diff options
Diffstat (limited to 'drivers/opus/celt')
68 files changed, 19260 insertions, 0 deletions
diff --git a/drivers/opus/celt/_kiss_fft_guts.h b/drivers/opus/celt/_kiss_fft_guts.h new file mode 100644 index 0000000000..21bea8a9b0 --- /dev/null +++ b/drivers/opus/celt/_kiss_fft_guts.h @@ -0,0 +1,183 @@ +/*Copyright (c) 2003-2004, Mark Borgerding + + All rights reserved. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are met: + + * Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above copyright notice, + this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" + AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE + LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + POSSIBILITY OF SUCH DAMAGE.*/ + +#ifndef KISS_FFT_GUTS_H +#define KISS_FFT_GUTS_H + +#define MIN(a,b) ((a)<(b) ? (a):(b)) +#define MAX(a,b) ((a)>(b) ? (a):(b)) + +/* kiss_fft.h + defines kiss_fft_scalar as either short or a float type + and defines + typedef struct { kiss_fft_scalar r; kiss_fft_scalar i; }kiss_fft_cpx; */ +#include "kiss_fft.h" + +/* + Explanation of macros dealing with complex math: + + C_MUL(m,a,b) : m = a*b + C_FIXDIV( c , div ) : if a fixed point impl., c /= div. noop otherwise + C_SUB( res, a,b) : res = a - b + C_SUBFROM( res , a) : res -= a + C_ADDTO( res , a) : res += a + * */ +#ifdef OPUS_FIXED_POINT +#include "arch.h" + + +#define SAMP_MAX 2147483647 +#define TWID_MAX 32767 +#define TRIG_UPSCALE 1 + +#define SAMP_MIN -SAMP_MAX + + +# define S_MUL(a,b) MULT16_32_Q15(b, a) + +# define C_MUL(m,a,b) \ + do{ (m).r = SUB32(S_MUL((a).r,(b).r) , S_MUL((a).i,(b).i)); \ + (m).i = ADD32(S_MUL((a).r,(b).i) , S_MUL((a).i,(b).r)); }while(0) + +# define C_MULC(m,a,b) \ + do{ (m).r = ADD32(S_MUL((a).r,(b).r) , S_MUL((a).i,(b).i)); \ + (m).i = SUB32(S_MUL((a).i,(b).r) , S_MUL((a).r,(b).i)); }while(0) + +# define C_MUL4(m,a,b) \ + do{ (m).r = SHR32(SUB32(S_MUL((a).r,(b).r) , S_MUL((a).i,(b).i)),2); \ + (m).i = SHR32(ADD32(S_MUL((a).r,(b).i) , S_MUL((a).i,(b).r)),2); }while(0) + +# define C_MULBYSCALAR( c, s ) \ + do{ (c).r = S_MUL( (c).r , s ) ;\ + (c).i = S_MUL( (c).i , s ) ; }while(0) + +# define DIVSCALAR(x,k) \ + (x) = S_MUL( x, (TWID_MAX-((k)>>1))/(k)+1 ) + +# define C_FIXDIV(c,div) \ + do { DIVSCALAR( (c).r , div); \ + DIVSCALAR( (c).i , div); }while (0) + +#define C_ADD( res, a,b)\ + do {(res).r=ADD32((a).r,(b).r); (res).i=ADD32((a).i,(b).i); \ + }while(0) +#define C_SUB( res, a,b)\ + do {(res).r=SUB32((a).r,(b).r); (res).i=SUB32((a).i,(b).i); \ + }while(0) +#define C_ADDTO( res , a)\ + do {(res).r = ADD32((res).r, (a).r); (res).i = ADD32((res).i,(a).i);\ + }while(0) + +#define C_SUBFROM( res , a)\ + do {(res).r = ADD32((res).r,(a).r); (res).i = SUB32((res).i,(a).i); \ + }while(0) + +#if defined(OPUS_ARM_INLINE_ASM) +#include "arm/kiss_fft_armv4.h" +#endif + +#if defined(OPUS_ARM_INLINE_EDSP) +#include "arm/kiss_fft_armv5e.h" +#endif + +#else /* not OPUS_FIXED_POINT*/ + +# define S_MUL(a,b) ( (a)*(b) ) +#define C_MUL(m,a,b) \ + do{ (m).r = (a).r*(b).r - (a).i*(b).i;\ + (m).i = (a).r*(b).i + (a).i*(b).r; }while(0) +#define C_MULC(m,a,b) \ + do{ (m).r = (a).r*(b).r + (a).i*(b).i;\ + (m).i = (a).i*(b).r - (a).r*(b).i; }while(0) + +#define C_MUL4(m,a,b) C_MUL(m,a,b) + +# define C_FIXDIV(c,div) /* NOOP */ +# define C_MULBYSCALAR( c, s ) \ + do{ (c).r *= (s);\ + (c).i *= (s); }while(0) +#endif + +#ifndef CHECK_OVERFLOW_OP +# define CHECK_OVERFLOW_OP(a,op,b) /* noop */ +#endif + +#ifndef C_ADD +#define C_ADD( res, a,b)\ + do { \ + CHECK_OVERFLOW_OP((a).r,+,(b).r)\ + CHECK_OVERFLOW_OP((a).i,+,(b).i)\ + (res).r=(a).r+(b).r; (res).i=(a).i+(b).i; \ + }while(0) +#define C_SUB( res, a,b)\ + do { \ + CHECK_OVERFLOW_OP((a).r,-,(b).r)\ + CHECK_OVERFLOW_OP((a).i,-,(b).i)\ + (res).r=(a).r-(b).r; (res).i=(a).i-(b).i; \ + }while(0) +#define C_ADDTO( res , a)\ + do { \ + CHECK_OVERFLOW_OP((res).r,+,(a).r)\ + CHECK_OVERFLOW_OP((res).i,+,(a).i)\ + (res).r += (a).r; (res).i += (a).i;\ + }while(0) + +#define C_SUBFROM( res , a)\ + do {\ + CHECK_OVERFLOW_OP((res).r,-,(a).r)\ + CHECK_OVERFLOW_OP((res).i,-,(a).i)\ + (res).r -= (a).r; (res).i -= (a).i; \ + }while(0) +#endif /* C_ADD defined */ + +#ifdef OPUS_FIXED_POINT +# define KISS_FFT_COS(phase) TRIG_UPSCALE*floor(MIN(32767,MAX(-32767,.5+32768 * cos (phase)))) +# define KISS_FFT_SIN(phase) TRIG_UPSCALE*floor(MIN(32767,MAX(-32767,.5+32768 * sin (phase)))) +# define KISS_FFT_COS(phase) floor(.5+TWID_MAX*cos (phase)) +# define KISS_FFT_SIN(phase) floor(.5+TWID_MAX*sin (phase)) +# define HALF_OF(x) ((x)>>1) +#elif defined(USE_SIMD) +# define KISS_FFT_COS(phase) _mm_set1_ps( cos(phase) ) +# define KISS_FFT_SIN(phase) _mm_set1_ps( sin(phase) ) +# define HALF_OF(x) ((x)*_mm_set1_ps(.5f)) +#else +# define KISS_FFT_COS(phase) (kiss_fft_scalar) cos(phase) +# define KISS_FFT_SIN(phase) (kiss_fft_scalar) sin(phase) +# define HALF_OF(x) ((x)*.5f) +#endif + +#define kf_cexp(x,phase) \ + do{ \ + (x)->r = KISS_FFT_COS(phase);\ + (x)->i = KISS_FFT_SIN(phase);\ + }while(0) + +#define kf_cexp2(x,phase) \ + do{ \ + (x)->r = TRIG_UPSCALE*celt_cos_norm((phase));\ + (x)->i = TRIG_UPSCALE*celt_cos_norm((phase)-32768);\ +}while(0) + +#endif /* KISS_FFT_GUTS_H */ diff --git a/drivers/opus/celt/arch.h b/drivers/opus/celt/arch.h new file mode 100644 index 0000000000..83e3705000 --- /dev/null +++ b/drivers/opus/celt/arch.h @@ -0,0 +1,214 @@ +/* Copyright (c) 2003-2008 Jean-Marc Valin + Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/** + @file arch.h + @brief Various architecture definitions for CELT +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef ARCH_H +#define ARCH_H + +#include "opus_types.h" +#include "opus_defines.h" + +# if !defined(__GNUC_PREREQ) +# if defined(__GNUC__)&&defined(__GNUC_MINOR__) +# define __GNUC_PREREQ(_maj,_min) \ + ((__GNUC__<<16)+__GNUC_MINOR__>=((_maj)<<16)+(_min)) +# else +# define __GNUC_PREREQ(_maj,_min) 0 +# endif +# endif + +#define CELT_SIG_SCALE 32768.f + +#define celt_fatal(str) _celt_fatal(str, __FILE__, __LINE__); +#ifdef ENABLE_ASSERTIONS +#include <stdio.h> +#include <stdlib.h> +#ifdef __GNUC__ +__attribute__((noreturn)) +#endif +static OPUS_INLINE void _celt_fatal(const char *str, const char *file, int line) +{ + fprintf (stderr, "Fatal (internal) error in %s, line %d: %s\n", file, line, str); + abort(); +} +#define celt_assert(cond) {if (!(cond)) {celt_fatal("assertion failed: " #cond);}} +#define celt_assert2(cond, message) {if (!(cond)) {celt_fatal("assertion failed: " #cond "\n" message);}} +#else +#define celt_assert(cond) +#define celt_assert2(cond, message) +#endif + +#define IMUL32(a,b) ((a)*(b)) + +#define ABS(x) ((x) < 0 ? (-(x)) : (x)) /**< Absolute integer value. */ +#define ABS16(x) ((x) < 0 ? (-(x)) : (x)) /**< Absolute 16-bit value. */ +#define MIN16(a,b) ((a) < (b) ? (a) : (b)) /**< Minimum 16-bit value. */ +#define MAX16(a,b) ((a) > (b) ? (a) : (b)) /**< Maximum 16-bit value. */ +#define ABS32(x) ((x) < 0 ? (-(x)) : (x)) /**< Absolute 32-bit value. */ +#define MIN32(a,b) ((a) < (b) ? (a) : (b)) /**< Minimum 32-bit value. */ +#define MAX32(a,b) ((a) > (b) ? (a) : (b)) /**< Maximum 32-bit value. */ +#define IMIN(a,b) ((a) < (b) ? (a) : (b)) /**< Minimum int value. */ +#define IMAX(a,b) ((a) > (b) ? (a) : (b)) /**< Maximum int value. */ +#define UADD32(a,b) ((a)+(b)) +#define USUB32(a,b) ((a)-(b)) + +#define PRINT_MIPS(file) + +#ifdef OPUS_FIXED_POINT + +typedef opus_int16 opus_val16; +typedef opus_int32 opus_val32; + +typedef opus_val32 celt_sig; +typedef opus_val16 celt_norm; +typedef opus_val32 celt_ener; + +#define Q15ONE 32767 + +#define SIG_SHIFT 12 + +#define NORM_SCALING 16384 + +#define DB_SHIFT 10 + +#define EPSILON 1 +#define VERY_SMALL 0 +#define VERY_LARGE16 ((opus_val16)32767) +#define Q15_ONE ((opus_val16)32767) + +#define SCALEIN(a) (a) +#define SCALEOUT(a) (a) + +#ifdef FIXED_DEBUG +#include "fixed_debug.h" +#else + +#include "fixed_generic.h" + +#ifdef OPUS_ARM_INLINE_EDSP +#include "arm/fixed_armv5e.h" +#elif defined (OPUS_ARM_INLINE_ASM) +#include "arm/fixed_armv4.h" +#elif defined (BFIN_ASM) +#include "fixed_bfin.h" +#elif defined (TI_C5X_ASM) +#include "fixed_c5x.h" +#elif defined (TI_C6X_ASM) +#include "fixed_c6x.h" +#endif + +#endif + +#else /* OPUS_FIXED_POINT */ + +typedef float opus_val16; +typedef float opus_val32; + +typedef float celt_sig; +typedef float celt_norm; +typedef float celt_ener; + +#define Q15ONE 1.0f + +#define NORM_SCALING 1.f + +#define EPSILON 1e-15f +#define VERY_SMALL 1e-30f +#define VERY_LARGE16 1e15f +#define Q15_ONE ((opus_val16)1.f) + +#define QCONST16(x,bits) (x) +#define QCONST32(x,bits) (x) + +#define NEG16(x) (-(x)) +#define NEG32(x) (-(x)) +#define EXTRACT16(x) (x) +#define EXTEND32(x) (x) +#define SHR16(a,shift) (a) +#define SHL16(a,shift) (a) +#define SHR32(a,shift) (a) +#define SHL32(a,shift) (a) +#define PSHR32(a,shift) (a) +#define VSHR32(a,shift) (a) + +#define PSHR(a,shift) (a) +#define SHR(a,shift) (a) +#define SHL(a,shift) (a) +#define SATURATE(x,a) (x) +#define SATURATE16(x) (x) + +#define ROUND16(a,shift) (a) +#define HALF16(x) (.5f*(x)) +#define HALF32(x) (.5f*(x)) + +#define ADD16(a,b) ((a)+(b)) +#define SUB16(a,b) ((a)-(b)) +#define ADD32(a,b) ((a)+(b)) +#define SUB32(a,b) ((a)-(b)) +#define MULT16_16_16(a,b) ((a)*(b)) +#define MULT16_16(a,b) ((opus_val32)(a)*(opus_val32)(b)) +#define MAC16_16(c,a,b) ((c)+(opus_val32)(a)*(opus_val32)(b)) + +#define MULT16_32_Q15(a,b) ((a)*(b)) +#define MULT16_32_Q16(a,b) ((a)*(b)) + +#define MULT32_32_Q31(a,b) ((a)*(b)) + +#define MAC16_32_Q15(c,a,b) ((c)+(a)*(b)) + +#define MULT16_16_Q11_32(a,b) ((a)*(b)) +#define MULT16_16_Q11(a,b) ((a)*(b)) +#define MULT16_16_Q13(a,b) ((a)*(b)) +#define MULT16_16_Q14(a,b) ((a)*(b)) +#define MULT16_16_Q15(a,b) ((a)*(b)) +#define MULT16_16_P15(a,b) ((a)*(b)) +#define MULT16_16_P13(a,b) ((a)*(b)) +#define MULT16_16_P14(a,b) ((a)*(b)) +#define MULT16_32_P16(a,b) ((a)*(b)) + +#define DIV32_16(a,b) (((opus_val32)(a))/(opus_val16)(b)) +#define DIV32(a,b) (((opus_val32)(a))/(opus_val32)(b)) + +#define SCALEIN(a) ((a)*CELT_SIG_SCALE) +#define SCALEOUT(a) ((a)*(1/CELT_SIG_SCALE)) + +#endif /* !OPUS_FIXED_POINT */ + +#ifndef GLOBAL_STACK_SIZE +#ifdef OPUS_FIXED_POINT +#define GLOBAL_STACK_SIZE 100000 +#else +#define GLOBAL_STACK_SIZE 100000 +#endif +#endif + +#endif /* ARCH_H */ diff --git a/drivers/opus/celt/arm/arm2gnu.pl b/drivers/opus/celt/arm/arm2gnu.pl new file mode 100755 index 0000000000..eab42efa2b --- /dev/null +++ b/drivers/opus/celt/arm/arm2gnu.pl @@ -0,0 +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"); +} diff --git a/drivers/opus/celt/arm/arm_celt_map.c b/drivers/opus/celt/arm/arm_celt_map.c new file mode 100644 index 0000000000..b187345154 --- /dev/null +++ b/drivers/opus/celt/arm/arm_celt_map.c @@ -0,0 +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 OPUS_HAVE_CONFIG_H +#include "opus_config.h" +#endif + +#include "pitch.h" + +#if defined(OPUS_HAVE_RTCD) + +# if defined(OPUS_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 diff --git a/drivers/opus/celt/arm/armcpu.c b/drivers/opus/celt/arm/armcpu.c new file mode 100644 index 0000000000..7f0af631b9 --- /dev/null +++ b/drivers/opus/celt/arm/armcpu.c @@ -0,0 +1,174 @@ +/* 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. +*/ + +/* Original code from libtheora modified to suit to Opus */ + +#ifdef OPUS_HAVE_CONFIG_H +#include "opus_config.h" +#endif + +#ifdef OPUS_HAVE_RTCD + +#include "armcpu.h" +#include "cpu_support.h" +#include "os_support.h" +#include "opus_types.h" + +#define OPUS_CPU_ARM_V4 (1) +#define OPUS_CPU_ARM_EDSP (1<<1) +#define OPUS_CPU_ARM_MEDIA (1<<2) +#define OPUS_CPU_ARM_NEON (1<<3) + +#if defined(_MSC_VER) +/*For GetExceptionCode() and EXCEPTION_ILLEGAL_INSTRUCTION.*/ +# define WIN32_LEAN_AND_MEAN +# define WIN32_EXTRA_LEAN +# include <windows.h> + +static OPUS_INLINE opus_uint32 opus_cpu_capabilities(void){ + opus_uint32 flags; + flags=0; + /* 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(OPUS_ARM_MAY_HAVE_EDSP) + __try{ + /*PLD [r13]*/ + __emit(0xF5DDF000); + flags|=OPUS_CPU_ARM_EDSP; + } + __except(GetExceptionCode()==EXCEPTION_ILLEGAL_INSTRUCTION){ + /*Ignore exception.*/ + } +# if defined(OPUS_ARM_MAY_HAVE_MEDIA) + __try{ + /*SHADD8 r3,r3,r3*/ + __emit(0xE6333F93); + flags|=OPUS_CPU_ARM_MEDIA; + } + __except(GetExceptionCode()==EXCEPTION_ILLEGAL_INSTRUCTION){ + /*Ignore exception.*/ + } +# if defined(OPUS_ARM_MAY_HAVE_NEON) + __try{ + /*VORR q0,q0,q0*/ + __emit(0xF2200150); + flags|=OPUS_CPU_ARM_NEON; + } + __except(GetExceptionCode()==EXCEPTION_ILLEGAL_INSTRUCTION){ + /*Ignore exception.*/ + } +# endif +# endif +# endif + return flags; +} + +#elif defined(__linux__) +/* Linux based */ +opus_uint32 opus_cpu_capabilities(void) +{ + opus_uint32 flags = 0; + FILE *cpuinfo; + + /* Reading /proc/self/auxv would be easier, but that doesn't work reliably on + * Android */ + cpuinfo = fopen("/proc/cpuinfo", "r"); + + if(cpuinfo != NULL) + { + /* 512 should be enough for anybody (it's even enough for all the flags that + * x86 has accumulated... so far). */ + char buf[512]; + + 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) + { + int version; + version = atoi(buf+17); + + if(version >= 6) + flags |= OPUS_CPU_ARM_MEDIA; + } +# endif + } + + fclose(cpuinfo); + } + return flags; +} +#else +/* The feature registers which can tell us what the processor supports are + * accessible in priveleged modes only, so we can't have a general user-space + * detection method like on x86.*/ +# error "Configured to use ARM asm but no CPU detection method available for " \ + "your platform. Reconfigure with --disable-rtcd (or send patches)." +#endif + +int opus_select_arch(void) +{ + opus_uint32 flags = opus_cpu_capabilities(); + int arch = 0; + + if(!(flags & OPUS_CPU_ARM_EDSP)) + return arch; + arch++; + + if(!(flags & OPUS_CPU_ARM_MEDIA)) + return arch; + arch++; + + if(!(flags & OPUS_CPU_ARM_NEON)) + return arch; + arch++; + + return arch; +} + +#endif diff --git a/drivers/opus/celt/arm/armcpu.h b/drivers/opus/celt/arm/armcpu.h new file mode 100644 index 0000000000..ac5744606e --- /dev/null +++ b/drivers/opus/celt/arm/armcpu.h @@ -0,0 +1,71 @@ +/* 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(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 diff --git a/drivers/opus/celt/arm/armopts.s b/drivers/opus/celt/arm/armopts.s new file mode 100644 index 0000000000..fb9196072a --- /dev/null +++ b/drivers/opus/celt/arm/armopts.s @@ -0,0 +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 * + +; Set the following to 1 if we have ARMv6 media instructions. +OPUS_ARM_MAY_HAVE_MEDIA * + +; Set the following to 1 if we have NEON (some ARMv7) +OPUS_ARM_MAY_HAVE_NEON * + +END diff --git a/drivers/opus/celt/arm/armopts.s.in b/drivers/opus/celt/arm/armopts.s.in new file mode 100644 index 0000000000..3d8aaf2754 --- /dev/null +++ b/drivers/opus/celt/arm/armopts.s.in @@ -0,0 +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 diff --git a/drivers/opus/celt/arm/celt_pitch_xcorr_arm.s b/drivers/opus/celt/arm/celt_pitch_xcorr_arm.s new file mode 100644 index 0000000000..09917b16bf --- /dev/null +++ b/drivers/opus/celt/arm/celt_pitch_xcorr_arm.s @@ -0,0 +1,545 @@ +; 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 +celt_pitch_xcorr_neon_process_remaining4_done + ; Reduce the sum to a single value. + VADD.S32 d0, d0, d1 + VPADDL.S32 d0, d0 + ADDS r12, r12, #4 + BLE celt_pitch_xcorr_neon_process_remaining_loop_done +; Sum terms 1 at a time. +celt_pitch_xcorr_neon_process_remaining_loop1 + VLD1.16 {d2[]}, [r4]! + VLD1.16 {d3[]}, [r5]! + SUBS r12, r12, #1 + VMLAL.S16 q0, d2, d3 + BGT celt_pitch_xcorr_neon_process_remaining_loop1 +celt_pitch_xcorr_neon_process_remaining_loop_done + VST1.32 {d0[0]}, [r2]! + VMAX.S32 d30, d30, d0 + SUBS r6, r6, #1 + ; _y++ + ADD r1, r1, #2 + ; if (--max_pitch > 0) goto celt_pitch_xcorr_neon_process_remaining + BGT celt_pitch_xcorr_neon_process_remaining +celt_pitch_xcorr_neon_done + VMOV.32 r0, d30[0] + LDMFD sp!, {r4-r6, pc} + 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 (must be 32-bit aligned) + ; r5 = opus_val16 *_y (must be 32-bit aligned) + ; r6...r9 = opus_val32 sum[4] + ; output: + ; r6...r9 = opus_val32 sum[4] + ; preserved: r0-r5 + ; internal usage + ; r2 = int j + ; r12,r14 = opus_val16 x[4] + ; r10,r11 = opus_val16 y[4] + STMFD sp!, {r2,r4,r5,lr} + LDR r10, [r5], #4 ; Load y[0...1] + SUBS r2, r3, #4 ; j = len-4 + LDR r11, [r5], #4 ; Load y[2...3] + BLE xcorr_kernel_edsp_process4_done + LDR r12, [r4], #4 ; Load x[0...1] + ; Stall +xcorr_kernel_edsp_process4 + ; The multiplies must issue from pipeline 0, and can't dual-issue with each + ; other. Every other instruction here dual-issues with a multiply, and is + ; thus "free". There should be no stalls in the body of the loop. + SMLABB r6, r12, r10, r6 ; sum[0] = MAC16_16(sum[0],x_0,y_0) + LDR r14, [r4], #4 ; Load x[2...3] + SMLABT r7, r12, r10, r7 ; sum[1] = MAC16_16(sum[1],x_0,y_1) + SUBS r2, r2, #4 ; j-=4 + SMLABB r8, r12, r11, r8 ; sum[2] = MAC16_16(sum[2],x_0,y_2) + SMLABT r9, r12, r11, r9 ; sum[3] = MAC16_16(sum[3],x_0,y_3) + SMLATT r6, r12, r10, r6 ; sum[0] = MAC16_16(sum[0],x_1,y_1) + LDR r10, [r5], #4 ; Load y[4...5] + SMLATB r7, r12, r11, r7 ; sum[1] = MAC16_16(sum[1],x_1,y_2) + SMLATT r8, r12, r11, r8 ; sum[2] = MAC16_16(sum[2],x_1,y_3) + SMLATB r9, r12, r10, r9 ; sum[3] = MAC16_16(sum[3],x_1,y_4) + LDRGT r12, [r4], #4 ; Load x[0...1] + SMLABB r6, r14, r11, r6 ; sum[0] = MAC16_16(sum[0],x_2,y_2) + SMLABT r7, r14, r11, r7 ; sum[1] = MAC16_16(sum[1],x_2,y_3) + SMLABB r8, r14, r10, r8 ; sum[2] = MAC16_16(sum[2],x_2,y_4) + SMLABT r9, r14, r10, r9 ; sum[3] = MAC16_16(sum[3],x_2,y_5) + SMLATT r6, r14, r11, r6 ; sum[0] = MAC16_16(sum[0],x_3,y_3) + LDR r11, [r5], #4 ; Load y[6...7] + SMLATB r7, r14, r10, r7 ; sum[1] = MAC16_16(sum[1],x_3,y_4) + SMLATT r8, r14, r10, r8 ; sum[2] = MAC16_16(sum[2],x_3,y_5) + SMLATB r9, r14, r11, r9 ; sum[3] = MAC16_16(sum[3],x_3,y_6) + BGT xcorr_kernel_edsp_process4 +xcorr_kernel_edsp_process4_done + ADDS r2, r2, #4 + BLE xcorr_kernel_edsp_done + LDRH r12, [r4], #2 ; r12 = *x++ + SUBS r2, r2, #1 ; j-- + ; Stall + SMLABB r6, r12, r10, r6 ; sum[0] = MAC16_16(sum[0],x,y_0) + 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 (must be 32-bit aligned) + ; r1 = opus_val16 *_y (only needs to be 16-bit aligned) + ; r2 = opus_val32 *xcorr + ; r3 = int len + ; output: + ; r0 = maxcorr + ; internal usage + ; r4 = opus_val16 *x + ; r5 = opus_val16 *y + ; r6 = opus_val32 sum0 + ; r7 = opus_val32 sum1 + ; r8 = opus_val32 sum2 + ; r9 = opus_val32 sum3 + ; r1 = int max_pitch + ; r12 = int j + STMFD sp!, {r4-r11, lr} + MOV r5, r1 + LDR r1, [sp, #36] + MOV r4, r0 + TST r5, #3 + ; maxcorr = 1 + MOV r0, #1 + BEQ celt_pitch_xcorr_edsp_process1u_done +; Compute one sum at the start to make y 32-bit aligned. + SUBS r12, r3, #4 + ; r14 = sum = 0 + MOV r14, #0 + LDRH r8, [r5], #2 + BLE celt_pitch_xcorr_edsp_process1u_loop4_done + LDR r6, [r4], #4 + MOV r8, r8, LSL #16 +celt_pitch_xcorr_edsp_process1u_loop4 + LDR r9, [r5], #4 + SMLABT r14, r6, r8, r14 ; sum = MAC16_16(sum, x_0, y_0) + LDR r7, [r4], #4 + SMLATB r14, r6, r9, r14 ; sum = MAC16_16(sum, x_1, y_1) + LDR r8, [r5], #4 + SMLABT r14, r7, r9, r14 ; sum = MAC16_16(sum, x_2, y_2) + SUBS r12, r12, #4 ; j-=4 + SMLATB r14, r7, r8, r14 ; sum = MAC16_16(sum, x_3, y_3) + LDRGT r6, [r4], #4 + BGT celt_pitch_xcorr_edsp_process1u_loop4 + MOV r8, r8, LSR #16 +celt_pitch_xcorr_edsp_process1u_loop4_done + ADDS r12, r12, #4 +celt_pitch_xcorr_edsp_process1u_loop1 + LDRGEH r6, [r4], #2 + ; Stall + SMLABBGE r14, r6, r8, r14 ; sum = MAC16_16(sum, *x, *y) + SUBGES r12, r12, #1 + LDRGTH r8, [r5], #2 + BGT celt_pitch_xcorr_edsp_process1u_loop1 + ; Restore _x + SUB r4, r4, r3, LSL #1 + ; Restore and advance _y + SUB r5, r5, r3, LSL #1 + ; maxcorr = max(maxcorr, sum) + CMP r0, r14 + ADD r5, r5, #2 + MOVLT r0, r14 + SUBS r1, r1, #1 + ; xcorr[i] = sum + STR r14, [r2], #4 + BLE celt_pitch_xcorr_edsp_done +celt_pitch_xcorr_edsp_process1u_done + ; if (max_pitch < 4) goto celt_pitch_xcorr_edsp_process2 + SUBS r1, r1, #4 + BLT celt_pitch_xcorr_edsp_process2 +celt_pitch_xcorr_edsp_process4 + ; xcorr_kernel_edsp parameters: + ; r3 = len, r4 = _x, r5 = _y, r6...r9 = sum[4] = {0, 0, 0, 0} + MOV r6, #0 + MOV r7, #0 + MOV r8, #0 + MOV r9, #0 + BL xcorr_kernel_edsp ; xcorr_kernel_edsp(_x, _y+i, xcorr+i, len) + ; maxcorr = max(maxcorr, sum0, sum1, sum2, sum3) + CMP r0, r6 + ; _y+=4 + ADD r5, r5, #8 + MOVLT r0, r6 + CMP r0, r7 + MOVLT r0, r7 + CMP r0, r8 + MOVLT r0, r8 + CMP r0, r9 + MOVLT r0, r9 + STMIA r2!, {r6-r9} + SUBS r1, r1, #4 + BGE celt_pitch_xcorr_edsp_process4 +celt_pitch_xcorr_edsp_process2 + ADDS r1, r1, #2 + BLT celt_pitch_xcorr_edsp_process1a + SUBS r12, r3, #4 + ; {r10, r11} = {sum0, sum1} = {0, 0} + MOV r10, #0 + MOV r11, #0 + LDR r8, [r5], #4 + BLE celt_pitch_xcorr_edsp_process2_loop_done + LDR r6, [r4], #4 + LDR r9, [r5], #4 +celt_pitch_xcorr_edsp_process2_loop4 + SMLABB r10, r6, r8, r10 ; sum0 = MAC16_16(sum0, x_0, y_0) + LDR r7, [r4], #4 + SMLABT r11, r6, r8, r11 ; sum1 = MAC16_16(sum1, x_0, y_1) + SUBS r12, r12, #4 ; j-=4 + SMLATT r10, r6, r8, r10 ; sum0 = MAC16_16(sum0, x_1, y_1) + LDR r8, [r5], #4 + SMLATB r11, r6, r9, r11 ; sum1 = MAC16_16(sum1, x_1, y_2) + LDRGT r6, [r4], #4 + SMLABB r10, r7, r9, r10 ; sum0 = MAC16_16(sum0, x_2, y_2) + SMLABT r11, r7, r9, r11 ; sum1 = MAC16_16(sum1, x_2, y_3) + SMLATT r10, r7, r9, r10 ; sum0 = MAC16_16(sum0, x_3, y_3) + LDRGT r9, [r5], #4 + SMLATB r11, r7, r8, r11 ; sum1 = MAC16_16(sum1, x_3, y_4) + BGT celt_pitch_xcorr_edsp_process2_loop4 +celt_pitch_xcorr_edsp_process2_loop_done + ADDS r12, r12, #2 + BLE celt_pitch_xcorr_edsp_process2_1 + LDR r6, [r4], #4 + ; Stall + SMLABB r10, r6, r8, r10 ; sum0 = MAC16_16(sum0, x_0, y_0) + LDR r9, [r5], #4 + SMLABT r11, r6, r8, r11 ; sum1 = MAC16_16(sum1, x_0, y_1) + SUB r12, r12, #2 + SMLATT r10, r6, r8, r10 ; sum0 = MAC16_16(sum0, x_1, y_1) + MOV r8, r9 + SMLATB r11, r6, r9, r11 ; sum1 = MAC16_16(sum1, x_1, y_2) +celt_pitch_xcorr_edsp_process2_1 + LDRH r6, [r4], #2 + ADDS r12, r12, #1 + ; Stall + SMLABB r10, r6, r8, r10 ; sum0 = MAC16_16(sum0, x_0, y_0) + LDRGTH r7, [r4], #2 + SMLABT r11, r6, r8, r11 ; sum1 = MAC16_16(sum1, x_0, y_1) + BLE celt_pitch_xcorr_edsp_process2_done + LDRH r9, [r5], #2 + SMLABT r10, r7, r8, r10 ; sum0 = MAC16_16(sum0, x_0, y_1) + SMLABB r11, r7, r9, r11 ; sum1 = MAC16_16(sum1, x_0, y_2) +celt_pitch_xcorr_edsp_process2_done + ; Restore _x + SUB r4, r4, r3, LSL #1 + ; Restore and advance _y + SUB r5, r5, r3, LSL #1 + ; maxcorr = max(maxcorr, sum0) + CMP r0, r10 + ADD r5, r5, #2 + MOVLT r0, r10 + SUB r1, r1, #2 + ; maxcorr = max(maxcorr, sum1) + CMP r0, r11 + ; xcorr[i] = sum + STR r10, [r2], #4 + MOVLT r0, r11 + STR r11, [r2], #4 +celt_pitch_xcorr_edsp_process1a + ADDS r1, r1, #1 + BLT celt_pitch_xcorr_edsp_done + SUBS r12, r3, #4 + ; r14 = sum = 0 + MOV r14, #0 + BLT celt_pitch_xcorr_edsp_process1a_loop_done + LDR r6, [r4], #4 + LDR r8, [r5], #4 + LDR r7, [r4], #4 + LDR r9, [r5], #4 +celt_pitch_xcorr_edsp_process1a_loop4 + SMLABB r14, r6, r8, r14 ; sum = MAC16_16(sum, x_0, y_0) + SUBS r12, r12, #4 ; j-=4 + SMLATT r14, r6, r8, r14 ; sum = MAC16_16(sum, x_1, y_1) + LDRGE r6, [r4], #4 + SMLABB r14, r7, r9, r14 ; sum = MAC16_16(sum, x_2, y_2) + LDRGE r8, [r5], #4 + SMLATT r14, r7, r9, r14 ; sum = MAC16_16(sum, x_3, y_3) + LDRGE r7, [r4], #4 + LDRGE r9, [r5], #4 + BGE celt_pitch_xcorr_edsp_process1a_loop4 +celt_pitch_xcorr_edsp_process1a_loop_done + ADDS r12, r12, #2 + LDRGE r6, [r4], #4 + LDRGE r8, [r5], #4 + ; Stall + SMLABBGE r14, r6, r8, r14 ; sum = MAC16_16(sum, x_0, y_0) + SUBGE r12, r12, #2 + SMLATTGE r14, r6, r8, r14 ; sum = MAC16_16(sum, x_1, y_1) + ADDS r12, r12, #1 + LDRGEH r6, [r4], #2 + LDRGEH r8, [r5], #2 + ; Stall + SMLABBGE r14, r6, r8, r14 ; sum = MAC16_16(sum, *x, *y) + ; maxcorr = max(maxcorr, sum) + CMP r0, r14 + ; xcorr[i] = sum + STR r14, [r2], #4 + MOVLT r0, r14 +celt_pitch_xcorr_edsp_done + LDMFD sp!, {r4-r11, pc} + ENDP + +ENDIF + +END diff --git a/drivers/opus/celt/arm/fixed_armv4.h b/drivers/opus/celt/arm/fixed_armv4.h new file mode 100644 index 0000000000..b690bc8cea --- /dev/null +++ b/drivers/opus/celt/arm/fixed_armv4.h @@ -0,0 +1,76 @@ +/* Copyright (C) 2013 Xiph.Org Foundation and contributors */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef FIXED_ARMv4_H +#define FIXED_ARMv4_H + +/** 16x32 multiplication, followed by a 16-bit shift right. Results fits in 32 bits */ +#undef MULT16_32_Q16 +static OPUS_INLINE opus_val32 MULT16_32_Q16_armv4(opus_val16 a, opus_val32 b) +{ + unsigned rd_lo; + int rd_hi; + __asm__( + "#MULT16_32_Q16\n\t" + "smull %0, %1, %2, %3\n\t" + : "=&r"(rd_lo), "=&r"(rd_hi) + : "%r"(b),"r"(a<<16) + ); + return rd_hi; +} +#define MULT16_32_Q16(a, b) (MULT16_32_Q16_armv4(a, b)) + + +/** 16x32 multiplication, followed by a 15-bit shift right. Results fits in 32 bits */ +#undef MULT16_32_Q15 +static OPUS_INLINE opus_val32 MULT16_32_Q15_armv4(opus_val16 a, opus_val32 b) +{ + unsigned rd_lo; + int rd_hi; + __asm__( + "#MULT16_32_Q15\n\t" + "smull %0, %1, %2, %3\n\t" + : "=&r"(rd_lo), "=&r"(rd_hi) + : "%r"(b), "r"(a<<16) + ); + /*We intentionally don't OR in the high bit of rd_lo for speed.*/ + return rd_hi<<1; +} +#define MULT16_32_Q15(a, b) (MULT16_32_Q15_armv4(a, b)) + + +/** 16x32 multiply, followed by a 15-bit shift right and 32-bit add. + b must fit in 31 bits. + Result fits in 32 bits. */ +#undef MAC16_32_Q15 +#define MAC16_32_Q15(c, a, b) ADD32(c, MULT16_32_Q15(a, b)) + + +/** 32x32 multiplication, followed by a 31-bit shift right. Results fits in 32 bits */ +#undef MULT32_32_Q31 +#define MULT32_32_Q31(a,b) (opus_val32)((((opus_int64)(a)) * ((opus_int64)(b)))>>31) + +#endif diff --git a/drivers/opus/celt/arm/fixed_armv5e.h b/drivers/opus/celt/arm/fixed_armv5e.h new file mode 100644 index 0000000000..1194a7d3ec --- /dev/null +++ b/drivers/opus/celt/arm/fixed_armv5e.h @@ -0,0 +1,116 @@ +/* Copyright (C) 2007-2009 Xiph.Org Foundation + Copyright (C) 2003-2008 Jean-Marc Valin + Copyright (C) 2007-2008 CSIRO + 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. +*/ + +#ifndef FIXED_ARMv5E_H +#define FIXED_ARMv5E_H + +#include "fixed_armv4.h" + +/** 16x32 multiplication, followed by a 16-bit shift right. Results fits in 32 bits */ +#undef MULT16_32_Q16 +static OPUS_INLINE opus_val32 MULT16_32_Q16_armv5e(opus_val16 a, opus_val32 b) +{ + int res; + __asm__( + "#MULT16_32_Q16\n\t" + "smulwb %0, %1, %2\n\t" + : "=r"(res) + : "r"(b),"r"(a) + ); + return res; +} +#define MULT16_32_Q16(a, b) (MULT16_32_Q16_armv5e(a, b)) + + +/** 16x32 multiplication, followed by a 15-bit shift right. Results fits in 32 bits */ +#undef MULT16_32_Q15 +static OPUS_INLINE opus_val32 MULT16_32_Q15_armv5e(opus_val16 a, opus_val32 b) +{ + int res; + __asm__( + "#MULT16_32_Q15\n\t" + "smulwb %0, %1, %2\n\t" + : "=r"(res) + : "r"(b), "r"(a) + ); + return res<<1; +} +#define MULT16_32_Q15(a, b) (MULT16_32_Q15_armv5e(a, b)) + + +/** 16x32 multiply, followed by a 15-bit shift right and 32-bit add. + b must fit in 31 bits. + Result fits in 32 bits. */ +#undef MAC16_32_Q15 +static OPUS_INLINE opus_val32 MAC16_32_Q15_armv5e(opus_val32 c, opus_val16 a, + opus_val32 b) +{ + int res; + __asm__( + "#MAC16_32_Q15\n\t" + "smlawb %0, %1, %2, %3;\n" + : "=r"(res) + : "r"(b<<1), "r"(a), "r"(c) + ); + return res; +} +#define MAC16_32_Q15(c, a, b) (MAC16_32_Q15_armv5e(c, a, b)) + +/** 16x16 multiply-add where the result fits in 32 bits */ +#undef MAC16_16 +static OPUS_INLINE opus_val32 MAC16_16_armv5e(opus_val32 c, opus_val16 a, + opus_val16 b) +{ + int res; + __asm__( + "#MAC16_16\n\t" + "smlabb %0, %1, %2, %3;\n" + : "=r"(res) + : "r"(a), "r"(b), "r"(c) + ); + return res; +} +#define MAC16_16(c, a, b) (MAC16_16_armv5e(c, a, b)) + +/** 16x16 multiplication where the result fits in 32 bits */ +#undef MULT16_16 +static OPUS_INLINE opus_val32 MULT16_16_armv5e(opus_val16 a, opus_val16 b) +{ + int res; + __asm__( + "#MULT16_16\n\t" + "smulbb %0, %1, %2;\n" + : "=r"(res) + : "r"(a), "r"(b) + ); + return res; +} +#define MULT16_16(a, b) (MULT16_16_armv5e(a, b)) + +#endif diff --git a/drivers/opus/celt/arm/kiss_fft_armv4.h b/drivers/opus/celt/arm/kiss_fft_armv4.h new file mode 100644 index 0000000000..773464628b --- /dev/null +++ b/drivers/opus/celt/arm/kiss_fft_armv4.h @@ -0,0 +1,121 @@ +/*Copyright (c) 2013, Xiph.Org Foundation and contributors. + + All rights reserved. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are met: + + * Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above copyright notice, + this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" + AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE + LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + POSSIBILITY OF SUCH DAMAGE.*/ + +#ifndef KISS_FFT_ARMv4_H +#define KISS_FFT_ARMv4_H + +#if !defined(KISS_FFT_GUTS_H) +#error "This file should only be included from _kiss_fft_guts.h" +#endif + +#ifdef OPUS_FIXED_POINT + +#undef C_MUL +#define C_MUL(m,a,b) \ + do{ \ + int br__; \ + int bi__; \ + int tt__; \ + __asm__ __volatile__( \ + "#C_MUL\n\t" \ + "ldrsh %[br], [%[bp], #0]\n\t" \ + "ldm %[ap], {r0,r1}\n\t" \ + "ldrsh %[bi], [%[bp], #2]\n\t" \ + "smull %[tt], %[mi], r1, %[br]\n\t" \ + "smlal %[tt], %[mi], r0, %[bi]\n\t" \ + "rsb %[bi], %[bi], #0\n\t" \ + "smull %[br], %[mr], r0, %[br]\n\t" \ + "mov %[tt], %[tt], lsr #15\n\t" \ + "smlal %[br], %[mr], r1, %[bi]\n\t" \ + "orr %[mi], %[tt], %[mi], lsl #17\n\t" \ + "mov %[br], %[br], lsr #15\n\t" \ + "orr %[mr], %[br], %[mr], lsl #17\n\t" \ + : [mr]"=r"((m).r), [mi]"=r"((m).i), \ + [br]"=&r"(br__), [bi]"=r"(bi__), [tt]"=r"(tt__) \ + : [ap]"r"(&(a)), [bp]"r"(&(b)) \ + : "r0", "r1" \ + ); \ + } \ + while(0) + +#undef C_MUL4 +#define C_MUL4(m,a,b) \ + do{ \ + int br__; \ + int bi__; \ + int tt__; \ + __asm__ __volatile__( \ + "#C_MUL4\n\t" \ + "ldrsh %[br], [%[bp], #0]\n\t" \ + "ldm %[ap], {r0,r1}\n\t" \ + "ldrsh %[bi], [%[bp], #2]\n\t" \ + "smull %[tt], %[mi], r1, %[br]\n\t" \ + "smlal %[tt], %[mi], r0, %[bi]\n\t" \ + "rsb %[bi], %[bi], #0\n\t" \ + "smull %[br], %[mr], r0, %[br]\n\t" \ + "mov %[tt], %[tt], lsr #17\n\t" \ + "smlal %[br], %[mr], r1, %[bi]\n\t" \ + "orr %[mi], %[tt], %[mi], lsl #15\n\t" \ + "mov %[br], %[br], lsr #17\n\t" \ + "orr %[mr], %[br], %[mr], lsl #15\n\t" \ + : [mr]"=r"((m).r), [mi]"=r"((m).i), \ + [br]"=&r"(br__), [bi]"=r"(bi__), [tt]"=r"(tt__) \ + : [ap]"r"(&(a)), [bp]"r"(&(b)) \ + : "r0", "r1" \ + ); \ + } \ + while(0) + +#undef C_MULC +#define C_MULC(m,a,b) \ + do{ \ + int br__; \ + int bi__; \ + int tt__; \ + __asm__ __volatile__( \ + "#C_MULC\n\t" \ + "ldrsh %[br], [%[bp], #0]\n\t" \ + "ldm %[ap], {r0,r1}\n\t" \ + "ldrsh %[bi], [%[bp], #2]\n\t" \ + "smull %[tt], %[mr], r0, %[br]\n\t" \ + "smlal %[tt], %[mr], r1, %[bi]\n\t" \ + "rsb %[bi], %[bi], #0\n\t" \ + "smull %[br], %[mi], r1, %[br]\n\t" \ + "mov %[tt], %[tt], lsr #15\n\t" \ + "smlal %[br], %[mi], r0, %[bi]\n\t" \ + "orr %[mr], %[tt], %[mr], lsl #17\n\t" \ + "mov %[br], %[br], lsr #15\n\t" \ + "orr %[mi], %[br], %[mi], lsl #17\n\t" \ + : [mr]"=r"((m).r), [mi]"=r"((m).i), \ + [br]"=&r"(br__), [bi]"=r"(bi__), [tt]"=r"(tt__) \ + : [ap]"r"(&(a)), [bp]"r"(&(b)) \ + : "r0", "r1" \ + ); \ + } \ + while(0) + +#endif /* OPUS_FIXED_POINT */ + +#endif /* KISS_FFT_ARMv4_H */ diff --git a/drivers/opus/celt/arm/kiss_fft_armv5e.h b/drivers/opus/celt/arm/kiss_fft_armv5e.h new file mode 100644 index 0000000000..1eff56a66a --- /dev/null +++ b/drivers/opus/celt/arm/kiss_fft_armv5e.h @@ -0,0 +1,118 @@ +/*Copyright (c) 2013, Xiph.Org Foundation and contributors. + + All rights reserved. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are met: + + * Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above copyright notice, + this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" + AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE + LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + POSSIBILITY OF SUCH DAMAGE.*/ + +#ifndef KISS_FFT_ARMv5E_H +#define KISS_FFT_ARMv5E_H + +#if !defined(KISS_FFT_GUTS_H) +#error "This file should only be included from _kiss_fft_guts.h" +#endif + +#ifdef OPUS_FIXED_POINT + +#if defined(__thumb__)||defined(__thumb2__) +#define LDRD_CONS "Q" +#else +#define LDRD_CONS "Uq" +#endif + +#undef C_MUL +#define C_MUL(m,a,b) \ + do{ \ + int mr1__; \ + int mr2__; \ + int mi__; \ + long long aval__; \ + int bval__; \ + __asm__( \ + "#C_MUL\n\t" \ + "ldrd %[aval], %H[aval], %[ap]\n\t" \ + "ldr %[bval], %[bp]\n\t" \ + "smulwb %[mi], %H[aval], %[bval]\n\t" \ + "smulwb %[mr1], %[aval], %[bval]\n\t" \ + "smulwt %[mr2], %H[aval], %[bval]\n\t" \ + "smlawt %[mi], %[aval], %[bval], %[mi]\n\t" \ + : [mr1]"=r"(mr1__), [mr2]"=r"(mr2__), [mi]"=r"(mi__), \ + [aval]"=&r"(aval__), [bval]"=r"(bval__) \ + : [ap]LDRD_CONS(a), [bp]"m"(b) \ + ); \ + (m).r = SHL32(SUB32(mr1__, mr2__), 1); \ + (m).i = SHL32(mi__, 1); \ + } \ + while(0) + +#undef C_MUL4 +#define C_MUL4(m,a,b) \ + do{ \ + int mr1__; \ + int mr2__; \ + int mi__; \ + long long aval__; \ + int bval__; \ + __asm__( \ + "#C_MUL4\n\t" \ + "ldrd %[aval], %H[aval], %[ap]\n\t" \ + "ldr %[bval], %[bp]\n\t" \ + "smulwb %[mi], %H[aval], %[bval]\n\t" \ + "smulwb %[mr1], %[aval], %[bval]\n\t" \ + "smulwt %[mr2], %H[aval], %[bval]\n\t" \ + "smlawt %[mi], %[aval], %[bval], %[mi]\n\t" \ + : [mr1]"=r"(mr1__), [mr2]"=r"(mr2__), [mi]"=r"(mi__), \ + [aval]"=&r"(aval__), [bval]"=r"(bval__) \ + : [ap]LDRD_CONS(a), [bp]"m"(b) \ + ); \ + (m).r = SHR32(SUB32(mr1__, mr2__), 1); \ + (m).i = SHR32(mi__, 1); \ + } \ + while(0) + +#undef C_MULC +#define C_MULC(m,a,b) \ + do{ \ + int mr__; \ + int mi1__; \ + int mi2__; \ + long long aval__; \ + int bval__; \ + __asm__( \ + "#C_MULC\n\t" \ + "ldrd %[aval], %H[aval], %[ap]\n\t" \ + "ldr %[bval], %[bp]\n\t" \ + "smulwb %[mr], %[aval], %[bval]\n\t" \ + "smulwb %[mi1], %H[aval], %[bval]\n\t" \ + "smulwt %[mi2], %[aval], %[bval]\n\t" \ + "smlawt %[mr], %H[aval], %[bval], %[mr]\n\t" \ + : [mr]"=r"(mr__), [mi1]"=r"(mi1__), [mi2]"=r"(mi2__), \ + [aval]"=&r"(aval__), [bval]"=r"(bval__) \ + : [ap]LDRD_CONS(a), [bp]"m"(b) \ + ); \ + (m).r = SHL32(mr__, 1); \ + (m).i = SHL32(SUB32(mi1__, mi2__), 1); \ + } \ + while(0) + +#endif /* OPUS_FIXED_POINT */ + +#endif /* KISS_FFT_GUTS_H */ diff --git a/drivers/opus/celt/arm/pitch_arm.h b/drivers/opus/celt/arm/pitch_arm.h new file mode 100644 index 0000000000..df5e82ef0b --- /dev/null +++ b/drivers/opus/celt/arm/pitch_arm.h @@ -0,0 +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(OPUS_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 diff --git a/drivers/opus/celt/bands.c b/drivers/opus/celt/bands.c new file mode 100644 index 0000000000..87280c8333 --- /dev/null +++ b/drivers/opus/celt/bands.c @@ -0,0 +1,1518 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Copyright (c) 2008-2009 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef OPUS_HAVE_CONFIG_H +#include "opus_config.h" +#endif + +#include <math.h> +#include "bands.h" +#include "opus_modes.h" +#include "vq.h" +#include "cwrs.h" +#include "stack_alloc.h" +#include "os_support.h" +#include "mathops.h" +#include "rate.h" +#include "quant_bands.h" +#include "pitch.h" + +int hysteresis_decision(opus_val16 val, const opus_val16 *thresholds, const opus_val16 *hysteresis, int N, int prev) +{ + int i; + for (i=0;i<N;i++) + { + if (val < thresholds[i]) + break; + } + if (i>prev && val < thresholds[prev]+hysteresis[prev]) + i=prev; + if (i<prev && val > thresholds[prev-1]-hysteresis[prev-1]) + i=prev; + return i; +} + +opus_uint32 celt_lcg_rand(opus_uint32 seed) +{ + return 1664525 * seed + 1013904223; +} + +/* This is a cos() approximation designed to be bit-exact on any platform. Bit exactness + with this approximation is important because it has an impact on the bit allocation */ +static opus_int16 bitexact_cos(opus_int16 x) +{ + opus_int32 tmp; + opus_int16 x2; + tmp = (4096+((opus_int32)(x)*(x)))>>13; + celt_assert(tmp<=32767); + x2 = tmp; + x2 = (32767-x2) + FRAC_MUL16(x2, (-7651 + FRAC_MUL16(x2, (8277 + FRAC_MUL16(-626, x2))))); + celt_assert(x2<=32766); + return 1+x2; +} + +static int bitexact_log2tan(int isin,int icos) +{ + int lc; + int ls; + lc=EC_ILOG(icos); + ls=EC_ILOG(isin); + icos<<=15-lc; + isin<<=15-ls; + return (ls-lc)*(1<<11) + +FRAC_MUL16(isin, FRAC_MUL16(isin, -2597) + 7932) + -FRAC_MUL16(icos, FRAC_MUL16(icos, -2597) + 7932); +} + +#ifdef OPUS_FIXED_POINT +/* Compute the amplitude (sqrt energy) in each of the bands */ +void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int M) +{ + int i, c, N; + const opus_int16 *eBands = m->eBands; + N = M*m->shortMdctSize; + c=0; do { + for (i=0;i<end;i++) + { + int j; + opus_val32 maxval=0; + opus_val32 sum = 0; + + j=M*eBands[i]; do { + maxval = MAX32(maxval, X[j+c*N]); + maxval = MAX32(maxval, -X[j+c*N]); + } while (++j<M*eBands[i+1]); + + if (maxval > 0) + { + int shift = celt_ilog2(maxval)-10; + j=M*eBands[i]; do { + sum = MAC16_16(sum, EXTRACT16(VSHR32(X[j+c*N],shift)), + EXTRACT16(VSHR32(X[j+c*N],shift))); + } while (++j<M*eBands[i+1]); + /* We're adding one here to ensure the normalized band isn't larger than unity norm */ + bandE[i+c*m->nbEBands] = EPSILON+VSHR32(EXTEND32(celt_sqrt(sum)),-shift); + } else { + bandE[i+c*m->nbEBands] = EPSILON; + } + /*printf ("%f ", bandE[i+c*m->nbEBands]);*/ + } + } while (++c<C); + /*printf ("\n");*/ +} + +/* Normalise each band such that the energy is one. */ +void normalise_bands(const CELTMode *m, const celt_sig * OPUS_RESTRICT freq, celt_norm * OPUS_RESTRICT X, const celt_ener *bandE, int end, int C, int M) +{ + int i, c, N; + const opus_int16 *eBands = m->eBands; + N = M*m->shortMdctSize; + c=0; do { + i=0; do { + opus_val16 g; + int j,shift; + opus_val16 E; + shift = celt_zlog2(bandE[i+c*m->nbEBands])-13; + E = VSHR32(bandE[i+c*m->nbEBands], shift); + g = EXTRACT16(celt_rcp(SHL32(E,3))); + j=M*eBands[i]; do { + X[j+c*N] = MULT16_16_Q15(VSHR32(freq[j+c*N],shift-1),g); + } while (++j<M*eBands[i+1]); + } while (++i<end); + } while (++c<C); +} + +#else /* OPUS_FIXED_POINT */ +/* Compute the amplitude (sqrt energy) in each of the bands */ +void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int M) +{ + int i, c, N; + const opus_int16 *eBands = m->eBands; + N = M*m->shortMdctSize; + c=0; do { + for (i=0;i<end;i++) + { + int j; + opus_val32 sum = 1e-27f; + for (j=M*eBands[i];j<M*eBands[i+1];j++) + sum += X[j+c*N]*X[j+c*N]; + bandE[i+c*m->nbEBands] = celt_sqrt(sum); + /*printf ("%f ", bandE[i+c*m->nbEBands]);*/ + } + } while (++c<C); + /*printf ("\n");*/ +} + +/* Normalise each band such that the energy is one. */ +void normalise_bands(const CELTMode *m, const celt_sig * OPUS_RESTRICT freq, celt_norm * OPUS_RESTRICT X, const celt_ener *bandE, int end, int C, int M) +{ + int i, c, N; + const opus_int16 *eBands = m->eBands; + N = M*m->shortMdctSize; + c=0; do { + for (i=0;i<end;i++) + { + int j; + opus_val16 g = 1.f/(1e-27f+bandE[i+c*m->nbEBands]); + for (j=M*eBands[i];j<M*eBands[i+1];j++) + X[j+c*N] = freq[j+c*N]*g; + } + } while (++c<C); +} + +#endif /* OPUS_FIXED_POINT */ + +/* De-normalise the energy to produce the synthesis from the unit-energy bands */ +void denormalise_bands(const CELTMode *m, const celt_norm * OPUS_RESTRICT X, + celt_sig * OPUS_RESTRICT freq, const opus_val16 *bandLogE, int start, int end, int C, int M) +{ + int i, c, N; + const opus_int16 *eBands = m->eBands; + N = M*m->shortMdctSize; + celt_assert2(C<=2, "denormalise_bands() not implemented for >2 channels"); + c=0; do { + celt_sig * OPUS_RESTRICT f; + const celt_norm * OPUS_RESTRICT x; + f = freq+c*N; + x = X+c*N+M*eBands[start]; + for (i=0;i<M*eBands[start];i++) + *f++ = 0; + for (i=start;i<end;i++) + { + int j, band_end; + opus_val16 g; + opus_val16 lg; +#ifdef OPUS_FIXED_POINT + int shift; +#endif + j=M*eBands[i]; + band_end = M*eBands[i+1]; + lg = ADD16(bandLogE[i+c*m->nbEBands], SHL16((opus_val16)eMeans[i],6)); +#ifndef OPUS_FIXED_POINT + g = celt_exp2(lg); +#else + /* Handle the integer part of the log energy */ + shift = 16-(lg>>DB_SHIFT); + if (shift>31) + { + shift=0; + g=0; + } else { + /* Handle the fractional part. */ + g = celt_exp2_frac(lg&((1<<DB_SHIFT)-1)); + } + /* Handle extreme gains with negative shift. */ + if (shift<0) + { + /* For shift < -2 we'd be likely to overflow, so we're capping + the gain here. This shouldn't happen unless the bitstream is + already corrupted. */ + if (shift < -2) + { + g = 32767; + shift = -2; + } + do { + *f++ = SHL32(MULT16_16(*x++, g), -shift); + } while (++j<band_end); + } else +#endif + /* Be careful of the fixed-point "else" just above when changing this code */ + do { + *f++ = SHR32(MULT16_16(*x++, g), shift); + } while (++j<band_end); + } + celt_assert(start <= end); + for (i=M*eBands[end];i<N;i++) + *f++ = 0; + } while (++c<C); +} + +/* This prevents energy collapse for transients with multiple short MDCTs */ +void anti_collapse(const CELTMode *m, celt_norm *X_, unsigned char *collapse_masks, int LM, int C, int size, + int start, int end, opus_val16 *logE, opus_val16 *prev1logE, + opus_val16 *prev2logE, int *pulses, opus_uint32 seed) +{ + int c, i, j, k; + for (i=start;i<end;i++) + { + int N0; + opus_val16 thresh, sqrt_1; + int depth; +#ifdef OPUS_FIXED_POINT + int shift; + opus_val32 thresh32; +#endif + + N0 = m->eBands[i+1]-m->eBands[i]; + /* depth in 1/8 bits */ + depth = (1+pulses[i])/((m->eBands[i+1]-m->eBands[i])<<LM); + +#ifdef OPUS_FIXED_POINT + thresh32 = SHR32(celt_exp2(-SHL16(depth, 10-BITRES)),1); + thresh = MULT16_32_Q15(QCONST16(0.5f, 15), MIN32(32767,thresh32)); + { + opus_val32 t; + t = N0<<LM; + shift = celt_ilog2(t)>>1; + t = SHL32(t, (7-shift)<<1); + sqrt_1 = celt_rsqrt_norm(t); + } +#else + thresh = .5f*celt_exp2(-.125f*depth); + sqrt_1 = celt_rsqrt(N0<<LM); +#endif + + c=0; do + { + celt_norm *X; + opus_val16 prev1; + opus_val16 prev2; + opus_val32 Ediff; + opus_val16 r; + int renormalize=0; + prev1 = prev1logE[c*m->nbEBands+i]; + prev2 = prev2logE[c*m->nbEBands+i]; + if (C==1) + { + prev1 = MAX16(prev1,prev1logE[m->nbEBands+i]); + prev2 = MAX16(prev2,prev2logE[m->nbEBands+i]); + } + Ediff = EXTEND32(logE[c*m->nbEBands+i])-EXTEND32(MIN16(prev1,prev2)); + Ediff = MAX32(0, Ediff); + +#ifdef OPUS_FIXED_POINT + if (Ediff < 16384) + { + opus_val32 r32 = SHR32(celt_exp2(-EXTRACT16(Ediff)),1); + r = 2*MIN16(16383,r32); + } else { + r = 0; + } + if (LM==3) + r = MULT16_16_Q14(23170, MIN32(23169, r)); + r = SHR16(MIN16(thresh, r),1); + r = SHR32(MULT16_16_Q15(sqrt_1, r),shift); +#else + /* r needs to be multiplied by 2 or 2*sqrt(2) depending on LM because + short blocks don't have the same energy as long */ + r = 2.f*celt_exp2(-Ediff); + if (LM==3) + r *= 1.41421356f; + r = MIN16(thresh, r); + r = r*sqrt_1; +#endif + X = X_+c*size+(m->eBands[i]<<LM); + for (k=0;k<1<<LM;k++) + { + /* Detect collapse */ + if (!(collapse_masks[i*C+c]&1<<k)) + { + /* Fill with noise */ + for (j=0;j<N0;j++) + { + seed = celt_lcg_rand(seed); + X[(j<<LM)+k] = (seed&0x8000 ? r : -r); + } + renormalize = 1; + } + } + /* We just added some energy, so we need to renormalise */ + if (renormalize) + renormalise_vector(X, N0<<LM, Q15ONE); + } while (++c<C); + } +} + +static void intensity_stereo(const CELTMode *m, celt_norm *X, celt_norm *Y, const celt_ener *bandE, int bandID, int N) +{ + int i = bandID; + int j; + opus_val16 a1, a2; + opus_val16 left, right; + opus_val16 norm; +#ifdef OPUS_FIXED_POINT + int shift = celt_zlog2(MAX32(bandE[i], bandE[i+m->nbEBands]))-13; +#endif + left = VSHR32(bandE[i],shift); + right = VSHR32(bandE[i+m->nbEBands],shift); + norm = EPSILON + celt_sqrt(EPSILON+MULT16_16(left,left)+MULT16_16(right,right)); + a1 = DIV32_16(SHL32(EXTEND32(left),14),norm); + a2 = DIV32_16(SHL32(EXTEND32(right),14),norm); + for (j=0;j<N;j++) + { + celt_norm r, l; + l = X[j]; + r = Y[j]; + X[j] = MULT16_16_Q14(a1,l) + MULT16_16_Q14(a2,r); + /* Side is not encoded, no need to calculate */ + } +} + +static void stereo_split(celt_norm *X, celt_norm *Y, int N) +{ + int j; + for (j=0;j<N;j++) + { + celt_norm r, l; + l = MULT16_16_Q15(QCONST16(.70710678f,15), X[j]); + r = MULT16_16_Q15(QCONST16(.70710678f,15), Y[j]); + X[j] = l+r; + Y[j] = r-l; + } +} + +static void stereo_merge(celt_norm *X, celt_norm *Y, opus_val16 mid, int N) +{ + int j; + opus_val32 xp=0, side=0; + opus_val32 El, Er; + opus_val16 mid2; +#ifdef OPUS_FIXED_POINT + int kl, kr; +#endif + opus_val32 t, lgain, rgain; + + /* Compute the norm of X+Y and X-Y as |X|^2 + |Y|^2 +/- sum(xy) */ + dual_inner_prod(Y, X, Y, N, &xp, &side); + /* Compensating for the mid normalization */ + xp = MULT16_32_Q15(mid, xp); + /* mid and side are in Q15, not Q14 like X and Y */ + mid2 = SHR32(mid, 1); + El = MULT16_16(mid2, mid2) + side - 2*xp; + Er = MULT16_16(mid2, mid2) + side + 2*xp; + if (Er < QCONST32(6e-4f, 28) || El < QCONST32(6e-4f, 28)) + { + for (j=0;j<N;j++) + Y[j] = X[j]; + return; + } + +#ifdef OPUS_FIXED_POINT + kl = celt_ilog2(El)>>1; + kr = celt_ilog2(Er)>>1; +#endif + t = VSHR32(El, (kl-7)<<1); + lgain = celt_rsqrt_norm(t); + t = VSHR32(Er, (kr-7)<<1); + rgain = celt_rsqrt_norm(t); + +#ifdef OPUS_FIXED_POINT + if (kl < 7) + kl = 7; + if (kr < 7) + kr = 7; +#endif + + for (j=0;j<N;j++) + { + celt_norm r, l; + /* Apply mid scaling (side is already scaled) */ + l = MULT16_16_Q15(mid, X[j]); + r = Y[j]; + X[j] = EXTRACT16(PSHR32(MULT16_16(lgain, SUB16(l,r)), kl+1)); + Y[j] = EXTRACT16(PSHR32(MULT16_16(rgain, ADD16(l,r)), kr+1)); + } +} + +/* Decide whether we should spread the pulses in the current frame */ +int spreading_decision(const CELTMode *m, celt_norm *X, int *average, + int last_decision, int *hf_average, int *tapset_decision, int update_hf, + int end, int C, int M) +{ + int i, c, N0; + int sum = 0, nbBands=0; + const opus_int16 * OPUS_RESTRICT eBands = m->eBands; + int decision; + int hf_sum=0; + + celt_assert(end>0); + + N0 = M*m->shortMdctSize; + + if (M*(eBands[end]-eBands[end-1]) <= 8) + return SPREAD_NONE; + c=0; do { + for (i=0;i<end;i++) + { + int j, N, tmp=0; + int tcount[3] = {0,0,0}; + celt_norm * OPUS_RESTRICT x = X+M*eBands[i]+c*N0; + N = M*(eBands[i+1]-eBands[i]); + if (N<=8) + continue; + /* Compute rough CDF of |x[j]| */ + for (j=0;j<N;j++) + { + opus_val32 x2N; /* Q13 */ + + x2N = MULT16_16(MULT16_16_Q15(x[j], x[j]), N); + if (x2N < QCONST16(0.25f,13)) + tcount[0]++; + if (x2N < QCONST16(0.0625f,13)) + tcount[1]++; + if (x2N < QCONST16(0.015625f,13)) + tcount[2]++; + } + + /* Only include four last bands (8 kHz and up) */ + if (i>m->nbEBands-4) + hf_sum += 32*(tcount[1]+tcount[0])/N; + tmp = (2*tcount[2] >= N) + (2*tcount[1] >= N) + (2*tcount[0] >= N); + sum += tmp*256; + nbBands++; + } + } while (++c<C); + + if (update_hf) + { + if (hf_sum) + hf_sum /= C*(4-m->nbEBands+end); + *hf_average = (*hf_average+hf_sum)>>1; + hf_sum = *hf_average; + if (*tapset_decision==2) + hf_sum += 4; + else if (*tapset_decision==0) + hf_sum -= 4; + if (hf_sum > 22) + *tapset_decision=2; + else if (hf_sum > 18) + *tapset_decision=1; + else + *tapset_decision=0; + } + /*printf("%d %d %d\n", hf_sum, *hf_average, *tapset_decision);*/ + celt_assert(nbBands>0); /* end has to be non-zero */ + sum /= nbBands; + /* Recursive averaging */ + sum = (sum+*average)>>1; + *average = sum; + /* Hysteresis */ + sum = (3*sum + (((3-last_decision)<<7) + 64) + 2)>>2; + if (sum < 80) + { + decision = SPREAD_AGGRESSIVE; + } else if (sum < 256) + { + decision = SPREAD_NORMAL; + } else if (sum < 384) + { + decision = SPREAD_LIGHT; + } else { + decision = SPREAD_NONE; + } +#ifdef FUZZING + decision = rand()&0x3; + *tapset_decision=rand()%3; +#endif + return decision; +} + +/* Indexing table for converting from natural Hadamard to ordery Hadamard + This is essentially a bit-reversed Gray, on top of which we've added + an inversion of the order because we want the DC at the end rather than + the beginning. The lines are for N=2, 4, 8, 16 */ +static const int ordery_table[] = { + 1, 0, + 3, 0, 2, 1, + 7, 0, 4, 3, 6, 1, 5, 2, + 15, 0, 8, 7, 12, 3, 11, 4, 14, 1, 9, 6, 13, 2, 10, 5, +}; + +static void deinterleave_hadamard(celt_norm *X, int N0, int stride, int hadamard) +{ + int i,j; + VARDECL(celt_norm, tmp); + int N; + SAVE_STACK; + N = N0*stride; + ALLOC(tmp, N, celt_norm); + celt_assert(stride>0); + if (hadamard) + { + const int *ordery = ordery_table+stride-2; + for (i=0;i<stride;i++) + { + for (j=0;j<N0;j++) + tmp[ordery[i]*N0+j] = X[j*stride+i]; + } + } else { + for (i=0;i<stride;i++) + for (j=0;j<N0;j++) + tmp[i*N0+j] = X[j*stride+i]; + } + for (j=0;j<N;j++) + X[j] = tmp[j]; + RESTORE_STACK; +} + +static void interleave_hadamard(celt_norm *X, int N0, int stride, int hadamard) +{ + int i,j; + VARDECL(celt_norm, tmp); + int N; + SAVE_STACK; + N = N0*stride; + ALLOC(tmp, N, celt_norm); + if (hadamard) + { + const int *ordery = ordery_table+stride-2; + for (i=0;i<stride;i++) + for (j=0;j<N0;j++) + tmp[j*stride+i] = X[ordery[i]*N0+j]; + } else { + for (i=0;i<stride;i++) + for (j=0;j<N0;j++) + tmp[j*stride+i] = X[i*N0+j]; + } + for (j=0;j<N;j++) + X[j] = tmp[j]; + RESTORE_STACK; +} + +void haar1(celt_norm *X, int N0, int stride) +{ + int i, j; + N0 >>= 1; + for (i=0;i<stride;i++) + for (j=0;j<N0;j++) + { + celt_norm tmp1, tmp2; + tmp1 = MULT16_16_Q15(QCONST16(.70710678f,15), X[stride*2*j+i]); + tmp2 = MULT16_16_Q15(QCONST16(.70710678f,15), X[stride*(2*j+1)+i]); + X[stride*2*j+i] = tmp1 + tmp2; + X[stride*(2*j+1)+i] = tmp1 - tmp2; + } +} + +static int compute_qn(int N, int b, int offset, int pulse_cap, int stereo) +{ + static const opus_int16 exp2_table8[8] = + {16384, 17866, 19483, 21247, 23170, 25267, 27554, 30048}; + int qn, qb; + int N2 = 2*N-1; + if (stereo && N==2) + N2--; + /* The upper limit ensures that in a stereo split with itheta==16384, we'll + always have enough bits left over to code at least one pulse in the + side; otherwise it would collapse, since it doesn't get folded. */ + qb = IMIN(b-pulse_cap-(4<<BITRES), (b+N2*offset)/N2); + + qb = IMIN(8<<BITRES, qb); + + if (qb<(1<<BITRES>>1)) { + qn = 1; + } else { + qn = exp2_table8[qb&0x7]>>(14-(qb>>BITRES)); + qn = (qn+1)>>1<<1; + } + celt_assert(qn <= 256); + return qn; +} + +struct band_ctx { + int encode; + const CELTMode *m; + int i; + int intensity; + int spread; + int tf_change; + ec_ctx *ec; + opus_int32 remaining_bits; + const celt_ener *bandE; + opus_uint32 seed; +}; + +struct split_ctx { + int inv; + int imid; + int iside; + int delta; + int itheta; + int qalloc; +}; + +static void compute_theta(struct band_ctx *ctx, struct split_ctx *sctx, + celt_norm *X, celt_norm *Y, int N, int *b, int B, int B0, + int LM, + int stereo, int *fill) +{ + int qn; + int itheta=0; + int delta; + int imid, iside; + int qalloc; + int pulse_cap; + int offset; + opus_int32 tell; + int inv=0; + int encode; + const CELTMode *m; + int i; + int intensity; + ec_ctx *ec; + const celt_ener *bandE; + + encode = ctx->encode; + m = ctx->m; + i = ctx->i; + intensity = ctx->intensity; + ec = ctx->ec; + bandE = ctx->bandE; + + /* Decide on the resolution to give to the split parameter theta */ + pulse_cap = m->logN[i]+LM*(1<<BITRES); + offset = (pulse_cap>>1) - (stereo&&N==2 ? QTHETA_OFFSET_TWOPHASE : QTHETA_OFFSET); + qn = compute_qn(N, *b, offset, pulse_cap, stereo); + if (stereo && i>=intensity) + qn = 1; + if (encode) + { + /* theta is the atan() of the ratio between the (normalized) + side and mid. With just that parameter, we can re-scale both + mid and side because we know that 1) they have unit norm and + 2) they are orthogonal. */ + itheta = stereo_itheta(X, Y, stereo, N); + } + tell = ec_tell_frac(ec); + if (qn!