shithub: libopusenc

Download patch

ref: 4099570a26f27bc11aa62cda2d6bd0898397f068
parent: 0c34772ec8717e922645d8c51f5f5c1988cb5875
author: Jean-Marc Valin <jmvalin@jmvalin.ca>
date: Sun Apr 30 22:01:44 EDT 2017

Adding resampler code (not used yet)

--- a/Makefile.am
+++ b/Makefile.am
@@ -1,5 +1,7 @@
 ACLOCAL_AMFLAGS = -I m4
 
+DEFS += -DSPX_RESAMPLE_EXPORT= -DRANDOM_PREFIX=libopusenc -DOUTSIDE_SPEEX -DFLOATING_POINT
+
 AM_CFLAGS = -I$(top_srcdir)/include $(DEPS_CFLAGS)
 
 dist_doc_DATA = COPYING AUTHORS README.md
@@ -10,7 +12,8 @@
 lib_LTLIBRARIES = libopusenc.la
 libopusenc_la_SOURCES = \
 	src/opus_header.c \
-	src/opusenc.c
+	src/opusenc.c \
+	src/resample.c
 libopusenc_la_LIBADD = $(DEPS_LIBS) $(lrintf_lib)
 libopusenc_la_LDFLAGS = -no-undefined \
  -version-info @OP_LT_CURRENT@:@OP_LT_REVISION@:@OP_LT_AGE@
--- a/configure.ac
+++ b/configure.ac
@@ -21,6 +21,9 @@
 
 AM_INIT_AUTOMAKE([1.11 foreign no-define dist-zip subdir-objects])
 AM_MAINTAINER_MODE([enable])
+
+AC_C_INLINE
+
 LT_INIT
 
 m4_ifdef([AM_SILENT_RULES], [AM_SILENT_RULES([yes])])
@@ -44,7 +47,7 @@
 AC_SUBST(OP_LT_AGE)
 
 CC_CHECK_CFLAGS_APPEND(
-  [-std=c89 -pedantic -Wall -Wextra -Wno-parentheses -Wno-long-long])
+  [-pedantic -Wall -Wextra -Wno-parentheses -Wno-long-long])
 
 # Platform-specific tweaks
 case $host in
--- /dev/null
+++ b/src/arch.h
@@ -1,0 +1,239 @@
+/* Copyright (C) 2003 Jean-Marc Valin */
+/**
+   @file arch.h
+   @brief Various architecture definitions Speex
+*/
+/*
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions
+   are met:
+   
+   - Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+   
+   - Redistributions in binary form must reproduce the above copyright
+   notice, this list of conditions and the following disclaimer in the
+   documentation and/or other materials provided with the distribution.
+   
+   - Neither the name of the Xiph.org Foundation nor the names of its
+   contributors may be used to endorse or promote products derived from
+   this software without specific prior written permission.
+   
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR
+   CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifndef ARCH_H
+#define ARCH_H
+
+#ifndef SPEEX_VERSION
+#define SPEEX_MAJOR_VERSION 1         /**< Major Speex version. */
+#define SPEEX_MINOR_VERSION 1         /**< Minor Speex version. */
+#define SPEEX_MICRO_VERSION 15        /**< Micro Speex version. */
+#define SPEEX_EXTRA_VERSION ""        /**< Extra Speex version. */
+#define SPEEX_VERSION "speex-1.2beta3"  /**< Speex version string. */
+#endif
+
+/* A couple test to catch stupid option combinations */
+#ifdef FIXED_POINT
+
+#ifdef FLOATING_POINT
+#error You cannot compile as floating point and fixed point at the same time
+#endif
+#if defined(__SSE__)
+#error SSE is only for floating-point
+#endif
+#if ((defined (ARM4_ASM)||defined (ARM4_ASM)) && defined(BFIN_ASM)) || (defined (ARM4_ASM)&&defined(ARM5E_ASM))
+#error Make up your mind. What CPU do you have?
+#endif
+#ifdef VORBIS_PSYCHO
+#error Vorbis-psy model currently not implemented in fixed-point
+#endif
+
+#else
+
+#ifndef FLOATING_POINT
+#error You now need to define either FIXED_POINT or FLOATING_POINT
+#endif
+#if defined (ARM4_ASM) || defined(ARM5E_ASM) || defined(BFIN_ASM)
+#error I suppose you can have a [ARM4/ARM5E/Blackfin] that has float instructions?
+#endif
+#ifdef FIXED_POINT_DEBUG
+#error "Don't you think enabling fixed-point is a good thing to do if you want to debug that?"
+#endif
+
+
+#endif
+
+#ifndef OUTSIDE_SPEEX
+#include "../include/speex/speex_types.h"
+#endif
+
+#define ABS(x) ((x) < 0 ? (-(x)) : (x))      /**< Absolute integer value. */
+#define ABS16(x) ((x) < 0 ? (-(x)) : (x))    /**< Absolute 16-bit value.  */
+#define MIN16(a,b) ((a) < (b) ? (a) : (b))   /**< Maximum 16-bit value.   */
+#define MAX16(a,b) ((a) > (b) ? (a) : (b))   /**< Maximum 16-bit value.   */
+#define ABS32(x) ((x) < 0 ? (-(x)) : (x))    /**< Absolute 32-bit value.  */
+#define MIN32(a,b) ((a) < (b) ? (a) : (b))   /**< Maximum 32-bit value.   */
+#define MAX32(a,b) ((a) > (b) ? (a) : (b))   /**< Maximum 32-bit value.   */
+
+#ifdef FIXED_POINT
+
+typedef spx_int16_t spx_word16_t;
+typedef spx_int32_t spx_word32_t;
+typedef spx_word32_t spx_mem_t;
+typedef spx_word16_t spx_coef_t;
+typedef spx_word16_t spx_lsp_t;
+typedef spx_word32_t spx_sig_t;
+
+#define Q15ONE 32767
+
+#define LPC_SCALING  8192
+#define SIG_SCALING  16384
+#define LSP_SCALING  8192.
+#define GAMMA_SCALING 32768.
+#define GAIN_SCALING 64
+#define GAIN_SCALING_1 0.015625
+
+#define LPC_SHIFT    13
+#define LSP_SHIFT    13
+#define SIG_SHIFT    14
+#define GAIN_SHIFT   6
+
+#define VERY_SMALL 0
+#define VERY_LARGE32 ((spx_word32_t)2147483647)
+#define VERY_LARGE16 ((spx_word16_t)32767)
+#define Q15_ONE ((spx_word16_t)32767)
+
+
+#ifdef FIXED_DEBUG
+#include "fixed_debug.h"
+#else
+
+#include "fixed_generic.h"
+
+#ifdef ARM5E_ASM
+#include "fixed_arm5e.h"
+#elif defined (ARM4_ASM)
+#include "fixed_arm4.h"
+#elif defined (BFIN_ASM)
+#include "fixed_bfin.h"
+#endif
+
+#endif
+
+
+#else
+
+typedef float spx_mem_t;
+typedef float spx_coef_t;
+typedef float spx_lsp_t;
+typedef float spx_sig_t;
+typedef float spx_word16_t;
+typedef float spx_word32_t;
+
+#define Q15ONE 1.0f
+#define LPC_SCALING  1.f
+#define SIG_SCALING  1.f
+#define LSP_SCALING  1.f
+#define GAMMA_SCALING 1.f
+#define GAIN_SCALING 1.f
+#define GAIN_SCALING_1 1.f
+
+
+#define VERY_SMALL 1e-15f
+#define VERY_LARGE32 1e15f
+#define VERY_LARGE16 1e15f
+#define Q15_ONE ((spx_word16_t)1.f)
+
+#define QCONST16(x,bits) (x)
+#define QCONST32(x,bits) (x)
+
+#define NEG16(x) (-(x))
+#define NEG32(x) (-(x))
+#define EXTRACT16(x) (x)
+#define EXTEND32(x) (x)
+#define SHR16(a,shift) (a)
+#define SHL16(a,shift) (a)
+#define SHR32(a,shift) (a)
+#define SHL32(a,shift) (a)
+#define PSHR16(a,shift) (a)
+#define PSHR32(a,shift) (a)
+#define VSHR32(a,shift) (a)
+#define SATURATE16(x,a) (x)
+#define SATURATE32(x,a) (x)
+
+#define PSHR(a,shift)       (a)
+#define SHR(a,shift)       (a)
+#define SHL(a,shift)       (a)
+#define SATURATE(x,a) (x)
+
+#define ADD16(a,b) ((a)+(b))
+#define SUB16(a,b) ((a)-(b))
+#define ADD32(a,b) ((a)+(b))
+#define SUB32(a,b) ((a)-(b))
+#define MULT16_16_16(a,b)     ((a)*(b))
+#define MULT16_16(a,b)     ((spx_word32_t)(a)*(spx_word32_t)(b))
+#define MAC16_16(c,a,b)     ((c)+(spx_word32_t)(a)*(spx_word32_t)(b))
+
+#define MULT16_32_Q11(a,b)     ((a)*(b))
+#define MULT16_32_Q13(a,b)     ((a)*(b))
+#define MULT16_32_Q14(a,b)     ((a)*(b))
+#define MULT16_32_Q15(a,b)     ((a)*(b))
+#define MULT16_32_P15(a,b)     ((a)*(b))
+
+#define MAC16_32_Q11(c,a,b)     ((c)+(a)*(b))
+#define MAC16_32_Q15(c,a,b)     ((c)+(a)*(b))
+
+#define MAC16_16_Q11(c,a,b)     ((c)+(a)*(b))
+#define MAC16_16_Q13(c,a,b)     ((c)+(a)*(b))
+#define MAC16_16_P13(c,a,b)     ((c)+(a)*(b))
+#define MULT16_16_Q11_32(a,b)     ((a)*(b))
+#define MULT16_16_Q13(a,b)     ((a)*(b))
+#define MULT16_16_Q14(a,b)     ((a)*(b))
+#define MULT16_16_Q15(a,b)     ((a)*(b))
+#define MULT16_16_P15(a,b)     ((a)*(b))
+#define MULT16_16_P13(a,b)     ((a)*(b))
+#define MULT16_16_P14(a,b)     ((a)*(b))
+
+#define DIV32_16(a,b)     (((spx_word32_t)(a))/(spx_word16_t)(b))
+#define PDIV32_16(a,b)     (((spx_word32_t)(a))/(spx_word16_t)(b))
+#define DIV32(a,b)     (((spx_word32_t)(a))/(spx_word32_t)(b))
+#define PDIV32(a,b)     (((spx_word32_t)(a))/(spx_word32_t)(b))
+
+
+#endif
+
+
+#if defined (CONFIG_TI_C54X) || defined (CONFIG_TI_C55X)
+
+/* 2 on TI C5x DSP */
+#define BYTES_PER_CHAR 2 
+#define BITS_PER_CHAR 16
+#define LOG2_BITS_PER_CHAR 4
+
+#else 
+
+#define BYTES_PER_CHAR 1
+#define BITS_PER_CHAR 8
+#define LOG2_BITS_PER_CHAR 3
+
+#endif
+
+
+
+#ifdef FIXED_DEBUG
+extern long long spx_mips;
+#endif
+
+
+#endif
--- /dev/null
+++ b/src/os_support.h
@@ -1,0 +1,167 @@
+/* Copyright (C) 2007 Jean-Marc Valin
+      
+   File: os_support.h
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+   1. Redistributions of source code must retain the above copyright notice,
+   this list of conditions and the following disclaimer.
+
+   2. Redistributions in binary form must reproduce the above copyright
+   notice, this list of conditions and the following disclaimer in the
+   documentation and/or other materials provided with the distribution.
+
+   3. The name of the author may not be used to endorse or promote products
+   derived from this software without specific prior written permission.
