Update Opus driver to 1.1.2

And opusfile to 0.7.

81 files changed, 6942 insertions, 2983 deletions

diff --git a/drivers/opus/celt/_kiss_fft_guts.h b/drivers/opus/celt/_kiss_fft_guts.h
index 2a4ee744ef..a41f9a1a05 100644
--- a/drivers/opus/celt/_kiss_fft_guts.h
+++ b/drivers/opus/celt/_kiss_fft_guts.h
@@ -65,10 +65,6 @@
       do{ (m).r = ADD32(S_MUL((a).r,(b).r) , S_MUL((a).i,(b).i)); \
           (m).i = SUB32(S_MUL((a).i,(b).r) , S_MUL((a).r,(b).i)); }while(0)
 
-#   define C_MUL4(m,a,b) \
-      do{ (m).r = SHR32(SUB32(S_MUL((a).r,(b).r) , S_MUL((a).i,(b).i)),2); \
-          (m).i = SHR32(ADD32(S_MUL((a).r,(b).i) , S_MUL((a).i,(b).r)),2); }while(0)
-
 #   define C_MULBYSCALAR( c, s ) \
       do{ (c).r =  S_MUL( (c).r , s ) ;\
           (c).i =  S_MUL( (c).i , s ) ; }while(0)
@@ -95,14 +91,17 @@
     }while(0)
 
 #if defined(OPUS_ARM_INLINE_ASM)
-#include "arm/kiss_fft_armv4.h"
+#include "opus/celt/arm/kiss_fft_armv4.h"
 #endif
 
 #if defined(OPUS_ARM_INLINE_EDSP)
-#include "arm/kiss_fft_armv5e.h"
+#include "opus/celt/arm/kiss_fft_armv5e.h"
+#endif
+#if defined(MIPSr1_ASM)
+#include "opus/celt/mips/kiss_fft_mipsr1.h"
 #endif
 
-#else  /* not OPUS_FIXED_POINT*/
+#else  /* not FIXED_POINT*/
 
 #   define S_MUL(a,b) ( (a)*(b) )
 #define C_MUL(m,a,b) \
@@ -153,8 +152,8 @@
 #endif /* C_ADD defined */
 
 #ifdef OPUS_FIXED_POINT
-#  define KISS_FFT_COS(phase)  TRIG_UPSCALE*floor(MIN(32767,MAX(-32767,.5+32768 * cos (phase))))
-#  define KISS_FFT_SIN(phase)  TRIG_UPSCALE*floor(MIN(32767,MAX(-32767,.5+32768 * sin (phase))))
+/*#  define KISS_FFT_COS(phase)  TRIG_UPSCALE*floor(MIN(32767,MAX(-32767,.5+32768 * cos (phase))))
+#  define KISS_FFT_SIN(phase)  TRIG_UPSCALE*floor(MIN(32767,MAX(-32767,.5+32768 * sin (phase))))*/
 #  define KISS_FFT_COS(phase)  floor(.5+TWID_MAX*cos (phase))
 #  define KISS_FFT_SIN(phase)  floor(.5+TWID_MAX*sin (phase))
 #  define HALF_OF(x) ((x)>>1)
diff --git a/drivers/opus/celt/arch.h b/drivers/opus/celt/arch.h
index d964f8d90c..7fb036e9fe 100644
--- a/drivers/opus/celt/arch.h
+++ b/drivers/opus/celt/arch.h
@@ -69,11 +69,8 @@ static OPUS_INLINE void _celt_fatal(const char *str, const char *file, int line)
 
 #define IMUL32(a,b) ((a)*(b))
 
-#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))   /**< Minimum 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))   /**< Minimum 32-bit value.   */
 #define MAX32(a,b) ((a) > (b) ? (a) : (b))   /**< Maximum 32-bit value.   */
 #define IMIN(a,b) ((a) < (b) ? (a) : (b))   /**< Minimum int value.   */
@@ -108,6 +105,13 @@ typedef opus_val32 celt_ener;
 #define SCALEIN(a)      (a)
 #define SCALEOUT(a)     (a)
 
+#define ABS16(x) ((x) < 0 ? (-(x)) : (x))
+#define ABS32(x) ((x) < 0 ? (-(x)) : (x))
+
+static OPUS_INLINE opus_int16 SAT16(opus_int32 x) {
+   return x > 32767 ? 32767 : x < -32768 ? -32768 : (opus_int16)x;
+}
+
 #ifdef FIXED_DEBUG
 #include "opus/celt/fixed_debug.h"
 #else
@@ -115,9 +119,9 @@ typedef opus_val32 celt_ener;
 #include "opus/celt/fixed_generic.h"
 
 #ifdef OPUS_ARM_INLINE_EDSP
-#include "arm/fixed_armv5e.h"
+#include "opus/celt/arm/fixed_armv5e.h"
 #elif defined (OPUS_ARM_INLINE_ASM)
-#include "arm/fixed_armv4.h"
+#include "opus/celt/arm/fixed_armv4.h"
 #elif defined (BFIN_ASM)
 #include "fixed_bfin.h"
 #elif defined (TI_C5X_ASM)
@@ -128,7 +132,7 @@ typedef opus_val32 celt_ener;
 
 #endif
 
-#else /* OPUS_FIXED_POINT */
+#else /* FIXED_POINT */
 
 typedef float opus_val16;
 typedef float opus_val32;
@@ -137,6 +141,22 @@ typedef float celt_sig;
 typedef float celt_norm;
 typedef float celt_ener;
 
+#ifdef FLOAT_APPROX
+/* This code should reliably detect NaN/inf even when -ffast-math is used.
+   Assumes IEEE 754 format. */
+static OPUS_INLINE int celt_isnan(float x)
+{
+   union {float f; opus_uint32 i;} in;
+   in.f = x;
+   return ((in.i>>23)&0xFF)==0xFF && (in.i&0x007FFFFF)!=0;
+}
+#else
+#ifdef __FAST_MATH__
+#error Cannot build libopus with -ffast-math unless FLOAT_APPROX is defined. This could result in crashes on extreme (e.g. NaN) input
+#endif
+#define celt_isnan(x) ((x)!=(x))
+#endif
+
 #define Q15ONE 1.0f
 
 #define NORM_SCALING 1.f
@@ -146,6 +166,10 @@ typedef float celt_ener;
 #define VERY_LARGE16 1e15f
 #define Q15_ONE ((opus_val16)1.f)
 
+/* This appears to be the same speed as C99's fabsf() but it's more portable. */
+#define ABS16(x) ((float)fabs(x))
+#define ABS32(x) ((float)fabs(x))
+
 #define QCONST16(x,bits) (x)
 #define QCONST32(x,bits) (x)
 
@@ -184,6 +208,7 @@ typedef float celt_ener;
 #define MULT32_32_Q31(a,b)     ((a)*(b))
 
 #define MAC16_32_Q15(c,a,b)     ((c)+(a)*(b))
+#define MAC16_32_Q16(c,a,b)     ((c)+(a)*(b))
 
 #define MULT16_16_Q11_32(a,b)     ((a)*(b))
 #define MULT16_16_Q11(a,b)     ((a)*(b))
@@ -201,7 +226,9 @@ typedef float celt_ener;
 #define SCALEIN(a)      ((a)*CELT_SIG_SCALE)
 #define SCALEOUT(a)     ((a)*(1/CELT_SIG_SCALE))
 
-#endif /* !OPUS_FIXED_POINT */
+#define SIG2WORD16(x) (x)
+
+#endif /* !FIXED_POINT */
 
 #ifndef GLOBAL_STACK_SIZE
 #ifdef OPUS_FIXED_POINT
diff --git a/drivers/opus/celt/arm/arm2gnu.pl b/drivers/opus/celt/arm/arm2gnu.pl
index eab42efa2b..6c922ac819 100755
--- a/drivers/opus/celt/arm/arm2gnu.pl
+++ b/drivers/opus/celt/arm/arm2gnu.pl
@@ -1,7 +1,33 @@
 #!/usr/bin/perl
+# Copyright (C) 2002-2013 Xiph.org Foundation
+#
+# 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.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+# OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+# EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+# PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+# PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+# LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+# NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+# SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 my $bigend;  # little/big endian
 my $nxstack;
+my $apple = 0;
+my $symprefix = "";
 
 $nxstack = 0;
 
@@ -10,11 +36,16 @@ eval 'exec /usr/local/bin/perl -S $0 ${1+"$@"}'
 
 while ($ARGV[0] =~ /^-/) {
     $_ = shift;
-  last if /^--/;
-    if (/^-n/) {
+  last if /^--$/;
+    if (/^-n$/) {
     $nflag++;
     next;
     }
+    if (/^--apple$/) {
+        $apple = 1;
+        $symprefix = "_";
+        next;
+    }
     die "I don't recognize this switch: $_\\n";
 }
 $printit++ unless $nflag;
@@ -25,6 +56,8 @@ $n=0;
 $thumb = 0;     # ARM mode by default, not Thumb.
 @proc_stack = ();
 
+printf ("    .syntax unified\n");
+
 LINE:
 while (<>) {
 
@@ -53,7 +86,7 @@ while (<>) {
     s/\bINCLUDE[ \t]*([^ \t\n]+)/.include \"$1\"/;
     s/\bGET[ \t]*([^ \t\n]+)/.include \"${ my $x=$1; $x =~ s|\.s|-gnu.S|; \$x }\"/;
     s/\bIMPORT\b/.extern/;
-    s/\bEXPORT\b/.global/;
+    s/\bEXPORT\b\s*/.global $symprefix/;
     s/^(\s+)\[/$1IF/;
     s/^(\s+)\|/$1ELSE/;
     s/^(\s+)\]/$1ENDIF/;
@@ -109,7 +142,7 @@ while (<>) {
             # won't match the original source file (we could use the .line
             # directive, which is documented to be obsolete, but then gdb will
             # show the wrong line in the translated source file).
-            s/$/;   .arch armv7-a\n   .fpu neon\n   .object_arch armv4t/;
+            s/$/;   .arch armv7-a\n   .fpu neon\n   .object_arch armv4t/ unless ($apple);
         }
     }
 
@@ -131,9 +164,13 @@ while (<>) {
         $prefix = "";
         if ($proc)
         {
-            $prefix = $prefix.sprintf("\t.type\t%s, %%function; ",$proc);
+            $prefix = $prefix.sprintf("\t.type\t%s, %%function; ",$proc) unless ($apple);
+            # Make sure we $prefix isn't empty here (for the $apple case).
+            # We handle mangling the label here, make sure it doesn't match
+            # the label handling below (if $prefix would be empty).
+            $prefix = "; ";
             push(@proc_stack, $proc);
-            s/^[A-Za-z_\.]\w+/$&:/;
+            s/^[A-Za-z_\.]\w+/$symprefix$&:/;
         }
         $prefix = $prefix."\t.thumb_func; " if ($thumb);
         s/\bPROC\b/@ $&/;
@@ -146,7 +183,7 @@ while (<>) {
         my $proc;
         s/\bENDP\b/@ $&/;
         $proc = pop(@proc_stack);
-        $_ = "\t.size $proc, .-$proc".$_ if ($proc);
+        $_ = "\t.size $proc, .-$proc".$_ if ($proc && !$apple);
     }
     s/\bSUBT\b/@ $&/;
     s/\bDATA\b/@ $&/;   # DATA directive is deprecated -- Asm guide, p.7-25
@@ -311,6 +348,6 @@ while (<>) {
 }
 #If we had a code section, mark that this object doesn't need an executable
 # stack.
-if ($nxstack) {
+if ($nxstack && !$apple) {
     printf ("    .section\t.note.GNU-stack,\"\",\%\%progbits\n");
 }
diff --git a/drivers/opus/celt/arm/arm_celt_map.c b/drivers/opus/celt/arm/arm_celt_map.c
index 31e7d5b319..92c83c1763 100644
--- a/drivers/opus/celt/arm/arm_celt_map.c
+++ b/drivers/opus/celt/arm/arm_celt_map.c
@@ -24,16 +24,15 @@
    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
-
-#ifdef OPUS_ENABLED
 #include "opus/opus_config.h"
-#endif
 
 #include "opus/celt/pitch.h"
+#include "opus/celt/kiss_fft.h"
+#include "opus/celt/mdct.h"
 
 #if defined(OPUS_HAVE_RTCD)
 
-# if defined(OPUS_FIXED_POINT)
+# if defined(FIXED_POINT)
 opus_val32 (*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *,
     const opus_val16 *, opus_val32 *, int , int) = {
   celt_pitch_xcorr_c,               /* ARMv4 */
@@ -41,9 +40,79 @@ opus_val32 (*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *,
   MAY_HAVE_MEDIA(celt_pitch_xcorr), /* Media */
   MAY_HAVE_NEON(celt_pitch_xcorr)   /* NEON */
 };
-# else
-#  error "Floating-point implementation is not supported by ARM asm yet." \
- "Reconfigure with --disable-rtcd or send patches."
-# endif
+# else /* !FIXED_POINT */
+#  if defined(OPUS_ARM_MAY_HAVE_NEON_INTR)
+void (*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *,
+    const opus_val16 *, opus_val32 *, int, int) = {
+  celt_pitch_xcorr_c,              /* ARMv4 */
+  celt_pitch_xcorr_c,              /* EDSP */
+  celt_pitch_xcorr_c,              /* Media */
+  celt_pitch_xcorr_float_neon      /* Neon */
+};
+#  endif
+# endif /* FIXED_POINT */
+
+# if defined(OPUS_ARM_MAY_HAVE_NEON_INTR)
+#  if defined(HAVE_ARM_NE10)
+#   if defined(CUSTOM_MODES)
+int (*const OPUS_FFT_ALLOC_ARCH_IMPL[OPUS_ARCHMASK+1])(kiss_fft_state *st) = {
+   opus_fft_alloc_arch_c,        /* ARMv4 */
+   opus_fft_alloc_arch_c,        /* EDSP */
+   opus_fft_alloc_arch_c,        /* Media */
+   opus_fft_alloc_arm_neon       /* Neon with NE10 library support */
+};
+
+void (*const OPUS_FFT_FREE_ARCH_IMPL[OPUS_ARCHMASK+1])(kiss_fft_state *st) = {
+   opus_fft_free_arch_c,         /* ARMv4 */
+   opus_fft_free_arch_c,         /* EDSP */
+   opus_fft_free_arch_c,         /* Media */
+   opus_fft_free_arm_neon        /* Neon with NE10 */
+};
+#   endif /* CUSTOM_MODES */
+
+void (*const OPUS_FFT[OPUS_ARCHMASK+1])(const kiss_fft_state *cfg,
+                                        const kiss_fft_cpx *fin,
+                                        kiss_fft_cpx *fout) = {
+   opus_fft_c,                   /* ARMv4 */
+   opus_fft_c,                   /* EDSP */
+   opus_fft_c,                   /* Media */
+   opus_fft_neon                 /* Neon with NE10 */
+};
+
+void (*const OPUS_IFFT[OPUS_ARCHMASK+1])(const kiss_fft_state *cfg,
+                                         const kiss_fft_cpx *fin,
+                                         kiss_fft_cpx *fout) = {
+   opus_ifft_c,                   /* ARMv4 */
+   opus_ifft_c,                   /* EDSP */
+   opus_ifft_c,                   /* Media */
+   opus_ifft_neon                 /* Neon with NE10 */
+};
+
+void (*const CLT_MDCT_FORWARD_IMPL[OPUS_ARCHMASK+1])(const mdct_lookup *l,
+                                                     kiss_fft_scalar *in,
+                                                     kiss_fft_scalar * OPUS_RESTRICT out,
+                                                     const opus_val16 *window,
+                                                     int overlap, int shift,
+                                                     int stride, int arch) = {
+   clt_mdct_forward_c,           /* ARMv4 */
+   clt_mdct_forward_c,           /* EDSP */
+   clt_mdct_forward_c,           /* Media */
+   clt_mdct_forward_neon         /* Neon with NE10 */
+};
+
+void (*const CLT_MDCT_BACKWARD_IMPL[OPUS_ARCHMASK+1])(const mdct_lookup *l,
+                                                      kiss_fft_scalar *in,
+                                                      kiss_fft_scalar * OPUS_RESTRICT out,
+                                                      const opus_val16 *window,
+                                                      int overlap, int shift,
+                                                      int stride, int arch) = {
+   clt_mdct_backward_c,           /* ARMv4 */
+   clt_mdct_backward_c,           /* EDSP */
+   clt_mdct_backward_c,           /* Media */
+   clt_mdct_backward_neon         /* Neon with NE10 */
+};
+
+#  endif /* HAVE_ARM_NE10 */
+# endif /* OPUS_ARM_MAY_HAVE_NEON_INTR */
 
-#endif
+#endif /* OPUS_HAVE_RTCD */
diff --git a/drivers/opus/celt/arm/armcpu.c b/drivers/opus/celt/arm/armcpu.c
index fb7f2421fe..ac8ad2ec14 100644
--- a/drivers/opus/celt/arm/armcpu.c
+++ b/drivers/opus/celt/arm/armcpu.c
@@ -26,10 +26,7 @@
 */
 
 /* Original code from libtheora modified to suit to Opus */
-
-#ifdef OPUS_ENABLED
 #include "opus/opus_config.h"
-#endif
 
 #ifdef OPUS_HAVE_RTCD
 
@@ -73,7 +70,7 @@ static OPUS_INLINE opus_uint32 opus_cpu_capabilities(void){
   __except(GetExceptionCode()==EXCEPTION_ILLEGAL_INSTRUCTION){
     /*Ignore exception.*/
   }
-#   if defined(OPUS_ARM_MAY_HAVE_NEON)
+#   if defined(OPUS_ARM_MAY_HAVE_NEON) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR)
   __try{
     /*VORR q0,q0,q0*/
     __emit(0xF2200150);
@@ -107,7 +104,7 @@ opus_uint32 opus_cpu_capabilities(void)
 
     while(fgets(buf, 512, cpuinfo) != NULL)
     {
-# if defined(OPUS_ARM_MAY_HAVE_EDSP) || defined(OPUS_ARM_MAY_HAVE_NEON)
+# if defined(OPUS_ARM_MAY_HAVE_EDSP) || defined(OPUS_ARM_MAY_HAVE_NEON) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR)
       /* Search for edsp and neon flag */
       if(memcmp(buf, "Features", 8) == 0)
       {
@@ -118,7 +115,7 @@ opus_uint32 opus_cpu_capabilities(void)
           flags |= OPUS_CPU_ARM_EDSP;
 #  endif
 
-#  if defined(OPUS_ARM_MAY_HAVE_NEON)
+#  if defined(OPUS_ARM_MAY_HAVE_NEON) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR)
         p = strstr(buf, " neon");
         if(p != NULL && (p[5] == ' ' || p[5] == '\n'))
           flags |= OPUS_CPU_ARM_NEON;
diff --git a/drivers/opus/celt/arm/celt_ne10_fft.c b/drivers/opus/celt/arm/celt_ne10_fft.c
new file mode 100644
index 0000000000..e57d23ee1d
--- /dev/null
+++ b/drivers/opus/celt/arm/celt_ne10_fft.c
@@ -0,0 +1,172 @@
+/* Copyright (c) 2015 Xiph.Org Foundation
+   Written by Viswanath Puttagunta */
+/**
+   @file celt_ne10_fft.c
+   @brief ARM Neon optimizations for fft using NE10 library
+ */
+
+/*
+   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.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifndef SKIP_CONFIG_H
+#include "opus/opus_config.h"
+#endif
+
+#include <NE10_init.h>
+#include <NE10_dsp.h>
+#include "opus/celt/os_support.h"
+#include "opus/celt/kiss_fft.h"
+#include "opus/celt/stack_alloc.h"
+
+#if !defined(FIXED_POINT)
+# define NE10_FFT_ALLOC_C2C_TYPE_NEON ne10_fft_alloc_c2c_float32_neon
+# define NE10_FFT_CFG_TYPE_T ne10_fft_cfg_float32_t
+# define NE10_FFT_STATE_TYPE_T ne10_fft_state_float32_t
+# define NE10_FFT_DESTROY_C2C_TYPE ne10_fft_destroy_c2c_float32
+# define NE10_FFT_CPX_TYPE_T ne10_fft_cpx_float32_t
+# define NE10_FFT_C2C_1D_TYPE_NEON ne10_fft_c2c_1d_float32_neon
+#else
+# define NE10_FFT_ALLOC_C2C_TYPE_NEON(nfft) ne10_fft_alloc_c2c_int32_neon(nfft)
+# define NE10_FFT_CFG_TYPE_T ne10_fft_cfg_int32_t
+# define NE10_FFT_STATE_TYPE_T ne10_fft_state_int32_t
+# define NE10_FFT_DESTROY_C2C_TYPE ne10_fft_destroy_c2c_int32
+# define NE10_FFT_DESTROY_C2C_TYPE ne10_fft_destroy_c2c_int32
+# define NE10_FFT_CPX_TYPE_T ne10_fft_cpx_int32_t
+# define NE10_FFT_C2C_1D_TYPE_NEON ne10_fft_c2c_1d_int32_neon
+#endif
+
+#if defined(CUSTOM_MODES)
+
+/* nfft lengths in NE10 that support scaled fft */
+# define NE10_FFTSCALED_SUPPORT_MAX 4
+static const int ne10_fft_scaled_support[NE10_FFTSCALED_SUPPORT_MAX] = {
+   480, 240, 120, 60
+};
+
+int opus_fft_alloc_arm_neon(kiss_fft_state *st)
+{
+   int i;
+   size_t memneeded = sizeof(struct arch_fft_state);
+
+   st->arch_fft = (arch_fft_state *)opus_alloc(memneeded);
+   if (!st->arch_fft)
+      return -1;
+
+   for (i = 0; i < NE10_FFTSCALED_SUPPORT_MAX; i++) {
+      if(st->nfft == ne10_fft_scaled_support[i])
+         break;
+   }
+   if (i == NE10_FFTSCALED_SUPPORT_MAX) {
+      /* This nfft length (scaled fft) is not supported in NE10 */
+      st->arch_fft->is_supported = 0;
+      st->arch_fft->priv = NULL;
+   }
+   else {
+      st->arch_fft->is_supported = 1;
+      st->arch_fft->priv = (void *)NE10_FFT_ALLOC_C2C_TYPE_NEON(st->nfft);
+      if (st->arch_fft->priv == NULL) {
+         return -1;
+      }
+   }
+   return 0;
+}
+
+void opus_fft_free_arm_neon(kiss_fft_state *st)
+{
+   NE10_FFT_CFG_TYPE_T cfg;
+
+   if (!st->arch_fft)
+      return;
+
+   cfg = (NE10_FFT_CFG_TYPE_T)st->arch_fft->priv;
+   if (cfg)
+      NE10_FFT_DESTROY_C2C_TYPE(cfg);
+   opus_free(st->arch_fft);
+}
+#endif
+
+void opus_fft_neon(const kiss_fft_state *st,
+                   const kiss_fft_cpx *fin,
+                   kiss_fft_cpx *fout)
+{
+   NE10_FFT_STATE_TYPE_T state;
+   NE10_FFT_CFG_TYPE_T cfg = &state;
+   VARDECL(NE10_FFT_CPX_TYPE_T, buffer);
+   SAVE_STACK;
+   ALLOC(buffer, st->nfft, NE10_FFT_CPX_TYPE_T);
+
+   if (!st->arch_fft->is_supported) {
+      /* This nfft length (scaled fft) not supported in NE10 */
+      opus_fft_c(st, fin, fout);
+   }
+   else {
+      memcpy((void *)cfg, st->arch_fft->priv, sizeof(NE10_FFT_STATE_TYPE_T));
+      state.buffer = (NE10_FFT_CPX_TYPE_T *)&buffer[0];
+#if !defined(FIXED_POINT)
+      state.is_forward_scaled = 1;
+
+      NE10_FFT_C2C_1D_TYPE_NEON((NE10_FFT_CPX_TYPE_T *)fout,
+                                (NE10_FFT_CPX_TYPE_T *)fin,
+                                cfg, 0);
+#else
+      NE10_FFT_C2C_1D_TYPE_NEON((NE10_FFT_CPX_TYPE_T *)fout,
+                                (NE10_FFT_CPX_TYPE_T *)fin,
+                                cfg, 0, 1);
+#endif
+   }
+   RESTORE_STACK;
+}
+
+void opus_ifft_neon(const kiss_fft_state *st,
+                    const kiss_fft_cpx *fin,
+                    kiss_fft_cpx *fout)
+{
+   NE10_FFT_STATE_TYPE_T state;
+   NE10_FFT_CFG_TYPE_T cfg = &state;
+   VARDECL(NE10_FFT_CPX_TYPE_T, buffer);
+   SAVE_STACK;
+   ALLOC(buffer, st->nfft, NE10_FFT_CPX_TYPE_T);
+
+   if (!st->arch_fft->is_supported) {
+      /* This nfft length (scaled fft) not supported in NE10 */
+      opus_ifft_c(st, fin, fout);
+   }
+   else {
+      memcpy((void *)cfg, st->arch_fft->priv, sizeof(NE10_FFT_STATE_TYPE_T));
+      state.buffer = (NE10_FFT_CPX_TYPE_T *)&buffer[0];
+#if !defined(FIXED_POINT)
+      state.is_backward_scaled = 0;
+
+      NE10_FFT_C2C_1D_TYPE_NEON((NE10_FFT_CPX_TYPE_T *)fout,
+                                (NE10_FFT_CPX_TYPE_T *)fin,
+                                cfg, 1);
+#else
+      NE10_FFT_C2C_1D_TYPE_NEON((NE10_FFT_CPX_TYPE_T *)fout,
+                                (NE10_FFT_CPX_TYPE_T *)fin,
+                                cfg, 1, 0);
+#endif
+   }
+   RESTORE_STACK;
+}
diff --git a/drivers/opus/celt/arm/celt_ne10_mdct.c b/drivers/opus/celt/arm/celt_ne10_mdct.c
new file mode 100644
index 0000000000..eb407b674f
--- /dev/null
+++ b/drivers/opus/celt/arm/celt_ne10_mdct.c
@@ -0,0 +1,256 @@
+/* Copyright (c) 2015 Xiph.Org Foundation
+   Written by Viswanath Puttagunta */
+/**
+   @file celt_ne10_mdct.c
+   @brief ARM Neon optimizations for mdct using NE10 library
+ */
+
+/*
+   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.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifndef SKIP_CONFIG_H
+#include "opus/opus_config.h"
+#endif
+
+#include "opus/celt/kiss_fft.h"
+#include "opus/celt/_kiss_fft_guts.h"
+#include "opus/celt/mdct.h"
+#include "opus/celt/stack_alloc.h"
+
+void clt_mdct_forward_neon(const mdct_lookup *l,
+                           kiss_fft_scalar *in,
+                           kiss_fft_scalar * OPUS_RESTRICT out,
+                           const opus_val16 *window,
+                           int overlap, int shift, int stride, int arch)
+{
+   int i;
+   int N, N2, N4;
+   VARDECL(kiss_fft_scalar, f);
+   VARDECL(kiss_fft_cpx, f2);
+   const kiss_fft_state *st = l->kfft[shift];
+   const kiss_twiddle_scalar *trig;
+
+   SAVE_STACK;
+
+   N = l->n;
+   trig = l->trig;
+   for (i=0;i<shift;i++)
+   {
+      N >>= 1;
+      trig += N;
+   }
+   N2 = N>>1;
+   N4 = N>>2;
+
+   ALLOC(f, N2, kiss_fft_scalar);
+   ALLOC(f2, N4, kiss_fft_cpx);
+
+   /* Consider the input to be composed of four blocks: [a, b, c, d] */
+   /* Window, shuffle, fold */
+   {
+      /* Temp pointers to make it really clear to the compiler what we're doing */
+      const kiss_fft_scalar * OPUS_RESTRICT xp1 = in+(overlap>>1);
+      const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+N2-1+(overlap>>1);
+      kiss_fft_scalar * OPUS_RESTRICT yp = f;
+      const opus_val16 * OPUS_RESTRICT wp1 = window+(overlap>>1);
+      const opus_val16 * OPUS_RESTRICT wp2 = window+(overlap>>1)-1;
+      for(i=0;i<((overlap+3)>>2);i++)
+      {
+         /* Real part arranged as -d-cR, Imag part arranged as -b+aR*/
+         *yp++ = MULT16_32_Q15(*wp2, xp1[N2]) + MULT16_32_Q15(*wp1,*xp2);
+         *yp++ = MULT16_32_Q15(*wp1, *xp1)    - MULT16_32_Q15(*wp2, xp2[-N2]);
+         xp1+=2;
+         xp2-=2;
+         wp1+=2;
+         wp2-=2;
+      }
+      wp1 = window;
+      wp2 = window+overlap-1;
+      for(;i<N4-((overlap+3)>>2);i++)
+      {
+         /* Real part arranged as a-bR, Imag part arranged as -c-dR */
+         *yp++ = *xp2;
+         *yp++ = *xp1;
+         xp1+=2;
+         xp2-=2;
+      }
+      for(;i<N4;i++)
+      {
+         /* Real part arranged as a-bR, Imag part arranged as -c-dR */
+         *yp++ =  -MULT16_32_Q15(*wp1, xp1[-N2]) + MULT16_32_Q15(*wp2, *xp2);
+         *yp++ = MULT16_32_Q15(*wp2, *xp1)     + MULT16_32_Q15(*wp1, xp2[N2]);
+         xp1+=2;
+         xp2-=2;
+         wp1+=2;
+         wp2-=2;
+      }
+   }
+   /* Pre-rotation */
+   {
+      kiss_fft_scalar * OPUS_RESTRICT yp = f;
+      const kiss_twiddle_scalar *t = &trig[0];
+      for(i=0;i<N4;i++)
+      {
+         kiss_fft_cpx yc;
+         kiss_twiddle_scalar t0, t1;
+         kiss_fft_scalar re, im, yr, yi;
+         t0 = t[i];
+         t1 = t[N4+i];
+         re = *yp++;
+         im = *yp++;
+         yr = S_MUL(re,t0)  -  S_MUL(im,t1);
+         yi = S_MUL(im,t0)  +  S_MUL(re,t1);
+         yc.r = yr;
+         yc.i = yi;
+         f2[i] = yc;
+      }
+   }
+
+   opus_fft(st, f2, (kiss_fft_cpx *)f, arch);
+
+   /* Post-rotate */
+   {
+      /* Temp pointers to make it really clear to the compiler what we're doing */
+      const kiss_fft_cpx * OPUS_RESTRICT fp = (kiss_fft_cpx *)f;
+      kiss_fft_scalar * OPUS_RESTRICT yp1 = out;
+      kiss_fft_scalar * OPUS_RESTRICT yp2 = out+stride*(N2-1);
+      const kiss_twiddle_scalar *t = &trig[0];
+      /* Temp pointers to make it really clear to the compiler what we're doing */
+      for(i=0;i<N4;i++)
+      {
+         kiss_fft_scalar yr, yi;
+         yr = S_MUL(fp->i,t[N4+i]) - S_MUL(fp->r,t[i]);
+         yi = S_MUL(fp->r,t[N4+i]) + S_MUL(fp->i,t[i]);
+         *yp1 = yr;
+         *yp2 = yi;
+         fp++;
+         yp1 += 2*stride;
+         yp2 -= 2*stride;
+      }
+   }
+   RESTORE_STACK;
+}
+
+void clt_mdct_backward_neon(const mdct_lookup *l,
+                            kiss_fft_scalar *in,
+                            kiss_fft_scalar * OPUS_RESTRICT out,
+                            const opus_val16 * OPUS_RESTRICT window,
+                            int overlap, int shift, int stride, int arch)
+{
+   int i;
+   int N, N2, N4;
+   VARDECL(kiss_fft_scalar, f);
+   const kiss_twiddle_scalar *trig;
+   const kiss_fft_state *st = l->kfft[shift];
+
+   N = l->n;
+   trig = l->trig;
+   for (i=0;i<shift;i++)
+   {
+      N >>= 1;
+      trig += N;
+   }
+   N2 = N>>1;
+   N4 = N>>2;
+
+   ALLOC(f, N2, kiss_fft_scalar);
+
+   /* Pre-rotate */
+   {
+      /* Temp pointers to make it really clear to the compiler what we're doing */
+      const kiss_fft_scalar * OPUS_RESTRICT xp1 = in;
+      const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+stride*(N2-1);
+      kiss_fft_scalar * OPUS_RESTRICT yp = f;
+      const kiss_twiddle_scalar * OPUS_RESTRICT t = &trig[0];
+      for(i=0;i<N4;i++)
+      {
+         kiss_fft_scalar yr, yi;
+         yr = S_MUL(*xp2, t[i]) + S_MUL(*xp1, t[N4+i]);
+         yi = S_MUL(*xp1, t[i]) - S_MUL(*xp2, t[N4+i]);
+         yp[2*i] = yr;
+         yp[2*i+1] = yi;
+         xp1+=2*stride;
+         xp2-=2*stride;
+      }
+   }
+
+   opus_ifft(st, (kiss_fft_cpx *)f, (kiss_fft_cpx*)(out+(overlap>>1)), arch);
+
+   /* Post-rotate and de-shuffle from both ends of the buffer at once to make
+      it in-place. */
+   {
+      kiss_fft_scalar * yp0 = out+(overlap>>1);
+      kiss_fft_scalar * yp1 = out+(overlap>>1)+N2-2;
+      const kiss_twiddle_scalar *t = &trig[0];
+      /* Loop to (N4+1)>>1 to handle odd N4. When N4 is odd, the
+         middle pair will be computed twice. */
+      for(i=0;i<(N4+1)>>1;i++)
+      {
+         kiss_fft_scalar re, im, yr, yi;
+         kiss_twiddle_scalar t0, t1;
+         re = yp0[0];
+         im = yp0[1];
+         t0 = t[i];
+         t1 = t[N4+i];
+         /* We'd scale up by 2 here, but instead it's done when mixing the windows */
+         yr = S_MUL(re,t0) + S_MUL(im,t1);
+         yi = S_MUL(re,t1) - S_MUL(im,t0);
+         re = yp1[0];
+         im = yp1[1];
+         yp0[0] = yr;
+         yp1[1] = yi;
+
+         t0 = t[(N4-i-1)];
+         t1 = t[(N2-i-1)];
+         /* We'd scale up by 2 here, but instead it's done when mixing the windows */
+         yr = S_MUL(re,t0) + S_MUL(im,t1);
+         yi = S_MUL(re,t1) - S_MUL(im,t0);
+         yp1[0] = yr;
+         yp0[1] = yi;
+         yp0 += 2;
+         yp1 -= 2;
+      }
+   }
+
+   /* Mirror on both sides for TDAC */
+   {
+      kiss_fft_scalar * OPUS_RESTRICT xp1 = out+overlap-1;
+      kiss_fft_scalar * OPUS_RESTRICT yp1 = out;
+      const opus_val16 * OPUS_RESTRICT wp1 = window;
+      const opus_val16 * OPUS_RESTRICT wp2 = window+overlap-1;
+
+      for(i = 0; i < overlap/2; i++)
+      {
+         kiss_fft_scalar x1, x2;
+         x1 = *xp1;
+         x2 = *yp1;
+         *yp1++ = MULT16_32_Q15(*wp2, x2) - MULT16_32_Q15(*wp1, x1);
+         *xp1-- = MULT16_32_Q15(*wp1, x2) + MULT16_32_Q15(*wp2, x1);
+         wp1++;
+         wp2--;
+      }
+   }
+   RESTORE_STACK;
+}
diff --git a/drivers/opus/celt/arm/celt_neon_intr.c b/drivers/opus/celt/arm/celt_neon_intr.c
new file mode 100644
index 0000000000..82b6958644
--- /dev/null
+++ b/drivers/opus/celt/arm/celt_neon_intr.c
@@ -0,0 +1,249 @@
+/* Copyright (c) 2014-2015 Xiph.Org Foundation
+   Written by Viswanath Puttagunta */
+/**
+   @file celt_neon_intr.c
+   @brief ARM Neon Intrinsic optimizations for celt
+ */
+
+/*
+   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.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+#include "opus/opus_config.h"
+
+#include <arm_neon.h>
+#include "opus/celt/pitch.h"
+
+#if !defined(FIXED_POINT)
+/*
+ * Function: xcorr_kernel_neon_float
+ * ---------------------------------
+ * Computes 4 correlation values and stores them in sum[4]
+ */
+static void xcorr_kernel_neon_float(const float32_t *x, const float32_t *y,
+      float32_t sum[4], int len) {
+   float32x4_t YY[3];
+   float32x4_t YEXT[3];
+   float32x4_t XX[2];
+   float32x2_t XX_2;
+   float32x4_t SUMM;
+   const float32_t *xi = x;
+   const float32_t *yi = y;
+
+   celt_assert(len>0);
+
+   YY[0] = vld1q_f32(yi);
+   SUMM = vdupq_n_f32(0);
+
+   /* Consume 8 elements in x vector and 12 elements in y
+    * vector. However, the 12'th element never really gets
+    * touched in this loop. So, if len == 8, then we only
+    * must access y[0] to y[10]. y[11] must not be accessed
+    * hence make sure len > 8 and not len >= 8
+    */
+   while (len > 8) {
+      yi += 4;
+      YY[1] = vld1q_f32(yi);
+      yi += 4;
+      YY[2] = vld1q_f32(yi);
+
+      XX[0] = vld1q_f32(xi);
+      xi += 4;
+      XX[1] = vld1q_f32(xi);
+      xi += 4;
+
+      SUMM = vmlaq_lane_f32(SUMM, YY[0], vget_low_f32(XX[0]), 0);
+      YEXT[0] = vextq_f32(YY[0], YY[1], 1);
+      SUMM = vmlaq_lane_f32(SUMM, YEXT[0], vget_low_f32(XX[0]), 1);
+      YEXT[1] = vextq_f32(YY[0], YY[1], 2);
+      SUMM = vmlaq_lane_f32(SUMM, YEXT[1], vget_high_f32(XX[0]), 0);
+      YEXT[2] = vextq_f32(YY[0], YY[1], 3);
+      SUMM = vmlaq_lane_f32(SUMM, YEXT[2], vget_high_f32(XX[0]), 1);
+
+      SUMM = vmlaq_lane_f32(SUMM, YY[1], vget_low_f32(XX[1]), 0);
+      YEXT[0] = vextq_f32(YY[1], YY[2], 1);
+      SUMM = vmlaq_lane_f32(SUMM, YEXT[0], vget_low_f32(XX[1]), 1);
+      YEXT[1] = vextq_f32(YY[1], YY[2], 2);
+      SUMM = vmlaq_lane_f32(SUMM, YEXT[1], vget_high_f32(XX[1]), 0);
+      YEXT[2] = vextq_f32(YY[1], YY[2], 3);
+      SUMM = vmlaq_lane_f32(SUMM, YEXT[2], vget_high_f32(XX[1]), 1);
+
+      YY[0] = YY[2];
+      len -= 8;
+   }
+
+   /* Consume 4 elements in x vector and 8 elements in y
+    * vector. However, the 8'th element in y never really gets
+    * touched in this loop. So, if len == 4, then we only
+    * must access y[0] to y[6]. y[7] must not be accessed
+    * hence make sure len>4 and not len>=4
+    */
+   if (len > 4) {
+      yi += 4;
+      YY[1] = vld1q_f32(yi);
+
+      XX[0] = vld1q_f32(xi);
+      xi += 4;
+
+      SUMM = vmlaq_lane_f32(SUMM, YY[0], vget_low_f32(XX[0]), 0);
+      YEXT[0] = vextq_f32(YY[0], YY[1], 1);
+      SUMM = vmlaq_lane_f32(SUMM, YEXT[0], vget_low_f32(XX[0]), 1);
+      YEXT[1] = vextq_f32(YY[0], YY[1], 2);
+      SUMM = vmlaq_lane_f32(SUMM, YEXT[1], vget_high_f32(XX[0]), 0);
+      YEXT[2] = vextq_f32(YY[0], YY[1], 3);
+      SUMM = vmlaq_lane_f32(SUMM, YEXT[2], vget_high_f32(XX[0]), 1);
+
+      YY[0] = YY[1];
+      len -= 4;
+   }
+
+   while (--len > 0) {
+      XX_2 = vld1_dup_f32(xi++);
+      SUMM = vmlaq_lane_f32(SUMM, YY[0], XX_2, 0);
+      YY[0]= vld1q_f32(++yi);
+   }
+
+   XX_2 = vld1_dup_f32(xi);
+   SUMM = vmlaq_lane_f32(SUMM, YY[0], XX_2, 0);
+
+   vst1q_f32(sum, SUMM);
+}
+
+/*
+ * Function: xcorr_kernel_neon_float_process1
+ * ---------------------------------
+ * Computes single correlation values and stores in *sum
+ */
+static void xcorr_kernel_neon_float_process1(const float32_t *x,
+      const float32_t *y, float32_t *sum, int len) {
+   float32x4_t XX[4];
+   float32x4_t YY[4];
+   float32x2_t XX_2;
+   float32x2_t YY_2;
+   float32x4_t SUMM;
+   float32x2_t SUMM_2[2];
+   const float32_t *xi = x;
+   const float32_t *yi = y;
+
+   SUMM = vdupq_n_f32(0);
+
+   /* Work on 16 values per iteration */
+   while (len >= 16) {
+      XX[0] = vld1q_f32(xi);
+      xi += 4;
+      XX[1] = vld1q_f32(xi);
+      xi += 4;
+      XX[2] = vld1q_f32(xi);
+      xi += 4;
+      XX[3] = vld1q_f32(xi);
+      xi += 4;
+
+      YY[0] = vld1q_f32(yi);
+      yi += 4;
+      YY[1] = vld1q_f32(yi);
+      yi += 4;
+      YY[2] = vld1q_f32(yi);
+      yi += 4;
+      YY[3] = vld1q_f32(yi);
+      yi += 4;
+
+      SUMM = vmlaq_f32(SUMM, YY[0], XX[0]);
+      SUMM = vmlaq_f32(SUMM, YY[1], XX[1]);
+      SUMM = vmlaq_f32(SUMM, YY[2], XX[2]);
+      SUMM = vmlaq_f32(SUMM, YY[3], XX[3]);
+      len -= 16;
+   }
+
+   /* Work on 8 values */
+   if (len >= 8) {
+      XX[0] = vld1q_f32(xi);
+      xi += 4;
+      XX[1] = vld1q_f32(xi);
+      xi += 4;
+
+      YY[0] = vld1q_f32(yi);
+      yi += 4;
+      YY[1] = vld1q_f32(yi);
+      yi += 4;
+
+      SUMM = vmlaq_f32(SUMM, YY[0], XX[0]);
+      SUMM = vmlaq_f32(SUMM, YY[1], XX[1]);
+      len -= 8;
+   }
+
+   /* Work on 4 values */
+   if (len >= 4) {
+      XX[0] = vld1q_f32(xi);
+      xi += 4;
+      YY[0] = vld1q_f32(yi);
+      yi += 4;
+      SUMM = vmlaq_f32(SUMM, YY[0], XX[0]);
+      len -= 4;
+   }
+
+   /* Start accumulating results */
+   SUMM_2[0] = vget_low_f32(SUMM);
+   if (len >= 2) {
+      /* While at it, consume 2 more values if available */
+      XX_2 = vld1_f32(xi);
+      xi += 2;
+      YY_2 = vld1_f32(yi);
+      yi += 2;
+      SUMM_2[0] = vmla_f32(SUMM_2[0], YY_2, XX_2);
+      len -= 2;
+   }
+   SUMM_2[1] = vget_high_f32(SUMM);
+   SUMM_2[0] = vadd_f32(SUMM_2[0], SUMM_2[1]);
+   SUMM_2[0] = vpadd_f32(SUMM_2[0], SUMM_2[0]);
+   /* Ok, now we have result accumulated in SUMM_2[0].0 */
+
+   if (len > 0) {
+      /* Case when you have one value left */
+      XX_2 = vld1_dup_f32(xi);
+      YY_2 = vld1_dup_f32(yi);
+      SUMM_2[0] = vmla_f32(SUMM_2[0], XX_2, YY_2);
+   }
+
+   vst1_lane_f32(sum, SUMM_2[0], 0);
+}
+
+void celt_pitch_xcorr_float_neon(const opus_val16 *_x, const opus_val16 *_y,
+                        opus_val32 *xcorr, int len, int max_pitch) {
+   int i;
+   celt_assert(max_pitch > 0);
+   celt_assert((((unsigned char *)_x-(unsigned char *)NULL)&3)==0);
+
+   for (i = 0; i < (max_pitch-3); i += 4) {
+      xcorr_kernel_neon_float((const float32_t *)_x, (const float32_t *)_y+i,
+            (float32_t *)xcorr+i, len);
+   }
+
+   /* In case max_pitch isn't multiple of 4
+    * compute single correlation value per iteration
+    */
+   for (; i < max_pitch; i++) {
+      xcorr_kernel_neon_float_process1((const float32_t *)_x,
+            (const float32_t *)_y+i, (float32_t *)xcorr+i, len);
+   }
+}
+#endif
diff --git a/drivers/opus/celt/arm/celt_pitch_xcorr_arm-gnu.S b/drivers/opus/celt/arm/celt_pitch_xcorr_arm-gnu.S
new file mode 100644
index 0000000000..5b2ee55a10
--- /dev/null
+++ b/drivers/opus/celt/arm/celt_pitch_xcorr_arm-gnu.S
@@ -0,0 +1,551 @@
+    .syntax unified
+@ Copyright (c) 2007-2008 CSIRO
+@ Copyright (c) 2007-2009 Xiph.Org Foundation
+@ Copyright (c) 2013      Parrot
+@ Written by Aurélien Zanelli
+@
+@ 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.
+@
+@ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+@ ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+@ LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+@ A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+@ OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+@ EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+@ PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+@ PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+@ LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+@ NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    .text;   .p2align 2;   .arch armv7-a
+   .fpu neon
+   .object_arch armv4t
+
+  .include "celt/arm/armopts-gnu.S"
+
+ .if OPUS_ARM_MAY_HAVE_EDSP
+  .global celt_pitch_xcorr_edsp
+ .endif
+
+ .if OPUS_ARM_MAY_HAVE_NEON
+  .global celt_pitch_xcorr_neon
+ .endif
+
+ .if OPUS_ARM_MAY_HAVE_NEON
+
+@ Compute sum[k]=sum(x[j]*y[j+k],j=0...len-1), k=0...3
+; xcorr_kernel_neon: @ PROC
+xcorr_kernel_neon_start:
+  @ input:
+  @   r3     = int         len
+  @   r4     = opus_val16 *x
+  @   r5     = opus_val16 *y
+  @   q0     = opus_val32  sum[4]
+  @ output:
+  @   q0     = opus_val32  sum[4]
+  @ preserved: r0-r3, r6-r11, d2, q4-q7, q9-q15
+  @ internal usage:
+  @   r12 = int j
+  @   d3  = y_3|y_2|y_1|y_0
+  @   q2  = y_B|y_A|y_9|y_8|y_7|y_6|y_5|y_4
+  @   q3  = x_7|x_6|x_5|x_4|x_3|x_2|x_1|x_0
+  @   q8  = scratch
+  @
+  @ Load y[0...3]
+  @ This requires len>0 to always be valid (which we assert in the C code).
+  VLD1.16      {d5}, [r5]!
+  SUBS         r12, r3, #8
+  BLE xcorr_kernel_neon_process4
+@ Process 8 samples at a time.
+@ This loop loads one y value more than we actually need. Therefore we have to
+@ stop as soon as there are 8 or fewer samples left (instead of 7), to avoid
+@ reading past the end of the array.
+xcorr_kernel_neon_process8:
+  @ This loop has 19 total instructions (10 cycles to issue, minimum), with
+  @ - 2 cycles of ARM insrtuctions,
+  @ - 10 cycles of load/store/byte permute instructions, and
+  @ - 9 cycles of data processing instructions.
+  @ On a Cortex A8, we dual-issue the maximum amount (9 cycles) between the
+  @ latter two categories, meaning the whole loop should run in 10 cycles per
+  @ iteration, barring cache misses.
+  @
+  @ Load x[0...7]
+  VLD1.16      {d6, d7}, [r4]!
+  @ Unlike VMOV, VAND is a data processsing instruction (and doesn't get
+  @ assembled to VMOV, like VORR would), so it dual-issues with the prior VLD1.
+  VAND         d3, d5, d5
+  SUBS         r12, r12, #8
+  @ Load y[4...11]
+  VLD1.16      {d4, d5}, [r5]!
+  VMLAL.S16    q0, d3, d6[0]
+  VEXT.16      d16, d3, d4, #1
+  VMLAL.S16    q0, d4, d7[0]
+  VEXT.16      d17, d4, d5, #1
+  VMLAL.S16    q0, d16, d6[1]
+  VEXT.16      d16, d3, d4, #2
+  VMLAL.S16    q0, d17, d7[1]
+  VEXT.16      d17, d4, d5, #2
+  VMLAL.S16    q0, d16, d6[2]
+  VEXT.16      d16, d3, d4, #3
+  VMLAL.S16    q0, d17, d7[2]
+  VEXT.16      d17, d4, d5, #3
+  VMLAL.S16    q0, d16, d6[3]
+  VMLAL.S16    q0, d17, d7[3]
+  BGT xcorr_kernel_neon_process8
+@ Process 4 samples here if we have > 4 left (still reading one extra y value).
+xcorr_kernel_neon_process4:
+  ADDS         r12, r12, #4
+  BLE xcorr_kernel_neon_process2
+  @ Load x[0...3]
+  VLD1.16      d6, [r4]!
+  @ Use VAND since it's a data processing instruction again.
+  VAND         d4, d5, d5
+  SUB          r12, r12, #4
+  @ Load y[4...7]
+  VLD1.16      d5, [r5]!
+  VMLAL.S16    q0, d4, d6[0]
+  VEXT.16      d16, d4, d5, #1
+  VMLAL.S16    q0, d16, d6[1]
+  VEXT.16      d16, d4, d5, #2
+  VMLAL.S16    q0, d16, d6[2]
+  VEXT.16      d16, d4, d5, #3
+  VMLAL.S16    q0, d16, d6[3]
+@ Process 2 samples here if we have > 2 left (still reading one extra y value).
+xcorr_kernel_neon_process2:
+  ADDS         r12, r12, #2
+  BLE xcorr_kernel_neon_process1
+  @ Load x[0...1]
+  VLD2.16      {d6[],d7[]}, [r4]!
+  @ Use VAND since it's a data processing instruction again.
+  VAND         d4, d5, d5
+  SUB          r12, r12, #2
+  @ Load y[4...5]
+  VLD1.32      {d5[]}, [r5]!
+  VMLAL.S16    q0, d4, d6
+  VEXT.16      d16, d4, d5, #1
+  @ Replace bottom copy of {y5,y4} in d5 with {y3,y2} from d4, using VSRI
+  @ instead of VEXT, since it's a data-processing instruction.
+  VSRI.64      d5, d4, #32
+  VMLAL.S16    q0, d16, d7
+@ Process 1 sample using the extra y value we loaded above.
+xcorr_kernel_neon_process1:
+  @ Load next *x
+  VLD1.16      {d6[]}, [r4]!
+  ADDS         r12, r12, #1
+  @ y[0...3] are left in d5 from prior iteration(s) (if any)
+  VMLAL.S16    q0, d5, d6
+  MOVLE        pc, lr
+@ Now process 1 last sample, not reading ahead.
+  @ Load last *y
+  VLD1.16      {d4[]}, [r5]!
+  VSRI.64      d4, d5, #16
+  @ Load last *x
+  VLD1.16      {d6[]}, [r4]!
+  VMLAL.S16    q0, d4, d6
+  MOV          pc, lr
+	.size xcorr_kernel_neon, .-xcorr_kernel_neon  @ ENDP
+
+@ opus_val32 celt_pitch_xcorr_neon(opus_val16 *_x, opus_val16 *_y,
+@  opus_val32 *xcorr, int len, int max_pitch)
+; celt_pitch_xcorr_neon: @ PROC
+  @ input:
+  @   r0  = opus_val16 *_x
+  @   r1  = opus_val16 *_y
+  @   r2  = opus_val32 *xcorr
+  @   r3  = int         len
+  @ output:
+  @   r0  = int         maxcorr
+  @ internal usage:
+  @   r4  = opus_val16 *x (for xcorr_kernel_neon())
+  @   r5  = opus_val16 *y (for xcorr_kernel_neon())
+  @   r6  = int         max_pitch
+  @   r12 = int         j
+  @   q15 = int         maxcorr[4] (q15 is not used by xcorr_kernel_neon())
+  STMFD        sp!, {r4-r6, lr}
+  LDR          r6, [sp, #16]
+  VMOV.S32     q15, #1
+  @ if (max_pitch < 4) goto celt_pitch_xcorr_neon_process4_done
+  SUBS         r6, r6, #4
+  BLT celt_pitch_xcorr_neon_process4_done
+celt_pitch_xcorr_neon_process4:
+  @ xcorr_kernel_neon parameters:
+  @ r3 = len, r4 = _x, r5 = _y, q0 = {0, 0, 0, 0}
+  MOV          r4, r0
+  MOV          r5, r1
+  VEOR         q0, q0, q0
+  @ xcorr_kernel_neon only modifies r4, r5, r12, and q0...q3.
+  @ So we don't save/restore any other registers.
+  BL xcorr_kernel_neon_start
+  SUBS         r6, r6, #4
+  VST1.32      {q0}, [r2]!
+  @ _y += 4
+  ADD          r1, r1, #8
+  VMAX.S32     q15, q15, q0
+  @ if (max_pitch < 4) goto celt_pitch_xcorr_neon_process4_done
+  BGE celt_pitch_xcorr_neon_process4
+@ We have less than 4 sums left to compute.
+celt_pitch_xcorr_neon_process4_done:
+  ADDS         r6, r6, #4
+  @ Reduce maxcorr to a single value
+  VMAX.S32     d30, d30, d31
+  VPMAX.S32    d30, d30, d30
+  @ if (max_pitch <= 0) goto celt_pitch_xcorr_neon_done
+  BLE celt_pitch_xcorr_neon_done
+@ Now compute each remaining sum one at a time.
+celt_pitch_xcorr_neon_process_remaining:
+  MOV          r4, r0
+  MOV          r5, r1
+  VMOV.I32     q0, #0
+  SUBS         r12, r3, #8
+  BLT celt_pitch_xcorr_neon_process_remaining4
+@ Sum terms 8 at a time.
+celt_pitch_xcorr_neon_process_remaining_loop8:
+  @ Load x[0...7]
+  VLD1.16      {q1}, [r4]!
+  @ Load y[0...7]
+  VLD1.16      {q2}, [r5]!
+  SUBS         r12, r12, #8
+  VMLAL.S16    q0, d4, d2
+  VMLAL.S16    q0, d5, d3
+  BGE celt_pitch_xcorr_neon_process_remaining_loop8
+@ Sum terms 4 at a time.
+celt_pitch_xcorr_neon_process_remaining4:
+  ADDS         r12, r12, #4
+  BLT celt_pitch_xcorr_neon_process_remaining4_done
+  @ Load x[0...3]
+  VLD1.16      {d2}, [r4]!
+  @ Load y[0...3]
+  VLD1.16      {d3}, [r5]!
+  SUB          r12, r12, #4
+  VMLAL.S16    q0, d3, d2
+celt_pitch_xcorr_neon_process_remaining4_done:
+  @ Reduce the sum to a single value.
+  VADD.S32     d0, d0, d1
+  VPADDL.S32   d0, d0
+  ADDS         r12, r12, #4
+  BLE celt_pitch_xcorr_neon_process_remaining_loop_done
+@ Sum terms 1 at a time.
+celt_pitch_xcorr_neon_process_remaining_loop1:
+  VLD1.16      {d2[]}, [r4]!
+  VLD1.16      {d3[]}, [r5]!
+  SUBS         r12, r12, #1
+  VMLAL.S16    q0, d2, d3
+  BGT celt_pitch_xcorr_neon_process_remaining_loop1
+celt_pitch_xcorr_neon_process_remaining_loop_done:
+  VST1.32      {d0[0]}, [r2]!
+  VMAX.S32     d30, d30, d0
+  SUBS         r6, r6, #1
+  @ _y++
+  ADD          r1, r1, #2
+  @ if (--max_pitch > 0) goto celt_pitch_xcorr_neon_process_remaining
+  BGT celt_pitch_xcorr_neon_process_remaining
+celt_pitch_xcorr_neon_done:
+  VMOV.32      r0, d30[0]
+  LDMFD        sp!, {r4-r6, pc}
+	.size celt_pitch_xcorr_neon, .-celt_pitch_xcorr_neon  @ ENDP
+
+ .endif
+
+ .if OPUS_ARM_MAY_HAVE_EDSP
+
+@ This will get used on ARMv7 devices without NEON, so it has been optimized
+@ to take advantage of dual-issuing where possible.
+; xcorr_kernel_edsp: @ PROC
+xcorr_kernel_edsp_start:
+  @ input:
+  @   r3      = int         len
+  @   r4      = opus_val16 *_x (must be 32-bit aligned)
+  @   r5      = opus_val16 *_y (must be 32-bit aligned)
+  @   r6...r9 = opus_val32  sum[4]
+  @ output:
+  @   r6...r9 = opus_val32  sum[4]
+  @ preserved: r0-r5
+  @ internal usage
+  @   r2      = int         j
+  @   r12,r14 = opus_val16  x[4]
+  @   r10,r11 = opus_val16  y[4]
+  STMFD        sp!, {r2,r4,r5,lr}
+  LDR          r10, [r5], #4      @ Load y[0...1]
+  SUBS         r2, r3, #4         @ j = len-4
+  LDR          r11, [r5], #4      @ Load y[2...3]
+  BLE xcorr_kernel_edsp_process4_done
+  LDR          r12, [r4], #4      @ Load x[0...1]
+  @ Stall
+xcorr_kernel_edsp_process4:
+  @ The multiplies must issue from pipeline 0, and can't dual-issue with each
+  @ other. Every other instruction here dual-issues with a multiply, and is
+  @ thus "free". There should be no stalls in the body of the loop.
+  SMLABB       r6, r12, r10, r6   @ sum[0] = MAC16_16(sum[0],x_0,y_0)
+  LDR          r14, [r4], #4      @ Load x[2...3]
+  SMLABT       r7, r12, r10, r7   @ sum[1] = MAC16_16(sum[1],x_0,y_1)
+  SUBS         r2, r2, #4         @ j-=4
+  SMLABB       r8, r12, r11, r8   @ sum[2] = MAC16_16(sum[2],x_0,y_2)
+  SMLABT       r9, r12, r11, r9   @ sum[3] = MAC16_16(sum[3],x_0,y_3)
+  SMLATT       r6, r12, r10, r6   @ sum[0] = MAC16_16(sum[0],x_1,y_1)
+  LDR          r10, [r5], #4      @ Load y[4...5]
+  SMLATB       r7, r12, r11, r7   @ sum[1] = MAC16_16(sum[1],x_1,y_2)
+  SMLATT       r8, r12, r11, r8   @ sum[2] = MAC16_16(sum[2],x_1,y_3)
+  SMLATB       r9, r12, r10, r9   @ sum[3] = MAC16_16(sum[3],x_1,y_4)
+  LDRGT        r12, [r4], #4      @ Load x[0...1]
+  SMLABB       r6, r14, r11, r6   @ sum[0] = MAC16_16(sum[0],x_2,y_2)
+  SMLABT       r7, r14, r11, r7   @ sum[1] = MAC16_16(sum[1],x_2,y_3)
+  SMLABB       r8, r14, r10, r8   @ sum[2] = MAC16_16(sum[2],x_2,y_4)
+  SMLABT       r9, r14, r10, r9   @ sum[3] = MAC16_16(sum[3],x_2,y_5)
+  SMLATT       r6, r14, r11, r6   @ sum[0] = MAC16_16(sum[0],x_3,y_3)
+  LDR          r11, [r5], #4      @ Load y[6...7]
+  SMLATB       r7, r14, r10, r7   @ sum[1] = MAC16_16(sum[1],x_3,y_4)
+  SMLATT       r8, r14, r10, r8   @ sum[2] = MAC16_16(sum[2],x_3,y_5)
+  SMLATB       r9, r14, r11, r9   @ sum[3] = MAC16_16(sum[3],x_3,y_6)
+  BGT xcorr_kernel_edsp_process4
+xcorr_kernel_edsp_process4_done:
+  ADDS         r2, r2, #4
+  BLE xcorr_kernel_edsp_done
+  LDRH         r12, [r4], #2      @ r12 = *x++
+  SUBS         r2, r2, #1         @ j--
+  @ Stall
+  SMLABB       r6, r12, r10, r6   @ sum[0] = MAC16_16(sum[0],x,y_0)
+  LDRHGT       r14, [r4], #2      @ r14 = *x++
+  SMLABT       r7, r12, r10, r7   @ sum[1] = MAC16_16(sum[1],x,y_1)
+  SMLABB       r8, r12, r11, r8   @ sum[2] = MAC16_16(sum[2],x,y_2)
+  SMLABT       r9, r12, r11, r9   @ sum[3] = MAC16_16(sum[3],x,y_3)
+  BLE xcorr_kernel_edsp_done
+  SMLABT       r6, r14, r10, r6   @ sum[0] = MAC16_16(sum[0],x,y_1)
+  SUBS         r2, r2, #1         @ j--
+  SMLABB       r7, r14, r11, r7   @ sum[1] = MAC16_16(sum[1],x,y_2)
+  LDRH         r10, [r5], #2      @ r10 = y_4 = *y++
+  SMLABT       r8, r14, r11, r8   @ sum[2] = MAC16_16(sum[2],x,y_3)
+  LDRHGT       r12, [r4], #2      @ r12 = *x++
+  SMLABB       r9, r14, r10, r9   @ sum[3] = MAC16_16(sum[3],x,y_4)
+  BLE xcorr_kernel_edsp_done
+  SMLABB       r6, r12, r11, r6   @ sum[0] = MAC16_16(sum[0],tmp,y_2)
+  CMP          r2, #1             @ j--
+  SMLABT       r7, r12, r11, r7   @ sum[1] = MAC16_16(sum[1],tmp,y_3)
+  LDRH         r2, [r5], #2       @ r2 = y_5 = *y++
+  SMLABB       r8, r12, r10, r8   @ sum[2] = MAC16_16(sum[2],tmp,y_4)
+  LDRHGT       r14, [r4]          @ r14 = *x
+  SMLABB       r9, r12, r2, r9    @ sum[3] = MAC16_16(sum[3],tmp,y_5)
+  BLE xcorr_kernel_edsp_done
+  SMLABT       r6, r14, r11, r6   @ sum[0] = MAC16_16(sum[0],tmp,y_3)
+  LDRH         r11, [r5]          @ r11 = y_6 = *y
+  SMLABB       r7, r14, r10, r7   @ sum[1] = MAC16_16(sum[1],tmp,y_4)
+  SMLABB       r8, r14, r2, r8    @ sum[2] = MAC16_16(sum[2],tmp,y_5)
+  SMLABB       r9, r14, r11, r9   @ sum[3] = MAC16_16(sum[3],tmp,y_6)
+xcorr_kernel_edsp_done:
+  LDMFD        sp!, {r2,r4,r5,pc}
+	.size xcorr_kernel_edsp, .-xcorr_kernel_edsp  @ ENDP
+
+; celt_pitch_xcorr_edsp: @ PROC
+  @ input:
+  @   r0  = opus_val16 *_x (must be 32-bit aligned)
+  @   r1  = opus_val16 *_y (only needs to be 16-bit aligned)
+  @   r2  = opus_val32 *xcorr
+  @   r3  = int         len
+  @ output:
+  @   r0  = maxcorr
+  @ internal usage
+  @   r4  = opus_val16 *x
+  @   r5  = opus_val16 *y
+  @   r6  = opus_val32  sum0
+  @   r7  = opus_val32  sum1
+  @   r8  = opus_val32  sum2
+  @   r9  = opus_val32  sum3
+  @   r1  = int         max_pitch
+  @   r12 = int         j
+  STMFD        sp!, {r4-r11, lr}
+  MOV          r5, r1
+  LDR          r1, [sp, #36]
+  MOV          r4, r0
+  TST          r5, #3
+  @ maxcorr = 1
+  MOV          r0, #1
+  BEQ          celt_pitch_xcorr_edsp_process1u_done
+@ Compute one sum at the start to make y 32-bit aligned.
+  SUBS         r12, r3, #4
+  @ r14 = sum = 0
+  MOV          r14, #0
+  LDRH         r8, [r5], #2
+  BLE celt_pitch_xcorr_edsp_process1u_loop4_done
+  LDR          r6, [r4], #4
+  MOV          r8, r8, LSL #16
+celt_pitch_xcorr_edsp_process1u_loop4:
+  LDR          r9, [r5], #4
+  SMLABT       r14, r6, r8, r14     @ sum = MAC16_16(sum, x_0, y_0)
+  LDR          r7, [r4], #4
+  SMLATB       r14, r6, r9, r14     @ sum = MAC16_16(sum, x_1, y_1)
+  LDR          r8, [r5], #4
+  SMLABT       r14, r7, r9, r14     @ sum = MAC16_16(sum, x_2, y_2)
+  SUBS         r12, r12, #4         @ j-=4
+  SMLATB       r14, r7, r8, r14     @ sum = MAC16_16(sum, x_3, y_3)
+  LDRGT        r6, [r4], #4
+  BGT celt_pitch_xcorr_edsp_process1u_loop4
+  MOV          r8, r8, LSR #16
+celt_pitch_xcorr_edsp_process1u_loop4_done:
+  ADDS         r12, r12, #4
+celt_pitch_xcorr_edsp_process1u_loop1:
+  LDRHGE       r6, [r4], #2
+  @ Stall
+  SMLABBGE     r14, r6, r8, r14    @ sum = MAC16_16(sum, *x, *y)
+  SUBSGE       r12, r12, #1
+  LDRHGT       r8, [r5], #2
+  BGT celt_pitch_xcorr_edsp_process1u_loop1
+  @ Restore _x
+  SUB          r4, r4, r3, LSL #1
+  @ Restore and advance _y
+  SUB          r5, r5, r3, LSL #1
+  @ maxcorr = max(maxcorr, sum)
+  CMP          r0, r14
+  ADD          r5, r5, #2
+  MOVLT        r0, r14
+  SUBS         r1, r1, #1
+  @ xcorr[i] = sum
+  STR          r14, [r2], #4
+  BLE celt_pitch_xcorr_edsp_done
+celt_pitch_xcorr_edsp_process1u_done:
+  @ if (max_pitch < 4) goto celt_pitch_xcorr_edsp_process2
+  SUBS         r1, r1, #4
+  BLT celt_pitch_xcorr_edsp_process2
+celt_pitch_xcorr_edsp_process4:
+  @ xcorr_kernel_edsp parameters:
+  @ r3 = len, r4 = _x, r5 = _y, r6...r9 = sum[4] = {0, 0, 0, 0}
+  MOV          r6, #0
+  MOV          r7, #0
+  MOV          r8, #0
+  MOV          r9, #0
+  BL xcorr_kernel_edsp_start  @ xcorr_kernel_edsp(_x, _y+i, xcorr+i, len)
+  @ maxcorr = max(maxcorr, sum0, sum1, sum2, sum3)
+  CMP          r0, r6
+  @ _y+=4
+  ADD          r5, r5, #8
+  MOVLT        r0, r6
+  CMP          r0, r7
+  MOVLT        r0, r7
+  CMP          r0, r8
+  MOVLT        r0, r8
+  CMP          r0, r9
+  MOVLT        r0, r9
+  STMIA        r2!, {r6-r9}
+  SUBS         r1, r1, #4
+  BGE celt_pitch_xcorr_edsp_process4
+celt_pitch_xcorr_edsp_process2:
+  ADDS         r1, r1, #2
+  BLT celt_pitch_xcorr_edsp_process1a
+  SUBS         r12, r3, #4
+  @ {r10, r11} = {sum0, sum1} = {0, 0}
+  MOV          r10, #0
+  MOV          r11, #0
+  LDR          r8, [r5], #4
+  BLE celt_pitch_xcorr_edsp_process2_loop_done
+  LDR          r6, [r4], #4
+  LDR          r9, [r5], #4
+celt_pitch_xcorr_edsp_process2_loop4:
+  SMLABB       r10, r6, r8, r10     @ sum0 = MAC16_16(sum0, x_0, y_0)
+  LDR          r7, [r4], #4
+  SMLABT       r11, r6, r8, r11     @ sum1 = MAC16_16(sum1, x_0, y_1)
+  SUBS         r12, r12, #4         @ j-=4
+  SMLATT       r10, r6, r8, r10     @ sum0 = MAC16_16(sum0, x_1, y_1)
+  LDR          r8, [r5], #4
+  SMLATB       r11, r6, r9, r11     @ sum1 = MAC16_16(sum1, x_1, y_2)
+  LDRGT        r6, [r4], #4
+  SMLABB       r10, r7, r9, r10     @ sum0 = MAC16_16(sum0, x_2, y_2)
+  SMLABT       r11, r7, r9, r11     @ sum1 = MAC16_16(sum1, x_2, y_3)
+  SMLATT       r10, r7, r9, r10     @ sum0 = MAC16_16(sum0, x_3, y_3)
+  LDRGT        r9, [r5], #4
+  SMLATB       r11, r7, r8, r11     @ sum1 = MAC16_16(sum1, x_3, y_4)
+  BGT celt_pitch_xcorr_edsp_process2_loop4
+celt_pitch_xcorr_edsp_process2_loop_done:
+  ADDS         r12, r12, #2
+  BLE  celt_pitch_xcorr_edsp_process2_1
+  LDR          r6, [r4], #4
+  @ Stall
+  SMLABB       r10, r6, r8, r10     @ sum0 = MAC16_16(sum0, x_0, y_0)
+  LDR          r9, [r5], #4
+  SMLABT       r11, r6, r8, r11     @ sum1 = MAC16_16(sum1, x_0, y_1)
+  SUB          r12, r12, #2
+  SMLATT       r10, r6, r8, r10     @ sum0 = MAC16_16(sum0, x_1, y_1)
+  MOV          r8, r9
+  SMLATB       r11, r6, r9, r11     @ sum1 = MAC16_16(sum1, x_1, y_2)
+celt_pitch_xcorr_edsp_process2_1:
+  LDRH         r6, [r4], #2
+  ADDS         r12, r12, #1
+  @ Stall
+  SMLABB       r10, r6, r8, r10     @ sum0 = MAC16_16(sum0, x_0, y_0)
+  LDRHGT       r7, [r4], #2
+  SMLABT       r11, r6, r8, r11     @ sum1 = MAC16_16(sum1, x_0, y_1)
+  BLE celt_pitch_xcorr_edsp_process2_done
+  LDRH         r9, [r5], #2
+  SMLABT       r10, r7, r8, r10     @ sum0 = MAC16_16(sum0, x_0, y_1)
+  SMLABB       r11, r7, r9, r11     @ sum1 = MAC16_16(sum1, x_0, y_2)
+celt_pitch_xcorr_edsp_process2_done:
+  @ Restore _x
+  SUB          r4, r4, r3, LSL #1
+  @ Restore and advance _y
+  SUB          r5, r5, r3, LSL #1
+  @ maxcorr = max(maxcorr, sum0)
+  CMP          r0, r10
+  ADD          r5, r5, #2
+  MOVLT        r0, r10
+  SUB          r1, r1, #2
+  @ maxcorr = max(maxcorr, sum1)
+  CMP          r0, r11
+  @ xcorr[i] = sum
+  STR          r10, [r2], #4
+  MOVLT        r0, r11
+  STR          r11, [r2], #4
+celt_pitch_xcorr_edsp_process1a:
+  ADDS         r1, r1, #1
+  BLT celt_pitch_xcorr_edsp_done
+  SUBS         r12, r3, #4
+  @ r14 = sum = 0
+  MOV          r14, #0
+  BLT celt_pitch_xcorr_edsp_process1a_loop_done
+  LDR          r6, [r4], #4
+  LDR          r8, [r5], #4
+  LDR          r7, [r4], #4
+  LDR          r9, [r5], #4
+celt_pitch_xcorr_edsp_process1a_loop4:
+  SMLABB       r14, r6, r8, r14     @ sum = MAC16_16(sum, x_0, y_0)
+  SUBS         r12, r12, #4         @ j-=4
+  SMLATT       r14, r6, r8, r14     @ sum = MAC16_16(sum, x_1, y_1)
+  LDRGE        r6, [r4], #4
+  SMLABB       r14, r7, r9, r14     @ sum = MAC16_16(sum, x_2, y_2)
+  LDRGE        r8, [r5], #4
+  SMLATT       r14, r7, r9, r14     @ sum = MAC16_16(sum, x_3, y_3)
+  LDRGE        r7, [r4], #4
+  LDRGE        r9, [r5], #4
+  BGE celt_pitch_xcorr_edsp_process1a_loop4
+celt_pitch_xcorr_edsp_process1a_loop_done:
+  ADDS         r12, r12, #2
+  LDRGE        r6, [r4], #4
+  LDRGE        r8, [r5], #4
+  @ Stall
+  SMLABBGE     r14, r6, r8, r14     @ sum = MAC16_16(sum, x_0, y_0)
+  SUBGE        r12, r12, #2
+  SMLATTGE     r14, r6, r8, r14     @ sum = MAC16_16(sum, x_1, y_1)
+  ADDS         r12, r12, #1
+  LDRHGE       r6, [r4], #2
+  LDRHGE       r8, [r5], #2
+  @ Stall
+  SMLABBGE     r14, r6, r8, r14     @ sum = MAC16_16(sum, *x, *y)
+  @ maxcorr = max(maxcorr, sum)
+  CMP          r0, r14
+  @ xcorr[i] = sum
+  STR          r14, [r2], #4
+  MOVLT        r0, r14
+celt_pitch_xcorr_edsp_done:
+  LDMFD        sp!, {r4-r11, pc}
+	.size celt_pitch_xcorr_edsp, .-celt_pitch_xcorr_edsp  @ ENDP
+
+ .endif
+
+@ END:
+    .section	.note.GNU-stack,"",%progbits
diff --git a/drivers/opus/celt/arm/celt_pitch_xcorr_arm.s b/drivers/opus/celt/arm/celt_pitch_xcorr_arm.s
index 09917b16bf..f96e0a88bb 100644
--- a/drivers/opus/celt/arm/celt_pitch_xcorr_arm.s
+++ b/drivers/opus/celt/arm/celt_pitch_xcorr_arm.s
@@ -42,6 +42,7 @@ IF OPUS_ARM_MAY_HAVE_NEON
 
 ; Compute sum[k]=sum(x[j]*y[j+k],j=0...len-1), k=0...3
 xcorr_kernel_neon PROC
+xcorr_kernel_neon_start
   ; input:
   ;   r3     = int         len
   ;   r4     = opus_val16 *x
@@ -181,7 +182,7 @@ celt_pitch_xcorr_neon_process4
   VEOR         q0, q0, q0
   ; xcorr_kernel_neon only modifies r4, r5, r12, and q0...q3.
   ; So we don't save/restore any other registers.
-  BL xcorr_kernel_neon
+  BL xcorr_kernel_neon_start
   SUBS         r6, r6, #4
   VST1.32      {q0}, [r2]!
   ; _y += 4
@@ -257,6 +258,7 @@ IF OPUS_ARM_MAY_HAVE_EDSP
 ; This will get used on ARMv7 devices without NEON, so it has been optimized
 ; to take advantage of dual-issuing where possible.
 xcorr_kernel_edsp PROC
+xcorr_kernel_edsp_start
   ; input:
   ;   r3      = int         len
   ;   r4      = opus_val16 *_x (must be 32-bit aligned)
@@ -309,7 +311,7 @@ xcorr_kernel_edsp_process4_done
   SUBS         r2, r2, #1         ; j--
   ; Stall
   SMLABB       r6, r12, r10, r6   ; sum[0] = MAC16_16(sum[0],x,y_0)
-  LDRGTH       r14, [r4], #2      ; r14 = *x++
+  LDRHGT       r14, [r4], #2      ; r14 = *x++
   SMLABT       r7, r12, r10, r7   ; sum[1] = MAC16_16(sum[1],x,y_1)
   SMLABB       r8, r12, r11, r8   ; sum[2] = MAC16_16(sum[2],x,y_2)
   SMLABT       r9, r12, r11, r9   ; sum[3] = MAC16_16(sum[3],x,y_3)
@@ -319,7 +321,7 @@ xcorr_kernel_edsp_process4_done
   SMLABB       r7, r14, r11, r7   ; sum[1] = MAC16_16(sum[1],x,y_2)
   LDRH         r10, [r5], #2      ; r10 = y_4 = *y++
   SMLABT       r8, r14, r11, r8   ; sum[2] = MAC16_16(sum[2],x,y_3)
-  LDRGTH       r12, [r4], #2      ; r12 = *x++
+  LDRHGT       r12, [r4], #2      ; r12 = *x++
   SMLABB       r9, r14, r10, r9   ; sum[3] = MAC16_16(sum[3],x,y_4)
   BLE xcorr_kernel_edsp_done
   SMLABB       r6, r12, r11, r6   ; sum[0] = MAC16_16(sum[0],tmp,y_2)
@@ -327,7 +329,7 @@ xcorr_kernel_edsp_process4_done
   SMLABT       r7, r12, r11, r7   ; sum[1] = MAC16_16(sum[1],tmp,y_3)
   LDRH         r2, [r5], #2       ; r2 = y_5 = *y++
   SMLABB       r8, r12, r10, r8   ; sum[2] = MAC16_16(sum[2],tmp,y_4)
-  LDRGTH       r14, [r4]          ; r14 = *x
+  LDRHGT       r14, [r4]          ; r14 = *x
   SMLABB       r9, r12, r2, r9    ; sum[3] = MAC16_16(sum[3],tmp,y_5)
   BLE xcorr_kernel_edsp_done
   SMLABT       r6, r14, r11, r6   ; sum[0] = MAC16_16(sum[0],tmp,y_3)
@@ -387,11 +389,11 @@ celt_pitch_xcorr_edsp_process1u_loop4
 celt_pitch_xcorr_edsp_process1u_loop4_done
   ADDS         r12, r12, #4
 celt_pitch_xcorr_edsp_process1u_loop1
-  LDRGEH       r6, [r4], #2
+  LDRHGE       r6, [r4], #2
   ; Stall
   SMLABBGE     r14, r6, r8, r14    ; sum = MAC16_16(sum, *x, *y)
-  SUBGES       r12, r12, #1
-  LDRGTH       r8, [r5], #2
+  SUBSGE       r12, r12, #1
+  LDRHGT       r8, [r5], #2
   BGT celt_pitch_xcorr_edsp_process1u_loop1
   ; Restore _x
   SUB          r4, r4, r3, LSL #1
@@ -416,7 +418,7 @@ celt_pitch_xcorr_edsp_process4
   MOV          r7, #0
   MOV          r8, #0
   MOV          r9, #0
-  BL xcorr_kernel_edsp  ; xcorr_kernel_edsp(_x, _y+i, xcorr+i, len)
+  BL xcorr_kernel_edsp_start  ; xcorr_kernel_edsp(_x, _y+i, xcorr+i, len)
   ; maxcorr = max(maxcorr, sum0, sum1, sum2, sum3)
   CMP          r0, r6
   ; _y+=4
@@ -474,7 +476,7 @@ celt_pitch_xcorr_edsp_process2_1
   ADDS         r12, r12, #1
   ; Stall
   SMLABB       r10, r6, r8, r10     ; sum0 = MAC16_16(sum0, x_0, y_0)
-  LDRGTH       r7, [r4], #2
+  LDRHGT       r7, [r4], #2
   SMLABT       r11, r6, r8, r11     ; sum1 = MAC16_16(sum1, x_0, y_1)
   BLE celt_pitch_xcorr_edsp_process2_done
   LDRH         r9, [r5], #2
@@ -527,8 +529,8 @@ celt_pitch_xcorr_edsp_process1a_loop_done
   SUBGE        r12, r12, #2
   SMLATTGE     r14, r6, r8, r14     ; sum = MAC16_16(sum, x_1, y_1)
   ADDS         r12, r12, #1
-  LDRGEH       r6, [r4], #2
-  LDRGEH       r8, [r5], #2
+  LDRHGE       r6, [r4], #2
+  LDRHGE       r8, [r5], #2
   ; Stall
   SMLABBGE     r14, r6, r8, r14     ; sum = MAC16_16(sum, *x, *y)
   ; maxcorr = max(maxcorr, sum)
diff --git a/drivers/opus/celt/arm/armopts.s b/drivers/opus/celt/arm/fft_arm.h
index fb9196072a..c99458b4af 100644
--- a/drivers/opus/celt/arm/armopts.s
+++ b/drivers/opus/celt/arm/fft_arm.h
@@ -1,4 +1,10 @@
-/* Copyright (C) 2013 Mozilla Corporation */
+/* Copyright (c) 2015 Xiph.Org Foundation
+   Written by Viswanath Puttagunta */
+/**
+   @file fft_arm.h
+   @brief ARM Neon Intrinsic optimizations for fft using NE10 library
+ */
+
 /*
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions
@@ -24,14 +30,43 @@
    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
 
