mirror of
https://github.com/hrydgard/ppsspp.git
synced 2024-11-23 05:19:56 +00:00
Implement sceSfmt19937 module.
This commit is contained in:
parent
77693633d7
commit
21b0c2992b
@ -1005,6 +1005,15 @@ add_library(kirk STATIC
|
||||
ext/libkirk/kirk_engine.h)
|
||||
include_directories(ext/libkirk)
|
||||
|
||||
add_library(sfmt19937 STATIC
|
||||
ext/sfmt19937/SFMT.c
|
||||
ext/sfmt19937/SFMT.h
|
||||
ext/sfmt19937/SFMT-common.h
|
||||
ext/sfmt19937/SFMT-params.h
|
||||
ext/sfmt19937/SFMT-params19937.h
|
||||
)
|
||||
include_directories(ext/sfmt19937)
|
||||
|
||||
add_library(xbrz STATIC
|
||||
ext/xbrz/xbrz.cpp
|
||||
ext/xbrz/xbrz.h
|
||||
@ -1252,6 +1261,8 @@ add_library(${CoreLibName} ${CoreLinkType}
|
||||
Core/HLE/sceRtc.h
|
||||
Core/HLE/sceSas.cpp
|
||||
Core/HLE/sceSas.h
|
||||
Core/HLE/sceSfmt19937.cpp
|
||||
Core/HLE/sceSfmt19937.h
|
||||
Core/HLE/sceSha256.cpp
|
||||
Core/HLE/sceSha256.h
|
||||
Core/HLE/sceSsl.cpp
|
||||
|
@ -169,6 +169,7 @@
|
||||
</ItemDefinitionGroup>
|
||||
<ItemGroup>
|
||||
<ClCompile Include="..\ext\disarm.cpp" />
|
||||
<ClCompile Include="..\ext\sfmt19937\SFMT.c" />
|
||||
<ClCompile Include="..\ext\snappy\snappy-c.cpp" />
|
||||
<ClCompile Include="..\ext\snappy\snappy.cpp" />
|
||||
<ClCompile Include="..\git-version.cpp" />
|
||||
@ -263,6 +264,12 @@
|
||||
<ClCompile Include="HLE\sceGameUpdate.cpp" />
|
||||
<ClCompile Include="HLE\sceRtc.cpp" />
|
||||
<ClCompile Include="HLE\sceSas.cpp" />
|
||||
<ClCompile Include="HLE\sceSfmt19937.cpp">
|
||||
<PreprocessorDefinitions Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">_XKEYCHECK_H;USING_WIN_UI;_CRT_SECURE_NO_WARNINGS;USE_FFMPEG;WIN32;_ARCH_32=1;_M_IX86=1;_DEBUG;_LIB;_UNICODE;UNICODE;%(PreprocessorDefinitions)</PreprocessorDefinitions>
|
||||
<PreprocessorDefinitions Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">_XKEYCHECK_H;USING_WIN_UI;_CRT_SECURE_NO_WARNINGS;USE_FFMPEG;WIN32;_ARCH_32=1;_M_IX86=1;_LIB;NDEBUG;_UNICODE;UNICODE;%(PreprocessorDefinitions)</PreprocessorDefinitions>
|
||||
<PreprocessorDefinitions Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">_XKEYCHECK_H;USING_WIN_UI;_CRT_SECURE_NO_WARNINGS;USE_FFMPEG;WIN32;_ARCH_64=1;_M_X64=1;_DEBUG;_LIB;_UNICODE;UNICODE;%(PreprocessorDefinitions)</PreprocessorDefinitions>
|
||||
<PreprocessorDefinitions Condition="'$(Configuration)|$(Platform)'=='Release|x64'">_XKEYCHECK_H;USING_WIN_UI;_CRT_SECURE_NO_WARNINGS;USE_FFMPEG;WIN32;_ARCH_64=1;_M_X64=1;_LIB;NDEBUG;_UNICODE;UNICODE;%(PreprocessorDefinitions)</PreprocessorDefinitions>
|
||||
</ClCompile>
|
||||
<ClCompile Include="HLE\sceSha256.cpp" />
|
||||
<ClCompile Include="HLE\sceSsl.cpp" />
|
||||
<ClCompile Include="HLE\sceUmd.cpp" />
|
||||
@ -408,6 +415,10 @@
|
||||
</ItemGroup>
|
||||
<ItemGroup>
|
||||
<ClInclude Include="..\ext\disarm.h" />
|
||||
<ClInclude Include="..\ext\sfmt19937\SFMT-common.h" />
|
||||
<ClInclude Include="..\ext\sfmt19937\SFMT-params.h" />
|
||||
<ClInclude Include="..\ext\sfmt19937\SFMT-params19937.h" />
|
||||
<ClInclude Include="..\ext\sfmt19937\SFMT.h" />