+
+   THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
+   IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
+   OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+   DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT,
+   INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+   (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+   SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+   HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
+   STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
+   ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+   POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifndef OS_SUPPORT_H
+#define OS_SUPPORT_H
+
+#include <string.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+#ifdef OS_SUPPORT_CUSTOM
+#include "os_support_custom.h"
+#endif
+
+/** Speex wrapper for calloc. To do your own dynamic allocation, all you need to do is replace this function, speex_realloc and speex_free 
+    NOTE: speex_alloc needs to CLEAR THE MEMORY */
+#ifndef OVERRIDE_SPEEX_ALLOC
+static inline void *speex_alloc (int size)
+{
+   /* WARNING: this is not equivalent to malloc(). If you want to use malloc() 
+      or your own allocator, YOU NEED TO CLEAR THE MEMORY ALLOCATED. Otherwise
+      you will experience strange bugs */
+   return calloc(size,1);
+}
+#endif
+
+/** Same as speex_alloc, except that the area is only needed inside a Speex call (might cause problem with wideband though) */
+#ifndef OVERRIDE_SPEEX_ALLOC_SCRATCH
+static inline void *speex_alloc_scratch (int size)
+{
+   /* Scratch space doesn't need to be cleared */
+   return calloc(size,1);
+}
+#endif
+
+/** Speex wrapper for realloc. To do your own dynamic allocation, all you need to do is replace this function, speex_alloc and speex_free */
+#ifndef OVERRIDE_SPEEX_REALLOC
+static inline void *speex_realloc (void *ptr, int size)
+{
+   return realloc(ptr, size);
+}
+#endif
+
+/** Speex wrapper for calloc. To do your own dynamic allocation, all you need to do is replace this function, speex_realloc and speex_alloc */
+#ifndef OVERRIDE_SPEEX_FREE
+static inline void speex_free (void *ptr)
+{
+   free(ptr);
+}
+#endif
+
+/** Same as speex_free, except that the area is only needed inside a Speex call (might cause problem with wideband though) */
+#ifndef OVERRIDE_SPEEX_FREE_SCRATCH
+static inline void speex_free_scratch (void *ptr)
+{
+   free(ptr);
+}
+#endif
+
+/** Copy n elements from src to dst. The 0* term provides compile-time type checking  */
+#ifndef OVERRIDE_SPEEX_COPY
+#define SPEEX_COPY(dst, src, n) (memcpy((dst), (src), (n)*sizeof(*(dst)) + 0*((dst)-(src)) ))
+#endif
+
+/** Copy n elements from src to dst, allowing overlapping regions. The 0* term
+    provides compile-time type checking */
+#ifndef OVERRIDE_SPEEX_MOVE
+#define SPEEX_MOVE(dst, src, n) (memmove((dst), (src), (n)*sizeof(*(dst)) + 0*((dst)-(src)) ))
+#endif
+
+/** For n elements worth of memory, set every byte to the value of c, starting at address dst */
+#ifndef OVERRIDE_SPEEX_MEMSET
+#define SPEEX_MEMSET(dst, c, n) (memset((dst), (c), (n)*sizeof(*(dst))))
+#endif
+
+
+#ifndef OVERRIDE_SPEEX_FATAL
+static inline void _speex_fatal(const char *str, const char *file, int line)
+{
+   fprintf (stderr, "Fatal (internal) error in %s, line %d: %s\n", file, line, str);
+   exit(1);
+}
+#endif
+
+#ifndef OVERRIDE_SPEEX_WARNING
+static inline void speex_warning(const char *str)
+{
+#ifndef DISABLE_WARNINGS
+   fprintf (stderr, "warning: %s\n", str);
+#endif
+}
+#endif
+
+#ifndef OVERRIDE_SPEEX_WARNING_INT
+static inline void speex_warning_int(const char *str, int val)
+{
+#ifndef DISABLE_WARNINGS
+   fprintf (stderr, "warning: %s %d\n", str, val);
+#endif
+}
+#endif
+
+#ifndef OVERRIDE_SPEEX_NOTIFY
+static inline void speex_notify(const char *str)
+{
+#ifndef DISABLE_NOTIFICATIONS
+   fprintf (stderr, "notification: %s\n", str);
+#endif
+}
+#endif
+
+#ifndef OVERRIDE_SPEEX_PUTC
+/** Speex wrapper for putc */
+static inline void _speex_putc(int ch, void *file)
+{
+   FILE *f = (FILE *)file;
+   fprintf(f, "%c", ch);
+}
+#endif
+
+#define speex_fatal(str) _speex_fatal(str, __FILE__, __LINE__);
+#define speex_assert(cond) {if (!(cond)) {speex_fatal("assertion failed: " #cond);}}
+
+#ifndef RELEASE
+static inline void print_vec(float *vec, int len, char *name)
+{
+   int i;
+   printf ("%s ", name);
+   for (i=0;i<len;i++)
+      printf (" %f", vec[i]);
+   printf ("\n");
+}
+#endif
+
+#endif
+
--- /dev/null
+++ b/src/resample.c
@@ -1,0 +1,1153 @@
+/* Copyright (C) 2007-2008 Jean-Marc Valin
+   Copyright (C) 2008      Thorvald Natvig
+      
+   File: resample.c
+   Arbitrary resampling code
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+   1. Redistributions of source code must retain the above copyright notice,
+   this list of conditions and the following disclaimer.
+
+   2. Redistributions in binary form must reproduce the above copyright
+   notice, this list of conditions and the following disclaimer in the
+   documentation and/or other materials provided with the distribution.
+
+   3. The name of the author may not be used to endorse or promote products
+   derived from this software without specific prior written permission.
+
+   THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
+   IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
+   OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+   DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT,
+   INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+   (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+   SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+   HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
+   STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
+   ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+   POSSIBILITY OF SUCH DAMAGE.
+*/
+
+/*
+   The design goals of this code are:
+      - Very fast algorithm
+      - SIMD-friendly algorithm
+      - Low memory requirement
+      - Good *perceptual* quality (and not best SNR)
+
+   Warning: This resampler is relatively new. Although I think I got rid of 
+   all the major bugs and I don't expect the API to change anymore, there
+   may be something I've missed. So use with caution.
+
+   This algorithm is based on this original resampling algorithm:
+   Smith, Julius O. Digital Audio Resampling Home Page
+   Center for Computer Research in Music and Acoustics (CCRMA), 
+   Stanford University, 2007.
+   Web published at http://www-ccrma.stanford.edu/~jos/resample/.
+
+   There is one main difference, though. This resampler uses cubic 
+   interpolation instead of linear interpolation in the above paper. This
+   makes the table much smaller and makes it possible to compute that table
+   on a per-stream basis. In turn, being able to tweak the table for each 
+   stream makes it possible to both reduce complexity on simple ratios 
+   (e.g. 2/3), and get rid of the rounding operations in the inner loop. 
+   The latter both reduces CPU time and makes the algorithm more SIMD-friendly.
+*/
+
+#ifdef HAVE_CONFIG_H
+# include "config.h"
+#endif
+
+#define RESAMPLE_HUGEMEM 1
+
+#ifdef OUTSIDE_SPEEX
+#include <stdlib.h>
+static void *speex_alloc (int size) {return calloc(size,1);}
+static void *speex_realloc (void *ptr, int size) {return realloc(ptr, size);}
+static void speex_free (void *ptr) {free(ptr);}
+#include "speex_resampler.h"
+#include "arch.h"
+#else /* OUTSIDE_SPEEX */
+               
+#include "../include/speex/speex_resampler.h"
+#include "arch.h"
+#include "os_support.h"
+#endif /* OUTSIDE_SPEEX */
+
+#include "stack_alloc.h"
+#include <math.h>
+
+#ifndef M_PI
+#define M_PI 3.14159263
+#endif
+
+#ifdef FIXED_POINT
+#define WORD2INT(x) ((x) < -32767 ? -32768 : ((x) > 32766 ? 32767 : (x)))  
+#else
+#define WORD2INT(x) ((x) < -32767.5f ? -32768 : ((x) > 32766.5f ? 32767 : floor(.5+(x))))  
+#endif
+               
+#define IMAX(a,b) ((a) > (b) ? (a) : (b))
+#define IMIN(a,b) ((a) < (b) ? (a) : (b))
+
+#ifndef NULL
+#define NULL 0
+#endif
+
+#if defined(FLOATING_POINT) && defined(__SSE__)
+# include "resample_sse.h"
+#endif
+
+/* Numer of elements to allocate on the stack */
+#ifdef VAR_ARRAYS
+#define FIXED_STACK_ALLOC 8192
+#else
+#define FIXED_STACK_ALLOC 1024
+#endif
+
+typedef int (*resampler_basic_func)(SpeexResamplerState *, spx_uint32_t , const spx_word16_t *, spx_uint32_t *, spx_word16_t *, spx_uint32_t *);
+
+struct SpeexResamplerState_ {
+   spx_uint32_t in_rate;
+   spx_uint32_t out_rate;
+   spx_uint32_t num_rate;
+   spx_uint32_t den_rate;
+   
+   int    quality;
+   spx_uint32_t nb_channels;
+   spx_uint32_t filt_len;
+   spx_uint32_t mem_alloc_size;
+   spx_uint32_t buffer_size;
+   int          int_advance;
+   int          frac_advance;
+   float  cutoff;
+   spx_uint32_t oversample;
+   int          initialised;
+   int          started;
+   
+   /* These are per-channel */
+   spx_int32_t  *last_sample;
+   spx_uint32_t *samp_frac_num;
+   spx_uint32_t *magic_samples;
+   
+   spx_word16_t *mem;
+   spx_word16_t *sinc_table;
+   spx_uint32_t sinc_table_length;
+   resampler_basic_func resampler_ptr;
+         
+   int    in_stride;
+   int    out_stride;
+} ;
+
+static const double kaiser12_table[68] = {
+   0.99859849, 1.00000000, 0.99859849, 0.99440475, 0.98745105, 0.97779076,
+   0.96549770, 0.95066529, 0.93340547, 0.91384741, 0.89213598, 0.86843014,
+   0.84290116, 0.81573067, 0.78710866, 0.75723148, 0.72629970, 0.69451601,
