Mercurial > audlegacy

changeset 4232:704607c1f858

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
1st attempt to integrate dithering and RG engine
author Eugene Zagidullin <e.asphyx@gmail.com>
date Wed, 30 Jan 2008 01:22:37 +0300
parents f6625617716b
children 74c6f3d3cf1d
files src/Makefile src/audacious/input.c src/audacious/main.c src/audacious/main.h src/audacious/plugin.h src/audacious/ui_preferences.c src/libSAD/Makefile src/libSAD/SFMT-alti.h src/libSAD/SFMT-params.h src/libSAD/SFMT-params19937.h src/libSAD/SFMT-sse2.h src/libSAD/SFMT.c src/libSAD/SFMT.h src/libSAD/dither.c src/libSAD/dither.h src/libSAD/dither_ops.c src/libSAD/dither_ops.h src/libSAD/libSAD.h src/libSAD/noicegen.c src/libSAD/noisegen.h
diffstat 20 files changed, 2301 insertions(+), 2 deletions(-) [+]
line wrap: on
line diff
--- a/src/Makefile	Mon Jan 28 02:50:54 2008 +0100
+++ b/src/Makefile	Wed Jan 30 01:22:37 2008 +0300
@@ -1,4 +1,4 @@
-SUBDIRS = libguess audacious libid3tag
+SUBDIRS = libguess audacious libid3tag libSAD
 
 include ../extra.mk
 
--- a/src/audacious/input.c	Mon Jan 28 02:50:54 2008 +0100
+++ b/src/audacious/input.c	Wed Jan 30 01:22:37 2008 +0300
@@ -250,6 +250,8 @@
                 right[i] = (*ptr16++);
             }
         break;
+    default:
+        AUDDBG("Incorrect sample format!\n");
     }
 }
 
--- a/src/audacious/main.c	Mon Jan 28 02:50:54 2008 +0100
+++ b/src/audacious/main.c	Wed Jan 30 01:22:37 2008 +0300
@@ -230,6 +230,7 @@
     .warn_about_broken_gtk_engines = TRUE,           /* warn about broken gtk themes */
     FALSE,          /* disable inline themes */
     TRUE,           /* remember jtf text entry */
+    16,             /* output bit depth */
 };
 
 typedef struct bmp_cfg_boolent_t {
@@ -378,6 +379,7 @@
     {"colorize_r", &cfg.colorize_r, TRUE},
     {"colorize_g", &cfg.colorize_g, TRUE},
     {"colorize_b", &cfg.colorize_b, TRUE},
+    {"output_bit_depth", &cfg.output_bit_depth, TRUE},
 };
 
 static gint ncfgient = G_N_ELEMENTS(bmp_numents);
--- a/src/audacious/main.h	Mon Jan 28 02:50:54 2008 +0100
+++ b/src/audacious/main.h	Wed Jan 30 01:22:37 2008 +0300
@@ -142,6 +142,7 @@
     gboolean warn_about_broken_gtk_engines;
     gboolean disable_inline_gtk;
     gboolean remember_jtf_entry;
+    gint output_bit_depth;
 };
 
 typedef struct _BmpConfig BmpConfig;
--- a/src/audacious/plugin.h	Mon Jan 28 02:50:54 2008 +0100
+++ b/src/audacious/plugin.h	Wed Jan 30 01:22:37 2008 +0300
@@ -62,12 +62,30 @@
 typedef enum {
     FMT_U8,
     FMT_S8,
+
     FMT_U16_LE,
     FMT_U16_BE,
     FMT_U16_NE,
     FMT_S16_LE,
     FMT_S16_BE,
-    FMT_S16_NE
+    FMT_S16_NE,
+
+    /* added in Audacious 1.5 */
+    FMT_U24_LE, /* stored in lower 3 bytes of 32-bit value, highest byte must be always set to 0 --asphyx */
+    FMT_U24_BE,
+    FMT_U24_NE,
+    FMT_S24_LE,
+    FMT_S24_BE,
+    FMT_S24_NE,
+
+    FMT_U32_LE,
+    FMT_U32_BE,
+    FMT_U32_NE,
+    FMT_S32_LE,
+    FMT_S32_BE,
+    FMT_S32_NE,
+
+    FMT_FLOAT,
 } AFormat;
 
 typedef enum {
--- a/src/audacious/ui_preferences.c	Mon Jan 28 02:50:54 2008 +0100
+++ b/src/audacious/ui_preferences.c	Wed Jan 30 01:22:37 2008 +0300
@@ -188,6 +188,7 @@
 static void mainwin_font_set_cb();
 static void playlist_font_set_cb();
 GtkWidget *ui_preferences_chardet_table_populate(void);
+static GtkWidget *ui_preferences_bit_depth(void);
 
 static PreferencesWidget appearance_misc_widgets[] = {
     {WIDGET_LABEL, N_("<b>_Fonts</b>"), NULL, NULL, NULL, FALSE},
@@ -215,6 +216,8 @@
         N_("When checked, Audacious will detect file formats on demand. This can result in a messier playlist, but delivers a major speed benefit."), FALSE},
     {WIDGET_CHK_BTN, N_("Detect file formats by extension."), &cfg.use_extension_probing, NULL,
         N_("When checked, Audacious will detect file formats based by extension. Only files with extensions of supported formats will be loaded."), FALSE},
+    {WIDGET_LABEL, N_("<b>Bit Depth</b>"), NULL, NULL, NULL, FALSE},
+    {WIDGET_CUSTOM, NULL, NULL, NULL, NULL, TRUE, ui_preferences_bit_depth},
 };
 
 static PreferencesWidget playback_page_widgets[] = {
@@ -1631,6 +1634,38 @@
     return widget;
 }
 
+static void
+on_bit_depth_cbox_changed(GtkWidget *cbox, gpointer data)
+{
+    gint active = gtk_combo_box_get_active(GTK_COMBO_BOX(cbox));
+    cfg.output_bit_depth = (active == 1) ? 24 : 16;
+}
+
+GtkWidget *
+ui_preferences_bit_depth(void)
+{
+    GtkWidget *box = gtk_hbox_new(FALSE, 10);
+    GtkWidget *label = gtk_label_new(_("Output bit depth:"));
+    gtk_misc_set_alignment(GTK_MISC(label), 1, 0.5);
+    gtk_box_pack_start(GTK_BOX(box), label, FALSE, FALSE, 0);
+
+    GtkWidget *combo = gtk_combo_box_new_text ();
+    gtk_combo_box_append_text(GTK_COMBO_BOX (combo), "16");
+    gtk_combo_box_append_text(GTK_COMBO_BOX (combo), "24");
+    gtk_box_pack_start(GTK_BOX(box), combo, FALSE, FALSE, 0);
+    
+    gint active = (cfg.output_bit_depth == 24) ? 1 : 0;
+    gtk_combo_box_set_active(GTK_COMBO_BOX(combo), active);
+    g_signal_connect(combo, "changed", G_CALLBACK(on_bit_depth_cbox_changed), NULL);
+    
+    gtk_tooltips_set_tip (tooltips, box,
+                          _("All streams will be converted to this bit depth.\n"
+                          "This should be the max supported bit depth of\nthe sound card or output plugin."),
+                          NULL);
+
+    return box;
+}
+
 /* it's at early stage */
 static void
 create_widgets(GtkBox *box, PreferencesWidget *widgets, gint amt)
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/libSAD/Makefile	Wed Jan 30 01:22:37 2008 +0300
@@ -0,0 +1,7 @@
+STATIC_LIB_NOINST = libSAD.a
+SRCS = dither.c \
+       dither_ops.c \
+       SFMT.c \
+       noicegen.c
+
+include ../../buildsys.mk
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/libSAD/SFMT-alti.h	Wed Jan 30 01:22:37 2008 +0300
@@ -0,0 +1,156 @@
+/** 
+ * @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)
+    ALWAYSINLINE;
+
+/**
+ * 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 = ALTI_SL1;
+    const vector unsigned int sr1 = ALTI_SR1;
+#ifdef ONLY64
+    const vector unsigned int mask = ALTI_MSK64;
+    const vector unsigned char perm_sl = ALTI_SL2_PERM64;
+    const vector unsigned char perm_sr = ALTI_SR2_PERM64;
+#else
+    const vector unsigned int mask = ALTI_MSK;
+    const vector unsigned char perm_sl = ALTI_SL2_PERM;
+    const vector unsigned char perm_sr = 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.