-; Set the following to 1 if we have EDSP instructions
-;  (LDRD/STRD, etc., ARMv5E and later).
-OPUS_ARM_MAY_HAVE_EDSP  * 
 
-; Set the following to 1 if we have ARMv6 media instructions.
-OPUS_ARM_MAY_HAVE_MEDIA * 
+#if !defined(FFT_ARM_H)
+#define FFT_ARM_H
+
+#include "opus/opus_config.h"
+#include "opus/celt/kiss_fft.h"
+
+#if defined(HAVE_ARM_NE10)
+
+int opus_fft_alloc_arm_neon(kiss_fft_state *st);
+void opus_fft_free_arm_neon(kiss_fft_state *st);
+
+void opus_fft_neon(const kiss_fft_state *st,
+                   const kiss_fft_cpx *fin,
+                   kiss_fft_cpx *fout);
+
+void opus_ifft_neon(const kiss_fft_state *st,
+                    const kiss_fft_cpx *fin,
+                    kiss_fft_cpx *fout);
+
+#if !defined(OPUS_HAVE_RTCD)
+#define OVERRIDE_OPUS_FFT (1)
+
+#define opus_fft_alloc_arch(_st, arch) \
+   ((void)(arch), opus_fft_alloc_arm_neon(_st))
+
+#define opus_fft_free_arch(_st, arch) \
+   ((void)(arch), opus_fft_free_arm_neon(_st))
+
+#define opus_fft(_st, _fin, _fout, arch) \
+   ((void)(arch), opus_fft_neon(_st, _fin, _fout))
+
+#define opus_ifft(_st, _fin, _fout, arch) \
+   ((void)(arch), opus_ifft_neon(_st, _fin, _fout))
+
+#endif /* OPUS_HAVE_RTCD */
 
-; Set the following to 1 if we have NEON (some ARMv7)
-OPUS_ARM_MAY_HAVE_NEON  * 
+#endif /* HAVE_ARM_NE10 */
 
-END
+#endif
diff --git a/drivers/opus/celt/arm/fixed_armv4.h b/drivers/opus/celt/arm/fixed_armv4.h
index b690bc8cea..efb3b1896a 100644
--- a/drivers/opus/celt/arm/fixed_armv4.h
+++ b/drivers/opus/celt/arm/fixed_armv4.h
@@ -68,6 +68,10 @@ static OPUS_INLINE opus_val32 MULT16_32_Q15_armv4(opus_val16 a, opus_val32 b)
 #undef MAC16_32_Q15
 #define MAC16_32_Q15(c, a, b) ADD32(c, MULT16_32_Q15(a, b))
 
+/** 16x32 multiply, followed by a 16-bit shift right and 32-bit add.
+    Result fits in 32 bits. */
+#undef MAC16_32_Q16
+#define MAC16_32_Q16(c, a, b) ADD32(c, MULT16_32_Q16(a, b))
 
 /** 32x32 multiplication, followed by a 31-bit shift right. Results fits in 32 bits */
 #undef MULT32_32_Q31
diff --git a/drivers/opus/celt/arm/fixed_armv5e.h b/drivers/opus/celt/arm/fixed_armv5e.h
index cb6e4c1da9..2db23262e8 100644
--- a/drivers/opus/celt/arm/fixed_armv5e.h
+++ b/drivers/opus/celt/arm/fixed_armv5e.h
@@ -82,6 +82,23 @@ static OPUS_INLINE opus_val32 MAC16_32_Q15_armv5e(opus_val32 c, opus_val16 a,
 }
 #define MAC16_32_Q15(c, a, b) (MAC16_32_Q15_armv5e(c, a, b))
 
+/** 16x32 multiply, followed by a 16-bit shift right and 32-bit add.
+    Result fits in 32 bits. */
+#undef MAC16_32_Q16
+static OPUS_INLINE opus_val32 MAC16_32_Q16_armv5e(opus_val32 c, opus_val16 a,
+ opus_val32 b)
+{
+  int res;
+  __asm__(
+      "#MAC16_32_Q16\n\t"
+      "smlawb %0, %1, %2, %3;\n"
+      : "=r"(res)
+      : "r"(b), "r"(a), "r"(c)
+  );
+  return res;
+}
+#define MAC16_32_Q16(c, a, b) (MAC16_32_Q16_armv5e(c, a, b))
+
 /** 16x16 multiply-add where the result fits in 32 bits */
 #undef MAC16_16
 static OPUS_INLINE opus_val32 MAC16_16_armv5e(opus_val32 c, opus_val16 a,
@@ -113,4 +130,22 @@ static OPUS_INLINE opus_val32 MULT16_16_armv5e(opus_val16 a, opus_val16 b)
 }
 #define MULT16_16(a, b) (MULT16_16_armv5e(a, b))
 
+#ifdef OPUS_ARM_INLINE_MEDIA
+
+#undef SIG2WORD16
+static OPUS_INLINE opus_val16 SIG2WORD16_armv6(opus_val32 x)
+{
+   celt_sig res;
+   __asm__(
+       "#SIG2WORD16\n\t"
+       "ssat %0, #16, %1, ASR #12\n\t"
+       : "=r"(res)
+       : "r"(x+2048)
+   );
+   return EXTRACT16(res);
+}
+#define SIG2WORD16(x) (SIG2WORD16_armv6(x))
+
+#endif /* OPUS_ARM_INLINE_MEDIA */
+
 #endif
diff --git a/drivers/opus/celt/arm/kiss_fft_armv4.h b/drivers/opus/celt/arm/kiss_fft_armv4.h
index 773464628b..0f5617324d 100644
--- a/drivers/opus/celt/arm/kiss_fft_armv4.h
+++ b/drivers/opus/celt/arm/kiss_fft_armv4.h
@@ -116,6 +116,6 @@
     } \
     while(0)
 
-#endif /* OPUS_FIXED_POINT */
+#endif /* FIXED_POINT */
 
 #endif /* KISS_FFT_ARMv4_H */
diff --git a/drivers/opus/celt/arm/kiss_fft_armv5e.h b/drivers/opus/celt/arm/kiss_fft_armv5e.h
index 1eff56a66a..0ab75fbc58 100644
--- a/drivers/opus/celt/arm/kiss_fft_armv5e.h
+++ b/drivers/opus/celt/arm/kiss_fft_armv5e.h
@@ -113,6 +113,6 @@
     } \
     while(0)
 
-#endif /* OPUS_FIXED_POINT */
+#endif /* FIXED_POINT */
 
 #endif /* KISS_FFT_GUTS_H */
diff --git a/drivers/opus/celt/arm/mdct_arm.h b/drivers/opus/celt/arm/mdct_arm.h
new file mode 100644
index 0000000000..faf3c8acab
--- /dev/null
+++ b/drivers/opus/celt/arm/mdct_arm.h
@@ -0,0 +1,60 @@
+/* Copyright (c) 2015 Xiph.Org Foundation
+   Written by Viswanath Puttagunta */
+/**
+   @file arm_mdct.h
+   @brief ARM Neon Intrinsic optimizations for mdct using NE10 library
+ */
+
+/*
+   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.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#if !defined(MDCT_ARM_H)
+#define MDCT_ARM_H
+
+#include "opus/opus_config.h"
+#include "opus/celt/mdct.h"
+
+#if defined(HAVE_ARM_NE10)
+/** Compute a forward MDCT and scale by 4/N, trashes the input array */
+void clt_mdct_forward_neon(const mdct_lookup *l, kiss_fft_scalar *in,
+                           kiss_fft_scalar * OPUS_RESTRICT out,
+                           const opus_val16 *window, int overlap,
+                           int shift, int stride, int arch);
+
+void clt_mdct_backward_neon(const mdct_lookup *l, kiss_fft_scalar *in,
+                            kiss_fft_scalar * OPUS_RESTRICT out,
+                            const opus_val16 *window, int overlap,
+                            int shift, int stride, int arch);
+
+#if !defined(OPUS_HAVE_RTCD)
+#define OVERRIDE_OPUS_MDCT (1)
+#define clt_mdct_forward(_l, _in, _out, _window, _int, _shift, _stride, _arch) \
+      clt_mdct_forward_neon(_l, _in, _out, _window, _int, _shift, _stride, _arch)
+#define clt_mdct_backward(_l, _in, _out, _window, _int, _shift, _stride, _arch) \
+      clt_mdct_backward_neon(_l, _in, _out, _window, _int, _shift, _stride, _arch)
+#endif /* OPUS_HAVE_RTCD */
+#endif /* HAVE_ARM_NE10 */
+
+#endif
diff --git a/drivers/opus/celt/arm/pitch_arm.h b/drivers/opus/celt/arm/pitch_arm.h
index 18d1f2e75e..996fbc7fd9 100644
--- a/drivers/opus/celt/arm/pitch_arm.h
+++ b/drivers/opus/celt/arm/pitch_arm.h
@@ -30,7 +30,7 @@
 
 # include "opus/celt/arm/armcpu.h"
 
-# if defined(OPUS_FIXED_POINT)
+# if defined(FIXED_POINT)
 
 #  if defined(OPUS_ARM_MAY_HAVE_NEON)
 opus_val32 celt_pitch_xcorr_neon(const opus_val16 *_x, const opus_val16 *_y,
@@ -52,6 +52,17 @@ opus_val32 celt_pitch_xcorr_edsp(const opus_val16 *_x, const opus_val16 *_y,
   ((void)(arch),PRESUME_NEON(celt_pitch_xcorr)(_x, _y, xcorr, len, max_pitch))
 #  endif
 
-# endif
+#else /* Start !FIXED_POINT */
+/* Float case */
+#if defined(OPUS_ARM_MAY_HAVE_NEON_INTR)
+void celt_pitch_xcorr_float_neon(const opus_val16 *_x, const opus_val16 *_y,
+                                 opus_val32 *xcorr, int len, int max_pitch);
+#if !defined(OPUS_HAVE_RTCD) || defined(OPUS_ARM_PRESUME_NEON_INTR)
+#define OVERRIDE_PITCH_XCORR (1)
+#   define celt_pitch_xcorr(_x, _y, xcorr, len, max_pitch, arch) \
+   ((void)(arch),celt_pitch_xcorr_float_neon(_x, _y, xcorr, len, max_pitch))
+#endif
+#endif
 
+#endif /* end !FIXED_POINT */
 #endif
diff --git a/drivers/opus/celt/bands.c b/drivers/opus/celt/bands.c
index 5a6b23d87f..bdd87dd327 100644
--- a/drivers/opus/celt/bands.c
+++ b/drivers/opus/celt/bands.c
@@ -26,14 +26,11 @@
    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
-
-#ifdef OPUS_ENABLED
 #include "opus/opus_config.h"
-#endif
 
 #include <math.h>
 #include "opus/celt/bands.h"
-#include "opus/celt/opus_modes.h"
+#include "opus/celt/modes.h"
 #include "opus/celt/vq.h"
 #include "opus/celt/cwrs.h"
 #include "opus/celt/stack_alloc.h"
@@ -92,11 +89,11 @@ static int bitexact_log2tan(int isin,int icos)
 
 #ifdef OPUS_FIXED_POINT
 /* Compute the amplitude (sqrt energy) in each of the bands */
-void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int M)
+void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int LM)
 {
    int i, c, N;
    const opus_int16 *eBands = m->eBands;
-   N = M*m->shortMdctSize;
+   N = m->shortMdctSize<<LM;
    c=0; do {
       for (i=0;i<end;i++)
       {
@@ -104,18 +101,23 @@ void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *band
          opus_val32 maxval=0;
          opus_val32 sum = 0;
 
-         j=M*eBands[i]; do {
-            maxval = MAX32(maxval, X[j+c*N]);
-            maxval = MAX32(maxval, -X[j+c*N]);
-         } while (++j<M*eBands[i+1]);
-
+         maxval = celt_maxabs32(&X[c*N+(eBands[i]<<LM)], (eBands[i+1]-eBands[i])<<LM);
          if (maxval > 0)
          {
-            int shift = celt_ilog2(maxval)-10;
-            j=M*eBands[i]; do {
-               sum = MAC16_16(sum, EXTRACT16(VSHR32(X[j+c*N],shift)),
-                                   EXTRACT16(VSHR32(X[j+c*N],shift)));
-            } while (++j<M*eBands[i+1]);
+            int shift = celt_ilog2(maxval) - 14 + (((m->logN[i]>>BITRES)+LM+1)>>1);
+            j=eBands[i]<<LM;
+            if (shift>0)
+            {
+               do {
+                  sum = MAC16_16(sum, EXTRACT16(SHR32(X[j+c*N],shift)),
+                        EXTRACT16(SHR32(X[j+c*N],shift)));
+               } while (++j<eBands[i+1]<<LM);
+            } else {
+               do {
+                  sum = MAC16_16(sum, EXTRACT16(SHL32(X[j+c*N],-shift)),
+                        EXTRACT16(SHL32(X[j+c*N],-shift)));
+               } while (++j<eBands[i+1]<<LM);
+            }
             /* We're adding one here to ensure the normalized band isn't larger than unity norm */
             bandE[i+c*m->nbEBands] = EPSILON+VSHR32(EXTEND32(celt_sqrt(sum)),-shift);
          } else {
@@ -148,20 +150,18 @@ void normalise_bands(const CELTMode *m, const celt_sig * OPUS_RESTRICT freq, cel
    } while (++c<C);
 }
 
-#else /* OPUS_FIXED_POINT */
+#else /* FIXED_POINT */
 /* Compute the amplitude (sqrt energy) in each of the bands */
-void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int M)
+void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int LM)
 {
    int i, c, N;
    const opus_int16 *eBands = m->eBands;
-   N = M*m->shortMdctSize;
+   N = m->shortMdctSize<<LM;
    c=0; do {
       for (i=0;i<end;i++)
       {
-         int j;
-         opus_val32 sum = 1e-27f;
-         for (j=M*eBands[i];j<M*eBands[i+1];j++)
-            sum += X[j+c*N]*X[j+c*N];
+         opus_val32 sum;
+         sum = 1e-27f + celt_inner_prod_c(&X[c*N+(eBands[i]<<LM)], &X[c*N+(eBands[i]<<LM)], (eBands[i+1]-eBands[i])<<LM);
          bandE[i+c*m->nbEBands] = celt_sqrt(sum);
          /*printf ("%f ", bandE[i+c*m->nbEBands]);*/
       }
@@ -186,78 +186,84 @@ void normalise_bands(const CELTMode *m, const celt_sig * OPUS_RESTRICT freq, cel
    } while (++c<C);
 }
 
-#endif /* OPUS_FIXED_POINT */
+#endif /* FIXED_POINT */
 
 /* De-normalise the energy to produce the synthesis from the unit-energy bands */
 void denormalise_bands(const CELTMode *m, const celt_norm * OPUS_RESTRICT X,
-      celt_sig * OPUS_RESTRICT freq, const opus_val16 *bandLogE, int start, int end, int C, int M)
+      celt_sig * OPUS_RESTRICT freq, const opus_val16 *bandLogE, int start,
+      int end, int M, int downsample, int silence)
 {
-   int i, c, N;
+   int i, N;
+   int bound;
+   celt_sig * OPUS_RESTRICT f;
+   const celt_norm * OPUS_RESTRICT x;
    const opus_int16 *eBands = m->eBands;
    N = M*m->shortMdctSize;
-   celt_assert2(C<=2, "denormalise_bands() not implemented for >2 channels");
-   c=0; do {
-      celt_sig * OPUS_RESTRICT f;
-      const celt_norm * OPUS_RESTRICT x;
-      f = freq+c*N;
-      x = X+c*N+M*eBands[start];
-      for (i=0;i<M*eBands[start];i++)
-         *f++ = 0;
-      for (i=start;i<end;i++)
-      {
-         int j, band_end;
-         opus_val16 g;
-         opus_val16 lg;
+   bound = M*eBands[end];
+   if (downsample!=1)
+      bound = IMIN(bound, N/downsample);
+   if (silence)
+   {
+      bound = 0;
+      start = end = 0;
+   }
+   f = freq;
+   x = X+M*eBands[start];
+   for (i=0;i<M*eBands[start];i++)
+      *f++ = 0;
+   for (i=start;i<end;i++)
+   {
+      int j, band_end;
+      opus_val16 g;
+      opus_val16 lg;
 #ifdef OPUS_FIXED_POINT
-         int shift;
+      int shift;
 #endif
-         j=M*eBands[i];
-         band_end = M*eBands[i+1];
-         lg = ADD16(bandLogE[i+c*m->nbEBands], SHL16((opus_val16)eMeans[i],6));
+      j=M*eBands[i];
+      band_end = M*eBands[i+1];
+      lg = ADD16(bandLogE[i], SHL16((opus_val16)eMeans[i],6));
 #ifndef OPUS_FIXED_POINT
-         g = celt_exp2(lg);
+      g = celt_exp2(lg);
 #else
-         /* Handle the integer part of the log energy */
-         shift = 16-(lg>>DB_SHIFT);
-         if (shift>31)
-         {
-            shift=0;
-            g=0;
-         } else {
-            /* Handle the fractional part. */
-            g = celt_exp2_frac(lg&((1<<DB_SHIFT)-1));
-         }
-         /* Handle extreme gains with negative shift. */
-         if (shift<0)
-         {
-            /* For shift < -2 we'd be likely to overflow, so we're capping
+      /* Handle the integer part of the log energy */
+      shift = 16-(lg>>DB_SHIFT);
+      if (shift>31)
+      {
+         shift=0;
+         g=0;
+      } else {
+         /* Handle the fractional part. */
+         g = celt_exp2_frac(lg&((1<<DB_SHIFT)-1));
+      }
+      /* Handle extreme gains with negative shift. */
+      if (shift<0)
+      {
+         /* For shift < -2 we'd be likely to overflow, so we're capping
                the gain here. This shouldn't happen unless the bitstream is
                already corrupted. */
-            if (shift < -2)
-            {
-               g = 32767;
-               shift = -2;
-            }
-            do {
-               *f++ = SHL32(MULT16_16(*x++, g), -shift);
-            } while (++j<band_end);
-         } else
+         if (shift < -2)
+         {
+            g = 32767;
+            shift = -2;
+         }
+         do {
+            *f++ = SHL32(MULT16_16(*x++, g), -shift);
+         } while (++j<band_end);
+      } else
 #endif
          /* Be careful of the fixed-point "else" just above when changing this code */
          do {
             *f++ = SHR32(MULT16_16(*x++, g), shift);
          } while (++j<band_end);
-      }
-      celt_assert(start <= end);
-      for (i=M*eBands[end];i<N;i++)
-         *f++ = 0;
-   } while (++c<C);
+   }
+   celt_assert(start <= end);
+   OPUS_CLEAR(&freq[bound], N-bound);
 }
 
 /* This prevents energy collapse for transients with multiple short MDCTs */
 void anti_collapse(const CELTMode *m, celt_norm *X_, unsigned char *collapse_masks, int LM, int C, int size,
-      int start, int end, opus_val16 *logE, opus_val16 *prev1logE,
-      opus_val16 *prev2logE, int *pulses, opus_uint32 seed)
+      int start, int end, const opus_val16 *logE, const opus_val16 *prev1logE,
+      const opus_val16 *prev2logE, const int *pulses, opus_uint32 seed, int arch)
 {
    int c, i, j, k;
    for (i=start;i<end;i++)
@@ -272,7 +278,8 @@ void anti_collapse(const CELTMode *m, celt_norm *X_, unsigned char *collapse_mas
 
       N0 = m->eBands[i+1]-m->eBands[i];
       /* depth in 1/8 bits */
-      depth = (1+pulses[i])/((m->eBands[i+1]-m->eBands[i])<<LM);
+      celt_assert(pulses[i]>=0);
+      depth = celt_udiv(1+pulses[i], (m->eBands[i+1]-m->eBands[i]))>>LM;
 
 #ifdef OPUS_FIXED_POINT
       thresh32 = SHR32(celt_exp2(-SHL16(depth, 10-BITRES)),1);
@@ -345,12 +352,12 @@ void anti_collapse(const CELTMode *m, celt_norm *X_, unsigned char *collapse_mas
          }
          /* We just added some energy, so we need to renormalise */
          if (renormalize)
-            renormalise_vector(X, N0<<LM, Q15ONE);
+            renormalise_vector(X, N0<<LM, Q15ONE, arch);
       } while (++c<C);
    }
 }
 
-static void intensity_stereo(const CELTMode *m, celt_norm *X, celt_norm *Y, const celt_ener *bandE, int bandID, int N)
+static void intensity_stereo(const CELTMode *m, celt_norm * OPUS_RESTRICT X, const celt_norm * OPUS_RESTRICT Y, const celt_ener *bandE, int bandID, int N)
 {
    int i = bandID;
    int j;
@@ -370,25 +377,25 @@ static void intensity_stereo(const CELTMode *m, celt_norm *X, celt_norm *Y, cons
       celt_norm r, l;
       l = X[j];
       r = Y[j];
-      X[j] = MULT16_16_Q14(a1,l) + MULT16_16_Q14(a2,r);
+      X[j] = EXTRACT16(SHR32(MAC16_16(MULT16_16(a1, l), a2, r), 14));
       /* Side is not encoded, no need to calculate */
    }
 }
 
-static void stereo_split(celt_norm *X, celt_norm *Y, int N)
+static void stereo_split(celt_norm * OPUS_RESTRICT X, celt_norm * OPUS_RESTRICT Y, int N)
 {
    int j;
    for (j=0;j<N;j++)
    {
-      celt_norm r, l;
-      l = MULT16_16_Q15(QCONST16(.70710678f,15), X[j]);
-      r = MULT16_16_Q15(QCONST16(.70710678f,15), Y[j]);
-      X[j] = l+r;
-      Y[j] = r-l;
+      opus_val32 r, l;
+      l = MULT16_16(QCONST16(.70710678f, 15), X[j]);
+      r = MULT16_16(QCONST16(.70710678f, 15), Y[j]);
+      X[j] = EXTRACT16(SHR32(ADD32(l, r), 15));
+      Y[j] = EXTRACT16(SHR32(SUB32(r, l), 15));
    }
 }
 
-static void stereo_merge(celt_norm *X, celt_norm *Y, opus_val16 mid, int N)
+static void stereo_merge(celt_norm * OPUS_RESTRICT X, celt_norm * OPUS_RESTRICT Y, opus_val16 mid, int N, int arch)
 {
    int j;
    opus_val32 xp=0, side=0;
@@ -400,7 +407,7 @@ static void stereo_merge(celt_norm *X, celt_norm *Y, opus_val16 mid, int N)
    opus_val32 t, lgain, rgain;
 
    /* Compute the norm of X+Y and X-Y as |X|^2 + |Y|^2 +/- sum(xy) */
-   dual_inner_prod(Y, X, Y, N, &xp, &side);
+   dual_inner_prod(Y, X, Y, N, &xp, &side, arch);
    /* Compensating for the mid normalization */
    xp = MULT16_32_Q15(mid, xp);
    /* mid and side are in Q15, not Q14 like X and Y */
@@ -409,8 +416,7 @@ static void stereo_merge(celt_norm *X, celt_norm *Y, opus_val16 mid, int N)
    Er = MULT16_16(mid2, mid2) + side + 2*xp;
    if (Er < QCONST32(6e-4f, 28) || El < QCONST32(6e-4f, 28))
    {
-      for (j=0;j<N;j++)
-         Y[j] = X[j];
+      OPUS_COPY(Y, X, N);
       return;
    }
 
@@ -434,7 +440,7 @@ static void stereo_merge(celt_norm *X, celt_norm *Y, opus_val16 mid, int N)
    {
       celt_norm r, l;
       /* Apply mid scaling (side is already scaled) */
-      l = MULT16_16_Q15(mid, X[j]);
+      l = MULT16_16_P15(mid, X[j]);
       r = Y[j];
       X[j] = EXTRACT16(PSHR32(MULT16_16(lgain, SUB16(l,r)), kl+1));
       Y[j] = EXTRACT16(PSHR32(MULT16_16(rgain, ADD16(l,r)), kr+1));
@@ -442,7 +448,7 @@ static void stereo_merge(celt_norm *X, celt_norm *Y, opus_val16 mid, int N)
 }
 
 /* Decide whether we should spread the pulses in the current frame */
-int spreading_decision(const CELTMode *m, celt_norm *X, int *average,
+int spreading_decision(const CELTMode *m, const celt_norm *X, int *average,
       int last_decision, int *hf_average, int *tapset_decision, int update_hf,
       int end, int C, int M)
 {
@@ -463,7 +469,7 @@ int spreading_decision(const CELTMode *m, celt_norm *X, int *average,
       {
          int j, N, tmp=0;
          int tcount[3] = {0,0,0};
-         celt_norm * OPUS_RESTRICT x = X+M*eBands[i]+c*N0;
+         const celt_norm * OPUS_RESTRICT x = X+M*eBands[i]+c*N0;
          N = M*(eBands[i+1]-eBands[i]);
          if (N<=8)
             continue;
@@ -483,7 +489,7 @@ int spreading_decision(const CELTMode *m, celt_norm *X, int *average,
 
          /* Only include four last bands (8 kHz and up) */
          if (i>m->nbEBands-4)
-            hf_sum += 32*(tcount[1]+tcount[0])/N;
+            hf_sum += celt_udiv(32*(tcount[1]+tcount[0]), N);
          tmp = (2*tcount[2] >= N) + (2*tcount[1] >= N) + (2*tcount[0] >= N);
          sum += tmp*256;
          nbBands++;
@@ -493,7 +499,7 @@ int spreading_decision(const CELTMode *m, celt_norm *X, int *average,
    if (update_hf)
    {
       if (hf_sum)
-         hf_sum /= C*(4-m->nbEBands+end);
+         hf_sum = celt_udiv(hf_sum, C*(4-m->nbEBands+end));
       *hf_average = (*hf_average+hf_sum)>>1;
       hf_sum = *hf_average;
       if (*tapset_decision==2)
@@ -509,7 +515,8 @@ int spreading_decision(const CELTMode *m, celt_norm *X, int *average,
    }
    /*printf("%d %d %d\n", hf_sum, *hf_average, *tapset_decision);*/
    celt_assert(nbBands>0); /* end has to be non-zero */
-   sum /= nbBands;
+   celt_assert(sum>=0);
+   sum = celt_udiv(sum, nbBands);
    /* Recursive averaging */
    sum = (sum+*average)>>1;
    *average = sum;
@@ -567,8 +574,7 @@ static void deinterleave_hadamard(celt_norm *X, int N0, int stride, int hadamard
          for (j=0;j<N0;j++)
             tmp[i*N0+j] = X[j*stride+i];
    }
-   for (j=0;j<N;j++)
-      X[j] = tmp[j];
+   OPUS_COPY(X, tmp, N);
    RESTORE_STACK;
 }
 
@@ -591,8 +597,7 @@ static void interleave_hadamard(celt_norm *X, int N0, int stride, int hadamard)
          for (j=0;j<N0;j++)
             tmp[j*stride+i] = X[i*N0+j];
    }
-   for (j=0;j<N;j++)
-      X[j] = tmp[j];
+   OPUS_COPY(X, tmp, N);
    RESTORE_STACK;
 }
 
@@ -603,11 +608,11 @@ void haar1(celt_norm *X, int N0, int stride)
    for (i=0;i<stride;i++)
       for (j=0;j<N0;j++)
       {
-         celt_norm tmp1, tmp2;
-         tmp1 = MULT16_16_Q15(QCONST16(.70710678f,15), X[stride*2*j+i]);
-         tmp2 = MULT16_16_Q15(QCONST16(.70710678f,15), X[stride*(2*j+1)+i]);
-         X[stride*2*j+i] = tmp1 + tmp2;
-         X[stride*(2*j+1)+i] = tmp1 - tmp2;
+         opus_val32 tmp1, tmp2;
+         tmp1 = MULT16_16(QCONST16(.70710678f,15), X[stride*2*j+i]);
+         tmp2 = MULT16_16(QCONST16(.70710678f,15), X[stride*(2*j+1)+i]);
+         X[stride*2*j+i] = EXTRACT16(PSHR32(ADD32(tmp1, tmp2), 15));
+         X[stride*(2*j+1)+i] = EXTRACT16(PSHR32(SUB32(tmp1, tmp2), 15));
       }
 }
 
@@ -622,7 +627,8 @@ static int compute_qn(int N, int b, int offset, int pulse_cap, int stereo)
    /* The upper limit ensures that in a stereo split with itheta==16384, we'll
        always have enough bits left over to code at least one pulse in the
        side; otherwise it would collapse, since it doesn't get folded. */
-   qb = IMIN(b-pulse_cap-(4<<BITRES), (b+N2*offset)/N2);
+   qb = celt_sudiv(b+N2*offset, N2);
+   qb = IMIN(b-pulse_cap-(4<<BITRES), qb);
 
    qb = IMIN(8<<BITRES, qb);
 
@@ -647,6 +653,7 @@ struct band_ctx {
    opus_int32 remaining_bits;
    const celt_ener *bandE;
    opus_uint32 seed;
+   int arch;
 };
 
 struct split_ctx {
@@ -698,7 +705,7 @@ static void compute_theta(struct band_ctx *ctx, struct split_ctx *sctx,
          side and mid. With just that parameter, we can re-scale both
          mid and side because we know that 1) they have unit norm and
          2) they are orthogonal. */
-      itheta = stereo_itheta(X, Y, stereo, N);
+      itheta = stereo_itheta(X, Y, stereo, N, ctx->arch);
    }
    tell = ec_tell_frac(ec);
    if (qn!=1)
@@ -769,7 +776,8 @@ static void compute_theta(struct band_ctx *ctx, struct split_ctx *sctx,
             ec_dec_update(ec, fl, fl+fs, ft);
          }
       }
-      itheta = (opus_int32)itheta*16384/qn;
+      celt_assert(itheta>=0);
+      itheta = celt_udiv((opus_int32)itheta*16384, qn);
       if (encode && stereo)
       {
          if (itheta==0)
@@ -1021,8 +1029,7 @@ static unsigned quant_partition(struct band_ctx *ctx, celt_norm *X,
             fill &= cm_mask;
             if (!fill)
             {
-               for (j=0;j<N;j++)
-                  X[j] = 0;
+               OPUS_CLEAR(X, N);
             } else {
                if (lowband == NULL)
                {
@@ -1046,7 +1053,7 @@ static unsigned quant_partition(struct band_ctx *ctx, celt_norm *X,
                   }
                   cm = fill;
                }
-               renormalise_vector(X, N, gain);
+               renormalise_vector(X, N, gain, ctx->arch);
             }
          }
       }
@@ -1084,7 +1091,7 @@ static unsigned quant_band(struct band_ctx *ctx, celt_norm *X,
 
    longBlocks = B0==1;
 
-   N_B /= B;
+   N_B = celt_udiv(N_B, B);
 
    /* Special case for one sample */
    if (N==1)
@@ -1098,9 +1105,7 @@ static unsigned quant_band(struct band_ctx *ctx, celt_norm *X,
 
    if (lowband_scratch && lowband && (recombine || ((N_B&1) == 0 && tf_change<0) || B0>1))
    {
-      int j;
-      for (j=0;j<N;j++)
-         lowband_scratch[j] = lowband[j];
+      OPUS_COPY(lowband_scratch, lowband, N);
       lowband = lowband_scratch;
    }
 
@@ -1340,7 +1345,7 @@ static unsigned quant_band_stereo(struct band_ctx *ctx, celt_norm *X, celt_norm
    if (resynth)
    {
       if (N!=2)
-         stereo_merge(X, Y, mid, N);
+         stereo_merge(X, Y, mid, N, ctx->arch);
       if (inv)
       {
          int j;
@@ -1353,9 +1358,11 @@ static unsigned quant_band_stereo(struct band_ctx *ctx, celt_norm *X, celt_norm
 
 
 void quant_all_bands(int encode, const CELTMode *m, int start, int end,
-      celt_norm *X_, celt_norm *Y_, unsigned char *collapse_masks, const celt_ener *bandE, int *pulses,
-      int shortBlocks, int spread, int dual_stereo, int intensity, int *tf_res,
-      opus_int32 total_bits, opus_int32 balance, ec_ctx *ec, int LM, int codedBands, opus_uint32 *seed)
+      celt_norm *X_, celt_norm *Y_, unsigned char *collapse_masks,
+      const celt_ener *bandE, int *pulses, int shortBlocks, int spread,
+      int dual_stereo, int intensity, int *tf_res, opus_int32 total_bits,
+      opus_int32 balance, ec_ctx *ec, int LM, int codedBands,
+      opus_uint32 *seed, int arch)
 {
    int i;
    opus_int32 remaining_bits;
@@ -1397,6 +1404,7 @@ void quant_all_bands(int encode, const CELTMode *m, int start, int end,
    ctx.m = m;
    ctx.seed = *seed;
    ctx.spread = spread;
+   ctx.arch = arch;
    for (i=start;i<end;i++)
    {
       opus_int32 tell;
@@ -1428,7 +1436,7 @@ void quant_all_bands(int encode, const CELTMode *m, int start, int end,
       ctx.remaining_bits = remaining_bits;
       if (i <= codedBands-1)
       {
-         curr_balance = balance / IMIN(3, codedBands-i);
+         curr_balance = celt_sudiv(balance, IMIN(3, codedBands-i));
          b = IMAX(0, IMIN(16383, IMIN(remaining_bits+1,pulses[i]+curr_balance)));
       } else {
          b = 0;
diff --git a/drivers/opus/celt/bands.h b/drivers/opus/celt/bands.h
index 1ef7cbc8ee..eb7ea4cc24 100644
--- a/drivers/opus/celt/bands.h
+++ b/drivers/opus/celt/bands.h
@@ -31,7 +31,7 @@
 #define BANDS_H
 
 #include "opus/celt/arch.h"
-#include "opus/celt/opus_modes.h"
+#include "opus/celt/modes.h"
 #include "opus/celt/entenc.h"
 #include "opus/celt/entdec.h"
 #include "opus/celt/rate.h"
@@ -41,7 +41,7 @@
  * @param X Spectrum
  * @param bandE Square root of the energy for each band (returned)
  */
-void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int M);
+void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int LM);
 
 /*void compute_noise_energies(const CELTMode *m, const celt_sig *X, const opus_val16 *tonality, celt_ener *bandE);*/
 
@@ -59,14 +59,15 @@ void normalise_bands(const CELTMode *m, const celt_sig * OPUS_RESTRICT freq, cel
  * @param bandE Square root of the energy for each band
  */
 void denormalise_bands(const CELTMode *m, const celt_norm * OPUS_RESTRICT X,
-      celt_sig * OPUS_RESTRICT freq, const opus_val16 *bandE, int start, int end, int C, int M);
+      celt_sig * OPUS_RESTRICT freq, const opus_val16 *bandE, int start,
+      int end, int M, int downsample, int silence);
 
 #define SPREAD_NONE       (0)
 #define SPREAD_LIGHT      (1)
 #define SPREAD_NORMAL     (2)
 #define SPREAD_AGGRESSIVE (3)
 
-int spreading_decision(const CELTMode *m, celt_norm *X, int *average,
+int spreading_decision(const CELTMode *m, const celt_norm *X, int *average,
       int last_decision, int *hf_average, int *tapset_decision, int update_hf,
       int end, int C, int M);
 
@@ -97,15 +98,20 @@ void haar1(celt_norm *X, int N0, int stride);
  * @param LM log2() of the number of 2.5 subframes in the frame
  * @param codedBands Last band to receive bits + 1
  * @param seed Random generator seed
+ * @param arch Run-time architecture (see opus_select_arch())
  */
 void quant_all_bands(int encode, const CELTMode *m, int start, int end,
-      celt_norm * X, celt_norm * Y, unsigned char *collapse_masks, const celt_ener *bandE, int *pulses,
-      int shortBlocks, int spread, int dual_stereo, int intensity, int *tf_res,
-      opus_int32 total_bits, opus_int32 balance, ec_ctx *ec, int M, int codedBands, opus_uint32 *seed);
-
-void anti_collapse(const CELTMode *m, celt_norm *X_, unsigned char *collapse_masks, int LM, int C, int size,
-      int start, int end, opus_val16 *logE, opus_val16 *prev1logE,
-      opus_val16 *prev2logE, int *pulses, opus_uint32 seed);
+      celt_norm * X, celt_norm * Y, unsigned char *collapse_masks,
+      const celt_ener *bandE, int *pulses, int shortBlocks, int spread,
+      int dual_stereo, int intensity, int *tf_res, opus_int32 total_bits,
+      opus_int32 balance, ec_ctx *ec, int M, int codedBands, opus_uint32 *seed,
+      int arch);
+
+void anti_collapse(const CELTMode *m, celt_norm *X_,
+      unsigned char *collapse_masks, int LM, int C, int size, int start,
+      int end, const opus_val16 *logE, const opus_val16 *prev1logE,
+      const opus_val16 *prev2logE, const int *pulses, opus_uint32 seed,
+      int arch);
 
 opus_uint32 celt_lcg_rand(opus_uint32 seed);
 
diff --git a/drivers/opus/celt/celt.c b/drivers/opus/celt/celt.c
index d99a91ba29..4a16651441 100644
--- a/drivers/opus/celt/celt.c
+++ b/drivers/opus/celt/celt.c
@@ -26,10 +26,7 @@
    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
-
-#ifdef OPUS_ENABLED
 #include "opus/opus_config.h"
-#endif
 
 #define CELT_C
 
@@ -39,7 +36,7 @@
 #include "opus/celt/celt.h"
 #include "opus/celt/pitch.h"
 #include "opus/celt/bands.h"
-#include "opus/celt/opus_modes.h"
+#include "opus/celt/modes.h"
 #include "opus/celt/entcode.h"
 #include "opus/celt/quant_bands.h"
 #include "opus/celt/rate.h"
@@ -54,6 +51,10 @@
 #define PACKAGE_VERSION "unknown"
 #endif
 
+#if defined(MIPSr1_ASM)
+#include "opus/celt/mips/celt_mipsr1.h"
+#endif
+
 
 int resampling_factor(opus_int32 rate)
 {
@@ -85,8 +86,71 @@ int resampling_factor(opus_int32 rate)
    return ret;
 }
 
-#ifndef OVERRIDE_COMB_FILTER_CONST
-static void comb_filter_const(opus_val32 *y, opus_val32 *x, int T, int N,
+#if !defined(OVERRIDE_COMB_FILTER_CONST) || defined(NON_STATIC_COMB_FILTER_CONST_C)
+/* This version should be faster on ARM */
+#ifdef OPUS_ARM_ASM
+#ifndef NON_STATIC_COMB_FILTER_CONST_C
+static
+#endif
+void comb_filter_const_c(opus_val32 *y, opus_val32 *x, int T, int N,
+      opus_val16 g10, opus_val16 g11, opus_val16 g12)
+{
+   opus_val32 x0, x1, x2, x3, x4;
+   int i;
+   x4 = SHL32(x[-T-2], 1);
+   x3 = SHL32(x[-T-1], 1);
+   x2 = SHL32(x[-T], 1);
+   x1 = SHL32(x[-T+1], 1);
+   for (i=0;i<N-4;i+=5)
+   {
+      opus_val32 t;
+      x0=SHL32(x[i-T+2],1);
+      t = MAC16_32_Q16(x[i], g10, x2);
+      t = MAC16_32_Q16(t, g11, ADD32(x1,x3));
+      t = MAC16_32_Q16(t, g12, ADD32(x0,x4));
+      y[i] = t;
+      x4=SHL32(x[i-T+3],1);
+      t = MAC16_32_Q16(x[i+1], g10, x1);
+      t = MAC16_32_Q16(t, g11, ADD32(x0,x2));
+      t = MAC16_32_Q16(t, g12, ADD32(x4,x3));
+      y[i+1] = t;
+      x3=SHL32(x[i-T+4],1);
+      t = MAC16_32_Q16(x[i+2], g10, x0);
+      t = MAC16_32_Q16(t, g11, ADD32(x4,x1));
+      t = MAC16_32_Q16(t, g12, ADD32(x3,x2));
+      y[i+2] = t;
+      x2=SHL32(x[i-T+5],1);
+      t = MAC16_32_Q16(x[i+3], g10, x4);
+      t = MAC16_32_Q16(t, g11, ADD32(x3,x0));
+      t = MAC16_32_Q16(t, g12, ADD32(x2,x1));
+      y[i+3] = t;
+      x1=SHL32(x[i-T+6],1);
+      t = MAC16_32_Q16(x[i+4], g10, x3);
+      t = MAC16_32_Q16(t, g11, ADD32(x2,x4));
+      t = MAC16_32_Q16(t, g12, ADD32(x1,x0));
+      y[i+4] = t;
+   }
+#ifdef CUSTOM_MODES
+   for (;i<N;i++)
+   {
+      opus_val32 t;
+      x0=SHL32(x[i-T+2],1);
+      t = MAC16_32_Q16(x[i], g10, x2);
+      t = MAC16_32_Q16(t, g11, ADD32(x1,x3));
+      t = MAC16_32_Q16(t, g12, ADD32(x0,x4));
+      y[i] = t;
+      x4=x3;
+      x3=x2;
+      x2=x1;
+      x1=x0;
+   }
+#endif
+}
+#else
+#ifndef NON_STATIC_COMB_FILTER_CONST_C
+static
+#endif
+void comb_filter_const_c(opus_val32 *y, opus_val32 *x, int T, int N,
       opus_val16 g10, opus_val16 g11, opus_val16 g12)
 {
    opus_val32 x0, x1, x2, x3, x4;
@@ -110,10 +174,12 @@ static void comb_filter_const(opus_val32 *y, opus_val32 *x, int T, int N,
 
 }
 #endif
+#endif
 
+#ifndef OVERRIDE_comb_filter
 void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N,
       opus_val16 g0, opus_val16 g1, int tapset0, int tapset1,
-      const opus_val16 *window, int overlap)
+      const opus_val16 *window, int overlap, int arch)
 {
    int i;
    /* printf ("%d %d %f %f\n", T0, T1, g0, g1); */
@@ -131,16 +197,19 @@ void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N,
          OPUS_MOVE(y, x, N);
       return;
    }
-   g00 = MULT16_16_Q15(g0, gains[tapset0][0]);
-   g01 = MULT16_16_Q15(g0, gains[tapset0][1]);
-   g02 = MULT16_16_Q15(g0, gains[tapset0][2]);
-   g10 = MULT16_16_Q15(g1, gains[tapset1][0]);
-   g11 = MULT16_16_Q15(g1, gains[tapset1][1]);
-   g12 = MULT16_16_Q15(g1, gains[tapset1][2]);
+   g00 = MULT16_16_P15(g0, gains[tapset0][0]);
+   g01 = MULT16_16_P15(g0, gains[tapset0][1]);
+   g02 = MULT16_16_P15(g0, gains[tapset0][2]);
+   g10 = MULT16_16_P15(g1, gains[tapset1][0]);
+   g11 = MULT16_16_P15(g1, gains[tapset1][1]);
+   g12 = MULT16_16_P15(g1, gains[tapset1][2]);
    x1 = x[-T1+1];
    x2 = x[-T1  ];
    x3 = x[-T1-1];
    x4 = x[-T1-2];
+   /* If the filter didn't change, we don't need the overlap */
+   if (g0==g1 && T0==T1 && tapset0==tapset1)
+      overlap=0;
    for (i=0;i<overlap;i++)
    {
       opus_val16 f;
@@ -168,8 +237,9 @@ void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N,
    }
 
    /* Compute the part with the constant filter. */
-   comb_filter_const(y+i, x+i, T1, N-i, g10, g11, g12);
+   comb_filter_const(y+i, x+i, T1, N-i, g10, g11, g12, arch);
 }
+#endif /* OVERRIDE_comb_filter */
 
 const signed char tf_select_table[4][8] = {
       {0, -1, 0, -1,    0,-1, 0,-1},
@@ -213,6 +283,9 @@ const char *opus_strerror(int error)
 const char *opus_get_version_string(void)
 {
     return "libopus " PACKAGE_VERSION
+    /* Applications may rely on the presence of this substring in the version
+       string to determine if they have a fixed-point or floating-point build
+       at runtime. */
 #ifdef OPUS_FIXED_POINT
           "-fixed"
 #endif
diff --git a/drivers/opus/celt/celt.h b/drivers/opus/celt/celt.h
index 2b7dc3fa72..72814dc0de 100644
--- a/drivers/opus/celt/celt.h
+++ b/drivers/opus/celt/celt.h
@@ -134,7 +134,8 @@ int celt_decoder_get_size(int channels);
 
 int celt_decoder_init(CELTDecoder *st, opus_int32 sampling_rate, int channels);
 
-int celt_decode_with_ec(OpusCustomDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, opus_val16 * OPUS_RESTRICT pcm, int frame_size, ec_dec *dec);
+int celt_decode_with_ec(OpusCustomDecoder * OPUS_RESTRICT st, const unsigned char *data,
+      int len, opus_val16 * OPUS_RESTRICT pcm, int frame_size, ec_dec *dec, int accum);
 
 #define celt_encoder_ctl opus_custom_encoder_ctl
 #define celt_decoder_ctl opus_custom_decoder_ctl
@@ -200,15 +201,25 @@ void celt_preemphasis(const opus_val16 * OPUS_RESTRICT pcmp, celt_sig * OPUS_RES
 
 void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N,
       opus_val16 g0, opus_val16 g1, int tapset0, int tapset1,
-      const opus_val16 *window, int overlap);
+      const opus_val16 *window, int overlap, int arch);
+
+#ifdef NON_STATIC_COMB_FILTER_CONST_C
+void comb_filter_const_c(opus_val32 *y, opus_val32 *x, int T, int N,
+                         opus_val16 g10, opus_val16 g11, opus_val16 g12);
+#endif
+
+#ifndef OVERRIDE_COMB_FILTER_CONST
+# define comb_filter_const(y, x, T, N, g10, g11, g12, arch)		\
+    ((void)(arch),comb_filter_const_c(y, x, T, N, g10, g11, g12))
+#endif
 
 void init_caps(const CELTMode *m,int *cap,int LM,int C);
 
 #ifdef RESYNTH
-void deemphasis(celt_sig *in[], opus_val16 *pcm, int N, int C, int downsample, const opus_val16 *coef, celt_sig *mem, celt_sig * OPUS_RESTRICT scratch);
-
-void compute_inv_mdcts(const CELTMode *mode, int shortBlocks, celt_sig *X,
-      celt_sig * OPUS_RESTRICT out_mem[], int C, int LM);
+void deemphasis(celt_sig *in[], opus_val16 *pcm, int N, int C, int downsample, const opus_val16 *coef, celt_sig *mem);
+void celt_synthesis(const CELTMode *mode, celt_norm *X, celt_sig * out_syn[],
+      opus_val16 *oldBandE, int start, int effEnd, int C, int CC, int isTransient,
+      int LM, int downsample, int silence);
 #endif
 
 #ifdef __cplusplus
diff --git a/drivers/opus/celt/celt_decoder.c b/drivers/opus/celt/celt_decoder.c
index 67c3789439..88072ed110 100644
--- a/drivers/opus/celt/celt_decoder.c
+++ b/drivers/opus/celt/celt_decoder.c
@@ -26,10 +26,7 @@
    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
-
-#ifdef OPUS_ENABLED
 #include "opus/opus_config.h"
-#endif
 
 #define CELT_DECODER_C
 
@@ -40,7 +37,7 @@
 #include "opus/celt/celt.h"
 #include "opus/celt/pitch.h"
 #include "opus/celt/bands.h"
-#include "opus/celt/opus_modes.h"
+#include "opus/celt/modes.h"
 #include "opus/celt/entcode.h"
 #include "opus/celt/quant_bands.h"
 #include "opus/celt/rate.h"
@@ -51,6 +48,9 @@
 #include "opus/celt/celt_lpc.h"
 #include "opus/celt/vq.h"
 
+#if defined(SMALL_FOOTPRINT) && defined(FIXED_POINT)
+#define NORM_ALIASING_HACK
+#endif
 /**********************************************************************/
 /*                                                                    */
 /*                             DECODER                                */
@@ -175,28 +175,24 @@ void opus_custom_decoder_destroy(CELTDecoder *st)
 }
 #endif /* CUSTOM_MODES */
 
-static OPUS_INLINE opus_val16 SIG2WORD16(celt_sig x)
-{
-#ifdef OPUS_FIXED_POINT
-   x = PSHR32(x, SIG_SHIFT);
-   x = MAX32(x, -32768);
-   x = MIN32(x, 32767);
-   return EXTRACT16(x);
-#else
-   return (opus_val16)x;
-#endif
-}
 
 #ifndef RESYNTH
 static
 #endif
-void deemphasis(celt_sig *in[], opus_val16 *pcm, int N, int C, int downsample, const opus_val16 *coef, celt_sig *mem, celt_sig * OPUS_RESTRICT scratch)
+void deemphasis(celt_sig *in[], opus_val16 *pcm, int N, int C, int downsample, const opus_val16 *coef,
+      celt_sig *mem, int accum)
 {
    int c;
    int Nd;
    int apply_downsampling=0;
    opus_val16 coef0;
-
+   VARDECL(celt_sig, scratch);
+   SAVE_STACK;
+#ifndef OPUS_FIXED_POINT
+   (void)accum;
+   celt_assert(accum==0);
+#endif
+   ALLOC(scratch, N, celt_sig);
    coef0 = coef[0];
    Nd = N/downsample;
    c=0; do {
@@ -234,11 +230,24 @@ void deemphasis(celt_sig *in[], opus_val16 *pcm, int N, int C, int downsample, c
          apply_downsampling=1;
       } else {
          /* Shortcut for the standard (non-custom modes) case */
-         for (j=0;j<N;j++)
+#ifdef OPUS_FIXED_POINT
+         if (accum)
          {
-            celt_sig tmp = x[j] + m + VERY_SMALL;
-            m = MULT16_32_Q15(coef0, tmp);
-            y[j*C] = SCALEOUT(SIG2WORD16(tmp));
+            for (j=0;j<N;j++)
+            {
+               celt_sig tmp = x[j] + m + VERY_SMALL;
+               m = MULT16_32_Q15(coef0, tmp);
+               y[j*C] = SAT16(ADD32(y[j*C], SCALEOUT(SIG2WORD16(tmp))));
+            }
+         } else
+#endif
+         {
+            for (j=0;j<N;j++)
+            {
+               celt_sig tmp = x[j] + m + VERY_SMALL;
+               m = MULT16_32_Q15(coef0, tmp);
+               y[j*C] = SCALEOUT(SIG2WORD16(tmp));
+            }
          }
       }
       mem[c] = m;
@@ -246,41 +255,95 @@ void deemphasis(celt_sig *in[], opus_val16 *pcm, int N, int C, int downsample, c
       if (apply_downsampling)
       {
          /* Perform down-sampling */
-         for (j=0;j<Nd;j++)
-            y[j*C] = SCALEOUT(SIG2WORD16(scratch[j*downsample]));
+#ifdef OPUS_FIXED_POINT
+         if (accum)
+         {
+            for (j=0;j<Nd;j++)
+               y[j*C] = SAT16(ADD32(y[j*C], SCALEOUT(SIG2WORD16(scratch[j*downsample]))));
+         } else
+#endif
+         {
+            for (j=0;j<Nd;j++)
+               y[j*C] = SCALEOUT(SIG2WORD16(scratch[j*downsample]));
+         }
       }
    } while (++c<C);
+   RESTORE_STACK;
 }
 
-/** Compute the IMDCT and apply window for all sub-frames and
-    all channels in a frame */
 #ifndef RESYNTH
 static
 #endif
-void compute_inv_mdcts(const CELTMode *mode, int shortBlocks, celt_sig *X,
-      celt_sig * OPUS_RESTRICT out_mem[], int C, int LM)
+void celt_synthesis(const CELTMode *mode, celt_norm *X, celt_sig * out_syn[],
+                    opus_val16 *oldBandE, int start, int effEnd, int C, int CC,
+                    int isTransient, int LM, int downsample,
+                    int silence, int arch)
 {
-   int b, c;
+   int c, i;
+   int M;
+   int b;
    int B;
-   int N;
+   int N, NB;
    int shift;
-   const int overlap = OVERLAP(mode);
+   int nbEBands;
+   int overlap;
+   VARDECL(celt_sig, freq);
+   SAVE_STACK;
+
+   overlap = mode->overlap;
+   nbEBands = mode->nbEBands;
+   N = mode->shortMdctSize<<LM;
+   ALLOC(freq, N, celt_sig); /**< Interleaved signal MDCTs */
+   M = 1<<LM;
 
-   if (shortBlocks)
+   if (isTransient)
    {
-      B = shortBlocks;
-      N = mode->shortMdctSize;
+      B = M;
+      NB = mode->shortMdctSize;
       shift = mode->maxLM;
    } else {
       B = 1;
-      N = mode->shortMdctSize<<LM;
+      NB = mode->shortMdctSize<<LM;
       shift = mode->maxLM-LM;
    }
-   c=0; do {
-      /* IMDCT on the interleaved the sub-frames, overlap-add is performed by the IMDCT */
+
+   if (CC==2&&C==1)
+   {
+      /* Copying a mono streams to two channels */
+      celt_sig *freq2;
+      denormalise_bands(mode, X, freq, oldBandE, start, effEnd, M,
+            downsample, silence);
+      /* Store a temporary copy in the output buffer because the IMDCT destroys its input. */
+      freq2 = out_syn[1]+overlap/2;
+      OPUS_COPY(freq2, freq, N);
       for (b=0;b<B;b++)
-         clt_mdct_backward(&mode->mdct, &X[b+c*N*B], out_mem[c]+N*b, mode->window, overlap, shift, B);
-   } while (++c<C);
+         clt_mdct_backward(&mode->mdct, &freq2[b], out_syn[0]+NB*b, mode->window, overlap, shift, B, arch);
+      for (b=0;b<B;b++)
+         clt_mdct_backward(&mode->mdct, &freq[b], out_syn[1]+NB*b, mode->window, overlap, shift, B, arch);
+   } else if (CC==1&&C==2)
+   {
+      /* Downmixing a stereo stream to mono */
+      celt_sig *freq2;
+      freq2 = out_syn[0]+overlap/2;
+      denormalise_bands(mode, X, freq, oldBandE, start, effEnd, M,
+            downsample, silence);
+      /* Use the output buffer as temp array before downmixing. */
+      denormalise_bands(mode, X+N, freq2, oldBandE+nbEBands, start, effEnd, M,
+            downsample, silence);
+      for (i=0;i<N;i++)
+         freq[i] = HALF32(ADD32(freq[i],freq2[i]));
+      for (b=0;b<B;b++)
+         clt_mdct_backward(&mode->mdct, &freq[b], out_syn[0]+NB*b, mode->window, overlap, shift, B, arch);
+   } else {
+      /* Normal case (mono or stereo) */
+      c=0; do {
+         denormalise_bands(mode, X+c*N, freq, oldBandE+c*nbEBands, start, effEnd, M,
+               downsample, silence);
+         for (b=0;b<B;b++)
+            clt_mdct_backward(&mode->mdct, &freq[b], out_syn[c]+NB*b, mode->window, overlap, shift, B, arch);
+      } while (++c<CC);
+   }
+   RESTORE_STACK;
 }
 
 static void tf_decode(int start, int end, int isTransient, int *tf_res, int LM, ec_dec *dec)
@@ -330,7 +393,23 @@ static void tf_decode(int start, int end, int isTransient, int *tf_res, int LM,
    pitch of 480 Hz. */
 #define PLC_PITCH_LAG_MIN (100)
 
-static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, opus_val16 * OPUS_RESTRICT pcm, int N, int LM)
+static int celt_plc_pitch_search(celt_sig *decode_mem[2], int C, int arch)
+{
+   int pitch_index;
+   VARDECL( opus_val16, lp_pitch_buf );
+   SAVE_STACK;
+   ALLOC( lp_pitch_buf, DECODE_BUFFER_SIZE>>1, opus_val16 );
+   pitch_downsample(decode_mem, lp_pitch_buf,
+         DECODE_BUFFER_SIZE, C, arch);
+   pitch_search(lp_pitch_buf+(PLC_PITCH_LAG_MAX>>1), lp_pitch_buf,
+         DECODE_BUFFER_SIZE-PLC_PITCH_LAG_MAX,
+         PLC_PITCH_LAG_MAX-PLC_PITCH_LAG_MIN, &pitch_index, arch);
+   pitch_index = PLC_PITCH_LAG_MAX-pitch_index;
+   RESTORE_STACK;
+   return pitch_index;
+}
+
+static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM)
 {
    int c;
    int i;
@@ -343,11 +422,9 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, opus_val16 * OPUS_R
    int nbEBands;
    int overlap;
    int start;
-   int downsample;
    int loss_count;
    int noise_based;
    const opus_int16 *eBands;
-   VARDECL(celt_sig, scratch);
    SAVE_STACK;
 
    mode = st->mode;
@@ -367,40 +444,37 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, opus_val16 * OPUS_R
 
    loss_count = st->loss_count;
    start = st->start;
-   downsample = st->downsample;
    noise_based = loss_count >= 5 || start != 0;
-   ALLOC(scratch, noise_based?N*C:N, celt_sig);
    if (noise_based)
    {
       /* Noise-based PLC/CNG */
-      celt_sig *freq;
+#ifdef NORM_ALIASING_HACK
+      celt_norm *X;
+#else
       VARDECL(celt_norm, X);
+#endif
       opus_uint32 seed;
-      opus_val16 *plcLogE;
       int end;
       int effEnd;
-
+      opus_val16 decay;
       end = st->end;
       effEnd = IMAX(start, IMIN(end, mode->effEBands));
 
-      /* Share the interleaved signal MDCT coefficient buffer with the
-         deemphasis scratch buffer. */
-      freq = scratch;
+#ifdef NORM_ALIASING_HACK
+      /* This is an ugly hack that breaks aliasing rules and would be easily broken,
+         but it saves almost 4kB of stack. */
+      X = (celt_norm*)(out_syn[C-1]+overlap/2);
+#else
       ALLOC(X, C*N, celt_norm);   /**< Interleaved normalised MDCTs */
+#endif
 
-      if (loss_count >= 5)
-         plcLogE = backgroundLogE;
-      else {
-         /* Energy decay */
-         opus_val16 decay = loss_count==0 ?
-               QCONST16(1.5f, DB_SHIFT) : QCONST16(.5f, DB_SHIFT);
-         c=0; do
-         {
-            for (i=start;i<end;i++)
-               oldBandE[c*nbEBands+i] -= decay;
-         } while (++c<C);
-         plcLogE = oldBandE;
-      }
+      /* Energy decay */
+      decay = loss_count==0 ? QCONST16(1.5f, DB_SHIFT) : QCONST16(.5f, DB_SHIFT);
+      c=0; do
+      {
+         for (i=start;i<end;i++)
+            oldBandE[c*nbEBands+i] = MAX16(backgroundLogE[c*nbEBands+i], oldBandE[c*nbEBands+i] - decay);
+      } while (++c<C);
       seed = st->rng;
       for (c=0;c<C;c++)
       {
@@ -416,25 +490,17 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, opus_val16 * OPUS_R
                seed = celt_lcg_rand(seed);
                X[boffs+j] = (celt_norm)((opus_int32)seed>>20);
             }
-            renormalise_vector(X+boffs, blen, Q15ONE);
+            renormalise_vector(X+boffs, blen, Q15ONE, st->arch);
          }
       }
       st->rng = seed;
 
-      denormalise_bands(mode, X, freq, plcLogE, start, effEnd, C, 1<<LM);
-
-      c=0; do {
-         int bound = eBands[effEnd]<<LM;
-         if (downsample!=1)
-            bound = IMIN(bound, N/downsample);
-         for (i=bound;i<N;i++)
-            freq[c*N+i] = 0;
-      } while (++c<C);
       c=0; do {
          OPUS_MOVE(decode_mem[c], decode_mem[c]+N,
                DECODE_BUFFER_SIZE-N+(overlap>>1));
       } while (++c<C);
-      compute_inv_mdcts(mode, 0, freq, out_syn, C, LM);
+
+      celt_synthesis(mode, X, out_syn, oldBandE, start, effEnd, C, C, 0, LM, st->downsample, 0, st->arch);
    } else {
       /* Pitch-based PLC */
       const opus_val16 *window;
@@ -445,15 +511,7 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, opus_val16 * OPUS_R
 
       if (loss_count == 0)
       {
-         VARDECL( opus_val16, lp_pitch_buf );
-         ALLOC( lp_pitch_buf, DECODE_BUFFER_SIZE>>1, opus_val16 );
-         pitch_downsample(decode_mem, lp_pitch_buf,
-               DECODE_BUFFER_SIZE, C, st->arch);
-         pitch_search(lp_pitch_buf+(PLC_PITCH_LAG_MAX>>1), lp_pitch_buf,
-               DECODE_BUFFER_SIZE-PLC_PITCH_LAG_MAX,
-               PLC_PITCH_LAG_MAX-PLC_PITCH_LAG_MIN, &pitch_index, st->arch);
-         pitch_index = PLC_PITCH_LAG_MAX-pitch_index;
-         st->last_pitch_index = pitch_index;
+         st->last_pitch_index = pitch_index = celt_plc_pitch_search(decode_mem, C, st->arch);
       } else {
          pitch_index = st->last_pitch_index;
          fade = QCONST16(.8f,15);
@@ -516,7 +574,7 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, opus_val16 * OPUS_R
             }
             /* Compute the excitation for exc_length samples before the loss. */
             celt_fir(exc+MAX_PERIOD-exc_length, lpc+c*LPC_ORDER,
-                  exc+MAX_PERIOD-exc_length, exc_length, LPC_ORDER, lpc_mem);
+                  exc+MAX_PERIOD-exc_length, exc_length, LPC_ORDER, lpc_mem, st->arch);
          }
 
          /* Check if the waveform is decaying, and if so how fast.
@@ -583,7 +641,7 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, opus_val16 * OPUS_R
                the signal domain. */
             celt_iir(buf+DECODE_BUFFER_SIZE-N, lpc+c*LPC_ORDER,
                   buf+DECODE_BUFFER_SIZE-N, extrapolation_len, LPC_ORDER,
-                  lpc_mem);
+                  lpc_mem, st->arch);
          }
 
          /* Check if the synthesis energy is higher than expected, which can
@@ -631,7 +689,7 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, opus_val16 * OPUS_R
          comb_filter(etmp, buf+DECODE_BUFFER_SIZE,
               st->postfilter_period, st->postfilter_period, overlap,
               -st->postfilter_gain, -st->postfilter_gain,
-              st->postfilter_tapset, st->postfilter_tapset, NULL, 0);
+              st->postfilter_tapset, st->postfilter_tapset, NULL, 0, st->arch);
 
          /* Simulate TDAC on the concealed audio so that it blends with the
             MDCT of the next frame. */
@@ -644,22 +702,23 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, opus_val16 * OPUS_R
       } while (++c<C);
    }
 
-   deemphasis(out_syn, pcm, N, C, downsample,
-         mode->preemph, st->preemph_memD, scratch);
-
    st->loss_count = loss_count+1;
 
    RESTORE_STACK;
 }
 
-int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, opus_val16 * OPUS_RESTRICT pcm, int frame_size, ec_dec *dec)
+int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data,
+      int len, opus_val16 * OPUS_RESTRICT pcm, int frame_size, ec_dec *dec, int accum)
 {
    int c, i, N;
    int spread_decision;
    opus_int32 bits;
    ec_dec _dec;
-   VARDECL(celt_sig, freq);
+#ifdef NORM_ALIASING_HACK
+   celt_norm *X;
+#else
    VARDECL(celt_norm, X);
+#endif
    VARDECL(int, fine_quant);
    VARDECL(int, pulses);
    VARDECL(int, cap);
@@ -677,6 +736,8 @@ int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *dat
    int intra_ener;
    const int CC = st->channels;
    int LM, M;
+   int start;
+   int end;
    int effEnd;
    int codedBands;
    int alloc_trim;
@@ -703,11 +764,10 @@ int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *dat
    nbEBands = mode->nbEBands;
    overlap = mode->overlap;
    eBands = mode->eBands;
+   start = st->start;
+   end = st->end;
    frame_size *= st->downsample;
 
-   c=0; do {
-      decode_mem[c] = st->_decode_mem + c*(DECODE_BUFFER_SIZE+overlap);
-   } while (++c<CC);
    lpc = (opus_val16*)(st->_decode_mem+(DECODE_BUFFER_SIZE+overlap)*CC);
    oldBandE = lpc+CC*LPC_ORDER;
    oldLogE = oldBandE + 2*nbEBands;
@@ -725,7 +785,7 @@ int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *dat
          if (data0<0)
             return OPUS_INVALID_PACKET;
       }
-      st->end = IMAX(1, mode->effEBands-2*(data0>>5));
+      st->end = end = IMAX(1, mode->effEBands-2*(data0>>5));
       LM = (data0>>3)&0x3;
       C = 1 + ((data0>>2)&0x1);
       data++;
@@ -752,14 +812,19 @@ int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *dat
       return OPUS_BAD_ARG;
 
    N = M*mode->shortMdctSize;
+   c=0; do {
+      decode_mem[c] = st->_decode_mem + c*(DECODE_BUFFER_SIZE+overlap);
+      out_syn[c] = decode_mem[c]+DECODE_BUFFER_SIZE-N;
+   } while (++c<CC);
 
-   effEnd = st->end;
+   effEnd = end;
    if (effEnd > mode->effEBands)
       effEnd = mode->effEBands;
 
    if (data == NULL || len<=1)
    {
-      celt_decode_lost(st, pcm, N, LM);
+      celt_decode_lost(st, N, LM);
+      deemphasis(out_syn, pcm, N, CC, st->downsample, mode->preemph, st->preemph_memD, accum);
       RESTORE_STACK;
       return frame_size/st->downsample;
    }
@@ -795,7 +860,7 @@ int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *dat
    postfilter_gain = 0;
    postfilter_pitch = 0;
    postfilter_tapset = 0;
-   if (st->start==0 && tell+16 <= total_bits)
+   if (start==0 && tell+16 <= total_bits)
    {
       if(ec_dec_bit_logp(dec, 1))
       {
@@ -826,11 +891,11 @@ int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *dat
    /* Decode the global flags (first symbols in the stream) */
    intra_ener = tell+3<=total_bits ? ec_dec_bit_logp(dec, 3) : 0;
    /* Get band energies */
-   unquant_coarse_energy(mode, st->start, st->end, oldBandE,
+   unquant_coarse_energy(mode, start, end, oldBandE,
          intra_ener, dec, C, LM);
 
    ALLOC(tf_res, nbEBands, int);
-   tf_decode(st->start, st->end, isTransient, tf_res, LM, dec);
+   tf_decode(start, end, isTransient, tf_res, LM, dec);
 
    tell = ec_tell(dec);
    spread_decision = SPREAD_NORMAL;
@@ -846,7 +911,7 @@ int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *dat
    dynalloc_logp = 6;
    total_bits<<=BITRES;
    tell = ec_tell_frac(dec);
-   for (i=st->start;i<st->end;i++)
+   for (i=start;i<end;i++)
    {
       int width, quanta;
       int dynalloc_loop_logp;
@@ -885,84 +950,62 @@ int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *dat
    ALLOC(pulses, nbEBands, int);
    ALLOC(fine_priority, nbEBands, int);
 
-   codedBands = compute_allocation(mode, st->start, st->end, offsets, cap,
+   codedBands = compute_allocation(mode, start, end, offsets, cap,
          alloc_trim, &intensity, &dual_stereo, bits, &balance, pulses,
          fine_quant, fine_priority, C, LM, dec, 0, 0, 0);
 
-   unquant_fine_energy(mode, st->start, st->end, oldBandE, fine_quant, dec, C);
+   unquant_fine_energy(mode, start, end, oldBandE, fine_quant, dec, C);
+
+   c=0; do {
+      OPUS_MOVE(decode_mem[c], decode_mem[c]+N, DECODE_BUFFER_SIZE-N+overlap/2);
+   } while (++c<CC);
 
    /* Decode fixed codebook */
    ALLOC(collapse_masks, C*nbEBands, unsigned char);
+
+#ifdef NORM_ALIASING_HACK
+   /* This is an ugly hack that breaks aliasing rules and would be easily broken,
+      but it saves almost 4kB of stack. */
+   X = (celt_norm*)(out_syn[CC-1]+overlap/2);
+#else
    ALLOC(X, C*N, celt_norm);   /**< Interleaved normalised MDCTs */
+#endif
 
-   quant_all_bands(0, mode, st->start, st->end, X, C==2 ? X+N : NULL, collapse_masks,
+   quant_all_bands(0, mode, start, end, X, C==2 ? X+N : NULL, collapse_masks,
          NULL, pulses, shortBlocks, spread_decision, dual_stereo, intensity, tf_res,
-         len*(8<<BITRES)-anti_collapse_rsv, balance, dec, LM, codedBands, &st->rng);
+         len*(8<<BITRES)-anti_collapse_rsv, balance, dec, LM, codedBands, &st->rng, st->arch);
 
    if (anti_collapse_rsv > 0)
    {
       anti_collapse_on = ec_dec_bits(dec, 1);
    }
 
-   unquant_energy_finalise(mode, st->start, st->end, oldBandE,
+   unquant_energy_finalise(mode, start, end, oldBandE,
          fine_quant, fine_priority, len*8-ec_tell(dec), dec, C);
 
    if (anti_collapse_on)
       anti_collapse(mode, X, collapse_masks, LM, C, N,
-            st->start, st->end, oldBandE, oldLogE, oldLogE2, pulses, st->rng);
-
-   ALLOC(freq, IMAX(CC,C)*N, celt_sig); /**< Interleaved signal MDCTs */
+            start, end, oldBandE, oldLogE, oldLogE2, pulses, st->rng, st->arch);
 
    if (silence)
    {
       for (i=0;i<C*nbEBands;i++)
          oldBandE[i] = -QCONST16(28.f,DB_SHIFT);
-      for (i=0;i<C*N;i++)
-         freq[i] = 0;
-   } else {
-      /* Synthesis */
-      denormalise_bands(mode, X, freq, oldBandE, st->start, effEnd, C, M);
    }
-   c=0; do {
-      OPUS_MOVE(decode_mem[c], decode_mem[c]+N, DECODE_BUFFER_SIZE-N+overlap/2);
-   } while (++c<CC);
 
-   c=0; do {
-      int bound = M*eBands[effEnd];
-      if (st->downsample!=1)
-         bound = IMIN(bound, N/st->downsample);
-      for (i=bound;i<N;i++)
-         freq[c*N+i] = 0;
-   } while (++c<C);
-
-   c=0; do {
-      out_syn[c] = decode_mem[c]+DECODE_BUFFER_SIZE-N;
-   } while (++c<CC);
-
-   if (CC==2&&C==1)
-   {
-      for (i=0;i<N;i++)
-         freq[N+i] = freq[i];
-   }
-   if (CC==1&&C==2)
-   {
-      for (i=0;i<N;i++)
-         freq[i] = HALF32(ADD32(freq[i],freq[N+i]));
-   }
-
-   /* Compute inverse MDCTs */
-   compute_inv_mdcts(mode, shortBlocks, freq, out_syn, CC, LM);
+   celt_synthesis(mode, X, out_syn, oldBandE, start, effEnd,
+                  C, CC, isTransient, LM, st->downsample, silence, st->arch);
 
    c=0; do {
       st->postfilter_period=IMAX(st->postfilter_period, COMBFILTER_MINPERIOD);
       st->postfilter_period_old=IMAX(st->postfilter_period_old, COMBFILTER_MINPERIOD);
       comb_filter(out_syn[c], out_syn[c], st->postfilter_period_old, st->postfilter_period, mode->shortMdctSize,
             st->postfilter_gain_old, st->postfilter_gain, st->postfilter_tapset_old, st->postfilter_tapset,
-            mode->window, overlap);
+            mode->window, overlap, st->arch);
       if (LM!=0)
          comb_filter(out_syn[c]+mode->shortMdctSize, out_syn[c]+mode->shortMdctSize, st->postfilter_period, postfilter_pitch, N-mode->shortMdctSize,
                st->postfilter_gain, postfilter_gain, st->postfilter_tapset, postfilter_tapset,
-               mode->window, overlap);
+               mode->window, overlap, st->arch);
 
    } while (++c<CC);
    st->postfilter_period_old = st->postfilter_period;
@@ -978,32 +1021,36 @@ int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *dat
       st->postfilter_tapset_old = st->postfilter_tapset;
    }
 
-   if (C==1) {
-      for (i=0;i<nbEBands;i++)
-         oldBandE[nbEBands+i]=oldBandE[i];
-   }
+   if (C==1)
+      OPUS_COPY(&oldBandE[nbEBands], oldBandE, nbEBands);
 
    /* In case start or end were to change */
    if (!isTransient)
    {
+      opus_val16 max_background_increase;
+      OPUS_COPY(oldLogE2, oldLogE, 2*nbEBands);
+      OPUS_COPY(oldLogE, oldBandE, 2*nbEBands);
+      /* In normal circumstances, we only allow the noise floor to increase by
+         up to 2.4 dB/second, but when we're in DTX, we allow up to 6 dB
+         increase for each update.*/
+      if (st->loss_count < 10)
+         max_background_increase = M*QCONST16(0.001f,DB_SHIFT);
+      else
+         max_background_increase = QCONST16(1.f,DB_SHIFT);
       for (i=0;i<2*nbEBands;i++)
-         oldLogE2[i] = oldLogE[i];
-      for (i=0;i<2*nbEBands;i++)
-         oldLogE[i] = oldBandE[i];
-      for (i=0;i<2*nbEBands;i++)
-         backgroundLogE[i] = MIN16(backgroundLogE[i] + M*QCONST16(0.001f,DB_SHIFT), oldBandE[i]);
+         backgroundLogE[i] = MIN16(backgroundLogE[i] + max_background_increase, oldBandE[i]);
    } else {
       for (i=0;i<2*nbEBands;i++)
          oldLogE[i] = MIN16(oldLogE[i], oldBandE[i]);
    }
    c=0; do
    {
-      for (i=0;i<st->start;i++)
+      for (i=0;i<start;i++)
       {
          oldBandE[c*nbEBands+i]=0;
          oldLogE[c*nbEBands+i]=oldLogE2[c*nbEBands+i]=-QCONST16(28.f,DB_SHIFT);
       }
-      for (i=st->end;i<nbEBands;i++)
+      for (i=end;i<nbEBands;i++)
       {
          oldBandE[c*nbEBands+i]=0;
          oldLogE[c*nbEBands+i]=oldLogE2[c*nbEBands+i]=-QCONST16(28.f,DB_SHIFT);
@@ -1011,8 +1058,7 @@ int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *dat
    } while (++c<2);
    st->rng = dec->rng;
 
-   /* We reuse freq[] as scratch space for the de-emphasis */
-   deemphasis(out_syn, pcm, N, CC, st->downsample, mode->preemph, st->preemph_memD, freq);
+   deemphasis(out_syn, pcm, N, CC, st->downsample, mode->preemph, st->preemph_memD, accum);
    st->loss_count = 0;
    RESTORE_STACK;
    if (ec_tell(dec) > 8*len)
@@ -1028,7 +1074,7 @@ int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *dat
 #ifdef OPUS_FIXED_POINT
 int opus_custom_decode(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, opus_int16 * OPUS_RESTRICT pcm, int frame_size)
 {
-   return celt_decode_with_ec(st, data, len, pcm, frame_size, NULL);
+   return celt_decode_with_ec(st, data, len, pcm, frame_size, NULL, 0);
 }
 
 #ifndef DISABLE_FLOAT_API
@@ -1045,7 +1091,7 @@ int opus_custom_decode_float(CELTDecoder * OPUS_RESTRICT st, const unsigned char
    N = frame_size;
 
    ALLOC(out, C*N, opus_int16);
-   ret=celt_decode_with_ec(st, data, len, out, frame_size, NULL);
+   ret=celt_decode_with_ec(st, data, len, out, frame_size, NULL, 0);
    if (ret>0)
       for (j=0;j<C*ret;j++)
          pcm[j]=out[j]*(1.f/32768.f);
@@ -1059,7 +1105,7 @@ int opus_custom_decode_float(CELTDecoder * OPUS_RESTRICT st, const unsigned char
 
 int opus_custom_decode_float(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, float * OPUS_RESTRICT pcm, int frame_size)
 {
-   return celt_decode_with_ec(st, data, len, pcm, frame_size, NULL);
+   return celt_decode_with_ec(st, data, len, pcm, frame_size, NULL, 0);
 }
 
 int opus_custom_decode(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, opus_int16 * OPUS_RESTRICT pcm, int frame_size)
@@ -1075,7 +1121,7 @@ int opus_custom_decode(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data
    N = frame_size;
    ALLOC(out, C*N, celt_sig);
 
-   ret=celt_decode_with_ec(st, data, len, out, frame_size, NULL);
+   ret=celt_decode_with_ec(st, data, len, out, frame_size, NULL, 0);
 
    if (ret>0)
       for (j=0;j<C*ret;j++)
diff --git a/drivers/opus/celt/celt_encoder.c b/drivers/opus/celt/celt_encoder.c
index 810ee5d743..9099bcd278 100644
--- a/drivers/opus/celt/celt_encoder.c
+++ b/drivers/opus/celt/celt_encoder.c
@@ -26,10 +26,7 @@
    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
-
-#ifdef OPUS_ENABLED
 #include "opus/opus_config.h"
-#endif
 
 #define CELT_ENCODER_C
 
@@ -40,7 +37,7 @@
 #include "opus/celt/celt.h"
 #include "opus/celt/pitch.h"
 #include "opus/celt/bands.h"
-#include "opus/celt/opus_modes.h"
+#include "opus/celt/modes.h"
 #include "opus/celt/entcode.h"
 #include "opus/celt/quant_bands.h"
 #include "opus/celt/rate.h"
@@ -57,7 +54,6 @@
  */
 struct OpusCustomEncoder {
    const OpusCustomMode *mode;     /**< Mode used by the encoder */
-   int overlap;
    int channels;
    int stream_channels;
 
