diff options
28 files changed, 2545 insertions, 417 deletions
diff --git a/dep/PackageList.txt b/dep/PackageList.txt index 3d72a9653de..97cd1287c08 100644 --- a/dep/PackageList.txt +++ b/dep/PackageList.txt @@ -37,8 +37,8 @@ OpenSSL (general-purpose cryptography library) Version: external SFMT (SIMD-oriented Fast Mersenne Twister) - Based on http://agner.org/random/ - Version: 2010-Aug-03 + https://github.com/MersenneTwister-Lab/SFMT + Version: 73bcba2e483640b8d50c1275514326dd4c1b0ca4 utf8-cpp (UTF-8 with C++ in a Portable Way) http://utfcpp.sourceforge.net/ diff --git a/dep/SFMT/CHANGE-LOG.txt b/dep/SFMT/CHANGE-LOG.txt new file mode 100644 index 00000000000..69d67f390dd --- /dev/null +++ b/dep/SFMT/CHANGE-LOG.txt @@ -0,0 +1,69 @@ +ver 1.5 +------- +bug fix to_res53 reported by Akio Takahashi. + +ver 1.4.1 +------- +fix MSC compile error reported by Dirk Steenpass. + +ver 1.4 +------- +internal state of SFMT is kept in structure. +function and macro names are changed to have sfmt_ prefix. +ALTIVEC OR BIGENDIAN ARE NOT TESTED AT ALL. + +ver 1.3.3 +------- +change condition compile of do_recursion in SFMT.c + +ver 1.3.2 +------- +bug fix to_res53_mix and genrand_res53_mix. + +ver 1.3.1 +------- +gcc compile option changed from -O9 to -O3. +add functions genrand_res53_mix and to_res53_mix. +bug fix about definition of ALWAYS_INLINE. +add new definition PRE_ALWAYS for MSC. + +ver 1.3 +------- +bug fixed: -DONLY64 without -DBIG_ENIAN64 had been generating +wrong sequence. +bug fixed: There is no documentation about BIG_ENDIAN64. +add automatic endian check by __BIG_ENDIAN__ predefined macro. +bug fixed: change == in check.sh to = +add SFMT-params216091.h +add AltiVec parameter format for systems which are not osx. +change Makefile for systems which are not osx and support AltiVec. +change sample2 of howto-compile for Free BSD. +change source files for BORLANDC and Visual Studio. +change period certification code more smart. +add params directory. + +ver 1.2.1 +------- +Fix typo in SFMT-alti.c SFMT-sse2.c +marge SFMT-alti.c and SFMT-alti.h into SFMT-alti.h +marge SFMT-sse2.c and SFMT-sse2.h into SFMT-sse2.h +This version is not released. + +ver 1.2 +------- +Support many periods: 2^{607}, 2^{1279}, 2^{2281}, 2^{4253}, 2^{11213}, +2^{19937}, 2^{44497}, 2^{86243}, 2^{132049} +Fix typo in LICENSE.txt. +Add cast to vec_perm for SFMT-alti.c, SFMT-alti64.c. +combine source codes. + +ver 1.1 +------- +The period certification method is changed from constant to function. +The convert functions from 32-bit and 64-bit integer to double are added. +The documentation is changed. +Sample programs are added. + +ver 1.0 +------- +The first version. diff --git a/dep/SFMT/CMakeLists.txt b/dep/SFMT/CMakeLists.txt index 1348d3203b1..84f5cc6afc4 100644 --- a/dep/SFMT/CMakeLists.txt +++ b/dep/SFMT/CMakeLists.txt @@ -1,4 +1,5 @@ # This file is part of the TrinityCore Project. See AUTHORS file for Copyright information +# Copyright 2019 Josua Mayer <josua.mayer97@gmail.com> # # This file is free software; as a special exception the author gives # unlimited permission to copy and/or distribute it, with or without @@ -8,8 +9,70 @@ # WITHOUT ANY WARRANTY, to the extent permitted by law; without even the # implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. -add_library(sfmt INTERFACE) +set(SFMT_SOURCES + SFMT.c + SFMT.h + SFMT-alti.h + SFMT-common.h + SFMT-neon.h + SFMT-params.h + SFMT-params607.h + SFMT-params1279.h + SFMT-params2281.h + SFMT-params4253.h + SFMT-params11213.h + SFMT-params19937.h + SFMT-params44497.h + SFMT-params86243.h + SFMT-params132049.h + SFMT-params216091.h + SFMT-sse2.h + SFMT-sse2-msc.h) + +add_library(sfmt STATIC ${SFMT_SOURCES}) target_include_directories(sfmt INTERFACE ${CMAKE_CURRENT_SOURCE_DIR}) + +# using the standard Mersenne exponent 19937 +target_compile_definitions(sfmt PUBLIC -DSFMT_MEXP=19937) + +# enable SIMD instructions if available +include(CheckCCompilerFlag) +if (CMAKE_SYSTEM_PROCESSOR MATCHES "(powerpc|ppc)64|(powerpc|ppc)64le") + check_c_compiler_flag("-maltivec" HAVE_ALTIVEC) + if (HAVE_ALTIVEC) + target_compile_options(sfmt PRIVATE -mabi=altivec -maltivec) + target_compile_definitions(sfmt PUBLIC -DHAVE_ALTIVEC) + else () + message(WARNING "Altivec not available - performance will be poor!") + endif () +elseif (CMAKE_SYSTEM_PROCESSOR MATCHES "arm|ARM") + check_c_compiler_flag(-mfpu=neon HAVE_NEON) + if (HAVE_NEON) + target_compile_options(sfmt PRIVATE -mfpu=neon -ftree-vectorize) + target_compile_definitions(sfmt PUBLIC -DHAVE_NEON) + else () + message(WARNING "Neon not available - performance will be poor!") + endif () +elseif (CMAKE_SYSTEM_PROCESSOR MATCHES "aarch64|AARCH64") + check_c_compiler_flag(-march=armv8-a+simd HAVE_NEON) + if (HAVE_NEON) + target_compile_options(sfmt PRIVATE -ftree-vectorize) + target_compile_definitions(sfmt PUBLIC -DHAVE_NEON) + else () + message(WARNING "Neon not available - performance will be poor!") + endif () +elseif (CMAKE_SYSTEM_PROCESSOR MATCHES "i686|amd64|x86_64|AMD64") + #SSE2 is always available + set(HAVE_SSE2 1) + + if (NOT CMAKE_C_COMPILER_ID MATCHES "MSVC") + target_compile_options(sfmt PRIVATE -msse2) + endif () + target_compile_definitions(sfmt PUBLIC -DHAVE_SSE2) +endif () + +# inherit trinitycore generic build options (e.g. fPIC) +target_link_libraries(sfmt PRIVATE trinity-dependency-interface) diff --git a/dep/SFMT/LICENSE.txt b/dep/SFMT/LICENSE.txt new file mode 100644 index 00000000000..6f9c4a6d6a1 --- /dev/null +++ b/dep/SFMT/LICENSE.txt @@ -0,0 +1,32 @@ +Copyright (c) 2006,2007 Mutsuo Saito, Makoto Matsumoto and Hiroshima +University. +Copyright (c) 2012 Mutsuo Saito, Makoto Matsumoto, Hiroshima University +and The University of Tokyo. +All rights reserved. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are +met: + + * Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above + copyright notice, this list of conditions and the following + disclaimer in the documentation and/or other materials provided + with the distribution. + * Neither the names of Hiroshima University, The University of + Tokyo nor the names of its contributors may be used to endorse + or promote products derived from this software without specific + prior written permission. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. diff --git a/dep/SFMT/README.txt b/dep/SFMT/README.txt new file mode 100644 index 00000000000..271fc40b651 --- /dev/null +++ b/dep/SFMT/README.txt @@ -0,0 +1,27 @@ + ================================================================= + SFMT ver. 1.5 + SIMD oriented Fast Mersenne Twister(SFMT) + + Mutsuo Saito (Hiroshima University) and + Makoto Matsumoto (The University of Tokyo) + + Copyright (C) 2006, 2007 Mutsuo Saito, Makoto Matsumoto and Hiroshima + University. + Copyright (c) 2012 Mutsuo Saito, Makoto Matsumoto, Hiroshima University + and The University of Tokyo. + All rights reserved. + + The (modified) BSD License is applied to this software, see LICENSE.txt + ================================================================= + CAUTION: + BIGENDIAN OR ALTIVEC FEATURES ARE NOT TESTED AT ALL. + + To see documents, see html/index.html. + + To make test program, see html/howto-compile.html + + If you want to redistribute and/or change source files, see LICENSE.txt. + + When you change these files and redistribute them, PLEASE write your + e-mail address in redistribution and write to contact YOU first if + users of your changed source encounter troubles. diff --git a/dep/SFMT/SFMT-alti.h b/dep/SFMT/SFMT-alti.h new file mode 100644 index 00000000000..81c98b12b49 --- /dev/null +++ b/dep/SFMT/SFMT-alti.h @@ -0,0 +1,156 @@ +#pragma once +/** + * @file SFMT-alti.h + * + * @brief SIMD oriented Fast Mersenne Twister(SFMT) + * pseudorandom number generator + * + * @author Mutsuo Saito (Hiroshima University) + * @author Makoto Matsumoto (Hiroshima University) + * + * Copyright (C) 2007 Mutsuo Saito, Makoto Matsumoto and Hiroshima + * University. All rights reserved. + * + * The new BSD License is applied to this software. + * see LICENSE.txt + */ + +#ifndef SFMT_ALTI_H +#define SFMT_ALTI_H + +inline static vector unsigned int vec_recursion(vector unsigned int a, + vector unsigned int b, + vector unsigned int c, + vector unsigned int d); + +/** + * This function represents the recursion formula in AltiVec and BIG ENDIAN. + * @param a a 128-bit part of the interal state array + * @param b a 128-bit part of the interal state array + * @param c a 128-bit part of the interal state array + * @param d a 128-bit part of the interal state array + * @return output + */ +inline static vector unsigned int vec_recursion(vector unsigned int a, + vector unsigned int b, + vector unsigned int c, + vector unsigned int d) { + + const vector unsigned int sl1 = SFMT_ALTI_SL1; + const vector unsigned int sr1 = SFMT_ALTI_SR1; +#ifdef ONLY64 + const vector unsigned int mask = SFMT_ALTI_MSK64; + const vector unsigned char perm_sl = SFMT_ALTI_SL2_PERM64; + const vector unsigned char perm_sr = SFMT_ALTI_SR2_PERM64; +#else + const vector unsigned int mask = SFMT_ALTI_MSK; + const vector unsigned char perm_sl = SFMT_ALTI_SL2_PERM; + const vector unsigned char perm_sr = SFMT_ALTI_SR2_PERM; +#endif + vector unsigned int v, w, x, y, z; + x = vec_perm(a, (vector unsigned int)perm_sl, perm_sl); + v = a; + y = vec_sr(b, sr1); + z = vec_perm(c, (vector unsigned int)perm_sr, perm_sr); + w = vec_sl(d, sl1); + z = vec_xor(z, w); + y = vec_and(y, mask); + v = vec_xor(v, x); + z = vec_xor(z, y); + z = vec_xor(z, v); + return z; +} + +/** + * This function fills the internal state array with pseudorandom + * integers. + */ +void sfmt_gen_rand_all(sfmt_t * sfmt) { + int i; + vector unsigned int r, r1, r2; + + r1 = sfmt->state[N - 2].s; + r2 = sfmt->state[N - 1].s; + for (i = 0; i < N - POS1; i++) { + r = vec_recursion(sfmt->state[i].s, sfmt->state[i + POS1].s, r1, r2); + sfmt->state[i].s = r; + r1 = r2; + r2 = r; + } + for (; i < N; i++) { + r = vec_recursion(sfmt->state[i].s, sfmt->state[i + POS1 - N].s, r1, r2); + sfmt->state[i].s = r; + r1 = r2; + r2 = r; + } +} + +/** + * This function fills the user-specified array with pseudorandom + * integers. + * + * @param array an 128-bit array to be filled by pseudorandom numbers. + * @param size number of 128-bit pesudorandom numbers to be generated. + */ +inline static void gen_rand_array(sfmt_t * sfmt, w128_t *array, int size) { + int i, j; + vector unsigned int r, r1, r2; + + r1 = sfmt->state[N - 2].s; + r2 = sfmt->state[N - 1].s; + for (i = 0; i < N - POS1; i++) { + r = vec_recursion(sfmt->state[i].s, sfmt->state[i + POS1].s, r1, r2); + array[i].s = r; + r1 = r2; + r2 = r; + } + for (; i < N; i++) { + r = vec_recursion(sfmt->state[i].s, array[i + POS1 - N].s, r1, r2); + array[i].s = r; + r1 = r2; + r2 = r; + } + /* main loop */ + for (; i < size - N; i++) { + r = vec_recursion(array[i - N].s, array[i + POS1 - N].s, r1, r2); + array[i].s = r; + r1 = r2; + r2 = r; + } + for (j = 0; j < 2 * N - size; j++) { + sfmt->state[j].s = array[j + size - N].s; + } + for (; i < size; i++) { + r = vec_recursion(array[i - N].s, array[i + POS1 - N].s, r1, r2); + array[i].s = r; + sfmt->state[j++].s = r; + r1 = r2; + r2 = r; + } +} + +#ifndef ONLY64 +#if defined(__APPLE__) +#define SFMT_ALTI_SWAP (vector unsigned char) \ + (4, 5, 6, 7, 0, 1, 2, 3, 12, 13, 14, 15, 8, 9, 10, 11) +#else +#define SFMT_ALTI_SWAP {4, 5, 6, 7, 0, 1, 2, 3, 12, 13, 14, 15, 8, 9, 10, 11} +#endif +/** + * This function swaps high and low 32-bit of 64-bit integers in user + * specified array. + * + * @param array an 128-bit array to be swaped. + * @param size size of 128-bit array. + */ +inline static void swap(w128_t *array, int size) { + int i; + const vector unsigned char perm = SFMT_ALTI_SWAP; + + for (i = 0; i < size; i++) { + array[i].s = vec_perm(array[i].s, (vector unsigned int)perm, perm); + } +} +#endif + +#endif diff --git a/dep/SFMT/SFMT-common.h b/dep/SFMT/SFMT-common.h new file mode 100644 index 00000000000..a5a9b05047a --- /dev/null +++ b/dep/SFMT/SFMT-common.h @@ -0,0 +1,164 @@ +#pragma once +/** + * @file SFMT-common.h + * + * @brief SIMD oriented Fast Mersenne Twister(SFMT) pseudorandom + * number generator with