
*
* rand()
*
* Same as rand(0, 2^64). Print 64 bits.
*
* rand(lim) (where 0 > lim)
*
* Same as rand(0, lim).
*
* randbit(x) (where x > 0)
*
* Same as rand(0, 2^x). Print x bits.
*
* randbit(skip) (where skip < 0)
*
* Skip random bits and return the bit skip count (-skip).
*/
/*
* INITIALIZATION AND SEEDS:
*
* All generators come already seeded with precomputed initial constants.
* Thus, it is not required to seed a generator before using it.
*
* The s100 generator may be initialized and seeded via srand().
*
* Using a seed of '0' will reload generators with their initial states.
*
* srand(0) restore subtractive 100 generator to the initial state
*
* The above single arg calls are fairly fast.
*
* Optimal seed range for the s100 generator:
*
* There is no limit on the size of a seed. On the other hand,
* extremely large seeds require large tables and long seed times.
* Using a seed in the range of [2^64, 2^64 * 100!) should be
* sufficient for most purposes. An easy way to stay within this
* range to to use seeds that are between 21 and 178 digits, or
* 64 to 588 bits long.
*
* To help make the generator produced by seed S, significantly
* different from S+1, seeds are scrambled prior to use. The
* function randreseed64() maps [0,2^64) into [0,2^64) in a 1-to-1
* and onto fashion.
*
* The purpose of the randreseed64() is not to add security. It
* simply helps remove the human perception of the relationship
* between the seed and the production of the generator.
*
* The randreseed64() process does not reduce the security of the
* generators. Every seed is converted into a different unique seed.
* No seed is ignored or favored.
*
******************************************************************************
*
* srand(seed)
*
* Seed the s100 generator.
*
* seed != 0:
* ---------
* Any buffered random bits are flushed. The subtractive table is loaded
* with the default subtractive table. The low order 64 bits of seed is
* xor-ed against each table value. The subtractive table is shuffled
* according to seed/2^64.
*
* The following calc code produces the same effect:
*
* (* reload default subtractive table xored with low 64 seed bits *)
* seed_xor = seed & ((1<<64)-1);
* for (i=0; i < 100; ++i) {
* subtractive[i] = xor(default_subtractive[i], seed_xor);
* }
*
* (* shuffle the subtractive table *)
* seed >>= 64;
* for (i=100; seed > 0 && i > 0; --i) {
* quomod(seed, i+1, seed, j);
* swap(subtractive[i], subtractive[j]);
* }
*
* Seed must be >= 0. All seed values < 0 are reserved for future use.
*
* The subtractive 100 pointers are reset to subtractive[36] and
* subtractive[99]. Last the shuffle table is loaded with successive
* values from the subtractive 100 generator.
*
* seed == 0:
* ---------
* Restore the initial state and modulus of the s100 generator.
* After this call, the s100 generator is restored to its initial
* state after calc started.
*
* The subtractive 100 pointers are reset to subtractive[36] and
* subtractive[99]. Last the shuffle table is loaded with successive
* values from the subtractive 100 generator.
*
******************************************************************************
*
* srand(mat100)
*
* Seed the s100 generator.
*
* Any buffered random bits are flushed. The subtractive table with the
* first 100 entries of the array mat100, mod 2^64.
*
* The subtractive 100 pointers are reset to subtractive[36] and
* subtractive[99]. Last the shuffle table is loaded with successive
* values from the subtractive 100 generator.
*
******************************************************************************
*
* srand()
*
* Return current s100 generator state. This call does not alter
* the generator state.
*
******************************************************************************
*
* srand(state)
*
* Restore the s100 state and return the previous state. Note that
* the argument state is a rand state value (isrand(state) is true).
* Any internally buffered random bits are restored.
*
* The states of the s100 generators can be saved by calling the seed
* function with no arguments, and later restored by calling the seed
* functions with that same return value.
*
* rand_state = srand();
* ... generate random bits ...
* prev_rand_state = srand(rand_state);
* ... generate the same random bits ...
* srand() == prev_rand_state; (* is true *)
*
* Saving the state just after seeding a generator and restoring it later
* as a very fast way to reseed a generator.
*/
/*
* TRUTH IN ADVERTISING:
*
* A "truth in advertising" issue is the use of the term
* 'pseudo-random'. All deterministic generators are pseudo random.
* This is opposed to true random generators based on some special
* physical device.
*
* A final "truth in advertising" issue deals with how the magic numbers
* found in this file were generated. Detains can be found in the
* various functions, while a overview can be found in the "SOURCE FOR
* MAGIC NUMBERS" section below.
*/
/*
* SOURCE OF MAGIC NUMBERS:
*
* Most of the magic constants used on this file ultimately are
* based on LavaRnd. LavaRnd produced them via a cryprographic
* of the digitization of chaotic system that consisted of a noisy
* digital camera and 6 Lava Lite(R) lamps.
*
* BTW: Lava Lite(R) is a trademark of Haggerty Enterprises, Inc.
*
* The first 100 groups of 64 bit bits were used to initialize init_s100.slot.
*
* The function, randreseed64(), uses 2 primes to scramble 64 bits
* into 64 bits. These primes were also extracted from the Rand
* Book of Random Numbers. See randreseed64() for details.
*
* The shuffle table size is longer than the 100 entries recommended
* by Knuth. We use a power of 2 shuffle table length so that the
* shuffle process can select a table entry from a new subtractive 100
* value by extracting its low order bits. The value 256 is convenient
* in that it is the size of a byte which allows for easy extraction.
*
* We use the upper byte of the subtractive 100 value to select the shuffle
* table entry because it allows all of 64 bits to play a part in the
* entry selection. If we were to select a lower 8 bits in the 64 bit
* value, carries that propagate above our 8 bits would not impact the
* s100 generator output.
*
******************************************************************************
*
* FOR THE PARANOID:
*
* The truly paranoid might suggest that my claims in the MAGIC NUMBERS
* section are a lie intended to entrap people. Well they are not, but
* if you that paranoid why would you use a non-cryprographically strong
* pseudo-random number generator in the first place? You would be
* better off using the random() builtin function.
*
* The two constants that were picked from the Rand Book of Random Numbers
* The random numbers from the Rand Book of Random Numbers can be
* verified by anyone who obtains the book. As these numbers were
* created before I (Landon Curt Noll) was born (you can look up
* my birth record if you want), I claim to have no possible influence
* on their generation.
*
* There is a very slight chance that the electronic copy of the
* Rand Book of Random Numbers that I was given access to differs
* from the printed text. I am willing to provide access to this
* electronic copy should anyone wants to compare it to the printed text.
*
* When using the s100 generator, one may select your own 100 subtractive
* values by calling:
*
* srand(mat100)
*
* and avoid using my magic numbers. The randreseed64 process is NOT
* applied to the matrix values. Of course, you must pick good subtractive
* 100 values yourself!
*
* One might object to the complexity of the seed scramble/mapping
* via the randreseed64() function. The randreseed64() function maps:
*
* 0 ==> 0
* 10239951819489363767 ==> 1363042948800878693
*
* so that srand(0) does the default action and randreseed64() remains
* an 1-to-1 and onto map. Thus calling srand(0) with the randreseed64()
* process would be the same as calling srand(4967126403401436567) without
* it. No extra security is gained or reduced by using the randreseed64()
* process. The meaning of seeds are exchanged, but not lost or favored
* (used by more than one input seed).
*/
#include <stdio.h>
#include "zrand.h"
#include "have_const.h"
/*
* default s100 generator state
*
* This is the state of the s100 generator after initialization, or srand(0),
* or srand(0) is called. The init_s100 value is never changed, only copied.
*/
static CONST RAND init_s100 = {
1, /* seeded */
0, /* no buffered bits */
#if FULL_BITS == SBITS /* buffer */
{0},
#elif 2*FULL_BITS == SBITS
{0, 0},
#else
/\../\ BASEB is assumed to be 16 or 32 /\../\ !!!