|
||||
<ClInclude Include="..\ext\snappy\snappy-internal.h" />
|
||||
<ClInclude Include="..\ext\snappy\snappy-sinksource.h" />
|
||||
<ClInclude Include="..\ext\snappy\snappy-stubs-internal.h" />
|
||||
@ -508,6 +519,7 @@
|
||||
<ClInclude Include="HLE\sceGameUpdate.h" />
|
||||
<ClInclude Include="HLE\sceRtc.h" />
|
||||
<ClInclude Include="HLE\sceSas.h" />
|
||||
<ClInclude Include="HLE\sceSfmt19937.h" />
|
||||
<ClInclude Include="HLE\sceSha256.h" />
|
||||
<ClInclude Include="HLE\sceSsl.h" />
|
||||
<ClInclude Include="HLE\sceUmd.h" />
|
||||
|
@ -55,6 +55,9 @@
|
||||
<Filter Include="Ext\udis86">
|
||||
<UniqueIdentifier>{435eb15d-386c-44df-97b4-343a1d6524ec}</UniqueIdentifier>
|
||||
</Filter>
|
||||
<Filter Include="Ext\sfmt19937">
|
||||
<UniqueIdentifier>{8f56c2fd-7706-49f7-8a1d-64dcffad9ad0}</UniqueIdentifier>
|
||||
</Filter>
|
||||
</ItemGroup>
|
||||
<ItemGroup>
|
||||
<ClCompile Include="ELF\ElfReader.cpp">
|
||||
@ -542,6 +545,12 @@
|
||||
<ClCompile Include="Util\AudioFormat.cpp">
|
||||
<Filter>Util</Filter>
|
||||
</ClCompile>
|
||||
<ClCompile Include="..\ext\sfmt19937\SFMT.c">
|
||||
<Filter>Ext\sfmt19937</Filter>
|
||||
</ClCompile>
|
||||
<ClCompile Include="HLE\sceSfmt19937.cpp">
|
||||
<Filter>HLE\Libraries</Filter>
|
||||
</ClCompile>
|
||||
</ItemGroup>
|
||||
<ItemGroup>
|
||||
<ClInclude Include="ELF\ElfReader.h">
|
||||
@ -1025,6 +1034,21 @@
|
||||
<ClInclude Include="Util\AudioFormatNEON.h">
|
||||
<Filter>Util</Filter>
|
||||
</ClInclude>
|
||||
<ClInclude Include="..\ext\sfmt19937\SFMT.h">
|
||||
<Filter>Ext\sfmt19937</Filter>
|
||||
</ClInclude>
|
||||
<ClInclude Include="..\ext\sfmt19937\SFMT-common.h">
|
||||
<Filter>Ext\sfmt19937</Filter>
|
||||
</ClInclude>
|
||||
<ClInclude Include="..\ext\sfmt19937\SFMT-params.h">
|
||||
<Filter>Ext\sfmt19937</Filter>
|
||||
</ClInclude>
|
||||
<ClInclude Include="..\ext\sfmt19937\SFMT-params19937.h">
|
||||
<Filter>Ext\sfmt19937</Filter>
|
||||
</ClInclude>
|
||||
<ClInclude Include="HLE\sceSfmt19937.h">
|
||||
<Filter>HLE\Libraries</Filter>
|
||||
</ClInclude>
|
||||
</ItemGroup>
|
||||
<ItemGroup>
|
||||
<None Include="CMakeLists.txt" />
|
||||
|
@ -804,3 +804,8 @@ template<int func(void *, u32, int)> void WrapI_VUI(){
|
||||
u32 retval = func(Memory::GetPointer(PARAM(0)), PARAM(1), PARAM(2));
|
||||
RETURN(retval);
|
||||
}
|
||||
|
||||
template<int func(u32, u64, int)> void WrapI_UU64I() {
|
||||
int retval = func(PARAM(0), PARAM64(2), PARAM(4));
|
||||
RETURN(retval);
|
||||
}
|
||||
|
@ -71,6 +71,7 @@
|
||||
#include "sceUsbGps.h"
|
||||
#include "sceSha256.h"
|
||||
#include "sceAdler.h"
|
||||
#include "sceSfmt19937.h"
|
||||
|
||||
#define N(s) s
|
||||
|
||||
@ -334,5 +335,6 @@ void RegisterAllModules() {
|
||||
Register_sceLibFttt();
|
||||
Register_sceSha256();
|
||||
Register_sceAdler();
|
||||
Register_sceSfmt19937();
|
||||
}
|
||||
|
||||
|
119
Core/HLE/sceSfmt19937.cpp
Normal file
119
Core/HLE/sceSfmt19937.cpp
Normal file
@ -0,0 +1,119 @@
|
||||
// Copyright (c) 2012- PPSSPP Project.