+ */
+inline static void gen_rand_all(void) {
+    int i;
+    vector unsigned int r, r1, r2;
+
+    r1 = sfmt[N - 2].s;
+    r2 = sfmt[N - 1].s;
+    for (i = 0; i < N - POS1; i++) {
+	r = vec_recursion(sfmt[i].s, sfmt[i + POS1].s, r1, r2);
+	sfmt[i].s = r;
+	r1 = r2;
+	r2 = r;
+    }
+    for (; i < N; i++) {
+	r = vec_recursion(sfmt[i].s, sfmt[i + POS1 - N].s, r1, r2);
+	sfmt[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(w128_t *array, int size) {
+    int i, j;
+    vector unsigned int r, r1, r2;
+
+    r1 = sfmt[N - 2].s;
+    r2 = sfmt[N - 1].s;
+    for (i = 0; i < N - POS1; i++) {
+	r = vec_recursion(sfmt[i].s, sfmt[i + POS1].s, r1, r2);
+	array[i].s = r;
+	r1 = r2;
+	r2 = r;
+    }
+    for (; i < N; i++) {
+	r = vec_recursion(sfmt[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[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[j++].s = r;
+	r1 = r2;
+	r2 = r;
+    }
+}
+
+#ifndef ONLY64
+#if defined(__APPLE__)
+#define ALTI_SWAP (vector unsigned char) \
+	(4, 5, 6, 7, 0, 1, 2, 3, 12, 13, 14, 15, 8, 9, 10, 11)
+#else
+#define 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 = ALTI_SWAP;
+
+    for (i = 0; i < size; i++) {
+	array[i].s = vec_perm(array[i].s, (vector unsigned int)perm, perm);
+    }
+}
+#endif
+
+#endif
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/libSAD/SFMT-params.h	Wed Jan 30 01:22:37 2008 +0300
@@ -0,0 +1,97 @@
+#ifndef SFMT_PARAMS_H
+#define SFMT_PARAMS_H
+
+#if !defined(MEXP)
+#ifdef __GNUC__
+  #warning "MEXP is not defined. I assume MEXP is 19937."
+#endif
+  #define MEXP 19937
+#endif
+/*-----------------
+  BASIC DEFINITIONS
+  -----------------*/
+/** Mersenne Exponent. The period of the sequence 
+ *  is a multiple of 2^MEXP-1.
+ * #define MEXP 19937 */
+/** SFMT generator has an internal state array of 128-bit integers,
+ * and N is its size. */
+#define N (MEXP / 128 + 1)
+/** N32 is the size of internal state array when regarded as an array
+ * of 32-bit integers.*/
+#define N32 (N * 4)
+/** N64 is the size of internal state array when regarded as an array
+ * of 64-bit integers.*/
+#define N64 (N * 2)
+
+/*----------------------
+  the parameters of SFMT
+  following definitions are in paramsXXXX.h file.
+  ----------------------*/
+/** the pick up position of the array.
+#define POS1 122 
+*/
+
+/** the parameter of shift left as four 32-bit registers.
+#define SL1 18
+ */
+
+/** the parameter of shift left as one 128-bit register. 
+ * The 128-bit integer is shifted by (SL2 * 8) bits. 
+#define SL2 1 
+*/
+
+/** the parameter of shift right as four 32-bit registers.
+#define SR1 11
+*/
+
+/** the parameter of shift right as one 128-bit register. 
+ * The 128-bit integer is shifted by (SL2 * 8) bits. 
+#define SR2 1 
+*/
+
+/** A bitmask, used in the recursion.  These parameters are introduced
+ * to break symmetry of SIMD.
+#define MSK1 0xdfffffefU
+#define MSK2 0xddfecb7fU
+#define MSK3 0xbffaffffU
+#define MSK4 0xbffffff6U 
+*/
+
+/** These definitions are part of a 128-bit period certification vector.
+#define PARITY1	0x00000001U
+#define PARITY2	0x00000000U
+#define PARITY3	0x00000000U
+#define PARITY4	0xc98e126aU
+*/
+
+#if MEXP == 607
+  #include "SFMT-params607.h"
+#elif MEXP == 1279
+  #include "SFMT-params1279.h"
+#elif MEXP == 2281
+  #include "SFMT-params2281.h"
+#elif MEXP == 4253
+  #include "SFMT-params4253.h"
+#elif MEXP == 11213
+  #include "SFMT-params11213.h"
+#elif MEXP == 19937
+  #include "SFMT-params19937.h"
+#elif MEXP == 44497
+  #include "SFMT-params44497.h"
+#elif MEXP == 86243
+  #include "SFMT-params86243.h"
+#elif MEXP == 132049
+  #include "SFMT-params132049.h"
+#elif MEXP == 216091
+  #include "SFMT-params216091.h"
+#else
+#ifdef __GNUC__
+  #error "MEXP is not valid."
+  #undef MEXP
+#else
+  #undef MEXP
+#endif
+
+#endif
+
+#endif /* SFMT_PARAMS_H */
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/libSAD/SFMT-params19937.h	Wed Jan 30 01:22:37 2008 +0300
@@ -0,0 +1,46 @@
+#ifndef SFMT_PARAMS19937_H
+#define SFMT_PARAMS19937_H
+
+#define POS1	122
+#define SL1	18
+#define SL2	1
+#define SR1	11
+#define SR2	1
+#define MSK1	0xdfffffefU
+#define MSK2	0xddfecb7fU
+#define MSK3	0xbffaffffU
+#define MSK4	0xbffffff6U
+#define PARITY1	0x00000001U
+#define PARITY2	0x00000000U
+#define PARITY3	0x00000000U
+#define PARITY4	0x13c9e684U
+
+
+/* PARAMETERS FOR ALTIVEC */
+#if defined(__APPLE__)	/* For OSX */
+    #define ALTI_SL1	(vector unsigned int)(SL1, SL1, SL1, SL1)
+    #define ALTI_SR1	(vector unsigned int)(SR1, SR1, SR1, SR1)
+    #define ALTI_MSK	(vector unsigned int)(MSK1, MSK2, MSK3, MSK4)
+    #define ALTI_MSK64 \
+	(vector unsigned int)(MSK2, MSK1, MSK4, MSK3)
+    #define ALTI_SL2_PERM \
+	(vector unsigned char)(1,2,3,23,5,6,7,0,9,10,11,4,13,14,15,8)
+    #define ALTI_SL2_PERM64 \
+	(vector unsigned char)(1,2,3,4,5,6,7,31,9,10,11,12,13,14,15,0)
+    #define ALTI_SR2_PERM \
+	(vector unsigned char)(7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14)
+    #define 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 ALTI_SL1	{SL1, SL1, SL1, SL1}
+    #define ALTI_SR1	{SR1, SR1, SR1, SR1}
+    #define ALTI_MSK	{MSK1, MSK2, MSK3, MSK4}
+    #define ALTI_MSK64	{MSK2, MSK1, MSK4, MSK3}
+    #define ALTI_SL2_PERM	{1,2,3,23,5,6,7,0,9,10,11,4,13,14,15,8}
+    #define ALTI_SL2_PERM64	{1,2,3,4,5,6,7,31,9,10,11,12,13,14,15,0}
+    #define ALTI_SR2_PERM	{7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14}
+    #define ALTI_SR2_PERM64	{15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14}
+#endif	/* For OSX */
+#define IDSTR	"SFMT-19937:122-18-1-11-1:dfffffef-ddfecb7f-bffaffff-bffffff6"
+
+#endif /* SFMT_PARAMS19937_H */
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/libSAD/SFMT-sse2.h	Wed Jan 30 01:22:37 2008 +0300
@@ -0,0 +1,121 @@
+/** 
+ * @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 __m128i mm_recursion(__m128i *a, __m128i *b, __m128i c,
+				   __m128i d, __m128i mask) ALWAYSINLINE;
+
+/**
+ * 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
+ * @param mask 128-bit mask
+ * @return output
+ */
+inline static __m128i mm_recursion(__m128i *a, __m128i *b, 
+				   __m128i c, __m128i d, __m128i mask) {
+    __m128i v, x, y, z;
+    
+    x = _mm_load_si128(a);
+    y = _mm_srli_epi32(*b, SR1);
+    z = _mm_srli_si128(c, SR2);
+    v = _mm_slli_epi32(d, SL1);
+    z = _mm_xor_si128(z, x);
+    z = _mm_xor_si128(z, v);
+    x = _mm_slli_si128(x, SL2);
+    y = _mm_and_si128(y, mask);
+    z = _mm_xor_si128(z, x);
+    z = _mm_xor_si128(z, y);
+    return z;
+}
+
+/**
+ * This function fills the internal state array with pseudorandom
+ * integers.
+ */
+inline static void gen_rand_all(void) {
+    int i;
+    __m128i r, r1, r2, mask;
+    mask = _mm_set_epi32(MSK4, MSK3, MSK2, MSK1);
+
+    r1 = _mm_load_si128(&sfmt[N - 2].si);
+    r2 = _mm_load_si128(&sfmt[N - 1].si);
+    for (i = 0; i < N - POS1; i++) {
+	r = mm_recursion(&sfmt[i].si, &sfmt[i + POS1].si, r1, r2, mask);
+	_mm_store_si128(&sfmt[i].si, r);
+	r1 = r2;
+	r2 = r;
+    }
+    for (; i < N; i++) {
+	r = mm_recursion(&sfmt[i].si, &sfmt[i + POS1 - N].si, r1, r2, mask);
+	_mm_store_si128(&sfmt[i].si, 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(w128_t *array, int size) {
+    int i, j;
+    __m128i r, r1, r2, mask;
+    mask = _mm_set_epi32(MSK4, MSK3, MSK2, MSK1);
+
+    r1 = _mm_load_si128(&sfmt[N - 2].si);
+    r2 = _mm_load_si128(&sfmt[N - 1].si);
+    for (i = 0; i < N - POS1; i++) {
+	r = mm_recursion(&sfmt[i].si, &sfmt[i + POS1].si, r1, r2, mask);
+	_mm_store_si128(&array[i].si, r);
+	r1 = r2;
+	r2 = r;
+    }
+    for (; i < N; i++) {
+	r = mm_recursion(&sfmt[i].si, &array[i + POS1 - N].si, r1, r2, mask);
+	_mm_store_si128(&array[i].si, r);
+	r1 = r2;
+	r2 = r;
+    }
+    /* main loop */
+    for (; i < size - N; i++) {
+	r = mm_recursion(&array[i - N].si, &array[i + POS1 - N].si, r1, r2,
+			 mask);
+	_mm_store_si128(&array[i].si, r);
+	r1 = r2;
+	r2 = r;
+    }
+    for (j = 0; j < 2 * N - size; j++) {
+	r = _mm_load_si128(&array[j + size - N].si);
+	_mm_store_si128(&sfmt[j].si, r);
+    }
+    for (; i < size; i++) {
+	r = mm_recursion(&array[i - N].si, &array[i + POS1 - N].si, r1, r2,
+			 mask);
+	_mm_store_si128(&array[i].si, r);
+	_mm_store_si128(&sfmt[j++].si, r);
+	r1 = r2;
+	r2 = r;
+    }
+}
+
+#endif
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/libSAD/SFMT.c	Wed Jan 30 01:22:37 2008 +0300
@@ -0,0 +1,618 @@
+/** 
+ * @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. All rights reserved.