=1) + { + if (encode) + itheta = (itheta*qn+8192)>>14; + + /* Entropy coding of the angle. We use a uniform pdf for the + time split, a step for stereo, and a triangular one for the rest. */ + if (stereo && N>2) + { + int p0 = 3; + int x = itheta; + int x0 = qn/2; + int ft = p0*(x0+1) + x0; + /* Use a probability of p0 up to itheta=8192 and then use 1 after */ + if (encode) + { + ec_encode(ec,x<=x0?p0*x:(x-1-x0)+(x0+1)*p0,x<=x0?p0*(x+1):(x-x0)+(x0+1)*p0,ft); + } else { + int fs; + fs=ec_decode(ec,ft); + if (fs<(x0+1)*p0) + x=fs/p0; + else + x=x0+1+(fs-(x0+1)*p0); + ec_dec_update(ec,x<=x0?p0*x:(x-1-x0)+(x0+1)*p0,x<=x0?p0*(x+1):(x-x0)+(x0+1)*p0,ft); + itheta = x; + } + } else if (B0>1 || stereo) { + /* Uniform pdf */ + if (encode) + ec_enc_uint(ec, itheta, qn+1); + else + itheta = ec_dec_uint(ec, qn+1); + } else { + int fs=1, ft; + ft = ((qn>>1)+1)*((qn>>1)+1); + if (encode) + { + int fl; + + fs = itheta <= (qn>>1) ? itheta + 1 : qn + 1 - itheta; + fl = itheta <= (qn>>1) ? itheta*(itheta + 1)>>1 : + ft - ((qn + 1 - itheta)*(qn + 2 - itheta)>>1); + + ec_encode(ec, fl, fl+fs, ft); + } else { + /* Triangular pdf */ + int fl=0; + int fm; + fm = ec_decode(ec, ft); + + if (fm < ((qn>>1)*((qn>>1) + 1)>>1)) + { + itheta = (isqrt32(8*(opus_uint32)fm + 1) - 1)>>1; + fs = itheta + 1; + fl = itheta*(itheta + 1)>>1; + } + else + { + itheta = (2*(qn + 1) + - isqrt32(8*(opus_uint32)(ft - fm - 1) + 1))>>1; + fs = qn + 1 - itheta; + fl = ft - ((qn + 1 - itheta)*(qn + 2 - itheta)>>1); + } + + ec_dec_update(ec, fl, fl+fs, ft); + } + } + itheta = (opus_int32)itheta*16384/qn; + if (encode && stereo) + { + if (itheta==0) + intensity_stereo(m, X, Y, bandE, i, N); + else + stereo_split(X, Y, N); + } + /* NOTE: Renormalising X and Y *may* help fixed-point a bit at very high rate. + Let's do that at higher complexity */ + } else if (stereo) { + if (encode) + { + inv = itheta > 8192; + if (inv) + { + int j; + for (j=0;j<N;j++) + Y[j] = -Y[j]; + } + intensity_stereo(m, X, Y, bandE, i, N); + } + if (*b>2<<BITRES && ctx->remaining_bits > 2<<BITRES) + { + if (encode) + ec_enc_bit_logp(ec, inv, 2); + else + inv = ec_dec_bit_logp(ec, 2); + } else + inv = 0; + itheta = 0; + } + qalloc = ec_tell_frac(ec) - tell; + *b -= qalloc; + + if (itheta == 0) + { + imid = 32767; + iside = 0; + *fill &= (1<<B)-1; + delta = -16384; + } else if (itheta == 16384) + { + imid = 0; + iside = 32767; + *fill &= ((1<<B)-1)<<B; + delta = 16384; + } else { + imid = bitexact_cos((opus_int16)itheta); + iside = bitexact_cos((opus_int16)(16384-itheta)); + /* This is the mid vs side allocation that minimizes squared error + in that band. */ + delta = FRAC_MUL16((N-1)<<7,bitexact_log2tan(iside,imid)); + } + + sctx->inv = inv; + sctx->imid = imid; + sctx->iside = iside; + sctx->delta = delta; + sctx->itheta = itheta; + sctx->qalloc = qalloc; +} +static unsigned quant_band_n1(struct band_ctx *ctx, celt_norm *X, celt_norm *Y, int b, + celt_norm *lowband_out) +{ +#ifdef RESYNTH + int resynth = 1; +#else + int resynth = !ctx->encode; +#endif + int c; + int stereo; + celt_norm *x = X; + int encode; + ec_ctx *ec; + + encode = ctx->encode; + ec = ctx->ec; + + stereo = Y != NULL; + c=0; do { + int sign=0; + if (ctx->remaining_bits>=1<<BITRES) + { + if (encode) + { + sign = x[0]<0; + ec_enc_bits(ec, sign, 1); + } else { + sign = ec_dec_bits(ec, 1); + } + ctx->remaining_bits -= 1<<BITRES; + b-=1<<BITRES; + } + if (resynth) + x[0] = sign ? -NORM_SCALING : NORM_SCALING; + x = Y; + } while (++c<1+stereo); + if (lowband_out) + lowband_out[0] = SHR16(X[0],4); + return 1; +} + +/* This function is responsible for encoding and decoding a mono partition. + It can split the band in two and transmit the energy difference with + the two half-bands. It can be called recursively so bands can end up being + split in 8 parts. */ +static unsigned quant_partition(struct band_ctx *ctx, celt_norm *X, + int N, int b, int B, celt_norm *lowband, + int LM, + opus_val16 gain, int fill) +{ + const unsigned char *cache; + int q; + int curr_bits; + int imid=0, iside=0; + int B0=B; + opus_val16 mid=0, side=0; + unsigned cm=0; +#ifdef RESYNTH + int resynth = 1; +#else + int resynth = !ctx->encode; +#endif + celt_norm *Y=NULL; + int encode; + const CELTMode *m; + int i; + int spread; + ec_ctx *ec; + + encode = ctx->encode; + m = ctx->m; + i = ctx->i; + spread = ctx->spread; + ec = ctx->ec; + + /* If we need 1.5 more bit than we can produce, split the band in two. */ + cache = m->cache.bits + m->cache.index[(LM+1)*m->nbEBands+i]; + if (LM != -1 && b > cache[cache[0]]+12 && N>2) + { + int mbits, sbits, delta; + int itheta; + int qalloc; + struct split_ctx sctx; + celt_norm *next_lowband2=NULL; + opus_int32 rebalance; + + N >>= 1; + Y = X+N; + LM -= 1; + if (B==1) + fill = (fill&1)|(fill<<1); + B = (B+1)>>1; + + compute_theta(ctx, &sctx, X, Y, N, &b, B, B0, + LM, 0, &fill); + imid = sctx.imid; + iside = sctx.iside; + delta = sctx.delta; + itheta = sctx.itheta; + qalloc = sctx.qalloc; +#ifdef OPUS_FIXED_POINT + mid = imid; + side = iside; +#else + mid = (1.f/32768)*imid; + side = (1.f/32768)*iside; +#endif + + /* Give more bits to low-energy MDCTs than they would otherwise deserve */ + if (B0>1 && (itheta&0x3fff)) + { + if (itheta > 8192) + /* Rough approximation for pre-echo masking */ + delta -= delta>>(4-LM); + else + /* Corresponds to a forward-masking slope of 1.5 dB per 10 ms */ + delta = IMIN(0, delta + (N<<BITRES>>(5-LM))); + } + mbits = IMAX(0, IMIN(b, (b-delta)/2)); + sbits = b-mbits; + ctx->remaining_bits -= qalloc; + + if (lowband) + next_lowband2 = lowband+N; /* >32-bit split case */ + + rebalance = ctx->remaining_bits; + if (mbits >= sbits) + { + cm = quant_partition(ctx, X, N, mbits, B, + lowband, LM, + MULT16_16_P15(gain,mid), fill); + rebalance = mbits - (rebalance-ctx->remaining_bits); + if (rebalance > 3<<BITRES && itheta!=0) + sbits += rebalance - (3<<BITRES); + cm |= quant_partition(ctx, Y, N, sbits, B, + next_lowband2, LM, + MULT16_16_P15(gain,side), fill>>B)<<(B0>>1); + } else { + cm = quant_partition(ctx, Y, N, sbits, B, + next_lowband2, LM, + MULT16_16_P15(gain,side), fill>>B)<<(B0>>1); + rebalance = sbits - (rebalance-ctx->remaining_bits); + if (rebalance > 3<<BITRES && itheta!=16384) + mbits += rebalance - (3<<BITRES); + cm |= quant_partition(ctx, X, N, mbits, B, + lowband, LM, + MULT16_16_P15(gain,mid), fill); + } + } else { + /* This is the basic no-split case */ + q = bits2pulses(m, i, LM, b); + curr_bits = pulses2bits(m, i, LM, q); + ctx->remaining_bits -= curr_bits; + + /* Ensures we can never bust the budget */ + while (ctx->remaining_bits < 0 && q > 0) + { + ctx->remaining_bits += curr_bits; + q--; + curr_bits = pulses2bits(m, i, LM, q); + ctx->remaining_bits -= curr_bits; + } + + if (q!=0) + { + int K = get_pulses(q); + + /* Finally do the actual quantization */ + if (encode) + { + cm = alg_quant(X, N, K, spread, B, ec +#ifdef RESYNTH + , gain +#endif + ); + } else { + cm = alg_unquant(X, N, K, spread, B, ec, gain); + } + } else { + /* If there's no pulse, fill the band anyway */ + int j; + if (resynth) + { + unsigned cm_mask; + /* B can be as large as 16, so this shift might overflow an int on a + 16-bit platform; use a long to get defined behavior.*/ + cm_mask = (unsigned)(1UL<<B)-1; + fill &= cm_mask; + if (!fill) + { + for (j=0;j<N;j++) + X[j] = 0; + } else { + if (lowband == NULL) + { + /* Noise */ + for (j=0;j<N;j++) + { + ctx->seed = celt_lcg_rand(ctx->seed); + X[j] = (celt_norm)((opus_int32)ctx->seed>>20); + } + cm = cm_mask; + } else { + /* Folded spectrum */ + for (j=0;j<N;j++) + { + opus_val16 tmp; + ctx->seed = celt_lcg_rand(ctx->seed); + /* About 48 dB below the "normal" folding level */ + tmp = QCONST16(1.0f/256, 10); + tmp = (ctx->seed)&0x8000 ? tmp : -tmp; + X[j] = lowband[j]+tmp; + } + cm = fill; + } + renormalise_vector(X, N, gain); + } + } + } + } + + return cm; +} + + +/* This function is responsible for encoding and decoding a band for the mono case. */ +static unsigned quant_band(struct band_ctx *ctx, celt_norm *X, + int N, int b, int B, celt_norm *lowband, + int LM, celt_norm *lowband_out, + opus_val16 gain, celt_norm *lowband_scratch, int fill) +{ + int N0=N; + int N_B=N; + int N_B0; + int B0=B; + int time_divide=0; + int recombine=0; + int longBlocks; + unsigned cm=0; +#ifdef RESYNTH + int resynth = 1; +#else + int resynth = !ctx->encode; +#endif + int k; + int encode; + int tf_change; + + encode = ctx->encode; + tf_change = ctx->tf_change; + + longBlocks = B0==1; + + N_B /= B; + + /* Special case for one sample */ + if (N==1) + { + return quant_band_n1(ctx, X, NULL, b, lowband_out); + } + + if (tf_change>0) + recombine = tf_change; + /* Band recombining to increase frequency resolution */ + + if (lowband_scratch && lowband && (recombine || ((N_B&1) == 0 && tf_change<0) || B0>1)) + { + int j; + for (j=0;j<N;j++) + lowband_scratch[j] = lowband[j]; + lowband = lowband_scratch; + } + + for (k=0;k<recombine;k++) + { + static const unsigned char bit_interleave_table[16]={ + 0,1,1,1,2,3,3,3,2,3,3,3,2,3,3,3 + }; + if (encode) + haar1(X, N>>k, 1<<k); + if (lowband) + haar1(lowband, N>>k, 1<<k); + fill = bit_interleave_table[fill&0xF]|bit_interleave_table[fill>>4]<<2; + } + B>>=recombine; + N_B<<=recombine; + + /* Increasing the time resolution */ + while ((N_B&1) == 0 && tf_change<0) + { + if (encode) + haar1(X, N_B, B); + if (lowband) + haar1(lowband, N_B, B); + fill |= fill<<B; + B <<= 1; + N_B >>= 1; + time_divide++; + tf_change++; + } + B0=B; + N_B0 = N_B; + + /* Reorganize the samples in time order instead of frequency order */ + if (B0>1) + { + if (encode) + deinterleave_hadamard(X, N_B>>recombine, B0<<recombine, longBlocks); + if (lowband) + deinterleave_hadamard(lowband, N_B>>recombine, B0<<recombine, longBlocks); + } + + cm = quant_partition(ctx, X, N, b, B, lowband, + LM, gain, fill); + + /* This code is used by the decoder and by the resynthesis-enabled encoder */ + if (resynth) + { + /* Undo the sample reorganization going from time order to frequency order */ + if (B0>1) + interleave_hadamard(X, N_B>>recombine, B0<<recombine, longBlocks); + + /* Undo time-freq changes that we did earlier */ + N_B = N_B0; + B = B0; + for (k=0;k<time_divide;k++) + { + B >>= 1; + N_B <<= 1; + cm |= cm>>B; + haar1(X, N_B, B); + } + + for (k=0;k<recombine;k++) + { + static const unsigned char bit_deinterleave_table[16]={ + 0x00,0x03,0x0C,0x0F,0x30,0x33,0x3C,0x3F, + 0xC0,0xC3,0xCC,0xCF,0xF0,0xF3,0xFC,0xFF + }; + cm = bit_deinterleave_table[cm]; + haar1(X, N0>>k, 1<<k); + } + B<<=recombine; + + /* Scale output for later folding */ + if (lowband_out) + { + int j; + opus_val16 n; + n = celt_sqrt(SHL32(EXTEND32(N0),22)); + for (j=0;j<N0;j++) + lowband_out[j] = MULT16_16_Q15(n,X[j]); + } + cm &= (1<<B)-1; + } + return cm; +} + + +/* This function is responsible for encoding and decoding a band for the stereo case. */ +static unsigned quant_band_stereo(struct band_ctx *ctx, celt_norm *X, celt_norm *Y, + int N, int b, int B, celt_norm *lowband, + int LM, celt_norm *lowband_out, + celt_norm *lowband_scratch, int fill) +{ + int imid=0, iside=0; + int inv = 0; + opus_val16 mid=0, side=0; + unsigned cm=0; +#ifdef RESYNTH + int resynth = 1; +#else + int resynth = !ctx->encode; +#endif + int mbits, sbits, delta; + int itheta; + int qalloc; + struct split_ctx sctx; + int orig_fill; + int encode; + ec_ctx *ec; + + encode = ctx->encode; + ec = ctx->ec; + + /* Special case for one sample */ + if (N==1) + { + return quant_band_n1(ctx, X, Y, b, lowband_out); + } + + orig_fill = fill; + + compute_theta(ctx, &sctx, X, Y, N, &b, B, B, + LM, 1, &fill); + inv = sctx.inv; + imid = sctx.imid; + iside = sctx.iside; + delta = sctx.delta; + itheta = sctx.itheta; + qalloc = sctx.qalloc; +#ifdef OPUS_FIXED_POINT + mid = imid; + side = iside; +#else + mid = (1.f/32768)*imid; + side = (1.f/32768)*iside; +#endif + + /* This is a special case for N=2 that only works for stereo and takes + advantage of the fact that mid and side are orthogonal to encode + the side with just one bit. */ + if (N==2) + { + int c; + int sign=0; + celt_norm *x2, *y2; + mbits = b; + sbits = 0; + /* Only need one bit for the side. */ + if (itheta != 0 && itheta != 16384) + sbits = 1<<BITRES; + mbits -= sbits; + c = itheta > 8192; + ctx->remaining_bits -= qalloc+sbits; + + x2 = c ? Y : X; + y2 = c ? X : Y; + if (sbits) + { + if (encode) + { + /* Here we only need to encode a sign for the side. */ + sign = x2[0]*y2[1] - x2[1]*y2[0] < 0; + ec_enc_bits(ec, sign, 1); + } else { + sign = ec_dec_bits(ec, 1); + } + } + sign = 1-2*sign; + /* We use orig_fill here because we want to fold the side, but if + itheta==16384, we'll have cleared the low bits of fill. */ + cm = quant_band(ctx, x2, N, mbits, B, lowband, + LM, lowband_out, Q15ONE, lowband_scratch, orig_fill); + /* We don't split N=2 bands, so cm is either 1 or 0 (for a fold-collapse), + and there's no need to worry about mixing with the other channel. */ + y2[0] = -sign*x2[1]; + y2[1] = sign*x2[0]; + if (resynth) + { + celt_norm tmp; + X[0] = MULT16_16_Q15(mid, X[0]); + X[1] = MULT16_16_Q15(mid, X[1]); + Y[0] = MULT16_16_Q15(side, Y[0]); + Y[1] = MULT16_16_Q15(side, Y[1]); + tmp = X[0]; + X[0] = SUB16(tmp,Y[0]); + Y[0] = ADD16(tmp,Y[0]); + tmp = X[1]; + X[1] = SUB16(tmp,Y[1]); + Y[1] = ADD16(tmp,Y[1]); + } + } else { + /* "Normal" split code */ + opus_int32 rebalance; + + mbits = IMAX(0, IMIN(b, (b-delta)/2)); + sbits = b-mbits; + ctx->remaining_bits -= qalloc; + + rebalance = ctx->remaining_bits; + if (mbits >= sbits) + { + /* In stereo mode, we do not apply a scaling to the mid because we need the normalized + mid for folding later. */ + cm = quant_band(ctx, X, N, mbits, B, + lowband, LM, lowband_out, + Q15ONE, lowband_scratch, fill); + rebalance = mbits - (rebalance-ctx->remaining_bits); + if (rebalance > 3<<BITRES && itheta!=0) + sbits += rebalance - (3<<BITRES); + + /* For a stereo split, the high bits of fill are always zero, so no + folding will be done to the side. */ + cm |= quant_band(ctx, Y, N, sbits, B, + NULL, LM, NULL, + side, NULL, fill>>B); + } else { + /* For a stereo split, the high bits of fill are always zero, so no + folding will be done to the side. */ + cm = quant_band(ctx, Y, N, sbits, B, + NULL, LM, NULL, + side, NULL, fill>>B); + rebalance = sbits - (rebalance-ctx->remaining_bits); + if (rebalance > 3<<BITRES && itheta!=16384) + mbits += rebalance - (3<<BITRES); + /* In stereo mode, we do not apply a scaling to the mid because we need the normalized + mid for folding later. */ + cm |= quant_band(ctx, X, N, mbits, B, + lowband, LM, lowband_out, + Q15ONE, lowband_scratch, fill); + } + } + + + /* This code is used by the decoder and by the resynthesis-enabled encoder */ + if (resynth) + { + if (N!=2) + stereo_merge(X, Y, mid, N); + if (inv) + { + int j; + for (j=0;j<N;j++) + Y[j] = -Y[j]; + } + } + return cm; +} + + +void quant_all_bands(int encode, const CELTMode *m, int start, int end, + celt_norm *X_, celt_norm *Y_, unsigned char *collapse_masks, const celt_ener *bandE, int *pulses, + int shortBlocks, int spread, int dual_stereo, int intensity, int *tf_res, + opus_int32 total_bits, opus_int32 balance, ec_ctx *ec, int LM, int codedBands, opus_uint32 *seed) +{ + int i; + opus_int32 remaining_bits; + const opus_int16 * OPUS_RESTRICT eBands = m->eBands; + celt_norm * OPUS_RESTRICT norm, * OPUS_RESTRICT norm2; + VARDECL(celt_norm, _norm); + celt_norm *lowband_scratch; + int B; + int M; + int lowband_offset; + int update_lowband = 1; + int C = Y_ != NULL ? 2 : 1; + int norm_offset; +#ifdef RESYNTH + int resynth = 1; +#else + int resynth = !encode; +#endif + struct band_ctx ctx; + SAVE_STACK; + + M = 1<<LM; + B = shortBlocks ? M : 1; + norm_offset = M*eBands[start]; + /* No need to allocate norm for the last band because we don't need an + output in that band. */ + ALLOC(_norm, C*(M*eBands[m->nbEBands-1]-norm_offset), celt_norm); + norm = _norm; + norm2 = norm + M*eBands[m->nbEBands-1]-norm_offset; + /* We can use the last band as scratch space because we don't need that + scratch space for the last band. */ + lowband_scratch = X_+M*eBands[m->nbEBands-1]; + + lowband_offset = 0; + ctx.bandE = bandE; + ctx.ec = ec; + ctx.encode = encode; + ctx.intensity = intensity; + ctx.m = m; + ctx.seed = *seed; + ctx.spread = spread; + for (i=start;i<end;i++) + { + opus_int32 tell; + int b; + int N; + opus_int32 curr_balance; + int effective_lowband=-1; + celt_norm * OPUS_RESTRICT X, * OPUS_RESTRICT Y; + int tf_change=0; + unsigned x_cm; + unsigned y_cm; + int last; + + ctx.i = i; + last = (i==end-1); + + X = X_+M*eBands[i]; + if (Y_!=NULL) + Y = Y_+M*eBands[i]; + else + Y = NULL; + N = M*eBands[i+1]-M*eBands[i]; + tell = ec_tell_frac(ec); + + /* Compute how many bits we want to allocate to this band */ + if (i != start) + balance -= tell; + remaining_bits = total_bits-tell-1; + ctx.remaining_bits = remaining_bits; + if (i <= codedBands-1) + { + curr_balance = balance / IMIN(3, codedBands-i); + b = IMAX(0, IMIN(16383, IMIN(remaining_bits+1,pulses[i]+curr_balance))); + } else { + b = 0; + } + + if (resynth && M*eBands[i]-N >= M*eBands[start] && (update_lowband || lowband_offset==0)) + lowband_offset = i; + + tf_change = tf_res[i]; + ctx.tf_change = tf_change; + if (i>=m->effEBands) + { + X=norm; + if (Y_!=NULL) + Y = norm; + lowband_scratch = NULL; + } + if (i==end-1) + lowband_scratch = NULL; + + /* Get a conservative estimate of the collapse_mask's for the bands we're + going to be folding from. */ + if (lowband_offset != 0 && (spread!=SPREAD_AGGRESSIVE || B>1 || tf_change<0)) + { + int fold_start; + int fold_end; + int fold_i; + /* This ensures we never repeat spectral content within one band */ + effective_lowband = IMAX(0, M*eBands[lowband_offset]-norm_offset-N); + fold_start = lowband_offset; + while(M*eBands[--fold_start] > effective_lowband+norm_offset); + fold_end = lowband_offset-1; + while(M*eBands[++fold_end] < effective_lowband+norm_offset+N); + x_cm = y_cm = 0; + fold_i = fold_start; do { + x_cm |= collapse_masks[fold_i*C+0]; + y_cm |= collapse_masks[fold_i*C+C-1]; + } while (++fold_i<fold_end); + } + /* Otherwise, we'll be using the LCG to fold, so all blocks will (almost + always) be non-zero. */ + else + x_cm = y_cm = (1<<B)-1; + + if (dual_stereo && i==intensity) + { + int j; + + /* Switch off dual stereo to do intensity. */ + dual_stereo = 0; + if (resynth) + for (j=0;j<M*eBands[i]-norm_offset;j++) + norm[j] = HALF32(norm[j]+norm2[j]); + } + if (dual_stereo) + { + x_cm = quant_band(&ctx, X, N, b/2, B, + effective_lowband != -1 ? norm+effective_lowband : NULL, LM, + last?NULL:norm+M*eBands[i]-norm_offset, Q15ONE, lowband_scratch, x_cm); + y_cm = quant_band(&ctx, Y, N, b/2, B, + effective_lowband != -1 ? norm2+effective_lowband : NULL, LM, + last?NULL:norm2+M*eBands[i]-norm_offset, Q15ONE, lowband_scratch, y_cm); + } else { + if (Y!=NULL) + { + x_cm = quant_band_stereo(&ctx, X, Y, N, b, B, + effective_lowband != -1 ? norm+effective_lowband : NULL, LM, + last?NULL:norm+M*eBands[i]-norm_offset, lowband_scratch, x_cm|y_cm); + } else { + x_cm = quant_band(&ctx, X, N, b, B, + effective_lowband != -1 ? norm+effective_lowband : NULL, LM, + last?NULL:norm+M*eBands[i]-norm_offset, Q15ONE, lowband_scratch, x_cm|y_cm); + } + y_cm = x_cm; + } + collapse_masks[i*C+0] = (unsigned char)x_cm; + collapse_masks[i*C+C-1] = (unsigned char)y_cm; + balance += pulses[i] + tell; + + /* Update the folding position only as long as we have 1 bit/sample depth. */ + update_lowband = b>(N<<BITRES); + } + *seed = ctx.seed; + + RESTORE_STACK; +} + diff --git a/drivers/opus/celt/bands.h b/drivers/opus/celt/bands.h new file mode 100644 index 0000000000..fe1e47097a --- /dev/null +++ b/drivers/opus/celt/bands.h @@ -0,0 +1,114 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Copyright (c) 2008-2009 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef BANDS_H +#define BANDS_H + +#include "arch.h" +#include "opus_modes.h" +#include "entenc.h" +#include "entdec.h" +#include "rate.h" + +/** Compute the amplitude (sqrt energy) in each of the bands + * @param m Mode data + * @param X Spectrum + * @param bandE Square root of the energy for each band (returned) + */ +void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int M); + +/*void compute_noise_energies(const CELTMode *m, const celt_sig *X, const opus_val16 *tonality, celt_ener *bandE);*/ + +/** Normalise each band of X such that the energy in each band is + equal to 1 + * @param m Mode data + * @param X Spectrum (returned normalised) + * @param bandE Square root of the energy for each band + */ +void normalise_bands(const CELTMode *m, const celt_sig * OPUS_RESTRICT freq, celt_norm * OPUS_RESTRICT X, const celt_ener *bandE, int end, int C, int M); + +/** Denormalise each band of X to restore full amplitude + * @param m Mode data + * @param X Spectrum (returned de-normalised) + * @param bandE Square root of the energy for each band + */ +void denormalise_bands(const CELTMode *m, const celt_norm * OPUS_RESTRICT X, + celt_sig * OPUS_RESTRICT freq, const opus_val16 *bandE, int start, int end, int C, int M); + +#define SPREAD_NONE (0) +#define SPREAD_LIGHT (1) +#define SPREAD_NORMAL (2) +#define SPREAD_AGGRESSIVE (3) + +int spreading_decision(const CELTMode *m, celt_norm *X, int *average, + int last_decision, int *hf_average, int *tapset_decision, int update_hf, + int end, int C, int M); + +#ifdef MEASURE_NORM_MSE +void measure_norm_mse(const CELTMode *m, float *X, float *X0, float *bandE, float *bandE0, int M, int N, int C); +#endif + +void haar1(celt_norm *X, int N0, int stride); + +/** Quantisation/encoding of the residual spectrum + * @param encode flag that indicates whether we're encoding (1) or decoding (0) + * @param m Mode data + * @param start First band to process + * @param end Last band to process + 1 + * @param X Residual (normalised) + * @param Y Residual (normalised) for second channel (or NULL for mono) + * @param collapse_masks Anti-collapse tracking mask + * @param bandE Square root of the energy for each band + * @param pulses Bit allocation (per band) for PVQ + * @param shortBlocks Zero for long blocks, non-zero for short blocks + * @param spread Amount of spreading to use + * @param dual_stereo Zero for MS stereo, non-zero for dual stereo + * @param intensity First band to use intensity stereo + * @param tf_res Time-frequency resolution change + * @param total_bits Total number of bits that can be used for the frame (including the ones already spent) + * @param balance Number of unallocated bits + * @param en Entropy coder state + * @param LM log2() of the number of 2.5 subframes in the frame + * @param codedBands Last band to receive bits + 1 + * @param seed Random generator seed + */ +void quant_all_bands(int encode, const CELTMode *m, int start, int end, + celt_norm * X, celt_norm * Y, unsigned char *collapse_masks, const celt_ener *bandE, int *pulses, + int shortBlocks, int spread, int dual_stereo, int intensity, int *tf_res, + opus_int32 total_bits, opus_int32 balance, ec_ctx *ec, int M, int codedBands, opus_uint32 *seed); + +void anti_collapse(const CELTMode *m, celt_norm *X_, unsigned char *collapse_masks, int LM, int C, int size, + int start, int end, opus_val16 *logE, opus_val16 *prev1logE, + opus_val16 *prev2logE, int *pulses, opus_uint32 seed); + +opus_uint32 celt_lcg_rand(opus_uint32 seed); + +int hysteresis_decision(opus_val16 val, const opus_val16 *thresholds, const opus_val16 *hysteresis, int N, int prev); + +#endif /* BANDS_H */ diff --git a/drivers/opus/celt/celt.c b/drivers/opus/celt/celt.c new file mode 100644 index 0000000000..b894e1e13f --- /dev/null +++ b/drivers/opus/celt/celt.c @@ -0,0 +1,223 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2010 Xiph.Org Foundation + Copyright (c) 2008 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef OPUS_HAVE_CONFIG_H +#include "opus_config.h" +#endif + +#define CELT_C + +#include "os_support.h" +#include "mdct.h" +#include <math.h> +#include "celt.h" +#include "pitch.h" +#include "bands.h" +#include "opus_modes.h" +#include "entcode.h" +#include "quant_bands.h" +#include "rate.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "float_cast.h" +#include <stdarg.h> +#include "celt_lpc.h" +#include "vq.h" + +#ifndef PACKAGE_VERSION +#define PACKAGE_VERSION "unknown" +#endif + + +int resampling_factor(opus_int32 rate) +{ + int ret; + switch (rate) + { + case 48000: + ret = 1; + break; + case 24000: + ret = 2; + break; + case 16000: + ret = 3; + break; + case 12000: + ret = 4; + break; + case 8000: + ret = 6; + break; + default: +#ifndef CUSTOM_MODES + celt_assert(0); +#endif + ret = 0; + break; + } + return ret; +} + +#ifndef OVERRIDE_COMB_FILTER_CONST +static void comb_filter_const(opus_val32 *y, opus_val32 *x, int T, int N, + opus_val16 g10, opus_val16 g11, opus_val16 g12) +{ + opus_val32 x0, x1, x2, x3, x4; + int i; + x4 = x[-T-2]; + x3 = x[-T-1]; + x2 = x[-T]; + x1 = x[-T+1]; + for (i=0;i<N;i++) + { + x0=x[i-T+2]; + y[i] = x[i] + + MULT16_32_Q15(g10,x2) + + MULT16_32_Q15(g11,ADD32(x1,x3)) + + MULT16_32_Q15(g12,ADD32(x0,x4)); + x4=x3; + x3=x2; + x2=x1; + x1=x0; + } + +} +#endif + +void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N, + opus_val16 g0, opus_val16 g1, int tapset0, int tapset1, + const opus_val16 *window, int overlap) +{ + int i; + /* printf ("%d %d %f %f\n", T0, T1, g0, g1); */ + opus_val16 g00, g01, g02, g10, g11, g12; + opus_val32 x0, x1, x2, x3, x4; + static const opus_val16 gains[3][3] = { + {QCONST16(0.3066406250f, 15), QCONST16(0.2170410156f, 15), QCONST16(0.1296386719f, 15)}, + {QCONST16(0.4638671875f, 15), QCONST16(0.2680664062f, 15), QCONST16(0.f, 15)}, + {QCONST16(0.7998046875f, 15), QCONST16(0.1000976562f, 15), QCONST16(0.f, 15)}}; + + if (g0==0 && g1==0) + { + /* OPT: Happens to work without the OPUS_MOVE(), but only because the current encoder already copies x to y */ + if (x!=y) + OPUS_MOVE(y, x, N); + return; + } + g00 = MULT16_16_Q15(g0, gains[tapset0][0]); + g01 = MULT16_16_Q15(g0, gains[tapset0][1]); + g02 = MULT16_16_Q15(g0, gains[tapset0][2]); + g10 = MULT16_16_Q15(g1, gains[tapset1][0]); + g11 = MULT16_16_Q15(g1, gains[tapset1][1]); + g12 = MULT16_16_Q15(g1, gains[tapset1][2]); + x1 = x[-T1+1]; + x2 = x[-T1 ]; + x3 = x[-T1-1]; + x4 = x[-T1-2]; + for (i=0;i<overlap;i++) + { + opus_val16 f; + x0=x[i-T1+2]; + f = MULT16_16_Q15(window[i],window[i]); + y[i] = x[i] + + MULT16_32_Q15(MULT16_16_Q15((Q15ONE-f),g00),x[i-T0]) + + MULT16_32_Q15(MULT16_16_Q15((Q15ONE-f),g01),ADD32(x[i-T0+1],x[i-T0-1])) + + MULT16_32_Q15(MULT16_16_Q15((Q15ONE-f),g02),ADD32(x[i-T0+2],x[i-T0-2])) + + MULT16_32_Q15(MULT16_16_Q15(f,g10),x2) + + MULT16_32_Q15(MULT16_16_Q15(f,g11),ADD32(x1,x3)) + + MULT16_32_Q15(MULT16_16_Q15(f,g12),ADD32(x0,x4)); + x4=x3; + x3=x2; + x2=x1; + x1=x0; + + } + if (g1==0) + { + /* OPT: Happens to work without the OPUS_MOVE(), but only because the current encoder already copies x to y */ + if (x!=y) + OPUS_MOVE(y+overlap, x+overlap, N-overlap); + return; + } + + /* Compute the part with the constant filter. */ + comb_filter_const(y+i, x+i, T1, N-i, g10, g11, g12); +} + +const signed char tf_select_table[4][8] = { + {0, -1, 0, -1, 0,-1, 0,-1}, + {0, -1, 0, -2, 1, 0, 1,-1}, + {0, -2, 0, -3, 2, 0, 1,-1}, + {0, -2, 0, -3, 3, 0, 1,-1}, +}; + + +void init_caps(const CELTMode *m,int *cap,int LM,int C) +{ + int i; + for (i=0;i<m->nbEBands;i++) + { + int N; + N=(m->eBands[i+1]-m->eBands[i])<<LM; + cap[i] = (m->cache.caps[m->nbEBands*(2*LM+C-1)+i]+64)*C*N>>2; + } +} + + + +const char *opus_strerror(int error) +{ + static const char * const error_strings[8] = { + "success", + "invalid argument", + "buffer too small", + "internal error", + "corrupted stream", + "request not implemented", + "invalid state", + "memory allocation failed" + }; + if (error > 0 || error < -7) + return "unknown error"; + else + return error_strings[-error]; +} + +const char *opus_get_version_string(void) +{ + return "libopus " PACKAGE_VERSION +#ifdef OPUS_FIXED_POINT + "-fixed" +#endif +#ifdef FUZZING + "-fuzzing" +#endif + ; +} diff --git a/drivers/opus/celt/celt.h b/drivers/opus/celt/celt.h new file mode 100644 index 0000000000..5deea1f0aa --- /dev/null +++ b/drivers/opus/celt/celt.h @@ -0,0 +1,218 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Copyright (c) 2008 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/** + @file celt.h + @brief Contains all the functions for encoding and decoding audio + */ + +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef CELT_H +#define CELT_H + +#include "opus_types.h" +#include "opus_defines.h" +#include "opus_custom.h" +#include "entenc.h" +#include "entdec.h" +#include "arch.h" + +#ifdef __cplusplus +extern "C" { +#endif + +#define CELTEncoder OpusCustomEncoder +#define CELTDecoder OpusCustomDecoder +#define CELTMode OpusCustomMode + +typedef struct { + int valid; + float tonality; + float tonality_slope; + float noisiness; + float activity; + float music_prob; + int bandwidth; +}AnalysisInfo; + +#define __celt_check_mode_ptr_ptr(ptr) ((ptr) + ((ptr) - (const CELTMode**)(ptr))) + +#define __celt_check_analysis_ptr(ptr) ((ptr) + ((ptr) - (const AnalysisInfo*)(ptr))) + +/* Encoder/decoder Requests */ + +/* Expose this option again when variable framesize actually works */ +#define OPUS_FRAMESIZE_VARIABLE 5010 /**< Optimize the frame size dynamically */ + + +#define CELT_SET_PREDICTION_REQUEST 10002 +/** Controls the use of interframe prediction. + 0=Independent frames + 1=Short term interframe prediction allowed + 2=Long term prediction allowed + */ +#define CELT_SET_PREDICTION(x) CELT_SET_PREDICTION_REQUEST, __opus_check_int(x) + +#define CELT_SET_INPUT_CLIPPING_REQUEST 10004 +#define CELT_SET_INPUT_CLIPPING(x) CELT_SET_INPUT_CLIPPING_REQUEST, __opus_check_int(x) + +#define CELT_GET_AND_CLEAR_ERROR_REQUEST 10007 +#define CELT_GET_AND_CLEAR_ERROR(x) CELT_GET_AND_CLEAR_ERROR_REQUEST, __opus_check_int_ptr(x) + +#define CELT_SET_CHANNELS_REQUEST 10008 +#define CELT_SET_CHANNELS(x) CELT_SET_CHANNELS_REQUEST, __opus_check_int(x) + + +/* Internal */ +#define CELT_SET_START_BAND_REQUEST 10010 +#define CELT_SET_START_BAND(x) CELT_SET_START_BAND_REQUEST, __opus_check_int(x) + +#define CELT_SET_END_BAND_REQUEST 10012 +#define CELT_SET_END_BAND(x) CELT_SET_END_BAND_REQUEST, __opus_check_int(x) + +#define CELT_GET_MODE_REQUEST 10015 +/** Get the CELTMode used by an encoder or decoder */ +#define CELT_GET_MODE(x) CELT_GET_MODE_REQUEST, __celt_check_mode_ptr_ptr(x) + +#define CELT_SET_SIGNALLING_REQUEST 10016 +#define CELT_SET_SIGNALLING(x) CELT_SET_SIGNALLING_REQUEST, __opus_check_int(x) + +#define CELT_SET_TONALITY_REQUEST 10018 +#define CELT_SET_TONALITY(x) CELT_SET_TONALITY_REQUEST, __opus_check_int(x) +#define CELT_SET_TONALITY_SLOPE_REQUEST 10020 +#define CELT_SET_TONALITY_SLOPE(x) CELT_SET_TONALITY_SLOPE_REQUEST, __opus_check_int(x) + +#define CELT_SET_ANALYSIS_REQUEST 10022 +#define CELT_SET_ANALYSIS(x) CELT_SET_ANALYSIS_REQUEST, __celt_check_analysis_ptr(x) + +#define OPUS_SET_LFE_REQUEST 10024 +#define OPUS_SET_LFE(x) OPUS_SET_LFE_REQUEST, __opus_check_int(x) + +#define OPUS_SET_ENERGY_MASK_REQUEST 10026 +#define OPUS_SET_ENERGY_MASK(x) OPUS_SET_ENERGY_MASK_REQUEST, __opus_check_val16_ptr(x) + +/* Encoder stuff */ + +int celt_encoder_get_size(int channels); + +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 arch); + + + +/* Decoder stuff */ + +int celt_decoder_get_size(int channels); + + +int celt_decoder_init(CELTDecoder *st, opus_int32 sampling_rate, int channels); + +int celt_decode_with_ec(OpusCustomDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, opus_val16 * OPUS_RESTRICT pcm, int frame_size, ec_dec *dec); + +#define celt_encoder_ctl opus_custom_encoder_ctl +#define celt_decoder_ctl opus_custom_decoder_ctl + + +#ifdef CUSTOM_MODES +#define OPUS_CUSTOM_NOSTATIC +#else +#define OPUS_CUSTOM_NOSTATIC static OPUS_INLINE +#endif + +static const unsigned char trim_icdf[11] = {126, 124, 119, 109, 87, 41, 19, 9, 4, 2, 0}; +/* Probs: NONE: 21.875%, LIGHT: 6.25%, NORMAL: 65.625%, AGGRESSIVE: 6.25% */ +static const unsigned char spread_icdf[4] = {25, 23, 2, 0}; + +static const unsigned char tapset_icdf[3]={2,1,0}; + +#ifdef CUSTOM_MODES +static const unsigned char toOpusTable[20] = { + 0xE0, 0xE8, 0xF0, 0xF8, + 0xC0, 0xC8, 0xD0, 0xD8, + 0xA0, 0xA8, 0xB0, 0xB8, + 0x00, 0x00, 0x00, 0x00, + 0x80, 0x88, 0x90, 0x98, +}; + +static const unsigned char fromOpusTable[16] = { + 0x80, 0x88, 0x90, 0x98, + 0x40, 0x48, 0x50, 0x58, + 0x20, 0x28, 0x30, 0x38, + 0x00, 0x08, 0x10, 0x18 +}; + +static OPUS_INLINE int toOpus(unsigned char c) +{ + int ret=0; + if (c<0xA0) + ret = toOpusTable[c>>3]; + if (ret == 0) + return -1; + else + return ret|(c&0x7); +} + +static OPUS_INLINE int fromOpus(unsigned char c) +{ + if (c<0x80) + return -1; + else + return fromOpusTable[(c>>3)-16] | (c&0x7); +} +#endif /* CUSTOM_MODES */ + +#define COMBFILTER_MAXPERIOD 1024 +#define COMBFILTER_MINPERIOD 15 + +extern const signed char tf_select_table[4][8]; + +int resampling_factor(opus_int32 rate); + +void celt_preemphasis(const opus_val16 * OPUS_RESTRICT pcmp, celt_sig * OPUS_RESTRICT inp, + int N, int CC, int upsample, const opus_val16 *coef, celt_sig *mem, int clip); + +void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N, + opus_val16 g0, opus_val16 g1, int tapset0, int tapset1, + const opus_val16 *window, int overlap); + +void init_caps(const CELTMode *m,int *cap,int LM,int C); + +#ifdef RESYNTH +void deemphasis(celt_sig *in[], opus_val16 *pcm, int N, int C, int downsample, const opus_val16 *coef, celt_sig *mem, celt_sig * OPUS_RESTRICT scratch); + +void compute_inv_mdcts(const CELTMode *mode, int shortBlocks, celt_sig *X, + celt_sig * OPUS_RESTRICT out_mem[], int C, int LM); +#endif + +#ifdef __cplusplus +} +#endif + +#endif /* CELT_H */ diff --git a/drivers/opus/celt/celt_decoder.c b/drivers/opus/celt/celt_decoder.c new file mode 100644 index 0000000000..93791feab4 --- /dev/null +++ b/drivers/opus/celt/celt_decoder.c @@ -0,0 +1,1195 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2010 Xiph.Org Foundation + Copyright (c) 2008 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef OPUS_HAVE_CONFIG_H +#include "opus_config.h" +#endif + +#define CELT_DECODER_C + +#include "cpu_support.h" +#include "os_support.h" +#include "mdct.h" +#include <math.h> +#include "celt.h" +#include "pitch.h" +#include "bands.h" +#include "opus_modes.h" +#include "entcode.h" +#include "quant_bands.h" +#include "rate.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "float_cast.h" +#include <stdarg.h> +#include "celt_lpc.h" +#include "vq.h" + +/**********************************************************************/ +/* */ +/* DECODER */ +/* */ +/**********************************************************************/ +#define DECODE_BUFFER_SIZE 2048 + +/** Decoder state + @brief Decoder state + */ +struct OpusCustomDecoder { + const OpusCustomMode *mode; + int overlap; + int channels; + int stream_channels; + + int downsample; + int start, end; + int signalling; + int arch; + + /* Everything beyond this point gets cleared on a reset */ +#define DECODER_RESET_START rng + + opus_uint32 rng; + int error; + int last_pitch_index; + int loss_count; + int postfilter_period; + int postfilter_period_old; + opus_val16 postfilter_gain; + opus_val16 postfilter_gain_old; + int postfilter_tapset; + int postfilter_tapset_old; + + celt_sig preemph_memD[2]; + + celt_sig _decode_mem[1]; /* Size = channels*(DECODE_BUFFER_SIZE+mode->overlap) */ + /* opus_val16 lpc[], Size = channels*LPC_ORDER */ + /* opus_val16 oldEBands[], Size = 2*mode->nbEBands */ + /* opus_val16 oldLogE[], Size = 2*mode->nbEBands */ + /* opus_val16 oldLogE2[], Size = 2*mode->nbEBands */ + /* opus_val16 backgroundLogE[], Size = 2*mode->nbEBands */ +}; + +int celt_decoder_get_size(int channels) +{ + const CELTMode *mode = opus_custom_mode_create(48000, 960, NULL); + return opus_custom_decoder_get_size(mode, channels); +} + +OPUS_CUSTOM_NOSTATIC int opus_custom_decoder_get_size(const CELTMode *mode, int channels) +{ + int size = sizeof(struct CELTDecoder) + + (channels*(DECODE_BUFFER_SIZE+mode->overlap)-1)*sizeof(celt_sig) + + channels*LPC_ORDER*sizeof(opus_val16) + + 4*2*mode->nbEBands*sizeof(opus_val16); + return size; +} + +#ifdef CUSTOM_MODES +CELTDecoder *opus_custom_decoder_create(const CELTMode *mode, int channels, int *error) +{ + int ret; + CELTDecoder *st = (CELTDecoder *)opus_alloc(opus_custom_decoder_get_size(mode, channels)); + ret = opus_custom_decoder_init(st, mode, channels); + if (ret != OPUS_OK) + { + opus_custom_decoder_destroy(st); + st = NULL; + } + if (error) + *error = ret; + return st; +} +#endif /* CUSTOM_MODES */ + +int celt_decoder_init(CELTDecoder *st, opus_int32 sampling_rate, int channels) +{ + int ret; + ret = opus_custom_decoder_init(st, opus_custom_mode_create(48000, 960, NULL), channels); + if (ret != OPUS_OK) + return ret; + st->downsample = resampling_factor(sampling_rate); + if (st->downsample==0) + return OPUS_BAD_ARG; + else + return OPUS_OK; +} + +OPUS_CUSTOM_NOSTATIC int opus_custom_decoder_init(CELTDecoder *st, const CELTMode *mode, int channels) +{ + if (channels < 0 || channels > 2) + return OPUS_BAD_ARG; + + if (st==NULL) + return OPUS_ALLOC_FAIL; + + OPUS_CLEAR((char*)st, opus_custom_decoder_get_size(mode, channels)); + + st->mode = mode; + st->overlap = mode->overlap; + st->stream_channels = st->channels = channels; + + st->downsample = 1; + st->start = 0; + st->end = st->mode->effEBands; + st->signalling = 1; + st->arch = opus_select_arch(); + + st->loss_count = 0; + + opus_custom_decoder_ctl(st, OPUS_RESET_STATE); + + return OPUS_OK; +} + +#ifdef CUSTOM_MODES +void opus_custom_decoder_destroy(CELTDecoder *st) +{ + opus_free(st); +} +#endif /* CUSTOM_MODES */ + +static OPUS_INLINE opus_val16 SIG2WORD16(celt_sig x) +{ +#ifdef OPUS_FIXED_POINT + x = PSHR32(x, SIG_SHIFT); + x = MAX32(x, -32768); + x = MIN32(x, 32767); + return EXTRACT16(x); +#else + return (opus_val16)x; +#endif +} + +#ifndef RESYNTH +static +#endif +void deemphasis(celt_sig *in[], opus_val16 *pcm, int N, int C, int downsample, const opus_val16 *coef, celt_sig *mem, celt_sig * OPUS_RESTRICT scratch) +{ + int c; + int Nd; + int apply_downsampling=0; + opus_val16 coef0; + + coef0 = coef[0]; + Nd = N/downsample; + c=0; do { + int j; + celt_sig * OPUS_RESTRICT x; + opus_val16 * OPUS_RESTRICT y; + celt_sig m = mem[c]; + x =in[c]; + y = pcm+c; +#ifdef CUSTOM_MODES + if (coef[1] != 0) + { + opus_val16 coef1 = coef[1]; + opus_val16 coef3 = coef[3]; + for (j=0;j<N;j++) + { + celt_sig tmp = x[j] + m + VERY_SMALL; + m = MULT16_32_Q15(coef0, tmp) + - MULT16_32_Q15(coef1, x[j]); + tmp = SHL32(MULT16_32_Q15(coef3, tmp), 2); + scratch[j] = tmp; + } + apply_downsampling=1; + } else +#endif + if (downsample>1) + { + /* Shortcut for the standard (non-custom modes) case */ + for (j=0;j<N;j++) + { + celt_sig tmp = x[j] + m + VERY_SMALL; + m = MULT16_32_Q15(coef0, tmp); + scratch[j] = tmp; + } + apply_downsampling=1; + } else { + /* Shortcut for the standard (non-custom modes) case */ + for (j=0;j<N;j++) + { + celt_sig tmp = x[j] + m + VERY_SMALL; + m = MULT16_32_Q15(coef0, tmp); + y[j*C] = SCALEOUT(SIG2WORD16(tmp)); + } + } + mem[c] = m; + + if (apply_downsampling) + { + /* Perform down-sampling */ + for (j=0;j<Nd;j++) + y[j*C] = SCALEOUT(SIG2WORD16(scratch[j*downsample])); + } + } while (++c<C); +} + +/** Compute the IMDCT and apply window for all sub-frames and + all channels in a frame */ +#ifndef RESYNTH +static +#endif +void compute_inv_mdcts(const CELTMode *mode, int shortBlocks, celt_sig *X, + celt_sig * OPUS_RESTRICT out_mem[], int C, int LM) +{ + int b, c; + int B; + int N; + int shift; + const int overlap = OVERLAP(mode); + + if (shortBlocks) + { + B = shortBlocks; + N = mode->shortMdctSize; + shift = mode->maxLM; + } else { + B = 1; + N = mode->shortMdctSize<<LM; + shift = mode->maxLM-LM; + } + c=0; do { + /* IMDCT on the interleaved the sub-frames, overlap-add is performed by the IMDCT */ + for (b=0;b<B;b++) + clt_mdct_backward(&mode->mdct, &X[b+c*N*B], out_mem[c]+N*b, mode->window, overlap, shift, B); + } while (++c<C); +} + +static void tf_decode(int start, int end, int isTransient, int *tf_res, int LM, ec_dec *dec) +{ + int i, curr, tf_select; + int tf_select_rsv; + int tf_changed; + int logp; + opus_uint32 budget; + opus_uint32 tell; + + budget = dec->storage*8; + tell = ec_tell(dec); + logp = isTransient ? 2 : 4; + tf_select_rsv = LM>0 && tell+logp+1<=budget; + budget -= tf_select_rsv; + tf_changed = curr = 0; + for (i=start;i<end;i++) + { + if (tell+logp<=budget) + { + curr ^= ec_dec_bit_logp(dec, logp); + tell = ec_tell(dec); + tf_changed |= curr; + } + tf_res[i] = curr; + logp = isTransient ? 4 : 5; + } + tf_select = 0; + if (tf_select_rsv && + tf_select_table[LM][4*isTransient+0+tf_changed] != + tf_select_table[LM][4*isTransient+2+tf_changed]) + { + tf_select = ec_dec_bit_logp(dec, 1); + } + for (i=start;i<end;i++) + { + tf_res[i] = tf_select_table[LM][4*isTransient+2*tf_select+tf_res[i]]; + } +} + +/* The maximum pitch lag to allow in the pitch-based PLC. It's possible to save + CPU time in the PLC pitch search by making this smaller than MAX_PERIOD. The + current value corresponds to a pitch of 66.67 Hz. */ +#define PLC_PITCH_LAG_MAX (720) +/* The minimum pitch lag to allow in the pitch-based PLC. This corresponds to a + pitch of 480 Hz. */ +#define PLC_PITCH_LAG_MIN (100) + +static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, opus_val16 * OPUS_RESTRICT pcm, int N, int LM) +{ + int c; + int i; + const int C = st->channels; + celt_sig *decode_mem[2]; + celt_sig *out_syn[2]; + opus_val16 *lpc; + opus_val16 *oldBandE, *oldLogE, *oldLogE2, *backgroundLogE; + const OpusCustomMode *mode; + int nbEBands; + int overlap; + int start; + int downsample; + int loss_count; + int noise_based; + const opus_int16 *eBands; + VARDECL(celt_sig, scratch); + SAVE_STACK; + + mode = st->mode; + nbEBands = mode->nbEBands; + overlap = mode->overlap; + eBands = mode->eBands; + + c=0; do { + decode_mem[c] = st->_decode_mem + c*(DECODE_BUFFER_SIZE+overlap); + out_syn[c] = decode_mem[c]+DECODE_BUFFER_SIZE-N; + } while (++c<C); + lpc = (opus_val16*)(st->_decode_mem+(DECODE_BUFFER_SIZE+overlap)*C); + oldBandE = lpc+C*LPC_ORDER; + oldLogE = oldBandE + 2*nbEBands; + oldLogE2 = oldLogE + 2*nbEBands; + backgroundLogE = oldLogE2 + 2*nbEBands; + + loss_count = st->loss_count; + start = st->start; + downsample = st->downsample; + noise_based = loss_count >= 5 || start != 0; + ALLOC(scratch, noise_based?N*C:N, celt_sig); + if (noise_based) + { + /* Noise-based PLC/CNG */ + celt_sig *freq; + VARDECL(celt_norm, X); + opus_uint32 seed; + opus_val16 *plcLogE; + int end; + int effEnd; + + end = st->end; + effEnd = IMAX(start, IMIN(end, mode->effEBands)); + + /* Share the interleaved signal MDCT coefficient buffer with the + deemphasis scratch buffer. */ + freq = scratch; + ALLOC(X, C*N, celt_norm); /**< Interleaved normalised MDCTs */ + + if (loss_count >= 5) + plcLogE = backgroundLogE; + else { + /* Energy decay */ + opus_val16 decay = loss_count==0 ? + QCONST16(1.5f, DB_SHIFT) : QCONST16(.5f, DB_SHIFT); + c=0; do + { + for (i=start;i<end;i++) + oldBandE[c*nbEBands+i] -= decay; + } while (++c<C); + plcLogE = oldBandE; + } + seed = st->rng; + for (c=0;c<C;c++) + { + for (i=start;i<effEnd;i++) + { + int j; + int boffs; + int blen; + boffs = N*c+(eBands[i]<<LM); + blen = (eBands[i+1]-eBands[i])<<LM; + for (j=0;j<blen;j++) + { + seed = celt_lcg_rand(seed); + X[boffs+j] = (celt_norm)((opus_int32)seed>>20); + } + renormalise_vector(X+boffs, blen, Q15ONE); + } + } + st->rng = seed; + + denormalise_bands(mode, X, freq, plcLogE, start, effEnd, C, 1<<LM); + + c=0; do { + int bound = eBands[effEnd]<<LM; + if (downsample!=1) + bound = IMIN(bound, N/downsample); + for (i=bound;i<N;i++) + freq[c*N+i] = 0; + } while (++c<C); + c=0; do { + OPUS_MOVE(decode_mem[c], decode_mem[c]+N, + DECODE_BUFFER_SIZE-N+(overlap>>1)); + } while (++c<C); + compute_inv_mdcts(mode, 0, freq, out_syn, C, LM); + } else { + /* Pitch-based PLC */ + const opus_val16 *window; + opus_val16 fade = Q15ONE; + int pitch_index; + VARDECL(opus_val32, etmp); + VARDECL(opus_val16, exc); + + if (loss_count == 0) + { + VARDECL( opus_val16, lp_pitch_buf ); + ALLOC( lp_pitch_buf, DECODE_BUFFER_SIZE>>1, opus_val16 ); + pitch_downsample(decode_mem, lp_pitch_buf, + DECODE_BUFFER_SIZE, C, st->arch); + pitch_search(lp_pitch_buf+(PLC_PITCH_LAG_MAX>>1), lp_pitch_buf, + DECODE_BUFFER_SIZE-PLC_PITCH_LAG_MAX, + PLC_PITCH_LAG_MAX-PLC_PITCH_LAG_MIN, &pitch_index, st->arch); + pitch_index = PLC_PITCH_LAG_MAX-pitch_index; + st->last_pitch_index = pitch_index; + } else { + pitch_index = st->last_pitch_index; + fade = QCONST16(.8f,15); + } + + ALLOC(etmp, overlap, opus_val32); + ALLOC(exc, MAX_PERIOD, opus_val16); + window = mode->window; + c=0; do { + opus_val16 decay; + opus_val16 attenuation; + opus_val32 S1=0; + celt_sig *buf; + int extrapolation_offset; + int extrapolation_len; + int exc_length; + int j; + + buf = decode_mem[c]; + for (i=0;i<MAX_PERIOD;i++) { + exc[i] = ROUND16(buf[DECODE_BUFFER_SIZE-MAX_PERIOD+i], SIG_SHIFT); + } + + if (loss_count == 0) + { + 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, st->arch); + /* Add a noise floor of -40 dB. */ +#ifdef OPUS_FIXED_POINT + ac[0] += SHR32(ac[0],13); +#else + ac[0] *= 1.0001f; +#endif + /* Use lag windowing to stabilize the Levinson-Durbin recursion. */ + for (i=1;i<=LPC_ORDER;i++) + { + /*ac[i] *= exp(-.5*(2*M_PI*.002*i)*(2*M_PI*.002*i));*/ +#ifdef OPUS_FIXED_POINT + ac[i] -= MULT16_32_Q15(2*i*i, ac[i]); +#else + ac[i] -= ac[i]*(0.008f*0.008f)*i*i; +#endif + } + _celt_lpc(lpc+c*LPC_ORDER, ac, LPC_ORDER); + } + /* We want the excitation for 2 pitch periods in order to look for a + decaying signal, but we can't get more than MAX_PERIOD. */ + exc_length = IMIN(2*pitch_index, MAX_PERIOD); + /* Initialize the LPC history with the samples just before the start + of the region for which we're computing the excitation. */ + { + opus_val16 lpc_mem[LPC_ORDER]; + for (i=0;i<LPC_ORDER;i++) + { + lpc_mem[i] = + ROUND16(buf[DECODE_BUFFER_SIZE-exc_length-1-i], SIG_SHIFT); + } + /* Compute the excitation for exc_length samples before the loss. */ + celt_fir(exc+MAX_PERIOD-exc_length, lpc+c*LPC_ORDER, + exc+MAX_PERIOD-exc_length, exc_length, LPC_ORDER, lpc_mem); + } + + /* Check if the waveform is decaying, and if so how fast. + We do this to avoid adding energy when concealing in a segment + with decaying energy. */ + { + opus_val32 E1=1, E2=1; + int decay_length; +#ifdef OPUS_FIXED_POINT + int shift = IMAX(0,2*celt_zlog2(celt_maxabs16(&exc[MAX_PERIOD-exc_length], exc_length))-20); +#endif + decay_length = exc_length>>1; + for (i=0;i<decay_length;i++) + { + opus_val16 e; + e = exc[MAX_PERIOD-decay_length+i]; + E1 += SHR32(MULT16_16(e, e), shift); + e = exc[MAX_PERIOD-2*decay_length+i]; + E2 += SHR32(MULT16_16(e, e), shift); + } + E1 = MIN32(E1, E2); + decay = celt_sqrt(frac_div32(SHR32(E1, 1), E2)); + } + + /* Move the decoder memory one frame to the left to give us room to + add the data for the new frame. We ignore the overlap that extends + past the end of the buffer, because we aren't going to use it. */ + OPUS_MOVE(buf, buf+N, DECODE_BUFFER_SIZE-N); + + /* Extrapolate from the end of the excitation with a period of + "pitch_index", scaling down each period by an additional factor of + "decay". */ + extrapolation_offset = MAX_PERIOD-pitch_index; + /* We need to extrapolate enough samples to cover a complete MDCT + window (including overlap/2 samples on both sides). */ + extrapolation_len = N+overlap; + /* We also apply fading if this is not the first loss. */ + attenuation = MULT16_16_Q15(fade, decay); + for (i=j=0;i<extrapolation_len;i++,j++) + { + opus_val16 tmp; + if (j >= pitch_index) { + j -= pitch_index; + attenuation = MULT16_16_Q15(attenuation, decay); + } + buf[DECODE_BUFFER_SIZE-N+i] = + SHL32(EXTEND32(MULT16_16_Q15(attenuation, + exc[extrapolation_offset+j])), SIG_SHIFT); + /* Compute the energy of the previously decoded signal whose + excitation we're copying. */ + tmp = ROUND16( + buf[DECODE_BUFFER_SIZE-MAX_PERIOD-N+extrapolation_offset+j], + SIG_SHIFT); + S1 += SHR32(MULT16_16(tmp, tmp), 8); + } + + { + opus_val16 lpc_mem[LPC_ORDER]; + /* Copy the last decoded samples (prior to the overlap region) to + synthesis filter memory so we can have a continuous signal. */ + for (i=0;i<LPC_ORDER;i++) + lpc_mem[i] = ROUND16(buf[DECODE_BUFFER_SIZE-N-1-i], SIG_SHIFT); + /* Apply the synthesis filter to convert the excitation back into + the signal domain. */ + celt_iir(buf+DECODE_BUFFER_SIZE-N, lpc+c*LPC_ORDER, + buf+DECODE_BUFFER_SIZE-N, extrapolation_len, LPC_ORDER, + lpc_mem); + } + + /* Check if the synthesis energy is higher than expected, which can + happen with the signal changes during our window. If so, + attenuate. */ + { + opus_val32 S2=0; + for (i=0;i<extrapolation_len;i++) + { + opus_val16 tmp = ROUND16(buf[DECODE_BUFFER_SIZE-N+i], SIG_SHIFT); + S2 += SHR32(MULT16_16(tmp, tmp), 8); + } + /* This checks for an "explosion" in the synthesis. */ +#ifdef OPUS_FIXED_POINT + if (!