|
||||
|
||||
// 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, version 2.0 or later versions.
|
||||
|
||||
// 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 2.0 for more details.
|
||||
|
||||
// A copy of the GPL 2.0 should have been included with the program.
|
||||
// If not, see http://www.gnu.org/licenses/
|
||||
|
||||
// Official git repository and contact information can be found at
|
||||
// https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.
|
||||
|
||||
#include "../ext/sfmt19937/sfmt.h"
|
||||
|
||||
#include "sceSfmt19937.h"
|
||||
#include "Common/Log.h"
|
||||
#include "Core/HLE/HLE.h"
|
||||
#include "Core/HLE/FunctionWrappers.h"
|
||||
|
||||
static int sceSfmt19937InitGenRand(u32 sfmt, u32 seed) {
|
||||
if (!Memory::IsValidAddress(sfmt)) {
|
||||
ERROR_LOG(HLE, "sceSfmt19937InitGenRand(sfmt=%08x, seed=%08x) - bad address(es)", sfmt, seed);
|
||||
return -1;
|
||||
}
|
||||
INFO_LOG(HLE, "sceSfmt19937InitGenRand(sfmt=%08x, seed=%08x)", sfmt, seed);
|
||||
|
||||
sfmt_t *psfmt = (sfmt_t *)Memory::GetPointerUnchecked(sfmt);
|
||||
sfmt_init_gen_rand(psfmt, seed);
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
static int sceSfmt19937InitByArray(u32 sfmt, u32 seeds, int seedslen) {
|
||||
if (!Memory::IsValidAddress(sfmt) || !Memory::IsValidAddress(seeds) || !Memory::IsValidAddress(seeds + 4 * (seedslen - 1))) {
|
||||
ERROR_LOG(HLE, "sceSfmt19937InitByArray(sfmt=%08x, seeds=%08x, seedslen=%08x) - bad address(es)", sfmt, seeds, seedslen);
|
||||
return -1;
|
||||
}
|
||||
INFO_LOG(HLE, "sceSfmt19937InitByArray(sfmt=%08x, seeds=%08x, seedslen=%08x)", sfmt, seeds, seedslen);
|
||||
|
||||
sfmt_t *psfmt = (sfmt_t *)Memory::GetPointerUnchecked(sfmt);
|
||||
u32 *pseeds = (u32 *)Memory::GetPointerUnchecked(seeds);
|
||||
sfmt_init_by_array(psfmt, pseeds, seedslen);
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
static u32 sceSfmt19937GenRand32(u32 sfmt) {
|
||||
if (!Memory::IsValidAddress(sfmt)) {
|
||||
ERROR_LOG(HLE, "sceSfmt19937GenRand32(sfmt=%08x) - bad address(es)", sfmt);
|
||||
return -1;
|
||||
}
|
||||
INFO_LOG(HLE, "sceSfmt19937GenRand32(sfmt=%08x)", sfmt);
|
||||
|
||||
sfmt_t *psfmt = (sfmt_t *)Memory::GetPointerUnchecked(sfmt);
|
||||
u32 ret = sfmt_genrand_uint32(psfmt);
|
||||
|
||||
return ret;
|
||||
}
|
||||
|
||||
static u64 sceSfmt19937GenRand64(u32 sfmt) {
|
||||
if (!Memory::IsValidAddress(sfmt)) {
|
||||
ERROR_LOG(HLE, "sceSfmt19937GenRand64(sfmt=%08x) - bad address(es)", sfmt);
|
||||
return -1;
|
||||
}
|
||||
INFO_LOG(HLE, "sceSfmt19937GenRand64(sfmt=%08x)", sfmt);
|
||||
|
||||
sfmt_t *psfmt = (sfmt_t *)Memory::GetPointerUnchecked(sfmt);
|
||||
u64 ret = sfmt_genrand_uint64(psfmt);
|
||||
|
||||
return ret;
|
||||
}