@@ -173,7 +169,6 @@ static int opus_custom_encoder_init_arch(CELTEncoder *st, const CELTMode *mode,
    OPUS_CLEAR((char*)st, opus_custom_encoder_get_size(mode, channels));
 
    st->mode = mode;
-   st->overlap = mode->overlap;
    st->stream_channels = st->channels = channels;
 
    st->upsample = 1;
@@ -276,8 +271,7 @@ static int transient_analysis(const opus_val32 * OPUS_RESTRICT in, int len, int
       }
       /*printf("\n");*/
       /* First few samples are bad because we don't propagate the memory */
-      for (i=0;i<12;i++)
-         tmp[i] = 0;
+      OPUS_CLEAR(tmp, 12);
 
 #ifdef OPUS_FIXED_POINT
       /* Normalize tmp to max range */
@@ -346,9 +340,9 @@ static int transient_analysis(const opus_val32 * OPUS_RESTRICT in, int len, int
       {
          int id;
 #ifdef OPUS_FIXED_POINT
-         id = IMAX(0,IMIN(127,MULT16_32_Q15(tmp[i],norm))); /* Do not round to nearest */
+         id = MAX32(0,MIN32(127,MULT16_32_Q15(tmp[i]+EPSILON,norm))); /* Do not round to nearest */
 #else
-         id = IMAX(0,IMIN(127,(int)floor(64*norm*tmp[i]))); /* Do not round to nearest */
+         id = (int)MAX32(0,MIN32(127,floor(64*norm*(tmp[i]+EPSILON)))); /* Do not round to nearest */
 #endif
          unmask += inv_table[id];
       }
@@ -366,7 +360,7 @@ static int transient_analysis(const opus_val32 * OPUS_RESTRICT in, int len, int
    /* Arbitrary metric for VBR boost */
    tf_max = MAX16(0,celt_sqrt(27*mask_metric)-42);
    /* *tf_estimate = 1 + MIN16(1, sqrt(MAX16(0, tf_max-30))/20); */
-   *tf_estimate = celt_sqrt(MAX16(0, SHL32(MULT16_16(QCONST16(0.0069,14),MIN16(163,tf_max)),14)-QCONST32(0.139,28)));
+   *tf_estimate = celt_sqrt(MAX32(0, SHL32(MULT16_16(QCONST16(0.0069,14),MIN16(163,tf_max)),14)-QCONST32(0.139,28)));
    /*printf("%d %f\n", tf_max, mask_metric);*/
    RESTORE_STACK;
 #ifdef FUZZING
@@ -378,8 +372,8 @@ static int transient_analysis(const opus_val32 * OPUS_RESTRICT in, int len, int
 
 /* Looks for sudden increases of energy to decide whether we need to patch
    the transient decision */
-int patch_transient_decision(opus_val16 *newE, opus_val16 *oldE, int nbEBands,
-      int end, int C)
+static int patch_transient_decision(opus_val16 *newE, opus_val16 *oldE, int nbEBands,
+      int start, int end, int C)
 {
    int i, c;
    opus_val32 mean_diff=0;
@@ -388,28 +382,28 @@ int patch_transient_decision(opus_val16 *newE, opus_val16 *oldE, int nbEBands,
       avoid false detection caused by irrelevant bands */
    if (C==1)
    {
-      spread_old[0] = oldE[0];
-      for (i=1;i<end;i++)
+      spread_old[start] = oldE[start];
+      for (i=start+1;i<end;i++)
          spread_old[i] = MAX16(spread_old[i-1]-QCONST16(1.0f, DB_SHIFT), oldE[i]);
    } else {
-      spread_old[0] = MAX16(oldE[0],oldE[nbEBands]);
-      for (i=1;i<end;i++)
+      spread_old[start] = MAX16(oldE[start],oldE[start+nbEBands]);
+      for (i=start+1;i<end;i++)
          spread_old[i] = MAX16(spread_old[i-1]-QCONST16(1.0f, DB_SHIFT),
                                MAX16(oldE[i],oldE[i+nbEBands]));
    }
-   for (i=end-2;i>=0;i--)
+   for (i=end-2;i>=start;i--)
       spread_old[i] = MAX16(spread_old[i], spread_old[i+1]-QCONST16(1.0f, DB_SHIFT));
    /* Compute mean increase */
    c=0; do {
-      for (i=2;i<end-1;i++)
+      for (i=IMAX(2,start);i<end-1;i++)
       {
          opus_val16 x1, x2;
-         x1 = MAX16(0, newE[i]);
+         x1 = MAX16(0, newE[i + c*nbEBands]);
          x2 = MAX16(0, spread_old[i]);
          mean_diff = ADD32(mean_diff, EXTEND32(MAX16(0, SUB16(x1, x2))));
       }
    } while (++c<C);
-   mean_diff = DIV32(mean_diff, C*(end-3));
+   mean_diff = DIV32(mean_diff, C*(end-1-IMAX(2,start)));
    /*printf("%f %f %d\n", mean_diff, max_diff, count);*/
    return mean_diff > QCONST16(1.f, DB_SHIFT);
 }
@@ -417,9 +411,10 @@ int patch_transient_decision(opus_val16 *newE, opus_val16 *oldE, int nbEBands,
 /** Apply window and compute the MDCT for all sub-frames and
     all channels in a frame */
 static void compute_mdcts(const CELTMode *mode, int shortBlocks, celt_sig * OPUS_RESTRICT in,
-                          celt_sig * OPUS_RESTRICT out, int C, int CC, int LM, int upsample)
+                          celt_sig * OPUS_RESTRICT out, int C, int CC, int LM, int upsample,
+                          int arch)
 {
-   const int overlap = OVERLAP(mode);
+   const int overlap = mode->overlap;
    int N;
    int B;
    int shift;
@@ -438,7 +433,9 @@ static void compute_mdcts(const CELTMode *mode, int shortBlocks, celt_sig * OPUS
       for (b=0;b<B;b++)
       {
          /* Interleaving the sub-frames while doing the MDCTs */
-         clt_mdct_forward(&mode->mdct, in+c*(B*N+overlap)+b*N, &out[b+c*N*B], mode->window, overlap, shift, B);
+         clt_mdct_forward(&mode->mdct, in+c*(B*N+overlap)+b*N,
+                          &out[b+c*N*B], mode->window, overlap, shift, B,
+                          arch);
       }
    } while (++c<CC);
    if (CC==2&&C==1)
@@ -453,8 +450,7 @@ static void compute_mdcts(const CELTMode *mode, int shortBlocks, celt_sig * OPUS
          int bound = B*N/upsample;
          for (i=0;i<bound;i++)
             out[c*B*N+i] *= upsample;
-         for (;i<B*N;i++)
-            out[c*B*N+i] = 0;
+         OPUS_CLEAR(&out[c*B*N+bound], B*N-bound);
       } while (++c<C);
    }
 }
@@ -469,26 +465,30 @@ void celt_preemphasis(const opus_val16 * OPUS_RESTRICT pcmp, celt_sig * OPUS_RES
    int Nu;
 
    coef0 = coef[0];
+   m = *mem;
 
+   /* Fast path for the normal 48kHz case and no clipping */
+   if (coef[1] == 0 && upsample == 1 && !clip)
+   {
+      for (i=0;i<N;i++)
+      {
+         opus_val16 x;
+         x = SCALEIN(pcmp[CC*i]);
+         /* Apply pre-emphasis */
+         inp[i] = SHL32(x, SIG_SHIFT) - m;
+         m = SHR32(MULT16_16(coef0, x), 15-SIG_SHIFT);
+      }
+      *mem = m;
+      return;
+   }
 
    Nu = N/upsample;
    if (upsample!=1)
    {
-      for (i=0;i<N;i++)
-         inp[i] = 0;
+      OPUS_CLEAR(inp, N);
    }
    for (i=0;i<Nu;i++)
-   {
-      celt_sig x;
-
-      x = SCALEIN(pcmp[CC*i]);
-#ifndef OPUS_FIXED_POINT
-      /* Replace NaNs with zeros */
-      if (!(x==x))
-         x = 0;
-#endif
-      inp[i*upsample] = x;
-   }
+      inp[i*upsample] = SCALEIN(pcmp[CC*i]);
 
 #ifndef OPUS_FIXED_POINT
    if (clip)
@@ -500,7 +500,6 @@ void celt_preemphasis(const opus_val16 * OPUS_RESTRICT pcmp, celt_sig * OPUS_RES
 #else
    (void)clip; /* Avoids a warning about clip being unused. */
 #endif
-   m = *mem;
 #ifdef CUSTOM_MODES
    if (coef[1] != 0)
    {
@@ -520,11 +519,11 @@ void celt_preemphasis(const opus_val16 * OPUS_RESTRICT pcmp, celt_sig * OPUS_RES
    {
       for (i=0;i<N;i++)
       {
-         celt_sig x;
-         x = SHL32(inp[i], SIG_SHIFT);
+         opus_val16 x;
+         x = inp[i];
          /* Apply pre-emphasis */
-         inp[i] = x + m;
-         m = - MULT16_32_Q15(coef0, x);
+         inp[i] = SHL32(x, SIG_SHIFT) - m;
+         m = SHR32(MULT16_16(coef0, x), 15-SIG_SHIFT);
       }
    }
    *mem = m;
@@ -575,15 +574,14 @@ static int tf_analysis(const CELTMode *m, int len, int isTransient,
    *tf_sum = 0;
    for (i=0;i<len;i++)
    {
-      int j, k, N;
+      int k, N;
       int narrow;
       opus_val32 L1, best_L1;
       int best_level=0;
       N = (m->eBands[i+1]-m->eBands[i])<<LM;
       /* band is too narrow to be split down to LM=-1 */
       narrow = (m->eBands[i+1]-m->eBands[i])==1;
-      for (j=0;j<N;j++)
-         tmp[j] = X[tf_chan*N0 + j+(m->eBands[i]<<LM)];
+      OPUS_COPY(tmp, &X[tf_chan*N0 + (m->eBands[i]<<LM)], N);
       /* Just add the right channel if we're in stereo */
       /*if (C==2)
          for (j=0;j<N;j++)
@@ -593,8 +591,7 @@ static int tf_analysis(const CELTMode *m, int len, int isTransient,
       /* Check the -1 case for transients */
       if (isTransient && !narrow)
       {
-         for (j=0;j<N;j++)
-            tmp_1[j] = tmp[j];
+         OPUS_COPY(tmp_1, tmp, N);
          haar1(tmp_1, N>>LM, 1<<LM);
          L1 = l1_metric(tmp_1, N, LM+1, bias);
          if (L1<best_L1)
@@ -754,12 +751,12 @@ static void tf_encode(int start, int end, int isTransient, int *tf_res, int LM,
 static int alloc_trim_analysis(const CELTMode *m, const celt_norm *X,
       const opus_val16 *bandLogE, int end, int LM, int C, int N0,
       AnalysisInfo *analysis, opus_val16 *stereo_saving, opus_val16 tf_estimate,
-      int intensity, opus_val16 surround_trim)
+      int intensity, opus_val16 surround_trim, int arch)
 {
    int i;
    opus_val32 diff=0;
    int c;
-   int trim_index = 5;
+   int trim_index;
    opus_val16 trim = QCONST16(5.f, 8);
    opus_val16 logXC, logXC2;
    if (C==2)
@@ -769,10 +766,9 @@ static int alloc_trim_analysis(const CELTMode *m, const celt_norm *X,
       /* Compute inter-channel correlation for low frequencies */
       for (i=0;i<8;i++)
       {
-         int j;
-         opus_val32 partial = 0;
-         for (j=m->eBands[i]<<LM;j<m->eBands[i+1]<<LM;j++)
-            partial = MAC16_16(partial, X[j], X[N0+j]);
+         opus_val32 partial;
+         partial = celt_inner_prod(&X[m->eBands[i]<<LM], &X[N0+(m->eBands[i]<<LM)],
+               (m->eBands[i+1]-m->eBands[i])<<LM, arch);
          sum = ADD16(sum, EXTRACT16(SHR32(partial, 18)));
       }
       sum = MULT16_16_Q15(QCONST16(1.f/8, 15), sum);
@@ -780,22 +776,13 @@ static int alloc_trim_analysis(const CELTMode *m, const celt_norm *X,
       minXC = sum;
       for (i=8;i<intensity;i++)
       {
-         int j;
-         opus_val32 partial = 0;
-         for (j=m->eBands[i]<<LM;j<m->eBands[i+1]<<LM;j++)
-            partial = MAC16_16(partial, X[j], X[N0+j]);
+         opus_val32 partial;
+         partial = celt_inner_prod(&X[m->eBands[i]<<LM], &X[N0+(m->eBands[i]<<LM)],
+               (m->eBands[i+1]-m->eBands[i])<<LM, arch);
          minXC = MIN16(minXC, ABS16(EXTRACT16(SHR32(partial, 18))));
       }
       minXC = MIN16(QCONST16(1.f, 10), ABS16(minXC));
       /*printf ("%f\n", sum);*/
-      if (sum > QCONST16(.995f,10))
-         trim_index-=4;
-      else if (sum > QCONST16(.92f,10))
-         trim_index-=3;
-      else if (sum > QCONST16(.85f,10))
-         trim_index-=2;
-      else if (sum > QCONST16(.8f,10))
-         trim_index-=1;
       /* mid-side savings estimations based on the LF average*/
       logXC = celt_log2(QCONST32(1.001f, 20)-MULT16_16(sum, sum));
       /* mid-side savings estimations based on min correlation */
@@ -819,14 +806,6 @@ static int alloc_trim_analysis(const CELTMode *m, const celt_norm *X,
    } while (++c<C);
    diff /= C*(end-1);
    /*printf("%f\n", diff);*/
-   if (diff > QCONST16(2.f, DB_SHIFT))
-      trim_index--;
-   if (diff > QCONST16(8.f, DB_SHIFT))
-      trim_index--;
-   if (diff < -QCONST16(4.f, DB_SHIFT))
-      trim_index++;
-   if (diff < -QCONST16(10.f, DB_SHIFT))
-      trim_index++;
    trim -= MAX16(-QCONST16(2.f, 8), MIN16(QCONST16(2.f, 8), SHR16(diff+QCONST16(1.f, DB_SHIFT),DB_SHIFT-8)/6 ));
    trim -= SHR16(surround_trim, DB_SHIFT-8);
    trim -= 2*SHR16(tf_estimate, 14-8);
@@ -836,6 +815,8 @@ static int alloc_trim_analysis(const CELTMode *m, const celt_norm *X,
       trim -= MAX16(-QCONST16(2.f, 8), MIN16(QCONST16(2.f, 8),
             (opus_val16)(QCONST16(2.f, 8)*(analysis->tonality_slope+.05f))));
    }
+#else
+   (void)analysis;
 #endif
 
 #ifdef OPUS_FIXED_POINT
@@ -843,10 +824,7 @@ static int alloc_trim_analysis(const CELTMode *m, const celt_norm *X,
 #else
    trim_index = (int)floor(.5f+trim);
 #endif
-   if (trim_index<0)
-      trim_index = 0;
-   if (trim_index>10)
-      trim_index = 10;
+   trim_index = IMAX(0, IMIN(10, trim_index));
    /*printf("%d\n", trim_index);*/
 #ifdef FUZZING
    trim_index = rand()%11;
@@ -886,6 +864,66 @@ static int stereo_analysis(const CELTMode *m, const celt_norm *X,
          > MULT16_32_Q15(m->eBands[13]<<(LM+1), sumLR);
 }
 
+#define MSWAP(a,b) do {opus_val16 tmp = a;a=b;b=tmp;} while(0)
+static opus_val16 median_of_5(const opus_val16 *x)
+{
+   opus_val16 t0, t1, t2, t3, t4;
+   t2 = x[2];
+   if (x[0] > x[1])
+   {
+      t0 = x[1];
+      t1 = x[0];
+   } else {
+      t0 = x[0];
+      t1 = x[1];
+   }
+   if (x[3] > x[4])
+   {
+      t3 = x[4];
+      t4 = x[3];
+   } else {
+      t3 = x[3];
+      t4 = x[4];
+   }
+   if (t0 > t3)
+   {
+      MSWAP(t0, t3);
+      MSWAP(t1, t4);
+   }
+   if (t2 > t1)
+   {
+      if (t1 < t3)
+         return MIN16(t2, t3);
+      else
+         return MIN16(t4, t1);
+   } else {
+      if (t2 < t3)
+         return MIN16(t1, t3);
+      else
+         return MIN16(t2, t4);
+   }
+}
+
+static opus_val16 median_of_3(const opus_val16 *x)
+{
+   opus_val16 t0, t1, t2;
+   if (x[0] > x[1])
+   {
+      t0 = x[1];
+      t1 = x[0];
+   } else {
+      t0 = x[0];
+      t1 = x[1];
+   }
+   t2 = x[2];
+   if (t1 < t2)
+      return t1;
+   else if (t0 < t2)
+      return t2;
+   else
+      return t0;
+}
+
 static opus_val16 dynalloc_analysis(const opus_val16 *bandLogE, const opus_val16 *bandLogE2,
       int nbEBands, int start, int end, int C, int *offsets, int lsb_depth, const opus_int16 *logN,
       int isTransient, int vbr, int constrained_vbr, const opus_int16 *eBands, int LM,
@@ -899,8 +937,7 @@ static opus_val16 dynalloc_analysis(const opus_val16 *bandLogE, const opus_val16
    SAVE_STACK;
    ALLOC(follower, C*nbEBands, opus_val16);
    ALLOC(noise_floor, C*nbEBands, opus_val16);
-   for (i=0;i<nbEBands;i++)
-      offsets[i] = 0;
+   OPUS_CLEAR(offsets, nbEBands);
    /* Dynamic allocation code */
    maxDepth=-QCONST16(31.9f, DB_SHIFT);
    for (i=0;i<end;i++)
@@ -922,7 +959,11 @@ static opus_val16 dynalloc_analysis(const opus_val16 *bandLogE, const opus_val16
       int last=0;
       c=0;do
       {
-         follower[c*nbEBands] = bandLogE2[c*nbEBands];
+         opus_val16 offset;
+         opus_val16 tmp;
+         opus_val16 *f;
+         f = &follower[c*nbEBands];
+         f[0] = bandLogE2[c*nbEBands];
          for (i=1;i<end;i++)
          {
             /* The last band to be at least 3 dB higher than the previous one
@@ -930,12 +971,26 @@ static opus_val16 dynalloc_analysis(const opus_val16 *bandLogE, const opus_val16
                bandlimited signals. */
             if (bandLogE2[c*nbEBands+i] > bandLogE2[c*nbEBands+i-1]+QCONST16(.5f,DB_SHIFT))
                last=i;
-            follower[c*nbEBands+i] = MIN16(follower[c*nbEBands+i-1]+QCONST16(1.5f,DB_SHIFT), bandLogE2[c*nbEBands+i]);
+            f[i] = MIN16(f[i-1]+QCONST16(1.5f,DB_SHIFT), bandLogE2[c*nbEBands+i]);
          }
          for (i=last-1;i>=0;i--)
-            follower[c*nbEBands+i] = MIN16(follower[c*nbEBands+i], MIN16(follower[c*nbEBands+i+1]+QCONST16(2.f,DB_SHIFT), bandLogE2[c*nbEBands+i]));
+            f[i] = MIN16(f[i], MIN16(f[i+1]+QCONST16(2.f,DB_SHIFT), bandLogE2[c*nbEBands+i]));
+
+         /* Combine with a median filter to avoid dynalloc triggering unnecessarily.
+            The "offset" value controls how conservative we are -- a higher offset
+            reduces the impact of the median filter and makes dynalloc use more bits. */
+         offset = QCONST16(1.f, DB_SHIFT);
+         for (i=2;i<end-2;i++)
+            f[i] = MAX16(f[i], median_of_5(&bandLogE2[c*nbEBands+i-2])-offset);
+         tmp = median_of_3(&bandLogE2[c*nbEBands])-offset;
+         f[0] = MAX16(f[0], tmp);
+         f[1] = MAX16(f[1], tmp);
+         tmp = median_of_3(&bandLogE2[c*nbEBands+end-3])-offset;
+         f[end-2] = MAX16(f[end-2], tmp);
+         f[end-1] = MAX16(f[end-1], tmp);
+
          for (i=0;i<end;i++)
-            follower[c*nbEBands+i] = MAX16(follower[c*nbEBands+i], noise_floor[i]);
+            f[i] = MAX16(f[i], noise_floor[i]);
       } while (++c<C);
       if (C==2)
       {
@@ -1016,9 +1071,11 @@ static int run_prefilter(CELTEncoder *st, celt_sig *in, celt_sig *prefilter_mem,
    opus_val16 pf_threshold;
    int pf_on;
    int qg;
+   int overlap;
    SAVE_STACK;
 
    mode = st->mode;
+   overlap = mode->overlap;
    ALLOC(_pre, CC*(N+COMBFILTER_MAXPERIOD), celt_sig);
 
    pre[0] = _pre;
@@ -1027,7 +1084,7 @@ static int run_prefilter(CELTEncoder *st, celt_sig *in, celt_sig *prefilter_mem,
 
    c=0; do {
       OPUS_COPY(pre[c], prefilter_mem+c*COMBFILTER_MAXPERIOD, COMBFILTER_MAXPERIOD);
-      OPUS_COPY(pre[c]+COMBFILTER_MAXPERIOD, in+c*(N+st->overlap)+st->overlap, N);
+      OPUS_COPY(pre[c]+COMBFILTER_MAXPERIOD, in+c*(N+overlap)+overlap, N);
    } while (++c<CC);
 
    if (enabled)
@@ -1044,7 +1101,7 @@ static int run_prefilter(CELTEncoder *st, celt_sig *in, celt_sig *prefilter_mem,
       pitch_index = COMBFILTER_MAXPERIOD-pitch_index;
 
       gain1 = remove_doubling(pitch_buf, COMBFILTER_MAXPERIOD, COMBFILTER_MINPERIOD,
-            N, &pitch_index, st->prefilter_period, st->prefilter_gain);
+            N, &pitch_index, st->prefilter_period, st->prefilter_gain, st->arch);
       if (pitch_index > COMBFILTER_MAXPERIOD-2)
          pitch_index = COMBFILTER_MAXPERIOD-2;
       gain1 = MULT16_16_Q15(QCONST16(.7f,15),gain1);
@@ -1100,18 +1157,18 @@ static int run_prefilter(CELTEncoder *st, celt_sig *in, celt_sig *prefilter_mem,
    /*printf("%d %f\n", pitch_index, gain1);*/
 
    c=0; do {
-      int offset = mode->shortMdctSize-st->overlap;
+      int offset = mode->shortMdctSize-overlap;
       st->prefilter_period=IMAX(st->prefilter_period, COMBFILTER_MINPERIOD);
-      OPUS_COPY(in+c*(N+st->overlap), st->in_mem+c*(st->overlap), st->overlap);
+      OPUS_COPY(in+c*(N+overlap), st->in_mem+c*(overlap), overlap);
       if (offset)
-         comb_filter(in+c*(N+st->overlap)+st->overlap, pre[c]+COMBFILTER_MAXPERIOD,
+         comb_filter(in+c*(N+overlap)+overlap, pre[c]+COMBFILTER_MAXPERIOD,
                st->prefilter_period, st->prefilter_period, offset, -st->prefilter_gain, -st->prefilter_gain,
-               st->prefilter_tapset, st->prefilter_tapset, NULL, 0);
+               st->prefilter_tapset, st->prefilter_tapset, NULL, 0, st->arch);
 
-      comb_filter(in+c*(N+st->overlap)+st->overlap+offset, pre[c]+COMBFILTER_MAXPERIOD+offset,
+      comb_filter(in+c*(N+overlap)+overlap+offset, pre[c]+COMBFILTER_MAXPERIOD+offset,
             st->prefilter_period, pitch_index, N-offset, -st->prefilter_gain, -gain1,
-            st->prefilter_tapset, prefilter_tapset, mode->window, st->overlap);
-      OPUS_COPY(st->in_mem+c*(st->overlap), in+c*(N+st->overlap)+N, st->overlap);
+            st->prefilter_tapset, prefilter_tapset, mode->window, overlap, st->arch);
+      OPUS_COPY(st->in_mem+c*(overlap), in+c*(N+overlap)+N, overlap);
 
       if (N>COMBFILTER_MAXPERIOD)
       {
@@ -1196,6 +1253,9 @@ static int compute_vbr(const CELTMode *mode, AnalysisInfo *analysis, opus_int32
       /*printf("%f %f ", analysis->tonality, tonal);*/
       target = tonal_target;
    }
+#else
+   (void)analysis;
+   (void)pitch_change;
 #endif
 
    if (has_surround_mask&&!lfe)
@@ -1273,6 +1333,8 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
    int LM, M;
    int tf_select;
    int nbFilledBytes, nbAvailableBytes;
+   int start;
+   int end;
    int effEnd;
    int codedBands;
    int tf_sum;
@@ -1316,6 +1378,8 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
    nbEBands = mode->nbEBands;
    overlap = mode->overlap;
    eBands = mode->eBands;
+   start = st->start;
+   end = st->end;
    tf_estimate = 0;
    if (nbCompressedBytes<2 || pcm==NULL)
    {
@@ -1335,8 +1399,8 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
    M=1<<LM;
    N = M*mode->shortMdctSize;
 
-   prefilter_mem = st->in_mem+CC*(st->overlap);
-   oldBandE = (opus_val16*)(st->in_mem+CC*(st->overlap+COMBFILTER_MAXPERIOD));
+   prefilter_mem = st->in_mem+CC*(overlap);
+   oldBandE = (opus_val16*)(st->in_mem+CC*(overlap+COMBFILTER_MAXPERIOD));
    oldLogE = oldBandE + CC*nbEBands;
    oldLogE2 = oldLogE + CC*nbEBands;
 
@@ -1352,8 +1416,8 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
 #ifdef CUSTOM_MODES
    if (st->signalling && enc==NULL)
    {
-      int tmp = (mode->effEBands-st->end)>>1;
-      st->end = IMAX(1, mode->effEBands-tmp);
+      int tmp = (mode->effEBands-end)>>1;
+      end = st->end = IMAX(1, mode->effEBands-tmp);
       compressed[0] = tmp<<5;
       compressed[0] |= LM<<3;
       compressed[0] |= (C==2)<<2;
@@ -1436,11 +1500,11 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
    }
    total_bits = nbCompressedBytes*8;
 
-   effEnd = st->end;
+   effEnd = end;
    if (effEnd > mode->effEBands)
       effEnd = mode->effEBands;
 
-   ALLOC(in, CC*(N+st->overlap), celt_sig);
+   ALLOC(in, CC*(N+overlap), celt_sig);
 
    sample_max=MAX32(st->overlap_max, celt_maxabs16(pcm, C*(N-overlap)/st->upsample));
    st->overlap_max=celt_maxabs16(pcm+C*(N-overlap)/st->upsample, C*overlap/st->upsample);
@@ -1474,8 +1538,12 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
       enc->nbits_total+=tell-ec_tell(enc);
    }
    c=0; do {
-      celt_preemphasis(pcm+c, in+c*(N+st->overlap)+st->overlap, N, CC, st->upsample,
-                  mode->preemph, st->preemph_memE+c, st->clip);
+      int need_clip=0;
+#ifndef OPUS_FIXED_POINT
+      need_clip = st->clip && sample_max>65536.f;
+#endif
+      celt_preemphasis(pcm+c, in+c*(N+overlap)+overlap, N, CC, st->upsample,
+                  mode->preemph, st->preemph_memE+c, need_clip);
    } while (++c<CC);
 
 
@@ -1484,7 +1552,7 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
    {
       int enabled;
       int qg;
-      enabled = ((st->lfe&&nbAvailableBytes>3) || nbAvailableBytes>12*C) && st->start==0 && !silence && !st->disable_pf
+      enabled = ((st->lfe&&nbAvailableBytes>3) || nbAvailableBytes>12*C) && start==0 && !silence && !st->disable_pf
             && st->complexity >= 5 && !(st->consec_transient && LM!=3 && st->variable_duration==OPUS_FRAMESIZE_VARIABLE);
 
       prefilter_tapset = st->tapset_decision;
@@ -1494,7 +1562,7 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
          pitch_change = 1;
       if (pf_on==0)
       {
-         if(st->start==0 && tell+16<=total_bits)
+         if(start==0 && tell+16<=total_bits)
             ec_enc_bit_logp(enc, 0, 1);
       } else {
          /*This block is not gated by a total bits check only because
@@ -1515,7 +1583,7 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
    shortBlocks = 0;
    if (st->complexity >= 1 && !st->lfe)
    {
-      isTransient = transient_analysis(in, N+st->overlap, CC,
+      isTransient = transient_analysis(in, N+overlap, CC,
             &tf_estimate, &tf_chan);
    }
    if (LM>0 && ec_tell(enc)+3<=total_bits)
@@ -1535,33 +1603,32 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
    ALLOC(bandLogE2, C*nbEBands, opus_val16);
    if (secondMdct)
    {
-      compute_mdcts(mode, 0, in, freq, C, CC, LM, st->upsample);
-      compute_band_energies(mode, freq, bandE, effEnd, C, M);
-      amp2Log2(mode, effEnd, st->end, bandE, bandLogE2, C);
+      compute_mdcts(mode, 0, in, freq, C, CC, LM, st->upsample, st->arch);
+      compute_band_energies(mode, freq, bandE, effEnd, C, LM);
+      amp2Log2(mode, effEnd, end, bandE, bandLogE2, C);
       for (i=0;i<C*nbEBands;i++)
          bandLogE2[i] += HALF16(SHL16(LM, DB_SHIFT));
    }
 
-   compute_mdcts(mode, shortBlocks, in, freq, C, CC, LM, st->upsample);
+   compute_mdcts(mode, shortBlocks, in, freq, C, CC, LM, st->upsample, st->arch);
    if (CC==2&&C==1)
       tf_chan = 0;
-   compute_band_energies(mode, freq, bandE, effEnd, C, M);
+   compute_band_energies(mode, freq, bandE, effEnd, C, LM);
 
    if (st->lfe)
    {
-      for (i=2;i<st->end;i++)
+      for (i=2;i<end;i++)
       {
          bandE[i] = IMIN(bandE[i], MULT16_32_Q15(QCONST16(1e-4f,15),bandE[0]));
          bandE[i] = MAX32(bandE[i], EPSILON);
       }
    }
-   amp2Log2(mode, effEnd, st->end, bandE, bandLogE, C);
+   amp2Log2(mode, effEnd, end, bandE, bandLogE, C);
 
    ALLOC(surround_dynalloc, C*nbEBands, opus_val16);
-   for(i=0;i<st->end;i++)
-      surround_dynalloc[i] = 0;
+   OPUS_CLEAR(surround_dynalloc, end);
    /* This computes how much masking takes place between surround channels */
-   if (st->start==0&&st->energy_mask&&!st->lfe)
+   if (start==0&&st->energy_mask&&!st->lfe)
    {
       int mask_end;
       int midband;
@@ -1584,6 +1651,7 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
             diff += MULT16_16(mask, 1+2*i-mask_end);
          }
       }
+      celt_assert(count>0);
       mask_avg = DIV32_16(mask_avg,count);
       mask_avg += QCONST16(.2f, DB_SHIFT);
       diff = diff*6/(C*(mask_end-1)*(mask_end+1)*mask_end);
@@ -1621,8 +1689,7 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
                disabling masking. */
             mask_avg = 0;
             diff = 0;
-            for(i=0;i<mask_end;i++)
-               surround_dynalloc[i] = 0;
+            OPUS_CLEAR(surround_dynalloc, mask_end);
          } else {
             for(i=0;i<mask_end;i++)
                surround_dynalloc[i] = MAX16(0, surround_dynalloc[i]-QCONST16(.25f, DB_SHIFT));
@@ -1640,14 +1707,14 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
       opus_val16 follow=-QCONST16(10.0f,DB_SHIFT);
       opus_val32 frame_avg=0;
       opus_val16 offset = shortBlocks?HALF16(SHL16(LM, DB_SHIFT)):0;
-      for(i=st->start;i<st->end;i++)
+      for(i=start;i<end;i++)
       {
          follow = MAX16(follow-QCONST16(1.f, DB_SHIFT), bandLogE[i]-offset);
          if (C==2)
             follow = MAX16(follow, bandLogE[i+nbEBands]-offset);
          frame_avg += follow;
       }
-      frame_avg /= (st->end-st->start);
+      frame_avg /= (end-start);
       temporal_vbr = SUB16(frame_avg,st->spec_avg);
       temporal_vbr = MIN16(QCONST16(3.f, DB_SHIFT), MAX16(-QCONST16(1.5f, DB_SHIFT), temporal_vbr));
       st->spec_avg += MULT16_16_Q15(QCONST16(.02f, 15), temporal_vbr);
@@ -1658,21 +1725,20 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
 
    if (!secondMdct)
    {
-      for (i=0;i<C*nbEBands;i++)
-         bandLogE2[i] = bandLogE[i];
+      OPUS_COPY(bandLogE2, bandLogE, C*nbEBands);
    }
 
    /* Last chance to catch any transient we might have missed in the
       time-domain analysis */
    if (LM>0 && ec_tell(enc)+3<=total_bits && !isTransient && st->complexity>=5 && !st->lfe)
    {
-      if (patch_transient_decision(bandLogE, oldBandE, nbEBands, st->end, C))
+      if (patch_transient_decision(bandLogE, oldBandE, nbEBands, start, end, C))
       {
          isTransient = 1;
          shortBlocks = M;
-         compute_mdcts(mode, shortBlocks, in, freq, C, CC, LM, st->upsample);
-         compute_band_energies(mode, freq, bandE, effEnd, C, M);
-         amp2Log2(mode, effEnd, st->end, bandE, bandLogE, C);
+         compute_mdcts(mode, shortBlocks, in, freq, C, CC, LM, st->upsample, st->arch);
+         compute_band_energies(mode, freq, bandE, effEnd, C, LM);
+         amp2Log2(mode, effEnd, end, bandE, bandLogE, C);
          /* Compensate for the scaling of short vs long mdcts */
          for (i=0;i<C*nbEBands;i++)
             bandLogE2[i] += HALF16(SHL16(LM, DB_SHIFT));
@@ -1690,7 +1756,7 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
 
    ALLOC(tf_res, nbEBands, int);
    /* Disable variable tf resolution for hybrid and at very low bitrate */
-   if (effectiveBytes>=15*C && st->start==0 && st->complexity>=2 && !st->lfe)
+   if (effectiveBytes>=15*C && start==0 && st->complexity>=2 && !st->lfe)
    {
       int lambda;
       if (effectiveBytes<40)
@@ -1703,22 +1769,22 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
          lambda = 3;
       lambda*=2;
       tf_select = tf_analysis(mode, effEnd, isTransient, tf_res, lambda, X, N, LM, &tf_sum, tf_estimate, tf_chan);
-      for (i=effEnd;i<st->end;i++)
+      for (i=effEnd;i<end;i++)
          tf_res[i] = tf_res[effEnd-1];
    } else {
       tf_sum = 0;
-      for (i=0;i<st->end;i++)
+      for (i=0;i<end;i++)
          tf_res[i] = isTransient;
       tf_select=0;
    }
 
    ALLOC(error, C*nbEBands, opus_val16);
-   quant_coarse_energy(mode, st->start, st->end, effEnd, bandLogE,
+   quant_coarse_energy(mode, start, end, effEnd, bandLogE,
          oldBandE, total_bits, error, enc,
          C, LM, nbAvailableBytes, st->force_intra,
          &st->delayedIntra, st->complexity >= 4, st->loss_rate, st->lfe);
 
-   tf_encode(st->start, st->end, isTransient, tf_res, LM, tf_select, enc);
+   tf_encode(start, end, isTransient, tf_res, LM, tf_select, enc);
 
    if (ec_tell(enc)+4<=total_bits)
    {
@@ -1726,7 +1792,7 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
       {
          st->tapset_decision = 0;
          st->spread_decision = SPREAD_NORMAL;
-      } else if (shortBlocks || st->complexity < 3 || nbAvailableBytes < 10*C || st->start != 0)
+      } else if (shortBlocks || st->complexity < 3 || nbAvailableBytes < 10*C || start != 0)
       {
          if (st->complexity == 0)
             st->spread_decision = SPREAD_NONE;
@@ -1760,7 +1826,7 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
 
    ALLOC(offsets, nbEBands, int);
 
-   maxDepth = dynalloc_analysis(bandLogE, bandLogE2, nbEBands, st->start, st->end, C, offsets,
+   maxDepth = dynalloc_analysis(bandLogE, bandLogE2, nbEBands, start, end, C, offsets,
          st->lsb_depth, mode->logN, isTransient, st->vbr, st->constrained_vbr,
          eBands, LM, effectiveBytes, &tot_boost, st->lfe, surround_dynalloc);
    /* For LFE, everything interesting is in the first band */
@@ -1773,7 +1839,7 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
    total_bits<<=BITRES;
    total_boost = 0;
    tell = ec_tell_frac(enc);
-   for (i=st->start;i<st->end;i++)
+   for (i=start;i<end;i++)
    {
       int width, quanta;
       int dynalloc_loop_logp;
@@ -1818,7 +1884,7 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
 
       st->intensity = hysteresis_decision((opus_val16)(equiv_rate/1000),
             intensity_thresholds, intensity_histeresis, 21, st->intensity);
-      st->intensity = IMIN(st->end,IMAX(st->start, st->intensity));
+      st->intensity = IMIN(end,IMAX(start, st->intensity));
    }
 
    alloc_trim = 5;
@@ -1828,7 +1894,8 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
          alloc_trim = 5;
       else
          alloc_trim = alloc_trim_analysis(mode, X, bandLogE,
-            st->end, LM, C, N, &st->analysis, &st->stereo_saving, tf_estimate, st->intensity, surround_trim);
+            end, LM, C, N, &st->analysis, &st->stereo_saving, tf_estimate,
+            st->intensity, surround_trim, st->arch);
       ec_enc_icdf(enc, alloc_trim, trim_icdf, 7);
       tell = ec_tell_frac(enc);
    }
@@ -1930,7 +1997,7 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
    bits = (((opus_int32)nbCompressedBytes*8)<<BITRES) - ec_tell_frac(enc) - 1;
    anti_collapse_rsv = isTransient&&LM>=2&&bits>=((LM+2)<<BITRES) ? (1<<BITRES) : 0;
    bits -= anti_collapse_rsv;
-   signalBandwidth = st->end-1;
+   signalBandwidth = end-1;
 #ifndef DISABLE_FLOAT_API
    if (st->analysis.valid)
    {
@@ -1950,7 +2017,7 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
 #endif
    if (st->lfe)
       signalBandwidth = 1;
-   codedBands = compute_allocation(mode, st->start, st->end, offsets, cap,
+   codedBands = compute_allocation(mode, start, end, offsets, cap,
          alloc_trim, &st->intensity, &dual_stereo, bits, &balance, pulses,
          fine_quant, fine_priority, C, LM, enc, 1, st->lastCodedBands, signalBandwidth);
    if (st->lastCodedBands)
@@ -1958,13 +2025,14 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
    else
       st->lastCodedBands = codedBands;
 
-   quant_fine_energy(mode, st->start, st->end, oldBandE, error, fine_quant, enc, C);
+   quant_fine_energy(mode, start, end, oldBandE, error, fine_quant, enc, C);
 
    /* Residual quantisation */
    ALLOC(collapse_masks, C*nbEBands, unsigned char);
-   quant_all_bands(1, mode, st->start, st->end, X, C==2 ? X+N : NULL, collapse_masks,
-         bandE, pulses, shortBlocks, st->spread_decision, dual_stereo, st->intensity, tf_res,
-         nbCompressedBytes*(8<<BITRES)-anti_collapse_rsv, balance, enc, LM, codedBands, &st->rng);
+   quant_all_bands(1, mode, start, end, X, C==2 ? X+N : NULL, collapse_masks,
+         bandE, pulses, shortBlocks, st->spread_decision,
+         dual_stereo, st->intensity, tf_res, nbCompressedBytes*(8<<BITRES)-anti_collapse_rsv,
+         balance, enc, LM, codedBands, &st->rng, st->arch);
 
    if (anti_collapse_rsv > 0)
    {
@@ -1974,7 +2042,7 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
 #endif
       ec_enc_bits(enc, anti_collapse_on, 1);
    }
-   quant_energy_finalise(mode, st->start, st->end, oldBandE, error, fine_quant, fine_priority, nbCompressedBytes*8-ec_tell(enc), enc, C);
+   quant_energy_finalise(mode, start, end, oldBandE, error, fine_quant, fine_priority, nbCompressedBytes*8-ec_tell(enc), enc, C);
 
    if (silence)
    {
@@ -1990,40 +2058,26 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
       if (anti_collapse_on)
       {
          anti_collapse(mode, X, collapse_masks, LM, C, N,
-               st->start, st->end, oldBandE, oldLogE, oldLogE2, pulses, st->rng);
-      }
-
-      if (silence)
-      {
-         for (i=0;i<C*N;i++)
-            freq[i] = 0;
-      } else {
-         /* Synthesis */
-         denormalise_bands(mode, X, freq, oldBandE, st->start, effEnd, C, M);
+               start, end, oldBandE, oldLogE, oldLogE2, pulses, st->rng);
       }
 
       c=0; do {
          OPUS_MOVE(st->syn_mem[c], st->syn_mem[c]+N, 2*MAX_PERIOD-N+overlap/2);
       } while (++c<CC);
 
-      if (CC==2&&C==1)
-      {
-         for (i=0;i<N;i++)
-            freq[N+i] = freq[i];
-      }
-
       c=0; do {
          out_mem[c] = st->syn_mem[c]+2*MAX_PERIOD-N;
       } while (++c<CC);
 
-      compute_inv_mdcts(mode, shortBlocks, freq, out_mem, CC, LM);
+      celt_synthesis(mode, X, out_mem, oldBandE, start, effEnd,
+                     C, CC, isTransient, LM, st->upsample, silence, st->arch);
 
       c=0; do {
          st->prefilter_period=IMAX(st->prefilter_period, COMBFILTER_MINPERIOD);
          st->prefilter_period_old=IMAX(st->prefilter_period_old, COMBFILTER_MINPERIOD);
          comb_filter(out_mem[c], out_mem[c], st->prefilter_period_old, st->prefilter_period, mode->shortMdctSize,
                st->prefilter_gain_old, st->prefilter_gain, st->prefilter_tapset_old, st->prefilter_tapset,
-               mode->window, st->overlap);
+               mode->window, overlap);
          if (LM!=0)
             comb_filter(out_mem[c]+mode->shortMdctSize, out_mem[c]+mode->shortMdctSize, st->prefilter_period, pitch_index, N-mode->shortMdctSize,
                   st->prefilter_gain, gain1, st->prefilter_tapset, prefilter_tapset,
@@ -2031,7 +2085,7 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
       } while (++c<CC);
 
       /* We reuse freq[] as scratch space for the de-emphasis */
-      deemphasis(out_mem, (opus_val16*)pcm, N, CC, st->upsample, mode->preemph, st->preemph_memD, freq);
+      deemphasis(out_mem, (opus_val16*)pcm, N, CC, st->upsample, mode->preemph, st->preemph_memD);
       st->prefilter_period_old = st->prefilter_period;
       st->prefilter_gain_old = st->prefilter_gain;
       st->prefilter_tapset_old = st->prefilter_tapset;
@@ -2051,16 +2105,13 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
 #endif
 
    if (CC==2&&C==1) {
-      for (i=0;i<nbEBands;i++)
-         oldBandE[nbEBands+i]=oldBandE[i];
+      OPUS_COPY(&oldBandE[nbEBands], oldBandE, nbEBands);
    }
 
    if (!isTransient)
    {
-      for (i=0;i<CC*nbEBands;i++)
-         oldLogE2[i] = oldLogE[i];
-      for (i=0;i<CC*nbEBands;i++)
-         oldLogE[i] = oldBandE[i];
+      OPUS_COPY(oldLogE2, oldLogE, CC*nbEBands);
+      OPUS_COPY(oldLogE, oldBandE, CC*nbEBands);
    } else {
       for (i=0;i<CC*nbEBands;i++)
          oldLogE[i] = MIN16(oldLogE[i], oldBandE[i]);
@@ -2068,12 +2119,12 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
    /* In case start or end were to change */
    c=0; do
    {
-      for (i=0;i<st->start;i++)
+      for (i=0;i<start;i++)
       {
          oldBandE[c*nbEBands+i]=0;
          oldLogE[c*nbEBands+i]=oldLogE2[c*nbEBands+i]=-QCONST16(28.f,DB_SHIFT);
       }
-      for (i=st->end;i<nbEBands;i++)
+      for (i=end;i<nbEBands;i++)
       {
          oldBandE[c*nbEBands+i]=0;
          oldLogE[c*nbEBands+i]=oldLogE2[c*nbEBands+i]=-QCONST16(28.f,DB_SHIFT);
@@ -2274,7 +2325,7 @@ int opus_custom_encoder_ctl(CELTEncoder * OPUS_RESTRICT st, int request, ...)
       {
          int i;
          opus_val16 *oldBandE, *oldLogE, *oldLogE2;
-         oldBandE = (opus_val16*)(st->in_mem+st->channels*(st->overlap+COMBFILTER_MAXPERIOD));
+         oldBandE = (opus_val16*)(st->in_mem+st->channels*(st->mode->overlap+COMBFILTER_MAXPERIOD));
          oldLogE = oldBandE + st->channels*st->mode->nbEBands;
          oldLogE2 = oldLogE + st->channels*st->mode->nbEBands;
          OPUS_CLEAR((char*)&st->ENCODER_RESET_START,
diff --git a/drivers/opus/celt/celt_lpc.c b/drivers/opus/celt/celt_lpc.c
index ad0a6dfd43..5b8c01021f 100644
--- a/drivers/opus/celt/celt_lpc.c
+++ b/drivers/opus/celt/celt_lpc.c
@@ -24,10 +24,7 @@
    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
-
-#ifdef OPUS_ENABLED
 #include "opus/opus_config.h"
-#endif
 
 #include "opus/celt/celt_lpc.h"
 #include "opus/celt/stack_alloc.h"
@@ -88,12 +85,15 @@ int          p
 #endif
 }
 
-void celt_fir(const opus_val16 *_x,
+
+void celt_fir_c(
+         const opus_val16 *_x,
          const opus_val16 *num,
          opus_val16 *_y,
          int N,
          int ord,
-         opus_val16 *mem)
+         opus_val16 *mem,
+         int arch)
 {
    int i,j;
    VARDECL(opus_val16, rnum);
@@ -111,6 +111,7 @@ void celt_fir(const opus_val16 *_x,
    for(i=0;i<ord;i++)
       mem[i] = _x[N-i-1];
 #ifdef SMALL_FOOTPRINT
+   (void)arch;
    for (i=0;i<N;i++)
    {
       opus_val32 sum = SHL32(EXTEND32(_x[i]), SIG_SHIFT);
@@ -124,7 +125,7 @@ void celt_fir(const opus_val16 *_x,
    for (i=0;i<N-3;i+=4)
    {
       opus_val32 sum[4]={0,0,0,0};
-      xcorr_kernel(rnum, x+i, sum, ord);
+      xcorr_kernel(rnum, x+i, sum, ord, arch);
       _y[i  ] = SATURATE16(ADD32(EXTEND32(_x[i  ]), PSHR32(sum[0], SIG_SHIFT)));
       _y[i+1] = SATURATE16(ADD32(EXTEND32(_x[i+1]), PSHR32(sum[1], SIG_SHIFT)));
       _y[i+2] = SATURATE16(ADD32(EXTEND32(_x[i+2]), PSHR32(sum[2], SIG_SHIFT)));
@@ -146,10 +147,12 @@ void celt_iir(const opus_val32 *_x,
          opus_val32 *_y,
          int N,
          int ord,
-         opus_val16 *mem)
+         opus_val16 *mem,
+         int arch)
 {
 #ifdef SMALL_FOOTPRINT
    int i,j;
+   (void)arch;
    for (i=0;i<N;i++)
    {
       opus_val32 sum = _x[i];
@@ -187,7 +190,7 @@ void celt_iir(const opus_val32 *_x,
       sum[1]=_x[i+1];
       sum[2]=_x[i+2];
       sum[3]=_x[i+3];
-      xcorr_kernel(rden, y+i, sum, ord);
+      xcorr_kernel(rden, y+i, sum, ord, arch);
 
       /* Patch up the result to compensate for the fact that this is an IIR */
       y[i+ord  ] = -ROUND16(sum[0],SIG_SHIFT);
diff --git a/drivers/opus/celt/celt_lpc.h b/drivers/opus/celt/celt_lpc.h
index 12bb78a6fd..dc0352d468 100644
--- a/drivers/opus/celt/celt_lpc.h
+++ b/drivers/opus/celt/celt_lpc.h
@@ -29,24 +29,37 @@
 #define PLC_H
 
 #include "opus/celt/arch.h"
+#include "opus/celt/cpu_support.h"
+
+#if defined(OPUS_X86_MAY_HAVE_SSE4_1)
+#include "opus/celt/x86/celt_lpc_sse.h"
+#endif
 
 #define LPC_ORDER 24
 
 void _celt_lpc(opus_val16 *_lpc, const opus_val32 *ac, int p);
 
-void celt_fir(const opus_val16 *x,
+void celt_fir_c(
+         const opus_val16 *x,
          const opus_val16 *num,
          opus_val16 *y,
          int N,
          int ord,
-         opus_val16 *mem);
+         opus_val16 *mem,
+         int arch);
+
+#if !defined(OVERRIDE_CELT_FIR)
+#define celt_fir(x, num, y, N, ord, mem, arch) \
+    (celt_fir_c(x, num, y, N, ord, mem, arch))
+#endif
 
 void celt_iir(const opus_val32 *x,
          const opus_val16 *den,
          opus_val32 *y,
          int N,
          int ord,
-         opus_val16 *mem);
+         opus_val16 *mem,
+         int arch);
 
 int _celt_autocorr(const opus_val16 *x, opus_val32 *ac,
          const opus_val16 *window, int overlap, int lag, int n, int arch);
diff --git a/drivers/opus/celt/cpu_support.h b/drivers/opus/celt/cpu_support.h
index f682a1d19e..2507e1861e 100644
--- a/drivers/opus/celt/cpu_support.h
+++ b/drivers/opus/celt/cpu_support.h
@@ -31,8 +31,9 @@
 #include "opus/opus_types.h"
 #include "opus/opus_defines.h"
 
-#if defined(OPUS_HAVE_RTCD) && defined(OPUS_ARM_ASM)
-#include "arm/armcpu.h"
+#if defined(OPUS_HAVE_RTCD) && \
+  (defined(OPUS_ARM_ASM) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR))
+#include "opus/celt/arm/armcpu.h"
 
 /* We currently support 4 ARM variants:
  * arch[0] -> ARMv4
@@ -42,6 +43,22 @@
  */
 #define OPUS_ARCHMASK 3
 
+#elif (defined(OPUS_X86_MAY_HAVE_SSE) && !defined(OPUS_X86_PRESUME_SSE)) || \
+  (defined(OPUS_X86_MAY_HAVE_SSE2) && !defined(OPUS_X86_PRESUME_SSE2)) || \
+  (defined(OPUS_X86_MAY_HAVE_SSE4_1) && !defined(OPUS_X86_PRESUME_SSE4_1)) || \
+  (defined(OPUS_X86_MAY_HAVE_AVX) && !defined(OPUS_X86_PRESUME_AVX))
+
+#include "opus/celt/x86/x86cpu.h"
+/* We currently support 5 x86 variants:
+ * arch[0] -> non-sse
+ * arch[1] -> sse
+ * arch[2] -> sse2
+ * arch[3] -> sse4.1
+ * arch[4] -> avx
+ */
+#define OPUS_ARCHMASK 7
+int opus_select_arch(void);
+
 #else
 #define OPUS_ARCHMASK 0
 
@@ -50,5 +67,4 @@ static OPUS_INLINE int opus_select_arch(void)
   return 0;
 }
 #endif
-
 #endif
diff --git a/drivers/opus/celt/cwrs.c b/drivers/opus/celt/cwrs.c
index bae9d21b97..983d4580a9 100644
--- a/drivers/opus/celt/cwrs.c
+++ b/drivers/opus/celt/cwrs.c
@@ -26,10 +26,7 @@
    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
-
-#ifdef OPUS_ENABLED
 #include "opus/opus_config.h"
-#endif
 
 #include "opus/celt/os_support.h"
 #include "opus/celt/cwrs.h"
@@ -460,10 +457,12 @@ void encode_pulses(const int *_y,int _n,int _k,ec_enc *_enc){
   ec_enc_uint(_enc,icwrs(_n,_y),CELT_PVQ_V(_n,_k));
 }
 
-static void cwrsi(int _n,int _k,opus_uint32 _i,int *_y){
+static opus_val32 cwrsi(int _n,int _k,opus_uint32 _i,int *_y){
   opus_uint32 p;
   int         s;
   int         k0;
+  opus_int16  val;
+  opus_val32  yy=0;
   celt_assert(_k>0);
   celt_assert(_n>1);
   while(_n>2){
@@ -487,7 +486,9 @@ static void cwrsi(int _n,int _k,opus_uint32 _i,int *_y){
       }
       else for(p=row[_k];p>_i;p=row[_k])_k--;
       _i-=p;
-      *_y++=(k0-_k+s)^s;
+      val=(k0-_k+s)^s;
+      *_y++=val;
+      yy=MAC16_16(yy,val,val);
     }
     /*Lots of dimensions case:*/
     else{
@@ -507,7 +508,9 @@ static void cwrsi(int _n,int _k,opus_uint32 _i,int *_y){
         do p=CELT_PVQ_U_ROW[--_k][_n];
         while(p>_i);
         _i-=p;
-        *_y++=(k0-_k+s)^s;
+        val=(k0-_k+s)^s;
+        *_y++=val;
+        yy=MAC16_16(yy,val,val);
       }
     }
     _n--;
@@ -519,14 +522,19 @@ static void cwrsi(int _n,int _k,opus_uint32 _i,int *_y){
   k0=_k;
   _k=(_i+1)>>1;
   if(_k)_i-=2*_k-1;
-  *_y++=(k0-_k+s)^s;
+  val=(k0-_k+s)^s;
+  *_y++=val;
+  yy=MAC16_16(yy,val,val);
   /*_n==1*/
   s=-(int)_i;
-  *_y=(_k+s)^s;
+  val=(_k+s)^s;
+  *_y=val;
+  yy=MAC16_16(yy,val,val);
+  return yy;
 }
 
-void decode_pulses(int *_y,int _n,int _k,ec_dec *_dec){
-  cwrsi(_n,_k,ec_dec_uint(_dec,CELT_PVQ_V(_n,_k)),_y);
+opus_val32 decode_pulses(int *_y,int _n,int _k,ec_dec *_dec){
+  return cwrsi(_n,_k,ec_dec_uint(_dec,CELT_PVQ_V(_n,_k)),_y);
 }
 
 #else /* SMALL_FOOTPRINT */
@@ -591,8 +599,10 @@ static opus_uint32 ncwrs_urow(unsigned _n,unsigned _k,opus_uint32 *_u){
   _y: Returns the vector of pulses.
   _u: Must contain entries [0..._k+1] of row _n of U() on input.
       Its contents will be destructively modified.*/
-static void cwrsi(int _n,int _k,opus_uint32 _i,int *_y,opus_uint32 *_u){
+static opus_val32 cwrsi(int _n,int _k,opus_uint32 _i,int *_y,opus_uint32 *_u){
   int j;
+  opus_int16 val;
+  opus_val32 yy=0;
   celt_assert(_n>0);
   j=0;
   do{
@@ -607,10 +617,13 @@ static void cwrsi(int _n,int _k,opus_uint32 _i,int *_y,opus_uint32 *_u){
     while(p>_i)p=_u[--_k];
     _i-=p;
     yj-=_k;
-    _y[j]=(yj+s)^s;
+    val=(yj+s)^s;
+    _y[j]=val;
+    yy=MAC16_16(yy,val,val);
     uprev(_u,_k+2,0);
   }
   while(++j<_n);
+  return yy;
 }
 
 /*Returns the index of the given combination of K elements chosen from a set
@@ -685,13 +698,15 @@ void encode_pulses(const int *_y,int _n,int _k,ec_enc *_enc){
   RESTORE_STACK;
 }
 
-void decode_pulses(int *_y,int _n,int _k,ec_dec *_dec){
+opus_val32 decode_pulses(int *_y,int _n,int _k,ec_dec *_dec){
   VARDECL(opus_uint32,u);
+  int ret;
   SAVE_STACK;
   celt_assert(_k>0);
   ALLOC(u,_k+2U,opus_uint32);
-  cwrsi(_n,_k,ec_dec_uint(_dec,ncwrs_urow(_n,_k,u)),_y,u);
+  ret = cwrsi(_n,_k,ec_dec_uint(_dec,ncwrs_urow(_n,_k,u)),_y,u);
   RESTORE_STACK;
+  return ret;
 }
 
 #endif /* SMALL_FOOTPRINT */
diff --git a/drivers/opus/celt/cwrs.h b/drivers/opus/celt/cwrs.h
index 5400afa6a4..f0ea844e5e 100644
--- a/drivers/opus/celt/cwrs.h
+++ b/drivers/opus/celt/cwrs.h
@@ -43,6 +43,6 @@ void get_required_bits(opus_int16 *bits, int N, int K, int frac);
 
 void encode_pulses(const int *_y, int N, int K, ec_enc *enc);
 
-void decode_pulses(int *_y, int N, int K, ec_dec *dec);
+opus_val32 decode_pulses(int *_y, int N, int K, ec_dec *dec);
 
 #endif /* CWRS_H */
diff --git a/drivers/opus/celt/entcode.c b/drivers/opus/celt/entcode.c
index 5c9874908d..d1c6b25c7f 100644
--- a/drivers/opus/celt/entcode.c
+++ b/drivers/opus/celt/entcode.c
@@ -24,10 +24,7 @@
    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
-
-#ifdef OPUS_ENABLED
 #include "opus/opus_config.h"
-#endif
 
 #include "opus/celt/entcode.h"
 #include "opus/celt/arch.h"
@@ -62,6 +59,27 @@ int ec_ilog(opus_uint32 _v){
 }
 #endif
 
+#if 1
+/* This is a faster version of ec_tell_frac() that takes advantage
+   of the low (1/8 bit) resolution to use just a linear function
+   followed by a lookup to determine the exact transition thresholds. */
+opus_uint32 ec_tell_frac(ec_ctx *_this){
+  static const unsigned correction[8] =
+    {35733, 38967, 42495, 46340,
+     50535, 55109, 60097, 65535};
+  opus_uint32 nbits;
+  opus_uint32 r;
+  int         l;
+  unsigned    b;
+  nbits=_this->nbits_total<<BITRES;
+  l=EC_ILOG(_this->rng);
+  r=_this->rng>>(l-16);
+  b = (r>>12)-8;
+  b += r>correction[b];
+  l = (l<<3)+b;
+  return nbits-l;
+}
+#else
 opus_uint32 ec_tell_frac(ec_ctx *_this){
   opus_uint32 nbits;
   opus_uint32 r;
@@ -91,3 +109,42 @@ opus_uint32 ec_tell_frac(ec_ctx *_this){
   }
   return nbits-l;
 }
+#endif
+
+#ifdef USE_SMALL_DIV_TABLE
+/* Result of 2^32/(2*i+1), except for i=0. */
+const opus_uint32 SMALL_DIV_TABLE[129] = {
+   0xFFFFFFFF, 0x55555555, 0x33333333, 0x24924924,
+   0x1C71C71C, 0x1745D174, 0x13B13B13, 0x11111111,
+   0x0F0F0F0F, 0x0D79435E, 0x0C30C30C, 0x0B21642C,
+   0x0A3D70A3, 0x097B425E, 0x08D3DCB0, 0x08421084,
+   0x07C1F07C, 0x07507507, 0x06EB3E45, 0x06906906,
+   0x063E7063, 0x05F417D0, 0x05B05B05, 0x0572620A,
+   0x05397829, 0x05050505, 0x04D4873E, 0x04A7904A,
+   0x047DC11F, 0x0456C797, 0x04325C53, 0x04104104,
+   0x03F03F03, 0x03D22635, 0x03B5CC0E, 0x039B0AD1,
+   0x0381C0E0, 0x0369D036, 0x03531DEC, 0x033D91D2,
+   0x0329161F, 0x03159721, 0x03030303, 0x02F14990,
+   0x02E05C0B, 0x02D02D02, 0x02C0B02C, 0x02B1DA46,
+   0x02A3A0FD, 0x0295FAD4, 0x0288DF0C, 0x027C4597,
+   0x02702702, 0x02647C69, 0x02593F69, 0x024E6A17,
+   0x0243F6F0, 0x0239E0D5, 0x02302302, 0x0226B902,
+   0x021D9EAD, 0x0214D021, 0x020C49BA, 0x02040810,
+   0x01FC07F0, 0x01F44659, 0x01ECC07B, 0x01E573AC,
+   0x01DE5D6E, 0x01D77B65, 0x01D0CB58, 0x01CA4B30,
+   0x01C3F8F0, 0x01BDD2B8, 0x01B7D6C3, 0x01B20364,
+   0x01AC5701, 0x01A6D01A, 0x01A16D3F, 0x019C2D14,
+   0x01970E4F, 0x01920FB4, 0x018D3018, 0x01886E5F,
+   0x0183C977, 0x017F405F, 0x017AD220, 0x01767DCE,
+   0x01724287, 0x016E1F76, 0x016A13CD, 0x01661EC6,
+   0x01623FA7, 0x015E75BB, 0x015AC056, 0x01571ED3,
+   0x01539094, 0x01501501, 0x014CAB88, 0x0149539E,
+   0x01460CBC, 0x0142D662, 0x013FB013, 0x013C995A,
+   0x013991C2, 0x013698DF, 0x0133AE45, 0x0130D190,
+   0x012E025C, 0x012B404A, 0x01288B01, 0x0125E227,
+   0x01234567, 0x0120B470, 0x011E2EF3, 0x011BB4A4,
+   0x01194538, 0x0116E068, 0x011485F0, 0x0112358E,
+   0x010FEF01, 0x010DB20A, 0x010B7E6E, 0x010953F3,
+   0x01073260, 0x0105197F, 0x0103091B, 0x01010101
+};
+#endif
diff --git a/drivers/opus/celt/entcode.h b/drivers/opus/celt/entcode.h
index c129f9b7d9..4d736d4dca 100644
--- a/drivers/opus/celt/entcode.h
+++ b/drivers/opus/celt/entcode.h
@@ -34,6 +34,12 @@
 # include <stddef.h>
 # include "opus/celt/ecintrin.h"
 
+extern const opus_uint32 SMALL_DIV_TABLE[129];
+
+#ifdef OPUS_ARM_ASM
+#define USE_SMALL_DIV_TABLE
+#endif
+
 /*OPT: ec_window must be at least 32 bits, but if you have fast arithmetic on a
    larger type, you can speed up the decoder by using it here.*/
 typedef opus_uint32           ec_window;
@@ -114,4 +120,33 @@ static OPUS_INLINE int ec_tell(ec_ctx *_this){
            rounding error is in the positive direction).*/
 opus_uint32 ec_tell_frac(ec_ctx *_this);
 
+/* Tested exhaustively for all n and for 1<=d<=256 */
+static OPUS_INLINE opus_uint32 celt_udiv(opus_uint32 n, opus_uint32 d) {
+   celt_assert(d>0);
+#ifdef USE_SMALL_DIV_TABLE
+   if (d>256)
+      return n/d;
+   else {
+      opus_uint32 t, q;
+      t = EC_ILOG(d&-d);
+      q = (opus_uint64)SMALL_DIV_TABLE[d>>t]*(n>>(t-1))>>32;
+      return q+(n-q*d >= d);
+   }
+#else
+   return n/d;
+#endif
+}
+
+static OPUS_INLINE opus_int32 celt_sudiv(opus_int32 n, opus_int32 d) {
+   celt_assert(d>0);
+#ifdef USE_SMALL_DIV_TABLE
+   if (n<0)
+      return -(opus_int32)celt_udiv(-n, d);
+   else
+      return celt_udiv(n, d);
+#else
+   return n/d;
+#endif
+}
+
 #endif
diff --git a/drivers/opus/celt/entdec.c b/drivers/opus/celt/entdec.c
index 0ec4d460d2..fdc49c92c1 100644
--- a/drivers/opus/celt/entdec.c
+++ b/drivers/opus/celt/entdec.c
@@ -24,10 +24,7 @@
    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
-
-#ifdef OPUS_ENABLED
 #include "opus/opus_config.h"
-#endif
 
 #include <stddef.h>
 #include "opus/celt/os_support.h"
@@ -138,7 +135,7 @@ void ec_dec_init(ec_dec *_this,unsigned char *_buf,opus_uint32 _storage){
 
 unsigned ec_decode(ec_dec *_this,unsigned _ft){
   unsigned s;
-  _this->ext=_this->rng/_ft;
+  _this->ext=celt_udiv(_this->rng,_ft);
   s=(unsigned)(_this->val/_this->ext);
   return _ft-EC_MINI(s+1,_ft);
 }
diff --git a/drivers/opus/celt/entenc.c b/drivers/opus/celt/entenc.c
index 085b2b2816..da7f2c8e62 100644
--- a/drivers/opus/celt/entenc.c
+++ b/drivers/opus/celt/entenc.c
@@ -24,10 +24,7 @@
    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
-
-#if defined(OPUS_ENABLED)
-# include "opus/opus_config.h"
-#endif
+#include "opus/opus_config.h"
 #include "opus/celt/os_support.h"
 #include "opus/celt/arch.h"
 #include "opus/celt/entenc.h"
@@ -98,7 +95,7 @@ static void ec_enc_carry_out(ec_enc *_this,int _c){
   else _this->ext++;
 }
 
-static void ec_enc_normalize(ec_enc *_this){
+static OPUS_INLINE void ec_enc_normalize(ec_enc *_this){
   /*If the range is too small, output some bits and rescale it.*/
   while(_this->rng<=EC_CODE_BOT){
     ec_enc_carry_out(_this,(int)(_this->val>>EC_CODE_SHIFT));
@@ -127,7 +124,7 @@ void ec_enc_init(ec_enc *_this,unsigned char *_buf,opus_uint32 _size){
 
 void ec_encode(ec_enc *_this,unsigned _fl,unsigned _fh,unsigned _ft){
   opus_uint32 r;
-  r=_this->rng/_ft;
+  r=celt_udiv(_this->rng,_ft);
   if(_fl>0){
     _this->val+=_this->rng-IMUL32(r,(_ft-_fl));
     _this->rng=IMUL32(r,(_fh-_fl));
diff --git a/drivers/opus/celt/fixed_debug.h b/drivers/opus/celt/fixed_debug.h
index 0ed16baa17..3b8d5ef7a9 100644
--- a/drivers/opus/celt/fixed_debug.h
+++ b/drivers/opus/celt/fixed_debug.h
@@ -496,6 +496,7 @@ static OPUS_INLINE int MULT16_32_PX_(int a, opus_int64 b, int Q, char *file, int
 
 #define MULT16_32_Q15(a,b) MULT16_32_QX(a,b,15)
 #define MAC16_32_Q15(c,a,b) (celt_mips-=2,ADD32((c),MULT16_32_Q15((a),(b))))
+#define MAC16_32_Q16(c,a,b) (celt_mips-=2,ADD32((c),MULT16_32_Q16((a),(b))))
 
 static OPUS_INLINE int SATURATE(int a, int b)
 {
@@ -767,6 +768,16 @@ static OPUS_INLINE int DIV32_(opus_int64 a, opus_int64 b, char *file, int line)
    return res;
 }
 
+static OPUS_INLINE opus_val16 SIG2WORD16_generic(celt_sig x)
+{
+   x = PSHR32(x, SIG_SHIFT);
+   x = MAX32(x, -32768);
+   x = MIN32(x, 32767);
+   return EXTRACT16(x);
+}
+#define SIG2WORD16(x) (SIG2WORD16_generic(x))
+
+
 #undef PRINT_MIPS
 #define PRINT_MIPS(file) do {fprintf (file, "total complexity = %llu MIPS\n", celt_mips);} while (0);
 
diff --git a/drivers/opus/celt/fixed_generic.h b/drivers/opus/celt/fixed_generic.h
index ecf018a244..71ab048e4a 100644
--- a/drivers/opus/celt/fixed_generic.h
+++ b/drivers/opus/celt/fixed_generic.h
@@ -113,7 +113,11 @@
 /** 16x32 multiply, followed by a 15-bit shift right and 32-bit add.
     b must fit in 31 bits.
     Result fits in 32 bits. */
-#define MAC16_32_Q15(c,a,b) ADD32(c,ADD32(MULT16_16((a),SHR((b),15)), SHR(MULT16_16((a),((b)&0x00007fff)),15)))
+#define MAC16_32_Q15(c,a,b) ADD32((c),ADD32(MULT16_16((a),SHR((b),15)), SHR(MULT16_16((a),((b)&0x00007fff)),15)))
+
+/** 16x32 multiplication, followed by a 16-bit shift right and 32-bit add.
+    Results fits in 32 bits */
+#define MAC16_32_Q16(c,a,b) ADD32((c),ADD32(MULT16_16((a),SHR((b),16)), SHR(MULT16_16SU((a),((b)&0x0000ffff)),16)))
 
 #define MULT16_16_Q11_32(a,b) (SHR(MULT16_16((a),(b)),11))
 #define MULT16_16_Q11(a,b) (SHR(MULT16_16((a),(b)),11))
@@ -131,4 +135,17 @@
 /** Divide a 32-bit value by a 32-bit value. Result fits in 32 bits */
 #define DIV32(a,b) (((opus_val32)(a))/((opus_val32)(b)))
 
+#if defined(MIPSr1_ASM)
+#include "opus/celt/mips/fixed_generic_mipsr1.h"
+#endif
+
+static OPUS_INLINE opus_val16 SIG2WORD16_generic(celt_sig x)
+{
+   x = PSHR32(x, SIG_SHIFT);
+   x = MAX32(x, -32768);
+   x = MIN32(x, 32767);
+   return EXTRACT16(x);
+}
+#define SIG2WORD16(x) (SIG2WORD16_generic(x))
+
 #endif
diff --git a/drivers/opus/celt/float_cast.h b/drivers/opus/celt/float_cast.h
index 86e80a93ff..36a1bfe671 100644
--- a/drivers/opus/celt/float_cast.h
+++ b/drivers/opus/celt/float_cast.h
@@ -90,14 +90,14 @@
 #include <math.h>
 #define float2int(x) lrint(x)
 
-#elif (defined(_MSC_VER) && _MSC_VER >= 1400) && (defined (WIN64) || defined (_WIN64))
+#elif (defined(_MSC_VER) && _MSC_VER >= 1400) && defined (_M_X64)
         #include <xmmintrin.h>
 
         __inline long int float2int(float value)
         {
                 return _mm_cvtss_si32(_mm_load_ss(&value));
         }
-#elif (defined(_MSC_VER) && _MSC_VER >= 1400) && (defined (WIN32) || defined (_WIN32))
+#elif (defined(_MSC_VER) && _MSC_VER >= 1400) && defined (_M_IX86)
         #include <math.h>
 
         /*      Win32 doesn't seem to have these functions.
diff --git a/drivers/opus/celt/kiss_fft.c b/drivers/opus/celt/kiss_fft.c
index 89a1790b24..cf8d049fa1 100644
--- a/drivers/opus/celt/kiss_fft.c
+++ b/drivers/opus/celt/kiss_fft.c
@@ -30,9 +30,7 @@
    heavily modified to better suit Opus */
 
 #ifndef SKIP_CONFIG_H
-#  ifdef OPUS_ENABLED
 #include "opus/opus_config.h"
-#  endif
 #endif
 
 #include "opus/celt/_kiss_fft_guts.h"
@@ -47,64 +45,56 @@
 
 static void kf_bfly2(
                      kiss_fft_cpx * Fout,
-                     const size_t fstride,
-                     const kiss_fft_state *st,
                      int m,
-                     int N,
-                     int mm
+                     int N
                     )
 {
    kiss_fft_cpx * Fout2;
-   const kiss_twiddle_cpx * tw1;
-   int i,j;
-   kiss_fft_cpx * Fout_beg = Fout;
-   for (i=0;i<N;i++)
+   int i;
+   (void)m;
+#ifdef CUSTOM_MODES
+   if (m==1)
    {
-      Fout = Fout_beg + i*mm;
-      Fout2 = Fout + m;
-      tw1 = st->twiddles;
-      for(j=0;j<m;j++)
+      celt_assert(m==1);
+      for (i=0;i<N;i++)
       {
          kiss_fft_cpx t;
-         Fout->r = SHR32(Fout->r, 1);Fout->i = SHR32(Fout->i, 1);
-         Fout2->r = SHR32(Fout2->r, 1);Fout2->i = SHR32(Fout2->i, 1);
-         C_MUL (t,  *Fout2 , *tw1);
-         tw1 += fstride;
+         Fout2 = Fout + 1;
+         t = *Fout2;
          C_SUB( *Fout2 ,  *Fout , t );
          C_ADDTO( *Fout ,  t );
-         ++Fout2;
-         ++Fout;
+         Fout += 2;
       }
-   }
-}
-
-static void ki_bfly2(
-                     kiss_fft_cpx * Fout,
-                     const size_t fstride,
-                     const kiss_fft_state *st,
-                     int m,
-                     int N,
-                     int mm
-                    )
-{
-   kiss_fft_cpx * Fout2;
-   const kiss_twiddle_cpx * tw1;
-   kiss_fft_cpx t;
-   int i,j;
-   kiss_fft_cpx * Fout_beg = Fout;
-   for (i=0;i<N;i++)
+   } else
+#endif
    {
-      Fout = Fout_beg + i*mm;
-      Fout2 = Fout + m;
-      tw1 = st->twiddles;
-      for(j=0;j<m;j++)
+      opus_val16 tw;
+      tw = QCONST16(0.7071067812f, 15);
+      /* We know that m==4 here because the radix-2 is just after a radix-4 */
+      celt_assert(m==4);
+      for (i=0;i<N;i++)
       {
-         C_MULC (t,  *Fout2 , *tw1);
-         tw1 += fstride;
-         C_SUB( *Fout2 ,  *Fout , t );
-         C_ADDTO( *Fout ,  t );
-         ++Fout2;
-         ++Fout;
+         kiss_fft_cpx t;
+         Fout2 = Fout + 4;
+         t = Fout2[0];
+         C_SUB( Fout2[0] ,  Fout[0] , t );
+         C_ADDTO( Fout[0] ,  t );
+
+         t.r = S_MUL(Fout2[1].r+Fout2[1].i, tw);
+         t.i = S_MUL(Fout2[1].i-Fout2[1].r, tw);
+         C_SUB( Fout2[1] ,  Fout[1] , t );
+         C_ADDTO( Fout[1] ,  t );
+
+         t.r = Fout2[2].i;
+         t.i = -Fout2[2].r;
+         C_SUB( Fout2[2] ,  Fout[2] , t );
+         C_ADDTO( Fout[2] ,  t );
+
+         t.r = S_MUL(Fout2[3].i-Fout2[3].r, tw);
+         t.i = S_MUL(-Fout2[3].i-Fout2[3].r, tw);
+         C_SUB( Fout2[3] ,  Fout[3] , t );
+         C_ADDTO( Fout[3] ,  t );
+         Fout += 8;
       }
    }
 }
@@ -118,89 +108,67 @@ static void kf_bfly4(
                      int mm
                     )
 {
-   const kiss_twiddle_cpx *tw1,*tw2,*tw3;
-   kiss_fft_cpx scratch[6];
-   const size_t m2=2*m;
-   const size_t m3=3*m;
-   int i, j;
+   int i;
 
-   kiss_fft_cpx * Fout_beg = Fout;
-   for (i=0;i<N;i++)
+   if (m==1)
    {
-      Fout = Fout_beg + i*mm;
-      tw3 = tw2 = tw1 = st->twiddles;
-      for (j=0;j<m;j++)
+      /* Degenerate case where all the twiddles are 1. */
+      for (i=0;i<N;i++)
       {
-         C_MUL4(scratch[0],Fout[m] , *tw1 );
-         C_MUL4(scratch[1],Fout[m2] , *tw2 );
-         C_MUL4(scratch[2],Fout[m3] , *tw3 );
-
-         Fout->r = PSHR32(Fout->r, 2);
-         Fout->i = PSHR32(Fout->i, 2);
-         C_SUB( scratch[5] , *Fout, scratch[1] );
-         C_ADDTO(*Fout, scratch[1]);
-         C_ADD( scratch[3] , scratch[0] , scratch[2] );
-         C_SUB( scratch[4] , scratch[0] , scratch[2] );
-         C_SUB( Fout[m2], *Fout, scratch[3] );
-         tw1 += fstride;
-         tw2 += fstride*2;
-         tw3 += fstride*3;
-         C_ADDTO( *Fout , scratch[3] );
-
-         Fout[m].r = scratch[5].r + scratch[4].i;
-         Fout[m].i = scratch[5].i - scratch[4].r;
-         Fout[m3].r = scratch[5].r - scratch[4].i;
-         Fout[m3].i = scratch[5].i + scratch[4].r;
-         ++Fout;
+         kiss_fft_cpx scratch0, scratch1;
+
+         C_SUB( scratch0 , *Fout, Fout[2] );
+         C_ADDTO(*Fout, Fout[2]);
+         C_ADD( scratch1 , Fout[1] , Fout[3] );
+         C_SUB( Fout[2], *Fout, scratch1 );
+         C_ADDTO( *Fout , scratch1 );
+         C_SUB( scratch1 , Fout[1] , Fout[3] );
+
+         Fout[1].r = scratch0.r + scratch1.i;
+         Fout[1].i = scratch0.i - scratch1.r;
+         Fout[3].r = scratch0.r - scratch1.i;
+         Fout[3].i = scratch0.i + scratch1.r;
+         Fout+=4;
       }
-   }
-}
-
-static void ki_bfly4(
-                     kiss_fft_cpx * Fout,
-                     const size_t fstride,
-                     const kiss_fft_state *st,
-                     int m,
-                     int N,
-                     int mm
-                    )
-{
-   const kiss_twiddle_cpx *tw1,*tw2,*tw3;
-   kiss_fft_cpx scratch[6];
-   const size_t m2=2*m;
-   const size_t m3=3*m;
-   int i, j;
-
-   kiss_fft_cpx * Fout_beg = Fout;
-   for (i=0;i<N;i++)
-   {
-      Fout = Fout_beg + i*mm;
-      tw3 = tw2 = tw1 = st->twiddles;
-      for (j=0;j<m;j++)
+   } else {
+      int j;
+      kiss_fft_cpx scratch[6];
+      const kiss_twiddle_cpx *tw1,*tw2,*tw3;
+      const int m2=2*m;
+      const int m3=3*m;
+      kiss_fft_cpx * Fout_beg = Fout;
+      for (i=0;i<N;i++)
       {
-         C_MULC(scratch[0],Fout[m] , *tw1 );
-         C_MULC(scratch[1],Fout[m2] , *tw2 );
-         C_MULC(scratch[2],Fout[m3] , *tw3 );
-
-         C_SUB( scratch[5] , *Fout, scratch[1] );
-         C_ADDTO(*Fout, scratch[1]);
-         C_ADD( scratch[3] , scratch[0] , scratch[2] );
-         C_SUB( scratch[4] , scratch[0] , scratch[2] );
-         C_SUB( Fout[m2], *Fout, scratch[3] );
-         tw1 += fstride;
-         tw2 += fstride*2;
-         tw3 += fstride*3;
-         C_ADDTO( *Fout , scratch[3] );
-
-         Fout[m].r = scratch[5].r - scratch[4].i;
-         Fout[m].i = scratch[5].i + scratch[4].r;
-         Fout[m3].r = scratch[5].r + scratch[4].i;
-         Fout[m3].i = scratch[5].i - scratch[4].r;
-         ++Fout;
+         Fout = Fout_beg + i*mm;
+         tw3 = tw2 = tw1 = st->twiddles;
+         /* m is guaranteed to be a multiple of 4. */
+         for (j=0;j<m;j++)
+         {
+            C_MUL(scratch[0],Fout[m] , *tw1 );
+            C_MUL(scratch[1],Fout[m2] , *tw2 );
+            C_MUL(scratch[2],Fout[m3] , *tw3 );
+
+            C_SUB( scratch[5] , *Fout, scratch[1] );
+            C_ADDTO(*Fout, scratch[1]);
+            C_ADD( scratch[3] , scratch[0] , scratch[2] );
+            C_SUB( scratch[4] , scratch[0] , scratch[2] );
+            C_SUB( Fout[m2], *Fout, scratch[3] );
+            tw1 += fstride;
+            tw2 += fstride*2;
+            tw3 += fstride*3;
+            C_ADDTO( *Fout , scratch[3] );
+
+            Fout[m].r = scratch[5].r + scratch[4].i;
+            Fout[m].i = scratch[5].i - scratch[4].r;
+            Fout[m3].r = scratch[5].r - scratch[4].i;
+            Fout[m3].i = scratch[5].i + scratch[4].r;
+            ++Fout;
+         }
       }
    }
 }
 