#endif
INIT_J, /* j */
INIT_K, /* k */
RAND_CONSEQ_USE, /* use this many before skipping values */
/* NOTE: Due to a SunOS cc bug, don't put spaces in the SVAL call! */
{ /* subtractive 100 table */
SVAL(c8c0370c,7db7dc19), SVAL(738e33b9,40a06fbb),
SVAL(481abb76,a859ed2b), SVAL(74106bb3,9ccdccb5),
SVAL(05a8eeb5,c3173bfc), SVAL(efd5100d,5a02e577),
SVAL(a69271f7,4030b24a), SVAL(641282fc,16fe22c5),
SVAL(7aa7267c,40438da3), SVAL(1fdf4abd,c2d878d1),
SVAL(d9899e7a,95702379), SVAL(5ea8e217,d02d7f08),
SVAL(770587fe,4d47a353), SVAL(de7d1bdd,0a33a2b8),
SVAL(4378c3c5,900e7c45), SVAL(77c94478,19a514f9),
SVAL(fc5edb22,843d1d32), SVAL(4fc42ce5,e8ee5e6e),
SVAL(c938713c,8488013e), SVAL(6a318f03,20ab0cac),
SVAL(73e6d1a3,ffc8bff3), SVAL(0cd3232a,8ca96aa7),
SVAL(605c8036,905f770d), SVAL(4d037b00,8b8d04a2),
SVAL(1ed81965,cb277294), SVAL(408d9c47,7a254ff3),
SVAL(8b68587a,e26c7377), SVAL(cff191a4,8a48832f),
SVAL(12d3df1d,8aeb6fe6), SVAL(b2bf907e,1feda37a),
SVAL(4e5f7719,3bb5f39f), SVAL(33ebcf6f,8f5d1581),
SVAL(203c8e48,d33654eb), SVAL(68d3656e,f19c8a4e),
SVAL(3ec20b04,986eb2af), SVAL(5d73a03b,062c3841),
SVAL(836ce709,5d4e49eb), SVAL(2310bc40,c3f49221),
SVAL(3868ee48,a6d0cbf6), SVAL(67578aa6,4a43deb1),
SVAL(6e3426c1,150dfc26), SVAL(c541ccaa,3131be30),
SVAL(f7e57432,cec7aab2), SVAL(2b35de99,8cb3c873),
SVAL(7b9f7764,8663a5d7), SVAL(23b00e6a,a771e5a6),
SVAL(859c775c,a9985d05), SVAL(99636ea1,6b692f1f),
SVAL(8700ac70,3730800d), SVAL(46142502,4298a753),
SVAL(ea4a411b,809e955f), SVAL(3119ad40,33709dfb),
SVAL(b76a6c6e,5f01cb7c), SVAL(6109dc8a,15984eaf),
SVAL(5d686db9,a5ca9505), SVAL(8e80d761,3b7e6add),
SVAL(79cbd718,de6f6fd3), SVAL(40e9cd15,1da0f699),
SVAL(e82158ba,b24f312d), SVAL(79a4c927,f5e5c36b),
SVAL(c25247c9,a0039333), SVAL(93687116,1766d81d),
SVAL(3c6a03b4,a6741327), SVAL(c8a7b6e8,c002f29a),
SVAL(0e2a67c6,7bbd5ea3), SVAL(0929042d,441eabc1),
SVAL(7dbe232a,25e82085), SVAL(8cfb26e5,44fbac3d),
SVAL(8e40384d,388ab983), SVAL(48dc1230,554632f8),
SVAL(ab405048,ab492397), SVAL(21c9e2f5,a118e387),
SVAL(484d1a8c,343b61b5), SVAL(d49e3dec,ab256f26),
SVAL(e615c7fd,78f2d2e3), SVAL(8442cc33,ce6cc2ed),
SVAL(0a3b93d8,44d4bbf6), SVAL(2d7e4efe,9301de77),
SVAL(33711b76,d8790d8a), SVAL(c07dc30e,44df77e7),
SVAL(b9132ed0,9ddd508f), SVAL(45d06cf8,c6fb43cc),
SVAL(22bed18a,d585dd7b), SVAL(61c6cced,10799ffa),
SVAL(d7f2393b,e4bd9aa9), SVAL(706753fb,cfd55094),
SVAL(f65a6713,ede6e446), SVAL(8bf6dfae,47c0d5c3),
SVAL(fb4dfc17,9f7927d6), SVAL(12ebbc16,e212c297),
SVAL(43c71283,a00a954c), SVAL(8957087a,e7bd40a5),
SVAL(b0859d71,08344837), SVAL(fbf4b9a3,aeb313f5),
SVAL(5e66e5be,ce81823a), SVAL(09a11c6e,58ad6da1),
SVAL(c76f4316,c608054f), SVAL(b5821361,46084099),
SVAL(4210008f,17a725ed), SVAL(e5ff8912,d347c481)
},
/* NOTE: Due to a SunOS cc bug, don't put spaces in the SVAL call! */
{ /* shuffle table */
SVAL(69a2296c,ec8abd57), SVAL(867e1869,99a6df81),
SVAL(c05ab96b,d849a48a), SVAL(7eb3ce0c,fa00554b),
SVAL(520d01f6,5a5a9acd), SVAL(d4ef1e33,36022d81),
SVAL(af44772b,c6f84f70), SVAL(647e85a6,a7c55173),
SVAL(26746cf1,959df8d1), SVAL(98681a90,4db28abd),
SVAL(b146c969,744c5cd2), SVAL(8ce69d1f,706f88c2),
SVAL(fd12eac4,21b4a748), SVAL(f12e70fe,2710eea5),
SVAL(0b8f7805,5901f2b5), SVAL(48860a76,4f2c115e),
SVAL(0edf6d2a,30767e2c), SVAL(8a6d7dc5,fce2713b),
SVAL(46a362ea,4e0e2346), SVAL(6c369a0a,359f5aa7),
SVAL(dfca81fe,41def54e), SVAL(4b733819,96c2bc4e),
SVAL(659e8b99,6f3f14f9), SVAL(8b97b934,93d47e6f),
SVAL(a73a8704,dfa10a55), SVAL(8d9eafe9,b06503da),
SVAL(2556fb88,f32336b0), SVAL(e71e9f75,1002a161),
SVAL(27a7be6e,200af907), SVAL(1b9b734e,d028e9a3),
SVAL(950cfeed,4c0be0d3), SVAL(f4c41694,2536d275),
SVAL(f05a58e8,5687b76e), SVAL(ba53ac01,71a62d54),
SVAL(4b14cbcb,285adc96), SVAL(fdf66edd,b00a5557),
SVAL(bb43d58d,185b6ea1), SVAL(905db9cd,f355c9a6),
SVAL(fc3a07fc,04429c8a), SVAL(65d7e365,aa3a4f7e),
SVAL(2d284c18,b243ac65), SVAL(72fba65d,44e417fd),
SVAL(422d50b4,5c934805), SVAL(b62a6053,d1587441),
SVAL(a5e71ce9,6f7ae035), SVAL(93abca2e,595c8dd8),
SVAL(534231af,e39afad5), SVAL(08d26cac,12eaad56),
SVAL(ec18bf8d,7fb1b1c2), SVAL(3d28ea16,faf6f09b),
SVAL(ea357a78,16697dd6), SVAL(51471ea1,420f3f51),
SVAL(5e051aeb,7f8946b4), SVAL(881be097,0cf0524c),
SVAL(d558b25b,1b31489e), SVAL(707d1a94,3a8b065c),
SVAL(37017e66,568ff836), SVAL(b9cd627c,24c2f747),
SVAL(1485549f,fb1d9ff6), SVAL(308d32d9,bdf2dc6f),
SVAL(4d4142ca,d543818a), SVAL(5d9c7aee,87ebba43),
SVAL(81c5bdd8,e17adb2f), SVAL(3dc9752e,c8d8677a),
SVAL(66b086e6,c34e4212), SVAL(3af7a90d,c62b25e3),
SVAL(f8349f79,35539315), SVAL(6bcfd9d5,a22917f0),
SVAL(8639bb76,5f5ee517), SVAL(d3c5e369,8095b092),
SVAL(8a33851e,7eb44748), SVAL(5e29d443,ea54bbcf),
SVAL(0f84651f,4d59a834), SVAL(85040bea,f1a5f951),
SVAL(3dba1c74,98002078), SVAL(5d70712b,f0b2cc15),
SVAL(fa3af8eb,cce8e5a7), SVAL(fb3e2237,04bba57d),