jump function. This file includes common functions + * used in random number generation and jump. + * + * @author Mutsuo Saito (Hiroshima University) + * @author Makoto Matsumoto (The University of Tokyo) + * + * Copyright (C) 2006, 2007 Mutsuo Saito, Makoto Matsumoto and Hiroshima + * University. + * Copyright (C) 2012 Mutsuo Saito, Makoto Matsumoto, Hiroshima + * University and The University of Tokyo. + * All rights reserved. + * + * The 3-clause BSD License is applied to this software, see + * LICENSE.txt + */ +#ifndef SFMT_COMMON_H +#define SFMT_COMMON_H + +#if defined(__cplusplus) +extern "C" { +#endif + +#include "SFMT.h" + +inline static void do_recursion(w128_t * r, w128_t * a, w128_t * b, + w128_t * c, w128_t * d); + +inline static void rshift128(w128_t *out, w128_t const *in, int shift); +inline static void lshift128(w128_t *out, w128_t const *in, int shift); + +/** + * This function simulates SIMD 128-bit right shift by the standard C. + * The 128-bit integer given in in is shifted by (shift * 8) bits. + * This function simulates the LITTLE ENDIAN SIMD. + * @param out the output of this function + * @param in the 128-bit data to be shifted + * @param shift the shift value + */ +#ifdef ONLY64 +inline static void rshift128(w128_t *out, w128_t const *in, int shift) { + uint64_t th, tl, oh, ol; + + th = ((uint64_t)in->u[2] << 32) | ((uint64_t)in->u[3]); + tl = ((uint64_t)in->u[0] << 32) | ((uint64_t)in->u[1]); + + oh = th >> (shift * 8); + ol = tl >> (shift * 8); + ol |= th << (64 - shift * 8); + out->u[0] = (uint32_t)(ol >> 32); + out->u[1] = (uint32_t)ol; + out->u[2] = (uint32_t)(oh >> 32); + out->u[3] = (uint32_t)oh; +} +#else +inline static void rshift128(w128_t *out, w128_t const *in, int shift) +{ + uint64_t th, tl, oh, ol; + + th = ((uint64_t)in->u[3] << 32) | ((uint64_t)in->u[2]); + tl = ((uint64_t)in->u[1] << 32) | ((uint64_t)in->u[0]); + + oh = th >> (shift * 8); + ol = tl >> (shift * 8); + ol |= th << (64 - shift * 8); + out->u[1] = (uint32_t)(ol >> 32); + out->u[0] = (uint32_t)ol; + out->u[3] = (uint32_t)(oh >> 32); + out->u[2] = (uint32_t)oh; +} +#endif +/** + * This function simulates SIMD 128-bit left shift by the standard C. + * The 128-bit integer given in in is shifted by (shift * 8) bits. + * This function simulates the LITTLE ENDIAN SIMD. + * @param out the output of this function + * @param in the 128-bit data to be shifted + * @param shift the shift value + */ +#ifdef ONLY64 +inline static void lshift128(w128_t *out, w128_t const *in, int shift) { + uint64_t th, tl, oh, ol; + + th = ((uint64_t)in->u[2] << 32) | ((uint64_t)in->u[3]); + tl = ((uint64_t)in->u[0] << 32) | ((uint64_t)in->u[1]); + + oh = th << (shift * 8); + ol = tl << (shift * 8); + oh |= tl >> (64 - shift * 8); + out->u[0] = (uint32_t)(ol >> 32); + out->u[1] = (uint32_t)ol; + out->u[2] = (uint32_t)(oh >> 32); + out->u[3] = (uint32_t)oh; +} +#else +inline static void lshift128(w128_t *out, w128_t const *in, int shift) +{ + uint64_t th, tl, oh, ol; + + th = ((uint64_t)in->u[3] << 32) | ((uint64_t)in->u[2]); + tl = ((uint64_t)in->u[1] << 32) | ((uint64_t)in->u[0]); + + oh = th << (shift * 8); + ol = tl << (shift * 8); + oh |= tl >> (64 - shift * 8); + out->u[1] = (uint32_t)(ol >> 32); + out->u[0] = (uint32_t)ol; + out->u[3] = (uint32_t)(oh >> 32); + out->u[2] = (uint32_t)oh; +} +#endif +/** + * This function represents the recursion formula. + * @param r output + * @param a a 128-bit part of the internal state array + * @param b a 128-bit part of the internal state array + * @param c a 128-bit part of the internal state array + * @param d a 128-bit part of the internal state array + */ +#ifdef ONLY64 +inline static void do_recursion(w128_t *r, w128_t *a, w128_t *b, w128_t *c, + w128_t *d) { + w128_t x; + w128_t y; + + lshift128(&x, a, SFMT_SL2); + rshift128(&y, c, SFMT_SR2); + r->u[0] = a->u[0] ^ x.u[0] ^ ((b->u[0] >> SFMT_SR1) & SFMT_MSK2) ^ y.u[0] + ^ (d->u[0] << SFMT_SL1); + r->u[1] = a->u[1] ^ x.u[1] ^ ((b->u[1] >> SFMT_SR1) & SFMT_MSK1) ^ y.u[1] + ^ (d->u[1] << SFMT_SL1); + r->u[2] = a->u[2] ^ x.u[2] ^ ((b->u[2] >> SFMT_SR1) & SFMT_MSK4) ^ y.u[2] + ^ (d->u[2] << SFMT_SL1); + r->u[3] = a->u[3] ^ x.u[3] ^ ((b->u[3] >> SFMT_SR1) & SFMT_MSK3) ^ y.u[3] + ^ (d->u[3] << SFMT_SL1); +} +#else +inline static void do_recursion(w128_t *r, w128_t *a, w128_t *b, + w128_t *c, w128_t *d) +{ + w128_t x; + w128_t y; + + lshift128(&x, a, SFMT_SL2); + rshift128(&y, c, SFMT_SR2); + r->u[0] = a->u[0] ^ x.u[0] ^ ((b->u[0] >> SFMT_SR1) & SFMT_MSK1) + ^ y.u[0] ^ (d->u[0] << SFMT_SL1); + r->u[1] = a->u[1] ^ x.u[1] ^ ((b->u[1] >> SFMT_SR1) & SFMT_MSK2) + ^ y.u[1] ^ (d->u[1] << SFMT_SL1); + r->u[2] = a->u[2] ^ x.u[2] ^ ((b->u[2] >> SFMT_SR1) & SFMT_MSK3) + ^ y.u[2] ^ (d->u[2] << SFMT_SL1); + r->u[3] = a->u[3] ^ x.u[3] ^ ((b->u[3] >> SFMT_SR1) & SFMT_MSK4) + ^ y.u[3] ^ (d->u[3] << SFMT_SL1); +} +#endif +#if defined(__cplusplus) +} +#endif + +#endif // SFMT_COMMON_H diff --git a/dep/SFMT/SFMT-hotfix1.diff b/dep/SFMT/SFMT-hotfix1.diff deleted file mode 100644 index ba7810dc100..00000000000 --- a/dep/SFMT/SFMT-hotfix1.diff +++ /dev/null @@ -1,14 +0,0 @@ -diff --git a/dep/SFMT/SFMT.h b/dep/SFMT/SFMT.h -index 3d15d65..ccf21ce 100644 ---- a/dep/SFMT/SFMT.h -+++ b/dep/SFMT/SFMT.h -@@ -173,7 +173,8 @@ public: - uint32_t statesize = SFMT_N*4; // Size of state vector - - // Fill state vector with random numbers from seed -- ((uint32_t*)state)[0] = y; -+ uint32_t* s = (uint32_t*)&state; -+ s[0] = y; - const uint32_t factor = 1812433253U;// Multiplication factor - - for (i = 1; i < statesize; i++) { diff --git a/dep/SFMT/SFMT-neon.h b/dep/SFMT/SFMT-neon.h new file mode 100644 index 00000000000..8790b8b60ee --- /dev/null +++ b/dep/SFMT/SFMT-neon.h @@ -0,0 +1,111 @@ +/** + * @file SFMT-neon.h + * @brief SIMD oriented Fast Mersenne Twister(SFMT) for ARM with 128b NEON + * + * @author Masaki Ota + * + * @note We assume LITTLE ENDIAN in this file + */ + +#ifndef SFMT_NEON_H +#define SFMT_NEON_H + +inline static void neon_recursion(uint32x4_t * r, uint32x4_t a, uint32x4_t b, + uint32x4_t c, uint32x4_t d); + + + +/** + * This function represents the recursion formula. + * @param r an output + * @param a a 128-bit part of the interal state array + * @param b a 128-bit part of the interal state array + * @param c a 128-bit part of the interal state array + * @param d a 128-bit part of the interal state array + */ +inline static void neon_recursion(uint32x4_t * r, uint32x4_t a, uint32x4_t b, + uint32x4_t c, uint32x4_t d) +{ + uint32x4_t v, x, y, z; + static const uint32x4_t vzero = {0,0,0,0}; + static const uint32x4_t vmask = {SFMT_MSK1, SFMT_MSK2, SFMT_MSK3, SFMT_MSK4}; + +#define rotate_bytes(A, B, C) vreinterpretq_u32_u8(vextq_u8(vreinterpretq_u8_u32(A),vreinterpretq_u8_u32(B),(C))) + + y = vshrq_n_u32(b, SFMT_SR1); + z = rotate_bytes(c, vzero, SFMT_SR2); + v = vshlq_n_u32(d, SFMT_SL1); + z = veorq_u32(z, a); + z = veorq_u32(z, v); + x = rotate_bytes(vzero, a, 16-SFMT_SL2); + y = vandq_u32(y, vmask); + z = veorq_u32(z, x); + z = veorq_u32(z, y); + *r = z; +} + +/** + * This function fills the internal state array with pseudorandom + * integers. + * @param sfmt SFMT internal state + */ +void sfmt_gen_rand_all(sfmt_t * sfmt) { + int i; + uint32x4_t r1, r2; + w128_t * pstate = sfmt->state; + + r1 = pstate[SFMT_N - 2].si; + r2 = pstate[SFMT_N - 1].si; + for (i = 0; i < SFMT_N - SFMT_POS1; i++) { + neon_recursion(&pstate[i].si, pstate[i].si, pstate[i + SFMT_POS1].si, r1, r2); + r1 = r2; + r2 = pstate[i].si; + } + for (; i < SFMT_N; i++) { + neon_recursion(&pstate[i].si, pstate[i].si, pstate[i + SFMT_POS1 - SFMT_N].si, r1, r2); + r1 = r2; + r2 = pstate[i].si; + } +} + +/** + * This function fills the user-specified array with pseudorandom + * integers. + * @param sfmt SFMT internal state. + * @param array an 128-bit array to be filled by pseudorandom numbers. + * @param size number of 128-bit pseudorandom numbers to be generated. + */ +static void gen_rand_array(sfmt_t * sfmt, w128_t * array, int size) +{ + int i, j; + uint32x4_t r1, r2; + w128_t * pstate = sfmt->state; + + r1 = pstate[SFMT_N - 2].si; + r2 = pstate[SFMT_N - 1].si; + for (i = 0; i < SFMT_N - SFMT_POS1; i++) { + neon_recursion(&array[i].si, pstate[i].si, pstate[i + SFMT_POS1].si, r1, r2); + r1 = r2; + r2 = array[i].si; + } + for (; i < SFMT_N; i++) { + neon_recursion(&array[i].si, pstate[i].si, array[i + SFMT_POS1 - SFMT_N].si, r1, r2); + r1 = r2; + r2 = array[i].si; + } + for (; i < size - SFMT_N; i++) { + neon_recursion(&array[i].si, array[i - SFMT_N].si, array[i + SFMT_POS1 - SFMT_N].si, r1, r2); + r1 = r2; + r2 = array[i].si; + } + for (j = 0; j < 2 * SFMT_N - size; j++) { + pstate[j] = array[j + size - SFMT_N]; + } + for (; i < size; i++, j++) { + neon_recursion(&array[i].si, array[i - SFMT_N].si, array[i + SFMT_POS1 - SFMT_N].si, r1, r2); + r1 = r2; + r2 = pstate[j].si = array[i].si; + } +} + +#endif diff --git a/dep/SFMT/SFMT-params.h b/dep/SFMT/SFMT-params.h new file mode 100644 index 00000000000..2fe663ab65b --- /dev/null +++ b/dep/SFMT/SFMT-params.h @@ -0,0 +1,98 @@ +#pragma once +#ifndef SFMT_PARAMS_H +#define SFMT_PARAMS_H + +#if !defined(SFMT_MEXP) +#if defined(__GNUC__) && !defined(__ICC) + #warning "SFMT_MEXP is not defined. I assume MEXP is 19937." +#endif + #define SFMT_MEXP 19937 +#endif +/*----------------- + BASIC DEFINITIONS + -----------------*/ +/** Mersenne Exponent. The period of the sequence + * is a multiple of 2^MEXP-1. + * #define SFMT_MEXP 19937 */ +/** SFMT generator has an internal state array of 128-bit integers, + * and N is its size. */ +#define SFMT_N (SFMT_MEXP / 128 + 1) +/** N32 is the size of internal state array when regarded as an array + * of 32-bit integers.*/ +#define SFMT_N32 (SFMT_N * 4) +/** N64 is the size of internal state array when regarded as an array + * of 64-bit integers.*/ +#define SFMT_N64 (SFMT_N * 2) + +/*---------------------- + the parameters of SFMT + following definitions are in paramsXXXX.h file. + ----------------------*/ +/** the pick up position of the array. +#define SFMT_POS1 122 +*/ + +/** the parameter of shift left as four 32-bit registers. +#define SFMT_SL1 18 + */ + +/** the parameter of shift left as one 128-bit register. + * The 128-bit integer is shifted by (SFMT_SL2 * 8) bits. +#define SFMT_SL2 1 +*/ + +/** the parameter of shift right as four 32-bit registers. +#define SFMT_SR1 11 +*/ + +/** the parameter of shift right as one 128-bit register. + * The 128-bit integer is shifted by (SFMT_SR2 * 8) bits. +#define SFMT_SR2 1 +*/ + +/** A bitmask, used in the recursion. These parameters are introduced + * to break symmetry of SIMD. +#define SFMT_MSK1 0xdfffffefU +#define SFMT_MSK2 0xddfecb7fU +#define SFMT_MSK3 0xbffaffffU +#define SFMT_MSK4 0xbffffff6U +*/ + +/** These definitions are part of a 128-bit period certification vector. +#define SFMT_PARITY1 0x00000001U +#define SFMT_PARITY2 0x00000000U +#define SFMT_PARITY3 0x00000000U +#define SFMT_PARITY4 0xc98e126aU +*/ + +#if SFMT_MEXP == 607 + #include "SFMT-params607.h" +#elif SFMT_MEXP == 1279 + #include "SFMT-params1279.h" +#elif SFMT_MEXP == 2281 + #include "SFMT-params2281.h" +#elif SFMT_MEXP == 4253 + #include "SFMT-params4253.h" +#elif SFMT_MEXP == 11213 + #include "SFMT-params11213.h" +#elif SFMT_MEXP == 19937 + #include "SFMT-params19937.h" +#elif SFMT_MEXP == 44497 + #include "SFMT-params44497.h" +#elif SFMT_MEXP == 86243 + #include "SFMT-params86243.h" +#elif SFMT_MEXP == 132049 + #include "SFMT-params132049.h" +#elif SFMT_MEXP == 216091 + #include "SFMT-params216091.h" +#else +#if defined(__GNUC__) && !defined(__ICC) + #error "SFMT_MEXP is not valid." + #undef SFMT_MEXP +#else + #undef SFMT_MEXP +#endif + +#endif + +#endif /* SFMT_PARAMS_H */ diff --git a/dep/SFMT/SFMT-params11213.h b/dep/SFMT/SFMT-params11213.h new file mode 100644 index 00000000000..57277477901 --- /dev/null +++ b/dep/SFMT/SFMT-params11213.h @@ -0,0 +1,50 @@ +#pragma once +#ifndef SFMT_PARAMS11213_H +#define SFMT_PARAMS11213_H + +#define SFMT_POS1 68 +#define SFMT_SL1 14 +#define SFMT_SL2 3 +#define SFMT_SR1 7 +#define SFMT_SR2 3 +#define SFMT_MSK1 0xeffff7fbU +#define SFMT_MSK2 0xffffffefU +#define SFMT_MSK3 0xdfdfbfffU +#define SFMT_MSK4 0x7fffdbfdU +#define SFMT_PARITY1 0x00000001U +#define SFMT_PARITY2 0x00000000U +#define SFMT_PARITY3 0xe8148000U +#define SFMT_PARITY4 0xd0c7afa3U + + +/* PARAMETERS FOR ALTIVEC */ +#if defined(__APPLE__) /* For OSX */ + #define SFMT_ALTI_SL1 \ + (vector unsigned int)(SFMT_SL1, SFMT_SL1, SFMT_SL1, SFMT_SL1) + #define SFMT_ALTI_SR1 \ + (vector unsigned int)(SFMT_SR1, SFMT_SR1, SFMT_SR1, SFMT_SR1) + #define SFMT_ALTI_MSK \ + (vector unsigned int)(SFMT_MSK1, SFMT_MSK2, SFMT_MSK3, SFMT_MSK4) + #define SFMT_ALTI_MSK64 \ + (vector unsigned int)(SFMT_MSK2, SFMT_MSK1, SFMT_MSK4, SFMT_MSK3) + #define SFMT_ALTI_SL2_PERM \ + (vector unsigned char)(3,21,21,21,7,0,1,2,11,4,5,6,15,8,9,10) + #define SFMT_ALTI_SL2_PERM64 \ + (vector unsigned char)(3,4,5,6,7,29,29,29,11,12,13,14,15,0,1,2) + #define SFMT_ALTI_SR2_PERM \ + (vector unsigned char)(5,6,7,0,9,10,11,4,13,14,15,8,19,19,19,12) + #define SFMT_ALTI_SR2_PERM64 \ + (vector unsigned char)(13,14,15,0,1,2,3,4,19,19,19,8,9,10,11,12) +#else /* For OTHER OSs(Linux?) */ + #define SFMT_ALTI_SL1 {SFMT_SL1, SFMT_SL1, SFMT_SL1, SFMT_SL1} + #define SFMT_ALTI_SR1 {SFMT_SR1, SFMT_SR1, SFMT_SR1, SFMT_SR1} + #define SFMT_ALTI_MSK {SFMT_MSK1, SFMT_MSK2, SFMT_MSK3, SFMT_MSK4} + #define SFMT_ALTI_MSK64 {SFMT_MSK2, SFMT_MSK1, SFMT_MSK4, SFMT_MSK3} + #define SFMT_ALTI_SL2_PERM {3,21,21,21,7,0,1,2,11,4,5,6,15,8,9,10} + #define SFMT_ALTI_SL2_PERM64 {3,4,5,6,7,29,29,29,11,12,13,14,15,0,1,2} + #define SFMT_ALTI_SR2_PERM {5,6,7,0,9,10,11,4,13,14,15,8,19,19,19,12} + #define SFMT_ALTI_SR2_PERM64 {13,14,15,0,1,2,3,4,19,19,19,8,9,10,11,12} +#endif /* For OSX */ +#define SFMT_IDSTR "SFMT-11213:68-14-3-7-3:effff7fb-ffffffef-dfdfbfff-7fffdbfd" + +#endif /* SFMT_PARAMS11213_H */ diff --git a/dep/SFMT/SFMT-params1279.h b/dep/SFMT/SFMT-params1279.h new file mode 100644 index 00000000000..12dbc132679 --- /dev/null +++ b/dep/SFMT/SFMT-params1279.h @@ -0,0 +1,50 @@ +#pragma once +#ifndef SFMT_PARAMS1279_H +#define SFMT_PARAMS1279_H + +#define SFMT_POS1 7 +#define SFMT_SL1 14 +#define SFMT_SL2 3 +#define SFMT_SR1 5 +#define SFMT_SR2 1 +#define SFMT_MSK1 0xf7fefffdU +#define SFMT_MSK2 0x7fefcfffU +#define SFMT_MSK3 0xaff3ef3fU +#define SFMT_MSK4 0xb5ffff7fU +#define SFMT_PARITY1 0x00000001U +#define SFMT_PARITY2 0x00000000U +#define SFMT_PARITY3 0x00000000U +#define SFMT_PARITY4 0x20000000U + + +/* PARAMETERS FOR ALTIVEC */ +#if defined(__APPLE__) /* For OSX */ + #define SFMT_ALTI_SL1 \ + (vector unsigned int)(SFMT_SL1, SFMT_SL1, SFMT_SL1, SFMT_SL1) + #define SFMT_ALTI_SR1 \ + (vector unsigned int)(SFMT_SR1, SFMT_SR1, SFMT_SR1, SFMT_SR1) + #define SFMT_ALTI_MSK \ + (vector unsigned int)(SFMT_MSK1, SFMT_MSK2, SFMT_MSK3, SFMT_MSK4) + #define SFMT_ALTI_MSK64 \ + (vector unsigned int)(SFMT_MSK2, SFMT_MSK1, SFMT_MSK4, SFMT_MSK3) + #define SFMT_ALTI_SL2_PERM \ + (vector unsigned char)(3,21,21,21,7,0,1,2,11,4,5,6,15,8,9,10) + #define SFMT_ALTI_SL2_PERM64 \ + (vector unsigned char)(3,4,5,6,7,29,29,29,11,12,13,14,15,0,1,2) + #define SFMT_ALTI_SR2_PERM \ + (vector unsigned char)(7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14) + #define SFMT_ALTI_SR2_PERM64 \ + (vector unsigned char)(15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14) +#else /* For OTHER OSs(Linux?) */ + #define SFMT_ALTI_SL1 {SFMT_SL1, SFMT_SL1, SFMT_SL1, SFMT_SL1} + #define SFMT_ALTI_SR1 {SFMT_SR1, SFMT_SR1, SFMT_SR1, SFMT_SR1} + #define SFMT_ALTI_MSK {SFMT_MSK1, SFMT_MSK2, SFMT_MSK3, SFMT_MSK4} + #define SFMT_ALTI_MSK64 {SFMT_MSK2, SFMT_MSK1, SFMT_MSK4, SFMT_MSK3} + #define SFMT_ALTI_SL2_PERM {3,21,21,21,7,0,1,2,11,4,5,6,15,8,9,10} + #define SFMT_ALTI_SL2_PERM64 {3,4,5,6,7,29,29,29,11,12,13,14,15,0,1,2} + #define SFMT_ALTI_SR2_PERM {7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14} + #define SFMT_ALTI_SR2_PERM64 {15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14} +#endif /* For OSX */ +#define SFMT_IDSTR "SFMT-1279:7-14-3-5-1:f7fefffd-7fefcfff-aff3ef3f-b5ffff7f" + +#endif /* SFMT_PARAMS1279_H */ diff --git a/dep/SFMT/SFMT-params132049.h b/dep/SFMT/SFMT-params132049.h new file mode 100644 index 00000000000..febc69f2f03 --- /dev/null +++ b/dep/SFMT/SFMT-params132049.h @@ -0,0 +1,50 @@ +#pragma once +#ifndef SFMT_PARAMS132049_H +#define SFMT_PARAMS132049_H + +#define SFMT_POS1 110 +#define SFMT_SL1 19 +#define SFMT_SL2 1 +#define SFMT_SR1 21 +#define SFMT_SR2 1 +#define SFMT_MSK1 0xffffbb5fU +#define SFMT_MSK2 0xfb6ebf95U +#define SFMT_MSK3 0xfffefffaU +#define SFMT_MSK4 0xcff77fffU +#define SFMT_PARITY1 0x00000001U +#define SFMT_PARITY2 0x00000000U +#define SFMT_PARITY3 0xcb520000U +#define SFMT_PARITY4 0xc7e91c7dU + + +/* PARAMETERS FOR ALTIVEC */ +#if defined(__APPLE__) /* For OSX */ + #define SFMT_ALTI_SL1 \ + (vector unsigned int)(SFMT_SL1, SFMT_SL1, SFMT_SL1, SFMT_SL1) + #define SFMT_ALTI_SR1 \ + (vector unsigned int)(SFMT_SR1, SFMT_SR1, SFMT_SR1, SFMT_SR1) + #define SFMT_ALTI_MSK \ + (vector unsigned int)(SFMT_MSK1, SFMT_MSK2, SFMT_MSK3, SFMT_MSK4) + #define SFMT_ALTI_MSK64 \ + (vector unsigned int)(SFMT_MSK2, SFMT_MSK1, SFMT_MSK4, SFMT_MSK3) + #define SFMT_ALTI_SL2_PERM \ + (vector unsigned char)(1,2,3,23,5,6,7,0,9,10,11,4,13,14,15,8) + #define SFMT_ALTI_SL2_PERM64 \ + (vector unsigned char)(1,2,3,4,5,6,7,31,9,10,11,12,13,14,15,0) + #define SFMT_ALTI_SR2_PERM \ + (vector unsigned char)(7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14) + #define SFMT_ALTI_SR2_PERM64 \ + (vector unsigned char)(15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14) +#else /* For OTHER OSs(Linux?) */ + #define SFMT_ALTI_SL1 {SFMT_SL1, SFMT_SL1, SFMT_SL1, SFMT_SL1} + #define SFMT_ALTI_SR1 {SFMT_SR1, SFMT_SR1, SFMT_SR1, SFMT_SR1} + #define SFMT_ALTI_MSK {SFMT_MSK1, SFMT_MSK2, SFMT_MSK3, SFMT_MSK4} + #define SFMT_ALTI_MSK64 {SFMT_MSK2, SFMT_MSK1, SFMT_MSK4, SFMT_MSK3} + #define SFMT_ALTI_SL2_PERM {1,2,3,23,5,6,7,0,9,10,11,4,13,14,15,8} + #define SFMT_ALTI_SL2_PERM64 {1,2,3,4,5,6,7,31,9,10,11,12,13,14,15,0} + #define SFMT_ALTI_SR2_PERM {7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14} + #define SFMT_ALTI_SR2_PERM64 {15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14} +#endif /* For OSX */ +#define SFMT_IDSTR "SFMT-132049:110-19-1-21-1:ffffbb5f-fb6ebf95-fffefffa-cff77fff" + +#endif /* SFMT_PARAMS132049_H */ diff --git a/dep/SFMT/SFMT-params19937.h b/dep/SFMT/SFMT-params19937.h new file mode 100644 index 00000000000..fc49fa14e50 --- /dev/null +++ b/dep/SFMT/SFMT-params19937.h @@ -0,0 +1,50 @@ +#pragma once +#ifndef SFMT_PARAMS19937_H +#define SFMT_PARAMS19937_H + +#define SFMT_POS1 122 +#define SFMT_SL1 18 +#define SFMT_SL2 1 +#define SFMT_SR1 11 +#define SFMT_SR2 1 +#define SFMT_MSK1 0xdfffffefU +#define SFMT_MSK2 0xddfecb7fU +#define SFMT_MSK3 0xbffaffffU +#define SFMT_MSK4 0xbffffff6U +#define SFMT_PARITY1 0x00000001U +#define SFMT_PARITY2 0x00000000U +#define SFMT_PARITY3 0x00000000U +#define SFMT_PARITY4 0x13c9e684U + + +/* PARAMETERS FOR ALTIVEC */ +#if defined(__APPLE__) /* For OSX */ + #define SFMT_ALTI_SL1 \ + (vector unsigned int)(SFMT_SL1, SFMT_SL1, SFMT_SL1, SFMT_SL1) + #define SFMT_ALTI_SR1 \ + (vector unsigned int)(SFMT_SR1, SFMT_SR1, SFMT_SR1, SFMT_SR1) + #define SFMT_ALTI_MSK \ + (vector unsigned int)(SFMT_MSK1, SFMT_MSK2, SFMT_MSK3, SFMT_MSK4) + #define SFMT_ALTI_MSK64 \ + (vector unsigned int)(SFMT_MSK2, SFMT_MSK1, SFMT_MSK4, SFMT_MSK3) + #define SFMT_ALTI_SL2_PERM \ + (vector unsigned char)(1,2,3,23,5,6,7,0,9,10,11,4,13,14,15,8) + #define SFMT_ALTI_SL2_PERM64 \ + (vector unsigned char)(1,2,3,4,5,6,7,31,9,10,11,12,13,14,15,0) + #define SFMT_ALTI_SR2_PERM \ + (vector unsigned char)(7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14) + #define SFMT_ALTI_SR2_PERM64 \ + (vector unsigned char)(15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14) +#else /* For OTHER OSs(Linux?) */ + #define SFMT_ALTI_SL1 {SFMT_SL1, SFMT_SL1, SFMT_SL1, SFMT_SL1} + #define SFMT_ALTI_SR1 {SFMT_SR1, SFMT_SR1, SFMT_SR1, SFMT_SR1} + #define SFMT_ALTI_MSK {SFMT_MSK1, SFMT_MSK2, SFMT_MSK3, SFMT_MSK4} + #define SFMT_ALTI_MSK64 {SFMT_MSK2, SFMT_MSK1, SFMT_MSK4, SFMT_MSK3} + #define SFMT_ALTI_SL2_PERM {1,2,3,23,5,6,7,0,9,10,11,4,13,14,15,8} + #define SFMT_ALTI_SL2_PERM64 {1,2,3,4,5,6,7,31,9,10,11,12,13,14,15,0} + #define SFMT_ALTI_SR2_PERM {7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14} + #define SFMT_ALTI_SR2_PERM64 {15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14} +#endif /* For OSX */ +#define SFMT_IDSTR "SFMT-19937:122-18-1-11-1:dfffffef-ddfecb7f-bffaffff-bffffff6" + +#endif /* SFMT_PARAMS19937_H */ diff --git a/dep/SFMT/SFMT-params216091.h b/dep/SFMT/SFMT-params216091.h new file mode 100644 index 00000000000..60d58574b83 --- /dev/null +++ b/dep/SFMT/SFMT-params216091.h @@ -0,0 +1,50 @@ +#pragma once +#ifndef SFMT_PARAMS216091_H +#define SFMT_PARAMS216091_H + +#define SFMT_POS1 627 +#define SFMT_SL1 11 +#define SFMT_SL2 3 +#define SFMT_SR1 10 +#define SFMT_SR2 1 +#define SFMT_MSK1 0xbff7bff7U +#define SFMT_MSK2 0xbfffffffU +#define SFMT_MSK3 0xbffffa7fU +#define SFMT_MSK4 0xffddfbfbU +#define SFMT_PARITY1 0xf8000001U +#define SFMT_PARITY2 0x89e80709U +#define SFMT_PARITY3 0x3bd2b64bU +#define SFMT_PARITY4 0x0c64b1e4U + + +/* PARAMETERS FOR ALTIVEC */ +#if defined(__APPLE__) /* For OSX */ + #define SFMT_ALTI_SL1 \ + (vector unsigned int)(SFMT_SL1, SFMT_SL1, SFMT_SL1, SFMT_SL1) + #define SFMT_ALTI_SR1 \ + (vector unsigned int)(SFMT_SR1, SFMT_SR1, SFMT_SR1, SFMT_SR1) + #define SFMT_ALTI_MSK \ + (vector unsigned int)(SFMT_MSK1, SFMT_MSK2, SFMT_MSK3, SFMT_MSK4) + #define SFMT_ALTI_MSK64 \ + (vector unsigned int)(SFMT_MSK2, SFMT_MSK1, SFMT_MSK4, SFMT_MSK3) + #define SFMT_ALTI_SL2_PERM \ + (vector unsigned char)(3,21,21,21,7,0,1,2,11,4,5,6,15,8,9,10) + #define SFMT_ALTI_SL2_PERM64 \ + (vector unsigned char)(3,4,5,6,7,29,29,29,11,12,13,14,15,0,1,2) + #define SFMT_ALTI_SR2_PERM \ + (vector unsigned char)(7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14) + #define SFMT_ALTI_SR2_PERM64 \ + (vector unsigned char)(15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14) +#else /* For OTHER OSs(Linux?) */ + #define SFMT_ALTI_SL1 {SFMT_SL1, SFMT_SL1, SFMT_SL1, SFMT_SL1} + #define SFMT_ALTI_SR1 {SFMT_SR1, SFMT_SR1, SFMT_SR1, SFMT_SR1} + #define SFMT_ALTI_MSK {SFMT_MSK1, SFMT_MSK2, SFMT_MSK3, SFMT_MSK4} + #define SFMT_ALTI_MSK64 {SFMT_MSK2, SFMT_MSK1, SFMT_MSK4, SFMT_MSK3} + #define SFMT_ALTI_SL2_PERM {3,21,21,21,7,0,1,2,11,4,5,6,15,8,9,10} + #define SFMT_ALTI_SL2_PERM64 {3,4,5,6,7,29,29,29,11,12,13,14,15,0,1,2} + #define SFMT_ALTI_SR2_PERM {7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14} + #define SFMT_ALTI_SR2_PERM64 {15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14} +#endif /* For OSX */ +#define SFMT_IDSTR "SFMT-216091:627-11-3-10-1:bff7bff7-bfffffff-bffffa7f-ffddfbfb" + +#endif /* SFMT_PARAMS216091_H */ diff --git a/dep/SFMT/SFMT-params2281.h b/dep/SFMT/SFMT-params2281.h new file mode 100644 index 00000000000..05729515ce0 --- /dev/null +++ b/dep/SFMT/SFMT-params2281.h @@ -0,0 +1,50 @@ +#pragma once +#ifndef SFMT_PARAMS2281_H +#define SFMT_PARAMS2281_H + +#define SFMT_POS1 12 +#define SFMT_SL1 19 +#define SFMT_SL2 1 +#define SFMT_SR1 5 +#define SFMT_SR2 1 +#define SFMT_MSK1 0xbff7ffbfU +#define SFMT_MSK2 0xfdfffffeU +#define SFMT_MSK3 0xf7ffef7fU +#define SFMT_MSK4 0xf2f7cbbfU +#define SFMT_PARITY1 0x00000001U +#define SFMT_PARITY2 0x00000000U +#define SFMT_PARITY3 0x00000000U +#define SFMT_PARITY4 0x41dfa600U + + +/* PARAMETERS FOR ALTIVEC */ +#if defined(__APPLE__) /* For OSX */ + #define SFMT_ALTI_SL1 \ + (vector unsigned int)(SFMT_SL1, SFMT_SL1, SFMT_SL1, SFMT_SL1) + #define SFMT_ALTI_SR1 \ + (vector unsigned int)(SFMT_SR1, SFMT_SR1, SFMT_SR1, SFMT_SR1) + #define SFMT_ALTI_MSK \ + (vector unsigned int)(SFMT_MSK1, SFMT_MSK2, SFMT_MSK3, SFMT_MSK4) + #define SFMT_ALTI_MSK64 \ + (vector unsigned int)(SFMT_MSK2, SFMT_MSK1, SFMT_MSK4, SFMT_MSK3) + #define SFMT_ALTI_SL2_PERM \ + (vector unsigned char)(1,2,3,23,5,6,7,0,9,10,11,4,13,14,15,8) + #define SFMT_ALTI_SL2_PERM64 \ + (vector unsigned char)(1,2,3,4,5,6,7,31,9,10,11,12,13,14,15,0) + #define SFMT_ALTI_SR2_PERM \ + (vector unsigned char)(7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14) + #define SFMT_ALTI_SR2_PERM64 \ + (vector unsigned char)(15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14) +#else /* For OTHER OSs(Linux?) */ + #define SFMT_ALTI_SL1 {SFMT_SL1, SFMT_SL1, SFMT_SL1, SFMT_SL1} + #define SFMT_ALTI_SR1 {SFMT_SR1, SFMT_SR1, SFMT_SR1, SFMT_SR1} + #define SFMT_ALTI_MSK {SFMT_MSK1, SFMT_MSK2, SFMT_MSK3, SFMT_MSK4} + #define SFMT_ALTI_MSK64 {SFMT_MSK2, SFMT_MSK1, SFMT_MSK4, SFMT_MSK3} + #define SFMT_ALTI_SL2_PERM {1,2,3,23,5,6,7,0,9,10,11,4,13,14,15,8} + #define SFMT_ALTI_SL2_PERM64 {1,2,3,4,5,6,7,31,9,10,11,12,13,14,15,0} + #define SFMT_ALTI_SR2_PERM {7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14} + #define SFMT_ALTI_SR2_PERM64 {15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14} +#endif /* For OSX */ +#define SFMT_IDSTR "SFMT-2281:12-19-1-5-1:bff7ffbf-fdfffffe-f7ffef7f-f2f7cbbf" + +#endif /* SFMT_PARAMS2281_H */ diff --git a/dep/SFMT/SFMT-params4253.h b/dep/SFMT/SFMT-params4253.h new file mode 100644 index 00000000000..2e7bfad1fed --- /dev/null +++ b/dep/SFMT/SFMT-params4253.h @@ -0,0 +1,50 @@ +#pragma once +#ifndef SFMT_PARAMS4253_H +#define SFMT_PARAMS4253_H + +#define SFMT_POS1 17 +#define SFMT_SL1 20 +#define SFMT_SL2 1 +#define SFMT_SR1 7 +#define SFMT_SR2 1 +#define SFMT_MSK1 0x9f7bffffU +#define SFMT_MSK2 0x9fffff5fU +#define SFMT_MSK3 0x3efffffbU +#define SFMT_MSK4 0xfffff7bbU +#define SFMT_PARITY1 0xa8000001U +#define SFMT_PARITY2 0xaf5390a3U +#define SFMT_PARITY3 0xb740b3f8U +#define SFMT_PARITY4 0x6c11486dU + + +/* PARAMETERS FOR ALTIVEC */ +#if defined(__APPLE__) /* For OSX */ + #define SFMT_ALTI_SL1 \ + (vector unsigned int)(SFMT_SL1, SFMT_SL1, SFMT_SL1, SFMT_SL1) + #define SFMT_ALTI_SR1 \ + (vector unsigned int)(SFMT_SR1, SFMT_SR1, SFMT_SR1, SFMT_SR1) + #define SFMT_ALTI_MSK \ + (vector unsigned int)(SFMT_MSK1, SFMT_MSK2, SFMT_MSK3, SFMT_MSK4) + #define SFMT_ALTI_MSK64 \ + (vector unsigned int)(SFMT_MSK2, SFMT_MSK1, SFMT_MSK4, SFMT_MSK3) + #define SFMT_ALTI_SL2_PERM \ + (vector unsigned char)(1,2,3,23,5,6,7,0,9,10,11,4,13,14,15,8) + #define SFMT_ALTI_SL2_PERM64 \ + (vector unsigned char)(1,2,3,4,5,6,7,31,9,10,11,12,13,14,15,0) + #define SFMT_ALTI_SR2_PERM \ + (vector unsigned char)(7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14) + #define SFMT_ALTI_SR2_PERM64 \ + (vector unsigned char)(15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14) +#else /* For OTHER OSs(Linux?) */ + #define SFMT_ALTI_SL1 {SFMT_SL1, SFMT_SL1, SFMT_SL1, SFMT_SL1} + #define SFMT_ALTI_SR1 {SFMT_SR1, SFMT_SR1, SFMT_SR1, SFMT_SR1} + #define SFMT_ALTI_MSK {SFMT_MSK1, SFMT_MSK2, SFMT_MSK3, SFMT_MSK4} + #define SFMT_ALTI_MSK64 {SFMT_MSK2, SFMT_MSK1, SFMT_MSK4, SFMT_MSK3} + #define SFMT_ALTI_SL2_PERM {1,2,3,23,5,6,7,0,9,10,11,4,13,14,15,8} + #define SFMT_ALTI_SL2_PERM64 {1,2,3,4,5,6,7,31,9,10,11,12,13,14,15,0} + #define SFMT_ALTI_SR2_PERM {7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14} + #define SFMT_ALTI_SR2_PERM64 {15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14} +#endif /* For OSX */ +#define SFMT_IDSTR "SFMT-4253:17-20-1-7-1:9f7bffff-9fffff5f-3efffffb-fffff7bb" + +#endif /* SFMT_PARAMS4253_H */ diff --git a/dep/SFMT/SFMT-params44497.h b/dep/SFMT/SFMT-params44497.h new file mode 100644 index 00000000000..f9829db0bbb --- /dev/null +++ b/dep/SFMT/SFMT-params44497.h @@ -0,0 +1,50 @@ +#pragma once +#ifndef SFMT_PARAMS44497_H +#define SFMT_PARAMS44497_H + +#define SFMT_POS1 330 +#define SFMT_SL1 5 +#define SFMT_SL2 3 +#define SFMT_SR1 9 +#define SFMT_SR2 3 +#define SFMT_MSK1 0xeffffffbU +#define SFMT_MSK2 0xdfbebfffU +#define SFMT_MSK3 0xbfbf7befU +#define SFMT_MSK4 0x9ffd7bffU +#define SFMT_PARITY1 0x00000001U +#define SFMT_PARITY2 0x00000000U +#define SFMT_PARITY3 0xa3ac4000U +#define SFMT_PARITY4 0xecc1327aU + + +/* PARAMETERS FOR ALTIVEC */ +#if defined(__APPLE__) /* For OSX */ + #define SFMT_ALTI_SL1 \ + (vector unsigned int)(SFMT_SL1, SFMT_SL1, SFMT_SL1, SFMT_SL1) + #define SFMT_ALTI_SR1 \ + (vector unsigned int)(SFMT_SR1, SFMT_SR1, SFMT_SR1, SFMT_SR1) + #define SFMT_ALTI_MSK \ + (vector unsigned int)(SFMT_MSK1, SFMT_MSK2, SFMT_MSK3, SFMT_MSK4) + #define SFMT_ALTI_MSK64 \ + (vector unsigned int)(SFMT_MSK2, SFMT_MSK1, SFMT_MSK4, SFMT_MSK3) + #define SFMT_ALTI_SL2_PERM \ + (vector unsigned char)(3,21,21,21,7,0,1,2,11,4,5,6,15,8,9,10) + #define SFMT_ALTI_SL2_PERM64 \ + (vector unsigned char)(3,4,5,6,7,29,29,29,11,12,13,14,15,0,1,2) + #define SFMT_ALTI_SR2_PERM \ + (vector unsigned char)(5,6,7,0,9,10,11,4,13,14,15,8,19,19,19,12) + #define SFMT_ALTI_SR2_PERM64 \ + (vector unsigned char)(13,14,15,0,1,2,3,4,19,19,19,8,9,10,11,12) +#else /* For OTHER OSs(Linux?) */ + #define SFMT_ALTI_SL1 {SFMT_SL1, SFMT_SL1, SFMT_SL1, SFMT_SL1} + #define SFMT_ALTI_SR1 {SFMT_SR1, SFMT_SR1, SFMT_SR1, SFMT_SR1} + #define SFMT_ALTI_MSK {SFMT_MSK1, SFMT_MSK2, SFMT_MSK3, SFMT_MSK4} + #define SFMT_ALTI_MSK64 {SFMT_MSK2, SFMT_MSK1, SFMT_MSK4, SFMT_MSK3} + #define SFMT_ALTI_SL2_PERM {3,21,21,21,7,0,1,2,11,4,5,6,15,8,9,10} + #define SFMT_ALTI_SL2_PERM64 {3,4,5,6,7,29,29,29,11,12,13,14,15,0,1,2} + #define SFMT_ALTI_SR2_PERM {5,6,7,0,9,10,11,4,13,14,15,8,19,19,19,12} + #define SFMT_ALTI_SR2_PERM64 {13,14,15,0,1,2,3,4,19,19,19,8,9,10,11,12} +#endif /* For OSX */ +#define SFMT_IDSTR "SFMT-44497:330-5-3-9-3:effffffb-dfbebfff-bfbf7bef-9ffd7bff" + +#endif /* SFMT_PARAMS44497_H */ diff --git a/dep/SFMT/SFMT-params607.h b/dep/SFMT/SFMT-params607.h new file mode 100644 index 00000000000..085367254af --- /dev/null +++ b/dep/SFMT/SFMT-params607.h @@ -0,0 +1,50 @@ +#pragma once +#ifndef SFMT_PARAMS607_H +#define SFMT_PARAMS607_H + +#define SFMT_POS1 2 +#define SFMT_SL1 15 +#define SFMT_SL2 3 +#define SFMT_SR1 13 +#define SFMT_SR2 3 +#define SFMT_MSK1 0xfdff37ffU +#define SFMT_MSK2 0xef7f3f7dU +#define SFMT_MSK3 0xff777b7dU +#define SFMT_MSK4 0x7ff7fb2fU +#define SFMT_PARITY1 0x00000001U +#define SFMT_PARITY2 0x00000000U +#define SFMT_PARITY3 0x00000000U +#define SFMT_PARITY4 0x5986f054U + + +/* PARAMETERS FOR ALTIVEC */ +#if defined(__APPLE__) /* For OSX */ + #define SFMT_ALTI_SL1 \ + (vector unsigned int)(SFMT_SL1, SFMT_SL1, SFMT_SL1, SFMT_SL1) + #define SFMT_ALTI_SR1 \ + (vector unsigned int)(SFMT_SR1, SFMT_SR1, SFMT_SR1, SFMT_SR1) + #define SFMT_ALTI_MSK \ + (vector unsigned int)(SFMT_MSK1, SFMT_MSK2, SFMT_MSK3, SFMT_MSK4) + #define SFMT_ALTI_MSK64 \ + (vector unsigned int)(SFMT_MSK2, SFMT_MSK1, SFMT_MSK4, SFMT_MSK3) + #define SFMT_ALTI_SL2_PERM \ + (vector unsigned char)(3,21,21,21,7,0,1,2,11,4,5,6,15,8,9,10) + #define SFMT_ALTI_SL2_PERM64 \ + (vector unsigned char)(3,4,5,6,7,29,29,29,11,12,13,14,15,0,1,2) + #define SFMT_ALTI_SR2_PERM \ + (vector unsigned char)(5,6,7,0,9,10,11,4,13,14,15,8,19,19,19,12) + #define SFMT_ALTI_SR2_PERM64 \ + (vector unsigned char)(13,14,15,0,1,2,3,4,19,19,19,8,9,10,11,12) +#else /* For OTHER OSs(Linux?) */ + #define SFMT_ALTI_SL1 {SFMT_SL1, SFMT_SL1, SFMT_SL1, SFMT_SL1} + #define SFMT_ALTI_SR1 {SFMT_SR1, SFMT_SR1, SFMT_SR1, SFMT_SR1} + #define SFMT_ALTI_MSK {SFMT_MSK1, SFMT_MSK2, SFMT_MSK3, SFMT_MSK4} + #define SFMT_ALTI_MSK64 {SFMT_MSK2, SFMT_MSK1, SFMT_MSK4, SFMT_MSK3} + #define SFMT_ALTI_SL2_PERM {3,21,21,21,7,0,1,2,11,4,5,6,15,8,9,10} + #define SFMT_ALTI_SL2_PERM64 {3,4,5,6,7,29,29,29,11,12,13,14,15,0,1,2} + #define SFMT_ALTI_SR2_PERM {5,6,7,0,9,10,11,4,13,14,15,8,19,19,19,12} + #define SFMT_ALTI_SR2_PERM64 {13,14,15,0,1,2,3,4,19,19,19,8,9,10,11,12} +#endif /* For OSX */ +#define SFMT_IDSTR "SFMT-607:2-15-3-13-3:fdff37ff-ef7f3f7d-ff777b7d-7ff7fb2f" + +#endif /* SFMT_PARAMS607_H */ diff --git a/dep/SFMT/SFMT-params86243.h b/dep/SFMT/SFMT-params86243.h new file mode 100644 index 00000000000..89773b156a9 --- /dev/null +++ b/dep/SFMT/SFMT-params86243.h @@ -0,0 +1,50 @@ +#pragma once +#ifndef SFMT_PARAMS86243_H +#define SFMT_PARAMS86243_H + +#define SFMT_POS1 366 +#define SFMT_SL1 6 +#define SFMT_SL2 7 +#define SFMT_SR1 19 +#define SFMT_SR2 1 +#define SFMT_MSK1 0xfdbffbffU +#define SFMT_MSK2 0xbff7ff3fU +#define SFMT_MSK3 0xfd77efffU +#define SFMT_MSK4 0xbf9ff3ffU +#define SFMT_PARITY1 0x00000001U +#define SFMT_PARITY2 0x00000000U +#define SFMT_PARITY3 0x00000000U +#define SFMT_PARITY4 0xe9528d85U + + +/* PARAMETERS FOR ALTIVEC */ +#if defined(__APPLE__) /* For OSX */ + #define SFMT_ALTI_SL1 \ + (vector unsigned int)(SFMT_SL1, SFMT_SL1, SFMT_SL1, SFMT_SL1) + #define SFMT_ALTI_SR1 \ + (vector unsigned int)(SFMT_SR1, SFMT_SR1, SFMT_SR1, SFMT_SR1) + #define SFMT_ALTI_MSK \ + (vector unsigned int)(SFMT_MSK1, SFMT_MSK2, SFMT_MSK3, SFMT_MSK4) + #define SFMT_ALTI_MSK64 \ + (vector unsigned int)(SFMT_MSK2, SFMT_MSK1, SFMT_MSK4, SFMT_MSK3) + #define SFMT_ALTI_SL2_PERM \ + (vector unsigned char)(25,25,25,25,3,25,25,25,7,0,1,2,11,4,5,6) + #define SFMT_ALTI_SL2_PERM64 \ + (vector unsigned char)(7,25,25,25,25,25,25,25,15,0,1,2,3,4,5,6) + #define SFMT_ALTI_SR2_PERM \ + (vector unsigned char)(7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14) + #define SFMT_ALTI_SR2_PERM64 \ + (vector unsigned char)(15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14) +#else /* For OTHER OSs(Linux?) */ + #define SFMT_ALTI_SL1 {SFMT_SL1, SFMT_SL1, SFMT_SL1, SFMT_SL1} + #define SFMT_ALTI_SR1 {SFMT_SR1, SFMT_SR1, SFMT_SR1, SFMT_SR1} + #define SFMT_ALTI_MSK {SFMT_MSK1, SFMT_MSK2, SFMT_MSK3, SFMT_MSK4} + #define SFMT_ALTI_MSK64 {SFMT_MSK2, SFMT_MSK1, SFMT_MSK4, SFMT_MSK3} + #define SFMT_ALTI_SL2_PERM {25,25,25,25,3,25,25,25,7,0,1,2,11,4,5,6} + #define SFMT_ALTI_SL2_PERM64 {7,25,25,25,25,25,25,25,15,0,1,2,3,4,5,6} + #define SFMT_ALTI_SR2_PERM {7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14} + #define SFMT_ALTI_SR2_PERM64 {15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14} +#endif /* For OSX */ +#define SFMT_IDSTR "SFMT-86243:366-6-7-19-1:fdbffbff-bff7ff3f-fd77efff-bf9ff3ff" + +#endif /* SFMT_PARAMS86243_H */ diff --git a/dep/SFMT/SFMT-sse2-msc.h b/dep/SFMT/SFMT-sse2-msc.h new file mode 100644 index 00000000000..9839df1a218 --- /dev/null +++ b/dep/SFMT/SFMT-sse2-msc.h @@ -0,0 +1,321 @@ +#pragma once +/** + * @file SFMT-sse2-msc.h + * @brief SIMD oriented Fast Mersenne Twister(SFMT) for Intel SSE2 for MSC + * + * @author Mutsuo Saito (Hiroshima University) + * @author Makoto Matsumoto (Hiroshima University) + * + * @note We assume LITTLE ENDIAN in this file + * + * Copyright (C) 2006, 2007 Mutsuo Saito, Makoto Matsumoto and Hiroshima + * University. All rights reserved. + * Copyright (C) 2013 Mutsuo Saito, Makoto Matsumoto and Hiroshima + * University. + * + * The new BSD License is applied to this software, see LICENSE.txt + */ + +#ifndef SFMT_SSE2_MSC_H +#define SFMT_SSE2_MSC_H + +/* This header file is included only when _MSC_VER is defined. */ +#if _MSC_VER > 1700 + +inline static __m128i __vectorcall mm_recursion(__m128i a, __m128i b, + __m128i c, __m128i d); + +/** + * This function represents the recursion formula. + * @param a a 128-bit part of the interal state array + * @param b a 128-bit part of the interal state array + * @param c a 128-bit part of the interal state array + * @param d a 128-bit part of the interal state array + * @return new value + */ +inline static __m128i __vectorcall mm_recursion(__m128i a, __m128i b, + __m128i c, __m128i d) +{ + __m128i v, x, y, z; + + y = _mm_srli_epi32(b, SFMT_SR1); + z = _mm_srli_si128(c, SFMT_SR2); + v = _mm_slli_epi32(d, SFMT_SL1); + z = _mm_xor_si128(z, a); + z = _mm_xor_si128(z, v); + x = _mm_slli_si128(a, SFMT_SL2); + y = _mm_and_si128(y, sse2_param_mask.si); + z = _mm_xor_si128(z, x); + return _mm_xor_si128(z, y); +} + +/** + * This function fills the internal state array with pseudorandom + * integers. + * @param sfmt SFMT internal state + */ +void sfmt_gen_rand_all(sfmt_t * sfmt) { + int i; + __m128i r1, r2; + w128_t * pstate = sfmt->state; + + r1 = pstate[SFMT_N - 2].si; + r2 = pstate[SFMT_N - 1].si; + for (i = 0; i < SFMT_N - SFMT_POS1; i++) { + pstate[i].si = mm_recursion(pstate[i].si, + pstate[i + SFMT_POS1].si, r1, r2); + r1 = r2; + r2 = pstate[i].si; + } + for (; i < SFMT_N; i++) { + pstate[i].si = mm_recursion(pstate[i].si, + pstate[i + SFMT_POS1 - SFMT_N].si, + r1, r2); + r1 = r2; + r2 = pstate[i].si; + } +} + +/** + * This function fills the user-specified array with pseudorandom + * integers. + * @param sfmt SFMT internal state. + * @param array an 128-bit array to be filled by pseudorandom numbers. + * @param size number of 128-bit