shithub: opus

Download patch

ref: 39386e0b85ec0f978aa104d312604badb9047d58
parent: 530198f955e49571b3f890b4da4d933a4cd5df4e
author: Timothy B. Terriberry <[email protected]>
date: Mon Nov 18 08:30:13 EST 2013 

Adds Neon assembly for correlation/convolution

Optimizing celt_pitch_xcorr()/xcorr_kernel() which also speeds up
FIRs, IIRs and auto-correlations

Signed-off-by: Jean-Marc Valin <[email protected]>


--- a/Makefile.am
+++ b/Makefile.am
@@ -29,7 +29,14 @@
 
 if CPU_ARM
 CELT_SOURCES += $(CELT_SOURCES_ARM)
+SILK_SOURCES += $(SILK_SOURCES_ARM)
+if OPUS_ARM_EXTERNAL_ASM
+CELT_SOURCES += $(CELT_SOURCES_ARM_ASM:.s=-gnu.S)
+BUILT_SOURCES = $(CELT_SOURCES_ARM_ASM:.s=-gnu.S) \
+ $(CELT_AM_SOURCES_ARM_ASM:.s.in=.s) \
+ $(CELT_AM_SOURCES_ARM_ASM.s.in=-gnu.S)
 endif
+endif
 
 include celt_headers.mk
 include silk_headers.mk
@@ -106,11 +113,12 @@
 endif
 
 EXTRA_DIST = version.mk \
-	     opus.pc.in \
+             opus.pc.in \
              opus-uninstalled.pc.in \
              opus.m4 \
              Makefile.unix \
              tests/run_vectors.sh \
+             celt/arm/arm2gnu.pl \
              win32/VS2010/silk_float.vcxproj \
              win32/VS2010/celt.vcxproj.filters \
              win32/VS2010/opus.vcxproj \
@@ -206,3 +214,14 @@
 
 
 .PHONY: opus check-opus install-opus docs install-docs
+
+# automake doesn't do dependency tracking for asm files, that I can tell
+$(CELT_SOURCES_ARM_ASM:%.s=%-gnu.S): celt/arm/armopts-gnu.S
+$(CELT_SOURCES_ARM_ASM:%.s=%-gnu.S): $(top_srcdir)/celt/arm/arm2gnu.pl
+
+# convert ARM asm to GNU as format
+%-gnu.S: $(top_srcdir)/%.s
+	$(top_srcdir)/celt/arm/arm2gnu.pl < $< > $@
+# For autoconf-modified sources (e.g., armopts.s)
+%-gnu.S: %.s
+	$(top_srcdir)/celt/arm/arm2gnu.pl < $< > $@
--- a/celt/_kiss_fft_guts.h
+++ b/celt/_kiss_fft_guts.h
@@ -94,11 +94,11 @@
     do {(res).r = ADD32((res).r,(a).r);  (res).i = SUB32((res).i,(a).i); \
     }while(0)
 
-#if defined(ARMv4_ASM)
+#if defined(OPUS_ARM_INLINE_ASM)
 #include "arm/kiss_fft_armv4.h"
 #endif
 
-#if defined(ARMv5E_ASM)
+#if defined(OPUS_ARM_INLINE_EDSP)
 #include "arm/kiss_fft_armv5e.h"
 #endif
 
--- a/celt/arch.h
+++ b/celt/arch.h
@@ -114,9 +114,9 @@
 
 #include "fixed_generic.h"
 
-#ifdef ARMv5E_ASM
+#ifdef OPUS_ARM_INLINE_EDSP
 #include "arm/fixed_armv5e.h"
-#elif defined (ARMv4_ASM)
+#elif defined (OPUS_ARM_INLINE_ASM)
 #include "arm/fixed_armv4.h"
 #elif defined (BFIN_ASM)
 #include "fixed_bfin.h"
--- /dev/null
+++ b/celt/arm/arm2gnu.pl
@@ -1,0 +1,316 @@
+#!/usr/bin/perl
+
+my $bigend;  # little/big endian
+my $nxstack;
+
+$nxstack = 0;
+
+eval 'exec /usr/local/bin/perl -S $0 ${1+"$@"}'
+    if $running_under_some_shell;
+
+while ($ARGV[0] =~ /^-/) {
+    $_ = shift;
+  last if /^--/;
+    if (/^-n/) {
+    $nflag++;
+    next;
+    }
+    die "I don't recognize this switch: $_\\n";
+}
+$printit++ unless $nflag;
+
+$\ = "\n";      # automatically add newline on print
+$n=0;
+
+$thumb = 0;     # ARM mode by default, not Thumb.
+@proc_stack = ();
+
+LINE:
+while (<>) {
+
+    # For ADRLs we need to add a new line after the substituted one.
+    $addPadding = 0;
+
+    # First, we do not dare to touch *anything* inside double quotes, do we?
+    # Second, if you want a dollar character in the string,
+    # insert two of them -- that's how ARM C and assembler treat strings.
+    s/^([A-Za-z_]\w*)[ \t]+DCB[ \t]*\"/$1:   .ascii \"/   && do { s/\$\$/\$/g; next };
+    s/\bDCB\b[ \t]*\"/.ascii \"/                          && do { s/\$\$/\$/g; next };
+    s/^(\S+)\s+RN\s+(\S+)/$1 .req r$2/                    && do { s/\$\$/\$/g; next };
+    # If there's nothing on a line but a comment, don't try to apply any further
+    #  substitutions (this is a cheap hack to avoid mucking up the license header)
+    s/^([ \t]*);/$1@/                                     && do { s/\$\$/\$/g; next };
+    # If substituted -- leave immediately !
+
+    s/@/,:/;
+    s/;/@/;
+    while ( /@.*'/ ) {
+      s/(@.*)'/$1/g;
+    }
+    s/\{FALSE\}/0/g;
+    s/\{TRUE\}/1/g;
+    s/\{(\w\w\w\w+)\}/$1/g;
+    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/^(\s+)\[/$1IF/;
+    s/^(\s+)\|/$1ELSE/;
+    s/^(\s+)\]/$1ENDIF/;
+    s/IF *:DEF:/ .ifdef/;
+    s/IF *:LNOT: *:DEF:/ .ifndef/;
+    s/ELSE/ .else/;
+    s/ENDIF/ .endif/;
+
+    if( /\bIF\b/ ) {
+      s/\bIF\b/ .if/;
+      s/=/==/;
+    }
+    if ( $n == 2) {
+        s/\$/\\/g;
+    }
+    if ($n == 1) {
+        s/\$//g;
+        s/label//g;
+    $n = 2;
+      }
+    if ( /MACRO/ ) {
+      s/MACRO *\n/.macro/;
+      $n=1;
+    }
+    if ( /\bMEND\b/ ) {
+      s/\bMEND\b/.endm/;
+      $n=0;
+    }
+
+    # ".rdata" doesn't work in 'as' version 2.13.2, as it is ".rodata" there.
+    #
+    if ( /\bAREA\b/ ) {
+        my $align;
+        $align = "2";
+        if ( /ALIGN=(\d+)/ ) {
+            $align = $1;
+        }
+        if ( /CODE/ ) {
+            $nxstack = 1;
+        }
+        s/^(.+)CODE(.+)READONLY(.*)/    .text/;
+        s/^(.+)DATA(.+)READONLY(.*)/    .section .rdata/;
+        s/^(.+)\|\|\.data\|\|(.+)/    .data/;
+        s/^(.+)\|\|\.bss\|\|(.+)/    .bss/;
+        s/$/;   .p2align $align/;
+        # Enable NEON instructions but don't produce a binary that requires
+        # ARMv7. RVCT does not have equivalent directives, so we just do this
+        # for all CODE areas.
+        if ( /.text/ ) {
+            # Separating .arch, .fpu, etc., by semicolons does not work (gas
+            # thinks the semicolon is part of the arch name, even when there's
+            # whitespace separating them). Sadly this means our line numbers
+            # won't match the original source file (we could use the .line
+            # directive, which is documented to be obsolete, but then gdb will
+            # show the wrong line in the translated source file).
+            s/$/;   .arch armv7-a\n   .fpu neon\n   .object_arch armv4t/;
+        }
+    }
+
+    s/\|\|\.constdata\$(\d+)\|\|/.L_CONST$1/;       # ||.constdata$3||
+    s/\|\|\.bss\$(\d+)\|\|/.L_BSS$1/;               # ||.bss$2||
+    s/\|\|\.data\$(\d+)\|\|/.L_DATA$1/;             # ||.data$2||
+    s/\|\|([a-zA-Z0-9_]+)\@([a-zA-Z0-9_]+)\|\|/@ $&/;
+    s/^(\s+)\%(\s)/    .space $1/;
+
+    s/\|(.+)\.(\d+)\|/\.$1_$2/;                     # |L80.123| -> .L80_123
+    s/\bCODE32\b/.code 32/ && do {$thumb = 0};
+    s/\bCODE16\b/.code 16/ && do {$thumb = 1};
+    if (/\bPROC\b/)
+    {
+        my $prefix;
+        my $proc;
+        /^([A-Za-z_\.]\w+)\b/;
+        $proc = $1;
+        $prefix = "";
+        if ($proc)
+        {
+            $prefix = $prefix.sprintf("\t.type\t%s, %%function; ",$proc);
+            push(@proc_stack, $proc);
+            s/^[A-Za-z_\.]\w+/$&:/;
+        }
+        $prefix = $prefix."\t.thumb_func; " if ($thumb);
+        s/\bPROC\b/@ $&/;
+        $_ = $prefix.$_;
+    }
+    s/^(\s*)(S|Q|SH|U|UQ|UH)ASX\b/$1$2ADDSUBX/;
+    s/^(\s*)(S|Q|SH|U|UQ|UH)SAX\b/$1$2SUBADDX/;
+    if (/\bENDP\b/)
+    {
+        my $proc;
+        s/\bENDP\b/@ $&/;
+        $proc = pop(@proc_stack);
+        $_ = "\t.size $proc, .-$proc".$_ if ($proc);
+    }
+    s/\bSUBT\b/@ $&/;
+    s/\bDATA\b/@ $&/;   # DATA directive is deprecated -- Asm guide, p.7-25
+    s/\bKEEP\b/@ $&/;
+    s/\bEXPORTAS\b/@ $&/;
+    s/\|\|(.)+\bEQU\b/@ $&/;
+    s/\|\|([\w\$]+)\|\|/$1/;
+    s/\bENTRY\b/@ $&/;
+    s/\bASSERT\b/@ $&/;
+    s/\bGBLL\b/@ $&/;
+    s/\bGBLA\b/@ $&/;
+    s/^\W+OPT\b/@ $&/;
+    s/:OR:/|/g;
+    s/:SHL:/<</g;
+    s/:SHR:/>>/g;
+    s/:AND:/&/g;
+    s/:LAND:/&&/g;
+    s/CPSR/cpsr/;
+    s/SPSR/spsr/;
+    s/ALIGN$/.balign 4/;
+    s/ALIGN\s+([0-9x]+)$/.balign $1/;
+    s/psr_cxsf/psr_all/;
+    s/LTORG/.ltorg/;
+    s/^([A-Za-z_]\w*)[ \t]+EQU/ .set $1,/;
+    s/^([A-Za-z_]\w*)[ \t]+SETL/ .set $1,/;
+    s/^([A-Za-z_]\w*)[ \t]+SETA/ .set $1,/;
+    s/^([A-Za-z_]\w*)[ \t]+\*/ .set $1,/;
+
+    #  {PC} + 0xdeadfeed  -->  . + 0xdeadfeed
+    s/\{PC\} \+/ \. +/;
+
+    # Single hex constant on the line !
+    #
+    # >>> NOTE <<<
+    #   Double-precision floats in gcc are always mixed-endian, which means
+    #   bytes in two words are little-endian, but words are big-endian.
+    #   So, 0x0000deadfeed0000 would be stored as 0x0000dead at low address
+    #   and 0xfeed0000 at high address.
+    #
+    s/\bDCFD\b[ \t]+0x([a-fA-F0-9]{8})([a-fA-F0-9]{8})/.long 0x$1, 0x$2/;
+    # Only decimal constants on the line, no hex !
+    s/\bDCFD\b[ \t]+([0-9\.\-]+)/.double $1/;
+
+    # Single hex constant on the line !
+#    s/\bDCFS\b[ \t]+0x([a-f0-9]{8})([a-f0-9]{8})/.long 0x$1, 0x$2/;
+    # Only decimal constants on the line, no hex !
+#    s/\bDCFS\b[ \t]+([0-9\.\-]+)/.double $1/;
+    s/\bDCFS[ \t]+0x/.word 0x/;
+    s/\bDCFS\b/.float/;
+
+    s/^([A-Za-z_]\w*)[ \t]+DCD/$1 .word/;
+    s/\bDCD\b/.word/;
+    s/^([A-Za-z_]\w*)[ \t]+DCW/$1 .short/;
+    s/\bDCW\b/.short/;
+    s/^([A-Za-z_]\w*)[ \t]+DCB/$1 .byte/;
+    s/\bDCB\b/.byte/;
+    s/^([A-Za-z_]\w*)[ \t]+\%/.comm $1,/;
+    s/^[A-Za-z_\.]\w+/$&:/;
+    s/^(\d+)/$1:/;
+    s/\%(\d+)/$1b_or_f/;
+    s/\%[Bb](\d+)/$1b/;
+    s/\%[Ff](\d+)/$1f/;
+    s/\%[Ff][Tt](\d+)/$1f/;
+    s/&([\dA-Fa-f]+)/0x$1/;
+    if ( /\b2_[01]+\b/ ) {
+      s/\b2_([01]+)\b/conv$1&&&&/g;
+      while ( /[01][01][01][01]&&&&/ ) {
+        s/0000&&&&/&&&&0/g;
+        s/0001&&&&/&&&&1/g;
+        s/0010&&&&/&&&&2/g;
+        s/0011&&&&/&&&&3/g;
+        s/0100&&&&/&&&&4/g;
+        s/0101&&&&/&&&&5/g;
+        s/0110&&&&/&&&&6/g;
+        s/0111&&&&/&&&&7/g;
+        s/1000&&&&/&&&&8/g;
+        s/1001&&&&/&&&&9/g;
+        s/1010&&&&/&&&&A/g;
+        s/1011&&&&/&&&&B/g;
+        s/1100&&&&/&&&&C/g;
+        s/1101&&&&/&&&&D/g;
+        s/1110&&&&/&&&&E/g;
+        s/1111&&&&/&&&&F/g;
+      }
+      s/000&&&&/&&&&0/g;
+      s/001&&&&/&&&&1/g;
+      s/010&&&&/&&&&2/g;
+      s/011&&&&/&&&&3/g;
+      s/100&&&&/&&&&4/g;
+      s/101&&&&/&&&&5/g;
+      s/110&&&&/&&&&6/g;
+      s/111&&&&/&&&&7/g;
+      s/00&&&&/&&&&0/g;
+      s/01&&&&/&&&&1/g;
+      s/10&&&&/&&&&2/g;
+      s/11&&&&/&&&&3/g;
+      s/0&&&&/&&&&0/g;
+      s/1&&&&/&&&&1/g;
+      s/conv&&&&/0x/g;
+    }
+
+    if ( /commandline/)
+    {
+        if( /-bigend/)
+        {
+            $bigend=1;
+        }
+    }
+
+    if ( /\bDCDU\b/ )
+    {
+        my $cmd=$_;
+        my $value;
+        my $prefix;
+        my $w1;
+        my $w2;
+        my $w3;
+        my $w4;
+
+        s/\s+DCDU\b/@ $&/;
+
+        $cmd =~ /\bDCDU\b\s+0x(\d+)/;
+        $value = $1;
+        $value =~ /(\w\w)(\w\w)(\w\w)(\w\w)/;
+        $w1 = $1;
+        $w2 = $2;
+        $w3 = $3;
+        $w4 = $4;
+
+        if( $bigend ne "")
+        {
+            # big endian
+            $prefix = "\t.byte\t0x".$w1.";".
+                      "\t.byte\t0x".$w2.";".
+                      "\t.byte\t0x".$w3.";".
+                      "\t.byte\t0x".$w4."; ";
+        }
+        else
+        {
+            # little endian
+            $prefix = "\t.byte\t0x".$w4.";".
+                      "\t.byte\t0x".$w3.";".
+                      "\t.byte\t0x".$w2.";".
+                      "\t.byte\t0x".$w1."; ";
+        }
+        $_=$prefix.$_;
+    }
+
+    if ( /\badrl\b/i )
+    {
+        s/\badrl\s+(\w+)\s*,\s*(\w+)/ldr $1,=$2/i;
+        $addPadding = 1;
+    }
+    s/\bEND\b/@ END/;
+} continue {
+    printf ("%s", $_) if $printit;
+    if ($addPadding != 0)
+    {
+        printf ("   mov r0,r0\n");
+        $addPadding = 0;
+    }
+}
+#If we had a code section, mark that this object doesn't need an executable
+# stack.
+if ($nxstack) {
+    printf ("    .section\t.note.GNU-stack,\"\",\%\%progbits\n");
+}
--- /dev/null
+++ b/celt/arm/arm_celt_map.c
@@ -1,0 +1,49 @@
+/* Copyright (c) 2010 Xiph.Org Foundation
+ * Copyright (c) 2013 Parrot */
+/*
+   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 HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#include "pitch.h"
+
+#if defined(OPUS_HAVE_RTCD)
+
+# 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 */
+  MAY_HAVE_EDSP(celt_pitch_xcorr),  /* EDSP */
+  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
+
+#endif
--- a/celt/arm/armcpu.c
+++ b/celt/arm/armcpu.c
@@ -55,7 +55,7 @@
   /* MSVC has no OPUS_INLINE __asm support for ARM, but it does let you __emit
    * instructions via their assembled hex code.
    * All of these instructions should be essentially nops. */
-# if defined(ARMv5E_ASM)
+# if defined(OPUS_ARM_MAY_HAVE_EDSP)
   __try{
     /*PLD [r13]*/
     __emit(0xF5DDF000);
@@ -64,7 +64,7 @@
   __except(GetExceptionCode()==EXCEPTION_ILLEGAL_INSTRUCTION){
     /*Ignore exception.*/
   }
-#  if defined(ARMv6E_ASM)
+#  if defined(OPUS_ARM_MAY_HAVE_MEDIA)
   __try{
     /*SHADD8 r3,r3,r3*/
     __emit(0xE6333F93);
@@ -73,7 +73,7 @@
   __except(GetExceptionCode()==EXCEPTION_ILLEGAL_INSTRUCTION){
     /*Ignore exception.*/
   }
-#   if defined(ARM_HAVE_NEON)
+#   if defined(OPUS_ARM_MAY_HAVE_NEON)
   __try{
     /*VORR q0,q0,q0*/
     __emit(0xF2200150);
@@ -107,19 +107,26 @@
 