(S1 > SHR32(S2,2))) +#else + /* The float test is written this way to catch NaNs in the output + of the IIR filter at the same time. */ + if (!(S1 > 0.2f*S2)) +#endif + { + for (i=0;i<extrapolation_len;i++) + buf[DECODE_BUFFER_SIZE-N+i] = 0; + } else if (S1 < S2) + { + opus_val16 ratio = celt_sqrt(frac_div32(SHR32(S1,1)+1,S2+1)); + for (i=0;i<overlap;i++) + { + opus_val16 tmp_g = Q15ONE + - MULT16_16_Q15(window[i], Q15ONE-ratio); + buf[DECODE_BUFFER_SIZE-N+i] = + MULT16_32_Q15(tmp_g, buf[DECODE_BUFFER_SIZE-N+i]); + } + for (i=overlap;i<extrapolation_len;i++) + { + buf[DECODE_BUFFER_SIZE-N+i] = + MULT16_32_Q15(ratio, buf[DECODE_BUFFER_SIZE-N+i]); + } + } + } + + /* Apply the pre-filter to the MDCT overlap for the next frame because + the post-filter will be re-applied in the decoder after the MDCT + overlap. */ + comb_filter(etmp, buf+DECODE_BUFFER_SIZE, + st->postfilter_period, st->postfilter_period, overlap, + -st->postfilter_gain, -st->postfilter_gain, + st->postfilter_tapset, st->postfilter_tapset, NULL, 0); + + /* Simulate TDAC on the concealed audio so that it blends with the + MDCT of the next frame. */ + for (i=0;i<overlap/2;i++) + { + buf[DECODE_BUFFER_SIZE+i] = + MULT16_32_Q15(window[i], etmp[overlap-1-i]) + + MULT16_32_Q15(window[overlap-i-1], etmp[i]); + } + } while (++c<C); + } + + deemphasis(out_syn, pcm, N, C, downsample, + mode->preemph, st->preemph_memD, scratch); + + st->loss_count = loss_count+1; + + RESTORE_STACK; +} + +int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, opus_val16 * OPUS_RESTRICT pcm, int frame_size, ec_dec *dec) +{ + int c, i, N; + int spread_decision; + opus_int32 bits; + ec_dec _dec; + VARDECL(celt_sig, freq); + VARDECL(celt_norm, X); + VARDECL(int, fine_quant); + VARDECL(int, pulses); + VARDECL(int, cap); + VARDECL(int, offsets); + VARDECL(int, fine_priority); + VARDECL(int, tf_res); + VARDECL(unsigned char, collapse_masks); + celt_sig *decode_mem[2]; + celt_sig *out_syn[2]; + opus_val16 *lpc; + opus_val16 *oldBandE, *oldLogE, *oldLogE2, *backgroundLogE; + + int shortBlocks; + int isTransient; + int intra_ener; + const int CC = st->channels; + int LM, M; + int effEnd; + int codedBands; + int alloc_trim; + int postfilter_pitch; + opus_val16 postfilter_gain; + int intensity=0; + int dual_stereo=0; + opus_int32 total_bits; + opus_int32 balance; + opus_int32 tell; + int dynalloc_logp; + int postfilter_tapset; + int anti_collapse_rsv; + int anti_collapse_on=0; + int silence; + int C = st->stream_channels; + const OpusCustomMode *mode; + int nbEBands; + int overlap; + const opus_int16 *eBands; + ALLOC_STACK; + + mode = st->mode; + nbEBands = mode->nbEBands; + overlap = mode->overlap; + eBands = mode->eBands; + frame_size *= st->downsample; + + c=0; do { + decode_mem[c] = st->_decode_mem + c*(DECODE_BUFFER_SIZE+overlap); + } while (++c<CC); + lpc = (opus_val16*)(st->_decode_mem+(DECODE_BUFFER_SIZE+overlap)*CC); + oldBandE = lpc+CC*LPC_ORDER; + oldLogE = oldBandE + 2*nbEBands; + oldLogE2 = oldLogE + 2*nbEBands; + backgroundLogE = oldLogE2 + 2*nbEBands; + +#ifdef CUSTOM_MODES + if (st->signalling && data!=NULL) + { + int data0=data[0]; + /* Convert "standard mode" to Opus header */ + if (mode->Fs==48000 && mode->shortMdctSize==120) + { + data0 = fromOpus(data0); + if (data0<0) + return OPUS_INVALID_PACKET; + } + st->end = IMAX(1, mode->effEBands-2*(data0>>5)); + LM = (data0>>3)&0x3; + C = 1 + ((data0>>2)&0x1); + data++; + len--; + if (LM>mode->maxLM) + return OPUS_INVALID_PACKET; + if (frame_size < mode->shortMdctSize<<LM) + return OPUS_BUFFER_TOO_SMALL; + else + frame_size = mode->shortMdctSize<<LM; + } else { +#else + { +#endif + for (LM=0;LM<=mode->maxLM;LM++) + if (mode->shortMdctSize<<LM==frame_size) + break; + if (LM>mode->maxLM) + return OPUS_BAD_ARG; + } + M=1<<LM; + + if (len<0 || len>1275 || pcm==NULL) + return OPUS_BAD_ARG; + + N = M*mode->shortMdctSize; + + effEnd = st->end; + if (effEnd > mode->effEBands) + effEnd = mode->effEBands; + + if (data == NULL || len<=1) + { + celt_decode_lost(st, pcm, N, LM); + RESTORE_STACK; + return frame_size/st->downsample; + } + + if (dec == NULL) + { + ec_dec_init(&_dec,(unsigned char*)data,len); + dec = &_dec; + } + + if (C==1) + { + for (i=0;i<nbEBands;i++) + oldBandE[i]=MAX16(oldBandE[i],oldBandE[nbEBands+i]); + } + + total_bits = len*8; + tell = ec_tell(dec); + + if (tell >= total_bits) + silence = 1; + else if (tell==1) + silence = ec_dec_bit_logp(dec, 15); + else + silence = 0; + if (silence) + { + /* Pretend we've read all the remaining bits */ + tell = len*8; + dec->nbits_total+=tell-ec_tell(dec); + } + + postfilter_gain = 0; + postfilter_pitch = 0; + postfilter_tapset = 0; + if (st->start==0 && tell+16 <= total_bits) + { + if(ec_dec_bit_logp(dec, 1)) + { + int qg, octave; + octave = ec_dec_uint(dec, 6); + postfilter_pitch = (16<<octave)+ec_dec_bits(dec, 4+octave)-1; + qg = ec_dec_bits(dec, 3); + if (ec_tell(dec)+2<=total_bits) + postfilter_tapset = ec_dec_icdf(dec, tapset_icdf, 2); + postfilter_gain = QCONST16(.09375f,15)*(qg+1); + } + tell = ec_tell(dec); + } + + if (LM > 0 && tell+3 <= total_bits) + { + isTransient = ec_dec_bit_logp(dec, 3); + tell = ec_tell(dec); + } + else + isTransient = 0; + + if (isTransient) + shortBlocks = M; + else + shortBlocks = 0; + + /* Decode the global flags (first symbols in the stream) */ + intra_ener = tell+3<=total_bits ? ec_dec_bit_logp(dec, 3) : 0; + /* Get band energies */ + unquant_coarse_energy(mode, st->start, st->end, oldBandE, + intra_ener, dec, C, LM); + + ALLOC(tf_res, nbEBands, int); + tf_decode(st->start, st->end, isTransient, tf_res, LM, dec); + + tell = ec_tell(dec); + spread_decision = SPREAD_NORMAL; + if (tell+4 <= total_bits) + spread_decision = ec_dec_icdf(dec, spread_icdf, 5); + + ALLOC(cap, nbEBands, int); + + init_caps(mode,cap,LM,C); + + ALLOC(offsets, nbEBands, int); + + dynalloc_logp = 6; + total_bits<<=BITRES; + tell = ec_tell_frac(dec); + for (i=st->start;i<st->end;i++) + { + int width, quanta; + int dynalloc_loop_logp; + int boost; + width = C*(eBands[i+1]-eBands[i])<<LM; + /* quanta is 6 bits, but no more than 1 bit/sample + and no less than 1/8 bit/sample */ + quanta = IMIN(width<<BITRES, IMAX(6<<BITRES, width)); + dynalloc_loop_logp = dynalloc_logp; + boost = 0; + while (tell+(dynalloc_loop_logp<<BITRES) < total_bits && boost < cap[i]) + { + int flag; + flag = ec_dec_bit_logp(dec, dynalloc_loop_logp); + tell = ec_tell_frac(dec); + if (!flag) + break; + boost += quanta; + total_bits -= quanta; + dynalloc_loop_logp = 1; + } + offsets[i] = boost; + /* Making dynalloc more likely */ + if (boost>0) + dynalloc_logp = IMAX(2, dynalloc_logp-1); + } + + ALLOC(fine_quant, nbEBands, int); + alloc_trim = tell+(6<<BITRES) <= total_bits ? + ec_dec_icdf(dec, trim_icdf, 7) : 5; + + bits = (((opus_int32)len*8)<<BITRES) - ec_tell_frac(dec) - 1; + anti_collapse_rsv = isTransient&&LM>=2&&bits>=((LM+2)<<BITRES) ? (1<<BITRES) : 0; + bits -= anti_collapse_rsv; + + ALLOC(pulses, nbEBands, int); + ALLOC(fine_priority, nbEBands, int); + + codedBands = compute_allocation(mode, st->start, st->end, offsets, cap, + alloc_trim, &intensity, &dual_stereo, bits, &balance, pulses, + fine_quant, fine_priority, C, LM, dec, 0, 0, 0); + + unquant_fine_energy(mode, st->start, st->end, oldBandE, fine_quant, dec, C); + + /* Decode fixed codebook */ + ALLOC(collapse_masks, C*nbEBands, unsigned char); + ALLOC(X, C*N, celt_norm); /**< Interleaved normalised MDCTs */ + + quant_all_bands(0, mode, st->start, st->end, X, C==2 ? X+N : NULL, collapse_masks, + NULL, pulses, shortBlocks, spread_decision, dual_stereo, intensity, tf_res, + len*(8<<BITRES)-anti_collapse_rsv, balance, dec, LM, codedBands, &st->rng); + + if (anti_collapse_rsv > 0) + { + anti_collapse_on = ec_dec_bits(dec, 1); + } + + unquant_energy_finalise(mode, st->start, st->end, oldBandE, + fine_quant, fine_priority, len*8-ec_tell(dec), dec, C); + + if (anti_collapse_on) + anti_collapse(mode, X, collapse_masks, LM, C, N, + st->start, st->end, oldBandE, oldLogE, oldLogE2, pulses, st->rng); + + ALLOC(freq, IMAX(CC,C)*N, celt_sig); /**< Interleaved signal MDCTs */ + + if (silence) + { + for (i=0;i<C*nbEBands;i++) + oldBandE[i] = -QCONST16(28.f,DB_SHIFT); + for (i=0;i<C*N;i++) + freq[i] = 0; + } else { + /* Synthesis */ + denormalise_bands(mode, X, freq, oldBandE, st->start, effEnd, C, M); + } + c=0; do { + OPUS_MOVE(decode_mem[c], decode_mem[c]+N, DECODE_BUFFER_SIZE-N+overlap/2); + } while (++c<CC); + + c=0; do { + int bound = M*eBands[effEnd]; + if (st->downsample!=1) + bound = IMIN(bound, N/st->downsample); + for (i=bound;i<N;i++) + freq[c*N+i] = 0; + } while (++c<C); + + c=0; do { + out_syn[c] = decode_mem[c]+DECODE_BUFFER_SIZE-N; + } while (++c<CC); + + if (CC==2&&C==1) + { + for (i=0;i<N;i++) + freq[N+i] = freq[i]; + } + if (CC==1&&C==2) + { + for (i=0;i<N;i++) + freq[i] = HALF32(ADD32(freq[i],freq[N+i])); + } + + /* Compute inverse MDCTs */ + compute_inv_mdcts(mode, shortBlocks, freq, out_syn, CC, LM); + + c=0; do { + st->postfilter_period=IMAX(st->postfilter_period, COMBFILTER_MINPERIOD); + st->postfilter_period_old=IMAX(st->postfilter_period_old, COMBFILTER_MINPERIOD); + comb_filter(out_syn[c], out_syn[c], st->postfilter_period_old, st->postfilter_period, mode->shortMdctSize, + st->postfilter_gain_old, st->postfilter_gain, st->postfilter_tapset_old, st->postfilter_tapset, + mode->window, overlap); + if (LM!=0) + comb_filter(out_syn[c]+mode->shortMdctSize, out_syn[c]+mode->shortMdctSize, st->postfilter_period, postfilter_pitch, N-mode->shortMdctSize, + st->postfilter_gain, postfilter_gain, st->postfilter_tapset, postfilter_tapset, + mode->window, overlap); + + } while (++c<CC); + st->postfilter_period_old = st->postfilter_period; + st->postfilter_gain_old = st->postfilter_gain; + st->postfilter_tapset_old = st->postfilter_tapset; + st->postfilter_period = postfilter_pitch; + st->postfilter_gain = postfilter_gain; + st->postfilter_tapset = postfilter_tapset; + if (LM!=0) + { + st->postfilter_period_old = st->postfilter_period; + st->postfilter_gain_old = st->postfilter_gain; + st->postfilter_tapset_old = st->postfilter_tapset; + } + + if (C==1) { + for (i=0;i<nbEBands;i++) + oldBandE[nbEBands+i]=oldBandE[i]; + } + + /* In case start or end were to change */ + if (!isTransient) + { + for (i=0;i<2*nbEBands;i++) + oldLogE2[i] = oldLogE[i]; + for (i=0;i<2*nbEBands;i++) + oldLogE[i] = oldBandE[i]; + for (i=0;i<2*nbEBands;i++) + backgroundLogE[i] = MIN16(backgroundLogE[i] + M*QCONST16(0.001f,DB_SHIFT), oldBandE[i]); + } else { + for (i=0;i<2*nbEBands;i++) + oldLogE[i] = MIN16(oldLogE[i], oldBandE[i]); + } + c=0; do + { + for (i=0;i<st->start;i++) + { + oldBandE[c*nbEBands+i]=0; + oldLogE[c*nbEBands+i]=oldLogE2[c*nbEBands+i]=-QCONST16(28.f,DB_SHIFT); + } + for (i=st->end;i<nbEBands;i++) + { + oldBandE[c*nbEBands+i]=0; + oldLogE[c*nbEBands+i]=oldLogE2[c*nbEBands+i]=-QCONST16(28.f,DB_SHIFT); + } + } while (++c<2); + st->rng = dec->rng; + + /* We reuse freq[] as scratch space for the de-emphasis */ + deemphasis(out_syn, pcm, N, CC, st->downsample, mode->preemph, st->preemph_memD, freq); + st->loss_count = 0; + RESTORE_STACK; + if (ec_tell(dec) > 8*len) + return OPUS_INTERNAL_ERROR; + if(ec_get_error(dec)) + st->error = 1; + return frame_size/st->downsample; +} + + +#ifdef CUSTOM_MODES + +#ifdef OPUS_FIXED_POINT +int opus_custom_decode(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, opus_int16 * OPUS_RESTRICT pcm, int frame_size) +{ + return celt_decode_with_ec(st, data, len, pcm, frame_size, NULL); +} + +#ifndef DISABLE_FLOAT_API +int opus_custom_decode_float(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, float * OPUS_RESTRICT pcm, int frame_size) +{ + int j, ret, C, N; + VARDECL(opus_int16, out); + ALLOC_STACK; + + if (pcm==NULL) + return OPUS_BAD_ARG; + + C = st->channels; + N = frame_size; + + ALLOC(out, C*N, opus_int16); + ret=celt_decode_with_ec(st, data, len, out, frame_size, NULL); + if (ret>0) + for (j=0;j<C*ret;j++) + pcm[j]=out[j]*(1.f/32768.f); + + RESTORE_STACK; + return ret; +} +#endif /* DISABLE_FLOAT_API */ + +#else + +int opus_custom_decode_float(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, float * OPUS_RESTRICT pcm, int frame_size) +{ + return celt_decode_with_ec(st, data, len, pcm, frame_size, NULL); +} + +int opus_custom_decode(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, opus_int16 * OPUS_RESTRICT pcm, int frame_size) +{ + int j, ret, C, N; + VARDECL(celt_sig, out); + ALLOC_STACK; + + if (pcm==NULL) + return OPUS_BAD_ARG; + + C = st->channels; + N = frame_size; + ALLOC(out, C*N, celt_sig); + + ret=celt_decode_with_ec(st, data, len, out, frame_size, NULL); + + if (ret>0) + for (j=0;j<C*ret;j++) + pcm[j] = FLOAT2INT16 (out[j]); + + RESTORE_STACK; + return ret; +} + +#endif +#endif /* CUSTOM_MODES */ + +int opus_custom_decoder_ctl(CELTDecoder * OPUS_RESTRICT st, int request, ...) +{ + va_list ap; + + va_start(ap, request); + switch (request) + { + case CELT_SET_START_BAND_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<0 || value>=st->mode->nbEBands) + goto bad_arg; + st->start = value; + } + break; + case CELT_SET_END_BAND_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<1 || value>st->mode->nbEBands) + goto bad_arg; + st->end = value; + } + break; + case CELT_SET_CHANNELS_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<1 || value>2) + goto bad_arg; + st->stream_channels = value; + } + break; + case CELT_GET_AND_CLEAR_ERROR_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (value==NULL) + goto bad_arg; + *value=st->error; + st->error = 0; + } + break; + case OPUS_GET_LOOKAHEAD_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (value==NULL) + goto bad_arg; + *value = st->overlap/st->downsample; + } + break; + case OPUS_RESET_STATE: + { + int i; + opus_val16 *lpc, *oldBandE, *oldLogE, *oldLogE2; + lpc = (opus_val16*)(st->_decode_mem+(DECODE_BUFFER_SIZE+st->overlap)*st->channels); + oldBandE = lpc+st->channels*LPC_ORDER; + oldLogE = oldBandE + 2*st->mode->nbEBands; + oldLogE2 = oldLogE + 2*st->mode->nbEBands; + OPUS_CLEAR((char*)&st->DECODER_RESET_START, + opus_custom_decoder_get_size(st->mode, st->channels)- + ((char*)&st->DECODER_RESET_START - (char*)st)); + for (i=0;i<2*st->mode->nbEBands;i++) + oldLogE[i]=oldLogE2[i]=-QCONST16(28.f,DB_SHIFT); + } + break; + case OPUS_GET_PITCH_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (value==NULL) + goto bad_arg; + *value = st->postfilter_period; + } + break; + case CELT_GET_MODE_REQUEST: + { + const CELTMode ** value = va_arg(ap, const CELTMode**); + if (value==0) + goto bad_arg; + *value=st->mode; + } + break; + case CELT_SET_SIGNALLING_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->signalling = value; + } + break; + case OPUS_GET_FINAL_RANGE_REQUEST: + { + opus_uint32 * value = va_arg(ap, opus_uint32 *); + if (value==0) + goto bad_arg; + *value=st->rng; + } + break; + default: + goto bad_request; + } + va_end(ap); + return OPUS_OK; +bad_arg: + va_end(ap); + return OPUS_BAD_ARG; +bad_request: + va_end(ap); + return OPUS_UNIMPLEMENTED; +} diff --git a/drivers/opus/celt/celt_encoder.c b/drivers/opus/celt/celt_encoder.c new file mode 100644 index 0000000000..a61e41f42d --- /dev/null +++ b/drivers/opus/celt/celt_encoder.c @@ -0,0 +1,2353 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2010 Xiph.Org Foundation + Copyright (c) 2008 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef OPUS_HAVE_CONFIG_H +#include "opus_config.h" +#endif + +#define CELT_ENCODER_C + +#include "cpu_support.h" +#include "os_support.h" +#include "mdct.h" +#include <math.h> +#include "celt.h" +#include "pitch.h" +#include "bands.h" +#include "opus_modes.h" +#include "entcode.h" +#include "quant_bands.h" +#include "rate.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "float_cast.h" +#include <stdarg.h> +#include "celt_lpc.h" +#include "vq.h" + + +/** Encoder state + @brief Encoder state + */ +struct OpusCustomEncoder { + const OpusCustomMode *mode; /**< Mode used by the encoder */ + int overlap; + int channels; + int stream_channels; + + int force_intra; + int clip; + int disable_pf; + int complexity; + int upsample; + int start, end; + + opus_int32 bitrate; + int vbr; + int signalling; + int constrained_vbr; /* If zero, VBR can do whatever it likes with the rate */ + int loss_rate; + int lsb_depth; + int variable_duration; + int lfe; + int arch; + + /* Everything beyond this point gets cleared on a reset */ +#define ENCODER_RESET_START rng + + opus_uint32 rng; + int spread_decision; + opus_val32 delayedIntra; + int tonal_average; + int lastCodedBands; + int hf_average; + int tapset_decision; + + int prefilter_period; + opus_val16 prefilter_gain; + int prefilter_tapset; +#ifdef RESYNTH + int prefilter_period_old; + opus_val16 prefilter_gain_old; + int prefilter_tapset_old; +#endif + int consec_transient; + AnalysisInfo analysis; + + opus_val32 preemph_memE[2]; + opus_val32 preemph_memD[2]; + + /* VBR-related parameters */ + opus_int32 vbr_reservoir; + opus_int32 vbr_drift; + opus_int32 vbr_offset; + opus_int32 vbr_count; + opus_val32 overlap_max; + opus_val16 stereo_saving; + int intensity; + opus_val16 *energy_mask; + opus_val16 spec_avg; + +#ifdef RESYNTH + /* +MAX_PERIOD/2 to make space for overlap */ + celt_sig syn_mem[2][2*MAX_PERIOD+MAX_PERIOD/2]; +#endif + + celt_sig in_mem[1]; /* Size = channels*mode->overlap */ + /* celt_sig prefilter_mem[], Size = channels*COMBFILTER_MAXPERIOD */ + /* opus_val16 oldBandE[], Size = channels*mode->nbEBands */ + /* opus_val16 oldLogE[], Size = channels*mode->nbEBands */ + /* opus_val16 oldLogE2[], Size = channels*mode->nbEBands */ +}; + +int celt_encoder_get_size(int channels) +{ + CELTMode *mode = opus_custom_mode_create(48000, 960, NULL); + return opus_custom_encoder_get_size(mode, channels); +} + +OPUS_CUSTOM_NOSTATIC int opus_custom_encoder_get_size(const CELTMode *mode, int channels) +{ + int size = sizeof(struct CELTEncoder) + + (channels*mode->overlap-1)*sizeof(celt_sig) /* celt_sig in_mem[channels*mode->overlap]; */ + + channels*COMBFILTER_MAXPERIOD*sizeof(celt_sig) /* celt_sig prefilter_mem[channels*COMBFILTER_MAXPERIOD]; */ + + 3*channels*mode->nbEBands*sizeof(opus_val16); /* opus_val16 oldBandE[channels*mode->nbEBands]; */ + /* opus_val16 oldLogE[channels*mode->nbEBands]; */ + /* opus_val16 oldLogE2[channels*mode->nbEBands]; */ + return size; +} + +#ifdef CUSTOM_MODES +CELTEncoder *opus_custom_encoder_create(const CELTMode *mode, int channels, int *error) +{ + int ret; + CELTEncoder *st = (CELTEncoder *)opus_alloc(opus_custom_encoder_get_size(mode, channels)); + /* init will handle the NULL case */ + ret = opus_custom_encoder_init(st, mode, channels); + if (ret != OPUS_OK) + { + opus_custom_encoder_destroy(st); + st = NULL; + } + if (error) + *error = ret; + return st; +} +#endif /* CUSTOM_MODES */ + +static int opus_custom_encoder_init_arch(CELTEncoder *st, const CELTMode *mode, + int channels, int arch) +{ + if (channels < 0 || channels > 2) + return OPUS_BAD_ARG; + + if (st==NULL || mode==NULL) + return OPUS_ALLOC_FAIL; + + OPUS_CLEAR((char*)st, opus_custom_encoder_get_size(mode, channels)); + + st->mode = mode; + st->overlap = mode->overlap; + st->stream_channels = st->channels = channels; + + st->upsample = 1; + st->start = 0; + st->end = st->mode->effEBands; + st->signalling = 1; + + st->arch = arch; + + st->constrained_vbr = 1; + st->clip = 1; + + st->bitrate = OPUS_BITRATE_MAX; + st->vbr = 0; + st->force_intra = 0; + st->complexity = 5; + st->lsb_depth=24; + + opus_custom_encoder_ctl(st, OPUS_RESET_STATE); + + return OPUS_OK; +} + +#ifdef CUSTOM_MODES +int opus_custom_encoder_init(CELTEncoder *st, const CELTMode *mode, int channels) +{ + return opus_custom_encoder_init_arch(st, mode, channels, opus_select_arch()); +} +#endif + +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) +{ + opus_free(st); +} +#endif /* CUSTOM_MODES */ + + +static int transient_analysis(const opus_val32 * OPUS_RESTRICT in, int len, int C, + opus_val16 *tf_estimate, int *tf_chan) +{ + int i; + VARDECL(opus_val16, tmp); + opus_val32 mem0,mem1; + int is_transient = 0; + opus_int32 mask_metric = 0; + int c; + opus_val16 tf_max; + int len2; + /* Table of 6*64/x, trained on real data to minimize the average error */ + static const unsigned char inv_table[128] = { + 255,255,156,110, 86, 70, 59, 51, 45, 40, 37, 33, 31, 28, 26, 25, + 23, 22, 21, 20, 19, 18, 17, 16, 16, 15, 15, 14, 13, 13, 12, 12, + 12, 12, 11, 11, 11, 10, 10, 10, 9, 9, 9, 9, 9, 9, 8, 8, + 8, 8, 8, 7, 7, 7, 7, 7, 7, 6, 6, 6, 6, 6, 6, 6, + 6, 6, 6, 6, 6, 6, 6, 6, 6, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, + 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 3, 3, + 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 2, + }; + SAVE_STACK; + ALLOC(tmp, len, opus_val16); + + len2=len/2; + for (c=0;c<C;c++) + { + opus_val32 mean; + opus_int32 unmask=0; + opus_val32 norm; + opus_val16 maxE; + mem0=0; + mem1=0; + /* High-pass filter: (1 - 2*z^-1 + z^-2) / (1 - z^-1 + .5*z^-2) */ + for (i=0;i<len;i++) + { + opus_val32 x,y; + x = SHR32(in[i+c*len],SIG_SHIFT); + y = ADD32(mem0, x); +#ifdef OPUS_FIXED_POINT + mem0 = mem1 + y - SHL32(x,1); + mem1 = x - SHR32(y,1); +#else + mem0 = mem1 + y - 2*x; + mem1 = x - .5f*y; +#endif + tmp[i] = EXTRACT16(SHR32(y,2)); + /*printf("%f ", tmp[i]);*/ + } + /*printf("\n");*/ + /* First few samples are bad because we don't propagate the memory */ + for (i=0;i<12;i++) + tmp[i] = 0; + +#ifdef OPUS_FIXED_POINT + /* Normalize tmp to max range */ + { + int shift=0; + shift = 14-celt_ilog2(1+celt_maxabs16(tmp, len)); + if (shift!=0) + { + for (i=0;i<len;i++) + tmp[i] = SHL16(tmp[i], shift); + } + } +#endif + + mean=0; + mem0=0; + /* Grouping by two to reduce complexity */ + /* Forward pass to compute the post-echo threshold*/ + for (i=0;i<len2;i++) + { + opus_val16 x2 = PSHR32(MULT16_16(tmp[2*i],tmp[2*i]) + MULT16_16(tmp[2*i+1],tmp[2*i+1]),16); + mean += x2; +#ifdef OPUS_FIXED_POINT + /* FIXME: Use PSHR16() instead */ + tmp[i] = mem0 + PSHR32(x2-mem0,4); +#else + tmp[i] = mem0 + MULT16_16_P15(QCONST16(.0625f,15),x2-mem0); +#endif + mem0 = tmp[i]; + } + + mem0=0; + maxE=0; + /* Backward pass to compute the pre-echo threshold */ + for (i=len2-1;i>=0;i--) + { +#ifdef OPUS_FIXED_POINT + /* FIXME: Use PSHR16() instead */ + tmp[i] = mem0 + PSHR32(tmp[i]-mem0,3); +#else + tmp[i] = mem0 + MULT16_16_P15(QCONST16(0.125f,15),tmp[i]-mem0); +#endif + mem0 = tmp[i]; + maxE = MAX16(maxE, mem0); + } + /*for (i=0;i<len2;i++)printf("%f ", tmp[i]/mean);printf("\n");*/ + + /* Compute the ratio of the "frame energy" over the harmonic mean of the energy. + This essentially corresponds to a bitrate-normalized temporal noise-to-mask + ratio */ + + /* As a compromise with the old transient detector, frame energy is the + geometric mean of the energy and half the max */ +#ifdef OPUS_FIXED_POINT + /* Costs two sqrt() to avoid overflows */ + mean = MULT16_16(celt_sqrt(mean), celt_sqrt(MULT16_16(maxE,len2>>1))); +#else + mean = celt_sqrt(mean * maxE*.5*len2); +#endif + /* Inverse of the mean energy in Q15+6 */ + norm = SHL32(EXTEND32(len2),6+14)/ADD32(EPSILON,SHR32(mean,1)); + /* Compute harmonic mean discarding the unreliable boundaries + The data is smooth, so we only take 1/4th of the samples */ + unmask=0; + for (i=12;i<len2-5;i+=4) + { + int id; +#ifdef OPUS_FIXED_POINT + id = IMAX(0,IMIN(127,MULT16_32_Q15(tmp[i],norm))); /* Do not round to nearest */ +#else + id = IMAX(0,IMIN(127,(int)floor(64*norm*tmp[i]))); /* Do not round to nearest */ +#endif + unmask += inv_table[id]; + } + /*printf("%d\n", unmask);*/ + /* Normalize, compensate for the 1/4th of the sample and the factor of 6 in the inverse table */ + unmask = 64*unmask*4/(6*(len2-17)); + if (unmask>mask_metric) + { + *tf_chan = c; + mask_metric = unmask; + } + } + is_transient = mask_metric>200; + + /* Arbitrary metric for VBR boost */ + tf_max = MAX16(0,celt_sqrt(27*mask_metric)-42); + /* *tf_estimate = 1 + MIN16(1, sqrt(MAX16(0, tf_max-30))/20); */ + *tf_estimate = celt_sqrt(MAX16(0, SHL32(MULT16_16(QCONST16(0.0069,14),MIN16(163,tf_max)),14)-QCONST32(0.139,28))); + /*printf("%d %f\n", tf_max, mask_metric);*/ + RESTORE_STACK; +#ifdef FUZZING + is_transient = rand()&0x1; +#endif + /*printf("%d %f %d\n", is_transient, (float)*tf_estimate, tf_max);*/ + return is_transient; +} + +/* Looks for sudden increases of energy to decide whether we need to patch + the transient decision */ +int patch_transient_decision(opus_val16 *newE, opus_val16 *oldE, int nbEBands, + int end, int C) +{ + int i, c; + opus_val32 mean_diff=0; + opus_val16 spread_old[26]; + /* Apply an aggressive (-6 dB/Bark) spreading function to the old frame to + avoid false detection caused by irrelevant bands */ + if (C==1) + { + spread_old[0] = oldE[0]; + for (i=1;i<end;i++) + spread_old[i] = MAX16(spread_old[i-1]-QCONST16(1.0f, DB_SHIFT), oldE[i]); + } else { + spread_old[0] = MAX16(oldE[0],oldE[nbEBands]); + for (i=1;i<end;i++) + spread_old[i] = MAX16(spread_old[i-1]-QCONST16(1.0f, DB_SHIFT), + MAX16(oldE[i],oldE[i+nbEBands])); + } + for (i=end-2;i>=0;i--) + spread_old[i] = MAX16(spread_old[i], spread_old[i+1]-QCONST16(1.0f, DB_SHIFT)); + /* Compute mean increase */ + c=0; do { + for (i=2;i<end-1;i++) + { + opus_val16 x1, x2; + x1 = MAX16(0, newE[i]); + x2 = MAX16(0, spread_old[i]); + mean_diff = ADD32(mean_diff, EXTEND32(MAX16(0, SUB16(x1, x2)))); + } + } while (++c<C); + mean_diff = DIV32(mean_diff, C*(end-3)); + /*printf("%f %f %d\n", mean_diff, max_diff, count);*/ + return mean_diff > QCONST16(1.f, DB_SHIFT); +} + +/** Apply window and compute the MDCT for all sub-frames and + all channels in a frame */ +static void compute_mdcts(const CELTMode *mode, int shortBlocks, celt_sig * OPUS_RESTRICT in, + celt_sig * OPUS_RESTRICT out, int C, int CC, int LM, int upsample) +{ + const int overlap = OVERLAP(mode); + int N; + int B; + int shift; + int i, b, c; + if (shortBlocks) + { + B = shortBlocks; + N = mode->shortMdctSize; + shift = mode->maxLM; + } else { + B = 1; + N = mode->shortMdctSize<<LM; + shift = mode->maxLM-LM; + } + c=0; do { + for (b=0;b<B;b++) + { + /* Interleaving the sub-frames while doing the MDCTs */ + clt_mdct_forward(&mode->mdct, in+c*(B*N+overlap)+b*N, &out[b+c*N*B], mode->window, overlap, shift, B); + } + } while (++c<CC); + if (CC==2&&C==1) + { + for (i=0;i<B*N;i++) + out[i] = ADD32(HALF32(out[i]), HALF32(out[B*N+i])); + } + if (upsample != 1) + { + c=0; do + { + int bound = B*N/upsample; + for (i=0;i<bound;i++) + out[c*B*N+i] *= upsample; + for (;i<B*N;i++) + out[c*B*N+i] = 0; + } while (++c<C); + } +} + + +void celt_preemphasis(const opus_val16 * OPUS_RESTRICT pcmp, celt_sig * OPUS_RESTRICT inp, + int N, int CC, int upsample, const opus_val16 *coef, celt_sig *mem, int clip) +{ + int i; + opus_val16 coef0; + celt_sig m; + int Nu; + + coef0 = coef[0]; + + + Nu = N/upsample; + if (upsample!=1) + { + for (i=0;i<N;i++) + inp[i] = 0; + } + for (i=0;i<Nu;i++) + { + celt_sig x; + + x = SCALEIN(pcmp[CC*i]); +#ifndef OPUS_FIXED_POINT + /* Replace NaNs with zeros */ + if (!(x==x)) + x = 0; +#endif + inp[i*upsample] = x; + } + +#ifndef OPUS_FIXED_POINT + if (clip) + { + /* Clip input to avoid encoding non-portable files */ + for (i=0;i<Nu;i++) + inp[i*upsample] = MAX32(-65536.f, MIN32(65536.f,inp[i*upsample])); + } +#else + (void)clip; /* Avoids a warning about clip being unused. */ +#endif + m = *mem; +#ifdef CUSTOM_MODES + if (coef[1] != 0) + { + opus_val16 coef1 = coef[1]; + opus_val16 coef2 = coef[2]; + for (i=0;i<N;i++) + { + celt_sig x, tmp; + x = inp[i]; + /* Apply pre-emphasis */ + tmp = MULT16_16(coef2, x); + inp[i] = tmp + m; + m = MULT16_32_Q15(coef1, inp[i]) - MULT16_32_Q15(coef0, tmp); + } + } else +#endif + { + for (i=0;i<N;i++) + { + celt_sig x; + x = SHL32(inp[i], SIG_SHIFT); + /* Apply pre-emphasis */ + inp[i] = x + m; + m = - MULT16_32_Q15(coef0, x); + } + } + *mem = m; +} + + + +static opus_val32 l1_metric(const celt_norm *tmp, int N, int LM, opus_val16 bias) +{ + int i; + opus_val32 L1; + L1 = 0; + for (i=0;i<N;i++) + L1 += EXTEND32(ABS16(tmp[i])); + /* When in doubt, prefer good freq resolution */ + L1 = MAC16_32_Q15(L1, LM*bias, L1); + return L1; + +} + +static int tf_analysis(const CELTMode *m, int len, int isTransient, + int *tf_res, int lambda, celt_norm *X, int N0, int LM, + int *tf_sum, opus_val16 tf_estimate, int tf_chan) +{ + int i; + VARDECL(int, metric); + int cost0; + int cost1; + VARDECL(int, path0); + VARDECL(int, path1); + VARDECL(celt_norm, tmp); + VARDECL(celt_norm, tmp_1); + int sel; + int selcost[2]; + int tf_select=0; + opus_val16 bias; + + SAVE_STACK; + bias = MULT16_16_Q14(QCONST16(.04f,15), MAX16(-QCONST16(.25f,14), QCONST16(.5f,14)-tf_estimate)); + /*printf("%f ", bias);*/ + + ALLOC(metric, len, int); + ALLOC(tmp, (m->eBands[len]-m->eBands[len-1])<<LM, celt_norm); + ALLOC(tmp_1, (m->eBands[len]-m->eBands[len-1])<<LM, celt_norm); + ALLOC(path0, len, int); + ALLOC(path1, len, int); + + *tf_sum = 0; + for (i=0;i<len;i++) + { + int j, k, N; + int narrow; + opus_val32 L1, best_L1; + int best_level=0; + N = (m->eBands[i+1]-m->eBands[i])<<LM; + /* band is too narrow to be split down to LM=-1 */ + narrow = (m->eBands[i+1]-m->eBands[i])==1; + for (j=0;j<N;j++) + tmp[j] = X[tf_chan*N0 + j+(m->eBands[i]<<LM)]; + /* Just add the right channel if we're in stereo */ + /*if (C==2) + for (j=0;j<N;j++) + tmp[j] = ADD16(SHR16(tmp[j], 1),SHR16(X[N0+j+(m->eBands[i]<<LM)], 1));*/ + L1 = l1_metric(tmp, N, isTransient ? LM : 0, bias); + best_L1 = L1; + /* Check the -1 case for transients */ + if (isTransient && !narrow) + { + for (j=0;j<N;j++) + tmp_1[j] = tmp[j]; + haar1(tmp_1, N>>LM, 1<<LM); + L1 = l1_metric(tmp_1, N, LM+1, bias); + if (L1<best_L1) + { + best_L1 = L1; + best_level = -1; + } + } + /*printf ("%f ", L1);*/ + for (k=0;k<LM+!(isTransient||narrow);k++) + { + int B; + + if (isTransient) + B = (LM-k-1); + else + B = k+1; + + haar1(tmp, N>>k, 1<<k); + + L1 = l1_metric(tmp, N, B, bias); + + if (L1 < best_L1) + { + best_L1 = L1; + best_level = k+1; + } + } + /*printf ("%d ", isTransient ? LM-best_level : best_level);*/ + /* metric is in Q1 to be able to select the mid-point (-0.5) for narrower bands */ + if (isTransient) + metric[i] = 2*best_level; + else + metric[i] = -2*best_level; + *tf_sum += (isTransient ? LM : 0) - metric[i]/2; + /* For bands that can't be split to -1, set the metric to the half-way point to avoid + biasing the decision */ + if (narrow && (metric[i]==0 || metric[i]==-2*LM)) + metric[i]-=1; + /*printf("%d ", metric[i]);*/ + } + /*printf("\n");*/ + /* Search for the optimal tf resolution, including tf_select */ + tf_select = 0; + for (sel=0;sel<2;sel++) + { + cost0 = 0; + cost1 = isTransient ? 0 : lambda; + for (i=1;i<len;i++) + { + int curr0, curr1; + curr0 = IMIN(cost0, cost1 + lambda); + curr1 = IMIN(cost0 + lambda, cost1); + cost0 = curr0 + abs(metric[i]-2*tf_select_table[LM][4*isTransient+2*sel+0]); + cost1 = curr1 + abs(metric[i]-2*tf_select_table[LM][4*isTransient+2*sel+1]); + } + cost0 = IMIN(cost0, cost1); + selcost[sel]=cost0; + } + /* For now, we're conservative and only allow tf_select=1 for transients. + * If tests confirm it's useful for non-transients, we could allow it. */ + if (selcost[1]<selcost[0] && isTransient) + tf_select=1; + cost0 = 0; + cost1 = isTransient ? 0 : lambda; + /* Viterbi forward pass */ + for (i=1;i<len;i++) + { + int curr0, curr1; + int from0, from1; + + from0 = cost0; + from1 = cost1 + lambda; + if (from0 < from1) + { + curr0 = from0; + path0[i]= 0; + } else { + curr0 = from1; + path0[i]= 1; + } + + from0 = cost0 + lambda; + from1 = cost1; + if (from0 < from1) + { + curr1 = from0; + path1[i]= 0; + } else { + curr1 = from1; + path1[i]= 1; + } + cost0 = curr0 + abs(metric[i]-2*tf_select_table[LM][4*isTransient+2*tf_select+0]); + cost1 = curr1 + abs(metric[i]-2*tf_select_table[LM][4*isTransient+2*tf_select+1]); + } + tf_res[len-1] = cost0 < cost1 ? 0 : 1; + /* Viterbi backward pass to check the decisions */ + for (i=len-2;i>=0;i--) + { + if (tf_res[i+1] == 1) + tf_res[i] = path1[i+1]; + else + tf_res[i] = path0[i+1]; + } + /*printf("%d %f\n", *tf_sum, tf_estimate);*/ + RESTORE_STACK; +#ifdef FUZZING + tf_select = rand()&0x1; + tf_res[0] = rand()&0x1; + for (i=1;i<len;i++) + tf_res[i] = tf_res[i-1] ^ ((rand()&0xF) == 0); +#endif + return tf_select; +} + +static void tf_encode(int start, int end, int isTransient, int *tf_res, int LM, int tf_select, ec_enc *enc) +{ + int curr, i; + int tf_select_rsv; + int tf_changed; + int logp; + opus_uint32 budget; + opus_uint32 tell; + budget = enc->storage*8; + tell = ec_tell(enc); + logp = isTransient ? 2 : 4; + /* Reserve space to code the tf_select decision. */ + tf_select_rsv = LM>0 && tell+logp+1 <= budget; + budget -= tf_select_rsv; + curr = tf_changed = 0; + for (i=start;i<end;i++) + { + if (tell+logp<=budget) + { + ec_enc_bit_logp(enc, tf_res[i] ^ curr, logp); + tell = ec_tell(enc); + curr = tf_res[i]; + tf_changed |= curr; + } + else + tf_res[i] = curr; + logp = isTransient ? 4 : 5; + } + /* Only code tf_select if it would actually make a difference. */ + if (tf_select_rsv && + tf_select_table[LM][4*isTransient+0+tf_changed]!= + tf_select_table[LM][4*isTransient+2+tf_changed]) + ec_enc_bit_logp(enc, tf_select, 1); + else + tf_select = 0; + for (i=start;i<end;i++) + tf_res[i] = tf_select_table[LM][4*isTransient+2*tf_select+tf_res[i]]; + /*for(i=0;i<end;i++)printf("%d ", isTransient ? tf_res[i] : LM+tf_res[i]);printf("\n");*/ +} + + +static int alloc_trim_analysis(const CELTMode *m, const celt_norm *X, + const opus_val16 *bandLogE, int end, int LM, int C, int N0, + AnalysisInfo *analysis, opus_val16 *stereo_saving, opus_val16 tf_estimate, + int intensity, opus_val16 surround_trim) +{ + int i; + opus_val32 diff=0; + int c; + int trim_index = 5; + opus_val16 trim = QCONST16(5.f, 8); + opus_val16 logXC, logXC2; + if (C==2) + { + opus_val16 sum = 0; /* Q10 */ + opus_val16 minXC; /* Q10 */ + /* Compute inter-channel correlation for low frequencies */ + for (i=0;i<8;i++) + { + int j; + opus_val32 partial = 0; + for (j=m->eBands[i]<<LM;j<m->eBands[i+1]<<LM;j++) + partial = MAC16_16(partial, X[j], X[N0+j]); + sum = ADD16(sum, EXTRACT16(SHR32(partial, 18))); + } + sum = MULT16_16_Q15(QCONST16(1.f/8, 15), sum); + sum = MIN16(QCONST16(1.f, 10), ABS16(sum)); + minXC = sum; + for (i=8;i<intensity;i++) + { + int j; + opus_val32 partial = 0; + for (j=m->eBands[i]<<LM;j<m->eBands[i+1]<<LM;j++) + partial = MAC16_16(partial, X[j], X[N0+j]); + minXC = MIN16(minXC, ABS16(EXTRACT16(SHR32(partial, 18)))); + } + minXC = MIN16(QCONST16(1.f, 10), ABS16(minXC)); + /*printf ("%f\n", sum);*/ + if (sum > QCONST16(.995f,10)) + trim_index-=4; + else if (sum > QCONST16(.92f,10)) + trim_index-=3; + else if (sum > QCONST16(.85f,10)) + trim_index-=2; + else if (sum > QCONST16(.8f,10)) + trim_index-=1; + /* mid-side savings estimations based on the LF average*/ + logXC = celt_log2(QCONST32(1.001f, 20)-MULT16_16(sum, sum)); + /* mid-side savings estimations based on min correlation */ + logXC2 = MAX16(HALF16(logXC), celt_log2(QCONST32(1.001f, 20)-MULT16_16(minXC, minXC))); +#ifdef OPUS_FIXED_POINT + /* Compensate for Q20 vs Q14 input and convert output to Q8 */ + logXC = PSHR32(logXC-QCONST16(6.f, DB_SHIFT),DB_SHIFT-8); + logXC2 = PSHR32(logXC2-QCONST16(6.f, DB_SHIFT),DB_SHIFT-8); +#endif + + trim += MAX16(-QCONST16(4.f, 8), MULT16_16_Q15(QCONST16(.75f,15),logXC)); + *stereo_saving = MIN16(*stereo_saving + QCONST16(0.25f, 8), -HALF16(logXC2)); + } + + /* Estimate spectral tilt */ + c=0; do { + for (i=0;i<end-1;i++) + { + diff += bandLogE[i+c*m->nbEBands]*(opus_int32)(2+2*i-end); + } + } while (++c<C); + diff /= C*(end-1); + /*printf("%f\n", diff);*/ + if (diff > QCONST16(2.f, DB_SHIFT)) + trim_index--; + if (diff > QCONST16(8.f, DB_SHIFT)) + trim_index--; + if (diff < -QCONST16(4.f, DB_SHIFT)) + trim_index++; + if (diff < -QCONST16(10.f, DB_SHIFT)) + trim_index++; + trim -= MAX16(-QCONST16(2.f, 8), MIN16(QCONST16(2.f, 8), SHR16(diff+QCONST16(1.f, DB_SHIFT),DB_SHIFT-8)/6 )); + trim -= SHR16(surround_trim, DB_SHIFT-8); + trim -= 2*SHR16(tf_estimate, 14-8); +#ifndef DISABLE_FLOAT_API + if (analysis->valid) + { + trim -= MAX16(-QCONST16(2.f, 8), MIN16(QCONST16(2.f, 8), + (opus_val16)(QCONST16(2.f, 8)*(analysis->tonality_slope+.05f)))); + } +#endif + +#ifdef OPUS_FIXED_POINT + trim_index = PSHR32(trim, 8); +#else + trim_index = (int)floor(.5f+trim); +#endif + if (trim_index<0) + trim_index = 0; + if (trim_index>10) + trim_index = 10; + /*printf("%d\n", trim_index);*/ +#ifdef FUZZING + trim_index = rand()%11; +#endif + return trim_index; +} + +static int stereo_analysis(const CELTMode *m, const celt_norm *X, + int LM, int N0) +{ + int i; + int thetas; + opus_val32 sumLR = EPSILON, sumMS = EPSILON; + + /* Use the L1 norm to model the entropy of the L/R signal vs the M/S signal */ + for (i=0;i<13;i++) + { + int j; + for (j=m->eBands[i]<<LM;j<m->eBands[i+1]<<LM;j++) + { + opus_val32 L, R, M, S; + /* We cast to 32-bit first because of the -32768 case */ + L = EXTEND32(X[j]); + R = EXTEND32(X[N0+j]); + M = ADD32(L, R); + S = SUB32(L, R); + sumLR = ADD32(sumLR, ADD32(ABS32(L), ABS32(R))); + sumMS = ADD32(sumMS, ADD32(ABS32(M), ABS32(S))); + } + } + sumMS = MULT16_32_Q15(QCONST16(0.707107f, 15), sumMS); + thetas = 13; + /* We don't need thetas for lower bands with LM<=1 */ + if (LM<=1) + thetas -= 8; + return MULT16_32_Q15((m->eBands[13]<<(LM+1))+thetas, sumMS) + > MULT16_32_Q15(m->eBands[13]<<(LM+1), sumLR); +} + +static opus_val16 dynalloc_analysis(const opus_val16 *bandLogE, const opus_val16 *bandLogE2, + int nbEBands, int start, int end, int C, int *offsets, int lsb_depth, const opus_int16 *logN, + int isTransient, int vbr, int constrained_vbr, const opus_int16 *eBands, int LM, + int effectiveBytes, opus_int32 *tot_boost_, int lfe, opus_val16 *surround_dynalloc) +{ + int i, c; + opus_int32 tot_boost=0; + opus_val16 maxDepth; + VARDECL(opus_val16, follower); + VARDECL(opus_val16, noise_floor); + SAVE_STACK; + ALLOC(follower, C*nbEBands, opus_val16); + ALLOC(noise_floor, C*nbEBands, opus_val16); + for (i=0;i<nbEBands;i++) + offsets[i] = 0; + /* Dynamic allocation code */ + maxDepth=-QCONST16(31.9f, DB_SHIFT); + for (i=0;i<end;i++) + { + /* Noise floor must take into account eMeans, the depth, the width of the bands + and the preemphasis filter (approx. square of bark band ID) */ + noise_floor[i] = MULT16_16(QCONST16(0.0625f, DB_SHIFT),logN[i]) + +QCONST16(.5f,DB_SHIFT)+SHL16(9-lsb_depth,DB_SHIFT)-SHL16(eMeans[i],6) + +MULT16_16(QCONST16(.0062,DB_SHIFT),(i+5)*(i+5)); + } + c=0;do + { + for (i=0;i<end;i++) + maxDepth = MAX16(maxDepth, bandLogE[c*nbEBands+i]-noise_floor[i]); + } while (++c<C); + /* Make sure that dynamic allocation can't make us bust the budget */ + if (effectiveBytes > 50 && LM>=1 && !lfe) + { + int last=0; + c=0;do + { + follower[c*nbEBands] = bandLogE2[c*nbEBands]; + for (i=1;i<end;i++) + { + /* The last band to be at least 3 dB higher than the previous one + is the last we'll consider. Otherwise, we run into problems on + bandlimited signals. */ + if (bandLogE2[c*nbEBands+i] > bandLogE2[c*nbEBands+i-1]+QCONST16(.5f,DB_SHIFT)) + last=i; + follower[c*nbEBands+i] = MIN16(follower[c*nbEBands+i-1]+QCONST16(1.5f,DB_SHIFT), bandLogE2[c*nbEBands+i]); + } + for (i=last-1;i>=0;i--) + follower[c*nbEBands+i] = MIN16(follower[c*nbEBands+i], MIN16(follower[c*nbEBands+i+1]+QCONST16(2.f,DB_SHIFT), bandLogE2[c*nbEBands+i])); + for (i=0;i<end;i++) + follower[c*nbEBands+i] = MAX16(follower[c*nbEBands+i], noise_floor[i]); + } while (++c<C); + if (C==2) + { + for (i=start;i<end;i++) + { + /* Consider 24 dB "cross-talk" */ + follower[nbEBands+i] = MAX16(follower[nbEBands+i], follower[ i]-QCONST16(4.f,DB_SHIFT)); + follower[ i] = MAX16(follower[ i], follower[nbEBands+i]-QCONST16(4.f,DB_SHIFT)); + follower[i] = HALF16(MAX16(0, bandLogE[i]-follower[i]) + MAX16(0, bandLogE[nbEBands+i]-follower[nbEBands+i])); + } + } else { + for (i=start;i<end;i++) + { + follower[i] = MAX16(0, bandLogE[i]-follower[i]); + } + } + for (i=start;i<end;i++) + follower[i] = MAX16(follower[i], surround_dynalloc[i]); + /* For non-transient CBR/CVBR frames, halve the dynalloc contribution */ + if ((!vbr || constrained_vbr)&&!isTransient) + { + for (i=start;i<end;i++) + follower[i] = HALF16(follower[i]); + } + for (i=start;i<end;i++) + { + int width; + int boost; + int boost_bits; + + if (i<8) + follower[i] *= 2; + if (i>=12) + follower[i] = HALF16(follower[i]); + follower[i] = MIN16(follower[i], QCONST16(4, DB_SHIFT)); + + width = C*(eBands[i+1]-eBands[i])<<LM; + if (width<6) + { + boost = (int)SHR32(EXTEND32(follower[i]),DB_SHIFT); + boost_bits = boost*width<<BITRES; + } else if (width > 48) { + boost = (int)SHR32(EXTEND32(follower[i])*8,DB_SHIFT); + boost_bits = (boost*width<<BITRES)/8; + } else { + boost = (int)SHR32(EXTEND32(follower[i])*width/6,DB_SHIFT); + boost_bits = boost*6<<BITRES; + } + /* For CBR and non-transient CVBR frames, limit dynalloc to 1/4 of the bits */ + if ((!vbr || (constrained_vbr&&!isTransient)) + && (tot_boost+boost_bits)>>BITRES>>3 > effectiveBytes/4) + { + opus_int32 cap = ((effectiveBytes/4)<<BITRES<<3); + offsets[i] = cap-tot_boost; + tot_boost = cap; + break; + } else { + offsets[i] = boost; + tot_boost += boost_bits; + } + } + } + *tot_boost_ = tot_boost; + RESTORE_STACK; + return maxDepth; +} + + +static int run_prefilter(CELTEncoder *st, celt_sig *in, celt_sig *prefilter_mem, int CC, int N, + int prefilter_tapset, int *pitch, opus_val16 *gain, int *qgain, int enabled, int nbAvailableBytes) +{ + int c; + VARDECL(celt_sig, _pre); + celt_sig *pre[2]; + const CELTMode *mode; + int pitch_index; + opus_val16 gain1; + opus_val16 pf_threshold; + int pf_on; + int qg; + SAVE_STACK; + + mode = st->mode; + ALLOC(_pre, CC*(N+COMBFILTER_MAXPERIOD), celt_sig); + + pre[0] = _pre; + pre[1] = _pre + (N+COMBFILTER_MAXPERIOD); + + + c=0; do { + OPUS_COPY(pre[c], prefilter_mem+c*COMBFILTER_MAXPERIOD, COMBFILTER_MAXPERIOD); + OPUS_COPY(pre[c]+COMBFILTER_MAXPERIOD, in+c*(N+st->overlap)+st->overlap, N); + } while (++c<CC); + + if (enabled) + { + VARDECL(opus_val16, pitch_buf); + ALLOC(pitch_buf, (COMBFILTER_MAXPERIOD+N)>>1, opus_val16); + + 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, + st->arch); + pitch_index = COMBFILTER_MAXPERIOD-pitch_index; + + gain1 = remove_doubling(pitch_buf, COMBFILTER_MAXPERIOD, COMBFILTER_MINPERIOD, + N, &pitch_index, st->prefilter_period, st->prefilter_gain); + if (pitch_index > COMBFILTER_MAXPERIOD-2) + pitch_index = COMBFILTER_MAXPERIOD-2; + gain1 = MULT16_16_Q15(QCONST16(.7f,15),gain1); + /*printf("%d %d %f %f\n", pitch_change, pitch_index, gain1, st->analysis.tonality);*/ + if (st->loss_rate>2) + gain1 = HALF32(gain1); + if (st->loss_rate>4) + gain1 = HALF32(gain1); + if (st->loss_rate>8) + gain1 = 0; + } else { + gain1 = 0; + pitch_index = COMBFILTER_MINPERIOD; + } + + /* Gain threshold for enabling the prefilter/postfilter */ + pf_threshold = QCONST16(.2f,15); + + /* Adjusting the threshold based on rate and continuity */ + if (abs(pitch_index-st->prefilter_period)*10>pitch_index) + pf_threshold += QCONST16(.2f,15); + if (nbAvailableBytes<25) + pf_threshold += QCONST16(.1f,15); + if (nbAvailableBytes<35) + pf_threshold += QCONST16(.1f,15); + if (st->prefilter_gain > QCONST16(.4f,15)) + pf_threshold -= QCONST16(.1f,15); + if (st->prefilter_gain > QCONST16(.55f,15)) + pf_threshold -= QCONST16(.1f,15); + + /* Hard threshold at 0.2 */ + pf_threshold = MAX16(pf_threshold, QCONST16(.2f,15)); + if (gain1<pf_threshold) + { + gain1 = 0; + pf_on = 0; + qg = 0; + } else { + /*This block is not gated by a total bits check only because + of the nbAvailableBytes check above.*/ + if (ABS16(gain1-st->prefilter_gain)<QCONST16(.1f,15)) + gain1=st->prefilter_gain; + +#ifdef OPUS_FIXED_POINT + qg = ((gain1+1536)>>10)/3-1; +#else + qg = (int)floor(.5f+gain1*32/3)-1; +#endif + qg = IMAX(0, IMIN(7, qg)); + gain1 = QCONST16(0.09375f,15)*(qg+1); + pf_on = 1; + } + /*printf("%d %f\n", pitch_index, gain1);*/ + + c=0; do { + int offset = mode->shortMdctSize-st->overlap; + st->prefilter_period=IMAX(st->prefilter_period, COMBFILTER_MINPERIOD); + OPUS_COPY(in+c*(N+st->overlap), st->in_mem+c*(st->overlap), st->overlap); + if (offset) + comb_filter(in+c*(N+st->overlap)+st->overlap, pre[c]+COMBFILTER_MAXPERIOD, + st->prefilter_period, st->prefilter_period, offset, -st->prefilter_gain, -st->prefilter_gain, + st->prefilter_tapset, st->prefilter_tapset, NULL, 0); + + comb_filter(in+c*(N+st->overlap)+st->overlap+offset, pre[c]+COMBFILTER_MAXPERIOD+offset, + st->prefilter_period, pitch_index, N-offset, -st->prefilter_gain, -gain1, + st->prefilter_tapset, prefilter_tapset, mode->window, st->overlap); + OPUS_COPY(st->in_mem+c*(st->overlap), in+c*(N+st->overlap)+N, st->overlap); + + if (N>COMBFILTER_MAXPERIOD) + { + OPUS_MOVE(prefilter_mem+c*COMBFILTER_MAXPERIOD, pre[c]+N, COMBFILTER_MAXPERIOD); + } else { + OPUS_MOVE(prefilter_mem+c*COMBFILTER_MAXPERIOD, prefilter_mem+c*COMBFILTER_MAXPERIOD+N, COMBFILTER_MAXPERIOD-N); + OPUS_MOVE(prefilter_mem+c*COMBFILTER_MAXPERIOD+COMBFILTER_MAXPERIOD-N, pre[c]+COMBFILTER_MAXPERIOD, N); + } + } while (++c<CC); + + RESTORE_STACK; + *gain = gain1; + *pitch = pitch_index; + *qgain = qg; + return pf_on; +} + +static int compute_vbr(const CELTMode *mode, AnalysisInfo *analysis, opus_int32 base_target, + int LM, opus_int32 bitrate, int lastCodedBands, int C, int intensity, + int constrained_vbr, opus_val16 stereo_saving, int tot_boost, + opus_val16 tf_estimate, int pitch_change, opus_val16 maxDepth, + int variable_duration, int lfe, int has_surround_mask, opus_val16 surround_masking, + opus_val16 temporal_vbr) +{ + /* The target rate in 8th bits per frame */ + opus_int32 target; + int coded_bins; + int coded_bands; + opus_val16 tf_calibration; + int nbEBands; + const opus_int16 *eBands; + + nbEBands = mode->nbEBands; + eBands = mode->eBands; + + coded_bands = lastCodedBands ? lastCodedBands : nbEBands; + coded_bins = eBands[coded_bands]<<LM; + if (C==2) + coded_bins += eBands[IMIN(intensity, coded_bands)]<<LM; + + target = base_target; + + /*printf("%f %f %f %f %d %d ", st->analysis.activity, st->analysis.tonality, tf_estimate, st->stereo_saving, tot_boost, coded_bands);*/ +#ifndef DISABLE_FLOAT_API + if (analysis->valid && analysis->activity<.4) + target -= (opus_int32)((coded_bins<<BITRES)*(.4f-analysis->activity)); +#endif + /* Stereo savings */ + if (C==2) + { + int coded_stereo_bands; + int coded_stereo_dof; + opus_val16 max_frac; + coded_stereo_bands = IMIN(intensity, coded_bands); + coded_stereo_dof = (eBands[coded_stereo_bands]<<LM)-coded_stereo_bands; + /* Maximum fraction of the bits we can save if the signal is mono. */ + max_frac = DIV32_16(MULT16_16(QCONST16(0.8f, 15), coded_stereo_dof), coded_bins); + stereo_saving = MIN16(stereo_saving, QCONST16(1.f, 8)); + /*printf("%d %d %d ", coded_stereo_dof, coded_bins, tot_boost);*/ + target -= (opus_int32)MIN32(MULT16_32_Q15(max_frac,target), + SHR32(MULT16_16(stereo_saving-QCONST16(0.1f,8),(coded_stereo_dof<<BITRES)),8)); + } + /* Boost the rate according to dynalloc (minus the dynalloc average for calibration). */ + target += tot_boost-(16<<LM); + /* Apply transient boost, compensating for average boost. */ + tf_calibration = variable_duration==OPUS_FRAMESIZE_VARIABLE ? + QCONST16(0.02f,14) : QCONST16(0.04f,14); + target += (opus_int32)SHL32(MULT16_32_Q15(tf_estimate-tf_calibration, target),1); + +#ifndef DISABLE_FLOAT_API + /* Apply tonality boost */ + if (analysis->valid && !lfe) + { + opus_int32 tonal_target; + float tonal; + + /* Tonality boost (compensating for the average). */ + tonal = MAX16(0.f,analysis->tonality-.15f)-0.09f; + tonal_target = target + (opus_int32)((coded_bins<<BITRES)*1.2f*tonal); + if (pitch_change) + tonal_target += (opus_int32)((coded_bins<<BITRES)*.8f); + /*printf("%f %f ", analysis->tonality, tonal);*/ + target = tonal_target; + } +#endif + + if (has_surround_mask&&!lfe) + { + opus_int32 surround_target = target + (opus_int32)SHR32(MULT16_16(surround_masking,coded_bins<<BITRES), DB_SHIFT); + /*printf("%f %d %d %d %d %d %d ", surround_masking, coded_bins, st->end, st->intensity, surround_target, target, st->bitrate);*/ + target = IMAX(target/4, surround_target); + } + + { + opus_int32 floor_depth; + int bins; + bins = eBands[nbEBands-2]<<LM; + /*floor_depth = SHR32(MULT16_16((C*bins<<BITRES),celt_log2(SHL32(MAX16(1,sample_max),13))), DB_SHIFT);*/ + floor_depth = (opus_int32)SHR32(MULT16_16((C*bins<<BITRES),maxDepth), DB_SHIFT); + floor_depth = IMAX(floor_depth, target>>2); + target = IMIN(target, floor_depth); + /*printf("%f %d\n", maxDepth, floor_depth);*/ + } + + if ((!has_surround_mask||lfe) && (constrained_vbr || bitrate<64000)) + { + opus_val16 rate_factor; +#ifdef OPUS_FIXED_POINT + rate_factor = MAX16(0,(bitrate-32000)); +#else + rate_factor = MAX16(0,(1.f/32768)*(bitrate-32000)); +#endif + if (constrained_vbr) + rate_factor = MIN16(rate_factor, QCONST16(0.67f, 15)); + target = base_target + (opus_int32)MULT16_32_Q15(rate_factor, target-base_target); + + } + + if (!has_surround_mask && tf_estimate < QCONST16(.2f, 14)) + { + opus_val16 amount; + opus_val16 tvbr_factor; + amount = MULT16_16_Q15(QCONST16(.0000031f, 30), IMAX(0, IMIN(32000, 96000-bitrate))); + tvbr_factor = SHR32(MULT16_16(temporal_vbr, amount), DB_SHIFT); + target += (opus_int32)MULT16_32_Q15(tvbr_factor, target); + } + + /* Don't allow more than doubling the rate */ + target = IMIN(2*base_target, target); + + return target; +} + +int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes, ec_enc *enc) +{ + int i, c, N; + opus_int32 bits; + ec_enc _enc; + VARDECL(celt_sig, in); + VARDECL(celt_sig, freq); + VARDECL(celt_norm, X); + VARDECL(celt_ener, bandE); + VARDECL(opus_val16, bandLogE); + VARDECL(opus_val16, bandLogE2); + VARDECL(int, fine_quant); + VARDECL(opus_val16, error); + VARDECL(int, pulses); + VARDECL(int, cap); + VARDECL(int, offsets); + VARDECL(int, fine_priority); + VARDECL(int, tf_res); + VARDECL(unsigned char, collapse_masks); + celt_sig *prefilter_mem; + opus_val16 *oldBandE, *oldLogE, *oldLogE2; + int shortBlocks=0; + int isTransient=0; + const int CC = st->channels; + const int C = st->stream_channels; + int LM, M; + int tf_select; + int nbFilledBytes, nbAvailableBytes; + int effEnd; + int codedBands; + int tf_sum; + int alloc_trim; + int pitch_index=COMBFILTER_MINPERIOD; + opus_val16 gain1 = 0; + int dual_stereo=0; + int effectiveBytes; + int dynalloc_logp; + opus_int32 vbr_rate; + opus_int32 total_bits; + opus_int32 total_boost; + opus_int32 balance; + opus_int32 tell; + int prefilter_tapset=0; + int pf_on; + int anti_collapse_rsv; + int anti_collapse_on=0; + int silence=0; + int tf_chan = 0; + opus_val16 tf_estimate; + int pitch_change=0; + opus_int32 tot_boost; + opus_val32 sample_max; + opus_val16 maxDepth; + const OpusCustomMode *mode; + int nbEBands; + int overlap; + const opus_int16 *eBands; + int secondMdct; + int signalBandwidth; + int transient_got_disabled=0; + opus_val16 surround_masking=0; + opus_val16 temporal_vbr=0; + opus_val16 surround_trim = 0; + opus_int32 equiv_rate = 510000; + VARDECL(opus_val16, surround_dynalloc); + ALLOC_STACK; + + mode = st->mode; + nbEBands = mode->nbEBands; + overlap = mode->overlap; + eBands = mode->eBands; + tf_estimate = 0; + if (nbCompressedBytes<2 || pcm==NULL) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + + frame_size *= st->upsample; + for (LM=0;LM<=mode->maxLM;LM++) + if (mode->shortMdctSize<<LM==frame_size) + break; + if (LM>mode->maxLM) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + M=1<<LM; + N = M*mode->shortMdctSize; + + prefilter_mem = st->in_mem+CC*(st->overlap); + oldBandE = (opus_val16*)(st->in_mem+CC*(st->overlap+COMBFILTER_MAXPERIOD)); + oldLogE = oldBandE + CC*nbEBands; + oldLogE2 = oldLogE + CC*nbEBands; + + if (enc==NULL) + { + tell=1; + nbFilledBytes=0; + } else { + tell=ec_tell(enc); + nbFilledBytes=(tell+4)>>3; + } + +#ifdef CUSTOM_MODES + if (st->signalling && enc==NULL) + { + int tmp = (mode->effEBands-st->end)>>1; + st->end = IMAX(1, mode->effEBands-tmp); + compressed[0] = tmp<<5; + compressed[0] |= LM<<3; + compressed[0] |= (C==2)<<2; + /* Convert "standard mode" to Opus header */ + if (mode->Fs==48000 && mode->shortMdctSize==120) + { + int c0 = toOpus(compressed[0]); + if (c0<0) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + compressed[0] = c0; + } + compressed++; + nbCompressedBytes--; + } +#else + celt_assert(st->signalling==0); +#endif + + /* Can't produce more than 1275 output bytes */ + nbCompressedBytes = IMIN(nbCompressedBytes,1275); + nbAvailableBytes = nbCompressedBytes - nbFilledBytes; + + if (st->vbr && st->bitrate!