SVAL(5d3b8785,8a950434), SVAL(ce3112bd,ba3f8dcf),
SVAL(44904f55,860d3051), SVAL(cec8fed4,4ed3e98b),
SVAL(4581698d,25d01ea4), SVAL(11eb6828,9a9548e0),
SVAL(796cb4c6,e911fac8), SVAL(2164cf26,b5fd813e),
SVAL(4ac8e0f5,d5de640f), SVAL(e9e757d7,8802ab4e),
SVAL(3c97de26,f49dfcbd), SVAL(c604881b,6ee6dbe6),
SVAL(a7c22a6e,57d6154e), SVAL(234e2370,877b3cc7),
SVAL(c0bdb72b,df1f8358), SVAL(6522e0fc,a95b7b55),
SVAL(ba174c90,22344162), SVAL(712c9b2d,75d48867),
SVAL(240f7e92,e59f3700), SVAL(e83cc2d4,ad95d763),
SVAL(8509445a,4336d717), SVAL(f1e572c5,dfff1804),
SVAL(ed10eb5d,623232dd), SVAL(9205ea1b,d4f957e8),
SVAL(4973a54f,2ff062f5), SVAL(26b018f1,e8c48cd5),
SVAL(56908401,d1c7ed9f), SVAL(2e48937b,df89a247),
SVAL(9d53069b,2be47129), SVAL(98069e3b,c048a2b0),
SVAL(f25b7d65,1cd83f93), SVAL(2b004e6c,e6f886c8),
SVAL(f618442a,5c635935), SVAL(a502ab5c,7198e052),
SVAL(c14241a4,a6c41b0b), SVAL(720e845a,7db9b18e),
SVAL(2abb13e9,4b713918), SVAL(90fc0c20,7f52467d),
SVAL(799c8ccd,7868d348), SVAL(f4817ced,912a0ea4),
SVAL(d68c0f4c,c4903a57), SVAL(a3171f29,e2b7934c),
SVAL(b1158baa,0b4ccc22), SVAL(f5d85553,49a29eda),
SVAL(59d1a078,959442ef), SVAL(db9b4a96,a67fd518),
SVAL(cc7ca9ee,d2870636), SVAL(548f021c,ecf59920),
SVAL(25b7f4b6,571bc8c5), SVAL(4fa52747,3a44f536),
SVAL(b246845f,df0ebdc2), SVAL(dd8d68ae,42058793),
SVAL(3ba13328,9f6c39fb), SVAL(8bfdfbf3,7b6b42af),
SVAL(fb34c5ca,7fb2b3b0), SVAL(2345dcec,d428e32a),
SVAL(6891e850,ad42b63e), SVAL(930642c8,362c1381),
SVAL(13871e9b,1886aff5), SVAL(d0cf2407,482bda55),
SVAL(125b5fc9,5069bc31), SVAL(9b71d0a9,f07dfa5d),
SVAL(55c044cc,6712e524), SVAL(f0377358,bb601978),
SVAL(152ad5f8,7fa51e8b), SVAL(e5ebf478,9fcdd9af),
SVAL(3d78e18c,66ebce7e), SVAL(8246db72,f36aa83f),
SVAL(cc6ddc6d,2c64c0a3), SVAL(a758d687,0d91851e),
SVAL(24b20a6f,9488ee36), SVAL(be11ccdf,09798197),
SVAL(11aca015,99c1f4e3), SVAL(40e89e36,6437ac05),
SVAL(c8bfc762,5af675f8), SVAL(6367c578,b577e759),
SVAL(00380346,615f0b74), SVAL(ee964cc4,8de07d81),
SVAL(17f6ac16,859d9261), SVAL(092f4a17,3a6e2f6c),
SVAL(79981a3d,b9024b95), SVAL(36db1660,04f7f540),
SVAL(c36252cf,65a2f1c8), SVAL(705b6fde,124c9bd2),
SVAL(31e58dda,85db40ce), SVAL(6342b1a5,9f5e8d6d),
SVAL(5c2c67d0,bd6d1d4d), SVAL(1fe5b46f,ba7e069d),
SVAL(21c46c6c,ac72e13c), SVAL(b80c5fd5,9eb8f52a),
SVAL(56c3aebf,a74c92bc), SVAL(c1aff1fc,bf8c4196),
SVAL(2b1df645,754ad208), SVAL(5c734600,d46eeb50),
SVAL(e0ff1b12,6a70a765), SVAL(d5416497,7a94547c),
SVAL(67b59d7c,4ea35206), SVAL(53be7146,779203b4),
SVAL(6b589fe5,414026b8), SVAL(9e81016c,3083bfee),
SVAL(b23526b9,3b4b7671), SVAL(4fa9ffb1,7ee300ba),
SVAL(6217e212,ad05fb21), SVAL(f5b3fcd3,b294e6c2),
SVAL(ac040bbe,216beb2a), SVAL(1f8d8a54,71d0e78c),
SVAL(b6d15b41,9cfec96b), SVAL(c5477845,d0508c78),
SVAL(5b486e81,b4bba621), SVAL(90c35c94,ef4c4121),
SVAL(efce7346,f6a6bc55), SVAL(a27828d9,25bdb9bb),
SVAL(e3a53095,a1f0b205), SVAL(1bfa6093,d9f208ab),
SVAL(fb078f6a,6842cdf4), SVAL(07806d72,97133a38),
SVAL(2c6c901b,a3ce9592), SVAL(1f0ab2cf,ebc1b789),
SVAL(2ce81415,e2d03d5e), SVAL(7da45d5b,aa9f2417),
SVAL(3be4f76d,dd800682), SVAL(dbf4e4a3,364d72d3),
SVAL(b538cccf,4fc59da5), SVAL(b0aa39d5,487f66ec),
SVAL(2fd28dfd,87927d3d), SVAL(d14e77f0,5900c6b1),
SVAL(2523fad2,5330c7b4), SVAL(991b5938,d82368a4),
SVAL(b7c11443,2b9c1302), SVAL(db842db6,1394b116),
SVAL(3641548d,78ed26d8), SVAL(274fa8ef,0a61dacf),
SVAL(a554ba63,112df6f1), SVAL(7b7fe985,6b50438d),
SVAL(c9fa0042,bb63bbad), SVAL(3abf45d0,e27f00da),
SVAL(d95faa15,9f87aabb), SVAL(4a95012e,3488e7ae),
SVAL(1be2bdb9,0c642d04), SVAL(145c8881,8b4abf3e),
SVAL(7f9fb635,544cf17f), SVAL(b8ab2f62,cc78db70),
SVAL(8ee64bcd,b4242f9a), SVAL(abd52858,95dad129),
SVAL(be722c2f,ccf31141), SVAL(7c330703,575e26a9),
SVAL(45d3e3b3,361b79e4), SVAL(241163a7,54b2e6a6),
SVAL(8f678d7d,f7cacb77), SVAL(988a68a4,83211d19),
SVAL(79599598,ba7836f6), SVAL(4850c887,eeda68bf),
SVAL(afa69a71,8052ce25), SVAL(8b21efc6,bdd73573),
SVAL(89dbae18,d0972493), SVAL(560776bf,537d9454),
SVAL(3c009f78,165310f2), SVAL(a3680021,0160c3af),
SVAL(3353ec3c,a643bd40), SVAL(7e593f99,911dab02),
SVAL(72d1ddd9,4f676e89), SVAL(fd18b8bd,6b43c0ea),
SVAL(43cacef2,ddbd697d), SVAL(2868a4d0,acefe884),
SVAL(5f377b63,a506f013), SVAL(eaa0975e,05ca662b),
SVAL(3740e6b8,eb433931), SVAL(ce85df00,08557948),
SVAL(784745fb,547e33f9), SVAL(4a1fc5d4,e5c6f598),
SVAL(85fa6fec,768430a7), SVAL(990d0c24,d2332a51),
SVAL(55245c2c,33b676d5), SVAL(b1091519,e2bcfa71),
SVAL(38521478,d23a28d8), SVAL(9b794f89,9a573010),
SVAL(61d225e8,699bb486), SVAL(21476d24,1c2158b0)
}
};
/*
* default subtractive 100 table
*
* The subtractive 100 table in init_s100 has been processed 256 times in order
* to preload the shuffle table. The array below is the table before
* this processing. These values have came from LavaRnd.
*
* This array is never changed, only copied.