     while(fgets(buf, 512, cpuinfo) != NULL)
     {
+# if defined(OPUS_ARM_MAY_HAVE_EDSP) || defined(OPUS_ARM_MAY_HAVE_NEON)
       /* Search for edsp and neon flag */
       if(memcmp(buf, "Features", 8) == 0)
       {
         char *p;
+#  if defined(OPUS_ARM_MAY_HAVE_EDSP)
         p = strstr(buf, " edsp");
         if(p != NULL && (p[5] == ' ' || p[5] == '\n'))
           flags |= OPUS_CPU_ARM_EDSP;
+#  endif
 
+#  if defined(OPUS_ARM_MAY_HAVE_NEON)
         p = strstr(buf, " neon");
         if(p != NULL && (p[5] == ' ' || p[5] == '\n'))
           flags |= OPUS_CPU_ARM_NEON;
+#  endif
       }
+# endif
 
+# if defined(OPUS_ARM_MAY_HAVE_MEDIA)
       /* Search for media capabilities (>= ARMv6) */
       if(memcmp(buf, "CPU architecture:", 17) == 0)
       {
@@ -129,6 +136,7 @@
         if(version >= 6)
           flags |= OPUS_CPU_ARM_MEDIA;
       }
+# endif
     }
 
     fclose(cpuinfo);
--- a/celt/arm/armcpu.h
+++ b/celt/arm/armcpu.h
@@ -25,11 +25,47 @@
    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
 
-/* Original code from libtheora modified to suit to Opus */
+#if !defined(ARMCPU_H)
+# define ARMCPU_H
 
-#ifndef ARMCPU_H
-#define ARMCPU_H
+# if defined(OPUS_ARM_MAY_HAVE_EDSP)
+#  define MAY_HAVE_EDSP(name) name ## _edsp
+# else
+#  define MAY_HAVE_EDSP(name) name ## _c
+# endif
 
+# if defined(OPUS_ARM_MAY_HAVE_MEDIA)
+#  define MAY_HAVE_MEDIA(name) name ## _media
+# else
+#  define MAY_HAVE_MEDIA(name) MAY_HAVE_EDSP(name)
+# endif
+
+# if defined(OPUS_ARM_MAY_HAVE_NEON)
+#  define MAY_HAVE_NEON(name) name ## _neon
+# else
+#  define MAY_HAVE_NEON(name) MAY_HAVE_MEDIA(name)
+# endif
+
+# if defined(OPUS_ARM_PRESUME_EDSP)
+#  define PRESUME_EDSP(name) name ## _edsp
+# else
+#  define PRESUME_EDSP(name) name ## _c
+# endif
+
+# if defined(OPUS_ARM_PRESUME_MEDIA)
+#  define PRESUME_MEDIA(name) name ## _media
+# else
+#  define PRESUME_MEDIA(name) PRESUME_EDSP(name)
+# endif
+
+# if defined(OPUS_ARM_PRESUME_NEON)
+#  define PRESUME_NEON(name) name ## _neon
+# else
+#  define PRESUME_NEON(name) PRESUME_MEDIA(name)
+# endif
+
+# if defined(OPUS_HAVE_RTCD)
 int opus_select_arch(void);
+# endif
 