=OPUS_BITRATE_MAX) + { + opus_int32 den=mode->Fs>>BITRES; + vbr_rate=(st->bitrate*frame_size+(den>>1))/den; +#ifdef CUSTOM_MODES + if (st->signalling) + vbr_rate -= 8<<BITRES; +#endif + effectiveBytes = vbr_rate>>(3+BITRES); + } else { + opus_int32 tmp; + vbr_rate = 0; + tmp = st->bitrate*frame_size; + if (tell>1) + tmp += tell; + if (st->bitrate!=OPUS_BITRATE_MAX) + nbCompressedBytes = IMAX(2, IMIN(nbCompressedBytes, + (tmp+4*mode->Fs)/(8*mode->Fs)-!!st->signalling)); + effectiveBytes = nbCompressedBytes; + } + if (st->bitrate != OPUS_BITRATE_MAX) + equiv_rate = st->bitrate - (40*C+20)*((400>>LM) - 50); + + if (enc==NULL) + { + ec_enc_init(&_enc, compressed, nbCompressedBytes); + enc = &_enc; + } + + if (vbr_rate>0) + { + /* Computes the max bit-rate allowed in VBR mode to avoid violating the + target rate and buffering. + We must do this up front so that bust-prevention logic triggers + correctly if we don't have enough bits. */ + if (st->constrained_vbr) + { + opus_int32 vbr_bound; + opus_int32 max_allowed; + /* We could use any multiple of vbr_rate as bound (depending on the + delay). + This is clamped to ensure we use at least two bytes if the encoder + was entirely empty, but to allow 0 in hybrid mode. */ + vbr_bound = vbr_rate; + max_allowed = IMIN(IMAX(tell==1?2:0, + (vbr_rate+vbr_bound-st->vbr_reservoir)>>(BITRES+3)), + nbAvailableBytes); + if(max_allowed < nbAvailableBytes) + { + nbCompressedBytes = nbFilledBytes+max_allowed; + nbAvailableBytes = max_allowed; + ec_enc_shrink(enc, nbCompressedBytes); + } + } + } + total_bits = nbCompressedBytes*8; + + effEnd = st->end; + if (effEnd > mode->effEBands) + effEnd = mode->effEBands; + + ALLOC(in, CC*(N+st->overlap), celt_sig); + + sample_max=MAX32(st->overlap_max, celt_maxabs16(pcm, C*(N-overlap)/st->upsample)); + st->overlap_max=celt_maxabs16(pcm+C*(N-overlap)/st->upsample, C*overlap/st->upsample); + sample_max=MAX32(sample_max, st->overlap_max); +#ifdef OPUS_FIXED_POINT + silence = (sample_max==0); +#else + silence = (sample_max <= (opus_val16)1/(1<<st->lsb_depth)); +#endif +#ifdef FUZZING + if ((rand()&0x3F)==0) + silence = 1; +#endif + if (tell==1) + ec_enc_bit_logp(enc, silence, 15); + else + silence=0; + if (silence) + { + /*In VBR mode there is no need to send more than the minimum. */ + if (vbr_rate>0) + { + effectiveBytes=nbCompressedBytes=IMIN(nbCompressedBytes, nbFilledBytes+2); + total_bits=nbCompressedBytes*8; + nbAvailableBytes=2; + ec_enc_shrink(enc, nbCompressedBytes); + } + /* Pretend we've filled all the remaining bits with zeros + (that's what the initialiser did anyway) */ + tell = nbCompressedBytes*8; + enc->nbits_total+=tell-ec_tell(enc); + } + c=0; do { + celt_preemphasis(pcm+c, in+c*(N+st->overlap)+st->overlap, N, CC, st->upsample, + mode->preemph, st->preemph_memE+c, st->clip); + } while (++c<CC); + + + + /* Find pitch period and gain */ + { + int enabled; + int qg; + enabled = ((st->lfe&&nbAvailableBytes>3) || nbAvailableBytes>12*C) && st->start==0 && !silence && !st->disable_pf + && st->complexity >= 5 && !(st->consec_transient && LM!=3 && st->variable_duration==OPUS_FRAMESIZE_VARIABLE); + + prefilter_tapset = st->tapset_decision; + pf_on = run_prefilter(st, in, prefilter_mem, CC, N, prefilter_tapset, &pitch_index, &gain1, &qg, enabled, nbAvailableBytes); + if ((gain1 > QCONST16(.4f,15) || st->prefilter_gain > QCONST16(.4f,15)) && (!st->analysis.valid || st->analysis.tonality > .3) + && (pitch_index > 1.26*st->prefilter_period || pitch_index < .79*st->prefilter_period)) + pitch_change = 1; + if (pf_on==0) + { + if(st->start==0 && tell+16<=total_bits) + ec_enc_bit_logp(enc, 0, 1); + } else { + /*This block is not gated by a total bits check only because + of the nbAvailableBytes check above.*/ + int octave; + ec_enc_bit_logp(enc, 1, 1); + pitch_index += 1; + octave = EC_ILOG(pitch_index)-5; + ec_enc_uint(enc, octave, 6); + ec_enc_bits(enc, pitch_index-(16<<octave), 4+octave); + pitch_index -= 1; + ec_enc_bits(enc, qg, 3); + ec_enc_icdf(enc, prefilter_tapset, tapset_icdf, 2); + } + } + + isTransient = 0; + shortBlocks = 0; + if (st->complexity >= 1 && !st->lfe) + { + isTransient = transient_analysis(in, N+st->overlap, CC, + &tf_estimate, &tf_chan); + } + if (LM>0 && ec_tell(enc)+3<=total_bits) + { + if (isTransient) + shortBlocks = M; + } else { + isTransient = 0; + transient_got_disabled=1; + } + + ALLOC(freq, CC*N, celt_sig); /**< Interleaved signal MDCTs */ + ALLOC(bandE,nbEBands*CC, celt_ener); + ALLOC(bandLogE,nbEBands*CC, opus_val16); + + secondMdct = shortBlocks && st->complexity>=8; + ALLOC(bandLogE2, C*nbEBands, opus_val16); + if (secondMdct) + { + compute_mdcts(mode, 0, in, freq, C, CC, LM, st->upsample); + compute_band_energies(mode, freq, bandE, effEnd, C, M); + amp2Log2(mode, effEnd, st->end, bandE, bandLogE2, C); + for (i=0;i<C*nbEBands;i++) + bandLogE2[i] += HALF16(SHL16(LM, DB_SHIFT)); + } + + compute_mdcts(mode, shortBlocks, in, freq, C, CC, LM, st->upsample); + if (CC==2&&C==1) + tf_chan = 0; + compute_band_energies(mode, freq, bandE, effEnd, C, M); + + if (st->lfe) + { + for (i=2;i<st->end;i++) + { + bandE[i] = IMIN(bandE[i], MULT16_32_Q15(QCONST16(1e-4f,15),bandE[0])); + bandE[i] = MAX32(bandE[i], EPSILON); + } + } + amp2Log2(mode, effEnd, st->end, bandE, bandLogE, C); + + ALLOC(surround_dynalloc, C*nbEBands, opus_val16); + for(i=0;i<st->end;i++) + surround_dynalloc[i] = 0; + /* This computes how much masking takes place between surround channels */ + if (st->start==0&&st->energy_mask&&!st->lfe) + { + int mask_end; + int midband; + int count_dynalloc; + opus_val32 mask_avg=0; + opus_val32 diff=0; + int count=0; + mask_end = IMAX(2,st->lastCodedBands); + for (c=0;c<C;c++) + { + for(i=0;i<mask_end;i++) + { + opus_val16 mask; + mask = MAX16(MIN16(st->energy_mask[nbEBands*c+i], + QCONST16(.25f, DB_SHIFT)), -QCONST16(2.0f, DB_SHIFT)); + if (mask > 0) + mask = HALF16(mask); + mask_avg += MULT16_16(mask, eBands[i+1]-eBands[i]); + count += eBands[i+1]-eBands[i]; + diff += MULT16_16(mask, 1+2*i-mask_end); + } + } + mask_avg = DIV32_16(mask_avg,count); + mask_avg += QCONST16(.2f, DB_SHIFT); + diff = diff*6/(C*(mask_end-1)*(mask_end+1)*mask_end); + /* Again, being conservative */ + diff = HALF32(diff); + diff = MAX32(MIN32(diff, QCONST32(.031f, DB_SHIFT)), -QCONST32(.031f, DB_SHIFT)); + /* Find the band that's in the middle of the coded spectrum */ + for (midband=0;eBands[midband+1] < eBands[mask_end]/2;midband++); + count_dynalloc=0; + for(i=0;i<mask_end;i++) + { + opus_val32 lin; + opus_val16 unmask; + lin = mask_avg + diff*(i-midband); + if (C==2) + unmask = MAX16(st->energy_mask[i], st->energy_mask[nbEBands+i]); + else + unmask = st->energy_mask[i]; + unmask = MIN16(unmask, QCONST16(.0f, DB_SHIFT)); + unmask -= lin; + if (unmask > QCONST16(.25f, DB_SHIFT)) + { + surround_dynalloc[i] = unmask - QCONST16(.25f, DB_SHIFT); + count_dynalloc++; + } + } + if (count_dynalloc>=3) + { + /* If we need dynalloc in many bands, it's probably because our + initial masking rate was too low. */ + mask_avg += QCONST16(.25f, DB_SHIFT); + if (mask_avg>0) + { + /* Something went really wrong in the original calculations, + disabling masking. */ + mask_avg = 0; + diff = 0; + for(i=0;i<mask_end;i++) + surround_dynalloc[i] = 0; + } else { + for(i=0;i<mask_end;i++) + surround_dynalloc[i] = MAX16(0, surround_dynalloc[i]-QCONST16(.25f, DB_SHIFT)); + } + } + mask_avg += QCONST16(.2f, DB_SHIFT); + /* Convert to 1/64th units used for the trim */ + surround_trim = 64*diff; + /*printf("%d %d ", mask_avg, surround_trim);*/ + surround_masking = mask_avg; + } + /* Temporal VBR (but not for LFE) */ + if (!st->lfe) + { + opus_val16 follow=-QCONST16(10.0f,DB_SHIFT); + opus_val32 frame_avg=0; + opus_val16 offset = shortBlocks?HALF16(SHL16(LM, DB_SHIFT)):0; + for(i=st->start;i<st->end;i++) + { + follow = MAX16(follow-QCONST16(1.f, DB_SHIFT), bandLogE[i]-offset); + if (C==2) + follow = MAX16(follow, bandLogE[i+nbEBands]-offset); + frame_avg += follow; + } + frame_avg /= (st->end-st->start); + temporal_vbr = SUB16(frame_avg,st->spec_avg); + temporal_vbr = MIN16(QCONST16(3.f, DB_SHIFT), MAX16(-QCONST16(1.5f, DB_SHIFT), temporal_vbr)); + st->spec_avg += MULT16_16_Q15(QCONST16(.02f, 15), temporal_vbr); + } + /*for (i=0;i<21;i++) + printf("%f ", bandLogE[i]); + printf("\n");*/ + + if (!secondMdct) + { + for (i=0;i<C*nbEBands;i++) + bandLogE2[i] = bandLogE[i]; + } + + /* Last chance to catch any transient we might have missed in the + time-domain analysis */ + if (LM>0 && ec_tell(enc)+3<=total_bits && !isTransient && st->complexity>=5 && !st->lfe) + { + if (patch_transient_decision(bandLogE, oldBandE, nbEBands, st->end, C)) + { + isTransient = 1; + shortBlocks = M; + compute_mdcts(mode, shortBlocks, in, freq, C, CC, LM, st->upsample); + compute_band_energies(mode, freq, bandE, effEnd, C, M); + amp2Log2(mode, effEnd, st->end, bandE, bandLogE, C); + /* Compensate for the scaling of short vs long mdcts */ + for (i=0;i<C*nbEBands;i++) + bandLogE2[i] += HALF16(SHL16(LM, DB_SHIFT)); + tf_estimate = QCONST16(.2f,14); + } + } + + if (LM>0 && ec_tell(enc)+3<=total_bits) + ec_enc_bit_logp(enc, isTransient, 3); + + ALLOC(X, C*N, celt_norm); /**< Interleaved normalised MDCTs */ + + /* Band normalisation */ + normalise_bands(mode, freq, X, bandE, effEnd, C, M); + + ALLOC(tf_res, nbEBands, int); + /* Disable variable tf resolution for hybrid and at very low bitrate */ + if (effectiveBytes>=15*C && st->start==0 && st->complexity>=2 && !st->lfe) + { + int lambda; + if (effectiveBytes<40) + lambda = 12; + else if (effectiveBytes<60) + lambda = 6; + else if (effectiveBytes<100) + lambda = 4; + else + lambda = 3; + lambda*=2; + tf_select = tf_analysis(mode, effEnd, isTransient, tf_res, lambda, X, N, LM, &tf_sum, tf_estimate, tf_chan); + for (i=effEnd;i<st->end;i++) + tf_res[i] = tf_res[effEnd-1]; + } else { + tf_sum = 0; + for (i=0;i<st->end;i++) + tf_res[i] = isTransient; + tf_select=0; + } + + ALLOC(error, C*nbEBands, opus_val16); + quant_coarse_energy(mode, st->start, st->end, effEnd, bandLogE, + oldBandE, total_bits, error, enc, + C, LM, nbAvailableBytes, st->force_intra, + &st->delayedIntra, st->complexity >= 4, st->loss_rate, st->lfe); + + tf_encode(st->start, st->end, isTransient, tf_res, LM, tf_select, enc); + + if (ec_tell(enc)+4<=total_bits) + { + if (st->lfe) + { + st->tapset_decision = 0; + st->spread_decision = SPREAD_NORMAL; + } else if (shortBlocks || st->complexity < 3 || nbAvailableBytes < 10*C || st->start != 0) + { + if (st->complexity == 0) + st->spread_decision = SPREAD_NONE; + else + st->spread_decision = SPREAD_NORMAL; + } else { + /* Disable new spreading+tapset estimator until we can show it works + better than the old one. So far it seems like spreading_decision() + works best. */ +#if 0 + if (st->analysis.valid) + { + static const opus_val16 spread_thresholds[3] = {-QCONST16(.6f, 15), -QCONST16(.2f, 15), -QCONST16(.07f, 15)}; + static const opus_val16 spread_histeresis[3] = {QCONST16(.15f, 15), QCONST16(.07f, 15), QCONST16(.02f, 15)}; + static const opus_val16 tapset_thresholds[2] = {QCONST16(.0f, 15), QCONST16(.15f, 15)}; + static const opus_val16 tapset_histeresis[2] = {QCONST16(.1f, 15), QCONST16(.05f, 15)}; + st->spread_decision = hysteresis_decision(-st->analysis.tonality, spread_thresholds, spread_histeresis, 3, st->spread_decision); + st->tapset_decision = hysteresis_decision(st->analysis.tonality_slope, tapset_thresholds, tapset_histeresis, 2, st->tapset_decision); + } else +#endif + { + st->spread_decision = spreading_decision(mode, X, + &st->tonal_average, st->spread_decision, &st->hf_average, + &st->tapset_decision, pf_on&&!shortBlocks, effEnd, C, M); + } + /*printf("%d %d\n", st->tapset_decision, st->spread_decision);*/ + /*printf("%f %d %f %d\n\n", st->analysis.tonality, st->spread_decision, st->analysis.tonality_slope, st->tapset_decision);*/ + } + ec_enc_icdf(enc, st->spread_decision, spread_icdf, 5); + } + + ALLOC(offsets, nbEBands, int); + + maxDepth = dynalloc_analysis(bandLogE, bandLogE2, nbEBands, st->start, st->end, C, offsets, + st->lsb_depth, mode->logN, isTransient, st->vbr, st->constrained_vbr, + eBands, LM, effectiveBytes, &tot_boost, st->lfe, surround_dynalloc); + /* For LFE, everything interesting is in the first band */ + if (st->lfe) + offsets[0] = IMIN(8, effectiveBytes/3); + ALLOC(cap, nbEBands, int); + init_caps(mode,cap,LM,C); + + dynalloc_logp = 6; + total_bits<<=BITRES; + total_boost = 0; + tell = ec_tell_frac(enc); + for (i=st->start;i<st->end;i++) + { + int width, quanta; + int dynalloc_loop_logp; + int boost; + int j; + width = C*(eBands[i+1]-eBands[i])<<LM; + /* quanta is 6 bits, but no more than 1 bit/sample + and no less than 1/8 bit/sample */ + quanta = IMIN(width<<BITRES, IMAX(6<<BITRES, width)); + dynalloc_loop_logp = dynalloc_logp; + boost = 0; + for (j = 0; tell+(dynalloc_loop_logp<<BITRES) < total_bits-total_boost + && boost < cap[i]; j++) + { + int flag; + flag = j<offsets[i]; + ec_enc_bit_logp(enc, flag, dynalloc_loop_logp); + tell = ec_tell_frac(enc); + if (!flag) + break; + boost += quanta; + total_boost += quanta; + dynalloc_loop_logp = 1; + } + /* Making dynalloc more likely */ + if (j) + dynalloc_logp = IMAX(2, dynalloc_logp-1); + offsets[i] = boost; + } + + if (C==2) + { + static const opus_val16 intensity_thresholds[21]= + /* 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 off*/ + { 1, 2, 3, 4, 5, 6, 7, 8,16,24,36,44,50,56,62,67,72,79,88,106,134}; + static const opus_val16 intensity_histeresis[21]= + { 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 3, 3, 4, 5, 6, 8, 8}; + + /* Always use MS for 2.5 ms frames until we can do a better analysis */ + if (LM!=0) + dual_stereo = stereo_analysis(mode, X, LM, N); + + st->intensity = hysteresis_decision((opus_val16)(equiv_rate/1000), + intensity_thresholds, intensity_histeresis, 21, st->intensity); + st->intensity = IMIN(st->end,IMAX(st->start, st->intensity)); + } + + alloc_trim = 5; + if (tell+(6<<BITRES) <= total_bits - total_boost) + { + if (st->lfe) + alloc_trim = 5; + else + alloc_trim = alloc_trim_analysis(mode, X, bandLogE, + st->end, LM, C, N, &st->analysis, &st->stereo_saving, tf_estimate, st->intensity, surround_trim); + ec_enc_icdf(enc, alloc_trim, trim_icdf, 7); + tell = ec_tell_frac(enc); + } + + /* Variable bitrate */ + if (vbr_rate>0) + { + opus_val16 alpha; + opus_int32 delta; + /* The target rate in 8th bits per frame */ + opus_int32 target, base_target; + opus_int32 min_allowed; + int lm_diff = mode->maxLM - LM; + + /* Don't attempt to use more than 510 kb/s, even for frames smaller than 20 ms. + The CELT allocator will just not be able to use more than that anyway. */ + nbCompressedBytes = IMIN(nbCompressedBytes,1275>>(3-LM)); + base_target = vbr_rate - ((40*C+20)<<BITRES); + + if (st->constrained_vbr) + base_target += (st->vbr_offset>>lm_diff); + + target = compute_vbr(mode, &st->analysis, base_target, LM, equiv_rate, + st->lastCodedBands, C, st->intensity, st->constrained_vbr, + st->stereo_saving, tot_boost, tf_estimate, pitch_change, maxDepth, + st->variable_duration, st->lfe, st->energy_mask!=NULL, surround_masking, + temporal_vbr); + + /* The current offset is removed from the target and the space used + so far is added*/ + target=target+tell; + /* In VBR mode the frame size must not be reduced so much that it would + result in the encoder running out of bits. + The margin of 2 bytes ensures that none of the bust-prevention logic + in the decoder will have triggered so far. */ + min_allowed = ((tell+total_boost+(1<<(BITRES+3))-1)>>(BITRES+3)) + 2 - nbFilledBytes; + + nbAvailableBytes = (target+(1<<(BITRES+2)))>>(BITRES+3); + nbAvailableBytes = IMAX(min_allowed,nbAvailableBytes); + nbAvailableBytes = IMIN(nbCompressedBytes,nbAvailableBytes+nbFilledBytes) - nbFilledBytes; + + /* By how much did we "miss" the target on that frame */ + delta = target - vbr_rate; + + target=nbAvailableBytes<<(BITRES+3); + + /*If the frame is silent we don't adjust our drift, otherwise + the encoder will shoot to very high rates after hitting a + span of silence, but we do allow the bitres to refill. + This means that we'll undershoot our target in CVBR/VBR modes + on files with lots of silence. */ + if(silence) + { + nbAvailableBytes = 2; + target = 2*8<<BITRES; + delta = 0; + } + + if (st->vbr_count < 970) + { + st->vbr_count++; + alpha = celt_rcp(SHL32(EXTEND32(st->vbr_count+20),16)); + } else + alpha = QCONST16(.001f,15); + /* How many bits have we used in excess of what we're allowed */ + if (st->constrained_vbr) + st->vbr_reservoir += target - vbr_rate; + /*printf ("%d\n", st->vbr_reservoir);*/ + + /* Compute the offset we need to apply in order to reach the target */ + if (st->constrained_vbr) + { + st->vbr_drift += (opus_int32)MULT16_32_Q15(alpha,(delta*(1<<lm_diff))-st->vbr_offset-st->vbr_drift); + st->vbr_offset = -st->vbr_drift; + } + /*printf ("%d\n", st->vbr_drift);*/ + + if (st->constrained_vbr && st->vbr_reservoir < 0) + { + /* We're under the min value -- increase rate */ + int adjust = (-st->vbr_reservoir)/(8<<BITRES); + /* Unless we're just coding silence */ + nbAvailableBytes += silence?0:adjust; + st->vbr_reservoir = 0; + /*printf ("+%d\n", adjust);*/ + } + nbCompressedBytes = IMIN(nbCompressedBytes,nbAvailableBytes+nbFilledBytes); + /*printf("%d\n", nbCompressedBytes*50*8);*/ + /* This moves the raw bits to take into account the new compressed size */ + ec_enc_shrink(enc, nbCompressedBytes); + } + + /* Bit allocation */ + ALLOC(fine_quant, nbEBands, int); + ALLOC(pulses, nbEBands, int); + ALLOC(fine_priority, nbEBands, int); + + /* bits = packet size - where we are - safety*/ + bits = (((opus_int32)nbCompressedBytes*8)<<BITRES) - ec_tell_frac(enc) - 1; + anti_collapse_rsv = isTransient&&LM>=2&&bits>=((LM+2)<<BITRES) ? (1<<BITRES) : 0; + bits -= anti_collapse_rsv; + signalBandwidth = st->end-1; +#ifndef DISABLE_FLOAT_API + if (st->analysis.valid) + { + int min_bandwidth; + if (equiv_rate < (opus_int32)32000*C) + min_bandwidth = 13; + else if (equiv_rate < (opus_int32)48000*C) + min_bandwidth = 16; + else if (equiv_rate < (opus_int32)60000*C) + min_bandwidth = 18; + else if (equiv_rate < (opus_int32)80000*C) + min_bandwidth = 19; + else + min_bandwidth = 20; + signalBandwidth = IMAX(st->analysis.bandwidth, min_bandwidth); + } +#endif + if (st->lfe) + signalBandwidth = 1; + codedBands = compute_allocation(mode, st->start, st->end, offsets, cap, + alloc_trim, &st->intensity, &dual_stereo, bits, &balance, pulses, + fine_quant, fine_priority, C, LM, enc, 1, st->lastCodedBands, signalBandwidth); + if (st->lastCodedBands) + st->lastCodedBands = IMIN(st->lastCodedBands+1,IMAX(st->lastCodedBands-1,codedBands)); + else + st->lastCodedBands = codedBands; + + quant_fine_energy(mode, st->start, st->end, oldBandE, error, fine_quant, enc, C); + + /* Residual quantisation */ + ALLOC(collapse_masks, C*nbEBands, unsigned char); + quant_all_bands(1, mode, st->start, st->end, X, C==2 ? X+N : NULL, collapse_masks, + bandE, pulses, shortBlocks, st->spread_decision, dual_stereo, st->intensity, tf_res, + nbCompressedBytes*(8<<BITRES)-anti_collapse_rsv, balance, enc, LM, codedBands, &st->rng); + + if (anti_collapse_rsv > 0) + { + anti_collapse_on = st->consec_transient<2; +#ifdef FUZZING + anti_collapse_on = rand()&0x1; +#endif + ec_enc_bits(enc, anti_collapse_on, 1); + } + quant_energy_finalise(mode, st->start, st->end, oldBandE, error, fine_quant, fine_priority, nbCompressedBytes*8-ec_tell(enc), enc, C); + + if (silence) + { + for (i=0;i<C*nbEBands;i++) + oldBandE[i] = -QCONST16(28.f,DB_SHIFT); + } + +#ifdef RESYNTH + /* Re-synthesis of the coded audio if required */ + { + celt_sig *out_mem[2]; + + if (anti_collapse_on) + { + anti_collapse(mode, X, collapse_masks, LM, C, N, + st->start, st->end, oldBandE, oldLogE, oldLogE2, pulses, st->rng); + } + + if (silence) + { + for (i=0;i<C*N;i++) + freq[i] = 0; + } else { + /* Synthesis */ + denormalise_bands(mode, X, freq, oldBandE, st->start, effEnd, C, M); + } + + c=0; do { + OPUS_MOVE(st->syn_mem[c], st->syn_mem[c]+N, 2*MAX_PERIOD-N+overlap/2); + } while (++c<CC); + + if (CC==2&&C==1) + { + for (i=0;i<N;i++) + freq[N+i] = freq[i]; + } + + c=0; do { + out_mem[c] = st->syn_mem[c]+2*MAX_PERIOD-N; + } while (++c<CC); + + compute_inv_mdcts(mode, shortBlocks, freq, out_mem, CC, LM); + + c=0; do { + st->prefilter_period=IMAX(st->prefilter_period, COMBFILTER_MINPERIOD); + st->prefilter_period_old=IMAX(st->prefilter_period_old, COMBFILTER_MINPERIOD); + comb_filter(out_mem[c], out_mem[c], st->prefilter_period_old, st->prefilter_period, mode->shortMdctSize, + st->prefilter_gain_old, st->prefilter_gain, st->prefilter_tapset_old, st->prefilter_tapset, + mode->window, st->overlap); + if (LM!=0) + comb_filter(out_mem[c]+mode->shortMdctSize, out_mem[c]+mode->shortMdctSize, st->prefilter_period, pitch_index, N-mode->shortMdctSize, + st->prefilter_gain, gain1, st->prefilter_tapset, prefilter_tapset, + mode->window, overlap); + } while (++c<CC); + + /* We reuse freq[] as scratch space for the de-emphasis */ + deemphasis(out_mem, (opus_val16*)pcm, N, CC, st->upsample, mode->preemph, st->preemph_memD, freq); + st->prefilter_period_old = st->prefilter_period; + st->prefilter_gain_old = st->prefilter_gain; + st->prefilter_tapset_old = st->prefilter_tapset; + } +#endif + + st->prefilter_period = pitch_index; + st->prefilter_gain = gain1; + st->prefilter_tapset = prefilter_tapset; +#ifdef RESYNTH + if (LM!=0) + { + st->prefilter_period_old = st->prefilter_period; + st->prefilter_gain_old = st->prefilter_gain; + st->prefilter_tapset_old = st->prefilter_tapset; + } +#endif + + if (CC==2&&C==1) { + for (i=0;i<nbEBands;i++) + oldBandE[nbEBands+i]=oldBandE[i]; + } + + if (!isTransient) + { + for (i=0;i<CC*nbEBands;i++) + oldLogE2[i] = oldLogE[i]; + for (i=0;i<CC*nbEBands;i++) + oldLogE[i] = oldBandE[i]; + } else { + for (i=0;i<CC*nbEBands;i++) + oldLogE[i] = MIN16(oldLogE[i], oldBandE[i]); + } + /* In case start or end were to change */ + c=0; do + { + for (i=0;i<st->start;i++) + { + oldBandE[c*nbEBands+i]=0; + oldLogE[c*nbEBands+i]=oldLogE2[c*nbEBands+i]=-QCONST16(28.f,DB_SHIFT); + } + for (i=st->end;i<nbEBands;i++) + { + oldBandE[c*nbEBands+i]=0; + oldLogE[c*nbEBands+i]=oldLogE2[c*nbEBands+i]=-QCONST16(28.f,DB_SHIFT); + } + } while (++c<CC); + + if (isTransient || transient_got_disabled) + st->consec_transient++; + else + st->consec_transient=0; + st->rng = enc->rng; + + /* If there's any room left (can only happen for very high rates), + it's already filled with zeros */ + ec_enc_done(enc); + +#ifdef CUSTOM_MODES + if (st->signalling) + nbCompressedBytes++; +#endif + + RESTORE_STACK; + if (ec_get_error(enc)) + return OPUS_INTERNAL_ERROR; + else + return nbCompressedBytes; +} + + +#ifdef CUSTOM_MODES + +#ifdef OPUS_FIXED_POINT +int opus_custom_encode(CELTEncoder * OPUS_RESTRICT st, const opus_int16 * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes) +{ + return celt_encode_with_ec(st, pcm, frame_size, compressed, nbCompressedBytes, NULL); +} + +#ifndef DISABLE_FLOAT_API +int opus_custom_encode_float(CELTEncoder * OPUS_RESTRICT st, const float * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes) +{ + int j, ret, C, N; + VARDECL(opus_int16, in); + ALLOC_STACK; + + if (pcm==NULL) + return OPUS_BAD_ARG; + + C = st->channels; + N = frame_size; + ALLOC(in, C*N, opus_int16); + + for (j=0;j<C*N;j++) + in[j] = FLOAT2INT16(pcm[j]); + + ret=celt_encode_with_ec(st,in,frame_size,compressed,nbCompressedBytes, NULL); +#ifdef RESYNTH + for (j=0;j<C*N;j++) + ((float*)pcm)[j]=in[j]*(1.f/32768.f); +#endif + RESTORE_STACK; + return ret; +} +#endif /* DISABLE_FLOAT_API */ +#else + +int opus_custom_encode(CELTEncoder * OPUS_RESTRICT st, const opus_int16 * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes) +{ + int j, ret, C, N; + VARDECL(celt_sig, in); + ALLOC_STACK; + + if (pcm==NULL) + return OPUS_BAD_ARG; + + C=st->channels; + N=frame_size; + ALLOC(in, C*N, celt_sig); + for (j=0;j<C*N;j++) { + in[j] = SCALEOUT(pcm[j]); + } + + ret = celt_encode_with_ec(st,in,frame_size,compressed,nbCompressedBytes, NULL); +#ifdef RESYNTH + for (j=0;j<C*N;j++) + ((opus_int16*)pcm)[j] = FLOAT2INT16(in[j]); +#endif + RESTORE_STACK; + return ret; +} + +int opus_custom_encode_float(CELTEncoder * OPUS_RESTRICT st, const float * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes) +{ + return celt_encode_with_ec(st, pcm, frame_size, compressed, nbCompressedBytes, NULL); +} + +#endif + +#endif /* CUSTOM_MODES */ + +int opus_custom_encoder_ctl(CELTEncoder * OPUS_RESTRICT st, int request, ...) +{ + va_list ap; + + va_start(ap, request); + switch (request) + { + case OPUS_SET_COMPLEXITY_REQUEST: + { + int value = va_arg(ap, opus_int32); + if (value<0 || value>10) + goto bad_arg; + st->complexity = value; + } + break; + case CELT_SET_START_BAND_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<0 || value>=st->mode->nbEBands) + goto bad_arg; + st->start = value; + } + break; + case CELT_SET_END_BAND_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<1 || value>st->mode->nbEBands) + goto bad_arg; + st->end = value; + } + break; + case CELT_SET_PREDICTION_REQUEST: + { + int value = va_arg(ap, opus_int32); + if (value<0 || value>2) + goto bad_arg; + st->disable_pf = value<=1; + st->force_intra = value==0; + } + break; + case OPUS_SET_PACKET_LOSS_PERC_REQUEST: + { + int value = va_arg(ap, opus_int32); + if (value<0 || value>100) + goto bad_arg; + st->loss_rate = value; + } + break; + case OPUS_SET_VBR_CONSTRAINT_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->constrained_vbr = value; + } + break; + case OPUS_SET_VBR_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->vbr = value; + } + break; + case OPUS_SET_BITRATE_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<=500 && value!=OPUS_BITRATE_MAX) + goto bad_arg; + value = IMIN(value, 260000*st->channels); + st->bitrate = value; + } + break; + case CELT_SET_CHANNELS_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<1 || value>2) + goto bad_arg; + st->stream_channels = value; + } + break; + case OPUS_SET_LSB_DEPTH_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<8 || value>24) + goto bad_arg; + st->lsb_depth=value; + } + break; + case OPUS_GET_LSB_DEPTH_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + *value=st->lsb_depth; + } + break; + case OPUS_SET_EXPERT_FRAME_DURATION_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->variable_duration = value; + } + break; + case OPUS_RESET_STATE: + { + int i; + opus_val16 *oldBandE, *oldLogE, *oldLogE2; + oldBandE = (opus_val16*)(st->in_mem+st->channels*(st->overlap+COMBFILTER_MAXPERIOD)); + oldLogE = oldBandE + st->channels*st->mode->nbEBands; + oldLogE2 = oldLogE + st->channels*st->mode->nbEBands; + OPUS_CLEAR((char*)&st->ENCODER_RESET_START, + opus_custom_encoder_get_size(st->mode, st->channels)- + ((char*)&st->ENCODER_RESET_START - (char*)st)); + for (i=0;i<st->channels*st->mode->nbEBands;i++) + oldLogE[i]=oldLogE2[i]=-QCONST16(28.f,DB_SHIFT); + st->vbr_offset = 0; + st->delayedIntra = 1; + st->spread_decision = SPREAD_NORMAL; + st->tonal_average = 256; + st->hf_average = 0; + st->tapset_decision = 0; + } + break; +#ifdef CUSTOM_MODES + case CELT_SET_INPUT_CLIPPING_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->clip = value; + } + break; +#endif + case CELT_SET_SIGNALLING_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->signalling = value; + } + break; + case CELT_SET_ANALYSIS_REQUEST: + { + AnalysisInfo *info = va_arg(ap, AnalysisInfo *); + if (info) + OPUS_COPY(&st->analysis, info, 1); + } + break; + case CELT_GET_MODE_REQUEST: + { + const CELTMode ** value = va_arg(ap, const CELTMode**); + if (value==0) + goto bad_arg; + *value=st->mode; + } + break; + case OPUS_GET_FINAL_RANGE_REQUEST: + { + opus_uint32 * value = va_arg(ap, opus_uint32 *); + if (value==0) + goto bad_arg; + *value=st->rng; + } + break; + case OPUS_SET_LFE_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->lfe = value; + } + break; + case OPUS_SET_ENERGY_MASK_REQUEST: + { + opus_val16 *value = va_arg(ap, opus_val16*); + st->energy_mask = value; + } + break; + default: + goto bad_request; + } + va_end(ap); + return OPUS_OK; +bad_arg: + va_end(ap); + return OPUS_BAD_ARG; +bad_request: + va_end(ap); + return OPUS_UNIMPLEMENTED; +} diff --git a/drivers/opus/celt/celt_lpc.c b/drivers/opus/celt/celt_lpc.c new file mode 100644 index 0000000000..1fa4406bc9 --- /dev/null +++ b/drivers/opus/celt/celt_lpc.c @@ -0,0 +1,309 @@ +/* Copyright (c) 2009-2010 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef OPUS_HAVE_CONFIG_H +#include "opus_config.h" +#endif + +#include "celt_lpc.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "pitch.h" + +void _celt_lpc( + opus_val16 *_lpc, /* out: [0...p-1] LPC coefficients */ +const opus_val32 *ac, /* in: [0...p] autocorrelation values */ +int p +) +{ + int i, j; + opus_val32 r; + opus_val32 error = ac[0]; +#ifdef OPUS_FIXED_POINT + opus_val32 lpc[LPC_ORDER]; +#else + float *lpc = _lpc; +#endif + + for (i = 0; i < p; i++) + lpc[i] = 0; + if (ac[0] != 0) + { + for (i = 0; i < p; i++) { + /* Sum up this iteration's reflection coefficient */ + opus_val32 rr = 0; + for (j = 0; j < i; j++) + rr += MULT32_32_Q31(lpc[j],ac[i - j]); + rr += SHR32(ac[i + 1],3); + r = -frac_div32(SHL32(rr,3), error); + /* Update LPC coefficients and total error */ + lpc[i] = SHR32(r,3); + for (j = 0; j < (i+1)>>1; j++) + { + opus_val32 tmp1, tmp2; + tmp1 = lpc[j]; + tmp2 = lpc[i-1-j]; + lpc[j] = tmp1 + MULT32_32_Q31(r,tmp2); + lpc[i-1-j] = tmp2 + MULT32_32_Q31(r,tmp1); + } + + error = error - MULT32_32_Q31(MULT32_32_Q31(r,r),error); + /* Bail out once we get 30 dB gain */ +#ifdef OPUS_FIXED_POINT + if (error<SHR32(ac[0],10)) + break; +#else + if (error<.001f*ac[0]) + break; +#endif + } + } +#ifdef OPUS_FIXED_POINT + for (i=0;i<p;i++) + _lpc[i] = ROUND16(lpc[i],16); +#endif +} + +void celt_fir(const opus_val16 *_x, + const opus_val16 *num, + opus_val16 *_y, + int N, + int ord, + opus_val16 *mem) +{ + int i,j; + VARDECL(opus_val16, rnum); + VARDECL(opus_val16, x); + SAVE_STACK; + + ALLOC(rnum, ord, opus_val16); + ALLOC(x, N+ord, opus_val16); + for(i=0;i<ord;i++) + rnum[i] = num[ord-i-1]; + for(i=0;i<ord;i++) + x[i] = mem[ord-i-1]; + for (i=0;i<N;i++) + x[i+ord]=_x[i]; + for(i=0;i<ord;i++) + mem[i] = _x[N-i-1]; +#ifdef SMALL_FOOTPRINT + for (i=0;i<N;i++) + { + opus_val32 sum = SHL32(EXTEND32(_x[i]), SIG_SHIFT); + for (j=0;j<ord;j++) + { + sum = MAC16_16(sum,rnum[j],x[i+j]); + } + _y[i] = SATURATE16(PSHR32(sum, SIG_SHIFT)); + } +#else + for (i=0;i<N-3;i+=4) + { + opus_val32 sum[4]={0,0,0,0}; + xcorr_kernel(rnum, x+i, sum, ord); + _y[i ] = SATURATE16(ADD32(EXTEND32(_x[i ]), PSHR32(sum[0], SIG_SHIFT))); + _y[i+1] = SATURATE16(ADD32(EXTEND32(_x[i+1]), PSHR32(sum[1], SIG_SHIFT))); + _y[i+2] = SATURATE16(ADD32(EXTEND32(_x[i+2]), PSHR32(sum[2], SIG_SHIFT))); + _y[i+3] = SATURATE16(ADD32(EXTEND32(_x[i+3]), PSHR32(sum[3], SIG_SHIFT))); + } + for (;i<N;i++) + { + opus_val32 sum = 0; + for (j=0;j<ord;j++) + sum = MAC16_16(sum,rnum[j],x[i+j]); + _y[i] = SATURATE16(ADD32(EXTEND32(_x[i]), PSHR32(sum, SIG_SHIFT))); + } +#endif + RESTORE_STACK; +} + +void celt_iir(const opus_val32 *_x, + const opus_val16 *den, + opus_val32 *_y, + int N, + int ord, + opus_val16 *mem) +{ +#ifdef SMALL_FOOTPRINT + int i,j; + for (i=0;i<N;i++) + { + opus_val32 sum = _x[i]; + for (j=0;j<ord;j++) + { + sum -= MULT16_16(den[j],mem[j]); + } + for (j=ord-1;j>=1;j--) + { + mem[j]=mem[j-1]; + } + mem[0] = ROUND16(sum,SIG_SHIFT); + _y[i] = sum; + } +#else + int i,j; + VARDECL(opus_val16, rden); + VARDECL(opus_val16, y); + SAVE_STACK; + + celt_assert((ord&3)==0); + ALLOC(rden, ord, opus_val16); + ALLOC(y, N+ord, opus_val16); + for(i=0;i<ord;i++) + rden[i] = den[ord-i-1]; + for(i=0;i<ord;i++) + y[i] = -mem[ord-i-1]; + for(;i<N+ord;i++) + y[i]=0; + for (i=0;i<N-3;i+=4) + { + /* Unroll by 4 as if it were an FIR filter */ + opus_val32 sum[4]; + sum[0]=_x[i]; + sum[1]=_x[i+1]; + sum[2]=_x[i+2]; + sum[3]=_x[i+3]; + xcorr_kernel(rden, y+i, sum, ord); + + /* Patch up the result to compensate for the fact that this is an IIR */ + y[i+ord ] = -ROUND16(sum[0],SIG_SHIFT); + _y[i ] = sum[0]; + sum[1] = MAC16_16(sum[1], y[i+ord ], den[0]); + y[i+ord+1] = -ROUND16(sum[1],SIG_SHIFT); + _y[i+1] = sum[1]; + sum[2] = MAC16_16(sum[2], y[i+ord+1], den[0]); + sum[2] = MAC16_16(sum[2], y[i+ord ], den[1]); + y[i+ord+2] = -ROUND16(sum[2],SIG_SHIFT); + _y[i+2] = sum[2]; + + sum[3] = MAC16_16(sum[3], y[i+ord+2], den[0]); + sum[3] = MAC16_16(sum[3], y[i+ord+1], den[1]); + sum[3] = MAC16_16(sum[3], y[i+ord ], den[2]); + y[i+ord+3] = -ROUND16(sum[3],SIG_SHIFT); + _y[i+3] = sum[3]; + } + for (;i<N;i++) + { + opus_val32 sum = _x[i]; + for (j=0;j<ord;j++) + sum -= MULT16_16(rden[j],y[i+j]); + y[i+ord] = ROUND16(sum,SIG_SHIFT); + _y[i] = sum; + } + for(i=0;i<ord;i++) + mem[i] = _y[N-i-1]; + RESTORE_STACK; +#endif +} + +int _celt_autocorr( + const opus_val16 *x, /* in: [0...n-1] samples x */ + opus_val32 *ac, /* out: [0...lag-1] ac values */ + const opus_val16 *window, + int overlap, + int lag, + int n, + int arch + ) +{ + opus_val32 d; + int i, k; + int fastN=n-lag; + int shift; + const opus_val16 *xptr; + VARDECL(opus_val16, xx); + SAVE_STACK; + ALLOC(xx, n, opus_val16); + celt_assert(n>0); + celt_assert(overlap>=0); + if (overlap == 0) + { + xptr = x; + } else { + for (i=0;i<n;i++) + xx[i] = x[i]; + for (i=0;i<overlap;i++) + { + xx[i] = MULT16_16_Q15(x[i],window[i]); + xx[n-i-1] = MULT16_16_Q15(x[n-i-1],window[i]); + } + xptr = xx; + } + shift=0; +#ifdef OPUS_FIXED_POINT + { + opus_val32 ac0; + ac0 = 1+(n<<7); + if (n&1) ac0 += SHR32(MULT16_16(xptr[0],xptr[0]),9); + for(i=(n&1);i<n;i+=2) + { + ac0 += SHR32(MULT16_16(xptr[i],xptr[i]),9); + ac0 += SHR32(MULT16_16(xptr[i+1],xptr[i+1]),9); + } + + shift = celt_ilog2(ac0)-30+10; + shift = (shift)/2; + if (shift>0) + { + for(i=0;i<n;i++) + xx[i] = PSHR32(xptr[i], shift); + xptr = xx; + } else + shift = 0; + } +#endif + 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++) + d = MAC16_16(d, xptr[i], xptr[i-k]); + ac[k] += d; + } +#ifdef OPUS_FIXED_POINT + shift = 2*shift; + if (shift<=0) + ac[0] += SHL32((opus_int32)1, -shift); + if (ac[0] < 268435456) + { + int shift2 = 29 - EC_ILOG(ac[0]); + for (i=0;i<=lag;i++) + ac[i] = SHL32(ac[i], shift2); + shift -= shift2; + } else if (ac[0] >= 536870912) + { + int shift2=1; + if (ac[0] >= 1073741824) + shift2++; + for (i=0;i<=lag;i++) + ac[i] = SHR32(ac[i], shift2); + shift += shift2; + } +#endif + + RESTORE_STACK; + return shift; +} diff --git a/drivers/opus/celt/celt_lpc.h b/drivers/opus/celt/celt_lpc.h new file mode 100644 index 0000000000..dc2a0a3d26 --- /dev/null +++ b/drivers/opus/celt/celt_lpc.h @@ -0,0 +1,54 @@ +/* Copyright (c) 2009-2010 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef PLC_H +#define PLC_H + +#include "arch.h" + +#define LPC_ORDER 24 + +void _celt_lpc(opus_val16 *_lpc, const opus_val32 *ac, int p); + +void celt_fir(const opus_val16 *x, + const opus_val16 *num, + opus_val16 *y, + int N, + int ord, + opus_val16 *mem); + +void celt_iir(const opus_val32 *x, + const opus_val16 *den, + opus_val32 *y, + int N, + 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 arch); + +#endif /* PLC_H */ diff --git a/drivers/opus/celt/cpu_support.h b/drivers/opus/celt/cpu_support.h new file mode 100644 index 0000000000..d68dbe62c5 --- /dev/null +++ b/drivers/opus/celt/cpu_support.h @@ -0,0 +1,54 @@ +/* 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. +*/ + +#ifndef CPU_SUPPORT_H +#define CPU_SUPPORT_H + +#include "opus_types.h" +#include "opus_defines.h" + +#if defined(OPUS_HAVE_RTCD) && defined(OPUS_ARM_ASM) +#include "arm/armcpu.h" + +/* We currently support 4 ARM variants: + * arch[0] -> ARMv4 + * arch[1] -> ARMv5E + * arch[2] -> ARMv6 + * arch[3] -> NEON + */ +#define OPUS_ARCHMASK 3 + +#else +#define OPUS_ARCHMASK 0 + +static OPUS_INLINE int opus_select_arch(void) +{ + return 0; +} +#endif + +#endif diff --git a/drivers/opus/celt/cwrs.c b/drivers/opus/celt/cwrs.c new file mode 100644 index 0000000000..b866aa9210 --- /dev/null +++ b/drivers/opus/celt/cwrs.c @@ -0,0 +1,697 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Copyright (c) 2007-2009 Timothy B. Terriberry + Written by Timothy B. Terriberry and Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef OPUS_HAVE_CONFIG_H +#include "opus_config.h" +#endif + +#include "os_support.h" +#include "cwrs.h" +#include "mathops.h" +#include "arch.h" + +#ifdef CUSTOM_MODES + +/*Guaranteed to return a conservatively large estimate of the binary logarithm + with frac bits of fractional precision. + Tested for all possible 32-bit inputs with frac=4, where the maximum + overestimation is 0.06254243 bits.*/ +int log2_frac(opus_uint32 val, int frac) +{ + int l; + l=EC_ILOG(val); + if(val&(val-1)){ + /*This is (val>>l-16), but guaranteed to round up, even if adding a bias + before the shift would cause overflow (e.g., for 0xFFFFxxxx). + Doesn't work for val=0, but that case fails the test above.*/ + if(l>16)val=((val-1)>>(l-16))+1; + else val<<=16-l; + l=(l-1)<<frac; + /*Note that we always need one iteration, since the rounding up above means + that we might need to adjust the integer part of the logarithm.*/ + do{ + int b; + b=(int)(val>>16); + l+=b<<frac; + val=(val+b)>>b; + val=(val*val+0x7FFF)>>15; + } + while(frac-->0); + /*If val is not exactly 0x8000, then we have to round up the remainder.*/ + return l+(val>0x8000); + } + /*Exact powers of two require no rounding.*/ + else return (l-1)<<frac; +} +#endif + +/*Although derived separately, the pulse vector coding scheme is equivalent to + a Pyramid Vector Quantizer \cite{Fis86}. + Some additional notes about an early version appear at + http://people.xiph.org/~tterribe/notes/cwrs.html, but the codebook ordering + and the definitions of some terms have evolved since that was written. + + The conversion from a pulse vector to an integer index (encoding) and back + (decoding) is governed by two related functions, V(N,K) and U(N,K). + + V(N,K) = the number of combinations, with replacement, of N items, taken K + at a time, when a sign bit is added to each item taken at least once (i.e., + the number of N-dimensional unit pulse vectors with K pulses). + One way to compute this is via + V(N,K) = K>0 ? sum(k=1...K,2**k*choose(N,k)*choose(K-1,k-1)) : 1, + where choose() is the binomial function. + A table of values for N<10 and K<10 looks like: + V[10][10] = { + {1, 0, 0, 0, 0, 0, 0, 0, 0, 0}, + {1, 2, 2, 2, 2, 2, 2, 2, 2, 2}, + {1, 4, 8, 12, 16, 20, 24, 28, 32, 36}, + {1, 6, 18, 38, 66, 102, 146, 198, 258, 326}, + {1, 8, 32, 88, 192, 360, 608, 952, 1408, 1992}, + {1, 10, 50, 170, 450, 1002, 1970, 3530, 5890, 9290}, + {1, 12, 72, 292, 912, 2364, 5336, 10836, 20256, 35436}, + {1, 14, 98, 462, 1666, 4942, 12642, 28814, 59906, 115598}, + {1, 16, 128, 688, 2816, 9424, 27008, 68464, 157184, 332688}, + {1, 18, 162, 978, 4482, 16722, 53154, 148626, 374274, 864146} + }; + + U(N,K) = the number of such combinations wherein N-1 objects are taken at + most K-1 at a time. + This is given by + U(N,K) = sum(k=0...K-1,V(N-1,k)) + = K>0 ? (V(N-1,K-1) + V(N,K-1))/2 : 0. + The latter expression also makes clear that U(N,K) is half the number of such + combinations wherein the first object is taken at least once. + Although it may not be clear from either of these definitions, U(N,K) is the + natural function to work with when enumerating the pulse vector codebooks, + not V(N,K). + U(N,K) is not well-defined for N=0, but with the extension + U(0,K) = K>0 ? 0 : 1, + the function becomes symmetric: U(N,K) = U(K,N), with a similar table: + U[10][10] = { + {1, 0, 0, 0, 0, 0, 0, 0, 0, 0}, + {0, 1, 1, 1, 1, 1, 1, 1, 1, 1}, + {0, 1, 3, 5, 7, 9, 11, 13, 15, 17}, + {0, 1, 5, 13, 25, 41, 61, 85, 113, 145}, + {0, 1, 7, 25, 63, 129, 231, 377, 575, 833}, + {0, 1, 9, 41, 129, 321, 681, 1289, 2241, 3649}, + {0, 1, 11, 61, 231, 681, 1683, 3653, 7183, 13073}, + {0, 1, 13, 85, 377, 1289, 3653, 8989, 19825, 40081}, + {0, 1, 15, 113, 575, 2241, 7183, 19825, 48639, 108545}, + {0, 1, 17, 145, 833, 3649, 13073, 40081, 108545, 265729} + }; + + With this extension, V(N,K) may be written in terms of U(N,K): + V(N,K) = U(N,K) + U(N,K+1) + for all N>=0, K>=0. + Thus U(N,K+1) represents the number of combinations where the first element + is positive or zero, and U(N,K) represents the number of combinations where + it is negative. + With a large enough table of U(N,K) values, we could write O(N) encoding + and O(min(N*log(K),N+K)) decoding routines, but such a table would be + prohibitively large for small embedded devices (K may be as large as 32767 + for small N, and N may be as large as 200). + + Both functions obey the same recurrence relation: + V(N,K) = V(N-1,K) + V(N,K-1) + V(N-1,K-1), + U(N,K) = U(N-1,K) + U(N,K-1) + U(N-1,K-1), + for all N>0, K>0, with different initial conditions at N=0 or K=0. + This allows us to construct a row of one of the tables above given the + previous row or the next row. + Thus we can derive O(NK) encoding and decoding routines with O(K) memory + using only addition and subtraction. + + When encoding, we build up from the U(2,K) row and work our way forwards. + When decoding, we need to start at the U(N,K) row and work our way backwards, + which requires a means of computing U(N,K). + U(N,K) may be computed from two previous values with the same N: + U(N,K) = ((2*N-1)*U(N,K-1) - U(N,K-2))/(K-1) + U(N,K-2) + for all N>1, and since U(N,K) is symmetric, a similar relation holds for two + previous values with the same K: + U(N,K>1) = ((2*K-1)*U(N-1,K) - U(N-2,K))/(N-1) + U(N-2,K) + for all K>1. + This allows us to construct an arbitrary row of the U(N,K) table by starting + with the first two values, which are constants. + This saves roughly 2/3 the work in our O(NK) decoding routine, but costs O(K) + multiplications. + Similar relations can be derived for V(N,K), but are not used here. + + For N>0 and K>0, U(N,K) and V(N,K) take on the form of an (N-1)-degree + polynomial for fixed N. + The first few are + U(1,K) = 1, + U(2,K) = 2*K-1, + U(3,K) = (2*K-2)*K+1, + U(4,K) = (((4*K-6)*K+8)*K-3)/3, + U(5,K) = ((((2*K-4)*K+10)*K-8)*K+3)/3, + and + V(1,K) = 2, + V(2,K) = 4*K, + V(3,K) = 4*K*K+2, + V(4,K) = 8*(K*K+2)*K/3, + V(5,K) = ((4*K*K+20)*K*K+6)/3, + for all K>0. + This allows us to derive O(N) encoding and O(N*log(K)) decoding routines for + small N (and indeed decoding is also O(N) for N<3). + + @ARTICLE{Fis86, + author="Thomas R. Fischer", + title="A Pyramid Vector Quantizer", + journal="IEEE Transactions on Information Theory", + volume="IT-32", + number=4, + pages="568--583", + month=Jul, + year=1986 + }*/ + +#if !defined(SMALL_FOOTPRINT) + +/*U(N,K) = U(K,N) := N>0?K>0?U(N-1,K)+U(N,K-1)+U(N-1,K-1):0:K>0?1:0*/ +# define CELT_PVQ_U(_n,_k) (CELT_PVQ_U_ROW[IMIN(_n,_k)][IMAX(_n,_k)]) +/*V(N,K) := U(N,K)+U(N,K+1) = the number of PVQ codewords for a band of size N + with K pulses allocated to it.*/ +# define CELT_PVQ_V(_n,_k) (CELT_PVQ_U(_n,_k)+CELT_PVQ_U(_n,(_k)+1)) + +/*For each V(N,K) supported, we will access element U(min(N,K+1),max(N,K+1)). + Thus, the number of entries in row I is the larger of the maximum number of + pulses we will ever allocate for a given N=I (K=128, or however many fit in + 32 bits, whichever is smaller), plus one, and the maximum N for which + K=I-1 pulses fit in 32 bits. + The largest band size in an Opus Custom mode is 208. + Otherwise, we can limit things to the set of N which can be achieved by + splitting a band from a standard Opus mode: 176, 144, 96, 88, 72, 64, 48, + 44, 36, 32, 24, 22, 18, 16, 8, 4, 2).