pseudorandom numbers to be generated. + */ +static void gen_rand_array(sfmt_t * sfmt, w128_t * array, int size) +{ + int i, j; + __m128i r1, r2; + w128_t * pstate = sfmt->state; + + r1 = pstate[SFMT_N - 2].si; + r2 = pstate[SFMT_N - 1].si; + for (i = 0; i < SFMT_N - SFMT_POS1; i++) { + array[i].si = mm_recursion(pstate[i].si, + pstate[i + SFMT_POS1].si, r1, r2); + r1 = r2; + r2 = array[i].si; + } + for (; i < SFMT_N; i++) { + array[i].si = mm_recursion(pstate[i].si, + array[i + SFMT_POS1 - SFMT_N].si, r1, r2); + r1 = r2; + r2 = array[i].si; + } + for (; i < size - SFMT_N; i++) { + array[i].si = mm_recursion(array[i - SFMT_N].si, + array[i + SFMT_POS1 - SFMT_N].si, r1, r2); + r1 = r2; + r2 = array[i].si; + } + for (j = 0; j < 2 * SFMT_N - size; j++) { + pstate[j] = array[j + size - SFMT_N]; + } + for (; i < size; i++, j++) { + array[i].si = mm_recursion(array[i - SFMT_N].si, + array[i + SFMT_POS1 - SFMT_N].si, r1, r2); + r1 = r2; + r2 = array[i].si; + pstate[j] = array[i]; + } +} + +#elif defined(SFMT_USE_MACRO_FUNCTION_FOR_MSC) +/** + * This function represents the recursion formula. + * @param r an output + * @param a a 128-bit part of the interal state array + * @param b a 128-bit part of the interal state array + * @param c a 128-bit part of the interal state array + * @param d a 128-bit part of the interal state array + */ +#define mm_recursion(r, a, b, c, d) \ +do { \ + __m128i v, x, y, z; \ + \ + y = _mm_srli_epi32((b), SFMT_SR1); \ + z = _mm_srli_si128((c), SFMT_SR2); \ + v = _mm_slli_epi32((d), SFMT_SL1); \ + z = _mm_xor_si128(z, (a)); \ + z = _mm_xor_si128(z, v); \ + x = _mm_slli_si128((a), SFMT_SL2); \ + y = _mm_and_si128(y, sse2_param_mask.si); \ + z = _mm_xor_si128(z, x); \ + r = _mm_xor_si128(z, y); \ +} while (0) + +/** + * This function fills the internal state array with pseudorandom + * integers. + * @param sfmt SFMT internal state + */ +void sfmt_gen_rand_all(sfmt_t * sfmt) { + int i; + __m128i r1, r2; + w128_t * pstate = sfmt->state; + + r1 = pstate[SFMT_N - 2].si; + r2 = pstate[SFMT_N - 1].si; + for (i = 0; i < SFMT_N - SFMT_POS1; i++) { + mm_recursion(pstate[i].si, pstate[i].si, + pstate[i + SFMT_POS1].si, r1, r2); + r1 = r2; + r2 = pstate[i].si; + } + for (; i < SFMT_N; i++) { + mm_recursion(pstate[i].si, pstate[i].si, + pstate[i + SFMT_POS1 - SFMT_N].si, + r1, r2); + r1 = r2; + r2 = pstate[i].si; + } +} + +/** + * This function fills the user-specified array with pseudorandom + * integers. + * @param sfmt SFMT internal state. + * @param array an 128-bit array to be filled by pseudorandom numbers. + * @param size number of 128-bit pseudorandom numbers to be generated. + */ +static void gen_rand_array(sfmt_t * sfmt, w128_t * array, int size) +{ + int i, j; + __m128i r1, r2; + w128_t * pstate = sfmt->state; + + r1 = pstate[SFMT_N - 2].si; + r2 = pstate[SFMT_N - 1].si; + for (i = 0; i < SFMT_N - SFMT_POS1; i++) { + mm_recursion(array[i].si, pstate[i].si, + pstate[i + SFMT_POS1].si, r1, r2); + r1 = r2; + r2 = array[i].si; + } + for (; i < SFMT_N; i++) { + mm_recursion(array[i].si, pstate[i].si, + array[i + SFMT_POS1 - SFMT_N].si, r1, r2); + r1 = r2; + r2 = array[i].si; + } + for (; i < size - SFMT_N; i++) { + mm_recursion(array[i].si, array[i - SFMT_N].si, + array[i + SFMT_POS1 - SFMT_N].si, r1, r2); + r1 = r2; + r2 = array[i].si; + } + for (j = 0; j < 2 * SFMT_N - size; j++) { + pstate[j] = array[j + size - SFMT_N]; + } + for (; i < size; i++, j++) { + mm_recursion(array[i].si, array[i - SFMT_N].si, + array[i + SFMT_POS1 - SFMT_N].si, r1, r2); + r1 = r2; + r2 = array[i].si; + pstate[j] = array[i]; + } +} +#else +inline static void mm_recursion(__m128i * r, __m128i a, __m128i b, + __m128i c, __m128i * d); + +/** + * This function represents the recursion formula. + * @param r an output + * @param a a 128-bit part of the interal state array + * @param b a 128-bit part of the interal state array + * @param c a 128-bit part of the interal state array + * @param d a 128-bit part of the interal state array + */ +inline static void mm_recursion(__m128i * r, __m128i a, __m128i b, + __m128i c, __m128i * d) +{ + __m128i v, x, y, z; + + y = _mm_srli_epi32(b, SFMT_SR1); + z = _mm_srli_si128(c, SFMT_SR2); + v = _mm_slli_epi32(*d, SFMT_SL1); + z = _mm_xor_si128(z, a); + z = _mm_xor_si128(z, v); + x = _mm_slli_si128(a, SFMT_SL2); + y = _mm_and_si128(y, sse2_param_mask.si); + z = _mm_xor_si128(z, x); + z = _mm_xor_si128(z, y); + *r = z; +} + +/** + * This function fills the internal state array with pseudorandom + * integers. + * @param sfmt SFMT internal state + */ +void sfmt_gen_rand_all(sfmt_t * sfmt) { + int i; + __m128i r1, r2; + w128_t * pstate = sfmt->state; + + r1 = pstate[SFMT_N - 2].si; + r2 = pstate[SFMT_N - 1].si; + for (i = 0; i < SFMT_N - SFMT_POS1; i++) { + mm_recursion(&pstate[i].si, pstate[i].si, + pstate[i + SFMT_POS1].si, r1, &r2); + r1 = r2; + r2 = pstate[i].si; + } + for (; i < SFMT_N; i++) { + mm_recursion(&pstate[i].si, pstate[i].si, + pstate[i + SFMT_POS1 - SFMT_N].si, + r1, &r2); + r1 = r2; + r2 = pstate[i].si; + } +} + +/** + * This function fills the user-specified array with pseudorandom + * integers. + * @param sfmt SFMT internal state. + * @param array an 128-bit array to be filled by pseudorandom numbers. + * @param size number of 128-bit pseudorandom numbers to be generated. + */ +static void gen_rand_array(sfmt_t * sfmt, w128_t * array, int size) +{ + int i, j; + __m128i r1, r2; + w128_t * pstate = sfmt->state; + + r1 = pstate[SFMT_N - 2].si; + r2 = pstate[SFMT_N - 1].si; + for (i = 0; i < SFMT_N - SFMT_POS1; i++) { + mm_recursion(&array[i].si, pstate[i].si, + pstate[i + SFMT_POS1].si, r1, &r2); + r1 = r2; + r2 = array[i].si; + } + for (; i < SFMT_N; i++) { + mm_recursion(&array[i].si, pstate[i].si, + array[i + SFMT_POS1 - SFMT_N].si, r1, &r2); + r1 = r2; + r2 = array[i].si; + } + for (; i < size - SFMT_N; i++) { + mm_recursion(&array[i].si, array[i - SFMT_N].si, + array[i + SFMT_POS1 - SFMT_N].si, r1, &r2); + r1 = r2; + r2 = array[i].si; + } + for (j = 0; j < 2 * SFMT_N - size; j++) { + pstate[j] = array[j + size - SFMT_N]; + } + for (; i < size; i++, j++) { + mm_recursion(&array[i].si, array[i - SFMT_N].si, + array[i + SFMT_POS1 - SFMT_N].si, r1, &r2); + r1 = r2; + r2 = array[i].si; + pstate[j] = array[i]; + } +} +#endif + +#endif diff --git a/dep/SFMT/SFMT-sse2.h b/dep/SFMT/SFMT-sse2.h new file mode 100644 index 00000000000..99a33ff1f8e --- /dev/null +++ b/dep/SFMT/SFMT-sse2.h @@ -0,0 +1,121 @@ +#pragma once +/** + * @file SFMT-sse2.h + * @brief SIMD oriented Fast Mersenne Twister(SFMT) for Intel SSE2 + * + * @author Mutsuo Saito (Hiroshima University) + * @author Makoto Matsumoto (Hiroshima University) + * + * @note We assume LITTLE ENDIAN in this file + * + * Copyright (C) 2006, 2007 Mutsuo Saito, Makoto Matsumoto and Hiroshima + * University. All rights reserved. + * + * The new BSD License is applied to this software, see LICENSE.txt + */ + +#ifndef SFMT_SSE2_H +#define SFMT_SSE2_H + +inline static void mm_recursion(__m128i * r, __m128i a, __m128i b, + __m128i c, __m128i d); + +/** + * This function represents the recursion formula. + * @param r an output + * @param a a 128-bit part of the interal state array + * @param b a 128-bit part of the interal state array + * @param c a 128-bit part of the interal state array + * @param d a 128-bit part of the interal state array + */ +inline static void mm_recursion(__m128i * r, __m128i a, __m128i b, + __m128i c, __m128i d) +{ + __m128i v, x, y, z; + + y = _mm_srli_epi32(b, SFMT_SR1); + z = _mm_srli_si128(c, SFMT_SR2); + v = _mm_slli_epi32(d, SFMT_SL1); + z = _mm_xor_si128(z, a); + z = _mm_xor_si128(z, v); + x = _mm_slli_si128(a, SFMT_SL2); + y = _mm_and_si128(y, sse2_param_mask.si); + z = _mm_xor_si128(z, x); + z = _mm_xor_si128(z, y); + *r = z; +} + +/** + * This function fills the internal state array with pseudorandom + * integers. + * @param sfmt SFMT internal state + */ +void sfmt_gen_rand_all(sfmt_t * sfmt) { + int i; + __m128i r1, r2; + w128_t * pstate = sfmt->state; + + r1 = pstate[SFMT_N - 2].si; + r2 = pstate[SFMT_N - 1].si; + for (i = 0; i < SFMT_N - SFMT_POS1; i++) { + mm_recursion(&pstate[i].si, pstate[i].si, + pstate[i + SFMT_POS1].si, r1, r2); + r1 = r2; + r2 = pstate[i].si; + } + for (; i < SFMT_N; i++) { + mm_recursion(&pstate[i].si, pstate[i].si, + pstate[i + SFMT_POS1 - SFMT_N].si, + r1, r2); + r1 = r2; + r2 = pstate[i].si; + } +} + +/** + * This function fills the user-specified array with pseudorandom + * integers. + * @param sfmt SFMT internal state. + * @param array an 128-bit array to be filled by pseudorandom numbers. + * @param size number of 128-bit pseudorandom numbers to be generated. + */ +static void gen_rand_array(sfmt_t * sfmt, w128_t * array, int size) +{ + int i, j; + __m128i r1, r2; + w128_t * pstate = sfmt->state; + + r1 = pstate[SFMT_N - 2].si; + r2 = pstate[SFMT_N - 1].si; + for (i = 0; i < SFMT_N - SFMT_POS1; i++) { + mm_recursion(&array[i].si, pstate[i].si, + pstate[i + SFMT_POS1].si, r1, r2); + r1 = r2; + r2 = array[i].si; + } + for (; i < SFMT_N; i++) { + mm_recursion(&array[i].si, pstate[i].si, + array[i + SFMT_POS1 - SFMT_N].si, r1, r2); + r1 = r2; + r2 = array[i].si; + } + for (; i < size - SFMT_N; i++) { + mm_recursion(&array[i].si, array[i - SFMT_N].si, + array[i + SFMT_POS1 - SFMT_N].si, r1, r2); + r1 = r2; + r2 = array[i].si; + } + for (j = 0; j < 2 * SFMT_N - size; j++) { + pstate[j] = array[j + size - SFMT_N]; + } + for (; i < size; i++, j++) { + mm_recursion(&array[i].si, array[i - SFMT_N].si, + array[i + SFMT_POS1 - SFMT_N].si, r1, r2); + r1 = r2; + r2 = array[i].si; + pstate[j] = array[i]; + } +} + + +#endif diff --git a/dep/SFMT/SFMT.c b/dep/SFMT/SFMT.c new file mode 100644 index 00000000000..b4ac9308bb5 --- /dev/null +++ b/dep/SFMT/SFMT.c @@ -0,0 +1,437 @@ +/** + * @file SFMT.c + * @brief SIMD oriented Fast Mersenne Twister(SFMT) + * + * @author Mutsuo Saito (Hiroshima University) + * @author Makoto Matsumoto (Hiroshima University) + * + * Copyright (C) 2006, 2007 Mutsuo Saito, Makoto Matsumoto and Hiroshima + * University. + * Copyright (C) 2012 Mutsuo Saito, Makoto Matsumoto, Hiroshima + * University and The University of Tokyo. + * Copyright (C) 2013 Mutsuo Saito, Makoto Matsumoto and Hiroshima + * University. + * All rights reserved. + * + * The 3-clause BSD License is applied to this software, see + * LICENSE.txt + */ + +#if defined(__cplusplus) +extern "C" { +#endif + +#include <string.h> +#include <assert.h> +#include "SFMT.h" +#include "SFMT-params.h" +#include "SFMT-common.h" + +#if defined(__BIG_ENDIAN__) && !defined(__amd64) && !defined(BIG_ENDIAN64) +#define BIG_ENDIAN64 1 +#endif +#if defined(HAVE_ALTIVEC) && !defined(BIG_ENDIAN64) +#define BIG_ENDIAN64 1 +#endif +#if defined(ONLY64) && !defined(BIG_ENDIAN64) + #if defined(__GNUC__) + #error "-DONLY64 must be specified with -DBIG_ENDIAN64" + #endif +#undef ONLY64 +#endif + +/*---------------- + STATIC FUNCTIONS + ----------------*/ +inline static int idxof(int i); +inline static void gen_rand_array(sfmt_t * sfmt, w128_t *array, int size); +inline static uint32_t func1(uint32_t x); +inline static uint32_t func2(uint32_t x); +static void period_certification(sfmt_t * sfmt); +#if defined(BIG_ENDIAN64) && !defined(ONLY64) +inline static void swap(w128_t *array, int size); +#endif + +#if defined(HAVE_ALTIVEC) + #include "SFMT-alti.h" +#elif defined(HAVE_SSE2) +/** + * parameters used by sse2. + */ + static const w128_t sse2_param_mask = {{SFMT_MSK1, SFMT_MSK2, + SFMT_MSK3, SFMT_MSK4}}; + #if defined(_MSC_VER) + #include "SFMT-sse2-msc.h" + #else + #include "SFMT-sse2.h" + #endif +#elif defined(HAVE_NEON) + #include "SFMT-neon.h" +#endif + +/** + * This function simulate a 64-bit index of LITTLE ENDIAN + * in BIG ENDIAN machine. + */ +#ifdef ONLY64 +inline static int idxof(int i) { + return i ^ 1; +} +#else +inline static int idxof(int i) { + return i; +} +#endif + +#if (!defined(HAVE_ALTIVEC)) && (!defined(HAVE_SSE2)) && (!defined(HAVE_NEON)) +/** + * This function fills the user-specified array with pseudorandom + * integers. + * + * @param sfmt SFMT internal state + * @param array an 128-bit array to be filled by pseudorandom numbers. + * @param size number