Initial import of dSFMT

src/shogun/lib/external/dSFMT/dSFMT-common.h

@@ -0,0 +1,115 @@
+#pragma once
+/**
+ * @file dSFMT-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 DSFMT_COMMON_H
+#define DSFMT_COMMON_H
+
+#include "dSFMT.h"
+
+#if defined(HAVE_SSE2)
+#  include <emmintrin.h>
+union X128I_T {
+    uint64_t u[2];
+    __m128i  i128;
+};
+union X128D_T {
+    double d[2];
+    __m128d d128;
+};
+/** mask data for sse2 */
+static const union X128I_T sse2_param_mask = {{DSFMT_MSK1, DSFMT_MSK2}};
+#endif
+
+#if defined(HAVE_ALTIVEC)
+inline static void do_recursion(w128_t *r, w128_t *a, w128_t * b,
+				w128_t *lung) {
+    const vector unsigned char sl1 = ALTI_SL1;
+    const vector unsigned char sl1_perm = ALTI_SL1_PERM;
+    const vector unsigned int sl1_msk = ALTI_SL1_MSK;
+    const vector unsigned char sr1 = ALTI_SR;
+    const vector unsigned char sr1_perm = ALTI_SR_PERM;
+    const vector unsigned int sr1_msk = ALTI_SR_MSK;
+    const vector unsigned char perm = ALTI_PERM;
+    const vector unsigned int msk1 = ALTI_MSK;
+    vector unsigned int w, x, y, z;
+
+    z = a->s;
+    w = lung->s;
+    x = vec_perm(w, (vector unsigned int)perm, perm);
+    y = vec_perm(z, (vector unsigned int)sl1_perm, sl1_perm);
+    y = vec_sll(y, sl1);
+    y = vec_and(y, sl1_msk);
+    w = vec_xor(x, b->s);
+    w = vec_xor(w, y);
+    x = vec_perm(w, (vector unsigned int)sr1_perm, sr1_perm);
+    x = vec_srl(x, sr1);
+    x = vec_and(x, sr1_msk);
+    y = vec_and(w, msk1);
+    z = vec_xor(z, y);
+    r->s = vec_xor(z, x);
+    lung->s = w;
+}
+#elif defined(HAVE_SSE2)
+/**
+ * This function represents the recursion formula.
+ * @param r output 128-bit
+ * @param a a 128-bit part of the internal state array
+ * @param b a 128-bit part of the internal state array
+ * @param d a 128-bit part of the internal state array (I/O)
+ */
+inline static void do_recursion(w128_t *r, w128_t *a, w128_t *b, w128_t *u) {
+    __m128i v, w, x, y, z;
+
+    x = a->si;
+    z = _mm_slli_epi64(x, DSFMT_SL1);
+    y = _mm_shuffle_epi32(u->si, SSE2_SHUFF);
+    z = _mm_xor_si128(z, b->si);
+    y = _mm_xor_si128(y, z);
+
+    v = _mm_srli_epi64(y, DSFMT_SR);
+    w = _mm_and_si128(y, sse2_param_mask.i128);
+    v = _mm_xor_si128(v, x);
+    v = _mm_xor_si128(v, w);
+    r->si = v;
+    u->si = y;
+}
+#else
+/**
+ * This function represents the recursion formula.
+ * @param r output 128-bit
+ * @param a a 128-bit part of the internal state array
+ * @param b a 128-bit part of the internal state array
+ * @param lung a 128-bit part of the internal state array (I/O)
+ */
+inline static void do_recursion(w128_t *r, w128_t *a, w128_t * b,
+				w128_t *lung) {
+    uint64_t t0, t1, L0, L1;
+
+    t0 = a->u[0];
+    t1 = a->u[1];
+    L0 = lung->u[0];
+    L1 = lung->u[1];
+    lung->u[0] = (t0 << DSFMT_SL1) ^ (L1 >> 32) ^ (L1 << 32) ^ b->u[0];
+    lung->u[1] = (t1 << DSFMT_SL1) ^ (L0 >> 32) ^ (L0 << 32) ^ b->u[1];
+    r->u[0] = (lung->u[0] >> DSFMT_SR) ^ (lung->u[0] & DSFMT_MSK1) ^ t0;
+    r->u[1] = (lung->u[1] >> DSFMT_SR) ^ (lung->u[1] & DSFMT_MSK2) ^ t1;
+}
+#endif
+#endif