+
 #ifndef RADIX_TWO_ONLY
 
 static void kf_bfly3(
@@ -220,14 +188,19 @@ static void kf_bfly3(
    kiss_twiddle_cpx epi3;
 
    kiss_fft_cpx * Fout_beg = Fout;
+#ifdef OPUS_FIXED_POINT
+   epi3.r = -16384;
+   epi3.i = -28378;
+#else
    epi3 = st->twiddles[fstride*m];
+#endif
    for (i=0;i<N;i++)
    {
       Fout = Fout_beg + i*mm;
       tw1=tw2=st->twiddles;
+      /* For non-custom modes, m is guaranteed to be a multiple of 4. */
       k=m;
       do {
-         C_FIXDIV(*Fout,3); C_FIXDIV(Fout[m],3); C_FIXDIV(Fout[m2],3);
 
          C_MUL(scratch[1],Fout[m] , *tw1);
          C_MUL(scratch[2],Fout[m2] , *tw2);
@@ -255,56 +228,8 @@ static void kf_bfly3(
    }
 }
 
-static void ki_bfly3(
-                     kiss_fft_cpx * Fout,
-                     const size_t fstride,
-                     const kiss_fft_state *st,
-                     int m,
-                     int N,
-                     int mm
-                    )
-{
-   int i, k;
-   const size_t m2 = 2*m;
-   const kiss_twiddle_cpx *tw1,*tw2;
-   kiss_fft_cpx scratch[5];
-   kiss_twiddle_cpx epi3;
-
-   kiss_fft_cpx * Fout_beg = Fout;
-   epi3 = st->twiddles[fstride*m];
-   for (i=0;i<N;i++)
-   {
-      Fout = Fout_beg + i*mm;
-      tw1=tw2=st->twiddles;
-      k=m;
-      do{
-
-         C_MULC(scratch[1],Fout[m] , *tw1);
-         C_MULC(scratch[2],Fout[m2] , *tw2);
-
-         C_ADD(scratch[3],scratch[1],scratch[2]);
-         C_SUB(scratch[0],scratch[1],scratch[2]);
-         tw1 += fstride;
-         tw2 += fstride*2;
-
-         Fout[m].r = Fout->r - HALF_OF(scratch[3].r);
-         Fout[m].i = Fout->i - HALF_OF(scratch[3].i);
-
-         C_MULBYSCALAR( scratch[0] , -epi3.i );
-
-         C_ADDTO(*Fout,scratch[3]);
-
-         Fout[m2].r = Fout[m].r + scratch[0].i;
-         Fout[m2].i = Fout[m].i - scratch[0].r;
-
-         Fout[m].r -= scratch[0].i;
-         Fout[m].i += scratch[0].r;
-
-         ++Fout;
-      }while(--k);
-   }
-}
 
+#ifndef OVERRIDE_kf_bfly5
 static void kf_bfly5(
                      kiss_fft_cpx * Fout,
                      const size_t fstride,
@@ -317,13 +242,19 @@ static void kf_bfly5(
    kiss_fft_cpx *Fout0,*Fout1,*Fout2,*Fout3,*Fout4;
    int i, u;
    kiss_fft_cpx scratch[13];
-   const kiss_twiddle_cpx * twiddles = st->twiddles;
    const kiss_twiddle_cpx *tw;
    kiss_twiddle_cpx ya,yb;
    kiss_fft_cpx * Fout_beg = Fout;
 
-   ya = twiddles[fstride*m];
-   yb = twiddles[fstride*2*m];
+#ifdef OPUS_FIXED_POINT
+   ya.r = 10126;
+   ya.i = -31164;
+   yb.r = -26510;
+   yb.i = -19261;
+#else
+   ya = st->twiddles[fstride*m];
+   yb = st->twiddles[fstride*2*m];
+#endif
    tw=st->twiddles;
 
    for (i=0;i<N;i++)
@@ -335,8 +266,8 @@ static void kf_bfly5(
       Fout3=Fout0+3*m;
       Fout4=Fout0+4*m;
 
+      /* For non-custom modes, m is guaranteed to be a multiple of 4. */
       for ( u=0; u<m; ++u ) {
-         C_FIXDIV( *Fout0,5); C_FIXDIV( *Fout1,5); C_FIXDIV( *Fout2,5); C_FIXDIV( *Fout3,5); C_FIXDIV( *Fout4,5);
          scratch[0] = *Fout0;
 
          C_MUL(scratch[1] ,*Fout1, tw[u*fstride]);
@@ -373,74 +304,8 @@ static void kf_bfly5(
       }
    }
 }
+#endif /* OVERRIDE_kf_bfly5 */
 
-static void ki_bfly5(
-                     kiss_fft_cpx * Fout,
-                     const size_t fstride,
-                     const kiss_fft_state *st,
-                     int m,
-                     int N,
-                     int mm
-                    )
-{
-   kiss_fft_cpx *Fout0,*Fout1,*Fout2,*Fout3,*Fout4;
-   int i, u;
-   kiss_fft_cpx scratch[13];
-   const kiss_twiddle_cpx * twiddles = st->twiddles;
-   const kiss_twiddle_cpx *tw;
-   kiss_twiddle_cpx ya,yb;
-   kiss_fft_cpx * Fout_beg = Fout;
-
-   ya = twiddles[fstride*m];
-   yb = twiddles[fstride*2*m];
-   tw=st->twiddles;
-
-   for (i=0;i<N;i++)
-   {
-      Fout = Fout_beg + i*mm;
-      Fout0=Fout;
-      Fout1=Fout0+m;
-      Fout2=Fout0+2*m;
-      Fout3=Fout0+3*m;
-      Fout4=Fout0+4*m;
-
-      for ( u=0; u<m; ++u ) {
-         scratch[0] = *Fout0;
-
-         C_MULC(scratch[1] ,*Fout1, tw[u*fstride]);
-         C_MULC(scratch[2] ,*Fout2, tw[2*u*fstride]);
-         C_MULC(scratch[3] ,*Fout3, tw[3*u*fstride]);
-         C_MULC(scratch[4] ,*Fout4, tw[4*u*fstride]);
-
-         C_ADD( scratch[7],scratch[1],scratch[4]);
-         C_SUB( scratch[10],scratch[1],scratch[4]);
-         C_ADD( scratch[8],scratch[2],scratch[3]);
-         C_SUB( scratch[9],scratch[2],scratch[3]);
-
-         Fout0->r += scratch[7].r + scratch[8].r;
-         Fout0->i += scratch[7].i + scratch[8].i;
-
-         scratch[5].r = scratch[0].r + S_MUL(scratch[7].r,ya.r) + S_MUL(scratch[8].r,yb.r);
-         scratch[5].i = scratch[0].i + S_MUL(scratch[7].i,ya.r) + S_MUL(scratch[8].i,yb.r);
-
-         scratch[6].r = -S_MUL(scratch[10].i,ya.i) - S_MUL(scratch[9].i,yb.i);
-         scratch[6].i =  S_MUL(scratch[10].r,ya.i) + S_MUL(scratch[9].r,yb.i);
-
-         C_SUB(*Fout1,scratch[5],scratch[6]);
-         C_ADD(*Fout4,scratch[5],scratch[6]);
-
-         scratch[11].r = scratch[0].r + S_MUL(scratch[7].r,yb.r) + S_MUL(scratch[8].r,ya.r);
-         scratch[11].i = scratch[0].i + S_MUL(scratch[7].i,yb.r) + S_MUL(scratch[8].i,ya.r);
-         scratch[12].r =  S_MUL(scratch[10].i,yb.i) - S_MUL(scratch[9].i,ya.i);
-         scratch[12].i = -S_MUL(scratch[10].r,yb.i) + S_MUL(scratch[9].r,ya.i);
-
-         C_ADD(*Fout2,scratch[11],scratch[12]);
-         C_SUB(*Fout3,scratch[11],scratch[12]);
-
-         ++Fout0;++Fout1;++Fout2;++Fout3;++Fout4;
-      }
-   }
-}
 
 #endif
 
@@ -488,6 +353,9 @@ static
 int kf_factor(int n,opus_int16 * facbuf)
 {
     int p=4;
+    int i;
+    int stages=0;
+    int nbak = n;
 
     /*factor out powers of 4, powers of 2, then any remaining primes */
     do {
@@ -509,9 +377,30 @@ int kf_factor(int n,opus_int16 * facbuf)
         {
            return 0;
         }
-        *facbuf++ = p;
-        *facbuf++ = n;
+        facbuf[2*stages] = p;
+        if (p==2 && stages > 1)
+        {
+           facbuf[2*stages] = 4;
+           facbuf[2] = 2;
+        }
+        stages++;
     } while (n > 1);
+    n = nbak;
+    /* Reverse the order to get the radix 4 at the end, so we can use the
+       fast degenerate case. It turns out that reversing the order also
+       improves the noise behaviour. */
+    for (i=0;i<stages/2;i++)
+    {
+       int tmp;
+       tmp = facbuf[2*i];
+       facbuf[2*i] = facbuf[2*(stages-i-1)];
+       facbuf[2*(stages-i-1)] = tmp;
+    }
+    for (i=0;i<stages;i++)
+    {
+        n /= facbuf[2*i];
+        facbuf[2*i+1] = n;
+    }
     return 1;
 }
 
@@ -532,13 +421,19 @@ static void compute_twiddles(kiss_twiddle_cpx *twiddles, int nfft)
 #endif
 }
 
+int opus_fft_alloc_arch_c(kiss_fft_state *st) {
+   (void)st;
+   return 0;
+}
+
 /*
  *
  * Allocates all necessary storage space for the fft and ifft.
  * The return value is a contiguous block of memory.  As such,
  * It can be freed with free().
  * */
-kiss_fft_state *opus_fft_alloc_twiddles(int nfft,void * mem,size_t * lenmem,  const kiss_fft_state *base)
+kiss_fft_state *opus_fft_alloc_twiddles(int nfft,void * mem,size_t * lenmem,
+                                        const kiss_fft_state *base, int arch)
 {
     kiss_fft_state *st=NULL;
     size_t memneeded = sizeof(struct kiss_fft_state); /* twiddle factors*/
@@ -555,14 +450,20 @@ kiss_fft_state *opus_fft_alloc_twiddles(int nfft,void * mem,size_t * lenmem,  co
         kiss_twiddle_cpx *twiddles;
 
         st->nfft=nfft;
-#ifndef OPUS_FIXED_POINT
+#ifdef OPUS_FIXED_POINT
+        st->scale_shift = celt_ilog2(st->nfft);
+        if (st->nfft == 1<<st->scale_shift)
+           st->scale = Q15ONE;
+        else
+           st->scale = (1073741824+st->nfft/2)/st->nfft>>(15-st->scale_shift);
+#else
         st->scale = 1.f/nfft;
 #endif
         if (base != NULL)
         {
            st->twiddles = base->twiddles;
            st->shift = 0;
-           while (nfft<<st->shift != base->nfft && st->shift < 32)
+           while (st->shift < 32 && nfft<<st->shift != base->nfft)
               st->shift++;
            if (st->shift>=32)
               goto fail;
@@ -581,22 +482,31 @@ kiss_fft_state *opus_fft_alloc_twiddles(int nfft,void * mem,size_t * lenmem,  co
         if (st->bitrev==NULL)
             goto fail;
         compute_bitrev_table(0, bitrev, 1,1, st->factors,st);
+
+        /* Initialize architecture specific fft parameters */
+        if (opus_fft_alloc_arch(st, arch))
+            goto fail;
     }
     return st;
 fail:
-    opus_fft_free(st);
+    opus_fft_free(st, arch);
     return NULL;
 }
 
-kiss_fft_state *opus_fft_alloc(int nfft,void * mem,size_t * lenmem )
+kiss_fft_state *opus_fft_alloc(int nfft,void * mem,size_t * lenmem, int arch)
 {
-   return opus_fft_alloc_twiddles(nfft, mem, lenmem, NULL);
+   return opus_fft_alloc_twiddles(nfft, mem, lenmem, NULL, arch);
+}
+
+void opus_fft_free_arch_c(kiss_fft_state *st) {
+   (void)st;
 }
 
-void opus_fft_free(const kiss_fft_state *cfg)
+void opus_fft_free(const kiss_fft_state *cfg, int arch)
 {
    if (cfg)
    {
+      opus_fft_free_arch((kiss_fft_state *)cfg, arch);
       opus_free((opus_int16*)cfg->bitrev);
       if (cfg->shift < 0)
          opus_free((kiss_twiddle_cpx*)cfg->twiddles);
@@ -606,7 +516,7 @@ void opus_fft_free(const kiss_fft_state *cfg)
 
 #endif /* CUSTOM_MODES */
 
-void opus_fft(const kiss_fft_state *st,const kiss_fft_cpx *fin,kiss_fft_cpx *fout)
+void opus_fft_impl(const kiss_fft_state *st,kiss_fft_cpx *fout)
 {
     int m2, m;
     int p;
@@ -618,17 +528,6 @@ void opus_fft(const kiss_fft_state *st,const kiss_fft_cpx *fin,kiss_fft_cpx *fou
     /* st->shift can be -1 */
     shift = st->shift>0 ? st->shift : 0;
 
-    celt_assert2 (fin != fout, "In-place FFT not supported");
-    /* Bit-reverse the input */
-    for (i=0;i<st->nfft;i++)
-    {
-       fout[st->bitrev[i]] = fin[i];
-#ifndef OPUS_FIXED_POINT
-       fout[st->bitrev[i]].r *= st->scale;
-       fout[st->bitrev[i]].i *= st->scale;
-#endif
-    }
-
     fstride[0] = 1;
     L=0;
     do {
@@ -647,7 +546,7 @@ void opus_fft(const kiss_fft_state *st,const kiss_fft_cpx *fin,kiss_fft_cpx *fou
        switch (st->factors[2*i])
        {
        case 2:
-          kf_bfly2(fout,fstride[i]<<shift,st,m, fstride[i], m2);
+          kf_bfly2(fout, m, fstride[i]);
           break;
        case 4:
           kf_bfly4(fout,fstride[i]<<shift,st,m, fstride[i], m2);
@@ -665,55 +564,39 @@ void opus_fft(const kiss_fft_state *st,const kiss_fft_cpx *fin,kiss_fft_cpx *fou
     }
 }
 
-void opus_ifft(const kiss_fft_state *st,const kiss_fft_cpx *fin,kiss_fft_cpx *fout)
+void opus_fft_c(const kiss_fft_state *st,const kiss_fft_cpx *fin,kiss_fft_cpx *fout)
 {
-   int m2, m;
-   int p;
-   int L;
-   int fstride[MAXFACTORS];
    int i;
-   int shift;
+   opus_val16 scale;
+#ifdef OPUS_FIXED_POINT
+   /* Allows us to scale with MULT16_32_Q16(), which is faster than
+      MULT16_32_Q15() on ARM. */
+   int scale_shift = st->scale_shift-1;
+#endif
+   scale = st->scale;
 
-   /* st->shift can be -1 */
-   shift = st->shift>0 ? st->shift : 0;
    celt_assert2 (fin != fout, "In-place FFT not supported");
    /* Bit-reverse the input */
    for (i=0;i<st->nfft;i++)
-      fout[st->bitrev[i]] = fin[i];
-
-   fstride[0] = 1;
-   L=0;
-   do {
-      p = st->factors[2*L];
-      m = st->factors[2*L+1];
-      fstride[L+1] = fstride[L]*p;
-      L++;
-   } while(m!=1);
-   m = st->factors[2*L-1];
-   for (i=L-1;i>=0;i--)
    {
-      if (i!=0)
-         m2 = st->factors[2*i-1];
-      else
-         m2 = 1;
-      switch (st->factors[2*i])
-      {
-      case 2:
-         ki_bfly2(fout,fstride[i]<<shift,st,m, fstride[i], m2);
-         break;
-      case 4:
-         ki_bfly4(fout,fstride[i]<<shift,st,m, fstride[i], m2);
-         break;
-#ifndef RADIX_TWO_ONLY
-      case 3:
-         ki_bfly3(fout,fstride[i]<<shift,st,m, fstride[i], m2);
-         break;
-      case 5:
-         ki_bfly5(fout,fstride[i]<<shift,st,m, fstride[i], m2);
-         break;
-#endif
-      }
-      m = m2;
+      kiss_fft_cpx x = fin[i];
+      fout[st->bitrev[i]].r = SHR32(MULT16_32_Q16(scale, x.r), scale_shift);
+      fout[st->bitrev[i]].i = SHR32(MULT16_32_Q16(scale, x.i), scale_shift);
    }
+   opus_fft_impl(st, fout);
 }
 
+
+void opus_ifft_c(const kiss_fft_state *st,const kiss_fft_cpx *fin,kiss_fft_cpx *fout)
+{
+   int i;
+   celt_assert2 (fin != fout, "In-place FFT not supported");
+   /* Bit-reverse the input */
+   for (i=0;i<st->nfft;i++)
+      fout[st->bitrev[i]] = fin[i];
+   for (i=0;i<st->nfft;i++)
+      fout[i].i = -fout[i].i;
+   opus_fft_impl(st, fout);
+   for (i=0;i<st->nfft;i++)
+      fout[i].i = -fout[i].i;
+}
diff --git a/drivers/opus/celt/kiss_fft.h b/drivers/opus/celt/kiss_fft.h
index db2532c692..6229044c43 100644
--- a/drivers/opus/celt/kiss_fft.h
+++ b/drivers/opus/celt/kiss_fft.h
@@ -32,6 +32,7 @@
 #include <stdlib.h>
 #include <math.h>
 #include "opus/celt/arch.h"
+#include "opus/celt/cpu_support.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -77,17 +78,28 @@ typedef struct {
  4*4*4*2
  */
 
+typedef struct arch_fft_state{
+   int is_supported;
+   void *priv;
+} arch_fft_state;
+
 typedef struct kiss_fft_state{
     int nfft;
-#ifndef OPUS_FIXED_POINT
-    kiss_fft_scalar scale;
+    opus_val16 scale;
+#ifdef OPUS_FIXED_POINT
+    int scale_shift;
 #endif
     int shift;
     opus_int16 factors[2*MAXFACTORS];
     const opus_int16 *bitrev;
     const kiss_twiddle_cpx *twiddles;
+    arch_fft_state *arch_fft;
 } kiss_fft_state;
 
+#if defined(HAVE_ARM_NE10)
+#include "opus/celt/arm/fft_arm.h"
+#endif
+
 /*typedef struct kiss_fft_state* kiss_fft_cfg;*/
 
 /**
@@ -113,9 +125,9 @@ typedef struct kiss_fft_state{
  *      buffer size in *lenmem.
  * */
 
-kiss_fft_state *opus_fft_alloc_twiddles(int nfft,void * mem,size_t * lenmem, const kiss_fft_state *base);
+kiss_fft_state *opus_fft_alloc_twiddles(int nfft,void * mem,size_t * lenmem, const kiss_fft_state *base, int arch);
 
-kiss_fft_state *opus_fft_alloc(int nfft,void * mem,size_t * lenmem);
+kiss_fft_state *opus_fft_alloc(int nfft,void * mem,size_t * lenmem, int arch);
 
 /**
  * opus_fft(cfg,in_out_buf)
@@ -127,10 +139,59 @@ kiss_fft_state *opus_fft_alloc(int nfft,void * mem,size_t * lenmem);
  * Note that each element is complex and can be accessed like
     f[k].r and f[k].i
  * */
-void opus_fft(const kiss_fft_state *cfg,const kiss_fft_cpx *fin,kiss_fft_cpx *fout);
-void opus_ifft(const kiss_fft_state *cfg,const kiss_fft_cpx *fin,kiss_fft_cpx *fout);
+void opus_fft_c(const kiss_fft_state *cfg,const kiss_fft_cpx *fin,kiss_fft_cpx *fout);
+void opus_ifft_c(const kiss_fft_state *cfg,const kiss_fft_cpx *fin,kiss_fft_cpx *fout);
+
+void opus_fft_impl(const kiss_fft_state *st,kiss_fft_cpx *fout);
+void opus_ifft_impl(const kiss_fft_state *st,kiss_fft_cpx *fout);
+
+void opus_fft_free(const kiss_fft_state *cfg, int arch);
+
+
+void opus_fft_free_arch_c(kiss_fft_state *st);
+int opus_fft_alloc_arch_c(kiss_fft_state *st);
+
+#if !defined(OVERRIDE_OPUS_FFT)
+/* Is run-time CPU detection enabled on this platform? */
+#if defined(OPUS_HAVE_RTCD) && (defined(HAVE_ARM_NE10))
+
+extern int (*const OPUS_FFT_ALLOC_ARCH_IMPL[OPUS_ARCHMASK+1])(
+ kiss_fft_state *st);
+
+#define opus_fft_alloc_arch(_st, arch) \
+         ((*OPUS_FFT_ALLOC_ARCH_IMPL[(arch)&OPUS_ARCHMASK])(_st))
+
+extern void (*const OPUS_FFT_FREE_ARCH_IMPL[OPUS_ARCHMASK+1])(
+ kiss_fft_state *st);
+#define opus_fft_free_arch(_st, arch) \
+         ((*OPUS_FFT_FREE_ARCH_IMPL[(arch)&OPUS_ARCHMASK])(_st))
+
+extern void (*const OPUS_FFT[OPUS_ARCHMASK+1])(const kiss_fft_state *cfg,
+ const kiss_fft_cpx *fin, kiss_fft_cpx *fout);
+#define opus_fft(_cfg, _fin, _fout, arch) \
+   ((*OPUS_FFT[(arch)&OPUS_ARCHMASK])(_cfg, _fin, _fout))
+
+extern void (*const OPUS_IFFT[OPUS_ARCHMASK+1])(const kiss_fft_state *cfg,
+ const kiss_fft_cpx *fin, kiss_fft_cpx *fout);
+#define opus_ifft(_cfg, _fin, _fout, arch) \
+   ((*OPUS_IFFT[(arch)&OPUS_ARCHMASK])(_cfg, _fin, _fout))
+
+#else /* else for if defined(OPUS_HAVE_RTCD) && (defined(HAVE_ARM_NE10)) */
+
+#define opus_fft_alloc_arch(_st, arch) \
+         ((void)(arch), opus_fft_alloc_arch_c(_st))
+
+#define opus_fft_free_arch(_st, arch) \
+         ((void)(arch), opus_fft_free_arch_c(_st))
+
+#define opus_fft(_cfg, _fin, _fout, arch) \
+         ((void)(arch), opus_fft_c(_cfg, _fin, _fout))
+
+#define opus_ifft(_cfg, _fin, _fout, arch) \
+         ((void)(arch), opus_ifft_c(_cfg, _fin, _fout))
 
-void opus_fft_free(const kiss_fft_state *cfg);
+#endif /* end if defined(OPUS_HAVE_RTCD) && (defined(HAVE_ARM_NE10)) */
+#endif /* end if !defined(OVERRIDE_OPUS_FFT) */
 
 #ifdef __cplusplus
 }
diff --git a/drivers/opus/celt/laplace.c b/drivers/opus/celt/laplace.c
index 9dc4d94d24..5bc956af45 100644
--- a/drivers/opus/celt/laplace.c
+++ b/drivers/opus/celt/laplace.c
@@ -25,10 +25,7 @@
    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
-
-#ifdef OPUS_ENABLED
 #include "opus/opus_config.h"
-#endif
 
 #include "opus/celt/laplace.h"
 #include "opus/celt/mathops.h"
diff --git a/drivers/opus/celt/mathops.c b/drivers/opus/celt/mathops.c
index 88e5aea129..3f7c1a8bdd 100644
--- a/drivers/opus/celt/mathops.c
+++ b/drivers/opus/celt/mathops.c
@@ -30,10 +30,7 @@
    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
-
-#ifdef OPUS_ENABLED
 #include "opus/opus_config.h"
-#endif
 
 #include "opus/celt/mathops.h"
 
diff --git a/drivers/opus/celt/mathops.h b/drivers/opus/celt/mathops.h
index 759f58d1b7..a008d71e18 100644
--- a/drivers/opus/celt/mathops.h
+++ b/drivers/opus/celt/mathops.h
@@ -254,5 +254,5 @@ static OPUS_INLINE opus_val16 celt_atan2p(opus_val16 y, opus_val16 x)
    }
 }
 
-#endif /* OPUS_FIXED_POINT */
+#endif /* FIXED_POINT */
 #endif /* MATHOPS_H */
diff --git a/drivers/opus/celt/mdct.c b/drivers/opus/celt/mdct.c
index ae34538d6c..641a20e49a 100644
--- a/drivers/opus/celt/mdct.c
+++ b/drivers/opus/celt/mdct.c
@@ -1,4 +1,4 @@
-	/* Copyright (c) 2007-2008 CSIRO
+/* Copyright (c) 2007-2008 CSIRO
    Copyright (c) 2007-2008 Xiph.Org Foundation
    Written by Jean-Marc Valin */
 /*
@@ -40,10 +40,8 @@
 */
 
 #ifndef SKIP_CONFIG_H
-#ifdef OPUS_ENABLED
 #include "opus/opus_config.h"
 #endif
-#endif
 
 #include "opus/celt/mdct.h"
 #include "opus/celt/kiss_fft.h"
@@ -53,76 +51,100 @@
 #include "opus/celt/mathops.h"
 #include "opus/celt/stack_alloc.h"
 
+#if defined(MIPSr1_ASM)
+#include "opus/celt/mips/mdct_mipsr1.h"
+#endif
+
+
 #ifdef CUSTOM_MODES
 
-int clt_mdct_init(celt_mdct_lookup *l,int N, int maxshift)
+int clt_mdct_init(mdct_lookup *l,int N, int maxshift, int arch)
 {
    int i;
-   int N4;
    kiss_twiddle_scalar *trig;
-#if defined(OPUS_FIXED_POINT)
+   int shift;
    int N2=N>>1;
-#endif
    l->n = N;
-   N4 = N>>2;
    l->maxshift = maxshift;
    for (i=0;i<=maxshift;i++)
    {
       if (i==0)
-         l->kfft[i] = opus_fft_alloc(N>>2>>i, 0, 0);
+         l->kfft[i] = opus_fft_alloc(N>>2>>i, 0, 0, arch);
       else
-         l->kfft[i] = opus_fft_alloc_twiddles(N>>2>>i, 0, 0, l->kfft[0]);
+         l->kfft[i] = opus_fft_alloc_twiddles(N>>2>>i, 0, 0, l->kfft[0], arch);
 #ifndef ENABLE_TI_DSPLIB55
       if (l->kfft[i]==NULL)
          return 0;
 #endif
    }
-   l->trig = trig = (kiss_twiddle_scalar*)opus_alloc((N4+1)*sizeof(kiss_twiddle_scalar));
+   l->trig = trig = (kiss_twiddle_scalar*)opus_alloc((N-(N2>>maxshift))*sizeof(kiss_twiddle_scalar));
    if (l->trig==NULL)
      return 0;
-   /* We have enough points that sine isn't necessary */
-#if defined(OPUS_FIXED_POINT)
-   for (i=0;i<=N4;i++)
-      trig[i] = TRIG_UPSCALE*celt_cos_norm(DIV32(ADD32(SHL32(EXTEND32(i),17),N2),N));
+   for (shift=0;shift<=maxshift;shift++)
+   {
+      /* We have enough points that sine isn't necessary */
+#if defined(FIXED_POINT)
+#if 1
+      for (i=0;i<N2;i++)
+         trig[i] = TRIG_UPSCALE*celt_cos_norm(DIV32(ADD32(SHL32(EXTEND32(i),17),N2+16384),N));
 #else
-   for (i=0;i<=N4;i++)
-      trig[i] = (kiss_twiddle_scalar)cos(2*PI*i/N);
+      for (i=0;i<N2;i++)
+         trig[i] = (kiss_twiddle_scalar)MAX32(-32767,MIN32(32767,floor(.5+32768*cos(2*M_PI*(i+.125)/N))));
 #endif
+#else
+      for (i=0;i<N2;i++)
+         trig[i] = (kiss_twiddle_scalar)cos(2*PI*(i+.125)/N);
+#endif
+      trig += N2;
+      N2 >>= 1;
+      N >>= 1;
+   }
    return 1;
 }
 
-void clt_mdct_clear(celt_mdct_lookup *l)
+void clt_mdct_clear(mdct_lookup *l, int arch)
 {
    int i;
    for (i=0;i<=l->maxshift;i++)
-      opus_fft_free(l->kfft[i]);
+      opus_fft_free(l->kfft[i], arch);
    opus_free((kiss_twiddle_scalar*)l->trig);
 }
 
 #endif /* CUSTOM_MODES */
 
 /* Forward MDCT trashes the input array */
-void clt_mdct_forward(const celt_mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out,
-      const opus_val16 *window, int overlap, int shift, int stride)
+#ifndef OVERRIDE_clt_mdct_forward
+void clt_mdct_forward_c(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out,
+      const opus_val16 *window, int overlap, int shift, int stride, int arch)
 {
    int i;
    int N, N2, N4;
-   kiss_twiddle_scalar sine;
    VARDECL(kiss_fft_scalar, f);
-   VARDECL(kiss_fft_scalar, f2);
+   VARDECL(kiss_fft_cpx, f2);
+   const kiss_fft_state *st = l->kfft[shift];
+   const kiss_twiddle_scalar *trig;
+   opus_val16 scale;
+#ifdef OPUS_FIXED_POINT
+   /* Allows us to scale with MULT16_32_Q16(), which is faster than
+      MULT16_32_Q15() on ARM. */
+   int scale_shift = st->scale_shift-1;
+#endif
    SAVE_STACK;
+   (void)arch;
+   scale = st->scale;
+
    N = l->n;
-   N >>= shift;
+   trig = l->trig;
+   for (i=0;i<shift;i++)
+   {
+      N >>= 1;
+      trig += N;
+   }
    N2 = N>>1;
    N4 = N>>2;
+
    ALLOC(f, N2, kiss_fft_scalar);
-   ALLOC(f2, N2, kiss_fft_scalar);
-   /* sin(x) ~= x here */
-#ifdef OPUS_FIXED_POINT
-   sine = TRIG_UPSCALE*(QCONST16(0.7853981f, 15)+N2)/N;
-#else
-   sine = (kiss_twiddle_scalar)2*PI*(.125f)/N;
-#endif
+   ALLOC(f2, N4, kiss_fft_cpx);
 
    /* Consider the input to be composed of four blocks: [a, b, c, d] */
    /* Window, shuffle, fold */
@@ -167,123 +189,131 @@ void clt_mdct_forward(const celt_mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_s
    /* Pre-rotation */
    {
       kiss_fft_scalar * OPUS_RESTRICT yp = f;
-      const kiss_twiddle_scalar *t = &l->trig[0];
+      const kiss_twiddle_scalar *t = &trig[0];
       for(i=0;i<N4;i++)
       {
+         kiss_fft_cpx yc;
+         kiss_twiddle_scalar t0, t1;
          kiss_fft_scalar re, im, yr, yi;
-         re = yp[0];
-         im = yp[1];
-         yr = -S_MUL(re,t[i<<shift])  -  S_MUL(im,t[(N4-i)<<shift]);
-         yi = -S_MUL(im,t[i<<shift])  +  S_MUL(re,t[(N4-i)<<shift]);
-         /* works because the cos is nearly one */
-         *yp++ = yr + S_MUL(yi,sine);
-         *yp++ = yi - S_MUL(yr,sine);
+         t0 = t[i];
+         t1 = t[N4+i];
+         re = *yp++;
+         im = *yp++;
+         yr = S_MUL(re,t0)  -  S_MUL(im,t1);
+         yi = S_MUL(im,t0)  +  S_MUL(re,t1);
+         yc.r = yr;
+         yc.i = yi;
+         yc.r = PSHR32(MULT16_32_Q16(scale, yc.r), scale_shift);
+         yc.i = PSHR32(MULT16_32_Q16(scale, yc.i), scale_shift);
+         f2[st->bitrev[i]] = yc;
       }
    }
 
-   /* N/4 complex FFT, down-scales by 4/N */
-   opus_fft(l->kfft[shift], (kiss_fft_cpx *)f, (kiss_fft_cpx *)f2);
+   /* N/4 complex FFT, does not downscale anymore */
+   opus_fft_impl(st, f2);
 
    /* Post-rotate */
    {
       /* Temp pointers to make it really clear to the compiler what we're doing */
-      const kiss_fft_scalar * OPUS_RESTRICT fp = f2;
+      const kiss_fft_cpx * OPUS_RESTRICT fp = f2;
       kiss_fft_scalar * OPUS_RESTRICT yp1 = out;
       kiss_fft_scalar * OPUS_RESTRICT yp2 = out+stride*(N2-1);
-      const kiss_twiddle_scalar *t = &l->trig[0];
+      const kiss_twiddle_scalar *t = &trig[0];
       /* Temp pointers to make it really clear to the compiler what we're doing */
       for(i=0;i<N4;i++)
       {
          kiss_fft_scalar yr, yi;
-         yr = S_MUL(fp[1],t[(N4-i)<<shift]) + S_MUL(fp[0],t[i<<shift]);
-         yi = S_MUL(fp[0],t[(N4-i)<<shift]) - S_MUL(fp[1],t[i<<shift]);
-         /* works because the cos is nearly one */
-         *yp1 = yr - S_MUL(yi,sine);
-         *yp2 = yi + S_MUL(yr,sine);;
-         fp += 2;
+         yr = S_MUL(fp->i,t[N4+i]) - S_MUL(fp->r,t[i]);
+         yi = S_MUL(fp->r,t[N4+i]) + S_MUL(fp->i,t[i]);
+         *yp1 = yr;
+         *yp2 = yi;
+         fp++;
          yp1 += 2*stride;
          yp2 -= 2*stride;
       }
    }
    RESTORE_STACK;
 }
+#endif /* OVERRIDE_clt_mdct_forward */
 
-void clt_mdct_backward(const celt_mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out,
-      const opus_val16 * OPUS_RESTRICT window, int overlap, int shift, int stride)
+#ifndef OVERRIDE_clt_mdct_backward
+void clt_mdct_backward_c(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out,
+      const opus_val16 * OPUS_RESTRICT window, int overlap, int shift, int stride, int arch)
 {
    int i;
    int N, N2, N4;
-   kiss_twiddle_scalar sine;
-   VARDECL(kiss_fft_scalar, f2);
-   SAVE_STACK;
+   const kiss_twiddle_scalar *trig;
+   (void) arch;
+
    N = l->n;
-   N >>= shift;
+   trig = l->trig;
+   for (i=0;i<shift;i++)
+   {
+      N >>= 1;
+      trig += N;
+   }
    N2 = N>>1;
    N4 = N>>2;
-   ALLOC(f2, N2, kiss_fft_scalar);
-   /* sin(x) ~= x here */
-#ifdef OPUS_FIXED_POINT
-   sine = TRIG_UPSCALE*(QCONST16(0.7853981f, 15)+N2)/N;
-#else
-   sine = (kiss_twiddle_scalar)2*PI*(.125f)/N;
-#endif
 
    /* Pre-rotate */
    {
       /* Temp pointers to make it really clear to the compiler what we're doing */
       const kiss_fft_scalar * OPUS_RESTRICT xp1 = in;
       const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+stride*(N2-1);
-      kiss_fft_scalar * OPUS_RESTRICT yp = f2;
-      const kiss_twiddle_scalar *t = &l->trig[0];
+      kiss_fft_scalar * OPUS_RESTRICT yp = out+(overlap>>1);
+      const kiss_twiddle_scalar * OPUS_RESTRICT t = &trig[0];
+      const opus_int16 * OPUS_RESTRICT bitrev = l->kfft[shift]->bitrev;
       for(i=0;i<N4;i++)
       {
+         int rev;
          kiss_fft_scalar yr, yi;
-         yr = -S_MUL(*xp2, t[i<<shift]) + S_MUL(*xp1,t[(N4-i)<<shift]);
-         yi =  -S_MUL(*xp2, t[(N4-i)<<shift]) - S_MUL(*xp1,t[i<<shift]);
-         /* works because the cos is nearly one */
-         *yp++ = yr - S_MUL(yi,sine);
-         *yp++ = yi + S_MUL(yr,sine);
+         rev = *bitrev++;
+         yr = S_MUL(*xp2, t[i]) + S_MUL(*xp1, t[N4+i]);
+         yi = S_MUL(*xp1, t[i]) - S_MUL(*xp2, t[N4+i]);
+         /* We swap real and imag because we use an FFT instead of an IFFT. */
+         yp[2*rev+1] = yr;
+         yp[2*rev] = yi;
+         /* Storing the pre-rotation directly in the bitrev order. */
          xp1+=2*stride;
          xp2-=2*stride;
       }
    }
 
-   /* Inverse N/4 complex FFT. This one should *not* downscale even in fixed-point */
-   opus_ifft(l->kfft[shift], (kiss_fft_cpx *)f2, (kiss_fft_cpx *)(out+(overlap>>1)));
+   opus_fft_impl(l->kfft[shift], (kiss_fft_cpx*)(out+(overlap>>1)));
 
    /* Post-rotate and de-shuffle from both ends of the buffer at once to make
       it in-place. */
    {
-      kiss_fft_scalar * OPUS_RESTRICT yp0 = out+(overlap>>1);
-      kiss_fft_scalar * OPUS_RESTRICT yp1 = out+(overlap>>1)+N2-2;
-      const kiss_twiddle_scalar *t = &l->trig[0];
+      kiss_fft_scalar * yp0 = out+(overlap>>1);
+      kiss_fft_scalar * yp1 = out+(overlap>>1)+N2-2;
+      const kiss_twiddle_scalar *t = &trig[0];
       /* Loop to (N4+1)>>1 to handle odd N4. When N4 is odd, the
          middle pair will be computed twice. */
       for(i=0;i<(N4+1)>>1;i++)
       {
          kiss_fft_scalar re, im, yr, yi;
          kiss_twiddle_scalar t0, t1;
-         re = yp0[0];
-         im = yp0[1];
-         t0 = t[i<<shift];
-         t1 = t[(N4-i)<<shift];
+         /* We swap real and imag because we're using an FFT instead of an IFFT. */
+         re = yp0[1];
+         im = yp0[0];
+         t0 = t[i];
+         t1 = t[N4+i];
          /* We'd scale up by 2 here, but instead it's done when mixing the windows */
-         yr = S_MUL(re,t0) - S_MUL(im,t1);
-         yi = S_MUL(im,t0) + S_MUL(re,t1);
-         re = yp1[0];
-         im = yp1[1];
-         /* works because the cos is nearly one */
-         yp0[0] = -(yr - S_MUL(yi,sine));
-         yp1[1] = yi + S_MUL(yr,sine);
+         yr = S_MUL(re,t0) + S_MUL(im,t1);
+         yi = S_MUL(re,t1) - S_MUL(im,t0);
+         /* We swap real and imag because we're using an FFT instead of an IFFT. */
+         re = yp1[1];
+         im = yp1[0];
+         yp0[0] = yr;
+         yp1[1] = yi;
 
-         t0 = t[(N4-i-1)<<shift];
-         t1 = t[(i+1)<<shift];
+         t0 = t[(N4-i-1)];
+         t1 = t[(N2-i-1)];
          /* We'd scale up by 2 here, but instead it's done when mixing the windows */
-         yr = S_MUL(re,t0) - S_MUL(im,t1);
-         yi = S_MUL(im,t0) + S_MUL(re,t1);
-         /* works because the cos is nearly one */
-         yp1[0] = -(yr - S_MUL(yi,sine));
-         yp0[1] = yi + S_MUL(yr,sine);
+         yr = S_MUL(re,t0) + S_MUL(im,t1);
+         yi = S_MUL(re,t1) - S_MUL(im,t0);
+         yp1[0] = yr;
+         yp0[1] = yi;
          yp0 += 2;
          yp1 -= 2;
       }
@@ -307,5 +337,5 @@ void clt_mdct_backward(const celt_mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_
          wp2--;
       }
    }
-   RESTORE_STACK;
 }
+#endif /* OVERRIDE_clt_mdct_backward */
diff --git a/drivers/opus/celt/mdct.h b/drivers/opus/celt/mdct.h
index 492afeae1e..2ec98a92e5 100644
--- a/drivers/opus/celt/mdct.h
+++ b/drivers/opus/celt/mdct.h
@@ -51,20 +51,62 @@ typedef struct {
    int maxshift;
    const kiss_fft_state *kfft[4];
    const kiss_twiddle_scalar * OPUS_RESTRICT trig;
-} celt_mdct_lookup;
+} mdct_lookup;
+
+#if defined(HAVE_ARM_NE10)
+#include "opus/celt/arm/mdct_arm.h"
+#endif
 
-int clt_mdct_init(celt_mdct_lookup *l,int N, int maxshift);
-void clt_mdct_clear(celt_mdct_lookup *l);
+
+int clt_mdct_init(mdct_lookup *l,int N, int maxshift, int arch);
+void clt_mdct_clear(mdct_lookup *l, int arch);
 
 /** Compute a forward MDCT and scale by 4/N, trashes the input array */
-void clt_mdct_forward(const celt_mdct_lookup *l, kiss_fft_scalar *in,
-      kiss_fft_scalar * OPUS_RESTRICT out,
-      const opus_val16 *window, int overlap, int shift, int stride);
+void clt_mdct_forward_c(const mdct_lookup *l, kiss_fft_scalar *in,
+                        kiss_fft_scalar * OPUS_RESTRICT out,
+                        const opus_val16 *window, int overlap,
+                        int shift, int stride, int arch);
 
 /** Compute a backward MDCT (no scaling) and performs weighted overlap-add
     (scales implicitly by 1/2) */
-void clt_mdct_backward(const celt_mdct_lookup *l, kiss_fft_scalar *in,
+void clt_mdct_backward_c(const mdct_lookup *l, kiss_fft_scalar *in,
       kiss_fft_scalar * OPUS_RESTRICT out,
-      const opus_val16 * OPUS_RESTRICT window, int overlap, int shift, int stride);
+      const opus_val16 * OPUS_RESTRICT window,
+      int overlap, int shift, int stride, int arch);
+
+#if !defined(OVERRIDE_OPUS_MDCT)
+/* Is run-time CPU detection enabled on this platform? */
+#if defined(OPUS_HAVE_RTCD) && defined(HAVE_ARM_NE10)
+
+extern void (*const CLT_MDCT_FORWARD_IMPL[OPUS_ARCHMASK+1])(
+      const mdct_lookup *l, kiss_fft_scalar *in,
+      kiss_fft_scalar * OPUS_RESTRICT out, const opus_val16 *window,
+      int overlap, int shift, int stride, int arch);
+
+#define clt_mdct_forward(_l, _in, _out, _window, _overlap, _shift, _stride, _arch) \
+   ((*CLT_MDCT_FORWARD_IMPL[(arch)&OPUS_ARCHMASK])(_l, _in, _out, \
+                                                   _window, _overlap, _shift, \
+                                                   _stride, _arch))
+
+extern void (*const CLT_MDCT_BACKWARD_IMPL[OPUS_ARCHMASK+1])(
+      const mdct_lookup *l, kiss_fft_scalar *in,
+      kiss_fft_scalar * OPUS_RESTRICT out, const opus_val16 *window,
+      int overlap, int shift, int stride, int arch);
+
+#define clt_mdct_backward(_l, _in, _out, _window, _overlap, _shift, _stride, _arch) \
+   (*CLT_MDCT_BACKWARD_IMPL[(arch)&OPUS_ARCHMASK])(_l, _in, _out, \
+                                                   _window, _overlap, _shift, \
+                                                   _stride, _arch)
+
+#else /* if defined(OPUS_HAVE_RTCD) && defined(HAVE_ARM_NE10) */
+
+#define clt_mdct_forward(_l, _in, _out, _window, _overlap, _shift, _stride, _arch) \
+   clt_mdct_forward_c(_l, _in, _out, _window, _overlap, _shift, _stride, _arch)
+
+#define clt_mdct_backward(_l, _in, _out, _window, _overlap, _shift, _stride, _arch) \
+   clt_mdct_backward_c(_l, _in, _out, _window, _overlap, _shift, _stride, _arch)
+
+#endif /* end if defined(OPUS_HAVE_RTCD) && defined(HAVE_ARM_NE10) && !defined(FIXED_POINT) */
+#endif /* end if !defined(OVERRIDE_OPUS_MDCT) */
 
 #endif
diff --git a/drivers/opus/celt/mips/celt_mipsr1.h b/drivers/opus/celt/mips/celt_mipsr1.h
new file mode 100644
index 0000000000..906848f1ec
--- /dev/null
+++ b/drivers/opus/celt/mips/celt_mipsr1.h
@@ -0,0 +1,148 @@
+/* Copyright (c) 2007-2008 CSIRO
+   Copyright (c) 2007-2010 Xiph.Org Foundation
+   Copyright (c) 2008 Gregory Maxwell
+   Written by Jean-Marc Valin and Gregory Maxwell */
+/*
+   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.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifndef __CELT_MIPSR1_H__
+#define __CELT_MIPSR1_H__
+#include "opus/opus_config.h"
+
+#define CELT_C
+
+#include "opus/celt/os_support.h"
+#include "opus/celt/mdct.h"
+#include <math.h>
+#include "opus/celt/celt.h"
+#include "opus/celt/pitch.h"
+#include "opus/celt/bands.h"
+#include "opus/celt/modes.h"
+#include "opus/celt/entcode.h"
+#include "opus/celt/quant_bands.h"
+#include "opus/celt/rate.h"
+#include "opus/celt/stack_alloc.h"
+#include "opus/celt/mathops.h"
+#include "opus/celt/float_cast.h"
+#include <stdarg.h>
+#include "opus/celt/celt_lpc.h"
+#include "opus/celt/vq.h"
+
+#define OVERRIDE_comb_filter
+void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N,
+      opus_val16 g0, opus_val16 g1, int tapset0, int tapset1,
+      const opus_val16 *window, int overlap, int arch)
+{
+   int i;
+   opus_val32 x0, x1, x2, x3, x4;
+
+   (void)arch;
+
+   /* printf ("%d %d %f %f\n", T0, T1, g0, g1); */
+   opus_val16 g00, g01, g02, g10, g11, g12;
+   static const opus_val16 gains[3][3] = {
+         {QCONST16(0.3066406250f, 15), QCONST16(0.2170410156f, 15), QCONST16(0.1296386719f, 15)},
+         {QCONST16(0.4638671875f, 15), QCONST16(0.2680664062f, 15), QCONST16(0.f, 15)},
+         {QCONST16(0.7998046875f, 15), QCONST16(0.1000976562f, 15), QCONST16(0.f, 15)}};
+
+   if (g0==0 && g1==0)
+   {
+      /* OPT: Happens to work without the OPUS_MOVE(), but only because the current encoder already copies x to y */
+      if (x!=y)
+         OPUS_MOVE(y, x, N);
+      return;
+   }
+
+   g00 = MULT16_16_P15(g0, gains[tapset0][0]);
+   g01 = MULT16_16_P15(g0, gains[tapset0][1]);
+   g02 = MULT16_16_P15(g0, gains[tapset0][2]);
+   g10 = MULT16_16_P15(g1, gains[tapset1][0]);
+   g11 = MULT16_16_P15(g1, gains[tapset1][1]);
+   g12 = MULT16_16_P15(g1, gains[tapset1][2]);
+   x1 = x[-T1+1];
+   x2 = x[-T1  ];
+   x3 = x[-T1-1];
+   x4 = x[-T1-2];
+   /* If the filter didn't change, we don't need the overlap */
+   if (g0==g1 && T0==T1 && tapset0==tapset1)
+      overlap=0;
+
+   for (i=0;i<overlap;i++)
+   {
+      opus_val16 f;
+      opus_val32 res;
+      f = MULT16_16_Q15(window[i],window[i]);
+      x0= x[i-T1+2];
+
+      asm volatile("MULT $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15((Q15ONE-f),g00)), "r" ((int)x[i-T0]));
+
+      asm volatile("MADD $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15((Q15ONE-f),g01)), "r" ((int)ADD32(x[i-T0-1],x[i-T0+1])));
+      asm volatile("MADD $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15((Q15ONE-f),g02)), "r" ((int)ADD32(x[i-T0-2],x[i-T0+2])));
+      asm volatile("MADD $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15(f,g10)), "r" ((int)x2));
+      asm volatile("MADD $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15(f,g11)), "r" ((int)ADD32(x3,x1)));
+      asm volatile("MADD $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15(f,g12)), "r" ((int)ADD32(x4,x0)));
+
+      asm volatile("EXTR.W %0,$ac1, %1" : "=r" (res): "i" (15));
+
+      y[i] = x[i] + res;
+
+      x4=x3;
+      x3=x2;
+      x2=x1;
+      x1=x0;
+   }
+
+   x4 = x[i-T1-2];
+   x3 = x[i-T1-1];
+   x2 = x[i-T1];
+   x1 = x[i-T1+1];
+
+   if (g1==0)
+   {
+      /* OPT: Happens to work without the OPUS_MOVE(), but only because the current encoder already copies x to y */
+      if (x!=y)
+         OPUS_MOVE(y+overlap, x+overlap, N-overlap);
+      return;
+   }
+
+   for (i=overlap;i<N;i++)
+   {
+      opus_val32 res;
+      x0=x[i-T1+2];
+
+      asm volatile("MULT $ac1, %0, %1" : : "r" ((int)g10), "r" ((int)x2));
+
+      asm volatile("MADD $ac1, %0, %1" : : "r" ((int)g11), "r" ((int)ADD32(x3,x1)));
+      asm volatile("MADD $ac1, %0, %1" : : "r" ((int)g12), "r" ((int)ADD32(x4,x0)));
+      asm volatile("EXTR.W %0,$ac1, %1" : "=r" (res): "i" (15));
+      y[i] = x[i] + res;
+      x4=x3;
+      x3=x2;
+      x2=x1;
+      x1=x0;
+   }
+}
+
+#endif /* __CELT_MIPSR1_H__ */
diff --git a/drivers/opus/celt/mips/fixed_generic_mipsr1.h b/drivers/opus/celt/mips/fixed_generic_mipsr1.h
new file mode 100644
index 0000000000..4a05efbf85
--- /dev/null
+++ b/drivers/opus/celt/mips/fixed_generic_mipsr1.h
@@ -0,0 +1,126 @@
+/* Copyright (C) 2007-2009 Xiph.Org Foundation
+   Copyright (C) 2003-2008 Jean-Marc Valin
+   Copyright (C) 2007-2008 CSIRO */
+/**
+   @file fixed_generic.h
+   @brief Generic fixed-point operations
+*/
+/*
+   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.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifndef CELT_FIXED_GENERIC_MIPSR1_H
+#define CELT_FIXED_GENERIC_MIPSR1_H
+
+#undef MULT16_32_Q15_ADD
+static inline int MULT16_32_Q15_ADD(int a, int b, int c, int d) {
+    int m;
+    asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a), "r" ((int)b));
+    asm volatile("madd $ac1, %0, %1" : : "r" ((int)c), "r" ((int)d));
+    asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m): "i" (15));
+    return m;
+}
+
+#undef MULT16_32_Q15_SUB
+static inline int MULT16_32_Q15_SUB(int a, int b, int c, int d) {
+    int m;
+    asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a), "r" ((int)b));
+    asm volatile("msub $ac1, %0, %1" : : "r" ((int)c), "r" ((int)d));
+    asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m): "i" (15));
+    return m;
+}
+
+#undef MULT16_16_Q15_ADD
+static inline int MULT16_16_Q15_ADD(int a, int b, int c, int d) {
+    int m;
+    asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a), "r" ((int)b));
+    asm volatile("madd $ac1, %0, %1" : : "r" ((int)c), "r" ((int)d));
+    asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m): "i" (15));
+    return m;
+}
+
+#undef MULT16_16_Q15_SUB
+static inline int MULT16_16_Q15_SUB(int a, int b, int c, int d) {
+    int m;
+    asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a), "r" ((int)b));
+    asm volatile("msub $ac1, %0, %1" : : "r" ((int)c), "r" ((int)d));
+    asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m): "i" (15));
+    return m;
+}
+
+
+#undef MULT16_32_Q16
+static inline int MULT16_32_Q16(int a, int b)
+{
+    int c;
+    asm volatile("MULT $ac1,%0, %1" : : "r" (a), "r" (b));
+    asm volatile("EXTR.W %0,$ac1, %1" : "=r" (c): "i" (16));
+    return c;
+}
+
+#undef MULT16_32_P16
+static inline int MULT16_32_P16(int a, int b)
+{
+    int c;
+    asm volatile("MULT $ac1, %0, %1" : : "r" (a), "r" (b));
+    asm volatile("EXTR_R.W %0,$ac1, %1" : "=r" (c): "i" (16));
+    return c;
+}
+
+#undef MULT16_32_Q15
+static inline int MULT16_32_Q15(int a, int b)
+{
+    int c;
+    asm volatile("MULT $ac1, %0, %1" : : "r" (a), "r" (b));
+    asm volatile("EXTR.W %0,$ac1, %1" : "=r" (c): "i" (15));
+    return c;
+}
+
+#undef MULT32_32_Q31
+static inline int MULT32_32_Q31(int a, int b)
+{
+    int r;
+    asm volatile("MULT $ac1, %0, %1" : : "r" (a), "r" (b));
+    asm volatile("EXTR.W %0,$ac1, %1" : "=r" (r): "i" (31));
+    return r;
+}
+
+#undef PSHR32
+static inline int PSHR32(int a, int shift)
+{
+    int r;
+    asm volatile ("SHRAV_R.W %0, %1, %2" :"=r" (r): "r" (a), "r" (shift));
+    return r;
+}
+
+#undef MULT16_16_P15
+static inline int MULT16_16_P15(int a, int b)
+{
+    int r;
+    asm volatile ("mul %0, %1, %2" :"=r" (r): "r" (a), "r" (b));
+    asm volatile ("SHRA_R.W %0, %1, %2" : "+r" (r):  "0" (r), "i"(15));
+    return r;
+}
+
+#endif /* CELT_FIXED_GENERIC_MIPSR1_H */
diff --git a/drivers/opus/celt/mips/kiss_fft_mipsr1.h b/drivers/opus/celt/mips/kiss_fft_mipsr1.h
new file mode 100644
index 0000000000..09dc9af464
--- /dev/null
+++ b/drivers/opus/celt/mips/kiss_fft_mipsr1.h
@@ -0,0 +1,167 @@
+/*Copyright (c) 2013, Xiph.Org Foundation and contributors.
+
+  All rights reserved.
+
+  Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are met:
+
+    * Redistributions of source code must retain the above copyright notice,
+       this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above copyright notice,
+       this list of conditions and the following disclaimer in the
+       documentation and/or other materials provided with the distribution.
+
+  THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+  AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+  IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+  ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
+  LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+  INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+  CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+  ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+  POSSIBILITY OF SUCH DAMAGE.*/
+
+#ifndef KISS_FFT_MIPSR1_H
+#define KISS_FFT_MIPSR1_H
+
+#if !defined(KISS_FFT_GUTS_H)
+#error "This file should only be included from _kiss_fft_guts.h"
+#endif
+
+#ifdef OPUS_FIXED_POINT
+
+#define S_MUL_ADD(a, b, c, d) (S_MUL(a,b)+S_MUL(c,d))
+#define S_MUL_SUB(a, b, c, d) (S_MUL(a,b)-S_MUL(c,d))
+
+#undef S_MUL_ADD
+static inline int S_MUL_ADD(int a, int b, int c, int d) {
+    int m;
+    asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a), "r" ((int)b));
+    asm volatile("madd $ac1, %0, %1" : : "r" ((int)c), "r" ((int)d));
+    asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m): "i" (15));
+    return m;
+}
+
+#undef S_MUL_SUB
+static inline int S_MUL_SUB(int a, int b, int c, int d) {
+    int m;
+    asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a), "r" ((int)b));
+    asm volatile("msub $ac1, %0, %1" : : "r" ((int)c), "r" ((int)d));
+    asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m): "i" (15));
+    return m;
+}
+
+#undef C_MUL
+#   define C_MUL(m,a,b) (m=C_MUL_fun(a,b))
+static inline kiss_fft_cpx C_MUL_fun(kiss_fft_cpx a, kiss_twiddle_cpx b) {
+    kiss_fft_cpx m;
+
+    asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a.r), "r" ((int)b.r));
+    asm volatile("msub $ac1, %0, %1" : : "r" ((int)a.i), "r" ((int)b.i));
+    asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m.r): "i" (15));
+    asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a.r), "r" ((int)b.i));
+    asm volatile("madd $ac1, %0, %1" : : "r" ((int)a.i), "r" ((int)b.r));
+    asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m.i): "i" (15));
+
+    return m;
+}
+#undef C_MULC
+#   define C_MULC(m,a,b) (m=C_MULC_fun(a,b))
+static inline kiss_fft_cpx C_MULC_fun(kiss_fft_cpx a, kiss_twiddle_cpx b) {
+    kiss_fft_cpx m;
+
+    asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a.r), "r" ((int)b.r));
+    asm volatile("madd $ac1, %0, %1" : : "r" ((int)a.i), "r" ((int)b.i));
+    asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m.r): "i" (15));
+    asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a.i), "r" ((int)b.r));
+    asm volatile("msub $ac1, %0, %1" : : "r" ((int)a.r), "r" ((int)b.i));
+    asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m.i): "i" (15));
+
+    return m;
+}
+
+#endif /* FIXED_POINT */
+
+#define OVERRIDE_kf_bfly5
+static void kf_bfly5(
+                     kiss_fft_cpx * Fout,
+                     const size_t fstride,
+                     const kiss_fft_state *st,
+                     int m,
+                     int N,
+                     int mm
+                    )
+{
+   kiss_fft_cpx *Fout0,*Fout1,*Fout2,*Fout3,*Fout4;
+   int i, u;
+   kiss_fft_cpx scratch[13];
+
+   const kiss_twiddle_cpx *tw;
+   kiss_twiddle_cpx ya,yb;
+   kiss_fft_cpx * Fout_beg = Fout;
+
+#ifdef OPUS_FIXED_POINT
+   ya.r = 10126;
+   ya.i = -31164;
+   yb.r = -26510;
+   yb.i = -19261;
+#else
+   ya = st->twiddles[fstride*m];
+   yb = st->twiddles[fstride*2*m];
+#endif
+
+   tw=st->twiddles;
+
+   for (i=0;i<N;i++)
+   {
+      Fout = Fout_beg + i*mm;
+      Fout0=Fout;
+      Fout1=Fout0+m;
+      Fout2=Fout0+2*m;
+      Fout3=Fout0+3*m;
+      Fout4=Fout0+4*m;
+
+      /* For non-custom modes, m is guaranteed to be a multiple of 4. */
+      for ( u=0; u<m; ++u ) {
+         scratch[0] = *Fout0;
+
+
+         C_MUL(scratch[1] ,*Fout1, tw[u*fstride]);
+         C_MUL(scratch[2] ,*Fout2, tw[2*u*fstride]);
+         C_MUL(scratch[3] ,*Fout3, tw[3*u*fstride]);
+         C_MUL(scratch[4] ,*Fout4, tw[4*u*fstride]);
+
+         C_ADD( scratch[7],scratch[1],scratch[4]);
+         C_SUB( scratch[10],scratch[1],scratch[4]);
+         C_ADD( scratch[8],scratch[2],scratch[3]);
+         C_SUB( scratch[9],scratch[2],scratch[3]);
+
+         Fout0->r += scratch[7].r + scratch[8].r;
+         Fout0->i += scratch[7].i + scratch[8].i;
+         scratch[5].r = scratch[0].r + S_MUL_ADD(scratch[7].r,ya.r,scratch[8].r,yb.r);
+         scratch[5].i = scratch[0].i + S_MUL_ADD(scratch[7].i,ya.r,scratch[8].i,yb.r);
+
+         scratch[6].r =  S_MUL_ADD(scratch[10].i,ya.i,scratch[9].i,yb.i);
+         scratch[6].i =  -S_MUL_ADD(scratch[10].r,ya.i,scratch[9].r,yb.i);
+
+         C_SUB(*Fout1,scratch[5],scratch[6]);
+         C_ADD(*Fout4,scratch[5],scratch[6]);
+
+         scratch[11].r = scratch[0].r + S_MUL_ADD(scratch[7].r,yb.r,scratch[8].r,ya.r);
+         scratch[11].i = scratch[0].i + S_MUL_ADD(scratch[7].i,yb.r,scratch[8].i,ya.r);
+
+         scratch[12].r =  S_MUL_SUB(scratch[9].i,ya.i,scratch[10].i,yb.i);
+         scratch[12].i =  S_MUL_SUB(scratch[10].r,yb.i,scratch[9].r,ya.i);
+
+         C_ADD(*Fout2,scratch[11],scratch[12]);
+         C_SUB(*Fout3,scratch[11],scratch[12]);
+
+         ++Fout0;++Fout1;++Fout2;++Fout3;++Fout4;
+      }
+   }
+}
+
+
+#endif /* KISS_FFT_MIPSR1_H */
diff --git a/drivers/opus/celt/mips/mdct_mipsr1.h b/drivers/opus/celt/mips/mdct_mipsr1.h
new file mode 100644
index 0000000000..0bb8b5a056
--- /dev/null
+++ b/drivers/opus/celt/mips/mdct_mipsr1.h
@@ -0,0 +1,286 @@
+/* Copyright (c) 2007-2008 CSIRO
+   Copyright (c) 2007-2008 Xiph.Org Foundation
+   Written by Jean-Marc Valin */
+/*
+   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.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+/* This is a simple MDCT implementation that uses a N/4 complex FFT
+   to do most of the work. It should be relatively straightforward to
+   plug in pretty much and FFT here.
+
+   This replaces the Vorbis FFT (and uses the exact same API), which
+   was a bit too messy and that was ending up duplicating code
+   (might as well use the same FFT everywhere).
+
+   The algorithm is similar to (and inspired from) Fabrice Bellard's
+   MDCT implementation in FFMPEG, but has differences in signs, ordering
+   and scaling in many places.
+*/
+#ifndef __MDCT_MIPSR1_H__
+#define __MDCT_MIPSR1_H__
+
+#ifndef SKIP_CONFIG_H
+#include "opus/opus_config.h"
+#endif
+
+#include "opus/celt/mdct.h"
+#include "opus/celt/kiss_fft.h"
+#include "opus/celt/_kiss_fft_guts.h"
+#include <math.h>
+#include "opus/celt/os_support.h"
+#include "opus/celt/mathops.h"
+#include "opus/celt/stack_alloc.h"
+
+/* Forward MDCT trashes the input array */
+#define OVERRIDE_clt_mdct_forward
+void clt_mdct_forward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out,
+      const opus_val16 *window, int overlap, int shift, int stride, int arch)
+{
+   int i;
+   int N, N2, N4;
+   VARDECL(kiss_fft_scalar, f);
+   VARDECL(kiss_fft_cpx, f2);
+   const kiss_fft_state *st = l->kfft[shift];
+   const kiss_twiddle_scalar *trig;
+   opus_val16 scale;
+#ifdef OPUS_FIXED_POINT
+   /* Allows us to scale with MULT16_32_Q16(), which is faster than
+      MULT16_32_Q15() on ARM. */
+   int scale_shift = st->scale_shift-1;
+#endif
+
+    (void)arch;
+
+   SAVE_STACK;
+   scale = st->scale;
+
+   N = l->n;
+   trig = l->trig;
+   for (i=0;i<shift;i++)
+   {
+      N >>= 1;
+      trig += N;
+   }
+   N2 = N>>1;
+   N4 = N>>2;
+
+   ALLOC(f, N2, kiss_fft_scalar);
+   ALLOC(f2, N4, kiss_fft_cpx);
+
+   /* Consider the input to be composed of four blocks: [a, b, c, d] */
+   /* Window, shuffle, fold */
+   {
+      /* Temp pointers to make it really clear to the compiler what we're doing */
+      const kiss_fft_scalar * OPUS_RESTRICT xp1 = in+(overlap>>1);
+      const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+N2-1+(overlap>>1);
+      kiss_fft_scalar * OPUS_RESTRICT yp = f;
+      const opus_val16 * OPUS_RESTRICT wp1 = window+(overlap>>1);
+      const opus_val16 * OPUS_RESTRICT wp2 = window+(overlap>>1)-1;
+      for(i=0;i<((overlap+3)>>2);i++)
+      {
+         /* Real part arranged as -d-cR, Imag part arranged as -b+aR*/
+          *yp++ = S_MUL_ADD(*wp2, xp1[N2],*wp1,*xp2);
+          *yp++ = S_MUL_SUB(*wp1, *xp1,*wp2, xp2[-N2]);
+         xp1+=2;
+         xp2-=2;
+         wp1+=2;
+         wp2-=2;
+      }
+      wp1 = window;
+      wp2 = window+overlap-1;
+      for(;i<N4-((overlap+3)>>2);i++)
+      {
+         /* Real part arranged as a-bR, Imag part arranged as -c-dR */
+         *yp++ = *xp2;
+         *yp++ = *xp1;
+         xp1+=2;
+         xp2-=2;
+      }
+      for(;i<N4;i++)
+      {
+         /* Real part arranged as a-bR, Imag part arranged as -c-dR */
+          *yp++ =  S_MUL_SUB(*wp2, *xp2, *wp1, xp1[-N2]);
+          *yp++ = S_MUL_ADD(*wp2, *xp1, *wp1, xp2[N2]);
+         xp1+=2;
+         xp2-=2;
+         wp1+=2;
+         wp2-=2;
+      }
+   }
+   /* Pre-rotation */
+   {
+      kiss_fft_scalar * OPUS_RESTRICT yp = f;
+      const kiss_twiddle_scalar *t = &trig[0];
+      for(i=0;i<N4;i++)
+      {
+         kiss_fft_cpx yc;
+         kiss_twiddle_scalar t0, t1;
+         kiss_fft_scalar re, im, yr, yi;
+         t0 = t[i];
+         t1 = t[N4+i];
+         re = *yp++;
+         im = *yp++;
+
+         yr = S_MUL_SUB(re,t0,im,t1);
+         yi = S_MUL_ADD(im,t0,re,t1);
+
+         yc.r = yr;
+         yc.i = yi;
+         yc.r = PSHR32(MULT16_32_Q16(scale, yc.r), scale_shift);
+         yc.i = PSHR32(MULT16_32_Q16(scale, yc.i), scale_shift);
+         f2[st->bitrev[i]] = yc;
+      }
+   }
+
+   /* N/4 complex FFT, does not downscale anymore */
+   opus_fft_impl(st, f2);
+
+   /* Post-rotate */
+   {
+      /* Temp pointers to make it really clear to the compiler what we're doing */
+      const kiss_fft_cpx * OPUS_RESTRICT fp = f2;
+      kiss_fft_scalar * OPUS_RESTRICT yp1 = out;
+      kiss_fft_scalar * OPUS_RESTRICT yp2 = out+stride*(N2-1);
+      const kiss_twiddle_scalar *t = &trig[0];
+      /* Temp pointers to make it really clear to the compiler what we're doing */
+      for(i=0;i<N4;i++)
+      {
+         kiss_fft_scalar yr, yi;
+         yr = S_MUL_SUB(fp->i,t[N4+i] , fp->r,t[i]);
+         yi = S_MUL_ADD(fp->r,t[N4+i] ,fp->i,t[i]);
+         *yp1 = yr;
+         *yp2 = yi;
+         fp++;
+         yp1 += 2*stride;
+         yp2 -= 2*stride;
+      }
+   }
+   RESTORE_STACK;
+}
+
+#define OVERRIDE_clt_mdct_backward
+void clt_mdct_backward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out,
+      const opus_val16 * OPUS_RESTRICT window, int overlap, int shift, int stride, int arch)
+{
+   int i;
+   int N, N2, N4;
+   const kiss_twiddle_scalar *trig;
+
+    (void)arch;
+
+   N = l->n;
+   trig = l->trig;
+   for (i=0;i<shift;i++)
+   {
+      N >>= 1;
+      trig += N;
+   }
+   N2 = N>>1;
+   N4 = N>>2;
+
+   /* Pre-rotate */
+   {
+      /* Temp pointers to make it really clear to the compiler what we're doing */
+      const kiss_fft_scalar * OPUS_RESTRICT xp1 = in;
+      const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+stride*(N2-1);
+      kiss_fft_scalar * OPUS_RESTRICT yp = out+(overlap>>1);
+      const kiss_twiddle_scalar * OPUS_RESTRICT t = &trig[0];
+      const opus_int16 * OPUS_RESTRICT bitrev = l->kfft[shift]->bitrev;
+      for(i=0;i<N4;i++)
+      {
+         int rev;
+         kiss_fft_scalar yr, yi;
+         rev = *bitrev++;
+         yr = S_MUL_ADD(*xp2, t[i] , *xp1, t[N4+i]);
+         yi = S_MUL_SUB(*xp1, t[i] , *xp2, t[N4+i]);
+         /* We swap real and imag because we use an FFT instead of an IFFT. */
+         yp[2*rev+1] = yr;
+         yp[2*rev] = yi;
+         /* Storing the pre-rotation directly in the bitrev order. */
+         xp1+=2*stride;
+         xp2-=2*stride;
+      }
+   }
+
+   opus_fft_impl(l->kfft[shift], (kiss_fft_cpx*)(out+(overlap>>1)));
+
+   /* Post-rotate and de-shuffle from both ends of the buffer at once to make
+      it in-place. */
+   {
+      kiss_fft_scalar * OPUS_RESTRICT yp0 = out+(overlap>>1);
+      kiss_fft_scalar * OPUS_RESTRICT yp1 = out+(overlap>>1)+N2-2;
+      const kiss_twiddle_scalar *t = &trig[0];
+      /* Loop to (N4+1)>>1 to handle odd N4. When N4 is odd, the
+         middle pair will be computed twice. */
+      for(i=0;i<(N4+1)>>1;i++)
+      {
+         kiss_fft_scalar re, im, yr, yi;
+         kiss_twiddle_scalar t0, t1;
+         /* We swap real and imag because we're using an FFT instead of an IFFT. */
+         re = yp0[1];
+         im = yp0[0];
+         t0 = t[i];
+         t1 = t[N4+i];
+         /* We'd scale up by 2 here, but instead it's done when mixing the windows */
+         yr = S_MUL_ADD(re,t0 , im,t1);
+         yi = S_MUL_SUB(re,t1 , im,t0);
+         /* We swap real and imag because we're using an FFT instead of an IFFT. */
+         re = yp1[1];
+         im = yp1[0];
+         yp0[0] = yr;
+         yp1[1] = yi;
+
+         t0 = t[(N4-i-1)];
+         t1 = t[(N2-i-1)];
+         /* We'd scale up by 2 here, but instead it's done when mixing the windows */
+         yr = S_MUL_ADD(re,t0,im,t1);
+         yi = S_MUL_SUB(re,t1,im,t0);
+         yp1[0] = yr;
+         yp0[1] = yi;
+         yp0 += 2;
+         yp1 -= 2;
+      }
+   }
+
+   /* Mirror on both sides for TDAC */
+   {
+      kiss_fft_scalar * OPUS_RESTRICT xp1 = out+overlap-1;
+      kiss_fft_scalar * OPUS_RESTRICT yp1 = out;
+      const opus_val16 * OPUS_RESTRICT wp1 = window;
+      const opus_val16 * OPUS_RESTRICT wp2 = window+overlap-1;
+
+      for(i = 0; i < overlap/2; i++)
+      {
+         kiss_fft_scalar x1, x2;
+         x1 = *xp1;
+         x2 = *yp1;
+         *yp1++ = MULT16_32_Q15(*wp2, x2) - MULT16_32_Q15(*wp1, x1);
+         *xp1-- = MULT16_32_Q15(*wp1, x2) + MULT16_32_Q15(*wp2, x1);
+         wp1++;
+         wp2--;
+      }
+   }
+}
+#endif /* __MDCT_MIPSR1_H__ */
diff --git a/drivers/opus/celt/mips/pitch_mipsr1.h b/drivers/opus/celt/mips/pitch_mipsr1.h
new file mode 100644
index 0000000000..a9500aff58
--- /dev/null
+++ b/drivers/opus/celt/mips/pitch_mipsr1.h
@@ -0,0 +1,161 @@
+/* Copyright (c) 2007-2008 CSIRO
+   Copyright (c) 2007-2009 Xiph.Org Foundation
+   Written by Jean-Marc Valin */
+/**
+   @file pitch.h
+   @brief Pitch analysis
+ */
+
+/*
+   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.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifndef PITCH_MIPSR1_H
+#define PITCH_MIPSR1_H
+
+#define OVERRIDE_DUAL_INNER_PROD
+static inline void dual_inner_prod(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02,
+      int N, opus_val32 *xy1, opus_val32 *xy2, int arch)
+{
+   int j;
+   opus_val32 xy01=0;
+   opus_val32 xy02=0;
+
+   (void)arch;
+
+   asm volatile("MULT $ac1, $0, $0");
+   asm volatile("MULT $ac2, $0, $0");
+   /* Compute the norm of X+Y and X-Y as |X|^2 + |Y|^2 +/- sum(xy) */
+   for (j=0;j<N;j++)
+   {
+      asm volatile("MADD $ac1, %0, %1" : : "r" ((int)x[j]), "r" ((int)y01[j]));
+      asm volatile("MADD $ac2, %0, %1" : : "r" ((int)x[j]), "r" ((int)y02[j]));
+      ++j;
+      asm volatile("MADD $ac1, %0, %1" : : "r" ((int)x[j]), "r" ((int)y01[j]));
+      asm volatile("MADD $ac2, %0, %1" : : "r" ((int)x[j]), "r" ((int)y02[j]));
+   }
+   asm volatile ("mflo %0, $ac1": "=r"(xy01));
+   asm volatile ("mflo %0, $ac2": "=r"(xy02));
+   *xy1 = xy01;
+   *xy2 = xy02;
+}
+
+static inline void xcorr_kernel_mips(const opus_val16 * x,
+      const opus_val16 * y, opus_val32 sum[4], int len)
+{
+   int j;
+   opus_val16 y_0, y_1, y_2, y_3;
+
+    opus_int64 sum_0, sum_1, sum_2, sum_3;
+    sum_0 =  (opus_int64)sum[0];
+    sum_1 =  (opus_int64)sum[1];
+    sum_2 =  (opus_int64)sum[2];
+    sum_3 =  (opus_int64)sum[3];
+
+    y_3=0; /* gcc doesn't realize that y_3 can't be used uninitialized */
+    y_0=*y++;
+    y_1=*y++;
+    y_2=*y++;
+    for (j=0;j<len-3;j+=4)
+    {
+        opus_val16 tmp;
+        tmp = *x++;
+        y_3=*y++;
+
+        sum_0 = __builtin_mips_madd( sum_0, tmp, y_0);
+        sum_1 = __builtin_mips_madd( sum_1, tmp, y_1);
+        sum_2 = __builtin_mips_madd( sum_2, tmp, y_2);
+        sum_3 = __builtin_mips_madd( sum_3, tmp, y_3);
+
+        tmp=*x++;
+        y_0=*y++;
+
+        sum_0 = __builtin_mips_madd( sum_0, tmp, y_1 );
+        sum_1 = __builtin_mips_madd( sum_1, tmp, y_2 );
+        sum_2 = __builtin_mips_madd( sum_2, tmp, y_3);
+        sum_3 = __builtin_mips_madd( sum_3, tmp, y_0);
+
+       tmp=*x++;
+       y_1=*y++;
+
+       sum_0 = __builtin_mips_madd( sum_0, tmp, y_2 );
+       sum_1 = __builtin_mips_madd( sum_1, tmp, y_3 );
+       sum_2 = __builtin_mips_madd( sum_2, tmp, y_0);
+       sum_3 = __builtin_mips_madd( sum_3, tmp, y_1);
+
+
+      tmp=*x++;
+      y_2=*y++;
+
+       sum_0 = __builtin_mips_madd( sum_0, tmp, y_3 );
+       sum_1 = __builtin_mips_madd( sum_1, tmp, y_0 );
+       sum_2 = __builtin_mips_madd( sum_2, tmp, y_1);
+       sum_3 = __builtin_mips_madd( sum_3, tmp, y_2);
+
+   }
+   if (j++<len)
+   {
+      opus_val16 tmp = *x++;
+      y_3=*y++;
+
+       sum_0 = __builtin_mips_madd( sum_0, tmp, y_0 );
+       sum_1 = __builtin_mips_madd( sum_1, tmp, y_1 );
+       sum_2 = __builtin_mips_madd( sum_2, tmp, y_2);
+       sum_3 = __builtin_mips_madd( sum_3, tmp, y_3);
+   }
+
+   if (j++<len)
+   {
+      opus_val16 tmp=*x++;
+      y_0=*y++;
+
+      sum_0 = __builtin_mips_madd( sum_0, tmp, y_1 );
+      sum_1 = __builtin_mips_madd( sum_1, tmp, y_2 );
+      sum_2 = __builtin_mips_madd( sum_2, tmp, y_3);
+      sum_3 = __builtin_mips_madd( sum_3, tmp, y_0);
+   }
+
+   if (j<len)
+   {
+      opus_val16 tmp=*x++;
+      y_1=*y++;
+
+       sum_0 = __builtin_mips_madd( sum_0, tmp, y_2 );
+       sum_1 = __builtin_mips_madd( sum_1, tmp, y_3 );
+       sum_2 = __builtin_mips_madd( sum_2, tmp, y_0);
+       sum_3 = __builtin_mips_madd( sum_3, tmp, y_1);
+
+   }
+
+   sum[0] = (opus_val32)sum_0;
+   sum[1] = (opus_val32)sum_1;
+   sum[2] = (opus_val32)sum_2;
+   sum[3] = (opus_val32)sum_3;
+}
+
+#define OVERRIDE_XCORR_KERNEL
+#define xcorr_kernel(x, y, sum, len, arch) \
+    ((void)(arch), xcorr_kernel_mips(x, y, sum, len))
+
+#endif /* PITCH_MIPSR1_H */
diff --git a/drivers/opus/celt/mips/vq_mipsr1.h b/drivers/opus/celt/mips/vq_mipsr1.h
new file mode 100644
index 0000000000..c68159972f
--- /dev/null
+++ b/drivers/opus/celt/mips/vq_mipsr1.h
@@ -0,0 +1,122 @@
+/* Copyright (c) 2007-2008 CSIRO
+   Copyright (c) 2007-2009 Xiph.Org Foundation
+   Written by Jean-Marc Valin */
+/*
+   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.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifndef __VQ_MIPSR1_H__
+#define __VQ_MIPSR1_H__
+#include "opus/opus_config.h"
+
+#include "opus/celt/mathops.h"
+#include "opus/celt/arch.h"
+
+static unsigned extract_collapse_mask(int *iy, int N, int B);
+static void normalise_residual(int * OPUS_RESTRICT iy, celt_norm * OPUS_RESTRICT X, int N, opus_val32 Ryy, opus_val16 gain);
+static void exp_rotation(celt_norm *X, int len, int dir, int stride, int K, int spread);
+static void renormalise_vector_mips(celt_norm *X, int N, opus_val16 gain, int arch);
+
+#define OVERRIDE_vq_exp_rotation1
+static void exp_rotation1(celt_norm *X, int len, int stride, opus_val16 c, opus_val16 s)
+{
+   int i;
+   opus_val16 ms;
+   celt_norm *Xptr;
+   Xptr = X;
+   ms = NEG16(s);
+   for (i=0;i<len-stride;i++)
+   {
+      celt_norm x1, x2;
+      x1 = Xptr[0];
+      x2 = Xptr[stride];
+      Xptr[stride] = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x2),  s, x1), 15));
+      *Xptr++      = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x1), ms, x2), 15));
+   }
+   Xptr = &X[len-2*stride-1];
+   for (i=len-2*stride-1;i>=0;i--)
+   {
+      celt_norm x1, x2;
+      x1 = Xptr[0];
+      x2 = Xptr[stride];
+      Xptr[stride] = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x2),  s, x1), 15));
+      *Xptr--      = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x1), ms, x2), 15));
+   }
+}
+
+#define OVERRIDE_renormalise_vector
+
+#define renormalise_vector(X, N, gain, arch) \
+ (renormalise_vector_mips(X, N, gain, arch))
+
+void renormalise_vector_mips(celt_norm *X, int N, opus_val16 gain, int arch)
+{
+   int i;
+#ifdef OPUS_FIXED_POINT
+   int k;
+#endif
+   opus_val32 E = EPSILON;
+   opus_val16 g;
+   opus_val32 t;
+   celt_norm *xptr = X;
+   int X0, X1;
+
+   (void)arch;
+
+   asm volatile("mult $ac1, $0, $0");
+   asm volatile("MTLO %0, $ac1" : :"r" (E));
+   /*if(N %4)
+       printf("error");*/
+   for (i=0;i<N-2;i+=2)
+   {
+      X0 = (int)*xptr++;
+      asm volatile("MADD $ac1, %0, %1" : : "r" (X0), "r" (X0));
+
+      X1 = (int)*xptr++;
+      asm volatile("MADD $ac1, %0, %1" : : "r" (X1), "r" (X1));
+   }
+
+   for (;i<N;i++)
+   {
+      X0 = (int)*xptr++;
+      asm volatile("MADD $ac1, %0, %1" : : "r" (X0), "r" (X0));
+   }
+
+   asm volatile("MFLO %0, $ac1" : "=r" (E));
+#ifdef OPUS_FIXED_POINT
+   k = celt_ilog2(E)>>1;
+#endif
+   t = VSHR32(E, 2*(k-7));
+   g = MULT16_16_P15(celt_rsqrt_norm(t),gain);
+
+   xptr = X;
+   for (i=0;i<N;i++)
+   {
+      *xptr = EXTRACT16(PSHR32(MULT16_16(g, *xptr), k+1));
+      xptr++;
+   }
+   /*return celt_sqrt(E);*/
+}
+
+#endif /* __VQ_MIPSR1_H__ */
diff --git a/drivers/opus/celt/modes.c b/drivers/opus/celt/modes.c
index aef2681443..d628f1468b 100644
--- a/drivers/opus/celt/modes.c
+++ b/drivers/opus/celt/modes.c
@@ -26,17 +26,15 @@
    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
-
-#ifdef OPUS_ENABLED
 #include "opus/opus_config.h"
-#endif
 