+ *
+ * The new BSD License is applied to this software, see LICENSE.txt
+ */
+#include <string.h>
+#include <assert.h>
+#include "SFMT.h"
+#include "SFMT-params.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
+/*------------------------------------------------------
+  128-bit SIMD data type for Altivec, SSE2 or 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];
+};
+/** 128-bit data type */
+typedef union W128_T w128_t;
+
+#elif defined(HAVE_SSE2)
+  #include <emmintrin.h>
+
+/** 128-bit data structure */
+union W128_T {
+    __m128i si;
+    uint32_t u[4];
+};
+/** 128-bit data type */
+typedef union W128_T w128_t;
+
+#else
+
+/** 128-bit data structure */
+struct W128_T {
+    uint32_t u[4];
+};
+/** 128-bit data type */
+typedef struct W128_T w128_t;
+
+#endif
+
+/*--------------------------------------
+  FILE GLOBAL VARIABLES
+  internal state, index counter and flag 
+  --------------------------------------*/
+/** the 128-bit internal state array */
+static w128_t sfmt[N];
+/** the 32bit integer pointer to the 128-bit internal state array */
+static uint32_t *psfmt32 = &sfmt[0].u[0];
+#if !defined(BIG_ENDIAN64) || defined(ONLY64)
+/** the 64bit integer pointer to the 128-bit internal state array */
+static uint64_t *psfmt64 = (uint64_t *)&sfmt[0].u[0];
+#endif
+/** index counter to the 32-bit internal state array */
+static int idx;
+/** a flag: it is 0 if and only if the internal state is not yet
+ * initialized. */
+static int initialized = 0;
+/** a parity check vector which certificate the period of 2^{MEXP} */
+static uint32_t parity[4] = {PARITY1, PARITY2, PARITY3, PARITY4};
+
+/*----------------
+  STATIC FUNCTIONS
+  ----------------*/
+inline static int idxof(int i);
+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);
+inline static void gen_rand_all(void);
+inline static void gen_rand_array(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(void);
+#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)
+  #include "SFMT-sse2.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
+/**
+ * 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, SL2);
+    rshift128(&y, c, SR2);
+    r->u[0] = a->u[0] ^ x.u[0] ^ ((b->u[0] >> SR1) & MSK2) ^ y.u[0] 
+	^ (d->u[0] << SL1);
+    r->u[1] = a->u[1] ^ x.u[1] ^ ((b->u[1] >> SR1) & MSK1) ^ y.u[1] 
+	^ (d->u[1] << SL1);
+    r->u[2] = a->u[2] ^ x.u[2] ^ ((b->u[2] >> SR1) & MSK4) ^ y.u[2] 
+	^ (d->u[2] << SL1);
+    r->u[3] = a->u[3] ^ x.u[3] ^ ((b->u[3] >> SR1) & MSK3) ^ y.u[3] 
+	^ (d->u[3] << 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, SL2);
+    rshift128(&y, c, SR2);
+    r->u[0] = a->u[0] ^ x.u[0] ^ ((b->u[0] >> SR1) & MSK1) ^ y.u[0] 
+	^ (d->u[0] << SL1);
+    r->u[1] = a->u[1] ^ x.u[1] ^ ((b->u[1] >> SR1) & MSK2) ^ y.u[1] 
+	^ (d->u[1] << SL1);
+    r->u[2] = a->u[2] ^ x.u[2] ^ ((b->u[2] >> SR1) & MSK3) ^ y.u[2] 
+	^ (d->u[2] << SL1);
+    r->u[3] = a->u[3] ^ x.u[3] ^ ((b->u[3] >> SR1) & MSK4) ^ y.u[3] 
+	^ (d->u[3] << SL1);
+}
+#endif
+
+#if (!defined(HAVE_ALTIVEC)) && (!defined(HAVE_SSE2))
+/**
+ * This function fills the internal state array with pseudorandom
+ * integers.
+ */
+inline static void gen_rand_all(void) {
+    int i;
+    w128_t *r1, *r2;
+
+    r1 = &sfmt[N - 2];
+    r2 = &sfmt[N - 1];
+    for (i = 0; i < N - POS1; i++) {
+	do_recursion(&sfmt[i], &sfmt[i], &sfmt[i + POS1], r1, r2);
+	r1 = r2;
+	r2 = &sfmt[i];
+    }
+    for (; i < N; i++) {
+	do_recursion(&sfmt[i], &sfmt[i], &sfmt[i + POS1 - N], r1, r2);
+	r1 = r2;
+	r2 = &sfmt[i];
+    }
+}
+
+/**
+ * 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 pseudorandom numbers to be generated.
+ */
+inline static void gen_rand_array(w128_t *array, int size) {
+    int i, j;
+    w128_t *r1, *r2;
+
+    r1 = &sfmt[N - 2];
+    r2 = &sfmt[N - 1];
+    for (i = 0; i < N - POS1; i++) {
+	do_recursion(&array[i], &sfmt[i], &sfmt[i + POS1], r1, r2);
+	r1 = r2;
+	r2 = &array[i];
+    }
+    for (; i < N; i++) {
+	do_recursion(&array[i], &sfmt[i], &array[i + POS1 - N], r1, r2);
+	r1 = r2;
+	r2 = &array[i];
+    }
+    for (; i < size - N; i++) {
+	do_recursion(&array[i], &array[i - N], &array[i + POS1 - N], r1, r2);
+	r1 = r2;
+	r2 = &array[i];
+    }
+    for (j = 0; j < 2 * N - size; j++) {
+	sfmt[j] = array[j + size - N];
+    }
+    for (; i < size; i++, j++) {
+	do_recursion(&array[i], &array[i - N], &array[i + POS1 - N], r1, r2);
+	r1 = r2;
+	r2 = &array[i];
+	sfmt[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}
+ */
+static void period_certification(void) {
+    int inner = 0;
+    int i, j;
+    uint32_t work;
+
+    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
+  ----------------*/
+/**
+ * This function returns the identification string.
+ * The string shows the word size, the Mersenne exponent,
+ * and all parameters of this generator.
+ */
+const char *get_idstring(void) {
+    return IDSTR;
+}
+
+/**
+ * This function returns the minimum size of array used for \b
+ * fill_array32() function.
+ * @return minimum size of array used for fill_array32() function.
+ */
+int get_min_array_size32(void) {
+    return N32;
+}
+
+/**
+ * This function returns the minimum size of array used for \b
+ * fill_array64() function.
+ * @return minimum size of array used for fill_array64() function.
+ */
+int get_min_array_size64(void) {
+    return N64;
+}
+
+#ifndef ONLY64
+/**
+ * This function generates and returns 32-bit pseudorandom number.
+ * init_gen_rand or init_by_array must be called before this function.
+ * @return 32-bit pseudorandom number
+ */
+uint32_t gen_rand32(void) {
+    uint32_t r;
+
+    assert(initialized);
+    if (idx >= N32) {
+	gen_rand_all();
+	idx = 0;
+    }
+    r = psfmt32[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. 
+ * @return 64-bit pseudorandom number
+ */
+uint64_t gen_rand64(void) {
+#if defined(BIG_ENDIAN64) && !defined(ONLY64)
+    uint32_t r1, r2;
+#else
+    uint64_t r;
+#endif
+
+    assert(initialized);
+    assert(idx % 2 == 0);
+
+    if (idx >= N32) {
+	gen_rand_all();
+	idx = 0;
+    }
+#if defined(BIG_ENDIAN64) && !defined(ONLY64)
+    r1 = psfmt32[idx];
+    r2 = psfmt32[idx + 1];
+    idx += 2;
+    return ((uint64_t)r2 << 32) | r1;
+#else
+    r = psfmt64[idx / 2];
+    idx += 2;
+    return r;
+#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 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 fill_array32(uint32_t *array, int size) {
+    assert(initialized);
+    assert(idx == N32);
+    assert(size % 4 == 0);
+    assert(size >= N32);
+
+    gen_rand_array((w128_t *)array, size / 4);
+    idx = 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.
+ *
+ * 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 fill_array64(uint64_t *array, int size) {
+    assert(initialized);
+    assert(idx == N32);
+    assert(size % 2 == 0);
+    assert(size >= N64);
+
+    gen_rand_array((w128_t *)array, size / 2);
+    idx = 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 seed a 32-bit integer used as the seed.