*/ +#if defined(CUSTOM_MODES) +static const opus_uint32 CELT_PVQ_U_DATA[1488]={ +#else +static const opus_uint32 CELT_PVQ_U_DATA[1272]={ +#endif + /*N=0, K=0...176:*/ + 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +#if defined(CUSTOM_MODES) + /*...208:*/ + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, +#endif + /*N=1, K=1...176:*/ + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, +#if defined(CUSTOM_MODES) + /*...208:*/ + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, +#endif + /*N=2, K=2...176:*/ + 3, 5, 7, 9, 11, 13, 15, 17, 19, 21, 23, 25, 27, 29, 31, 33, 35, 37, 39, 41, + 43, 45, 47, 49, 51, 53, 55, 57, 59, 61, 63, 65, 67, 69, 71, 73, 75, 77, 79, + 81, 83, 85, 87, 89, 91, 93, 95, 97, 99, 101, 103, 105, 107, 109, 111, 113, + 115, 117, 119, 121, 123, 125, 127, 129, 131, 133, 135, 137, 139, 141, 143, + 145, 147, 149, 151, 153, 155, 157, 159, 161, 163, 165, 167, 169, 171, 173, + 175, 177, 179, 181, 183, 185, 187, 189, 191, 193, 195, 197, 199, 201, 203, + 205, 207, 209, 211, 213, 215, 217, 219, 221, 223, 225, 227, 229, 231, 233, + 235, 237, 239, 241, 243, 245, 247, 249, 251, 253, 255, 257, 259, 261, 263, + 265, 267, 269, 271, 273, 275, 277, 279, 281, 283, 285, 287, 289, 291, 293, + 295, 297, 299, 301, 303, 305, 307, 309, 311, 313, 315, 317, 319, 321, 323, + 325, 327, 329, 331, 333, 335, 337, 339, 341, 343, 345, 347, 349, 351, +#if defined(CUSTOM_MODES) + /*...208:*/ + 353, 355, 357, 359, 361, 363, 365, 367, 369, 371, 373, 375, 377, 379, 381, + 383, 385, 387, 389, 391, 393, 395, 397, 399, 401, 403, 405, 407, 409, 411, + 413, 415, +#endif + /*N=3, K=3...176:*/ + 13, 25, 41, 61, 85, 113, 145, 181, 221, 265, 313, 365, 421, 481, 545, 613, + 685, 761, 841, 925, 1013, 1105, 1201, 1301, 1405, 1513, 1625, 1741, 1861, + 1985, 2113, 2245, 2381, 2521, 2665, 2813, 2965, 3121, 3281, 3445, 3613, 3785, + 3961, 4141, 4325, 4513, 4705, 4901, 5101, 5305, 5513, 5725, 5941, 6161, 6385, + 6613, 6845, 7081, 7321, 7565, 7813, 8065, 8321, 8581, 8845, 9113, 9385, 9661, + 9941, 10225, 10513, 10805, 11101, 11401, 11705, 12013, 12325, 12641, 12961, + 13285, 13613, 13945, 14281, 14621, 14965, 15313, 15665, 16021, 16381, 16745, + 17113, 17485, 17861, 18241, 18625, 19013, 19405, 19801, 20201, 20605, 21013, + 21425, 21841, 22261, 22685, 23113, 23545, 23981, 24421, 24865, 25313, 25765, + 26221, 26681, 27145, 27613, 28085, 28561, 29041, 29525, 30013, 30505, 31001, + 31501, 32005, 32513, 33025, 33541, 34061, 34585, 35113, 35645, 36181, 36721, + 37265, 37813, 38365, 38921, 39481, 40045, 40613, 41185, 41761, 42341, 42925, + 43513, 44105, 44701, 45301, 45905, 46513, 47125, 47741, 48361, 48985, 49613, + 50245, 50881, 51521, 52165, 52813, 53465, 54121, 54781, 55445, 56113, 56785, + 57461, 58141, 58825, 59513, 60205, 60901, 61601, +#if defined(CUSTOM_MODES) + /*...208:*/ + 62305, 63013, 63725, 64441, 65161, 65885, 66613, 67345, 68081, 68821, 69565, + 70313, 71065, 71821, 72581, 73345, 74113, 74885, 75661, 76441, 77225, 78013, + 78805, 79601, 80401, 81205, 82013, 82825, 83641, 84461, 85285, 86113, +#endif + /*N=4, K=4...176:*/ + 63, 129, 231, 377, 575, 833, 1159, 1561, 2047, 2625, 3303, 4089, 4991, 6017, + 7175, 8473, 9919, 11521, 13287, 15225, 17343, 19649, 22151, 24857, 27775, + 30913, 34279, 37881, 41727, 45825, 50183, 54809, 59711, 64897, 70375, 76153, + 82239, 88641, 95367, 102425, 109823, 117569, 125671, 134137, 142975, 152193, + 161799, 171801, 182207, 193025, 204263, 215929, 228031, 240577, 253575, + 267033, 280959, 295361, 310247, 325625, 341503, 357889, 374791, 392217, + 410175, 428673, 447719, 467321, 487487, 508225, 529543, 551449, 573951, + 597057, 620775, 645113, 670079, 695681, 721927, 748825, 776383, 804609, + 833511, 863097, 893375, 924353, 956039, 988441, 1021567, 1055425, 1090023, + 1125369, 1161471, 1198337, 1235975, 1274393, 1313599, 1353601, 1394407, + 1436025, 1478463, 1521729, 1565831, 1610777, 1656575, 1703233, 1750759, + 1799161, 1848447, 1898625, 1949703, 2001689, 2054591, 2108417, 2163175, + 2218873, 2275519, 2333121, 2391687, 2451225, 2511743, 2573249, 2635751, + 2699257, 2763775, 2829313, 2895879, 2963481, 3032127, 3101825, 3172583, + 3244409, 3317311, 3391297, 3466375, 3542553, 3619839, 3698241, 3777767, + 3858425, 3940223, 4023169, 4107271, 4192537, 4278975, 4366593, 4455399, + 4545401, 4636607, 4729025, 4822663, 4917529, 5013631, 5110977, 5209575, + 5309433, 5410559, 5512961, 5616647, 5721625, 5827903, 5935489, 6044391, + 6154617, 6266175, 6379073, 6493319, 6608921, 6725887, 6844225, 6963943, + 7085049, 7207551, +#if defined(CUSTOM_MODES) + /*...208:*/ + 7331457, 7456775, 7583513, 7711679, 7841281, 7972327, 8104825, 8238783, + 8374209, 8511111, 8649497, 8789375, 8930753, 9073639, 9218041, 9363967, + 9511425, 9660423, 9810969, 9963071, 10116737, 10271975, 10428793, 10587199, + 10747201, 10908807, 11072025, 11236863, 11403329, 11571431, 11741177, + 11912575, +#endif + /*N=5, K=5...176:*/ + 321, 681, 1289, 2241, 3649, 5641, 8361, 11969, 16641, 22569, 29961, 39041, + 50049, 63241, 78889, 97281, 118721, 143529, 172041, 204609, 241601, 283401, + 330409, 383041, 441729, 506921, 579081, 658689, 746241, 842249, 947241, + 1061761, 1186369, 1321641, 1468169, 1626561, 1797441, 1981449, 2179241, + 2391489, 2618881, 2862121, 3121929, 3399041, 3694209, 4008201, 4341801, + 4695809, 5071041, 5468329, 5888521, 6332481, 6801089, 7295241, 7815849, + 8363841, 8940161, 9545769, 10181641, 10848769, 11548161, 12280841, 13047849, + 13850241, 14689089, 15565481, 16480521, 17435329, 18431041, 19468809, + 20549801, 21675201, 22846209, 24064041, 25329929, 26645121, 28010881, + 29428489, 30899241, 32424449, 34005441, 35643561, 37340169, 39096641, + 40914369, 42794761, 44739241, 46749249, 48826241, 50971689, 53187081, + 55473921, 57833729, 60268041, 62778409, 65366401, 68033601, 70781609, + 73612041, 76526529, 79526721, 82614281, 85790889, 89058241, 92418049, + 95872041, 99421961, 103069569, 106816641, 110664969, 114616361, 118672641, + 122835649, 127107241, 131489289, 135983681, 140592321, 145317129, 150160041, + 155123009, 160208001, 165417001, 170752009, 176215041, 181808129, 187533321, + 193392681, 199388289, 205522241, 211796649, 218213641, 224775361, 231483969, + 238341641, 245350569, 252512961, 259831041, 267307049, 274943241, 282741889, + 290705281, 298835721, 307135529, 315607041, 324252609, 333074601, 342075401, + 351257409, 360623041, 370174729, 379914921, 389846081, 399970689, 410291241, + 420810249, 431530241, 442453761, 453583369, 464921641, 476471169, 488234561, + 500214441, 512413449, 524834241, 537479489, 550351881, 563454121, 576788929, + 590359041, 604167209, 618216201, 632508801, +#if defined(CUSTOM_MODES) + /*...208:*/ + 647047809, 661836041, 676876329, 692171521, 707724481, 723538089, 739615241, + 755958849, 772571841, 789457161, 806617769, 824056641, 841776769, 859781161, + 878072841, 896654849, 915530241, 934702089, 954173481, 973947521, 994027329, + 1014416041, 1035116809, 1056132801, 1077467201, 1099123209, 1121104041, + 1143412929, 1166053121, 1189027881, 1212340489, 1235994241, +#endif + /*N=6, K=6...96:*/ + 1683, 3653, 7183, 13073, 22363, 36365, 56695, 85305, 124515, 177045, 246047, + 335137, 448427, 590557, 766727, 982729, 1244979, 1560549, 1937199, 2383409, + 2908411, 3522221, 4235671, 5060441, 6009091, 7095093, 8332863, 9737793, + 11326283, 13115773, 15124775, 17372905, 19880915, 22670725, 25765455, + 29189457, 32968347, 37129037, 41699767, 46710137, 52191139, 58175189, + 64696159, 71789409, 79491819, 87841821, 96879431, 106646281, 117185651, + 128542501, 140763503, 153897073, 167993403, 183104493, 199284183, 216588185, + 235074115, 254801525, 275831935, 298228865, 322057867, 347386557, 374284647, + 402823977, 433078547, 465124549, 499040399, 534906769, 572806619, 612825229, + 655050231, 699571641, 746481891, 795875861, 847850911, 902506913, 959946283, + 1020274013, 1083597703, 1150027593, 1219676595, 1292660325, 1369097135, + 1449108145, 1532817275, 1620351277, 1711839767, 1807415257, 1907213187, + 2011371957, 2120032959, +#if defined(CUSTOM_MODES) + /*...109:*/ + 2233340609U, 2351442379U, 2474488829U, 2602633639U, 2736033641U, 2874848851U, + 3019242501U, 3169381071U, 3325434321U, 3487575323U, 3655980493U, 3830829623U, + 4012305913U, +#endif + /*N=7, K=7...54*/ + 8989, 19825, 40081, 75517, 134245, 227305, 369305, 579125, 880685, 1303777, + 1884961, 2668525, 3707509, 5064793, 6814249, 9041957, 11847485, 15345233, + 19665841, 24957661, 31388293, 39146185, 48442297, 59511829, 72616013, + 88043969, 106114625, 127178701, 151620757, 179861305, 212358985, 249612805, + 292164445, 340600625, 395555537, 457713341, 527810725, 606639529, 695049433, + 793950709, 904317037, 1027188385, 1163673953, 1314955181, 1482288821, + 1667010073, 1870535785, 2094367717, +#if defined(CUSTOM_MODES) + /*...60:*/ + 2340095869U, 2609401873U, 2904062449U, 3225952925U, 3577050821U, 3959439497U, +#endif + /*N=8, K=8...37*/ + 48639, 108545, 224143, 433905, 795455, 1392065, 2340495, 3800305, 5984767, + 9173505, 13726991, 20103025, 28875327, 40754369, 56610575, 77500017, + 104692735, 139703809, 184327311, 240673265, 311207743, 398796225, 506750351, + 638878193, 799538175, 993696769, 1226990095, 1505789553, 1837271615, + 2229491905U, +#if defined(CUSTOM_MODES) + /*...40:*/ + 2691463695U, 3233240945U, 3866006015U, +#endif + /*N=9, K=9...28:*/ + 265729, 598417, 1256465, 2485825, 4673345, 8405905, 14546705, 24331777, + 39490049, 62390545, 96220561, 145198913, 214828609, 312193553, 446304145, + 628496897, 872893441, 1196924561, 1621925137, 2173806145U, +#if defined(CUSTOM_MODES) + /*...29:*/ + 2883810113U, +#endif + /*N=10, K=10...24:*/ + 1462563, 3317445, 7059735, 14218905, 27298155, 50250765, 89129247, 152951073, + 254831667, 413442773, 654862247, 1014889769, 1541911931, 2300409629U, + 3375210671U, + /*N=11, K=11...19:*/ + 8097453, 18474633, 39753273, 81270333, 158819253, 298199265, 540279585, + 948062325, 1616336765, +#if defined(CUSTOM_MODES) + /*...20:*/ + 2684641785U, +#endif + /*N=12, K=12...18:*/ + 45046719, 103274625, 224298231, 464387817, 921406335, 1759885185, + 3248227095U, + /*N=13, K=13...16:*/ + 251595969, 579168825, 1267854873, 2653649025U, + /*N=14, K=14:*/ + 1409933619 +}; + +#if defined(CUSTOM_MODES) +static const opus_uint32 *const CELT_PVQ_U_ROW[15]={ + CELT_PVQ_U_DATA+ 0,CELT_PVQ_U_DATA+ 208,CELT_PVQ_U_DATA+ 415, + CELT_PVQ_U_DATA+ 621,CELT_PVQ_U_DATA+ 826,CELT_PVQ_U_DATA+1030, + CELT_PVQ_U_DATA+1233,CELT_PVQ_U_DATA+1336,CELT_PVQ_U_DATA+1389, + CELT_PVQ_U_DATA+1421,CELT_PVQ_U_DATA+1441,CELT_PVQ_U_DATA+1455, + CELT_PVQ_U_DATA+1464,CELT_PVQ_U_DATA+1470,CELT_PVQ_U_DATA+1473 +}; +#else +static const opus_uint32 *const CELT_PVQ_U_ROW[15]={ + CELT_PVQ_U_DATA+ 0,CELT_PVQ_U_DATA+ 176,CELT_PVQ_U_DATA+ 351, + CELT_PVQ_U_DATA+ 525,CELT_PVQ_U_DATA+ 698,CELT_PVQ_U_DATA+ 870, + CELT_PVQ_U_DATA+1041,CELT_PVQ_U_DATA+1131,CELT_PVQ_U_DATA+1178, + CELT_PVQ_U_DATA+1207,CELT_PVQ_U_DATA+1226,CELT_PVQ_U_DATA+1240, + CELT_PVQ_U_DATA+1248,CELT_PVQ_U_DATA+1254,CELT_PVQ_U_DATA+1257 +}; +#endif + +#if defined(CUSTOM_MODES) +void get_required_bits(opus_int16 *_bits,int _n,int _maxk,int _frac){ + int k; + /*_maxk==0 => there's nothing to do.*/ + celt_assert(_maxk>0); + _bits[0]=0; + for(k=1;k<=_maxk;k++)_bits[k]=log2_frac(CELT_PVQ_V(_n,k),_frac); +} +#endif + +static opus_uint32 icwrs(int _n,const int *_y){ + opus_uint32 i; + int j; + int k; + celt_assert(_n>=2); + j=_n-1; + i=_y[j]<0; + k=abs(_y[j]); + do{ + j--; + i+=CELT_PVQ_U(_n-j,k); + k+=abs(_y[j]); + if(_y[j]<0)i+=CELT_PVQ_U(_n-j,k+1); + } + while(j>0); + return i; +} + +void encode_pulses(const int *_y,int _n,int _k,ec_enc *_enc){ + celt_assert(_k>0); + ec_enc_uint(_enc,icwrs(_n,_y),CELT_PVQ_V(_n,_k)); +} + +static void cwrsi(int _n,int _k,opus_uint32 _i,int *_y){ + opus_uint32 p; + int s; + int k0; + celt_assert(_k>0); + celt_assert(_n>1); + while(_n>2){ + opus_uint32 q; + /*Lots of pulses case:*/ + if(_k>=_n){ + const opus_uint32 *row; + row=CELT_PVQ_U_ROW[_n]; + /*Are the pulses in this dimension negative?*/ + p=row[_k+1]; + s=-(_i>=p); + _i-=p&s; + /*Count how many pulses were placed in this dimension.*/ + k0=_k; + q=row[_n]; + if(q>_i){ + celt_assert(p>q); + _k=_n; + do p=CELT_PVQ_U_ROW[--_k][_n]; + while(p>_i); + } + else for(p=row[_k];p>_i;p=row[_k])_k--; + _i-=p; + *_y++=(k0-_k+s)^s; + } + /*Lots of dimensions case:*/ + else{ + /*Are there any pulses in this dimension at all?*/ + p=CELT_PVQ_U_ROW[_k][_n]; + q=CELT_PVQ_U_ROW[_k+1][_n]; + if(p<=_i&&_i<q){ + _i-=p; + *_y++=0; + } + else{ + /*Are the pulses in this dimension negative?*/ + s=-(_i>=q); + _i-=q&s; + /*Count how many pulses were placed in this dimension.*/ + k0=_k; + do p=CELT_PVQ_U_ROW[--_k][_n]; + while(p>_i); + _i-=p; + *_y++=(k0-_k+s)^s; + } + } + _n--; + } + /*_n==2*/ + p=2*_k+1; + s=-(_i>=p); + _i-=p&s; + k0=_k; + _k=(_i+1)>>1; + if(_k)_i-=2*_k-1; + *_y++=(k0-_k+s)^s; + /*_n==1*/ + s=-(int)_i; + *_y=(_k+s)^s; +} + +void decode_pulses(int *_y,int _n,int _k,ec_dec *_dec){ + cwrsi(_n,_k,ec_dec_uint(_dec,CELT_PVQ_V(_n,_k)),_y); +} + +#else /* SMALL_FOOTPRINT */ + +/*Computes the next row/column of any recurrence that obeys the relation + u[i][j]=u[i-1][j]+u[i][j-1]+u[i-1][j-1]. + _ui0 is the base case for the new row/column.*/ +static OPUS_INLINE void unext(opus_uint32 *_ui,unsigned _len,opus_uint32 _ui0){ + opus_uint32 ui1; + unsigned j; + /*This do-while will overrun the array if we don't have storage for at least + 2 values.*/ + j=1; do { + ui1=UADD32(UADD32(_ui[j],_ui[j-1]),_ui0); + _ui[j-1]=_ui0; + _ui0=ui1; + } while (++j<_len); + _ui[j-1]=_ui0; +} + +/*Computes the previous row/column of any recurrence that obeys the relation + u[i-1][j]=u[i][j]-u[i][j-1]-u[i-1][j-1]. + _ui0 is the base case for the new row/column.*/ +static OPUS_INLINE void uprev(opus_uint32 *_ui,unsigned _n,opus_uint32 _ui0){ + opus_uint32 ui1; + unsigned j; + /*This do-while will overrun the array if we don't have storage for at least + 2 values.*/ + j=1; do { + ui1=USUB32(USUB32(_ui[j],_ui[j-1]),_ui0); + _ui[j-1]=_ui0; + _ui0=ui1; + } while (++j<_n); + _ui[j-1]=_ui0; +} + +/*Compute V(_n,_k), as well as U(_n,0..._k+1). + _u: On exit, _u[i] contains U(_n,i) for i in [0..._k+1].*/ +static opus_uint32 ncwrs_urow(unsigned _n,unsigned _k,opus_uint32 *_u){ + opus_uint32 um2; + unsigned len; + unsigned k; + len=_k+2; + /*We require storage at least 3 values (e.g., _k>0).*/ + celt_assert(len>=3); + _u[0]=0; + _u[1]=um2=1; + /*If _n==0, _u[0] should be 1 and the rest should be 0.*/ + /*If _n==1, _u[i] should be 1 for i>1.*/ + celt_assert(_n>=2); + /*If _k==0, the following do-while loop will overflow the buffer.*/ + celt_assert(_k>0); + k=2; + do _u[k]=(k<<1)-1; + while(++k<len); + for(k=2;k<_n;k++)unext(_u+1,_k+1,1); + return _u[_k]+_u[_k+1]; +} + +/*Returns the _i'th combination of _k elements chosen from a set of size _n + with associated sign bits. + _y: Returns the vector of pulses. + _u: Must contain entries [0..._k+1] of row _n of U() on input. + Its contents will be destructively modified.*/ +static void cwrsi(int _n,int _k,opus_uint32 _i,int *_y,opus_uint32 *_u){ + int j; + celt_assert(_n>0); + j=0; + do{ + opus_uint32 p; + int s; + int yj; + p=_u[_k+1]; + s=-(_i>=p); + _i-=p&s; + yj=_k; + p=_u[_k]; + while(p>_i)p=_u[--_k]; + _i-=p; + yj-=_k; + _y[j]=(yj+s)^s; + uprev(_u,_k+2,0); + } + while(++j<_n); +} + +/*Returns the index of the given combination of K elements chosen from a set + of size 1 with associated sign bits. + _y: The vector of pulses, whose sum of absolute values is K. + _k: Returns K.*/ +static OPUS_INLINE opus_uint32 icwrs1(const int *_y,int *_k){ + *_k=abs(_y[0]); + return _y[0]<0; +} + +/*Returns the index of the given combination of K elements chosen from a set + of size _n with associated sign bits. + _y: The vector of pulses, whose sum of absolute values must be _k. + _nc: Returns V(_n,_k).*/ +static OPUS_INLINE opus_uint32 icwrs(int _n,int _k,opus_uint32 *_nc,const int *_y, + opus_uint32 *_u){ + opus_uint32 i; + int j; + int k; + /*We can't unroll the first two iterations of the loop unless _n>=2.*/ + celt_assert(_n>=2); + _u[0]=0; + for(k=1;k<=_k+1;k++)_u[k]=(k<<1)-1; + i=icwrs1(_y+_n-1,&k); + j=_n-2; + i+=_u[k]; + k+=abs(_y[j]); + if(_y[j]<0)i+=_u[k+1]; + while(j-->0){ + unext(_u,_k+2,0); + i+=_u[k]; + k+=abs(_y[j]); + if(_y[j]<0)i+=_u[k+1]; + } + *_nc=_u[k]+_u[k+1]; + return i; +} + +#ifdef CUSTOM_MODES +void get_required_bits(opus_int16 *_bits,int _n,int _maxk,int _frac){ + int k; + /*_maxk==0 => there's nothing to do.*/ + celt_assert(_maxk>0); + _bits[0]=0; + if (_n==1) + { + for (k=1;k<=_maxk;k++) + _bits[k] = 1<<_frac; + } + else { + VARDECL(opus_uint32,u); + SAVE_STACK; + ALLOC(u,_maxk+2U,opus_uint32); + ncwrs_urow(_n,_maxk,u); + for(k=1;k<=_maxk;k++) + _bits[k]=log2_frac(u[k]+u[k+1],_frac); + RESTORE_STACK; + } +} +#endif /* CUSTOM_MODES */ + +void encode_pulses(const int *_y,int _n,int _k,ec_enc *_enc){ + opus_uint32 i; + VARDECL(opus_uint32,u); + opus_uint32 nc; + SAVE_STACK; + celt_assert(_k>0); + ALLOC(u,_k+2U,opus_uint32); + i=icwrs(_n,_k,&nc,_y,u); + ec_enc_uint(_enc,i,nc); + RESTORE_STACK; +} + +void decode_pulses(int *_y,int _n,int _k,ec_dec *_dec){ + VARDECL(opus_uint32,u); + SAVE_STACK; + celt_assert(_k>0); + ALLOC(u,_k+2U,opus_uint32); + cwrsi(_n,_k,ec_dec_uint(_dec,ncwrs_urow(_n,_k,u)),_y,u); + RESTORE_STACK; +} + +#endif /* SMALL_FOOTPRINT */ diff --git a/drivers/opus/celt/cwrs.h b/drivers/opus/celt/cwrs.h new file mode 100644 index 0000000000..7dfbd076d1 --- /dev/null +++ b/drivers/opus/celt/cwrs.h @@ -0,0 +1,48 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Copyright (c) 2007-2009 Timothy B. Terriberry + Written by Timothy B. Terriberry and Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef CWRS_H +#define CWRS_H + +#include "arch.h" +#include "stack_alloc.h" +#include "entenc.h" +#include "entdec.h" + +#ifdef CUSTOM_MODES +int log2_frac(opus_uint32 val, int frac); +#endif + +void get_required_bits(opus_int16 *bits, int N, int K, int frac); + +void encode_pulses(const int *_y, int N, int K, ec_enc *enc); + +void decode_pulses(int *_y, int N, int K, ec_dec *dec); + +#endif /* CWRS_H */ diff --git a/drivers/opus/celt/ecintrin.h b/drivers/opus/celt/ecintrin.h new file mode 100644 index 0000000000..2263cff6bd --- /dev/null +++ b/drivers/opus/celt/ecintrin.h @@ -0,0 +1,87 @@ +/* Copyright (c) 2003-2008 Timothy B. Terriberry + Copyright (c) 2008 Xiph.Org Foundation */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +/*Some common macros for potential platform-specific optimization.*/ +#include "opus_types.h" +#include <math.h> +#include <limits.h> +#include "arch.h" +#if !defined(_ecintrin_H) +# define _ecintrin_H (1) + +/*Some specific platforms may have optimized intrinsic or OPUS_INLINE assembly + versions of these functions which can substantially improve performance. + We define macros for them to allow easy incorporation of these non-ANSI + features.*/ + +/*Modern gcc (4.x) can compile the naive versions of min and max with cmov if + given an appropriate architecture, but the branchless bit-twiddling versions + are just as fast, and do not require any special target architecture. + Earlier gcc versions (3.x) compiled both code to the same assembly + instructions, because of the way they represented ((_b)>(_a)) internally.*/ +# define EC_MINI(_a,_b) ((_a)+(((_b)-(_a))&-((_b)<(_a)))) + +/*Count leading zeros. + This macro should only be used for implementing ec_ilog(), if it is defined. + All other code should use EC_ILOG() instead.*/ +#if defined(_MSC_VER) && (_MSC_VER >= 1400) +# include <intrin.h> +/*In _DEBUG mode this is not an intrinsic by default.*/ +# pragma intrinsic(_BitScanReverse) + +static __inline int ec_bsr(unsigned long _x){ + unsigned long ret; + _BitScanReverse(&ret,_x); + return (int)ret; +} +# define EC_CLZ0 (1) +# define EC_CLZ(_x) (-ec_bsr(_x)) +#elif defined(ENABLE_TI_DSPLIB) +# include "dsplib.h" +# define EC_CLZ0 (31) +# define EC_CLZ(_x) (_lnorm(_x)) +#elif __GNUC_PREREQ(3,4) +# if INT_MAX>=2147483647 +# define EC_CLZ0 ((int)sizeof(unsigned)*CHAR_BIT) +# define EC_CLZ(_x) (__builtin_clz(_x)) +# elif LONG_MAX>=2147483647L +# define EC_CLZ0 ((int)sizeof(unsigned long)*CHAR_BIT) +# define EC_CLZ(_x) (__builtin_clzl(_x)) +# endif +#endif + +#if defined(EC_CLZ) +/*Note that __builtin_clz is not defined when _x==0, according to the gcc + documentation (and that of the BSR instruction that implements it on x86). + The majority of the time we can never pass it zero. + When we need to, it can be special cased.*/ +# define EC_ILOG(_x) (EC_CLZ0-EC_CLZ(_x)) +#else +int ec_ilog(opus_uint32 _v); +# define EC_ILOG(_x) (ec_ilog(_x)) +#endif +#endif diff --git a/drivers/opus/celt/entcode.c b/drivers/opus/celt/entcode.c new file mode 100644 index 0000000000..fd817a9db5 --- /dev/null +++ b/drivers/opus/celt/entcode.c @@ -0,0 +1,93 @@ +/* Copyright (c) 2001-2011 Timothy B. Terriberry +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef OPUS_HAVE_CONFIG_H +#include "opus_config.h" +#endif + +#include "entcode.h" +#include "arch.h" + +#if !defined(EC_CLZ) +/*This is a fallback for systems where we don't know how to access + a BSR or CLZ instruction (see ecintrin.h). + If you are optimizing Opus on a new platform and it has a native CLZ or + BZR (e.g. cell, MIPS, x86, etc) then making it available to Opus will be + an easy performance win.*/ +int ec_ilog(opus_uint32 _v){ + /*On a Pentium M, this branchless version tested as the fastest on + 1,000,000,000 random 32-bit integers, edging out a similar version with + branches, and a 256-entry LUT version.*/ + int ret; + int m; + ret=!!_v; + m=!!(_v&0xFFFF0000)<<4; + _v>>=m; + ret|=m; + m=!!(_v&0xFF00)<<3; + _v>>=m; + ret|=m; + m=!!(_v&0xF0)<<2; + _v>>=m; + ret|=m; + m=!!(_v&0xC)<<1; + _v>>=m; + ret|=m; + ret+=!!(_v&0x2); + return ret; +} +#endif + +opus_uint32 ec_tell_frac(ec_ctx *_this){ + opus_uint32 nbits; + opus_uint32 r; + int l; + int i; + /*To handle the non-integral number of bits still left in the encoder/decoder + state, we compute the worst-case number of bits of val that must be + encoded to ensure that the value is inside the range for any possible + subsequent bits. + The computation here is independent of val itself (the decoder does not + even track that value), even though the real number of bits used after + ec_enc_done() may be 1 smaller if rng is a power of two and the + corresponding trailing bits of val are all zeros. + If we did try to track that special case, then coding a value with a + probability of 1/(1<<n) might sometimes appear to use more than n bits. + This may help explain the surprising result that a newly initialized + encoder or decoder claims to have used 1 bit.*/ + nbits=_this->nbits_total<<BITRES; + l=EC_ILOG(_this->rng); + r=_this->rng>>(l-16); + for(i=BITRES;i-->0;){ + int b; + r=r*r>>15; + b=(int)(r>>16); + l=l<<1|b; + r>>=b; + } + return nbits-l; +} diff --git a/drivers/opus/celt/entcode.h b/drivers/opus/celt/entcode.h new file mode 100644 index 0000000000..dd13e49e50 --- /dev/null +++ b/drivers/opus/celt/entcode.h @@ -0,0 +1,117 @@ +/* Copyright (c) 2001-2011 Timothy B. Terriberry + Copyright (c) 2008-2009 Xiph.Org Foundation */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#include "opus_types.h" +#include "opus_defines.h" + +#if !defined(_entcode_H) +# define _entcode_H (1) +# include <limits.h> +# include <stddef.h> +# include "ecintrin.h" + +/*OPT: ec_window must be at least 32 bits, but if you have fast arithmetic on a + larger type, you can speed up the decoder by using it here.*/ +typedef opus_uint32 ec_window; +typedef struct ec_ctx ec_ctx; +typedef struct ec_ctx ec_enc; +typedef struct ec_ctx ec_dec; + +# define EC_WINDOW_SIZE ((int)sizeof(ec_window)*CHAR_BIT) + +/*The number of bits to use for the range-coded part of unsigned integers.*/ +# define EC_UINT_BITS (8) + +/*The resolution of fractional-precision bit usage measurements, i.e., + 3 => 1/8th bits.*/ +# define BITRES 3 + +/*The entropy encoder/decoder context. + We use the same structure for both, so that common functions like ec_tell() + can be used on either one.*/ +struct ec_ctx{ + /*Buffered input/output.*/ + unsigned char *buf; + /*The size of the buffer.*/ + opus_uint32 storage; + /*The offset at which the last byte containing raw bits was read/written.*/ + opus_uint32 end_offs; + /*Bits that will be read from/written at the end.*/ + ec_window end_window; + /*Number of valid bits in end_window.*/ + int nend_bits; + /*The total number of whole bits read/written. + This does not include partial bits currently in the range coder.*/ + int nbits_total; + /*The offset at which the next range coder byte will be read/written.*/ + opus_uint32 offs; + /*The number of values in the current range.*/ + opus_uint32 rng; + /*In the decoder: the difference between the top of the current range and + the input value, minus one. + In the encoder: the low end of the current range.*/ + opus_uint32 val; + /*In the decoder: the saved normalization factor from ec_decode(). + In the encoder: the number of oustanding carry propagating symbols.*/ + opus_uint32 ext; + /*A buffered input/output symbol, awaiting carry propagation.*/ + int rem; + /*Nonzero if an error occurred.*/ + int error; +}; + +static OPUS_INLINE opus_uint32 ec_range_bytes(ec_ctx *_this){ + return _this->offs; +} + +static OPUS_INLINE unsigned char *ec_get_buffer(ec_ctx *_this){ + return _this->buf; +} + +static OPUS_INLINE int ec_get_error(ec_ctx *_this){ + return _this->error; +} + +/*Returns the number of bits "used" by the encoded or decoded symbols so far. + This same number can be computed in either the encoder or the decoder, and is + suitable for making coding decisions. + Return: The number of bits. + This will always be slightly larger than the exact value (e.g., all + rounding error is in the positive direction).*/ +static OPUS_INLINE int ec_tell(ec_ctx *_this){ + return _this->nbits_total-EC_ILOG(_this->rng); +} + +/*Returns the number of bits "used" by the encoded or decoded symbols so far. + This same number can be computed in either the encoder or the decoder, and is + suitable for making coding decisions. + Return: The number of bits scaled by 2**BITRES. + This will always be slightly larger than the exact value (e.g., all + rounding error is in the positive direction).*/ +opus_uint32 ec_tell_frac(ec_ctx *_this); + +#endif diff --git a/drivers/opus/celt/entdec.c b/drivers/opus/celt/entdec.c new file mode 100644 index 0000000000..383da571c9 --- /dev/null +++ b/drivers/opus/celt/entdec.c @@ -0,0 +1,245 @@ +/* Copyright (c) 2001-2011 Timothy B. Terriberry + Copyright (c) 2008-2009 Xiph.Org Foundation */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef OPUS_HAVE_CONFIG_H +#include "opus_config.h" +#endif + +#include <stddef.h> +#include "os_support.h" +#include "arch.h" +#include "entdec.h" +#include "mfrngcod.h" + +/*A range decoder. + This is an entropy decoder based upon \cite{Mar79}, which is itself a + rediscovery of the FIFO arithmetic code introduced by \cite{Pas76}. + It is very similar to arithmetic encoding, except that encoding is done with + digits in any base, instead of with bits, and so it is faster when using + larger bases (i.e.: a byte). + The author claims an average waste of $\frac{1}{2}\log_b(2b)$ bits, where $b$ + is the base, longer than the theoretical optimum, but to my knowledge there + is no published justification for this claim. + This only seems true when using near-infinite precision arithmetic so that + the process is carried out with no rounding errors. + + An excellent description of implementation details is available at + http://www.arturocampos.com/ac_range.html + A recent work \cite{MNW98} which proposes several changes to arithmetic + encoding for efficiency actually re-discovers many of the principles + behind range encoding, and presents a good theoretical analysis of them. + + End of stream is handled by writing out the smallest number of bits that + ensures that the stream will be correctly decoded regardless of the value of + any subsequent bits. + ec_tell() can be used to determine how many bits were needed to decode + all the symbols thus far; other data can be packed in the remaining bits of + the input buffer. + @PHDTHESIS{Pas76, + author="Richard Clark Pasco", + title="Source coding algorithms for fast data compression", + school="Dept. of Electrical Engineering, Stanford University", + address="Stanford, CA", + month=May, + year=1976 + } + @INPROCEEDINGS{Mar79, + author="Martin, G.N.N.", + title="Range encoding: an algorithm for removing redundancy from a digitised + message", + booktitle="Video & Data Recording Conference", + year=1979, + address="Southampton", + month=Jul + } + @ARTICLE{MNW98, + author="Alistair Moffat and Radford Neal and Ian H. Witten", + title="Arithmetic Coding Revisited", + journal="{ACM} Transactions on Information Systems", + year=1998, + volume=16, + number=3, + pages="256--294", + month=Jul, + URL="http://www.stanford.edu/class/ee398a/handouts/papers/Moffat98ArithmCoding.pdf" + }*/ + +static int ec_read_byte(ec_dec *_this){ + return _this->offs<_this->storage?_this->buf[_this->offs++]:0; +} + +static int ec_read_byte_from_end(ec_dec *_this){ + return _this->end_offs<_this->storage? + _this->buf[_this->storage-++(_this->end_offs)]:0; +} + +/*Normalizes the contents of val and rng so that rng lies entirely in the + high-order symbol.*/ +static void ec_dec_normalize(ec_dec *_this){ + /*If the range is too small, rescale it and input some bits.*/ + while(_this->rng<=EC_CODE_BOT){ + int sym; + _this->nbits_total+=EC_SYM_BITS; + _this->rng<<=EC_SYM_BITS; + /*Use up the remaining bits from our last symbol.*/ + sym=_this->rem; + /*Read the next value from the input.*/ + _this->rem=ec_read_byte(_this); + /*Take the rest of the bits we need from this new symbol.*/ + sym=(sym<<EC_SYM_BITS|_this->rem)>>(EC_SYM_BITS-EC_CODE_EXTRA); + /*And subtract them from val, capped to be less than EC_CODE_TOP.*/ + _this->val=((_this->val<<EC_SYM_BITS)+(EC_SYM_MAX&~sym))&(EC_CODE_TOP-1); + } +} + +void ec_dec_init(ec_dec *_this,unsigned char *_buf,opus_uint32 _storage){ + _this->buf=_buf; + _this->storage=_storage; + _this->end_offs=0; + _this->end_window=0; + _this->nend_bits=0; + /*This is the offset from which ec_tell() will subtract partial bits. + The final value after the ec_dec_normalize() call will be the same as in + the encoder, but we have to compensate for the bits that are added there.*/ + _this->nbits_total=EC_CODE_BITS+1 + -((EC_CODE_BITS-EC_CODE_EXTRA)/EC_SYM_BITS)*EC_SYM_BITS; + _this->offs=0; + _this->rng=1U<<EC_CODE_EXTRA; + _this->rem=ec_read_byte(_this); + _this->val=_this->rng-1-(_this->rem>>(EC_SYM_BITS-EC_CODE_EXTRA)); + _this->error=0; + /*Normalize the interval.*/ + ec_dec_normalize(_this); +} + +unsigned ec_decode(ec_dec *_this,unsigned _ft){ + unsigned s; + _this->ext=_this->rng/_ft; + s=(unsigned)(_this->val/_this->ext); + return _ft-EC_MINI(s+1,_ft); +} + +unsigned ec_decode_bin(ec_dec *_this,unsigned _bits){ + unsigned s; + _this->ext=_this->rng>>_bits; + s=(unsigned)(_this->val/_this->ext); + return (1U<<_bits)-EC_MINI(s+1U,1U<<_bits); +} + +void ec_dec_update(ec_dec *_this,unsigned _fl,unsigned _fh,unsigned _ft){ + opus_uint32 s; + s=IMUL32(_this->ext,_ft-_fh); + _this->val-=s; + _this->rng=_fl>0?IMUL32(_this->ext,_fh-_fl):_this->rng-s; + ec_dec_normalize(_this); +} + +/*The probability of having a "one" is 1/(1<<_logp).*/ +int ec_dec_bit_logp(ec_dec *_this,unsigned _logp){ + opus_uint32 r; + opus_uint32 d; + opus_uint32 s; + int ret; + r=_this->rng; + d=_this->val; + s=r>>_logp; + ret=d<s; + if(!ret)_this->val=d-s; + _this->rng=ret?s:r-s; + ec_dec_normalize(_this); + return ret; +} + +int ec_dec_icdf(ec_dec *_this,const unsigned char *_icdf,unsigned _ftb){ + opus_uint32 r; + opus_uint32 d; + opus_uint32 s; + opus_uint32 t; + int ret; + s=_this->rng; + d=_this->val; + r=s>>_ftb; + ret=-1; + do{ + t=s; + s=IMUL32(r,_icdf[++ret]); + } + while(d<s); + _this->val=d-s; + _this->rng=t-s; + ec_dec_normalize(_this); + return ret; +} + +opus_uint32 ec_dec_uint(ec_dec *_this,opus_uint32 _ft){ + unsigned ft; + unsigned s; + int ftb; + /*In order to optimize EC_ILOG(), it is undefined for the value 0.*/ + celt_assert(_ft>1); + _ft--; + ftb=EC_ILOG(_ft); + if(ftb>EC_UINT_BITS){ + opus_uint32 t; + ftb-=EC_UINT_BITS; + ft=(unsigned)(_ft>>ftb)+1; + s=ec_decode(_this,ft); + ec_dec_update(_this,s,s+1,ft); + t=(opus_uint32)s<<ftb|ec_dec_bits(_this,ftb); + if(t<=_ft)return t; + _this->error=1; + return _ft; + } + else{ + _ft++; + s=ec_decode(_this,(unsigned)_ft); + ec_dec_update(_this,s,s+1,(unsigned)_ft); + return s; + } +} + +opus_uint32 ec_dec_bits(ec_dec *_this,unsigned _bits){ + ec_window window; + int available; + opus_uint32 ret; + window=_this->end_window; + available=_this->nend_bits; + if((unsigned)available<_bits){ + do{ + window|=(ec_window)ec_read_byte_from_end(_this)<<available; + available+=EC_SYM_BITS; + } + while(available<=EC_WINDOW_SIZE-EC_SYM_BITS); + } + ret=(opus_uint32)window&(((opus_uint32)1<<_bits)-1U); + window>>=_bits; + available-=_bits; + _this->end_window=window; + _this->nend_bits=available; + _this->nbits_total+=_bits; + return ret; +} diff --git a/drivers/opus/celt/entdec.h b/drivers/opus/celt/entdec.h new file mode 100644 index 0000000000..d8ab318730 --- /dev/null +++ b/drivers/opus/celt/entdec.h @@ -0,0 +1,100 @@ +/* Copyright (c) 2001-2011 Timothy B. Terriberry + Copyright (c) 2008-2009 Xiph.Org Foundation */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#if !defined(_entdec_H) +# define _entdec_H (1) +# include <limits.h> +# include "entcode.h" + +/*Initializes the decoder. + _buf: The input buffer to use. + Return: 0 on success, or a negative value on error.*/ +void ec_dec_init(ec_dec *_this,unsigned char *_buf,opus_uint32 _storage); + +/*Calculates the cumulative frequency for the next symbol. + This can then be fed into the probability model to determine what that + symbol is, and the additional frequency information required to advance to + the next symbol. + This function cannot be called more than once without a corresponding call to + ec_dec_update(), or decoding will not proceed correctly. + _ft: The total frequency of the symbols in the alphabet the next symbol was + encoded with. + Return: A cumulative frequency representing the encoded symbol. + If the cumulative frequency of all the symbols before the one that + was encoded was fl, and the cumulative frequency of all the symbols + up to and including the one encoded is fh, then the returned value + will fall in the range [fl,fh).*/ +unsigned ec_decode(ec_dec *_this,unsigned _ft); + +/*Equivalent to ec_decode() with _ft==1<<_bits.*/ +unsigned ec_decode_bin(ec_dec *_this,unsigned _bits); + +/*Advance the decoder past the next symbol using the frequency information the + symbol was encoded with. + Exactly one call to ec_decode() must have been made so that all necessary + intermediate calculations are performed. + _fl: The cumulative frequency of all symbols that come before the symbol + decoded. + _fh: The cumulative frequency of all symbols up to and including the symbol + decoded. + Together with _fl, this defines the range [_fl,_fh) in which the value + returned above must fall. + _ft: The total frequency of the symbols in the alphabet the symbol decoded + was encoded in. + This must be the same as passed to the preceding call to ec_decode().*/ +void ec_dec_update(ec_dec *_this,unsigned _fl,unsigned _fh,unsigned _ft); + +/* Decode a bit that has a 1/(1<<_logp) probability of being a one */ +int ec_dec_bit_logp(ec_dec *_this,unsigned _logp); + +/*Decodes a symbol given an "inverse" CDF table. + No call to ec_dec_update() is necessary after this call. + _icdf: The "inverse" CDF, such that symbol s falls in the range + [s>0?ft-_icdf[s-1]:0,ft-_icdf[s]), where ft=1<<_ftb. + The values must be monotonically non-increasing, and the last value + must be 0. + _ftb: The number of bits of precision in the cumulative distribution. + Return: The decoded symbol s.*/ +int ec_dec_icdf(ec_dec *_this,const unsigned char *_icdf,unsigned _ftb); + +/*Extracts a raw unsigned integer with a non-power-of-2 range from the stream. + The bits must have been encoded with ec_enc_uint(). + No call to ec_dec_update() is necessary after this call. + _ft: The number of integers that can be decoded (one more than the max). + This must be at least one, and no more than 2**32-1. + Return: The decoded bits.*/ +opus_uint32 ec_dec_uint(ec_dec *_this,opus_uint32 _ft); + +/*Extracts a sequence of raw bits from the stream. + The bits must have been encoded with ec_enc_bits(). + No call to ec_dec_update() is necessary after this call. + _ftb: The number of bits to extract. + This must be between 0 and 25, inclusive. + Return: The decoded bits.*/ +opus_uint32 ec_dec_bits(ec_dec *_this,unsigned _ftb); + +#endif diff --git a/drivers/opus/celt/entenc.c b/drivers/opus/celt/entenc.c new file mode 100644 index 0000000000..299329c63f --- /dev/null +++ b/drivers/opus/celt/entenc.c @@ -0,0 +1,294 @@ +/* Copyright (c) 2001-2011 Timothy B. Terriberry + Copyright (c) 2008-2009 Xiph.Org Foundation */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#if defined(OPUS_HAVE_CONFIG_H) +# include "opus_config.h" +#endif +#include "os_support.h" +#include "arch.h" +#include "entenc.h" +#include "mfrngcod.h" + +/*A range encoder. + See entdec.c and the references for implementation details \cite{Mar79,MNW98}. + + @INPROCEEDINGS{Mar79, + author="Martin, G.N.N.", + title="Range encoding: an algorithm for removing redundancy from a digitised + message", + booktitle="Video \& Data Recording Conference", + year=1979, + address="Southampton", + month=Jul + } + @ARTICLE{MNW98, + author="Alistair Moffat and Radford Neal and Ian H. Witten", + title="Arithmetic Coding Revisited", + journal="{ACM} Transactions on Information Systems", + year=1998, + volume=16, + number=3, + pages="256--294", + month=Jul, + URL="http://www.stanford.edu/class/ee398/handouts/papers/Moffat98ArithmCoding.pdf" + }*/ + +static int ec_write_byte(ec_enc *_this,unsigned _value){ + if(_this->offs+_this->end_offs>=_this->storage)return -1; + _this->buf[_this->offs++]=(unsigned char)_value; + return 0; +} + +static int ec_write_byte_at_end(ec_enc *_this,unsigned _value){ + if(_this->offs+_this->end_offs>=_this->storage)return -1; + _this->buf[_this->storage-++(_this->end_offs)]=(unsigned char)_value; + return 0; +} + +/*Outputs a symbol, with a carry bit. + If there is a potential to propagate a carry over several symbols, they are + buffered until it can be determined whether or not an actual carry will + occur. + If the counter for the buffered symbols overflows, then the stream becomes + undecodable. + This gives a theoretical limit of a few billion symbols in a single packet on + 32-bit systems. + The alternative is to truncate the range in order to force a carry, but + requires similar carry tracking in the decoder, needlessly slowing it down.*/ +static void ec_enc_carry_out(ec_enc *_this,int _c){ + if(_c!=EC_SYM_MAX){ + /*No further carry propagation possible, flush buffer.*/ + int carry; + carry=_c>>EC_SYM_BITS; + /*Don't output a byte on the first write. + This compare should be taken care of by branch-prediction thereafter.*/ + if(_this->rem>=0)_this->error|=ec_write_byte(_this,_this->rem+carry); + if(_this->ext>0){ + unsigned sym; + sym=(EC_SYM_MAX+carry)&EC_SYM_MAX; + do _this->error|=ec_write_byte(_this,sym); + while(--(_this->ext)>0); + } + _this->rem=_c&EC_SYM_MAX; + } + else _this->ext++; +} + +static void ec_enc_normalize(ec_enc *_this){ + /*If the range is too small, output some bits and rescale it.*/ + while(_this->rng<=EC_CODE_BOT){ + ec_enc_carry_out(_this,(int)(_this->val>>EC_CODE_SHIFT)); + /*Move the next-to-high-order symbol into the high-order position.*/ + _this->val=(_this->val<<EC_SYM_BITS)&(EC_CODE_TOP-1); + _this->rng<<=EC_SYM_BITS; + _this->nbits_total+=EC_SYM_BITS; + } +} + +void ec_enc_init(ec_enc *_this,unsigned char *_buf,opus_uint32 _size){ + _this->buf=_buf; + _this->end_offs=0; + _this->end_window=0; + _this->nend_bits=0; + /*This is the offset from which ec_tell() will subtract partial bits.*/ + _this->nbits_total=EC_CODE_BITS+1; + _this->offs=0; + _this->rng=EC_CODE_TOP; + _this->rem=-1; + _this->val=0; + _this->ext=0; + _this->storage=_size; + _this->error=0; +} + +void ec_encode(ec_enc *_this,unsigned _fl,unsigned _fh,unsigned _ft){ + opus_uint32 r; + r=_this->rng/_ft; + if(_fl>0){ + _this->val+=_this->rng-IMUL32(r,(_ft-_fl)); + _this->rng=IMUL32(r,(_fh-_fl)); + } + else _this->rng-=IMUL32(r,(_ft-_fh)); + ec_enc_normalize(_this); +} + +void ec_encode_bin(ec_enc *_this,unsigned _fl,unsigned _fh,unsigned _bits){ + opus_uint32 r; + r=_this->rng>>_bits; + if(_fl>0){ + _this->val+=_this->rng-IMUL32(r,((1U<<_bits)-_fl)); + _this->rng=IMUL32(r,(_fh-_fl)); + } + else _this->rng-=IMUL32(r,((1U<<_bits)-_fh)); + ec_enc_normalize(_this); +} + +/*The probability of having a "one" is 1/(1<<_logp).*/ +void ec_enc_bit_logp(ec_enc *_this,int _val,unsigned _logp){ + opus_uint32 r; + opus_uint32 s; + opus_uint32 l; + r=_this->rng; + l=_this->val; + s=r>>_logp; + r-=s; + if(_val)_this->val=l+r; + _this->rng=_val?s:r; + ec_enc_normalize(_this); +} + +void ec_enc_icdf(ec_enc *_this,int _s,const unsigned char *_icdf,unsigned _ftb){ + opus_uint32 r; + r=_this->rng>>_ftb; + if(_s>0){ + _this->val+=_this->rng-IMUL32(r,_icdf[_s-1]); + _this->rng=IMUL32(r,_icdf[_s-1]-_icdf[_s]); + } + else _this->rng-=IMUL32(r,_icdf[_s]); + ec_enc_normalize(_this); +} + +void ec_enc_uint(ec_enc *_this,opus_uint32 _fl,opus_uint32 _ft){ + unsigned ft; + unsigned fl; + int ftb; + /*In order to optimize EC_ILOG(), it is undefined for the value 0.*/ + celt_assert(_ft>1); + _ft--; + ftb=EC_ILOG(_ft); + if(ftb>EC_UINT_BITS){ + ftb-=EC_UINT_BITS; + ft=(_ft>>ftb)+1; + fl=(unsigned)(_fl>>ftb); + ec_encode(_this,fl,fl+1,ft); + ec_enc_bits(_this,_fl&(((opus_uint32)1<<ftb)-1U),ftb); + } + else ec_encode(_this,_fl,_fl+1,_ft+1); +} + +void ec_enc_bits(ec_enc *_this,opus_uint32 _fl,unsigned _bits){ + ec_window window; + int used; + window=_this->end_window; + used=_this->nend_bits; + celt_assert(_bits>0); + if(used+_bits>EC_WINDOW_SIZE){ + do{ + _this->error|=ec_write_byte_at_end(_this,(unsigned)window&EC_SYM_MAX); + window>>=EC_SYM_BITS; + used-=EC_SYM_BITS; + } + while(used>=EC_SYM_BITS); + } + window|=(ec_window)_fl<<used; + used+=_bits; + _this->end_window=window; + _this->nend_bits=used; + _this->nbits_total+=_bits; +} + +void ec_enc_patch_initial_bits(ec_enc *_this,unsigned _val,unsigned _nbits){ + int shift; + unsigned mask; + celt_assert(_nbits<=EC_SYM_BITS); + shift=EC_SYM_BITS-_nbits; + mask=((1<<_nbits)-1)<<shift; + if(_this->offs>0){ + /*The first byte has been finalized.*/ + _this->buf[0]=(unsigned char)((_this->buf[0]&~mask)|_val<<shift); + } + else if(_this->rem>=0){ + /*The first byte is still awaiting carry propagation.*/ + _this->rem=(_this->rem&~mask)|_val<<shift; + } + else if(_this->rng<=(EC_CODE_TOP>>_nbits)){ + /*The renormalization loop has never been run.*/ + _this->val=(_this->val&~((opus_uint32)mask<<EC_CODE_SHIFT))| + (opus_uint32)_val<<(EC_CODE_SHIFT+shift); + } + /*The encoder hasn't even encoded _nbits of data yet.*/ + else _this->error=-1; +} + +void ec_enc_shrink(ec_enc *_this,opus_uint32 _size){ + celt_assert(_this->offs+_this->end_offs<=_size); + OPUS_MOVE(_this->buf+_size-_this->end_offs, + _this->buf+_this->storage-_this->end_offs,_this->end_offs); + _this->storage=_size; +} + +void ec_enc_done(ec_enc *_this){ + ec_window window; + int used; + opus_uint32 msk; + opus_uint32 end; + int l; + /*We output the minimum number of bits that ensures that the symbols encoded + thus far will be decoded correctly regardless of the bits that follow.*/ + l=EC_CODE_BITS-EC_ILOG(_this->rng); + msk=(EC_CODE_TOP-1)>>l; + end=(_this->val+msk)&~msk; + if((end|msk)>=_this->val+_this->rng){ + l++; + msk>>=1; + end=(_this->val+msk)&~msk; + } + while(l>0){ + ec_enc_carry_out(_this,(int)(end>>EC_CODE_SHIFT)); + end=(end<<EC_SYM_BITS)&(EC_CODE_TOP-1); + l-=EC_SYM_BITS; + } + /*If we have a buffered byte flush it into the output buffer.*/ + if(_this->rem>=0||_this->ext>0)ec_enc_carry_out(_this,0); + /*If we have buffered extra bits, flush them as well.*/ + window=_this->end_window; + used=_this->nend_bits; + while(used>=EC_SYM_BITS){ + _this->error|=ec_write_byte_at_end(_this,(unsigned)window&EC_SYM_MAX); + window>>=EC_SYM_BITS; + used-=EC_SYM_BITS; + } + /*Clear any excess space and add any remaining extra bits to the last byte.*/ + if(!_this->error){ + OPUS_CLEAR(_this->buf+_this->offs, + _this->storage-_this->offs-_this->end_offs); + if(used>0){ + /*If there's no range coder data at all, give up.*/ + if(_this->end_offs>=_this->storage)_this->error=-1; + else{ + l=-l; + /*If we've busted, don't add too many extra bits to the last byte; it + would corrupt the range coder data, and that's more important.*/ + if(_this->offs+_this->end_offs>=_this->storage&&l<used){ + window&=(1<<l)-1; + _this->error=-1; + } + _this->buf[_this->storage-_this->end_offs-1]|=(unsigned char)window; + } + } + } +} diff --git a/drivers/opus/celt/entenc.h b/drivers/opus/celt/entenc.h new file mode 100644 index 0000000000..796bc4d572 --- /dev/null +++ b/drivers/opus/celt/entenc.h @@ -0,0 +1,110 @@ +/* Copyright (c) 2001-2011 Timothy B. Terriberry + Copyright (c) 2008-2009 Xiph.Org Foundation */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#if !defined(_entenc_H) +# define _entenc_H (1) +# include <stddef.h> +# include "entcode.h" + +/*Initializes the encoder. + _buf: The buffer to store output bytes in. + _size: The size of the buffer, in chars.*/ +void ec_enc_init(ec_enc *_this,unsigned char *_buf,opus_uint32 _size); +/*Encodes a symbol given its frequency information. + The frequency information must be discernable by the decoder, assuming it + has read only the previous symbols from the stream. + It is allowable to change the frequency information, or even the entire + source alphabet, so long as the decoder can tell from the context of the + previously encoded information that it is supposed to do so as well. + _fl: The cumulative frequency of all symbols that come before the one to be + encoded. + _fh: The cumulative frequency of all symbols up to and including the one to + be encoded. + Together with _fl, this defines the range [_fl,_fh) in which the + decoded value will fall. + _ft: The sum of the frequencies of all the symbols*/ +void ec_encode(ec_enc *_this,unsigned _fl,unsigned _fh,unsigned _ft); + +/*Equivalent to ec_encode() with _ft==1<<_bits.*/ +void ec_encode_bin(ec_enc *_this,unsigned _fl,unsigned _fh,unsigned _bits); + +/* Encode a bit that has a 1/(1<<_logp) probability of being a one */ +void ec_enc_bit_logp(ec_enc *_this,int _val,unsigned _logp); + +/*Encodes a symbol given an "inverse" CDF table. + _s: The index of the symbol to encode. + _icdf: The "inverse" CDF, such that symbol _s falls in the range + [_s>0?ft-_icdf[_s-1]:0,ft-_icdf[_s]), where ft=1<<_ftb. + The values must be monotonically non-increasing, and the last value + must be 0. + _ftb: The number of bits of precision in the cumulative distribution.*/ +void ec_enc_icdf(ec_enc *_this,int _s,const unsigned char *_icdf,unsigned _ftb); + +/*Encodes a raw unsigned integer in the stream. + _fl: The integer to encode. + _ft: The number of integers that can be encoded (one more than the max). + This must be at least one, and no more than 2**32-1.*/ +void ec_enc_uint(ec_enc *_this,opus_uint32 _fl,opus_uint32 _ft); + +/*Encodes a sequence of raw bits in the stream. + _fl: The bits to encode. + _ftb: The number of bits to encode. + This must be between 1 and 25, inclusive.*/ +void ec_enc_bits(ec_enc *_this,opus_uint32 _fl,unsigned _ftb); + +/*Overwrites a few bits at the very start of an existing stream, after they + have already been encoded. + This makes it possible to have a few flags up front, where it is easy for + decoders to access them without parsing the whole stream, even if their + values are not determined until late in the encoding process, without having + to buffer all the intermediate symbols in the encoder. + In order for this to work, at least _nbits bits must have already been + encoded using probabilities that are an exact power of two. + The encoder can verify the number of encoded bits is sufficient, but cannot + check this latter condition. + _val: The bits to encode (in the least _nbits significant bits). + They will be decoded in order from most-significant to least. + _nbits: The number of bits to overwrite. + This must be no more than 8.*/ +void ec_enc_patch_initial_bits(ec_enc *_this,unsigned _val,unsigned _nbits); + +/*Compacts the data to fit in the target size. + This moves up the raw bits at the end of the current buffer so they are at + the end of the new buffer size. + The caller must ensure that the amount of data that's already been written + will fit in the new size. + _size: The number of bytes in the new buffer. + This must be large enough to contain the bits already written, and + must be no larger than the existing size.*/ +void ec_enc_shrink(ec_enc *_this,opus_uint32 _size); + +/*Indicates that there are no more symbols to encode. + All reamining output bytes are flushed to the output buffer. + ec_enc_init() must be called before the encoder can be used again.