*/
static CONST FULL def_subtract[SCNT] = {
/* NOTE: Due to a SunOS cc bug, don't put spaces in the SVAL call! */
SVAL(c8c0370c,7db7dc19), SVAL(738e33b9,40a06fbb),
SVAL(481abb76,a859ed2b), SVAL(74106bb3,9ccdccb5),
SVAL(05a8eeb5,c3173bfc), SVAL(efd5100d,5a02e577),
SVAL(a69271f7,4030b24a), SVAL(641282fc,16fe22c5),
SVAL(7aa7267c,40438da3), SVAL(1fdf4abd,c2d878d1),
SVAL(d9899e7a,95702379), SVAL(5ea8e217,d02d7f08),
SVAL(770587fe,4d47a353), SVAL(de7d1bdd,0a33a2b8),
SVAL(4378c3c5,900e7c45), SVAL(77c94478,19a514f9),
SVAL(fc5edb22,843d1d32), SVAL(4fc42ce5,e8ee5e6e),
SVAL(c938713c,8488013e), SVAL(6a318f03,20ab0cac),
SVAL(73e6d1a3,ffc8bff3), SVAL(0cd3232a,8ca96aa7),
SVAL(605c8036,905f770d), SVAL(4d037b00,8b8d04a2),
SVAL(1ed81965,cb277294), SVAL(408d9c47,7a254ff3),
SVAL(8b68587a,e26c7377), SVAL(cff191a4,8a48832f),
SVAL(12d3df1d,8aeb6fe6), SVAL(b2bf907e,1feda37a),
SVAL(4e5f7719,3bb5f39f), SVAL(33ebcf6f,8f5d1581),
SVAL(203c8e48,d33654eb), SVAL(68d3656e,f19c8a4e),
SVAL(3ec20b04,986eb2af), SVAL(5d73a03b,062c3841),
SVAL(836ce709,5d4e49eb), SVAL(2310bc40,c3f49221),
SVAL(3868ee48,a6d0cbf6), SVAL(67578aa6,4a43deb1),
SVAL(6e3426c1,150dfc26), SVAL(c541ccaa,3131be30),
SVAL(f7e57432,cec7aab2), SVAL(2b35de99,8cb3c873),
SVAL(7b9f7764,8663a5d7), SVAL(23b00e6a,a771e5a6),
SVAL(859c775c,a9985d05), SVAL(99636ea1,6b692f1f),
SVAL(8700ac70,3730800d), SVAL(46142502,4298a753),
SVAL(ea4a411b,809e955f), SVAL(3119ad40,33709dfb),
SVAL(b76a6c6e,5f01cb7c), SVAL(6109dc8a,15984eaf),
SVAL(5d686db9,a5ca9505), SVAL(8e80d761,3b7e6add),
SVAL(79cbd718,de6f6fd3), SVAL(40e9cd15,1da0f699),
SVAL(e82158ba,b24f312d), SVAL(79a4c927,f5e5c36b),
SVAL(c25247c9,a0039333), SVAL(93687116,1766d81d),
SVAL(3c6a03b4,a6741327), SVAL(c8a7b6e8,c002f29a),
SVAL(0e2a67c6,7bbd5ea3), SVAL(0929042d,441eabc1),
SVAL(7dbe232a,25e82085), SVAL(8cfb26e5,44fbac3d),
SVAL(8e40384d,388ab983), SVAL(48dc1230,554632f8),
SVAL(ab405048,ab492397), SVAL(21c9e2f5,a118e387),
SVAL(484d1a8c,343b61b5), SVAL(d49e3dec,ab256f26),
SVAL(e615c7fd,78f2d2e3), SVAL(8442cc33,ce6cc2ed),
SVAL(0a3b93d8,44d4bbf6), SVAL(2d7e4efe,9301de77),
SVAL(33711b76,d8790d8a), SVAL(c07dc30e,44df77e7),
SVAL(b9132ed0,9ddd508f), SVAL(45d06cf8,c6fb43cc),
SVAL(22bed18a,d585dd7b), SVAL(61c6cced,10799ffa),
SVAL(d7f2393b,e4bd9aa9), SVAL(706753fb,cfd55094),
SVAL(f65a6713,ede6e446), SVAL(8bf6dfae,47c0d5c3),
SVAL(fb4dfc17,9f7927d6), SVAL(12ebbc16,e212c297),
SVAL(43c71283,a00a954c), SVAL(8957087a,e7bd40a5),
SVAL(b0859d71,08344837), SVAL(fbf4b9a3,aeb313f5),
SVAL(5e66e5be,ce81823a), SVAL(09a11c6e,58ad6da1),
SVAL(c76f4316,c608054f), SVAL(b5821361,46084099),
SVAL(4210008f,17a725ed), SVAL(e5ff8912,d347c481)
};
/*
* Linear Congruential Constants
*
* a = 6316878969928993981 = 0x57aa0ff473c0ccbd
* c = 1363042948800878693 = 0x12ea805718e09865
*
* These constants are used in the randreseed64(). See below.
*/
/* NOTE: Due to a SunOS cc bug, don't put spaces in the SHVAL call! */
static CONST HALF a_vec[SHALFS] = {SHVAL(57aa,0ff4,73c0,ccbd)};
static CONST HALF c_vec[SHALFS] = {SHVAL(12ea,8057,18e0,9865)};
static CONST ZVALUE a_val = {(HALF *)a_vec, SHALFS, 0};
static CONST ZVALUE c_val = {(HALF *)c_vec, SHALFS, 0};
/*
* current s100 generator state
*/
static RAND s100;
/*
* declare static functions
*/
static void randreseed64(ZVALUE seed, ZVALUE *res);
static int slotcp(BITSTR *bitstr, FULL *src, int count);
static void slotcp64(BITSTR *bitstr, FULL *src);
/*
* randreseed64 - scramble seed in 64 bit chunks
*
* given:
* a seed
*
* returns:
* a scrambled seed, or 0 if seed was 0
*
* It is 'nice' when a seed of "n" produces a 'significantly different'
* sequence than a seed of "n+1". Generators, by convention, assign
* special significance to the seed of '0'. It is an unfortunate that
* people often pick small seed values, particularly when large seed
* are of significance to the generators found in this file.
*
* We will process seed 64 bits at a time until the entire seed has been
* exhausted. If a 64 bit chunk is 0, then 0 is returned. If the 64 bit
* chunk is non-zero, we will produce a different and unique new scrambled
* chunk. In particular, if the seed is 0 we will return 0. If the seed
* is non-zero, we will return a different value (though chunks of 64
* zeros will remain zero). This scrambling will effectively eliminate
* the human perceptions that are noted above.
*
* It should be noted that the purpose of this process is to scramble a seed
* ONLY. We do not care if these generators produce good random numbers.
* We only want to help eliminate the human factors and perceptions
* noted above.
*
* This function scrambles all 64 bit chunks of a seed, by mapping [0,2^64)
* into [0,2^64). This map is one-to-one and onto. Mapping is performed
* using a linear congruence generator of the form:
*
* X1 <-- (a*X0 + c) % m
*
* with the exception that:
*
* 0 ==> 0 (so that srand(0) acts as default)
* randreseed64() is an 1-to-1 and onto map
*
* The generator are based on the linear congruential generators found in
* Knuth's "The Art of Computer Programming - Seminumerical Algorithms",
* vol 2, 2nd edition (1981), Section 3.6, pages 170-171.
*
* Because we process 64 bits we will take:
*
* m = 2^64 (based on note ii)
*
* We will scan the Rand Book of Random Numbers to look for an 'a' such that:
*
* a mod 8 == 5 (based on note iii)
* 0.01*m < a < 0.99*m (based on note iv)
* 0.01*2^64 < a < 0.99*2^64
*
* To help keep the generators independent, we want:
*
* a is prime
*
* The choice of an adder 'c' is considered immaterial according (based
* in note v). Knuth suggests 'c==1' or 'c==a'. We elect to select 'c'
* using the same process as we used to select 'a'. The choice is
* 'immaterial' after all, and as long as:
*
* gcd(c, m) == 1 (based on note v)
* gcd(c, 2^64) == 1
*
* the concerns are met. It can be shown that if we have:
*
* gcd(a, c) == 1
*
* then the adders and multipliers will be more independent.
*
* We will obtain the values 'a' and 'c for our generator from the
* Rand Book of Random Numbers. Because m=2^64 is 20 decimal digits long,
* we will search the Rand Book of Random Numbers 20 at a time. We will
* skip any of the 100 values that were used to initialize the subtractive 100
* generators. The values obtained from the Rand Book of Random Numbers are:
*
* a = 6316878969928993981
* c = 1363042948800878693
*
* As we stated before, we must map 0 ==> 0. The linear congruence
* generator would normally map as follows:
*
* 0 ==> 1363042948800878693 (0 ==> c)
*
* We can determine which 0<=y<m will produce a given value z by inverting the
* linear congruence generator:
*
* z = (a*y + c) % m
*
* z = a*y % m + c % m
*
* z-c % m = a*y % m
*
* (z-c % m) * minv(a,m) = (a*y % m) * minv(a,m)
* [minv(a,m) is the multiplicative inverse of a % m]
*
* ((z-c % m) * minv(a,m)) % m = ((a*y % m) * minv(a,m)) % m
*
* ((z-c % m) * minv(a,m)) % m = y % m
* [a % m * minv(a,m) = 1 % m by definition]
*
* ((z-c) * minv(a,m)) % m = y % m
*
* ((z-c) * minv(a,m)) % m = y
* [we defined y to be 0<=y<m]
*
* To determine which value maps back into 0, we let z = 0 and compute:
*
* ((0-c) * minv(a,m)) % m ==> 10239951819489363767
*
* and thus we find that the congruence generator would also normally map:
*
* 10239951819489363767 ==> 0
*
* To overcome this, and preserve the 1-to-1 and onto map, we force:
*
* 0 ==> 0
* 10239951819489363767 ==> 1363042948800878693
*
* To repeat, this function converts a values into a seed value. With the
* except of 'seed == 0', every value is mapped into a unique seed value.