src/shogun/lib/external/dSFMT/dSFMT-params.h

@@ -0,0 +1,87 @@
+#ifndef DSFMT_PARAMS_H
+#define DSFMT_PARAMS_H
+
+#include "dSFMT.h"
+
+/*----------------------
+  the parameters of DSFMT
+  following definitions are in dSFMT-paramsXXXX.h file.
+  ----------------------*/
+/** the pick up position of the array.
+#define DSFMT_POS1 122 
+*/
+
+/** the parameter of shift left as four 32-bit registers.
+#define DSFMT_SL1 18
+ */
+
+/** the parameter of shift right as four 32-bit registers.
+#define DSFMT_SR1 12
+*/
+
+/** A bitmask, used in the recursion.  These parameters are introduced
+ * to break symmetry of SIMD.
+#define DSFMT_MSK1 (uint64_t)0xdfffffefULL
+#define DSFMT_MSK2 (uint64_t)0xddfecb7fULL
+*/
+
+/** These definitions are part of a 128-bit period certification vector.
+#define DSFMT_PCV1	UINT64_C(0x00000001)
+#define DSFMT_PCV2	UINT64_C(0x00000000)
+*/
+
+#define DSFMT_LOW_MASK  UINT64_C(0x000FFFFFFFFFFFFF)
+#define DSFMT_HIGH_CONST UINT64_C(0x3FF0000000000000)
+#define DSFMT_SR	12
+
+/* for sse2 */
+#if defined(HAVE_SSE2)
+  #define SSE2_SHUFF 0x1b
+#elif defined(HAVE_ALTIVEC)
+  #if defined(__APPLE__)  /* For OSX */
+    #define ALTI_SR (vector unsigned char)(4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4)
+    #define ALTI_SR_PERM \
+        (vector unsigned char)(15,0,1,2,3,4,5,6,15,8,9,10,11,12,13,14)
+    #define ALTI_SR_MSK \
+        (vector unsigned int)(0x000fffffU,0xffffffffU,0x000fffffU,0xffffffffU)
+    #define ALTI_PERM \
+        (vector unsigned char)(12,13,14,15,8,9,10,11,4,5,6,7,0,1,2,3)
+  #else
+    #define ALTI_SR      {4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4}
+    #define ALTI_SR_PERM {15,0,1,2,3,4,5,6,15,8,9,10,11,12,13,14}
+    #define ALTI_SR_MSK  {0x000fffffU,0xffffffffU,0x000fffffU,0xffffffffU}
+    #define ALTI_PERM    {12,13,14,15,8,9,10,11,4,5,6,7,0,1,2,3}
+  #endif
+#endif
+
+#if DSFMT_MEXP == 521
+  #include "dSFMT-params521.h"
+#elif DSFMT_MEXP == 1279
+  #include "dSFMT-params1279.h"
+#elif DSFMT_MEXP == 2203
+  #include "dSFMT-params2203.h"
+#elif DSFMT_MEXP == 4253
+  #include "dSFMT-params4253.h"
+#elif DSFMT_MEXP == 11213
+  #include "dSFMT-params11213.h"
+#elif DSFMT_MEXP == 19937
+  #include "dSFMT-params19937.h"
+#elif DSFMT_MEXP == 44497
+  #include "dSFMT-params44497.h"
+#elif DSFMT_MEXP == 86243
+  #include "dSFMT-params86243.h"
+#elif DSFMT_MEXP == 132049
+  #include "dSFMT-params132049.h"
+#elif DSFMT_MEXP == 216091
+  #include "dSFMT-params216091.h"
+#else
+#ifdef __GNUC__
+  #error "DSFMT_MEXP is not valid."
+  #undef DSFMT_MEXP
+#else
+  #undef DSFMT_MEXP
+#endif
+
+#endif
+
+#endif /* DSFMT_PARAMS_H */