*/ +void ec_enc_done(ec_enc *_this); + +#endif diff --git a/drivers/opus/celt/fixed_debug.h b/drivers/opus/celt/fixed_debug.h new file mode 100644 index 0000000000..80bc94910f --- /dev/null +++ b/drivers/opus/celt/fixed_debug.h @@ -0,0 +1,773 @@ +/* Copyright (C) 2003-2008 Jean-Marc Valin + Copyright (C) 2007-2012 Xiph.Org Foundation */ +/** + @file fixed_debug.h + @brief Fixed-point operations with debugging +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef FIXED_DEBUG_H +#define FIXED_DEBUG_H + +#include <stdio.h> +#include "opus_defines.h" + +#ifdef CELT_C +OPUS_EXPORT opus_int64 celt_mips=0; +#else +extern opus_int64 celt_mips; +#endif + +#define MULT16_16SU(a,b) ((opus_val32)(opus_val16)(a)*(opus_val32)(opus_uint16)(b)) +#define MULT32_32_Q31(a,b) ADD32(ADD32(SHL32(MULT16_16(SHR32((a),16),SHR((b),16)),1), SHR32(MULT16_16SU(SHR32((a),16),((b)&0x0000ffff)),15)), SHR32(MULT16_16SU(SHR32((b),16),((a)&0x0000ffff)),15)) + +/** 16x32 multiplication, followed by a 16-bit shift right. Results fits in 32 bits */ +#define MULT16_32_Q16(a,b) ADD32(MULT16_16((a),SHR32((b),16)), SHR32(MULT16_16SU((a),((b)&0x0000ffff)),16)) + +#define MULT16_32_P16(a,b) MULT16_32_PX(a,b,16) + +#define QCONST16(x,bits) ((opus_val16)(.5+(x)*(((opus_val32)1)<<(bits)))) +#define QCONST32(x,bits) ((opus_val32)(.5+(x)*(((opus_val32)1)<<(bits)))) + +#define VERIFY_SHORT(x) ((x)<=32767&&(x)>=-32768) +#define VERIFY_INT(x) ((x)<=2147483647LL&&(x)>=-2147483648LL) +#define VERIFY_UINT(x) ((x)<=(2147483647LLU<<1)) + +#define SHR(a,b) SHR32(a,b) +#define PSHR(a,b) PSHR32(a,b) + +static OPUS_INLINE short NEG16(int x) +{ + int res; + if (!VERIFY_SHORT(x)) + { + fprintf (stderr, "NEG16: input is not short: %d\n", (int)x); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = -x; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "NEG16: output is not short: %d\n", (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips++; + return res; +} +static OPUS_INLINE int NEG32(opus_int64 x) +{ + opus_int64 res; + if (!VERIFY_INT(x)) + { + fprintf (stderr, "NEG16: input is not int: %d\n", (int)x); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = -x; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "NEG16: output is not int: %d\n", (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=2; + return res; +} + +#define EXTRACT16(x) EXTRACT16_(x, __FILE__, __LINE__) +static OPUS_INLINE short EXTRACT16_(int x, char *file, int line) +{ + int res; + if (!VERIFY_SHORT(x)) + { + fprintf (stderr, "EXTRACT16: input is not short: %d in %s: line %d\n", x, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = x; + celt_mips++; + return res; +} + +#define EXTEND32(x) EXTEND32_(x, __FILE__, __LINE__) +static OPUS_INLINE int EXTEND32_(int x, char *file, int line) +{ + int res; + if (!VERIFY_SHORT(x)) + { + fprintf (stderr, "EXTEND32: input is not short: %d in %s: line %d\n", x, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = x; + celt_mips++; + return res; +} + +#define SHR16(a, shift) SHR16_(a, shift, __FILE__, __LINE__) +static OPUS_INLINE short SHR16_(int a, int shift, char *file, int line) +{ + int res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(shift)) + { + fprintf (stderr, "SHR16: inputs are not short: %d >> %d in %s: line %d\n", a, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a>>shift; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "SHR16: output is not short: %d in %s: line %d\n", res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips++; + return res; +} +#define SHL16(a, shift) SHL16_(a, shift, __FILE__, __LINE__) +static OPUS_INLINE short SHL16_(int a, int shift, char *file, int line) +{ + int res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(shift)) + { + fprintf (stderr, "SHL16: inputs are not short: %d %d in %s: line %d\n", a, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a<<shift; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "SHL16: output is not short: %d in %s: line %d\n", res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips++; + return res; +} + +static OPUS_INLINE int SHR32(opus_int64 a, int shift) +{ + opus_int64 res; + if (!VERIFY_INT(a) || !VERIFY_SHORT(shift)) + { + fprintf (stderr, "SHR32: inputs are not int: %d %d\n", (int)a, shift); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a>>shift; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "SHR32: output is not int: %d\n", (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=2; + return res; +} +#define SHL32(a, shift) SHL32_(a, shift, __FILE__, __LINE__) +static OPUS_INLINE int SHL32_(opus_int64 a, int shift, char *file, int line) +{ + opus_int64 res; + if (!VERIFY_INT(a) || !VERIFY_SHORT(shift)) + { + fprintf (stderr, "SHL32: inputs are not int: %lld %d in %s: line %d\n", a, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a<<shift; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "SHL32: output is not int: %lld<<%d = %lld in %s: line %d\n", a, shift, res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=2; + return res; +} + +#define PSHR32(a,shift) (celt_mips--,SHR32(ADD32((a),(((opus_val32)(1)<<((shift))>>1))),shift)) +#define VSHR32(a, shift) (((shift)>0) ? SHR32(a, shift) : SHL32(a, -(shift))) + +#define ROUND16(x,a) (celt_mips--,EXTRACT16(PSHR32((x),(a)))) +#define HALF16(x) (SHR16(x,1)) +#define HALF32(x) (SHR32(x,1)) + +//#define SHR(a,shift) ((a) >> (shift)) +//#define SHL(a,shift) ((a) << (shift)) + +#define ADD16(a, b) ADD16_(a, b, __FILE__, __LINE__) +static OPUS_INLINE short ADD16_(int a, int b, char *file, int line) +{ + int res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "ADD16: inputs are not short: %d %d in %s: line %d\n", a, b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a+b; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "ADD16: output is not short: %d+%d=%d in %s: line %d\n", a,b,res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips++; + return res; +} + +#define SUB16(a, b) SUB16_(a, b, __FILE__, __LINE__) +static OPUS_INLINE short SUB16_(int a, int b, char *file, int line) +{ + int res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "SUB16: inputs are not short: %d %d in %s: line %d\n", a, b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a-b; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "SUB16: output is not short: %d in %s: line %d\n", res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips++; + return res; +} + +#define ADD32(a, b) ADD32_(a, b, __FILE__, __LINE__) +static OPUS_INLINE int ADD32_(opus_int64 a, opus_int64 b, char *file, int line) +{ + opus_int64 res; + if (!VERIFY_INT(a) || !VERIFY_INT(b)) + { + fprintf (stderr, "ADD32: inputs are not int: %d %d in %s: line %d\n", (int)a, (int)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a+b; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "ADD32: output is not int: %d in %s: line %d\n", (int)res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=2; + return res; +} + +#define SUB32(a, b) SUB32_(a, b, __FILE__, __LINE__) +static OPUS_INLINE int SUB32_(opus_int64 a, opus_int64 b, char *file, int line) +{ + opus_int64 res; + if (!VERIFY_INT(a) || !VERIFY_INT(b)) + { + fprintf (stderr, "SUB32: inputs are not int: %d %d in %s: line %d\n", (int)a, (int)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a-b; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "SUB32: output is not int: %d in %s: line %d\n", (int)res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=2; + return res; +} + +#undef UADD32 +#define UADD32(a, b) UADD32_(a, b, __FILE__, __LINE__) +static OPUS_INLINE unsigned int UADD32_(opus_uint64 a, opus_uint64 b, char *file, int line) +{ + opus_uint64 res; + if (!VERIFY_UINT(a) || !VERIFY_UINT(b)) + { + fprintf (stderr, "UADD32: inputs are not uint32: %llu %llu in %s: line %d\n", a, b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a+b; + if (!VERIFY_UINT(res)) + { + fprintf (stderr, "UADD32: output is not uint32: %llu in %s: line %d\n", res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=2; + return res; +} + +#undef USUB32 +#define USUB32(a, b) USUB32_(a, b, __FILE__, __LINE__) +static OPUS_INLINE unsigned int USUB32_(opus_uint64 a, opus_uint64 b, char *file, int line) +{ + opus_uint64 res; + if (!VERIFY_UINT(a) || !VERIFY_UINT(b)) + { + fprintf (stderr, "USUB32: inputs are not uint32: %llu %llu in %s: line %d\n", a, b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + if (a<b) + { + fprintf (stderr, "USUB32: inputs underflow: %llu < %llu in %s: line %d\n", a, b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a-b; + if (!VERIFY_UINT(res)) + { + fprintf (stderr, "USUB32: output is not uint32: %llu - %llu = %llu in %s: line %d\n", a, b, res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=2; + return res; +} + +/* result fits in 16 bits */ +static OPUS_INLINE short MULT16_16_16(int a, int b) +{ + int res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_16: inputs are not short: %d %d\n", a, b); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a*b; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "MULT16_16_16: output is not short: %d\n", res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips++; + return res; +} + +#define MULT16_16(a, b) MULT16_16_(a, b, __FILE__, __LINE__) +static OPUS_INLINE int MULT16_16_(int a, int b, char *file, int line) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16: inputs are not short: %d %d in %s: line %d\n", a, b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((opus_int64)a)*b; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "MULT16_16: output is not int: %d in %s: line %d\n", (int)res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips++; + return res; +} + +#define MAC16_16(c,a,b) (celt_mips-=2,ADD32((c),MULT16_16((a),(b)))) + +#define MULT16_32_QX(a, b, Q) MULT16_32_QX_(a, b, Q, __FILE__, __LINE__) +static OPUS_INLINE int MULT16_32_QX_(int a, opus_int64 b, int Q, char *file, int line) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_INT(b)) + { + fprintf (stderr, "MULT16_32_Q%d: inputs are not short+int: %d %d in %s: line %d\n", Q, (int)a, (int)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + if (ABS32(b)>=((opus_val32)(1)<<(15+Q))) + { + fprintf (stderr, "MULT16_32_Q%d: second operand too large: %d %d in %s: line %d\n", Q, (int)a, (int)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = (((opus_int64)a)*(opus_int64)b) >> Q; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "MULT16_32_Q%d: output is not int: %d*%d=%d in %s: line %d\n", Q, (int)a, (int)b,(int)res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + if (Q==15) + celt_mips+=3; + else + celt_mips+=4; + return res; +} + +#define MULT16_32_PX(a, b, Q) MULT16_32_PX_(a, b, Q, __FILE__, __LINE__) +static OPUS_INLINE int MULT16_32_PX_(int a, opus_int64 b, int Q, char *file, int line) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_INT(b)) + { + fprintf (stderr, "MULT16_32_P%d: inputs are not short+int: %d %d in %s: line %d\n\n", Q, (int)a, (int)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + if (ABS32(b)>=((opus_int64)(1)<<(15+Q))) + { + fprintf (stderr, "MULT16_32_Q%d: second operand too large: %d %d in %s: line %d\n\n", Q, (int)a, (int)b,file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((((opus_int64)a)*(opus_int64)b) + (((opus_val32)(1)<<Q)>>1))>> Q; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "MULT16_32_P%d: output is not int: %d*%d=%d in %s: line %d\n\n", Q, (int)a, (int)b,(int)res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + if (Q==15) + celt_mips+=4; + else + celt_mips+=5; + return res; +} + +#define MULT16_32_Q15(a,b) MULT16_32_QX(a,b,15) +#define MAC16_32_Q15(c,a,b) (celt_mips-=2,ADD32((c),MULT16_32_Q15((a),(b)))) + +static OPUS_INLINE int SATURATE(int a, int b) +{ + if (a>b) + a=b; + if (a<-b) + a = -b; + celt_mips+=3; + return a; +} + +static OPUS_INLINE opus_int16 SATURATE16(opus_int32 a) +{ + celt_mips+=3; + if (a>32767) + return 32767; + else if (a<-32768) + return -32768; + else return a; +} + +static OPUS_INLINE int MULT16_16_Q11_32(int a, int b) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_Q11: inputs are not short: %d %d\n", a, b); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((opus_int64)a)*b; + res >>= 11; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "MULT16_16_Q11: output is not short: %d*%d=%d\n", (int)a, (int)b, (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=3; + return res; +} +static OPUS_INLINE short MULT16_16_Q13(int a, int b) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_Q13: inputs are not short: %d %d\n", a, b); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((opus_int64)a)*b; + res >>= 13; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "MULT16_16_Q13: output is not short: %d*%d=%d\n", a, b, (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=3; + return res; +} +static OPUS_INLINE short MULT16_16_Q14(int a, int b) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_Q14: inputs are not short: %d %d\n", a, b); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((opus_int64)a)*b; + res >>= 14; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "MULT16_16_Q14: output is not short: %d\n", (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=3; + return res; +} + +#define MULT16_16_Q15(a, b) MULT16_16_Q15_(a, b, __FILE__, __LINE__) +static OPUS_INLINE short MULT16_16_Q15_(int a, int b, char *file, int line) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_Q15: inputs are not short: %d %d in %s: line %d\n", a, b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((opus_int64)a)*b; + res >>= 15; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "MULT16_16_Q15: output is not short: %d in %s: line %d\n", (int)res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=1; + return res; +} + +static OPUS_INLINE short MULT16_16_P13(int a, int b) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_P13: inputs are not short: %d %d\n", a, b); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((opus_int64)a)*b; + res += 4096; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "MULT16_16_P13: overflow: %d*%d=%d\n", a, b, (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res >>= 13; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "MULT16_16_P13: output is not short: %d*%d=%d\n", a, b, (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=4; + return res; +} +static OPUS_INLINE short MULT16_16_P14(int a, int b) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_P14: inputs are not short: %d %d\n", a, b); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((opus_int64)a)*b; + res += 8192; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "MULT16_16_P14: overflow: %d*%d=%d\n", a, b, (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res >>= 14; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "MULT16_16_P14: output is not short: %d*%d=%d\n", a, b, (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=4; + return res; +} +static OPUS_INLINE short MULT16_16_P15(int a, int b) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_P15: inputs are not short: %d %d\n", a, b); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((opus_int64)a)*b; + res += 16384; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "MULT16_16_P15: overflow: %d*%d=%d\n", a, b, (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res >>= 15; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "MULT16_16_P15: output is not short: %d*%d=%d\n", a, b, (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=2; + return res; +} + +#define DIV32_16(a, b) DIV32_16_(a, b, __FILE__, __LINE__) + +static OPUS_INLINE int DIV32_16_(opus_int64 a, opus_int64 b, char *file, int line) +{ + opus_int64 res; + if (b==0) + { + fprintf(stderr, "DIV32_16: divide by zero: %d/%d in %s: line %d\n", (int)a, (int)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + return 0; + } + if (!VERIFY_INT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "DIV32_16: inputs are not int/short: %d %d in %s: line %d\n", (int)a, (int)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a/b; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "DIV32_16: output is not short: %d / %d = %d in %s: line %d\n", (int)a,(int)b,(int)res, file, line); + if (res>32767) + res = 32767; + if (res<-32768) + res = -32768; +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=35; + return res; +} + +#define DIV32(a, b) DIV32_(a, b, __FILE__, __LINE__) +static OPUS_INLINE int DIV32_(opus_int64 a, opus_int64 b, char *file, int line) +{ + opus_int64 res; + if (b==0) + { + fprintf(stderr, "DIV32: divide by zero: %d/%d in %s: line %d\n", (int)a, (int)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + return 0; + } + + if (!VERIFY_INT(a) || !VERIFY_INT(b)) + { + fprintf (stderr, "DIV32: inputs are not int/short: %d %d in %s: line %d\n", (int)a, (int)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a/b; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "DIV32: output is not int: %d in %s: line %d\n", (int)res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=70; + return res; +} + +#undef PRINT_MIPS +#define PRINT_MIPS(file) do {fprintf (file, "total complexity = %llu MIPS\n", celt_mips);} while (0); + +#endif diff --git a/drivers/opus/celt/fixed_generic.h b/drivers/opus/celt/fixed_generic.h new file mode 100644 index 0000000000..ecf018a244 --- /dev/null +++ b/drivers/opus/celt/fixed_generic.h @@ -0,0 +1,134 @@ +/* Copyright (C) 2007-2009 Xiph.Org Foundation + Copyright (C) 2003-2008 Jean-Marc Valin + Copyright (C) 2007-2008 CSIRO */ +/** + @file fixed_generic.h + @brief Generic fixed-point operations +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef FIXED_GENERIC_H +#define FIXED_GENERIC_H + +/** Multiply a 16-bit signed value by a 16-bit unsigned value. The result is a 32-bit signed value */ +#define MULT16_16SU(a,b) ((opus_val32)(opus_val16)(a)*(opus_val32)(opus_uint16)(b)) + +/** 16x32 multiplication, followed by a 16-bit shift right. Results fits in 32 bits */ +#define MULT16_32_Q16(a,b) ADD32(MULT16_16((a),SHR((b),16)), SHR(MULT16_16SU((a),((b)&0x0000ffff)),16)) + +/** 16x32 multiplication, followed by a 16-bit shift right (round-to-nearest). Results fits in 32 bits */ +#define MULT16_32_P16(a,b) ADD32(MULT16_16((a),SHR((b),16)), PSHR(MULT16_16SU((a),((b)&0x0000ffff)),16)) + +/** 16x32 multiplication, followed by a 15-bit shift right. Results fits in 32 bits */ +#define MULT16_32_Q15(a,b) ADD32(SHL(MULT16_16((a),SHR((b),16)),1), SHR(MULT16_16SU((a),((b)&0x0000ffff)),15)) + +/** 32x32 multiplication, followed by a 31-bit shift right. Results fits in 32 bits */ +#define MULT32_32_Q31(a,b) ADD32(ADD32(SHL(MULT16_16(SHR((a),16),SHR((b),16)),1), SHR(MULT16_16SU(SHR((a),16),((b)&0x0000ffff)),15)), SHR(MULT16_16SU(SHR((b),16),((a)&0x0000ffff)),15)) + +/** Compile-time conversion of float constant to 16-bit value */ +#define QCONST16(x,bits) ((opus_val16)(.5+(x)*(((opus_val32)1)<<(bits)))) + +/** Compile-time conversion of float constant to 32-bit value */ +#define QCONST32(x,bits) ((opus_val32)(.5+(x)*(((opus_val32)1)<<(bits)))) + +/** Negate a 16-bit value */ +#define NEG16(x) (-(x)) +/** Negate a 32-bit value */ +#define NEG32(x) (-(x)) + +/** Change a 32-bit value into a 16-bit value. The value is assumed to fit in 16-bit, otherwise the result is undefined */ +#define EXTRACT16(x) ((opus_val16)(x)) +/** Change a 16-bit value into a 32-bit value */ +#define EXTEND32(x) ((opus_val32)(x)) + +/** Arithmetic shift-right of a 16-bit value */ +#define SHR16(a,shift) ((a) >> (shift)) +/** Arithmetic shift-left of a 16-bit value */ +#define SHL16(a,shift) ((opus_int16)((opus_uint16)(a)<<(shift))) +/** Arithmetic shift-right of a 32-bit value */ +#define SHR32(a,shift) ((a) >> (shift)) +/** Arithmetic shift-left of a 32-bit value */ +#define SHL32(a,shift) ((opus_int32)((opus_uint32)(a)<<(shift))) + +/** 32-bit arithmetic shift right with rounding-to-nearest instead of rounding down */ +#define PSHR32(a,shift) (SHR32((a)+((EXTEND32(1)<<((shift))>>1)),shift)) +/** 32-bit arithmetic shift right where the argument can be negative */ +#define VSHR32(a, shift) (((shift)>0) ? SHR32(a, shift) : SHL32(a, -(shift))) + +/** "RAW" macros, should not be used outside of this header file */ +#define SHR(a,shift) ((a) >> (shift)) +#define SHL(a,shift) SHL32(a,shift) +#define PSHR(a,shift) (SHR((a)+((EXTEND32(1)<<((shift))>>1)),shift)) +#define SATURATE(x,a) (((x)>(a) ? (a) : (x)<-(a) ? -(a) : (x))) + +#define SATURATE16(x) (EXTRACT16((x)>32767 ? 32767 : (x)<-32768 ? -32768 : (x))) + +/** Shift by a and round-to-neareast 32-bit value. Result is a 16-bit value */ +#define ROUND16(x,a) (EXTRACT16(PSHR32((x),(a)))) +/** Divide by two */ +#define HALF16(x) (SHR16(x,1)) +#define HALF32(x) (SHR32(x,1)) + +/** Add two 16-bit values */ +#define ADD16(a,b) ((opus_val16)((opus_val16)(a)+(opus_val16)(b))) +/** Subtract two 16-bit values */ +#define SUB16(a,b) ((opus_val16)(a)-(opus_val16)(b)) +/** Add two 32-bit values */ +#define ADD32(a,b) ((opus_val32)(a)+(opus_val32)(b)) +/** Subtract two 32-bit values */ +#define SUB32(a,b) ((opus_val32)(a)-(opus_val32)(b)) + +/** 16x16 multiplication where the result fits in 16 bits */ +#define MULT16_16_16(a,b) ((((opus_val16)(a))*((opus_val16)(b)))) + +/* (opus_val32)(opus_val16) gives TI compiler a hint that it's 16x16->32 multiply */ +/** 16x16 multiplication where the result fits in 32 bits */ +#define MULT16_16(a,b) (((opus_val32)(opus_val16)(a))*((opus_val32)(opus_val16)(b))) + +/** 16x16 multiply-add where the result fits in 32 bits */ +#define MAC16_16(c,a,b) (ADD32((c),MULT16_16((a),(b)))) +/** 16x32 multiply, followed by a 15-bit shift right and 32-bit add. + b must fit in 31 bits. + Result fits in 32 bits. */ +#define MAC16_32_Q15(c,a,b) ADD32(c,ADD32(MULT16_16((a),SHR((b),15)), SHR(MULT16_16((a),((b)&0x00007fff)),15))) + +#define MULT16_16_Q11_32(a,b) (SHR(MULT16_16((a),(b)),11)) +#define MULT16_16_Q11(a,b) (SHR(MULT16_16((a),(b)),11)) +#define MULT16_16_Q13(a,b) (SHR(MULT16_16((a),(b)),13)) +#define MULT16_16_Q14(a,b) (SHR(MULT16_16((a),(b)),14)) +#define MULT16_16_Q15(a,b) (SHR(MULT16_16((a),(b)),15)) + +#define MULT16_16_P13(a,b) (SHR(ADD32(4096,MULT16_16((a),(b))),13)) +#define MULT16_16_P14(a,b) (SHR(ADD32(8192,MULT16_16((a),(b))),14)) +#define MULT16_16_P15(a,b) (SHR(ADD32(16384,MULT16_16((a),(b))),15)) + +/** Divide a 32-bit value by a 16-bit value. Result fits in 16 bits */ +#define DIV32_16(a,b) ((opus_val16)(((opus_val32)(a))/((opus_val16)(b)))) + +/** Divide a 32-bit value by a 32-bit value. Result fits in 32 bits */ +#define DIV32(a,b) (((opus_val32)(a))/((opus_val32)(b))) + +#endif diff --git a/drivers/opus/celt/float_cast.h b/drivers/opus/celt/float_cast.h new file mode 100644 index 0000000000..ede6574860 --- /dev/null +++ b/drivers/opus/celt/float_cast.h @@ -0,0 +1,140 @@ +/* Copyright (C) 2001 Erik de Castro Lopo <erikd AT mega-nerd DOT com> */ +/* + 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. +*/ + +/* Version 1.1 */ + +#ifndef FLOAT_CAST_H +#define FLOAT_CAST_H + + +#include "arch.h" + +/*============================================================================ +** On Intel Pentium processors (especially PIII and probably P4), converting +** from float to int is very slow. To meet the C specs, the code produced by +** most C compilers targeting Pentium needs to change the FPU rounding mode +** before the float to int conversion is performed. +** +** Changing the FPU rounding mode causes the FPU pipeline to be flushed. It +** is this flushing of the pipeline which is so slow. +** +** Fortunately the ISO C99 specifications define the functions lrint, lrintf, +** llrint and llrintf which fix this problem as a side effect. +** +** On Unix-like systems, the configure process should have detected the +** presence of these functions. If they weren't found we have to replace them +** here with a standard C cast. +*/ + +/* +** The C99 prototypes for lrint and lrintf are as follows: +** +** long int lrintf (float x) ; +** long int lrint (double x) ; +*/ + +/* The presence of the required functions are detected during the configure +** process and the values HAVE_LRINT and HAVE_LRINTF are set accordingly in +** the config.h file. +*/ + +#if (HAVE_LRINTF) + +/* These defines enable functionality introduced with the 1999 ISO C +** standard. They must be defined before the inclusion of math.h to +** engage them. If optimisation is enabled, these functions will be +** inlined. With optimisation switched off, you have to link in the +** maths library using -lm. +*/ + +#define _ISOC9X_SOURCE 1 +#define _ISOC99_SOURCE 1 + +#define __USE_ISOC9X 1 +#define __USE_ISOC99 1 + +#include <math.h> +#define float2int(x) lrintf(x) + +#elif (defined(HAVE_LRINT)) + +#define _ISOC9X_SOURCE 1 +#define _ISOC99_SOURCE 1 + +#define __USE_ISOC9X 1 +#define __USE_ISOC99 1 + +#include <math.h> +#define float2int(x) lrint(x) + +#elif (defined(_MSC_VER) && _MSC_VER >= 1400) && (defined (WIN64) || defined (_WIN64)) + #include <xmmintrin.h> + + __inline long int float2int(float value) + { + return _mm_cvtss_si32(_mm_load_ss(&value)); + } +#elif (defined(_MSC_VER) && _MSC_VER >= 1400) && (defined (WIN32) || defined (_WIN32)) + #include <math.h> + + /* Win32 doesn't seem to have these functions. + ** Therefore implement OPUS_INLINE versions of these functions here. + */ + + __inline long int + float2int (float flt) + { int intgr; + + _asm + { fld flt + fistp intgr + } ; + + return intgr ; + } + +#else + +#if (defined(__GNUC__) && defined(__STDC__) && __STDC__ && __STDC_VERSION__ >= 199901L) + /* supported by gcc in C99 mode, but not by all other compilers */ + #warning "Don't have the functions lrint() and lrintf ()." + #warning "Replacing these functions with a standard C cast." +#endif /* __STDC_VERSION__ >= 199901L */ + #include <math.h> + #define float2int(flt) ((int)(floor(.5+flt))) +#endif + +#ifndef DISABLE_FLOAT_API +static OPUS_INLINE opus_int16 FLOAT2INT16(float x) +{ + x = x*CELT_SIG_SCALE; + x = MAX32(x, -32768); + x = MIN32(x, 32767); + return (opus_int16)float2int(x); +} +#endif /* DISABLE_FLOAT_API */ + +#endif /* FLOAT_CAST_H */ diff --git a/drivers/opus/celt/kiss_fft.c b/drivers/opus/celt/kiss_fft.c new file mode 100644 index 0000000000..333be975d1 --- /dev/null +++ b/drivers/opus/celt/kiss_fft.c @@ -0,0 +1,719 @@ +/*Copyright (c) 2003-2004, Mark Borgerding + Lots of modifications by Jean-Marc Valin + Copyright (c) 2005-2007, Xiph.Org Foundation + Copyright (c) 2008, Xiph.Org Foundation, CSIRO + + All rights reserved. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are met: + + * Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above copyright notice, + this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" + AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE + LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + POSSIBILITY OF SUCH DAMAGE.*/ + +/* This code is originally from Mark Borgerding's KISS-FFT but has been + heavily modified to better suit Opus */ + +#ifndef SKIP_CONFIG_H +# ifdef OPUS_HAVE_CONFIG_H +#include "opus_config.h" +# endif +#endif + +#include "_kiss_fft_guts.h" +#include "arch.h" +#include "os_support.h" +#include "mathops.h" +#include "stack_alloc.h" + +/* The guts header contains all the multiplication and addition macros that are defined for + complex numbers. It also delares the kf_ internal functions. +*/ + +static void kf_bfly2( + kiss_fft_cpx * Fout, + const size_t fstride, + const kiss_fft_state *st, + int m, + int N, + int mm + ) +{ + kiss_fft_cpx * Fout2; + const kiss_twiddle_cpx * tw1; + int i,j; + kiss_fft_cpx * Fout_beg = Fout; + for (i=0;i<N;i++) + { + Fout = Fout_beg + i*mm; + Fout2 = Fout + m; + tw1 = st->twiddles; + for(j=0;j<m;j++) + { + kiss_fft_cpx t; + Fout->r = SHR32(Fout->r, 1);Fout->i = SHR32(Fout->i, 1); + Fout2->r = SHR32(Fout2->r, 1);Fout2->i = SHR32(Fout2->i, 1); + C_MUL (t, *Fout2 , *tw1); + tw1 += fstride; + C_SUB( *Fout2 , *Fout , t ); + C_ADDTO( *Fout , t ); + ++Fout2; + ++Fout; + } + } +} + +static void ki_bfly2( + kiss_fft_cpx * Fout, + const size_t fstride, + const kiss_fft_state *st, + int m, + int N, + int mm + ) +{ + kiss_fft_cpx * Fout2; + const kiss_twiddle_cpx * tw1; + kiss_fft_cpx t; + int i,j; + kiss_fft_cpx * Fout_beg = Fout; + for (i=0;i<N;i++) + { + Fout = Fout_beg + i*mm; + Fout2 = Fout + m; + tw1 = st->twiddles; + for(j=0;j<m;j++) + { + C_MULC (t, *Fout2 , *tw1); + tw1 += fstride; + C_SUB( *Fout2 , *Fout , t ); + C_ADDTO( *Fout , t ); + ++Fout2; + ++Fout; + } + } +} + +static void kf_bfly4( + kiss_fft_cpx * Fout, + const size_t fstride, + const kiss_fft_state *st, + int m, + int N, + int mm + ) +{ + const kiss_twiddle_cpx *tw1,*tw2,*tw3; + kiss_fft_cpx scratch[6]; + const size_t m2=2*m; + const size_t m3=3*m; + int i, j; + + kiss_fft_cpx * Fout_beg = Fout; + for (i=0;i<N;i++) + { + Fout = Fout_beg + i*mm; + tw3 = tw2 = tw1 = st->twiddles; + for (j=0;j<m;j++) + { + C_MUL4(scratch[0],Fout[m] , *tw1 ); + C_MUL4(scratch[1],Fout[m2] , *tw2 ); + C_MUL4(scratch[2],Fout[m3] , *tw3 ); + + Fout->r = PSHR32(Fout->r, 2); + Fout->i = PSHR32(Fout->i, 2); + C_SUB( scratch[5] , *Fout, scratch[1] ); + C_ADDTO(*Fout, scratch[1]); + C_ADD( scratch[3] , scratch[0] , scratch[2] ); + C_SUB( scratch[4] , scratch[0] , scratch[2] ); + C_SUB( Fout[m2], *Fout, scratch[3] ); + tw1 += fstride; + tw2 += fstride*2; + tw3 += fstride*3; + C_ADDTO( *Fout , scratch[3] ); + + Fout[m].r = scratch[5].r + scratch[4].i; + Fout[m].i = scratch[5].i - scratch[4].r; + Fout[m3].r = scratch[5].r - scratch[4].i; + Fout[m3].i = scratch[5].i + scratch[4].r; + ++Fout; + } + } +} + +static void ki_bfly4( + kiss_fft_cpx * Fout, + const size_t fstride, + const kiss_fft_state *st, + int m, + int N, + int mm + ) +{ + const kiss_twiddle_cpx *tw1,*tw2,*tw3; + kiss_fft_cpx scratch[6]; + const size_t m2=2*m; + const size_t m3=3*m; + int i, j; + + kiss_fft_cpx * Fout_beg = Fout; + for (i=0;i<N;i++) + { + Fout = Fout_beg + i*mm; + tw3 = tw2 = tw1 = st->twiddles; + for (j=0;j<m;j++) + { + C_MULC(scratch[0],Fout[m] , *tw1 ); + C_MULC(scratch[1],Fout[m2] , *tw2 ); + C_MULC(scratch[2],Fout[m3] , *tw3 ); + + C_SUB( scratch[5] , *Fout, scratch[1] ); + C_ADDTO(*Fout, scratch[1]); + C_ADD( scratch[3] , scratch[0] , scratch[2] ); + C_SUB( scratch[4] , scratch[0] , scratch[2] ); + C_SUB( Fout[m2], *Fout, scratch[3] ); + tw1 += fstride; + tw2 += fstride*2; + tw3 += fstride*3; + C_ADDTO( *Fout , scratch[3] ); + + Fout[m].r = scratch[5].r - scratch[4].i; + Fout[m].i = scratch[5].i + scratch[4].r; + Fout[m3].r = scratch[5].r + scratch[4].i; + Fout[m3].i = scratch[5].i - scratch[4].r; + ++Fout; + } + } +} + +#ifndef RADIX_TWO_ONLY + +static void kf_bfly3( + kiss_fft_cpx * Fout, + const size_t fstride, + const kiss_fft_state *st, + int m, + int N, + int mm + ) +{ + int i; + size_t k; + const size_t m2 = 2*m; + const kiss_twiddle_cpx *tw1,*tw2; + kiss_fft_cpx scratch[5]; + kiss_twiddle_cpx epi3; + + kiss_fft_cpx * Fout_beg = Fout; + epi3 = st->twiddles[fstride*m]; + for (i=0;i<N;i++) + { + Fout = Fout_beg + i*mm; + tw1=tw2=st->twiddles; + k=m; + do { + C_FIXDIV(*Fout,3); C_FIXDIV(Fout[m],3); C_FIXDIV(Fout[m2],3); + + C_MUL(scratch[1],Fout[m] , *tw1); + C_MUL(scratch[2],Fout[m2] , *tw2); + + C_ADD(scratch[3],scratch[1],scratch[2]); + C_SUB(scratch[0],scratch[1],scratch[2]); + tw1 += fstride; + tw2 += fstride*2; + + Fout[m].r = Fout->r - HALF_OF(scratch[3].r); + Fout[m].i = Fout->i - HALF_OF(scratch[3].i); + + C_MULBYSCALAR( scratch[0] , epi3.i ); + + C_ADDTO(*Fout,scratch[3]); + + Fout[m2].r = Fout[m].r + scratch[0].i; + Fout[m2].i = Fout[m].i - scratch[0].r; + + Fout[m].r -= scratch[0].i; + Fout[m].i += scratch[0].r; + + ++Fout; + } while(--k); + } +} + +static void ki_bfly3( + kiss_fft_cpx * Fout, + const size_t fstride, + const kiss_fft_state *st, + int m, + int N, + int mm + ) +{ + int i, k; + const size_t m2 = 2*m; + const kiss_twiddle_cpx *tw1,*tw2; + kiss_fft_cpx scratch[5]; + kiss_twiddle_cpx epi3; + + kiss_fft_cpx * Fout_beg = Fout; + epi3 = st->twiddles[fstride*m]; + for (i=0;i<N;i++) + { + Fout = Fout_beg + i*mm; + tw1=tw2=st->twiddles; + k=m; + do{ + + C_MULC(scratch[1],Fout[m] , *tw1); + C_MULC(scratch[2],Fout[m2] , *tw2); + + C_ADD(scratch[3],scratch[1],scratch[2]); + C_SUB(scratch[0],scratch[1],scratch[2]); + tw1 += fstride; + tw2 += fstride*2; + + Fout[m].r = Fout->r - HALF_OF(scratch[3].r); + Fout[m].i = Fout->i - HALF_OF(scratch[3].i); + + C_MULBYSCALAR( scratch[0] , -epi3.i ); + + C_ADDTO(*Fout,scratch[3]); + + Fout[m2].r = Fout[m].r + scratch[0].i; + Fout[m2].i = Fout[m].i - scratch[0].r; + + Fout[m].r -= scratch[0].i; + Fout[m].i += scratch[0].r; + + ++Fout; + }while(--k); + } +} + +static void kf_bfly5( + kiss_fft_cpx * Fout, + const size_t fstride, + const kiss_fft_state *st, + int m, + int N, + int mm + ) +{ + kiss_fft_cpx *Fout0,*Fout1,*Fout2,*Fout3,*Fout4; + int i, u; + kiss_fft_cpx scratch[13]; + const kiss_twiddle_cpx * twiddles = st->twiddles; + const kiss_twiddle_cpx *tw; + kiss_twiddle_cpx ya,yb; + kiss_fft_cpx * Fout_beg = Fout; + + ya = twiddles[fstride*m]; + yb = twiddles[fstride*2*m]; + tw=st->twiddles; + + for (i=0;i<N;i++) + { + Fout = Fout_beg + i*mm; + Fout0=Fout; + Fout1=Fout0+m; + Fout2=Fout0+2*m; + Fout3=Fout0+3*m; + Fout4=Fout0+4*m; + + for ( u=0; u<m; ++u ) { + C_FIXDIV( *Fout0,5); C_FIXDIV( *Fout1,5); C_FIXDIV( *Fout2,5); C_FIXDIV( *Fout3,5); C_FIXDIV( *Fout4,5); + scratch[0] = *Fout0; + + C_MUL(scratch[1] ,*Fout1, tw[u*fstride]); + C_MUL(scratch[2] ,*Fout2, tw[2*u*fstride]); + C_MUL(scratch[3] ,*Fout3, tw[3*u*fstride]); + C_MUL(scratch[4] ,*Fout4, tw[4*u*fstride]); + + C_ADD( scratch[7],scratch[1],scratch[4]); + C_SUB( scratch[10],scratch[1],scratch[4]); + C_ADD( scratch[8],scratch[2],scratch[3]); + C_SUB( scratch[9],scratch[2],scratch[3]); + + Fout0->r += scratch[7].r + scratch[8].r; + Fout0->i += scratch[7].i + scratch[8].i; + + scratch[5].r = scratch[0].r + S_MUL(scratch[7].r,ya.r) + S_MUL(scratch[8].r,yb.r); + scratch[5].i = scratch[0].i + S_MUL(scratch[7].i,ya.r) + S_MUL(scratch[8].i,yb.r); + + scratch[6].r = S_MUL(scratch[10].i,ya.i) + S_MUL(scratch[9].i,yb.i); + scratch[6].i = -S_MUL(scratch[10].r,ya.i) - S_MUL(scratch[9].r,yb.i); + + C_SUB(*Fout1,scratch[5],scratch[6]); + C_ADD(*Fout4,scratch[5],scratch[6]); + + scratch[11].r = scratch[0].r + S_MUL(scratch[7].r,yb.r) + S_MUL(scratch[8].r,ya.r); + scratch[11].i = scratch[0].i + S_MUL(scratch[7].i,yb.r) + S_MUL(scratch[8].i,ya.r); + scratch[12].r = - S_MUL(scratch[10].i,yb.i) + S_MUL(scratch[9].i,ya.i); + scratch[12].i = S_MUL(scratch[10].r,yb.i) - S_MUL(scratch[9].r,ya.i); + + C_ADD(*Fout2,scratch[11],scratch[12]); + C_SUB(*Fout3,scratch[11],scratch[12]); + + ++Fout0;++Fout1;++Fout2;++Fout3;++Fout4; + } + } +} + +static void ki_bfly5( + kiss_fft_cpx * Fout, + const size_t fstride, + const kiss_fft_state *st, + int m, + int N, + int mm + ) +{ + kiss_fft_cpx *Fout0,*Fout1,*Fout2,*Fout3,*Fout4; + int i, u; + kiss_fft_cpx scratch[13]; + const kiss_twiddle_cpx * twiddles = st->twiddles; + const kiss_twiddle_cpx *tw; + kiss_twiddle_cpx ya,yb; + kiss_fft_cpx * Fout_beg = Fout; + + ya = twiddles[fstride*m]; + yb = twiddles[fstride*2*m]; + tw=st->twiddles; + + for (i=0;i<N;i++) + { + Fout = Fout_beg + i*mm; + Fout0=Fout; + Fout1=Fout0+m; + Fout2=Fout0+2*m; + Fout3=Fout0+3*m; + Fout4=Fout0+4*m; + + for ( u=0; u<m; ++u ) { + scratch[0] = *Fout0; + + C_MULC(scratch[1] ,*Fout1, tw[u*fstride]); + C_MULC(scratch[2] ,*Fout2, tw[2*u*fstride]); + C_MULC(scratch[3] ,*Fout3, tw[3*u*fstride]); + C_MULC(scratch[4] ,*Fout4, tw[4*u*fstride]); + + C_ADD( scratch[7],scratch[1],scratch[4]); + C_SUB( scratch[10],scratch[1],scratch[4]); + C_ADD( scratch[8],scratch[2],scratch[3]); + C_SUB( scratch[9],scratch[2],scratch[3]); + + Fout0->r += scratch[7].r + scratch[8].r; + Fout0->i += scratch[7].i + scratch[8].i; + + scratch[5].r = scratch[0].r + S_MUL(scratch[7].r,ya.r) + S_MUL(scratch[8].r,yb.r); + scratch[5].i = scratch[0].i + S_MUL(scratch[7].i,ya.r) + S_MUL(scratch[8].i,yb.r); + + scratch[6].r = -S_MUL(scratch[10].i,ya.i) - S_MUL(scratch[9].i,yb.i); + scratch[6].i = S_MUL(scratch[10].r,ya.i) + S_MUL(scratch[9].r,yb.i); + + C_SUB(*Fout1,scratch[5],scratch[6]); + C_ADD(*Fout4,scratch[5],scratch[6]); + + scratch[11].r = scratch[0].r + S_MUL(scratch[7].r,yb.r) + S_MUL(scratch[8].r,ya.r); + scratch[11].i = scratch[0].i + S_MUL(scratch[7].i,yb.r) + S_MUL(scratch[8].i,ya.r); + scratch[12].r = S_MUL(scratch[10].i,yb.i) - S_MUL(scratch[9].i,ya.i); + scratch[12].i = -S_MUL(scratch[10].r,yb.i) + S_MUL(scratch[9].r,ya.i); + + C_ADD(*Fout2,scratch[11],scratch[12]); + C_SUB(*Fout3,scratch[11],scratch[12]); + + ++Fout0;++Fout1;++Fout2;++Fout3;++Fout4; + } + } +} + +#endif + + +#ifdef CUSTOM_MODES + +static +void compute_bitrev_table( + int Fout, + opus_int16 *f, + const size_t fstride, + int in_stride, + opus_int16 * factors, + const kiss_fft_state *st + ) +{ + const int p=*factors++; /* the radix */ + const int m=*factors++; /* stage's fft length/p */ + + /*printf ("fft %d %d %d %d %d %d\n", p*m, m, p, s2, fstride*in_stride, N);*/ + if (m==1) + { + int j; + for (j=0;j<p;j++) + { + *f = Fout+j; + f += fstride*in_stride; + } + } else { + int j; + for (j=0;j<p;j++) + { + compute_bitrev_table( Fout , f, fstride*p, in_stride, factors,st); + f += fstride*in_stride; + Fout += m; + } + } +} + +/* facbuf is populated by p1,m1,p2,m2, ... + where + p[i] * m[i] = m[i-1] + m0 = n */ +static +int kf_factor(int n,opus_int16 * facbuf) +{ + int p=4; + + /*factor out powers of 4, powers of 2, then any remaining primes */ + do { + while (n % p) { + switch (p) { + case 4: p = 2; break; + case 2: p = 3; break; + default: p += 2; break; + } + if (p>32000 || (opus_int32)p*(opus_int32)p > n) + p = n; /* no more factors, skip to end */ + } + n /= p; +#ifdef RADIX_TWO_ONLY + if (p!=2 && p != 4) +#else + if (p>5) +#endif + { + return 0; + } + *facbuf++ = p; + *facbuf++ = n; + } while (n > 1); + return 1; +} + +static void compute_twiddles(kiss_twiddle_cpx *twiddles, int nfft) +{ + int i; +#ifdef OPUS_FIXED_POINT + for (i=0;i<nfft;++i) { + opus_val32 phase = -i; + kf_cexp2(twiddles+i, DIV32(SHL32(phase,17),nfft)); + } +#else + for (i=0;i<nfft;++i) { + const double pi=3.14159265358979323846264338327; + double phase = ( -2*pi /nfft ) * i; + kf_cexp(twiddles+i, phase ); + } +#endif +} + +/* + * + * Allocates all necessary storage space for the fft and ifft. + * The return value is a contiguous block of memory. As such, + * It can be freed with free(). + * */ +kiss_fft_state *opus_fft_alloc_twiddles(int nfft,void * mem,size_t * lenmem, const kiss_fft_state *base) +{ + kiss_fft_state *st=NULL; + size_t memneeded = sizeof(struct kiss_fft_state); /* twiddle factors*/ + + if ( lenmem==NULL ) { + st = ( kiss_fft_state*)KISS_FFT_MALLOC( memneeded ); + }else{ + if (mem != NULL && *lenmem >= memneeded) + st = (kiss_fft_state*)mem; + *lenmem = memneeded; + } + if (st) { + opus_int16 *bitrev; + kiss_twiddle_cpx *twiddles; + + st->nfft=nfft; +#ifndef OPUS_FIXED_POINT + st->scale = 1.f/nfft; +#endif + if (base != NULL) + { + st->twiddles = base->twiddles; + st->shift = 0; + while (nfft<<st->shift != base->nfft && st->shift < 32) + st->shift++; + if (st->shift>=32) + goto fail; + } else { + st->twiddles = twiddles = (kiss_twiddle_cpx*)KISS_FFT_MALLOC(sizeof(kiss_twiddle_cpx)*nfft); + compute_twiddles(twiddles, nfft); + st->shift = -1; + } + if (!kf_factor(nfft,st->factors)) + { + goto fail; + } + + /* bitrev */ + st->bitrev = bitrev = (opus_int16*)KISS_FFT_MALLOC(sizeof(opus_int16)*nfft); + if (st->bitrev==NULL) + goto fail; + compute_bitrev_table(0, bitrev, 1,1, st->factors,st); + } + return st; +fail: + opus_fft_free(st); + return NULL; +} + +kiss_fft_state *opus_fft_alloc(int nfft,void * mem,size_t * lenmem ) +{ + return opus_fft_alloc_twiddles(nfft, mem, lenmem, NULL); +} + +void opus_fft_free(const kiss_fft_state *cfg) +{ + if (cfg) + { + opus_free((opus_int16*)cfg->bitrev); + if (cfg->shift < 0) + opus_free((kiss_twiddle_cpx*)cfg->twiddles); + opus_free((kiss_fft_state*)cfg); + } +} + +#endif /* CUSTOM_MODES */ + +void opus_fft(const kiss_fft_state *st,const kiss_fft_cpx *fin,kiss_fft_cpx *fout) +{ + int m2, m; + int p; + int L; + int fstride[MAXFACTORS]; + int i; + int shift; + + /* st->shift can be -1 */ + shift = st->shift>0 ? st->shift : 0; + + celt_assert2 (fin != fout, "In-place FFT not supported"); + /* Bit-reverse the input */ + for (i=0;i<st->nfft;i++) + { + fout[st->bitrev[i]] = fin[i]; +#ifndef OPUS_FIXED_POINT + fout[st->bitrev[i]].r *= st->scale; + fout[st->bitrev[i]].i *= st->scale; +#endif + } + + fstride[0] = 1; + L=0; + do { + p = st->factors[2*L]; + m = st->factors[2*L+1]; + fstride[L+1] = fstride[L]*p; + L++; + } while(m!=1); + m = st->factors[2*L-1]; + for (i=L-1;i>=0;i--) + { + if (i!=0) + m2 = st->factors[2*i-1]; + else + m2 = 1; + switch (st->factors[2*i]) + { + case 2: + kf_bfly2(fout,fstride[i]<<shift,st,m, fstride[i], m2); + break; + case 4: + kf_bfly4(fout,fstride[i]<<shift,st,m, fstride[i], m2); + break; + #ifndef RADIX_TWO_ONLY + case 3: + kf_bfly3(fout,fstride[i]<<shift,st,m, fstride[i], m2); + break; + case 5: + kf_bfly5(fout,fstride[i]<<shift,st,m, fstride[i], m2); + break; + #endif + } + m = m2; + } +} + +void opus_ifft(const kiss_fft_state *st,const kiss_fft_cpx *fin,kiss_fft_cpx *fout) +{ + int m2, m; + int p; + int L; + int fstride[MAXFACTORS]; + int i; + int shift; + + /* st->shift can be -1 */ + shift = st->shift>0 ? st->shift : 0; + celt_assert2 (fin != fout, "In-place FFT not supported"); + /* Bit-reverse the input */ + for (i=0;i<st->nfft;i++) + fout[st->bitrev[i]] = fin[i]; + + fstride[0] = 1; + L=0; + do { + p = st->factors[2*L]; + m = st->factors[2*L+1]; + fstride[L+1] = fstride[L]*p; + L++; + } while(m!=1); + m = st->factors[2*L-1]; + for (i=L-1;i>=0;i--) + { + if (i!=0) + m2 = st->factors[2*i-1]; + else + m2 = 1; + switch (st->factors[2*i]) + { + case 2: + ki_bfly2(fout,fstride[i]<<shift,st,m, fstride[i], m2); + break; + case 4: + ki_bfly4(fout,fstride[i]<<shift,st,m, fstride[i], m2); + break; +#ifndef RADIX_TWO_ONLY + case 3: + ki_bfly3(fout,fstride[i]<<shift,st,m, fstride[i], m2); + break; + case 5: + ki_bfly5(fout,fstride[i]<<shift,st,m, fstride[i], m2); + break; +#endif + } + m = m2; + } +} + diff --git a/drivers/opus/celt/kiss_fft.h b/drivers/opus/celt/kiss_fft.h new file mode 100644 index 0000000000..aa22b3a419 --- /dev/null +++ b/drivers/opus/celt/kiss_fft.h @@ -0,0 +1,139 @@ +/*Copyright (c) 2003-2004, Mark Borgerding + Lots of modifications by Jean-Marc Valin + Copyright (c) 2005-2007, Xiph.Org Foundation + Copyright (c) 2008, Xiph.Org Foundation, CSIRO + + All rights reserved. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are met: + + * Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above copyright notice, + this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" + AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE + LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + POSSIBILITY OF SUCH DAMAGE.*/ + +#ifndef KISS_FFT_H +#define KISS_FFT_H + +#include <stdlib.h> +#include <math.h> +#include "arch.h" + +#ifdef __cplusplus +extern "C" { +#endif + +#ifdef USE_SIMD +# include <xmmintrin.h> +# define kiss_fft_scalar __m128 +#define KISS_FFT_MALLOC(nbytes) memalign(16,nbytes) +#else +#define KISS_FFT_MALLOC opus_alloc +#endif + +#ifdef OPUS_FIXED_POINT +#include "arch.h" + +# define kiss_fft_scalar opus_int32 +# define kiss_twiddle_scalar opus_int16 + + +#else +# ifndef kiss_fft_scalar +/* default is float */ +# define kiss_fft_scalar float +# define kiss_twiddle_scalar float +# define KF_SUFFIX _celt_single +# endif +#endif + +typedef struct { + kiss_fft_scalar r; + kiss_fft_scalar i; +}kiss_fft_cpx; + +typedef struct { + kiss_twiddle_scalar r; + kiss_twiddle_scalar i; +}kiss_twiddle_cpx; + +#define MAXFACTORS 8 +/* e.g. an fft of length 128 has 4 factors + as far as kissfft is concerned + 4*4*4*2 + */ + +typedef struct kiss_fft_state{ + int nfft; +#ifndef OPUS_FIXED_POINT + kiss_fft_scalar scale; +#endif + int shift; + opus_int16 factors[2*MAXFACTORS]; + const opus_int16 *bitrev; + const kiss_twiddle_cpx *twiddles; +} kiss_fft_state; + +/*typedef struct kiss_fft_state* kiss_fft_cfg;*/ + +/** + * opus_fft_alloc + * + * Initialize a FFT (or IFFT) algorithm's cfg/state buffer. + * + * typical usage: kiss_fft_cfg mycfg=opus_fft_alloc(1024,0,NULL,NULL); + * + * The return value from fft_alloc is a cfg buffer used internally + * by the fft routine or NULL. + * + * If lenmem is NULL, then opus_fft_alloc will allocate a cfg buffer using malloc. + * The returned value should be free()d when done to avoid memory leaks. + * + * The state can be placed in a user supplied buffer 'mem': + * If lenmem is not NULL and mem is not NULL and *lenmem is large enough, + * then the function places the cfg in mem and the size used in *lenmem + * and returns mem. + * + * If lenmem is not NULL and ( mem is NULL or *lenmem is not large enough), + * then the function returns NULL and places the minimum cfg + * buffer size in *lenmem. + * */ + +kiss_fft_state *opus_fft_alloc_twiddles(int nfft,void * mem,size_t * lenmem, const kiss_fft_state *base); + +kiss_fft_state *opus_fft_alloc(int nfft,void * mem,size_t * lenmem); + +/** + * opus_fft(cfg,in_out_buf) + * + * Perform an FFT on a complex input buffer. + * for a forward FFT, + * fin should be f[0] , f[1] , ... ,f[nfft-1] + * fout will be F[0] , F[1] , ... ,F[nfft-1] + * Note that each element is complex and can be accessed like + f[k].r and f[k].i + * */ +void opus_fft(const kiss_fft_state *cfg,const kiss_fft_cpx *fin,kiss_fft_cpx *fout); +void opus_ifft(const kiss_fft_state *cfg,const kiss_fft_cpx *fin,kiss_fft_cpx *fout); + +void opus_fft_free(const kiss_fft_state *cfg); + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/drivers/opus/celt/laplace.c b/drivers/opus/celt/laplace.c new file mode 100644 index 0000000000..c6d293f298 --- /dev/null +++ b/drivers/opus/celt/laplace.c @@ -0,0 +1,134 @@ +/* Copyright (c) 2007 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef OPUS_HAVE_CONFIG_H +#include "opus_config.h" +#endif + +#include "laplace.h" +#include "mathops.h" + +/* The minimum probability of an energy delta (out of 32768). */ +#define LAPLACE_LOG_MINP (0) +#define LAPLACE_MINP (1<<LAPLACE_LOG_MINP) +/* The minimum number of guaranteed representable energy deltas (in one + direction). */ +#define LAPLACE_NMIN (16) + +/* When called, decay is positive and at most 11456. */ +static unsigned ec_laplace_get_freq1(unsigned fs0, int decay) +{ + unsigned ft; + ft = 32768 - LAPLACE_MINP*(2*LAPLACE_NMIN) - fs0; + return ft*(opus_int32)(16384-decay)>>15; +} + +void ec_laplace_encode(ec_enc *enc, int *value, unsigned fs, int decay) +{ + unsigned fl; + int val = *value; + fl = 0; + if (val) + { + int s; + int i; + s = -(val<0); + val = (val+s)^s; + fl = fs; + fs = ec_laplace_get_freq1(fs, decay); + /* Search the decaying part of the PDF.*/ + for (i=1; fs > 0 && i < val; i++) + { + fs *= 2; + fl += fs+2*LAPLACE_MINP; + fs = (fs*(opus_int32)decay)>>15; + } + /* Everything beyond that has probability LAPLACE_MINP. */ + if (!fs) + { + int di; + int ndi_max; + ndi_max = (32768-fl+LAPLACE_MINP-1)>>LAPLACE_LOG_MINP; + ndi_max = (ndi_max-s)>>1; + di = IMIN(val - i, ndi_max - 1); + fl += (2*di+1+s)*LAPLACE_MINP; + fs = IMIN(LAPLACE_MINP, 32768-fl); + *value = (i+di+s)^s; + } + else + { + fs += LAPLACE_MINP; + fl += fs&~s; + } + celt_assert(fl+fs<=32768); + celt_assert(fs>0); + } + ec_encode_bin(enc, fl, fl+fs, 15); +} + +int ec_laplace_decode(ec_dec *dec, unsigned fs, int decay) +{ + int val=0; + unsigned fl; + unsigned fm; + fm = ec_decode_bin(dec, 15); + fl = 0; + if (fm >= fs) + { + val++; + fl = fs; + fs = ec_laplace_get_freq1(fs, decay)+LAPLACE_MINP; + /* Search the decaying part of the PDF.*/ + while(fs > LAPLACE_MINP && fm >= fl+2*fs) + { + fs *= 2; + fl += fs; + fs = ((fs-2*LAPLACE_MINP)*(opus_int32)decay)>>15; + fs += LAPLACE_MINP; + val++; + } + /* Everything beyond that has probability LAPLACE_MINP. */ + if (fs <= LAPLACE_MINP) + { + int di; + di = (fm-fl)>>(LAPLACE_LOG_MINP+1); + val += di; + fl += 2*di*LAPLACE_MINP; + } + if (fm < fl+fs) + val = -val; + else + fl += fs; + } + celt_assert(fl<32768); + celt_assert(fs>0); + celt_assert(fl<=fm); + celt_assert(fm<IMIN(fl+fs,32768)); + ec_dec_update(dec, fl, IMIN(fl+fs,32768), 32768); + return val; +} diff --git a/drivers/opus/celt/laplace.h b/drivers/opus/celt/laplace.h new file mode 100644 index 0000000000..46c14b5da5 --- /dev/null +++ b/drivers/opus/celt/laplace.h @@ -0,0 +1,48 @@ +/* Copyright (c) 2007 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#include "entenc.h" +#include "entdec.h" + +/** Encode a value that is assumed to be the realisation of a + Laplace-distributed random process + @param enc Entropy encoder state + @param value Value to encode + @param fs Probability of 0, multiplied by 32768 + @param decay Probability of the value +/- 1, multiplied by 16384 +*/ +void ec_laplace_encode(ec_enc *enc, int *value, unsigned fs, int decay); + +/** Decode a value that is assumed to be the realisation of a + Laplace-distributed random process + @param dec Entropy decoder state + @param fs Probability of 0, multiplied by 32768 + @param decay Probability of the value +/- 1, multiplied by 16384 + @return Value decoded + */ +int ec_laplace_decode(ec_dec *dec, unsigned fs, int decay); diff --git a/drivers/opus/celt/mathops.c b/drivers/opus/celt/mathops.c new file mode 100644 index 0000000000..49be746d8c --- /dev/null +++ b/drivers/opus/celt/mathops.c @@ -0,0 +1,208 @@ +/* Copyright (c) 2002-2008 Jean-Marc Valin + Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/** + @file mathops.h + @brief Various math functions +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef OPUS_HAVE_CONFIG_H +#include "opus_config.h" +#endif + +#include "mathops.h" + +/*Compute floor(sqrt(_val)) with exact arithmetic. + This has been tested on all possible 32-bit inputs.*/ +unsigned isqrt32(opus_uint32 _val){ + unsigned b; + unsigned g; + int bshift; + /*Uses the second method from + http://www.azillionmonkeys.com/qed/sqroot.html + The main idea is to search for the largest binary digit b such that + (g+b)*(g+b) <= _val, and add it to the solution g.*/ + g=0; + bshift=(EC_ILOG(_val)-1)>>1; + b=1U<<bshift; + do{ + opus_uint32 t; + t=(((opus_uint32)g<<1)+b)<<bshift; + if(t<=_val){ + g+=b; + _val-=t; + } + b>>=1; + bshift--; + } + while(bshift>=0); + return g; +} + +#ifdef OPUS_FIXED_POINT + +opus_val32 frac_div32(opus_val32 a, opus_val32 b) +{ + opus_val16 rcp; + opus_val32 result, rem; + int shift = celt_ilog2(b)-29; + a = VSHR32(a,shift); + b = VSHR32(b,shift); + /* 16-bit reciprocal */ + rcp = ROUND16(celt_rcp(ROUND16(b,16)),3); + result = MULT16_32_Q15(rcp, a); + rem = PSHR32(a,2)-MULT32_32_Q31(result, b); + result = ADD32(result, SHL32(MULT16_32_Q15(rcp, rem),2)); + if (result >= 536870912) /* 2^29 */ + return 2147483647; /* 2^31 - 1 */ + else if (result <= -536870912) /* -2^29 */ + return -2147483647; /* -2^31 */ + else + return SHL32(result, 2); +} + +/** Reciprocal sqrt approximation in the range [0.25,1) (Q16 in, Q14 out) */ +opus_val16 celt_rsqrt_norm(opus_val32 x) +{ + opus_val16 n; + opus_val16 r; + opus_val16 r2; + opus_val16 y; + /* Range of n is [-16384,32767] ([-0.5,1) in Q15). */ + n = x-32768; + /* Get a rough initial guess for the root. + The optimal minimax quadratic approximation (using relative error) is + r = 1.437799046117536+n*(-0.823394375837328+n*0.4096419668459485). + Coefficients here, and the final result r, are Q14.*/ + r = ADD16(23557, MULT16_16_Q15(n, ADD16(-13490, MULT16_16_Q15(n, 6713)))); + /* We want y = x*r*r-1 in Q15, but x is 32-bit Q16 and r is Q14. + We can compute the result from n and r using Q15 multiplies with some + adjustment, carefully done to avoid overflow. + Range of y is [-1564,1594]. */ + r2 = MULT16_16_Q15(r, r); + y = SHL16(SUB16(ADD16(MULT16_16_Q15(r2, n), r2), 16384), 1); + /* Apply a 2nd-order Householder iteration: r += r*y*(y*0.375-0.5). + This yields the Q14 reciprocal square root of the Q16 x, with a maximum + relative error of 1.04956E-4, a (relative) RMSE of 2.80979E-5, and a + peak absolute error of 2.26591/16384. */ + return ADD16(r, MULT16_16_Q15(r, MULT16_16_Q15(y, + SUB16(MULT16_16_Q15(y, 12288), 16384)))); +} + +/** Sqrt approximation (QX input, QX/2 output) */ +opus_val32 celt_sqrt(opus_val32 x) +{ + int k; + opus_val16 n; + opus_val32 rt; + static const opus_val16 C[5] = {23175, 11561, -3011, 1699, -664}; + if (x==0) + return 0; + else if (x>=1073741824) + return 32767; + k = (celt_ilog2(x)>>1)-7; + x = VSHR32(x, 2*k); + n = x-32768; + rt = ADD16(C[0], MULT16_16_Q15(n, ADD16(C[1], MULT16_16_Q15(n, ADD16(C[2], + MULT16_16_Q15(n, ADD16(C[3], MULT16_16_Q15(n, (C[4]))))))))); + rt = VSHR32(rt,7-k); + return rt; +} + +#define L1 32767 +#define L2 -7651 +#define L3 8277 +#define L4 -626 + +static OPUS_INLINE opus_val16 _celt_cos_pi_2(opus_val16 x) +{ + opus_val16 x2; + + x2 = MULT16_16_P15(x,x); + return ADD16(1,MIN16(32766,ADD32(SUB16(L1,x2), MULT16_16_P15(x2, ADD32(L2, MULT16_16_P15(x2, ADD32(L3, MULT16_16_P15(L4, x2 + )))))))); +} + +#undef L1 +#undef L2 +#undef L3 +#undef L4 + +opus_val16 celt_cos_norm(opus_val32 x) +{ + x = x&0x0001ffff; + if (x>SHL32(EXTEND32(1), 16)) + x = SUB32(SHL32(EXTEND32(1), 17),x); + if (x&0x00007fff) + { + if (x<SHL32(EXTEND32(1), 15)) + { + return _celt_cos_pi_2(EXTRACT16(x)); + } else { + return NEG32(_celt_cos_pi_2(EXTRACT16(65536-x))); + } + } else { + if (x&0x0000ffff) + return 0; + else if (x&0x0001ffff) + return -32767; + else + return 32767; + } +} + +/** Reciprocal approximation (Q15 input, Q16 output) */ +opus_val32 celt_rcp(opus_val32 x) +{ + int i; + opus_val16 n; + opus_val16 r; + celt_assert2(x>0, "celt_rcp() only defined for positive values"); + i = celt_ilog2(x); + /* n is Q15 with range [0,1). */ + n = VSHR32(x,i-15)-32768; + /* Start with a linear approximation: + r = 1.8823529411764706-0.9411764705882353*n. + The coefficients and the result are Q14 in the range [15420,30840].