 #endif
--- /dev/null
+++ b/celt/arm/armopts.s.in
@@ -1,0 +1,37 @@
+/* Copyright (C) 2013 Mozilla Corporation */
+/*
+   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.
+*/
+
+; Set the following to 1 if we have EDSP instructions
+;  (LDRD/STRD, etc., ARMv5E and later).
+OPUS_ARM_MAY_HAVE_EDSP  * @OPUS_ARM_MAY_HAVE_EDSP@
+
+; Set the following to 1 if we have ARMv6 media instructions.
+OPUS_ARM_MAY_HAVE_MEDIA * @OPUS_ARM_MAY_HAVE_MEDIA@
+
+; Set the following to 1 if we have NEON (some ARMv7)
+OPUS_ARM_MAY_HAVE_NEON  * @OPUS_ARM_MAY_HAVE_NEON@
+
+END
--- /dev/null
+++ b/celt/arm/celt_pitch_xcorr_arm.s
@@ -1,0 +1,598 @@
+; 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.
+
+  AREA  |.text|, CODE, READONLY
+
+  GET    celt/arm/armopts.s
+
+IF OPUS_ARM_MAY_HAVE_EDSP
+  EXPORT celt_pitch_xcorr_edsp
+ENDIF
+
+IF OPUS_ARM_MAY_HAVE_NEON
+  EXPORT 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
+;  ; 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
+;  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
+;  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
+;  ; Reduce the sum to a single value.
+;  VADD.S32     d0, d0, d1
+;  VPADDL.S32   d0, d0
+;celt_pitch_xcorr_neon_process_remaining4_done
+;  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}
+;  ENDP
+
+xcorr_kernel_neon PROC
+  ; input:
+  ; r0 = opus_val16 *x
+  ; r1 = opus_val16 *y
+  ; r2 = int        len
+  ; q0 = opus_val32 sum (sum[3] | sum[2] | sum[1] | sum[0])
+
+  ; output:
+  ; q0 = sum
+
+  ; internal usage:
+  ; r3 = j
+  ; d2 = x_3|x_2|x_1|x_0  d3 = y_3|y_2|y_1|y_0
+  ; d4 = y_7|y_6|y_5|y_4  d5 = y_4|y_3|y_2|y_1
+  ; d6 = y_5|y_4|y_3|y_2  d7 = y_6|y_5|y_4|y_3
+  ; We will build d5, d6 and d7 vector from d3 and d4
+
+
+  VLD1.16   {d3}, [r1]!      ; Load y[3] downto y[0] to d3 lane (yy0)
+  SUB       r3, r2, #1
+  MOVS      r3, r3, lsr #2   ; j=(len-1)>>2
+  BEQ       xcorr_kernel_neon_process4_done
+
+  ; Process 4 x samples at a time
+  ; For this, we will need 4 y vectors
+xcorr_kernel_neon_process4
+  SUBS      r3, r3, #1       ; j--
+  VLD1.16   d4, [r1]!        ; Load y[7] downto y[4] to d4 lane
+  VLD1.16   d2, [r0]!        ; Load x[3] downto x[0] to d2 lane
+  VEXT.16   d5, d3, d4, #1   ; Build y[4] downto y[1] vector (yy1)
+  VEXT.16   d6, d3, d4, #2   ; Build y[5] downto y[2] vector (yy2)
+  VEXT.16   d7, d3, d4, #3   ; Build y[6] downto y[3] vector (yy3)
+
+  VMLAL.S16 q0, d3, d2[0]    ; MAC16_16(sum, x[0], yy0)
+  VMLAL.S16 q0, d5, d2[1]    ; MAC16_16(sum, x[1], yy1)
+  VMLAL.S16 q0, d6, d2[2]    ; MAC16_16(sum, x[2], yy2)
+  VMLAL.S16 q0, d7, d2[3]    ; MAC16_16(sum, x[3], yy3)
+
+  VMOV.S16  d3, d4           ; Next y vector should be in d3 (yy0)
+
+  BNE xcorr_kernel_neon_process4
+
+xcorr_kernel_neon_process4_done
+  ;Process len-1 to len
+  VLD1.16   {d2[]}, [r0]!    ; Load *x and duplicate to d2 lane
+
+  SUB       r3, r2, #1
+  ANDS      r3, r3, #3       ; j=(len-1)&3
+  VMLAL.S16 q0, d3, d2       ; MAC16_16(sum, *x, yy0)
+  BEQ xcorr_kernel_neon_done
+
+xcorr_kernel_neon_process_remaining
+  SUBS      r3, r3, #1       ; j--
+  VLD1.16   {d4[]}, [r1]!    ; Load y value and duplicate to d4 lane
+  VLD1.16   {d2[]}, [r0]!    ; Load *x and duplicate to d2 lane
+  VEXT.16   d3, d3, d4, #1   ; Build y vector from previous and d4
+  VMLAL.S16 q0, d3, d2       ; MAC16_16(sum, *x, yy0)
+  BNE xcorr_kernel_neon_process_remaining
+
+xcorr_kernel_neon_done
+  MOV       pc, lr
+  ENDP
+
+celt_pitch_xcorr_neon PROC
+  ; input:
+  ; r0 = opus_val16 *_x
+  ; r1 = opus_val16 *_y
+  ; r2 = opus_val32 *xcorr
+  ; r3 = int        len
+
+  ; output:
+  ; r0 = maxcorr
+
+  STMFD     sp!, {r4-r9, lr}
+
+  LDR       r4, [sp, #28]        ; r4 = int max_pitch
+  MOV       r5, r0               ; r5 = _x
+  MOV       r6, r1               ; r6 = _y
+  MOV       r7, r2               ; r7 = xcorr
+  MOV       r2, r3               ; r2 = len
+
+  VMOV.S32  d16, #1              ; d16 = {1, 1}  (not used by xcorr_kernel_neon)
+  MOV       r8, #0               ; r8 = i = 0
+  CMP       r4, #3               ; max_pitch-3 <= 0  ---> pitch_xcorr_neon_process4_done
+  BLE       celt_pitch_xcorr_neon_process4_done
+
+  SUB       r9, r4, #3           ; r9 = max_pitch-3
+
+celt_pitch_xcorr_neon_process4
+  MOV       r0, r5               ; r0 = _x
+  ADD       r1, r6 ,r8, LSL #1   ; r1 = _y + i
+  VMOV.I32  q0, #0               ; q0 = opus_val32 sum[4] = {0, 0, 0, 0}
+
+                                 ; xcorr_kernel_neon don't touch r2 (len)
+                                 ; So we don't store it
+  BL xcorr_kernel_neon           ; xcorr_kernel_neon(_x, _y+i, sum, len)
+
+  VST1.32   {q0}, [r7]!          ; Store sum to xcorr
+  VPMAX.S32 d0, d0, d1           ; d0 = max(sum[3], sum[2]) | max(sum[1], sum[0])
+  ADD       r8, r8, #4           ; i+=4
+  VPMAX.S32 d0, d0, d0           ; d0 = max(sum[3], sum[2], sum[1], sum[0])
+  CMP       r8, r9               ; i < max_pitch-3 ----> pitch_xcorr_neon_process4
+  VMAX.S32  d16, d16, d0         ; d16 = maxcorr = max(maxcorr, sum)
+
+  BLT       celt_pitch_xcorr_neon_process4
+
+celt_pitch_xcorr_neon_process4_done
+  CMP       r8, r4;
+  BGE       celt_pitch_xcorr_neon_done
+
+celt_pitch_xcorr_neon_process_remaining
+  MOV       r0, r5               ; r0 = _x
+  ADD       r1, r6, r8, LSL #1   ; r1 = _y + i
+  VMOV.I32  q0, #0
+  MOVS      r3, r2, LSR #2       ; r3 = j = len
+  BEQ       inner_loop_neon_process4_done
+
+inner_loop_neon_process4
+  VLD1.16   {d2}, [r0]!          ; Load x
+  VLD1.16   {d3}, [r1]!          ; Load y
+  SUBS      r3, r3, #1
+  VMLAL.S16 q0, d2, d3
+  BNE       inner_loop_neon_process4
+
+  VPADD.S32 d0, d0, d1          ; Reduce sum
+  VPADD.S32 d0, d0, d0
+
+inner_loop_neon_process4_done
+  ANDS      r3, r2, #3
+  BEQ       inner_loop_neon_done
+
+inner_loop_neon_process_remaining
+  VLD1.16   {d2[]}, [r0]!
+  VLD1.16   {d3[]}, [r1]!
+  SUBS      r3, r3, #1
+  VMLAL.S16 q0, d2, d3
+  BNE       inner_loop_neon_process_remaining
+
+inner_loop_neon_done
+  VST1.32   {d0[0]}, [r7]!
+  VMAX.S32  d16, d16, d0
+
+  ADD       r8, r8, #1
+  CMP       r8, r4
+  BCC       celt_pitch_xcorr_neon_process_remaining
+
+celt_pitch_xcorr_neon_done
+  VMOV      d0, d16
+  VMOV.32   r0, d0[0]
+  LDMFD     sp!, {r4-r9, pc}
+  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
+  ; input:
+  ;   r3      = int         len
+  ;   r4      = opus_val16 *_x
+  ;   r5      = opus_val16 *_y
+  ;   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}
+  SUBS         r2, r3, #4         ; j = len-4
+  LDRD         r10, r11, [r5], #8 ; Load y[0...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)
+  LDRGTH       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)
+  LDRGTH       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)
+  LDRGTH       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}
+  ENDP
+
+celt_pitch_xcorr_edsp PROC
+  ; input:
+  ;   r0  = opus_val16 *_x
+  ;   r1  = opus_val16 *_y
+  ;   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
+  ; maxcorr = 1
+  MOV          r0, #1
+  ; if (max_pitch < 4) goto celt_pitch_xcorr_edsp_process4_done
+  SUBS         r1, r1, #4
+  BLT celt_pitch_xcorr_edsp_process4_done
+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  ; 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
+  STRD         r6, r7, [r2], #8
+  MOVLT        r0, r7
+  CMP          r0, r8
+  STRD         r8, r9, [r2], #8
+  MOVLT        r0, r8
+  CMP          r0, r9
+  MOVLT        r0, r9
+  SUBS         r1, r1, #4
+  BGE celt_pitch_xcorr_edsp_process4
+celt_pitch_xcorr_edsp_process4_done
+  ADDS         r1, r1, #4
+  BLE celt_pitch_xcorr_edsp_done
+; Now compute each remaining sum one at a time.
+celt_pitch_xcorr_edsp_process_remaining
+  SUBS         r12, r3, #4
+  ; r14 = sum = 0
+  MOV          r14, #0
+  BLT celt_pitch_xcorr_edsp_process_remaining_loop_done
+  LDRD         r6, r7, [r4], #8
+  LDRD         r8, r9, [r5], #8
+  ; Stall
+celt_pitch_xcorr_edsp_process_remaining_loop4
+  SMLABB       r14, r6, r8, r14     ; sum = MAC16_16(sum, x_0, y_0)
+  SUBS         r12, r12, #4         ; j--
+  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_process_remaining_loop4
+celt_pitch_xcorr_edsp_process_remaining_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
+  LDRGEH       r6, [r4], #2
+  LDRGEH       r8, [r5], #2
+  ; Restore _x
+  SUB          r4, r4, r3, LSL #1
+  ; Stall
+  SMLABBGE     r14, r6, r8, r14     ; sum = MAC16_16(sum, *x, *y)
+  ; Restore and advance _y
+  SUB          r5, r5, r3, LSL #1
+  ; maxcorr = max(maxcorr, sum)
+  ; Stall
+  CMP          r0, r14
+  ADD          r5, r5, #2
+  MOVLT        r0, r14
+  SUBS         r1, r1, #1
+  ; xcorr[i] = sum
+  STR          r14, [r2], #4
+  BGT celt_pitch_xcorr_edsp_process_remaining
+celt_pitch_xcorr_edsp_done
+  LDMFD        sp!, {r4-r11, pc}
+  ENDP
+
+ENDIF
+
+END
--- /dev/null
+++ b/celt/arm/pitch_arm.h
@@ -1,0 +1,57 @@
+/* Copyright (c) 2010 Xiph.Org Foundation
+ * Copyright (c) 2013 Parrot */
+/*
+   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(PITCH_ARM_H)
+# define PITCH_ARM_H
+
+# include "armcpu.h"
+
+# 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,
+    opus_val32 *xcorr, int len, int max_pitch);
+#  endif
+
+#  if defined(OPUS_ARM_MAY_HAVE_MEDIA)
+#   define celt_pitch_xcorr_media MAY_HAVE_EDSP(celt_pitch_xcorr)
+#  endif
+
+#  if defined(OPUS_ARM_MAY_HAVE_EDSP)
+opus_val32 celt_pitch_xcorr_edsp(const opus_val16 *_x, const opus_val16 *_y,
+    opus_val32 *xcorr, int len, int max_pitch);
+#  endif
+
+#  if !defined(OPUS_HAVE_RTCD)
+#   define OVERRIDE_PITCH_XCORR (1)
+#   define celt_pitch_xcorr(_x, _y, xcorr, len, max_pitch, arch) \
+  ((void)(arch),PRESUME_NEON(celt_pitch_xcorr)(_x, _y, xcorr, len, max_pitch))
+#  endif
+
+# endif
+
+#endif
--- a/celt/celt.h
+++ b/celt/celt.h
@@ -122,7 +122,8 @@
 
 int celt_encode_with_ec(OpusCustomEncoder * OPUS_RESTRICT st, const opus_val16 * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes, ec_enc *enc);
 