+   0.66208321, 0.62920216, 0.59606986, 0.56287762, 0.52980938, 0.49704014,
+   0.46473455, 0.43304576, 0.40211431, 0.37206735, 0.34301800, 0.31506490,
+   0.28829195, 0.26276832, 0.23854851, 0.21567274, 0.19416736, 0.17404546,
+   0.15530766, 0.13794294, 0.12192957, 0.10723616, 0.09382272, 0.08164178,
+   0.07063950, 0.06075685, 0.05193064, 0.04409466, 0.03718069, 0.03111947,
+   0.02584161, 0.02127838, 0.01736250, 0.01402878, 0.01121463, 0.00886058,
+   0.00691064, 0.00531256, 0.00401805, 0.00298291, 0.00216702, 0.00153438,
+   0.00105297, 0.00069463, 0.00043489, 0.00025272, 0.00013031, 0.0000527734,
+   0.00001000, 0.00000000};
+/*
+static const double kaiser12_table[36] = {
+   0.99440475, 1.00000000, 0.99440475, 0.97779076, 0.95066529, 0.91384741,
+   0.86843014, 0.81573067, 0.75723148, 0.69451601, 0.62920216, 0.56287762,
+   0.49704014, 0.43304576, 0.37206735, 0.31506490, 0.26276832, 0.21567274,
+   0.17404546, 0.13794294, 0.10723616, 0.08164178, 0.06075685, 0.04409466,
+   0.03111947, 0.02127838, 0.01402878, 0.00886058, 0.00531256, 0.00298291,
+   0.00153438, 0.00069463, 0.00025272, 0.0000527734, 0.00000500, 0.00000000};
+*/
+static const double kaiser10_table[36] = {
+   0.99537781, 1.00000000, 0.99537781, 0.98162644, 0.95908712, 0.92831446,
+   0.89005583, 0.84522401, 0.79486424, 0.74011713, 0.68217934, 0.62226347,
+   0.56155915, 0.50119680, 0.44221549, 0.38553619, 0.33194107, 0.28205962,
+   0.23636152, 0.19515633, 0.15859932, 0.12670280, 0.09935205, 0.07632451,
+   0.05731132, 0.04193980, 0.02979584, 0.02044510, 0.01345224, 0.00839739,
+   0.00488951, 0.00257636, 0.00115101, 0.00035515, 0.00000000, 0.00000000};
+
+static const double kaiser8_table[36] = {
+   0.99635258, 1.00000000, 0.99635258, 0.98548012, 0.96759014, 0.94302200,
+   0.91223751, 0.87580811, 0.83439927, 0.78875245, 0.73966538, 0.68797126,
+   0.63451750, 0.58014482, 0.52566725, 0.47185369, 0.41941150, 0.36897272,
+   0.32108304, 0.27619388, 0.23465776, 0.19672670, 0.16255380, 0.13219758,
+   0.10562887, 0.08273982, 0.06335451, 0.04724088, 0.03412321, 0.02369490,
+   0.01563093, 0.00959968, 0.00527363, 0.00233883, 0.00050000, 0.00000000};
+   
+static const double kaiser6_table[36] = {
+   0.99733006, 1.00000000, 0.99733006, 0.98935595, 0.97618418, 0.95799003,
+   0.93501423, 0.90755855, 0.87598009, 0.84068475, 0.80211977, 0.76076565,
+   0.71712752, 0.67172623, 0.62508937, 0.57774224, 0.53019925, 0.48295561,
+   0.43647969, 0.39120616, 0.34752997, 0.30580127, 0.26632152, 0.22934058,
+   0.19505503, 0.16360756, 0.13508755, 0.10953262, 0.08693120, 0.06722600,
+   0.05031820, 0.03607231, 0.02432151, 0.01487334, 0.00752000, 0.00000000};
+
+struct FuncDef {
+   const double *table;
+   int oversample;
+};
+      
+static const struct FuncDef _KAISER12 = {kaiser12_table, 64};
+#define KAISER12 (&_KAISER12)
+/*static struct FuncDef _KAISER12 = {kaiser12_table, 32};
+#define KAISER12 (&_KAISER12)*/
+static const struct FuncDef _KAISER10 = {kaiser10_table, 32};
+#define KAISER10 (&_KAISER10)
+static const struct FuncDef _KAISER8 = {kaiser8_table, 32};
+#define KAISER8 (&_KAISER8)
+static const struct FuncDef _KAISER6 = {kaiser6_table, 32};
+#define KAISER6 (&_KAISER6)
+
+struct QualityMapping {
+   int base_length;
+   int oversample;
+   float downsample_bandwidth;
+   float upsample_bandwidth;
+   const struct FuncDef *window_func;
+};
+
+
+/* This table maps conversion quality to internal parameters. There are two
+   reasons that explain why the up-sampling bandwidth is larger than the 
+   down-sampling bandwidth:
+   1) When up-sampling, we can assume that the spectrum is already attenuated
+      close to the Nyquist rate (from an A/D or a previous resampling filter)
+   2) Any aliasing that occurs very close to the Nyquist rate will be masked
+      by the sinusoids/noise just below the Nyquist rate (guaranteed only for
+      up-sampling).
+*/
+static const struct QualityMapping quality_map[11] = {
+   {  8,  4, 0.830f, 0.860f, KAISER6 }, /* Q0 */
+   { 16,  4, 0.850f, 0.880f, KAISER6 }, /* Q1 */
+   { 32,  4, 0.882f, 0.910f, KAISER6 }, /* Q2 */  /* 82.3% cutoff ( ~60 dB stop) 6  */
+   { 48,  8, 0.895f, 0.917f, KAISER8 }, /* Q3 */  /* 84.9% cutoff ( ~80 dB stop) 8  */
+   { 64,  8, 0.921f, 0.940f, KAISER8 }, /* Q4 */  /* 88.7% cutoff ( ~80 dB stop) 8  */
+   { 80, 16, 0.922f, 0.940f, KAISER10}, /* Q5 */  /* 89.1% cutoff (~100 dB stop) 10 */
+   { 96, 16, 0.940f, 0.945f, KAISER10}, /* Q6 */  /* 91.5% cutoff (~100 dB stop) 10 */
+   {128, 16, 0.950f, 0.950f, KAISER10}, /* Q7 */  /* 93.1% cutoff (~100 dB stop) 10 */
+   {160, 16, 0.960f, 0.960f, KAISER10}, /* Q8 */  /* 94.5% cutoff (~100 dB stop) 10 */
+   {192, 32, 0.968f, 0.968f, KAISER12}, /* Q9 */  /* 95.5% cutoff (~100 dB stop) 10 */
+   {256, 32, 0.975f, 0.975f, KAISER12}, /* Q10 */ /* 96.6% cutoff (~100 dB stop) 10 */
+};
+/*8,24,40,56,80,104,128,160,200,256,320*/
+static double compute_func(float x, const struct FuncDef *func)
+{
+   float y, frac;
+   double interp[4];
+   int ind; 
+   y = x*func->oversample;
+   ind = (int)floor(y);
+   frac = (y-ind);
+   /* CSE with handle the repeated powers */
+   interp[3] =  -0.1666666667*frac + 0.1666666667*(frac*frac*frac);
+   interp[2] = frac + 0.5*(frac*frac) - 0.5*(frac*frac*frac);
+   /*interp[2] = 1.f - 0.5f*frac - frac*frac + 0.5f*frac*frac*frac;*/
+   interp[0] = -0.3333333333*frac + 0.5*(frac*frac) - 0.1666666667*(frac*frac*frac);
+   /* Just to make sure we don't have rounding problems */
+   interp[1] = 1.f-interp[3]-interp[2]-interp[0];
+   
+   /*sum = frac*accum[1] + (1-frac)*accum[2];*/
+   return interp[0]*func->table[ind] + interp[1]*func->table[ind+1] + interp[2]*func->table[ind+2] + interp[3]*func->table[ind+3];
+}
+
+#if 0
+#include <stdio.h>
+int main(int argc, char **argv)
+{
+   int i;
+   for (i=0;i<256;i++)
+   {
+      printf ("%f\n", compute_func(i/256., KAISER12));
+   }
+   return 0;
+}
+#endif
+
+#ifdef FIXED_POINT
+/* The slow way of computing a sinc for the table. Should improve that some day */
+static spx_word16_t sinc(float cutoff, float x, int N, const struct FuncDef *window_func)
+{
+   /*fprintf (stderr, "%f ", x);*/
+   float xx = x * cutoff;
+   if (fabs(x)<1e-6f)
+      return WORD2INT(32768.*cutoff);
+   else if (fabs(x) > .5f*N)
+      return 0;
+   /*FIXME: Can it really be any slower than this? */
+   return WORD2INT(32768.*cutoff*sin(M_PI*xx)/(M_PI*xx) * compute_func(fabs(2.*x/N), window_func));
+}
+#else
+/* The slow way of computing a sinc for the table. Should improve that some day */
+static spx_word16_t sinc(float cutoff, float x, int N, const struct FuncDef *window_func)
+{
+   /*fprintf (stderr, "%f ", x);*/
+   float xx = x * cutoff;
+   if (fabs(x)<1e-6)
+      return cutoff;
+   else if (fabs(x) > .5*N)
+      return 0;
+   /*FIXME: Can it really be any slower than this? */
+   return cutoff*sin(M_PI*xx)/(M_PI*xx) * compute_func(fabs(2.*x/N), window_func);
+}
+#endif
+
+#ifdef FIXED_POINT
+static void cubic_coef(spx_word16_t x, spx_word16_t interp[4])
+{
+   /* Compute interpolation coefficients. I'm not sure whether this corresponds to cubic interpolation
+   but I know it's MMSE-optimal on a sinc */
+   spx_word16_t x2, x3;
+   x2 = MULT16_16_P15(x, x);
+   x3 = MULT16_16_P15(x, x2);
+   interp[0] = PSHR32(MULT16_16(QCONST16(-0.16667f, 15),x) + MULT16_16(QCONST16(0.16667f, 15),x3),15);
+   interp[1] = EXTRACT16(EXTEND32(x) + SHR32(SUB32(EXTEND32(x2),EXTEND32(x3)),1));
+   interp[3] = PSHR32(MULT16_16(QCONST16(-0.33333f, 15),x) + MULT16_16(QCONST16(.5f,15),x2) - MULT16_16(QCONST16(0.16667f, 15),x3),15);
+   /* Just to make sure we don't have rounding problems */
+   interp[2] = Q15_ONE-interp[0]-interp[1]-interp[3];
+   if (interp[2]<32767)
+      interp[2]+=1;
+}
+#else
+static void cubic_coef(spx_word16_t frac, spx_word16_t interp[4])
+{
+   /* Compute interpolation coefficients. I'm not sure whether this corresponds to cubic interpolation
+   but I know it's MMSE-optimal on a sinc */
+   interp[0] =  -0.16667f*frac + 0.16667f*frac*frac*frac;
+   interp[1] = frac + 0.5f*frac*frac - 0.5f*frac*frac*frac;
+   /*interp[2] = 1.f - 0.5f*frac - frac*frac + 0.5f*frac*frac*frac;*/
+   interp[3] = -0.33333f*frac + 0.5f*frac*frac - 0.16667f*frac*frac*frac;
+   /* Just to make sure we don't have rounding problems */
+   interp[2] = 1.-interp[0]-interp[1]-interp[3];
+}
+#endif
+
+static int resampler_basic_direct_single(SpeexResamplerState *st, spx_uint32_t channel_index, const spx_word16_t *in, spx_uint32_t *in_len, spx_word16_t *out, spx_uint32_t *out_len)
+{
+   const int N = st->filt_len;
+   int out_sample = 0;
+   int last_sample = st->last_sample[channel_index];
+   spx_uint32_t samp_frac_num = st->samp_frac_num[channel_index];
+   const spx_word16_t *sinc_table = st->sinc_table;
+   const int out_stride = st->out_stride;
+   const int int_advance = st->int_advance;
+   const int frac_advance = st->frac_advance;
+   const spx_uint32_t den_rate = st->den_rate;
+   spx_word32_t sum;
+
+   while (!(last_sample >= (spx_int32_t)*in_len || out_sample >= (spx_int32_t)*out_len))
+   {
+      const spx_word16_t *sinct = & sinc_table[samp_frac_num*N];
+      const spx_word16_t *iptr = & in[last_sample];
+
+#ifndef OVERRIDE_INNER_PRODUCT_SINGLE
+      int j;
+      sum = 0;
+      for(j=0;j<N;j++) sum += MULT16_16(sinct[j], iptr[j]);
+
+/*    This code is slower on most DSPs which have only 2 accumulators.
+      Plus this this forces truncation to 32 bits and you lose the HW guard bits.
+      I think we can trust the compiler and let it vectorize and/or unroll itself.