|
||||
|
||||
static int sceSfmt19937FillArray32(u32 sfmt, u32 array, int arraylen) {
|
||||
if (!Memory::IsValidAddress(sfmt) || !Memory::IsValidAddress(array) || !Memory::IsValidAddress(array + 4 * (arraylen - 1))) {
|
||||
ERROR_LOG(HLE, "sceSfmt19937FillArray32(sfmt=%08x, ar=%08x, arlen=%08x) - bad address(es)", sfmt, array, arraylen);
|
||||
return -1;
|
||||
}
|
||||
INFO_LOG(HLE, "sceSfmt19937FillArray32(sfmt=%08x, ar=%08x, arlen=%08x)", sfmt, array, arraylen);
|
||||
|
||||
sfmt_t *psfmt = (sfmt_t *)Memory::GetPointerUnchecked(sfmt);
|
||||
u32 *parray = (u32 *)Memory::GetPointerUnchecked(array);
|
||||
sfmt_fill_array32(psfmt, parray, arraylen);
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
static int sceSfmt19937FillArray64(u32 sfmt, u64 array, int arraylen) {
|
||||
if (!Memory::IsValidAddress(sfmt) || !Memory::IsValidAddress(array) || !Memory::IsValidAddress(array + 8 * (arraylen - 1))) {
|
||||
ERROR_LOG(HLE, "sceSfmt19937FillArray64(sfmt=%08x, ar=%08x, arlen=%08x) - bad address(es)", sfmt, array, arraylen);
|
||||
return -1;
|
||||
}
|
||||
INFO_LOG(HLE, "sceSfmt19937FillArray64(sfmt=%08x, ar=%08x, arlen=%08x)", sfmt, array, arraylen);
|
||||
|
||||
sfmt_t *psfmt = (sfmt_t *)Memory::GetPointerUnchecked(sfmt);
|
||||
u64 *parray = (u64 *)Memory::GetPointerUnchecked(array);
|
||||
sfmt_fill_array64(psfmt, parray, arraylen);
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
const HLEFunction sceSfmt19937[] =
|
||||
{
|
||||
{ 0x161ACEB2, WrapI_UU<sceSfmt19937InitGenRand>, "sceSfmt19937InitGenRand" },
|
||||
{ 0xDD5A5D6C, WrapI_UUI<sceSfmt19937InitByArray>, "sceSfmt19937InitByArray" },
|
||||
{ 0xB33FE749, WrapU_U<sceSfmt19937GenRand32>, "sceSfmt19937GenRand32" },
|
||||
{ 0xD5AC9F99, WrapU64_U<sceSfmt19937GenRand64>, "sceSfmt19937GenRand64" },
|
||||
{ 0xDB025BFA, WrapI_UUI<sceSfmt19937FillArray32>, "sceSfmt19937FillArray32" },
|
||||
{ 0xEE2938C4, WrapI_UU64I<sceSfmt19937FillArray64>, "sceSfmt19937FillArray64" },
|
||||
};
|
||||
|
||||
void Register_sceSfmt19937()
|
||||
{
|
||||
RegisterModule("sceSfmt19937", ARRAY_SIZE(sceSfmt19937), sceSfmt19937);
|
||||
}
|
20
Core/HLE/sceSfmt19937.h
Normal file
20
Core/HLE/sceSfmt19937.h
Normal file
@ -0,0 +1,20 @@
|
||||
// Copyright (c) 2012- PPSSPP Project.
|
||||
|
||||
// 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, version 2.0 or later versions.
|
||||
|
||||
// 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 2.0 for more details.
|
||||
|
||||
// A copy of the GPL 2.0 should have been included with the program.
|
||||
// If not, see http://www.gnu.org/licenses/
|
||||
|
||||
// Official git repository and contact information can be found at
|
||||
// https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.
