11176bdada62cabc6ec4b0308a930e83b679d5d36John Reck/*
21176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * Copyright © 2012 Siarhei Siamashka <siarhei.siamashka@gmail.com>
31176bdada62cabc6ec4b0308a930e83b679d5d36John Reck *
41176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * Based on the public domain implementation of small noncryptographic PRNG
51176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * authored by Bob Jenkins: http://burtleburtle.net/bob/rand/smallprng.html
61176bdada62cabc6ec4b0308a930e83b679d5d36John Reck *
71176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * Permission is hereby granted, free of charge, to any person obtaining a
81176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * copy of this software and associated documentation files (the "Software"),
91176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * to deal in the Software without restriction, including without limitation
101176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * the rights to use, copy, modify, merge, publish, distribute, sublicense,
111176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * and/or sell copies of the Software, and to permit persons to whom the
121176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * Software is furnished to do so, subject to the following conditions:
131176bdada62cabc6ec4b0308a930e83b679d5d36John Reck *
141176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * The above copyright notice and this permission notice (including the next
151176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * paragraph) shall be included in all copies or substantial portions of the
161176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * Software.
171176bdada62cabc6ec4b0308a930e83b679d5d36John Reck *
181176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
191176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
201176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
211176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
221176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
231176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
241176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * DEALINGS IN THE SOFTWARE.
251176bdada62cabc6ec4b0308a930e83b679d5d36John Reck */
261176bdada62cabc6ec4b0308a930e83b679d5d36John Reck
271176bdada62cabc6ec4b0308a930e83b679d5d36John Reck#ifndef __UTILS_PRNG_H__
281176bdada62cabc6ec4b0308a930e83b679d5d36John Reck#define __UTILS_PRNG_H__
291176bdada62cabc6ec4b0308a930e83b679d5d36John Reck
301176bdada62cabc6ec4b0308a930e83b679d5d36John Reck/*
311176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * This file provides a fast SIMD-optimized noncryptographic PRNG (pseudorandom
321176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * number generator), with the output good enough to pass "Big Crush" tests
331176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * from TestU01 (http://en.wikipedia.org/wiki/TestU01).
341176bdada62cabc6ec4b0308a930e83b679d5d36John Reck *
351176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * SIMD code uses http://gcc.gnu.org/onlinedocs/gcc/Vector-Extensions.html
361176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * which is a GCC specific extension. There is also a slower alternative
371176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * code path, which should work with any C compiler.
381176bdada62cabc6ec4b0308a930e83b679d5d36John Reck *
391176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * The "prng_t" structure keeps the internal state of the random number
401176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * generator. It is possible to have multiple instances of the random number
411176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * generator active at the same time, in this case each of them needs to have
421176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * its own "prng_t". All the functions take a pointer to "prng_t"
431176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * as the first argument.
441176bdada62cabc6ec4b0308a930e83b679d5d36John Reck *
451176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * Functions:
461176bdada62cabc6ec4b0308a930e83b679d5d36John Reck *
471176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * ----------------------------------------------------------------------------
481176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * void prng_srand_r (prng_t *prng, uint32_t seed);
491176bdada62cabc6ec4b0308a930e83b679d5d36John Reck *
501176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * Initialize the pseudorandom number generator. The sequence of preudorandom
511176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * numbers is deterministic and only depends on "seed". Any two generators
521176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * initialized with the same seed will produce exactly the same sequence.
531176bdada62cabc6ec4b0308a930e83b679d5d36John Reck *
541176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * ----------------------------------------------------------------------------
551176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * uint32_t prng_rand_r (prng_t *prng);
561176bdada62cabc6ec4b0308a930e83b679d5d36John Reck *
571176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * Generate a single uniformly distributed 32-bit pseudorandom value.
581176bdada62cabc6ec4b0308a930e83b679d5d36John Reck *
591176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * ----------------------------------------------------------------------------
601176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * void prng_randmemset_r (prng_t                  *prng,
611176bdada62cabc6ec4b0308a930e83b679d5d36John Reck *                         void                    *buffer,
621176bdada62cabc6ec4b0308a930e83b679d5d36John Reck *                         size_t                   size,
631176bdada62cabc6ec4b0308a930e83b679d5d36John Reck *                         prng_randmemset_flags_t  flags);
641176bdada62cabc6ec4b0308a930e83b679d5d36John Reck *
651176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * Fills the memory buffer "buffer" with "size" bytes of pseudorandom data.
661176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * The "flags" argument may be used to tweak some statistics properties:
671176bdada62cabc6ec4b0308a930e83b679d5d36John Reck *    RANDMEMSET_MORE_00 - set ~25% of bytes to 0x00
681176bdada62cabc6ec4b0308a930e83b679d5d36John Reck *    RANDMEMSET_MORE_FF - set ~25% of bytes to 0xFF
691176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * The flags can be combined. This allows a bit better simulation of typical
701176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * pixel data, which normally contains a lot of fully transparent or fully
711176bdada62cabc6ec4b0308a930e83b679d5d36John Reck * opaque pixels.