+      spx_word32_t accum[4] = {0,0,0,0};
+      for(j=0;j<N;j+=4) {
+        accum[0] += MULT16_16(sinct[j], iptr[j]);
+        accum[1] += MULT16_16(sinct[j+1], iptr[j+1]);
+        accum[2] += MULT16_16(sinct[j+2], iptr[j+2]);
+        accum[3] += MULT16_16(sinct[j+3], iptr[j+3]);
+      }
+      sum = accum[0] + accum[1] + accum[2] + accum[3];
+*/
+#else
+      sum = inner_product_single(sinct, iptr, N);
+#endif
+
+      out[out_stride * out_sample++] = SATURATE32(PSHR32(sum, 15), 32767);
+      last_sample += int_advance;
+      samp_frac_num += frac_advance;
+      if (samp_frac_num >= den_rate)
+      {
+         samp_frac_num -= den_rate;
+         last_sample++;
+      }
+   }
+
+   st->last_sample[channel_index] = last_sample;
+   st->samp_frac_num[channel_index] = samp_frac_num;
+   return out_sample;
+}
+
+#ifdef FIXED_POINT
+#else
+/* This is the same as the previous function, except with a double-precision accumulator */
+static int resampler_basic_direct_double(SpeexResamplerState *st, spx_uint32_t channel_index, const spx_word16_t *in, spx_uint32_t *in_len, spx_word16_t *out, spx_uint32_t *out_len)
+{
+   const int N = st->filt_len;
+   int out_sample = 0;
+   int last_sample = st->last_sample[channel_index];
+   spx_uint32_t samp_frac_num = st->samp_frac_num[channel_index];
+   const spx_word16_t *sinc_table = st->sinc_table;
+   const int out_stride = st->out_stride;
+   const int int_advance = st->int_advance;
+   const int frac_advance = st->frac_advance;
+   const spx_uint32_t den_rate = st->den_rate;
+   double sum;
+
+   while (!(last_sample >= (spx_int32_t)*in_len || out_sample >= (spx_int32_t)*out_len))
+   {
+      const spx_word16_t *sinct = & sinc_table[samp_frac_num*N];
+      const spx_word16_t *iptr = & in[last_sample];
+
+#ifndef OVERRIDE_INNER_PRODUCT_DOUBLE
+      int j;
+      double accum[4] = {0,0,0,0};
+
+      for(j=0;j<N;j+=4) {
+        accum[0] += sinct[j]*iptr[j];
+        accum[1] += sinct[j+1]*iptr[j+1];
+        accum[2] += sinct[j+2]*iptr[j+2];
+        accum[3] += sinct[j+3]*iptr[j+3];
+      }
+      sum = accum[0] + accum[1] + accum[2] + accum[3];
+#else
+      sum = inner_product_double(sinct, iptr, N);
+#endif
+
+      out[out_stride * out_sample++] = PSHR32(sum, 15);
+      last_sample += int_advance;
+      samp_frac_num += frac_advance;
+      if (samp_frac_num >= den_rate)
+      {
+         samp_frac_num -= den_rate;
+         last_sample++;
+      }
+   }
+
+   st->last_sample[channel_index] = last_sample;
+   st->samp_frac_num[channel_index] = samp_frac_num;
+   return out_sample;
+}
+#endif
+
+static int resampler_basic_interpolate_single(SpeexResamplerState *st, spx_uint32_t channel_index, const spx_word16_t *in, spx_uint32_t *in_len, spx_word16_t *out, spx_uint32_t *out_len)
+{
+   const int N = st->filt_len;
+   int out_sample = 0;
+   int last_sample = st->last_sample[channel_index];
+   spx_uint32_t samp_frac_num = st->samp_frac_num[channel_index];
+   const int out_stride = st->out_stride;
+   const int int_advance = st->int_advance;
+   const int frac_advance = st->frac_advance;
+   const spx_uint32_t den_rate = st->den_rate;
+   spx_word32_t sum;
+
+   while (!(last_sample >= (spx_int32_t)*in_len || out_sample >= (spx_int32_t)*out_len))
+   {
+      const spx_word16_t *iptr = & in[last_sample];
+
+      const int offset = samp_frac_num*st->oversample/st->den_rate;
+#ifdef FIXED_POINT
+      const spx_word16_t frac = PDIV32(SHL32((samp_frac_num*st->oversample) % st->den_rate,15),st->den_rate);
+#else
+      const spx_word16_t frac = ((float)((samp_frac_num*st->oversample) % st->den_rate))/st->den_rate;
+#endif
+      spx_word16_t interp[4];
+
+
+#ifndef OVERRIDE_INTERPOLATE_PRODUCT_SINGLE
+      int j;
+      spx_word32_t accum[4] = {0,0,0,0};
+
+      for(j=0;j<N;j++) {
+        const spx_word16_t curr_in=iptr[j];
+        accum[0] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset-2]);
+        accum[1] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset-1]);
+        accum[2] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset]);
+        accum[3] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset+1]);
+      }
+
+      cubic_coef(frac, interp);
+      sum = MULT16_32_Q15(interp[0],SHR32(accum[0], 1)) + MULT16_32_Q15(interp[1],SHR32(accum[1], 1)) + MULT16_32_Q15(interp[2],SHR32(accum[2], 1)) + MULT16_32_Q15(interp[3],SHR32(accum[3], 1));
+#else
+      cubic_coef(frac, interp);
+      sum = interpolate_product_single(iptr, st->sinc_table + st->oversample + 4 - offset - 2, N, st->oversample, interp);
+#endif
+      
+      out[out_stride * out_sample++] = SATURATE32(PSHR32(sum, 14), 32767);
+      last_sample += int_advance;
+      samp_frac_num += frac_advance;
+      if (samp_frac_num >= den_rate)
+      {
+         samp_frac_num -= den_rate;
+         last_sample++;
+      }
+   }
+
+   st->last_sample[channel_index] = last_sample;
+   st->samp_frac_num[channel_index] = samp_frac_num;
+   return out_sample;
+}
+
+#ifdef FIXED_POINT
+#else
+/* This is the same as the previous function, except with a double-precision accumulator */
+static int resampler_basic_interpolate_double(SpeexResamplerState *st, spx_uint32_t channel_index, const spx_word16_t *in, spx_uint32_t *in_len, spx_word16_t *out, spx_uint32_t *out_len)
+{
+   const int N = st->filt_len;
+   int out_sample = 0;
+   int last_sample = st->last_sample[channel_index];
+   spx_uint32_t samp_frac_num = st->samp_frac_num[channel_index];
+   const int out_stride = st->out_stride;
+   const int int_advance = st->int_advance;
+   const int frac_advance = st->frac_advance;
+   const spx_uint32_t den_rate = st->den_rate;
+   spx_word32_t sum;
+
+   while (!(last_sample >= (spx_int32_t)*in_len || out_sample >= (spx_int32_t)*out_len))
+   {
+      const spx_word16_t *iptr = & in[last_sample];
+
+      const int offset = samp_frac_num*st->oversample/st->den_rate;
+#ifdef FIXED_POINT
+      const spx_word16_t frac = PDIV32(SHL32((samp_frac_num*st->oversample) % st->den_rate,15),st->den_rate);
+#else
+      const spx_word16_t frac = ((float)((samp_frac_num*st->oversample) % st->den_rate))/st->den_rate;
+#endif
+      spx_word16_t interp[4];
+
+
+#ifndef OVERRIDE_INTERPOLATE_PRODUCT_DOUBLE
+      int j;
+      double accum[4] = {0,0,0,0};
+
+      for(j=0;j<N;j++) {
+        const double curr_in=iptr[j];
+        accum[0] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset-2]);
+        accum[1] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset-1]);
+        accum[2] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset]);
+        accum[3] += MULT16_16(curr_in,st->sinc_table[4+(j+1)*st->oversample-offset+1]);
+      }
+
+      cubic_coef(frac, interp);
+      sum = MULT16_32_Q15(interp[0],accum[0]) + MULT16_32_Q15(interp[1],accum[1]) + MULT16_32_Q15(interp[2],accum[2]) + MULT16_32_Q15(interp[3],accum[3]);
+#else
+      cubic_coef(frac, interp);
+      sum = interpolate_product_double(iptr, st->sinc_table + st->oversample + 4 - offset - 2, N, st->oversample, interp);
+#endif
+      
+      out[out_stride * out_sample++] = PSHR32(sum,15);
+      last_sample += int_advance;
+      samp_frac_num += frac_advance;
+      if (samp_frac_num >= den_rate)
+      {
+         samp_frac_num -= den_rate;
+         last_sample++;
+      }
+   }
+
+   st->last_sample[channel_index] = last_sample;
+   st->samp_frac_num[channel_index] = samp_frac_num;
+   return out_sample;
+}
+#endif
+
+static void update_filter(SpeexResamplerState *st)
+{
+   spx_uint32_t old_length;
+   
+   old_length = st->filt_len;
+   st->oversample = quality_map[st->quality].oversample;
+   st->filt_len = quality_map[st->quality].base_length;
+   
+   if (st->num_rate > st->den_rate)
+   {
+      /* down-sampling */
+      st->cutoff = quality_map[st->quality].downsample_bandwidth * st->den_rate / st->num_rate;
+      /* FIXME: divide the numerator and denominator by a certain amount if they're too large */
+      st->filt_len = st->filt_len*st->num_rate / st->den_rate;
+      /* Round up to make sure we have a multiple of 8 */
+      st->filt_len = ((st->filt_len-1)&(~0x7))+8;
+      if (2*st->den_rate < st->num_rate)
+         st->oversample >>= 1;
+      if (4*st->den_rate < st->num_rate)
+         st->oversample >>= 1;
+      if (8*st->den_rate < st->num_rate)
+         st->oversample >>= 1;
+      if (16*st->den_rate < st->num_rate)
+         st->oversample >>= 1;
+      if (st->oversample < 1)
+         st->oversample = 1;
+   } else {
+      /* up-sampling */
+      st->cutoff = quality_map[st->quality].upsample_bandwidth;
+   }
+
+#ifdef RESAMPLE_HUGEMEM
+   if (st->den_rate <= 16*(st->oversample+8))
+#else
+   /* Choose the resampling type that requires the least amount of memory */
+   if (st->den_rate <= (st->oversample+8))
+#endif
+   {
+      spx_uint32_t i;
+      if (!st->sinc_table)
+         st->sinc_table = (spx_word16_t *)speex_alloc(st->filt_len*st->den_rate*sizeof(spx_word16_t));
+      else if (st->sinc_table_length < st->filt_len*st->den_rate)
+      {
+         st->sinc_table = (spx_word16_t *)speex_realloc(st->sinc_table,st->filt_len*st->den_rate*sizeof(spx_word16_t));
+         st->sinc_table_length = st->filt_len*st->den_rate;
+      }
+      for (i=0;i<st->den_rate;i++)
+      {
+         spx_int32_t j;
+         for (j=0;j<st->filt_len;j++)
+         {
+            st->sinc_table[i*st->filt_len+j] = sinc(st->cutoff,((j-(spx_int32_t)st->filt_len/2+1)-((float)i)/st->den_rate), st->filt_len, quality_map[st->quality].window_func);
+         }
+      }
+#ifdef FIXED_POINT
+      st->resampler_ptr = resampler_basic_direct_single;
+#else
+      if (st->quality>8)
+         st->resampler_ptr = resampler_basic_direct_double;
+      else
+         st->resampler_ptr = resampler_basic_direct_single;