* This mapping need not be complex, random or secure. All we attempt
* to do here is to allow humans who pick small or successive seed values
* to obtain reasonably different sequences from the generators below.
*
* NOTE: This is NOT a pseudo random number generator. This function is
* intended to be used internally by ss100rand() and sshufrand().
*/
static void
randreseed64(ZVALUE seed, ZVALUE *res)
{
ZVALUE t; /* temp value */
ZVALUE chunk; /* processed 64 bit chunk value */
ZVALUE seed64; /* seed mod 2^64 */
HALF *v64; /* 64 bit array of HALFs */
long chunknum; /* 64 bit chunk number */
/*
* quickly return 0 if seed is 0
*/
if (ziszero(seed) || seed.len <= 0) {
itoz(0, res);
return;
}
/*
* allocate result
*/
seed.sign = 0; /* use the value of seed */
res->len = (int)(((seed.len+SHALFS-1) / SHALFS) * SHALFS);
res->v = alloc(res->len);
res->sign = 0;
memset(res->v, 0, res->len*sizeof(HALF)); /* default value is 0 */
/*
* process 64 bit chunks until done
*/
chunknum = 0;
while (!zislezero(seed)) {
/*
* grab the lower 64 bits of seed
*/
if (zge64b(seed)) {
v64 = alloc(SHALFS);
memcpy(v64, seed.v, SHALFS*sizeof(HALF));
seed64.v = v64;
seed64.len = SHALFS;
seed64.sign = 0;
} else {
zcopy(seed, &seed64);
}
zshiftr(seed, SBITS);
ztrim(&seed);
ztrim(&seed64);
/*
* do nothing if chunk is zero
*/
if (ziszero(seed64)) {
++chunknum;
zfree(seed64);
continue;
}
/*
* Compute the linear congruence generator map:
*
* X1 <-- (a*X0 + c) mod m
*
* in other words:
*
* chunk == (a_val*seed + c_val) mod 2^64
*/
zmul(seed64, a_val, &t);
zfree(seed64);
zadd(t, c_val, &chunk);
zfree(t);
/*
* form chunk mod 2^64
*/
if (chunk.len > SHALFS) {
/* result is too large, reduce to 64 bits */
v64 = alloc(SHALFS);
memcpy(v64, chunk.v, SHALFS*sizeof(HALF));
free(chunk.v);
chunk.v = v64;
chunk.len = SHALFS;
ztrim(&chunk);
}
/*
* Normally, the above equation would map:
*
* f(0) == 1363042948800878693
* f(10239951819489363767) == 0
*
* However, we have already forced f(0) == 0. To preserve the
* 1-to-1 and onto map property, we force:
*
* f(10239951819489363767) ==> 1363042948800878693
*/
if (ziszero(chunk)) {
/* load 1363042948800878693 instead of 0 */
zcopy(c_val, &chunk);
memcpy(res->v+(chunknum*SHALFS), c_val.v,
c_val.len*sizeof(HALF));
/*
* load the 64 bit chunk into the result
*/
} else {
memcpy(res->v+(chunknum*SHALFS), chunk.v,
chunk.len*sizeof(HALF));
}
++chunknum;
zfree(chunk);
}
ztrim(res);
}
/*
* zsrand - seed the s100 generator
*
* given:
* pseed - ptr to seed of the generator or NULL
* pmat100 - subtractive 100 state table or NULL
*
* returns:
* previous s100 state
*/
RAND *
zsrand(CONST ZVALUE *pseed, CONST MATRIX *pmat100)
{
RAND *ret; /* previous s100 state */
CONST VALUE *v; /* value from a passed matrix */
ZVALUE zscram; /* scrambled 64 bit seed */
ZVALUE ztmp; /* temp holding value for zscram */
ZVALUE seed; /* to hold *pseed */
FULL shufxor[SLEN]; /* zshufxor as an 64 bit array of FULLs */
long indx; /* index to shuffle slots for seeding */
int i;
/*
* firewall
*/
if (pseed != NULL && zisneg(*pseed)) {
math_error("neg seeds for srand reserved for future use");
/*NOTREACHED*/
}
/*
* initialize state if first call
*/
if (!s100.seeded) {
s100 = init_s100;
}
/*
* save the current state to return later
*/
ret = (RAND *)malloc(sizeof(RAND));
if (ret == NULL) {
math_error("cannot allocate RAND state");
/*NOTREACHED*/
}
*ret = s100;
/*
* if call was srand(), just return current state
*/
if (pseed == NULL && pmat100 == NULL) {
return ret;
}
/*
* if call is srand(0), initialize and return quickly
*/
if (pmat100 == NULL && ziszero(*pseed)) {
s100 = init_s100;
return ret;
}
/*
* clear buffered bits, initialize pointers
*/
s100.seeded = 0; /* not seeded now */
s100.j = INIT_J;
s100.k = INIT_K;
s100.bits = 0;
memset(s100.buffer, 0, sizeof(s100.buffer));
/*
* load subtractive table
*
* We will load the default subtractive table unless we are passed a
* matrix. If we are passed a matrix, we will load the first 100
* values mod 2^64 instead.
*/
if (pmat100 == NULL) {
memcpy(s100.slot, def_subtract, sizeof(def_subtract));
} else {
/*
* use the first 100 entries from the matrix arg
*/
if (pmat100->m_size < S100) {
math_error("matrix for srand has < 100 elements");
/*NOTREACHED*/
}
for (v=pmat100->m_table, i=0; i < S100; ++i, ++v) {
/* reject if not integer */
if (v->v_type != V_NUM || qisfrac(v->v_num)) {
math_error("matrix for srand must contain ints");
/*NOTREACHED*/
}
/* load table element from matrix element mod 2^64 */
SLOAD(s100, i, v->v_num->num);
}
}
/*
* scramble the seed in 64 bit chunks
*/
if (pseed != NULL) {
seed.sign = pseed->sign;
seed.len = pseed->len;
seed.v = alloc(seed.len);
zcopyval(*pseed, seed);
randreseed64(seed, &zscram);
zfree(seed);
}
/*
* xor subtractive table with the rehashed lower 64 bits of seed
*/
if (pseed != NULL && !ziszero(zscram)) {
/* xor subtractive table with lower 64 bits of seed */
SMOD64(shufxor, zscram);
for (i=0; i < S100; ++i) {
SXOR(s100, i, shufxor);
}
}
/*
* shuffle subtractive 100 table according to seed, if passed
*/
if (pseed != NULL && zge64b(zscram)) {
/* prepare the seed for subtractive slot shuffling */
zshiftr(zscram, 64);
ztrim(&zscram);
/* shuffle subtractive table */
for (i=S100-1; i > 0 && !zislezero(zscram); --i) {
/* determine what we will swap with */
indx = zdivi(zscram, i+1, &ztmp);
zfree(zscram);
zscram = ztmp;
/* do nothing if swap with itself */
if (indx == i) {
continue;
}
/* swap slot[i] with slot[indx] */
SSWAP(s100, i, indx);
}
zfree(zscram);
} else if (pseed != NULL) {
zfree(zscram);
}
/*
* load the shuffle table
*
* We will generate SHUFCNT entries from the subtractive 100 slots
* and fill the shuffle table in consecutive order.
*/
for (i=0; i < SHUFCNT; ++i) {
/*
* skip values if required
*
* See:
* Knuth's "The Art of Computer Programming -
* Seminumerical Algorithms", Vol 2, 3rd edition (1998),
* Section 3.6, page 188".