+ */
+void init_gen_rand(uint32_t seed) {
+    int i;
+
+    psfmt32[idxof(0)] = seed;
+    for (i = 1; i < N32; i++) {
+	psfmt32[idxof(i)] = 1812433253UL * (psfmt32[idxof(i - 1)] 
+					    ^ (psfmt32[idxof(i - 1)] >> 30))
+	    + i;
+    }
+    idx = N32;
+    period_certification();
+    initialized = 1;
+}
+
+/**
+ * This function initializes the internal state array,
+ * with an array of 32-bit integers used as the seeds
+ * @param init_key the array of 32-bit integers, used as a seed.
+ * @param key_length the length of init_key.
+ */
+void init_by_array(uint32_t *init_key, int key_length) {
+    int i, j, count;
+    uint32_t r;
+    int lag;
+    int mid;
+    int size = N * 4;
+
+    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));
+    if (key_length + 1 > N32) {
+	count = key_length + 1;
+    } else {
+	count = N32;
+    }
+    r = func1(psfmt32[idxof(0)] ^ psfmt32[idxof(mid)] 
+	      ^ psfmt32[idxof(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) % N32)] 
+		  ^ psfmt32[idxof((i + N32 - 1) % N32)]);
+	psfmt32[idxof((i + mid) % N32)] += r;
+	r += init_key[j] + i;
+	psfmt32[idxof((i + mid + lag) % N32)] += r;
+	psfmt32[idxof(i)] = r;
+	i = (i + 1) % N32;
+    }
+    for (; j < count; j++) {
+	r = func1(psfmt32[idxof(i)] ^ psfmt32[idxof((i + mid) % N32)] 
+		  ^ psfmt32[idxof((i + N32 - 1) % N32)]);
+	psfmt32[idxof((i + mid) % N32)] += r;
+	r += i;
+	psfmt32[idxof((i + mid + lag) % N32)] += r;
+	psfmt32[idxof(i)] = r;
+	i = (i + 1) % N32;
+    }
+    for (j = 0; j < N32; j++) {
+	r = func2(psfmt32[idxof(i)] + psfmt32[idxof((i + mid) % N32)] 
+		  + psfmt32[idxof((i + N32 - 1) % N32)]);
+	psfmt32[idxof((i + mid) % N32)] ^= r;
+	r -= i;
+	psfmt32[idxof((i + mid + lag) % N32)] ^= r;
+	psfmt32[idxof(i)] = r;
+	i = (i + 1) % N32;
+    }
+
+    idx = N32;
+    period_certification();
+    initialized = 1;
+}
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/libSAD/SFMT.h	Wed Jan 30 01:22:37 2008 +0300
@@ -0,0 +1,125 @@
+/** 
+ * @file SFMT.h 
+ *
+ * @brief SIMD oriented Fast Mersenne Twister(SFMT) pseudorandom
+ * number generator
+ *
+ * @author Mutsuo Saito (Hiroshima University)
+ * @author Makoto Matsumoto (Hiroshima University)
+ *
+ * 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
+ *
+ * @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 <stdio.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
+
+#if defined(__GNUC__)
+#define ALWAYSINLINE __attribute__((always_inline))
+#endif
+
+uint32_t gen_rand32(void);
+uint64_t gen_rand64(void);
+void fill_array32(uint32_t *array, int size);
+void fill_array64(uint64_t *array, int size);
+void init_gen_rand(uint32_t seed);
+void init_by_array(uint32_t *init_key, int key_length);
+const char *get_idstring(void);
+int get_min_array_size32(void);
+int get_min_array_size64(void);
+
+/* These real versions are due to Isaku Wada */
+/** generates a random number on [0,1]-real-interval */
+inline static double 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 */
+inline static double genrand_real1(void)
+{
+    return to_real1(gen_rand32());
+}
+
+/** generates a random number on [0,1)-real-interval */
+inline static double to_real2(uint32_t v)
+{
+    return v * (1.0/4294967296.0); 
+    /* divided by 2^32 */
+}
+
+/** generates a random number on [0,1)-real-interval */
+inline static double genrand_real2(void)
+{
+    return to_real2(gen_rand32());
+}
+
+/** generates a random number on (0,1)-real-interval */
+inline static double 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 */
+inline static double genrand_real3(void)
+{
+    return to_real3(gen_rand32());
+}
+/** These real versions are due to Isaku Wada */
+
+/** generates a random number on [0,1) with 53-bit resolution*/
+inline static double to_res53(uint64_t v) 
+{ 
+    return v * (1.0/18446744073709551616.0L);
+}
+
+/** generates a random number on [0,1) with 53-bit resolution*/
+inline static double genrand_res53(void) 
+{ 
+    return to_res53(gen_rand64());
+} 
+#endif
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/libSAD/dither.c	Wed Jan 30 01:22:37 2008 +0300
@@ -0,0 +1,492 @@
+/* Scale & Dither library (libSAD)
+ * High-precision bit depth converter with ReplayGain support
+ *
+ * (c)2007 by Eugene Zagidullin (e.asphyx@gmail.com)
+ * 
+ * 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, write to the Free Software 
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+
+#include "common.h"
+#include "buffer.h"
+#include "noisegen.h"
+
+#include <assert.h>
+#include <math.h>
+
+/* 
+ * Supported conversions:
+ *
+ *					O U T P U T
+ *   ,------------------+-----------------------------------------------.
+ *   |			|S8 U8 S16 U16 S24 U24 S32 U32 FLOAT FIXED-POINT|
+ *   +------------------+-----------------------------------------------+
+ *   | S8		|X  X  X   X   X   X   X   X   -     -          |
+ *   | U8		|X  X  X   X   X   X   X   X   -     -          |
+ * I | S16		|X  X  X   X   X   X   X   X   -     -          |
+ * N | U16		|X  X  X   X   X   X   X   X   -     -          |
+ * P | S24		|X  X  X   X   X   X   X   X   -     -          |
+ * U | U24		|X  X  X   X   X   X   X   X   -     -          |
+ * T | S32		|X  X  X   X   X   X   X   X   -     -          |
+ *   | U32		|X  X  X   X   X   X   X   X   -     -          |
+ *   | FLOAT		|X  X  X   X   X   X   X   X   X     -          |
+ *   | FIXED-POINT	|X  X  X   X   X   X   X   X   X     -          |
+ *   `------------------+-----------------------------------------------'
+ */
+
+#define SCALE(x,s) (s != 1.0 ? x * s : x)
+#define MAXINT(a) (1L << ((a)-1))
+#define CLIP(x,m) (x > m-1 ? m-1 : (x < -m ? -m : x))
+
+/* private object */
+typedef struct {
+  SAD_sample_format input_sample_format;
+  SAD_sample_format output_sample_format;
+  int input_bits;
+  int input_fracbits;
+  int output_bits;
+  int output_fracbits;
+  int channels;
+  SAD_channels_order input_chorder;
+  SAD_channels_order output_chorder;
+  SAD_get_sample_proc get_sample;
+  SAD_put_sample_proc put_sample;
+  int dither;
+  int hardlimit;
+  float scale;
+  float rg_scale;
+} SAD_state_priv;
+
+/* error code */
+
+//static SAD_error SAD_last_error = SAD_ERROR_OK;
+
+static inline double compute_hardlimit (double sample, float scale) {
+  sample *= scale;
+  const double k = 0.5;    /* -6dBFS */
+  if (sample > k) {
+    return tanh((sample - k) / (1 - k)) * (1 - k) + k;
+  }
+  else if (sample < -k) {
+    return tanh((sample + k) / (1 - k)) * (1 - k) - k;
+  }
+  return sample;
+}
+
+/* 
+ * Dither fixed-point normalized or integer sample to n-bits integer
+ * samples < -1 and > 1 will be clipped
+ */
+
+static inline sad_sint32 __dither_sample_fixed_to_int (sad_sint32 sample, int inbits, int fracbits, int outbits, float scale, int dither,
+							int hardlimit)
+{
+  int n_bits_to_loose, bitwidth, precision_loss;
+  sad_sint32 maxint = MAXINT(outbits);
+
+  n_bits_to_loose = 0;
+  bitwidth = inbits;