-int celt_encoder_init(CELTEncoder *st, opus_int32 sampling_rate, int channels);
+int celt_encoder_init(CELTEncoder *st, opus_int32 sampling_rate, int channels,
+                      int arch);
 
 
 
--- a/celt/celt_decoder.c
+++ b/celt/celt_decoder.c
@@ -447,10 +447,11 @@
       {
          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);
+         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);
+               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;
       } else {
@@ -481,7 +482,8 @@
             opus_val32 ac[LPC_ORDER+1];
             /* Compute LPC coefficients for the last MAX_PERIOD samples before
                the first loss so we can work in the excitation-filter domain. */
-            _celt_autocorr(exc, ac, window, overlap, LPC_ORDER, MAX_PERIOD);
+            _celt_autocorr(exc, ac, window, overlap,
+                   LPC_ORDER, MAX_PERIOD, st->arch);
             /* Add a noise floor of -40 dB. */
 #ifdef FIXED_POINT
             ac[0] += SHR32(ac[0],13);
--- a/celt/celt_encoder.c
+++ b/celt/celt_encoder.c
@@ -161,18 +161,9 @@
 }
 #endif /* CUSTOM_MODES */
 
-int celt_encoder_init(CELTEncoder *st, opus_int32 sampling_rate, int channels)
+static int opus_custom_encoder_init_arch(CELTEncoder *st, const CELTMode *mode,
+                                         int channels, int arch)
 {
-   int ret;
-   ret = opus_custom_encoder_init(st, opus_custom_mode_create(48000, 960, NULL), channels);
-   if (ret != OPUS_OK)
-      return ret;
-   st->upsample = resampling_factor(sampling_rate);
-   return OPUS_OK;
-}
-
-OPUS_CUSTOM_NOSTATIC int opus_custom_encoder_init(CELTEncoder *st, const CELTMode *mode, int channels)
-{
    if (channels < 0 || channels > 2)
       return OPUS_BAD_ARG;
 
@@ -190,7 +181,7 @@
    st->end = st->mode->effEBands;
    st->signalling = 1;
 
-   st->arch = opus_select_arch();
+   st->arch = arch;
 
    st->constrained_vbr = 1;
    st->clip = 1;
@@ -206,6 +197,23 @@
    return OPUS_OK;
 }
 
+OPUS_CUSTOM_NOSTATIC int opus_custom_encoder_init(CELTEncoder *st, const CELTMode *mode, int channels)
+{
+   return opus_custom_encoder_init_arch(st, mode, channels, opus_select_arch());
+}
+
+int celt_encoder_init(CELTEncoder *st, opus_int32 sampling_rate, int channels,
+                      int arch)
+{
+   int ret;
+   ret = opus_custom_encoder_init_arch(st,
+           opus_custom_mode_create(48000, 960, NULL), channels, arch);
+   if (ret != OPUS_OK)
+      return ret;
+   st->upsample = resampling_factor(sampling_rate);
+   return OPUS_OK;
+}
+
 #ifdef CUSTOM_MODES
 void opus_custom_encoder_destroy(CELTEncoder *st)
 {
@@ -1023,11 +1031,12 @@
       VARDECL(opus_val16, pitch_buf);
       ALLOC(pitch_buf, (COMBFILTER_MAXPERIOD+N)>>1, opus_val16);
 
-      pitch_downsample(pre, pitch_buf, COMBFILTER_MAXPERIOD+N, CC);
+      pitch_downsample(pre, pitch_buf, COMBFILTER_MAXPERIOD+N, CC, st->arch);
       /* Don't search for the fir last 1.5 octave of the range because
          there's too many false-positives due to short-term correlation */
       pitch_search(pitch_buf+(COMBFILTER_MAXPERIOD>>1), pitch_buf, N,
-            COMBFILTER_MAXPERIOD-3*COMBFILTER_MINPERIOD, &pitch_index);
+            COMBFILTER_MAXPERIOD-3*COMBFILTER_MINPERIOD, &pitch_index,
+            st->arch);
       pitch_index = COMBFILTER_MAXPERIOD-pitch_index;
 
       gain1 = remove_doubling(pitch_buf, COMBFILTER_MAXPERIOD, COMBFILTER_MINPERIOD,
--- a/celt/celt_lpc.c
+++ b/celt/celt_lpc.c
@@ -226,7 +226,8 @@
                    const opus_val16       *window,
                    int          overlap,
                    int          lag,
-                   int          n
+                   int          n,
+                   int          arch
                   )
 {
    opus_val32 d;
@@ -275,7 +276,7 @@
          shift = 0;
    }
 #endif
-   celt_pitch_xcorr(xptr, xptr, ac, fastN, lag+1);
+   celt_pitch_xcorr(xptr, xptr, ac, fastN, lag+1, arch);
    for (k=0;k<=lag;k++)
    {
       for (i = k+fastN, d = 0; i < n; i++)
--- a/celt/celt_lpc.h
+++ b/celt/celt_lpc.h
@@ -48,6 +48,7 @@
          int ord,
          opus_val16 *mem);
 
-int _celt_autocorr(const opus_val16 *x, opus_val32 *ac, const opus_val16 *window, int overlap, int lag, int n);
+int _celt_autocorr(const opus_val16 *x, opus_val32 *ac,
+         const opus_val16 *window, int overlap, int lag, int n, int arch);
 
 #endif /* PLC_H */
--- a/celt/cpu_support.h
+++ b/celt/cpu_support.h
@@ -31,7 +31,7 @@
 #include "opus_types.h"
 #include "opus_defines.h"
 
-#if defined(OPUS_HAVE_RTCD) && defined(ARMv4_ASM)
+#if defined(OPUS_HAVE_RTCD) && defined(OPUS_ARM_ASM)
 #include "arm/armcpu.h"
 
 /* We currently support 4 ARM variants:
--- a/celt/pitch.c
+++ b/celt/pitch.c
@@ -145,7 +145,7 @@
 
 
 void pitch_downsample(celt_sig * OPUS_RESTRICT x[], opus_val16 * OPUS_RESTRICT x_lp,
-      int len, int C)
+      int len, int C, int arch)
 {
    int i;
    opus_val32 ac[5];
@@ -180,7 +180,7 @@
    }
 
    _celt_autocorr(x_lp, ac, NULL, 0,
-                  4, len>>1);
+                  4, len>>1, arch);
 
    /* Noise floor -40 dB */
 #ifdef FIXED_POINT
@@ -250,7 +250,7 @@
 #else
 void
 #endif
-celt_pitch_xcorr(const opus_val16 *_x, const opus_val16 *_y, opus_val32 *xcorr, int len, int max_pitch)
+celt_pitch_xcorr_c(const opus_val16 *_x, const opus_val16 *_y, opus_val32 *xcorr, int len, int max_pitch)
 {
    int i,j;
 #ifdef FIXED_POINT
@@ -289,7 +289,7 @@
 
 #endif
 void pitch_search(const opus_val16 * OPUS_RESTRICT x_lp, opus_val16 * OPUS_RESTRICT y,
-                  int len, int max_pitch, int *pitch)
+                  int len, int max_pitch, int *pitch, int arch)
 {
    int i, j;
    int lag;
@@ -342,7 +342,7 @@
 #ifdef FIXED_POINT
    maxcorr =
 #endif
-   celt_pitch_xcorr(x_lp4, y_lp4, xcorr, len>>2, max_pitch>>2);
+   celt_pitch_xcorr(x_lp4, y_lp4, xcorr, len>>2, max_pitch>>2, arch);
 
    find_best_pitch(xcorr, y_lp4, len>>2, max_pitch>>2, best_pitch
 #ifdef FIXED_POINT
--- a/celt/pitch.h
+++ b/celt/pitch.h
@@ -35,16 +35,21 @@
 #define PITCH_H
 
 #include "modes.h"
+#include "cpu_support.h"
 
 #if defined(__SSE__) && !defined(FIXED_POINT)
 #include "x86/pitch_sse.h"
 #endif
 
+#if defined(OPUS_ARM_ASM) && defined(FIXED_POINT)
+# include "arm/pitch_arm.h"
+#endif
+
 void pitch_downsample(celt_sig * OPUS_RESTRICT x[], opus_val16 * OPUS_RESTRICT x_lp,
-      int len, int C);
+      int len, int C, int arch);
 
 void pitch_search(const opus_val16 * OPUS_RESTRICT x_lp, opus_val16 * OPUS_RESTRICT y,
-                  int len, int max_pitch, int *pitch);
+                  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);
@@ -140,6 +145,52 @@
 #else
 void
 #endif
-celt_pitch_xcorr(const opus_val16 *_x, const opus_val16 *_y, opus_val32 *xcorr, int len, int max_pitch);
+celt_pitch_xcorr_c(const opus_val16 *_x, const opus_val16 *_y,
+      opus_val32 *xcorr, int len, int max_pitch);
+
+#if !defined(OVERRIDE_PITCH_XCORR)
+/*Is run-time CPU detection enabled on this platform?*/
+# if defined(OPUS_HAVE_RTCD)
+extern
+#  if defined(FIXED_POINT)
+opus_val32
+#  else
+void
+#  endif
+(*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *,
+      const opus_val16 *, opus_val32 *, int, int);
+
+#  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))
+# endif
+#else
+
+/*static inline opus_val32 real_celt_pitch_xcorr(const opus_val16 *_x,
+ const opus_val16 *_y,opus_val32 *xcorr,int len,int max_pitch,int arch){
+  opus_val32 *xcorr_tmp;
+  opus_val32  ret_tmp;
+  opus_val32  ret;
+  int         i;
+  xcorr_tmp=(opus_val32 *)malloc(max_pitch*sizeof(*xcorr));
+  ret_tmp=celt_pitch_xcorr_c(_x,_y,xcorr_tmp,len,max_pitch);
+  ret=celt_pitch_xcorr(_x,_y,xcorr,len,max_pitch,arch);
+  for(i=0;i<max_pitch;i++)if(xcorr[i]!=xcorr_tmp[i]){
+    fprintf(stderr,"xcorr[%i] (0x%08X) != xcorr_tmp[%i] (0x%08X)\n",
+     i,xcorr[i],i,xcorr_tmp[i]);
+  }
+  if(ret!=ret_tmp){
+    fprintf(stderr,"ret (0x%08X) != ret_tmp (0x%08X)\n",ret,ret_tmp);
+  }
+  return ret_tmp;
+}
+
+#undef celt_pitch_xcorr
+#define celt_pitch_xcorr real_celt_pitch_xcorr*/
+
+#endif
 
 #endif
--- a/celt_headers.mk
+++ b/celt_headers.mk
@@ -35,4 +35,5 @@
 celt/arm/fixed_armv5e.h \
 celt/arm/kiss_fft_armv4.h \
 celt/arm/kiss_fft_armv5e.h \
+celt/arm/pitch_arm.h \
 celt/x86/pitch_sse.h
--- a/celt_sources.mk
+++ b/celt_sources.mk
@@ -18,4 +18,11 @@
 celt/vq.c
 
 CELT_SOURCES_ARM = \
-celt/arm/armcpu.c
+celt/arm/armcpu.c \
+celt/arm/arm_celt_map.c
+
+CELT_SOURCES_ARM_ASM = \
+celt/arm/celt_pitch_xcorr_arm.s
+
+CELT_AM_SOURCES_ARM_ASM = \
+celt/arm/armopts.s.in
--- a/configure.ac
+++ b/configure.ac
@@ -42,6 +42,8 @@
 AC_C_CONST
 AC_C_INLINE
 
+AM_PROG_AS
+
 AC_DEFINE([OPUS_BUILD], [], [This is a build of OPUS])
 