*/
if (s100.need_to_skip <= 0) {
int sk;
/* skip the require number of values */
for (sk=0; sk < RAND_SKIP; ++sk) {
/* bump j and k */
if (++s100.j >= S100) {
s100.j = 0;
}
if (++s100.k >= S100) {
s100.k = 0;
}
/* slot[k] -= slot[j] */
SSUB(s100, s100.k, s100.j);
/* NOTE: don't shuffle, no shuffle table yet */
}
/* reset the skip count */
s100.need_to_skip = RAND_CONSEQ_USE;
if (conf->calc_debug & CALCDBG_RAND) {
printf("rand: skipped %d states\n", RAND_SKIP);
}
/* no skip, just descrment use counter */
} else {
--s100.need_to_skip;
}
/* bump j and k */
if (++s100.j >= S100) {
s100.j = 0;
}
if (++s100.k >= S100) {
s100.k = 0;
}
/* slot[k] -= slot[j] */
SSUB(s100, s100.k, s100.j);
/* shuf[i] = slot[k] */
SSHUF(s100, i, s100.k);
}
/*
* note that we are seeded
*/
s100.seeded = 1;
/*
* return the previous state
*/
return ret;
}
/*
* zsetrand - set the s100 generator state
*
* given:
* state - the state to copy
*
* returns:
* previous s100 state
*/
RAND *
zsetrand(CONST RAND *state)
{
RAND *ret; /* previous s100 state */
/*
* initialize state if first call
*/
if (!s100.seeded) {
s100 = init_s100;
}
/*
* save the current state to return later
*/
ret = randcopy(&s100);
/*
* load the new state
*/
s100 = *state;
/*
* return the previous state
*/
return ret;
}
/*
* slotcp - copy up to 64 bits from a 64 bit array of FULLs to some HALFs
*
* We will copy data from an array of FULLs into an array of HALFs.
* The destination within the HALFs is some bit location found in bitstr.
* We will attempt to copy 64 bits, but if there is not enough room
* (bits not yet loaded) in the destination bit string we will transfer
* what we can.
*
* The src slot is 64 bits long and is stored as an array of FULLs.
* When FULL_BITS is 64 the element is 1 FULL, otherwise FULL_BITS
* is 32 bits and the element is 2 FULLs. The most significant bit
* in the array (highest bit in the last FULL of the array) is to
* be transfered to the most significant bit in the destination.
*
* given:
* bitstr - most significant destination bit in a bit string
* src - low order FULL in a 64 bit slot
* count - number of bits to transfer (must be 0 < count <= 64)
*
* returns:
* number of bits transfered
*/
static int
slotcp(BITSTR *bitstr, FULL *src, int count)
{
HALF *dh; /* most significant HALF in dest */
int dnxtbit; /* next bit beyond most signif in dh */
int need; /* number of bits we need to transfer */
int ret; /* bits transfered */
/*
* determine how many bits we actually need to transfer
*/
dh = bitstr->loc;
dnxtbit = bitstr->bit+1;
count &= (SBITS-1);
need = (bitstr->len < count) ? bitstr->len : count;
/*
* prepare for the return
*
* Note the final bitstr location after we have moved the
* position down 'need' bits.
*/
bitstr->len -= need;
bitstr->loc -= need / BASEB;
bitstr->bit -= need % BASEB;
if (bitstr->bit < 0) {
--bitstr->loc;
bitstr->bit += BASEB;
}
ret = need;
/*
* deal with aligned copies quickly
*/
if (dnxtbit == BASEB) {
if (need == SBITS) {
#if 2*FULL_BITS == SBITS
*dh-- = (HALF)(src[SLEN-1] >> BASEB);
*dh-- = (HALF)(src[SLEN-1]);
#endif
*dh-- = (HALF)(src[0] >> BASEB);
*dh = (HALF)(src[0]);
#if 2*FULL_BITS == SBITS
} else if (need > FULL_BITS+BASEB) {
*dh-- = (HALF)(src[SLEN-1] >> BASEB);
*dh-- = (HALF)(src[SLEN-1]);
*dh-- = (HALF)(src[0] >> BASEB);
*dh = ((HALF)src[0] &
highhalf[need-FULL_BITS-BASEB]);
} else if (need > FULL_BITS) {
*dh-- = (HALF)(src[SLEN-1] >> BASEB);
*dh-- = (HALF)(src[SLEN-1]);
*dh = ((HALF)(src[0] >> BASEB) &
highhalf[need-FULL_BITS]);
#endif
} else if (need > BASEB) {
*dh-- = (HALF)(src[SLEN-1] >> BASEB);
*dh = ((HALF)(src[SLEN-1]) & highhalf[need-BASEB]);
} else {
*dh = ((HALF)(src[SLEN-1] >> BASEB) & highhalf[need]);
}
return ret;
}
/*
* load the most significant HALF
*/
if (need >= dnxtbit) {
/* fill up the most significant HALF */
*dh-- |= (HALF)(src[SLEN-1] >> (FULL_BITS-dnxtbit));
need -= dnxtbit;
} else if (need > 0) {
/* we exhaust our need before 1st half is filled */
*dh |= (HALF)((src[SLEN-1] >> (FULL_BITS-need)) <<
(dnxtbit-need));
return ret; /* our need has been filled */
} else {
return ret; /* our need has been filled */
}
/*
* load the 2nd most significant HALF
*/
if (need > BASEB) {
/* fill up the 2nd most significant HALF */
*dh-- = (HALF)(src[SLEN-1] >> (BASEB-dnxtbit));
need -= BASEB;
} else if (need > 0) {
/* we exhaust our need before 2nd half is filled */
*dh |= ((HALF)(src[SLEN-1] >> (BASEB-dnxtbit)) &
highhalf[need]);
return ret; /* our need has been filled */
} else {
return ret; /* our need has been filled */
}
/*
* load the 3rd most significant HALF
*
* At this point we know that our 3rd HALF will force us
* to cross into a second FULL for systems with 32 bit FULLs.
* We know this because the aligned case has been previously
* taken care of above.
*
* For systems that have 64 bit FULLs (and 32 bit HALFs) this
* is will be our least significant HALF. We also know that
* the need must be < BASEB.
*/
#if FULL_BITS == SBITS
*dh |= (((HALF)src[0] & highhalf[dnxtbit+need]) << dnxtbit);
#else
if (need > BASEB) {
/* load the remaining bits from the most signif FULL */
*dh-- = ((((HALF)src[SLEN-1] & lowhalf[BASEB-dnxtbit])
<< dnxtbit) | (HALF)(src[0] >> (FULL_BITS-dnxtbit)));
need -= BASEB;
} else if (need > 0) {
/* load the remaining bits from the most signif FULL */
*dh-- |= (((((HALF)src[SLEN-1] & lowhalf[BASEB-dnxtbit])
<< dnxtbit) | (HALF)(src[0] >> (FULL_BITS-dnxtbit))) &
highhalf[need]);
return ret; /* our need has been filled */
} else {
return ret; /* our need has been filled */
}
/*
* load the 4th most significant HALF
*
* At this point, only 32 bit FULLs are operating.
*/
if (need > BASEB) {
/* fill up the 2nd most significant HALF */
*dh-- = (HALF)(src[0] >> (BASEB-dnxtbit));
/* no need todo: need -= BASEB, because we are nearly done */
} else if (need > 0) {
/* we exhaust our need before 2nd half is filled */
*dh |= ((HALF)(src[0] >> (BASEB-dnxtbit)) &
highhalf[need]);
return ret; /* our need has been filled */
} else {
return ret; /* our need has been filled */
}
/*
* load the 5th and least significant HALF
*
* At this point we know that the need will be satisfied.
*/
*dh |= (((HALF)src[0] & lowhalf[BASEB-dnxtbit]) << dnxtbit);
#endif
return ret; /* our need has been filled */
}
/*
* slotcp64 - copy 64 bits from a 64 bit array of FULLs to some HALFs
*
* We will copy data from an array of FULLs into an array of HALFs.
* The destination within the HALFs is some bit location found in bitstr.
* Unlike slotcp(), this function will always copy 64 bits.
*
* The src slot is 64 bits long and is stored as an array of FULLs.
* When FULL_BITS is 64 this array is 1 FULL, otherwise FULL_BITS
* is 32 bits and the array is 2 FULLs. The most significant bit
* in the array (highest bit in the last FULL of the array) is to
* be transfered to the most significant bit in the destination.
*
* given:
* bitstr - most significant destination bit in a bit string
* src - low order FULL in a 64 bit slot
*
* returns:
* number of bits transfered
*/
static void
slotcp64(BITSTR *bitstr, FULL *src)
{
HALF *dh = bitstr->loc; /* most significant HALF in dest */
int dnxtbit = bitstr->bit+1; /* next bit beyond most signif in dh */
/*
* prepare for the return
*
* Since we are moving the point 64 bits down, we know that
* the bit location (bitstr->bit) will remain the same.