721176bdada62cabc6ec4b0308a930e83b679d5d36John Reck */
731176bdada62cabc6ec4b0308a930e83b679d5d36John Reck
741176bdada62cabc6ec4b0308a930e83b679d5d36John Reck#ifdef HAVE_CONFIG_H
751176bdada62cabc6ec4b0308a930e83b679d5d36John Reck#include <config.h>
761176bdada62cabc6ec4b0308a930e83b679d5d36John Reck#endif
771176bdada62cabc6ec4b0308a930e83b679d5d36John Reck
781176bdada62cabc6ec4b0308a930e83b679d5d36John Reck#include "pixman-private.h"
791176bdada62cabc6ec4b0308a930e83b679d5d36John Reck
801176bdada62cabc6ec4b0308a930e83b679d5d36John Reck/*****************************************************************************/
811176bdada62cabc6ec4b0308a930e83b679d5d36John Reck
821176bdada62cabc6ec4b0308a930e83b679d5d36John Reck#if defined(__GNUC__) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 7))
831176bdada62cabc6ec4b0308a930e83b679d5d36John Reck#define GCC_VECTOR_EXTENSIONS_SUPPORTED
841176bdada62cabc6ec4b0308a930e83b679d5d36John Recktypedef uint32_t uint32x4 __attribute__ ((vector_size(16)));
851176bdada62cabc6ec4b0308a930e83b679d5d36John Recktypedef uint8_t  uint8x16 __attribute__ ((vector_size(16)));
861176bdada62cabc6ec4b0308a930e83b679d5d36John Reck#endif
871176bdada62cabc6ec4b0308a930e83b679d5d36John Reck
881176bdada62cabc6ec4b0308a930e83b679d5d36John Recktypedef struct
891176bdada62cabc6ec4b0308a930e83b679d5d36John Reck{
901176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    uint32_t a, b, c, d;
911176bdada62cabc6ec4b0308a930e83b679d5d36John Reck} smallprng_t;
921176bdada62cabc6ec4b0308a930e83b679d5d36John Reck
931176bdada62cabc6ec4b0308a930e83b679d5d36John Recktypedef struct
941176bdada62cabc6ec4b0308a930e83b679d5d36John Reck{
951176bdada62cabc6ec4b0308a930e83b679d5d36John Reck#ifdef GCC_VECTOR_EXTENSIONS_SUPPORTED
961176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    uint32x4 a, b, c, d;
971176bdada62cabc6ec4b0308a930e83b679d5d36John Reck#else
981176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    smallprng_t p1, p2, p3, p4;
991176bdada62cabc6ec4b0308a930e83b679d5d36John Reck#endif
1001176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    smallprng_t p0;
1011176bdada62cabc6ec4b0308a930e83b679d5d36John Reck} prng_t;
1021176bdada62cabc6ec4b0308a930e83b679d5d36John Reck
1031176bdada62cabc6ec4b0308a930e83b679d5d36John Recktypedef union
1041176bdada62cabc6ec4b0308a930e83b679d5d36John Reck{
1051176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    uint8_t  b[16];
1061176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    uint32_t w[4];
1071176bdada62cabc6ec4b0308a930e83b679d5d36John Reck#ifdef GCC_VECTOR_EXTENSIONS_SUPPORTED
1081176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    uint8x16 vb;
1091176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    uint32x4 vw;
1101176bdada62cabc6ec4b0308a930e83b679d5d36John Reck#endif
1111176bdada62cabc6ec4b0308a930e83b679d5d36John Reck} prng_rand_128_data_t;
1121176bdada62cabc6ec4b0308a930e83b679d5d36John Reck
1131176bdada62cabc6ec4b0308a930e83b679d5d36John Reck/*****************************************************************************/
1141176bdada62cabc6ec4b0308a930e83b679d5d36John Reck
1151176bdada62cabc6ec4b0308a930e83b679d5d36John Reckstatic force_inline uint32_t
1161176bdada62cabc6ec4b0308a930e83b679d5d36John Recksmallprng_rand_r (smallprng_t *x)
1171176bdada62cabc6ec4b0308a930e83b679d5d36John Reck{
1181176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    uint32_t e = x->a - ((x->b << 27) + (x->b >> (32 - 27)));
1191176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    x->a = x->b ^ ((x->c << 17) ^ (x->c >> (32 - 17)));
1201176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    x->b = x->c + x->d;
1211176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    x->c = x->d + e;
1221176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    x->d = e + x->a;
1231176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    return x->d;
1241176bdada62cabc6ec4b0308a930e83b679d5d36John Reck}
1251176bdada62cabc6ec4b0308a930e83b679d5d36John Reck