 #Use a hacked up version of autoconf's AC_C_RESTRICT because it's not
@@ -54,13 +56,13 @@
    for ac_kw in __restrict __restrict__ _Restrict restrict; do
      AC_COMPILE_IFELSE([AC_LANG_PROGRAM(
       [[typedef int * int_ptr;
-	int foo (int_ptr $ac_kw ip, int * $ac_kw baz[]) {
-	return ip[0];
+        int foo (int_ptr $ac_kw ip, int * $ac_kw baz[]) {
+        return ip[0];
        }]],
       [[int s[1];
-	int * $ac_kw t = s;
-	t[0] = 0;
-	return foo(t, (void *)0)]])],
+        int * $ac_kw t = s;
+        t[0] = 0;
+        return foo(t, (void *)0)]])],
       [ac_cv_c_restrict=$ac_kw])
      test "$ac_cv_c_restrict" != no && break
    done
@@ -165,7 +167,7 @@
 #i[[3456]]86 | x86_64 | powerpc64 | powerpc32 | ia64)
 #  has_float_approx=yes
 #  ;;
-#esac    
+#esac
 
 AC_ARG_ENABLE([float-approx],
     [AS_HELP_STRING([--enable-float-approx], [enable fast approximations for floating point])],
@@ -183,55 +185,167 @@
     [AS_HELP_STRING([--disable-asm], [Disable assembly optimizations])],,
     [enable_asm=yes])
 
+AC_ARG_ENABLE([rtcd],
+    [AS_HELP_STRING([--disable-rtcd], [Disable run-time CPU capabilities detection])],,
+    [enable_rtcd=yes])
+
 rtcd_support=no
 cpu_arm=no
 
-AS_IF([test "$enable_asm" = "yes"],[
-    asm_optimization="no asm for your platform, please send patches"
+AS_IF([test x"${enable_asm}" = x"yes"],[
+    inline_optimization="No ASM for your platform, please send patches"
     case $host_cpu in
-    arm*)
-        cpu_arm=yes
-        AS_GCC_INLINE_ASSEMBLY([asm_optimization="ARM"],
-            [asm_optimization="disabled"])
-        if test "$asm_optimization" = "ARM" ; then
-            rtcd_support=yes
-            AC_DEFINE([ARMv4_ASM], 1, [Use generic ARMv4 asm optimizations])
-            AS_ASM_ARM_EDSP([ARMv5E_ASM=1],[ARMv5E_ASM=0])
-            if test "$ARMv5E_ASM" = "1" ; then
-                AC_DEFINE([ARMv5E_ASM], [1], [Use ARMv5E asm optimizations])
-                asm_optimization="$asm_optimization (EDSP)"
-            fi
-            AS_ASM_ARM_MEDIA([ARMv6_ASM=1],[ARMv6_ASM=0])
-            if test "$ARMv6_ASM" = "1" ; then
-                AC_DEFINE([ARMv6_ASM], [1], [Use ARMv6 asm optimizations])
-                asm_optimization="$asm_optimization (Media)"
-            fi
-            AS_ASM_ARM_NEON([ARM_HAVE_NEON=1],[ARM_HAVE_NEON=0])
-            if test "$ARM_HAVE_NEON" = "1" ; then
-              AC_DEFINE([ARM_HAVE_NEON], 1, [Use ARM NEON optimizations])
-              asm_optimization="$asm_optimization (NEON)"
-            fi
-        fi
+      arm*)
+        dnl Currently we only have asm for fixed-point
+        AS_IF([test "$enable_float" != "yes"],[
+            cpu_arm=yes
+            AC_DEFINE([OPUS_ARM_ASM], [],  [Make use of ARM asm optimization])
+            AS_GCC_INLINE_ASSEMBLY(
+                [inline_optimization="ARM"],
+                [inline_optimization="disabled"]
+            )
+            AS_ASM_ARM_EDSP([OPUS_ARM_INLINE_EDSP=1],[OPUS_ARM_INLINE_EDSP=0])
+            AS_ASM_ARM_MEDIA([OPUS_ARM_INLINE_MEDIA=1],
+                [OPUS_ARM_INLINE_MEDIA=0])
+            AS_ASM_ARM_NEON([OPUS_ARM_INLINE_NEON=1],[OPUS_ARM_INLINE_NEON=0])
+            AS_IF([test x"$inline_optimization" = x"ARM"],[
+                AM_CONDITIONAL([OPUS_ARM_INLINE_ASM],[true])
+                AC_DEFINE([OPUS_ARM_INLINE_ASM], 1,
+                    [Use generic ARMv4 inline asm optimizations])
+                AS_IF([test x"$OPUS_ARM_INLINE_EDSP" = x"1"],[
+                    AC_DEFINE([OPUS_ARM_INLINE_EDSP], [1],
+                        [Use ARMv5E inline asm optimizations])
+                    inline_optimization="$inline_optimization (EDSP)"
+                ])
+                AS_IF([test x"$OPUS_ARM_INLINE_MEDIA" = x"1"],[
+                    AC_DEFINE([OPUS_ARM_INLINE_MEDIA], [1],
+                        [Use ARMv6 inline asm optimizations])
+                    inline_optimization="$inline_optimization (Media)"
+                ])
+                AS_IF([test x"$OPUS_ARM_INLINE_NEON" = x"1"],[
+                    AC_DEFINE([OPUS_ARM_INLINE_NEON], 1,
+                        [Use ARM NEON inline asm optimizations])
+                    inline_optimization="$inline_optimization (NEON)"
+                ])
+            ])
+            dnl We need Perl to translate RVCT-syntax asm to gas syntax.
+            AC_CHECK_PROG([HAVE_PERL], perl, yes, no)
+            AS_IF([test x"$HAVE_PERL" = x"yes"],[
+                AM_CONDITIONAL([OPUS_ARM_EXTERNAL_ASM],[true])
+                asm_optimization="ARM"
+                AS_IF([test x"$OPUS_ARM_INLINE_EDSP" = x"1"], [
+                    OPUS_ARM_PRESUME_EDSP=1
+                    OPUS_ARM_MAY_HAVE_EDSP=1
+                ],
+                [
+                    OPUS_ARM_PRESUME_EDSP=0
+                    OPUS_ARM_MAY_HAVE_EDSP=0
+                ])
+                AS_IF([test x"$OPUS_ARM_INLINE_MEDIA" = x"1"], [
+                    OPUS_ARM_PRESUME_MEDIA=1
+                    OPUS_ARM_MAY_HAVE_MEDIA=1
+                ],
+                [
+                    OPUS_ARM_PRESUME_MEDIA=0
+                    OPUS_ARM_MAY_HAVE_MEDIA=0
+                ])
+                AS_IF([test x"$OPUS_ARM_INLINE_NEON" = x"1"], [
+                    OPUS_ARM_PRESUME_NEON=1
+                    OPUS_ARM_MAY_HAVE_NEON=1
+                ],
+                [
+                    OPUS_ARM_PRESUME_NEON=0
+                    OPUS_ARM_MAY_HAVE_NEON=0
+                ])
+                AS_IF([test x"$enable_rtcd" = x"yes"],[
+                    AS_IF([test x"$OPUS_ARM_MAY_HAVE_EDSP" != x"1"],[
+                        AC_MSG_NOTICE(
+                          [Trying to force-enable armv5e EDSP instructions...])
+                        AS_ASM_ARM_EDSP_FORCE([OPUS_ARM_MAY_HAVE_EDSP=1])
+                    ])
+                    AS_IF([test x"$OPUS_ARM_MAY_HAVE_MEDIA" != x"1"],[
+                        AC_MSG_NOTICE(
+                          [Trying to force-enable ARMv6 media instructions...])
+                        AS_ASM_ARM_MEDIA_FORCE([OPUS_ARM_MAY_HAVE_MEDIA=1])
+                    ])
+                    AS_IF([test x"$OPUS_ARM_MAY_HAVE_NEON" != x"1"],[
+                        AC_MSG_NOTICE(
+                          [Trying to force-enable NEON instructions...])
+                        AS_ASM_ARM_NEON_FORCE([OPUS_ARM_MAY_HAVE_NEON=1])
+                    ])
+                ])
+                rtcd_support=
+                AS_IF([test x"$OPUS_ARM_MAY_HAVE_EDSP" = x"1"],[
+                    AC_DEFINE(OPUS_ARM_MAY_HAVE_EDSP, 1,
+                        [Define if assembler supports EDSP instructions])
+                    AS_IF([test x"$OPUS_ARM_PRESUME_EDSP" = x"1"],[
+                        AC_DEFINE(OPUS_ARM_PRESUME_EDSP, 1,
+                          [Define if binary requires EDSP instruction support])
+                        asm_optimization="$asm_optimization (EDSP)"
+                    ],
+                        [rtcd_support="$rtcd_support (EDSP)"]
+                    )
+                ])
+                AC_SUBST(OPUS_ARM_MAY_HAVE_EDSP)
+                AS_IF([test x"$OPUS_ARM_MAY_HAVE_MEDIA" = x"1"],[
+                    AC_DEFINE(OPUS_ARM_MAY_HAVE_MEDIA, 1,
+                      [Define if assembler supports ARMv6 media instructions])
+                    AS_IF([test x"$OPUS_ARM_PRESUME_MEDIA" = x"1"],[
+                        AC_DEFINE(OPUS_ARM_PRESUME_MEDIA, 1,
+                          [Define if binary requires ARMv6 media instruction support])
+                        asm_optimization="$asm_optimization (Media)"
+                    ],
+                        [rtcd_support="$rtcd_support (Media)"]
+                    )
+                ])
+                AC_SUBST(OPUS_ARM_MAY_HAVE_MEDIA)
+                AS_IF([test x"$OPUS_ARM_MAY_HAVE_NEON" = x"1"],[
+                    AC_DEFINE(OPUS_ARM_MAY_HAVE_NEON, 1,
+                      [Define if compiler supports NEON instructions])
+                    AS_IF([test x"$OPUS_ARM_PRESUME_NEON" = x"1"], [
+                        AC_DEFINE(OPUS_ARM_PRESUME_NEON, 1,
+                          [Define if binary requires NEON instruction support])
+                        asm_optimization="$asm_optimization (NEON)"
+                    ],
+                        [rtcd_support="$rtcd_support (NEON)"]
+                    )
+                ])
+                AC_SUBST(OPUS_ARM_MAY_HAVE_NEON)
+                dnl Make sure turning on RTCD gets us at least one
+                dnl instruction set.
+                AS_IF([test x"$rtcd_support" != x""],
+                    [rtcd_support=ARM"$rtcd_support"],
+                    [rtcd_support="no"]
+                )
+            ],
+            [
+                AC_MSG_WARN(
+                  [*** ARM assembly requires perl -- disabling optimizations])
+                asm_optimization="(missing perl dependency for ARM)"
+            ])
+        ])
         ;;
     esac
 ],[
-    asm_optimization="disabled"
+   inline_optimization="disabled"
+   asm_optimization="disabled"
 ])
 
 AM_CONDITIONAL([CPU_ARM], [test "$cpu_arm" = "yes"])
+AM_CONDITIONAL([OPUS_ARM_INLINE_ASM],
+    [test x"${inline_optimization:0:3}" = x"ARM"])
+AM_CONDITIONAL([OPUS_ARM_EXTERNAL_ASM],
+    [test x"${asm_optimization:0:3}" = x"ARM"])
 
-AC_ARG_ENABLE([rtcd],
-    [AS_HELP_STRING([--disable-rtcd], [Disable run-time CPU capabilities detection])],,
-    [enable_rtcd=yes])
-
-AS_IF([test "$enable_rtcd" = "yes"],[
-    AS_IF([test "$rtcd_support" = "yes"],[
-        AC_DEFINE([OPUS_HAVE_RTCD], [1], [Use run-time CPU capabilities detection])
-    ],[
-        rtcd_support="no rtcd for your platform, please send patches"
+AS_IF([test x"$enable_rtcd" = x"yes"],[
+    AS_IF([test x"$rtcd_support" != x"no"],[
+        AC_DEFINE([OPUS_HAVE_RTCD], [1],
+            [Use run-time CPU capabilities detection])
+        OPUS_HAVE_RTCD=1
+        AC_SUBST(OPUS_HAVE_RTCD)
     ])
 ],[
-    rtcd_support="no"
+    rtcd_support="disabled"
 ])
 