of 128-bit pseudorandom numbers to be generated. + */ +inline static void gen_rand_array(sfmt_t * sfmt, w128_t *array, int size) { + int i, j; + w128_t *r1, *r2; + + r1 = &sfmt->state[SFMT_N - 2]; + r2 = &sfmt->state[SFMT_N - 1]; + for (i = 0; i < SFMT_N - SFMT_POS1; i++) { + do_recursion(&array[i], &sfmt->state[i], &sfmt->state[i + SFMT_POS1], r1, r2); + r1 = r2; + r2 = &array[i]; + } + for (; i < SFMT_N; i++) { + do_recursion(&array[i], &sfmt->state[i], + &array[i + SFMT_POS1 - SFMT_N], r1, r2); + r1 = r2; + r2 = &array[i]; + } + for (; i < size - SFMT_N; i++) { + do_recursion(&array[i], &array[i - SFMT_N], + &array[i + SFMT_POS1 - SFMT_N], r1, r2); + r1 = r2; + r2 = &array[i]; + } + for (j = 0; j < 2 * SFMT_N - size; j++) { + sfmt->state[j] = array[j + size - SFMT_N]; + } + for (; i < size; i++, j++) { + do_recursion(&array[i], &array[i - SFMT_N], + &array[i + SFMT_POS1 - SFMT_N], r1, r2); + r1 = r2; + r2 = &array[i]; + sfmt->state[j] = array[i]; + } +} +#endif + +#if defined(BIG_ENDIAN64) && !defined(ONLY64) && !defined(HAVE_ALTIVEC) +inline static void swap(w128_t *array, int size) { + int i; + uint32_t x, y; + + for (i = 0; i < size; i++) { + x = array[i].u[0]; + y = array[i].u[2]; + array[i].u[0] = array[i].u[1]; + array[i].u[2] = array[i].u[3]; + array[i].u[1] = x; + array[i].u[3] = y; + } +} +#endif +/** + * This function represents a function used in the initialization + * by init_by_array + * @param x 32-bit integer + * @return 32-bit integer + */ +static uint32_t func1(uint32_t x) { + return (x ^ (x >> 27)) * (uint32_t)1664525UL; +} + +/** + * This function represents a function used in the initialization + * by init_by_array + * @param x 32-bit integer + * @return 32-bit integer + */ +static uint32_t func2(uint32_t x) { + return (x ^ (x >> 27)) * (uint32_t)1566083941UL; +} + +/** + * This function certificate the period of 2^{MEXP} + * @param sfmt SFMT internal state + */ +static void period_certification(sfmt_t * sfmt) { + uint32_t inner = 0; + int i, j; + uint32_t work; + uint32_t *psfmt32 = &sfmt->state[0].u[0]; + const uint32_t parity[4] = {SFMT_PARITY1, SFMT_PARITY2, + SFMT_PARITY3, SFMT_PARITY4}; + + for (i = 0; i < 4; i++) { + inner ^= psfmt32[idxof(i)] & parity[i]; + } + for (i = 16; i > 0; i >>= 1) { + inner ^= inner >> i; + } + inner &= 1; + /* check OK */ + if (inner == 1) { + return; + } + /* check NG, and modification */ + for (i = 0; i < 4; i++) { + work = 1; + for (j = 0; j < 32; j++) { + if ((work & parity[i]) != 0) { + psfmt32[idxof(i)] ^= work; + return; + } + work = work << 1; + } + } +} + +/*---------------- + PUBLIC FUNCTIONS + ----------------*/ +#define UNUSED_VARIABLE(x) (void)(x) +/** + * This function returns the identification string. + * The string shows the word size, the Mersenne exponent, + * and all parameters of this generator. + * @param sfmt SFMT internal state + */ +const char *sfmt_get_idstring(sfmt_t * sfmt) { + UNUSED_VARIABLE(sfmt); + return SFMT_IDSTR; +} + +/** + * This function returns the minimum size of array used for \b + * fill_array32() function. + * @param sfmt SFMT internal state + * @return minimum size of array used for fill_array32() function. + */ +int sfmt_get_min_array_size32(sfmt_t * sfmt) { + UNUSED_VARIABLE(sfmt); + return SFMT_N32; +} + +/** + * This function returns the minimum size of array used for \b + * fill_array64() function. + * @param sfmt SFMT internal state + * @return minimum size of array used for fill_array64() function. + */ +int sfmt_get_min_array_size64(sfmt_t * sfmt) { + UNUSED_VARIABLE(sfmt); + return SFMT_N64; +} + +#if !defined(HAVE_SSE2) && !defined(HAVE_ALTIVEC) && !defined(HAVE_NEON) +/** + * This function fills the internal state array with pseudorandom + * integers. + * @param sfmt SFMT internal state + */ +void sfmt_gen_rand_all(sfmt_t * sfmt) { + int i; + w128_t *r1, *r2; + + r1 = &sfmt->state[SFMT_N - 2]; + r2 = &sfmt->state[SFMT_N - 1]; + for (i = 0; i < SFMT_N - SFMT_POS1; i++) { + do_recursion(&sfmt->state[i], &sfmt->state[i], + &sfmt->state[i + SFMT_POS1], r1, r2); + r1 = r2; + r2 = &sfmt->state[i]; + } + for (; i < SFMT_N; i++) { + do_recursion(&sfmt->state[i], &sfmt->state[i], + &sfmt->state[i + SFMT_POS1 - SFMT_N], r1, r2); + r1 = r2; + r2 = &sfmt->state[i]; + } +} +#endif + +#ifndef ONLY64 +/** + * This function generates pseudorandom 32-bit integers in the + * specified array[] by one call. The number of pseudorandom integers + * is specified by the argument size, which must be at least 624 and a + * multiple of four. The generation by this function is much faster + * than the following gen_rand function. + * + * For initialization, init_gen_rand or init_by_array must be called + * before the first call of this function. This function can not be + * used after calling gen_rand function, without initialization. + * + * @param sfmt SFMT internal state + * @param array an array where pseudorandom 32-bit integers are filled + * by this function. The pointer to the array must be \b "aligned" + * (namely, must be a multiple of 16) in the SIMD version, since it + * refers to the address of a 128-bit integer. In the standard C + * version, the pointer is arbitrary. + * + * @param size the number of 32-bit pseudorandom integers to be + * generated. size must be a multiple of 4, and greater than or equal + * to (MEXP / 128 + 1) * 4. + * + * @note \b memalign or \b posix_memalign is available to get aligned + * memory. Mac OSX doesn't have these functions, but \b malloc of OSX + * returns the pointer to the aligned memory block. + */ +void sfmt_fill_array32(sfmt_t * sfmt, uint32_t *array, int size) { + assert(sfmt->idx == SFMT_N32); + assert(size % 4 == 0); + assert(size >= SFMT_N32); + + gen_rand_array(sfmt, (w128_t *)array, size / 4); + sfmt->idx = SFMT_N32; +} +#endif + +/** + * This function generates pseudorandom 64-bit integers in the + * specified array[] by one call. The number of pseudorandom integers + * is specified by the argument size, which must be at least 312 and a + * multiple of two. The generation by this function is much faster + * than the following gen_rand function. + * + * @param sfmt SFMT internal state + * For initialization, init_gen_rand or init_by_array must be called + * before the first call of this function. This function can not be + * used after calling gen_rand function, without initialization. + * + * @param array an array where pseudorandom 64-bit integers are filled + * by this function. The pointer to the array must be "aligned" + * (namely, must be a multiple of 16) in the SIMD version, since it + * refers to the address of a 128-bit integer. In the standard C + * version, the pointer is arbitrary. + * + * @param size the number of 64-bit pseudorandom integers to be + * generated. size must be a multiple of 2, and greater than or equal + * to (MEXP / 128 + 1) * 2 + * + * @note \b memalign or \b posix_memalign is available to get aligned + * memory. Mac OSX doesn't have these functions, but \b malloc of OSX + * returns the pointer to the aligned memory block. + */ +void sfmt_fill_array64(sfmt_t * sfmt, uint64_t *array, int size) { + assert(sfmt->idx == SFMT_N32); + assert(size % 2 == 0); + assert(size >= SFMT_N64); + + gen_rand_array(sfmt, (w128_t *)array, size / 2); + sfmt->idx = SFMT_N32; + +#if defined(BIG_ENDIAN64) && !defined(ONLY64) + swap((w128_t *)array, size /2); +#endif +} + +/** + * This function initializes the internal state array with a 32-bit + * integer seed. + * + * @param sfmt SFMT internal state + * @param seed a 32-bit integer used as the seed. + */ +void sfmt_init_gen_rand(sfmt_t * sfmt, uint32_t seed) { + int i; + + uint32_t *psfmt32 = &sfmt->state[0].u[0]; + + psfmt32[idxof(0)] = seed; + for (i = 1; i < SFMT_N32; i++) { + psfmt32[idxof(i)] = 1812433253UL * (psfmt32[idxof(i - 1)] + ^ (psfmt32[idxof(i - 1)] >> 30)) + + i; + } + sfmt->idx = SFMT_N32; + period_certification(sfmt); +} + +/** + * This function initializes the internal state array, + * with an array of 32-bit integers used as the seeds + * @param sfmt SFMT internal state + * @param init_key the array of 32-bit integers, used as a seed. + * @param key_length the length of init_key. + */ +void sfmt_init_by_array(sfmt_t * sfmt, uint32_t *init_key, int key_length) { + int i, j, count; + uint32_t r; + int lag; + int mid; + int size = SFMT_N * 4; + uint32_t *psfmt32 = &sfmt->state[0].u[0]; + + if (size >= 623) { + lag = 11; + } else if (size >= 68) { + lag = 7; + } else if (size >= 39) { + lag = 5; + } else { + lag = 3; + } + mid = (size - lag) / 2; + + memset(sfmt, 0x8b, sizeof(sfmt_t)); + if (key_length + 1 > SFMT_N32) { + count = key_length + 1; + } else { + count = SFMT_N32; + } + r = func1(psfmt32[idxof(0)] ^ psfmt32[idxof(mid)] + ^ psfmt32[idxof(SFMT_N32 - 1)]); + psfmt32[idxof(mid)] += r; + r += key_length; + psfmt32[idxof(mid + lag)] += r; + psfmt32[idxof(0)] = r; + + count--; + for (i = 1, j = 0; (j < count) && (j < key_length); j++) { + r = func1(psfmt32[idxof(i)] ^ psfmt32[idxof((i + mid) % SFMT_N32)] + ^ psfmt32[idxof((i + SFMT_N32 - 1) % SFMT_N32)]); + psfmt32[idxof((i + mid) % SFMT_N32)] += r; + r += init_key[j] + i; + psfmt32[idxof((i + mid + lag) % SFMT_N32)] += r; + psfmt32[idxof(i)] = r; + i = (i + 1) % SFMT_N32; + } + for (; j < count; j++) { + r = func1(psfmt32[idxof(i)] ^ psfmt32[idxof((i + mid) % SFMT_N32)] + ^ psfmt32[idxof((i + SFMT_N32 - 1) % SFMT_N32)]); + psfmt32[idxof((i + mid) % SFMT_N32)] += r; + r += i; + psfmt32[idxof((i + mid + lag) % SFMT_N32)] += r; + psfmt32[idxof(i)] = r; + i = (i + 1) % SFMT_N32; + } + for (j = 0; j < SFMT_N32; j++) { + r = func2(psfmt32[idxof(i)] + psfmt32[idxof((i + mid) % SFMT_N32)] + + psfmt32[idxof((i + SFMT_N32 - 1) % SFMT_N32)]); + psfmt32[idxof((i + mid) % SFMT_N32)] ^= r; + r -= i; + psfmt32[idxof((i + mid + lag) % SFMT_N32)] ^= r; + psfmt32[idxof(i)] = r; + i = (i + 1) % SFMT_N32; + } + + sfmt->idx = SFMT_N32; + period_certification(sfmt); +} +#if defined(__cplusplus) +} +#endif diff --git a/dep/SFMT/SFMT.h b/dep/SFMT/SFMT.h index ccf21cecd5f..79e012d633c 100644 --- a/dep/SFMT/SFMT.h +++ b/dep/SFMT/SFMT.h @@ -1,363 +1,304 @@ -/* - * Copyright notice - * ================ - * GNU General Public License http://www.gnu.org/licenses/gpl.html - * This C++ implementation of SFMT contains parts of the original C code - * which was published under the following BSD license, which is therefore - * in effect in addition to the GNU General Public License. - * Copyright (c) 2006, 2007 by Mutsuo Saito, Makoto Matsumoto and Hiroshima University. - * Copyright (c) 2008 by Agner Fog. - * Copyright (c) 2008-2013 Trinity Core - * - * BSD License: - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are met: - * > Redistributions of source code must retain the above copyright notice, - * this list of conditions and the following disclaimer. - * > Redistributions in binary form must reproduce the above copyright notice, - * this list of conditions and the following disclaimer in the documentation - * and/or other materials provided with the distribution. - * > Neither the name of the Hiroshima University nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +#pragma once +/** + * @file SFMT.h + * + * @brief SIMD oriented Fast Mersenne Twister(SFMT) pseudorandom + * number generator using C structure. + * + * @author Mutsuo Saito (Hiroshima University) + * @author Makoto Matsumoto (The University of Tokyo) + * + * Copyright (C) 2006, 2007 Mutsuo Saito, Makoto Matsumoto and Hiroshima + * University. + * Copyright (C) 2012 Mutsuo Saito, Makoto Matsumoto, Hiroshima + * University and The University of Tokyo. + * All rights reserved. + * + * The 3-clause BSD License is applied to this software, see + * LICENSE.txt + * + * @note We assume that your system has inttypes.h. If your system + * doesn't have inttypes.h, you have to typedef uint32_t and uint64_t, + * and you have to define PRIu64 and PRIx64 in this file as follows: + * @verbatim + typedef unsigned int uint32_t + typedef unsigned long long uint64_t + #define PRIu64 "llu" + #define PRIx64 "llx" +@endverbatim + * uint32_t must be exactly 32-bit unsigned integer type (no more, no + * less), and uint64_t must be exactly 64-bit unsigned integer type. + * PRIu64 