 #include "opus/celt/celt.h"
-#include "opus/celt/opus_modes.h"
+#include "opus/celt/modes.h"
 #include "opus/celt/rate.h"
 #include "opus/celt/os_support.h"
 #include "opus/celt/stack_alloc.h"
 #include "opus/celt/quant_bands.h"
+#include "opus/celt/cpu_support.h"
 
 static const opus_int16 eband5ms[] = {
 /*0  200 400 600 800  1k 1.2 1.4 1.6  2k 2.4 2.8 3.2  4k 4.8 5.6 6.8  8k 9.6 12k 15.6 */
@@ -229,6 +227,7 @@ CELTMode *opus_custom_mode_create(opus_int32 Fs, int frame_size, int *error)
    opus_val16 *window;
    opus_int16 *logN;
    int LM;
+   int arch = opus_select_arch();
    ALLOC_STACK;
 #if !defined(VAR_ARRAYS) && !defined(USE_ALLOCA)
    if (global_stack==NULL)
@@ -389,7 +388,7 @@ CELTMode *opus_custom_mode_create(opus_int32 Fs, int frame_size, int *error)
    compute_pulse_cache(mode, mode->maxLM);
 
    if (clt_mdct_init(&mode->mdct, 2*mode->shortMdctSize*mode->nbShortMdcts,
-           mode->maxLM) == 0)
+           mode->maxLM, arch) == 0)
       goto failure;
 
    if (error)
@@ -408,6 +407,8 @@ failure:
 #ifdef CUSTOM_MODES
 void opus_custom_mode_destroy(CELTMode *mode)
 {
+   int arch = opus_select_arch();
+
    if (mode == NULL)
       return;
 #ifndef CUSTOM_MODES_ONLY
@@ -431,7 +432,7 @@ void opus_custom_mode_destroy(CELTMode *mode)
    opus_free((opus_int16*)mode->cache.index);
    opus_free((unsigned char*)mode->cache.bits);
    opus_free((unsigned char*)mode->cache.caps);
-   clt_mdct_clear(&mode->mdct);
+   clt_mdct_clear(&mode->mdct, arch);
 
    opus_free((CELTMode *)mode);
 }
diff --git a/drivers/opus/celt/opus_modes.h b/drivers/opus/celt/modes.h
index 38e5844535..a5f1c30ac9 100644
--- a/drivers/opus/celt/opus_modes.h
+++ b/drivers/opus/celt/modes.h
@@ -27,8 +27,8 @@
    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
 
-#ifndef OPUS_MODES_H
-#define OPUS_MODES_H
+#ifndef MODES_H
+#define MODES_H
 
 #include "opus/opus_types.h"
 #include "opus/celt/celt.h"
@@ -39,14 +39,6 @@
 
 #define MAX_PERIOD 1024
 
-#ifndef OVERLAP
-#define OVERLAP(mode) ((mode)->overlap)
-#endif
-
-#ifndef FRAMESIZE
-#define FRAMESIZE(mode) ((mode)->mdctSize)
-#endif
-
 typedef struct {
    int size;
    const opus_int16 *index;
@@ -75,7 +67,7 @@ struct OpusCustomMode {
    const opus_int16 *logN;
 
    const opus_val16 *window;
-   celt_mdct_lookup mdct;
+   mdct_lookup mdct;
    PulseCache cache;
 };
 
diff --git a/drivers/opus/celt/opus_custom_demo.c b/drivers/opus/celt/opus_custom_demo.c
index b3129de779..956bec4e45 100644
--- a/drivers/opus/celt/opus_custom_demo.c
+++ b/drivers/opus/celt/opus_custom_demo.c
@@ -25,10 +25,7 @@
    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
-
-#ifdef OPUS_ENABLED
 #include "opus/opus_config.h"
-#endif
 
 #include "opus/opus_custom.h"
 #include "opus/celt/arch.h"
@@ -52,7 +49,7 @@ int main(int argc, char *argv[])
    int bytes_per_packet;
    unsigned char data[MAX_PACKET];
    int complexity;
-#if !(defined (OPUS_FIXED_POINT) && !defined(CUSTOM_MODES)) && defined(RESYNTH)
+#if !(defined (FIXED_POINT) && !defined(CUSTOM_MODES)) && defined(RESYNTH)
    int i;
    double rmsd = 0;
 #endif
@@ -174,7 +171,7 @@ int main(int argc, char *argv[])
       for (i=0;i<ret*channels;i++)
          out[i] = in[i];
 #endif
-#if !(defined (OPUS_FIXED_POINT) && !defined(CUSTOM_MODES)) && defined(RESYNTH)
+#if !(defined (FIXED_POINT) && !defined(CUSTOM_MODES)) && defined(RESYNTH)
       for (i=0;i<ret*channels;i++)
       {
          rmsd += (in[i]-out[i])*1.0*(in[i]-out[i]);
@@ -194,7 +191,7 @@ int main(int argc, char *argv[])
    opus_custom_mode_destroy(mode);
    free(in);
    free(out);
-#if !(defined (OPUS_FIXED_POINT) && !defined(CUSTOM_MODES)) && defined(RESYNTH)
+#if !(defined (FIXED_POINT) && !defined(CUSTOM_MODES)) && defined(RESYNTH)
    if (rmsd > 0)
    {
       rmsd = sqrt(rmsd/(1.0*frame_size*channels*count));
diff --git a/drivers/opus/celt/os_support.h b/drivers/opus/celt/os_support.h
index e1cf884467..5e36b3ae56 100644
--- a/drivers/opus/celt/os_support.h
+++ b/drivers/opus/celt/os_support.h
@@ -67,18 +67,18 @@ static OPUS_INLINE void opus_free (void *ptr)
 }
 #endif
 
-/** Copy n bytes of memory from src to dst. The 0* term provides compile-time type checking  */
+/** Copy n elements from src to dst. The 0* term provides compile-time type checking  */
 #ifndef OVERRIDE_OPUS_COPY
 #define OPUS_COPY(dst, src, n) (memcpy((dst), (src), (n)*sizeof(*(dst)) + 0*((dst)-(src)) ))
 #endif
 
-/** Copy n bytes of memory from src to dst, allowing overlapping regions. The 0* term
+/** Copy n elements from src to dst, allowing overlapping regions. The 0* term
     provides compile-time type checking */
 #ifndef OVERRIDE_OPUS_MOVE
 #define OPUS_MOVE(dst, src, n) (memmove((dst), (src), (n)*sizeof(*(dst)) + 0*((dst)-(src)) ))
 #endif
 
-/** Set n elements of dst to zero, starting at address s */
+/** Set n elements of dst to zero */
 #ifndef OVERRIDE_OPUS_CLEAR
 #define OPUS_CLEAR(dst, n) (memset((dst), 0, (n)*sizeof(*(dst))))
 #endif
diff --git a/drivers/opus/celt/pitch.c b/drivers/opus/celt/pitch.c
index c7c2b98c00..1bcbabe943 100644
--- a/drivers/opus/celt/pitch.c
+++ b/drivers/opus/celt/pitch.c
@@ -30,14 +30,11 @@
    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
-
-#ifdef OPUS_ENABLED
 #include "opus/opus_config.h"
-#endif
 
 #include "opus/celt/pitch.h"
 #include "opus/celt/os_support.h"
-#include "opus/celt/opus_modes.h"
+#include "opus/celt/modes.h"
 #include "opus/celt/stack_alloc.h"
 #include "opus/celt/mathops.h"
 #include "opus/celt/celt_lpc.h"
@@ -214,25 +211,35 @@ void pitch_downsample(celt_sig * OPUS_RESTRICT x[], opus_val16 * OPUS_RESTRICT x
    celt_fir5(x_lp, lpc2, x_lp, len>>1, mem);
 }
 
-#if 0 /* This is a simple version of the pitch correlation that should work
-         well on DSPs like Blackfin and TI C5x/C6x */
-
+/* Pure C implementation. */
 #ifdef OPUS_FIXED_POINT
 opus_val32
 #else
 void
 #endif
-celt_pitch_xcorr(opus_val16 *x, opus_val16 *y, opus_val32 *xcorr, int len, int max_pitch)
+#if defined(OVERRIDE_PITCH_XCORR)
+celt_pitch_xcorr_c(const opus_val16 *_x, const opus_val16 *_y,
+      opus_val32 *xcorr, int len, int max_pitch)
+#else
+celt_pitch_xcorr(const opus_val16 *_x, const opus_val16 *_y,
+      opus_val32 *xcorr, int len, int max_pitch, int arch)
+#endif
 {
+
+#if 0 /* This is a simple version of the pitch correlation that should work
+         well on DSPs like Blackfin and TI C5x/C6x */
    int i, j;
 #ifdef OPUS_FIXED_POINT
    opus_val32 maxcorr=1;
 #endif
+#if !defined(OVERRIDE_PITCH_XCORR)
+   (void)arch;
+#endif
    for (i=0;i<max_pitch;i++)
    {
       opus_val32 sum = 0;
       for (j=0;j<len;j++)
-         sum = MAC16_16(sum, x[j],y[i+j]);
+         sum = MAC16_16(sum, _x[j], _y[i+j]);
       xcorr[i] = sum;
 #ifdef OPUS_FIXED_POINT
       maxcorr = MAX32(maxcorr, sum);
@@ -241,30 +248,25 @@ celt_pitch_xcorr(opus_val16 *x, opus_val16 *y, opus_val32 *xcorr, int len, int m
 #ifdef OPUS_FIXED_POINT
    return maxcorr;
 #endif
-}
 
 #else /* Unrolled version of the pitch correlation -- runs faster on x86 and ARM */
-
-#ifdef OPUS_FIXED_POINT
-opus_val32
-#else
-void
-#endif
-celt_pitch_xcorr_c(const opus_val16 *_x, const opus_val16 *_y, opus_val32 *xcorr, int len, int max_pitch)
-{
-   int i,j;
+   int i;
    /*The EDSP version requires that max_pitch is at least 1, and that _x is
       32-bit aligned.
      Since it's hard to put asserts in assembly, put them here.*/
-   celt_assert(max_pitch>0);
-   celt_assert((((unsigned char *)_x-(unsigned char *)NULL)&3)==0);
 #ifdef OPUS_FIXED_POINT
    opus_val32 maxcorr=1;
 #endif
+   celt_assert(max_pitch>0);
+   celt_assert((((unsigned char *)_x-(unsigned char *)NULL)&3)==0);
    for (i=0;i<max_pitch-3;i+=4)
    {
       opus_val32 sum[4]={0,0,0,0};
-      xcorr_kernel(_x, _y+i, sum, len);
+#if defined(OVERRIDE_PITCH_XCORR)
+      xcorr_kernel_c(_x, _y+i, sum, len);
+#else
+      xcorr_kernel(_x, _y+i, sum, len, arch);
+#endif
       xcorr[i]=sum[0];
       xcorr[i+1]=sum[1];
       xcorr[i+2]=sum[2];
@@ -279,9 +281,12 @@ celt_pitch_xcorr_c(const opus_val16 *_x, const opus_val16 *_y, opus_val32 *xcorr
    /* In case max_pitch isn't a multiple of 4, do non-unrolled version. */
    for (;i<max_pitch;i++)
    {
-      opus_val32 sum = 0;
-      for (j=0;j<len;j++)
-         sum = MAC16_16(sum, _x[j],_y[i+j]);
+      opus_val32 sum;
+#if defined(OVERRIDE_PITCH_XCORR)
+      sum = celt_inner_prod_c(_x, _y+i, len);
+#else
+      sum = celt_inner_prod(_x, _y+i, len, arch);
+#endif
       xcorr[i] = sum;
 #ifdef OPUS_FIXED_POINT
       maxcorr = MAX32(maxcorr, sum);
@@ -290,9 +295,9 @@ celt_pitch_xcorr_c(const opus_val16 *_x, const opus_val16 *_y, opus_val32 *xcorr
 #ifdef OPUS_FIXED_POINT
    return maxcorr;
 #endif
+#endif
 }
 
-#endif
 void pitch_search(const opus_val16 * OPUS_RESTRICT x_lp, opus_val16 * OPUS_RESTRICT y,
                   int len, int max_pitch, int *pitch, int arch)
 {
@@ -361,12 +366,17 @@ void pitch_search(const opus_val16 * OPUS_RESTRICT x_lp, opus_val16 * OPUS_RESTR
 #endif
    for (i=0;i<max_pitch>>1;i++)
    {
-      opus_val32 sum=0;
+      opus_val32 sum;
       xcorr[i] = 0;
       if (abs(i-2*best_pitch[0])>2 && abs(i-2*best_pitch[1])>2)
          continue;
+#ifdef OPUS_FIXED_POINT
+      sum = 0;
       for (j=0;j<len>>1;j++)
          sum += SHR32(MULT16_16(x_lp[j],y[i+j]), shift);
+#else
+      sum = celt_inner_prod_c(x_lp, y+i, len>>1);
+#endif
       xcorr[i] = MAX32(-1, sum);
 #ifdef OPUS_FIXED_POINT
       maxcorr = MAX32(maxcorr, sum);
@@ -401,7 +411,7 @@ void pitch_search(const opus_val16 * OPUS_RESTRICT x_lp, opus_val16 * OPUS_RESTR
 
 static const int second_check[16] = {0, 0, 3, 2, 3, 2, 5, 2, 3, 2, 3, 2, 5, 2, 3, 2};
 opus_val16 remove_doubling(opus_val16 *x, int maxperiod, int minperiod,
-      int N, int *T0_, int prev_period, opus_val16 prev_gain)
+      int N, int *T0_, int prev_period, opus_val16 prev_gain, int arch)
 {
    int k, i, T, T0;
    opus_val16 g, g0;
@@ -426,7 +436,7 @@ opus_val16 remove_doubling(opus_val16 *x, int maxperiod, int minperiod,
 
    T = T0 = *T0_;
    ALLOC(yy_lookup, maxperiod+1, opus_val32);
-   dual_inner_prod(x, x, x-T0, N, &xx, &xy);
+   dual_inner_prod(x, x, x-T0, N, &xx, &xy, arch);
    yy_lookup[0] = xx;
    yy=xx;
    for (i=1;i<=maxperiod;i++)
@@ -456,7 +466,7 @@ opus_val16 remove_doubling(opus_val16 *x, int maxperiod, int minperiod,
       opus_val16 g1;
       opus_val16 cont=0;
       opus_val16 thresh;
-      T1 = (2*T0+k)/(2*k);
+      T1 = celt_udiv(2*T0+k, 2*k);
       if (T1 < minperiod)
          break;
       /* Look for another strong correlation at T1b */
@@ -468,9 +478,9 @@ opus_val16 remove_doubling(opus_val16 *x, int maxperiod, int minperiod,
             T1b = T0+T1;
       } else
       {
-         T1b = (2*second_check[k]*T0+k)/(2*k);
+         T1b = celt_udiv(2*second_check[k]*T0+k, 2*k);
       }
-      dual_inner_prod(x, &x[-T1], &x[-T1b], N, &xy, &xy2);
+      dual_inner_prod(x, &x[-T1], &x[-T1b], N, &xy, &xy2, arch);
       xy += xy2;
       yy = yy_lookup[T1] + yy_lookup[T1b];
 #ifdef OPUS_FIXED_POINT
@@ -513,13 +523,7 @@ opus_val16 remove_doubling(opus_val16 *x, int maxperiod, int minperiod,
       pg = SHR32(frac_div32(best_xy,best_yy+1),16);
 
    for (k=0;k<3;k++)
-   {
-      int T1 = T+k-1;
-      xy = 0;
-      for (i=0;i<N;i++)
-         xy = MAC16_16(xy, x[i], x[i-T1]);
-      xcorr[k] = xy;
-   }
+      xcorr[k] = celt_inner_prod(x, x-(T+k-1), N, arch);
    if ((xcorr[2]-xcorr[0]) > MULT16_32_Q15(QCONST16(.7f,15),xcorr[1]-xcorr[0]))
       offset = 1;
    else if ((xcorr[0]-xcorr[2]) > MULT16_32_Q15(QCONST16(.7f,15),xcorr[1]-xcorr[2]))
diff --git a/drivers/opus/celt/pitch.h b/drivers/opus/celt/pitch.h
index f599f5fc76..ecae7667b8 100644
--- a/drivers/opus/celt/pitch.h
+++ b/drivers/opus/celt/pitch.h
@@ -34,15 +34,21 @@
 #ifndef PITCH_H
 #define PITCH_H
 
-#include "opus/celt/opus_modes.h"
+#include "opus/celt/modes.h"
 #include "opus/celt/cpu_support.h"
 
-#if defined(__SSE__) && !defined(OPUS_FIXED_POINT)
-#include "x86/pitch_sse.h"
+#if (defined(OPUS_X86_MAY_HAVE_SSE) && !defined(FIXED_POINT)) \
+  || ((defined(OPUS_X86_MAY_HAVE_SSE4_1) || defined(OPUS_X86_MAY_HAVE_SSE2)) && defined(FIXED_POINT))
+#include "opus/celt/x86/pitch_sse.h"
 #endif
 
-#if defined(OPUS_ARM_ASM) && defined(OPUS_FIXED_POINT)
-# include "arm/pitch_arm.h"
+#if defined(MIPSr1_ASM)
+#include "opus/celt/mips/pitch_mipsr1.h"
+#endif
+
+#if ((defined(OPUS_ARM_ASM) && defined(FIXED_POINT)) \
+  || defined(OPUS_ARM_MAY_HAVE_NEON_INTR))
+# include "opus/celt/arm/pitch_arm.h"
 #endif
 
 void pitch_downsample(celt_sig * OPUS_RESTRICT x[], opus_val16 * OPUS_RESTRICT x_lp,
@@ -52,12 +58,12 @@ void pitch_search(const opus_val16 * OPUS_RESTRICT x_lp, opus_val16 * OPUS_RESTR
                   int len, int max_pitch, int *pitch, int arch);
 
 opus_val16 remove_doubling(opus_val16 *x, int maxperiod, int minperiod,
-      int N, int *T0, int prev_period, opus_val16 prev_gain);
+      int N, int *T0, int prev_period, opus_val16 prev_gain, int arch);
+
 
 /* OPT: This is the kernel you really want to optimize. It gets used a lot
    by the prefilter and by the PLC. */
-#ifndef OVERRIDE_XCORR_KERNEL
-static OPUS_INLINE void xcorr_kernel(const opus_val16 * x, const opus_val16 * y, opus_val32 sum[4], int len)
+static OPUS_INLINE void xcorr_kernel_c(const opus_val16 * x, const opus_val16 * y, opus_val32 sum[4], int len)
 {
    int j;
    opus_val16 y_0, y_1, y_2, y_3;
@@ -122,10 +128,14 @@ static OPUS_INLINE void xcorr_kernel(const opus_val16 * x, const opus_val16 * y,
       sum[3] = MAC16_16(sum[3],tmp,y_1);
    }
 }
+
+#ifndef OVERRIDE_XCORR_KERNEL
+#define xcorr_kernel(x, y, sum, len, arch) \
+    ((void)(arch),xcorr_kernel_c(x, y, sum, len))
 #endif /* OVERRIDE_XCORR_KERNEL */
 
-#ifndef OVERRIDE_DUAL_INNER_PROD
-static OPUS_INLINE void dual_inner_prod(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02,
+
+static OPUS_INLINE void dual_inner_prod_c(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02,
       int N, opus_val32 *xy1, opus_val32 *xy2)
 {
    int i;
@@ -139,8 +149,35 @@ static OPUS_INLINE void dual_inner_prod(const opus_val16 *x, const opus_val16 *y
    *xy1 = xy01;
    *xy2 = xy02;
 }
+
+#ifndef OVERRIDE_DUAL_INNER_PROD
+# define dual_inner_prod(x, y01, y02, N, xy1, xy2, arch) \
+    ((void)(arch),dual_inner_prod_c(x, y01, y02, N, xy1, xy2))
+#endif
+
+/*We make sure a C version is always available for cases where the overhead of
+  vectorization and passing around an arch flag aren't worth it.*/
+static OPUS_INLINE opus_val32 celt_inner_prod_c(const opus_val16 *x,
+      const opus_val16 *y, int N)
+{
+   int i;
+   opus_val32 xy=0;
+   for (i=0;i<N;i++)
+      xy = MAC16_16(xy, x[i], y[i]);
+   return xy;
+}
+
+#if !defined(OVERRIDE_CELT_INNER_PROD)
+# define celt_inner_prod(x, y, N, arch) \
+    ((void)(arch),celt_inner_prod_c(x, y, N))
 #endif
 
+#ifdef NON_STATIC_COMB_FILTER_CONST_C
+void comb_filter_const_c(opus_val32 *y, opus_val32 *x, int T, int N,
+     opus_val16 g10, opus_val16 g11, opus_val16 g12);
+#endif
+
+
 #ifdef OPUS_FIXED_POINT
 opus_val32
 #else
@@ -151,9 +188,11 @@ celt_pitch_xcorr_c(const opus_val16 *_x, const opus_val16 *_y,
 
 #if !defined(OVERRIDE_PITCH_XCORR)
 /*Is run-time CPU detection enabled on this platform?*/
-# if defined(OPUS_HAVE_RTCD)
+# if defined(OPUS_HAVE_RTCD) && (defined(OPUS_ARM_ASM) \
+   || (defined(OPUS_ARM_MAY_HAVE_NEON_INTR) \
+   && !defined(OPUS_ARM_PRESUME_NEON_INTR)))
 extern
-#  if defined(OPUS_FIXED_POINT)
+#  if defined(FIXED_POINT)
 opus_val32
 #  else
 void
@@ -161,12 +200,20 @@ void
 (*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *,
       const opus_val16 *, opus_val32 *, int, int);
 
+#  define OVERRIDE_PITCH_XCORR
 #  define celt_pitch_xcorr(_x, _y, xcorr, len, max_pitch, arch) \
   ((*CELT_PITCH_XCORR_IMPL[(arch)&OPUS_ARCHMASK])(_x, _y, \
         xcorr, len, max_pitch))
 # else
-#  define celt_pitch_xcorr(_x, _y, xcorr, len, max_pitch, arch) \
-  ((void)(arch),celt_pitch_xcorr_c(_x, _y, xcorr, len, max_pitch))
+
+#ifdef OPUS_FIXED_POINT
+opus_val32
+#else
+void
+#endif
+celt_pitch_xcorr(const opus_val16 *_x, const opus_val16 *_y,
+      opus_val32 *xcorr, int len, int max_pitch, int arch);
+
 # endif
 #endif
 
diff --git a/drivers/opus/celt/quant_bands.c b/drivers/opus/celt/quant_bands.c
index e64ed28f22..b38bfa8b8e 100644
--- a/drivers/opus/celt/quant_bands.c
+++ b/drivers/opus/celt/quant_bands.c
@@ -25,10 +25,7 @@
    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
-
-#ifdef OPUS_ENABLED
 #include "opus/opus_config.h"
-#endif
 
 #include "opus/celt/quant_bands.h"
 #include "opus/celt/laplace.h"
@@ -292,7 +289,7 @@ void quant_coarse_energy(const CELTMode *m, int start, int end, int effEnd,
 #endif
    }
    if (lfe)
-      max_decay=3;
+      max_decay = QCONST16(3.f,DB_SHIFT);
    enc_start_state = *enc;
 
    ALLOC(oldEBands_intra, C*m->nbEBands, opus_val16);
diff --git a/drivers/opus/celt/quant_bands.h b/drivers/opus/celt/quant_bands.h
index 85d011e6e0..31d4b94e41 100644
--- a/drivers/opus/celt/quant_bands.h
+++ b/drivers/opus/celt/quant_bands.h
@@ -30,7 +30,7 @@
 #define QUANT_BANDS
 
 #include "opus/celt/arch.h"
-#include "opus/celt/opus_modes.h"
+#include "opus/celt/modes.h"
 #include "opus/celt/entenc.h"
 #include "opus/celt/entdec.h"
 #include "opus/celt/mathops.h"
diff --git a/drivers/opus/celt/rate.c b/drivers/opus/celt/rate.c
index ecc0ab2a4f..e69fb3f030 100644
--- a/drivers/opus/celt/rate.c
+++ b/drivers/opus/celt/rate.c
@@ -25,13 +25,10 @@
    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
-
-#ifdef OPUS_ENABLED
 #include "opus/opus_config.h"
-#endif
 
 #include <math.h>
-#include "opus/celt/opus_modes.h"
+#include "opus/celt/modes.h"
 #include "opus/celt/cwrs.h"
 #include "opus/celt/arch.h"
 #include "opus/celt/os_support.h"
@@ -131,7 +128,7 @@ void compute_pulse_cache(CELTMode *m, int LM)
    for (i=0;i<nbEntries;i++)
    {
       unsigned char *ptr = bits+entryI[i];
-      opus_int16 tmp[MAX_PULSES+1];
+      opus_int16 tmp[CELT_MAX_PULSES+1];
       get_required_bits(tmp, entryN[i], get_pulses(entryK[i]), BITRES);
       for (j=1;j<=entryK[i];j++)
          ptr[j] = tmp[get_pulses(j)]-1;
@@ -333,7 +330,7 @@ static OPUS_INLINE int interp_bits2pulses(const CELTMode *m, int start, int end,
       /*Figure out how many left-over bits we would be adding to this band.
         This can include bits we've stolen back from higher, skipped bands.*/
       left = total-psum;
-      percoeff = left/(m->eBands[codedBands]-m->eBands[start]);
+      percoeff = celt_udiv(left, m->eBands[codedBands]-m->eBands[start]);
       left -= (m->eBands[codedBands]-m->eBands[start])*percoeff;
       rem = IMAX(left-(m->eBands[j]-m->eBands[start]),0);
       band_width = m->eBands[codedBands]-m->eBands[j];
@@ -414,7 +411,7 @@ static OPUS_INLINE int interp_bits2pulses(const CELTMode *m, int start, int end,
 
    /* Allocate the remaining bits */
    left = total-psum;
-   percoeff = left/(m->eBands[codedBands]-m->eBands[start]);
+   percoeff = celt_udiv(left, m->eBands[codedBands]-m->eBands[start]);
    left -= (m->eBands[codedBands]-m->eBands[start])*percoeff;
    for (j=start;j<codedBands;j++)
       bits[j] += ((int)percoeff*(m->eBands[j+1]-m->eBands[j]));
@@ -465,7 +462,8 @@ static OPUS_INLINE int interp_bits2pulses(const CELTMode *m, int start, int end,
             offset += NClogN>>3;
 
          /* Divide with rounding */
-         ebits[j] = IMAX(0, (bits[j] + offset + (den<<(BITRES-1))) / (den<<BITRES));
+         ebits[j] = IMAX(0, (bits[j] + offset + (den<<(BITRES-1))));
+         ebits[j] = celt_udiv(ebits[j], den)>>BITRES;
 
          /* Make sure not to bust */
          if (C*ebits[j] > (bits[j]>>BITRES))
diff --git a/drivers/opus/celt/rate.h b/drivers/opus/celt/rate.h
index e12dd29db8..cf8085fccf 100644
--- a/drivers/opus/celt/rate.h
+++ b/drivers/opus/celt/rate.h
@@ -32,7 +32,7 @@
 #define MAX_PSEUDO 40
 #define LOG_MAX_PSEUDO 6
 
-#define MAX_PULSES 128
+#define CELT_MAX_PULSES 128
 
 #define MAX_FINE_BITS 8
 
@@ -41,7 +41,7 @@
 #define QTHETA_OFFSET_TWOPHASE 16
 
 #include "opus/celt/cwrs.h"
-#include "opus/celt/opus_modes.h"
+#include "opus/celt/modes.h"
 
 void compute_pulse_cache(CELTMode *m, int LM);
 
diff --git a/drivers/opus/celt/stack_alloc.h b/drivers/opus/celt/stack_alloc.h
index 464a6d0b7f..7cd6c56ccf 100644
--- a/drivers/opus/celt/stack_alloc.h
+++ b/drivers/opus/celt/stack_alloc.h
@@ -36,21 +36,17 @@
 #include "opus/opus_defines.h"
 
 #if (!defined (VAR_ARRAYS) && !defined (USE_ALLOCA) && !defined (NONTHREADSAFE_PSEUDOSTACK))
-#define VAR_ARRAYS
+#error "Opus requires one of VAR_ARRAYS, USE_ALLOCA, or NONTHREADSAFE_PSEUDOSTACK be defined to select the temporary allocation mode."
 #endif
 
 #ifdef USE_ALLOCA
 # ifdef WIN32
 #  include <malloc.h>
 # else
-#  ifdef OPUS_HAVE_ALLOCA_H
+#  ifdef HAVE_ALLOCA_H
 #   include <alloca.h>
 #  else
-#   ifdef __linux__
-#    include <alloca.h>
-#   else
-#    include <stdlib.h>
-#   endif
+#   include <stdlib.h>
 #  endif
 # endif
 #endif
@@ -120,9 +116,11 @@
 #else
 
 #ifdef CELT_C
+char *scratch_ptr=0;
 char *global_stack=0;
 #else
 extern char *global_stack;
+extern char *scratch_ptr;
 #endif /* CELT_C */
 
 #ifdef ENABLE_VALGRIND
@@ -144,8 +142,12 @@ extern char *global_stack_top;
 
 #define ALIGN(stack, size) ((stack) += ((size) - (long)(stack)) & ((size) - 1))
 #define PUSH(stack, size, type) (ALIGN((stack),sizeof(type)/sizeof(char)),(stack)+=(size)*(sizeof(type)/sizeof(char)),(type*)((stack)-(size)*(sizeof(type)/sizeof(char))))
+#if 0 /* Set this to 1 to instrument pseudostack usage */
+#define RESTORE_STACK (printf("%ld %s:%d\n", global_stack-scratch_ptr, __FILE__, __LINE__),global_stack = _saved_stack)
+#else
 #define RESTORE_STACK (global_stack = _saved_stack)
-#define ALLOC_STACK char *_saved_stack; (global_stack = (global_stack==0) ? opus_alloc_scratch(GLOBAL_STACK_SIZE) : global_stack); _saved_stack = global_stack;
+#endif
+#define ALLOC_STACK char *_saved_stack; (global_stack = (global_stack==0) ? (scratch_ptr=opus_alloc_scratch(GLOBAL_STACK_SIZE)) : global_stack); _saved_stack = global_stack;
 
 #endif /* ENABLE_VALGRIND */
 
diff --git a/drivers/opus/celt/static_modes_fixed.h b/drivers/opus/celt/static_modes_fixed.h
index 1d92fc4b27..352642f507 100644
--- a/drivers/opus/celt/static_modes_fixed.h
+++ b/drivers/opus/celt/static_modes_fixed.h
@@ -1,9 +1,14 @@
 /* The contents of this file was automatically generated by dump_modes.c
    with arguments: 48000 960
    It contains static definitions for some pre-defined modes. */
-#include "opus/celt/opus_modes.h"
+#include "opus/celt/modes.h"
 #include "opus/celt/rate.h"
 
+#ifdef HAVE_ARM_NE10
+#define OVERRIDE_FFT 1
+#include "opus/celt/static_modes_fixed_arm_ne10.h"
+#endif
+
 #ifndef DEF_WINDOW120
 #define DEF_WINDOW120
 static const opus_val16 window120[120] = {
@@ -341,84 +346,84 @@ static const kiss_twiddle_cpx fft_twiddles48000_960[480] = {
 #ifndef FFT_BITREV480
 #define FFT_BITREV480
 static const opus_int16 fft_bitrev480[480] = {
-0, 120, 240, 360, 30, 150, 270, 390, 60, 180, 300, 420, 90, 210, 330,
-450, 15, 135, 255, 375, 45, 165, 285, 405, 75, 195, 315, 435, 105, 225,
-345, 465, 5, 125, 245, 365, 35, 155, 275, 395, 65, 185, 305, 425, 95,
-215, 335, 455, 20, 140, 260, 380, 50, 170, 290, 410, 80, 200, 320, 440,
-110, 230, 350, 470, 10, 130, 250, 370, 40, 160, 280, 400, 70, 190, 310,
-430, 100, 220, 340, 460, 25, 145, 265, 385, 55, 175, 295, 415, 85, 205,
-325, 445, 115, 235, 355, 475, 1, 121, 241, 361, 31, 151, 271, 391, 61,
-181, 301, 421, 91, 211, 331, 451, 16, 136, 256, 376, 46, 166, 286, 406,
-76, 196, 316, 436, 106, 226, 346, 466, 6, 126, 246, 366, 36, 156, 276,
-396, 66, 186, 306, 426, 96, 216, 336, 456, 21, 141, 261, 381, 51, 171,
-291, 411, 81, 201, 321, 441, 111, 231, 351, 471, 11, 131, 251, 371, 41,
-161, 281, 401, 71, 191, 311, 431, 101, 221, 341, 461, 26, 146, 266, 386,
-56, 176, 296, 416, 86, 206, 326, 446, 116, 236, 356, 476, 2, 122, 242,
-362, 32, 152, 272, 392, 62, 182, 302, 422, 92, 212, 332, 452, 17, 137,
-257, 377, 47, 167, 287, 407, 77, 197, 317, 437, 107, 227, 347, 467, 7,
-127, 247, 367, 37, 157, 277, 397, 67, 187, 307, 427, 97, 217, 337, 457,
-22, 142, 262, 382, 52, 172, 292, 412, 82, 202, 322, 442, 112, 232, 352,
-472, 12, 132, 252, 372, 42, 162, 282, 402, 72, 192, 312, 432, 102, 222,
-342, 462, 27, 147, 267, 387, 57, 177, 297, 417, 87, 207, 327, 447, 117,
-237, 357, 477, 3, 123, 243, 363, 33, 153, 273, 393, 63, 183, 303, 423,
-93, 213, 333, 453, 18, 138, 258, 378, 48, 168, 288, 408, 78, 198, 318,
-438, 108, 228, 348, 468, 8, 128, 248, 368, 38, 158, 278, 398, 68, 188,
-308, 428, 98, 218, 338, 458, 23, 143, 263, 383, 53, 173, 293, 413, 83,
-203, 323, 443, 113, 233, 353, 473, 13, 133, 253, 373, 43, 163, 283, 403,
-73, 193, 313, 433, 103, 223, 343, 463, 28, 148, 268, 388, 58, 178, 298,
-418, 88, 208, 328, 448, 118, 238, 358, 478, 4, 124, 244, 364, 34, 154,
-274, 394, 64, 184, 304, 424, 94, 214, 334, 454, 19, 139, 259, 379, 49,
-169, 289, 409, 79, 199, 319, 439, 109, 229, 349, 469, 9, 129, 249, 369,
-39, 159, 279, 399, 69, 189, 309, 429, 99, 219, 339, 459, 24, 144, 264,
-384, 54, 174, 294, 414, 84, 204, 324, 444, 114, 234, 354, 474, 14, 134,
-254, 374, 44, 164, 284, 404, 74, 194, 314, 434, 104, 224, 344, 464, 29,
-149, 269, 389, 59, 179, 299, 419, 89, 209, 329, 449, 119, 239, 359, 479,
+0, 96, 192, 288, 384, 32, 128, 224, 320, 416, 64, 160, 256, 352, 448,
+8, 104, 200, 296, 392, 40, 136, 232, 328, 424, 72, 168, 264, 360, 456,
+16, 112, 208, 304, 400, 48, 144, 240, 336, 432, 80, 176, 272, 368, 464,
+24, 120, 216, 312, 408, 56, 152, 248, 344, 440, 88, 184, 280, 376, 472,
+4, 100, 196, 292, 388, 36, 132, 228, 324, 420, 68, 164, 260, 356, 452,
+12, 108, 204, 300, 396, 44, 140, 236, 332, 428, 76, 172, 268, 364, 460,
+20, 116, 212, 308, 404, 52, 148, 244, 340, 436, 84, 180, 276, 372, 468,
+28, 124, 220, 316, 412, 60, 156, 252, 348, 444, 92, 188, 284, 380, 476,
+1, 97, 193, 289, 385, 33, 129, 225, 321, 417, 65, 161, 257, 353, 449,
+9, 105, 201, 297, 393, 41, 137, 233, 329, 425, 73, 169, 265, 361, 457,
+17, 113, 209, 305, 401, 49, 145, 241, 337, 433, 81, 177, 273, 369, 465,
+25, 121, 217, 313, 409, 57, 153, 249, 345, 441, 89, 185, 281, 377, 473,
+5, 101, 197, 293, 389, 37, 133, 229, 325, 421, 69, 165, 261, 357, 453,
+13, 109, 205, 301, 397, 45, 141, 237, 333, 429, 77, 173, 269, 365, 461,
+21, 117, 213, 309, 405, 53, 149, 245, 341, 437, 85, 181, 277, 373, 469,
+29, 125, 221, 317, 413, 61, 157, 253, 349, 445, 93, 189, 285, 381, 477,
+2, 98, 194, 290, 386, 34, 130, 226, 322, 418, 66, 162, 258, 354, 450,
+10, 106, 202, 298, 394, 42, 138, 234, 330, 426, 74, 170, 266, 362, 458,
+18, 114, 210, 306, 402, 50, 146, 242, 338, 434, 82, 178, 274, 370, 466,
+26, 122, 218, 314, 410, 58, 154, 250, 346, 442, 90, 186, 282, 378, 474,
+6, 102, 198, 294, 390, 38, 134, 230, 326, 422, 70, 166, 262, 358, 454,
+14, 110, 206, 302, 398, 46, 142, 238, 334, 430, 78, 174, 270, 366, 462,
+22, 118, 214, 310, 406, 54, 150, 246, 342, 438, 86, 182, 278, 374, 470,
+30, 126, 222, 318, 414, 62, 158, 254, 350, 446, 94, 190, 286, 382, 478,
+3, 99, 195, 291, 387, 35, 131, 227, 323, 419, 67, 163, 259, 355, 451,
+11, 107, 203, 299, 395, 43, 139, 235, 331, 427, 75, 171, 267, 363, 459,
+19, 115, 211, 307, 403, 51, 147, 243, 339, 435, 83, 179, 275, 371, 467,
+27, 123, 219, 315, 411, 59, 155, 251, 347, 443, 91, 187, 283, 379, 475,
+7, 103, 199, 295, 391, 39, 135, 231, 327, 423, 71, 167, 263, 359, 455,
+15, 111, 207, 303, 399, 47, 143, 239, 335, 431, 79, 175, 271, 367, 463,
+23, 119, 215, 311, 407, 55, 151, 247, 343, 439, 87, 183, 279, 375, 471,
+31, 127, 223, 319, 415, 63, 159, 255, 351, 447, 95, 191, 287, 383, 479,
 };
 #endif
 
 #ifndef FFT_BITREV240
 #define FFT_BITREV240
 static const opus_int16 fft_bitrev240[240] = {
-0, 60, 120, 180, 15, 75, 135, 195, 30, 90, 150, 210, 45, 105, 165,
-225, 5, 65, 125, 185, 20, 80, 140, 200, 35, 95, 155, 215, 50, 110,
-170, 230, 10, 70, 130, 190, 25, 85, 145, 205, 40, 100, 160, 220, 55,
-115, 175, 235, 1, 61, 121, 181, 16, 76, 136, 196, 31, 91, 151, 211,
-46, 106, 166, 226, 6, 66, 126, 186, 21, 81, 141, 201, 36, 96, 156,
-216, 51, 111, 171, 231, 11, 71, 131, 191, 26, 86, 146, 206, 41, 101,
-161, 221, 56, 116, 176, 236, 2, 62, 122, 182, 17, 77, 137, 197, 32,
-92, 152, 212, 47, 107, 167, 227, 7, 67, 127, 187, 22, 82, 142, 202,
-37, 97, 157, 217, 52, 112, 172, 232, 12, 72, 132, 192, 27, 87, 147,
-207, 42, 102, 162, 222, 57, 117, 177, 237, 3, 63, 123, 183, 18, 78,
-138, 198, 33, 93, 153, 213, 48, 108, 168, 228, 8, 68, 128, 188, 23,
-83, 143, 203, 38, 98, 158, 218, 53, 113, 173, 233, 13, 73, 133, 193,
-28, 88, 148, 208, 43, 103, 163, 223, 58, 118, 178, 238, 4, 64, 124,
-184, 19, 79, 139, 199, 34, 94, 154, 214, 49, 109, 169, 229, 9, 69,
-129, 189, 24, 84, 144, 204, 39, 99, 159, 219, 54, 114, 174, 234, 14,
-74, 134, 194, 29, 89, 149, 209, 44, 104, 164, 224, 59, 119, 179, 239,
+0, 48, 96, 144, 192, 16, 64, 112, 160, 208, 32, 80, 128, 176, 224,
+4, 52, 100, 148, 196, 20, 68, 116, 164, 212, 36, 84, 132, 180, 228,
+8, 56, 104, 152, 200, 24, 72, 120, 168, 216, 40, 88, 136, 184, 232,
+12, 60, 108, 156, 204, 28, 76, 124, 172, 220, 44, 92, 140, 188, 236,
+1, 49, 97, 145, 193, 17, 65, 113, 161, 209, 33, 81, 129, 177, 225,
+5, 53, 101, 149, 197, 21, 69, 117, 165, 213, 37, 85, 133, 181, 229,
+9, 57, 105, 153, 201, 25, 73, 121, 169, 217, 41, 89, 137, 185, 233,
+13, 61, 109, 157, 205, 29, 77, 125, 173, 221, 45, 93, 141, 189, 237,
+2, 50, 98, 146, 194, 18, 66, 114, 162, 210, 34, 82, 130, 178, 226,
+6, 54, 102, 150, 198, 22, 70, 118, 166, 214, 38, 86, 134, 182, 230,
+10, 58, 106, 154, 202, 26, 74, 122, 170, 218, 42, 90, 138, 186, 234,
+14, 62, 110, 158, 206, 30, 78, 126, 174, 222, 46, 94, 142, 190, 238,
+3, 51, 99, 147, 195, 19, 67, 115, 163, 211, 35, 83, 131, 179, 227,
+7, 55, 103, 151, 199, 23, 71, 119, 167, 215, 39, 87, 135, 183, 231,
+11, 59, 107, 155, 203, 27, 75, 123, 171, 219, 43, 91, 139, 187, 235,
+15, 63, 111, 159, 207, 31, 79, 127, 175, 223, 47, 95, 143, 191, 239,
 };
 #endif
 
 #ifndef FFT_BITREV120
 #define FFT_BITREV120
 static const opus_int16 fft_bitrev120[120] = {
-0, 30, 60, 90, 15, 45, 75, 105, 5, 35, 65, 95, 20, 50, 80,
-110, 10, 40, 70, 100, 25, 55, 85, 115, 1, 31, 61, 91, 16, 46,
-76, 106, 6, 36, 66, 96, 21, 51, 81, 111, 11, 41, 71, 101, 26,
-56, 86, 116, 2, 32, 62, 92, 17, 47, 77, 107, 7, 37, 67, 97,
-22, 52, 82, 112, 12, 42, 72, 102, 27, 57, 87, 117, 3, 33, 63,
-93, 18, 48, 78, 108, 8, 38, 68, 98, 23, 53, 83, 113, 13, 43,
-73, 103, 28, 58, 88, 118, 4, 34, 64, 94, 19, 49, 79, 109, 9,
-39, 69, 99, 24, 54, 84, 114, 14, 44, 74, 104, 29, 59, 89, 119,
+0, 24, 48, 72, 96, 8, 32, 56, 80, 104, 16, 40, 64, 88, 112,
+4, 28, 52, 76, 100, 12, 36, 60, 84, 108, 20, 44, 68, 92, 116,
+1, 25, 49, 73, 97, 9, 33, 57, 81, 105, 17, 41, 65, 89, 113,
+5, 29, 53, 77, 101, 13, 37, 61, 85, 109, 21, 45, 69, 93, 117,
+2, 26, 50, 74, 98, 10, 34, 58, 82, 106, 18, 42, 66, 90, 114,
+6, 30, 54, 78, 102, 14, 38, 62, 86, 110, 22, 46, 70, 94, 118,
+3, 27, 51, 75, 99, 11, 35, 59, 83, 107, 19, 43, 67, 91, 115,
+7, 31, 55, 79, 103, 15, 39, 63, 87, 111, 23, 47, 71, 95, 119,
 };
 #endif
 
 #ifndef FFT_BITREV60
 #define FFT_BITREV60
 static const opus_int16 fft_bitrev60[60] = {
-0, 15, 30, 45, 5, 20, 35, 50, 10, 25, 40, 55, 1, 16, 31,
-46, 6, 21, 36, 51, 11, 26, 41, 56, 2, 17, 32, 47, 7, 22,
-37, 52, 12, 27, 42, 57, 3, 18, 33, 48, 8, 23, 38, 53, 13,
-28, 43, 58, 4, 19, 34, 49, 9, 24, 39, 54, 14, 29, 44, 59,
+0, 12, 24, 36, 48, 4, 16, 28, 40, 52, 8, 20, 32, 44, 56,
+1, 13, 25, 37, 49, 5, 17, 29, 41, 53, 9, 21, 33, 45, 57,
+2, 14, 26, 38, 50, 6, 18, 30, 42, 54, 10, 22, 34, 46, 58,
+3, 15, 27, 39, 51, 7, 19, 31, 43, 55, 11, 23, 35, 47, 59,
 };
 #endif
 
@@ -426,10 +431,17 @@ static const opus_int16 fft_bitrev60[60] = {
 #define FFT_STATE48000_960_0
 static const kiss_fft_state fft_state48000_960_0 = {
 480,    /* nfft */
+17476,    /* scale */
+8,      /* scale_shift */
 -1,     /* shift */
-{4, 120, 4, 30, 2, 15, 3, 5, 5, 1, 0, 0, 0, 0, 0, 0, }, /* factors */
+{5, 96, 3, 32, 4, 8, 2, 4, 4, 1, 0, 0, 0, 0, 0, 0, },    /* factors */
 fft_bitrev480,  /* bitrev */
 fft_twiddles48000_960,  /* bitrev */
+#ifdef OVERRIDE_FFT
+(arch_fft_state *)&cfg_arch_480,
+#else
+NULL,
+#endif
 };
 #endif
 
@@ -437,10 +449,17 @@ fft_twiddles48000_960,  /* bitrev */
 #define FFT_STATE48000_960_1
 static const kiss_fft_state fft_state48000_960_1 = {
 240,    /* nfft */
+17476,    /* scale */
+7,      /* scale_shift */
 1,      /* shift */
-{4, 60, 4, 15, 3, 5, 5, 1, 0, 0, 0, 0, 0, 0, 0, 0, },   /* factors */
+{5, 48, 3, 16, 4, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, },    /* factors */
 fft_bitrev240,  /* bitrev */
 fft_twiddles48000_960,  /* bitrev */
+#ifdef OVERRIDE_FFT
+(arch_fft_state *)&cfg_arch_240,
+#else
+NULL,
+#endif
 };
 #endif
 
@@ -448,10 +467,17 @@ fft_twiddles48000_960,  /* bitrev */
 #define FFT_STATE48000_960_2
 static const kiss_fft_state fft_state48000_960_2 = {
 120,    /* nfft */
+17476,    /* scale */
+6,      /* scale_shift */
 2,      /* shift */
-{4, 30, 2, 15, 3, 5, 5, 1, 0, 0, 0, 0, 0, 0, 0, 0, },   /* factors */
+{5, 24, 3, 8, 2, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, },    /* factors */
 fft_bitrev120,  /* bitrev */
 fft_twiddles48000_960,  /* bitrev */
+#ifdef OVERRIDE_FFT
+(arch_fft_state *)&cfg_arch_120,
+#else
+NULL,
+#endif
 };
 #endif
 
@@ -459,10 +485,17 @@ fft_twiddles48000_960,  /* bitrev */
 #define FFT_STATE48000_960_3
 static const kiss_fft_state fft_state48000_960_3 = {
 60,     /* nfft */
+17476,    /* scale */
+5,      /* scale_shift */
 3,      /* shift */
-{4, 15, 3, 5, 5, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },    /* factors */
+{5, 12, 3, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },    /* factors */
 fft_bitrev60,   /* bitrev */
 fft_twiddles48000_960,  /* bitrev */
+#ifdef OVERRIDE_FFT
+(arch_fft_state *)&cfg_arch_60,
+#else
+NULL,
+#endif
 };
 #endif
 
@@ -470,104 +503,368 @@ fft_twiddles48000_960,  /* bitrev */
 
 #ifndef MDCT_TWIDDLES960
 #define MDCT_TWIDDLES960
-static const opus_val16 mdct_twiddles960[481] = {
-32767, 32767, 32767, 32767, 32766,
-32763, 32762, 32759, 32757, 32753,
-32751, 32747, 32743, 32738, 32733,
-32729, 32724, 32717, 32711, 32705,
-32698, 32690, 32683, 32676, 32667,
-32658, 32650, 32640, 32631, 32620,
-32610, 32599, 32588, 32577, 32566,
-32554, 32541, 32528, 32515, 32502,
-32487, 32474, 32459, 32444, 32429,
-32413, 32397, 32381, 32364, 32348,
-32331, 32313, 32294, 32277, 32257,
-32239, 32219, 32200, 32180, 32159,
-32138, 32118, 32096, 32074, 32051,
-32029, 32006, 31984, 31960, 31936,
-31912, 31888, 31863, 31837, 31812,
-31786, 31760, 31734, 31707, 31679,
-31652, 31624, 31596, 31567, 31539,
-31508, 31479, 31450, 31419, 31388,
-31357, 31326, 31294, 31262, 31230,
-31198, 31164, 31131, 31097, 31063,
-31030, 30994, 30959, 30924, 30889,
-30853, 30816, 30779, 30743, 30705,
-30668, 30629, 30592, 30553, 30515,
-30475, 30435, 30396, 30356, 30315,
-30274, 30233, 30191, 30149, 30107,
-30065, 30022, 29979, 29936, 29891,
-29847, 29803, 29758, 29713, 29668,
-29622, 29577, 29529, 29483, 29436,
-29390, 29341, 29293, 29246, 29197,
-29148, 29098, 29050, 29000, 28949,
-28899, 28848, 28797, 28746, 28694,
-28642, 28590, 28537, 28485, 28432,
-28378, 28324, 28271, 28217, 28162,
-28106, 28051, 27995, 27940, 27884,
-27827, 27770, 27713, 27657, 27598,
-27540, 27481, 27423, 27365, 27305,
-27246, 27187, 27126, 27066, 27006,
-26945, 26883, 26822, 26760, 26698,
-26636, 26574, 26510, 26448, 26383,
-26320, 26257, 26191, 26127, 26062,
-25997, 25931, 25866, 25800, 25734,
-25667, 25601, 25533, 25466, 25398,
-25330, 25262, 25194, 25125, 25056,
-24987, 24917, 24848, 24778, 24707,
-24636, 24566, 24495, 24424, 24352,
-24280, 24208, 24135, 24063, 23990,
-23917, 23842, 23769, 23695, 23622,
-23546, 23472, 23398, 23322, 23246,
-23171, 23095, 23018, 22942, 22866,
-22788, 22711, 22634, 22557, 22478,
-22400, 22322, 22244, 22165, 22085,
-22006, 21927, 21846, 21766, 21687,
-21606, 21524, 21443, 21363, 21282,
-21199, 21118, 21035, 20954, 20870,
-20788, 20705, 20621, 20538, 20455,
-20371, 20286, 20202, 20118, 20034,
-19947, 19863, 19777, 19692, 19606,
-19520, 19434, 19347, 19260, 19174,
-19088, 18999, 18911, 18825, 18737,
-18648, 18560, 18472, 18384, 18294,
-18205, 18116, 18025, 17936, 17846,
-17757, 17666, 17576, 17485, 17395,
-17303, 17212, 17122, 17030, 16937,
-16846, 16755, 16662, 16569, 16477,
-16385, 16291, 16198, 16105, 16012,
-15917, 15824, 15730, 15636, 15541,
-15447, 15352, 15257, 15162, 15067,
-14973, 14875, 14781, 14685, 14589,
-14493, 14396, 14300, 14204, 14107,
-14010, 13914, 13815, 13718, 13621,
-13524, 13425, 13328, 13230, 13133,
-13033, 12935, 12836, 12738, 12638,
-12540, 12441, 12341, 12241, 12142,
-12044, 11943, 11843, 11744, 11643,
-11542, 11442, 11342, 11241, 11139,
-11039, 10939, 10836, 10736, 10635,
-10534, 10431, 10330, 10228, 10127,
-10024, 9921, 9820, 9718, 9614,
-9512, 9410, 9306, 9204, 9101,
-8998, 8895, 8791, 8689, 8585,
-8481, 8377, 8274, 8171, 8067,
-7962, 7858, 7753, 7650, 7545,
-7441, 7336, 7231, 7129, 7023,
-6917, 6813, 6709, 6604, 6498,
-6393, 6288, 6182, 6077, 5973,
-5867, 5760, 5656, 5549, 5445,
-5339, 5232, 5127, 5022, 4914,
-4809, 4703, 4596, 4490, 4384,
-4278, 4171, 4065, 3958, 3852,
-3745, 3640, 3532, 3426, 3318,
-3212, 3106, 2998, 2891, 2786,
-2679, 2570, 2465, 2358, 2251,
-2143, 2037, 1929, 1823, 1715,
-1609, 1501, 1393, 1287, 1180,
-1073, 964, 858, 751, 644,
-535, 429, 322, 214, 107,
-0, };
+static const opus_val16 mdct_twiddles960[1800] = {
+32767, 32767, 32767, 32766, 32765,
+32763, 32761, 32759, 32756, 32753,
+32750, 32746, 32742, 32738, 32733,
+32728, 32722, 32717, 32710, 32704,
+32697, 32690, 32682, 32674, 32666,
+32657, 32648, 32639, 32629, 32619,
+32609, 32598, 32587, 32576, 32564,
+32552, 32539, 32526, 32513, 32500,
+32486, 32472, 32457, 32442, 32427,
+32411, 32395, 32379, 32362, 32345,
+32328, 32310, 32292, 32274, 32255,
+32236, 32217, 32197, 32177, 32157,
+32136, 32115, 32093, 32071, 32049,
+32027, 32004, 31981, 31957, 31933,
+31909, 31884, 31859, 31834, 31809,
+31783, 31756, 31730, 31703, 31676,
+31648, 31620, 31592, 31563, 31534,
+31505, 31475, 31445, 31415, 31384,
+31353, 31322, 31290, 31258, 31226,
+31193, 31160, 31127, 31093, 31059,
+31025, 30990, 30955, 30920, 30884,
+30848, 30812, 30775, 30738, 30701,
+30663, 30625, 30587, 30548, 30509,
+30470, 30430, 30390, 30350, 30309,
+30269, 30227, 30186, 30144, 30102,
+30059, 30016, 29973, 29930, 29886,
+29842, 29797, 29752, 29707, 29662,
+29616, 29570, 29524, 29477, 29430,
+29383, 29335, 29287, 29239, 29190,
+29142, 29092, 29043, 28993, 28943,
+28892, 28842, 28791, 28739, 28688,
+28636, 28583, 28531, 28478, 28425,
+28371, 28317, 28263, 28209, 28154,
+28099, 28044, 27988, 27932, 27876,
+27820, 27763, 27706, 27648, 27591,
+27533, 27474, 27416, 27357, 27298,
+27238, 27178, 27118, 27058, 26997,
+26936, 26875, 26814, 26752, 26690,
+26628, 26565, 26502, 26439, 26375,
+26312, 26247, 26183, 26119, 26054,
+25988, 25923, 25857, 25791, 25725,
+25658, 25592, 25524, 25457, 25389,
+25322, 25253, 25185, 25116, 25047,
+24978, 24908, 24838, 24768, 24698,
+24627, 24557, 24485, 24414, 24342,
+24270, 24198, 24126, 24053, 23980,
+23907, 23834, 23760, 23686, 23612,
+23537, 23462, 23387, 23312, 23237,
+23161, 23085, 23009, 22932, 22856,
+22779, 22701, 22624, 22546, 22468,
+22390, 22312, 22233, 22154, 22075,
+21996, 21916, 21836, 21756, 21676,
+21595, 21515, 21434, 21352, 21271,
+21189, 21107, 21025, 20943, 20860,
+20777, 20694, 20611, 20528, 20444,
+20360, 20276, 20192, 20107, 20022,
+19937, 19852, 19767, 19681, 19595,
+19509, 19423, 19336, 19250, 19163,
+19076, 18988, 18901, 18813, 18725,
+18637, 18549, 18460, 18372, 18283,
+18194, 18104, 18015, 17925, 17835,
+17745, 17655, 17565, 17474, 17383,
+17292, 17201, 17110, 17018, 16927,
+16835, 16743, 16650, 16558, 16465,
+16372, 16279, 16186, 16093, 15999,
+15906, 15812, 15718, 15624, 15529,
+15435, 15340, 15245, 15150, 15055,
+14960, 14864, 14769, 14673, 14577,
+14481, 14385, 14288, 14192, 14095,
+13998, 13901, 13804, 13706, 13609,
+13511, 13414, 13316, 13218, 13119,
+13021, 12923, 12824, 12725, 12626,
+12527, 12428, 12329, 12230, 12130,
+12030, 11930, 11831, 11730, 11630,
+11530, 11430, 11329, 11228, 11128,
+11027, 10926, 10824, 10723, 10622,
+10520, 10419, 10317, 10215, 10113,
+10011, 9909, 9807, 9704, 9602,
+9499, 9397, 9294, 9191, 9088,
+8985, 8882, 8778, 8675, 8572,
+8468, 8364, 8261, 8157, 8053,
+7949, 7845, 7741, 7637, 7532,
+7428, 7323, 7219, 7114, 7009,
+6905, 6800, 6695, 6590, 6485,
+6380, 6274, 6169, 6064, 5958,
+5853, 5747, 5642, 5536, 5430,
+5325, 5219, 5113, 5007, 4901,
+4795, 4689, 4583, 4476, 4370,
+4264, 4157, 4051, 3945, 3838,
+3732, 3625, 3518, 3412, 3305,
+3198, 3092, 2985, 2878, 2771,
+2664, 2558, 2451, 2344, 2237,
+2130, 2023, 1916, 1809, 1702,
+1594, 1487, 1380, 1273, 1166,
+1059, 952, 844, 737, 630,
+523, 416, 308, 201, 94,
+-13, -121, -228, -335, -442,
+-550, -657, -764, -871, -978,
+-1086, -1193, -1300, -1407, -1514,
+-1621, -1728, -1835, -1942, -2049,
+-2157, -2263, -2370, -2477, -2584,
+-2691, -2798, -2905, -3012, -3118,
+-3225, -3332, -3439, -3545, -3652,
+-3758, -3865, -3971, -4078, -4184,
+-4290, -4397, -4503, -4609, -4715,
+-4821, -4927, -5033, -5139, -5245,
+-5351, -5457, -5562, -5668, -5774,
+-5879, -5985, -6090, -6195, -6301,
+-6406, -6511, -6616, -6721, -6826,
+-6931, -7036, -7140, -7245, -7349,
+-7454, -7558, -7663, -7767, -7871,
+-7975, -8079, -8183, -8287, -8390,
+-8494, -8597, -8701, -8804, -8907,
+-9011, -9114, -9217, -9319, -9422,
+-9525, -9627, -9730, -9832, -9934,
+-10037, -10139, -10241, -10342, -10444,
+-10546, -10647, -10748, -10850, -10951,
+-11052, -11153, -11253, -11354, -11455,
+-11555, -11655, -11756, -11856, -11955,
+-12055, -12155, -12254, -12354, -12453,
+-12552, -12651, -12750, -12849, -12947,
+-13046, -13144, -13242, -13340, -13438,
+-13536, -13633, -13731, -13828, -13925,
+-14022, -14119, -14216, -14312, -14409,
+-14505, -14601, -14697, -14793, -14888,
+-14984, -15079, -15174, -15269, -15364,
+-15459, -15553, -15647, -15741, -15835,
+-15929, -16023, -16116, -16210, -16303,
+-16396, -16488, -16581, -16673, -16766,
+-16858, -16949, -17041, -17133, -17224,
+-17315, -17406, -17497, -17587, -17678,
+-17768, -17858, -17948, -18037, -18127,
+-18216, -18305, -18394, -18483, -18571,
+-18659, -18747, -18835, -18923, -19010,
+-19098, -19185, -19271, -19358, -19444,
+-19531, -19617, -19702, -19788, -19873,
+-19959, -20043, -20128, -20213, -20297,
+-20381, -20465, -20549, -20632, -20715,
+-20798, -20881, -20963, -21046, -21128,
+-21210, -21291, -21373, -21454, -21535,
+-21616, -21696, -21776, -21856, -21936,
+-22016, -22095, -22174, -22253, -22331,
+-22410, -22488, -22566, -22643, -22721,
+-22798, -22875, -22951, -23028, -23104,
+-23180, -23256, -23331, -23406, -23481,
+-23556, -23630, -23704, -23778, -23852,
+-23925, -23998, -24071, -24144, -24216,
+-24288, -24360, -24432, -24503, -24574,
+-24645, -24716, -24786, -24856, -24926,
+-24995, -25064, -25133, -25202, -25270,
+-25339, -25406, -25474, -25541, -25608,
+-25675, -25742, -25808, -25874, -25939,
+-26005, -26070, -26135, -26199, -26264,
+-26327, -26391, -26455, -26518, -26581,
+-26643, -26705, -26767, -26829, -26891,
+-26952, -27013, -27073, -27133, -27193,
+-27253, -27312, -27372, -27430, -27489,
+-27547, -27605, -27663, -27720, -27777,
+-27834, -27890, -27946, -28002, -28058,
+-28113, -28168, -28223, -28277, -28331,
+-28385, -28438, -28491, -28544, -28596,
+-28649, -28701, -28752, -28803, -28854,
+-28905, -28955, -29006, -29055, -29105,
+-29154, -29203, -29251, -29299, -29347,
+-29395, -29442, -29489, -29535, -29582,
+-29628, -29673, -29719, -29764, -29808,
+-29853, -29897, -29941, -29984, -30027,
+-30070, -30112, -30154, -30196, -30238,
+-30279, -30320, -30360, -30400, -30440,
+-30480, -30519, -30558, -30596, -30635,
+-30672, -30710, -30747, -30784, -30821,
+-30857, -30893, -30929, -30964, -30999,
+-31033, -31068, -31102, -31135, -31168,
+-31201, -31234, -31266, -31298, -31330,
+-31361, -31392, -31422, -31453, -31483,
+-31512, -31541, -31570, -31599, -31627,
+-31655, -31682, -31710, -31737, -31763,
+-31789, -31815, -31841, -31866, -31891,
+-31915, -31939, -31963, -31986, -32010,
+-32032, -32055, -32077, -32099, -32120,
+-32141, -32162, -32182, -32202, -32222,
+-32241, -32260, -32279, -32297, -32315,
+-32333, -32350, -32367, -32383, -32399,
+-32415, -32431, -32446, -32461, -32475,
+-32489, -32503, -32517, -32530, -32542,
+-32555, -32567, -32579, -32590, -32601,
+-32612, -32622, -32632, -32641, -32651,
+-32659, -32668, -32676, -32684, -32692,
+-32699, -32706, -32712, -32718, -32724,
+-32729, -32734, -32739, -32743, -32747,
+-32751, -32754, -32757, -32760, -32762,
+-32764, -32765, -32767, -32767, -32767,
+32767, 32767, 32765, 32761, 32756,
+32750, 32742, 32732, 32722, 32710,
+32696, 32681, 32665, 32647, 32628,
+32608, 32586, 32562, 32538, 32512,
+32484, 32455, 32425, 32393, 32360,
+32326, 32290, 32253, 32214, 32174,
+32133, 32090, 32046, 32001, 31954,
+31906, 31856, 31805, 31753, 31700,
+31645, 31588, 31530, 31471, 31411,
+31349, 31286, 31222, 31156, 31089,
+31020, 30951, 30880, 30807, 30733,
+30658, 30582, 30504, 30425, 30345,
+30263, 30181, 30096, 30011, 29924,
+29836, 29747, 29656, 29564, 29471,
+29377, 29281, 29184, 29086, 28987,
+28886, 28784, 28681, 28577, 28471,
+28365, 28257, 28147, 28037, 27925,
+27812, 27698, 27583, 27467, 27349,
+27231, 27111, 26990, 26868, 26744,
+26620, 26494, 26367, 26239, 26110,
+25980, 25849, 25717, 25583, 25449,
+25313, 25176, 25038, 24900, 24760,
+24619, 24477, 24333, 24189, 24044,
+23898, 23751, 23602, 23453, 23303,
+23152, 22999, 22846, 22692, 22537,
+22380, 22223, 22065, 21906, 21746,
+21585, 21423, 21261, 21097, 20933,
+20767, 20601, 20434, 20265, 20096,
+19927, 19756, 19584, 19412, 19239,
+19065, 18890, 18714, 18538, 18361,
+18183, 18004, 17824, 17644, 17463,
+17281, 17098, 16915, 16731, 16546,
+16361, 16175, 15988, 15800, 15612,
+15423, 15234, 15043, 14852, 14661,
+14469, 14276, 14083, 13889, 13694,
+13499, 13303, 13107, 12910, 12713,
+12515, 12317, 12118, 11918, 11718,
+11517, 11316, 11115, 10913, 10710,
+10508, 10304, 10100, 9896, 9691,
+9486, 9281, 9075, 8869, 8662,
+8455, 8248, 8040, 7832, 7623,
+7415, 7206, 6996, 6787, 6577,
+6366, 6156, 5945, 5734, 5523,
+5311, 5100, 4888, 4675, 4463,
+4251, 4038, 3825, 3612, 3399,
+3185, 2972, 2758, 2544, 2330,
+2116, 1902, 1688, 1474, 1260,
+1045, 831, 617, 402, 188,
+-27, -241, -456, -670, -885,
+-1099, -1313, -1528, -1742, -1956,
+-2170, -2384, -2598, -2811, -3025,
+-3239, -3452, -3665, -3878, -4091,
+-4304, -4516, -4728, -4941, -5153,
+-5364, -5576, -5787, -5998, -6209,
+-6419, -6629, -6839, -7049, -7258,
+-7467, -7676, -7884, -8092, -8300,
+-8507, -8714, -8920, -9127, -9332,
+-9538, -9743, -9947, -10151, -10355,
+-10558, -10761, -10963, -11165, -11367,
+-11568, -11768, -11968, -12167, -12366,
+-12565, -12762, -12960, -13156, -13352,
+-13548, -13743, -13937, -14131, -14324,
+-14517, -14709, -14900, -15091, -15281,
+-15470, -15659, -15847, -16035, -16221,
+-16407, -16593, -16777, -16961, -17144,
+-17326, -17508, -17689, -17869, -18049,
+-18227, -18405, -18582, -18758, -18934,
+-19108, -19282, -19455, -19627, -19799,
+-19969, -20139, -20308, -20475, -20642,
+-20809, -20974, -21138, -21301, -21464,
+-21626, -21786, -21946, -22105, -22263,
+-22420, -22575, -22730, -22884, -23037,
+-23189, -23340, -23490, -23640, -23788,
+-23935, -24080, -24225, -24369, -24512,
+-24654, -24795, -24934, -25073, -25211,
+-25347, -25482, -25617, -25750, -25882,
+-26013, -26143, -26272, -26399, -26526,
+-26651, -26775, -26898, -27020, -27141,
+-27260, -27379, -27496, -27612, -27727,
+-27841, -27953, -28065, -28175, -28284,
+-28391, -28498, -28603, -28707, -28810,
+-28911, -29012, -29111, -29209, -29305,
+-29401, -29495, -29587, -29679, -29769,
+-29858, -29946, -30032, -30118, -30201,
+-30284, -30365, -30445, -30524, -30601,
+-30677, -30752, -30825, -30897, -30968,
+-31038, -31106, -31172, -31238, -31302,
+-31365, -31426, -31486, -31545, -31602,
+-31658, -31713, -31766, -31818, -31869,
+-31918, -31966, -32012, -32058, -32101,
+-32144, -32185, -32224, -32262, -32299,
+-32335, -32369, -32401, -32433, -32463,
+-32491, -32518, -32544, -32568, -32591,
+-32613, -32633, -32652, -32669, -32685,
+-32700, -32713, -32724, -32735, -32744,
+-32751, -32757, -32762, -32766, -32767,
+32767, 32764, 32755, 32741, 32720,
+32694, 32663, 32626, 32583, 32535,
+32481, 32421, 32356, 32286, 32209,
+32128, 32041, 31948, 31850, 31747,
+31638, 31523, 31403, 31278, 31148,
+31012, 30871, 30724, 30572, 30415,
+30253, 30086, 29913, 29736, 29553,
+29365, 29172, 28974, 28771, 28564,
+28351, 28134, 27911, 27684, 27452,
+27216, 26975, 26729, 26478, 26223,
+25964, 25700, 25432, 25159, 24882,
+24601, 24315, 24026, 23732, 23434,
+23133, 22827, 22517, 22204, 21886,
+21565, 21240, 20912, 20580, 20244,
+19905, 19563, 19217, 18868, 18516,
+18160, 17802, 17440, 17075, 16708,
+16338, 15964, 15588, 15210, 14829,
+14445, 14059, 13670, 13279, 12886,
+12490, 12093, 11693, 11291, 10888,
+10482, 10075, 9666, 9255, 8843,
+8429, 8014, 7597, 7180, 6760,
+6340, 5919, 5496, 5073, 4649,
+4224, 3798, 3372, 2945, 2517,
+2090, 1661, 1233, 804, 375,
+-54, -483, -911, -1340, -1768,
+-2197, -2624, -3052, -3479, -3905,
+-4330, -4755, -5179, -5602, -6024,
+-6445, -6865, -7284, -7702, -8118,
+-8533, -8946, -9358, -9768, -10177,
+-10584, -10989, -11392, -11793, -12192,
+-12589, -12984, -13377, -13767, -14155,
+-14541, -14924, -15305, -15683, -16058,
+-16430, -16800, -17167, -17531, -17892,
+-18249, -18604, -18956, -19304, -19649,
+-19990, -20329, -20663, -20994, -21322,
+-21646, -21966, -22282, -22595, -22904,
+-23208, -23509, -23806, -24099, -24387,
+-24672, -24952, -25228, -25499, -25766,
+-26029, -26288, -26541, -26791, -27035,
+-27275, -27511, -27741, -27967, -28188,
+-28405, -28616, -28823, -29024, -29221,
+-29412, -29599, -29780, -29957, -30128,
+-30294, -30455, -30611, -30761, -30906,
+-31046, -31181, -31310, -31434, -31552,
+-31665, -31773, -31875, -31972, -32063,
+-32149, -32229, -32304, -32373, -32437,
+-32495, -32547, -32594, -32635, -32671,
+-32701, -32726, -32745, -32758, -32766,
+32767, 32754, 32717, 32658, 32577,
+32473, 32348, 32200, 32029, 31837,
+31624, 31388, 31131, 30853, 30553,
+30232, 29891, 29530, 29148, 28746,
+28324, 27883, 27423, 26944, 26447,
+25931, 25398, 24847, 24279, 23695,
+23095, 22478, 21846, 21199, 20538,
+19863, 19174, 18472, 17757, 17030,
+16291, 15541, 14781, 14010, 13230,
+12441, 11643, 10837, 10024, 9204,
+8377, 7545, 6708, 5866, 5020,
+4171, 3319, 2464, 1608, 751,
+-107, -965, -1822, -2678, -3532,
+-4383, -5232, -6077, -6918, -7754,
+-8585, -9409, -10228, -11039, -11843,
+-12639, -13426, -14204, -14972, -15730,
+-16477, -17213, -17937, -18648, -19347,
+-20033, -20705, -21363, -22006, -22634,
+-23246, -23843, -24423, -24986, -25533,
+-26062, -26573, -27066, -27540, -27995,
+-28431, -28848, -29245, -29622, -29979,
+-30315, -30630, -30924, -31197, -31449,
+-31679, -31887, -32074, -32239, -32381,
+-32501, -32600, -32675, -32729, -32759,
+};
 #endif
 