 AC_ARG_ENABLE([assertions],
@@ -300,9 +414,14 @@
 
 AC_SUBST([PC_BUILD])
 
-
-AC_CONFIG_FILES([Makefile opus.pc opus-uninstalled.pc
-                 doc/Makefile doc/Doxyfile])
+AC_CONFIG_FILES([
+    Makefile
+    opus.pc
+    opus-uninstalled.pc
+    celt/arm/armopts.s
+    doc/Makefile
+    doc/Doxyfile
+])
 AC_CONFIG_HEADERS([config.h])
 
 AC_OUTPUT
@@ -316,13 +435,14 @@
       C99 var arrays: ................ ${has_var_arrays}
       C99 lrintf: .................... ${ac_cv_func_lrintf}
       Use alloca: .................... ${use_alloca}
-    
+
     General configuration:
-    
+
       Floating point support: ........ ${enable_float}
       Fast float approximations: ..... ${enable_float_approx}
       Fixed point debugging: ......... ${enable_fixed_point_debug}
-      Assembly optimization: ......... ${asm_optimization}
+      Inline Assembly Optimizations: . ${inline_optimization}
+      External Assembly Optimizations: ${asm_optimization}
       Run-time CPU detection: ........ ${rtcd_support}
       Custom modes: .................. ${enable_custom_modes}
       Assertion checking: ............ ${enable_assertions}
--- a/m4/as-gcc-inline-assembly.m4
+++ b/m4/as-gcc-inline-assembly.m4
@@ -42,7 +42,17 @@
                      $2])
 ])
 
+AC_DEFUN([AS_ASM_ARM_NEON_FORCE],
+[
+  AC_MSG_CHECKING([if assembler supports NEON instructions on ARM])
 
+  AC_COMPILE_IFELSE([AC_LANG_PROGRAM([],[__asm__(".arch armv7-a\n.fpu neon\n.object_arch armv4t\nvorr d0,d0,d0")])],
+                    [AC_MSG_RESULT([yes])
+                     $1],
+                    [AC_MSG_RESULT([no])
+                     $2])
+])
+
 AC_DEFUN([AS_ASM_ARM_MEDIA],
 [
   AC_MSG_CHECKING([if assembler supports ARMv6 media instructions on ARM])
@@ -54,12 +64,33 @@
                      $2])
 ])
 
+AC_DEFUN([AS_ASM_ARM_MEDIA_FORCE],
+[
+  AC_MSG_CHECKING([if assembler supports ARMv6 media instructions on ARM])
 
+  AC_COMPILE_IFELSE([AC_LANG_PROGRAM([],[__asm__(".arch armv6\n.object_arch armv4t\nshadd8 r3,r3,r3")])],
+                    [AC_MSG_RESULT([yes])
+                     $1],
+                    [AC_MSG_RESULT([no])
+                     $2])
+])
+
 AC_DEFUN([AS_ASM_ARM_EDSP],
 [
   AC_MSG_CHECKING([if assembler supports EDSP instructions on ARM])
 
   AC_COMPILE_IFELSE([AC_LANG_PROGRAM([],[__asm__("qadd r3,r3,r3")])],
+                    [AC_MSG_RESULT([yes])
+                     $1],
+                    [AC_MSG_RESULT([no])
+                     $2])
+])
+
+AC_DEFUN([AS_ASM_ARM_EDSP_FORCE],
+[
+  AC_MSG_CHECKING([if assembler supports EDSP instructions on ARM])
+
+  AC_COMPILE_IFELSE([AC_LANG_PROGRAM([],[__asm__(".arch armv5te\n.object_arch armv4t\nqadd r3,r3,r3")])],
                     [AC_MSG_RESULT([yes])
                      $1],
                     [AC_MSG_RESULT([no])
--- a/silk/API.h
+++ b/silk/API.h
@@ -64,6 +64,7 @@
 /*************************/
 opus_int silk_InitEncoder(                              /* O    Returns error code                              */
     void                            *encState,          /* I/O  State                                           */
+    int                              arch,              /* I    Run-time architecture                           */
     silk_EncControlStruct           *encStatus          /* O    Encoder Status                                  */
 );
 
--- a/silk/SigProc_FIX.h
+++ b/silk/SigProc_FIX.h
@@ -227,7 +227,8 @@
     opus_int                    *scale,             /* O    Scaling of the correlation vector                           */
     const opus_int16            *inputData,         /* I    Input data to correlate                                     */
     const opus_int              inputDataSize,      /* I    Length of input                                             */
-    const opus_int              correlationCount    /* I    Number of correlation taps to compute                       */
+    const opus_int              correlationCount,   /* I    Number of correlation taps to compute                       */
+    int                         arch                /* I    Run-time architecture                                       */
 );
 
 void silk_decode_pitch(
@@ -249,7 +250,8 @@
     const opus_int              search_thres2_Q13,  /* I    Final threshold for lag candidates 0 - 1                    */
     const opus_int              Fs_kHz,             /* I    Sample frequency (kHz)                                      */
     const opus_int              complexity,         /* I    Complexity setting, 0-2, where 2 is highest                 */
-    const opus_int              nb_subfr            /* I    number of 5 ms subframes                                    */
+    const opus_int              nb_subfr,           /* I    number of 5 ms subframes                                    */
+    int                         arch                /* I    Run-time architecture                                       */
 );
 
 /* Compute Normalized Line Spectral Frequencies (NLSFs) from whitening filter coefficients      */
@@ -309,7 +311,8 @@
     const opus_int32            minInvGain_Q30,     /* I    Inverse of max prediction gain                              */
     const opus_int              subfr_length,       /* I    Input signal subframe length (incl. D preceding samples)    */
     const opus_int              nb_subfr,           /* I    Number of subframes stacked in x                            */
-    const opus_int              D                   /* I    Order                                                       */
+    const opus_int              D,                  /* I    Order                                                       */
+    int                         arch                /* I    Run-time architecture                                       */
 );
 
 /* Copy and multiply a vector by a constant */
@@ -576,11 +579,11 @@
 #include "MacroCount.h"
 #include "MacroDebug.h"
 
-#ifdef ARMv4_ASM
+#ifdef OPUS_ARM_INLINE_ASM
 #include "arm/SigProc_FIX_armv4.h"
 #endif
 
-#ifdef ARMv5E_ASM
+#ifdef OPUS_ARM_INLINE_EDSP
 #include "arm/SigProc_FIX_armv5e.h"
 #endif
 
--- a/silk/enc_API.c
+++ b/silk/enc_API.c
@@ -69,6 +69,7 @@
 /*************************/
 opus_int silk_InitEncoder(                              /* O    Returns error code                              */
     void                            *encState,          /* I/O  State                                           */
+    int                              arch,              /* I    Run-time architecture                           */
     silk_EncControlStruct           *encStatus          /* O    Encoder Status                                  */
 )
 {
@@ -80,7 +81,7 @@
     /* Reset encoder */
     silk_memset( psEnc, 0, sizeof( silk_encoder ) );
     for( n = 0; n < ENCODER_NUM_CHANNELS; n++ ) {
-        if( ret += silk_init_encoder( &psEnc->state_Fxx[ n ] ) ) {
+        if( ret += silk_init_encoder( &psEnc->state_Fxx[ n ], arch ) ) {
             silk_assert( 0 );
         }
     }
@@ -174,7 +175,7 @@
 
     if( encControl->nChannelsInternal > psEnc->nChannelsInternal ) {
         /* Mono -> Stereo transition: init state of second channel and stereo state */
-        ret += silk_init_encoder( &psEnc->state_Fxx[ 1 ] );
+        ret += silk_init_encoder( &psEnc->state_Fxx[ 1 ], psEnc->state_Fxx[ 0 ].sCmn.arch );
         silk_memset( psEnc->sStereo.pred_prev_Q13, 0, sizeof( psEnc->sStereo.pred_prev_Q13 ) );
         silk_memset( psEnc->sStereo.sSide, 0, sizeof( psEnc->sStereo.sSide ) );
         psEnc->sStereo.mid_side_amp_Q0[ 0 ] = 0;
@@ -206,9 +207,8 @@
         }
         /* Reset Encoder */
         for( n = 0; n < encControl->nChannelsInternal; n++ ) {
-            if( (ret = silk_init_encoder( &psEnc->state_Fxx[ n ] ) ) != 0 ) {
-                silk_assert( 0 );
-            }
+            ret = silk_init_encoder( &psEnc->state_Fxx[ n ], psEnc->state_Fxx[ n ].sCmn.arch );
+            silk_assert( !ret );
         }
         tmp_payloadSize_ms = encControl->payloadSize_ms;
         encControl->payloadSize_ms = 10;
--- a/silk/fixed/autocorr_FIX.c
+++ b/silk/fixed/autocorr_FIX.c
@@ -38,10 +38,11 @@
     opus_int                    *scale,             /* O    Scaling of the correlation vector                           */
     const opus_int16            *inputData,         /* I    Input data to correlate                                     */
     const opus_int              inputDataSize,      /* I    Length of input                                             */
-    const opus_int              correlationCount    /* I    Number of correlation taps to compute                       */
+    const opus_int              correlationCount,   /* I    Number of correlation taps to compute                       */
+    int                         arch                /* I    Run-time architecture                                       */
 )
 {
     opus_int   corrCount;
     corrCount = silk_min_int( inputDataSize, correlationCount );
-    *scale = _celt_autocorr(inputData, results, NULL, 0, corrCount-1, inputDataSize);
+    *scale = _celt_autocorr(inputData, results, NULL, 0, corrCount-1, inputDataSize, arch);
 }
--- a/silk/fixed/burg_modified_FIX.c
+++ b/silk/fixed/burg_modified_FIX.c
@@ -50,7 +50,8 @@
     const opus_int32            minInvGain_Q30,     /* I    Inverse of max prediction gain                              */
     const opus_int              subfr_length,       /* I    Input signal subframe length (incl. D preceding samples)    */
     const opus_int              nb_subfr,           /* I    Number of subframes stacked in x                            */
-    const opus_int              D                   /* I    Order                                                       */
+    const opus_int              D,                  /* I    Order                                                       */
+    int                         arch                /* I    Run-time architecture                                       */
 )
 {
     opus_int         k, n, s, lz, rshifts, rshifts_extra, reached_max_gain;
@@ -98,7 +99,7 @@
             int i;
             opus_int32 d;
             x_ptr = x + s * subfr_length;
-            celt_pitch_xcorr(x_ptr, x_ptr + 1, xcorr, subfr_length - D, D );
+            celt_pitch_xcorr(x_ptr, x_ptr + 1, xcorr, subfr_length - D, D, arch );
             for( n = 1; n < D + 1; n++ ) {
                for ( i = n + subfr_length - D, d = 0; i < subfr_length; i++ )
                   d = MAC16_16( d, x_ptr[ i ], x_ptr[ i - n ] );
--- a/silk/fixed/encode_frame_FIX.c
+++ b/silk/fixed/encode_frame_FIX.c
@@ -132,12 +132,12 @@
         /*****************************************/
         /* Find pitch lags, initial LPC analysis */
         /*****************************************/
-        silk_find_pitch_lags_FIX( psEnc, &sEncCtrl, res_pitch, x_frame );
+        silk_find_pitch_lags_FIX( psEnc, &sEncCtrl, res_pitch, x_frame, psEnc->sCmn.arch );
 
         /************************/
         /* Noise shape analysis */
         /************************/
-        silk_noise_shape_analysis_FIX( psEnc, &sEncCtrl, res_pitch_frame, x_frame );
+        silk_noise_shape_analysis_FIX( psEnc, &sEncCtrl, res_pitch_frame, x_frame, psEnc->sCmn.arch );
 