and PRIx64 are used for printf function to print 64-bit + * unsigned int and 64-bit unsigned int in hexadecimal format. */ -#ifndef SFMT_H -#define SFMT_H - -#include <emmintrin.h> // Define SSE2 intrinsics -#include "randomc.h" // Define integer types etc -#include <time.h> -#include <new> - -// Choose one of the possible Mersenne exponents. -// Higher values give longer cycle length and use more memory: -//#define MEXP 607 -//#define MEXP 1279 -//#define MEXP 2281 -//#define MEXP 4253 - #define MEXP 11213 -//#define MEXP 19937 -//#define MEXP 44497 +#ifndef SFMTST_H +#define SFMTST_H +#if defined(__cplusplus) +extern "C" { +#endif -// Define constants for the selected Mersenne exponent: -#if MEXP == 44497 -#define SFMT_N 348 // Size of state vector -#define SFMT_M 330 // Position of intermediate feedback -#define SFMT_SL1 5 // Left shift of W[N-1], 32-bit words -#define SFMT_SL2 3 // Left shift of W[0], *8, 128-bit words -#define SFMT_SR1 9 // Right shift of W[M], 32-bit words -#define SFMT_SR2 3 // Right shift of W[N-2], *8, 128-bit words -#define SFMT_MASK 0xeffffffb,0xdfbebfff,0xbfbf7bef,0x9ffd7bff // AND mask -#define SFMT_PARITY 1,0,0xa3ac4000,0xecc1327a // Period certification vector +#include <stdio.h> +#include <assert.h> + +#if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) + #include <inttypes.h> +#elif defined(_MSC_VER) || defined(__BORLANDC__) + typedef unsigned int uint32_t; + typedef unsigned __int64 uint64_t; + #define inline __inline +#else + #include <inttypes.h> + #if defined(__GNUC__) + #define inline __inline__ + #endif +#endif -#elif MEXP == 19937 -#define SFMT_N 156 // Size of state vector -#define SFMT_M 122 // Position of intermediate feedback -#define SFMT_SL1 18 // Left shift of W[N-1], 32-bit words -#define SFMT_SL2 1 // Left shift of W[0], *8, 128-bit words -#define SFMT_SR1 11 // Right shift of W[M], 32-bit words -#define SFMT_SR2 1 // Right shift of W[N-2], *8, 128-bit words -#define SFMT_MASK 0xdfffffef,0xddfecb7f,0xbffaffff,0xbffffff6 // AND mask -#define SFMT_PARITY 1,0,0,0x13c9e684 // Period certification vector +#ifndef PRIu64 + #if defined(_MSC_VER) || defined(__BORLANDC__) + #define PRIu64 "I64u" + #define PRIx64 "I64x" + #else + #define PRIu64 "llu" + #define PRIx64 "llx" + #endif +#endif -#elif MEXP == 11213 -#define SFMT_N 88 // Size of state vector -#define SFMT_M 68 // Position of intermediate feedback -#define SFMT_SL1 14 // Left shift of W[N-1], 32-bit words -#define SFMT_SL2 3 // Left shift of W[0], *8, 128-bit words -#define SFMT_SR1 7 // Right shift of W[M], 32-bit words -#define SFMT_SR2 3 // Right shift of W[N-2], *8, 128-bit words -#define SFMT_MASK 0xeffff7fb,0xffffffef,0xdfdfbfff,0x7fffdbfd // AND mask -#define SFMT_PARITY 1,0,0xe8148000,0xd0c7afa3 // Period certification vector +#include "SFMT-params.h" + +/*------------------------------------------ + 128-bit SIMD like data type for standard C + ------------------------------------------*/ +#if defined(HAVE_ALTIVEC) + #if !defined(__APPLE__) + #include <altivec.h> + #endif +/** 128-bit data structure */ +union W128_T { + vector unsigned int s; + uint32_t u[4]; + uint64_t u64[2]; +}; +#elif defined(HAVE_NEON) + #include <arm_neon.h> + +/** 128-bit data structure */ +union W128_T { + uint32_t u[4]; + uint64_t u64[2]; + uint32x4_t si; +}; +#elif defined(HAVE_SSE2) + #include <emmintrin.h> + +/** 128-bit data structure */ +union W128_T { + uint32_t u[4]; + uint64_t u64[2]; + __m128i si; +}; +#else +/** 128-bit data structure */ +union W128_T { + uint32_t u[4]; + uint64_t u64[2]; +}; +#endif -#elif MEXP == 4253 -#define SFMT_N 34 // Size of state vector -#define SFMT_M 17 // Position of intermediate feedback -#define SFMT_SL1 20 // Left shift of W[N-1], 32-bit words -#define SFMT_SL2 1 // Left shift of W[0], *8, 128-bit words -#define SFMT_SR1 7 // Right shift of W[M], 32-bit words -#define SFMT_SR2 1 // Right shift of W[N-2], *8, 128-bit words -#define SFMT_MASK 0x9f7bffff, 0x9fffff5f, 0x3efffffb, 0xfffff7bb // AND mask -#define SFMT_PARITY 0xa8000001, 0xaf5390a3, 0xb740b3f8, 0x6c11486d // Period certification vector +/** 128-bit data type */ +typedef union W128_T w128_t; -#elif MEXP == 2281 -#define SFMT_N 18 // Size of state vector -#define SFMT_M 12 // Position of intermediate feedback -#define SFMT_SL1 19 // Left shift of W[N-1], 32-bit words -#define SFMT_SL2 1 // Left shift of W[0], *8, 128-bit words -#define SFMT_SR1 5 // Right shift of W[M], 32-bit words -#define SFMT_SR2 1 // Right shift of W[N-2], *8, 128-bit words -#define SFMT_MASK 0xbff7ffbf, 0xfdfffffe, 0xf7ffef7f, 0xf2f7cbbf // AND mask -#define SFMT_PARITY 0x00000001, 0x00000000, 0x00000000, 0x41dfa600 // Period certification vector +/** + * SFMT internal state + */ +struct SFMT_T { + /** the 128-bit internal state array */ + w128_t state[SFMT_N]; + /** index counter to the 32-bit internal state array */ + int idx; +}; -#elif MEXP == 1279 -#define SFMT_N 10 // Size of state vector -#define SFMT_M 7 // Position of intermediate feedback -#define SFMT_SL1 14 // Left shift of W[N-1], 32-bit words -#define SFMT_SL2 3 // Left shift of W[0], *8, 128-bit words -#define SFMT_SR1 5 // Right shift of W[M], 32-bit words -#define SFMT_SR2 1 // Right shift of W[N-2], *8, 128-bit words -#define SFMT_MASK 0xf7fefffd, 0x7fefcfff, 0xaff3ef3f, 0xb5ffff7f // AND mask -#define SFMT_PARITY 0x00000001, 0x00000000, 0x00000000, 0x20000000 // Period certification vector +typedef struct SFMT_T sfmt_t; + +void sfmt_fill_array32(sfmt_t * sfmt, uint32_t * array, int size); +void sfmt_fill_array64(sfmt_t * sfmt, uint64_t * array, int size); +void sfmt_init_gen_rand(sfmt_t * sfmt, uint32_t seed); +void sfmt_init_by_array(sfmt_t * sfmt, uint32_t * init_key, int key_length); +const char * sfmt_get_idstring(sfmt_t * sfmt); +int sfmt_get_min_array_size32(sfmt_t * sfmt); +int sfmt_get_min_array_size64(sfmt_t * sfmt); +void sfmt_gen_rand_all(sfmt_t * sfmt); + +#ifndef ONLY64 +/** + * This function generates and returns 32-bit pseudorandom number. + * init_gen_rand or init_by_array must be called before this function. + * @param sfmt SFMT internal state + * @return 32-bit pseudorandom number + */ +inline static uint32_t sfmt_genrand_uint32(sfmt_t * sfmt) { + uint32_t r; + uint32_t * psfmt32 = &sfmt->state[0].u[0]; -#elif MEXP == 607 -#define SFMT_N 5 // Size of state vector -#define SFMT_M 2 // Position of intermediate feedback -#define SFMT_SL1 15 // Left shift of W[N-1], 32-bit words -#define SFMT_SL2 3 // Left shift of W[0], *8, 128-bit words -#define SFMT_SR1 13 // Right shift of W[M], 32-bit words -#define SFMT_SR2 3 // Right shift of W[N-2], *8, 128-bit words -#define SFMT_MASK 0xfdff37ff, 0xef7f3f7d, 0xff777b7d, 0x7ff7fb2f // AND mask -#define SFMT_PARITY 0x00000001, 0x00000000, 0x00000000, 0x5986f054 // Period certification vector + if (sfmt->idx >= SFMT_N32) { + sfmt_gen_rand_all(sfmt); + sfmt->idx = 0; + } + r = psfmt32[sfmt->idx++]; + return r; +} #endif +/** + * This function generates and returns 64-bit pseudorandom number. + * init_gen_rand or init_by_array must be called before this function. + * The function gen_rand64 should not be called after gen_rand32, + * unless an initialization is again executed. + * @param sfmt SFMT internal state + * @return 64-bit pseudorandom number + */ +inline static uint64_t sfmt_genrand_uint64(sfmt_t * sfmt) { +#if defined(BIG_ENDIAN64) && !defined(ONLY64) + uint32_t * psfmt32 = &sfmt->state[0].u[0]; + uint32_t r1, r2; +#else + uint64_t r; +#endif + uint64_t * psfmt64 = &sfmt->state[0].u64[0]; + assert(sfmt->idx % 2 == 0); -// Functions used by SFMTRand::RandomInitByArray (UNUSED AND COMMENTED OUT) -/* -static uint32_t func1(uint32_t x) { - return (x ^ (x >> 27)) * 1664525U; + if (sfmt->idx >= SFMT_N32) { + sfmt_gen_rand_all(sfmt); + sfmt->idx = 0; + } +#if defined(BIG_ENDIAN64) && !defined(ONLY64) + r1 = psfmt32[sfmt->idx]; + r2 = psfmt32[sfmt->idx + 1]; + sfmt->idx += 2; + return ((uint64_t)r2 << 32) | r1; +#else + r = psfmt64[sfmt->idx / 2]; + sfmt->idx += 2; + return r; +#endif } -static uint32_t func2(uint32_t x) { - return (x ^ (x >> 27)) * 1566083941U; +/* ================================================= + The following real versions are due to Isaku Wada + ================================================= */ +/** + * converts an unsigned 32-bit number to a double on [0,1]-real-interval. + * @param v 32-bit unsigned integer + * @return double on [0,1]-real-interval + */ +inline static double sfmt_to_real1(uint32_t v) +{ + return v * (1.0/4294967295.0); + /* divided by 2^32-1 */ } -*/ -// Subfunction for the sfmt algorithm -static inline __m128i sfmt_recursion(__m128i const &a, __m128i const &b, -__m128i const &c, __m128i const &d, __m128i const &mask) { - __m128i a1, b1, c1, d1, z1, z2; - b1 = _mm_srli_epi32(b, SFMT_SR1); - a1 = _mm_slli_si128(a, SFMT_SL2); - c1 = _mm_srli_si128(c, SFMT_SR2); - d1 = _mm_slli_epi32(d, SFMT_SL1); - b1 = _mm_and_si128(b1, mask); - z1 = _mm_xor_si128(a, a1); - z2 = _mm_xor_si128(b1, d1); - z1 = _mm_xor_si128(z1, c1); - z2 = _mm_xor_si128(z1, z2); - return z2; +/** + * generates a random number on [0,1]-real-interval + * @param sfmt SFMT internal state + * @return double on [0,1]-real-interval + */ +inline static double sfmt_genrand_real1(sfmt_t * sfmt) +{ + return sfmt_to_real1(sfmt_genrand_uint32(sfmt)); } -namespace boost { - template <typename T> class thread_specific_ptr; +/** + * converts an unsigned 32-bit integer to a double on [0,1)-real-interval. + * @param v 32-bit unsigned integer + * @return double on [0,1)-real-interval + */ +inline static double sfmt_to_real2(uint32_t v) +{ + return v * (1.0/4294967296.0); + /* divided by 2^32 */ } -// Class for SFMT generator -class SFMTRand { // Encapsulate random number generator - friend class boost::thread_specific_ptr<SFMTRand>; - -public: - SFMTRand() - { - LastInterval = 0; - RandomInit((int)(time(0))); - } - - void RandomInit(int seed) // Re-seed - { - // Re-seed - uint32_t i; // Loop counter - uint32_t y = seed; // Temporary - uint32_t statesize = SFMT_N*4; // Size of state vector - - // Fill state vector with random numbers from seed - uint32_t* s = (uint32_t*)&state; - s[0] = y; - const uint32_t factor = 1812433253U;// Multiplication factor - - for (i = 1; i < statesize; i++) { - y = factor * (y ^ (y >> 30)) + i; - ((uint32_t*)state)[i] = y; - } - - // Further initialization and period certification - Init2(); - } - - int32_t IRandom(int32_t min, int32_t max) // Output random integer - { - // Output random integer in the interval min <= x <= max - // Slightly inaccurate if (max-min+1) is not a power of 2 - if (max <= min) { - if (max == min) return min; else return 0x80000000; - } - // Assume 64 bit integers supported. Use multiply and shift method - uint32_t interval; // Length of interval - uint64_t longran; // Random bits * interval - uint32_t iran; // Longran / 2^32 - - interval = (uint32_t)(max - min + 1); - longran = (uint64_t)BRandom() * interval; - iran = (uint32_t)(longran >> 32); - // Convert back to signed and return result - return (int32_t)iran + min; - } - - uint32_t URandom(uint32_t min, uint32_t max) - { - // Output random integer in the interval min <= x <= max - // Slightly inaccurate if (max-min+1) is not a power of 2 - if (max <= min) { - if (max == min) return min; else return 0; - } - // Assume 64 bit integers supported. Use multiply and shift method - uint32_t interval; // Length of interval - uint64_t longran; // Random bits * interval - uint32_t iran; // Longran / 2^32 - - interval = (uint32_t)(max - min + 1); - longran = (uint64_t)BRandom() * interval; - iran = (uint32_t)(longran >> 32); - // Convert back to signed and return result - return iran + min; - } - - double Random() // Output random floating point number - { - // Output random floating point number - if (ix >= SFMT_N*4-1) { - // Make sure we have at least two 32-bit numbers - Generate(); - } - uint64_t r = *(uint64_t*)((uint32_t*)state+ix); - ix += 2; - // 52 bits resolution for compatibility with assembly version: - return (int64_t)(r >> 12) * (1./(67108864.0*67108864.0)); - } - - uint32_t BRandom() // Output random bits - { - // Output 32 random bits - uint32_t y; - - if (ix >= SFMT_N*4) { - Generate(); - } - y = ((uint32_t*)state)[ix++]; - return y; - } - - void* operator new(size_t size, std::nothrow_t const&) - { - return _mm_malloc(size, 16); - } - - void operator delete(void* ptr, std::nothrow_t const&) - { - _mm_free(ptr); - } - - void* operator new(size_t size) - { - return _mm_malloc(size, 16); - } - - void operator delete(void* ptr) - { - _mm_free(ptr); - } - - void* operator new[](size_t size, std::nothrow_t const&) - { - return _mm_malloc(size, 16); - } - - void operator delete[](void* ptr, std::nothrow_t const&) - { - _mm_free(ptr); - } - - void* operator new[](size_t size) - { - return _mm_malloc(size, 16); - } +/** + * generates a random number on [0,1)-real-interval + * @param sfmt SFMT internal state + * @return double on [0,1)-real-interval + */ +inline static double sfmt_genrand_real2(sfmt_t * sfmt) +{ + return sfmt_to_real2(sfmt_genrand_uint32(sfmt)); +} - void operator delete[](void* ptr) - { - _mm_free(ptr); - } +/** + * converts an unsigned 32-bit integer to a double on (0,1)-real-interval. + * @param v 32-bit unsigned integer + * @return double on (0,1)-real-interval + */ +inline static double sfmt_to_real3(uint32_t v) +{ + return (((double)v) + 0.5)*(1.0/4294967296.0); + /* divided by 2^32 */ +} -private: - void Init2() // Various initializations and period certification - { - // Various initializations and period certification - uint32_t i, j, temp; - - // Initialize mask - static const uint32_t maskinit[4] = {SFMT_MASK}; - mask = _mm_loadu_si128((__m128i*)maskinit); +/** + * generates a random number on (0,1)-real-interval + * @param sfmt SFMT internal state + * @return double on (0,1)-real-interval + */ +inline static double sfmt_genrand_real3(sfmt_t * sfmt) +{ + return sfmt_to_real3(sfmt_genrand_uint32(sfmt)); +} - // Period certification - // Define period certification vector - static const uint32_t parityvec[4] = {SFMT_PARITY}; +/** + * converts an unsigned 32-bit integer to double on [0,1) + * with 53-bit resolution. + * @param v 32-bit unsigned integer + * @return double on [0,1)-real-interval with 53-bit resolution. + */ +inline static double sfmt_to_res53(uint64_t v) +{ + return (v >> 11) * (1.0/9007199254740992.0); +} - // Check if parityvec & state[0] has odd parity - temp = 0; - for (i = 0; i < 4; i++) - temp ^= parityvec[i] & ((uint32_t*)state)[i]; +/** + * generates a random number on [0,1) with 53-bit resolution + * @param sfmt SFMT internal state + * @return double on [0,1) with 53-bit resolution + */ +inline static double sfmt_genrand_res53(sfmt_t * sfmt) +{ + return sfmt_to_res53(sfmt_genrand_uint64(sfmt)); +} - for (i = 16; i > 0; i >>= 1) temp ^= temp >> i; - if (!(temp & 1)) { - // parity is even. Certification failed - // Find a nonzero bit in period certification vector - for (i = 0; i < 4; i++) { - if (parityvec[i]) { - for (j = 1; j; j <<= 1) { - if (parityvec[i] & j) { - // Flip the corresponding bit in state[0] to change parity - ((uint32_t*)state)[i] ^= j; - // Done. Exit i and j loops - i = 5; break; - } - } - } - } - } - // Generate first random numbers and set ix = 0 - Generate(); - } +/* ================================================= + The following function are added by Saito. + ================================================= */ +/** + * generates a random number on [0,1) with 53-bit resolution from two + * 32 bit integers + */ +inline static double sfmt_to_res53_mix(uint32_t x, uint32_t y) +{ + return sfmt_to_res53(x | ((uint64_t)y << 32)); +} - void Generate() // Fill state array with new random numbers - { - // Fill state array with new random numbers - int i; - __m128i r, r1, r2; +/** + * generates a random number on [0,1) with 53-bit resolution + * using two 32bit integers. + * @param sfmt SFMT internal state + * @return double on [0,1) with 53-bit resolution + */ +inline static double sfmt_genrand_res53_mix(sfmt_t * sfmt) +{ + uint32_t x, y; - r1 = state[SFMT_N - 2]; - r2 = state[SFMT_N - 1]; - for (i = 0; i < SFMT_N - SFMT_M; i++) { - r = sfmt_recursion(state[i], state[i + SFMT_M], r1, r2, mask); - state[i] = r; - r1 = r2; - r2 = r; - } - for (; i < SFMT_N; i++) { - r = sfmt_recursion(state[i], state[i + SFMT_M - SFMT_N], r1, r2, mask); - state[i] = r; - r1 = r2; - r2 = r; - } - ix = 0; - } + x = sfmt_genrand_uint32(sfmt); + y = sfmt_genrand_uint32(sfmt); + return sfmt_to_res53_mix(x, y); +} - __m128i mask; // AND mask - __m128i state[SFMT_N]; // State vector for SFMT generator - uint32_t ix; // Index into state array - uint32_t LastInterval; // Last interval length for IRandom - uint32_t RLimit; // Rejection limit used by IRandom -}; +#if defined(__cplusplus) +} +#endif -#endif // SFMT_H +#endif diff --git a/dep/SFMT/randomc.h b/dep/SFMT/randomc.h deleted file mode 100644 index eed2c81a205..00000000000 --- a/dep/SFMT/randomc.h +++ /dev/null @@ -1,65 +0,0 @@ -/* - * Copyright notice - * ================ - * GNU General Public License http://www.gnu.org/licenses/gpl.html - * This C++ implementation of SFMT contains parts of the original C code - * which was published under the following BSD license, which is therefore - * in effect in addition to the GNU General Public License. - * Copyright (c) 2006, 2007 by Mutsuo Saito, Makoto Matsumoto and Hiroshima University. - * Copyright (c) 2008 by Agner Fog. - * Copyright (c) 2008-2013 Trinity Core - * - * BSD License: - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are met: - * > Redistributions of source code must retain the above copyright notice, - * this list of conditions and the following disclaimer. - * > Redistributions in binary form must reproduce the above copyright notice, - * this list of conditions and the following disclaimer in the documentation - * and/or other materials provided with the distribution. - * > Neither the name of the Hiroshima University nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - */ - -#ifndef RANDOMC_H -#define RANDOMC_H - -// Define integer types with known size: int32_t, uint32_t, int64_t, uint64_t. -// If this doesn't work then insert compiler-specific definitions here: -#if defined(__GNUC__) - // Compilers supporting C99 or C++0x have inttypes.h defining these integer types - #include <inttypes.h> - #define INT64_SUPPORTED // Remove this if the compiler doesn't support 64-bit integers -#elif defined(_WIN16) || defined(__MSDOS__) || defined(_MSDOS) - // 16 bit systems use long int for 32 bit integer - typedef signed long int int32_t; - typedef unsigned long int uint32_t; -#elif defined(_MSC_VER) - // Microsoft have their own definition - typedef signed __int32 int32_t; - typedef unsigned __int32 uint32_t; - typedef signed __int64 int64_t; - typedef unsigned __int64 uint64_t; - #define INT64_SUPPORTED // Remove this if the compiler doesn't support 64-bit integers -#else - // This works with most compilers - typedef signed int int32_t; - typedef unsigned int uint32_t; - typedef long long int64_t; - typedef unsigned long long uint64_t; - #define INT64_SUPPORTED // Remove this if the compiler doesn't support 64-bit integers -#endif - -#endif // RANDOMC_H diff --git a/src/common/Utilities/Random.cpp b/src/common/Utilities/Random.cpp index 028163f7456..974f13db0be 100644 --- a/src/common/Utilities/Random.cpp +++ b/src/common/Utilities/Random.cpp @@ -17,7 +17,7 @@ #include "Random.h" #include "Errors.h" -#include "SFMT.h" +#include "SFMTRand.h" #include <boost/thread/tss.hpp> #include <random> diff --git a/src/common/Utilities/SFMTRand.cpp b/src/common/Utilities/SFMTRand.cpp new file mode 100644 index 00000000000..ddc65c8a772 --- /dev/null +++ b/src/common/Utilities/SFMTRand.cpp @@ -0,0 +1,120 @@ +/* + * This file is part of the TrinityCore Project. See AUTHORS file for Copyright information + * + * This program is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License as published by the + * Free Software Foundation; either version 2 of the License, or (at your + * option) any later version. + * + * This program 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 General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along + * with this program. If not, see <http://www.gnu.org/licenses/>. + */ + +#include "SFMTRand.h" + +#include <exception> +#include <emmintrin.h> +#include <ctime> + +SFMTRand::SFMTRand() +{ + RandomInit((uint32_t)(time(0))); +} + +void SFMTRand::RandomInit(uint32_t seed) // Re-seed +{ + sfmt_init_gen_rand(&state, seed); +} + +int32_t SFMTRand::IRandom(int32_t min, int32_t max) // Output random integer +{ + // Output random integer in the interval min <= x <= max + // Slightly inaccurate if (max-min+1) is not a power of 2 + if (max <= min) { + if (max == min) return min; else return 0x80000000; + } + // Assume 64 bit integers supported. Use multiply and shift method + uint32_t interval; // Length of interval + uint64_t longran; // Random bits * interval + uint32_t iran; // Longran / 2^32 + + interval = (uint32_t)(max - min + 1); + longran = (uint64_t)BRandom() * interval; + iran = (uint32_t)(longran >> 32); + // Convert back to signed and return result + return (int32_t)iran + min; +} + +uint32_t SFMTRand::URandom(uint32_t min, uint32_t max) +{ + // Output random integer in the interval min <= x <= max + // Slightly inaccurate if (max-min+1) is not a power of 2 + if (max <= min) { + if (max == min) return min; else return 0; + } + // Assume 64 bit integers supported. Use multiply and shift method + uint32_t interval; // Length of interval + uint64_t longran; // Random bits * interval + uint32_t iran; // Longran / 2^32 + + interval = (uint32_t)(max - min + 1); + longran = (uint64_t)BRandom() * interval; + iran = (uint32_t)(longran >> 32); + // Convert back to signed and return result + return iran + min; +} + +double SFMTRand::Random() // Output random floating point number +{ + return sfmt_genrand_real1(&state); +} + +uint32_t SFMTRand::BRandom() // Output random bits +{ + return sfmt_genrand_uint32(&state); +} + +void* SFMTRand::operator new(size_t size, std::nothrow_t const&) +{ + return _mm_malloc(size, 16); +} + +void SFMTRand::operator delete(void* ptr, std::nothrow_t const&) +{ + _mm_free(ptr); +} + +void* SFMTRand::operator new(size_t size) +{ + return _mm_malloc(size, 16); +} + +void SFMTRand::operator delete(void* ptr) +{ + _mm_free(ptr); +} + +void* SFMTRand::operator new[](size_t size, std::nothrow_t const&) +{ + return _mm_malloc(size, 16); +} + +void SFMTRand::operator delete[](void* ptr, std::nothrow_t const&) +{ + _mm_free(ptr); +} + +void* SFMTRand::operator new[](size_t size) +{ + return _mm_malloc(size, 16); +} + +void SFMTRand::operator delete[](void* ptr) +{ + _mm_free(ptr); +} diff --git a/src/common/Utilities/SFMTRand.h b/src/common/Utilities/SFMTRand.h new file mode 100644 index 00000000000..1189accae2f --- /dev/null +++ b/src/common/Utilities/SFMTRand.h @@ -0,0 +1,47 @@ +/* + * This file is part of the TrinityCore Project. See AUTHORS file for Copyright information + * + * This program is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License as published by the + * Free Software Foundation; either version 2 of the License, or (at your + * option) any later version. + * + * This program 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 General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along + * with this program. If not, see <http://www.gnu.org/licenses/>. + */ + +#ifndef SFMTRand_h__ +#define SFMTRand_h__ + +#include <SFMT.h> +#include <new> + +/* + * C++ Wrapper for SFMT + */ +class SFMTRand { +public: + SFMTRand(); + void RandomInit(uint32_t seed); // Re-seed + int32_t IRandom(int32_t min, int32_t max); // Output random integer + uint32_t URandom(uint32_t min, uint32_t max); + double Random(); // Output random floating point number + uint32_t BRandom(); // Output random bits + void* operator new(size_t size, std::nothrow_t const&); + void operator delete(void* ptr, std::nothrow_t const&); + void* operator new(size_t size); + void operator delete(void* ptr); + void* operator new[](size_t size, std::nothrow_t const&); + void operator delete[](void* ptr, std::nothrow_t const&); + void* operator new[](size_t size); + void operator delete[](void* ptr); +private: + sfmt_t state; +}; + +#endif // SFMTRand_h__ |