*/ + r = ADD16(30840, MULT16_16_Q15(-15420, n)); + /* Perform two Newton iterations: + r -= r*((r*n)-1.Q15) + = r*((r*n)+(r-1.Q15)). */ + r = SUB16(r, MULT16_16_Q15(r, + ADD16(MULT16_16_Q15(r, n), ADD16(r, -32768)))); + /* We subtract an extra 1 in the second iteration to avoid overflow; it also + neatly compensates for truncation error in the rest of the process. */ + r = SUB16(r, ADD16(1, MULT16_16_Q15(r, + ADD16(MULT16_16_Q15(r, n), ADD16(r, -32768))))); + /* r is now the Q15 solution to 2/(n+1), with a maximum relative error + of 7.05346E-5, a (relative) RMSE of 2.14418E-5, and a peak absolute + error of 1.24665/32768. */ + return VSHR32(EXTEND32(r),i-16); +} + +#endif diff --git a/drivers/opus/celt/mathops.h b/drivers/opus/celt/mathops.h new file mode 100644 index 0000000000..4a6bc539bc --- /dev/null +++ b/drivers/opus/celt/mathops.h @@ -0,0 +1,258 @@ +/* Copyright (c) 2002-2008 Jean-Marc Valin + Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/** + @file mathops.h + @brief Various math functions +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef MATHOPS_H +#define MATHOPS_H + +#include "arch.h" +#include "entcode.h" +#include "os_support.h" + +/* Multiplies two 16-bit fractional values. Bit-exactness of this macro is important */ +#define FRAC_MUL16(a,b) ((16384+((opus_int32)(opus_int16)(a)*(opus_int16)(b)))>>15) + +unsigned isqrt32(opus_uint32 _val); + +#ifndef OVERRIDE_CELT_MAXABS16 +static OPUS_INLINE opus_val32 celt_maxabs16(const opus_val16 *x, int len) +{ + int i; + opus_val16 maxval = 0; + opus_val16 minval = 0; + for (i=0;i<len;i++) + { + maxval = MAX16(maxval, x[i]); + minval = MIN16(minval, x[i]); + } + return MAX32(EXTEND32(maxval),-EXTEND32(minval)); +} +#endif + +#ifndef OVERRIDE_CELT_MAXABS32 +#ifdef OPUS_FIXED_POINT +static OPUS_INLINE opus_val32 celt_maxabs32(const opus_val32 *x, int len) +{ + int i; + opus_val32 maxval = 0; + opus_val32 minval = 0; + for (i=0;i<len;i++) + { + maxval = MAX32(maxval, x[i]); + minval = MIN32(minval, x[i]); + } + return MAX32(maxval, -minval); +} +#else +#define celt_maxabs32(x,len) celt_maxabs16(x,len) +#endif +#endif + + +#ifndef OPUS_FIXED_POINT + +#define PI 3.141592653f +#define celt_sqrt(x) ((float)sqrt(x)) +#define celt_rsqrt(x) (1.f/celt_sqrt(x)) +#define celt_rsqrt_norm(x) (celt_rsqrt(x)) +#define celt_cos_norm(x) ((float)cos((.5f*PI)*(x))) +#define celt_rcp(x) (1.f/(x)) +#define celt_div(a,b) ((a)/(b)) +#define frac_div32(a,b) ((float)(a)/(b)) + +#ifdef FLOAT_APPROX + +/* Note: This assumes radix-2 floating point with the exponent at bits 23..30 and an offset of 127 + denorm, +/- inf and NaN are *not* handled */ + +/** Base-2 log approximation (log2(x)). */ +static OPUS_INLINE float celt_log2(float x) +{ + int integer; + float frac; + union { + float f; + opus_uint32 i; + } in; + in.f = x; + integer = (in.i>>23)-127; + in.i -= integer<<23; + frac = in.f - 1.5f; + frac = -0.41445418f + frac*(0.95909232f + + frac*(-0.33951290f + frac*0.16541097f)); + return 1+integer+frac; +} + +/** Base-2 exponential approximation (2^x). */ +static OPUS_INLINE float celt_exp2(float x) +{ + int integer; + float frac; + union { + float f; + opus_uint32 i; + } res; + integer = floor(x); + if (integer < -50) + return 0; + frac = x-integer; + /* K0 = 1, K1 = log(2), K2 = 3-4*log(2), K3 = 3*log(2) - 2 */ + res.f = 0.99992522f + frac * (0.69583354f + + frac * (0.22606716f + 0.078024523f*frac)); + res.i = (res.i + (integer<<23)) & 0x7fffffff; + return res.f; +} + +#else +#define celt_log2(x) ((float)(1.442695040888963387*log(x))) +#define celt_exp2(x) ((float)exp(0.6931471805599453094*(x))) +#endif + +#endif + +#ifdef OPUS_FIXED_POINT + +#include "os_support.h" + +#ifndef OVERRIDE_CELT_ILOG2 +/** Integer log in base2. Undefined for zero and negative numbers */ +static OPUS_INLINE opus_int16 celt_ilog2(opus_int32 x) +{ + celt_assert2(x>0, "celt_ilog2() only defined for strictly positive numbers"); + return EC_ILOG(x)-1; +} +#endif + + +/** Integer log in base2. Defined for zero, but not for negative numbers */ +static OPUS_INLINE opus_int16 celt_zlog2(opus_val32 x) +{ + return x <= 0 ? 0 : celt_ilog2(x); +} + +opus_val16 celt_rsqrt_norm(opus_val32 x); + +opus_val32 celt_sqrt(opus_val32 x); + +opus_val16 celt_cos_norm(opus_val32 x); + +/** Base-2 logarithm approximation (log2(x)). (Q14 input, Q10 output) */ +static OPUS_INLINE opus_val16 celt_log2(opus_val32 x) +{ + int i; + opus_val16 n, frac; + /* -0.41509302963303146, 0.9609890551383969, -0.31836011537636605, + 0.15530808010959576, -0.08556153059057618 */ + static const opus_val16 C[5] = {-6801+(1<<(13-DB_SHIFT)), 15746, -5217, 2545, -1401}; + if (x==0) + return -32767; + i = celt_ilog2(x); + n = VSHR32(x,i-15)-32768-16384; + frac = ADD16(C[0], MULT16_16_Q15(n, ADD16(C[1], MULT16_16_Q15(n, ADD16(C[2], MULT16_16_Q15(n, ADD16(C[3], MULT16_16_Q15(n, C[4])))))))); + return SHL16(i-13,DB_SHIFT)+SHR16(frac,14-DB_SHIFT); +} + +/* + K0 = 1 + K1 = log(2) + K2 = 3-4*log(2) + K3 = 3*log(2) - 2 +*/ +#define D0 16383 +#define D1 22804 +#define D2 14819 +#define D3 10204 + +static OPUS_INLINE opus_val32 celt_exp2_frac(opus_val16 x) +{ + opus_val16 frac; + frac = SHL16(x, 4); + return ADD16(D0, MULT16_16_Q15(frac, ADD16(D1, MULT16_16_Q15(frac, ADD16(D2 , MULT16_16_Q15(D3,frac)))))); +} +/** Base-2 exponential approximation (2^x). (Q10 input, Q16 output) */ +static OPUS_INLINE opus_val32 celt_exp2(opus_val16 x) +{ + int integer; + opus_val16 frac; + integer = SHR16(x,10); + if (integer>14) + return 0x7f000000; + else if (integer < -15) + return 0; + frac = celt_exp2_frac(x-SHL16(integer,10)); + return VSHR32(EXTEND32(frac), -integer-2); +} + +opus_val32 celt_rcp(opus_val32 x); + +#define celt_div(a,b) MULT32_32_Q31((opus_val32)(a),celt_rcp(b)) + +opus_val32 frac_div32(opus_val32 a, opus_val32 b); + +#define M1 32767 +#define M2 -21 +#define M3 -11943 +#define M4 4936 + +/* Atan approximation using a 4th order polynomial. Input is in Q15 format + and normalized by pi/4. Output is in Q15 format */ +static OPUS_INLINE opus_val16 celt_atan01(opus_val16 x) +{ + return MULT16_16_P15(x, ADD32(M1, MULT16_16_P15(x, ADD32(M2, MULT16_16_P15(x, ADD32(M3, MULT16_16_P15(M4, x))))))); +} + +#undef M1 +#undef M2 +#undef M3 +#undef M4 + +/* atan2() approximation valid for positive input values */ +static OPUS_INLINE opus_val16 celt_atan2p(opus_val16 y, opus_val16 x) +{ + if (y < x) + { + opus_val32 arg; + arg = celt_div(SHL32(EXTEND32(y),15),x); + if (arg >= 32767) + arg = 32767; + return SHR16(celt_atan01(EXTRACT16(arg)),1); + } else { + opus_val32 arg; + arg = celt_div(SHL32(EXTEND32(x),15),y); + if (arg >= 32767) + arg = 32767; + return 25736-SHR16(celt_atan01(EXTRACT16(arg)),1); + } +} + +#endif /* OPUS_FIXED_POINT */ +#endif /* MATHOPS_H */ diff --git a/drivers/opus/celt/mdct.c b/drivers/opus/celt/mdct.c new file mode 100644 index 0000000000..d08d026fac --- /dev/null +++ b/drivers/opus/celt/mdct.c @@ -0,0 +1,311 @@ + /* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2008 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +/* This is a simple MDCT implementation that uses a N/4 complex FFT + to do most of the work. It should be relatively straightforward to + plug in pretty much and FFT here. + + This replaces the Vorbis FFT (and uses the exact same API), which + was a bit too messy and that was ending up duplicating code + (might as well use the same FFT everywhere). + + The algorithm is similar to (and inspired from) Fabrice Bellard's + MDCT implementation in FFMPEG, but has differences in signs, ordering + and scaling in many places. +*/ + +#ifndef SKIP_CONFIG_H +#ifdef OPUS_HAVE_CONFIG_H +#include "opus_config.h" +#endif +#endif + +#include "mdct.h" +#include "kiss_fft.h" +#include "_kiss_fft_guts.h" +#include <math.h> +#include "os_support.h" +#include "mathops.h" +#include "stack_alloc.h" + +#ifdef CUSTOM_MODES + +int clt_mdct_init(celt_mdct_lookup *l,int N, int maxshift) +{ + int i; + int N4; + kiss_twiddle_scalar *trig; +#if defined(OPUS_FIXED_POINT) + int N2=N>>1; +#endif + l->n = N; + N4 = N>>2; + l->maxshift = maxshift; + for (i=0;i<=maxshift;i++) + { + if (i==0) + l->kfft[i] = opus_fft_alloc(N>>2>>i, 0, 0); + else + l->kfft[i] = opus_fft_alloc_twiddles(N>>2>>i, 0, 0, l->kfft[0]); +#ifndef ENABLE_TI_DSPLIB55 + if (l->kfft[i]==NULL) + return 0; +#endif + } + l->trig = trig = (kiss_twiddle_scalar*)opus_alloc((N4+1)*sizeof(kiss_twiddle_scalar)); + if (l->trig==NULL) + return 0; + /* We have enough points that sine isn't necessary */ +#if defined(OPUS_FIXED_POINT) + for (i=0;i<=N4;i++) + trig[i] = TRIG_UPSCALE*celt_cos_norm(DIV32(ADD32(SHL32(EXTEND32(i),17),N2),N)); +#else + for (i=0;i<=N4;i++) + trig[i] = (kiss_twiddle_scalar)cos(2*PI*i/N); +#endif + return 1; +} + +void clt_mdct_clear(celt_mdct_lookup *l) +{ + int i; + for (i=0;i<=l->maxshift;i++) + opus_fft_free(l->kfft[i]); + opus_free((kiss_twiddle_scalar*)l->trig); +} + +#endif /* CUSTOM_MODES */ + +/* Forward MDCT trashes the input array */ +void clt_mdct_forward(const celt_mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 *window, int overlap, int shift, int stride) +{ + int i; + int N, N2, N4; + kiss_twiddle_scalar sine; + VARDECL(kiss_fft_scalar, f); + VARDECL(kiss_fft_scalar, f2); + SAVE_STACK; + N = l->n; + N >>= shift; + N2 = N>>1; + N4 = N>>2; + ALLOC(f, N2, kiss_fft_scalar); + ALLOC(f2, N2, kiss_fft_scalar); + /* sin(x) ~= x here */ +#ifdef OPUS_FIXED_POINT + sine = TRIG_UPSCALE*(QCONST16(0.7853981f, 15)+N2)/N; +#else + sine = (kiss_twiddle_scalar)2*PI*(.125f)/N; +#endif + + /* Consider the input to be composed of four blocks: [a, b, c, d] */ + /* Window, shuffle, fold */ + { + /* Temp pointers to make it really clear to the compiler what we're doing */ + const kiss_fft_scalar * OPUS_RESTRICT xp1 = in+(overlap>>1); + const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+N2-1+(overlap>>1); + kiss_fft_scalar * OPUS_RESTRICT yp = f; + const opus_val16 * OPUS_RESTRICT wp1 = window+(overlap>>1); + const opus_val16 * OPUS_RESTRICT wp2 = window+(overlap>>1)-1; + for(i=0;i<((overlap+3)>>2);i++) + { + /* Real part arranged as -d-cR, Imag part arranged as -b+aR*/ + *yp++ = MULT16_32_Q15(*wp2, xp1[N2]) + MULT16_32_Q15(*wp1,*xp2); + *yp++ = MULT16_32_Q15(*wp1, *xp1) - MULT16_32_Q15(*wp2, xp2[-N2]); + xp1+=2; + xp2-=2; + wp1+=2; + wp2-=2; + } + wp1 = window; + wp2 = window+overlap-1; + for(;i<N4-((overlap+3)>>2);i++) + { + /* Real part arranged as a-bR, Imag part arranged as -c-dR */ + *yp++ = *xp2; + *yp++ = *xp1; + xp1+=2; + xp2-=2; + } + for(;i<N4;i++) + { + /* Real part arranged as a-bR, Imag part arranged as -c-dR */ + *yp++ = -MULT16_32_Q15(*wp1, xp1[-N2]) + MULT16_32_Q15(*wp2, *xp2); + *yp++ = MULT16_32_Q15(*wp2, *xp1) + MULT16_32_Q15(*wp1, xp2[N2]); + xp1+=2; + xp2-=2; + wp1+=2; + wp2-=2; + } + } + /* Pre-rotation */ + { + kiss_fft_scalar * OPUS_RESTRICT yp = f; + const kiss_twiddle_scalar *t = &l->trig[0]; + for(i=0;i<N4;i++) + { + kiss_fft_scalar re, im, yr, yi; + re = yp[0]; + im = yp[1]; + yr = -S_MUL(re,t[i<<shift]) - S_MUL(im,t[(N4-i)<<shift]); + yi = -S_MUL(im,t[i<<shift]) + S_MUL(re,t[(N4-i)<<shift]); + /* works because the cos is nearly one */ + *yp++ = yr + S_MUL(yi,sine); + *yp++ = yi - S_MUL(yr,sine); + } + } + + /* N/4 complex FFT, down-scales by 4/N */ + opus_fft(l->kfft[shift], (kiss_fft_cpx *)f, (kiss_fft_cpx *)f2); + + /* Post-rotate */ + { + /* Temp pointers to make it really clear to the compiler what we're doing */ + const kiss_fft_scalar * OPUS_RESTRICT fp = f2; + kiss_fft_scalar * OPUS_RESTRICT yp1 = out; + kiss_fft_scalar * OPUS_RESTRICT yp2 = out+stride*(N2-1); + const kiss_twiddle_scalar *t = &l->trig[0]; + /* Temp pointers to make it really clear to the compiler what we're doing */ + for(i=0;i<N4;i++) + { + kiss_fft_scalar yr, yi; + yr = S_MUL(fp[1],t[(N4-i)<<shift]) + S_MUL(fp[0],t[i<<shift]); + yi = S_MUL(fp[0],t[(N4-i)<<shift]) - S_MUL(fp[1],t[i<<shift]); + /* works because the cos is nearly one */ + *yp1 = yr - S_MUL(yi,sine); + *yp2 = yi + S_MUL(yr,sine);; + fp += 2; + yp1 += 2*stride; + yp2 -= 2*stride; + } + } + RESTORE_STACK; +} + +void clt_mdct_backward(const celt_mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 * OPUS_RESTRICT window, int overlap, int shift, int stride) +{ + int i; + int N, N2, N4; + kiss_twiddle_scalar sine; + VARDECL(kiss_fft_scalar, f2); + SAVE_STACK; + N = l->n; + N >>= shift; + N2 = N>>1; + N4 = N>>2; + ALLOC(f2, N2, kiss_fft_scalar); + /* sin(x) ~= x here */ +#ifdef OPUS_FIXED_POINT + sine = TRIG_UPSCALE*(QCONST16(0.7853981f, 15)+N2)/N; +#else + sine = (kiss_twiddle_scalar)2*PI*(.125f)/N; +#endif + + /* Pre-rotate */ + { + /* Temp pointers to make it really clear to the compiler what we're doing */ + const kiss_fft_scalar * OPUS_RESTRICT xp1 = in; + const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+stride*(N2-1); + kiss_fft_scalar * OPUS_RESTRICT yp = f2; + const kiss_twiddle_scalar *t = &l->trig[0]; + for(i=0;i<N4;i++) + { + kiss_fft_scalar yr, yi; + yr = -S_MUL(*xp2, t[i<<shift]) + S_MUL(*xp1,t[(N4-i)<<shift]); + yi = -S_MUL(*xp2, t[(N4-i)<<shift]) - S_MUL(*xp1,t[i<<shift]); + /* works because the cos is nearly one */ + *yp++ = yr - S_MUL(yi,sine); + *yp++ = yi + S_MUL(yr,sine); + xp1+=2*stride; + xp2-=2*stride; + } + } + + /* Inverse N/4 complex FFT. This one should *not* downscale even in fixed-point */ + opus_ifft(l->kfft[shift], (kiss_fft_cpx *)f2, (kiss_fft_cpx *)(out+(overlap>>1))); + + /* Post-rotate and de-shuffle from both ends of the buffer at once to make + it in-place. */ + { + kiss_fft_scalar * OPUS_RESTRICT yp0 = out+(overlap>>1); + kiss_fft_scalar * OPUS_RESTRICT yp1 = out+(overlap>>1)+N2-2; + const kiss_twiddle_scalar *t = &l->trig[0]; + /* Loop to (N4+1)>>1 to handle odd N4. When N4 is odd, the + middle pair will be computed twice. */ + for(i=0;i<(N4+1)>>1;i++) + { + kiss_fft_scalar re, im, yr, yi; + kiss_twiddle_scalar t0, t1; + re = yp0[0]; + im = yp0[1]; + t0 = t[i<<shift]; + t1 = t[(N4-i)<<shift]; + /* We'd scale up by 2 here, but instead it's done when mixing the windows */ + yr = S_MUL(re,t0) - S_MUL(im,t1); + yi = S_MUL(im,t0) + S_MUL(re,t1); + re = yp1[0]; + im = yp1[1]; + /* works because the cos is nearly one */ + yp0[0] = -(yr - S_MUL(yi,sine)); + yp1[1] = yi + S_MUL(yr,sine); + + t0 = t[(N4-i-1)<<shift]; + t1 = t[(i+1)<<shift]; + /* We'd scale up by 2 here, but instead it's done when mixing the windows */ + yr = S_MUL(re,t0) - S_MUL(im,t1); + yi = S_MUL(im,t0) + S_MUL(re,t1); + /* works because the cos is nearly one */ + yp1[0] = -(yr - S_MUL(yi,sine)); + yp0[1] = yi + S_MUL(yr,sine); + yp0 += 2; + yp1 -= 2; + } + } + + /* Mirror on both sides for TDAC */ + { + kiss_fft_scalar * OPUS_RESTRICT xp1 = out+overlap-1; + kiss_fft_scalar * OPUS_RESTRICT yp1 = out; + const opus_val16 * OPUS_RESTRICT wp1 = window; + const opus_val16 * OPUS_RESTRICT wp2 = window+overlap-1; + + for(i = 0; i < overlap/2; i++) + { + kiss_fft_scalar x1, x2; + x1 = *xp1; + x2 = *yp1; + *yp1++ = MULT16_32_Q15(*wp2, x2) - MULT16_32_Q15(*wp1, x1); + *xp1-- = MULT16_32_Q15(*wp1, x2) + MULT16_32_Q15(*wp2, x1); + wp1++; + wp2--; + } + } + RESTORE_STACK; +} diff --git a/drivers/opus/celt/mdct.h b/drivers/opus/celt/mdct.h new file mode 100644 index 0000000000..4e7a199246 --- /dev/null +++ b/drivers/opus/celt/mdct.h @@ -0,0 +1,70 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2008 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +/* This is a simple MDCT implementation that uses a N/4 complex FFT + to do most of the work. It should be relatively straightforward to + plug in pretty much and FFT here. + + This replaces the Vorbis FFT (and uses the exact same API), which + was a bit too messy and that was ending up duplicating code + (might as well use the same FFT everywhere). + + The algorithm is similar to (and inspired from) Fabrice Bellard's + MDCT implementation in FFMPEG, but has differences in signs, ordering + and scaling in many places. +*/ + +#ifndef MDCT_H +#define MDCT_H + +#include "opus_defines.h" +#include "kiss_fft.h" +#include "arch.h" + +typedef struct { + int n; + int maxshift; + const kiss_fft_state *kfft[4]; + const kiss_twiddle_scalar * OPUS_RESTRICT trig; +} celt_mdct_lookup; + +int clt_mdct_init(celt_mdct_lookup *l,int N, int maxshift); +void clt_mdct_clear(celt_mdct_lookup *l); + +/** Compute a forward MDCT and scale by 4/N, trashes the input array */ +void clt_mdct_forward(const celt_mdct_lookup *l, kiss_fft_scalar *in, + kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 *window, int overlap, int shift, int stride); + +/** Compute a backward MDCT (no scaling) and performs weighted overlap-add + (scales implicitly by 1/2) */ +void clt_mdct_backward(const celt_mdct_lookup *l, kiss_fft_scalar *in, + kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 * OPUS_RESTRICT window, int overlap, int shift, int stride); + +#endif diff --git a/drivers/opus/celt/mfrngcod.h b/drivers/opus/celt/mfrngcod.h new file mode 100644 index 0000000000..809152a59a --- /dev/null +++ b/drivers/opus/celt/mfrngcod.h @@ -0,0 +1,48 @@ +/* Copyright (c) 2001-2008 Timothy B. Terriberry + Copyright (c) 2008-2009 Xiph.Org Foundation */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#if !defined(_mfrngcode_H) +# define _mfrngcode_H (1) +# include "entcode.h" + +/*Constants used by the entropy encoder/decoder.*/ + +/*The number of bits to output at a time.*/ +# define EC_SYM_BITS (8) +/*The total number of bits in each of the state registers.*/ +# define EC_CODE_BITS (32) +/*The maximum symbol value.*/ +# define EC_SYM_MAX ((1U<<EC_SYM_BITS)-1) +/*Bits to shift by to move a symbol into the high-order position.*/ +# define EC_CODE_SHIFT (EC_CODE_BITS-EC_SYM_BITS-1) +/*Carry bit of the high-order range symbol.*/ +# define EC_CODE_TOP (((opus_uint32)1U)<<(EC_CODE_BITS-1)) +/*Low-order bit of the high-order range symbol.*/ +# define EC_CODE_BOT (EC_CODE_TOP>>EC_SYM_BITS) +/*The number of bits available for the last, partial symbol in the code field.*/ +# define EC_CODE_EXTRA ((EC_CODE_BITS-2)%EC_SYM_BITS+1) +#endif diff --git a/drivers/opus/celt/modes.c b/drivers/opus/celt/modes.c new file mode 100644 index 0000000000..3794074aaa --- /dev/null +++ b/drivers/opus/celt/modes.c @@ -0,0 +1,438 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Copyright (c) 2008 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef OPUS_HAVE_CONFIG_H +#include "opus_config.h" +#endif + +#include "celt.h" +#include "opus_modes.h" +#include "rate.h" +#include "os_support.h" +#include "stack_alloc.h" +#include "quant_bands.h" + +static const opus_int16 eband5ms[] = { +/*0 200 400 600 800 1k 1.2 1.4 1.6 2k 2.4 2.8 3.2 4k 4.8 5.6 6.8 8k 9.6 12k 15.6 */ + 0, 1, 2, 3, 4, 5, 6, 7, 8, 10, 12, 14, 16, 20, 24, 28, 34, 40, 48, 60, 78, 100 +}; + +/* Alternate tuning (partially derived from Vorbis) */ +#define BITALLOC_SIZE 11 +/* Bit allocation table in units of 1/32 bit/sample (0.1875 dB SNR) */ +static const unsigned char band_allocation[] = { +/*0 200 400 600 800 1k 1.2 1.4 1.6 2k 2.4 2.8 3.2 4k 4.8 5.6 6.8 8k 9.6 12k 15.6 */ + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 90, 80, 75, 69, 63, 56, 49, 40, 34, 29, 20, 18, 10, 0, 0, 0, 0, 0, 0, 0, 0, +110,100, 90, 84, 78, 71, 65, 58, 51, 45, 39, 32, 26, 20, 12, 0, 0, 0, 0, 0, 0, +118,110,103, 93, 86, 80, 75, 70, 65, 59, 53, 47, 40, 31, 23, 15, 4, 0, 0, 0, 0, +126,119,112,104, 95, 89, 83, 78, 72, 66, 60, 54, 47, 39, 32, 25, 17, 12, 1, 0, 0, +134,127,120,114,103, 97, 91, 85, 78, 72, 66, 60, 54, 47, 41, 35, 29, 23, 16, 10, 1, +144,137,130,124,113,107,101, 95, 88, 82, 76, 70, 64, 57, 51, 45, 39, 33, 26, 15, 1, +152,145,138,132,123,117,111,105, 98, 92, 86, 80, 74, 67, 61, 55, 49, 43, 36, 20, 1, +162,155,148,142,133,127,121,115,108,102, 96, 90, 84, 77, 71, 65, 59, 53, 46, 30, 1, +172,165,158,152,143,137,131,125,118,112,106,100, 94, 87, 81, 75, 69, 63, 56, 45, 20, +200,200,200,200,200,200,200,200,198,193,188,183,178,173,168,163,158,153,148,129,104, +}; + +#ifndef CUSTOM_MODES_ONLY + #ifdef OPUS_FIXED_POINT + #include "static_modes_fixed.h" + #else + #include "static_modes_float.h" + #endif +#endif /* CUSTOM_MODES_ONLY */ + +#ifndef M_PI +#define M_PI 3.141592653 +#endif + +#ifdef CUSTOM_MODES + +/* Defining 25 critical bands for the full 0-20 kHz audio bandwidth + Taken from http://ccrma.stanford.edu/~jos/bbt/Bark_Frequency_Scale.html */ +#define BARK_BANDS 25 +static const opus_int16 bark_freq[BARK_BANDS+1] = { + 0, 100, 200, 300, 400, + 510, 630, 770, 920, 1080, + 1270, 1480, 1720, 2000, 2320, + 2700, 3150, 3700, 4400, 5300, + 6400, 7700, 9500, 12000, 15500, + 20000}; + +static opus_int16 *compute_ebands(opus_int32 Fs, int frame_size, int res, int *nbEBands) +{ + opus_int16 *eBands; + int i, j, lin, low, high, nBark, offset=0; + + /* All modes that have 2.5 ms short blocks use the same definition */ + if (Fs == 400*(opus_int32)frame_size) + { + *nbEBands = sizeof(eband5ms)/sizeof(eband5ms[0])-1; + eBands = opus_alloc(sizeof(opus_int16)*(*nbEBands+1)); + for (i=0;i<*nbEBands+1;i++) + eBands[i] = eband5ms[i]; + return eBands; + } + /* Find the number of critical bands supported by our sampling rate */ + for (nBark=1;nBark<BARK_BANDS;nBark++) + if (bark_freq[nBark+1]*2 >= Fs) + break; + + /* Find where the linear part ends (i.e. where the spacing is more than min_width */ + for (lin=0;lin<nBark;lin++) + if (bark_freq[lin+1]-bark_freq[lin] >= res) + break; + + low = (bark_freq[lin]+res/2)/res; + high = nBark-lin; + *nbEBands = low+high; + eBands = opus_alloc(sizeof(opus_int16)*(*nbEBands+2)); + + if (eBands==NULL) + return NULL; + + /* Linear spacing (min_width) */ + for (i=0;i<low;i++) + eBands[i] = i; + if (low>0) + offset = eBands[low-1]*res - bark_freq[lin-1]; + /* Spacing follows critical bands */ + for (i=0;i<high;i++) + { + int target = bark_freq[lin+i]; + /* Round to an even value */ + eBands[i+low] = (target+offset/2+res)/(2*res)*2; + offset = eBands[i+low]*res - target; + } + /* Enforce the minimum spacing at the boundary */ + for (i=0;i<*nbEBands;i++) + if (eBands[i] < i) + eBands[i] = i; + /* Round to an even value */ + eBands[*nbEBands] = (bark_freq[nBark]+res)/(2*res)*2; + if (eBands[*nbEBands] > frame_size) + eBands[*nbEBands] = frame_size; + for (i=1;i<*nbEBands-1;i++) + { + if (eBands[i+1]-eBands[i] < eBands[i]-eBands[i-1]) + { + eBands[i] -= (2*eBands[i]-eBands[i-1]-eBands[i+1])/2; + } + } + /* Remove any empty bands. */ + for (i=j=0;i<*nbEBands;i++) + if(eBands[i+1]>eBands[j]) + eBands[++j]=eBands[i+1]; + *nbEBands=j; + + for (i=1;i<*nbEBands;i++) + { + /* Every band must be smaller than the last band. */ + celt_assert(eBands[i]-eBands[i-1]<=eBands[*nbEBands]-eBands[*nbEBands-1]); + /* Each band must be no larger than twice the size of the previous one. */ + celt_assert(eBands[i+1]-eBands[i]<=2*(eBands[i]-eBands[i-1])); + } + + return eBands; +} + +static void compute_allocation_table(CELTMode *mode) +{ + int i, j; + unsigned char *allocVectors; + int maxBands = sizeof(eband5ms)/sizeof(eband5ms[0])-1; + + mode->nbAllocVectors = BITALLOC_SIZE; + allocVectors = opus_alloc(sizeof(unsigned char)*(BITALLOC_SIZE*mode->nbEBands)); + if (allocVectors==NULL) + return; + + /* Check for standard mode */ + if (mode->Fs == 400*(opus_int32)mode->shortMdctSize) + { + for (i=0;i<BITALLOC_SIZE*mode->nbEBands;i++) + allocVectors[i] = band_allocation[i]; + mode->allocVectors = allocVectors; + return; + } + /* If not the standard mode, interpolate */ + /* Compute per-codec-band allocation from per-critical-band matrix */ + for (i=0;i<BITALLOC_SIZE;i++) + { + for (j=0;j<mode->nbEBands;j++) + { + int k; + for (k=0;k<maxBands;k++) + { + if (400*(opus_int32)eband5ms[k] > mode->eBands[j]*(opus_int32)mode->Fs/mode->shortMdctSize) + break; + } + if (k>maxBands-1) + allocVectors[i*mode->nbEBands+j] = band_allocation[i*maxBands + maxBands-1]; + else { + opus_int32 a0, a1; + a1 = mode->eBands[j]*(opus_int32)mode->Fs/mode->shortMdctSize - 400*(opus_int32)eband5ms[k-1]; + a0 = 400*(opus_int32)eband5ms[k] - mode->eBands[j]*(opus_int32)mode->Fs/mode->shortMdctSize; + allocVectors[i*mode->nbEBands+j] = (a0*band_allocation[i*maxBands+k-1] + + a1*band_allocation[i*maxBands+k])/(a0+a1); + } + } + } + + /*printf ("\n"); + for (i=0;i<BITALLOC_SIZE;i++) + { + for (j=0;j<mode->nbEBands;j++) + printf ("%d ", allocVectors[i*mode->nbEBands+j]); + printf ("\n"); + } + exit(0);*/ + + mode->allocVectors = allocVectors; +} + +#endif /* CUSTOM_MODES */ + +CELTMode *opus_custom_mode_create(opus_int32 Fs, int frame_size, int *error) +{ + int i; +#ifdef CUSTOM_MODES + CELTMode *mode=NULL; + int res; + opus_val16 *window; + opus_int16 *logN; + int LM; + ALLOC_STACK; +#if !defined(VAR_ARRAYS) && !defined(USE_ALLOCA) + if (global_stack==NULL) + goto failure; +#endif +#endif + +#ifndef CUSTOM_MODES_ONLY + for (i=0;i<TOTAL_MODES;i++) + { + int j; + for (j=0;j<4;j++) + { + if (Fs == static_mode_list[i]->Fs && + (frame_size<<j) == static_mode_list[i]->shortMdctSize*static_mode_list[i]->nbShortMdcts) + { + if (error) + *error = OPUS_OK; + return (CELTMode*)static_mode_list[i]; + } + } + } +#endif /* CUSTOM_MODES_ONLY */ + +#ifndef CUSTOM_MODES + if (error) + *error = OPUS_BAD_ARG; + return NULL; +#else + + /* The good thing here is that permutation of the arguments will automatically be invalid */ + + if (Fs < 8000 || Fs > 96000) + { + if (error) + *error = OPUS_BAD_ARG; + return NULL; + } + if (frame_size < 40 || frame_size > 1024 || frame_size%2!=0) + { + if (error) + *error = OPUS_BAD_ARG; + return NULL; + } + /* Frames of less than 1ms are not supported. */ + if ((opus_int32)frame_size*1000 < Fs) + { + if (error) + *error = OPUS_BAD_ARG; + return NULL; + } + + if ((opus_int32)frame_size*75 >= Fs && (frame_size%16)==0) + { + LM = 3; + } else if ((opus_int32)frame_size*150 >= Fs && (frame_size%8)==0) + { + LM = 2; + } else if ((opus_int32)frame_size*300 >= Fs && (frame_size%4)==0) + { + LM = 1; + } else + { + LM = 0; + } + + /* Shorts longer than 3.3ms are not supported. */ + if ((opus_int32)(frame_size>>LM)*300 > Fs) + { + if (error) + *error = OPUS_BAD_ARG; + return NULL; + } + + mode = opus_alloc(sizeof(CELTMode)); + if (mode==NULL) + goto failure; + mode->Fs = Fs; + + /* Pre/de-emphasis depends on sampling rate. The "standard" pre-emphasis + is defined as A(z) = 1 - 0.85*z^-1 at 48 kHz. Other rates should + approximate that. */ + if(Fs < 12000) /* 8 kHz */ + { + mode->preemph[0] = QCONST16(0.3500061035f, 15); + mode->preemph[1] = -QCONST16(0.1799926758f, 15); + mode->preemph[2] = QCONST16(0.2719968125f, SIG_SHIFT); /* exact 1/preemph[3] */ + mode->preemph[3] = QCONST16(3.6765136719f, 13); + } else if(Fs < 24000) /* 16 kHz */ + { + mode->preemph[0] = QCONST16(0.6000061035f, 15); + mode->preemph[1] = -QCONST16(0.1799926758f, 15); + mode->preemph[2] = QCONST16(0.4424998650f, SIG_SHIFT); /* exact 1/preemph[3] */ + mode->preemph[3] = QCONST16(2.2598876953f, 13); + } else if(Fs < 40000) /* 32 kHz */ + { + mode->preemph[0] = QCONST16(0.7799987793f, 15); + mode->preemph[1] = -QCONST16(0.1000061035f, 15); + mode->preemph[2] = QCONST16(0.7499771125f, SIG_SHIFT); /* exact 1/preemph[3] */ + mode->preemph[3] = QCONST16(1.3333740234f, 13); + } else /* 48 kHz */ + { + mode->preemph[0] = QCONST16(0.8500061035f, 15); + mode->preemph[1] = QCONST16(0.0f, 15); + mode->preemph[2] = QCONST16(1.f, SIG_SHIFT); + mode->preemph[3] = QCONST16(1.f, 13); + } + + mode->maxLM = LM; + mode->nbShortMdcts = 1<<LM; + mode->shortMdctSize = frame_size/mode->nbShortMdcts; + res = (mode->Fs+mode->shortMdctSize)/(2*mode->shortMdctSize); + + mode->eBands = compute_ebands(Fs, mode->shortMdctSize, res, &mode->nbEBands); + if (mode->eBands==NULL) + goto failure; +#if !defined(SMALL_FOOTPRINT) + /* Make sure we don't allocate a band larger than our PVQ table. + 208 should be enough, but let's be paranoid. */ + if ((mode->eBands[mode->nbEBands] - mode->eBands[mode->nbEBands-1])<<LM > + 208) { + goto failure; + } +#endif + + mode->effEBands = mode->nbEBands; + while (mode->eBands[mode->effEBands] > mode->shortMdctSize) + mode->effEBands--; + + /* Overlap must be divisible by 4 */ + mode->overlap = ((mode->shortMdctSize>>2)<<2); + + compute_allocation_table(mode); + if (mode->allocVectors==NULL) + goto failure; + + window = (opus_val16*)opus_alloc(mode->overlap*sizeof(opus_val16)); + if (window==NULL) + goto failure; + +#ifndef OPUS_FIXED_POINT + for (i=0;i<mode->overlap;i++) + window[i] = Q15ONE*sin(.5*M_PI* sin(.5*M_PI*(i+.5)/mode->overlap) * sin(.5*M_PI*(i+.5)/mode->overlap)); +#else + for (i=0;i<mode->overlap;i++) + window[i] = MIN32(32767,floor(.5+32768.*sin(.5*M_PI* sin(.5*M_PI*(i+.5)/mode->overlap) * sin(.5*M_PI*(i+.5)/mode->overlap)))); +#endif + mode->window = window; + + logN = (opus_int16*)opus_alloc(mode->nbEBands*sizeof(opus_int16)); + if (logN==NULL) + goto failure; + + for (i=0;i<mode->nbEBands;i++) + logN[i] = log2_frac(mode->eBands[i+1]-mode->eBands[i], BITRES); + mode->logN = logN; + + compute_pulse_cache(mode, mode->maxLM); + + if (clt_mdct_init(&mode->mdct, 2*mode->shortMdctSize*mode->nbShortMdcts, + mode->maxLM) == 0) + goto failure; + + if (error) + *error = OPUS_OK; + + return mode; +failure: + if (error) + *error = OPUS_ALLOC_FAIL; + if (mode!=NULL) + opus_custom_mode_destroy(mode); + return NULL; +#endif /* !CUSTOM_MODES */ +} + +#ifdef CUSTOM_MODES +void opus_custom_mode_destroy(CELTMode *mode) +{ + if (mode == NULL) + return; +#ifndef CUSTOM_MODES_ONLY + { + int i; + for (i=0;i<TOTAL_MODES;i++) + { + if (mode == static_mode_list[i]) + { + return; + } + } + } +#endif /* CUSTOM_MODES_ONLY */ + opus_free((opus_int16*)mode->eBands); + opus_free((opus_int16*)mode->allocVectors); + + opus_free((opus_val16*)mode->window); + opus_free((opus_int16*)mode->logN); + + opus_free((opus_int16*)mode->cache.index); + opus_free((unsigned char*)mode->cache.bits); + opus_free((unsigned char*)mode->cache.caps); + clt_mdct_clear(&mode->mdct); + + opus_free((CELTMode *)mode); +} +#endif diff --git a/drivers/opus/celt/opus_custom_demo.c b/drivers/opus/celt/opus_custom_demo.c new file mode 100644 index 0000000000..8c7f58b6e6 --- /dev/null +++ b/drivers/opus/celt/opus_custom_demo.c @@ -0,0 +1,210 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef OPUS_HAVE_CONFIG_H +#include "opus_config.h" +#endif + +#include "opus_custom.h" +#include "arch.h" +#include <stdio.h> +#include <stdlib.h> +#include <math.h> +#include <string.h> + +#define MAX_PACKET 1275 + +int main(int argc, char *argv[]) +{ + int err; + char *inFile, *outFile; + FILE *fin, *fout; + OpusCustomMode *mode=NULL; + OpusCustomEncoder *enc; + OpusCustomDecoder *dec; + int len; + opus_int32 frame_size, channels, rate; + int bytes_per_packet; + unsigned char data[MAX_PACKET]; + int complexity; +#if !(defined (OPUS_FIXED_POINT) && !defined(CUSTOM_MODES)) && defined(RESYNTH) + int i; + double rmsd = 0; +#endif + int count = 0; + opus_int32 skip; + opus_int16 *in, *out; + if (argc != 9 && argc != 8 && argc != 7) + { + fprintf (stderr, "Usage: test_opus_custom <rate> <channels> <frame size> " + " <bytes per packet> [<complexity> [packet loss rate]] " + "<input> <output>\n"); + return 1; + } + + rate = (opus_int32)atol(argv[1]); + channels = atoi(argv[2]); + frame_size = atoi(argv[3]); + mode = opus_custom_mode_create(rate, frame_size, NULL); + if (mode == NULL) + { + fprintf(stderr, "failed to create a mode\n"); + return 1; + } + + bytes_per_packet = atoi(argv[4]); + if (bytes_per_packet < 0 || bytes_per_packet > MAX_PACKET) + { + fprintf (stderr, "bytes per packet must be between 0 and %d\n", + MAX_PACKET); + return 1; + } + + inFile = argv[argc-2]; + fin = fopen(inFile, "rb"); + if (!fin) + { + fprintf (stderr, "Could not open input file %s\n", argv[argc-2]); + return 1; + } + outFile = argv[argc-1]; + fout = fopen(outFile, "wb+"); + if (!fout) + { + fprintf (stderr, "Could not open output file %s\n", argv[argc-1]); + fclose(fin); + return 1; + } + + enc = opus_custom_encoder_create(mode, channels, &err); + if (err != 0) + { + fprintf(stderr, "Failed to create the encoder: %s\n", opus_strerror(err)); + fclose(fin); + fclose(fout); + return 1; + } + dec = opus_custom_decoder_create(mode, channels, &err); + if (err != 0) + { + fprintf(stderr, "Failed to create the decoder: %s\n", opus_strerror(err)); + fclose(fin); + fclose(fout); + return 1; + } + opus_custom_decoder_ctl(dec, OPUS_GET_LOOKAHEAD(&skip)); + + if (argc>7) + { + complexity=atoi(argv[5]); + opus_custom_encoder_ctl(enc,OPUS_SET_COMPLEXITY(complexity)); + } + + in = (opus_int16*)malloc(frame_size*channels*sizeof(opus_int16)); + out = (opus_int16*)malloc(frame_size*channels*sizeof(opus_int16)); + + while (!feof(fin)) + { + int ret; + err = fread(in, sizeof(short), frame_size*channels, fin); + if (feof(fin)) + break; + len = opus_custom_encode(enc, in, frame_size, data, bytes_per_packet); + if (len <= 0) + fprintf (stderr, "opus_custom_encode() failed: %s\n", opus_strerror(len)); + + /* This is for simulating bit errors */ +#if 0 + int errors = 0; + int eid = 0; + /* This simulates random bit error */ + for (i=0;i<len*8;i++) + { + if (rand()%atoi(argv[8])==0) + { + if (i<64) + { + errors++; + eid = i; + } + data[i/8] ^= 1<<(7-(i%8)); + } + } + if (errors == 1) + data[eid/8] ^= 1<<(7-(eid%8)); + else if (errors%2 == 1) + data[rand()%8] ^= 1<<rand()%8; +#endif + +#if 1 /* Set to zero to use the encoder's output instead */ + /* This is to simulate packet loss */ + if (argc==9 && rand()%1000<atoi(argv[argc-3])) + /*if (errors && (errors%2==0))*/ + ret = opus_custom_decode(dec, NULL, len, out, frame_size); + else + ret = opus_custom_decode(dec, data, len, out, frame_size); + if (ret < 0) + fprintf(stderr, "opus_custom_decode() failed: %s\n", opus_strerror(ret)); +#else + for (i=0;i<ret*channels;i++) + out[i] = in[i]; +#endif +#if !(defined (OPUS_FIXED_POINT) && !defined(CUSTOM_MODES)) && defined(RESYNTH) + for (i=0;i<ret*channels;i++) + { + rmsd += (in[i]-out[i])*1.0*(in[i]-out[i]); + /*out[i] -= in[i];*/ + } +#endif + count++; + fwrite(out+skip*channels, sizeof(short), (ret-skip)*channels, fout); + skip = 0; + } + PRINT_MIPS(stderr); + + opus_custom_encoder_destroy(enc); + opus_custom_decoder_destroy(dec); + fclose(fin); + fclose(fout); + opus_custom_mode_destroy(mode); + free(in); + free(out); +#if !(defined (OPUS_FIXED_POINT) && !defined(CUSTOM_MODES)) && defined(RESYNTH) + if (rmsd > 0) + { + rmsd = sqrt(rmsd/(1.0*frame_size*channels*count)); + fprintf (stderr, "Error: encoder doesn't match decoder\n"); + fprintf (stderr, "RMS mismatch is %f\n", rmsd); + return 1; + } else { + fprintf (stderr, "Encoder matches decoder!!\n"); + } +#endif + return 0; +} + diff --git a/drivers/opus/celt/opus_modes.h b/drivers/opus/celt/opus_modes.h new file mode 100644 index 0000000000..a1df46265e --- /dev/null +++ b/drivers/opus/celt/opus_modes.h @@ -0,0 +1,83 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Copyright (c) 2008 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef OPUS_MODES_H +#define OPUS_MODES_H + +#include "opus_types.h" +#include "celt.h" +#include "arch.h" +#include "mdct.h" +#include "entenc.h" +#include "entdec.h" + +#define MAX_PERIOD 1024 + +#ifndef OVERLAP +#define OVERLAP(mode) ((mode)->overlap) +#endif + +#ifndef FRAMESIZE +#define FRAMESIZE(mode) ((mode)->mdctSize) +#endif + +typedef struct { + int size; + const opus_int16 *index; + const unsigned char *bits; + const unsigned char *caps; +} PulseCache; + +/** Mode definition (opaque) + @brief Mode definition + */ +struct OpusCustomMode { + opus_int32 Fs; + int overlap; + + int nbEBands; + int effEBands; + opus_val16 preemph[4]; + const opus_int16 *eBands; /**< Definition for each "pseudo-critical band" */ + + int maxLM; + int nbShortMdcts; + int shortMdctSize; + + int nbAllocVectors; /**< Number of lines in the matrix below */ + const unsigned char *allocVectors; /**< Number of bits in each band for several rates */ + const opus_int16 *logN; + + const opus_val16 *window; + celt_mdct_lookup mdct; + PulseCache cache; +}; + + +#endif diff --git a/drivers/opus/celt/os_support.h b/drivers/opus/celt/os_support.h new file mode 100644 index 0000000000..5e47e3cff9 --- /dev/null +++ b/drivers/opus/celt/os_support.h @@ -0,0 +1,92 @@ +/* Copyright (C) 2007 Jean-Marc Valin + + File: os_support.h + This is the (tiny) OS abstraction layer. Aside from math.h, this is the + only place where system headers are allowed. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are + met: + + 1. Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + + 2. Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR + IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES + OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, + INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR + SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, + STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN + ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef OS_SUPPORT_H +#define OS_SUPPORT_H + +#ifdef CUSTOM_SUPPORT +# include "custom_support.h" +#endif + +#include "opus_types.h" +#include "opus_defines.h" + +#include <string.h> +#include <stdio.h> +#include <stdlib.h> + +/** Opus wrapper for malloc(). To do your own dynamic allocation, all you need to do is replace this function and opus_free */ +#ifndef OVERRIDE_OPUS_ALLOC +static OPUS_INLINE void *opus_alloc (size_t size) +{ + return malloc(size); +} +#endif + +/** Same as celt_alloc(), except that the area is only needed inside a CELT call (might cause problem with wideband though) */ +#ifndef OVERRIDE_OPUS_ALLOC_SCRATCH +static OPUS_INLINE void *opus_alloc_scratch (size_t size) +{ + /* Scratch space doesn't need to be cleared */ + return opus_alloc(size); +} +#endif + +/** Opus wrapper for free(). To do your own dynamic allocation, all you need to do is replace this function and opus_alloc */ +#ifndef OVERRIDE_OPUS_FREE +static OPUS_INLINE void opus_free (void *ptr) +{ + free(ptr); +} +#endif + +/** Copy n bytes of memory from src to dst. The 0* term provides compile-time type checking */ +#ifndef OVERRIDE_OPUS_COPY +#define OPUS_COPY(dst, src, n) (memcpy((dst), (src), (n)*sizeof(*(dst)) + 0*((dst)-(src)) )) +#endif + +/** Copy n bytes of memory from src to dst, allowing overlapping regions. The 0* term + provides compile-time type checking */ +#ifndef OVERRIDE_OPUS_MOVE +#define OPUS_MOVE(dst, src, n) (memmove((dst), (src), (n)*sizeof(*(dst)) + 0*((dst)-(src)) )) +#endif + +/** Set n elements of dst to zero, starting at address s */ +#ifndef OVERRIDE_OPUS_CLEAR +#define OPUS_CLEAR(dst, n) (memset((dst), 0, (n)*sizeof(*(dst)))) +#endif + +/*#ifdef __GNUC__ +#pragma GCC poison printf sprintf +#pragma GCC poison malloc free realloc calloc +#endif*/ + +#endif /* OS_SUPPORT_H */ + diff --git a/drivers/opus/celt/pitch.c b/drivers/opus/celt/pitch.c new file mode 100644 index 0000000000..48cd02fb2b --- /dev/null +++ b/drivers/opus/celt/pitch.c @@ -0,0 +1,537 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/** + @file pitch.c + @brief Pitch analysis + */ + +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef OPUS_HAVE_CONFIG_H +#include "opus_config.h" +#endif + +#include "pitch.h" +#include "os_support.h" +#include "opus_modes.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "celt_lpc.h" + +static void find_best_pitch(opus_val32 *xcorr, opus_val16 *y, int len, + int max_pitch, int *best_pitch +#ifdef OPUS_FIXED_POINT + , int yshift, opus_val32 maxcorr +#endif + ) +{ + int i, j; + opus_val32 Syy=1; + opus_val16 best_num[2]; + opus_val32 best_den[2]; +#ifdef OPUS_FIXED_POINT + int xshift; + + xshift = celt_ilog2(maxcorr)-14; +#endif + + best_num[0] = -1; + best_num[1] = -1; + best_den[0] = 0; + best_den[1] = 0; + best_pitch[0] = 0; + best_pitch[1] = 1; + for (j=0;j<len;j++) + Syy = ADD32(Syy, SHR32(MULT16_16(y[j],y[j]), yshift)); + for (i=0;i<max_pitch;i++) + { + if (xcorr[i]>0) + { + opus_val16 num; + opus_val32 xcorr16; + xcorr16 = EXTRACT16(VSHR32(xcorr[i], xshift)); +#ifndef OPUS_FIXED_POINT + /* Considering the range of xcorr16, this should avoid both underflows + and overflows (inf) when squaring xcorr16 */ + xcorr16 *= 1e-12f; +#endif + num = MULT16_16_Q15(xcorr16,xcorr16); + if (MULT16_32_Q15(num,best_den[1]) > MULT16_32_Q15(best_num[1],Syy)) + { + if (MULT16_32_Q15(num,best_den[0]) > MULT16_32_Q15(best_num[0],Syy)) + { + best_num[1] = best_num[0]; + best_den[1] = best_den[0]; + best_pitch[1] = best_pitch[0]; + best_num[0] = num; + best_den[0] = Syy; + best_pitch[0] = i; + } else { + best_num[1] = num; + best_den[1] = Syy; + best_pitch[1] = i; + } + } + } + Syy += SHR32(MULT16_16(y[i+len],y[i+len]),yshift) - SHR32(MULT16_16(y[i],y[i]),yshift); + Syy = MAX32(1, Syy); + } +} + +static void celt_fir5(const opus_val16 *x, + const opus_val16 *num, + opus_val16 *y, + int N, + opus_val16 *mem) +{ + int i; + opus_val16 num0, num1, num2, num3, num4; + opus_val32 mem0, mem1, mem2, mem3, mem4; + num0=num[0]; + num1=num[1]; + num2=num[2]; + num3=num[3]; + num4=num[4]; + mem0=mem[0]; + mem1=mem[1]; + mem2=mem[2]; + mem3=mem[3]; + mem4=mem[4]; + for (i=0;i<N;i++) + { + opus_val32 sum = SHL32(EXTEND32(x[i]), SIG_SHIFT); + sum = MAC16_16(sum,num0,mem0); + sum = MAC16_16(sum,num1,mem1); + sum = MAC16_16(sum,num2,mem2); + sum = MAC16_16(sum,num3,mem3); + sum = MAC16_16(sum,num4,mem4); + mem4 = mem3; + mem3 = mem2; + mem2 = mem1; + mem1 = mem0; + mem0 = x[i]; + y[i] = ROUND16(sum, SIG_SHIFT); + } + mem[0]=mem0; + mem[1]=mem1; + mem[2]=mem2; + mem[3]=mem3; + mem[4]=mem4; +} + + +void pitch_downsample(celt_sig * OPUS_RESTRICT x[], opus_val16 * OPUS_RESTRICT x_lp, + int len, int C, int arch) +{ + int i; + opus_val32 ac[5]; + opus_val16 tmp=Q15ONE; + opus_val16 lpc[4], mem[5]={0,0,0,0,0}; + opus_val16 lpc2[5]; + opus_val16 c1 = QCONST16(.8f,15); +#ifdef OPUS_FIXED_POINT + int shift; + opus_val32 maxabs = celt_maxabs32(x[0], len); + if (C==2) + { + opus_val32 maxabs_1 = celt_maxabs32(x[1], len); + maxabs = MAX32(maxabs, maxabs_1); + } + if (maxabs<1) + maxabs=1; + shift = celt_ilog2(maxabs)-10; + if (shift<0) + shift=0; + if (C==2) + shift++; +#endif + for (i=1;i<len>>1;i++) + x_lp[i] = SHR32(HALF32(HALF32(x[0][(2*i-1)]+x[0][(2*i+1)])+x[0][2*i]), shift); + x_lp[0] = SHR32(HALF32(HALF32(x[0][1])+x[0][0]), shift); + if (C==2) + { + for (i=1;i<len>>1;i++) + x_lp[i] += SHR32(HALF32(HALF32(x[1][(2*i-1)]+x[1][(2*i+1)])+x[1][2*i]), shift); + x_lp[0] += SHR32(HALF32(HALF32(x[1][1])+x[1][0]), shift); + } + + _celt_autocorr(x_lp, ac, NULL, 0, + 4, len>>1, arch); + + /* Noise floor -40 dB */ +#ifdef OPUS_FIXED_POINT + ac[0] += SHR32(ac[0],13); +#else + ac[0] *= 1.0001f; +#endif + /* Lag windowing */ + for (i=1;i<=4;i++) + { + /*ac[i] *= exp(-.5*(2*M_PI*.002*i)*(2*M_PI*.002*i));*/ +#ifdef OPUS_FIXED_POINT + ac[i] -= MULT16_32_Q15(2*i*i, ac[i]); +#else + ac[i] -= ac[i]*(.008f*i)*(.008f*i); +#endif + } + + _celt_lpc(lpc, ac, 4); + for (i=0;i<4;i++) + { + tmp = MULT16_16_Q15(QCONST16(.9f,15), tmp); + lpc[i] = MULT16_16_Q15(lpc[i], tmp); + } + /* Add a zero */ + lpc2[0] = lpc[0] + QCONST16(.8f,SIG_SHIFT); + lpc2[1] = lpc[1] + MULT16_16_Q15(c1,lpc[0]); + lpc2[2] = lpc[2] + MULT16_16_Q15(c1,lpc[1]); + lpc2[3] = lpc[3] + MULT16_16_Q15(c1,lpc[2]); + lpc2[4] = MULT16_16_Q15(c1,lpc[3]); + celt_fir5(x_lp, lpc2, x_lp, len>>1, mem); +} + +#if 0 /* This is a simple version of the pitch correlation that should work + well on DSPs like Blackfin and TI C5x/C6x */ + +#ifdef OPUS_FIXED_POINT +opus_val32 +#else +void +#endif +celt_pitch_xcorr(opus_val16 *x, opus_val16 *y, opus_val32 *xcorr, int len, int max_pitch) +{ + int i, j; +#ifdef OPUS_FIXED_POINT + opus_val32 maxcorr=1; +#endif + for (i=0;i<max_pitch;i++) + { + opus_val32 sum = 0; + for (j=0;j<len;j++) + sum = MAC16_16(sum, x[j],y[i+j]); + xcorr[i] = sum; +#ifdef OPUS_FIXED_POINT + maxcorr = MAX32(maxcorr, sum); +#endif + } +#ifdef OPUS_FIXED_POINT + return maxcorr; +#endif +} + +#else /* Unrolled version of the pitch correlation -- runs faster on x86 and ARM */ + +#ifdef OPUS_FIXED_POINT +opus_val32 +#else +void +#endif +celt_pitch_xcorr_c(const opus_val16 *_x, const opus_val16 *_y, opus_val32 *xcorr, int len, int max_pitch) +{ + int i,j; + /*The EDSP version requires that max_pitch is at least 1, and that _x is + 32-bit aligned. + Since it's hard to put asserts in assembly, put them here.*/ + celt_assert(max_pitch>0); + celt_assert((((unsigned char *)_x-(unsigned char *)NULL)&3)==0); +#ifdef OPUS_FIXED_POINT + opus_val32 maxcorr=1; +#endif + for (i=0;i<max_pitch-3;i+=4) + { + opus_val32 sum[4]={0,0,0,0}; + xcorr_kernel(_x, _y+i, sum, len); + xcorr[i]=sum[0]; + xcorr[i+1]=sum[1]; + xcorr[i+2]=sum[2]; + xcorr[i+3]=sum[3]; +#ifdef OPUS_FIXED_POINT + sum[0] = MAX32(sum[0], sum[1]); + sum[2] = MAX32(sum[2], sum[3]); + sum[0] = MAX32(sum[0], sum[2]); + maxcorr = MAX32(maxcorr, sum[0]); +#endif + } + /* In case max_pitch isn't a multiple of 4, do non-unrolled version. */ + for (;i<max_pitch;i++) + { + opus_val32 sum = 0; + for (j=0;j<len;j++) + sum = MAC16_16(sum, _x[j],_y[i+j]); + xcorr[i] = sum; +#ifdef OPUS_FIXED_POINT + maxcorr = MAX32(maxcorr, sum); +#endif + } +#ifdef OPUS_FIXED_POINT + return maxcorr; +#endif +} + +#endif +void pitch_search(const opus_val16 * OPUS_RESTRICT x_lp, opus_val16 * OPUS_RESTRICT y, + int len, int max_pitch, int *pitch, int arch) +{ + int i, j; + int lag; + int best_pitch[2]={0,0}; + VARDECL(opus_val16, x_lp4); + VARDECL(opus_val16, y_lp4); + VARDECL(opus_val32, xcorr); +#ifdef OPUS_FIXED_POINT + opus_val32 maxcorr; + opus_val32 xmax, ymax; + int shift=0; +#endif + int offset; + + SAVE_STACK; + + celt_assert(len>0); + celt_assert(max_pitch>0); + lag = len+max_pitch; + + ALLOC(x_lp4, len>>2, opus_val16); + ALLOC(y_lp4, lag>>2, opus_val16); + ALLOC(xcorr, max_pitch>>1, opus_val32); + + /* Downsample by 2 again */ + for (j=0;j<len>>2;j++) + x_lp4[j] = x_lp[2*j]; + for (j=0;j<lag>>2;j++) + y_lp4[j] = y[2*j]; + +#ifdef OPUS_FIXED_POINT + xmax = celt_maxabs16(x_lp4, len>>2); + ymax = celt_maxabs16(y_lp4, lag>>2); + shift = celt_ilog2(MAX32(1, MAX32(xmax, ymax)))-11; + if (shift>0) + { + for (j=0;j<len>>2;j++) + x_lp4[j] = SHR16(x_lp4[j], shift); + for (j=0;j<lag>>2;j++) + y_lp4[j] = SHR16(y_lp4[j], shift); + /* Use double the shift for a MAC */ + shift *= 2; + } else { + shift = 0; + } +#endif + + /* Coarse search with 4x decimation */ + +#ifdef OPUS_FIXED_POINT + maxcorr = +#endif + 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 OPUS_FIXED_POINT + , 0, maxcorr +#endif + ); + + /* Finer search with 2x decimation */ +#ifdef OPUS_FIXED_POINT + maxcorr=1; +#endif + for (i=0;i<max_pitch>>1;i++) + { + opus_val32 sum=0; + xcorr[i] = 0; + if (abs(i-2*best_pitch[0])>2 && abs(i-2*best_pitch[1])>2) + continue; + for (j=0;j<len>>1;j++) + sum += SHR32(MULT16_16(x_lp[j],y[i+j]), shift); + xcorr[i] = MAX32(-1, sum); +#ifdef OPUS_FIXED_POINT + maxcorr = MAX32(maxcorr, sum); +#endif + } + find_best_pitch(xcorr, y, len>>1, max_pitch>>1, best_pitch +#ifdef OPUS_FIXED_POINT + , shift+1, maxcorr +#endif + ); + + /* Refine by pseudo-interpolation */ + if (best_pitch[0]>0 && best_pitch[0]<(max_pitch>>1)-1) + { + opus_val32 a, b, c; + a = xcorr[best_pitch[0]-1]; + b = xcorr[best_pitch[0]]; + c = xcorr[best_pitch[0]+1]; + if ((c-a) > MULT16_32_Q15(QCONST16(.7f,15),b-a)) + offset = 1; + else if ((a-c) > MULT16_32_Q15(QCONST16(.7f,15),b-c)) + offset = -1; + else + offset = 0; + } else { + offset = 0; + } + *pitch = 2*best_pitch[0]-offset; + + RESTORE_STACK; +} + +static const int second_check[16] = {0, 0, 3, 2, 3, 2, 5, 2, 3, 2, 3, 2, 5, 2, 3, 2}; +opus_val16 remove_doubling(opus_val16 *x, int maxperiod, int minperiod, + int N, int *T0_, int prev_period, opus_val16 prev_gain) +{ + int k, i, T, T0; + opus_val16 g, g0; + opus_val16 pg; + opus_val32 xy,xx,yy,xy2; + opus_val32 xcorr[3]; + opus_val32 best_xy, best_yy; + int offset; + int minperiod0; + VARDECL(opus_val32, yy_lookup); + SAVE_STACK; + + minperiod0 = minperiod; + maxperiod /= 2; + minperiod /= 2; + *T0_ /= 2; + prev_period /= 2; + N /= 2; + x += maxperiod; + if (*T0_>=maxperiod) + *T0_=maxperiod-1; + + T = T0 = *T0_; + ALLOC(yy_lookup, maxperiod+1, opus_val32); + dual_inner_prod(x, x, x-T0, N, &xx, &xy); + yy_lookup[0] = xx; + yy=xx; + for (i=1;i<=maxperiod;i++) + { + yy = yy+MULT16_16(x[-i],x[-i])-MULT16_16(x[N-i],x[N-i]); + yy_lookup[i] = MAX32(0, yy); + } + yy = yy_lookup[T0]; + best_xy = xy; + best_yy = yy; +#ifdef OPUS_FIXED_POINT + { + opus_val32 x2y2; + int sh, t; + x2y2 = 1+HALF32(MULT32_32_Q31(xx,yy)); + sh = celt_ilog2(x2y2)>>1; + t = VSHR32(x2y2, 2*(sh-7)); + g = g0 = VSHR32(MULT16_32_Q15(celt_rsqrt_norm(t), xy),sh+1); + } +#else + g = g0 = xy/celt_sqrt(1+xx*yy); +#endif + /* Look for any pitch at T/k */ + for (k=2;k<=15;k++) + { + int T1, T1b; + opus_val16 g1; + opus_val16 cont=0; + opus_val16 thresh; + T1 = (2*T0+k)/(2*k); + if (T1 < minperiod) + break; + /* Look for another strong correlation at T1b */ + if (k==2) + { + if (T1+T0>maxperiod) + T1b = T0; + else + T1b = T0+T1; + } else + { + T1b = (2*second_check[k]*T0+k)/(2*k); + } + dual_inner_prod(x, &x[-T1], &x[-T1b], N, &xy, &xy2); + xy += xy2; + yy = yy_lookup[T1] + yy_lookup[T1b]; +#ifdef OPUS_FIXED_POINT + { + opus_val32 x2y2; + int sh, t; + x2y2 = 1+MULT32_32_Q31(xx,yy); + sh = celt_ilog2(x2y2)>>1; + t = VSHR32(x2y2, 2*(sh-7)); + g1 = VSHR32(MULT16_32_Q15(celt_rsqrt_norm(t), xy),sh+1); + } +#else + g1 = xy/celt_sqrt(1+2.f*xx*1.f*yy); +#endif + if (abs(T1-prev_period)<=1) + cont = prev_gain; + else if (abs(T1-prev_period)<=2 && 5*k*k < T0) + cont = HALF32(prev_gain); + else + cont = 0; + thresh = MAX16(QCONST16(.3f,15), MULT16_16_Q15(QCONST16(.7f,15),g0)-cont); + /* Bias against very high pitch (very short period) to avoid false-positives + due to short-term correlation */ + if (T1<3*minperiod) + thresh = MAX16(QCONST16(.4f,15), MULT16_16_Q15(QCONST16(.85f,15),g0)-cont); + else if (T1<2*minperiod) + thresh = MAX16(QCONST16(.5f,15), MULT16_16_Q15(QCONST16(.9f,15),g0)-cont); + if (g1 > thresh) + { + best_xy = xy; + best_yy = yy; + T = T1; + g = g1; + } + } + best_xy = MAX32(0, best_xy); + if (best_yy <= best_xy) + pg = Q15ONE; + else + pg = SHR32(frac_div32(best_xy,best_yy+1),16); + + for (k=0;k<3;k++) + { + int T1 = T+k-1; + xy = 0; + for (i=0;i<N;i++) + xy = MAC16_16(xy, x[i], x[i-T1]); + xcorr[k] = xy; + } + if ((xcorr[2]-xcorr[0]) > MULT16_32_Q15(QCONST16(.7f,15),xcorr[1]-xcorr[0])) + offset = 1; + else if ((xcorr[0]-xcorr[2]) > MULT16_32_Q15(QCONST16(.7f,15),xcorr[1]-xcorr[2])) + offset = -1; + else + offset = 0; + if (pg > g) + pg = g; + *T0_ = 2*T+offset; + + if (*T0_<minperiod0) + *T0_=minperiod0; + RESTORE_STACK; + return pg; +} diff --git a/drivers/opus/celt/pitch.h b/drivers/opus/celt/pitch.h new file mode 100644 index 0000000000..3a7d305425 --- /dev/null +++ b/drivers/opus/celt/pitch.h @@ -0,0 +1,173 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/** + @file pitch.h + @brief Pitch analysis + */ + +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef PITCH_H +#define PITCH_H + +#include "opus_modes.h" +#include "cpu_support.h" + +#if defined(__SSE__) && !defined(OPUS_FIXED_POINT) +#include "x86/pitch_sse.h" +#endif + +#if defined(OPUS_ARM_ASM) && defined(OPUS_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 arch); + +void pitch_search(const opus_val16 * OPUS_RESTRICT x_lp, opus_val16 * OPUS_RESTRICT y, + 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); + +/* OPT: This is the kernel you really want to optimize. It gets used a lot + by the prefilter and by the PLC. */ +#ifndef OVERRIDE_XCORR_KERNEL +static OPUS_INLINE void xcorr_kernel(const opus_val16 * x, const opus_val16 * y, opus_val32 sum[4], int len) +{ + int j; + opus_val16 y_0, y_1, y_2, y_3; + celt_assert(len>=3); + y_3=0; /* gcc doesn't realize that y_3 can't be used uninitialized */ + y_0=*y++; + y_1=*y++; + y_2=*y++; + for (j=0;j<len-3;j+=4) + { + opus_val16 tmp; + tmp = *x++; + y_3=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_0); + sum[1] = MAC16_16(sum[1],tmp,y_1); + sum[2] = MAC16_16(sum[2],tmp,y_2); + sum[3] = MAC16_16(sum[3],tmp,y_3); + tmp=*x++; + y_0=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_1); + sum[1] = MAC16_16(sum[1],tmp,y_2); + sum[2] = MAC16_16(sum[2],tmp,y_3); + sum[3] = MAC16_16(sum[3],tmp,y_0); + tmp=*x++; + y_1=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_2); + sum[1] = MAC16_16(sum[1],tmp,y_3); + sum[2] = MAC16_16(sum[2],tmp,y_0); + sum[3] = MAC16_16(sum[3],tmp,y_1); + tmp=*x++; + y_2=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_3); + sum[1] = MAC16_16(sum[1],tmp,y_0); + sum[2] = MAC16_16(sum[2],tmp,y_1); + sum[3] = MAC16_16(sum[3],tmp,y_2); + } + if (j++<len) + { + opus_val16 tmp = *x++; + y_3=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_0); + sum[1] = MAC16_16(sum[1],tmp,y_1); + sum[2] = MAC16_16(sum[2],tmp,y_2); + sum[3] = MAC16_16(sum[3],tmp,y_3); + } + if (j++<len) + { + opus_val16 tmp=*x++; + y_0=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_1); + sum[1] = MAC16_16(sum[1],tmp,y_2); + sum[2] = MAC16_16(sum[2],tmp,y_3); + sum[3] = MAC16_16(sum[3],tmp,y_0); + } + if (j<len) + { + opus_val16 tmp=*x++; + y_1=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_2); + sum[1] = MAC16_16(sum[1],tmp,y_3); + sum[2] = MAC16_16(sum[2],tmp,y_0); + sum[3] = MAC16_16(sum[3],tmp,y_1); + } +} +#endif /* OVERRIDE_XCORR_KERNEL */ + +#ifndef OVERRIDE_DUAL_INNER_PROD +static OPUS_INLINE void dual_inner_prod(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02, + int N, opus_val32 *xy1, opus_val32 *xy2) +{ + int i; + opus_val32 xy01=0; + opus_val32 xy02=0; + for (i=0;i<N;i++) + { + xy01 = MAC16_16(xy01, x[i], y01[i]); + xy02 = MAC16_16(xy02, x[i], y02[i]); + } + *xy1 = xy01; + *xy2 = xy02; +} +#endif + +#ifdef OPUS_FIXED_POINT +opus_val32 +#else +void +#endif +celt_pitch_xcorr_c(const opus_val16 *_x, const opus_val16 *_y, + opus_val32 *xcorr, int len, int max_pitch); + +#if !defined(OVERRIDE_PITCH_XCORR) +/*Is run-time CPU detection enabled on this platform?