+  precision_loss = FALSE;
+
+/*#ifdef DEEP_DEBUG
+  printf("f: __dither_sample_fixed_to_int\n");
+#endif*/
+
+  if (fracbits == 0) {
+    if (inbits<29) {
+      /* convert to 4.28 fixed-point */
+      n_bits_to_loose = 29 - inbits;
+      sample <<= n_bits_to_loose;
+      bitwidth += n_bits_to_loose;
+    }
+
+    n_bits_to_loose += inbits - outbits;
+
+    if (inbits > outbits) {
+      precision_loss = TRUE;
+#ifdef PRECISION_DEBUG
+      printf("Precision loss, reason: bitwidth loss %d --> %d\n", inbits, outbits);
+#endif
+    }
+  } else {
+    n_bits_to_loose = fracbits + 1 - outbits;
+    bitwidth = fracbits;
+    precision_loss = TRUE;
+#ifdef PRECISION_DEBUG
+    printf("Precision loss, reason: fixed-point input\n", inbits, outbits);
+#endif
+  }
+  
+  assert(n_bits_to_loose >=0 );
+
+  if (hardlimit) {
+    sample = (sad_sint32)(compute_hardlimit((double)sample/(double)MAXINT(bitwidth), scale) * (double)MAXINT(bitwidth));
+#ifdef PRECISION_DEBUG
+    printf("Precision loss, reason: hard limiter\n", inbits, outbits);
+#endif
+    precision_loss = TRUE;
+  } else {
+    sample = SCALE(sample, scale);
+  }
+ 
+  if (scale != 1.0){
+    precision_loss = TRUE;
+#ifdef PRECISION_DEBUG
+    printf("Precision loss, reason: scale\n", inbits, outbits);
+#endif
+  }
+
+  if (precision_loss && (n_bits_to_loose >= 1)) sample += (1L << (n_bits_to_loose - 1));
+
+#ifdef DITHER_DEBUG
+  sad_sint32 val_wo_dither = sample, >> n_bits_to_loose;
+  val_wo_dither = CLIP(val_wo_dither, maxint);
+#endif
+  if (dither && precision_loss && (n_bits_to_loose >= 1)) {
+    sad_sint32 dither_num = triangular_dither_noise(n_bits_to_loose + 1);
+    sample += dither_num;
+  }
+
+  sample >>= n_bits_to_loose;
+
+  /* Clipping */
+#ifdef CLIPPING_DEBUG
+  sad_sint32 val_wo_clip = sample;
+#endif
+  sample = CLIP(sample, maxint);
+#ifdef CLIPPING_DEBUG
+  if (val_wo_clip != sample) {
+    printf("Clipping: %d --> %d\n", val_wo_clip, sample);
+  }
+#endif
+#ifdef DITHER_DEBUG
+  if (dither && precision_loss && (n_bits_to_loose >= 1)) printf("%d --> %d, noise: %d\n", val_wo_dither, sample, sample - val_wo_dither);
+#endif
+  return sample;
+}
+
+/* 
+ * Dither floating-point normalized sample to n-bits integer
+ * samples < -1 and > 1 will be clipped
+ */
+static inline sad_sint32 __dither_sample_float_to_int (float sample, int nbits, float scale, int dither, int hardlimit) {
+
+#ifdef DEEP_DEBUG
+  printf("f: __dither_sample_float_to_int\n");
+#endif
+
+  sad_sint32 maxint = MAXINT(nbits);
+
+  if (hardlimit) {
+    sample = compute_hardlimit((double)sample, scale);
+  } else {
+    sample = SCALE(sample, scale);
+  }
+
+  sample *= maxint;
+  /* we want to round precisely */
+  sample = (sample < 0 ? sample - 0.5 : sample + 0.5);
+
+#ifdef DITHER_DEBUG
+  sad_sint32 val_wo_dither = (sad_sint32) sample;
+  val_wo_dither = CLIP(val_wo_dither, maxint);
+#endif
+  if (dither) {
+    float dither_num = triangular_dither_noise_f();
+    sample += dither_num;
+  }
+
+  /* Round and clipping */
+  sad_sint32 value = (sad_sint32) sample;
+#ifdef CLIPPING_DEBUG
+  sad_sint32 val_wo_clip = value;
+#endif
+  value = CLIP(value, maxint);
+#ifdef CLIPPING_DEBUG
+  if (val_wo_clip != value) {
+    printf("Clipping: %d --> %d\n", val_wo_clip, value);
+  }
+#endif
+
+#ifdef DITHER_DEBUG
+  printf("%d --> %d, noise: %d\n", val_wo_dither, value, value - val_wo_dither);
+#endif
+  return value;
+}
+
+static inline float __dither_sample_float_to_float (float sample, float scale, int hardlimit) {
+#ifdef DEEP_DEBUG
+  printf("f: __dither_sample_float_to_float\n");
+#endif
+  if (hardlimit) {
+    sample = compute_hardlimit((double)sample, scale);
+  } else {
+    sample = SCALE(sample, scale);
+  }
+  return sample;
+}
+
+static inline float __dither_sample_fixed_to_float (sad_sint32 sample, int inbits, int fracbits, float scale, int hardlimit) {
+  float fsample;
+
+#ifdef DEEP_DEBUG
+  printf("f: __dither_sample_fixed_to_float\n");
+#endif
+  if (fracbits == 0) {
+     fsample = (float)sample / (float)MAXINT(inbits);
+  } else {
+     fsample = (float)sample / (float)MAXINT(fracbits+1);
+  }
+  return __dither_sample_float_to_float (fsample, scale, hardlimit);
+}
+
+
+
+
+
+SAD_dither_t* SAD_dither_init(SAD_buffer_format *inbuf_format, SAD_buffer_format *outbuf_format, int *error) {
+  SAD_state_priv *priv;
+
+  DEBUG_MSG("f: SAD_dither_init\n",0);
+
+  priv = calloc(sizeof(SAD_state_priv), 1);
+
+  /* Check buffer formats and assign buffer ops */
+  SAD_buffer_ops* inops = SAD_assign_buf_ops(inbuf_format);
+
+  if (inbuf_format->sample_format != SAD_SAMPLE_FLOAT) {
+    if (inops != NULL) {
+      priv->get_sample = inops->get_sample;
+    } else {
+      free(priv);
+      *error = SAD_ERROR_INCORRECT_INPUT_SAMPLEFORMAT;
+      return NULL;
+    }
+  }
+
+  SAD_buffer_ops* outops = SAD_assign_buf_ops(outbuf_format);
+
+  if (outbuf_format->sample_format != SAD_SAMPLE_FLOAT) {
+    if (outops != NULL) {
+      priv->put_sample = outops->put_sample;
+    } else {
+      free(priv);
+      *error = SAD_ERROR_INCORRECT_OUTPUT_SAMPLEFORMAT;
+      return NULL;
+    }
+  }
+
+  priv->input_fracbits = 0;
+  priv->output_fracbits = 0;
+  priv->input_sample_format = inbuf_format->sample_format;
+  priv->output_sample_format = outbuf_format->sample_format;
+  priv->input_chorder = inbuf_format->channels_order;
+  priv->output_chorder = outbuf_format->channels_order;
+  priv->channels = inbuf_format->channels;
+  priv->scale = 1.0;
+  priv->rg_scale = 1.0;
+  priv->dither = TRUE;
+  priv->hardlimit = FALSE;
+
+  switch(outbuf_format->sample_format){
+    case SAD_SAMPLE_S8:
+    case SAD_SAMPLE_U8: priv->output_bits = 8; break;
+    case SAD_SAMPLE_S16:
+    case SAD_SAMPLE_U16: priv->output_bits = 16; break;
+    case SAD_SAMPLE_S24:
+    case SAD_SAMPLE_U24: priv->output_bits = 24; break;
+    case SAD_SAMPLE_S32:
+    case SAD_SAMPLE_U32: priv->output_bits = 32; break;
+    case SAD_SAMPLE_FLOAT: break;
+    default:
+      free(priv);
+      *error = SAD_ERROR_INCORRECT_OUTPUT_SAMPLEFORMAT;
+      return NULL;
+  }
+
+  switch(inbuf_format->sample_format){
+    case SAD_SAMPLE_S8:
+    case SAD_SAMPLE_U8: priv->input_bits = 8; break;
+    case SAD_SAMPLE_S16:
+    case SAD_SAMPLE_U16: priv->input_bits = 16; break;
+    case SAD_SAMPLE_S24:
+    case SAD_SAMPLE_U24: priv->input_bits = 24; break;
+    case SAD_SAMPLE_S32:
+    case SAD_SAMPLE_U32: priv->input_bits = 32; break;
+    case SAD_SAMPLE_FIXED32: priv->input_fracbits = inbuf_format->fracbits; break;
+    case SAD_SAMPLE_FLOAT: break;
+    default:
+      free(priv);
+      *error = SAD_ERROR_INCORRECT_INPUT_SAMPLEFORMAT;
+      return NULL;
+  }
+
+  *error = SAD_ERROR_OK;
+  return (SAD_dither_t*)priv;
+}
+
+int SAD_dither_free(SAD_dither_t* state) {
+  DEBUG_MSG("f: SAD_dither_free\n",0);
+  free(state);
+  return SAD_ERROR_OK;
+}
+
+/* 
+ * Depend on format->channels_order inbuf and outbuf will be treated as