 static const CELTMode mode48000_960_120 = {
diff --git a/drivers/opus/celt/static_modes_fixed_arm_ne10.h b/drivers/opus/celt/static_modes_fixed_arm_ne10.h
new file mode 100644
index 0000000000..b8ef0cee98
--- /dev/null
+++ b/drivers/opus/celt/static_modes_fixed_arm_ne10.h
@@ -0,0 +1,388 @@
+/* The contents of this file was automatically generated by
+ * dump_mode_arm_ne10.c with arguments: 48000 960
+ * It contains static definitions for some pre-defined modes. */
+#include <NE10_init.h>
+
+#ifndef NE10_FFT_PARAMS48000_960
+#define NE10_FFT_PARAMS48000_960
+static const ne10_int32_t ne10_factors_480[64] = {
+4, 40, 4, 30, 2, 15, 5, 3, 3, 1, 1, 0, 0, 0, 0,
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, };
+static const ne10_int32_t ne10_factors_240[64] = {
+3, 20, 4, 15, 5, 3, 3, 1, 1, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, };
+static const ne10_int32_t ne10_factors_120[64] = {
+3, 10, 2, 15, 5, 3, 3, 1, 1, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, };
+static const ne10_int32_t ne10_factors_60[64] = {
+2, 5, 5, 3, 3, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, };
+static const ne10_fft_cpx_int32_t ne10_twiddles_480[480] = {
+{0,0}, {2147483647,0}, {2147483647,0},
+{2147483647,0}, {1961823921,-873460313}, {1436946998,-1595891394},
+{2147483647,0}, {1436946998,-1595891394}, {-224473265,-2135719496},
+{2147483647,0}, {663608871,-2042378339}, {-1737350854,-1262259096},
+{2147483647,0}, {-224473265,-2135719496}, {-2100555935,446487152},
+{2147483647,0}, {2100555974,-446486968}, {1961823921,-873460313},
+{1737350743,-1262259248}, {1436946998,-1595891394}, {1073741769,-1859775424},
+{663608871,-2042378339}, {224473078,-2135719516}, {-224473265,-2135719496},
+{-663609049,-2042378281}, {-1073741932,-1859775330}, {-1436947137,-1595891268},
+{-1737350854,-1262259096}, {-1961823997,-873460141}, {-2100556013,-446486785},
+{2147483647,0}, {2144540595,-112390613}, {2135719506,-224473172},
+{2121044558,-335940465}, {2100555974,-446486968}, {2074309912,-555809682},
+{2042378310,-663608960}, {2004848691,-769589332}, {1961823921,-873460313},
+{1913421927,-974937199}, {1859775377,-1073741851}, {1801031311,-1169603450},
+{1737350743,-1262259248}, {1668908218,-1351455280}, {1595891331,-1436947067},
+{1518500216,-1518500282}, {1436946998,-1595891394}, {1351455207,-1668908277},
+{1262259172,-1737350799}, {1169603371,-1801031362}, {1073741769,-1859775424},
+{974937230,-1913421912}, {873460227,-1961823959}, {769589125,-2004848771},
+{663608871,-2042378339}, {555809715,-2074309903}, {446486876,-2100555994},
+{335940246,-2121044593}, {224473078,-2135719516}, {112390647,-2144540593},
+{2147483647,0}, {2135719506,-224473172}, {2100555974,-446486968},
+{2042378310,-663608960}, {1961823921,-873460313}, {1859775377,-1073741851},
+{1737350743,-1262259248}, {1595891331,-1436947067}, {1436946998,-1595891394},
+{1262259172,-1737350799}, {1073741769,-1859775424}, {873460227,-1961823959},
+{663608871,-2042378339}, {446486876,-2100555994}, {224473078,-2135719516},
+{-94,-2147483647}, {-224473265,-2135719496}, {-446487060,-2100555955},
+{-663609049,-2042378281}, {-873460398,-1961823883}, {-1073741932,-1859775330},
+{-1262259116,-1737350839}, {-1436947137,-1595891268}, {-1595891628,-1436946738},
+{-1737350854,-1262259096}, {-1859775343,-1073741910}, {-1961823997,-873460141},
+{-2042378447,-663608538}, {-2100556013,-446486785}, {-2135719499,-224473240},
+{2147483647,0}, {2121044558,-335940465}, {2042378310,-663608960},
+{1913421927,-974937199}, {1737350743,-1262259248}, {1518500216,-1518500282},
+{1262259172,-1737350799}, {974937230,-1913421912}, {663608871,-2042378339},
+{335940246,-2121044593}, {-94,-2147483647}, {-335940431,-2121044564},
+{-663609049,-2042378281}, {-974937397,-1913421827}, {-1262259116,-1737350839},
+{-1518500258,-1518500240}, {-1737350854,-1262259096}, {-1913422071,-974936918},
+{-2042378447,-663608538}, {-2121044568,-335940406}, {-2147483647,188},
+{-2121044509,335940777}, {-2042378331,663608895}, {-1913421900,974937252},
+{-1737350633,1262259400}, {-1518499993,1518500506}, {-1262258813,1737351059},
+{-974936606,1913422229}, {-663609179,2042378239}, {-335940566,2121044542},
+{2147483647,0}, {2147299667,-28109693}, {2146747758,-56214570},
+{2145828015,-84309815}, {2144540595,-112390613}, {2142885719,-140452154},
+{2140863671,-168489630}, {2138474797,-196498235}, {2135719506,-224473172},
+{2132598271,-252409646}, {2129111626,-280302871}, {2125260168,-308148068},
+{2121044558,-335940465}, {2116465518,-363675300}, {2111523833,-391347822},
+{2106220349,-418953288}, {2100555974,-446486968}, {2094531681,-473944146},
+{2088148500,-501320115}, {2081407525,-528610186}, {2074309912,-555809682},
+{2066856885,-582913912}, {2059049696,-609918325}, {2050889698,-636818231},
+{2042378310,-663608960}, {2033516972,-690285983}, {2024307180,-716844791},
+{2014750533,-743280770}, {2004848691,-769589332}, {1994603329,-795766029},
+{1984016179,-821806435}, {1973089077,-847706028}, {1961823921,-873460313},
+{1950222618,-899064934}, {1938287127,-924515564}, {1926019520,-949807783},
+{1913421927,-974937199}, {1900496481,-999899565}, {1887245364,-1024690661},
+{1873670877,-1049306180}, {1859775377,-1073741851}, {1845561215,-1097993541},
+{1831030826,-1122057097}, {1816186632,-1145928502}, {1801031311,-1169603450},
+{1785567394,-1193077993}, {1769797456,-1216348214}, {1753724345,-1239409914},
+{1737350743,-1262259248}, {1720679456,-1284892300}, {1703713340,-1307305194},
+{1686455222,-1329494189}, {1668908218,-1351455280}, {1651075255,-1373184807},
+{1632959307,-1394679144}, {1614563642,-1415934412}, {1595891331,-1436947067},
+{1576945572,-1457713510}, {1557729613,-1478230181}, {1538246655,-1498493658},
+{1518500216,-1518500282}, {1498493590,-1538246721}, {1478230113,-1557729677},
+{1457713441,-1576945636}, {1436946998,-1595891394}, {1415934341,-1614563704},
+{1394679073,-1632959368}, {1373184735,-1651075315}, {1351455207,-1668908277},
+{1329494115,-1686455280}, {1307305120,-1703713397}, {1284892225,-1720679512},
+{1262259172,-1737350799}, {1239409837,-1753724400}, {1216348136,-1769797510},
+{1193077915,-1785567446}, {1169603371,-1801031362}, {1145928423,-1816186682},
+{1122057017,-1831030875}, {1097993571,-1845561197}, {1073741769,-1859775424},
+{1049305987,-1873670985}, {1024690635,-1887245378}, {999899482,-1900496524},
+{974937230,-1913421912}, {949807699,-1926019561}, {924515422,-1938287195},
+{899064965,-1950222603}, {873460227,-1961823959}, {847705824,-1973089164},
+{821806407,-1984016190}, {795765941,-1994603364}, {769589125,-2004848771},
+{743280682,-2014750566}, {716844642,-2024307233}, {690286016,-2033516961},
+{663608871,-2042378339}, {636818019,-2050889764}, {609918296,-2059049705},
+{582913822,-2066856911}, {555809715,-2074309903}, {528610126,-2081407540},
+{501319962,-2088148536}, {473944148,-2094531680}, {446486876,-2100555994},
+{418953102,-2106220386}, {391347792,-2111523838}, {363675176,-2116465540},
+{335940246,-2121044593}, {308148006,-2125260177}, {280302715,-2129111646},
+{252409648,-2132598271}, {224473078,-2135719516}, {196498046,-2138474814},
+{168489600,-2140863674}, {140452029,-2142885728}, {112390647,-2144540593},
+{84309753,-2145828017}, {56214412,-2146747762}, {28109695,-2147299667},
+{2147483647,0}, {2146747758,-56214570}, {2144540595,-112390613},
+{2140863671,-168489630}, {2135719506,-224473172}, {2129111626,-280302871},
+{2121044558,-335940465}, {2111523833,-391347822}, {2100555974,-446486968},
+{2088148500,-501320115}, {2074309912,-555809682}, {2059049696,-609918325},
+{2042378310,-663608960}, {2024307180,-716844791}, {2004848691,-769589332},
+{1984016179,-821806435}, {1961823921,-873460313}, {1938287127,-924515564},
+{1913421927,-974937199}, {1887245364,-1024690661}, {1859775377,-1073741851},
+{1831030826,-1122057097}, {1801031311,-1169603450}, {1769797456,-1216348214},
+{1737350743,-1262259248}, {1703713340,-1307305194}, {1668908218,-1351455280},
+{1632959307,-1394679144}, {1595891331,-1436947067}, {1557729613,-1478230181},
+{1518500216,-1518500282}, {1478230113,-1557729677}, {1436946998,-1595891394},
+{1394679073,-1632959368}, {1351455207,-1668908277}, {1307305120,-1703713397},
+{1262259172,-1737350799}, {1216348136,-1769797510}, {1169603371,-1801031362},
+{1122057017,-1831030875}, {1073741769,-1859775424}, {1024690635,-1887245378},
+{974937230,-1913421912}, {924515422,-1938287195}, {873460227,-1961823959},
+{821806407,-1984016190}, {769589125,-2004848771}, {716844642,-2024307233},
+{663608871,-2042378339}, {609918296,-2059049705}, {555809715,-2074309903},
+{501319962,-2088148536}, {446486876,-2100555994}, {391347792,-2111523838},
+{335940246,-2121044593}, {280302715,-2129111646}, {224473078,-2135719516},
+{168489600,-2140863674}, {112390647,-2144540593}, {56214412,-2146747762},
+{-94,-2147483647}, {-56214600,-2146747757}, {-112390835,-2144540584},
+{-168489787,-2140863659}, {-224473265,-2135719496}, {-280302901,-2129111622},
+{-335940431,-2121044564}, {-391347977,-2111523804}, {-446487060,-2100555955},
+{-501320144,-2088148493}, {-555809896,-2074309855}, {-609918476,-2059049651},
+{-663609049,-2042378281}, {-716844819,-2024307170}, {-769589300,-2004848703},
+{-821806581,-1984016118}, {-873460398,-1961823883}, {-924515591,-1938287114},
+{-974937397,-1913421827}, {-1024690575,-1887245411}, {-1073741932,-1859775330},
+{-1122057395,-1831030643}, {-1169603421,-1801031330}, {-1216348291,-1769797403},
+{-1262259116,-1737350839}, {-1307305268,-1703713283}, {-1351455453,-1668908078},
+{-1394679021,-1632959413}, {-1436947137,-1595891268}, {-1478230435,-1557729372},
+{-1518500258,-1518500240}, {-1557729742,-1478230045}, {-1595891628,-1436946738},
+{-1632959429,-1394679001}, {-1668908417,-1351455035}, {-1703713298,-1307305248},
+{-1737350854,-1262259096}, {-1769797708,-1216347848}, {-1801031344,-1169603400},
+{-1831030924,-1122056937}, {-1859775343,-1073741910}, {-1887245423,-1024690552},
+{-1913422071,-974936918}, {-1938287125,-924515568}, {-1961823997,-873460141},
+{-1984016324,-821806084}, {-2004848713,-769589276}, {-2024307264,-716844553},
+{-2042378447,-663608538}, {-2059049731,-609918206}, {-2074309994,-555809377},
+{-2088148499,-501320119}, {-2100556013,-446486785}, {-2111523902,-391347448},
+{-2121044568,-335940406}, {-2129111659,-280302621}, {-2135719499,-224473240},
+{-2140863681,-168489506}, {-2144540612,-112390298}, {-2146747758,-56214574},
+{2147483647,0}, {2145828015,-84309815}, {2140863671,-168489630},
+{2132598271,-252409646}, {2121044558,-335940465}, {2106220349,-418953288},
+{2088148500,-501320115}, {2066856885,-582913912}, {2042378310,-663608960},
+{2014750533,-743280770}, {1984016179,-821806435}, {1950222618,-899064934},
+{1913421927,-974937199}, {1873670877,-1049306180}, {1831030826,-1122057097},
+{1785567394,-1193077993}, {1737350743,-1262259248}, {1686455222,-1329494189},
+{1632959307,-1394679144}, {1576945572,-1457713510}, {1518500216,-1518500282},
+{1457713441,-1576945636}, {1394679073,-1632959368}, {1329494115,-1686455280},
+{1262259172,-1737350799}, {1193077915,-1785567446}, {1122057017,-1831030875},
+{1049305987,-1873670985}, {974937230,-1913421912}, {899064965,-1950222603},
+{821806407,-1984016190}, {743280682,-2014750566}, {663608871,-2042378339},
+{582913822,-2066856911}, {501319962,-2088148536}, {418953102,-2106220386},
+{335940246,-2121044593}, {252409648,-2132598271}, {168489600,-2140863674},
+{84309753,-2145828017}, {-94,-2147483647}, {-84309940,-2145828010},
+{-168489787,-2140863659}, {-252409834,-2132598249}, {-335940431,-2121044564},
+{-418953286,-2106220349}, {-501320144,-2088148493}, {-582914003,-2066856860},
+{-663609049,-2042378281}, {-743280858,-2014750501}, {-821806581,-1984016118},
+{-899065136,-1950222525}, {-974937397,-1913421827}, {-1049306374,-1873670768},
+{-1122057395,-1831030643}, {-1193078284,-1785567199}, {-1262259116,-1737350839},
+{-1329494061,-1686455323}, {-1394679021,-1632959413}, {-1457713485,-1576945595},
+{-1518500258,-1518500240}, {-1576945613,-1457713466}, {-1632959429,-1394679001},
+{-1686455338,-1329494041}, {-1737350854,-1262259096}, {-1785567498,-1193077837},
+{-1831030924,-1122056937}, {-1873671031,-1049305905}, {-1913422071,-974936918},
+{-1950222750,-899064648}, {-1984016324,-821806084}, {-2014750687,-743280354},
+{-2042378447,-663608538}, {-2066856867,-582913978}, {-2088148499,-501320119},
+{-2106220354,-418953261}, {-2121044568,-335940406}, {-2132598282,-252409555},
+{-2140863681,-168489506}, {-2145828021,-84309659}, {-2147483647,188},
+{-2145828006,84310034}, {-2140863651,168489881}, {-2132598237,252409928},
+{-2121044509,335940777}, {-2106220281,418953629}, {-2088148411,501320484},
+{-2066856765,582914339}, {-2042378331,663608895}, {-2014750557,743280706},
+{-1984016181,821806431}, {-1950222593,899064989}, {-1913421900,974937252},
+{-1873670848,1049306232}, {-1831030728,1122057257}, {-1785567289,1193078149},
+{-1737350633,1262259400}, {-1686455106,1329494336}, {-1632959185,1394679287},
+{-1576945358,1457713742}, {-1518499993,1518500506}, {-1457713209,1576945850},
+{-1394678735,1632959656}, {-1329493766,1686455555}, {-1262258813,1737351059},
+{-1193077546,1785567692}, {-1122056638,1831031107}, {-1049305599,1873671202},
+{-974936606,1913422229}, {-899064330,1950222896}, {-821805761,1984016458},
+{-743280025,2014750808}, {-663609179,2042378239}, {-582914134,2066856823},
+{-501320277,2088148461}, {-418953420,2106220322}, {-335940566,2121044542},
+{-252409716,2132598263}, {-168489668,2140863668}, {-84309821,2145828015},
+};
+static const ne10_fft_cpx_int32_t ne10_twiddles_240[240] = {
+{0,0}, {2147483647,0}, {2147483647,0},
+{2147483647,0}, {1961823921,-873460313}, {1436946998,-1595891394},
+{2147483647,0}, {1436946998,-1595891394}, {-224473265,-2135719496},
+{2147483647,0}, {663608871,-2042378339}, {-1737350854,-1262259096},
+{2147483647,0}, {-224473265,-2135719496}, {-2100555935,446487152},
+{2147483647,0}, {2135719506,-224473172}, {2100555974,-446486968},
+{2042378310,-663608960}, {1961823921,-873460313}, {1859775377,-1073741851},
+{1737350743,-1262259248}, {1595891331,-1436947067}, {1436946998,-1595891394},
+{1262259172,-1737350799}, {1073741769,-1859775424}, {873460227,-1961823959},
+{663608871,-2042378339}, {446486876,-2100555994}, {224473078,-2135719516},
+{2147483647,0}, {2100555974,-446486968}, {1961823921,-873460313},
+{1737350743,-1262259248}, {1436946998,-1595891394}, {1073741769,-1859775424},
+{663608871,-2042378339}, {224473078,-2135719516}, {-224473265,-2135719496},
+{-663609049,-2042378281}, {-1073741932,-1859775330}, {-1436947137,-1595891268},
+{-1737350854,-1262259096}, {-1961823997,-873460141}, {-2100556013,-446486785},
+{2147483647,0}, {2042378310,-663608960}, {1737350743,-1262259248},
+{1262259172,-1737350799}, {663608871,-2042378339}, {-94,-2147483647},
+{-663609049,-2042378281}, {-1262259116,-1737350839}, {-1737350854,-1262259096},
+{-2042378447,-663608538}, {-2147483647,188}, {-2042378331,663608895},
+{-1737350633,1262259400}, {-1262258813,1737351059}, {-663609179,2042378239},
+{2147483647,0}, {2146747758,-56214570}, {2144540595,-112390613},
+{2140863671,-168489630}, {2135719506,-224473172}, {2129111626,-280302871},
+{2121044558,-335940465}, {2111523833,-391347822}, {2100555974,-446486968},
+{2088148500,-501320115}, {2074309912,-555809682}, {2059049696,-609918325},
+{2042378310,-663608960}, {2024307180,-716844791}, {2004848691,-769589332},
+{1984016179,-821806435}, {1961823921,-873460313}, {1938287127,-924515564},
+{1913421927,-974937199}, {1887245364,-1024690661}, {1859775377,-1073741851},
+{1831030826,-1122057097}, {1801031311,-1169603450}, {1769797456,-1216348214},
+{1737350743,-1262259248}, {1703713340,-1307305194}, {1668908218,-1351455280},
+{1632959307,-1394679144}, {1595891331,-1436947067}, {1557729613,-1478230181},
+{1518500216,-1518500282}, {1478230113,-1557729677}, {1436946998,-1595891394},
+{1394679073,-1632959368}, {1351455207,-1668908277}, {1307305120,-1703713397},
+{1262259172,-1737350799}, {1216348136,-1769797510}, {1169603371,-1801031362},
+{1122057017,-1831030875}, {1073741769,-1859775424}, {1024690635,-1887245378},
+{974937230,-1913421912}, {924515422,-1938287195}, {873460227,-1961823959},
+{821806407,-1984016190}, {769589125,-2004848771}, {716844642,-2024307233},
+{663608871,-2042378339}, {609918296,-2059049705}, {555809715,-2074309903},
+{501319962,-2088148536}, {446486876,-2100555994}, {391347792,-2111523838},
+{335940246,-2121044593}, {280302715,-2129111646}, {224473078,-2135719516},
+{168489600,-2140863674}, {112390647,-2144540593}, {56214412,-2146747762},
+{2147483647,0}, {2144540595,-112390613}, {2135719506,-224473172},
+{2121044558,-335940465}, {2100555974,-446486968}, {2074309912,-555809682},
+{2042378310,-663608960}, {2004848691,-769589332}, {1961823921,-873460313},
+{1913421927,-974937199}, {1859775377,-1073741851}, {1801031311,-1169603450},
+{1737350743,-1262259248}, {1668908218,-1351455280}, {1595891331,-1436947067},
+{1518500216,-1518500282}, {1436946998,-1595891394}, {1351455207,-1668908277},
+{1262259172,-1737350799}, {1169603371,-1801031362}, {1073741769,-1859775424},
+{974937230,-1913421912}, {873460227,-1961823959}, {769589125,-2004848771},
+{663608871,-2042378339}, {555809715,-2074309903}, {446486876,-2100555994},
+{335940246,-2121044593}, {224473078,-2135719516}, {112390647,-2144540593},
+{-94,-2147483647}, {-112390835,-2144540584}, {-224473265,-2135719496},
+{-335940431,-2121044564}, {-446487060,-2100555955}, {-555809896,-2074309855},
+{-663609049,-2042378281}, {-769589300,-2004848703}, {-873460398,-1961823883},
+{-974937397,-1913421827}, {-1073741932,-1859775330}, {-1169603421,-1801031330},
+{-1262259116,-1737350839}, {-1351455453,-1668908078}, {-1436947137,-1595891268},
+{-1518500258,-1518500240}, {-1595891628,-1436946738}, {-1668908417,-1351455035},
+{-1737350854,-1262259096}, {-1801031344,-1169603400}, {-1859775343,-1073741910},
+{-1913422071,-974936918}, {-1961823997,-873460141}, {-2004848713,-769589276},
+{-2042378447,-663608538}, {-2074309994,-555809377}, {-2100556013,-446486785},
+{-2121044568,-335940406}, {-2135719499,-224473240}, {-2144540612,-112390298},
+{2147483647,0}, {2140863671,-168489630}, {2121044558,-335940465},
+{2088148500,-501320115}, {2042378310,-663608960}, {1984016179,-821806435},
+{1913421927,-974937199}, {1831030826,-1122057097}, {1737350743,-1262259248},
+{1632959307,-1394679144}, {1518500216,-1518500282}, {1394679073,-1632959368},
+{1262259172,-1737350799}, {1122057017,-1831030875}, {974937230,-1913421912},
+{821806407,-1984016190}, {663608871,-2042378339}, {501319962,-2088148536},
+{335940246,-2121044593}, {168489600,-2140863674}, {-94,-2147483647},
+{-168489787,-2140863659}, {-335940431,-2121044564}, {-501320144,-2088148493},
+{-663609049,-2042378281}, {-821806581,-1984016118}, {-974937397,-1913421827},
+{-1122057395,-1831030643}, {-1262259116,-1737350839}, {-1394679021,-1632959413},
+{-1518500258,-1518500240}, {-1632959429,-1394679001}, {-1737350854,-1262259096},
+{-1831030924,-1122056937}, {-1913422071,-974936918}, {-1984016324,-821806084},
+{-2042378447,-663608538}, {-2088148499,-501320119}, {-2121044568,-335940406},
+{-2140863681,-168489506}, {-2147483647,188}, {-2140863651,168489881},
+{-2121044509,335940777}, {-2088148411,501320484}, {-2042378331,663608895},
+{-1984016181,821806431}, {-1913421900,974937252}, {-1831030728,1122057257},
+{-1737350633,1262259400}, {-1632959185,1394679287}, {-1518499993,1518500506},
+{-1394678735,1632959656}, {-1262258813,1737351059}, {-1122056638,1831031107},
+{-974936606,1913422229}, {-821805761,1984016458}, {-663609179,2042378239},
+{-501320277,2088148461}, {-335940566,2121044542}, {-168489668,2140863668},
+};
+static const ne10_fft_cpx_int32_t ne10_twiddles_120[120] = {
+{0,0}, {2147483647,0}, {2147483647,0},
+{2147483647,0}, {1961823921,-873460313}, {1436946998,-1595891394},
+{2147483647,0}, {1436946998,-1595891394}, {-224473265,-2135719496},
+{2147483647,0}, {663608871,-2042378339}, {-1737350854,-1262259096},
+{2147483647,0}, {-224473265,-2135719496}, {-2100555935,446487152},
+{2147483647,0}, {2100555974,-446486968}, {1961823921,-873460313},
+{1737350743,-1262259248}, {1436946998,-1595891394}, {1073741769,-1859775424},
+{663608871,-2042378339}, {224473078,-2135719516}, {-224473265,-2135719496},
+{-663609049,-2042378281}, {-1073741932,-1859775330}, {-1436947137,-1595891268},
+{-1737350854,-1262259096}, {-1961823997,-873460141}, {-2100556013,-446486785},
+{2147483647,0}, {2144540595,-112390613}, {2135719506,-224473172},
+{2121044558,-335940465}, {2100555974,-446486968}, {2074309912,-555809682},
+{2042378310,-663608960}, {2004848691,-769589332}, {1961823921,-873460313},
+{1913421927,-974937199}, {1859775377,-1073741851}, {1801031311,-1169603450},
+{1737350743,-1262259248}, {1668908218,-1351455280}, {1595891331,-1436947067},
+{1518500216,-1518500282}, {1436946998,-1595891394}, {1351455207,-1668908277},
+{1262259172,-1737350799}, {1169603371,-1801031362}, {1073741769,-1859775424},
+{974937230,-1913421912}, {873460227,-1961823959}, {769589125,-2004848771},
+{663608871,-2042378339}, {555809715,-2074309903}, {446486876,-2100555994},
+{335940246,-2121044593}, {224473078,-2135719516}, {112390647,-2144540593},
+{2147483647,0}, {2135719506,-224473172}, {2100555974,-446486968},
+{2042378310,-663608960}, {1961823921,-873460313}, {1859775377,-1073741851},
+{1737350743,-1262259248}, {1595891331,-1436947067}, {1436946998,-1595891394},
+{1262259172,-1737350799}, {1073741769,-1859775424}, {873460227,-1961823959},
+{663608871,-2042378339}, {446486876,-2100555994}, {224473078,-2135719516},
+{-94,-2147483647}, {-224473265,-2135719496}, {-446487060,-2100555955},
+{-663609049,-2042378281}, {-873460398,-1961823883}, {-1073741932,-1859775330},
+{-1262259116,-1737350839}, {-1436947137,-1595891268}, {-1595891628,-1436946738},
+{-1737350854,-1262259096}, {-1859775343,-1073741910}, {-1961823997,-873460141},
+{-2042378447,-663608538}, {-2100556013,-446486785}, {-2135719499,-224473240},
+{2147483647,0}, {2121044558,-335940465}, {2042378310,-663608960},
+{1913421927,-974937199}, {1737350743,-1262259248}, {1518500216,-1518500282},
+{1262259172,-1737350799}, {974937230,-1913421912}, {663608871,-2042378339},
+{335940246,-2121044593}, {-94,-2147483647}, {-335940431,-2121044564},
+{-663609049,-2042378281}, {-974937397,-1913421827}, {-1262259116,-1737350839},
+{-1518500258,-1518500240}, {-1737350854,-1262259096}, {-1913422071,-974936918},
+{-2042378447,-663608538}, {-2121044568,-335940406}, {-2147483647,188},
+{-2121044509,335940777}, {-2042378331,663608895}, {-1913421900,974937252},
+{-1737350633,1262259400}, {-1518499993,1518500506}, {-1262258813,1737351059},
+{-974936606,1913422229}, {-663609179,2042378239}, {-335940566,2121044542},
+};
+static const ne10_fft_cpx_int32_t ne10_twiddles_60[60] = {
+{0,0}, {2147483647,0}, {2147483647,0},
+{2147483647,0}, {1961823921,-873460313}, {1436946998,-1595891394},
+{2147483647,0}, {1436946998,-1595891394}, {-224473265,-2135719496},
+{2147483647,0}, {663608871,-2042378339}, {-1737350854,-1262259096},
+{2147483647,0}, {-224473265,-2135719496}, {-2100555935,446487152},
+{2147483647,0}, {2135719506,-224473172}, {2100555974,-446486968},
+{2042378310,-663608960}, {1961823921,-873460313}, {1859775377,-1073741851},
+{1737350743,-1262259248}, {1595891331,-1436947067}, {1436946998,-1595891394},
+{1262259172,-1737350799}, {1073741769,-1859775424}, {873460227,-1961823959},
+{663608871,-2042378339}, {446486876,-2100555994}, {224473078,-2135719516},
+{2147483647,0}, {2100555974,-446486968}, {1961823921,-873460313},
+{1737350743,-1262259248}, {1436946998,-1595891394}, {1073741769,-1859775424},
+{663608871,-2042378339}, {224473078,-2135719516}, {-224473265,-2135719496},
+{-663609049,-2042378281}, {-1073741932,-1859775330}, {-1436947137,-1595891268},
+{-1737350854,-1262259096}, {-1961823997,-873460141}, {-2100556013,-446486785},
+{2147483647,0}, {2042378310,-663608960}, {1737350743,-1262259248},
+{1262259172,-1737350799}, {663608871,-2042378339}, {-94,-2147483647},
+{-663609049,-2042378281}, {-1262259116,-1737350839}, {-1737350854,-1262259096},
+{-2042378447,-663608538}, {-2147483647,188}, {-2042378331,663608895},
+{-1737350633,1262259400}, {-1262258813,1737351059}, {-663609179,2042378239},
+};
+static const ne10_fft_state_int32_t ne10_fft_state_int32_t_480 = {
+120,
+(ne10_int32_t *)ne10_factors_480,
+(ne10_fft_cpx_int32_t *)ne10_twiddles_480,
+NULL,
+(ne10_fft_cpx_int32_t *)&ne10_twiddles_480[120],
+};
+static const arch_fft_state cfg_arch_480 = {
+1,
+(void *)&ne10_fft_state_int32_t_480,
+};
+
+static const ne10_fft_state_int32_t ne10_fft_state_int32_t_240 = {
+60,
+(ne10_int32_t *)ne10_factors_240,
+(ne10_fft_cpx_int32_t *)ne10_twiddles_240,
+NULL,
+(ne10_fft_cpx_int32_t *)&ne10_twiddles_240[60],
+};
+static const arch_fft_state cfg_arch_240 = {
+1,
+(void *)&ne10_fft_state_int32_t_240,
+};
+
+static const ne10_fft_state_int32_t ne10_fft_state_int32_t_120 = {
+30,
+(ne10_int32_t *)ne10_factors_120,
+(ne10_fft_cpx_int32_t *)ne10_twiddles_120,
+NULL,
+(ne10_fft_cpx_int32_t *)&ne10_twiddles_120[30],
+};
+static const arch_fft_state cfg_arch_120 = {
+1,
+(void *)&ne10_fft_state_int32_t_120,
+};
+
+static const ne10_fft_state_int32_t ne10_fft_state_int32_t_60 = {
+15,
+(ne10_int32_t *)ne10_factors_60,
+(ne10_fft_cpx_int32_t *)ne10_twiddles_60,
+NULL,
+(ne10_fft_cpx_int32_t *)&ne10_twiddles_60[15],
+};
+static const arch_fft_state cfg_arch_60 = {
+1,
+(void *)&ne10_fft_state_int32_t_60,
+};
+
+#endif  /* end NE10_FFT_PARAMS48000_960 */
diff --git a/drivers/opus/celt/static_modes_float.h b/drivers/opus/celt/static_modes_float.h
index 362be6cca8..f7d501a993 100644
--- a/drivers/opus/celt/static_modes_float.h
+++ b/drivers/opus/celt/static_modes_float.h
@@ -1,9 +1,14 @@
 /* The contents of this file was automatically generated by dump_modes.c
    with arguments: 48000 960
    It contains static definitions for some pre-defined modes. */
-#include "opus/celt/opus_modes.h"
+#include "opus/celt/modes.h"
 #include "opus/celt/rate.h"
 
+#ifdef HAVE_ARM_NE10
+#define OVERRIDE_FFT 1
+#include "opus/celt/static_modes_float_arm_ne10.h"
+#endif
+
 #ifndef DEF_WINDOW120
 #define DEF_WINDOW120
 static const opus_val16 window120[120] = {
@@ -341,84 +346,84 @@ static const kiss_twiddle_cpx fft_twiddles48000_960[480] = {
 #ifndef FFT_BITREV480
 #define FFT_BITREV480
 static const opus_int16 fft_bitrev480[480] = {
-0, 120, 240, 360, 30, 150, 270, 390, 60, 180, 300, 420, 90, 210, 330,
-450, 15, 135, 255, 375, 45, 165, 285, 405, 75, 195, 315, 435, 105, 225,
-345, 465, 5, 125, 245, 365, 35, 155, 275, 395, 65, 185, 305, 425, 95,
-215, 335, 455, 20, 140, 260, 380, 50, 170, 290, 410, 80, 200, 320, 440,
-110, 230, 350, 470, 10, 130, 250, 370, 40, 160, 280, 400, 70, 190, 310,
-430, 100, 220, 340, 460, 25, 145, 265, 385, 55, 175, 295, 415, 85, 205,
-325, 445, 115, 235, 355, 475, 1, 121, 241, 361, 31, 151, 271, 391, 61,
-181, 301, 421, 91, 211, 331, 451, 16, 136, 256, 376, 46, 166, 286, 406,
-76, 196, 316, 436, 106, 226, 346, 466, 6, 126, 246, 366, 36, 156, 276,
-396, 66, 186, 306, 426, 96, 216, 336, 456, 21, 141, 261, 381, 51, 171,
-291, 411, 81, 201, 321, 441, 111, 231, 351, 471, 11, 131, 251, 371, 41,
-161, 281, 401, 71, 191, 311, 431, 101, 221, 341, 461, 26, 146, 266, 386,
-56, 176, 296, 416, 86, 206, 326, 446, 116, 236, 356, 476, 2, 122, 242,
-362, 32, 152, 272, 392, 62, 182, 302, 422, 92, 212, 332, 452, 17, 137,
-257, 377, 47, 167, 287, 407, 77, 197, 317, 437, 107, 227, 347, 467, 7,
-127, 247, 367, 37, 157, 277, 397, 67, 187, 307, 427, 97, 217, 337, 457,
-22, 142, 262, 382, 52, 172, 292, 412, 82, 202, 322, 442, 112, 232, 352,
-472, 12, 132, 252, 372, 42, 162, 282, 402, 72, 192, 312, 432, 102, 222,
-342, 462, 27, 147, 267, 387, 57, 177, 297, 417, 87, 207, 327, 447, 117,
-237, 357, 477, 3, 123, 243, 363, 33, 153, 273, 393, 63, 183, 303, 423,
-93, 213, 333, 453, 18, 138, 258, 378, 48, 168, 288, 408, 78, 198, 318,
-438, 108, 228, 348, 468, 8, 128, 248, 368, 38, 158, 278, 398, 68, 188,
-308, 428, 98, 218, 338, 458, 23, 143, 263, 383, 53, 173, 293, 413, 83,
-203, 323, 443, 113, 233, 353, 473, 13, 133, 253, 373, 43, 163, 283, 403,
-73, 193, 313, 433, 103, 223, 343, 463, 28, 148, 268, 388, 58, 178, 298,
-418, 88, 208, 328, 448, 118, 238, 358, 478, 4, 124, 244, 364, 34, 154,
-274, 394, 64, 184, 304, 424, 94, 214, 334, 454, 19, 139, 259, 379, 49,
-169, 289, 409, 79, 199, 319, 439, 109, 229, 349, 469, 9, 129, 249, 369,
-39, 159, 279, 399, 69, 189, 309, 429, 99, 219, 339, 459, 24, 144, 264,
-384, 54, 174, 294, 414, 84, 204, 324, 444, 114, 234, 354, 474, 14, 134,
-254, 374, 44, 164, 284, 404, 74, 194, 314, 434, 104, 224, 344, 464, 29,
-149, 269, 389, 59, 179, 299, 419, 89, 209, 329, 449, 119, 239, 359, 479,
+0, 96, 192, 288, 384, 32, 128, 224, 320, 416, 64, 160, 256, 352, 448,
+8, 104, 200, 296, 392, 40, 136, 232, 328, 424, 72, 168, 264, 360, 456,
+16, 112, 208, 304, 400, 48, 144, 240, 336, 432, 80, 176, 272, 368, 464,
+24, 120, 216, 312, 408, 56, 152, 248, 344, 440, 88, 184, 280, 376, 472,
+4, 100, 196, 292, 388, 36, 132, 228, 324, 420, 68, 164, 260, 356, 452,
+12, 108, 204, 300, 396, 44, 140, 236, 332, 428, 76, 172, 268, 364, 460,
+20, 116, 212, 308, 404, 52, 148, 244, 340, 436, 84, 180, 276, 372, 468,
+28, 124, 220, 316, 412, 60, 156, 252, 348, 444, 92, 188, 284, 380, 476,
+1, 97, 193, 289, 385, 33, 129, 225, 321, 417, 65, 161, 257, 353, 449,
+9, 105, 201, 297, 393, 41, 137, 233, 329, 425, 73, 169, 265, 361, 457,
+17, 113, 209, 305, 401, 49, 145, 241, 337, 433, 81, 177, 273, 369, 465,
+25, 121, 217, 313, 409, 57, 153, 249, 345, 441, 89, 185, 281, 377, 473,
+5, 101, 197, 293, 389, 37, 133, 229, 325, 421, 69, 165, 261, 357, 453,
+13, 109, 205, 301, 397, 45, 141, 237, 333, 429, 77, 173, 269, 365, 461,
+21, 117, 213, 309, 405, 53, 149, 245, 341, 437, 85, 181, 277, 373, 469,
+29, 125, 221, 317, 413, 61, 157, 253, 349, 445, 93, 189, 285, 381, 477,
+2, 98, 194, 290, 386, 34, 130, 226, 322, 418, 66, 162, 258, 354, 450,
+10, 106, 202, 298, 394, 42, 138, 234, 330, 426, 74, 170, 266, 362, 458,
+18, 114, 210, 306, 402, 50, 146, 242, 338, 434, 82, 178, 274, 370, 466,
+26, 122, 218, 314, 410, 58, 154, 250, 346, 442, 90, 186, 282, 378, 474,
+6, 102, 198, 294, 390, 38, 134, 230, 326, 422, 70, 166, 262, 358, 454,
+14, 110, 206, 302, 398, 46, 142, 238, 334, 430, 78, 174, 270, 366, 462,
+22, 118, 214, 310, 406, 54, 150, 246, 342, 438, 86, 182, 278, 374, 470,
+30, 126, 222, 318, 414, 62, 158, 254, 350, 446, 94, 190, 286, 382, 478,
+3, 99, 195, 291, 387, 35, 131, 227, 323, 419, 67, 163, 259, 355, 451,
+11, 107, 203, 299, 395, 43, 139, 235, 331, 427, 75, 171, 267, 363, 459,
+19, 115, 211, 307, 403, 51, 147, 243, 339, 435, 83, 179, 275, 371, 467,
+27, 123, 219, 315, 411, 59, 155, 251, 347, 443, 91, 187, 283, 379, 475,
+7, 103, 199, 295, 391, 39, 135, 231, 327, 423, 71, 167, 263, 359, 455,
+15, 111, 207, 303, 399, 47, 143, 239, 335, 431, 79, 175, 271, 367, 463,
+23, 119, 215, 311, 407, 55, 151, 247, 343, 439, 87, 183, 279, 375, 471,
+31, 127, 223, 319, 415, 63, 159, 255, 351, 447, 95, 191, 287, 383, 479,
 };
 #endif
 
 #ifndef FFT_BITREV240
 #define FFT_BITREV240
 static const opus_int16 fft_bitrev240[240] = {
-0, 60, 120, 180, 15, 75, 135, 195, 30, 90, 150, 210, 45, 105, 165,
-225, 5, 65, 125, 185, 20, 80, 140, 200, 35, 95, 155, 215, 50, 110,
-170, 230, 10, 70, 130, 190, 25, 85, 145, 205, 40, 100, 160, 220, 55,
-115, 175, 235, 1, 61, 121, 181, 16, 76, 136, 196, 31, 91, 151, 211,
-46, 106, 166, 226, 6, 66, 126, 186, 21, 81, 141, 201, 36, 96, 156,
-216, 51, 111, 171, 231, 11, 71, 131, 191, 26, 86, 146, 206, 41, 101,
-161, 221, 56, 116, 176, 236, 2, 62, 122, 182, 17, 77, 137, 197, 32,
-92, 152, 212, 47, 107, 167, 227, 7, 67, 127, 187, 22, 82, 142, 202,
-37, 97, 157, 217, 52, 112, 172, 232, 12, 72, 132, 192, 27, 87, 147,
-207, 42, 102, 162, 222, 57, 117, 177, 237, 3, 63, 123, 183, 18, 78,
-138, 198, 33, 93, 153, 213, 48, 108, 168, 228, 8, 68, 128, 188, 23,
-83, 143, 203, 38, 98, 158, 218, 53, 113, 173, 233, 13, 73, 133, 193,
-28, 88, 148, 208, 43, 103, 163, 223, 58, 118, 178, 238, 4, 64, 124,
-184, 19, 79, 139, 199, 34, 94, 154, 214, 49, 109, 169, 229, 9, 69,
-129, 189, 24, 84, 144, 204, 39, 99, 159, 219, 54, 114, 174, 234, 14,
-74, 134, 194, 29, 89, 149, 209, 44, 104, 164, 224, 59, 119, 179, 239,
+0, 48, 96, 144, 192, 16, 64, 112, 160, 208, 32, 80, 128, 176, 224,
+4, 52, 100, 148, 196, 20, 68, 116, 164, 212, 36, 84, 132, 180, 228,
+8, 56, 104, 152, 200, 24, 72, 120, 168, 216, 40, 88, 136, 184, 232,
+12, 60, 108, 156, 204, 28, 76, 124, 172, 220, 44, 92, 140, 188, 236,
+1, 49, 97, 145, 193, 17, 65, 113, 161, 209, 33, 81, 129, 177, 225,
+5, 53, 101, 149, 197, 21, 69, 117, 165, 213, 37, 85, 133, 181, 229,
+9, 57, 105, 153, 201, 25, 73, 121, 169, 217, 41, 89, 137, 185, 233,
+13, 61, 109, 157, 205, 29, 77, 125, 173, 221, 45, 93, 141, 189, 237,
+2, 50, 98, 146, 194, 18, 66, 114, 162, 210, 34, 82, 130, 178, 226,
+6, 54, 102, 150, 198, 22, 70, 118, 166, 214, 38, 86, 134, 182, 230,
+10, 58, 106, 154, 202, 26, 74, 122, 170, 218, 42, 90, 138, 186, 234,
+14, 62, 110, 158, 206, 30, 78, 126, 174, 222, 46, 94, 142, 190, 238,
+3, 51, 99, 147, 195, 19, 67, 115, 163, 211, 35, 83, 131, 179, 227,
+7, 55, 103, 151, 199, 23, 71, 119, 167, 215, 39, 87, 135, 183, 231,
+11, 59, 107, 155, 203, 27, 75, 123, 171, 219, 43, 91, 139, 187, 235,
+15, 63, 111, 159, 207, 31, 79, 127, 175, 223, 47, 95, 143, 191, 239,
 };
 #endif
 
 #ifndef FFT_BITREV120
 #define FFT_BITREV120
 static const opus_int16 fft_bitrev120[120] = {
-0, 30, 60, 90, 15, 45, 75, 105, 5, 35, 65, 95, 20, 50, 80,
-110, 10, 40, 70, 100, 25, 55, 85, 115, 1, 31, 61, 91, 16, 46,
-76, 106, 6, 36, 66, 96, 21, 51, 81, 111, 11, 41, 71, 101, 26,
-56, 86, 116, 2, 32, 62, 92, 17, 47, 77, 107, 7, 37, 67, 97,
-22, 52, 82, 112, 12, 42, 72, 102, 27, 57, 87, 117, 3, 33, 63,
-93, 18, 48, 78, 108, 8, 38, 68, 98, 23, 53, 83, 113, 13, 43,
-73, 103, 28, 58, 88, 118, 4, 34, 64, 94, 19, 49, 79, 109, 9,
-39, 69, 99, 24, 54, 84, 114, 14, 44, 74, 104, 29, 59, 89, 119,
+0, 24, 48, 72, 96, 8, 32, 56, 80, 104, 16, 40, 64, 88, 112,
+4, 28, 52, 76, 100, 12, 36, 60, 84, 108, 20, 44, 68, 92, 116,
+1, 25, 49, 73, 97, 9, 33, 57, 81, 105, 17, 41, 65, 89, 113,
+5, 29, 53, 77, 101, 13, 37, 61, 85, 109, 21, 45, 69, 93, 117,
+2, 26, 50, 74, 98, 10, 34, 58, 82, 106, 18, 42, 66, 90, 114,
+6, 30, 54, 78, 102, 14, 38, 62, 86, 110, 22, 46, 70, 94, 118,
+3, 27, 51, 75, 99, 11, 35, 59, 83, 107, 19, 43, 67, 91, 115,
+7, 31, 55, 79, 103, 15, 39, 63, 87, 111, 23, 47, 71, 95, 119,
 };
 #endif
 
 #ifndef FFT_BITREV60
 #define FFT_BITREV60
 static const opus_int16 fft_bitrev60[60] = {
-0, 15, 30, 45, 5, 20, 35, 50, 10, 25, 40, 55, 1, 16, 31,
-46, 6, 21, 36, 51, 11, 26, 41, 56, 2, 17, 32, 47, 7, 22,
-37, 52, 12, 27, 42, 57, 3, 18, 33, 48, 8, 23, 38, 53, 13,
-28, 43, 58, 4, 19, 34, 49, 9, 24, 39, 54, 14, 29, 44, 59,
+0, 12, 24, 36, 48, 4, 16, 28, 40, 52, 8, 20, 32, 44, 56,
+1, 13, 25, 37, 49, 5, 17, 29, 41, 53, 9, 21, 33, 45, 57,
+2, 14, 26, 38, 50, 6, 18, 30, 42, 54, 10, 22, 34, 46, 58,
+3, 15, 27, 39, 51, 7, 19, 31, 43, 55, 11, 23, 35, 47, 59,
 };
 #endif
 
@@ -428,9 +433,14 @@ static const kiss_fft_state fft_state48000_960_0 = {
 480,    /* nfft */
 0.002083333f,   /* scale */
 -1,     /* shift */
-{4, 120, 4, 30, 2, 15, 3, 5, 5, 1, 0, 0, 0, 0, 0, 0, }, /* factors */
+{5, 96, 3, 32, 4, 8, 2, 4, 4, 1, 0, 0, 0, 0, 0, 0, },   /* factors */
 fft_bitrev480,  /* bitrev */
 fft_twiddles48000_960,  /* bitrev */
+#ifdef OVERRIDE_FFT
+(arch_fft_state *)&cfg_arch_480,
+#else
+NULL,
+#endif
 };
 #endif
 
@@ -440,9 +450,14 @@ static const kiss_fft_state fft_state48000_960_1 = {
 240,    /* nfft */
 0.004166667f,   /* scale */
 1,      /* shift */
-{4, 60, 4, 15, 3, 5, 5, 1, 0, 0, 0, 0, 0, 0, 0, 0, },   /* factors */
+{5, 48, 3, 16, 4, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, },    /* factors */
 fft_bitrev240,  /* bitrev */
 fft_twiddles48000_960,  /* bitrev */
+#ifdef OVERRIDE_FFT
+(arch_fft_state *)&cfg_arch_240,
+#else
+NULL,
+#endif
 };
 #endif
 
@@ -452,9 +467,14 @@ static const kiss_fft_state fft_state48000_960_2 = {
 120,    /* nfft */
 0.008333333f,   /* scale */
 2,      /* shift */
-{4, 30, 2, 15, 3, 5, 5, 1, 0, 0, 0, 0, 0, 0, 0, 0, },   /* factors */
+{5, 24, 3, 8, 2, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, },    /* factors */
 fft_bitrev120,  /* bitrev */
 fft_twiddles48000_960,  /* bitrev */
+#ifdef OVERRIDE_FFT
+(arch_fft_state *)&cfg_arch_120,
+#else
+NULL,
+#endif
 };
 #endif
 
@@ -464,9 +484,14 @@ static const kiss_fft_state fft_state48000_960_3 = {
 60,     /* nfft */
 0.016666667f,   /* scale */
 3,      /* shift */
-{4, 15, 3, 5, 5, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },    /* factors */
+{5, 12, 3, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },    /* factors */
 fft_bitrev60,   /* bitrev */
 fft_twiddles48000_960,  /* bitrev */
+#ifdef OVERRIDE_FFT
+(arch_fft_state *)&cfg_arch_60,
+#else
+NULL,
+#endif
 };
 #endif
 
@@ -474,104 +499,368 @@ fft_twiddles48000_960,  /* bitrev */
 
 #ifndef MDCT_TWIDDLES960
 #define MDCT_TWIDDLES960
-static const opus_val16 mdct_twiddles960[481] = {
-1.0000000f, 0.99999465f, 0.99997858f, 0.99995181f, 0.99991433f,
-0.99986614f, 0.99980724f, 0.99973764f, 0.99965732f, 0.99956631f,
-0.99946459f, 0.99935216f, 0.99922904f, 0.99909521f, 0.99895068f,
-0.99879546f, 0.99862953f, 0.99845292f, 0.99826561f, 0.99806761f,
-0.99785892f, 0.99763955f, 0.99740949f, 0.99716875f, 0.99691733f,
-0.99665524f, 0.99638247f, 0.99609903f, 0.99580493f, 0.99550016f,
-0.99518473f, 0.99485864f, 0.99452190f, 0.99417450f, 0.99381646f,
-0.99344778f, 0.99306846f, 0.99267850f, 0.99227791f, 0.99186670f,
-0.99144486f, 0.99101241f, 0.99056934f, 0.99011566f, 0.98965139f,
-0.98917651f, 0.98869104f, 0.98819498f, 0.98768834f, 0.98717112f,
-0.98664333f, 0.98610497f, 0.98555606f, 0.98499659f, 0.98442657f,
-0.98384600f, 0.98325491f, 0.98265328f, 0.98204113f, 0.98141846f,
-0.98078528f, 0.98014159f, 0.97948742f, 0.97882275f, 0.97814760f,
-0.97746197f, 0.97676588f, 0.97605933f, 0.97534232f, 0.97461487f,
-0.97387698f, 0.97312866f, 0.97236992f, 0.97160077f, 0.97082121f,
-0.97003125f, 0.96923091f, 0.96842019f, 0.96759909f, 0.96676764f,
-0.96592582f, 0.96507367f, 0.96421118f, 0.96333837f, 0.96245523f,
-0.96156180f, 0.96065806f, 0.95974403f, 0.95881973f, 0.95788517f,
-0.95694034f, 0.95598526f, 0.95501995f, 0.95404440f, 0.95305864f,
-0.95206267f, 0.95105651f, 0.95004016f, 0.94901364f, 0.94797697f,
-0.94693013f, 0.94587315f, 0.94480604f, 0.94372882f, 0.94264149f,
-0.94154406f, 0.94043656f, 0.93931897f, 0.93819133f, 0.93705365f,
-0.93590592f, 0.93474818f, 0.93358042f, 0.93240268f, 0.93121493f,
-0.93001722f, 0.92880955f, 0.92759193f, 0.92636438f, 0.92512690f,
-0.92387953f, 0.92262225f, 0.92135509f, 0.92007809f, 0.91879121f,
-0.91749449f, 0.91618795f, 0.91487161f, 0.91354545f, 0.91220952f,
-0.91086382f, 0.90950836f, 0.90814316f, 0.90676824f, 0.90538363f,
-0.90398929f, 0.90258528f, 0.90117161f, 0.89974828f, 0.89831532f,
-0.89687273f, 0.89542055f, 0.89395877f, 0.89248742f, 0.89100652f,
-0.88951606f, 0.88801610f, 0.88650661f, 0.88498764f, 0.88345918f,
-0.88192125f, 0.88037390f, 0.87881711f, 0.87725090f, 0.87567531f,
-0.87409035f, 0.87249599f, 0.87089232f, 0.86927933f, 0.86765699f,
-0.86602540f, 0.86438453f, 0.86273437f, 0.86107503f, 0.85940641f,
-0.85772862f, 0.85604161f, 0.85434547f, 0.85264014f, 0.85092572f,
-0.84920218f, 0.84746955f, 0.84572781f, 0.84397704f, 0.84221721f,
-0.84044838f, 0.83867056f, 0.83688375f, 0.83508799f, 0.83328325f,
-0.83146961f, 0.82964704f, 0.82781562f, 0.82597530f, 0.82412620f,
-0.82226820f, 0.82040144f, 0.81852589f, 0.81664154f, 0.81474847f,
-0.81284665f, 0.81093620f, 0.80901698f, 0.80708914f, 0.80515262f,
-0.80320752f, 0.80125378f, 0.79929149f, 0.79732067f, 0.79534125f,
-0.79335335f, 0.79135691f, 0.78935204f, 0.78733867f, 0.78531691f,
-0.78328674f, 0.78124818f, 0.77920122f, 0.77714595f, 0.77508232f,
-0.77301043f, 0.77093026f, 0.76884183f, 0.76674517f, 0.76464026f,
-0.76252720f, 0.76040593f, 0.75827656f, 0.75613907f, 0.75399349f,
-0.75183978f, 0.74967807f, 0.74750833f, 0.74533054f, 0.74314481f,
-0.74095112f, 0.73874950f, 0.73653993f, 0.73432251f, 0.73209718f,
-0.72986405f, 0.72762307f, 0.72537438f, 0.72311787f, 0.72085359f,
-0.71858162f, 0.71630192f, 0.71401459f, 0.71171956f, 0.70941701f,
-0.70710677f, 0.70478900f, 0.70246363f, 0.70013079f, 0.69779041f,
-0.69544260f, 0.69308738f, 0.69072466f, 0.68835458f, 0.68597709f,
-0.68359229f, 0.68120013f, 0.67880072f, 0.67639404f, 0.67398011f,
-0.67155892f, 0.66913059f, 0.66669509f, 0.66425240f, 0.66180265f,
-0.65934581f, 0.65688191f, 0.65441092f, 0.65193298f, 0.64944801f,
-0.64695613f, 0.64445727f, 0.64195160f, 0.63943902f, 0.63691954f,
-0.63439328f, 0.63186019f, 0.62932037f, 0.62677377f, 0.62422055f,
-0.62166055f, 0.61909394f, 0.61652065f, 0.61394081f, 0.61135435f,
-0.60876139f, 0.60616195f, 0.60355593f, 0.60094349f, 0.59832457f,
-0.59569929f, 0.59306758f, 0.59042957f, 0.58778523f, 0.58513460f,
-0.58247766f, 0.57981452f, 0.57714518f, 0.57446961f, 0.57178793f,
-0.56910013f, 0.56640624f, 0.56370623f, 0.56100023f, 0.55828818f,
-0.55557020f, 0.55284627f, 0.55011641f, 0.54738067f, 0.54463901f,
-0.54189157f, 0.53913828f, 0.53637921f, 0.53361450f, 0.53084398f,
-0.52806787f, 0.52528601f, 0.52249852f, 0.51970543f, 0.51690688f,
-0.51410279f, 0.51129310f, 0.50847793f, 0.50565732f, 0.50283139f,
-0.49999997f, 0.49716321f, 0.49432122f, 0.49147383f, 0.48862118f,
-0.48576340f, 0.48290042f, 0.48003216f, 0.47715876f, 0.47428025f,
-0.47139677f, 0.46850813f, 0.46561448f, 0.46271584f, 0.45981235f,
-0.45690383f, 0.45399042f, 0.45107214f, 0.44814915f, 0.44522124f,
-0.44228868f, 0.43935137f, 0.43640926f, 0.43346247f, 0.43051104f,
-0.42755511f, 0.42459449f, 0.42162932f, 0.41865964f, 0.41568558f,
-0.41270697f, 0.40972393f, 0.40673661f, 0.40374494f, 0.40074884f,
-0.39774844f, 0.39474390f, 0.39173501f, 0.38872193f, 0.38570469f,
-0.38268343f, 0.37965796f, 0.37662842f, 0.37359496f, 0.37055739f,
-0.36751585f, 0.36447038f, 0.36142122f, 0.35836797f, 0.35531089f,
-0.35225000f, 0.34918544f, 0.34611704f, 0.34304493f, 0.33996926f,
-0.33688983f, 0.33380680f, 0.33072019f, 0.32763015f, 0.32453650f,
-0.32143936f, 0.31833890f, 0.31523503f, 0.31212767f, 0.30901696f,
-0.30590306f, 0.30278577f, 0.29966524f, 0.29654150f, 0.29341470f,
-0.29028464f, 0.28715147f, 0.28401522f, 0.28087605f, 0.27773376f,
-0.27458861f, 0.27144052f, 0.26828940f, 0.26513541f, 0.26197859f,
-0.25881907f, 0.25565666f, 0.25249152f, 0.24932367f, 0.24615327f,
-0.24298012f, 0.23980436f, 0.23662604f, 0.23344530f, 0.23026206f,
-0.22707623f, 0.22388809f, 0.22069744f, 0.21750443f, 0.21430908f,
-0.21111156f, 0.20791165f, 0.20470953f, 0.20150520f, 0.19829884f,
-0.19509024f, 0.19187955f, 0.18866692f, 0.18545227f, 0.18223552f,
-0.17901681f, 0.17579631f, 0.17257380f, 0.16934945f, 0.16612328f,
-0.16289546f, 0.15966577f, 0.15643437f, 0.15320141f, 0.14996669f,
-0.14673037f, 0.14349260f, 0.14025329f, 0.13701235f, 0.13376995f,
-0.13052612f, 0.12728101f, 0.12403442f, 0.12078650f, 0.11753740f,
-0.11428693f, 0.11103523f, 0.10778234f, 0.10452842f, 0.10127326f,
-0.098017137f, 0.094759842f, 0.091501652f, 0.088242363f, 0.084982129f,
-0.081721103f, 0.078459084f, 0.075196224f, 0.071932560f, 0.068668243f,
-0.065403073f, 0.062137201f, 0.058870665f, 0.055603617f, 0.052335974f,
-0.049067651f, 0.045798921f, 0.042529582f, 0.039259788f, 0.035989573f,
-0.032719092f, 0.029448142f, 0.026176876f, 0.022905329f, 0.019633657f,
-0.016361655f, 0.013089478f, 0.0098171604f, 0.0065449764f, 0.0032724839f,
--4.3711390e-08f, };
+static const opus_val16 mdct_twiddles960[1800] = {
+0.99999994f, 0.99999321f, 0.99997580f, 0.99994773f, 0.99990886f,
+0.99985933f, 0.99979913f, 0.99972820f, 0.99964654f, 0.99955416f,
+0.99945110f, 0.99933738f, 0.99921292f, 0.99907774f, 0.99893188f,
+0.99877530f, 0.99860805f, 0.99843007f, 0.99824142f, 0.99804211f,
+0.99783206f, 0.99761140f, 0.99737996f, 0.99713790f, 0.99688518f,
+0.99662173f, 0.99634761f, 0.99606287f, 0.99576741f, 0.99546129f,
+0.99514455f, 0.99481714f, 0.99447906f, 0.99413031f, 0.99377096f,
+0.99340093f, 0.99302030f, 0.99262899f, 0.99222708f, 0.99181455f,
+0.99139136f, 0.99095762f, 0.99051321f, 0.99005818f, 0.98959261f,
+0.98911643f, 0.98862964f, 0.98813224f, 0.98762429f, 0.98710573f,
+0.98657662f, 0.98603696f, 0.98548669f, 0.98492593f, 0.98435456f,
+0.98377270f, 0.98318028f, 0.98257732f, 0.98196387f, 0.98133987f,
+0.98070538f, 0.98006040f, 0.97940493f, 0.97873890f, 0.97806245f,
+0.97737551f, 0.97667813f, 0.97597027f, 0.97525197f, 0.97452319f,
+0.97378403f, 0.97303438f, 0.97227436f, 0.97150391f, 0.97072303f,
+0.96993178f, 0.96913016f, 0.96831810f, 0.96749574f, 0.96666300f,
+0.96581990f, 0.96496642f, 0.96410263f, 0.96322852f, 0.96234411f,
+0.96144938f, 0.96054435f, 0.95962906f, 0.95870346f, 0.95776761f,
+0.95682150f, 0.95586514f, 0.95489854f, 0.95392174f, 0.95293468f,
+0.95193744f, 0.95093000f, 0.94991243f, 0.94888461f, 0.94784665f,
+0.94679856f, 0.94574034f, 0.94467193f, 0.94359344f, 0.94250488f,
+0.94140619f, 0.94029742f, 0.93917859f, 0.93804967f, 0.93691075f,
+0.93576175f, 0.93460274f, 0.93343377f, 0.93225473f, 0.93106574f,
+0.92986679f, 0.92865789f, 0.92743903f, 0.92621022f, 0.92497152f,
+0.92372292f, 0.92246443f, 0.92119598f, 0.91991776f, 0.91862965f,
+0.91733170f, 0.91602397f, 0.91470635f, 0.91337901f, 0.91204184f,
+0.91069490f, 0.90933824f, 0.90797186f, 0.90659571f, 0.90520984f,
+0.90381432f, 0.90240908f, 0.90099424f, 0.89956969f, 0.89813554f,
+0.89669174f, 0.89523834f, 0.89377540f, 0.89230281f, 0.89082074f,
+0.88932908f, 0.88782793f, 0.88631725f, 0.88479710f, 0.88326746f,
+0.88172835f, 0.88017982f, 0.87862182f, 0.87705445f, 0.87547767f,
+0.87389153f, 0.87229604f, 0.87069118f, 0.86907703f, 0.86745358f,
+0.86582077f, 0.86417878f, 0.86252749f, 0.86086690f, 0.85919720f,
+0.85751826f, 0.85583007f, 0.85413277f, 0.85242635f, 0.85071075f,
+0.84898609f, 0.84725231f, 0.84550947f, 0.84375757f, 0.84199661f,
+0.84022665f, 0.83844769f, 0.83665979f, 0.83486289f, 0.83305705f,
+0.83124226f, 0.82941860f, 0.82758605f, 0.82574469f, 0.82389444f,
+0.82203537f, 0.82016748f, 0.81829083f, 0.81640542f, 0.81451124f,
+0.81260836f, 0.81069672f, 0.80877650f, 0.80684757f, 0.80490994f,
+0.80296379f, 0.80100900f, 0.79904562f, 0.79707366f, 0.79509324f,
+0.79310423f, 0.79110676f, 0.78910083f, 0.78708643f, 0.78506362f,
+0.78303236f, 0.78099275f, 0.77894479f, 0.77688843f, 0.77482378f,
+0.77275085f, 0.77066964f, 0.76858020f, 0.76648247f, 0.76437658f,
+0.76226246f, 0.76014024f, 0.75800985f, 0.75587130f, 0.75372469f,
+0.75157005f, 0.74940729f, 0.74723655f, 0.74505776f, 0.74287105f,
+0.74067634f, 0.73847371f, 0.73626316f, 0.73404479f, 0.73181850f,
+0.72958434f, 0.72734243f, 0.72509271f, 0.72283524f, 0.72057003f,
+0.71829706f, 0.71601641f, 0.71372813f, 0.71143216f, 0.70912862f,
+0.70681745f, 0.70449871f, 0.70217246f, 0.69983864f, 0.69749737f,
+0.69514859f, 0.69279242f, 0.69042879f, 0.68805778f, 0.68567938f,
+0.68329364f, 0.68090063f, 0.67850029f, 0.67609268f, 0.67367786f,
+0.67125577f, 0.66882652f, 0.66639012f, 0.66394657f, 0.66149592f,
+0.65903819f, 0.65657341f, 0.65410155f, 0.65162271f, 0.64913690f,
+0.64664418f, 0.64414448f, 0.64163786f, 0.63912445f, 0.63660413f,
+0.63407701f, 0.63154310f, 0.62900239f, 0.62645501f, 0.62390089f,
+0.62134010f, 0.61877263f, 0.61619854f, 0.61361790f, 0.61103064f,
+0.60843682f, 0.60583651f, 0.60322970f, 0.60061646f, 0.59799677f,
+0.59537065f, 0.59273821f, 0.59009939f, 0.58745426f, 0.58480281f,
+0.58214509f, 0.57948118f, 0.57681108f, 0.57413477f, 0.57145232f,
+0.56876373f, 0.56606907f, 0.56336832f, 0.56066155f, 0.55794877f,
+0.55523002f, 0.55250537f, 0.54977477f, 0.54703826f, 0.54429591f,
+0.54154772f, 0.53879374f, 0.53603399f, 0.53326851f, 0.53049731f,
+0.52772039f, 0.52493787f, 0.52214974f, 0.51935595f, 0.51655668f,
+0.51375180f, 0.51094145f, 0.50812566f, 0.50530440f, 0.50247771f,
+0.49964568f, 0.49680826f, 0.49396557f, 0.49111754f, 0.48826426f,
+0.48540577f, 0.48254207f, 0.47967321f, 0.47679919f, 0.47392011f,
+0.47103590f, 0.46814668f, 0.46525243f, 0.46235323f, 0.45944905f,
+0.45653993f, 0.45362595f, 0.45070711f, 0.44778344f, 0.44485497f,
+0.44192174f, 0.43898380f, 0.43604112f, 0.43309379f, 0.43014181f,
+0.42718524f, 0.42422408f, 0.42125839f, 0.41828820f, 0.41531351f,
+0.41233435f, 0.40935081f, 0.40636289f, 0.40337059f, 0.40037400f,
+0.39737311f, 0.39436796f, 0.39135858f, 0.38834500f, 0.38532731f,
+0.38230544f, 0.37927949f, 0.37624949f, 0.37321547f, 0.37017745f,
+0.36713544f, 0.36408952f, 0.36103970f, 0.35798600f, 0.35492846f,
+0.35186714f, 0.34880206f, 0.34573323f, 0.34266070f, 0.33958447f,
+0.33650464f, 0.33342120f, 0.33033419f, 0.32724363f, 0.32414958f,
+0.32105204f, 0.31795108f, 0.31484672f, 0.31173897f, 0.30862790f,
+0.30551350f, 0.30239585f, 0.29927495f, 0.29615086f, 0.29302359f,
+0.28989318f, 0.28675964f, 0.28362307f, 0.28048345f, 0.27734083f,
+0.27419522f, 0.27104670f, 0.26789525f, 0.26474094f, 0.26158381f,
+0.25842386f, 0.25526115f, 0.25209570f, 0.24892756f, 0.24575676f,
+0.24258332f, 0.23940729f, 0.23622867f, 0.23304754f, 0.22986393f,
+0.22667783f, 0.22348931f, 0.22029841f, 0.21710514f, 0.21390954f,
+0.21071166f, 0.20751151f, 0.20430915f, 0.20110460f, 0.19789790f,
+0.19468907f, 0.19147816f, 0.18826519f, 0.18505022f, 0.18183327f,
+0.17861435f, 0.17539354f, 0.17217083f, 0.16894630f, 0.16571994f,
+0.16249183f, 0.15926196f, 0.15603039f, 0.15279715f, 0.14956227f,
+0.14632578f, 0.14308774f, 0.13984816f, 0.13660708f, 0.13336454f,
+0.13012058f, 0.12687522f, 0.12362850f, 0.12038045f, 0.11713112f,
+0.11388054f, 0.11062872f, 0.10737573f, 0.10412160f, 0.10086634f,
+0.097609997f, 0.094352618f, 0.091094226f, 0.087834857f, 0.084574550f,
+0.081313334f, 0.078051247f, 0.074788325f, 0.071524605f, 0.068260118f,
+0.064994894f, 0.061728980f, 0.058462404f, 0.055195201f, 0.051927410f,
+0.048659060f, 0.045390189f, 0.042120833f, 0.038851023f, 0.035580799f,
+0.032310195f, 0.029039243f, 0.025767982f, 0.022496443f, 0.019224664f,
+0.015952680f, 0.012680525f, 0.0094082337f, 0.0061358409f, 0.0028633832f,
+-0.00040910527f, -0.0036815894f, -0.0069540343f, -0.010226404f, -0.013498665f,
+-0.016770782f, -0.020042717f, -0.023314439f, -0.026585912f, -0.029857099f,
+-0.033127967f, -0.036398482f, -0.039668605f, -0.042938303f, -0.046207540f,
+-0.049476285f, -0.052744497f, -0.056012146f, -0.059279196f, -0.062545612f,
+-0.065811358f, -0.069076397f, -0.072340697f, -0.075604223f, -0.078866936f,
+-0.082128808f, -0.085389800f, -0.088649876f, -0.091909006f, -0.095167145f,
+-0.098424271f, -0.10168034f, -0.10493532f, -0.10818918f, -0.11144188f,
+-0.11469338f, -0.11794366f, -0.12119267f, -0.12444039f, -0.12768677f,
+-0.13093179f, -0.13417540f, -0.13741758f, -0.14065829f, -0.14389749f,
+-0.14713514f, -0.15037122f, -0.15360570f, -0.15683852f, -0.16006967f,
+-0.16329910f, -0.16652679f, -0.16975269f, -0.17297678f, -0.17619900f,
+-0.17941935f, -0.18263777f, -0.18585424f, -0.18906870f, -0.19228116f,
+-0.19549155f, -0.19869985f, -0.20190603f, -0.20511003f, -0.20831184f,
+-0.21151142f, -0.21470875f, -0.21790376f, -0.22109644f, -0.22428675f,
+-0.22747467f, -0.23066014f, -0.23384315f, -0.23702365f, -0.24020162f,
+-0.24337701f, -0.24654980f, -0.24971995f, -0.25288740f, -0.25605217f,
+-0.25921419f, -0.26237345f, -0.26552987f, -0.26868346f, -0.27183419f,
+-0.27498198f, -0.27812684f, -0.28126872f, -0.28440759f, -0.28754342f,
+-0.29067615f, -0.29380578f, -0.29693225f, -0.30005556f, -0.30317566f,
+-0.30629250f, -0.30940607f, -0.31251630f, -0.31562322f, -0.31872672f,
+-0.32182685f, -0.32492352f, -0.32801670f, -0.33110636f, -0.33419248f,
+-0.33727503f, -0.34035397f, -0.34342924f, -0.34650084f, -0.34956875f,
+-0.35263291f, -0.35569328f, -0.35874987f, -0.36180258f, -0.36485144f,
+-0.36789638f, -0.37093741f, -0.37397444f, -0.37700745f, -0.38003644f,
+-0.38306138f, -0.38608220f, -0.38909888f, -0.39211139f, -0.39511973f,
+-0.39812380f, -0.40112361f, -0.40411916f, -0.40711036f, -0.41009718f,
+-0.41307965f, -0.41605768f, -0.41903123f, -0.42200032f, -0.42496487f,
+-0.42792490f, -0.43088034f, -0.43383113f, -0.43677729f, -0.43971881f,
+-0.44265559f, -0.44558764f, -0.44851488f, -0.45143735f, -0.45435500f,
+-0.45726776f, -0.46017563f, -0.46307856f, -0.46597654f, -0.46886954f,
+-0.47175750f, -0.47464043f, -0.47751826f, -0.48039100f, -0.48325855f,
+-0.48612097f, -0.48897815f, -0.49183011f, -0.49467680f, -0.49751821f,
+-0.50035429f, -0.50318497f, -0.50601029f, -0.50883019f, -0.51164466f,
+-0.51445359f, -0.51725709f, -0.52005500f, -0.52284735f, -0.52563411f,
+-0.52841520f, -0.53119069f, -0.53396046f, -0.53672451f, -0.53948283f,
+-0.54223537f, -0.54498214f, -0.54772300f, -0.55045801f, -0.55318713f,
+-0.55591035f, -0.55862761f, -0.56133890f, -0.56404412f, -0.56674337f,
+-0.56943649f, -0.57212353f, -0.57480448f, -0.57747924f, -0.58014780f,
+-0.58281022f, -0.58546633f, -0.58811617f, -0.59075975f, -0.59339696f,
+-0.59602785f, -0.59865236f, -0.60127044f, -0.60388207f, -0.60648727f,
+-0.60908598f, -0.61167812f, -0.61426371f, -0.61684275f, -0.61941516f,
+-0.62198097f, -0.62454009f, -0.62709254f, -0.62963831f, -0.63217729f,
+-0.63470948f, -0.63723493f, -0.63975352f, -0.64226526f, -0.64477009f,
+-0.64726806f, -0.64975911f, -0.65224314f, -0.65472025f, -0.65719032f,
+-0.65965337f, -0.66210932f, -0.66455823f, -0.66700000f, -0.66943461f,
+-0.67186207f, -0.67428231f, -0.67669535f, -0.67910111f, -0.68149966f,
+-0.68389088f, -0.68627477f, -0.68865126f, -0.69102043f, -0.69338220f,
+-0.69573659f, -0.69808346f, -0.70042288f, -0.70275480f, -0.70507920f,
+-0.70739603f, -0.70970529f, -0.71200693f, -0.71430099f, -0.71658736f,
+-0.71886611f, -0.72113711f, -0.72340041f, -0.72565591f, -0.72790372f,
+-0.73014367f, -0.73237586f, -0.73460019f, -0.73681659f, -0.73902518f,
+-0.74122584f, -0.74341851f, -0.74560326f, -0.74778003f, -0.74994880f,
+-0.75210953f, -0.75426215f, -0.75640678f, -0.75854325f, -0.76067162f,
+-0.76279181f, -0.76490390f, -0.76700771f, -0.76910341f, -0.77119076f,
+-0.77326995f, -0.77534080f, -0.77740335f, -0.77945763f, -0.78150350f,
+-0.78354102f, -0.78557014f, -0.78759086f, -0.78960317f, -0.79160696f,
+-0.79360235f, -0.79558921f, -0.79756755f, -0.79953730f, -0.80149853f,
+-0.80345118f, -0.80539525f, -0.80733067f, -0.80925739f, -0.81117553f,
+-0.81308490f, -0.81498563f, -0.81687760f, -0.81876087f, -0.82063532f,
+-0.82250100f, -0.82435787f, -0.82620591f, -0.82804507f, -0.82987541f,
+-0.83169687f, -0.83350939f, -0.83531296f, -0.83710766f, -0.83889335f,
+-0.84067005f, -0.84243774f, -0.84419644f, -0.84594607f, -0.84768665f,
+-0.84941816f, -0.85114056f, -0.85285389f, -0.85455805f, -0.85625303f,
+-0.85793889f, -0.85961550f, -0.86128294f, -0.86294121f, -0.86459017f,
+-0.86622989f, -0.86786032f, -0.86948150f, -0.87109333f, -0.87269586f,
+-0.87428904f, -0.87587279f, -0.87744725f, -0.87901229f, -0.88056785f,
+-0.88211405f, -0.88365078f, -0.88517809f, -0.88669586f, -0.88820416f,
+-0.88970292f, -0.89119220f, -0.89267188f, -0.89414203f, -0.89560264f,
+-0.89705360f, -0.89849502f, -0.89992678f, -0.90134889f, -0.90276134f,
+-0.90416414f, -0.90555727f, -0.90694070f, -0.90831441f, -0.90967834f,
+-0.91103262f, -0.91237706f, -0.91371179f, -0.91503674f, -0.91635185f,
+-0.91765714f, -0.91895264f, -0.92023826f, -0.92151409f, -0.92277998f,
+-0.92403603f, -0.92528218f, -0.92651838f, -0.92774469f, -0.92896110f,
+-0.93016750f, -0.93136400f, -0.93255049f, -0.93372697f, -0.93489349f,
+-0.93604994f, -0.93719643f, -0.93833286f, -0.93945926f, -0.94057560f,
+-0.94168180f, -0.94277799f, -0.94386405f, -0.94494003f, -0.94600588f,
+-0.94706154f, -0.94810712f, -0.94914252f, -0.95016778f, -0.95118284f,
+-0.95218778f, -0.95318246f, -0.95416695f, -0.95514119f, -0.95610523f,
+-0.95705903f, -0.95800257f, -0.95893586f, -0.95985889f, -0.96077162f,
+-0.96167403f, -0.96256620f, -0.96344805f, -0.96431959f, -0.96518075f,
+-0.96603161f, -0.96687216f, -0.96770233f, -0.96852213f, -0.96933156f,
+-0.97013056f, -0.97091925f, -0.97169751f, -0.97246534f, -0.97322279f,
+-0.97396982f, -0.97470641f, -0.97543252f, -0.97614825f, -0.97685349f,
+-0.97754824f, -0.97823256f, -0.97890645f, -0.97956979f, -0.98022264f,
+-0.98086500f, -0.98149687f, -0.98211825f, -0.98272908f, -0.98332942f,
+-0.98391914f, -0.98449844f, -0.98506713f, -0.98562527f, -0.98617285f,
+-0.98670989f, -0.98723638f, -0.98775226f, -0.98825759f, -0.98875231f,
+-0.98923647f, -0.98971003f, -0.99017298f, -0.99062532f, -0.99106705f,
+-0.99149817f, -0.99191868f, -0.99232858f, -0.99272782f, -0.99311644f,
+-0.99349445f, -0.99386179f, -0.99421853f, -0.99456459f, -0.99489999f,
+-0.99522477f, -0.99553883f, -0.99584228f, -0.99613506f, -0.99641716f,
+-0.99668860f, -0.99694937f, -0.99719942f, -0.99743885f, -0.99766755f,
+-0.99788558f, -0.99809295f, -0.99828959f, -0.99847561f, -0.99865085f,
+-0.99881548f, -0.99896932f, -0.99911255f, -0.99924499f, -0.99936682f,
+-0.99947786f, -0.99957830f, -0.99966794f, -0.99974692f, -0.99981517f,
+-0.99987274f, -0.99991959f, -0.99995571f, -0.99998116f, -0.99999589f,
+0.99999964f, 0.99997288f, 0.99990326f, 0.99979085f, 0.99963558f,
+0.99943751f, 0.99919659f, 0.99891287f, 0.99858636f, 0.99821711f,
+0.99780506f, 0.99735034f, 0.99685282f, 0.99631262f, 0.99572974f,
+0.99510419f, 0.99443603f, 0.99372530f, 0.99297196f, 0.99217612f,
+0.99133772f, 0.99045694f, 0.98953366f, 0.98856801f, 0.98756003f,
+0.98650974f, 0.98541719f, 0.98428243f, 0.98310548f, 0.98188645f,
+0.98062533f, 0.97932225f, 0.97797716f, 0.97659022f, 0.97516143f,
+0.97369087f, 0.97217858f, 0.97062469f, 0.96902919f, 0.96739221f,
+0.96571374f, 0.96399397f, 0.96223283f, 0.96043050f, 0.95858705f,
+0.95670253f, 0.95477700f, 0.95281059f, 0.95080340f, 0.94875544f,
+0.94666684f, 0.94453770f, 0.94236809f, 0.94015813f, 0.93790787f,
+0.93561745f, 0.93328691f, 0.93091643f, 0.92850608f, 0.92605597f,
+0.92356616f, 0.92103678f, 0.91846794f, 0.91585976f, 0.91321236f,
+0.91052586f, 0.90780038f, 0.90503591f, 0.90223277f, 0.89939094f,
+0.89651060f, 0.89359182f, 0.89063478f, 0.88763964f, 0.88460642f,
+0.88153529f, 0.87842643f, 0.87527996f, 0.87209594f, 0.86887461f,
+0.86561602f, 0.86232042f, 0.85898781f, 0.85561842f, 0.85221243f,
+0.84876984f, 0.84529096f, 0.84177583f, 0.83822471f, 0.83463764f,
+0.83101481f, 0.82735640f, 0.82366252f, 0.81993335f, 0.81616908f,
+0.81236988f, 0.80853581f, 0.80466717f, 0.80076402f, 0.79682660f,
+0.79285502f, 0.78884947f, 0.78481019f, 0.78073722f, 0.77663082f,
+0.77249116f, 0.76831841f, 0.76411277f, 0.75987434f, 0.75560343f,
+0.75130010f, 0.74696463f, 0.74259710f, 0.73819780f, 0.73376691f,
+0.72930455f, 0.72481096f, 0.72028631f, 0.71573079f, 0.71114463f,
+0.70652801f, 0.70188117f, 0.69720417f, 0.69249737f, 0.68776089f,
+0.68299496f, 0.67819971f, 0.67337549f, 0.66852236f, 0.66364062f,
+0.65873051f, 0.65379208f, 0.64882571f, 0.64383155f, 0.63880974f,
+0.63376063f, 0.62868434f, 0.62358117f, 0.61845124f, 0.61329484f,
+0.60811216f, 0.60290343f, 0.59766883f, 0.59240872f, 0.58712316f,
+0.58181250f, 0.57647687f, 0.57111657f, 0.56573176f, 0.56032276f,
+0.55488980f, 0.54943299f, 0.54395270f, 0.53844911f, 0.53292239f,
+0.52737290f, 0.52180082f, 0.51620632f, 0.51058978f, 0.50495136f,
+0.49929130f, 0.49360985f, 0.48790723f, 0.48218375f, 0.47643960f,
+0.47067502f, 0.46489030f, 0.45908567f, 0.45326138f, 0.44741765f,
+0.44155475f, 0.43567297f, 0.42977250f, 0.42385364f, 0.41791660f,
+0.41196167f, 0.40598908f, 0.39999911f, 0.39399201f, 0.38796803f,
+0.38192743f, 0.37587047f, 0.36979741f, 0.36370850f, 0.35760403f,
+0.35148421f, 0.34534934f, 0.33919969f, 0.33303553f, 0.32685706f,
+0.32066461f, 0.31445843f, 0.30823877f, 0.30200592f, 0.29576012f,
+0.28950164f, 0.28323078f, 0.27694780f, 0.27065292f, 0.26434645f,
+0.25802869f, 0.25169984f, 0.24536023f, 0.23901010f, 0.23264973f,
+0.22627939f, 0.21989937f, 0.21350993f, 0.20711134f, 0.20070387f,
+0.19428782f, 0.18786344f, 0.18143101f, 0.17499080f, 0.16854310f,
+0.16208819f, 0.15562633f, 0.14915779f, 0.14268288f, 0.13620184f,
+0.12971498f, 0.12322257f, 0.11672486f, 0.11022217f, 0.10371475f,
+0.097202882f, 0.090686858f, 0.084166944f, 0.077643424f, 0.071116582f,
+0.064586692f, 0.058054037f, 0.051518895f, 0.044981543f, 0.038442269f,
+0.031901345f, 0.025359053f, 0.018815678f, 0.012271495f, 0.0057267868f,
+-0.00081816671f, -0.0073630852f, -0.013907688f, -0.020451695f, -0.026994826f,
+-0.033536803f, -0.040077340f, -0.046616159f, -0.053152986f, -0.059687532f,
+-0.066219524f, -0.072748676f, -0.079274714f, -0.085797355f, -0.092316322f,
+-0.098831341f, -0.10534211f, -0.11184838f, -0.11834986f, -0.12484626f,
+-0.13133731f, -0.13782275f, -0.14430228f, -0.15077563f, -0.15724251f,
+-0.16370267f, -0.17015581f, -0.17660165f, -0.18303993f, -0.18947038f,
+-0.19589271f, -0.20230664f, -0.20871192f, -0.21510825f, -0.22149536f,
+-0.22787298f, -0.23424086f, -0.24059868f, -0.24694622f, -0.25328314f,
+-0.25960925f, -0.26592422f, -0.27222782f, -0.27851975f, -0.28479972f,
+-0.29106751f, -0.29732284f, -0.30356544f, -0.30979502f, -0.31601134f,
+-0.32221413f, -0.32840309f, -0.33457801f, -0.34073856f, -0.34688455f,
+-0.35301566f, -0.35913166f, -0.36523229f, -0.37131724f, -0.37738630f,
+-0.38343921f, -0.38947567f, -0.39549544f, -0.40149832f, -0.40748394f,
+-0.41345215f, -0.41940263f, -0.42533514f, -0.43124944f, -0.43714526f,
+-0.44302234f, -0.44888046f, -0.45471936f, -0.46053877f, -0.46633846f,
+-0.47211814f, -0.47787762f, -0.48361665f, -0.48933494f, -0.49503228f,
+-0.50070840f, -0.50636309f, -0.51199609f, -0.51760709f, -0.52319598f,
+-0.52876246f, -0.53430629f, -0.53982723f, -0.54532504f, -0.55079949f,
+-0.55625033f, -0.56167740f, -0.56708032f, -0.57245898f, -0.57781315f,
+-0.58314258f, -0.58844697f, -0.59372622f, -0.59897995f, -0.60420811f,
+-0.60941035f, -0.61458647f, -0.61973625f, -0.62485951f, -0.62995601f,
+-0.63502556f, -0.64006782f, -0.64508271f, -0.65007001f, -0.65502942f,
+-0.65996075f, -0.66486382f, -0.66973841f, -0.67458433f, -0.67940134f,
+-0.68418926f, -0.68894786f, -0.69367695f, -0.69837630f, -0.70304573f,
+-0.70768511f, -0.71229410f, -0.71687263f, -0.72142041f, -0.72593731f,
+-0.73042315f, -0.73487765f, -0.73930067f, -0.74369204f, -0.74805158f,
+-0.75237900f, -0.75667429f, -0.76093709f, -0.76516730f, -0.76936477f,
+-0.77352923f, -0.77766061f, -0.78175867f, -0.78582323f, -0.78985411f,
+-0.79385114f, -0.79781419f, -0.80174309f, -0.80563760f, -0.80949765f,
+-0.81332302f, -0.81711352f, -0.82086903f, -0.82458937f, -0.82827437f,
+-0.83192390f, -0.83553779f, -0.83911592f, -0.84265804f, -0.84616417f,
+-0.84963393f, -0.85306740f, -0.85646427f, -0.85982448f, -0.86314780f,
+-0.86643422f, -0.86968350f, -0.87289548f, -0.87607014f, -0.87920725f,
+-0.88230664f, -0.88536829f, -0.88839203f, -0.89137769f, -0.89432514f,
+-0.89723432f, -0.90010506f, -0.90293723f, -0.90573072f, -0.90848541f,
+-0.91120118f, -0.91387796f, -0.91651553f, -0.91911387f, -0.92167282f,
+-0.92419231f, -0.92667222f, -0.92911243f, -0.93151283f, -0.93387336f,
+-0.93619382f, -0.93847424f, -0.94071442f, -0.94291431f, -0.94507378f,
+-0.94719279f, -0.94927126f, -0.95130903f, -0.95330608f, -0.95526224f,
+-0.95717752f, -0.95905179f, -0.96088499f, -0.96267700f, -0.96442777f,
+-0.96613729f, -0.96780539f, -0.96943200f, -0.97101706f, -0.97256058f,
+-0.97406244f, -0.97552258f, -0.97694093f, -0.97831738f, -0.97965199f,
+-0.98094457f, -0.98219514f, -0.98340368f, -0.98457009f, -0.98569429f,
+-0.98677629f, -0.98781598f, -0.98881340f, -0.98976845f, -0.99068111f,
+-0.99155134f, -0.99237907f, -0.99316430f, -0.99390697f, -0.99460709f,
+-0.99526459f, -0.99587947f, -0.99645168f, -0.99698120f, -0.99746799f,
+-0.99791211f, -0.99831343f, -0.99867201f, -0.99898779f, -0.99926084f,
+-0.99949104f, -0.99967843f, -0.99982297f, -0.99992472f, -0.99998361f,
+0.99999869f, 0.99989158f, 0.99961317f, 0.99916345f, 0.99854255f,
+0.99775058f, 0.99678761f, 0.99565387f, 0.99434954f, 0.99287480f,
+0.99122995f, 0.98941529f, 0.98743105f, 0.98527765f, 0.98295540f,
+0.98046476f, 0.97780609f, 0.97497988f, 0.97198665f, 0.96882683f,
+0.96550101f, 0.96200979f, 0.95835376f, 0.95453346f, 0.95054960f,
+0.94640291f, 0.94209403f, 0.93762374f, 0.93299282f, 0.92820197f,
+0.92325211f, 0.91814411f, 0.91287869f, 0.90745693f, 0.90187967f,
+0.89614785f, 0.89026248f, 0.88422459f, 0.87803519f, 0.87169534f,
+0.86520612f, 0.85856867f, 0.85178405f, 0.84485358f, 0.83777827f,
+0.83055943f, 0.82319832f, 0.81569612f, 0.80805415f, 0.80027372f,
+0.79235619f, 0.78430289f, 0.77611518f, 0.76779449f, 0.75934225f,
+0.75075996f, 0.74204898f, 0.73321080f, 0.72424710f, 0.71515924f,
+0.70594883f, 0.69661748f, 0.68716675f, 0.67759830f, 0.66791373f,
+0.65811473f, 0.64820296f, 0.63818014f, 0.62804794f, 0.61780810f,
+0.60746247f, 0.59701276f, 0.58646071f, 0.57580817f, 0.56505698f,
+0.55420899f, 0.54326600f, 0.53222996f, 0.52110273f, 0.50988621f,
+0.49858227f, 0.48719296f, 0.47572014f, 0.46416581f, 0.45253196f,
+0.44082057f, 0.42903364f, 0.41717321f, 0.40524128f, 0.39323992f,
+0.38117120f, 0.36903715f, 0.35683987f, 0.34458145f, 0.33226398f,
+0.31988961f, 0.30746040f, 0.29497850f, 0.28244606f, 0.26986524f,
+0.25723818f, 0.24456702f, 0.23185398f, 0.21910121f, 0.20631088f,
+0.19348522f, 0.18062639f, 0.16773662f, 0.15481812f, 0.14187308f,
+0.12890373f, 0.11591230f, 0.10290100f, 0.089872077f, 0.076827750f,
+0.063770257f, 0.050701842f, 0.037624735f, 0.024541186f, 0.011453429f,
+-0.0016362892f, -0.014725727f, -0.027812643f, -0.040894791f, -0.053969935f,
+-0.067035832f, -0.080090240f, -0.093130924f, -0.10615565f, -0.11916219f,
+-0.13214831f, -0.14511178f, -0.15805040f, -0.17096193f, -0.18384418f,
+-0.19669491f, -0.20951195f, -0.22229309f, -0.23503613f, -0.24773891f,
+-0.26039925f, -0.27301496f, -0.28558388f, -0.29810387f, -0.31057280f,
+-0.32298848f, -0.33534884f, -0.34765175f, -0.35989508f, -0.37207675f,
+-0.38419467f, -0.39624676f, -0.40823093f, -0.42014518f, -0.43198743f,
+-0.44375566f, -0.45544785f, -0.46706200f, -0.47859612f, -0.49004826f,
+-0.50141639f, -0.51269865f, -0.52389306f, -0.53499764f, -0.54601061f,
+-0.55693001f, -0.56775403f, -0.57848072f, -0.58910829f, -0.59963489f,
+-0.61005878f, -0.62037814f, -0.63059121f, -0.64069623f, -0.65069145f,
+-0.66057515f, -0.67034572f, -0.68000144f, -0.68954057f, -0.69896162f,
+-0.70826286f, -0.71744281f, -0.72649974f, -0.73543227f, -0.74423873f,
+-0.75291771f, -0.76146764f, -0.76988715f, -0.77817470f, -0.78632891f,
+-0.79434842f, -0.80223179f, -0.80997771f, -0.81758487f, -0.82505190f,
+-0.83237761f, -0.83956063f, -0.84659988f, -0.85349399f, -0.86024189f,
+-0.86684239f, -0.87329435f, -0.87959671f, -0.88574833f, -0.89174819f,
+-0.89759529f, -0.90328854f, -0.90882701f, -0.91420978f, -0.91943592f,
+-0.92450452f, -0.92941469f, -0.93416560f, -0.93875647f, -0.94318646f,
+-0.94745487f, -0.95156091f, -0.95550388f, -0.95928317f, -0.96289814f,
+-0.96634805f, -0.96963239f, -0.97275060f, -0.97570217f, -0.97848648f,
+-0.98110318f, -0.98355180f, -0.98583186f, -0.98794299f, -0.98988485f,
+-0.99165714f, -0.99325943f, -0.99469161f, -0.99595332f, -0.99704438f,
+-0.99796462f, -0.99871385f, -0.99929196f, -0.99969882f, -0.99993443f,
+0.99999464f, 0.99956632f, 0.99845290f, 0.99665523f, 0.99417448f,
+0.99101239f, 0.98717111f, 0.98265326f, 0.97746199f, 0.97160077f,
+0.96507365f, 0.95788515f, 0.95004016f, 0.94154406f, 0.93240267f,
+0.92262226f, 0.91220951f, 0.90117162f, 0.88951606f, 0.87725091f,
+0.86438453f, 0.85092574f, 0.83688372f, 0.82226819f, 0.80708915f,
+0.79135692f, 0.77508235f, 0.75827658f, 0.74095112f, 0.72311783f,
+0.70478898f, 0.68597710f, 0.66669506f, 0.64695615f, 0.62677377f,
+0.60616189f, 0.58513457f, 0.56370622f, 0.54189157f, 0.51970547f,
+0.49716324f, 0.47428027f, 0.45107225f, 0.42755505f, 0.40374488f,
+0.37965798f, 0.35531086f, 0.33072025f, 0.30590299f, 0.28087607f,
+0.25565663f, 0.23026201f, 0.20470956f, 0.17901683f, 0.15320139f,
+0.12728097f, 0.10127331f, 0.075196236f, 0.049067631f, 0.022905400f,
+-0.0032725304f, -0.029448219f, -0.055603724f, -0.081721120f, -0.10778251f,
+-0.13377003f, -0.15966587f, -0.18545228f, -0.21111161f, -0.23662624f,
+-0.26197869f, -0.28715160f, -0.31212771f, -0.33688989f, -0.36142120f,
+-0.38570482f, -0.40972409f, -0.43346253f, -0.45690393f, -0.48003218f,
+-0.50283146f, -0.52528608f, -0.54738069f, -0.56910020f, -0.59042966f,
+-0.61135447f, -0.63186026f, -0.65193301f, -0.67155898f, -0.69072473f,
+-0.70941705f, -0.72762316f, -0.74533063f, -0.76252723f, -0.77920127f,
+-0.79534131f, -0.81093621f, -0.82597536f, -0.84044844f, -0.85434550f,
+-0.86765707f, -0.88037395f, -0.89248747f, -0.90398932f, -0.91487163f,
+-0.92512697f, -0.93474823f, -0.94372886f, -0.95206273f, -0.95974404f,
+-0.96676767f, -0.97312868f, -0.97882277f, -0.98384601f, -0.98819500f,
+-0.99186671f, -0.99485862f, -0.99716878f, -0.99879545f, -0.99973762f,
+};
 #endif
 