+#endif
+      /*fprintf (stderr, "resampler uses direct sinc table and normalised cutoff %f\n", cutoff);*/
+   } else {
+      spx_int32_t i;
+      if (!st->sinc_table)
+         st->sinc_table = (spx_word16_t *)speex_alloc((st->filt_len*st->oversample+8)*sizeof(spx_word16_t));
+      else if (st->sinc_table_length < st->filt_len*st->oversample+8)
+      {
+         st->sinc_table = (spx_word16_t *)speex_realloc(st->sinc_table,(st->filt_len*st->oversample+8)*sizeof(spx_word16_t));
+         st->sinc_table_length = st->filt_len*st->oversample+8;
+      }
+      for (i=-4;i<(spx_int32_t)(st->oversample*st->filt_len+4);i++)
+         st->sinc_table[i+4] = sinc(st->cutoff,(i/(float)st->oversample - st->filt_len/2), st->filt_len, quality_map[st->quality].window_func);
+#ifdef FIXED_POINT
+      st->resampler_ptr = resampler_basic_interpolate_single;
+#else
+      if (st->quality>8)
+         st->resampler_ptr = resampler_basic_interpolate_double;
+      else
+         st->resampler_ptr = resampler_basic_interpolate_single;
+#endif
+      /*fprintf (stderr, "resampler uses interpolated sinc table and normalised cutoff %f\n", cutoff);*/
+   }
+   st->int_advance = st->num_rate/st->den_rate;
+   st->frac_advance = st->num_rate%st->den_rate;
+
+   
+   /* Here's the place where we update the filter memory to take into account
+      the change in filter length. It's probably the messiest part of the code
+      due to handling of lots of corner cases. */
+   if (!st->mem)
+   {
+      spx_uint32_t i;
+      st->mem_alloc_size = st->filt_len-1 + st->buffer_size;
+      st->mem = (spx_word16_t*)speex_alloc(st->nb_channels*st->mem_alloc_size * sizeof(spx_word16_t));
+      for (i=0;i<st->nb_channels*st->mem_alloc_size;i++)
+         st->mem[i] = 0;
+      /*speex_warning("init filter");*/
+   } else if (!st->started)
+   {
+      spx_uint32_t i;
+      st->mem_alloc_size = st->filt_len-1 + st->buffer_size;
+      st->mem = (spx_word16_t*)speex_realloc(st->mem, st->nb_channels*st->mem_alloc_size * sizeof(spx_word16_t));
+      for (i=0;i<st->nb_channels*st->mem_alloc_size;i++)
+         st->mem[i] = 0;
+      /*speex_warning("reinit filter");*/
+   } else if (st->filt_len > old_length)
+   {
+      spx_int32_t i;
+      /* Increase the filter length */
+      /*speex_warning("increase filter size");*/
+      int old_alloc_size = st->mem_alloc_size;
+      if ((st->filt_len-1 + st->buffer_size) > st->mem_alloc_size)
+      {
+         st->mem_alloc_size = st->filt_len-1 + st->buffer_size;
+         st->mem = (spx_word16_t*)speex_realloc(st->mem, st->nb_channels*st->mem_alloc_size * sizeof(spx_word16_t));
+      }
+      for (i=st->nb_channels-1;i>=0;i--)
+      {
+         spx_int32_t j;
+         spx_uint32_t olen = old_length;
+         /*if (st->magic_samples[i])*/
+         {
+            /* Try and remove the magic samples as if nothing had happened */
+            
+            /* FIXME: This is wrong but for now we need it to avoid going over the array bounds */
+            olen = old_length + 2*st->magic_samples[i];
+            for (j=old_length-2+st->magic_samples[i];j>=0;j--)
+               st->mem[i*st->mem_alloc_size+j+st->magic_samples[i]] = st->mem[i*old_alloc_size+j];
+            for (j=0;j<st->magic_samples[i];j++)
+               st->mem[i*st->mem_alloc_size+j] = 0;
+            st->magic_samples[i] = 0;
+         }
+         if (st->filt_len > olen)
+         {
+            /* If the new filter length is still bigger than the "augmented" length */
+            /* Copy data going backward */
+            for (j=0;j<olen-1;j++)
+               st->mem[i*st->mem_alloc_size+(st->filt_len-2-j)] = st->mem[i*st->mem_alloc_size+(olen-2-j)];
+            /* Then put zeros for lack of anything better */
+            for (;j<st->filt_len-1;j++)
+               st->mem[i*st->mem_alloc_size+(st->filt_len-2-j)] = 0;
+            /* Adjust last_sample */
+            st->last_sample[i] += (st->filt_len - olen)/2;
+         } else {
+            /* Put back some of the magic! */
+            st->magic_samples[i] = (olen - st->filt_len)/2;
+            for (j=0;j<st->filt_len-1+st->magic_samples[i];j++)
+               st->mem[i*st->mem_alloc_size+j] = st->mem[i*st->mem_alloc_size+j+st->magic_samples[i]];
+         }
+      }
+   } else if (st->filt_len < old_length)
+   {
+      spx_uint32_t i;
+      /* Reduce filter length, this a bit tricky. We need to store some of the memory as "magic"
+         samples so they can be used directly as input the next time(s) */
+      for (i=0;i<st->nb_channels;i++)
+      {
+         spx_uint32_t j;
+         spx_uint32_t old_magic = st->magic_samples[i];
+         st->magic_samples[i] = (old_length - st->filt_len)/2;
+         /* We must copy some of the memory that's no longer used */
+         /* Copy data going backward */
+         for (j=0;j<st->filt_len-1+st->magic_samples[i]+old_magic;j++)
+            st->mem[i*st->mem_alloc_size+j] = st->mem[i*st->mem_alloc_size+j+st->magic_samples[i]];
+         st->magic_samples[i] += old_magic;
+      }
+   }
+
+}
+
+SPX_RESAMPLE_EXPORT SpeexResamplerState *speex_resampler_init(spx_uint32_t nb_channels, spx_uint32_t in_rate, spx_uint32_t out_rate, int quality, int *err)
+{
+   return speex_resampler_init_frac(nb_channels, in_rate, out_rate, in_rate, out_rate, quality, err);
+}
+
+SPX_RESAMPLE_EXPORT SpeexResamplerState *speex_resampler_init_frac(spx_uint32_t nb_channels, spx_uint32_t ratio_num, spx_uint32_t ratio_den, spx_uint32_t in_rate, spx_uint32_t out_rate, int quality, int *err)
+{
+   spx_uint32_t i;
+   SpeexResamplerState *st;
+   if (quality > 10 || quality < 0)
+   {
+      if (err)
+         *err = RESAMPLER_ERR_INVALID_ARG;
+      return NULL;
+   }
+   st = (SpeexResamplerState *)speex_alloc(sizeof(SpeexResamplerState));
+   st->initialised = 0;
+   st->started = 0;
+   st->in_rate = 0;
+   st->out_rate = 0;
+   st->num_rate = 0;
+   st->den_rate = 0;
+   st->quality = -1;
+   st->sinc_table_length = 0;
+   st->mem_alloc_size = 0;
+   st->filt_len = 0;
+   st->mem = 0;
+   st->resampler_ptr = 0;
+         
+   st->cutoff = 1.f;
+   st->nb_channels = nb_channels;
+   st->in_stride = 1;
+   st->out_stride = 1;
+   
+#ifdef FIXED_POINT
+   st->buffer_size = 160;
+#else
+   st->buffer_size = 160;
+#endif
+   
+   /* Per channel data */
+   st->last_sample = (spx_int32_t*)speex_alloc(nb_channels*sizeof(spx_int32_t));
+   st->magic_samples = (spx_uint32_t*)speex_alloc(nb_channels*sizeof(spx_uint32_t));
+   st->samp_frac_num = (spx_uint32_t*)speex_alloc(nb_channels*sizeof(spx_uint32_t));
+   for (i=0;i<nb_channels;i++)
+   {
+      st->last_sample[i] = 0;
+      st->magic_samples[i] = 0;
+      st->samp_frac_num[i] = 0;
+   }
+
+   speex_resampler_set_quality(st, quality);
+   speex_resampler_set_rate_frac(st, ratio_num, ratio_den, in_rate, out_rate);
+
+   
+   update_filter(st);
+   
+   st->initialised = 1;
+   if (err)
+      *err = RESAMPLER_ERR_SUCCESS;
+
+   return st;
+}
+
+SPX_RESAMPLE_EXPORT void speex_resampler_destroy(SpeexResamplerState *st)
+{
+   speex_free(st->mem);
+   speex_free(st->sinc_table);
+   speex_free(st->last_sample);
+   speex_free(st->magic_samples);
+   speex_free(st->samp_frac_num);
+   speex_free(st);
+}
+
+static int speex_resampler_process_native(SpeexResamplerState *st, spx_uint32_t channel_index, spx_uint32_t *in_len, spx_word16_t *out, spx_uint32_t *out_len)
+{
+   int j=0;
+   const int N = st->filt_len;
+   int out_sample = 0;
+   spx_word16_t *mem = st->mem + channel_index * st->mem_alloc_size;
+   spx_uint32_t ilen;
+   
+   st->started = 1;
+   
+   /* Call the right resampler through the function ptr */
+   out_sample = st->resampler_ptr(st, channel_index, mem, in_len, out, out_len);
+   
+   if (st->last_sample[channel_index] < (spx_int32_t)*in_len)
+      *in_len = st->last_sample[channel_index];
+   *out_len = out_sample;
+   st->last_sample[channel_index] -= *in_len;
+   
+   ilen = *in_len;
+
+   for(j=0;j<N-1;++j)
+     mem[j] = mem[j+ilen];
+
+   return RESAMPLER_ERR_SUCCESS;
+}
+
+static int speex_resampler_magic(SpeexResamplerState *st, spx_uint32_t channel_index, spx_word16_t **out, spx_uint32_t out_len) {
+   spx_uint32_t tmp_in_len = st->magic_samples[channel_index];
+   spx_word16_t *mem = st->mem + channel_index * st->mem_alloc_size;
+   const int N = st->filt_len;
+   
+   speex_resampler_process_native(st, channel_index, &tmp_in_len, *out, &out_len);
+
+   st->magic_samples[channel_index] -= tmp_in_len;
+   
+   /* If we couldn't process all "magic" input samples, save the rest for next time */
+   if (st->magic_samples[channel_index])
+   {
+      spx_uint32_t i;
+      for (i=0;i<st->magic_samples[channel_index];i++)
+         mem[N-1+i]=mem[N-1+i+tmp_in_len];
+   }
+   *out += out_len*st->out_stride;
+   return out_len;
+}
+
+#ifdef FIXED_POINT
+SPX_RESAMPLE_EXPORT int speex_resampler_process_int(SpeexResamplerState *st, spx_uint32_t channel_index, const spx_int16_t *in, spx_uint32_t *in_len, spx_int16_t *out, spx_uint32_t *out_len)
+#else
+SPX_RESAMPLE_EXPORT int speex_resampler_process_float(SpeexResamplerState *st, spx_uint32_t channel_index, const float *in, spx_uint32_t *in_len, float *out, spx_uint32_t *out_len)
+#endif
+{
+   int j;
+   spx_uint32_t ilen = *in_len;
+   spx_uint32_t olen = *out_len;
+   spx_word16_t *x = st->mem + channel_index * st->mem_alloc_size;
+   const int filt_offs = st->filt_len - 1;
+   const spx_uint32_t xlen = st->mem_alloc_size - filt_offs;
+   const int istride = st->in_stride;
+
+   if (st->magic_samples[channel_index]) 
+      olen -= speex_resampler_magic(st, channel_index, &out, olen);
+   if (! st->magic_samples[channel_index]) {
+      while (ilen && olen) {
+        spx_uint32_t ichunk = (ilen > xlen) ? xlen : ilen;