src/shogun/lib/external/dSFMT/dSFMT-params11213.h

@@ -0,0 +1,40 @@
+#ifndef DSFMT_PARAMS11213_H
+#define DSFMT_PARAMS11213_H
+
+/* #define DSFMT_N	107 */
+/* #define DSFMT_MAXDEGREE	11256 */
+#define DSFMT_POS1	37
+#define DSFMT_SL1	19
+#define DSFMT_MSK1	UINT64_C(0x000ffffffdf7fffd)
+#define DSFMT_MSK2	UINT64_C(0x000dfffffff6bfff)
+#define DSFMT_MSK32_1	0x000fffffU
+#define DSFMT_MSK32_2	0xfdf7fffdU
+#define DSFMT_MSK32_3	0x000dffffU
+#define DSFMT_MSK32_4	0xfff6bfffU
+#define DSFMT_FIX1	UINT64_C(0xd0ef7b7c75b06793)
+#define DSFMT_FIX2	UINT64_C(0x9c50ff4caae0a641)
+#define DSFMT_PCV1	UINT64_C(0x8234c51207c80000)
+#define DSFMT_PCV2	UINT64_C(0x0000000000000001)
+#define DSFMT_IDSTR	"dSFMT2-11213:37-19:ffffffdf7fffd-dfffffff6bfff"
+
+
+/* PARAMETERS FOR ALTIVEC */
+#if defined(__APPLE__)	/* For OSX */
+    #define ALTI_SL1 	(vector unsigned char)(3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3)
+    #define ALTI_SL1_PERM \
+	(vector unsigned char)(2,3,4,5,6,7,30,30,10,11,12,13,14,15,0,1)
+    #define ALTI_SL1_MSK \
+	(vector unsigned int)(0xffffffffU,0xfff80000U,0xffffffffU,0xfff80000U)
+    #define ALTI_MSK	(vector unsigned int)(DSFMT_MSK32_1, \
+			DSFMT_MSK32_2, DSFMT_MSK32_3, DSFMT_MSK32_4)
+#else	/* For OTHER OSs(Linux?) */
+    #define ALTI_SL1 	{3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3}
+    #define ALTI_SL1_PERM \
+	{2,3,4,5,6,7,30,30,10,11,12,13,14,15,0,1}
+    #define ALTI_SL1_MSK \
+	{0xffffffffU,0xfff80000U,0xffffffffU,0xfff80000U}
+    #define ALTI_MSK \
+	{DSFMT_MSK32_1, DSFMT_MSK32_2, DSFMT_MSK32_3, DSFMT_MSK32_4}
+#endif
+
+#endif /* DSFMT_PARAMS11213_H */

src/shogun/lib/external/dSFMT/dSFMT-params1279.h

@@ -0,0 +1,40 @@
+#ifndef DSFMT_PARAMS1279_H
+#define DSFMT_PARAMS1279_H
+
+/* #define DSFMT_N	12 */
+/* #define DSFMT_MAXDEGREE	1376 */
+#define DSFMT_POS1	9
+#define DSFMT_SL1	19
+#define DSFMT_MSK1	UINT64_C(0x000efff7ffddffee)
+#define DSFMT_MSK2	UINT64_C(0x000fbffffff77fff)
+#define DSFMT_MSK32_1	0x000efff7U
+#define DSFMT_MSK32_2	0xffddffeeU
+#define DSFMT_MSK32_3	0x000fbfffU
+#define DSFMT_MSK32_4	0xfff77fffU
+#define DSFMT_FIX1	UINT64_C(0xb66627623d1a31be)
+#define DSFMT_FIX2	UINT64_C(0x04b6c51147b6109b)
+#define DSFMT_PCV1	UINT64_C(0x7049f2da382a6aeb)
+#define DSFMT_PCV2	UINT64_C(0xde4ca84a40000001)
+#define DSFMT_IDSTR	"dSFMT2-1279:9-19:efff7ffddffee-fbffffff77fff"
+
+
+/* PARAMETERS FOR ALTIVEC */
+#if defined(__APPLE__)	/* For OSX */
+    #define ALTI_SL1 	(vector unsigned char)(3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3)
+    #define ALTI_SL1_PERM \
+	(vector unsigned char)(2,3,4,5,6,7,30,30,10,11,12,13,14,15,0,1)
+    #define ALTI_SL1_MSK \
+	(vector unsigned int)(0xffffffffU,0xfff80000U,0xffffffffU,0xfff80000U)
+    #define ALTI_MSK	(vector unsigned int)(DSFMT_MSK32_1, \
+			DSFMT_MSK32_2, DSFMT_MSK32_3, DSFMT_MSK32_4)
+#else	/* For OTHER OSs(Linux?) */
+    #define ALTI_SL1 	{3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3}
+    #define ALTI_SL1_PERM \
+	{2,3,4,5,6,7,30,30,10,11,12,13,14,15,0,1}
+    #define ALTI_SL1_MSK \
+	{0xffffffffU,0xfff80000U,0xffffffffU,0xfff80000U}
+    #define ALTI_MSK \
+	{DSFMT_MSK32_1, DSFMT_MSK32_2, DSFMT_MSK32_3, DSFMT_MSK32_4}
+#endif
+
+#endif /* DSFMT_PARAMS1279_H */