         /***************************************************/
         /* Find linear prediction coefficients (LPC + LTP) */
--- a/silk/fixed/find_LPC_FIX.c
+++ b/silk/fixed/find_LPC_FIX.c
@@ -60,13 +60,13 @@
     psEncC->indices.NLSFInterpCoef_Q2 = 4;
 
     /* Burg AR analysis for the full frame */
-    silk_burg_modified( &res_nrg, &res_nrg_Q, a_Q16, x, minInvGain_Q30, subfr_length, psEncC->nb_subfr, psEncC->predictLPCOrder );
+    silk_burg_modified( &res_nrg, &res_nrg_Q, a_Q16, x, minInvGain_Q30, subfr_length, psEncC->nb_subfr, psEncC->predictLPCOrder, psEncC->arch );
 
     if( psEncC->useInterpolatedNLSFs && !psEncC->first_frame_after_reset && psEncC->nb_subfr == MAX_NB_SUBFR ) {
         VARDECL( opus_int16, LPC_res );
 
         /* Optimal solution for last 10 ms */
-        silk_burg_modified( &res_tmp_nrg, &res_tmp_nrg_Q, a_tmp_Q16, x + 2 * subfr_length, minInvGain_Q30, subfr_length, 2, psEncC->predictLPCOrder );
+        silk_burg_modified( &res_tmp_nrg, &res_tmp_nrg_Q, a_tmp_Q16, x + 2 * subfr_length, minInvGain_Q30, subfr_length, 2, psEncC->predictLPCOrder, psEncC->arch );
 
         /* subtract residual energy here, as that's easier than adding it to the    */
         /* residual energy of the first 10 ms in each iteration of the search below */
--- a/silk/fixed/find_pitch_lags_FIX.c
+++ b/silk/fixed/find_pitch_lags_FIX.c
@@ -38,7 +38,8 @@
     silk_encoder_state_FIX          *psEnc,                                 /* I/O  encoder state                                                               */
     silk_encoder_control_FIX        *psEncCtrl,                             /* I/O  encoder control                                                             */
     opus_int16                      res[],                                  /* O    residual                                                                    */
-    const opus_int16                x[]                                     /* I    Speech signal                                                               */
+    const opus_int16                x[],                                    /* I    Speech signal                                                               */
+    int                             arch                                    /* I    Run-time architecture                                                       */
 )
 {
     opus_int   buf_len, i, scale;
@@ -86,7 +87,7 @@
     silk_apply_sine_window( Wsig_ptr, x_buf_ptr, 2, psEnc->sCmn.la_pitch );
 
     /* Calculate autocorrelation sequence */
-    silk_autocorr( auto_corr, &scale, Wsig, psEnc->sCmn.pitch_LPC_win_length, psEnc->sCmn.pitchEstimationLPCOrder + 1 );
+    silk_autocorr( auto_corr, &scale, Wsig, psEnc->sCmn.pitch_LPC_win_length, psEnc->sCmn.pitchEstimationLPCOrder + 1, arch );
 
     /* Add white noise, as fraction of energy */
     auto_corr[ 0 ] = silk_SMLAWB( auto_corr[ 0 ], auto_corr[ 0 ], SILK_FIX_CONST( FIND_PITCH_WHITE_NOISE_FRACTION, 16 ) ) + 1;
@@ -127,7 +128,8 @@
         /*****************************************/
         if( silk_pitch_analysis_core( res, psEncCtrl->pitchL, &psEnc->sCmn.indices.lagIndex, &psEnc->sCmn.indices.contourIndex,
                 &psEnc->LTPCorr_Q15, psEnc->sCmn.prevLag, psEnc->sCmn.pitchEstimationThreshold_Q16,
-                (opus_int)thrhld_Q13, psEnc->sCmn.fs_kHz, psEnc->sCmn.pitchEstimationComplexity, psEnc->sCmn.nb_subfr ) == 0 )
+                (opus_int)thrhld_Q13, psEnc->sCmn.fs_kHz, psEnc->sCmn.pitchEstimationComplexity, psEnc->sCmn.nb_subfr,
+                psEnc->sCmn.arch) == 0 )
         {
             psEnc->sCmn.indices.signalType = TYPE_VOICED;
         } else {
--- a/silk/fixed/main_FIX.h
+++ b/silk/fixed/main_FIX.h
@@ -73,7 +73,8 @@
 
 /* Initializes the Silk encoder state */
 opus_int silk_init_encoder(
-    silk_encoder_state_Fxx          *psEnc                                  /* I/O  Pointer to Silk FIX encoder state                                           */
+    silk_encoder_state_Fxx          *psEnc,                                 /* I/O  Pointer to Silk FIX encoder state                                           */
+    int                              arch                                   /* I    Run-time architecture                                                       */
 );
 
 /* Control the Silk encoder */
@@ -104,7 +105,8 @@
     silk_encoder_state_FIX          *psEnc,                                 /* I/O  Encoder state FIX                                                           */
     silk_encoder_control_FIX        *psEncCtrl,                             /* I/O  Encoder control FIX                                                         */
     const opus_int16                *pitch_res,                             /* I    LPC residual from pitch analysis                                            */
-    const opus_int16                *x                                      /* I    Input signal [ frame_length + la_shape ]                                    */
+    const opus_int16                *x,                                     /* I    Input signal [ frame_length + la_shape ]                                    */
+    int                              arch                                   /* I    Run-time architecture                                                       */
 );
 
 /* Autocorrelations for a warped frequency axis */
@@ -132,7 +134,8 @@
     silk_encoder_state_FIX          *psEnc,                                 /* I/O  encoder state                                                               */
     silk_encoder_control_FIX        *psEncCtrl,                             /* I/O  encoder control                                                             */
     opus_int16                      res[],                                  /* O    residual                                                                    */
-    const opus_int16                x[]                                     /* I    Speech signal                                                               */
+    const opus_int16                x[],                                    /* I    Speech signal                                                               */
+    int                             arch                                    /* I    Run-time architecture                                                       */
 );
 
 /* Find LPC and LTP coefficients */
--- a/silk/fixed/noise_shape_analysis_FIX.c
+++ b/silk/fixed/noise_shape_analysis_FIX.c
@@ -145,7 +145,8 @@
     silk_encoder_state_FIX          *psEnc,                                 /* I/O  Encoder state FIX                                                           */
     silk_encoder_control_FIX        *psEncCtrl,                             /* I/O  Encoder control FIX                                                         */
     const opus_int16                *pitch_res,                             /* I    LPC residual from pitch analysis                                            */
-    const opus_int16                *x                                      /* I    Input signal [ frame_length + la_shape ]                                    */
+    const opus_int16                *x,                                     /* I    Input signal [ frame_length + la_shape ]                                    */
+    int                              arch                                   /* I    Run-time architecture                                                       */
 )
 {
     silk_shape_state_FIX *psShapeSt = &psEnc->sShape;
@@ -281,7 +282,7 @@
             silk_warped_autocorrelation_FIX( auto_corr, &scale, x_windowed, warping_Q16, psEnc->sCmn.shapeWinLength, psEnc->sCmn.shapingLPCOrder );
         } else {
             /* Calculate regular auto correlation */
-            silk_autocorr( auto_corr, &scale, x_windowed, psEnc->sCmn.shapeWinLength, psEnc->sCmn.shapingLPCOrder + 1 );
+            silk_autocorr( auto_corr, &scale, x_windowed, psEnc->sCmn.shapeWinLength, psEnc->sCmn.shapingLPCOrder + 1, arch );
         }
 
         /* Add white noise, as a fraction of energy */
--- a/silk/fixed/pitch_analysis_core_FIX.c
+++ b/silk/fixed/pitch_analysis_core_FIX.c
@@ -62,7 +62,8 @@
     opus_int          start_lag,                       /* I lag offset to search around */
     opus_int          sf_length,                       /* I length of a 5 ms subframe   */
     opus_int          nb_subfr,                        /* I number of subframes         */
-    opus_int          complexity                       /* I Complexity setting          */
+    opus_int          complexity,                      /* I Complexity setting          */
+    int               arch                             /* I Run-time architecture       */
 );
 
 static void silk_P_Ana_calc_energy_st3(
@@ -88,7 +89,8 @@
     const opus_int              search_thres2_Q13,  /* I    Final threshold for lag candidates 0 - 1                    */
     const opus_int              Fs_kHz,             /* I    Sample frequency (kHz)                                      */
     const opus_int              complexity,         /* I    Complexity setting, 0-2, where 2 is highest                 */
-    const opus_int              nb_subfr            /* I    number of 5 ms subframes                                    */
+    const opus_int              nb_subfr,           /* I    number of 5 ms subframes                                    */
+    int                         arch                /* I    Run-time architecture                                       */
 )
 {
     VARDECL( opus_int16, frame_8kHz );
@@ -189,7 +191,7 @@
         silk_assert( basis_ptr >= frame_4kHz );
         silk_assert( basis_ptr + SF_LENGTH_8KHZ <= frame_4kHz + frame_length_4kHz );
 
-        celt_pitch_xcorr( target_ptr, target_ptr - MAX_LAG_4KHZ, xcorr32, SF_LENGTH_8KHZ, MAX_LAG_4KHZ - MIN_LAG_4KHZ + 1 );
+        celt_pitch_xcorr( target_ptr, target_ptr - MAX_LAG_4KHZ, xcorr32, SF_LENGTH_8KHZ, MAX_LAG_4KHZ - MIN_LAG_4KHZ + 1, arch );
 
         /* Calculate first vector products before loop */
         cross_corr = xcorr32[ MAX_LAG_4KHZ - MIN_LAG_4KHZ ];
@@ -516,7 +518,7 @@
         /* Calculate the correlations and energies needed in stage 3 */
         ALLOC( energies_st3, nb_subfr * nb_cbk_search, silk_pe_stage3_vals );
         ALLOC( cross_corr_st3, nb_subfr * nb_cbk_search, silk_pe_stage3_vals );
-        silk_P_Ana_calc_corr_st3(  cross_corr_st3, input_frame_ptr, start_lag, sf_length, nb_subfr, complexity );
+        silk_P_Ana_calc_corr_st3(  cross_corr_st3, input_frame_ptr, start_lag, sf_length, nb_subfr, complexity, arch );
         silk_P_Ana_calc_energy_st3( energies_st3, input_frame_ptr, start_lag, sf_length, nb_subfr, complexity );
 
         lag_counter = 0;
@@ -597,7 +599,8 @@
     opus_int          start_lag,                       /* I lag offset to search around */
     opus_int          sf_length,                       /* I length of a 5 ms subframe   */
     opus_int          nb_subfr,                        /* I number of subframes         */
-    opus_int          complexity                       /* I Complexity setting          */
+    opus_int          complexity,                      /* I Complexity setting          */
+    int               arch                             /* I Run-time architecture       */
 )
 {
     const opus_int16 *target_ptr;
@@ -634,7 +637,7 @@
         lag_low  = matrix_ptr( Lag_range_ptr, k, 0, 2 );
         lag_high = matrix_ptr( Lag_range_ptr, k, 1, 2 );
         silk_assert(lag_high-lag_low+1 <= SCRATCH_SIZE);
-        celt_pitch_xcorr( target_ptr, target_ptr - start_lag - lag_high, xcorr32, sf_length, lag_high - lag_low + 1 );
+        celt_pitch_xcorr( target_ptr, target_ptr - start_lag - lag_high, xcorr32, sf_length, lag_high - lag_low + 1, arch );
         for( j = lag_low; j <= lag_high; j++ ) {
             silk_assert( lag_counter < SCRATCH_SIZE );
             scratch_mem[ lag_counter ] = xcorr32[ lag_high - j ];
--- a/silk/float/SigProc_FLP.h
+++ b/silk/float/SigProc_FLP.h
@@ -94,7 +94,8 @@
     const silk_float    search_thres2,      /* I    Final threshold for lag candidates 0 - 1                    */
     const opus_int      Fs_kHz,             /* I    sample frequency (kHz)                                      */
     const opus_int      complexity,         /* I    Complexity setting, 0-2, where 2 is highest                 */
-    const opus_int      nb_subfr            /* I    Number of 5 ms subframes                                    */
+    const opus_int      nb_subfr,           /* I    Number of 5 ms subframes                                    */
+    int                 arch                /* I    Run-time architecture                                       */
 );
 
 void silk_insertion_sort_decreasing_FLP(
--- a/silk/float/encode_frame_FLP.c
+++ b/silk/float/encode_frame_FLP.c
@@ -129,7 +129,7 @@
         /*****************************************/
         /* Find pitch lags, initial LPC analysis */
         /*****************************************/
-        silk_find_pitch_lags_FLP( psEnc, &sEncCtrl, res_pitch, x_frame );
+        silk_find_pitch_lags_FLP( psEnc, &sEncCtrl, res_pitch, x_frame, psEnc->sCmn.arch );
 