+ * smth* or smth** if channels_order = SAD_CHORDER_INTERLEAVED or SAD_CHORDER_SEPARATED
+ * accordingly
+ *
+ * frame is aggregate of format->channels samples
+ */
+
+#define GET_FLOAT_SAMPLE(b,o,n,c,i) (o == SAD_CHORDER_INTERLEAVED ? (((float*)b)[i*n+c]) : (((float**)b)[c][i]))
+#define PUT_FLOAT_SAMPLE(b,o,n,c,i,s) { \
+    if (o == SAD_CHORDER_INTERLEAVED) { \
+      ((float*)b)[i*n+c] = s;		\
+    } else {				\
+      ((float**)b)[c][i] = s;		\
+    }					\
+  }
+
+int SAD_dither_process_buffer (SAD_dither_t *state, void *inbuf, void *outbuf, int frames)
+{
+  SAD_state_priv *priv = (SAD_state_priv*) state;
+  int i, ch;
+  int channels = priv->channels;
+  int inbits = priv->input_bits;
+  int outbits = priv->output_bits;
+  int fracbits = priv->input_fracbits;
+  float scale = priv->scale * priv->rg_scale;
+  int dither = priv->dither;
+  int hardlimit = priv->hardlimit;
+  SAD_channels_order input_chorder = priv->input_chorder;
+  SAD_channels_order output_chorder = priv->output_chorder;
+
+  SAD_get_sample_proc get_sample = priv->get_sample;
+  SAD_put_sample_proc put_sample = priv->put_sample;
+
+#ifdef DEEP_DEBUG
+  printf("f: SAD_process_buffer\n");
+#endif
+
+  if (priv->input_sample_format == SAD_SAMPLE_FLOAT) {
+      if (priv->output_sample_format == SAD_SAMPLE_FLOAT) {
+          /* process buffer */
+          for(i=0; i<frames; i++) {
+	      for(ch=0; ch<channels; ch++) {
+	          float sample = GET_FLOAT_SAMPLE(inbuf, input_chorder, channels, ch ,i);
+	          sample = __dither_sample_float_to_float(sample, scale, hardlimit);
+                  PUT_FLOAT_SAMPLE(outbuf, output_chorder, channels, ch ,i, sample);
+	      }
+	  }
+      } else {
+          if (put_sample == NULL) return SAD_ERROR_CORRUPTED_PRIVATE_DATA;
+          /* process buffer */
+          for(i=0; i<frames; i++) {
+	      for(ch=0; ch<channels; ch++) {
+	          float sample = GET_FLOAT_SAMPLE(inbuf, input_chorder, channels, ch ,i);
+	          sad_sint32 isample = __dither_sample_float_to_int(sample, outbits, scale, dither, hardlimit);
+                  put_sample (outbuf, isample, channels, ch, i);
+	      }
+	  }
+      }
+  } else {
+      if (priv->output_sample_format == SAD_SAMPLE_FLOAT) {
+          if (get_sample == NULL) return SAD_ERROR_CORRUPTED_PRIVATE_DATA;
+          /* process buffer */
+          for(i=0; i<frames; i++) {
+	      for(ch=0; ch<channels; ch++) {
+  	          sad_sint32 sample = get_sample (inbuf, channels, ch, i);
+                  float fsample = __dither_sample_fixed_to_float (sample, inbits, fracbits, scale, hardlimit);
+                  PUT_FLOAT_SAMPLE(outbuf, output_chorder, channels, ch ,i, fsample);
+	      }
+	  }
+      } else {
+          if (put_sample == NULL || get_sample == NULL) return SAD_ERROR_CORRUPTED_PRIVATE_DATA;
+          /* process buffer */
+          for(i=0; i<frames; i++) {
+	      for(ch=0; ch<channels; ch++){
+  	          sad_sint32 sample = get_sample (inbuf, channels, ch, i);
+	          sad_sint32 isample = __dither_sample_fixed_to_int (sample, inbits, fracbits, outbits, scale, dither, hardlimit);
+                  put_sample (outbuf, isample, channels, ch, i);
+	      }
+	  }
+      }
+  }
+
+  return SAD_ERROR_OK;
+}
+
+int SAD_dither_apply_replaygain (SAD_dither_t *state, SAD_replaygain_info *rg_info, SAD_replaygain_mode *mode) {
+  SAD_state_priv *priv = (SAD_state_priv*) state;
+  float scale = -1.0, peak = 0.0;
+
+  DEBUG_MSG("f: SAD_dither_apply_replaygain\n",0);
+  
+  if(!rg_info->present) {
+    priv->rg_scale = 1.0;
+    priv->hardlimit = FALSE;
+    return SAD_ERROR_OK;
+  }
+
+  switch(mode->mode) {
+    case SAD_RG_ALBUM:
+      scale = db2scale(rg_info->album_gain);
+      peak = rg_info->album_peak;
+      if (peak == 0.0) {
+        scale = db2scale(rg_info->track_gain); // fallback to per-track mode
+	peak = rg_info->track_peak;
+	DEBUG_MSG("f: SAD_dither_apply_replaygain: fallback to track mode\n",0);
+      }
+      break;
+    case SAD_RG_TRACK:
+      scale = db2scale(rg_info->track_gain);
+      peak = rg_info->track_peak;
+      break;
+    case SAD_RG_NONE:
+      scale = -1.0;
+  }
+  
+  if (scale != -1.0 && peak != 0.0) {
+    DEBUG_MSG("f: SAD_dither_apply_replaygain: applying\n",0);
+    scale *= mode->preamp;
+    // Clipping prevention
+    if(mode->clipping_prevention) {
+#ifdef DEBUG
+      if(scale * peak > 1.0) DEBUG_MSG("f: SAD_dither_apply_replaygain: clipping prevented\n",0);
+#endif
+      scale = scale * peak > 1.0 ? 1.0 / peak : scale;
+    }
+    scale = scale > 15.0 ? 15.0 : scale; // safety
+    priv->rg_scale = scale;
+    priv->hardlimit = mode->hard_limit; // apply settings
+  } else {
+    priv->rg_scale = 1.0;
+    priv->hardlimit = FALSE;
+  }
+
+  return SAD_ERROR_OK;
+}
+
+int SAD_dither_set_scale (SAD_dither_t *state, float scale) {
+  SAD_state_priv *priv = (SAD_state_priv*) state;
+  priv->scale = scale;
+  return SAD_ERROR_OK;
+}
+
+int SAD_dither_set_dither (SAD_dither_t *state, int dither) {
+  SAD_state_priv *priv = (SAD_state_priv*) state;
+  priv->dither = dither;
+  return SAD_ERROR_OK;
+}
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/libSAD/dither.h	Wed Jan 30 01:22:37 2008 +0300
@@ -0,0 +1,28 @@
+#ifndef DITHER_H
+#define DITHER_H
+
+#include "common.h"
+
+#define SAD_ERROR_INCORRECT_INPUT_SAMPLEFORMAT -2
+#define SAD_ERROR_INCORRECT_OUTPUT_SAMPLEFORMAT -3
+#define SAD_ERROR_CORRUPTED_PRIVATE_DATA -4
+
+typedef sad_sint32 (*SAD_get_sample_proc) (void *buf, int nch, int ch, int i);
+typedef void (*SAD_put_sample_proc) (void *buf, sad_sint32 sample, int nch, int ch, int i);
+
+typedef struct {
+  SAD_get_sample_proc get_sample;
+  SAD_put_sample_proc put_sample;
+} SAD_buffer_ops;
+
+/* private data */
+typedef struct {} SAD_dither_t;
+
+SAD_dither_t* SAD_dither_init(SAD_buffer_format *inbuf_format, SAD_buffer_format *outbuf_format, int *error);
+int SAD_dither_free(SAD_dither_t* state);
+int SAD_dither_process_buffer (SAD_dither_t *state, void *inbuf, void *outbuf, int frames);
+int SAD_dither_apply_replaygain (SAD_dither_t *state, SAD_replaygain_info *rg_info, SAD_replaygain_mode *mode);
+int SAD_dither_set_scale (SAD_dither_t *state, float scale);
+int SAD_dither_set_dither (SAD_dither_t *state, int dither);
+
+#endif
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/libSAD/dither_ops.c	Wed Jan 30 01:22:37 2008 +0300
@@ -0,0 +1,301 @@
+/* Scale & Dither library (libSAD)
+ * High-precision bit depth converter with ReplayGain support
+ *
+ * (c)2007 by Eugene Zagidullin (e.asphyx@gmail.com)
+ * 
+ * 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, write to the Free Software 
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+#include "common.h"
+#include "buffer.h"
+#include "dither.h"
+
+/* buffer ops: */
+/********** 
+ * 8-bit  *
+ **********/
+
+/* signed */
+static sad_sint32 get_s8_i_sample (void *buf, int nch, int ch, int i) {