+        spx_uint32_t ochunk = olen;
+ 
+        if (in) {
+           for(j=0;j<ichunk;++j)
+              x[j+filt_offs]=in[j*istride];
+        } else {
+          for(j=0;j<ichunk;++j)
+            x[j+filt_offs]=0;
+        }
+        speex_resampler_process_native(st, channel_index, &ichunk, out, &ochunk);
+        ilen -= ichunk;
+        olen -= ochunk;
+        out += ochunk * st->out_stride;
+        if (in)
+           in += ichunk * istride;
+      }
+   }
+   *in_len -= ilen;
+   *out_len -= olen;
+   return RESAMPLER_ERR_SUCCESS;
+}
+
+#ifdef FIXED_POINT
+SPX_RESAMPLE_EXPORT int speex_resampler_process_float(SpeexResamplerState *st, spx_uint32_t channel_index, const float *in, spx_uint32_t *in_len, float *out, spx_uint32_t *out_len)
+#else
+SPX_RESAMPLE_EXPORT int speex_resampler_process_int(SpeexResamplerState *st, spx_uint32_t channel_index, const spx_int16_t *in, spx_uint32_t *in_len, spx_int16_t *out, spx_uint32_t *out_len)
+#endif
+{
+   int j;
+   const int istride_save = st->in_stride;
+   const int ostride_save = st->out_stride;
+   spx_uint32_t ilen = *in_len;
+   spx_uint32_t olen = *out_len;
+   spx_word16_t *x = st->mem + channel_index * st->mem_alloc_size;
+   const spx_uint32_t xlen = st->mem_alloc_size - (st->filt_len - 1);
+#ifdef VAR_ARRAYS
+   const unsigned int ylen = (olen < FIXED_STACK_ALLOC) ? olen : FIXED_STACK_ALLOC;
+   VARDECL(spx_word16_t *ystack);
+   ALLOC(ystack, ylen, spx_word16_t);
+#else
+   const unsigned int ylen = FIXED_STACK_ALLOC;
+   spx_word16_t ystack[FIXED_STACK_ALLOC];
+#endif
+
+   st->out_stride = 1;
+   
+   while (ilen && olen) {
+     spx_word16_t *y = ystack;
+     spx_uint32_t ichunk = (ilen > xlen) ? xlen : ilen;
+     spx_uint32_t ochunk = (olen > ylen) ? ylen : olen;
+     spx_uint32_t omagic = 0;
+
+     if (st->magic_samples[channel_index]) {
+       omagic = speex_resampler_magic(st, channel_index, &y, ochunk);
+       ochunk -= omagic;
+       olen -= omagic;
+     }
+     if (! st->magic_samples[channel_index]) {
+       if (in) {
+         for(j=0;j<ichunk;++j)
+#ifdef FIXED_POINT
+           x[j+st->filt_len-1]=WORD2INT(in[j*istride_save]);
+#else
+           x[j+st->filt_len-1]=in[j*istride_save];
+#endif
+       } else {
+         for(j=0;j<ichunk;++j)
+           x[j+st->filt_len-1]=0;
+       }
+
+       speex_resampler_process_native(st, channel_index, &ichunk, y, &ochunk);
+     } else {
+       ichunk = 0;
+       ochunk = 0;
+     }
+
+     for (j=0;j<ochunk+omagic;++j)
+#ifdef FIXED_POINT
+        out[j*ostride_save] = ystack[j];
+#else
+        out[j*ostride_save] = WORD2INT(ystack[j]);
+#endif
+     
+     ilen -= ichunk;
+     olen -= ochunk;
+     out += (ochunk+omagic) * ostride_save;
+     if (in)
+       in += ichunk * istride_save;
+   }
+   st->out_stride = ostride_save;
+   *in_len -= ilen;
+   *out_len -= olen;
+
+   return RESAMPLER_ERR_SUCCESS;
+}
+
+SPX_RESAMPLE_EXPORT int speex_resampler_process_interleaved_float(SpeexResamplerState *st, const float *in, spx_uint32_t *in_len, float *out, spx_uint32_t *out_len)
+{
+   spx_uint32_t i;
+   int istride_save, ostride_save;
+   spx_uint32_t bak_out_len = *out_len;
+   spx_uint32_t bak_in_len = *in_len;
+   istride_save = st->in_stride;
+   ostride_save = st->out_stride;
+   st->in_stride = st->out_stride = st->nb_channels;
+   for (i=0;i<st->nb_channels;i++)
+   {
+      *out_len = bak_out_len;
+      *in_len = bak_in_len;
+      if (in != NULL)
+         speex_resampler_process_float(st, i, in+i, in_len, out+i, out_len);
+      else
+         speex_resampler_process_float(st, i, NULL, in_len, out+i, out_len);
+   }
+   st->in_stride = istride_save;
+   st->out_stride = ostride_save;
+   return RESAMPLER_ERR_SUCCESS;
+}
+               
+SPX_RESAMPLE_EXPORT int speex_resampler_process_interleaved_int(SpeexResamplerState *st, const spx_int16_t *in, spx_uint32_t *in_len, spx_int16_t *out, spx_uint32_t *out_len)
+{
+   spx_uint32_t i;
+   int istride_save, ostride_save;
+   spx_uint32_t bak_out_len = *out_len;
+   spx_uint32_t bak_in_len = *in_len;
+   istride_save = st->in_stride;
+   ostride_save = st->out_stride;
+   st->in_stride = st->out_stride = st->nb_channels;
+   for (i=0;i<st->nb_channels;i++)
+   {
+      *out_len = bak_out_len;
+      *in_len = bak_in_len;
+      if (in != NULL)
+         speex_resampler_process_int(st, i, in+i, in_len, out+i, out_len);
+      else
+         speex_resampler_process_int(st, i, NULL, in_len, out+i, out_len);
+   }
+   st->in_stride = istride_save;
+   st->out_stride = ostride_save;
+   return RESAMPLER_ERR_SUCCESS;
+}
+
+SPX_RESAMPLE_EXPORT int speex_resampler_set_rate(SpeexResamplerState *st, spx_uint32_t in_rate, spx_uint32_t out_rate)
+{
+   return speex_resampler_set_rate_frac(st, in_rate, out_rate, in_rate, out_rate);
+}
+
+SPX_RESAMPLE_EXPORT void speex_resampler_get_rate(SpeexResamplerState *st, spx_uint32_t *in_rate, spx_uint32_t *out_rate)
+{
+   *in_rate = st->in_rate;
+   *out_rate = st->out_rate;
+}
+
+SPX_RESAMPLE_EXPORT int speex_resampler_set_rate_frac(SpeexResamplerState *st, spx_uint32_t ratio_num, spx_uint32_t ratio_den, spx_uint32_t in_rate, spx_uint32_t out_rate)
+{
+   spx_uint32_t fact;
+   spx_uint32_t old_den;
+   spx_uint32_t i;
+   if (st->in_rate == in_rate && st->out_rate == out_rate && st->num_rate == ratio_num && st->den_rate == ratio_den)
+      return RESAMPLER_ERR_SUCCESS;
+   
+   old_den = st->den_rate;
+   st->in_rate = in_rate;
+   st->out_rate = out_rate;
+   st->num_rate = ratio_num;
+   st->den_rate = ratio_den;
+   /* FIXME: This is terribly inefficient, but who cares (at least for now)? */
+   for (fact=2;fact<=IMIN(st->num_rate, st->den_rate);fact++)
+   {
+      while ((st->num_rate % fact == 0) && (st->den_rate % fact == 0))
+      {
+         st->num_rate /= fact;
+         st->den_rate /= fact;
+      }
+   }
+      
+   if (old_den > 0)
+   {
+      for (i=0;i<st->nb_channels;i++)
+      {
+         st->samp_frac_num[i]=st->samp_frac_num[i]*st->den_rate/old_den;
+         /* Safety net */
+         if (st->samp_frac_num[i] >= st->den_rate)
+            st->samp_frac_num[i] = st->den_rate-1;
+      }
+   }
+   
+   if (st->initialised)
+      update_filter(st);
+   return RESAMPLER_ERR_SUCCESS;
+}
+
+SPX_RESAMPLE_EXPORT void speex_resampler_get_ratio(SpeexResamplerState *st, spx_uint32_t *ratio_num, spx_uint32_t *ratio_den)
+{
+   *ratio_num = st->num_rate;
+   *ratio_den = st->den_rate;
+}
+
+SPX_RESAMPLE_EXPORT int speex_resampler_set_quality(SpeexResamplerState *st, int quality)
+{
+   if (quality > 10 || quality < 0)
+      return RESAMPLER_ERR_INVALID_ARG;
+   if (st->quality == quality)
+      return RESAMPLER_ERR_SUCCESS;
+   st->quality = quality;
+   if (st->initialised)
+      update_filter(st);
+   return RESAMPLER_ERR_SUCCESS;
+}
+
+SPX_RESAMPLE_EXPORT void speex_resampler_get_quality(SpeexResamplerState *st, int *quality)
+{
+   *quality = st->quality;
+}
+
+SPX_RESAMPLE_EXPORT void speex_resampler_set_input_stride(SpeexResamplerState *st, spx_uint32_t stride)
+{
+   st->in_stride = stride;
+}
+
+SPX_RESAMPLE_EXPORT void speex_resampler_get_input_stride(SpeexResamplerState *st, spx_uint32_t *stride)
+{
+   *stride = st->in_stride;
+}
+
+SPX_RESAMPLE_EXPORT void speex_resampler_set_output_stride(SpeexResamplerState *st, spx_uint32_t stride)
+{
+   st->out_stride = stride;
+}
+
+SPX_RESAMPLE_EXPORT void speex_resampler_get_output_stride(SpeexResamplerState *st, spx_uint32_t *stride)
+{
+   *stride = st->out_stride;
+}
+
+SPX_RESAMPLE_EXPORT int speex_resampler_get_input_latency(SpeexResamplerState *st)
+{
+  return st->filt_len / 2;
+}
+
+SPX_RESAMPLE_EXPORT int speex_resampler_get_output_latency(SpeexResamplerState *st)
+{
+  return ((st->filt_len / 2) * st->den_rate + (st->num_rate >> 1)) / st->num_rate;
+}
+
+SPX_RESAMPLE_EXPORT int speex_resampler_skip_zeros(SpeexResamplerState *st)
+{
+   spx_uint32_t i;
+   for (i=0;i<st->nb_channels;i++)
+      st->last_sample[i] = st->filt_len/2;
+   return RESAMPLER_ERR_SUCCESS;
+}
+
+SPX_RESAMPLE_EXPORT int speex_resampler_reset_mem(SpeexResamplerState *st)
+{
+   spx_uint32_t i;
+   for (i=0;i<st->nb_channels;i++)
+   {
+      st->last_sample[i] = 0;
+      st->magic_samples[i] = 0;
+      st->samp_frac_num[i] = 0;
+   }
+   for (i=0;i<st->nb_channels*(st->filt_len-1);i++)
+      st->mem[i] = 0;
+   return RESAMPLER_ERR_SUCCESS;
+}
+
+SPX_RESAMPLE_EXPORT const char *speex_resampler_strerror(int err)
+{
+   switch (err)
+   {
+      case RESAMPLER_ERR_SUCCESS:
+         return "Success.";
+      case RESAMPLER_ERR_ALLOC_FAILED:
+         return "Memory allocation failed.";
+      case RESAMPLER_ERR_BAD_STATE:
+         return "Bad resampler state.";
+      case RESAMPLER_ERR_INVALID_ARG:
+         return "Invalid argument.";
+      case RESAMPLER_ERR_PTR_OVERLAP:
+         return "Input and output buffers overlap.";
+      default:
+         return "Unknown error. Bad error code or strange version mismatch.";
+   }
+}
--- /dev/null
+++ b/src/resample_sse.h
@@ -1,0 +1,152 @@
+/* Copyright (C) 2007-2008 Jean-Marc Valin
+ * Copyright (C) 2008 Thorvald Natvig
+ */
+/**
+   @file resample_sse.h
+   @brief Resampler functions (SSE version)
+*/
+/*
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions
+   are met:
+   
+   - Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+   
+   - Redistributions in binary form must reproduce the above copyright
+   notice, this list of conditions and the following disclaimer in the
+   documentation and/or other materials provided with the distribution.
+   
+   - Neither the name of the Xiph.org Foundation nor the names of its
+   contributors may be used to endorse or promote products derived from
+   this software without specific prior written permission.