         /************************/
         /* Noise shape analysis */
--- a/silk/float/find_pitch_lags_FLP.c
+++ b/silk/float/find_pitch_lags_FLP.c
@@ -37,7 +37,8 @@
     silk_encoder_state_FLP          *psEnc,                             /* I/O  Encoder state FLP                           */
     silk_encoder_control_FLP        *psEncCtrl,                         /* I/O  Encoder control FLP                         */
     silk_float                      res[],                              /* O    Residual                                    */
-    const silk_float                x[]                                 /* I    Speech signal                               */
+    const silk_float                x[],                                /* I    Speech signal                               */
+    int                             arch                                /* I    Run-time architecture                       */
 )
 {
     opus_int   buf_len;
@@ -116,7 +117,7 @@
         /*****************************************/
         if( silk_pitch_analysis_core_FLP( res, psEncCtrl->pitchL, &psEnc->sCmn.indices.lagIndex,
             &psEnc->sCmn.indices.contourIndex, &psEnc->LTPCorr, psEnc->sCmn.prevLag, psEnc->sCmn.pitchEstimationThreshold_Q16 / 65536.0f,
-            thrhld, psEnc->sCmn.fs_kHz, psEnc->sCmn.pitchEstimationComplexity, psEnc->sCmn.nb_subfr ) == 0 )
+            thrhld, psEnc->sCmn.fs_kHz, psEnc->sCmn.pitchEstimationComplexity, psEnc->sCmn.nb_subfr, arch ) == 0 )
         {
             psEnc->sCmn.indices.signalType = TYPE_VOICED;
         } else {
--- a/silk/float/main_FLP.h
+++ b/silk/float/main_FLP.h
@@ -71,7 +71,8 @@
 
 /* Initializes the Silk encoder state */
 opus_int silk_init_encoder(
-    silk_encoder_state_FLP          *psEnc                              /* I/O  Encoder state FLP                           */
+    silk_encoder_state_FLP          *psEnc,                             /* I/O  Encoder state FLP                           */
+    int                              arch                               /* I    Run-tim architecture                        */
 );
 
 /* Control the Silk encoder */
@@ -129,7 +130,8 @@
     silk_encoder_state_FLP          *psEnc,                             /* I/O  Encoder state FLP                           */
     silk_encoder_control_FLP        *psEncCtrl,                         /* I/O  Encoder control FLP                         */
     silk_float                      res[],                              /* O    Residual                                    */
-    const silk_float                x[]                                 /* I    Speech signal                               */
+    const silk_float                x[],                                /* I    Speech signal                               */
+    int                             arch                                /* I    Run-time architecture                       */
 );
 
 /* Find LPC and LTP coefficients */
--- a/silk/float/pitch_analysis_core_FLP.c
+++ b/silk/float/pitch_analysis_core_FLP.c
@@ -48,7 +48,8 @@
     opus_int            start_lag,          /* I start lag                                                      */
     opus_int            sf_length,          /* I sub frame length                                               */
     opus_int            nb_subfr,           /* I number of subframes                                            */
-    opus_int            complexity          /* I Complexity setting                                             */
+    opus_int            complexity,         /* I Complexity setting                                             */
+    int                 arch                /* I Run-time architecture                                          */
 );
 
 static void silk_P_Ana_calc_energy_st3(
@@ -74,7 +75,8 @@
     const silk_float    search_thres2,      /* I    Final threshold for lag candidates 0 - 1                    */
     const opus_int      Fs_kHz,             /* I    sample frequency (kHz)                                      */
     const opus_int      complexity,         /* I    Complexity setting, 0-2, where 2 is highest                 */
-    const opus_int      nb_subfr            /* I    Number of 5 ms subframes                                    */
+    const opus_int      nb_subfr,           /* I    Number of 5 ms subframes                                    */
+    int                 arch                /* I    Run-time architecture                                       */
 )
 {
     opus_int   i, k, d, j;
@@ -176,7 +178,7 @@
         silk_assert( basis_ptr >= frame_4kHz );
         silk_assert( basis_ptr + sf_length_8kHz <= frame_4kHz + frame_length_4kHz );
 
-        celt_pitch_xcorr( target_ptr, target_ptr-max_lag_4kHz, xcorr, sf_length_8kHz, max_lag_4kHz - min_lag_4kHz + 1 );
+        celt_pitch_xcorr( target_ptr, target_ptr-max_lag_4kHz, xcorr, sf_length_8kHz, max_lag_4kHz - min_lag_4kHz + 1, arch );
 
         /* Calculate first vector products before loop */
         cross_corr = xcorr[ max_lag_4kHz - min_lag_4kHz ];
@@ -409,7 +411,7 @@
         CCmax = -1000.0f;
 
         /* Calculate the correlations and energies needed in stage 3 */
-        silk_P_Ana_calc_corr_st3( cross_corr_st3, frame, start_lag, sf_length, nb_subfr, complexity );
+        silk_P_Ana_calc_corr_st3( cross_corr_st3, frame, start_lag, sf_length, nb_subfr, complexity, arch );
         silk_P_Ana_calc_energy_st3( energies_st3, frame, start_lag, sf_length, nb_subfr, complexity );
 
         lag_counter = 0;
@@ -493,7 +495,8 @@
     opus_int            start_lag,          /* I start lag                                                      */
     opus_int            sf_length,          /* I sub frame length                                               */
     opus_int            nb_subfr,           /* I number of subframes                                            */
-    opus_int            complexity          /* I Complexity setting                                             */
+    opus_int            complexity,         /* I Complexity setting                                             */
+    int                 arch                /* I Run-time architecture                                          */
 )
 {
     const silk_float *target_ptr;
@@ -527,7 +530,7 @@
         lag_low  = matrix_ptr( Lag_range_ptr, k, 0, 2 );
         lag_high = matrix_ptr( Lag_range_ptr, k, 1, 2 );
         silk_assert(lag_high-lag_low+1 <= SCRATCH_SIZE);
-        celt_pitch_xcorr( target_ptr, target_ptr - start_lag - lag_high, xcorr, sf_length, lag_high - lag_low + 1 );
+        celt_pitch_xcorr( target_ptr, target_ptr - start_lag - lag_high, xcorr, sf_length, lag_high - lag_low + 1, arch );
         for( j = lag_low; j <= lag_high; j++ ) {
             silk_assert( lag_counter < SCRATCH_SIZE );
             scratch_mem[ lag_counter ] = xcorr[ lag_high - j ];
--- a/silk/init_encoder.c
+++ b/silk/init_encoder.c
@@ -34,12 +34,14 @@
 #include "main_FLP.h"
 #endif
 #include "tuning_parameters.h"
+#include "cpu_support.h"
 
 /*********************************/
 /* Initialize Silk Encoder state */
 /*********************************/
 opus_int silk_init_encoder(
-    silk_encoder_state_Fxx          *psEnc                                  /* I/O  Pointer to Silk FIX encoder state                                           */
+    silk_encoder_state_Fxx          *psEnc,                                 /* I/O  Pointer to Silk FIX encoder state                                           */
+    int                              arch                                   /* I    Run-time architecture                                                       */
 )
 {
     opus_int ret = 0;
@@ -46,6 +48,8 @@
 
     /* Clear the entire encoder state */
     silk_memset( psEnc, 0, sizeof( silk_encoder_state_Fxx ) );
+
+    psEnc->sCmn.arch = arch;
 
     psEnc->sCmn.variable_HP_smth1_Q15 = silk_LSHIFT( silk_lin2log( SILK_FIX_CONST( VARIABLE_HP_MIN_CUTOFF_HZ, 16 ) ) - ( 16 << 7 ), 8 );
     psEnc->sCmn.variable_HP_smth2_Q15 = psEnc->sCmn.variable_HP_smth1_Q15;
--- a/silk/macros.h
+++ b/silk/macros.h
@@ -103,11 +103,11 @@
     (*((Matrix_base_adr) + ((row)+(M)*(column))))
 #endif
 
-#ifdef ARMv4_ASM
+#ifdef OPUS_ARM_INLINE_ASM
 #include "arm/macros_armv4.h"
 #endif
 
-#ifdef ARMv5E_ASM
+#ifdef OPUS_ARM_INLINE_EDSP
 #include "arm/macros_armv5e.h"
 #endif
 
--- a/silk/structs.h
+++ b/silk/structs.h
@@ -191,6 +191,8 @@
     SideInfoIndices              indices;
     opus_int8                    pulses[ MAX_FRAME_LENGTH ];
 
+    int                          arch;
+
     /* Input/output buffering */
     opus_int16                   inputBuf[ MAX_FRAME_LENGTH + 2 ];  /* Buffer containing input signal                                   */
     opus_int                     inputBufIx;
--- a/src/opus_decoder.c
+++ b/src/opus_decoder.c
@@ -75,7 +75,6 @@
 #endif
 
    opus_uint32  rangeFinal;
-   int arch;
 };
 
 #ifdef FIXED_POINT
@@ -125,7 +124,6 @@
    st->Fs = Fs;
    st->DecControl.API_sampleRate = st->Fs;
    st->DecControl.nChannelsAPI      = st->channels;
-   st->arch = opus_select_arch();
 
    /* Reset decoder */
    ret = silk_InitDecoder( silk_dec );
--- a/src/opus_encoder.c
+++ b/src/opus_encoder.c
@@ -104,7 +104,7 @@
     int          analysis_offset;
 #endif
     opus_uint32  rangeFinal;
-    int arch;
+    int          arch;
 };
 
 /* Transition tables for the voice and music. First column is the
@@ -188,7 +188,7 @@
 
     st->arch = opus_select_arch();
 
-    ret = silk_InitEncoder( silk_enc, &st->silk_mode );
+    ret = silk_InitEncoder( silk_enc, st->arch, &st->silk_mode );
     if(ret)return OPUS_INTERNAL_ERROR;
 
     /* default SILK parameters */
@@ -209,7 +209,7 @@
 
     /* Create CELT encoder */
     /* Initialize CELT encoder */
-    err = celt_encoder_init(celt_enc, Fs, channels);
+    err = celt_encoder_init(celt_enc, Fs, channels, st->arch);
     if(err!=OPUS_OK)return OPUS_INTERNAL_ERROR;
 
     celt_encoder_ctl(celt_enc, CELT_SET_SIGNALLING(0));
@@ -1219,7 +1219,7 @@
     if (st->mode != MODE_CELT_ONLY && st->prev_mode == MODE_CELT_ONLY)
     {
         silk_EncControlStruct dummy;
-        silk_InitEncoder( silk_enc, &dummy);
+        silk_InitEncoder( silk_enc, st->arch, &dummy);
         prefill=1;
     }
 
@@ -2418,7 +2418,7 @@
                  ((char*)&st->OPUS_ENCODER_RESET_START - (char*)st));
 
            celt_encoder_ctl(celt_enc, OPUS_RESET_STATE);
-           silk_InitEncoder( silk_enc, &dummy );
+           silk_InitEncoder( silk_enc, st->arch, &dummy );
            st->stream_channels = st->channels;
            st->hybrid_stereo_width_Q14 = 1 << 14;
            st->prev_HB_gain = Q15ONE;