1261176bdada62cabc6ec4b0308a930e83b679d5d36John Reck/* Generate 4 bytes (32-bits) of random data */
1271176bdada62cabc6ec4b0308a930e83b679d5d36John Reckstatic force_inline uint32_t
1281176bdada62cabc6ec4b0308a930e83b679d5d36John Reckprng_rand_r (prng_t *x)
1291176bdada62cabc6ec4b0308a930e83b679d5d36John Reck{
1301176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    return smallprng_rand_r (&x->p0);
1311176bdada62cabc6ec4b0308a930e83b679d5d36John Reck}
1321176bdada62cabc6ec4b0308a930e83b679d5d36John Reck
1331176bdada62cabc6ec4b0308a930e83b679d5d36John Reck/* Generate 16 bytes (128-bits) of random data */
1341176bdada62cabc6ec4b0308a930e83b679d5d36John Reckstatic force_inline void
1351176bdada62cabc6ec4b0308a930e83b679d5d36John Reckprng_rand_128_r (prng_t *x, prng_rand_128_data_t *data)
1361176bdada62cabc6ec4b0308a930e83b679d5d36John Reck{
1371176bdada62cabc6ec4b0308a930e83b679d5d36John Reck#ifdef GCC_VECTOR_EXTENSIONS_SUPPORTED
1381176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    uint32x4 e = x->a - ((x->b << 27) + (x->b >> (32 - 27)));
1391176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    x->a = x->b ^ ((x->c << 17) ^ (x->c >> (32 - 17)));
1401176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    x->b = x->c + x->d;
1411176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    x->c = x->d + e;
1421176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    x->d = e + x->a;
1431176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    data->vw = x->d;
1441176bdada62cabc6ec4b0308a930e83b679d5d36John Reck#else
1451176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    data->w[0] = smallprng_rand_r (&x->p1);
1461176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    data->w[1] = smallprng_rand_r (&x->p2);
1471176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    data->w[2] = smallprng_rand_r (&x->p3);
1481176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    data->w[3] = smallprng_rand_r (&x->p4);
1491176bdada62cabc6ec4b0308a930e83b679d5d36John Reck#endif
1501176bdada62cabc6ec4b0308a930e83b679d5d36John Reck}
1511176bdada62cabc6ec4b0308a930e83b679d5d36John Reck
1521176bdada62cabc6ec4b0308a930e83b679d5d36John Recktypedef enum
1531176bdada62cabc6ec4b0308a930e83b679d5d36John Reck{
1541176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    RANDMEMSET_MORE_00        = 1, /* ~25% chance for 0x00 bytes */
1551176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    RANDMEMSET_MORE_FF        = 2, /* ~25% chance for 0xFF bytes */
1561176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    RANDMEMSET_MORE_00000000  = 4, /* ~25% chance for 0x00000000 clusters */
1571176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    RANDMEMSET_MORE_FFFFFFFF  = 8, /* ~25% chance for 0xFFFFFFFF clusters */
1581176bdada62cabc6ec4b0308a930e83b679d5d36John Reck    RANDMEMSET_MORE_00_AND_FF = (RANDMEMSET_MORE_00 | RANDMEMSET_MORE_00000000 |
1591176bdada62cabc6ec4b0308a930e83b679d5d36John Reck                                 RANDMEMSET_MORE_FF | RANDMEMSET_MORE_FFFFFFFF)
1601176bdada62cabc6ec4b0308a930e83b679d5d36John Reck} prng_randmemset_flags_t;
1611176bdada62cabc6ec4b0308a930e83b679d5d36John Reck
1621176bdada62cabc6ec4b0308a930e83b679d5d36John Reck/* Set the 32-bit seed for PRNG */
1631176bdada62cabc6ec4b0308a930e83b679d5d36John Reckvoid prng_srand_r (prng_t *prng, uint32_t seed);
1641176bdada62cabc6ec4b0308a930e83b679d5d36John Reck
1651176bdada62cabc6ec4b0308a930e83b679d5d36John Reck/* Fill memory buffer with random data */
1661176bdada62cabc6ec4b0308a930e83b679d5d36John Reckvoid prng_randmemset_r (prng_t                  *prng,
1671176bdada62cabc6ec4b0308a930e83b679d5d36John Reck                        void                    *buffer,
1681176bdada62cabc6ec4b0308a930e83b679d5d36John Reck                        size_t                   size,
1691176bdada62cabc6ec4b0308a930e83b679d5d36John Reck                        prng_randmemset_flags_t  flags);
1701176bdada62cabc6ec4b0308a930e83b679d5d36John Reck
1711176bdada62cabc6ec4b0308a930e83b679d5d36John Reck#endif
172