*/
bitstr->len -= SBITS;
bitstr->loc -= SBITS/BASEB;
/*
* deal with aligned copies quickly
*/
if (dnxtbit == BASEB) {
#if 2*FULL_BITS == SBITS
*dh-- = (HALF)(src[SLEN-1] >> BASEB);
*dh-- = (HALF)(src[SLEN-1]);
#endif
*dh-- = (HALF)(src[0] >> BASEB);
*dh = (HALF)(src[0]);
return;
}
/*
* load the most significant HALF
*/
*dh-- |= (HALF)(src[SLEN-1] >> (FULL_BITS-dnxtbit));
/*
* load the 2nd most significant HALF
*/
*dh-- = (HALF)(src[SLEN-1] >> (BASEB-dnxtbit));
/*
* load the 3rd most significant HALF
*
* At this point we know that our 3rd HALF will force us
* to cross into a second FULL for systems with 32 bit FULLs.
* We know this because the aligned case has been previously
* taken care of above.
*
* For systems that have 64 bit FULLs (and 32 bit HALFs) this
* is will be our least significant HALF.
*/
#if FULL_BITS == SBITS
*dh |= (((HALF)src[0] & lowhalf[BASEB-dnxtbit]) << dnxtbit);
#else
/* load the remaining bits from the most signif FULL */
*dh-- = ((((HALF)src[SLEN-1] & lowhalf[BASEB-dnxtbit])
<< dnxtbit) | (HALF)(src[0] >> (FULL_BITS-dnxtbit)));
/*
* load the 4th most significant HALF
*
* At this point, only 32 bit FULLs are operating.
*/
*dh-- = (HALF)(src[0] >> (BASEB-dnxtbit));
/*
* load the 5th and least significant HALF
*
* At this point we know that the need will be satisfied.
*/
*dh |= (((HALF)src[0] & lowhalf[BASEB-dnxtbit]) << dnxtbit);
#endif
}
/*
* zrandskip - skip s s100 bits
*
* given:
* count - number of bits to be skipped
*/
void
zrandskip(long cnt)
{
int indx; /* shuffle entry index */
/*
* initialize state if first call
*/
if (!s100.seeded) {
s100 = init_s100;
}
/*
* skip required bits in the buffer
*/
if (s100.bits > 0 && s100.bits <= cnt) {
/* just toss the buffer bits */
cnt -= s100.bits;
s100.bits = 0;
memset(s100.buffer, 0, sizeof(s100.buffer));
} else if (s100.bits > 0 && s100.bits > cnt) {
/* buffer contains more bits than we need to toss */
#if FULL_BITS == SBITS
s100.buffer[0] <<= cnt;
#else
if (cnt >= FULL_BITS) {
s100.buffer[SLEN-1] = (s100.buffer[0] << cnt);
s100.buffer[0] = 0;
} else {
s100.buffer[SLEN-1] =
((s100.buffer[SLEN-1] << cnt) |
(s100.buffer[0] >> (FULL_BITS-cnt)));
s100.buffer[0] <<= cnt;
}
#endif
s100.bits -= cnt;
return; /* skip need satisfied */
}
/*
* skip 64 bits at a time
*/
while (cnt >= SBITS) {
/*
* skip values if required
*
* NOTE: This skip loop is part of the algorithm, not part
* of the builtin function request.
*
* See:
* Knuth's "The Art of Computer Programming -
* Seminumerical Algorithms", Vol 2, 3rd edition (1998),
* Section 3.6, page 188".
*/
if (s100.need_to_skip <= 0) {
int sk;
/* skip the require number of values */
for (sk=0; sk < RAND_SKIP; ++sk) {
/* bump j and k */
if (++s100.j >= S100) {
s100.j = 0;
}
if (++s100.k >= S100) {
s100.k = 0;
}
/* slot[k] -= slot[j] */
SSUB(s100, s100.k, s100.j);
/* store s100.k into s100.slot[indx] */
indx = SINDX(s100, s100.k);
SSHUF(s100, indx, s100.k);
}
/* reset the skip count */
s100.need_to_skip = RAND_CONSEQ_USE;
if (conf->calc_debug & CALCDBG_RAND) {
printf("rand: skipped %d states\n", RAND_SKIP);
}
/* no skip, just descrment use counter */
} else {
--s100.need_to_skip;
}
/* bump j and k */
if (++s100.j >= S100) {
s100.j = 0;
}
if (++s100.k >= S100) {
s100.k = 0;
}
/* slot[k] -= slot[j] */
SSUB(s100, s100.k, s100.j);
/* we will ignore the output value of s100.slot[indx] */
indx = SINDX(s100, s100.k);
cnt -= SBITS;
/* store s100.k into s100.slot[indx] */
SSHUF(s100, indx, s100.k);
}
/*
* skip the final bits
*/
if (cnt > 0) {
/*
* skip values if required
*
* NOTE: This skip loop is part of the algorithm, not part
* of the builtin function request.
*
* See:
* Knuth's "The Art of Computer Programming -
* Seminumerical Algorithms", Vol 2, 3rd edition (1998),
* Section 3.6, page 188".
*/
if (s100.need_to_skip <= 0) {
int sk;
/* skip the require number of values */
for (sk=0; sk < RAND_SKIP; ++sk) {
/* bump j and k */
if (++s100.j >= S100) {
s100.j = 0;
}
if (++s100.k >= S100) {
s100.k = 0;
}
/* slot[k] -= slot[j] */
SSUB(s100, s100.k, s100.j);
/* store s100.k into s100.slot[indx] */
indx = SINDX(s100, s100.k);
SSHUF(s100, indx, s100.k);
}
/* reset the skip count */
s100.need_to_skip = RAND_CONSEQ_USE;
if (conf->calc_debug & CALCDBG_RAND) {
printf("rand: skipped %d states\n", RAND_SKIP);
}
/* no skip, just descrment use counter */
} else {
--s100.need_to_skip;
}
/* bump j and k */
if (++s100.j >= S100) {
s100.j = 0;
}
if (++s100.k >= S100) {
s100.k = 0;
}
/* slot[k] -= slot[j] */
SSUB(s100, s100.k, s100.j);
/* we will ignore the output value of s100.slot[indx] */
indx = SINDX(s100, s100.k);
/*
* We know the buffer is empty, so fill it
* with any unused bits. Copy SBITS-trans bits
* from slot[indx] into buffer.
*/
s100.bits = (int)(SBITS-cnt);
memcpy(s100.buffer, &s100.shuf[indx*SLEN],
sizeof(s100.buffer));
/*
* shift the buffer bits all the way up to
* the most significant bit
*/
#if FULL_BITS == SBITS
s100.buffer[0] <<= cnt;
#else
if (cnt >= FULL_BITS) {
s100.buffer[SLEN-1] = (s100.buffer[0] << cnt);
s100.buffer[0] = 0;
} else {
s100.buffer[SLEN-1] =
((s100.buffer[SLEN-1] << cnt) |
(s100.buffer[0] >> (FULL_BITS-cnt)));
s100.buffer[0] <<= cnt;
}
#endif
/* store s100.k into s100.slot[indx] */
SSHUF(s100, indx, s100.k);
}
}
/*
* zrand - crank the s100 generator for some bits
*
* We will load the ZVALUE with random bits starting from the
* most significant and ending with the lowest bit in the
* least significant HALF.
*
* given:
* count - number of bits required
* res - where to place the random bits as ZVALUE
*/
void
zrand(long cnt, ZVALUE *res)
{
BITSTR dest; /* destination bit string */
int trans; /* bits transfered */
int indx; /* shuffle entry index */
/*
* firewall
*/
if (cnt <= 0) {
if (cnt == 0) {
/* zero length random number is always 0 */
itoz(0, res);
return;
} else {
math_error("negative zrand bit count");
/*NOTREACHED*/
}
#if LONG_BITS > 32
} else if (cnt > (1L<<31)) {
math_error("huge rand bit count in internal zrand function");
/*NOTREACHED*/
#endif
}
/*
* initialize state if first call
*/
if (!s100.seeded) {
s100 = init_s100;
}
/*
* allocate storage
*/
res->len = (LEN)((cnt+BASEB-1)/BASEB);
res->v = alloc((LEN)((cnt+BASEB-1)/BASEB));
/*
* dest bit string
*/
dest.len = (int)cnt;
dest.loc = res->v + (((cnt+BASEB-1)/BASEB)-1);
dest.bit = (int)((cnt-1) % BASEB);
memset(res->v, 0, (LEN)((cnt+BASEB-1)/BASEB)*sizeof(HALF));
/*
* load from buffer first
*/
if (s100.bits > 0) {
/*
* We know there are only s100.bits in the buffer, so
* transfer as much as we can (treating it as a slot)
* and return the bit transfer count.
*/
trans = slotcp(&dest, s100.buffer, s100.bits);
/*
* If we need to keep bits in the buffer,
* shift the buffer bits all the way up to
* the most significant unused bit.