+  return ((sad_sint8*)buf)[i*nch+ch];
+}
+
+static sad_sint32 get_s8_s_sample (void *buf, int nch, int ch, int i) {
+  return ((sad_sint8**)buf)[ch][i];
+}
+
+static void put_s8_i_sample (void *buf, sad_sint32 sample, int nch, int ch, int i) {
+  ((sad_sint8*)buf)[i*nch+ch] = (sad_sint8)sample;
+}
+
+static void put_s8_s_sample (void *buf, sad_sint32 sample, int nch, int ch, int i) {
+  ((sad_sint8**)buf)[ch][i] = (sad_sint8)sample;
+}
+
+/* unsigned */
+static sad_sint32 get_u8_i_sample (void *buf, int nch, int ch, int i) {
+  return (sad_sint32)(((sad_uint8*)buf)[i*nch+ch]) - 128;
+}
+
+static sad_sint32 get_u8_s_sample (void *buf, int nch, int ch, int i) {
+  return (sad_sint32)(((sad_uint8**)buf)[ch][i]) - 128;
+}
+
+static void put_u8_i_sample (void *buf, sad_sint32 sample, int nch, int ch, int i) {
+  ((sad_uint8*)buf)[i*nch+ch] = (sad_uint8)sample + 128;
+}
+
+static void put_u8_s_sample (void *buf, sad_sint32 sample, int nch, int ch, int i) {
+  ((sad_uint8**)buf)[ch][i] = (sad_uint8)sample + 128;
+}
+
+static SAD_buffer_ops buf_s8_i_ops = {
+  &get_s8_i_sample,
+  &put_s8_i_sample
+};
+
+static SAD_buffer_ops buf_s8_s_ops = {
+  &get_s8_s_sample,
+  &put_s8_s_sample
+};
+
+static SAD_buffer_ops buf_u8_i_ops = {
+  &get_u8_i_sample,
+  &put_u8_i_sample
+};
+
+static SAD_buffer_ops buf_u8_s_ops = {
+  &get_u8_s_sample,
+  &put_u8_s_sample
+};
+
+/**********
+ * 16-bit *
+ **********/
+
+/* signed */
+static sad_sint32 get_s16_i_sample (void *buf, int nch, int ch, int i) {
+  return (sad_sint32)(((sad_sint16*)buf)[i*nch+ch]);
+}
+
+static sad_sint32 get_s16_s_sample (void *buf, int nch, int ch, int i) {
+  return (sad_sint32)(((sad_sint16**)buf)[ch][i]);
+}
+
+static void put_s16_i_sample (void *buf, sad_sint32 sample, int nch, int ch, int i) {
+/*#ifdef DEEP_DEBUG
+  printf("f: put_s16_i_sample\n");
+#endif*/
+  ((sad_sint16*)buf)[i*nch+ch] = (sad_sint16)sample;
+}
+
+static void put_s16_s_sample (void *buf, sad_sint32 sample, int nch, int ch, int i) {
+/*#ifdef DEEP_DEBUG
+  printf("f: put_s16_s_sample\n");
+#endif*/
+  ((sad_sint16**)buf)[ch][i] = (sad_sint16)sample;
+}
+
+/* unsigned */
+static sad_sint32 get_u16_i_sample (void *buf, int nch, int ch, int i) {
+  return ((sad_sint32)(((sad_uint16*)buf)[i*nch+ch])) - 32768;
+}
+
+static sad_sint32 get_u16_s_sample (void *buf, int nch, int ch, int i) {
+  return ((sad_sint32)(((sad_uint16**)buf)[ch][i])) - 32768;
+}
+
+static void put_u16_i_sample (void *buf, sad_sint32 sample, int nch, int ch, int i) {
+  ((sad_uint16*)buf)[i*nch+ch] = (sad_uint16)(sample + 32768);
+}
+
+static void put_u16_s_sample (void *buf, sad_sint32 sample, int nch, int ch, int i) {
+  ((sad_uint16**)buf)[ch][i] = (sad_uint16)(sample + 32768);
+}
+
+static SAD_buffer_ops buf_s16_i_ops = {
+  &get_s16_i_sample,
+  &put_s16_i_sample
+};
+
+static SAD_buffer_ops buf_s16_s_ops = {
+  &get_s16_s_sample,
+  &put_s16_s_sample
+};
+
+static SAD_buffer_ops buf_u16_i_ops = {
+  &get_u16_i_sample,
+  &put_u16_i_sample
+};
+
+static SAD_buffer_ops buf_u16_s_ops = {
+  &get_u16_s_sample,
+  &put_u16_s_sample
+};
+
+/**********
+ * 24-bit *
+ **********/
+
+/*expand 24-bit signed value to 32-bit*/
+#define EXPAND_24_TO_32(x) (((sad_sint32)(((x) & 0x00ffffff) << 8)) >> 8)
+
+/* signed */
+static sad_sint32 get_s24_i_sample (void *buf, int nch, int ch, int i) {
+  return (sad_sint32)EXPAND_24_TO_32(((sad_sint32*)buf)[i*nch+ch]);
+}
+
+static sad_sint32 get_s24_s_sample (void *buf, int nch, int ch, int i) {
+  return (sad_sint32)EXPAND_24_TO_32(((sad_sint32**)buf)[ch][i]);
+}
+
+static void put_s24_i_sample (void *buf, sad_sint32 sample, int nch, int ch, int i) {
+  ((sad_sint32*)buf)[i*nch+ch] = (sad_sint32)sample & 0x00ffffff;
+}
+
+static void put_s24_s_sample (void *buf, sad_sint32 sample, int nch, int ch, int i) {
+  ((sad_sint32**)buf)[ch][i] = (sad_sint32)sample & 0x00ffffff;
+}
+
+/* unsigned */
+static sad_sint32 get_u24_i_sample (void *buf, int nch, int ch, int i) {
+  return (sad_sint32)EXPAND_24_TO_32(((sad_uint32*)buf)[i*nch+ch]) - 8388608;
+}
+
+static sad_sint32 get_u24_s_sample (void *buf, int nch, int ch, int i) {
+  return (sad_sint32)EXPAND_24_TO_32(((sad_uint32**)buf)[ch][i]) - 8388608;
+}
+
+static void put_u24_i_sample (void *buf, sad_sint32 sample, int nch, int ch, int i) {
+  ((sad_uint32*)buf)[i*nch+ch] = ((sad_uint32)sample + 8388608) & 0x00ffffff;
+}
+
+static void put_u24_s_sample (void *buf, sad_sint32 sample, int nch, int ch, int i) {
+  ((sad_uint32**)buf)[ch][i] = ((sad_uint32)sample + 8388608) & 0x00ffffff;
+}
+
+static SAD_buffer_ops buf_s24_i_ops = {
+  &get_s24_i_sample,
+  &put_s24_i_sample
+};
+
+static SAD_buffer_ops buf_s24_s_ops = {
+  &get_s24_s_sample,
+  &put_s24_s_sample
+};
+
+static SAD_buffer_ops buf_u24_i_ops = {
+  &get_u24_i_sample,
+  &put_u24_i_sample
+};
+
+static SAD_buffer_ops buf_u24_s_ops = {
+  &get_u24_s_sample,
+  &put_u24_s_sample
+};
+
+/**********
+ * 32-bit *
+ **********/
+
+/* signed */
+static sad_sint32 get_s32_i_sample (void *buf, int nch, int ch, int i) {
+  return ((sad_sint32*)buf)[i*nch+ch];
+}
+
+static sad_sint32 get_s32_s_sample (void *buf, int nch, int ch, int i) {
+  return ((sad_sint32**)buf)[ch][i];
+}
+
+static void put_s32_i_sample (void *buf, sad_sint32 sample, int nch, int ch, int i) {
+  ((sad_sint32*)buf)[i*nch+ch] = (sad_sint32)sample;
+}
+
+static void put_s32_s_sample (void *buf, sad_sint32 sample, int nch, int ch, int i) {
+  ((sad_sint32**)buf)[ch][i] = (sad_sint32)sample;
+}
+
+/* unsigned */
+static sad_sint32 get_u32_i_sample (void *buf, int nch, int ch, int i) {
+  return ((sad_sint32*)buf)[i*nch+ch] - (sad_sint32)(1L<<31);
+}
+
+static sad_sint32 get_u32_s_sample (void *buf, int nch, int ch, int i) {
+  return ((sad_sint32**)buf)[ch][i] - (sad_sint32)(1L<<31);
+}
+
+static void put_u32_i_sample (void *buf, sad_sint32 sample, int nch, int ch, int i) {
+  ((sad_uint32*)buf)[i*nch+ch] = (sad_uint32)(sample + (sad_sint32)(1L<<31));
+}
+
+static void put_u32_s_sample (void *buf, sad_sint32 sample, int nch, int ch, int i) {
+  ((sad_uint32**)buf)[ch][i] = (sad_uint32)(sample + (sad_sint32)(1L<<31));
+}
+
+static SAD_buffer_ops buf_s32_i_ops = {
+  &get_s32_i_sample,
+  &put_s32_i_sample
+};
+
+static SAD_buffer_ops buf_s32_s_ops = {
+  &get_s32_s_sample,
+  &put_s32_s_sample
+};
+
+static SAD_buffer_ops buf_u32_i_ops = {
+  &get_u32_i_sample,
+  &put_u32_i_sample
+};
+
+static SAD_buffer_ops buf_u32_s_ops = {
+  &get_u32_s_sample,
+  &put_u32_s_sample
+};
+
+static SAD_buffer_ops *SAD_buffer_optable[SAD_SAMPLE_MAX][SAD_CHORDER_MAX] = {
+  {&buf_s8_i_ops, &buf_s8_s_ops},	/* SAD_SAMPLE_S8     */
+  {&buf_u8_i_ops, &buf_u8_s_ops},	/* SAD_SAMPLE_U8     */
+  {&buf_s16_i_ops, &buf_s16_s_ops},	/* SAD_SAMPLE_S16    */
+  {NULL,           NULL},		/* SAD_SAMPLE_S16_LE */
+  {NULL,           NULL},		/* SAD_SAMPLE_S16_BE */
+  {&buf_u16_i_ops, &buf_u16_s_ops},	/* SAD_SAMPLE_U16    */