+   
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR
+   CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#include <xmmintrin.h>
+
+#define OVERRIDE_INNER_PRODUCT_SINGLE
+static inline float inner_product_single(const float *a, const float *b, unsigned int len)
+{
+   int i;
+   float ret;
+   if (1)
+   {
+      __m128 sum = _mm_setzero_ps();
+      for (i=0;i<len;i+=8)
+      {
+         sum = _mm_add_ps(sum, _mm_mul_ps(_mm_loadu_ps(a+i), _mm_loadu_ps(b+i)));
+         sum = _mm_add_ps(sum, _mm_mul_ps(_mm_loadu_ps(a+i+4), _mm_loadu_ps(b+i+4)));
+      }
+      sum = _mm_add_ps(sum, _mm_movehl_ps(sum, sum));
+      sum = _mm_add_ss(sum, _mm_shuffle_ps(sum, sum, 0x55));
+      _mm_store_ss(&ret, sum);
+   }
+   else
+   {
+      ret = 0;
+      for (i=0;i<len;i++) ret += a[i] * b[i];
+   }
+   return ret;
+}
+
+#define OVERRIDE_INTERPOLATE_PRODUCT_SINGLE
+static inline float interpolate_product_single(const float *a, const float *b, unsigned int len, const spx_uint32_t oversample, float *frac) {
+  int i;
+  float ret;
+  if (1)
+  {
+    __m128 sum = _mm_setzero_ps();
+    __m128 f = _mm_loadu_ps(frac);
+    for(i=0;i<len;i+=2)
+    {
+      sum = _mm_add_ps(sum, _mm_mul_ps(_mm_load1_ps(a+i), _mm_loadu_ps(b+i*oversample)));
+      sum = _mm_add_ps(sum, _mm_mul_ps(_mm_load1_ps(a+i+1), _mm_loadu_ps(b+(i+1)*oversample)));
+    }
+    sum = _mm_mul_ps(f, sum);
+    sum = _mm_add_ps(sum, _mm_movehl_ps(sum, sum));
+    sum = _mm_add_ss(sum, _mm_shuffle_ps(sum, sum, 0x55));
+    _mm_store_ss(&ret, sum);
+  }
+  else
+  {
+    float accum[4] = {0,0,0,0};
+    for(i=0;i<len;i++)
+    {
+      const float curr_in=a[i];
+      accum[0] += curr_in * b[i * oversample + 0];
+      accum[1] += curr_in * b[i * oversample + 1];
+      accum[2] += curr_in * b[i * oversample + 2];
+      accum[3] += curr_in * b[i * oversample + 3];
+    }
+    ret = accum[0] * frac[0] + accum[1] * frac[1] + accum[2] * frac[2] + accum[3] * frac[3];
+  }
+  return ret;
+}
+
+#ifdef __SSE2__
+#include <emmintrin.h>
+#define OVERRIDE_INNER_PRODUCT_DOUBLE
+
+static inline double inner_product_double(const float *a, const float *b, unsigned int len)
+{
+   int i;
+   double ret;
+   __m128d sum = _mm_setzero_pd();
+   __m128 t;
+   for (i=0;i<len;i+=8)
+   {
+      t = _mm_mul_ps(_mm_loadu_ps(a+i), _mm_loadu_ps(b+i));
+      sum = _mm_add_pd(sum, _mm_cvtps_pd(t));
+      sum = _mm_add_pd(sum, _mm_cvtps_pd(_mm_movehl_ps(t, t)));
+
+      t = _mm_mul_ps(_mm_loadu_ps(a+i+4), _mm_loadu_ps(b+i+4));
+      sum = _mm_add_pd(sum, _mm_cvtps_pd(t));
+      sum = _mm_add_pd(sum, _mm_cvtps_pd(_mm_movehl_ps(t, t)));
+   }
+   sum = _mm_add_sd(sum, (__m128d) _mm_movehl_ps((__m128) sum, (__m128) sum));
+   _mm_store_sd(&ret, sum);
+   return ret;
+}
+
+#define OVERRIDE_INTERPOLATE_PRODUCT_DOUBLE
+static inline double interpolate_product_double(const float *a, const float *b, unsigned int len, const spx_uint32_t oversample, float *frac) {
+  int i;
+  double ret;
+  __m128d sum;
+  __m128d sum1 = _mm_setzero_pd();
+  __m128d sum2 = _mm_setzero_pd();
+  __m128 f = _mm_loadu_ps(frac);
+  __m128d f1 = _mm_cvtps_pd(f);
+  __m128d f2 = _mm_cvtps_pd(_mm_movehl_ps(f,f));
+  __m128 t;
+  for(i=0;i<len;i+=2)
+  {
+    t = _mm_mul_ps(_mm_load1_ps(a+i), _mm_loadu_ps(b+i*oversample));
+    sum1 = _mm_add_pd(sum1, _mm_cvtps_pd(t));
+    sum2 = _mm_add_pd(sum2, _mm_cvtps_pd(_mm_movehl_ps(t, t)));
+
+    t = _mm_mul_ps(_mm_load1_ps(a+i+1), _mm_loadu_ps(b+(i+1)*oversample));
+    sum1 = _mm_add_pd(sum1, _mm_cvtps_pd(t));
+    sum2 = _mm_add_pd(sum2, _mm_cvtps_pd(_mm_movehl_ps(t, t)));
+  }
+  sum1 = _mm_mul_pd(f1, sum1);
+  sum2 = _mm_mul_pd(f2, sum2);
+  sum = _mm_add_pd(sum1, sum2);
+  sum = _mm_add_sd(sum, (__m128d) _mm_movehl_ps((__m128) sum, (__m128) sum));
+  _mm_store_sd(&ret, sum);
+  return ret;
+}
+
+#endif
--- /dev/null
+++ b/src/speex_resampler.h
@@ -1,0 +1,344 @@
+/* Copyright (C) 2007 Jean-Marc Valin
+      
+   File: speex_resampler.h
+   Resampling code
+      
+   The design goals of this code are:
+      - Very fast algorithm
+      - Low memory requirement
+      - Good *perceptual* quality (and not best SNR)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+   1. Redistributions of source code must retain the above copyright notice,
+   this list of conditions and the following disclaimer.
+
+   2. Redistributions in binary form must reproduce the above copyright
+   notice, this list of conditions and the following disclaimer in the
+   documentation and/or other materials provided with the distribution.
+
+   3. The name of the author may not be used to endorse or promote products
+   derived from this software without specific prior written permission.
+
+   THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
+   IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
+   OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+   DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT,
+   INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+   (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+   SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+   HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
+   STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
+   ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+   POSSIBILITY OF SUCH DAMAGE.
+*/
+
+
+#ifndef SPEEX_RESAMPLER_H
+#define SPEEX_RESAMPLER_H
+
+#ifdef OUTSIDE_SPEEX
+
+/********* WARNING: MENTAL SANITY ENDS HERE *************/
+
+/* If the resampler is defined outside of Speex, we change the symbol names so that 
+   there won't be any clash if linking with Speex later on. */
+
+/* #define RANDOM_PREFIX your software name here */
+#ifndef RANDOM_PREFIX
+#error "Please define RANDOM_PREFIX (above) to something specific to your project to prevent symbol name clashes"
+#endif
+
+#define CAT_PREFIX2(a,b) a ## b
+#define CAT_PREFIX(a,b) CAT_PREFIX2(a, b)
+      
+#define speex_resampler_init CAT_PREFIX(RANDOM_PREFIX,_resampler_init)
+#define speex_resampler_init_frac CAT_PREFIX(RANDOM_PREFIX,_resampler_init_frac)
+#define speex_resampler_destroy CAT_PREFIX(RANDOM_PREFIX,_resampler_destroy)
+#define speex_resampler_process_float CAT_PREFIX(RANDOM_PREFIX,_resampler_process_float)
+#define speex_resampler_process_int CAT_PREFIX(RANDOM_PREFIX,_resampler_process_int)
+#define speex_resampler_process_interleaved_float CAT_PREFIX(RANDOM_PREFIX,_resampler_process_interleaved_float)
+#define speex_resampler_process_interleaved_int CAT_PREFIX(RANDOM_PREFIX,_resampler_process_interleaved_int)
+#define speex_resampler_set_rate CAT_PREFIX(RANDOM_PREFIX,_resampler_set_rate)
+#define speex_resampler_get_rate CAT_PREFIX(RANDOM_PREFIX,_resampler_get_rate)
+#define speex_resampler_set_rate_frac CAT_PREFIX(RANDOM_PREFIX,_resampler_set_rate_frac)
+#define speex_resampler_get_ratio CAT_PREFIX(RANDOM_PREFIX,_resampler_get_ratio)
+#define speex_resampler_set_quality CAT_PREFIX(RANDOM_PREFIX,_resampler_set_quality)
+#define speex_resampler_get_quality CAT_PREFIX(RANDOM_PREFIX,_resampler_get_quality)
+#define speex_resampler_set_input_stride CAT_PREFIX(RANDOM_PREFIX,_resampler_set_input_stride)
+#define speex_resampler_get_input_stride CAT_PREFIX(RANDOM_PREFIX,_resampler_get_input_stride)
+#define speex_resampler_set_output_stride CAT_PREFIX(RANDOM_PREFIX,_resampler_set_output_stride)
+#define speex_resampler_get_output_stride CAT_PREFIX(RANDOM_PREFIX,_resampler_get_output_stride)
+#define speex_resampler_get_input_latency CAT_PREFIX(RANDOM_PREFIX,_resampler_get_input_latency)
+#define speex_resampler_get_output_latency CAT_PREFIX(RANDOM_PREFIX,_resampler_get_output_latency)
+#define speex_resampler_skip_zeros CAT_PREFIX(RANDOM_PREFIX,_resampler_skip_zeros)
+#define speex_resampler_reset_mem CAT_PREFIX(RANDOM_PREFIX,_resampler_reset_mem)
+#define speex_resampler_strerror CAT_PREFIX(RANDOM_PREFIX,_resampler_strerror)
+
+#define spx_int16_t short
+#define spx_int32_t int
+#define spx_uint16_t unsigned short
+#define spx_uint32_t unsigned int
+      
+#else /* OUTSIDE_SPEEX */
+
+#ifdef _BUILD_SPEEX
+# include "speex_types.h"
+#else
+# include <speex/speex_types.h>
+#endif
+
+#endif /* OUTSIDE_SPEEX */
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define SPEEX_RESAMPLER_QUALITY_MAX 10
+#define SPEEX_RESAMPLER_QUALITY_MIN 0
+#define SPEEX_RESAMPLER_QUALITY_DEFAULT 4
+#define SPEEX_RESAMPLER_QUALITY_VOIP 3
+#define SPEEX_RESAMPLER_QUALITY_DESKTOP 5
+
+enum {
+   RESAMPLER_ERR_SUCCESS         = 0,
+   RESAMPLER_ERR_ALLOC_FAILED    = 1,
+   RESAMPLER_ERR_BAD_STATE       = 2,
+   RESAMPLER_ERR_INVALID_ARG     = 3,
+   RESAMPLER_ERR_PTR_OVERLAP     = 4,
+   
+   RESAMPLER_ERR_MAX_ERROR
+};
+
+struct SpeexResamplerState_;
+typedef struct SpeexResamplerState_ SpeexResamplerState;
+
+/** Create a new resampler with integer input and output rates.
+ * @param nb_channels Number of channels to be processed
+ * @param in_rate Input sampling rate (integer number of Hz).
+ * @param out_rate Output sampling rate (integer number of Hz).
+ * @param quality Resampling quality between 0 and 10, where 0 has poor quality
+ * and 10 has very high quality.