src/shogun/lib/external/dSFMT/dSFMT-params132049.h

@@ -0,0 +1,40 @@
+#ifndef DSFMT_PARAMS132049_H
+#define DSFMT_PARAMS132049_H
+
+/* #define DSFMT_N	1269 */
+/* #define DSFMT_MAXDEGREE	132104 */
+#define DSFMT_POS1	371
+#define DSFMT_SL1	23
+#define DSFMT_MSK1	UINT64_C(0x000fb9f4eff4bf77)
+#define DSFMT_MSK2	UINT64_C(0x000fffffbfefff37)
+#define DSFMT_MSK32_1	0x000fb9f4U
+#define DSFMT_MSK32_2	0xeff4bf77U
+#define DSFMT_MSK32_3	0x000fffffU
+#define DSFMT_MSK32_4	0xbfefff37U
+#define DSFMT_FIX1	UINT64_C(0x4ce24c0e4e234f3b)
+#define DSFMT_FIX2	UINT64_C(0x62612409b5665c2d)
+#define DSFMT_PCV1	UINT64_C(0x181232889145d000)
+#define DSFMT_PCV2	UINT64_C(0x0000000000000001)
+#define DSFMT_IDSTR	"dSFMT2-132049:371-23:fb9f4eff4bf77-fffffbfefff37"
+
+
+/* PARAMETERS FOR ALTIVEC */
+#if defined(__APPLE__)	/* For OSX */
+    #define ALTI_SL1 	(vector unsigned char)(7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7)
+    #define ALTI_SL1_PERM \
+	(vector unsigned char)(2,3,4,5,6,7,30,30,10,11,12,13,14,15,0,1)
+    #define ALTI_SL1_MSK \
+	(vector unsigned int)(0xffffffffU,0xff800000U,0xffffffffU,0xff800000U)
+    #define ALTI_MSK	(vector unsigned int)(DSFMT_MSK32_1, \
+			DSFMT_MSK32_2, DSFMT_MSK32_3, DSFMT_MSK32_4)
+#else	/* For OTHER OSs(Linux?) */
+    #define ALTI_SL1 	{7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7}
+    #define ALTI_SL1_PERM \
+	{2,3,4,5,6,7,30,30,10,11,12,13,14,15,0,1}
+    #define ALTI_SL1_MSK \
+	{0xffffffffU,0xff800000U,0xffffffffU,0xff800000U}
+    #define ALTI_MSK \
+	{DSFMT_MSK32_1, DSFMT_MSK32_2, DSFMT_MSK32_3, DSFMT_MSK32_4}
+#endif
+
+#endif /* DSFMT_PARAMS132049_H */

src/shogun/lib/external/dSFMT/dSFMT-params19937.h

@@ -0,0 +1,40 @@
+#ifndef DSFMT_PARAMS19937_H
+#define DSFMT_PARAMS19937_H
+
+/* #define DSFMT_N	191 */
+/* #define DSFMT_MAXDEGREE	19992 */
+#define DSFMT_POS1	117
+#define DSFMT_SL1	19
+#define DSFMT_MSK1	UINT64_C(0x000ffafffffffb3f)
+#define DSFMT_MSK2	UINT64_C(0x000ffdfffc90fffd)
+#define DSFMT_MSK32_1	0x000ffaffU
+#define DSFMT_MSK32_2	0xfffffb3fU
+#define DSFMT_MSK32_3	0x000ffdffU
+#define DSFMT_MSK32_4	0xfc90fffdU
+#define DSFMT_FIX1	UINT64_C(0x90014964b32f4329)
+#define DSFMT_FIX2	UINT64_C(0x3b8d12ac548a7c7a)
+#define DSFMT_PCV1	UINT64_C(0x3d84e1ac0dc82880)
+#define DSFMT_PCV2	UINT64_C(0x0000000000000001)
+#define DSFMT_IDSTR	"dSFMT2-19937:117-19:ffafffffffb3f-ffdfffc90fffd"
+
+
+/* PARAMETERS FOR ALTIVEC */
+#if defined(__APPLE__)	/* For OSX */
+    #define ALTI_SL1 	(vector unsigned char)(3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3)
+    #define ALTI_SL1_PERM \
+	(vector unsigned char)(2,3,4,5,6,7,30,30,10,11,12,13,14,15,0,1)
+    #define ALTI_SL1_MSK \
+	(vector unsigned int)(0xffffffffU,0xfff80000U,0xffffffffU,0xfff80000U)
+    #define ALTI_MSK	(vector unsigned int)(DSFMT_MSK32_1, \
+			DSFMT_MSK32_2, DSFMT_MSK32_3, DSFMT_MSK32_4)
+#else	/* For OTHER OSs(Linux?) */
+    #define ALTI_SL1 	{3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3}
+    #define ALTI_SL1_PERM \
+	{2,3,4,5,6,7,30,30,10,11,12,13,14,15,0,1}
+    #define ALTI_SL1_MSK \
+	{0xffffffffU,0xfff80000U,0xffffffffU,0xfff80000U}
+    #define ALTI_MSK \
+	{DSFMT_MSK32_1, DSFMT_MSK32_2, DSFMT_MSK32_3, DSFMT_MSK32_4}
+#endif
+
+#endif /* DSFMT_PARAMS19937_H */