+  {NULL,           NULL}, 		/* SAD_SAMPLE_U16_LE */
+  {NULL,           NULL}, 		/* SAD_SAMPLE_U16_BE */
+  {&buf_s24_i_ops, &buf_s24_s_ops},	/* SAD_SAMPLE_S24    */
+  {NULL,           NULL}, 		/* SAD_SAMPLE_S24_LE */
+  {NULL,           NULL}, 		/* SAD_SAMPLE_S24_BE */
+  {&buf_u24_i_ops, &buf_u24_s_ops},	/* SAD_SAMPLE_U24    */
+  {NULL,           NULL}, 		/* SAD_SAMPLE_U24_LE */
+  {NULL,           NULL}, 		/* SAD_SAMPLE_U24_BE */
+  {&buf_s32_i_ops, &buf_s32_s_ops}, 	/* SAD_SAMPLE_S32    */
+  {NULL,           NULL}, 		/* SAD_SAMPLE_S32_LE */
+  {NULL,           NULL}, 		/* SAD_SAMPLE_S32_BE */
+  {&buf_u32_i_ops, &buf_u32_s_ops}, 	/* SAD_SAMPLE_U32    */
+  {NULL,           NULL}, 		/* SAD_SAMPLE_U32_LE */
+  {NULL,           NULL}, 		/* SAD_SAMPLE_U32_BE */
+  {&buf_s32_i_ops, &buf_s32_s_ops}, 	/* SAD_SAMPLE_FIXED32*/
+  {NULL,           NULL}		/* SAD_SAMPLE_FLOAT  */
+};
+
+SAD_buffer_ops* SAD_assign_buf_ops (SAD_buffer_format *format) {
+#ifdef DEBUG
+  printf("f: SAD_assign_buf_ops\n");
+#endif
+  if (format->sample_format < SAD_SAMPLE_MAX && format->channels_order < SAD_CHORDER_MAX) {
+    return SAD_buffer_optable[format->sample_format][format->channels_order];
+  } else {
+    return NULL;
+  }
+}
+
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/libSAD/dither_ops.h	Wed Jan 30 01:22:37 2008 +0300
@@ -0,0 +1,28 @@
+/* Scale & Dither library (libSAD)
+ * High-precision bit depth converter with ReplayGain support
+ *
+ * (c)2007 by Eugene Zagidullin (e.asphyx@gmail.com)
+ * 
+ * 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, write to the Free Software 
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+#ifndef BUFFER_H
+#define BUFFER_H
+
+#include "common.h"
+#include "dither.h"
+
+SAD_buffer_ops* SAD_assign_buf_ops (SAD_buffer_format *format);
+
+#endif /*BUFFER_H*/
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/libSAD/libSAD.h	Wed Jan 30 01:22:37 2008 +0300
@@ -0,0 +1,180 @@
+/* Scale & Dither library (libSAD)
+ * High-precision bit depth converter with ReplayGain support
+ *
+ * (c)2007 by Eugene Zagidullin (e.asphyx@gmail.com)
+ * 
+ * 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, write to the Free Software 
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+#ifndef COMMON_H
+#define COMMON_H
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <sys/types.h>
+#include <math.h>
+
+#ifndef TRUE
+#define TRUE 1
+#endif
+
+#ifndef FALSE
+#define FALSE 0
+#endif
+
+#include "../config.h"
+
+#ifdef ENABLE_NLS
+#  include <libintl.h>
+#  include <locale.h>
+#  define _(String) dgettext(PACKAGE,String)
+#else 
+#  define _(String) (String)
+#endif
+
+typedef u_int8_t sad_uint8;
+typedef int8_t sad_sint8;
+
+#if SIZEOF_SHORT == 2
+  typedef unsigned short sad_uint16;
+  typedef short sad_sint16;
+#else
+#  error "Unsupported sizeof(short)"
+#endif
+
+#if SIZEOF_INT == 4
+  typedef unsigned int sad_uint32;
+  typedef int sad_sint32;
+#else
+#  error "Unsupported sizeof(int)"
+#endif
+
+typedef size_t sad_size_t;
+
+typedef int SAD_error;
+
+typedef enum {
+  SAD_SAMPLE_S8,
+  SAD_SAMPLE_U8,
+  SAD_SAMPLE_S16, 
+  SAD_SAMPLE_S16_LE, 
+  SAD_SAMPLE_S16_BE, 
+  SAD_SAMPLE_U16,
+  SAD_SAMPLE_U16_LE,
+  SAD_SAMPLE_U16_BE,
+  SAD_SAMPLE_S24,
+  SAD_SAMPLE_S24_LE,
+  SAD_SAMPLE_S24_BE,
+  SAD_SAMPLE_U24,
+  SAD_SAMPLE_U24_LE,
+  SAD_SAMPLE_U24_BE,
+  SAD_SAMPLE_S32,
+  SAD_SAMPLE_S32_LE,
+  SAD_SAMPLE_S32_BE,
+  SAD_SAMPLE_U32,
+  SAD_SAMPLE_U32_LE,
+  SAD_SAMPLE_U32_BE,
+  SAD_SAMPLE_FIXED32,
+  SAD_SAMPLE_FLOAT,
+  SAD_SAMPLE_MAX /*EOF*/
+} SAD_sample_format;
+
+/* sample format -> sample size */
+static inline sad_size_t sf2ss(SAD_sample_format fmt) {
+  switch(fmt) {
+    case SAD_SAMPLE_S8:
+    case SAD_SAMPLE_U8: return sizeof(sad_sint8);
+    case SAD_SAMPLE_S16:
+    case SAD_SAMPLE_S16_LE:
+    case SAD_SAMPLE_S16_BE:
+    case SAD_SAMPLE_U16:
+    case SAD_SAMPLE_U16_LE:
+    case SAD_SAMPLE_U16_BE: return sizeof(sad_sint16);
+    case SAD_SAMPLE_S24:
+    case SAD_SAMPLE_S24_LE:
+    case SAD_SAMPLE_S24_BE:
+    case SAD_SAMPLE_U24:
+    case SAD_SAMPLE_U24_LE:
+    case SAD_SAMPLE_U24_BE:
+    case SAD_SAMPLE_S32:
+    case SAD_SAMPLE_S32_LE:
+    case SAD_SAMPLE_S32_BE:
+    case SAD_SAMPLE_U32:
+    case SAD_SAMPLE_U32_LE:
+    case SAD_SAMPLE_U32_BE:
+    case SAD_SAMPLE_FIXED32: return sizeof(sad_sint32);
+    case SAD_SAMPLE_FLOAT: return sizeof(float);
+    default: return 0;
+  }
+}
+
+typedef enum {
+  SAD_CHORDER_INTERLEAVED,
+  SAD_CHORDER_SEPARATED,
+  SAD_CHORDER_MAX /*EOF*/
+} SAD_channels_order;
+
+typedef struct {
+  SAD_sample_format sample_format;
+  int fracbits; /* for fixed-point only */
+  int channels;
+  SAD_channels_order channels_order;
+  int samplerate;
+} SAD_buffer_format;
+
+static inline sad_size_t bytes2frames(SAD_buffer_format *fmt, sad_size_t bytes) {
+  return bytes / sf2ss(fmt->sample_format) / fmt->channels;
+}
+
+static inline sad_size_t frames2bytes(SAD_buffer_format *fmt, sad_size_t frames) {
+  return frames * sf2ss(fmt->sample_format) * fmt->channels;
+}
+
+static inline float db2scale(float db) {
+  return pow(10, db / 20);
+}
+
+enum {
+  SAD_RG_NONE,
+  SAD_RG_TRACK,
+  SAD_RG_ALBUM
+};
+
+#define SAD_ERROR_OK 0
+#define SAD_ERROR_FAIL -1
+
+#ifdef DEBUG
+#define DEBUG_MSG(f,x) {printf("debug: "f, x);}
+#else
+#define DEBUG_MSG(f,x) {}
+#endif
+
+typedef struct {
+  int present;
+  float track_gain; // in dB !!!
+  float track_peak;
+  float album_gain;
+  float album_peak;
+} SAD_replaygain_info;
+
+typedef struct {
+  int mode;
+  int clipping_prevention;
+  int hard_limit;
+  float preamp;
+} SAD_replaygain_mode;
+
+
+#endif /* COMMON_H */
+
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/libSAD/noicegen.c	Wed Jan 30 01:22:37 2008 +0300
@@ -0,0 +1,35 @@
+#include <stdio.h>
+#include <assert.h>
+#include "../../config.h"
+
+#define MEXP 19937
+
+#ifdef HAVE_SSE2
+#define SSE2 1
+#endif
+#ifdef HAVE_ALTIVEC
+#define ALTIVEC 1
+#endif
+
+#include "SFMT.h"
+#include "SFMT.c"
+
+#include "noicegen.h"
+
+int triangular_dither_noise(int nbits)
+{
+    // parameter nbits : the peak-to-peak amplitude desired (in bits)
+    //  use with nbits set to    2 + nber of bits to be trimmed.
+    // (because triangular is made from two uniformly distributed processes,
+    // it starts at 2 bits peak-to-peak amplitude)
+    // see The Theory of Dithered Quantization by Robert Alexander Wannamaker
+    // for complete proof of why that's optimal
+
+    int v = (gen_rand32() / 2 - gen_rand32() / 2);   // in ]-2^31, 2^31[
+    //int signe = (v>0) ? 1 : -1;
+    int P = 1 << (32 - nbits); // the power of 2
+    v /= P;
+    // now v in ]-2^(nbits-1), 2^(nbits-1) [ 
+
+    return v;
+}
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/libSAD/noisegen.h	Wed Jan 30 01:22:37 2008 +0300
@@ -0,0 +1,7 @@
+#ifndef NOISGEN_H
+#define NOISGEN_H
+
+int triangular_dither_noise(int nbits);
+float triangular_dither_noise_f(void);
+
+#endif /*NOISGEN_H*/