 static const CELTMode mode48000_960_120 = {
diff --git a/drivers/opus/celt/static_modes_float_arm_ne10.h b/drivers/opus/celt/static_modes_float_arm_ne10.h
new file mode 100644
index 0000000000..934a82a420
--- /dev/null
+++ b/drivers/opus/celt/static_modes_float_arm_ne10.h
@@ -0,0 +1,404 @@
+/* The contents of this file was automatically generated by
+ * dump_mode_arm_ne10.c with arguments: 48000 960
+ * It contains static definitions for some pre-defined modes. */
+#include <NE10_init.h>
+
+#ifndef NE10_FFT_PARAMS48000_960
+#define NE10_FFT_PARAMS48000_960
+static const ne10_int32_t ne10_factors_480[64] = {
+4, 40, 4, 30, 2, 15, 5, 3, 3, 1, 1, 0, 0, 0, 0,
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, };
+static const ne10_int32_t ne10_factors_240[64] = {
+3, 20, 4, 15, 5, 3, 3, 1, 1, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, };
+static const ne10_int32_t ne10_factors_120[64] = {
+3, 10, 2, 15, 5, 3, 3, 1, 1, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, };
+static const ne10_int32_t ne10_factors_60[64] = {
+2, 5, 5, 3, 3, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+0, 0, 0, 0, };
+static const ne10_fft_cpx_float32_t ne10_twiddles_480[480] = {
+{1.0000000f,0.0000000f}, {1.0000000f,-0.0000000f}, {1.0000000f,-0.0000000f},
+{1.0000000f,-0.0000000f}, {0.91354543f,-0.40673664f}, {0.66913056f,-0.74314487f},
+{1.0000000f,-0.0000000f}, {0.66913056f,-0.74314487f}, {-0.10452851f,-0.99452192f},
+{1.0000000f,-0.0000000f}, {0.30901697f,-0.95105654f}, {-0.80901700f,-0.58778518f},
+{1.0000000f,-0.0000000f}, {-0.10452851f,-0.99452192f}, {-0.97814757f,0.20791179f},
+{1.0000000f,-0.0000000f}, {0.97814763f,-0.20791170f}, {0.91354543f,-0.40673664f},
+{0.80901700f,-0.58778524f}, {0.66913056f,-0.74314487f}, {0.49999997f,-0.86602545f},
+{0.30901697f,-0.95105654f}, {0.10452842f,-0.99452192f}, {-0.10452851f,-0.99452192f},
+{-0.30901703f,-0.95105648f}, {-0.50000006f,-0.86602533f}, {-0.66913068f,-0.74314475f},
+{-0.80901700f,-0.58778518f}, {-0.91354549f,-0.40673658f}, {-0.97814763f,-0.20791161f},
+{1.0000000f,-0.0000000f}, {0.99862951f,-0.052335959f}, {0.99452192f,-0.10452846f},
+{0.98768836f,-0.15643448f}, {0.97814763f,-0.20791170f}, {0.96592581f,-0.25881904f},
+{0.95105648f,-0.30901700f}, {0.93358040f,-0.35836795f}, {0.91354543f,-0.40673664f},
+{0.89100653f,-0.45399052f}, {0.86602545f,-0.50000000f}, {0.83867055f,-0.54463905f},
+{0.80901700f,-0.58778524f}, {0.77714598f,-0.62932038f}, {0.74314475f,-0.66913062f},
+{0.70710677f,-0.70710683f}, {0.66913056f,-0.74314487f}, {0.62932038f,-0.77714598f},
+{0.58778524f,-0.80901700f}, {0.54463899f,-0.83867055f}, {0.49999997f,-0.86602545f},
+{0.45399052f,-0.89100653f}, {0.40673661f,-0.91354549f}, {0.35836786f,-0.93358046f},
+{0.30901697f,-0.95105654f}, {0.25881907f,-0.96592581f}, {0.20791166f,-0.97814763f},
+{0.15643437f,-0.98768836f}, {0.10452842f,-0.99452192f}, {0.052335974f,-0.99862951f},
+{1.0000000f,-0.0000000f}, {0.99452192f,-0.10452846f}, {0.97814763f,-0.20791170f},
+{0.95105648f,-0.30901700f}, {0.91354543f,-0.40673664f}, {0.86602545f,-0.50000000f},
+{0.80901700f,-0.58778524f}, {0.74314475f,-0.66913062f}, {0.66913056f,-0.74314487f},
+{0.58778524f,-0.80901700f}, {0.49999997f,-0.86602545f}, {0.40673661f,-0.91354549f},
+{0.30901697f,-0.95105654f}, {0.20791166f,-0.97814763f}, {0.10452842f,-0.99452192f},
+{-4.3711388e-08f,-1.0000000f}, {-0.10452851f,-0.99452192f}, {-0.20791174f,-0.97814757f},
+{-0.30901703f,-0.95105648f}, {-0.40673670f,-0.91354543f}, {-0.50000006f,-0.86602533f},
+{-0.58778518f,-0.80901700f}, {-0.66913068f,-0.74314475f}, {-0.74314493f,-0.66913044f},
+{-0.80901700f,-0.58778518f}, {-0.86602539f,-0.50000006f}, {-0.91354549f,-0.40673658f},
+{-0.95105654f,-0.30901679f}, {-0.97814763f,-0.20791161f}, {-0.99452192f,-0.10452849f},
+{1.0000000f,-0.0000000f}, {0.98768836f,-0.15643448f}, {0.95105648f,-0.30901700f},
+{0.89100653f,-0.45399052f}, {0.80901700f,-0.58778524f}, {0.70710677f,-0.70710683f},
+{0.58778524f,-0.80901700f}, {0.45399052f,-0.89100653f}, {0.30901697f,-0.95105654f},
+{0.15643437f,-0.98768836f}, {-4.3711388e-08f,-1.0000000f}, {-0.15643445f,-0.98768836f},
+{-0.30901703f,-0.95105648f}, {-0.45399061f,-0.89100647f}, {-0.58778518f,-0.80901700f},
+{-0.70710677f,-0.70710677f}, {-0.80901700f,-0.58778518f}, {-0.89100659f,-0.45399037f},
+{-0.95105654f,-0.30901679f}, {-0.98768836f,-0.15643445f}, {-1.0000000f,8.7422777e-08f},
+{-0.98768830f,0.15643461f}, {-0.95105654f,0.30901697f}, {-0.89100653f,0.45399055f},
+{-0.80901694f,0.58778536f}, {-0.70710665f,0.70710689f}, {-0.58778507f,0.80901712f},
+{-0.45399022f,0.89100665f}, {-0.30901709f,0.95105648f}, {-0.15643452f,0.98768830f},
+{1.0000000f,-0.0000000f}, {0.99991435f,-0.013089596f}, {0.99965733f,-0.026176950f},
+{0.99922901f,-0.039259817f}, {0.99862951f,-0.052335959f}, {0.99785894f,-0.065403134f},
+{0.99691731f,-0.078459099f}, {0.99580491f,-0.091501623f}, {0.99452192f,-0.10452846f},
+{0.99306846f,-0.11753740f}, {0.99144489f,-0.13052620f}, {0.98965138f,-0.14349262f},
+{0.98768836f,-0.15643448f}, {0.98555607f,-0.16934951f}, {0.98325491f,-0.18223552f},
+{0.98078525f,-0.19509032f}, {0.97814763f,-0.20791170f}, {0.97534233f,-0.22069745f},
+{0.97236991f,-0.23344538f}, {0.96923089f,-0.24615330f}, {0.96592581f,-0.25881904f},
+{0.96245521f,-0.27144045f}, {0.95881975f,-0.28401536f}, {0.95501995f,-0.29654160f},
+{0.95105648f,-0.30901700f}, {0.94693011f,-0.32143945f}, {0.94264150f,-0.33380687f},
+{0.93819129f,-0.34611708f}, {0.93358040f,-0.35836795f}, {0.92880952f,-0.37055743f},
+{0.92387956f,-0.38268346f}, {0.91879117f,-0.39474389f}, {0.91354543f,-0.40673664f},
+{0.90814316f,-0.41865975f}, {0.90258527f,-0.43051112f}, {0.89687270f,-0.44228873f},
+{0.89100653f,-0.45399052f}, {0.88498765f,-0.46561453f}, {0.87881708f,-0.47715878f},
+{0.87249601f,-0.48862126f}, {0.86602545f,-0.50000000f}, {0.85940641f,-0.51129311f},
+{0.85264015f,-0.52249855f}, {0.84572786f,-0.53361452f}, {0.83867055f,-0.54463905f},
+{0.83146960f,-0.55557024f}, {0.82412618f,-0.56640625f}, {0.81664151f,-0.57714522f},
+{0.80901700f,-0.58778524f}, {0.80125380f,-0.59832460f}, {0.79335332f,-0.60876143f},
+{0.78531694f,-0.61909395f}, {0.77714598f,-0.62932038f}, {0.76884180f,-0.63943899f},
+{0.76040596f,-0.64944810f}, {0.75183982f,-0.65934587f}, {0.74314475f,-0.66913062f},
+{0.73432249f,-0.67880076f}, {0.72537434f,-0.68835455f}, {0.71630192f,-0.69779050f},
+{0.70710677f,-0.70710683f}, {0.69779044f,-0.71630198f}, {0.68835455f,-0.72537440f},
+{0.67880070f,-0.73432255f}, {0.66913056f,-0.74314487f}, {0.65934581f,-0.75183982f},
+{0.64944804f,-0.76040596f}, {0.63943899f,-0.76884186f}, {0.62932038f,-0.77714598f},
+{0.61909395f,-0.78531694f}, {0.60876137f,-0.79335338f}, {0.59832460f,-0.80125386f},
+{0.58778524f,-0.80901700f}, {0.57714516f,-0.81664151f}, {0.56640625f,-0.82412618f},
+{0.55557019f,-0.83146960f}, {0.54463899f,-0.83867055f}, {0.53361452f,-0.84572786f},
+{0.52249849f,-0.85264015f}, {0.51129311f,-0.85940641f}, {0.49999997f,-0.86602545f},
+{0.48862118f,-0.87249601f}, {0.47715876f,-0.87881708f}, {0.46561447f,-0.88498765f},
+{0.45399052f,-0.89100653f}, {0.44228867f,-0.89687276f}, {0.43051103f,-0.90258533f},
+{0.41865975f,-0.90814316f}, {0.40673661f,-0.91354549f}, {0.39474380f,-0.91879129f},
+{0.38268343f,-0.92387956f}, {0.37055740f,-0.92880958f}, {0.35836786f,-0.93358046f},
+{0.34611705f,-0.93819135f}, {0.33380681f,-0.94264150f}, {0.32143947f,-0.94693011f},
+{0.30901697f,-0.95105654f}, {0.29654151f,-0.95501995f}, {0.28401533f,-0.95881975f},
+{0.27144039f,-0.96245527f}, {0.25881907f,-0.96592581f}, {0.24615327f,-0.96923089f},
+{0.23344530f,-0.97236991f}, {0.22069745f,-0.97534233f}, {0.20791166f,-0.97814763f},
+{0.19509023f,-0.98078531f}, {0.18223552f,-0.98325491f}, {0.16934945f,-0.98555607f},
+{0.15643437f,-0.98768836f}, {0.14349259f,-0.98965138f}, {0.13052613f,-0.99144489f},
+{0.11753740f,-0.99306846f}, {0.10452842f,-0.99452192f}, {0.091501534f,-0.99580491f},
+{0.078459084f,-0.99691731f}, {0.065403074f,-0.99785894f}, {0.052335974f,-0.99862951f},
+{0.039259788f,-0.99922901f}, {0.026176875f,-0.99965733f}, {0.013089597f,-0.99991435f},
+{1.0000000f,-0.0000000f}, {0.99965733f,-0.026176950f}, {0.99862951f,-0.052335959f},
+{0.99691731f,-0.078459099f}, {0.99452192f,-0.10452846f}, {0.99144489f,-0.13052620f},
+{0.98768836f,-0.15643448f}, {0.98325491f,-0.18223552f}, {0.97814763f,-0.20791170f},
+{0.97236991f,-0.23344538f}, {0.96592581f,-0.25881904f}, {0.95881975f,-0.28401536f},
+{0.95105648f,-0.30901700f}, {0.94264150f,-0.33380687f}, {0.93358040f,-0.35836795f},
+{0.92387956f,-0.38268346f}, {0.91354543f,-0.40673664f}, {0.90258527f,-0.43051112f},
+{0.89100653f,-0.45399052f}, {0.87881708f,-0.47715878f}, {0.86602545f,-0.50000000f},
+{0.85264015f,-0.52249855f}, {0.83867055f,-0.54463905f}, {0.82412618f,-0.56640625f},
+{0.80901700f,-0.58778524f}, {0.79335332f,-0.60876143f}, {0.77714598f,-0.62932038f},
+{0.76040596f,-0.64944810f}, {0.74314475f,-0.66913062f}, {0.72537434f,-0.68835455f},
+{0.70710677f,-0.70710683f}, {0.68835455f,-0.72537440f}, {0.66913056f,-0.74314487f},
+{0.64944804f,-0.76040596f}, {0.62932038f,-0.77714598f}, {0.60876137f,-0.79335338f},
+{0.58778524f,-0.80901700f}, {0.56640625f,-0.82412618f}, {0.54463899f,-0.83867055f},
+{0.52249849f,-0.85264015f}, {0.49999997f,-0.86602545f}, {0.47715876f,-0.87881708f},
+{0.45399052f,-0.89100653f}, {0.43051103f,-0.90258533f}, {0.40673661f,-0.91354549f},
+{0.38268343f,-0.92387956f}, {0.35836786f,-0.93358046f}, {0.33380681f,-0.94264150f},
+{0.30901697f,-0.95105654f}, {0.28401533f,-0.95881975f}, {0.25881907f,-0.96592581f},
+{0.23344530f,-0.97236991f}, {0.20791166f,-0.97814763f}, {0.18223552f,-0.98325491f},
+{0.15643437f,-0.98768836f}, {0.13052613f,-0.99144489f}, {0.10452842f,-0.99452192f},
+{0.078459084f,-0.99691731f}, {0.052335974f,-0.99862951f}, {0.026176875f,-0.99965733f},
+{-4.3711388e-08f,-1.0000000f}, {-0.026176963f,-0.99965733f}, {-0.052336060f,-0.99862951f},
+{-0.078459173f,-0.99691731f}, {-0.10452851f,-0.99452192f}, {-0.13052621f,-0.99144489f},
+{-0.15643445f,-0.98768836f}, {-0.18223560f,-0.98325491f}, {-0.20791174f,-0.97814757f},
+{-0.23344538f,-0.97236991f}, {-0.25881916f,-0.96592581f}, {-0.28401542f,-0.95881969f},
+{-0.30901703f,-0.95105648f}, {-0.33380687f,-0.94264150f}, {-0.35836795f,-0.93358040f},
+{-0.38268352f,-0.92387950f}, {-0.40673670f,-0.91354543f}, {-0.43051112f,-0.90258527f},
+{-0.45399061f,-0.89100647f}, {-0.47715873f,-0.87881708f}, {-0.50000006f,-0.86602533f},
+{-0.52249867f,-0.85264009f}, {-0.54463905f,-0.83867055f}, {-0.56640631f,-0.82412612f},
+{-0.58778518f,-0.80901700f}, {-0.60876143f,-0.79335332f}, {-0.62932050f,-0.77714586f},
+{-0.64944804f,-0.76040596f}, {-0.66913068f,-0.74314475f}, {-0.68835467f,-0.72537428f},
+{-0.70710677f,-0.70710677f}, {-0.72537446f,-0.68835449f}, {-0.74314493f,-0.66913044f},
+{-0.76040596f,-0.64944804f}, {-0.77714604f,-0.62932026f}, {-0.79335332f,-0.60876143f},
+{-0.80901700f,-0.58778518f}, {-0.82412624f,-0.56640613f}, {-0.83867055f,-0.54463899f},
+{-0.85264021f,-0.52249849f}, {-0.86602539f,-0.50000006f}, {-0.87881714f,-0.47715873f},
+{-0.89100659f,-0.45399037f}, {-0.90258527f,-0.43051112f}, {-0.91354549f,-0.40673658f},
+{-0.92387956f,-0.38268328f}, {-0.93358040f,-0.35836792f}, {-0.94264150f,-0.33380675f},
+{-0.95105654f,-0.30901679f}, {-0.95881975f,-0.28401530f}, {-0.96592587f,-0.25881892f},
+{-0.97236991f,-0.23344538f}, {-0.97814763f,-0.20791161f}, {-0.98325491f,-0.18223536f},
+{-0.98768836f,-0.15643445f}, {-0.99144489f,-0.13052608f}, {-0.99452192f,-0.10452849f},
+{-0.99691737f,-0.078459039f}, {-0.99862957f,-0.052335810f}, {-0.99965733f,-0.026176952f},
+{1.0000000f,-0.0000000f}, {0.99922901f,-0.039259817f}, {0.99691731f,-0.078459099f},
+{0.99306846f,-0.11753740f}, {0.98768836f,-0.15643448f}, {0.98078525f,-0.19509032f},
+{0.97236991f,-0.23344538f}, {0.96245521f,-0.27144045f}, {0.95105648f,-0.30901700f},
+{0.93819129f,-0.34611708f}, {0.92387956f,-0.38268346f}, {0.90814316f,-0.41865975f},
+{0.89100653f,-0.45399052f}, {0.87249601f,-0.48862126f}, {0.85264015f,-0.52249855f},
+{0.83146960f,-0.55557024f}, {0.80901700f,-0.58778524f}, {0.78531694f,-0.61909395f},
+{0.76040596f,-0.64944810f}, {0.73432249f,-0.67880076f}, {0.70710677f,-0.70710683f},
+{0.67880070f,-0.73432255f}, {0.64944804f,-0.76040596f}, {0.61909395f,-0.78531694f},
+{0.58778524f,-0.80901700f}, {0.55557019f,-0.83146960f}, {0.52249849f,-0.85264015f},
+{0.48862118f,-0.87249601f}, {0.45399052f,-0.89100653f}, {0.41865975f,-0.90814316f},
+{0.38268343f,-0.92387956f}, {0.34611705f,-0.93819135f}, {0.30901697f,-0.95105654f},
+{0.27144039f,-0.96245527f}, {0.23344530f,-0.97236991f}, {0.19509023f,-0.98078531f},
+{0.15643437f,-0.98768836f}, {0.11753740f,-0.99306846f}, {0.078459084f,-0.99691731f},
+{0.039259788f,-0.99922901f}, {-4.3711388e-08f,-1.0000000f}, {-0.039259877f,-0.99922901f},
+{-0.078459173f,-0.99691731f}, {-0.11753749f,-0.99306846f}, {-0.15643445f,-0.98768836f},
+{-0.19509032f,-0.98078525f}, {-0.23344538f,-0.97236991f}, {-0.27144048f,-0.96245521f},
+{-0.30901703f,-0.95105648f}, {-0.34611711f,-0.93819129f}, {-0.38268352f,-0.92387950f},
+{-0.41865984f,-0.90814310f}, {-0.45399061f,-0.89100647f}, {-0.48862135f,-0.87249595f},
+{-0.52249867f,-0.85264009f}, {-0.55557036f,-0.83146954f}, {-0.58778518f,-0.80901700f},
+{-0.61909389f,-0.78531694f}, {-0.64944804f,-0.76040596f}, {-0.67880076f,-0.73432249f},
+{-0.70710677f,-0.70710677f}, {-0.73432249f,-0.67880070f}, {-0.76040596f,-0.64944804f},
+{-0.78531694f,-0.61909389f}, {-0.80901700f,-0.58778518f}, {-0.83146966f,-0.55557019f},
+{-0.85264021f,-0.52249849f}, {-0.87249607f,-0.48862115f}, {-0.89100659f,-0.45399037f},
+{-0.90814322f,-0.41865960f}, {-0.92387956f,-0.38268328f}, {-0.93819135f,-0.34611690f},
+{-0.95105654f,-0.30901679f}, {-0.96245521f,-0.27144048f}, {-0.97236991f,-0.23344538f},
+{-0.98078531f,-0.19509031f}, {-0.98768836f,-0.15643445f}, {-0.99306846f,-0.11753736f},
+{-0.99691737f,-0.078459039f}, {-0.99922901f,-0.039259743f}, {-1.0000000f,8.7422777e-08f},
+{-0.99922901f,0.039259918f}, {-0.99691731f,0.078459218f}, {-0.99306846f,0.11753753f},
+{-0.98768830f,0.15643461f}, {-0.98078525f,0.19509049f}, {-0.97236985f,0.23344554f},
+{-0.96245515f,0.27144065f}, {-0.95105654f,0.30901697f}, {-0.93819135f,0.34611705f},
+{-0.92387956f,0.38268346f}, {-0.90814316f,0.41865975f}, {-0.89100653f,0.45399055f},
+{-0.87249601f,0.48862129f}, {-0.85264015f,0.52249861f}, {-0.83146960f,0.55557030f},
+{-0.80901694f,0.58778536f}, {-0.78531688f,0.61909401f}, {-0.76040590f,0.64944816f},
+{-0.73432243f,0.67880082f}, {-0.70710665f,0.70710689f}, {-0.67880058f,0.73432261f},
+{-0.64944792f,0.76040608f}, {-0.61909378f,0.78531706f}, {-0.58778507f,0.80901712f},
+{-0.55557001f,0.83146977f}, {-0.52249837f,0.85264033f}, {-0.48862100f,0.87249613f},
+{-0.45399022f,0.89100665f}, {-0.41865945f,0.90814328f}, {-0.38268313f,0.92387968f},
+{-0.34611672f,0.93819147f}, {-0.30901709f,0.95105648f}, {-0.27144054f,0.96245521f},
+{-0.23344545f,0.97236991f}, {-0.19509038f,0.98078525f}, {-0.15643452f,0.98768830f},
+{-0.11753743f,0.99306846f}, {-0.078459114f,0.99691731f}, {-0.039259821f,0.99922901f},
+};
+static const ne10_fft_cpx_float32_t ne10_twiddles_240[240] = {
+{1.0000000f,0.0000000f}, {1.0000000f,-0.0000000f}, {1.0000000f,-0.0000000f},
+{1.0000000f,-0.0000000f}, {0.91354543f,-0.40673664f}, {0.66913056f,-0.74314487f},
+{1.0000000f,-0.0000000f}, {0.66913056f,-0.74314487f}, {-0.10452851f,-0.99452192f},
+{1.0000000f,-0.0000000f}, {0.30901697f,-0.95105654f}, {-0.80901700f,-0.58778518f},
+{1.0000000f,-0.0000000f}, {-0.10452851f,-0.99452192f}, {-0.97814757f,0.20791179f},
+{1.0000000f,-0.0000000f}, {0.99452192f,-0.10452846f}, {0.97814763f,-0.20791170f},
+{0.95105648f,-0.30901700f}, {0.91354543f,-0.40673664f}, {0.86602545f,-0.50000000f},
+{0.80901700f,-0.58778524f}, {0.74314475f,-0.66913062f}, {0.66913056f,-0.74314487f},
+{0.58778524f,-0.80901700f}, {0.49999997f,-0.86602545f}, {0.40673661f,-0.91354549f},
+{0.30901697f,-0.95105654f}, {0.20791166f,-0.97814763f}, {0.10452842f,-0.99452192f},
+{1.0000000f,-0.0000000f}, {0.97814763f,-0.20791170f}, {0.91354543f,-0.40673664f},
+{0.80901700f,-0.58778524f}, {0.66913056f,-0.74314487f}, {0.49999997f,-0.86602545f},
+{0.30901697f,-0.95105654f}, {0.10452842f,-0.99452192f}, {-0.10452851f,-0.99452192f},
+{-0.30901703f,-0.95105648f}, {-0.50000006f,-0.86602533f}, {-0.66913068f,-0.74314475f},
+{-0.80901700f,-0.58778518f}, {-0.91354549f,-0.40673658f}, {-0.97814763f,-0.20791161f},
+{1.0000000f,-0.0000000f}, {0.95105648f,-0.30901700f}, {0.80901700f,-0.58778524f},
+{0.58778524f,-0.80901700f}, {0.30901697f,-0.95105654f}, {-4.3711388e-08f,-1.0000000f},
+{-0.30901703f,-0.95105648f}, {-0.58778518f,-0.80901700f}, {-0.80901700f,-0.58778518f},
+{-0.95105654f,-0.30901679f}, {-1.0000000f,8.7422777e-08f}, {-0.95105654f,0.30901697f},
+{-0.80901694f,0.58778536f}, {-0.58778507f,0.80901712f}, {-0.30901709f,0.95105648f},
+{1.0000000f,-0.0000000f}, {0.99965733f,-0.026176950f}, {0.99862951f,-0.052335959f},
+{0.99691731f,-0.078459099f}, {0.99452192f,-0.10452846f}, {0.99144489f,-0.13052620f},
+{0.98768836f,-0.15643448f}, {0.98325491f,-0.18223552f}, {0.97814763f,-0.20791170f},
+{0.97236991f,-0.23344538f}, {0.96592581f,-0.25881904f}, {0.95881975f,-0.28401536f},
+{0.95105648f,-0.30901700f}, {0.94264150f,-0.33380687f}, {0.93358040f,-0.35836795f},
+{0.92387956f,-0.38268346f}, {0.91354543f,-0.40673664f}, {0.90258527f,-0.43051112f},
+{0.89100653f,-0.45399052f}, {0.87881708f,-0.47715878f}, {0.86602545f,-0.50000000f},
+{0.85264015f,-0.52249855f}, {0.83867055f,-0.54463905f}, {0.82412618f,-0.56640625f},
+{0.80901700f,-0.58778524f}, {0.79335332f,-0.60876143f}, {0.77714598f,-0.62932038f},
+{0.76040596f,-0.64944810f}, {0.74314475f,-0.66913062f}, {0.72537434f,-0.68835455f},
+{0.70710677f,-0.70710683f}, {0.68835455f,-0.72537440f}, {0.66913056f,-0.74314487f},
+{0.64944804f,-0.76040596f}, {0.62932038f,-0.77714598f}, {0.60876137f,-0.79335338f},
+{0.58778524f,-0.80901700f}, {0.56640625f,-0.82412618f}, {0.54463899f,-0.83867055f},
+{0.52249849f,-0.85264015f}, {0.49999997f,-0.86602545f}, {0.47715876f,-0.87881708f},
+{0.45399052f,-0.89100653f}, {0.43051103f,-0.90258533f}, {0.40673661f,-0.91354549f},
+{0.38268343f,-0.92387956f}, {0.35836786f,-0.93358046f}, {0.33380681f,-0.94264150f},
+{0.30901697f,-0.95105654f}, {0.28401533f,-0.95881975f}, {0.25881907f,-0.96592581f},
+{0.23344530f,-0.97236991f}, {0.20791166f,-0.97814763f}, {0.18223552f,-0.98325491f},
+{0.15643437f,-0.98768836f}, {0.13052613f,-0.99144489f}, {0.10452842f,-0.99452192f},
+{0.078459084f,-0.99691731f}, {0.052335974f,-0.99862951f}, {0.026176875f,-0.99965733f},
+{1.0000000f,-0.0000000f}, {0.99862951f,-0.052335959f}, {0.99452192f,-0.10452846f},
+{0.98768836f,-0.15643448f}, {0.97814763f,-0.20791170f}, {0.96592581f,-0.25881904f},
+{0.95105648f,-0.30901700f}, {0.93358040f,-0.35836795f}, {0.91354543f,-0.40673664f},
+{0.89100653f,-0.45399052f}, {0.86602545f,-0.50000000f}, {0.83867055f,-0.54463905f},
+{0.80901700f,-0.58778524f}, {0.77714598f,-0.62932038f}, {0.74314475f,-0.66913062f},
+{0.70710677f,-0.70710683f}, {0.66913056f,-0.74314487f}, {0.62932038f,-0.77714598f},
+{0.58778524f,-0.80901700f}, {0.54463899f,-0.83867055f}, {0.49999997f,-0.86602545f},
+{0.45399052f,-0.89100653f}, {0.40673661f,-0.91354549f}, {0.35836786f,-0.93358046f},
+{0.30901697f,-0.95105654f}, {0.25881907f,-0.96592581f}, {0.20791166f,-0.97814763f},
+{0.15643437f,-0.98768836f}, {0.10452842f,-0.99452192f}, {0.052335974f,-0.99862951f},
+{-4.3711388e-08f,-1.0000000f}, {-0.052336060f,-0.99862951f}, {-0.10452851f,-0.99452192f},
+{-0.15643445f,-0.98768836f}, {-0.20791174f,-0.97814757f}, {-0.25881916f,-0.96592581f},
+{-0.30901703f,-0.95105648f}, {-0.35836795f,-0.93358040f}, {-0.40673670f,-0.91354543f},
+{-0.45399061f,-0.89100647f}, {-0.50000006f,-0.86602533f}, {-0.54463905f,-0.83867055f},
+{-0.58778518f,-0.80901700f}, {-0.62932050f,-0.77714586f}, {-0.66913068f,-0.74314475f},
+{-0.70710677f,-0.70710677f}, {-0.74314493f,-0.66913044f}, {-0.77714604f,-0.62932026f},
+{-0.80901700f,-0.58778518f}, {-0.83867055f,-0.54463899f}, {-0.86602539f,-0.50000006f},
+{-0.89100659f,-0.45399037f}, {-0.91354549f,-0.40673658f}, {-0.93358040f,-0.35836792f},
+{-0.95105654f,-0.30901679f}, {-0.96592587f,-0.25881892f}, {-0.97814763f,-0.20791161f},
+{-0.98768836f,-0.15643445f}, {-0.99452192f,-0.10452849f}, {-0.99862957f,-0.052335810f},
+{1.0000000f,-0.0000000f}, {0.99691731f,-0.078459099f}, {0.98768836f,-0.15643448f},
+{0.97236991f,-0.23344538f}, {0.95105648f,-0.30901700f}, {0.92387956f,-0.38268346f},
+{0.89100653f,-0.45399052f}, {0.85264015f,-0.52249855f}, {0.80901700f,-0.58778524f},
+{0.76040596f,-0.64944810f}, {0.70710677f,-0.70710683f}, {0.64944804f,-0.76040596f},
+{0.58778524f,-0.80901700f}, {0.52249849f,-0.85264015f}, {0.45399052f,-0.89100653f},
+{0.38268343f,-0.92387956f}, {0.30901697f,-0.95105654f}, {0.23344530f,-0.97236991f},
+{0.15643437f,-0.98768836f}, {0.078459084f,-0.99691731f}, {-4.3711388e-08f,-1.0000000f},
+{-0.078459173f,-0.99691731f}, {-0.15643445f,-0.98768836f}, {-0.23344538f,-0.97236991f},
+{-0.30901703f,-0.95105648f}, {-0.38268352f,-0.92387950f}, {-0.45399061f,-0.89100647f},
+{-0.52249867f,-0.85264009f}, {-0.58778518f,-0.80901700f}, {-0.64944804f,-0.76040596f},
+{-0.70710677f,-0.70710677f}, {-0.76040596f,-0.64944804f}, {-0.80901700f,-0.58778518f},
+{-0.85264021f,-0.52249849f}, {-0.89100659f,-0.45399037f}, {-0.92387956f,-0.38268328f},
+{-0.95105654f,-0.30901679f}, {-0.97236991f,-0.23344538f}, {-0.98768836f,-0.15643445f},
+{-0.99691737f,-0.078459039f}, {-1.0000000f,8.7422777e-08f}, {-0.99691731f,0.078459218f},
+{-0.98768830f,0.15643461f}, {-0.97236985f,0.23344554f}, {-0.95105654f,0.30901697f},
+{-0.92387956f,0.38268346f}, {-0.89100653f,0.45399055f}, {-0.85264015f,0.52249861f},
+{-0.80901694f,0.58778536f}, {-0.76040590f,0.64944816f}, {-0.70710665f,0.70710689f},
+{-0.64944792f,0.76040608f}, {-0.58778507f,0.80901712f}, {-0.52249837f,0.85264033f},
+{-0.45399022f,0.89100665f}, {-0.38268313f,0.92387968f}, {-0.30901709f,0.95105648f},
+{-0.23344545f,0.97236991f}, {-0.15643452f,0.98768830f}, {-0.078459114f,0.99691731f},
+};
+static const ne10_fft_cpx_float32_t ne10_twiddles_120[120] = {
+{1.0000000f,0.0000000f}, {1.0000000f,-0.0000000f}, {1.0000000f,-0.0000000f},
+{1.0000000f,-0.0000000f}, {0.91354543f,-0.40673664f}, {0.66913056f,-0.74314487f},
+{1.0000000f,-0.0000000f}, {0.66913056f,-0.74314487f}, {-0.10452851f,-0.99452192f},
+{1.0000000f,-0.0000000f}, {0.30901697f,-0.95105654f}, {-0.80901700f,-0.58778518f},
+{1.0000000f,-0.0000000f}, {-0.10452851f,-0.99452192f}, {-0.97814757f,0.20791179f},
+{1.0000000f,-0.0000000f}, {0.97814763f,-0.20791170f}, {0.91354543f,-0.40673664f},
+{0.80901700f,-0.58778524f}, {0.66913056f,-0.74314487f}, {0.49999997f,-0.86602545f},
+{0.30901697f,-0.95105654f}, {0.10452842f,-0.99452192f}, {-0.10452851f,-0.99452192f},
+{-0.30901703f,-0.95105648f}, {-0.50000006f,-0.86602533f}, {-0.66913068f,-0.74314475f},
+{-0.80901700f,-0.58778518f}, {-0.91354549f,-0.40673658f}, {-0.97814763f,-0.20791161f},
+{1.0000000f,-0.0000000f}, {0.99862951f,-0.052335959f}, {0.99452192f,-0.10452846f},
+{0.98768836f,-0.15643448f}, {0.97814763f,-0.20791170f}, {0.96592581f,-0.25881904f},
+{0.95105648f,-0.30901700f}, {0.93358040f,-0.35836795f}, {0.91354543f,-0.40673664f},
+{0.89100653f,-0.45399052f}, {0.86602545f,-0.50000000f}, {0.83867055f,-0.54463905f},
+{0.80901700f,-0.58778524f}, {0.77714598f,-0.62932038f}, {0.74314475f,-0.66913062f},
+{0.70710677f,-0.70710683f}, {0.66913056f,-0.74314487f}, {0.62932038f,-0.77714598f},
+{0.58778524f,-0.80901700f}, {0.54463899f,-0.83867055f}, {0.49999997f,-0.86602545f},
+{0.45399052f,-0.89100653f}, {0.40673661f,-0.91354549f}, {0.35836786f,-0.93358046f},
+{0.30901697f,-0.95105654f}, {0.25881907f,-0.96592581f}, {0.20791166f,-0.97814763f},
+{0.15643437f,-0.98768836f}, {0.10452842f,-0.99452192f}, {0.052335974f,-0.99862951f},
+{1.0000000f,-0.0000000f}, {0.99452192f,-0.10452846f}, {0.97814763f,-0.20791170f},
+{0.95105648f,-0.30901700f}, {0.91354543f,-0.40673664f}, {0.86602545f,-0.50000000f},
+{0.80901700f,-0.58778524f}, {0.74314475f,-0.66913062f}, {0.66913056f,-0.74314487f},
+{0.58778524f,-0.80901700f}, {0.49999997f,-0.86602545f}, {0.40673661f,-0.91354549f},
+{0.30901697f,-0.95105654f}, {0.20791166f,-0.97814763f}, {0.10452842f,-0.99452192f},
+{-4.3711388e-08f,-1.0000000f}, {-0.10452851f,-0.99452192f}, {-0.20791174f,-0.97814757f},
+{-0.30901703f,-0.95105648f}, {-0.40673670f,-0.91354543f}, {-0.50000006f,-0.86602533f},
+{-0.58778518f,-0.80901700f}, {-0.66913068f,-0.74314475f}, {-0.74314493f,-0.66913044f},
+{-0.80901700f,-0.58778518f}, {-0.86602539f,-0.50000006f}, {-0.91354549f,-0.40673658f},
+{-0.95105654f,-0.30901679f}, {-0.97814763f,-0.20791161f}, {-0.99452192f,-0.10452849f},
+{1.0000000f,-0.0000000f}, {0.98768836f,-0.15643448f}, {0.95105648f,-0.30901700f},
+{0.89100653f,-0.45399052f}, {0.80901700f,-0.58778524f}, {0.70710677f,-0.70710683f},
+{0.58778524f,-0.80901700f}, {0.45399052f,-0.89100653f}, {0.30901697f,-0.95105654f},
+{0.15643437f,-0.98768836f}, {-4.3711388e-08f,-1.0000000f}, {-0.15643445f,-0.98768836f},
+{-0.30901703f,-0.95105648f}, {-0.45399061f,-0.89100647f}, {-0.58778518f,-0.80901700f},
+{-0.70710677f,-0.70710677f}, {-0.80901700f,-0.58778518f}, {-0.89100659f,-0.45399037f},
+{-0.95105654f,-0.30901679f}, {-0.98768836f,-0.15643445f}, {-1.0000000f,8.7422777e-08f},
+{-0.98768830f,0.15643461f}, {-0.95105654f,0.30901697f}, {-0.89100653f,0.45399055f},
+{-0.80901694f,0.58778536f}, {-0.70710665f,0.70710689f}, {-0.58778507f,0.80901712f},
+{-0.45399022f,0.89100665f}, {-0.30901709f,0.95105648f}, {-0.15643452f,0.98768830f},
+};
+static const ne10_fft_cpx_float32_t ne10_twiddles_60[60] = {
+{1.0000000f,0.0000000f}, {1.0000000f,-0.0000000f}, {1.0000000f,-0.0000000f},
+{1.0000000f,-0.0000000f}, {0.91354543f,-0.40673664f}, {0.66913056f,-0.74314487f},
+{1.0000000f,-0.0000000f}, {0.66913056f,-0.74314487f}, {-0.10452851f,-0.99452192f},
+{1.0000000f,-0.0000000f}, {0.30901697f,-0.95105654f}, {-0.80901700f,-0.58778518f},
+{1.0000000f,-0.0000000f}, {-0.10452851f,-0.99452192f}, {-0.97814757f,0.20791179f},
+{1.0000000f,-0.0000000f}, {0.99452192f,-0.10452846f}, {0.97814763f,-0.20791170f},
+{0.95105648f,-0.30901700f}, {0.91354543f,-0.40673664f}, {0.86602545f,-0.50000000f},
+{0.80901700f,-0.58778524f}, {0.74314475f,-0.66913062f}, {0.66913056f,-0.74314487f},
+{0.58778524f,-0.80901700f}, {0.49999997f,-0.86602545f}, {0.40673661f,-0.91354549f},
+{0.30901697f,-0.95105654f}, {0.20791166f,-0.97814763f}, {0.10452842f,-0.99452192f},
+{1.0000000f,-0.0000000f}, {0.97814763f,-0.20791170f}, {0.91354543f,-0.40673664f},
+{0.80901700f,-0.58778524f}, {0.66913056f,-0.74314487f}, {0.49999997f,-0.86602545f},
+{0.30901697f,-0.95105654f}, {0.10452842f,-0.99452192f}, {-0.10452851f,-0.99452192f},
+{-0.30901703f,-0.95105648f}, {-0.50000006f,-0.86602533f}, {-0.66913068f,-0.74314475f},
+{-0.80901700f,-0.58778518f}, {-0.91354549f,-0.40673658f}, {-0.97814763f,-0.20791161f},
+{1.0000000f,-0.0000000f}, {0.95105648f,-0.30901700f}, {0.80901700f,-0.58778524f},
+{0.58778524f,-0.80901700f}, {0.30901697f,-0.95105654f}, {-4.3711388e-08f,-1.0000000f},
+{-0.30901703f,-0.95105648f}, {-0.58778518f,-0.80901700f}, {-0.80901700f,-0.58778518f},
+{-0.95105654f,-0.30901679f}, {-1.0000000f,8.7422777e-08f}, {-0.95105654f,0.30901697f},
+{-0.80901694f,0.58778536f}, {-0.58778507f,0.80901712f}, {-0.30901709f,0.95105648f},
+};
+static const ne10_fft_state_float32_t ne10_fft_state_float32_t_480 = {
+120,
+(ne10_int32_t *)ne10_factors_480,
+(ne10_fft_cpx_float32_t *)ne10_twiddles_480,
+NULL,
+(ne10_fft_cpx_float32_t *)&ne10_twiddles_480[120],
+/* is_forward_scaled = true */
+(ne10_int32_t) 1,
+/* is_backward_scaled = false */
+(ne10_int32_t) 0,
+};
+static const arch_fft_state cfg_arch_480 = {
+1,
+(void *)&ne10_fft_state_float32_t_480,
+};
+
+static const ne10_fft_state_float32_t ne10_fft_state_float32_t_240 = {
+60,
+(ne10_int32_t *)ne10_factors_240,
+(ne10_fft_cpx_float32_t *)ne10_twiddles_240,
+NULL,
+(ne10_fft_cpx_float32_t *)&ne10_twiddles_240[60],
+/* is_forward_scaled = true */
+(ne10_int32_t) 1,
+/* is_backward_scaled = false */
+(ne10_int32_t) 0,
+};
+static const arch_fft_state cfg_arch_240 = {
+1,
+(void *)&ne10_fft_state_float32_t_240,
+};
+
+static const ne10_fft_state_float32_t ne10_fft_state_float32_t_120 = {
+30,
+(ne10_int32_t *)ne10_factors_120,
+(ne10_fft_cpx_float32_t *)ne10_twiddles_120,
+NULL,
+(ne10_fft_cpx_float32_t *)&ne10_twiddles_120[30],
+/* is_forward_scaled = true */
+(ne10_int32_t) 1,
+/* is_backward_scaled = false */
+(ne10_int32_t) 0,
+};
+static const arch_fft_state cfg_arch_120 = {
+1,
+(void *)&ne10_fft_state_float32_t_120,
+};
+
+static const ne10_fft_state_float32_t ne10_fft_state_float32_t_60 = {
+15,
+(ne10_int32_t *)ne10_factors_60,
+(ne10_fft_cpx_float32_t *)ne10_twiddles_60,
+NULL,
+(ne10_fft_cpx_float32_t *)&ne10_twiddles_60[15],
+/* is_forward_scaled = true */
+(ne10_int32_t) 1,
+/* is_backward_scaled = false */
+(ne10_int32_t) 0,
+};
+static const arch_fft_state cfg_arch_60 = {
+1,
+(void *)&ne10_fft_state_float32_t_60,
+};
+
+#endif  /* end NE10_FFT_PARAMS48000_960 */
diff --git a/drivers/opus/celt/tests/test_unit_cwrs32.c b/drivers/opus/celt/tests/test_unit_cwrs32.c
deleted file mode 100644
index db43e3392b..0000000000
--- a/drivers/opus/celt/tests/test_unit_cwrs32.c
+++ /dev/null
@@ -1,161 +0,0 @@
-/* Copyright (c) 2008-2011 Xiph.Org Foundation, Mozilla Corporation,
-                           Gregory Maxwell
-   Written by Jean-Marc Valin, Gregory Maxwell, and Timothy B. Terriberry */
-/*
-   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.
-
-   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
-   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
-   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
-   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
-   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
-   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
-   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
-   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-*/
-
-#ifdef OPUS_ENABLED
-#include "opus/opus_config.h"
-#endif
-
-#include <stdio.h>
-#include <string.h>
-
-#ifndef CUSTOM_MODES
-#define CUSTOM_MODES
-#else
-#define TEST_CUSTOM_MODES
-#endif
-
-#define CELT_C
-#include "opus/celt/stack_alloc.h"
-#include "opus/celt/entenc.c"
-#include "opus/celt/entdec.c"
-#include "opus/celt/entcode.c"
-#include "opus/celt/cwrs.c"
-#include "opus/celt/mathops.c"
-#include "opus/celt/rate.h"
-
-#define NMAX (240)
-#define KMAX (128)
-
-#ifdef TEST_CUSTOM_MODES
-
-#define NDIMS (44)
-static const int pn[NDIMS]={
-   2,   3,   4,   5,   6,   7,   8,   9,  10,
-  11,  12,  13,  14,  15,  16,  18,  20,  22,
-  24,  26,  28,  30,  32,  36,  40,  44,  48,
-  52,  56,  60,  64,  72,  80,  88,  96, 104,
- 112, 120, 128, 144, 160, 176, 192, 208
-};
-static const int pkmax[NDIMS]={
- 128, 128, 128, 128,  88,  52,  36,  26,  22,
-  18,  16,  15,  13,  12,  12,  11,  10,   9,
-   9,   8,   8,   7,   7,   7,   7,   6,   6,
-   6,   6,   6,   5,   5,   5,   5,   5,   5,
-   4,   4,   4,   4,   4,   4,   4,   4
-};
-
-#else /* TEST_CUSTOM_MODES */
-
-#define NDIMS (22)
-static const int pn[NDIMS]={
-   2,   3,   4,   6,   8,   9,  11,  12,  16,
-  18,  22,  24,  32,  36,  44,  48,  64,  72,
-  88,  96, 144, 176
-};
-static const int pkmax[NDIMS]={
- 128, 128, 128,  88,  36,  26,  18,  16,  12,
-  11,   9,   9,   7,   7,   6,   6,   5,   5,
-   5,   5,   4,   4
-};
-
-#endif
-
-int main(void){
-  int t;
-  int n;
-  ALLOC_STACK;
-  for(t=0;t<NDIMS;t++){
-    int pseudo;
-    n=pn[t];
-    for(pseudo=1;pseudo<41;pseudo++)
-    {
-      int k;
-#if defined(SMALL_FOOTPRINT)
-      opus_uint32 uu[KMAX+2U];
-#endif
-      opus_uint32 inc;
-      opus_uint32 nc;
-      opus_uint32 i;
-      k=get_pulses(pseudo);
-      if (k>pkmax[t])break;
-      printf("Testing CWRS with N=%i, K=%i...\n",n,k);
-#if defined(SMALL_FOOTPRINT)
-      nc=ncwrs_urow(n,k,uu);
-#else
-      nc=CELT_PVQ_V(n,k);
-#endif
-      inc=nc/20000;
-      if(inc<1)inc=1;
-      for(i=0;i<nc;i+=inc){
-#if defined(SMALL_FOOTPRINT)
-        opus_uint32 u[KMAX+2U];
-#endif
-        int         y[NMAX];
-        int         sy;
-        opus_uint32 v;
-        opus_uint32 ii;
-        int         j;
-#if defined(SMALL_FOOTPRINT)
-        memcpy(u,uu,(k+2U)*sizeof(*u));
-        cwrsi(n,k,i,y,u);
-#else
-        cwrsi(n,k,i,y);
-#endif
-        sy=0;
-        for(j=0;j<n;j++)sy+=ABS(y[j]);
-        if(sy!=k){
-          fprintf(stderr,"N=%d Pulse count mismatch in cwrsi (%d!=%d).\n",
-           n,sy,k);
-          return 99;
-        }
-        /*printf("%6u of %u:",i,nc);
-        for(j=0;j<n;j++)printf(" %+3i",y[j]);
-        printf(" ->");*/
-#if defined(SMALL_FOOTPRINT)
-        ii=icwrs(n,k,&v,y,u);
-#else
-        ii=icwrs(n,y);
-        v=CELT_PVQ_V(n,k);
-#endif
-        if(ii!=i){
-          fprintf(stderr,"Combination-index mismatch (%lu!=%lu).\n",
-           (long)ii,(long)i);
-          return 1;
-        }
-        if(v!=nc){
-          fprintf(stderr,"Combination count mismatch (%lu!=%lu).\n",
-           (long)v,(long)nc);
-          return 2;
-        }
-        /*printf(" %6u\n",i);*/
-      }
-      /*printf("\n");*/
-    }
-  }
-  return 0;
-}
diff --git a/drivers/opus/celt/tests/test_unit_dft.c b/drivers/opus/celt/tests/test_unit_dft.c
deleted file mode 100644
index 9c0db3e9ac..0000000000
--- a/drivers/opus/celt/tests/test_unit_dft.c
+++ /dev/null
@@ -1,164 +0,0 @@
-/* Copyright (c) 2008 Xiph.Org Foundation
-   Written by Jean-Marc Valin */
-/*
-   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.
-
-   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
-   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
-   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
-   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
-   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
-   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
-   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
-   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-*/
-
-#ifdef OPUS_ENABLED
-#include "opus/opus_config.h"
-#endif
-
-#define SKIP_CONFIG_H
-
-#ifndef CUSTOM_MODES
-#define CUSTOM_MODES
-#endif
-
-#include <stdio.h>
-
-#define CELT_C
-#include "opus/celt/stack_alloc.h"
-#include "opus/celt/kiss_fft.h"
-#include "opus/celt/kiss_fft.c"
-#include "opus/celt/mathops.c"
-#include "opus/celt/entcode.c"
-
-
-#ifndef M_PI
-#define M_PI 3.141592653
-#endif
-
-int ret = 0;
-
-void check(kiss_fft_cpx  * in,kiss_fft_cpx  * out,int nfft,int isinverse)
-{
-    int bin,k;
-    double errpow=0,sigpow=0, snr;
-
-    for (bin=0;bin<nfft;++bin) {
-        double ansr = 0;
-        double ansi = 0;
-        double difr;
-        double difi;
-
-        for (k=0;k<nfft;++k) {
-            double phase = -2*M_PI*bin*k/nfft;
-            double re = cos(phase);
-            double im = sin(phase);
-            if (isinverse)
-                im = -im;
-
-            if (!isinverse)
-            {
-               re /= nfft;
-               im /= nfft;
-            }
-
-            ansr += in[k].r * re - in[k].i * im;
-            ansi += in[k].r * im + in[k].i * re;
-        }
-        /*printf ("%d %d ", (int)ansr, (int)ansi);*/
-        difr = ansr - out[bin].r;
-        difi = ansi - out[bin].i;
-        errpow += difr*difr + difi*difi;
-        sigpow += ansr*ansr+ansi*ansi;
-    }
-    snr = 10*log10(sigpow/errpow);
-    printf("nfft=%d inverse=%d,snr = %f\n",nfft,isinverse,snr );
-    if (snr<60) {
-       printf( "** poor snr: %f ** \n", snr);
-       ret = 1;
-    }
-}
-
-void test1d(int nfft,int isinverse)
-{
-    size_t buflen = sizeof(kiss_fft_cpx)*nfft;
-
-    kiss_fft_cpx  * in = (kiss_fft_cpx*)malloc(buflen);
-    kiss_fft_cpx  * out= (kiss_fft_cpx*)malloc(buflen);
-    kiss_fft_state *cfg = opus_fft_alloc(nfft,0,0);
-    int k;
-
-    for (k=0;k<nfft;++k) {
-        in[k].r = (rand() % 32767) - 16384;
-        in[k].i = (rand() % 32767) - 16384;
-    }
-
-    for (k=0;k<nfft;++k) {
-       in[k].r *= 32768;
-       in[k].i *= 32768;
-    }
-
-    if (isinverse)
-    {
-       for (k=0;k<nfft;++k) {
-          in[k].r /= nfft;
-          in[k].i /= nfft;
-       }
-    }
-
-    /*for (k=0;k<nfft;++k) printf("%d %d ", in[k].r, in[k].i);printf("\n");*/
-
-    if (isinverse)
-       opus_ifft(cfg,in,out);
-    else
-       opus_fft(cfg,in,out);
-
-    /*for (k=0;k<nfft;++k) printf("%d %d ", out[k].r, out[k].i);printf("\n");*/
-
-    check(in,out,nfft,isinverse);
-
-    free(in);
-    free(out);
-    free(cfg);
-}
-
-int main(int argc,char ** argv)
-{
-    ALLOC_STACK;
-    if (argc>1) {
-        int k;
-        for (k=1;k<argc;++k) {
-            test1d(atoi(argv[k]),0);
-            test1d(atoi(argv[k]),1);
-        }
-    }else{
-        test1d(32,0);
-        test1d(32,1);
-        test1d(128,0);
-        test1d(128,1);
-        test1d(256,0);
-        test1d(256,1);
-#ifndef RADIX_TWO_ONLY
-        test1d(36,0);
-        test1d(36,1);
-        test1d(50,0);
-        test1d(50,1);
-        test1d(120,0);
-        test1d(120,1);
-#endif
-    }
-    return ret;
-}
diff --git a/drivers/opus/celt/tests/test_unit_entropy.c b/drivers/opus/celt/tests/test_unit_entropy.c
deleted file mode 100644
index 8c2defa8b3..0000000000
--- a/drivers/opus/celt/tests/test_unit_entropy.c
+++ /dev/null
@@ -1,382 +0,0 @@
-/* Copyright (c) 2007-2011 Xiph.Org Foundation, Mozilla Corporation,
-                           Gregory Maxwell
-   Written by Jean-Marc Valin, Gregory Maxwell, and Timothy B. Terriberry */
-/*
-   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.
-
-   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
-   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
-   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
-   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
-   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
-   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
-   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
-   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-*/
-
-#ifdef OPUS_ENABLED
-#include "opus/opus_config.h"
-#endif
-
-#include <stdlib.h>
-#include <stdio.h>
-#include <math.h>
-#include <time.h>
-#include "opus/celt/entcode.h"
-#include "opus/celt/entenc.h"
-#include "opus/celt/entdec.h"
-#include <string.h>
-
-#include "opus/celt/entenc.c"
-#include "opus/celt/entdec.c"
-#include "opus/celt/entcode.c"
-
-#ifndef M_LOG2E
-# define M_LOG2E    1.4426950408889634074
-#endif
-#define DATA_SIZE 10000000
-#define DATA_SIZE2 10000
-
-int main(int _argc,char **_argv){
-  ec_enc         enc;
-  ec_dec         dec;
-  long           nbits;
-  long           nbits2;
-  double         entropy;
-  int            ft;
-  int            ftb;
-  int            sz;
-  int            i;
-  int            ret;
-  unsigned int   sym;
-  unsigned int   seed;
-  unsigned char *ptr;
-  const char    *env_seed;
-  ret=0;
-  entropy=0;
-    if (_argc > 2) {
-	fprintf(stderr, "Usage: %s [<seed>]\n", _argv[0]);
-	return 1;
-    }
-  env_seed = getenv("SEED");
-  if (_argc > 1)
-    seed = atoi(_argv[1]);
-  else if (env_seed)
-    seed = atoi(env_seed);
-  else
-    seed = time(NULL);
-  /*Testing encoding of raw bit values.*/
-  ptr = (unsigned char *)malloc(DATA_SIZE);
-  ec_enc_init(&enc,ptr, DATA_SIZE);
-  for(ft=2;ft<1024;ft++){
-    for(i=0;i<ft;i++){
-      entropy+=log(ft)*M_LOG2E;
-      ec_enc_uint(&enc,i,ft);
-    }
-  }
-  /*Testing encoding of raw bit values.*/
-  for(ftb=1;ftb<16;ftb++){
-    for(i=0;i<(1<<ftb);i++){
-      entropy+=ftb;
-      nbits=ec_tell(&enc);
-      ec_enc_bits(&enc,i,ftb);
-      nbits2=ec_tell(&enc);
-      if(nbits2-nbits!=ftb){
-        fprintf(stderr,"Used %li bits to encode %i bits directly.\n",
-         nbits2-nbits,ftb);
-        ret=-1;
-      }
-    }
-  }
-  nbits=ec_tell_frac(&enc);
-  ec_enc_done(&enc);
-  fprintf(stderr,
-   "Encoded %0.2lf bits of entropy to %0.2lf bits (%0.3lf%% wasted).\n",
-   entropy,ldexp(nbits,-3),100*(nbits-ldexp(entropy,3))/nbits);
-  fprintf(stderr,"Packed to %li bytes.\n",(long)ec_range_bytes(&enc));
-  ec_dec_init(&dec,ptr,DATA_SIZE);
-  for(ft=2;ft<1024;ft++){
-    for(i=0;i<ft;i++){
-      sym=ec_dec_uint(&dec,ft);
-      if(sym!=(unsigned)i){
-        fprintf(stderr,"Decoded %i instead of %i with ft of %i.\n",sym,i,ft);
-        ret=-1;
-      }
-    }
-  }
-  for(ftb=1;ftb<16;ftb++){
-    for(i=0;i<(1<<ftb);i++){
-      sym=ec_dec_bits(&dec,ftb);
-      if(sym!=(unsigned)i){
-        fprintf(stderr,"Decoded %i instead of %i with ftb of %i.\n",sym,i,ftb);
-        ret=-1;
-      }
-    }
-  }
-  nbits2=ec_tell_frac(&dec);
-  if(nbits!=nbits2){
-    fprintf(stderr,
-     "Reported number of bits used was %0.2lf, should be %0.2lf.\n",
-     ldexp(nbits2,-3),ldexp(nbits,-3));
-    ret=-1;
-  }
-  /*Testing an encoder bust prefers range coder data over raw bits.
-    This isn't a general guarantee, will only work for data that is buffered in
-     the encoder state and not yet stored in the user buffer, and should never
-     get used in practice.
-    It's mostly here for code coverage completeness.*/
-  /*Start with a 16-bit buffer.*/
-  ec_enc_init(&enc,ptr,2);
-  /*Write 7 raw bits.*/
-  ec_enc_bits(&enc,0x55,7);
-  /*Write 12.3 bits of range coder data.*/
-  ec_enc_uint(&enc,1,2);
-  ec_enc_uint(&enc,1,3);
-  ec_enc_uint(&enc,1,4);
-  ec_enc_uint(&enc,1,5);
-  ec_enc_uint(&enc,2,6);
-  ec_enc_uint(&enc,6,7);
-  ec_enc_done(&enc);
-  ec_dec_init(&dec,ptr,2);
-  if(!enc.error
-   /*The raw bits should have been overwritten by the range coder data.*/
-   ||ec_dec_bits(&dec,7)!=0x05
-   /*And all the range coder data should have been encoded correctly.*/
-   ||ec_dec_uint(&dec,2)!=1
-   ||ec_dec_uint(&dec,3)!=1
-   ||ec_dec_uint(&dec,4)!=1
-   ||ec_dec_uint(&dec,5)!=1
-   ||ec_dec_uint(&dec,6)!=2
-   ||ec_dec_uint(&dec,7)!=6){
-    fprintf(stderr,"Encoder bust overwrote range coder data with raw bits.\n");
-    ret=-1;
-  }
-  srand(seed);
-  fprintf(stderr,"Testing random streams... Random seed: %u (%.4X)\n", seed, rand() % 65536);
-  for(i=0;i<409600;i++){
-    unsigned *data;
-    unsigned *tell;
-    unsigned tell_bits;
-    int       j;
-    int zeros;
-    ft=rand()/((RAND_MAX>>(rand()%11U))+1U)+10;
-    sz=rand()/((RAND_MAX>>(rand()%9U))+1U);
-    data=(unsigned *)malloc(sz*sizeof(*data));
-    tell=(unsigned *)malloc((sz+1)*sizeof(*tell));
-    ec_enc_init(&enc,ptr,DATA_SIZE2);
-    zeros = rand()%13==0;
-    tell[0]=ec_tell_frac(&enc);
-    for(j=0;j<sz;j++){
-      if (zeros)
-        data[j]=0;
-      else
-        data[j]=rand()%ft;
-      ec_enc_uint(&enc,data[j],ft);
-      tell[j+1]=ec_tell_frac(&enc);
-    }
-    if (rand()%2==0)
-      while(ec_tell(&enc)%8 != 0)
-        ec_enc_uint(&enc, rand()%2, 2);
-    tell_bits = ec_tell(&enc);
-    ec_enc_done(&enc);
-    if(tell_bits!=(unsigned)ec_tell(&enc)){
-      fprintf(stderr,"ec_tell() changed after ec_enc_done(): %i instead of %i (Random seed: %u)\n",
-       ec_tell(&enc),tell_bits,seed);
-      ret=-1;
-    }
-    if ((tell_bits+7)/8 < ec_range_bytes(&enc))
-    {
-      fprintf (stderr, "ec_tell() lied, there's %i bytes instead of %d (Random seed: %u)\n",
-               ec_range_bytes(&enc), (tell_bits+7)/8,seed);
-      ret=-1;
-    }
-    ec_dec_init(&dec,ptr,DATA_SIZE2);
-    if(ec_tell_frac(&dec)!=tell[0]){
-      fprintf(stderr,
-       "Tell mismatch between encoder and decoder at symbol %i: %i instead of %i (Random seed: %u).\n",
-       0,ec_tell_frac(&dec),tell[0],seed);
-    }
-    for(j=0;j<sz;j++){
-      sym=ec_dec_uint(&dec,ft);
-      if(sym!=data[j]){
-        fprintf(stderr,
-         "Decoded %i instead of %i with ft of %i at position %i of %i (Random seed: %u).\n",
-         sym,data[j],ft,j,sz,seed);
-        ret=-1;
-      }
-      if(ec_tell_frac(&dec)!=tell[j+1]){
-        fprintf(stderr,
-         "Tell mismatch between encoder and decoder at symbol %i: %i instead of %i (Random seed: %u).\n",
-         j+1,ec_tell_frac(&dec),tell[j+1],seed);
-      }
-    }
-    free(tell);
-    free(data);
-  }
-  /*Test compatibility between multiple different encode/decode routines.*/
-  for(i=0;i<409600;i++){
-    unsigned *logp1;
-    unsigned *data;
-    unsigned *tell;
-    unsigned *enc_method;
-    int       j;
-    sz=rand()/((RAND_MAX>>(rand()%9U))+1U);
-    logp1=(unsigned *)malloc(sz*sizeof(*logp1));
-    data=(unsigned *)malloc(sz*sizeof(*data));
-    tell=(unsigned *)malloc((sz+1)*sizeof(*tell));
-    enc_method=(unsigned *)malloc(sz*sizeof(*enc_method));
-    ec_enc_init(&enc,ptr,DATA_SIZE2);
-    tell[0]=ec_tell_frac(&enc);
-    for(j=0;j<sz;j++){
-      data[j]=rand()/((RAND_MAX>>1)+1);
-      logp1[j]=(rand()%15)+1;
-      enc_method[j]=rand()/((RAND_MAX>>2)+1);
-      switch(enc_method[j]){
-        case 0:{
-          ec_encode(&enc,data[j]?(1<<logp1[j])-1:0,
-           (1<<logp1[j])-(data[j]?0:1),1<<logp1[j]);
-        }break;
-        case 1:{
-          ec_encode_bin(&enc,data[j]?(1<<logp1[j])-1:0,
-           (1<<logp1[j])-(data[j]?0:1),logp1[j]);
-        }break;
-        case 2:{
-          ec_enc_bit_logp(&enc,data[j],logp1[j]);
-        }break;
-        case 3:{
-          unsigned char icdf[2];
-          icdf[0]=1;
-          icdf[1]=0;
-          ec_enc_icdf(&enc,data[j],icdf,logp1[j]);
-        }break;
-      }
-      tell[j+1]=ec_tell_frac(&enc);
-    }
-    ec_enc_done(&enc);
-    if((ec_tell(&enc)+7U)/8U<ec_range_bytes(&enc)){
-      fprintf(stderr,"tell() lied, there's %i bytes instead of %d (Random seed: %u)\n",
-       ec_range_bytes(&enc),(ec_tell(&enc)+7)/8,seed);
-      ret=-1;
-    }
-    ec_dec_init(&dec,ptr,DATA_SIZE2);
-    if(ec_tell_frac(&dec)!=tell[0]){
-      fprintf(stderr,
-       "Tell mismatch between encoder and decoder at symbol %i: %i instead of %i (Random seed: %u).\n",
-       0,ec_tell_frac(&dec),tell[0],seed);
-    }
-    for(j=0;j<sz;j++){
-      int fs;
-      int dec_method;
-      dec_method=rand()/((RAND_MAX>>2)+1);
-      switch(dec_method){
-        case 0:{
-          fs=ec_decode(&dec,1<<logp1[j]);
-          sym=fs>=(1<<logp1[j])-1;
-          ec_dec_update(&dec,sym?(1<<logp1[j])-1:0,
-           (1<<logp1[j])-(sym?0:1),1<<logp1[j]);
-        }break;
-        case 1:{
-          fs=ec_decode_bin(&dec,logp1[j]);
-          sym=fs>=(1<<logp1[j])-1;
-          ec_dec_update(&dec,sym?(1<<logp1[j])-1:0,
-           (1<<logp1[j])-(sym?0:1),1<<logp1[j]);
-        }break;
-        case 2:{
-          sym=ec_dec_bit_logp(&dec,logp1[j]);
-        }break;
-        case 3:{
-          unsigned char icdf[2];
-          icdf[0]=1;
-          icdf[1]=0;
-          sym=ec_dec_icdf(&dec,icdf,logp1[j]);
-        }break;
-      }
-      if(sym!=data[j]){
-        fprintf(stderr,
-         "Decoded %i instead of %i with logp1 of %i at position %i of %i (Random seed: %u).\n",
-         sym,data[j],logp1[j],j,sz,seed);
-        fprintf(stderr,"Encoding method: %i, decoding method: %i\n",
-         enc_method[j],dec_method);
-        ret=-1;
-      }
-      if(ec_tell_frac(&dec)!=tell[j+1]){
-        fprintf(stderr,
-         "Tell mismatch between encoder and decoder at symbol %i: %i instead of %i (Random seed: %u).\n",
-         j+1,ec_tell_frac(&dec),tell[j+1],seed);
-      }
-    }
-    free(enc_method);
-    free(tell);
-    free(data);
-    free(logp1);
-  }
-  ec_enc_init(&enc,ptr,DATA_SIZE2);
-  ec_enc_bit_logp(&enc,0,1);
-  ec_enc_bit_logp(&enc,0,1);
-  ec_enc_bit_logp(&enc,0,1);
-  ec_enc_bit_logp(&enc,0,1);
-  ec_enc_bit_logp(&enc,0,2);
-  ec_enc_patch_initial_bits(&enc,3,2);
-  if(enc.error){
-    fprintf(stderr,"patch_initial_bits failed");
-    ret=-1;
-  }
-  ec_enc_patch_initial_bits(&enc,0,5);
-  if(!enc.error){
-    fprintf(stderr,"patch_initial_bits didn't fail when it should have");
-    ret=-1;
-  }
-  ec_enc_done(&enc);
-  if(ec_range_bytes(&enc)!=1||ptr[0]!=192){
-    fprintf(stderr,"Got %d when expecting 192 for patch_initial_bits",ptr[0]);
-    ret=-1;
-  }
-  ec_enc_init(&enc,ptr,DATA_SIZE2);
-  ec_enc_bit_logp(&enc,0,1);
-  ec_enc_bit_logp(&enc,0,1);
-  ec_enc_bit_logp(&enc,1,6);
-  ec_enc_bit_logp(&enc,0,2);
-  ec_enc_patch_initial_bits(&enc,0,2);
-  if(enc.error){
-    fprintf(stderr,"patch_initial_bits failed");
-    ret=-1;
-  }
-  ec_enc_done(&enc);
-  if(ec_range_bytes(&enc)!=2||ptr[0]!=63){
-    fprintf(stderr,"Got %d when expecting 63 for patch_initial_bits",ptr[0]);
-    ret=-1;
-  }
-  ec_enc_init(&enc,ptr,2);
-  ec_enc_bit_logp(&enc,0,2);
-  for(i=0;i<48;i++){
-    ec_enc_bits(&enc,0,1);
-  }
-  ec_enc_done(&enc);
-  if(!enc.error){
-    fprintf(stderr,"Raw bits overfill didn't fail when it should have");
-    ret=-1;
-  }
-  ec_enc_init(&enc,ptr,2);
-  for(i=0;i<17;i++){
-    ec_enc_bits(&enc,0,1);
-  }
-  ec_enc_done(&enc);
-  if(!enc.error){
-    fprintf(stderr,"17 raw bits encoded in two bytes");
-    ret=-1;
-  }
-  free(ptr);
-  return ret;
-}
diff --git a/drivers/opus/celt/tests/test_unit_laplace.c b/drivers/opus/celt/tests/test_unit_laplace.c
deleted file mode 100644
index e4fe83c4d6..0000000000
--- a/drivers/opus/celt/tests/test_unit_laplace.c
+++ /dev/null
@@ -1,92 +0,0 @@
-/* Copyright (c) 2008-2011 Xiph.Org Foundation, Mozilla Corporation
-   Written by Jean-Marc Valin and Timothy B. Terriberry */
-/*
-   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.
-
-   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
-   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
-   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
-   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
-   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
-   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
-   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
-   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-*/
-
-#ifdef OPUS_ENABLED
-#include "opus/opus_config.h"
-#endif
-
-#include <stdio.h>
-#include <stdlib.h>
-#include "opus/celt/laplace.h"
-#define CELT_C
-#include "opus/celt/stack_alloc.h"
-
-#include "opus/celt/entenc.c"
-#include "opus/celt/entdec.c"
-#include "opus/celt/entcode.c"
-#include "opus/celt/laplace.c"
-
-#define DATA_SIZE 40000
-
-int ec_laplace_get_start_freq(int decay)
-{
-   opus_uint32 ft = 32768 - LAPLACE_MINP*(2*LAPLACE_NMIN+1);
-   int fs = (ft*(16384-decay))/(16384+decay);
-   return fs+LAPLACE_MINP;
-}
-
-int main(void)
-{
-   int i;
-   int ret = 0;
-   ec_enc enc;
-   ec_dec dec;
-   unsigned char *ptr;
-   int val[10000], decay[10000];
-   ALLOC_STACK;
-   ptr = (unsigned char *)malloc(DATA_SIZE);
-   ec_enc_init(&enc,ptr,DATA_SIZE);
-
-   val[0] = 3; decay[0] = 6000;
-   val[1] = 0; decay[1] = 5800;
-   val[2] = -1; decay[2] = 5600;
-   for (i=3;i<10000;i++)
-   {
-      val[i] = rand()%15-7;
-      decay[i] = rand()%11000+5000;
-   }
-   for (i=0;i<10000;i++)
-      ec_laplace_encode(&enc, &val[i],
-            ec_laplace_get_start_freq(decay[i]), decay[i]);
-
-   ec_enc_done(&enc);
-
-   ec_dec_init(&dec,ec_get_buffer(&enc),ec_range_bytes(&enc));
-
-   for (i=0;i<10000;i++)
-   {
-      int d = ec_laplace_decode(&dec,
-            ec_laplace_get_start_freq(decay[i]), decay[i]);
-      if (d != val[i])
-      {
-         fprintf (stderr, "Got %d instead of %d\n", d, val[i]);
-         ret = 1;
-      }
-   }
-
-   return ret;
-}
diff --git a/drivers/opus/celt/tests/test_unit_mathops.c b/drivers/opus/celt/tests/test_unit_mathops.c
deleted file mode 100644
index 0e3f300d40..0000000000
--- a/drivers/opus/celt/tests/test_unit_mathops.c
+++ /dev/null
@@ -1,275 +0,0 @@
-/* Copyright (c) 2008-2011 Xiph.Org Foundation, Mozilla Corporation,
-                           Gregory Maxwell
-   Written by Jean-Marc Valin, Gregory Maxwell, and Timothy B. Terriberry */
-/*
-   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.
-
-   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
-   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
-   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
-   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
-   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
-   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
-   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
-   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-*/
-
-#ifdef OPUS_ENABLED
-#include "opus/opus_config.h"
-#endif
-
-#ifndef CUSTOM_MODES
-#define CUSTOM_MODES
-#endif
-
-#define CELT_C
-
-#include "opus/celt/mathops.c"
-#include "opus/celt/entenc.c"
-#include "opus/celt/entdec.c"
-#include "opus/celt/entcode.c"
-#include "opus/celt/bands.c"
-#include "opus/celt/quant_bands.c"
-#include "opus/celt/laplace.c"
-#include "opus/celt/vq.c"
-#include "opus/celt/cwrs.c"
-#include <stdio.h>
-#include <math.h>
-
-#ifdef OPUS_FIXED_POINT
-#define WORD "%d"
-#else
-#define WORD "%f"
-#endif
-
-int ret = 0;
-
-void testdiv(void)
-{
-   opus_int32 i;
-   for (i=1;i<=327670;i++)
-   {
-      double prod;
-      opus_val32 val;
-      val = celt_rcp(i);
-#ifdef OPUS_FIXED_POINT
-      prod = (1./32768./65526.)*val*i;
-#else
-      prod = val*i;
-#endif
-      if (fabs(prod-1) > .00025)
-      {
-         fprintf (stderr, "div failed: 1/%d="WORD" (product = %f)\n", i, val, prod);
-         ret = 1;
-      }
-   }
-}
-
-void testsqrt(void)
-{
-   opus_int32 i;
-   for (i=1;i<=1000000000;i++)
-   {
-      double ratio;
-      opus_val16 val;
-      val = celt_sqrt(i);
-      ratio = val/sqrt(i);
-      if (fabs(ratio - 1) > .0005 && fabs(val-sqrt(i)) > 2)
-      {
-         fprintf (stderr, "sqrt failed: sqrt(%d)="WORD" (ratio = %f)\n", i, val, ratio);
-         ret = 1;
-      }
-      i+= i>>10;
-   }
-}
-
-void testbitexactcos(void)
-{
-   int i;
-   opus_int32 min_d,max_d,last,chk;
-   chk=max_d=0;
-   last=min_d=32767;
-   for(i=64;i<=16320;i++)
-   {
-      opus_int32 d;
-      opus_int32 q=bitexact_cos(i);
-      chk ^= q*i;
-      d = last - q;
-      if (d>max_d)max_d=d;
-      if (d<min_d)min_d=d;
-      last = q;
-   }
-   if ((chk!=89408644)||(max_d!=5)||(min_d!=0)||(bitexact_cos(64)!=32767)||
-       (bitexact_cos(16320)!=200)||(bitexact_cos(8192)!=23171))
-   {
-      fprintf (stderr, "bitexact_cos failed\n");
-      ret = 1;
-   }
-}
-
-void testbitexactlog2tan(void)
-{
-   int i,fail;
-   opus_int32 min_d,max_d,last,chk;
-   fail=chk=max_d=0;
-   last=min_d=15059;
-   for(i=64;i<8193;i++)
-   {
-      opus_int32 d;
-      opus_int32 mid=bitexact_cos(i);
-      opus_int32 side=bitexact_cos(16384-i);
-      opus_int32 q=bitexact_log2tan(mid,side);
-      chk ^= q*i;
-      d = last - q;
-      if (q!=-1*bitexact_log2tan(side,mid))
-        fail = 1;
-      if (d>max_d)max_d=d;
-      if (d<min_d)min_d=d;
-      last = q;
-   }
-   if ((chk!=15821257)||(max_d!=61)||(min_d!=-2)||fail||
-       (bitexact_log2tan(32767,200)!=15059)||(bitexact_log2tan(30274,12540)!=2611)||
-       (bitexact_log2tan(23171,23171)!=0))
-   {
-      fprintf (stderr, "bitexact_log2tan failed\n");
-      ret = 1;
-   }
-}
-
-#ifndef OPUS_FIXED_POINT
-void testlog2(void)
-{
-   float x;
-   for (x=0.001;x<1677700.0;x+=(x/8.0))
-   {
-      float error = fabs((1.442695040888963387*log(x))-celt_log2(x));
-      if (error>0.0009)
-      {
-         fprintf (stderr, "celt_log2 failed: fabs((1.442695040888963387*log(x))-celt_log2(x))>0.001 (x = %f, error = %f)\n", x,error);
-         ret = 1;
-      }
-   }
-}
-
-void testexp2(void)
-{
-   float x;
-   for (x=-11.0;x<24.0;x+=0.0007)
-   {
-      float error = fabs(x-(1.442695040888963387*log(celt_exp2(x))));
-      if (error>0.0002)
-      {
-         fprintf (stderr, "celt_exp2 failed: fabs(x-(1.442695040888963387*log(celt_exp2(x))))>0.0005 (x = %f, error = %f)\n", x,error);
-         ret = 1;
-      }
-   }
-}
-
-void testexp2log2(void)
-{
-   float x;
-   for (x=-11.0;x<24.0;x+=0.0007)
-   {
-      float error = fabs(x-(celt_log2(celt_exp2(x))));
-      if (error>0.001)
-      {
-         fprintf (stderr, "celt_log2/celt_exp2 failed: fabs(x-(celt_log2(celt_exp2(x))))>0.001 (x = %f, error = %f)\n", x,error);
-         ret = 1;
-      }
-   }
-}
-#else
-void testlog2(void)
-{
-   opus_val32 x;
-   for (x=8;x<1073741824;x+=(x>>3))
-   {
-      float error = fabs((1.442695040888963387*log(x/16384.0))-celt_log2(x)/1024.0);
-      if (error>0.003)
-      {
-         fprintf (stderr, "celt_log2 failed: x = %ld, error = %f\n", (long)x,error);
-         ret = 1;
-      }
-   }
-}
-
-void testexp2(void)
-{
-   opus_val16 x;
-   for (x=-32768;x<15360;x++)
-   {
-      float error1 = fabs(x/1024.0-(1.442695040888963387*log(celt_exp2(x)/65536.0)));
-      float error2 = fabs(exp(0.6931471805599453094*x/1024.0)-celt_exp2(x)/65536.0);
-      if (error1>0.0002&&error2>0.00004)
-      {
-    	 fprintf (stderr, "celt_exp2 failed: x = "WORD", error1 = %f, error2 = %f\n", x,error1,error2);
-         ret = 1;
-      }
-   }
-}
-
-void testexp2log2(void)
-{
-   opus_val32 x;
-   for (x=8;x<65536;x+=(x>>3))
-   {
-      float error = fabs(x-0.25*celt_exp2(celt_log2(x)))/16384;
-      if (error>0.004)
-      {
-         fprintf (stderr, "celt_log2/celt_exp2 failed: fabs(x-(celt_exp2(celt_log2(x))))>0.001 (x = %ld, error = %f)\n", (long)x,error);
-         ret = 1;
-      }
-   }
-}
-
-void testilog2(void)
-{
-   opus_val32 x;
-   for (x=1;x<=268435455;x+=127)
-   {
-      opus_val32 lg;
-      opus_val32 y;
-
-      lg = celt_ilog2(x);
-      if (lg<0 || lg>=31)
-      {
-         printf("celt_ilog2 failed: 0<=celt_ilog2(x)<31 (x = %d, celt_ilog2(x) = %d)\n",x,lg);
-         ret = 1;
-      }
-      y = 1<<lg;
-
-      if (x<y || (x>>1)>=y)
-      {
-         printf("celt_ilog2 failed: 2**celt_ilog2(x)<=x<2**(celt_ilog2(x)+1) (x = %d, 2**celt_ilog2(x) = %d)\n",x,y);
-         ret = 1;
-      }
-   }
-}
-#endif
-
-int main(void)
-{
-   testbitexactcos();
-   testbitexactlog2tan();
-   testdiv();
-   testsqrt();
-   testlog2();
-   testexp2();
-   testexp2log2();
-#ifdef OPUS_FIXED_POINT
-   testilog2();
-#endif
-   return ret;
-}
diff --git a/drivers/opus/celt/tests/test_unit_mdct.c b/drivers/opus/celt/tests/test_unit_mdct.c
deleted file mode 100644
index 0be03db2e8..0000000000
--- a/drivers/opus/celt/tests/test_unit_mdct.c
+++ /dev/null
@@ -1,210 +0,0 @@
-/* Copyright (c) 2008-2011 Xiph.Org Foundation
-   Written by Jean-Marc Valin */
-/*
-   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.
-
-   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
-   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
-   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
-   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
-   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
-   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
-   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
-   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-*/
-
-#ifdef OPUS_ENABLED
-#include "opus/opus_config.h"
-#endif
-
-#define SKIP_CONFIG_H
-
-#ifndef CUSTOM_MODES
-#define CUSTOM_MODES
-#endif
-
-#include <stdio.h>
-
-#define CELT_C
-#include "opus/celt/mdct.h"
-#include "opus/celt/stack_alloc.h"
-
-#include "opus/celt/kiss_fft.c"
-#include "opus/celt/mdct.c"
-#include "opus/celt/mathops.c"
-#include "opus/celt/entcode.c"
-
-#ifndef M_PI
-#define M_PI 3.141592653
-#endif
-
-int ret = 0;
-void check(kiss_fft_scalar  * in,kiss_fft_scalar  * out,int nfft,int isinverse)
-{
-    int bin,k;
-    double errpow=0,sigpow=0;
-    double snr;
-    for (bin=0;bin<nfft/2;++bin) {
-        double ansr = 0;
-        double difr;
-
-        for (k=0;k<nfft;++k) {
-           double phase = 2*M_PI*(k+.5+.25*nfft)*(bin+.5)/nfft;
-           double re = cos(phase);
-
-           re /= nfft/4;
-
-           ansr += in[k] * re;
-        }
-        /*printf ("%f %f\n", ansr, out[bin]);*/
-        difr = ansr - out[bin];
-        errpow += difr*difr;
-        sigpow += ansr*ansr;
-    }
-    snr = 10*log10(sigpow/errpow);
-    printf("nfft=%d inverse=%d,snr = %f\n",nfft,isinverse,snr );
-    if (snr<60) {
-       printf( "** poor snr: %f **\n", snr);
-       ret = 1;
-    }
-}
-
-void check_inv(kiss_fft_scalar  * in,kiss_fft_scalar  * out,int nfft,int isinverse)
-{
-   int bin,k;
-   double errpow=0,sigpow=0;
-   double snr;
-   for (bin=0;bin<nfft;++bin) {
-      double ansr = 0;
-      double difr;
-
-      for (k=0;k<nfft/2;++k) {
-         double phase = 2*M_PI*(bin+.5+.25*nfft)*(k+.5)/nfft;
-         double re = cos(phase);
-
-         /*re *= 2;*/
-
-         ansr += in[k] * re;
-      }
-      /*printf ("%f %f\n", ansr, out[bin]);*/
-      difr = ansr - out[bin];
-      errpow += difr*difr;
-      sigpow += ansr*ansr;
-   }
-   snr = 10*log10(sigpow/errpow);
-   printf("nfft=%d inverse=%d,snr = %f\n",nfft,isinverse,snr );
-   if (snr<60) {
-      printf( "** poor snr: %f **\n", snr);
-      ret = 1;
-   }
-}
-
-
-void test1d(int nfft,int isinverse)
-{
-    celt_mdct_lookup cfg;
-    size_t buflen = sizeof(kiss_fft_scalar)*nfft;
-
-    kiss_fft_scalar  * in = (kiss_fft_scalar*)malloc(buflen);
-    kiss_fft_scalar  * in_copy = (kiss_fft_scalar*)malloc(buflen);
-    kiss_fft_scalar  * out= (kiss_fft_scalar*)malloc(buflen);
-    opus_val16  * window= (opus_val16*)malloc(sizeof(opus_val16)*nfft/2);
-    int k;
-
-    clt_mdct_init(&cfg, nfft, 0);
-    for (k=0;k<nfft;++k) {
-        in[k] = (rand() % 32768) - 16384;
-    }
-
-    for (k=0;k<nfft/2;++k) {
-       window[k] = Q15ONE;
-    }
-    for (k=0;k<nfft;++k) {
-       in[k] *= 32768;
-    }
-
-    if (isinverse)
-    {
-       for (k=0;k<nfft;++k) {
-          in[k] /= nfft;
-       }
-    }
-
-    for (k=0;k<nfft;++k)
-       in_copy[k] = in[k];
-    /*for (k=0;k<nfft;++k) printf("%d %d ", in[k].r, in[k].i);printf("\n");*/
-
-    if (isinverse)
-    {
-       for (k=0;k<nfft;++k)
-          out[k] = 0;
-       clt_mdct_backward(&cfg,in,out, window, nfft/2, 0, 1);
-       /* apply TDAC because clt_mdct_backward() no longer does that */
-       for (k=0;k<nfft/4;++k)
-          out[nfft-k-1] = out[nfft/2+k];
-       check_inv(in,out,nfft,isinverse);
-    } else {
-       clt_mdct_forward(&cfg,in,out,window, nfft/2, 0, 1);
-       check(in_copy,out,nfft,isinverse);
-    }
-    /*for (k=0;k<nfft;++k) printf("%d %d ", out[k].r, out[k].i);printf("\n");*/
-
-
-    free(in);
-    free(out);
-    clt_mdct_clear(&cfg);
-}
-
-int main(int argc,char ** argv)
-{
-    ALLOC_STACK;
-    if (argc>1) {
-        int k;
-        for (k=1;k<argc;++k) {
-            test1d(atoi(argv[k]),0);
-            test1d(atoi(argv[k]),1);
-        }
-    }else{
-        test1d(32,0);
-        test1d(32,1);
-        test1d(256,0);
-        test1d(256,1);
-        test1d(512,0);
-        test1d(512,1);
-        test1d(1024,0);
-        test1d(1024,1);
-        test1d(2048,0);
-        test1d(2048,1);
-#ifndef RADIX_TWO_ONLY
-        test1d(36,0);
-        test1d(36,1);
-        test1d(40,0);
-        test1d(40,1);
-        test1d(60,0);
-        test1d(60,1);
-        test1d(120,0);
-        test1d(120,1);
-        test1d(240,0);
-        test1d(240,1);
-        test1d(480,0);
-        test1d(480,1);
-        test1d(960,0);
-        test1d(960,1);
-        test1d(1920,0);
-        test1d(1920,1);
-#endif
-    }
-    return ret;
-}
diff --git a/drivers/opus/celt/tests/test_unit_rotation.c b/drivers/opus/celt/tests/test_unit_rotation.c
deleted file mode 100644
index a57bf11e79..0000000000
--- a/drivers/opus/celt/tests/test_unit_rotation.c
+++ /dev/null
@@ -1,90 +0,0 @@
-/* Copyright (c) 2008-2011 Xiph.Org Foundation
-   Written by Jean-Marc Valin */
-/*
-   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.
-
-   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
-   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
-   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
-   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
-   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
-   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
-   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
-   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-*/
-
-#ifdef OPUS_ENABLED
-#include "opus/opus_config.h"
-#endif
-
-#ifndef CUSTOM_MODES
-#define CUSTOM_MODES
-#endif
-
-#define CELT_C
-
-#include <stdio.h>
-#include <stdlib.h>
-#include "opus/celt/vq.c"
-#include "opus/celt/cwrs.c"
-#include "opus/celt/entcode.c"
-#include "opus/celt/entenc.c"
-#include "opus/celt/entdec.c"
-#include "opus/celt/mathops.c"
-#include "opus/celt/bands.h"
-#include <math.h>
-#define MAX_SIZE 100
-
-int ret=0;
-void test_rotation(int N, int K)
-{
-   int i;
-   double err = 0, ener = 0, snr, snr0;
-   opus_val16 x0[MAX_SIZE];
-   opus_val16 x1[MAX_SIZE];
-   for (i=0;i<N;i++)
-      x1[i] = x0[i] = rand()%32767-16384;
-   exp_rotation(x1, N, 1, 1, K, SPREAD_NORMAL);
-   for (i=0;i<N;i++)
-   {
-      err += (x0[i]-(double)x1[i])*(x0[i]-(double)x1[i]);
-      ener += x0[i]*(double)x0[i];
-   }
-   snr0 = 20*log10(ener/err);
-   err = ener = 0;
-   exp_rotation(x1, N, -1, 1, K, SPREAD_NORMAL);
-   for (i=0;i<N;i++)
-   {
-      err += (x0[i]-(double)x1[i])*(x0[i]-(double)x1[i]);
-      ener += x0[i]*(double)x0[i];
-   }
-   snr = 20*log10(ener/err);
-   printf ("SNR for size %d (%d pulses) is %f (was %f without inverse)\n", N, K, snr, snr0);
-   if (snr < 60 || snr0 > 20)
-   {
-      fprintf(stderr, "FAIL!\n");
-      ret = 1;
-   }
-}
-
-int main(void)
-{
-   ALLOC_STACK;
-   test_rotation(15, 3);
-   test_rotation(23, 5);
-   test_rotation(50, 3);
-   test_rotation(80, 1);
-   return ret;
-}
diff --git a/drivers/opus/celt/vq.c b/drivers/opus/celt/vq.c
index 29a98a3a63..3061e3006f 100644
--- a/drivers/opus/celt/vq.c
+++ b/drivers/opus/celt/vq.c
@@ -25,10 +25,7 @@
    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
-
-#ifdef OPUS_ENABLED
 #include "opus/opus_config.h"
-#endif
 