*/
if (trans < s100.bits) {
#if FULL_BITS == SBITS
s100.buffer[0] <<= trans;
#else
if (trans >= FULL_BITS) {
s100.buffer[SLEN-1] =
(s100.buffer[0] << (trans-FULL_BITS));
s100.buffer[0] = 0;
} else {
s100.buffer[SLEN-1] =
((s100.buffer[SLEN-1] << trans) |
(s100.buffer[0] >> (FULL_BITS-trans)));
s100.buffer[0] <<= trans;
}
#endif
}
/* note that we have fewer bits in the buffer */
s100.bits -= trans;
}
/*
* spin the generator until we need less than 64 bits
*
* The buffer did not contain enough bits, so we crank the
* s100 generator and load then 64 bits at a time.
*/
while (dest.len >= SBITS) {
/*
* skip values if required
*
* See:
* Knuth's "The Art of Computer Programming -
* Seminumerical Algorithms", Vol 2, 3rd edition (1998),
* Section 3.6, page 188".
*/
if (s100.need_to_skip <= 0) {
int sk;
/* skip the require number of values */
for (sk=0; sk < RAND_SKIP; ++sk) {
/* bump j and k */
if (++s100.j >= S100) {
s100.j = 0;
}
if (++s100.k >= S100) {
s100.k = 0;
}
/* slot[k] -= slot[j] */
SSUB(s100, s100.k, s100.j);
/* select slot index to output */
indx = SINDX(s100, s100.k);
/* store s100.k into s100.slot[indx] */
SSHUF(s100, indx, s100.k);
}
/* reset the skip count */
s100.need_to_skip = RAND_CONSEQ_USE;
if (conf->calc_debug & CALCDBG_RAND) {
printf("rand: skipped %d states\n", RAND_SKIP);
}
/* no skip, just descrment use counter */
} else {
--s100.need_to_skip;
}
/* bump j and k */
if (++s100.j >= S100) {
s100.j = 0;
}
if (++s100.k >= S100) {
s100.k = 0;
}
/* slot[k] -= slot[j] */
SSUB(s100, s100.k, s100.j);
/* select slot index to output */
indx = SINDX(s100, s100.k);
/* move up to 64 bits from slot[indx] to dest */
slotcp64(&dest, &s100.shuf[indx*SLEN]);
/* store s100.k into s100.slot[indx] */
SSHUF(s100, indx, s100.k);
}
/*
* spin the generator one last time to fill out the remaining need
*/
if (dest.len > 0) {
/*
* skip values if required
*
* See:
* Knuth's "The Art of Computer Programming -
* Seminumerical Algorithms", Vol 2, 3rd edition (1998),
* Section 3.6, page 188".
*/
if (s100.need_to_skip <= 0) {
int sk;
/* skip the require number of values */
for (sk=0; sk < RAND_SKIP; ++sk) {
/* bump j and k */
if (++s100.j >= S100) {
s100.j = 0;
}
if (++s100.k >= S100) {
s100.k = 0;
}
/* slot[k] -= slot[j] */
SSUB(s100, s100.k, s100.j);
/* select slot index to output */
indx = SINDX(s100, s100.k);
/* store s100.k into s100.slot[indx] */
SSHUF(s100, indx, s100.k);
}
/* reset the skip count */
s100.need_to_skip = RAND_CONSEQ_USE;
if (conf->calc_debug & CALCDBG_RAND) {
printf("rand: skipped %d states\n", RAND_SKIP);
}
/* no skip, just descrment use counter */
} else {
--s100.need_to_skip;
}
/* bump j and k */
if (++s100.j >= S100) {
s100.j = 0;
}
if (++s100.k >= S100) {
s100.k = 0;
}
/* slot[k] -= slot[j] */
SSUB(s100, s100.k, s100.j);
/* select slot index to output */
indx = SINDX(s100, s100.k);
/* move up to 64 bits from slot[indx] to dest */
trans = slotcp(&dest, &s100.shuf[indx*SLEN], dest.len);
/* buffer up unused bits if we are done */
if (trans != SBITS) {
/*
* We know the buffer is empty, so fill it
* with any unused bits. Copy SBITS-trans bits
* from slot[indx] into buffer.
*/
s100.bits = SBITS-trans;
memcpy(s100.buffer, &s100.shuf[indx*SLEN],
sizeof(s100.buffer));
/*
* shift the buffer bits all the way up to
* the most significant bit
*/
#if FULL_BITS == SBITS
s100.buffer[0] <<= trans;
#else
if (trans >= FULL_BITS) {
s100.buffer[SLEN-1] =
(s100.buffer[0] << (trans-FULL_BITS));
s100.buffer[0] = 0;
} else {
s100.buffer[SLEN-1] =
((s100.buffer[SLEN-1] << trans) |
(s100.buffer[0] >> (FULL_BITS-trans)));
s100.buffer[0] <<= trans;
}
#endif
}
/* store s100.k into s100.slot[indx] */
SSHUF(s100, indx, s100.k);
}
res->sign = 0;
ztrim(res);
}
/*
* zrandrange - generate an s100 random value in the range [low, high)
*
* given:
* low - low value of range
* high - beyond end of range
* res - where to place the random bits as ZVALUE
*/
void
zrandrange(CONST ZVALUE low, CONST ZVALUE high, ZVALUE *res)
{
ZVALUE range; /* high-low */
ZVALUE rval; /* random value [0, 2^bitlen) */
ZVALUE rangem1; /* range - 1 */
long bitlen; /* smallest power of 2 >= diff */
/*
* firewall
*/
if (zrel(low, high) >= 0) {
math_error("srand low range >= high range");
/*NOTREACHED*/
}
/*
* determine the size of the random number needed
*/
zsub(high, low, &range);
if (zisone(range)) {
zfree(range);
*res = low;
return;
}
zsub(range, _one_, &rangem1);
bitlen = 1+zhighbit(rangem1);
zfree(rangem1);
/*
* generate a random value between [0, diff)
*
* We will not fall into the trap of thinking that we can simply take
* a value mod 'range'. Consider the case where 'range' is '80'
* and we are given pseudo-random numbers [0,100). If we took them
* mod 80, then the numbers [0,20) would be produced more frequently
* because the numbers [81,100) mod 80 wrap back into [0,20).
*/
rval.v = NULL;
do {
if (rval.v != NULL) {
zfree(rval);
}
zrand(bitlen, &rval);
} while (zrel(rval, range) >= 0);
/*
* add in low value to produce the range [0+low, diff+low)
* which is the range [low, high)
*/
zadd(rval, low, res);
zfree(rval);
zfree(range);
}
/*
* irand - generate an s100 random long in the range [0, s)
*
* given:
* s - limit of the range
*
* returns:
* random long in the range [0, s)
*/
long
irand(long s)
{
ZVALUE z1, z2;
long res;
if (s <= 0) {
math_error("Non-positive argument for irand()");
/*NOTREACHED*/
}
if (s == 1)
return 0;
itoz(s, &z1);
zrandrange(_zero_, z1, &z2);
res = ztoi(z2);
zfree(z1);
zfree(z2);
return res;
}
/*
* randcopy - make a copy of an s100 state
*
* given:
* state - the state to copy
*
* returns:
* a malloced copy of the state
*/
RAND *
randcopy(CONST RAND *state)
{
RAND *ret; /* return copy of state */
/*
* malloc state
*/
ret = (RAND *)malloc(sizeof(RAND));
if (ret == NULL) {
math_error("can't allocate RAND state");
/*NOTREACHED*/
}
memcpy(ret, state, sizeof(RAND));
/*
* return copy
*/
return ret;
}
/*
* randfree - free an s100 state
*
* given:
* state - the state to free
*/
void
randfree(RAND *state)
{
/* free it */
free(state);
}
/*
* randcmp - compare two s100 states
*
* given:
* s1 - first state to compare
* s2 - second state to compare
*
* return:
* TRUE if states differ
*/
BOOL
randcmp(CONST RAND *s1, CONST RAND *s2)
{
/*
* assume uninitialized state == the default seeded state
*/
if (!s1->seeded) {
if (!s2->seeded) {
/* uninitialized == uninitialized */
return FALSE;
} else {
/* uninitialized only equals default state */
return randcmp(s2, &init_s100);
}
} else if (!s2->seeded) {
/* uninitialized only equals default state */
return randcmp(s1, &init_s100);
}
/* compare states */
return (BOOL)(memcmp(s1, s2, sizeof(RAND)) != 0);
}
/*
* randprint - print an s100 state
*
* given:
* state - state to print
* flags - print flags passed from printvalue() in value.c
*/
/*ARGSUSED*/
void
randprint(CONST RAND *state, int flags)
{
math_str("RAND state");
}
Reference in New Issue View Git Blame Copy Permalink