|
||||
|
||||
#pragma once
|
||||
|
||||
void Register_sceSfmt19937();
|
@ -119,6 +119,7 @@ EXEC_AND_LIB_FILES := \
|
||||
$(SRC)/ext/libkirk/bn.c \
|
||||
$(SRC)/ext/libkirk/ec.c \
|
||||
$(SRC)/ext/libkirk/kirk_engine.c \
|
||||
$(SRC)/ext/sfmt19937/SFMT.c \
|
||||
$(SRC)/ext/snappy/snappy-c.cpp \
|
||||
$(SRC)/ext/snappy/snappy.cpp \
|
||||
$(SRC)/ext/udis86/decode.c \
|
||||
@ -268,6 +269,7 @@ EXEC_AND_LIB_FILES := \
|
||||
$(SRC)/Core/HLE/sceRtc.cpp \
|
||||
$(SRC)/Core/HLE/scePsmf.cpp \
|
||||
$(SRC)/Core/HLE/sceSas.cpp \
|
||||
$(SRC)/Core/HLE/sceSfmt19937.cpp \
|
||||
$(SRC)/Core/HLE/sceSha256.cpp \
|
||||
$(SRC)/Core/HLE/sceSsl.cpp \
|
||||
$(SRC)/Core/HLE/sceUmd.cpp \
|
||||
|
164
ext/sfmt19937/SFMT-common.h
Normal file
164
ext/sfmt19937/SFMT-common.h
Normal file
@ -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
|
||||
#endif
|
||||
|
||||
#if defined(__cplusplus)
|
||||
}
|
||||
#endif
|
98
ext/sfmt19937/SFMT-params.h
Normal file
98
ext/sfmt19937/SFMT-params.h
Normal file
@ -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_SL2 * 8) bits.
|
||||
#define SFMT_SR21 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 */
|
50
ext/sfmt19937/SFMT-params19937.h
Normal file
50
ext/sfmt19937/SFMT-params19937.h
Normal file
@ -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 */
|
433
ext/sfmt19937/SFMT.c
Normal file
433
ext/sfmt19937/SFMT.c
Normal file
@ -0,0 +1,433 @@
|
||||
/**
|
||||
* @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
|
||||
|
||||
/**
|
||||
* parameters used by sse2.
|
||||
*/
|
||||
static const w128_t sse2_param_mask = {{SFMT_MSK1, SFMT_MSK2,
|
||||
SFMT_MSK3, SFMT_MSK4}};
|
||||
/*----------------
|
||||
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)
|
||||
#if defined(_MSC_VER)
|
||||
#include "SFMT-sse2-msc.h"
|
||||
#else
|
||||
#include "SFMT-sse2.h"
|
||||
#endif
|
||||
#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))
|
||||
/**
|
||||
* 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) {
|
||||
int 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)
|
||||
/**
|
||||
* 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
|
295
ext/sfmt19937/SFMT.h
Normal file
295
ext/sfmt19937/SFMT.h
Normal file
@ -0,0 +1,295 @@
|
||||
#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 SFMTST_H
|
||||
#define SFMTST_H
|
||||
#if defined(__cplusplus)
|
||||
extern "C" {
|
||||
#endif
|
||||
|
||||
#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
|
||||
|
||||
#ifndef PRIu64
|
||||
#if defined(_MSC_VER) || defined(__BORLANDC__)
|
||||
#define PRIu64 "I64u"
|
||||
#define PRIx64 "I64x"
|
||||
#else
|
||||
#define PRIu64 "llu"
|
||||
#define PRIx64 "llx"
|
||||
#endif
|
||||
#endif
|
||||
|
||||
#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_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
|
||||
|
||||
/** 128-bit data type */
|
||||
typedef union W128_T w128_t;
|
||||
|
||||
/**
|
||||
* 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;
|
||||
};
|
||||
|
||||
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];
|
||||
|
||||
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);
|
||||
|
||||
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
|
||||
}
|
||||
|
||||
/* =================================================
|
||||
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 */
|
||||
}
|
||||
|
||||
/**
|
||||
* 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));
|
||||
}
|
||||
|
||||
/**
|
||||
* 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 */
|
||||
}
|
||||
|
||||
/**
|
||||
* 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));
|
||||
}
|
||||
|
||||
/**
|
||||
* 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 */
|
||||
}
|
||||
|
||||
/**
|
||||
* 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));
|
||||
}
|
||||
|
||||
/**
|
||||
* 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 * (1.0/18446744073709551616.0);
|
||||
}
|
||||
|
||||
/**
|
||||
* 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));
|
||||
}
|
||||
|
||||
|
||||
/* =================================================
|
||||
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));
|
||||
}
|
||||
|
||||
/**
|
||||
* 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;
|
||||
|
||||
x = sfmt_genrand_uint32(sfmt);
|
||||
y = sfmt_genrand_uint32(sfmt);
|
||||
return sfmt_to_res53_mix(x, y);
|
||||
}
|
||||
|
||||
#if defined(__cplusplus)
|
||||
}
|
||||
#endif
|
||||
|
||||
#endif
|
Loading…
Reference in New Issue
Block a user