 #include "opus/celt/mathops.h"
 #include "opus/celt/cwrs.h"
@@ -37,19 +34,23 @@
 #include "opus/celt/os_support.h"
 #include "opus/celt/bands.h"
 #include "opus/celt/rate.h"
+#include "opus/celt/pitch.h"
 
+#ifndef OVERRIDE_vq_exp_rotation1
 static void exp_rotation1(celt_norm *X, int len, int stride, opus_val16 c, opus_val16 s)
 {
    int i;
+   opus_val16 ms;
    celt_norm *Xptr;
    Xptr = X;
+   ms = NEG16(s);
    for (i=0;i<len-stride;i++)
    {
       celt_norm x1, x2;
       x1 = Xptr[0];
       x2 = Xptr[stride];
-      Xptr[stride] = EXTRACT16(SHR32(MULT16_16(c,x2) + MULT16_16(s,x1), 15));
-      *Xptr++      = EXTRACT16(SHR32(MULT16_16(c,x1) - MULT16_16(s,x2), 15));
+      Xptr[stride] = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x2),  s, x1), 15));
+      *Xptr++      = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x1), ms, x2), 15));
    }
    Xptr = &X[len-2*stride-1];
    for (i=len-2*stride-1;i>=0;i--)
@@ -57,10 +58,11 @@ static void exp_rotation1(celt_norm *X, int len, int stride, opus_val16 c, opus_
       celt_norm x1, x2;
       x1 = Xptr[0];
       x2 = Xptr[stride];
-      Xptr[stride] = EXTRACT16(SHR32(MULT16_16(c,x2) + MULT16_16(s,x1), 15));
-      *Xptr--      = EXTRACT16(SHR32(MULT16_16(c,x1) - MULT16_16(s,x2), 15));
+      Xptr[stride] = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x2),  s, x1), 15));
+      *Xptr--      = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x1), ms, x2), 15));
    }
 }
+#endif /* OVERRIDE_vq_exp_rotation1 */
 
 static void exp_rotation(celt_norm *X, int len, int dir, int stride, int K, int spread)
 {
@@ -91,7 +93,7 @@ static void exp_rotation(celt_norm *X, int len, int dir, int stride, int K, int
    }
    /*NOTE: As a minor optimization, we could be passing around log2(B), not B, for both this and for
       extract_collapse_mask().*/
-   len /= stride;
+   len = celt_udiv(len, stride);
    for (i=0;i<stride;i++)
    {
       if (dir < 0)
@@ -140,13 +142,15 @@ static unsigned extract_collapse_mask(int *iy, int N, int B)
       return 1;
    /*NOTE: As a minor optimization, we could be passing around log2(B), not B, for both this and for
       exp_rotation().*/
-   N0 = N/B;
+   N0 = celt_udiv(N, B);
    collapse_mask = 0;
    i=0; do {
       int j;
+      unsigned tmp=0;
       j=0; do {
-         collapse_mask |= (iy[i*N0+j]!=0)<<i;
+         tmp |= iy[i*N0+j];
       } while (++j<N0);
+      collapse_mask |= (tmp!=0)<<i;
    } while (++i<B);
    return collapse_mask;
 }
@@ -322,7 +326,6 @@ unsigned alg_quant(celt_norm *X, int N, int K, int spread, int B, ec_enc *enc
 unsigned alg_unquant(celt_norm *X, int N, int K, int spread, int B,
       ec_dec *dec, opus_val16 gain)
 {
-   int i;
    opus_val32 Ryy;
    unsigned collapse_mask;
    VARDECL(int, iy);
@@ -331,12 +334,7 @@ unsigned alg_unquant(celt_norm *X, int N, int K, int spread, int B,
    celt_assert2(K>0, "alg_unquant() needs at least one pulse");
    celt_assert2(N>1, "alg_unquant() needs at least two dimensions");
    ALLOC(iy, N, int);
-   decode_pulses(iy, N, K, dec);
-   Ryy = 0;
-   i=0;
-   do {
-      Ryy = MAC16_16(Ryy, iy[i], iy[i]);
-   } while (++i < N);
+   Ryy = decode_pulses(iy, N, K, dec);
    normalise_residual(iy, X, N, Ryy, gain);
    exp_rotation(X, N, -1, B, K, spread);
    collapse_mask = extract_collapse_mask(iy, N, B);
@@ -344,21 +342,18 @@ unsigned alg_unquant(celt_norm *X, int N, int K, int spread, int B,
    return collapse_mask;
 }
 
-void renormalise_vector(celt_norm *X, int N, opus_val16 gain)
+#ifndef OVERRIDE_renormalise_vector
+void renormalise_vector(celt_norm *X, int N, opus_val16 gain, int arch)
 {
    int i;
 #ifdef OPUS_FIXED_POINT
    int k;
 #endif
-   opus_val32 E = EPSILON;
+   opus_val32 E;
    opus_val16 g;
    opus_val32 t;
-   celt_norm *xptr = X;
-   for (i=0;i<N;i++)
-   {
-      E = MAC16_16(E, *xptr, *xptr);
-      xptr++;
-   }
+   celt_norm *xptr;
+   E = EPSILON + celt_inner_prod(X, X, N, arch);
 #ifdef OPUS_FIXED_POINT
    k = celt_ilog2(E)>>1;
 #endif
@@ -373,8 +368,9 @@ void renormalise_vector(celt_norm *X, int N, opus_val16 gain)
    }
    /*return celt_sqrt(E);*/
 }
+#endif /* OVERRIDE_renormalise_vector */
 
-int stereo_itheta(celt_norm *X, celt_norm *Y, int stereo, int N)
+int stereo_itheta(const celt_norm *X, const celt_norm *Y, int stereo, int N, int arch)
 {
    int i;
    int itheta;
@@ -393,14 +389,8 @@ int stereo_itheta(celt_norm *X, celt_norm *Y, int stereo, int N)
          Eside = MAC16_16(Eside, s, s);
       }
    } else {
-      for (i=0;i<N;i++)
-      {
-         celt_norm m, s;
-         m = X[i];
-         s = Y[i];
-         Emid = MAC16_16(Emid, m, m);
-         Eside = MAC16_16(Eside, s, s);
-      }
+      Emid += celt_inner_prod(X, X, N, arch);
+      Eside += celt_inner_prod(Y, Y, N, arch);
    }
    mid = celt_sqrt(Emid);
    side = celt_sqrt(Eside);
diff --git a/drivers/opus/celt/vq.h b/drivers/opus/celt/vq.h
index b52b1a0982..10b5e55306 100644
--- a/drivers/opus/celt/vq.h
+++ b/drivers/opus/celt/vq.h
@@ -35,7 +35,12 @@
 
 #include "opus/celt/entenc.h"
 #include "opus/celt/entdec.h"
-#include "opus/celt/opus_modes.h"
+#include "opus/celt/modes.h"
+
+#if defined(MIPSr1_ASM)
+#include "opus/celt/mips/vq_mipsr1.h"
+#endif
+
 
 /** Algebraic pulse-vector quantiser. The signal x is replaced by the sum of
   * the pitch and a combination of pulses such that its norm is still equal
@@ -63,8 +68,8 @@ unsigned alg_quant(celt_norm *X, int N, int K, int spread, int B,
 unsigned alg_unquant(celt_norm *X, int N, int K, int spread, int B,
       ec_dec *dec, opus_val16 gain);
 
-void renormalise_vector(celt_norm *X, int N, opus_val16 gain);
+void renormalise_vector(celt_norm *X, int N, opus_val16 gain, int arch);
 
-int stereo_itheta(celt_norm *X, celt_norm *Y, int stereo, int N);
+int stereo_itheta(const celt_norm *X, const celt_norm *Y, int stereo, int N, int arch);
 
 #endif /* VQ_H */
diff --git a/drivers/opus/celt/x86/celt_lpc_sse.c b/drivers/opus/celt/x86/celt_lpc_sse.c
new file mode 100644
index 0000000000..cfc86dc66c
--- /dev/null
+++ b/drivers/opus/celt/x86/celt_lpc_sse.c
@@ -0,0 +1,129 @@
+/* Copyright (c) 2014, Cisco Systems, INC
+   Written by XiangMingZhu WeiZhou MinPeng YanWang
+
+   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.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+#include "opus/opus_config.h"
+
+#include <xmmintrin.h>
+#include <emmintrin.h>
+#include <smmintrin.h>
+#include "opus/celt/celt_lpc.h"
+#include "opus/celt/stack_alloc.h"
+#include "opus/celt/mathops.h"
+#include "opus/celt/pitch.h"
+#include "opus/celt/x86/x86cpu.h"
+
+#if defined(FIXED_POINT)
+
+void celt_fir_sse4_1(const opus_val16 *_x,
+         const opus_val16 *num,
+         opus_val16 *_y,
+         int N,
+         int ord,
+         opus_val16 *mem,
+         int arch)
+{
+    int i,j;
+    VARDECL(opus_val16, rnum);
+    VARDECL(opus_val16, x);
+
+    __m128i vecNoA;
+    opus_int32 noA ;
+    SAVE_STACK;
+
+   ALLOC(rnum, ord, opus_val16);
+   ALLOC(x, N+ord, opus_val16);
+   for(i=0;i<ord;i++)
+      rnum[i] = num[ord-i-1];
+   for(i=0;i<ord;i++)
+      x[i] = mem[ord-i-1];
+
+   for (i=0;i<N-7;i+=8)
+   {
+       x[i+ord  ]=_x[i  ];
+       x[i+ord+1]=_x[i+1];
+       x[i+ord+2]=_x[i+2];
+       x[i+ord+3]=_x[i+3];
+       x[i+ord+4]=_x[i+4];
+       x[i+ord+5]=_x[i+5];
+       x[i+ord+6]=_x[i+6];
+       x[i+ord+7]=_x[i+7];
+   }
+
+   for (;i<N-3;i+=4)
+   {
+       x[i+ord  ]=_x[i  ];
+       x[i+ord+1]=_x[i+1];
+       x[i+ord+2]=_x[i+2];
+       x[i+ord+3]=_x[i+3];
+   }
+
+   for (;i<N;i++)
+         x[i+ord]=_x[i];
+
+   for(i=0;i<ord;i++)
+      mem[i] = _x[N-i-1];
+#ifdef SMALL_FOOTPRINT
+   for (i=0;i<N;i++)
+   {
+      opus_val32 sum = SHL32(EXTEND32(_x[i]), SIG_SHIFT);
+      for (j=0;j<ord;j++)
+      {
+         sum = MAC16_16(sum,rnum[j],x[i+j]);
+      }
+      _y[i] = SATURATE16(PSHR32(sum, SIG_SHIFT));
+   }
+#else
+   noA = EXTEND32(1) << SIG_SHIFT >> 1;
+   vecNoA = _mm_set_epi32(noA, noA, noA, noA);
+
+   for (i=0;i<N-3;i+=4)
+   {
+      opus_val32 sums[4] = {0};
+      __m128i vecSum, vecX;
+
+      xcorr_kernel(rnum, x+i, sums, ord, arch);
+
+      vecSum = _mm_loadu_si128((__m128i *)sums);
+      vecSum = _mm_add_epi32(vecSum, vecNoA);
+      vecSum = _mm_srai_epi32(vecSum, SIG_SHIFT);
+      vecX = OP_CVTEPI16_EPI32_M64(_x + i);
+      vecSum = _mm_add_epi32(vecSum, vecX);
+      vecSum = _mm_packs_epi32(vecSum, vecSum);
+      _mm_storel_epi64((__m128i *)(_y + i), vecSum);
+   }
+   for (;i<N;i++)
+   {
+      opus_val32 sum = 0;
+      for (j=0;j<ord;j++)
+         sum = MAC16_16(sum, rnum[j], x[i + j]);
+      _y[i] = SATURATE16(ADD32(EXTEND32(_x[i]), PSHR32(sum, SIG_SHIFT)));
+   }
+
+#endif
+   RESTORE_STACK;
+}
+
+#endif
diff --git a/drivers/opus/celt/tests/test_unit_types.c b/drivers/opus/celt/x86/celt_lpc_sse.h
index 12d5ca72f8..3787afd3ff 100644
--- a/drivers/opus/celt/tests/test_unit_types.c
+++ b/drivers/opus/celt/x86/celt_lpc_sse.h
@@ -1,6 +1,6 @@
-/* Copyright (c) 2008-2011 Xiph.Org Foundation
-   Written by Jean-Marc Valin */
-/*
+/* Copyright (c) 2014, Cisco Systems, INC
+   Written by XiangMingZhu WeiZhou MinPeng YanWang
+
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions
    are met:
@@ -25,26 +25,41 @@
    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
 
-#ifdef OPUS_ENABLED
+#ifndef CELT_LPC_SSE_H
+#define CELT_LPC_SSE_H
 #include "opus/opus_config.h"
+
+#if defined(OPUS_X86_MAY_HAVE_SSE4_1) && defined(FIXED_POINT)
+#define OVERRIDE_CELT_FIR
+
+void celt_fir_sse4_1(
+         const opus_val16 *x,
+         const opus_val16 *num,
+         opus_val16 *y,
+         int N,
+         int ord,
+         opus_val16 *mem,
+         int arch);
+
+#if defined(OPUS_X86_PRESUME_SSE4_1)
+#define celt_fir(x, num, y, N, ord, mem, arch) \
+    ((void)arch, celt_fir_sse4_1(x, num, y, N, ord, mem, arch))
+
+#else
+
+extern void (*const CELT_FIR_IMPL[OPUS_ARCHMASK + 1])(
+         const opus_val16 *x,
+         const opus_val16 *num,
+         opus_val16 *y,
+         int N,
+         int ord,
+         opus_val16 *mem,
+         int arch);
+
+#  define celt_fir(x, num, y, N, ord, mem, arch) \
+    ((*CELT_FIR_IMPL[(arch) & OPUS_ARCHMASK])(x, num, y, N, ord, mem, arch))
+
+#endif
 #endif
 
-#include "opus/opus_types.h"
-#include <stdio.h>
-
-int main(void)
-{
-   opus_int16 i = 1;
-   i <<= 14;
-   if (i>>14 != 1)
-   {
-      fprintf(stderr, "opus_int16 isn't 16 bits\n");
-      return 1;
-   }
-   if (sizeof(opus_int16)*2 != sizeof(opus_int32))
-   {
-      fprintf(stderr, "16*2 != 32\n");
-      return 1;
-   }
-   return 0;
-}
+#endif
diff --git a/drivers/opus/celt/x86/pitch_sse.c b/drivers/opus/celt/x86/pitch_sse.c
new file mode 100644
index 0000000000..9f22ffd9ab
--- /dev/null
+++ b/drivers/opus/celt/x86/pitch_sse.c
@@ -0,0 +1,182 @@
+/* Copyright (c) 2014, Cisco Systems, INC
+   Written by XiangMingZhu WeiZhou MinPeng YanWang
+
+   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.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+#include "opus/opus_config.h"
+
+#include "opus/silk/macros.h"
+#include "opus/celt/celt_lpc.h"
+#include "opus/celt/stack_alloc.h"
+#include "opus/celt/mathops.h"
+#include "opus/celt/pitch.h"
+
+#if defined(OPUS_X86_MAY_HAVE_SSE) && !defined(FIXED_POINT)
+
+#include <xmmintrin.h>
+#include "opus/celt/arch.h"
+
+void xcorr_kernel_sse(const opus_val16 *x, const opus_val16 *y, opus_val32 sum[4], int len)
+{
+   int j;
+   __m128 xsum1, xsum2;
+   xsum1 = _mm_loadu_ps(sum);
+   xsum2 = _mm_setzero_ps();
+
+   for (j = 0; j < len-3; j += 4)
+   {
+      __m128 x0 = _mm_loadu_ps(x+j);
+      __m128 yj = _mm_loadu_ps(y+j);
+      __m128 y3 = _mm_loadu_ps(y+j+3);
+
+      xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0x00),yj));
+      xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0x55),
+                                          _mm_shuffle_ps(yj,y3,0x49)));
+      xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0xaa),
+                                          _mm_shuffle_ps(yj,y3,0x9e)));
+      xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0xff),y3));
+   }
+   if (j < len)
+   {
+      xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j)));
+      if (++j < len)
+      {
+         xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j)));
+         if (++j < len)
+         {
+            xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j)));
+         }
+      }
+   }
+   _mm_storeu_ps(sum,_mm_add_ps(xsum1,xsum2));
+}
+
+
+void dual_inner_prod_sse(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02,
+      int N, opus_val32 *xy1, opus_val32 *xy2)
+{
+   int i;
+   __m128 xsum1, xsum2;
+   xsum1 = _mm_setzero_ps();
+   xsum2 = _mm_setzero_ps();
+   for (i=0;i<N-3;i+=4)
+   {
+      __m128 xi = _mm_loadu_ps(x+i);
+      __m128 y1i = _mm_loadu_ps(y01+i);
+      __m128 y2i = _mm_loadu_ps(y02+i);
+      xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(xi, y1i));
+      xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(xi, y2i));
+   }
+   /* Horizontal sum */
+   xsum1 = _mm_add_ps(xsum1, _mm_movehl_ps(xsum1, xsum1));
+   xsum1 = _mm_add_ss(xsum1, _mm_shuffle_ps(xsum1, xsum1, 0x55));
+   _mm_store_ss(xy1, xsum1);
+   xsum2 = _mm_add_ps(xsum2, _mm_movehl_ps(xsum2, xsum2));
+   xsum2 = _mm_add_ss(xsum2, _mm_shuffle_ps(xsum2, xsum2, 0x55));
+   _mm_store_ss(xy2, xsum2);
+   for (;i<N;i++)
+   {
+      *xy1 = MAC16_16(*xy1, x[i], y01[i]);
+      *xy2 = MAC16_16(*xy2, x[i], y02[i]);
+   }
+}
+
+opus_val32 celt_inner_prod_sse(const opus_val16 *x, const opus_val16 *y,
+      int N)
+{
+   int i;
+   float xy;
+   __m128 sum;
+   sum = _mm_setzero_ps();
+   /* FIXME: We should probably go 8-way and use 2 sums. */
+   for (i=0;i<N-3;i+=4)
+   {
+      __m128 xi = _mm_loadu_ps(x+i);
+      __m128 yi = _mm_loadu_ps(y+i);
+      sum = _mm_add_ps(sum,_mm_mul_ps(xi, yi));
+   }
+   /* Horizontal 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(&xy, sum);
+   for (;i<N;i++)
+   {
+      xy = MAC16_16(xy, x[i], y[i]);
+   }
+   return xy;
+}
+
+void comb_filter_const_sse(opus_val32 *y, opus_val32 *x, int T, int N,
+      opus_val16 g10, opus_val16 g11, opus_val16 g12)
+{
+   int i;
+   __m128 x0v;
+   __m128 g10v, g11v, g12v;
+   g10v = _mm_load1_ps(&g10);
+   g11v = _mm_load1_ps(&g11);
+   g12v = _mm_load1_ps(&g12);
+   x0v = _mm_loadu_ps(&x[-T-2]);
+   for (i=0;i<N-3;i+=4)
+   {
+      __m128 yi, yi2, x1v, x2v, x3v, x4v;
+      const opus_val32 *xp = &x[i-T-2];
+      yi = _mm_loadu_ps(x+i);
+      x4v = _mm_loadu_ps(xp+4);
+#if 0
+      /* Slower version with all loads */
+      x1v = _mm_loadu_ps(xp+1);
+      x2v = _mm_loadu_ps(xp+2);
+      x3v = _mm_loadu_ps(xp+3);
+#else
+      x2v = _mm_shuffle_ps(x0v, x4v, 0x4e);
+      x1v = _mm_shuffle_ps(x0v, x2v, 0x99);
+      x3v = _mm_shuffle_ps(x2v, x4v, 0x99);
+#endif
+
+      yi = _mm_add_ps(yi, _mm_mul_ps(g10v,x2v));
+#if 0 /* Set to 1 to make it bit-exact with the non-SSE version */
+      yi = _mm_add_ps(yi, _mm_mul_ps(g11v,_mm_add_ps(x3v,x1v)));
+      yi = _mm_add_ps(yi, _mm_mul_ps(g12v,_mm_add_ps(x4v,x0v)));
+#else
+      /* Use partial sums */
+      yi2 = _mm_add_ps(_mm_mul_ps(g11v,_mm_add_ps(x3v,x1v)),
+                       _mm_mul_ps(g12v,_mm_add_ps(x4v,x0v)));
+      yi = _mm_add_ps(yi, yi2);
+#endif
+      x0v=x4v;
+      _mm_storeu_ps(y+i, yi);
+   }
+#ifdef CUSTOM_MODES
+   for (;i<N;i++)
+   {
+      y[i] = x[i]
+               + MULT16_32_Q15(g10,x[i-T])
+               + MULT16_32_Q15(g11,ADD32(x[i-T+1],x[i-T-1]))
+               + MULT16_32_Q15(g12,ADD32(x[i-T+2],x[i-T-2]));
+   }
+#endif
+}
+
+
+#endif
diff --git a/drivers/opus/celt/x86/pitch_sse.h b/drivers/opus/celt/x86/pitch_sse.h
index 1542b87232..2b4b54f61f 100644
--- a/drivers/opus/celt/x86/pitch_sse.h
+++ b/drivers/opus/celt/x86/pitch_sse.h
@@ -1,4 +1,5 @@
-/* Copyright (c) 2013 Jean-Marc Valin and John Ridges */
+/* Copyright (c) 2013 Jean-Marc Valin and John Ridges
+   Copyright (c) 2014, Cisco Systems, INC MingXiang WeiZhou MinPeng YanWang*/
 /**
    @file pitch_sse.h
    @brief Pitch analysis
@@ -31,126 +32,158 @@
 
 #ifndef PITCH_SSE_H
 #define PITCH_SSE_H
+#include "opus/opus_config.h"
 
-#include <xmmintrin.h>
-#include "opus/celt/arch.h"
+#if defined(OPUS_X86_MAY_HAVE_SSE4_1) && defined(FIXED_POINT)
+void xcorr_kernel_sse4_1(
+                    const opus_int16 *x,
+                    const opus_int16 *y,
+                    opus_val32       sum[4],
+                    int              len);
+#endif
+
+#if defined(OPUS_X86_MAY_HAVE_SSE) && !defined(FIXED_POINT)
+void xcorr_kernel_sse(
+                    const opus_val16 *x,
+                    const opus_val16 *y,
+                    opus_val32       sum[4],
+                    int              len);
+#endif
 
+#if defined(OPUS_X86_PRESUME_SSE4_1) && defined(FIXED_POINT)
 #define OVERRIDE_XCORR_KERNEL
-static OPUS_INLINE void xcorr_kernel(const opus_val16 *x, const opus_val16 *y, opus_val32 sum[4], int len)
-{
-   int j;
-   __m128 xsum1, xsum2;
-   xsum1 = _mm_loadu_ps(sum);
-   xsum2 = _mm_setzero_ps();
-
-   for (j = 0; j < len-3; j += 4)
-   {
-      __m128 x0 = _mm_loadu_ps(x+j);
-      __m128 yj = _mm_loadu_ps(y+j);
-      __m128 y3 = _mm_loadu_ps(y+j+3);
-
-      xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0x00),yj));
-      xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0x55),
-                                          _mm_shuffle_ps(yj,y3,0x49)));
-      xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0xaa),
-                                          _mm_shuffle_ps(yj,y3,0x9e)));
-      xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0xff),y3));
-   }
-   if (j < len)
-   {
-      xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j)));
-      if (++j < len)
-      {
-         xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j)));
-         if (++j < len)
-         {
-            xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j)));
-         }
-      }
-   }
-   _mm_storeu_ps(sum,_mm_add_ps(xsum1,xsum2));
-}
+#define xcorr_kernel(x, y, sum, len, arch) \
+    ((void)arch, xcorr_kernel_sse4_1(x, y, sum, len))
 
-#define OVERRIDE_DUAL_INNER_PROD
-static OPUS_INLINE void dual_inner_prod(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02,
-      int N, opus_val32 *xy1, opus_val32 *xy2)
-{
-   int i;
-   __m128 xsum1, xsum2;
-   xsum1 = _mm_setzero_ps();
-   xsum2 = _mm_setzero_ps();
-   for (i=0;i<N-3;i+=4)
-   {
-      __m128 xi = _mm_loadu_ps(x+i);
-      __m128 y1i = _mm_loadu_ps(y01+i);
-      __m128 y2i = _mm_loadu_ps(y02+i);
-      xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(xi, y1i));
-      xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(xi, y2i));
-   }
-   /* Horizontal sum */
-   xsum1 = _mm_add_ps(xsum1, _mm_movehl_ps(xsum1, xsum1));
-   xsum1 = _mm_add_ss(xsum1, _mm_shuffle_ps(xsum1, xsum1, 0x55));
-   _mm_store_ss(xy1, xsum1);
-   xsum2 = _mm_add_ps(xsum2, _mm_movehl_ps(xsum2, xsum2));
-   xsum2 = _mm_add_ss(xsum2, _mm_shuffle_ps(xsum2, xsum2, 0x55));
-   _mm_store_ss(xy2, xsum2);
-   for (;i<N;i++)
-   {
-      *xy1 = MAC16_16(*xy1, x[i], y01[i]);
-      *xy2 = MAC16_16(*xy2, x[i], y02[i]);
-   }
-}
+#elif defined(OPUS_X86_PRESUME_SSE) && !defined(FIXED_POINT)
+#define OVERRIDE_XCORR_KERNEL
+#define xcorr_kernel(x, y, sum, len, arch) \
+    ((void)arch, xcorr_kernel_sse(x, y, sum, len))
+
+#elif (defined(OPUS_X86_MAY_HAVE_SSE4_1) && defined(FIXED_POINT)) || (defined(OPUS_X86_MAY_HAVE_SSE) && !defined(FIXED_POINT))
+
+extern void (*const XCORR_KERNEL_IMPL[OPUS_ARCHMASK + 1])(
+                    const opus_val16 *x,
+                    const opus_val16 *y,
+                    opus_val32       sum[4],
+                    int              len);
+
+#define OVERRIDE_XCORR_KERNEL
+#define xcorr_kernel(x, y, sum, len, arch) \
+    ((*XCORR_KERNEL_IMPL[(arch) & OPUS_ARCHMASK])(x, y, sum, len))
+
+#endif
+
+#if defined(OPUS_X86_MAY_HAVE_SSE4_1) && defined(FIXED_POINT)
+opus_val32 celt_inner_prod_sse4_1(
+    const opus_int16 *x,
+    const opus_int16 *y,
+    int               N);
+#endif
+
+#if defined(OPUS_X86_MAY_HAVE_SSE2) && defined(FIXED_POINT)
+opus_val32 celt_inner_prod_sse2(
+    const opus_int16 *x,
+    const opus_int16 *y,
+    int               N);
+#endif
+
+#if defined(OPUS_X86_MAY_HAVE_SSE2) && !defined(FIXED_POINT)
+opus_val32 celt_inner_prod_sse(
+    const opus_val16 *x,
+    const opus_val16 *y,
+    int               N);
+#endif
+
+
+#if defined(OPUS_X86_PRESUME_SSE4_1) && defined(FIXED_POINT)
+#define OVERRIDE_CELT_INNER_PROD
+#define celt_inner_prod(x, y, N, arch) \
+	((void)arch, celt_inner_prod_sse4_1(x, y, N))
+
+#elif defined(OPUS_X86_PRESUME_SSE2) && defined(FIXED_POINT) && !defined(OPUS_X86_MAY_HAVE_SSE4_1)
+#define OVERRIDE_CELT_INNER_PROD
+#define celt_inner_prod(x, y, N, arch) \
+	((void)arch, celt_inner_prod_sse2(x, y, N))
+
+#elif defined(OPUS_X86_PRESUME_SSE) && !defined(FIXED_POINT)
+#define OVERRIDE_CELT_INNER_PROD
+#define celt_inner_prod(x, y, N, arch) \
+	((void)arch, celt_inner_prod_sse(x, y, N))
+
+
+#elif ((defined(OPUS_X86_MAY_HAVE_SSE4_1) || defined(OPUS_X86_MAY_HAVE_SSE2)) && defined(FIXED_POINT)) || \
+	(defined(OPUS_X86_MAY_HAVE_SSE) && !defined(FIXED_POINT))
+
+extern opus_val32 (*const CELT_INNER_PROD_IMPL[OPUS_ARCHMASK + 1])(
+                    const opus_val16 *x,
+                    const opus_val16 *y,
+                    int               N);
+
+#define OVERRIDE_CELT_INNER_PROD
+#define celt_inner_prod(x, y, N, arch) \
+    ((*CELT_INNER_PROD_IMPL[(arch) & OPUS_ARCHMASK])(x, y, N))
 
-#define OVERRIDE_COMB_FILTER_CONST
-static OPUS_INLINE void comb_filter_const(opus_val32 *y, opus_val32 *x, int T, int N,
-      opus_val16 g10, opus_val16 g11, opus_val16 g12)
-{
-   int i;
-   __m128 x0v;
-   __m128 g10v, g11v, g12v;
-   g10v = _mm_load1_ps(&g10);
-   g11v = _mm_load1_ps(&g11);
-   g12v = _mm_load1_ps(&g12);
-   x0v = _mm_loadu_ps(&x[-T-2]);
-   for (i=0;i<N-3;i+=4)
-   {
-      __m128 yi, yi2, x1v, x2v, x3v, x4v;
-      const opus_val32 *xp = &x[i-T-2];
-      yi = _mm_loadu_ps(x+i);
-      x4v = _mm_loadu_ps(xp+4);
-#if 0
-      /* Slower version with all loads */
-      x1v = _mm_loadu_ps(xp+1);
-      x2v = _mm_loadu_ps(xp+2);
-      x3v = _mm_loadu_ps(xp+3);
-#else
-      x2v = _mm_shuffle_ps(x0v, x4v, 0x4e);
-      x1v = _mm_shuffle_ps(x0v, x2v, 0x99);
-      x3v = _mm_shuffle_ps(x2v, x4v, 0x99);
 #endif
 
-      yi = _mm_add_ps(yi, _mm_mul_ps(g10v,x2v));
-#if 0 /* Set to 1 to make it bit-exact with the non-SSE version */
-      yi = _mm_add_ps(yi, _mm_mul_ps(g11v,_mm_add_ps(x3v,x1v)));
-      yi = _mm_add_ps(yi, _mm_mul_ps(g12v,_mm_add_ps(x4v,x0v)));
+#if defined(OPUS_X86_MAY_HAVE_SSE) && !defined(FIXED_POINT)
+
+#define OVERRIDE_DUAL_INNER_PROD
+#define OVERRIDE_COMB_FILTER_CONST
+
+#undef dual_inner_prod
+#undef comb_filter_const
+
+void dual_inner_prod_sse(const opus_val16 *x,
+	const opus_val16 *y01,
+	const opus_val16 *y02,
+	int               N,
+	opus_val32       *xy1,
+	opus_val32       *xy2);
+
+void comb_filter_const_sse(opus_val32 *y,
+	opus_val32 *x,
+	int         T,
+	int         N,
+	opus_val16  g10,
+	opus_val16  g11,
+	opus_val16  g12);
+
+
+#if defined(OPUS_X86_PRESUME_SSE)
+# define dual_inner_prod(x, y01, y02, N, xy1, xy2, arch) \
+    ((void)(arch),dual_inner_prod_sse(x, y01, y02, N, xy1, xy2))
+
+# define comb_filter_const(y, x, T, N, g10, g11, g12, arch) \
+    ((void)(arch),comb_filter_const_sse(y, x, T, N, g10, g11, g12))
 #else
-      /* Use partial sums */
-      yi2 = _mm_add_ps(_mm_mul_ps(g11v,_mm_add_ps(x3v,x1v)),
-                       _mm_mul_ps(g12v,_mm_add_ps(x4v,x0v)));
-      yi = _mm_add_ps(yi, yi2);
+
+extern void (*const DUAL_INNER_PROD_IMPL[OPUS_ARCHMASK + 1])(
+              const opus_val16 *x,
+              const opus_val16 *y01,
+              const opus_val16 *y02,
+              int               N,
+              opus_val32       *xy1,
+              opus_val32       *xy2);
+
+#define dual_inner_prod(x, y01, y02, N, xy1, xy2, arch)			\
+    ((*DUAL_INNER_PROD_IMPL[(arch) & OPUS_ARCHMASK])(x, y01, y02, N, xy1, xy2))
+
+extern void (*const COMB_FILTER_CONST_IMPL[OPUS_ARCHMASK + 1])(
+              opus_val32 *y,
+              opus_val32 *x,
+              int         T,
+              int         N,
+              opus_val16  g10,
+              opus_val16  g11,
+              opus_val16  g12);
+
+#define comb_filter_const(y, x, T, N, g10, g11, g12, arch)				\
+    ((*COMB_FILTER_CONST_IMPL[(arch) & OPUS_ARCHMASK])(y, x, T, N, g10, g11, g12))
+
+#define NON_STATIC_COMB_FILTER_CONST_C
+
 #endif
-      x0v=x4v;
-      _mm_storeu_ps(y+i, yi);
-   }
-#ifdef CUSTOM_MODES
-   for (;i<N;i++)
-   {
-      y[i] = x[i]
-               + MULT16_32_Q15(g10,x[i-T])
-               + MULT16_32_Q15(g11,ADD32(x[i-T+1],x[i-T-1]))
-               + MULT16_32_Q15(g12,ADD32(x[i-T+2],x[i-T-2]));
-   }
 #endif
-}
 
 #endif
diff --git a/drivers/opus/celt/x86/pitch_sse2.c b/drivers/opus/celt/x86/pitch_sse2.c
new file mode 100644
index 0000000000..b5a78b4acd
--- /dev/null
+++ b/drivers/opus/celt/x86/pitch_sse2.c
@@ -0,0 +1,92 @@
+/* Copyright (c) 2014, Cisco Systems, INC
+   Written by XiangMingZhu WeiZhou MinPeng YanWang
+
+   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.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+#include "opus/opus_config.h"
+
+#include <xmmintrin.h>
+#include <emmintrin.h>
+
+#include "opus/silk/macros.h"
+#include "opus/celt/celt_lpc.h"
+#include "opus/celt/stack_alloc.h"
+#include "opus/celt/mathops.h"
+#include "opus/celt/pitch.h"
+
+#if defined(OPUS_X86_MAY_HAVE_SSE2) && defined(FIXED_POINT)
+opus_val32 celt_inner_prod_sse2(const opus_val16 *x, const opus_val16 *y,
+      int N)
+{
+    opus_int  i, dataSize16;
+    opus_int32 sum;
+
+    __m128i inVec1_76543210, inVec1_FEDCBA98, acc1;
+    __m128i inVec2_76543210, inVec2_FEDCBA98, acc2;
+
+    sum = 0;
+    dataSize16 = N & ~15;
+
+    acc1 = _mm_setzero_si128();
+    acc2 = _mm_setzero_si128();
+
+    for (i=0;i<dataSize16;i+=16)
+    {
+        inVec1_76543210 = _mm_loadu_si128((__m128i *)(&x[i + 0]));
+        inVec2_76543210 = _mm_loadu_si128((__m128i *)(&y[i + 0]));
+
+        inVec1_FEDCBA98 = _mm_loadu_si128((__m128i *)(&x[i + 8]));
+        inVec2_FEDCBA98 = _mm_loadu_si128((__m128i *)(&y[i + 8]));
+
+        inVec1_76543210 = _mm_madd_epi16(inVec1_76543210, inVec2_76543210);
+        inVec1_FEDCBA98 = _mm_madd_epi16(inVec1_FEDCBA98, inVec2_FEDCBA98);
+
+        acc1 = _mm_add_epi32(acc1, inVec1_76543210);
+        acc2 = _mm_add_epi32(acc2, inVec1_FEDCBA98);
+    }
+
+    acc1 = _mm_add_epi32( acc1, acc2 );
+
+    if (N - i >= 8)
+    {
+        inVec1_76543210 = _mm_loadu_si128((__m128i *)(&x[i + 0]));
+        inVec2_76543210 = _mm_loadu_si128((__m128i *)(&y[i + 0]));
+
+        inVec1_76543210 = _mm_madd_epi16(inVec1_76543210, inVec2_76543210);
+
+        acc1 = _mm_add_epi32(acc1, inVec1_76543210);
+        i += 8;
+    }
+
+    acc1 = _mm_add_epi32(acc1, _mm_unpackhi_epi64( acc1, acc1));
+    acc1 = _mm_add_epi32(acc1, _mm_shufflelo_epi16( acc1, 0x0E));
+    sum += _mm_cvtsi128_si32(acc1);
+
+    for (;i<N;i++) {
+        sum = silk_SMLABB(sum, x[i], y[i]);
+    }
+
+    return sum;
+}
+#endif
diff --git a/drivers/opus/celt/x86/pitch_sse4_1.c b/drivers/opus/celt/x86/pitch_sse4_1.c
new file mode 100644
index 0000000000..57fb332210
--- /dev/null
+++ b/drivers/opus/celt/x86/pitch_sse4_1.c
@@ -0,0 +1,192 @@
+/* Copyright (c) 2014, Cisco Systems, INC
+   Written by XiangMingZhu WeiZhou MinPeng YanWang
+
+   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.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+#include "opus/opus_config.h"
+
+#include <xmmintrin.h>
+#include <emmintrin.h>
+
+#include "opus/silk/macros.h"
+#include "opus/celt/celt_lpc.h"
+#include "opus/celt/stack_alloc.h"
+#include "opus/celt/mathops.h"
+#include "opus/celt/pitch.h"
+
+#if defined(OPUS_X86_MAY_HAVE_SSE4_1) && defined(FIXED_POINT)
+#include <smmintrin.h>
+#include "opus/celt/x86/x86cpu.h"
+
+opus_val32 celt_inner_prod_sse4_1(const opus_val16 *x, const opus_val16 *y,
+      int N)
+{
+    opus_int  i, dataSize16;
+    opus_int32 sum;
+    __m128i inVec1_76543210, inVec1_FEDCBA98, acc1;
+    __m128i inVec2_76543210, inVec2_FEDCBA98, acc2;
+    __m128i inVec1_3210, inVec2_3210;
+
+    sum = 0;
+    dataSize16 = N & ~15;
+
+    acc1 = _mm_setzero_si128();
+    acc2 = _mm_setzero_si128();
+
+    for (i=0;i<dataSize16;i+=16) {
+        inVec1_76543210 = _mm_loadu_si128((__m128i *)(&x[i + 0]));
+        inVec2_76543210 = _mm_loadu_si128((__m128i *)(&y[i + 0]));
+
+        inVec1_FEDCBA98 = _mm_loadu_si128((__m128i *)(&x[i + 8]));
+        inVec2_FEDCBA98 = _mm_loadu_si128((__m128i *)(&y[i + 8]));
+
+        inVec1_76543210 = _mm_madd_epi16(inVec1_76543210, inVec2_76543210);
+        inVec1_FEDCBA98 = _mm_madd_epi16(inVec1_FEDCBA98, inVec2_FEDCBA98);
+
+        acc1 = _mm_add_epi32(acc1, inVec1_76543210);
+        acc2 = _mm_add_epi32(acc2, inVec1_FEDCBA98);
+    }
+
+    acc1 = _mm_add_epi32(acc1, acc2);
+
+    if (N - i >= 8)
+    {
+        inVec1_76543210 = _mm_loadu_si128((__m128i *)(&x[i + 0]));
+        inVec2_76543210 = _mm_loadu_si128((__m128i *)(&y[i + 0]));
+
+        inVec1_76543210 = _mm_madd_epi16(inVec1_76543210, inVec2_76543210);
+
+        acc1 = _mm_add_epi32(acc1, inVec1_76543210);
+        i += 8;
+    }
+
+    if (N - i >= 4)
+    {
+        inVec1_3210 = OP_CVTEPI16_EPI32_M64(&x[i + 0]);
+        inVec2_3210 = OP_CVTEPI16_EPI32_M64(&y[i + 0]);
+
+        inVec1_3210 = _mm_mullo_epi32(inVec1_3210, inVec2_3210);
+
+        acc1 = _mm_add_epi32(acc1, inVec1_3210);
+        i += 4;
+    }
+
+    acc1 = _mm_add_epi32(acc1, _mm_unpackhi_epi64(acc1, acc1));
+    acc1 = _mm_add_epi32(acc1, _mm_shufflelo_epi16(acc1, 0x0E));
+
+    sum += _mm_cvtsi128_si32(acc1);
+
+    for (;i<N;i++)
+    {
+        sum = silk_SMLABB(sum, x[i], y[i]);
+    }
+
+    return sum;
+}
+
+void xcorr_kernel_sse4_1(const opus_val16 * x, const opus_val16 * y, opus_val32 sum[ 4 ], int len)
+{
+    int j;
+
+    __m128i vecX, vecX0, vecX1, vecX2, vecX3;
+    __m128i vecY0, vecY1, vecY2, vecY3;
+    __m128i sum0, sum1, sum2, sum3, vecSum;
+    __m128i initSum;
+
+    celt_assert(len >= 3);
+
+    sum0 = _mm_setzero_si128();
+    sum1 = _mm_setzero_si128();
+    sum2 = _mm_setzero_si128();
+    sum3 = _mm_setzero_si128();
+
+    for (j=0;j<(len-7);j+=8)
+    {
+        vecX = _mm_loadu_si128((__m128i *)(&x[j + 0]));
+        vecY0 = _mm_loadu_si128((__m128i *)(&y[j + 0]));
+        vecY1 = _mm_loadu_si128((__m128i *)(&y[j + 1]));
+        vecY2 = _mm_loadu_si128((__m128i *)(&y[j + 2]));
+        vecY3 = _mm_loadu_si128((__m128i *)(&y[j + 3]));
+
+        sum0 = _mm_add_epi32(sum0, _mm_madd_epi16(vecX, vecY0));
+        sum1 = _mm_add_epi32(sum1, _mm_madd_epi16(vecX, vecY1));
+        sum2 = _mm_add_epi32(sum2, _mm_madd_epi16(vecX, vecY2));
+        sum3 = _mm_add_epi32(sum3, _mm_madd_epi16(vecX, vecY3));
+    }
+
+    sum0 = _mm_add_epi32(sum0, _mm_unpackhi_epi64( sum0, sum0));
+    sum0 = _mm_add_epi32(sum0, _mm_shufflelo_epi16( sum0, 0x0E));
+
+    sum1 = _mm_add_epi32(sum1, _mm_unpackhi_epi64( sum1, sum1));
+    sum1 = _mm_add_epi32(sum1, _mm_shufflelo_epi16( sum1, 0x0E));
+
+    sum2 = _mm_add_epi32(sum2, _mm_unpackhi_epi64( sum2, sum2));
+    sum2 = _mm_add_epi32(sum2, _mm_shufflelo_epi16( sum2, 0x0E));
+
+    sum3 = _mm_add_epi32(sum3, _mm_unpackhi_epi64( sum3, sum3));
+    sum3 = _mm_add_epi32(sum3, _mm_shufflelo_epi16( sum3, 0x0E));
+
+    vecSum = _mm_unpacklo_epi64(_mm_unpacklo_epi32(sum0, sum1),
+          _mm_unpacklo_epi32(sum2, sum3));
+
+    for (;j<(len-3);j+=4)
+    {
+        vecX = OP_CVTEPI16_EPI32_M64(&x[j + 0]);
+        vecX0 = _mm_shuffle_epi32(vecX, 0x00);
+        vecX1 = _mm_shuffle_epi32(vecX, 0x55);
+        vecX2 = _mm_shuffle_epi32(vecX, 0xaa);
+        vecX3 = _mm_shuffle_epi32(vecX, 0xff);
+
+        vecY0 = OP_CVTEPI16_EPI32_M64(&y[j + 0]);
+        vecY1 = OP_CVTEPI16_EPI32_M64(&y[j + 1]);
+        vecY2 = OP_CVTEPI16_EPI32_M64(&y[j + 2]);
+        vecY3 = OP_CVTEPI16_EPI32_M64(&y[j + 3]);
+
+        sum0 = _mm_mullo_epi32(vecX0, vecY0);
+        sum1 = _mm_mullo_epi32(vecX1, vecY1);
+        sum2 = _mm_mullo_epi32(vecX2, vecY2);
+        sum3 = _mm_mullo_epi32(vecX3, vecY3);
+
+        sum0 = _mm_add_epi32(sum0, sum1);
+        sum2 = _mm_add_epi32(sum2, sum3);
+        vecSum = _mm_add_epi32(vecSum, sum0);
+        vecSum = _mm_add_epi32(vecSum, sum2);
+    }
+
+    for (;j<len;j++)
+    {
+        vecX = OP_CVTEPI16_EPI32_M64(&x[j + 0]);
+        vecX0 = _mm_shuffle_epi32(vecX, 0x00);
+
+        vecY0 = OP_CVTEPI16_EPI32_M64(&y[j + 0]);
+
+        sum0 = _mm_mullo_epi32(vecX0, vecY0);
+        vecSum = _mm_add_epi32(vecSum, sum0);
+    }
+
+    initSum = _mm_loadu_si128((__m128i *)(&sum[0]));
+    initSum = _mm_add_epi32(initSum, vecSum);
+    _mm_storeu_si128((__m128i *)sum, initSum);
+}
+#endif
diff --git a/drivers/opus/celt/x86/x86_celt_map.c b/drivers/opus/celt/x86/x86_celt_map.c
new file mode 100644
index 0000000000..5146ea2b38
--- /dev/null
+++ b/drivers/opus/celt/x86/x86_celt_map.c
@@ -0,0 +1,152 @@
+/* Copyright (c) 2014, Cisco Systems, INC
+   Written by XiangMingZhu WeiZhou MinPeng YanWang
+
+   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.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+#include "opus/opus_config.h"
+
+#include "opus/celt/x86/x86cpu.h"
+#include "opus/celt/celt_lpc.h"
+#include "opus/celt/pitch.h"
+#include "opus/celt/x86/pitch_sse.h"
+
+#if defined(OPUS_HAVE_RTCD)
+
+# if defined(FIXED_POINT)
+
+#if defined(OPUS_X86_MAY_HAVE_SSE4_1) && !defined(OPUS_X86_PRESUME_SSE4_1)
+
+void (*const CELT_FIR_IMPL[OPUS_ARCHMASK + 1])(
+         const opus_val16 *x,
+         const opus_val16 *num,
+         opus_val16       *y,
+         int              N,
+         int              ord,
+         opus_val16       *mem,
+         int              arch
+) = {
+  celt_fir_c,                /* non-sse */
+  celt_fir_c,
+  celt_fir_c,
+  MAY_HAVE_SSE4_1(celt_fir), /* sse4.1  */
+  MAY_HAVE_SSE4_1(celt_fir)  /* avx  */
+};
+
+void (*const XCORR_KERNEL_IMPL[OPUS_ARCHMASK + 1])(
+         const opus_val16 *x,
+         const opus_val16 *y,
+         opus_val32       sum[4],
+         int              len
+) = {
+  xcorr_kernel_c,                /* non-sse */
+  xcorr_kernel_c,
+  xcorr_kernel_c,
+  MAY_HAVE_SSE4_1(xcorr_kernel), /* sse4.1  */
+  MAY_HAVE_SSE4_1(xcorr_kernel)  /* avx  */
+};
+
+#endif
+
+#if (defined(OPUS_X86_MAY_HAVE_SSE4_1) && !defined(OPUS_X86_PRESUME_SSE4_1)) ||  \
+	(!defined(OPUS_X86_MAY_HAVE_SSE_4_1) && defined(OPUS_X86_MAY_HAVE_SSE2) && !defined(OPUS_X86_PRESUME_SSE2))
+
+opus_val32 (*const CELT_INNER_PROD_IMPL[OPUS_ARCHMASK + 1])(
+         const opus_val16 *x,
+         const opus_val16 *y,
+         int              N
+) = {
+  celt_inner_prod_c,                /* non-sse */
+  celt_inner_prod_c,
+  MAY_HAVE_SSE2(celt_inner_prod),
+  MAY_HAVE_SSE4_1(celt_inner_prod), /* sse4.1  */
+  MAY_HAVE_SSE4_1(celt_inner_prod)  /* avx  */
+};
+
+#endif
+
+# else
+
+#if defined(OPUS_X86_MAY_HAVE_SSE) && !defined(OPUS_X86_PRESUME_SSE)
+
+void (*const XCORR_KERNEL_IMPL[OPUS_ARCHMASK + 1])(
+         const opus_val16 *x,
+         const opus_val16 *y,
+         opus_val32       sum[4],
+         int              len
+) = {
+  xcorr_kernel_c,                /* non-sse */
+  MAY_HAVE_SSE(xcorr_kernel),
+  MAY_HAVE_SSE(xcorr_kernel),
+  MAY_HAVE_SSE(xcorr_kernel),
+  MAY_HAVE_SSE(xcorr_kernel)
+};
+
+opus_val32 (*const CELT_INNER_PROD_IMPL[OPUS_ARCHMASK + 1])(
+         const opus_val16 *x,
+         const opus_val16 *y,
+         int              N
+) = {
+  celt_inner_prod_c,                /* non-sse */
+  MAY_HAVE_SSE(celt_inner_prod),
+  MAY_HAVE_SSE(celt_inner_prod),
+  MAY_HAVE_SSE(celt_inner_prod),
+  MAY_HAVE_SSE(celt_inner_prod)
+};
+
+void (*const DUAL_INNER_PROD_IMPL[OPUS_ARCHMASK + 1])(
+                    const opus_val16 *x,
+                    const opus_val16 *y01,
+                    const opus_val16 *y02,
+                    int               N,
+                    opus_val32       *xy1,
+                    opus_val32       *xy2
+) = {
+  dual_inner_prod_c,                /* non-sse */
+  MAY_HAVE_SSE(dual_inner_prod),
+  MAY_HAVE_SSE(dual_inner_prod),
+  MAY_HAVE_SSE(dual_inner_prod),
+  MAY_HAVE_SSE(dual_inner_prod)
+};
+
+void (*const COMB_FILTER_CONST_IMPL[OPUS_ARCHMASK + 1])(
+              opus_val32 *y,
+              opus_val32 *x,
+              int         T,
+              int         N,
+              opus_val16  g10,
+              opus_val16  g11,
+              opus_val16  g12
+) = {
+  comb_filter_const_c,                /* non-sse */
+  MAY_HAVE_SSE(comb_filter_const),
+  MAY_HAVE_SSE(comb_filter_const),
+  MAY_HAVE_SSE(comb_filter_const),
+  MAY_HAVE_SSE(comb_filter_const)
+};
+
+
+#endif
+
+#endif
+#endif
diff --git a/drivers/opus/celt/x86/x86cpu.c b/drivers/opus/celt/x86/x86cpu.c
new file mode 100644
index 0000000000..91b9b185c2
--- /dev/null
+++ b/drivers/opus/celt/x86/x86cpu.c
@@ -0,0 +1,154 @@
+/* Copyright (c) 2014, Cisco Systems, INC
+   Written by XiangMingZhu WeiZhou MinPeng YanWang
+
+   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.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+#include "opus/opus_config.h"
+
+#include "opus/celt/cpu_support.h"
+#include "opus/silk/macros.h"
+#include "opus/silk/main.h"
+#include "opus/celt/pitch.h"
+#include "opus/celt/x86/x86cpu.h"
+
+#if (defined(OPUS_X86_MAY_HAVE_SSE) && !defined(OPUS_X86_PRESUME_SSE)) || \
+  (defined(OPUS_X86_MAY_HAVE_SSE2) && !defined(OPUS_X86_PRESUME_SSE2)) || \
+  (defined(OPUS_X86_MAY_HAVE_SSE4_1) && !defined(OPUS_X86_PRESUME_SSE4_1)) || \
+  (defined(OPUS_X86_MAY_HAVE_AVX) && !defined(OPUS_X86_PRESUME_AVX))
+
+
+#if defined(_MSC_VER)
+
+#include <intrin.h>
+static _inline void cpuid(unsigned int CPUInfo[4], unsigned int InfoType)
+{
+	__cpuid((int*)CPUInfo, InfoType);
+}
+
+#else
+
+#if defined(CPU_INFO_BY_C)
+#include <cpuid.h>
+#endif
+
+static void cpuid(unsigned int CPUInfo[4], unsigned int InfoType)
+{
+#if defined(CPU_INFO_BY_ASM)
+#if defined(__i386__) && defined(__PIC__)
+/* %ebx is PIC register in 32-bit, so mustn't clobber it. */
+    __asm__ __volatile__ (
+        "xchg %%ebx, %1\n"
+        "cpuid\n"
+        "xchg %%ebx, %1\n":
+        "=a" (CPUInfo[0]),
+        "=r" (CPUInfo[1]),
+        "=c" (CPUInfo[2]),
+        "=d" (CPUInfo[3]) :
+        "0" (InfoType)
+    );
+#else
+    __asm__ __volatile__ (
+        "cpuid":
+        "=a" (CPUInfo[0]),
+        "=b" (CPUInfo[1]),
+        "=c" (CPUInfo[2]),
+        "=d" (CPUInfo[3]) :
+        "0" (InfoType)
+    );
+#endif
+#elif defined(CPU_INFO_BY_C)
+    __get_cpuid(InfoType, &(CPUInfo[0]), &(CPUInfo[1]), &(CPUInfo[2]), &(CPUInfo[3]));
+#endif
+}
+
+#endif
+
+typedef struct CPU_Feature{
+    /*  SIMD: 128-bit */
+    int HW_SSE;
+    int HW_SSE2;
+    int HW_SSE41;
+    /*  SIMD: 256-bit */
+    int HW_AVX;
+} CPU_Feature;
+
+static void opus_cpu_feature_check(CPU_Feature *cpu_feature)
+{
+    unsigned int info[4] = {0};
+    unsigned int nIds = 0;
+
+    cpuid(info, 0);
+    nIds = info[0];
+
+    if (nIds >= 1){
+        cpuid(info, 1);
+        cpu_feature->HW_SSE = (info[3] & (1 << 25)) != 0;
+        cpu_feature->HW_SSE2 = (info[3] & (1 << 26)) != 0;
+        cpu_feature->HW_SSE41 = (info[2] & (1 << 19)) != 0;
+        cpu_feature->HW_AVX = (info[2] & (1 << 28)) != 0;
+    }
+    else {
+        cpu_feature->HW_SSE = 0;
+        cpu_feature->HW_SSE2 = 0;
+        cpu_feature->HW_SSE41 = 0;
+        cpu_feature->HW_AVX = 0;
+    }
+}
+
+int opus_select_arch(void)
+{
+    CPU_Feature cpu_feature;
+    int arch;
+
+    opus_cpu_feature_check(&cpu_feature);
+
+    arch = 0;
+    if (!cpu_feature.HW_SSE)
+    {
+       return arch;
+    }
+    arch++;
+
+    if (!cpu_feature.HW_SSE2)
+    {
+       return arch;
+    }
+    arch++;
+
+    if (!cpu_feature.HW_SSE41)
+    {
+        return arch;
+    }
+    arch++;
+
+    if (!cpu_feature.HW_AVX)
+    {
+        return arch;
+    }
+    arch++;
+
+    return arch;
+}
+
+#endif
diff --git a/drivers/opus/celt/x86/x86cpu.h b/drivers/opus/celt/x86/x86cpu.h
new file mode 100644
index 0000000000..04fd48aac4
--- /dev/null
+++ b/drivers/opus/celt/x86/x86cpu.h
@@ -0,0 +1,93 @@
+/* Copyright (c) 2014, Cisco Systems, INC
+   Written by XiangMingZhu WeiZhou MinPeng YanWang
+
+   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.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#if !defined(X86CPU_H)
+# define X86CPU_H
+
+# if defined(OPUS_X86_MAY_HAVE_SSE)
+#  define MAY_HAVE_SSE(name) name ## _sse
+# else
+#  define MAY_HAVE_SSE(name) name ## _c
+# endif
+
+# if defined(OPUS_X86_MAY_HAVE_SSE2)
+#  define MAY_HAVE_SSE2(name) name ## _sse2
+# else
+#  define MAY_HAVE_SSE2(name) name ## _c
+# endif
+
+# if defined(OPUS_X86_MAY_HAVE_SSE4_1)
+#  define MAY_HAVE_SSE4_1(name) name ## _sse4_1
+# else
+#  define MAY_HAVE_SSE4_1(name) name ## _c
+# endif
+
+# if defined(OPUS_X86_MAY_HAVE_AVX)
+#  define MAY_HAVE_AVX(name) name ## _avx
+# else
+#  define MAY_HAVE_AVX(name) name ## _c
+# endif
+
+# if defined(OPUS_HAVE_RTCD)
+int opus_select_arch(void);
+# endif
+
+/*gcc appears to emit MOVDQA's to load the argument of an _mm_cvtepi8_epi32()
+  or _mm_cvtepi16_epi32() when optimizations are disabled, even though the
+  actual PMOVSXWD instruction takes an m32 or m64. Unlike a normal memory
+  reference, these require 16-byte alignment and load a full 16 bytes (instead
+  of 4 or 8), possibly reading out of bounds.
+
+  We can insert an explicit MOVD or MOVQ using _mm_cvtsi32_si128() or
+  _mm_loadl_epi64(), which should have the same semantics as an m32 or m64
+  reference in the PMOVSXWD instruction itself, but gcc is not smart enough to
+  optimize this out when optimizations ARE enabled.
+
+  Clang, in contrast, requires us to do this always for _mm_cvtepi8_epi32
+  (which is fair, since technically the compiler is always allowed to do the
+  dereference before invoking the function implementing the intrinsic).
+  However, it is smart enough to eliminate the extra MOVD instruction.
+  For _mm_cvtepi16_epi32, it does the right thing, though does *not* optimize out
+  the extra MOVQ if it's specified explicitly */
+
+# if defined(__clang__) || !defined(__OPTIMIZE__)
+#  define OP_CVTEPI8_EPI32_M32(x) \
+ (_mm_cvtepi8_epi32(_mm_cvtsi32_si128(*(int *)(x))))
+# else
+#  define OP_CVTEPI8_EPI32_M32(x) \
+ (_mm_cvtepi8_epi32(*(__m128i *)(x)))
+#endif
+
+# if !defined(__OPTIMIZE__)
+#  define OP_CVTEPI16_EPI32_M64(x) \
+ (_mm_cvtepi16_epi32(_mm_loadl_epi64((__m128i *)(x))))
+# else
+#  define OP_CVTEPI16_EPI32_M64(x) \
+ (_mm_cvtepi16_epi32(*(__m128i *)(x)))
+# endif
+
+#endif