+ * @return Newly created resampler state
+ * @retval NULL Error: not enough memory
+ */
+SpeexResamplerState *speex_resampler_init(spx_uint32_t nb_channels, 
+                                          spx_uint32_t in_rate, 
+                                          spx_uint32_t out_rate, 
+                                          int quality,
+                                          int *err);
+
+/** Create a new resampler with fractional input/output rates. The sampling 
+ * rate ratio is an arbitrary rational number with both the numerator and 
+ * denominator being 32-bit integers.
+ * @param nb_channels Number of channels to be processed
+ * @param ratio_num Numerator of the sampling rate ratio
+ * @param ratio_den Denominator of the sampling rate ratio
+ * @param in_rate Input sampling rate rounded to the nearest integer (in Hz).
+ * @param out_rate Output sampling rate rounded to the nearest integer (in Hz).
+ * @param quality Resampling quality between 0 and 10, where 0 has poor quality
+ * and 10 has very high quality.
+ * @return Newly created resampler state
+ * @retval NULL Error: not enough memory
+ */
+SpeexResamplerState *speex_resampler_init_frac(spx_uint32_t nb_channels, 
+                                               spx_uint32_t ratio_num, 
+                                               spx_uint32_t ratio_den, 
+                                               spx_uint32_t in_rate, 
+                                               spx_uint32_t out_rate, 
+                                               int quality,
+                                               int *err);
+
+/** Destroy a resampler state.
+ * @param st Resampler state
+ */
+void speex_resampler_destroy(SpeexResamplerState *st);
+
+/** Resample a float array. The input and output buffers must *not* overlap.
+ * @param st Resampler state
+ * @param channel_index Index of the channel to process for the multi-channel 
+ * base (0 otherwise)
+ * @param in Input buffer
+ * @param in_len Number of input samples in the input buffer. Returns the 
+ * number of samples processed
+ * @param out Output buffer
+ * @param out_len Size of the output buffer. Returns the number of samples written
+ */
+int speex_resampler_process_float(SpeexResamplerState *st, 
+                                   spx_uint32_t channel_index, 
+                                   const float *in, 
+                                   spx_uint32_t *in_len, 
+                                   float *out, 
+                                   spx_uint32_t *out_len);
+
+/** Resample an int array. The input and output buffers must *not* overlap.
+ * @param st Resampler state
+ * @param channel_index Index of the channel to process for the multi-channel 
+ * base (0 otherwise)
+ * @param in Input buffer
+ * @param in_len Number of input samples in the input buffer. Returns the number
+ * of samples processed
+ * @param out Output buffer
+ * @param out_len Size of the output buffer. Returns the number of samples written
+ */
+int speex_resampler_process_int(SpeexResamplerState *st, 
+                                 spx_uint32_t channel_index, 
+                                 const spx_int16_t *in, 
+                                 spx_uint32_t *in_len, 
+                                 spx_int16_t *out, 
+                                 spx_uint32_t *out_len);
+
+/** Resample an interleaved float array. The input and output buffers must *not* overlap.
+ * @param st Resampler state
+ * @param in Input buffer
+ * @param in_len Number of input samples in the input buffer. Returns the number
+ * of samples processed. This is all per-channel.
+ * @param out Output buffer
+ * @param out_len Size of the output buffer. Returns the number of samples written.
+ * This is all per-channel.
+ */
+int speex_resampler_process_interleaved_float(SpeexResamplerState *st, 
+                                               const float *in, 
+                                               spx_uint32_t *in_len, 
+                                               float *out, 
+                                               spx_uint32_t *out_len);
+
+/** Resample an interleaved int array. The input and output buffers must *not* overlap.
+ * @param st Resampler state
+ * @param in Input buffer
+ * @param in_len Number of input samples in the input buffer. Returns the number
+ * of samples processed. This is all per-channel.
+ * @param out Output buffer
+ * @param out_len Size of the output buffer. Returns the number of samples written.
+ * This is all per-channel.
+ */
+int speex_resampler_process_interleaved_int(SpeexResamplerState *st, 
+                                             const spx_int16_t *in, 
+                                             spx_uint32_t *in_len, 
+                                             spx_int16_t *out, 
+                                             spx_uint32_t *out_len);
+
+/** Set (change) the input/output sampling rates (integer value).
+ * @param st Resampler state
+ * @param in_rate Input sampling rate (integer number of Hz).
+ * @param out_rate Output sampling rate (integer number of Hz).
+ */
+int speex_resampler_set_rate(SpeexResamplerState *st, 
+                              spx_uint32_t in_rate, 
+                              spx_uint32_t out_rate);
+
+/** Get the current input/output sampling rates (integer value).
+ * @param st Resampler state
+ * @param in_rate Input sampling rate (integer number of Hz) copied.
+ * @param out_rate Output sampling rate (integer number of Hz) copied.
+ */
+void speex_resampler_get_rate(SpeexResamplerState *st, 
+                              spx_uint32_t *in_rate, 
+                              spx_uint32_t *out_rate);
+
+/** Set (change) the input/output sampling rates and resampling ratio 
+ * (fractional values in Hz supported).
+ * @param st Resampler state
+ * @param ratio_num Numerator of the sampling rate ratio
+ * @param ratio_den Denominator of the sampling rate ratio
+ * @param in_rate Input sampling rate rounded to the nearest integer (in Hz).
+ * @param out_rate Output sampling rate rounded to the nearest integer (in Hz).
+ */
+int speex_resampler_set_rate_frac(SpeexResamplerState *st, 
+                                   spx_uint32_t ratio_num, 
+                                   spx_uint32_t ratio_den, 
+                                   spx_uint32_t in_rate, 
+                                   spx_uint32_t out_rate);
+
+/** Get the current resampling ratio. This will be reduced to the least
+ * common denominator.
+ * @param st Resampler state
+ * @param ratio_num Numerator of the sampling rate ratio copied
+ * @param ratio_den Denominator of the sampling rate ratio copied
+ */
+void speex_resampler_get_ratio(SpeexResamplerState *st, 
+                               spx_uint32_t *ratio_num, 
+                               spx_uint32_t *ratio_den);
+
+/** Set (change) the conversion quality.
+ * @param st Resampler state
+ * @param quality Resampling quality between 0 and 10, where 0 has poor 
+ * quality and 10 has very high quality.
+ */
+int speex_resampler_set_quality(SpeexResamplerState *st, 
+                                 int quality);
+
+/** Get the conversion quality.
+ * @param st Resampler state
+ * @param quality Resampling quality between 0 and 10, where 0 has poor 
+ * quality and 10 has very high quality.
+ */
+void speex_resampler_get_quality(SpeexResamplerState *st, 
+                                 int *quality);
+
+/** Set (change) the input stride.
+ * @param st Resampler state
+ * @param stride Input stride
+ */
+void speex_resampler_set_input_stride(SpeexResamplerState *st, 
+                                      spx_uint32_t stride);
+
+/** Get the input stride.
+ * @param st Resampler state
+ * @param stride Input stride copied
+ */
+void speex_resampler_get_input_stride(SpeexResamplerState *st, 
+                                      spx_uint32_t *stride);
+
+/** Set (change) the output stride.
+ * @param st Resampler state
+ * @param stride Output stride
+ */
+void speex_resampler_set_output_stride(SpeexResamplerState *st, 
+                                      spx_uint32_t stride);
+
+/** Get the output stride.
+ * @param st Resampler state copied
+ * @param stride Output stride
+ */
+void speex_resampler_get_output_stride(SpeexResamplerState *st, 
+                                      spx_uint32_t *stride);
+
+/** Get the latency introduced by the resampler measured in input samples.
+ * @param st Resampler state
+ */
+int speex_resampler_get_input_latency(SpeexResamplerState *st);
+
+/** Get the latency introduced by the resampler measured in output samples.
+ * @param st Resampler state
+ */
+int speex_resampler_get_output_latency(SpeexResamplerState *st);
+
+/** Make sure that the first samples to go out of the resamplers don't have 
+ * leading zeros. This is only useful before starting to use a newly created 
+ * resampler. It is recommended to use that when resampling an audio file, as
+ * it will generate a file with the same length. For real-time processing,
+ * it is probably easier not to use this call (so that the output duration
+ * is the same for the first frame).
+ * @param st Resampler state
+ */
+int speex_resampler_skip_zeros(SpeexResamplerState *st);
+
+/** Reset a resampler so a new (unrelated) stream can be processed.
+ * @param st Resampler state
+ */
+int speex_resampler_reset_mem(SpeexResamplerState *st);
+
+/** Returns the English meaning for an error code
+ * @param err Error code
+ * @return English string
+ */
+const char *speex_resampler_strerror(int err);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif
--- /dev/null
+++ b/src/stack_alloc.h
@@ -1,0 +1,116 @@
+/* Copyright (C) 2002 Jean-Marc Valin */
+/**
+   @file stack_alloc.h
+   @brief Temporary memory allocation on stack
+*/
+/*
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions
+   are met:
+
+   - Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+
+   - Redistributions in binary form must reproduce the above copyright
+   notice, this list of conditions and the following disclaimer in the
+   documentation and/or other materials provided with the distribution.
+
+   - Neither the name of the Xiph.org Foundation nor the names of its
+   contributors may be used to endorse or promote products derived from
+   this software without specific prior written permission.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR
+   CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifndef STACK_ALLOC_H
+#define STACK_ALLOC_H
+
+#ifdef WIN32
+# include <malloc.h>
+# ifndef alloca
+#   define alloca(_x) _alloca(_x);
+# endif
+#else
+#ifdef HAVE_ALLOCA_H
+#  include <alloca.h>
+# else
+#  include <stdlib.h>
+# endif
+#endif
+
+/**
+ * @def ALIGN(stack, size)
+ *
+ * Aligns the stack to a 'size' boundary
+ *
+ * @param stack Stack
+ * @param size  New size boundary
+ */
+
+/**
+ * @def PUSH(stack, size, type)
+ *
+ * Allocates 'size' elements of type 'type' on the stack
+ *
+ * @param stack Stack
+ * @param size  Number of elements
+ * @param type  Type of element
+ */
+
+/**
+ * @def VARDECL(var)
+ *
+ * Declare variable on stack
+ *
+ * @param var Variable to declare
+ */
+
+/**
+ * @def ALLOC(var, size, type)
+ *
+ * Allocate 'size' elements of 'type' on stack
+ *
+ * @param var  Name of variable to allocate
+ * @param size Number of elements
+ * @param type Type of element
+ */
+
+#ifdef ENABLE_VALGRIND
+
+#include <valgrind/memcheck.h>
+
+#define ALIGN(stack, size) ((stack) += ((size) - (long)(stack)) & ((size) - 1))
+
+#define PUSH(stack, size, type) (VALGRIND_MAKE_NOACCESS(stack, 1000),ALIGN((stack),sizeof(type)),VALGRIND_MAKE_WRITABLE(stack, ((size)*sizeof(type))),(stack)+=((size)*sizeof(type)),(type*)((stack)-((size)*sizeof(type))))
+
+#else
+
+#define ALIGN(stack, size) ((stack) += ((size) - (long)(stack)) & ((size) - 1))
+
+#define PUSH(stack, size, type) (ALIGN((stack),sizeof(type)),(stack)+=((size)*sizeof(type)),(type*)((stack)-((size)*sizeof(type))))
+
+#endif
+
+#if defined(VAR_ARRAYS)
+#define VARDECL(var)
+#define ALLOC(var, size, type) type var[size]
+#elif defined(USE_ALLOCA)
+#define VARDECL(var) var
+#define ALLOC(var, size, type) var = alloca(sizeof(type)*(size))
+#else
+#define VARDECL(var) var
+#define ALLOC(var, size, type) var = PUSH(stack, size, type)
+#endif
+
+
+#endif