*/ +# if defined(OPUS_HAVE_RTCD) +extern +# if defined(OPUS_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 +#endif + +#endif diff --git a/drivers/opus/celt/quant_bands.c b/drivers/opus/celt/quant_bands.c new file mode 100644 index 0000000000..0a170e850d --- /dev/null +++ b/drivers/opus/celt/quant_bands.c @@ -0,0 +1,556 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef OPUS_HAVE_CONFIG_H +#include "opus_config.h" +#endif + +#include "quant_bands.h" +#include "laplace.h" +#include <math.h> +#include "os_support.h" +#include "arch.h" +#include "mathops.h" +#include "stack_alloc.h" +#include "rate.h" + +#ifdef OPUS_FIXED_POINT +/* Mean energy in each band quantized in Q4 */ +const signed char eMeans[25] = { + 103,100, 92, 85, 81, + 77, 72, 70, 78, 75, + 73, 71, 78, 74, 69, + 72, 70, 74, 76, 71, + 60, 60, 60, 60, 60 +}; +#else +/* Mean energy in each band quantized in Q4 and converted back to float */ +const opus_val16 eMeans[25] = { + 6.437500f, 6.250000f, 5.750000f, 5.312500f, 5.062500f, + 4.812500f, 4.500000f, 4.375000f, 4.875000f, 4.687500f, + 4.562500f, 4.437500f, 4.875000f, 4.625000f, 4.312500f, + 4.500000f, 4.375000f, 4.625000f, 4.750000f, 4.437500f, + 3.750000f, 3.750000f, 3.750000f, 3.750000f, 3.750000f +}; +#endif +/* prediction coefficients: 0.9, 0.8, 0.65, 0.5 */ +#ifdef OPUS_FIXED_POINT +static const opus_val16 pred_coef[4] = {29440, 26112, 21248, 16384}; +static const opus_val16 beta_coef[4] = {30147, 22282, 12124, 6554}; +static const opus_val16 beta_intra = 4915; +#else +static const opus_val16 pred_coef[4] = {29440/32768., 26112/32768., 21248/32768., 16384/32768.}; +static const opus_val16 beta_coef[4] = {30147/32768., 22282/32768., 12124/32768., 6554/32768.}; +static const opus_val16 beta_intra = 4915/32768.; +#endif + +/*Parameters of the Laplace-like probability models used for the coarse energy. + There is one pair of parameters for each frame size, prediction type + (inter/intra), and band number. + The first number of each pair is the probability of 0, and the second is the + decay rate, both in Q8 precision.*/ +static const unsigned char e_prob_model[4][2][42] = { + /*120 sample frames.*/ + { + /*Inter*/ + { + 72, 127, 65, 129, 66, 128, 65, 128, 64, 128, 62, 128, 64, 128, + 64, 128, 92, 78, 92, 79, 92, 78, 90, 79, 116, 41, 115, 40, + 114, 40, 132, 26, 132, 26, 145, 17, 161, 12, 176, 10, 177, 11 + }, + /*Intra*/ + { + 24, 179, 48, 138, 54, 135, 54, 132, 53, 134, 56, 133, 55, 132, + 55, 132, 61, 114, 70, 96, 74, 88, 75, 88, 87, 74, 89, 66, + 91, 67, 100, 59, 108, 50, 120, 40, 122, 37, 97, 43, 78, 50 + } + }, + /*240 sample frames.*/ + { + /*Inter*/ + { + 83, 78, 84, 81, 88, 75, 86, 74, 87, 71, 90, 73, 93, 74, + 93, 74, 109, 40, 114, 36, 117, 34, 117, 34, 143, 17, 145, 18, + 146, 19, 162, 12, 165, 10, 178, 7, 189, 6, 190, 8, 177, 9 + }, + /*Intra*/ + { + 23, 178, 54, 115, 63, 102, 66, 98, 69, 99, 74, 89, 71, 91, + 73, 91, 78, 89, 86, 80, 92, 66, 93, 64, 102, 59, 103, 60, + 104, 60, 117, 52, 123, 44, 138, 35, 133, 31, 97, 38, 77, 45 + } + }, + /*480 sample frames.*/ + { + /*Inter*/ + { + 61, 90, 93, 60, 105, 42, 107, 41, 110, 45, 116, 38, 113, 38, + 112, 38, 124, 26, 132, 27, 136, 19, 140, 20, 155, 14, 159, 16, + 158, 18, 170, 13, 177, 10, 187, 8, 192, 6, 175, 9, 159, 10 + }, + /*Intra*/ + { + 21, 178, 59, 110, 71, 86, 75, 85, 84, 83, 91, 66, 88, 73, + 87, 72, 92, 75, 98, 72, 105, 58, 107, 54, 115, 52, 114, 55, + 112, 56, 129, 51, 132, 40, 150, 33, 140, 29, 98, 35, 77, 42 + } + }, + /*960 sample frames.*/ + { + /*Inter*/ + { + 42, 121, 96, 66, 108, 43, 111, 40, 117, 44, 123, 32, 120, 36, + 119, 33, 127, 33, 134, 34, 139, 21, 147, 23, 152, 20, 158, 25, + 154, 26, 166, 21, 173, 16, 184, 13, 184, 10, 150, 13, 139, 15 + }, + /*Intra*/ + { + 22, 178, 63, 114, 74, 82, 84, 83, 92, 82, 103, 62, 96, 72, + 96, 67, 101, 73, 107, 72, 113, 55, 118, 52, 125, 52, 118, 52, + 117, 55, 135, 49, 137, 39, 157, 32, 145, 29, 97, 33, 77, 40 + } + } +}; + +static const unsigned char small_energy_icdf[3]={2,1,0}; + +static opus_val32 loss_distortion(const opus_val16 *eBands, opus_val16 *oldEBands, int start, int end, int len, int C) +{ + int c, i; + opus_val32 dist = 0; + c=0; do { + for (i=start;i<end;i++) + { + opus_val16 d = SUB16(SHR16(eBands[i+c*len], 3), SHR16(oldEBands[i+c*len], 3)); + dist = MAC16_16(dist, d,d); + } + } while (++c<C); + return MIN32(200,SHR32(dist,2*DB_SHIFT-6)); +} + +static int quant_coarse_energy_impl(const CELTMode *m, int start, int end, + const opus_val16 *eBands, opus_val16 *oldEBands, + opus_int32 budget, opus_int32 tell, + const unsigned char *prob_model, opus_val16 *error, ec_enc *enc, + int C, int LM, int intra, opus_val16 max_decay, int lfe) +{ + int i, c; + int badness = 0; + opus_val32 prev[2] = {0,0}; + opus_val16 coef; + opus_val16 beta; + + if (tell+3 <= budget) + ec_enc_bit_logp(enc, intra, 3); + if (intra) + { + coef = 0; + beta = beta_intra; + } else { + beta = beta_coef[LM]; + coef = pred_coef[LM]; + } + + /* Encode at a fixed coarse resolution */ + for (i=start;i<end;i++) + { + c=0; + do { + int bits_left; + int qi, qi0; + opus_val32 q; + opus_val16 x; + opus_val32 f, tmp; + opus_val16 oldE; + opus_val16 decay_bound; + x = eBands[i+c*m->nbEBands]; + oldE = MAX16(-QCONST16(9.f,DB_SHIFT), oldEBands[i+c*m->nbEBands]); +#ifdef OPUS_FIXED_POINT + f = SHL32(EXTEND32(x),7) - PSHR32(MULT16_16(coef,oldE), 8) - prev[c]; + /* Rounding to nearest integer here is really important! */ + qi = (f+QCONST32(.5f,DB_SHIFT+7))>>(DB_SHIFT+7); + decay_bound = EXTRACT16(MAX32(-QCONST16(28.f,DB_SHIFT), + SUB32((opus_val32)oldEBands[i+c*m->nbEBands],max_decay))); +#else + f = x-coef*oldE-prev[c]; + /* Rounding to nearest integer here is really important! */ + qi = (int)floor(.5f+f); + decay_bound = MAX16(-QCONST16(28.f,DB_SHIFT), oldEBands[i+c*m->nbEBands]) - max_decay; +#endif + /* Prevent the energy from going down too quickly (e.g. for bands + that have just one bin) */ + if (qi < 0 && x < decay_bound) + { + qi += (int)SHR16(SUB16(decay_bound,x), DB_SHIFT); + if (qi > 0) + qi = 0; + } + qi0 = qi; + /* If we don't have enough bits to encode all the energy, just assume + something safe. */ + tell = ec_tell(enc); + bits_left = budget-tell-3*C*(end-i); + if (i!=start && bits_left < 30) + { + if (bits_left < 24) + qi = IMIN(1, qi); + if (bits_left < 16) + qi = IMAX(-1, qi); + } + if (lfe && i>=2) + qi = IMIN(qi, 0); + if (budget-tell >= 15) + { + int pi; + pi = 2*IMIN(i,20); + ec_laplace_encode(enc, &qi, + prob_model[pi]<<7, prob_model[pi+1]<<6); + } + else if(budget-tell >= 2) + { + qi = IMAX(-1, IMIN(qi, 1)); + ec_enc_icdf(enc, 2*qi^-(qi<0), small_energy_icdf, 2); + } + else if(budget-tell >= 1) + { + qi = IMIN(0, qi); + ec_enc_bit_logp(enc, -qi, 1); + } + else + qi = -1; + error[i+c*m->nbEBands] = PSHR32(f,7) - SHL16(qi,DB_SHIFT); + badness += abs(qi0-qi); + q = (opus_val32)SHL32(EXTEND32(qi),DB_SHIFT); + + tmp = PSHR32(MULT16_16(coef,oldE),8) + prev[c] + SHL32(q,7); +#ifdef OPUS_FIXED_POINT + tmp = MAX32(-QCONST32(28.f, DB_SHIFT+7), tmp); +#endif + oldEBands[i+c*m->nbEBands] = PSHR32(tmp, 7); + prev[c] = prev[c] + SHL32(q,7) - MULT16_16(beta,PSHR32(q,8)); + } while (++c < C); + } + return lfe ? 0 : badness; +} + +void quant_coarse_energy(const CELTMode *m, int start, int end, int effEnd, + const opus_val16 *eBands, opus_val16 *oldEBands, opus_uint32 budget, + opus_val16 *error, ec_enc *enc, int C, int LM, int nbAvailableBytes, + int force_intra, opus_val32 *delayedIntra, int two_pass, int loss_rate, int lfe) +{ + int intra; + opus_val16 max_decay; + VARDECL(opus_val16, oldEBands_intra); + VARDECL(opus_val16, error_intra); + ec_enc enc_start_state; + opus_uint32 tell; + int badness1=0; + opus_int32 intra_bias; + opus_val32 new_distortion; + SAVE_STACK; + + intra = force_intra || (!two_pass && *delayedIntra>2*C*(end-start) && nbAvailableBytes > (end-start)*C); + intra_bias = (opus_int32)((budget**delayedIntra*loss_rate)/(C*512)); + new_distortion = loss_distortion(eBands, oldEBands, start, effEnd, m->nbEBands, C); + + tell = ec_tell(enc); + if (tell+3 > budget) + two_pass = intra = 0; + + max_decay = QCONST16(16.f,DB_SHIFT); + if (end-start>10) + { +#ifdef OPUS_FIXED_POINT + max_decay = MIN32(max_decay, SHL32(EXTEND32(nbAvailableBytes),DB_SHIFT-3)); +#else + max_decay = MIN32(max_decay, .125f*nbAvailableBytes); +#endif + } + if (lfe) + max_decay=3; + enc_start_state = *enc; + + ALLOC(oldEBands_intra, C*m->nbEBands, opus_val16); + ALLOC(error_intra, C*m->nbEBands, opus_val16); + OPUS_COPY(oldEBands_intra, oldEBands, C*m->nbEBands); + + if (two_pass || intra) + { + badness1 = quant_coarse_energy_impl(m, start, end, eBands, oldEBands_intra, budget, + tell, e_prob_model[LM][1], error_intra, enc, C, LM, 1, max_decay, lfe); + } + + if (!intra) + { + unsigned char *intra_buf; + ec_enc enc_intra_state; + opus_int32 tell_intra; + opus_uint32 nstart_bytes; + opus_uint32 nintra_bytes; + opus_uint32 save_bytes; + int badness2; + VARDECL(unsigned char, intra_bits); + + tell_intra = ec_tell_frac(enc); + + enc_intra_state = *enc; + + nstart_bytes = ec_range_bytes(&enc_start_state); + nintra_bytes = ec_range_bytes(&enc_intra_state); + intra_buf = ec_get_buffer(&enc_intra_state) + nstart_bytes; + save_bytes = nintra_bytes-nstart_bytes; + if (save_bytes == 0) + save_bytes = ALLOC_NONE; + ALLOC(intra_bits, save_bytes, unsigned char); + /* Copy bits from intra bit-stream */ + OPUS_COPY(intra_bits, intra_buf, nintra_bytes - nstart_bytes); + + *enc = enc_start_state; + + badness2 = quant_coarse_energy_impl(m, start, end, eBands, oldEBands, budget, + tell, e_prob_model[LM][intra], error, enc, C, LM, 0, max_decay, lfe); + + if (two_pass && (badness1 < badness2 || (badness1 == badness2 && ((opus_int32)ec_tell_frac(enc))+intra_bias > tell_intra))) + { + *enc = enc_intra_state; + /* Copy intra bits to bit-stream */ + OPUS_COPY(intra_buf, intra_bits, nintra_bytes - nstart_bytes); + OPUS_COPY(oldEBands, oldEBands_intra, C*m->nbEBands); + OPUS_COPY(error, error_intra, C*m->nbEBands); + intra = 1; + } + } else { + OPUS_COPY(oldEBands, oldEBands_intra, C*m->nbEBands); + OPUS_COPY(error, error_intra, C*m->nbEBands); + } + + if (intra) + *delayedIntra = new_distortion; + else + *delayedIntra = ADD32(MULT16_32_Q15(MULT16_16_Q15(pred_coef[LM], pred_coef[LM]),*delayedIntra), + new_distortion); + + RESTORE_STACK; +} + +void quant_fine_energy(const CELTMode *m, int start, int end, opus_val16 *oldEBands, opus_val16 *error, int *fine_quant, ec_enc *enc, int C) +{ + int i, c; + + /* Encode finer resolution */ + for (i=start;i<end;i++) + { + opus_int16 frac = 1<<fine_quant[i]; + if (fine_quant[i] <= 0) + continue; + c=0; + do { + int q2; + opus_val16 offset; +#ifdef OPUS_FIXED_POINT + /* Has to be without rounding */ + q2 = (error[i+c*m->nbEBands]+QCONST16(.5f,DB_SHIFT))>>(DB_SHIFT-fine_quant[i]); +#else + q2 = (int)floor((error[i+c*m->nbEBands]+.5f)*frac); +#endif + if (q2 > frac-1) + q2 = frac-1; + if (q2<0) + q2 = 0; + ec_enc_bits(enc, q2, fine_quant[i]); +#ifdef OPUS_FIXED_POINT + offset = SUB16(SHR32(SHL32(EXTEND32(q2),DB_SHIFT)+QCONST16(.5f,DB_SHIFT),fine_quant[i]),QCONST16(.5f,DB_SHIFT)); +#else + offset = (q2+.5f)*(1<<(14-fine_quant[i]))*(1.f/16384) - .5f; +#endif + oldEBands[i+c*m->nbEBands] += offset; + error[i+c*m->nbEBands] -= offset; + /*printf ("%f ", error[i] - offset);*/ + } while (++c < C); + } +} + +void quant_energy_finalise(const CELTMode *m, int start, int end, opus_val16 *oldEBands, opus_val16 *error, int *fine_quant, int *fine_priority, int bits_left, ec_enc *enc, int C) +{ + int i, prio, c; + + /* Use up the remaining bits */ + for (prio=0;prio<2;prio++) + { + for (i=start;i<end && bits_left>=C ;i++) + { + if (fine_quant[i] >= MAX_FINE_BITS || fine_priority[i]!=prio) + continue; + c=0; + do { + int q2; + opus_val16 offset; + q2 = error[i+c*m->nbEBands]<0 ? 0 : 1; + ec_enc_bits(enc, q2, 1); +#ifdef OPUS_FIXED_POINT + offset = SHR16(SHL16(q2,DB_SHIFT)-QCONST16(.5f,DB_SHIFT),fine_quant[i]+1); +#else + offset = (q2-.5f)*(1<<(14-fine_quant[i]-1))*(1.f/16384); +#endif + oldEBands[i+c*m->nbEBands] += offset; + bits_left--; + } while (++c < C); + } + } +} + +void unquant_coarse_energy(const CELTMode *m, int start, int end, opus_val16 *oldEBands, int intra, ec_dec *dec, int C, int LM) +{ + const unsigned char *prob_model = e_prob_model[LM][intra]; + int i, c; + opus_val32 prev[2] = {0, 0}; + opus_val16 coef; + opus_val16 beta; + opus_int32 budget; + opus_int32 tell; + + if (intra) + { + coef = 0; + beta = beta_intra; + } else { + beta = beta_coef[LM]; + coef = pred_coef[LM]; + } + + budget = dec->storage*8; + + /* Decode at a fixed coarse resolution */ + for (i=start;i<end;i++) + { + c=0; + do { + int qi; + opus_val32 q; + opus_val32 tmp; + /* It would be better to express this invariant as a + test on C at function entry, but that isn't enough + to make the static analyzer happy. */ + celt_assert(c<2); + tell = ec_tell(dec); + if(budget-tell>=15) + { + int pi; + pi = 2*IMIN(i,20); + qi = ec_laplace_decode(dec, + prob_model[pi]<<7, prob_model[pi+1]<<6); + } + else if(budget-tell>=2) + { + qi = ec_dec_icdf(dec, small_energy_icdf, 2); + qi = (qi>>1)^-(qi&1); + } + else if(budget-tell>=1) + { + qi = -ec_dec_bit_logp(dec, 1); + } + else + qi = -1; + q = (opus_val32)SHL32(EXTEND32(qi),DB_SHIFT); + + oldEBands[i+c*m->nbEBands] = MAX16(-QCONST16(9.f,DB_SHIFT), oldEBands[i+c*m->nbEBands]); + tmp = PSHR32(MULT16_16(coef,oldEBands[i+c*m->nbEBands]),8) + prev[c] + SHL32(q,7); +#ifdef OPUS_FIXED_POINT + tmp = MAX32(-QCONST32(28.f, DB_SHIFT+7), tmp); +#endif + oldEBands[i+c*m->nbEBands] = PSHR32(tmp, 7); + prev[c] = prev[c] + SHL32(q,7) - MULT16_16(beta,PSHR32(q,8)); + } while (++c < C); + } +} + +void unquant_fine_energy(const CELTMode *m, int start, int end, opus_val16 *oldEBands, int *fine_quant, ec_dec *dec, int C) +{ + int i, c; + /* Decode finer resolution */ + for (i=start;i<end;i++) + { + if (fine_quant[i] <= 0) + continue; + c=0; + do { + int q2; + opus_val16 offset; + q2 = ec_dec_bits(dec, fine_quant[i]); +#ifdef OPUS_FIXED_POINT + offset = SUB16(SHR32(SHL32(EXTEND32(q2),DB_SHIFT)+QCONST16(.5f,DB_SHIFT),fine_quant[i]),QCONST16(.5f,DB_SHIFT)); +#else + offset = (q2+.5f)*(1<<(14-fine_quant[i]))*(1.f/16384) - .5f; +#endif + oldEBands[i+c*m->nbEBands] += offset; + } while (++c < C); + } +} + +void unquant_energy_finalise(const CELTMode *m, int start, int end, opus_val16 *oldEBands, int *fine_quant, int *fine_priority, int bits_left, ec_dec *dec, int C) +{ + int i, prio, c; + + /* Use up the remaining bits */ + for (prio=0;prio<2;prio++) + { + for (i=start;i<end && bits_left>=C ;i++) + { + if (fine_quant[i] >= MAX_FINE_BITS || fine_priority[i]!=prio) + continue; + c=0; + do { + int q2; + opus_val16 offset; + q2 = ec_dec_bits(dec, 1); +#ifdef OPUS_FIXED_POINT + offset = SHR16(SHL16(q2,DB_SHIFT)-QCONST16(.5f,DB_SHIFT),fine_quant[i]+1); +#else + offset = (q2-.5f)*(1<<(14-fine_quant[i]-1))*(1.f/16384); +#endif + oldEBands[i+c*m->nbEBands] += offset; + bits_left--; + } while (++c < C); + } + } +} + +void amp2Log2(const CELTMode *m, int effEnd, int end, + celt_ener *bandE, opus_val16 *bandLogE, int C) +{ + int c, i; + c=0; + do { + for (i=0;i<effEnd;i++) + bandLogE[i+c*m->nbEBands] = + celt_log2(SHL32(bandE[i+c*m->nbEBands],2)) + - SHL16((opus_val16)eMeans[i],6); + for (i=effEnd;i<end;i++) + bandLogE[c*m->nbEBands+i] = -QCONST16(14.f,DB_SHIFT); + } while (++c < C); +} diff --git a/drivers/opus/celt/quant_bands.h b/drivers/opus/celt/quant_bands.h new file mode 100644 index 0000000000..840df8723f --- /dev/null +++ b/drivers/opus/celt/quant_bands.h @@ -0,0 +1,66 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef QUANT_BANDS +#define QUANT_BANDS + +#include "arch.h" +#include "opus_modes.h" +#include "entenc.h" +#include "entdec.h" +#include "mathops.h" + +#ifdef OPUS_FIXED_POINT +extern const signed char eMeans[25]; +#else +extern const opus_val16 eMeans[25]; +#endif + +void amp2Log2(const CELTMode *m, int effEnd, int end, + celt_ener *bandE, opus_val16 *bandLogE, int C); + +void log2Amp(const CELTMode *m, int start, int end, + celt_ener *eBands, const opus_val16 *oldEBands, int C); + +void quant_coarse_energy(const CELTMode *m, int start, int end, int effEnd, + const opus_val16 *eBands, opus_val16 *oldEBands, opus_uint32 budget, + opus_val16 *error, ec_enc *enc, int C, int LM, + int nbAvailableBytes, int force_intra, opus_val32 *delayedIntra, + int two_pass, int loss_rate, int lfe); + +void quant_fine_energy(const CELTMode *m, int start, int end, opus_val16 *oldEBands, opus_val16 *error, int *fine_quant, ec_enc *enc, int C); + +void quant_energy_finalise(const CELTMode *m, int start, int end, opus_val16 *oldEBands, opus_val16 *error, int *fine_quant, int *fine_priority, int bits_left, ec_enc *enc, int C); + +void unquant_coarse_energy(const CELTMode *m, int start, int end, opus_val16 *oldEBands, int intra, ec_dec *dec, int C, int LM); + +void unquant_fine_energy(const CELTMode *m, int start, int end, opus_val16 *oldEBands, int *fine_quant, ec_dec *dec, int C); + +void unquant_energy_finalise(const CELTMode *m, int start, int end, opus_val16 *oldEBands, int *fine_quant, int *fine_priority, int bits_left, ec_dec *dec, int C); + +#endif /* QUANT_BANDS */ diff --git a/drivers/opus/celt/rate.c b/drivers/opus/celt/rate.c new file mode 100644 index 0000000000..cca585ad95 --- /dev/null +++ b/drivers/opus/celt/rate.c @@ -0,0 +1,638 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef OPUS_HAVE_CONFIG_H +#include "opus_config.h" +#endif + +#include <math.h> +#include "opus_modes.h" +#include "cwrs.h" +#include "arch.h" +#include "os_support.h" + +#include "entcode.h" +#include "rate.h" + +static const unsigned char LOG2_FRAC_TABLE[24]={ + 0, + 8,13, + 16,19,21,23, + 24,26,27,28,29,30,31,32, + 32,33,34,34,35,36,36,37,37 +}; + +#ifdef CUSTOM_MODES + +/*Determines if V(N,K) fits in a 32-bit unsigned integer. + N and K are themselves limited to 15 bits.*/ +static int fits_in32(int _n, int _k) +{ + static const opus_int16 maxN[15] = { + 32767, 32767, 32767, 1476, 283, 109, 60, 40, + 29, 24, 20, 18, 16, 14, 13}; + static const opus_int16 maxK[15] = { + 32767, 32767, 32767, 32767, 1172, 238, 95, 53, + 36, 27, 22, 18, 16, 15, 13}; + if (_n>=14) + { + if (_k>=14) + return 0; + else + return _n <= maxN[_k]; + } else { + return _k <= maxK[_n]; + } +} + +void compute_pulse_cache(CELTMode *m, int LM) +{ + int C; + int i; + int j; + int curr=0; + int nbEntries=0; + int entryN[100], entryK[100], entryI[100]; + const opus_int16 *eBands = m->eBands; + PulseCache *cache = &m->cache; + opus_int16 *cindex; + unsigned char *bits; + unsigned char *cap; + + cindex = (opus_int16 *)opus_alloc(sizeof(cache->index[0])*m->nbEBands*(LM+2)); + cache->index = cindex; + + /* Scan for all unique band sizes */ + for (i=0;i<=LM+1;i++) + { + for (j=0;j<m->nbEBands;j++) + { + int k; + int N = (eBands[j+1]-eBands[j])<<i>>1; + cindex[i*m->nbEBands+j] = -1; + /* Find other bands that have the same size */ + for (k=0;k<=i;k++) + { + int n; + for (n=0;n<m->nbEBands && (k!=i || n<j);n++) + { + if (N == (eBands[n+1]-eBands[n])<<k>>1) + { + cindex[i*m->nbEBands+j] = cindex[k*m->nbEBands+n]; + break; + } + } + } + if (cache->index[i*m->nbEBands+j] == -1 && N!=0) + { + int K; + entryN[nbEntries] = N; + K = 0; + while (fits_in32(N,get_pulses(K+1)) && K<MAX_PSEUDO) + K++; + entryK[nbEntries] = K; + cindex[i*m->nbEBands+j] = curr; + entryI[nbEntries] = curr; + + curr += K+1; + nbEntries++; + } + } + } + bits = (unsigned char *)opus_alloc(sizeof(unsigned char)*curr); + cache->bits = bits; + cache->size = curr; + /* Compute the cache for all unique sizes */ + for (i=0;i<nbEntries;i++) + { + unsigned char *ptr = bits+entryI[i]; + opus_int16 tmp[MAX_PULSES+1]; + get_required_bits(tmp, entryN[i], get_pulses(entryK[i]), BITRES); + for (j=1;j<=entryK[i];j++) + ptr[j] = tmp[get_pulses(j)]-1; + ptr[0] = entryK[i]; + } + + /* Compute the maximum rate for each band at which we'll reliably use as + many bits as we ask for. */ + cache->caps = cap = (unsigned char *)opus_alloc(sizeof(cache->caps[0])*(LM+1)*2*m->nbEBands); + for (i=0;i<=LM;i++) + { + for (C=1;C<=2;C++) + { + for (j=0;j<m->nbEBands;j++) + { + int N0; + int max_bits; + N0 = m->eBands[j+1]-m->eBands[j]; + /* N=1 bands only have a sign bit and fine bits. */ + if (N0<<i == 1) + max_bits = C*(1+MAX_FINE_BITS)<<BITRES; + else + { + const unsigned char *pcache; + opus_int32 num; + opus_int32 den; + int LM0; + int N; + int offset; + int ndof; + int qb; + int k; + LM0 = 0; + /* Even-sized bands bigger than N=2 can be split one more time. + As of commit 44203907 all bands >1 are even, including custom modes.*/ + if (N0 > 2) + { + N0>>=1; + LM0--; + } + /* N0=1 bands can't be split down to N<2. */ + else if (N0 <= 1) + { + LM0=IMIN(i,1); + N0<<=LM0; + } + /* Compute the cost for the lowest-level PVQ of a fully split + band. */ + pcache = bits + cindex[(LM0+1)*m->nbEBands+j]; + max_bits = pcache[pcache[0]]+1; + /* Add in the cost of coding regular splits. */ + N = N0; + for(k=0;k<i-LM0;k++){ + max_bits <<= 1; + /* Offset the number of qtheta bits by log2(N)/2 + + QTHETA_OFFSET compared to their "fair share" of + total/N */ + offset = ((m->logN[j]+((LM0+k)<<BITRES))>>1)-QTHETA_OFFSET; + /* The number of qtheta bits we'll allocate if the remainder + is to be max_bits. + The average measured cost for theta is 0.89701 times qb, + approximated here as 459/512. */ + num=459*(opus_int32)((2*N-1)*offset+max_bits); + den=((opus_int32)(2*N-1)<<9)-459; + qb = IMIN((num+(den>>1))/den, 57); + celt_assert(qb >= 0); + max_bits += qb; + N <<= 1; + } + /* Add in the cost of a stereo split, if necessary. */ + if (C==2) + { + max_bits <<= 1; + offset = ((m->logN[j]+(i<<BITRES))>>1)-(N==2?QTHETA_OFFSET_TWOPHASE:QTHETA_OFFSET); + ndof = 2*N-1-(N==2); + /* The average measured cost for theta with the step PDF is + 0.95164 times qb, approximated here as 487/512. */ + num = (N==2?512:487)*(opus_int32)(max_bits+ndof*offset); + den = ((opus_int32)ndof<<9)-(N==2?512:487); + qb = IMIN((num+(den>>1))/den, (N==2?64:61)); + celt_assert(qb >= 0); + max_bits += qb; + } + /* Add the fine bits we'll use. */ + /* Compensate for the extra DoF in stereo */ + ndof = C*N + ((C==2 && N>2) ? 1 : 0); + /* Offset the number of fine bits by log2(N)/2 + FINE_OFFSET + compared to their "fair share" of total/N */ + offset = ((m->logN[j] + (i<<BITRES))>>1)-FINE_OFFSET; + /* N=2 is the only point that doesn't match the curve */ + if (N==2) + offset += 1<<BITRES>>2; + /* The number of fine bits we'll allocate if the remainder is + to be max_bits. */ + num = max_bits+ndof*offset; + den = (ndof-1)<<BITRES; + qb = IMIN((num+(den>>1))/den, MAX_FINE_BITS); + celt_assert(qb >= 0); + max_bits += C*qb<<BITRES; + } + max_bits = (4*max_bits/(C*((m->eBands[j+1]-m->eBands[j])<<i)))-64; + celt_assert(max_bits >= 0); + celt_assert(max_bits < 256); + *cap++ = (unsigned char)max_bits; + } + } + } +} + +#endif /* CUSTOM_MODES */ + +#define ALLOC_STEPS 6 + +static OPUS_INLINE int interp_bits2pulses(const CELTMode *m, int start, int end, int skip_start, + const int *bits1, const int *bits2, const int *thresh, const int *cap, opus_int32 total, opus_int32 *_balance, + int skip_rsv, int *intensity, int intensity_rsv, int *dual_stereo, int dual_stereo_rsv, int *bits, + int *ebits, int *fine_priority, int C, int LM, ec_ctx *ec, int encode, int prev, int signalBandwidth) +{ + opus_int32 psum; + int lo, hi; + int i, j; + int logM; + int stereo; + int codedBands=-1; + int alloc_floor; + opus_int32 left, percoeff; + int done; + opus_int32 balance; + SAVE_STACK; + + alloc_floor = C<<BITRES; + stereo = C>1; + + logM = LM<<BITRES; + lo = 0; + hi = 1<<ALLOC_STEPS; + for (i=0;i<ALLOC_STEPS;i++) + { + int mid = (lo+hi)>>1; + psum = 0; + done = 0; + for (j=end;j-->start;) + { + int tmp = bits1[j] + (mid*(opus_int32)bits2[j]>>ALLOC_STEPS); + if (tmp >= thresh[j] || done) + { + done = 1; + /* Don't allocate more than we can actually use */ + psum += IMIN(tmp, cap[j]); + } else { + if (tmp >= alloc_floor) + psum += alloc_floor; + } + } + if (psum > total) + hi = mid; + else + lo = mid; + } + psum = 0; + /*printf ("interp bisection gave %d\n", lo);*/ + done = 0; + for (j=end;j-->start;) + { + int tmp = bits1[j] + (lo*bits2[j]>>ALLOC_STEPS); + if (tmp < thresh[j] && !done) + { + if (tmp >= alloc_floor) + tmp = alloc_floor; + else + tmp = 0; + } else + done = 1; + /* Don't allocate more than we can actually use */ + tmp = IMIN(tmp, cap[j]); + bits[j] = tmp; + psum += tmp; + } + + /* Decide which bands to skip, working backwards from the end. */ + for (codedBands=end;;codedBands--) + { + int band_width; + int band_bits; + int rem; + j = codedBands-1; + /* Never skip the first band, nor a band that has been boosted by + dynalloc. + In the first case, we'd be coding a bit to signal we're going to waste + all the other bits. + In the second case, we'd be coding a bit to redistribute all the bits + we just signaled should be cocentrated in this band. */ + if (j<=skip_start) + { + /* Give the bit we reserved to end skipping back. */ + total += skip_rsv; + break; + } + /*Figure out how many left-over bits we would be adding to this band. + This can include bits we've stolen back from higher, skipped bands.*/ + left = total-psum; + percoeff = left/(m->eBands[codedBands]-m->eBands[start]); + left -= (m->eBands[codedBands]-m->eBands[start])*percoeff; + rem = IMAX(left-(m->eBands[j]-m->eBands[start]),0); + band_width = m->eBands[codedBands]-m->eBands[j]; + band_bits = (int)(bits[j] + percoeff*band_width + rem); + /*Only code a skip decision if we're above the threshold for this band. + Otherwise it is force-skipped. + This ensures that we have enough bits to code the skip flag.*/ + if (band_bits >= IMAX(thresh[j], alloc_floor+(1<<BITRES))) + { + if (encode) + { + /*This if() block is the only part of the allocation function that + is not a mandatory part of the bitstream: any bands we choose to + skip here must be explicitly signaled.*/ + /*Choose a threshold with some hysteresis to keep bands from + fluctuating in and out.*/ +#ifdef FUZZING + if ((rand()&0x1) == 0) +#else + if (codedBands<=start+2 || (band_bits > ((j<prev?7:9)*band_width<<LM<<BITRES)>>4 && j<=signalBandwidth)) +#endif + { + ec_enc_bit_logp(ec, 1, 1); + break; + } + ec_enc_bit_logp(ec, 0, 1); + } else if (ec_dec_bit_logp(ec, 1)) { + break; + } + /*We used a bit to skip this band.*/ + psum += 1<<BITRES; + band_bits -= 1<<BITRES; + } + /*Reclaim the bits originally allocated to this band.*/ + psum -= bits[j]+intensity_rsv; + if (intensity_rsv > 0) + intensity_rsv = LOG2_FRAC_TABLE[j-start]; + psum += intensity_rsv; + if (band_bits >= alloc_floor) + { + /*If we have enough for a fine energy bit per channel, use it.*/ + psum += alloc_floor; + bits[j] = alloc_floor; + } else { + /*Otherwise this band gets nothing at all.*/ + bits[j] = 0; + } + } + + celt_assert(codedBands > start); + /* Code the intensity and dual stereo parameters. */ + if (intensity_rsv > 0) + { + if (encode) + { + *intensity = IMIN(*intensity, codedBands); + ec_enc_uint(ec, *intensity-start, codedBands+1-start); + } + else + *intensity = start+ec_dec_uint(ec, codedBands+1-start); + } + else + *intensity = 0; + if (*intensity <= start) + { + total += dual_stereo_rsv; + dual_stereo_rsv = 0; + } + if (dual_stereo_rsv > 0) + { + if (encode) + ec_enc_bit_logp(ec, *dual_stereo, 1); + else + *dual_stereo = ec_dec_bit_logp(ec, 1); + } + else + *dual_stereo = 0; + + /* Allocate the remaining bits */ + left = total-psum; + percoeff = left/(m->eBands[codedBands]-m->eBands[start]); + left -= (m->eBands[codedBands]-m->eBands[start])*percoeff; + for (j=start;j<codedBands;j++) + bits[j] += ((int)percoeff*(m->eBands[j+1]-m->eBands[j])); + for (j=start;j<codedBands;j++) + { + int tmp = (int)IMIN(left, m->eBands[j+1]-m->eBands[j]); + bits[j] += tmp; + left -= tmp; + } + /*for (j=0;j<end;j++)printf("%d ", bits[j]);printf("\n");*/ + + balance = 0; + for (j=start;j<codedBands;j++) + { + int N0, N, den; + int offset; + int NClogN; + opus_int32 excess, bit; + + celt_assert(bits[j] >= 0); + N0 = m->eBands[j+1]-m->eBands[j]; + N=N0<<LM; + bit = (opus_int32)bits[j]+balance; + + if (N>1) + { + excess = MAX32(bit-cap[j],0); + bits[j] = bit-excess; + + /* Compensate for the extra DoF in stereo */ + den=(C*N+ ((C==2 && N>2 && !*dual_stereo && j<*intensity) ? 1 : 0)); + + NClogN = den*(m->logN[j] + logM); + + /* Offset for the number of fine bits by log2(N)/2 + FINE_OFFSET + compared to their "fair share" of total/N */ + offset = (NClogN>>1)-den*FINE_OFFSET; + + /* N=2 is the only point that doesn't match the curve */ + if (N==2) + offset += den<<BITRES>>2; + + /* Changing the offset for allocating the second and third + fine energy bit */ + if (bits[j] + offset < den*2<<BITRES) + offset += NClogN>>2; + else if (bits[j] + offset < den*3<<BITRES) + offset += NClogN>>3; + + /* Divide with rounding */ + ebits[j] = IMAX(0, (bits[j] + offset + (den<<(BITRES-1))) / (den<<BITRES)); + + /* Make sure not to bust */ + if (C*ebits[j] > (bits[j]>>BITRES)) + ebits[j] = bits[j] >> stereo >> BITRES; + + /* More than that is useless because that's about as far as PVQ can go */ + ebits[j] = IMIN(ebits[j], MAX_FINE_BITS); + + /* If we rounded down or capped this band, make it a candidate for the + final fine energy pass */ + fine_priority[j] = ebits[j]*(den<<BITRES) >= bits[j]+offset; + + /* Remove the allocated fine bits; the rest are assigned to PVQ */ + bits[j] -= C*ebits[j]<<BITRES; + + } else { + /* For N=1, all bits go to fine energy except for a single sign bit */ + excess = MAX32(0,bit-(C<<BITRES)); + bits[j] = bit-excess; + ebits[j] = 0; + fine_priority[j] = 1; + } + + /* Fine energy can't take advantage of the re-balancing in + quant_all_bands(). + Instead, do the re-balancing here.*/ + if(excess > 0) + { + int extra_fine; + int extra_bits; + extra_fine = IMIN(excess>>(stereo+BITRES),MAX_FINE_BITS-ebits[j]); + ebits[j] += extra_fine; + extra_bits = extra_fine*C<<BITRES; + fine_priority[j] = extra_bits >= excess-balance; + excess -= extra_bits; + } + balance = excess; + + celt_assert(bits[j] >= 0); + celt_assert(ebits[j] >= 0); + } + /* Save any remaining bits over the cap for the rebalancing in + quant_all_bands(). */ + *_balance = balance; + + /* The skipped bands use all their bits for fine energy. */ + for (;j<end;j++) + { + ebits[j] = bits[j] >> stereo >> BITRES; + celt_assert(C*ebits[j]<<BITRES == bits[j]); + bits[j] = 0; + fine_priority[j] = ebits[j]<1; + } + RESTORE_STACK; + return codedBands; +} + +int compute_allocation(const CELTMode *m, int start, int end, const int *offsets, const int *cap, int alloc_trim, int *intensity, int *dual_stereo, + opus_int32 total, opus_int32 *balance, int *pulses, int *ebits, int *fine_priority, int C, int LM, ec_ctx *ec, int encode, int prev, int signalBandwidth) +{ + int lo, hi, len, j; + int codedBands; + int skip_start; + int skip_rsv; + int intensity_rsv; + int dual_stereo_rsv; + VARDECL(int, bits1); + VARDECL(int, bits2); + VARDECL(int, thresh); + VARDECL(int, trim_offset); + SAVE_STACK; + + total = IMAX(total, 0); + len = m->nbEBands; + skip_start = start; + /* Reserve a bit to signal the end of manually skipped bands. */ + skip_rsv = total >= 1<<BITRES ? 1<<BITRES : 0; + total -= skip_rsv; + /* Reserve bits for the intensity and dual stereo parameters. */ + intensity_rsv = dual_stereo_rsv = 0; + if (C==2) + { + intensity_rsv = LOG2_FRAC_TABLE[end-start]; + if (intensity_rsv>total) + intensity_rsv = 0; + else + { + total -= intensity_rsv; + dual_stereo_rsv = total>=1<<BITRES ? 1<<BITRES : 0; + total -= dual_stereo_rsv; + } + } + ALLOC(bits1, len, int); + ALLOC(bits2, len, int); + ALLOC(thresh, len, int); + ALLOC(trim_offset, len, int); + + for (j=start;j<end;j++) + { + /* Below this threshold, we're sure not to allocate any PVQ bits */ + thresh[j] = IMAX((C)<<BITRES, (3*(m->eBands[j+1]-m->eBands[j])<<LM<<BITRES)>>4); + /* Tilt of the allocation curve */ + trim_offset[j] = C*(m->eBands[j+1]-m->eBands[j])*(alloc_trim-5-LM)*(end-j-1) + *(1<<(LM+BITRES))>>6; + /* Giving less resolution to single-coefficient bands because they get + more benefit from having one coarse value per coefficient*/ + if ((m->eBands[j+1]-m->eBands[j])<<LM==1) + trim_offset[j] -= C<<BITRES; + } + lo = 1; + hi = m->nbAllocVectors - 1; + do + { + int done = 0; + int psum = 0; + int mid = (lo+hi) >> 1; + for (j=end;j-->start;) + { + int bitsj; + int N = m->eBands[j+1]-m->eBands[j]; + bitsj = C*N*m->allocVectors[mid*len+j]<<LM>>2; + if (bitsj > 0) + bitsj = IMAX(0, bitsj + trim_offset[j]); + bitsj += offsets[j]; + if (bitsj >= thresh[j] || done) + { + done = 1; + /* Don't allocate more than we can actually use */ + psum += IMIN(bitsj, cap[j]); + } else { + if (bitsj >= C<<BITRES) + psum += C<<BITRES; + } + } + if (psum > total) + hi = mid - 1; + else + lo = mid + 1; + /*printf ("lo = %d, hi = %d\n", lo, hi);*/ + } + while (lo <= hi); + hi = lo--; + /*printf ("interp between %d and %d\n", lo, hi);*/ + for (j=start;j<end;j++) + { + int bits1j, bits2j; + int N = m->eBands[j+1]-m->eBands[j]; + bits1j = C*N*m->allocVectors[lo*len+j]<<LM>>2; + bits2j = hi>=m->nbAllocVectors ? + cap[j] : C*N*m->allocVectors[hi*len+j]<<LM>>2; + if (bits1j > 0) + bits1j = IMAX(0, bits1j + trim_offset[j]); + if (bits2j > 0) + bits2j = IMAX(0, bits2j + trim_offset[j]); + if (lo > 0) + bits1j += offsets[j]; + bits2j += offsets[j]; + if (offsets[j]>0) + skip_start = j; + bits2j = IMAX(0,bits2j-bits1j); + bits1[j] = bits1j; + bits2[j] = bits2j; + } + codedBands = interp_bits2pulses(m, start, end, skip_start, bits1, bits2, thresh, cap, + total, balance, skip_rsv, intensity, intensity_rsv, dual_stereo, dual_stereo_rsv, + pulses, ebits, fine_priority, C, LM, ec, encode, prev, signalBandwidth); + RESTORE_STACK; + return codedBands; +} + diff --git a/drivers/opus/celt/rate.h b/drivers/opus/celt/rate.h new file mode 100644 index 0000000000..7ced23ea09 --- /dev/null +++ b/drivers/opus/celt/rate.h @@ -0,0 +1,101 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef RATE_H +#define RATE_H + +#define MAX_PSEUDO 40 +#define LOG_MAX_PSEUDO 6 + +#define MAX_PULSES 128 + +#define MAX_FINE_BITS 8 + +#define FINE_OFFSET 21 +#define QTHETA_OFFSET 4 +#define QTHETA_OFFSET_TWOPHASE 16 + +#include "cwrs.h" +#include "opus_modes.h" + +void compute_pulse_cache(CELTMode *m, int LM); + +static OPUS_INLINE int get_pulses(int i) +{ + return i<8 ? i : (8 + (i&7)) << ((i>>3)-1); +} + +static OPUS_INLINE int bits2pulses(const CELTMode *m, int band, int LM, int bits) +{ + int i; + int lo, hi; + const unsigned char *cache; + + LM++; + cache = m->cache.bits + m->cache.index[LM*m->nbEBands+band]; + + lo = 0; + hi = cache[0]; + bits--; + for (i=0;i<LOG_MAX_PSEUDO;i++) + { + int mid = (lo+hi+1)>>1; + /* OPT: Make sure this is implemented with a conditional move */ + if ((int)cache[mid] >= bits) + hi = mid; + else + lo = mid; + } + if (bits- (lo == 0 ? -1 : (int)cache[lo]) <= (int)cache[hi]-bits) + return lo; + else + return hi; +} + +static OPUS_INLINE int pulses2bits(const CELTMode *m, int band, int LM, int pulses) +{ + const unsigned char *cache; + + LM++; + cache = m->cache.bits + m->cache.index[LM*m->nbEBands+band]; + return pulses == 0 ? 0 : cache[pulses]+1; +} + +/** Compute the pulse allocation, i.e. how many pulses will go in each + * band. + @param m mode + @param offsets Requested increase or decrease in the number of bits for + each band + @param total Number of bands + @param pulses Number of pulses per band (returned) + @return Total number of bits allocated +*/ +int compute_allocation(const CELTMode *m, int start, int end, const int *offsets, const int *cap, int alloc_trim, int *intensity, int *dual_stero, + opus_int32 total, opus_int32 *balance, int *pulses, int *ebits, int *fine_priority, int C, int LM, ec_ctx *ec, int encode, int prev, int signalBandwidth); + +#endif diff --git a/drivers/opus/celt/stack_alloc.h b/drivers/opus/celt/stack_alloc.h new file mode 100644 index 0000000000..d500c4dab9 --- /dev/null +++ b/drivers/opus/celt/stack_alloc.h @@ -0,0 +1,182 @@ +/* Copyright (C) 2002-2003 Jean-Marc Valin + Copyright (C) 2007-2009 Xiph.Org Foundation */ +/** + @file stack_alloc.h + @brief Temporary memory allocation on stack +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef STACK_ALLOC_H +#define STACK_ALLOC_H + +#include "opus_types.h" +#include "opus_defines.h" + +#if (!defined (VAR_ARRAYS) && !defined (USE_ALLOCA) && !defined (NONTHREADSAFE_PSEUDOSTACK)) +#define VAR_ARRAYS +#endif + +#ifdef USE_ALLOCA +# ifdef WIN32 +# include <malloc.h> +# else +# ifdef OPUS_HAVE_ALLOCA_H +# include <alloca.h> +# else +# ifdef __linux__ +# include <alloca.h> +# else +# include <stdlib.h> +# endif +# endif +# endif +#endif + +/** + * @def ALIGN(stack, size) + * + * Aligns the stack to a 'size' boundary + * + * @param stack Stack + * @param size New size boundary + */ + +/** + * @def PUSH(stack, size, type) + * + * Allocates 'size' elements of type 'type' on the stack + * + * @param stack Stack + * @param size Number of elements + * @param type Type of element + */ + +/** + * @def VARDECL(var) + * + * Declare variable on stack + * + * @param var Variable to declare + */ + +/** + * @def ALLOC(var, size, type) + * + * Allocate 'size' elements of 'type' on stack + * + * @param var Name of variable to allocate + * @param size Number of elements + * @param type Type of element + */ + +#if defined(VAR_ARRAYS) + +#define VARDECL(type, var) +#define ALLOC(var, size, type) type var[size] +#define SAVE_STACK +#define RESTORE_STACK +#define ALLOC_STACK +/* C99 does not allow VLAs of size zero */ +#define ALLOC_NONE 1 + +#elif defined(USE_ALLOCA) + +#define VARDECL(type, var) type *var + +# ifdef WIN32 +# define ALLOC(var, size, type) var = ((type*)_alloca(sizeof(type)*(size))) +# else +# define ALLOC(var, size, type) var = ((type*)alloca(sizeof(type)*(size))) +# endif + +#define SAVE_STACK +#define RESTORE_STACK +#define ALLOC_STACK +#define ALLOC_NONE 0 + +#else + +#ifdef CELT_C +char *global_stack=0; +#else +extern char *global_stack; +#endif /* CELT_C */ + +#ifdef ENABLE_VALGRIND + +#include <valgrind/memcheck.h> + +#ifdef CELT_C +char *global_stack_top=0; +#else +extern char *global_stack_top; +#endif /* CELT_C */ + +#define ALIGN(stack, size) ((stack) += ((size) - (long)(stack)) & ((size) - 1)) +#define PUSH(stack, size, type) (VALGRIND_MAKE_MEM_NOACCESS(stack, global_stack_top-stack),ALIGN((stack),sizeof(type)/sizeof(char)),VALGRIND_MAKE_MEM_UNDEFINED(stack, ((size)*sizeof(type)/sizeof(char))),(stack)+=(2*(size)*sizeof(type)/sizeof(char)),(type*)((stack)-(2*(size)*sizeof(type)/sizeof(char)))) +#define RESTORE_STACK ((global_stack = _saved_stack),VALGRIND_MAKE_MEM_NOACCESS(global_stack, global_stack_top-global_stack)) +#define ALLOC_STACK char *_saved_stack; ((global_stack = (global_stack==0) ? ((global_stack_top=opus_alloc_scratch(GLOBAL_STACK_SIZE*2)+(GLOBAL_STACK_SIZE*2))-(GLOBAL_STACK_SIZE*2)) : global_stack),VALGRIND_MAKE_MEM_NOACCESS(global_stack, global_stack_top-global_stack)); _saved_stack = global_stack; + +#else + +#define ALIGN(stack, size) ((stack) += ((size) - (long)(stack)) & ((size) - 1)) +#define PUSH(stack, size, type) (ALIGN((stack),sizeof(type)/sizeof(char)),(stack)+=(size)*(sizeof(type)/sizeof(char)),(type*)((stack)-(size)*(sizeof(type)/sizeof(char)))) +#define RESTORE_STACK (global_stack = _saved_stack) +#define ALLOC_STACK char *_saved_stack; (global_stack = (global_stack==0) ? opus_alloc_scratch(GLOBAL_STACK_SIZE) : global_stack); _saved_stack = global_stack; + +#endif /* ENABLE_VALGRIND */ + +#include "os_support.h" +#define VARDECL(type, var) type *var +#define ALLOC(var, size, type) var = PUSH(global_stack, size, type) +#define SAVE_STACK char *_saved_stack = global_stack; +#define ALLOC_NONE 0 + +#endif /* VAR_ARRAYS */ + + +#ifdef ENABLE_VALGRIND + +#include <valgrind/memcheck.h> +#define OPUS_CHECK_ARRAY(ptr, len) VALGRIND_CHECK_MEM_IS_DEFINED(ptr, len*sizeof(*ptr)) +#define OPUS_CHECK_VALUE(value) VALGRIND_CHECK_VALUE_IS_DEFINED(value) +#define OPUS_CHECK_ARRAY_COND(ptr, len) VALGRIND_CHECK_MEM_IS_DEFINED(ptr, len*sizeof(*ptr)) +#define OPUS_CHECK_VALUE_COND(value) VALGRIND_CHECK_VALUE_IS_DEFINED(value) +#define OPUS_PRINT_INT(value) do {fprintf(stderr, #value " = %d at %s:%d\n", value, __FILE__, __LINE__);}while(0) +#define OPUS_FPRINTF fprintf + +#else + +static OPUS_INLINE int _opus_false(void) {return 0;} +#define OPUS_CHECK_ARRAY(ptr, len) _opus_false() +#define OPUS_CHECK_VALUE(value) _opus_false() +#define OPUS_PRINT_INT(value) do{}while(0) +#define OPUS_FPRINTF (void) + +#endif + + +#endif /* STACK_ALLOC_H */ diff --git a/drivers/opus/celt/static_modes_fixed.h b/drivers/opus/celt/static_modes_fixed.h new file mode 100644 index 0000000000..d23e2a66f5 --- /dev/null +++ b/drivers/opus/celt/static_modes_fixed.h @@ -0,0 +1,595 @@ +/* The contents of this file was automatically generated by dump_modes.c + with arguments: 48000 960 + It contains static definitions for some pre-defined modes. */ +#include "opus_modes.h" +#include "rate.h" + +#ifndef DEF_WINDOW120 +#define DEF_WINDOW120 +static const opus_val16 window120[120] = { +2, 20, 55, 108, 178, +266, 372, 494, 635, 792, +966, 1157, 1365, 1590, 1831, +2089, 2362, 2651, 2956, 3276, +3611, 3961, 4325, 4703, 5094, +5499, 5916, 6346, 6788, 7241, +7705, 8179, 8663, 9156, 9657, +10167, 10684, 11207, 11736, 12271, +12810, 13353, 13899, 14447, 14997, +15547, 16098, 16648, 17197, 17744, +18287, 18827, 19363, 19893, 20418, +20936, 21447, 21950, 22445, 22931, +23407, 23874, 24330, 24774, 25208, +25629, 26039, 26435, 26819, 27190, +27548, 27893, 28224, 28541, 28845, +29135, 29411, 29674, 29924, 30160, +30384, 30594, 30792, 30977, 31151, +31313, 31463, 31602, 31731, 31849, +31958, 32057, 32148, 32229, 32303, +32370, 32429, 32481, 32528, 32568, +32604, 32634, 32661, 32683, 32701, +32717, 32729, 32740, 32748, 32754, +32758, 32762, 32764, 32766, 32767, +32767, 32767, 32767, 32767, 32767, +}; +#endif + +#ifndef DEF_LOGN400 +#define DEF_LOGN400 +static const opus_int16 logN400[21] = { +0, 0, 0, 0, 0, 0, 0, 0, 8, 8, 8, 8, 16, 16, 16, 21, 21, 24, 29, 34, 36, }; +#endif + +#ifndef DEF_PULSE_CACHE50 +#define DEF_PULSE_CACHE50 +static const opus_int16 cache_index50[105] = { +-1, -1, -1, -1, -1, -1, -1, -1, 0, 0, 0, 0, 41, 41, 41, +82, 82, 123, 164, 200, 222, 0, 0, 0, 0, 0, 0, 0, 0, 41, +41, 41, 41, 123, 123, 123, 164, 164, 240, 266, 283, 295, 41, 41, 41, +41, 41, 41, 41, 41, 123, 123, 123, 123, 240, 240, 240, 266, 266, 305, +318, 328, 336, 123, 123, 123, 123, 123, 123, 123, 123, 240, 240, 240, 240, +305, 305, 305, 318, 318, 343, 351, 358, 364, 240, 240, 240, 240, 240, 240, +240, 240, 305, 305, 305, 305, 343, 343, 343, 351, 351, 370, 376, 382, 387, +}; +static const unsigned char cache_bits50[392] = { +40, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, +7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, +7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 40, 15, 23, 28, +31, 34, 36, 38, 39, 41, 42, 43, 44, 45, 46, 47, 47, 49, 50, +51, 52, 53, 54, 55, 55, 57, 58, 59, 60, 61, 62, 63, 63, 65, +66, 67, 68, 69, 70, 71, 71, 40, 20, 33, 41, 48, 53, 57, 61, +64, 66, 69, 71, 73, 75, 76, 78, 80, 82, 85, 87, 89, 91, 92, +94, 96, 98, 101, 103, 105, 107, 108, 110, 112, 114, 117, 119, 121, 123, +124, 126, 128, 40, 23, 39, 51, 60, 67, 73, 79, 83, 87, 91, 94, +97, 100, 102, 105, 107, 111, 115, 118, 121, 124, 126, 129, 131, 135, 139, +142, 145, 148, 150, 153, 155, 159, 163, 166, 169, 172, 174, 177, 179, 35, +28, 49, 65, 78, 89, 99, 107, 114, 120, 126, 132, 136, 141, 145, 149, +153, 159, 165, 171, 176, 180, 185, 189, 192, 199, 205, 211, 216, 220, 225, +229, 232, 239, 245, 251, 21, 33, 58, 79, 97, 112, 125, 137, 148, 157, +166, 174, 182, 189, 195, 201, 207, 217, 227, 235, 243, 251, 17, 35, 63, +86, 106, 123, 139, 152, 165, 177, 187, 197, 206, 214, 222, 230, 237, 250, +25, 31, 55, 75, 91, 105, 117, 128, 138, 146, 154, 161, 168, 174, 180, +185, 190, 200, 208, 215, 222, 229, 235, 240, 245, 255, 16, 36, 65, 89, +110, 128, 144, 159, 173, 185, 196, 207, 217, 226, 234, 242, 250, 11, 41, +74, 103, 128, 151, 172, 191, 209, 225, 241, 255, 9, 43, 79, 110, 138, +163, 186, 207, 227, 246, 12, 39, 71, 99, 123, 144, 164, 182, 198, 214, +228, 241, 253, 9, 44, 81, 113, 142, 168, 192, 214, 235, 255, 7, 49, +90, 127, 160, 191, 220, 247, 6, 51, 95, 134, 170, 203, 234, 7, 47, +87, 123, 155, 184, 212, 237, 6, 52, 97, 137, 174, 208, 240, 5, 57, +106, 151, 192, 231, 5, 59, 111, 158, 202, 243, 5, 55, 103, 147, 187, +224, 5, 60, 113, 161, 206, 248, 4, 65, 122, 175, 224, 4, 67, 127, +182, 234, }; +static const unsigned char cache_caps50[168] = { +224, 224, 224, 224, 224, 224, 224, 224, 160, 160, 160, 160, 185, 185, 185, +178, 178, 168, 134, 61, 37, 224, 224, 224, 224, 224, 224, 224, 224, 240, +240, 240, 240, 207, 207, 207, 198, 198, 183, 144, 66, 40, 160, 160, 160, +160, 160, 160, 160, 160, 185, 185, 185, 185, 193, 193, 193, 183, 183, 172, +138, 64, 38, 240, 240, 240, 240, 240, 240, 240, 240, 207, 207, 207, 207, +204, 204, 204, 193, 193, 180, 143, 66, 40, 185, 185, 185, 185, 185, 185, +185, 185, 193, 193, 193, 193, 193, 193, 193, 183, 183, 172, 138, 65, 39, +207, 207, 207, 207, 207, 207, 207, 207, 204, 204, 204, 204, 201, 201, 201, +188, 188, 176, 141, 66, 40, 193, 193, 193, 193, 193, 193, 193, 193, 193, +193, 193, 193, 194, 194, 194, 184, 184, 173, 139, 65, 39, 204, 204, 204, +204, 204, 204, 204, 204, 201, 201, 201, 201, 198, 198, 198, 187, 187, 175, +140, 66, 40, }; +#endif + +#ifndef FFT_TWIDDLES48000_960 +#define FFT_TWIDDLES48000_960 +static const kiss_twiddle_cpx fft_twiddles48000_960[480] = { +{32767, 0}, {32766, -429}, +{32757, -858}, {32743, -1287}, +{32724, -1715}, {32698, -2143}, +{32667, -2570}, {32631, -2998}, +{32588, -3425}, {32541, -3851}, +{32488, -4277}, {32429, -4701}, +{32364, -5125}, {32295, -5548}, +{32219, -5971}, {32138, -6393}, +{32051, -6813}, {31960, -7231}, +{31863, -7650}, {31760, -8067}, +{31652, -8481}, {31539, -8895}, +{31419, -9306}, {31294, -9716}, +{31165, -10126}, {31030, -10532}, +{30889, -10937}, {30743, -11340}, +{30592, -11741}, {30436, -12141}, +{30274, -12540}, {30107, -12935}, +{29936, -13328}, {29758, -13718}, +{29577, -14107}, {29390, -14493}, +{29197, -14875}, {29000, -15257}, +{28797, -15635}, {28590, -16010}, +{28379, -16384}, {28162, -16753}, +{27940, -17119}, {27714, -17484}, +{27482, -17845}, {27246, -18205}, +{27006, -18560}, {26760, -18911}, +{26510, -19260}, {26257, -19606}, +{25997, -19947}, {25734, -20286}, +{25466, -20621}, {25194, -20952}, +{24918, -21281}, {24637, -21605}, +{24353, -21926}, {24063, -22242}, +{23770, -22555}, {23473, -22865}, +{23171, -23171}, {22866, -23472}, +{22557, -23769}, {22244, -24063}, +{21927, -24352}, {21606, -24636}, +{21282, -24917}, {20954, -25194}, +{20622, -25465}, {20288, -25733}, +{19949, -25997}, {19607, -26255}, +{19261, -26509}, {18914, -26760}, +{18561, -27004}, {18205, -27246}, +{17846, -27481}, {17485, -27713}, +{17122, -27940}, {16755, -28162}, +{16385, -28378}, {16012, -28590}, +{15636, -28797}, {15258, -28999}, +{14878, -29197}, {14494, -29389}, +{14108, -29576}, {13720, -29757}, +{13329, -29934}, {12937, -30107}, +{12540, -30274}, {12142, -30435}, +{11744, -30592}, {11342, -30743}, +{10939, -30889}, {10534, -31030}, +{10127, -31164}, {9718, -31294}, +{9307, -31418}, {8895, -31537}, +{8482, -31652}, {8067, -31759}, +{7650, -31862}, {7233, -31960}, +{6815, -32051}, {6393, -32138}, +{5973, -32219}, {5549, -32294}, +{5127, -32364}, {4703, -32429}, +{4278, -32487}, {3852, -32541}, +{3426, -32588}, {2999, -32630}, +{2572, -32667}, {2144, -32698}, +{1716, -32724}, {1287, -32742}, +{860, -32757}, {430, -32766}, +{0, -32767}, {-429, -32766}, +{-858, -32757}, {-1287, -32743}, +{-1715, -32724}, {-2143, -32698}, +{-2570, -32667}, {-2998, -32631}, +{-3425, -32588}, {-3851, -32541}, +{-4277, -32488}, {-4701, -32429}, +{-5125, -32364}, {-5548, -32295}, +{-5971, -32219}, {-6393, -32138}, +{-6813, -32051}, {-7231, -31960}, +{-7650, -31863}, {-8067, -31760}, +{-8481, -31652}, {-8895, -31539}, +{-9306, -31419}, {-9716, -31294}, +{-10126, -31165}, {-10532, -31030}, +{-10937, -30889}, {-11340, -30743}, +{-11741, -30592}, {-12141, -30436}, +{-12540, -30274}, {-12935, -30107}, +{-13328, -29936}, {-13718, -29758}, +{-14107, -29577}, {-14493, -29390}, +{-14875, -29197}, {-15257, -29000}, +{-15635, -28797}, {-16010, -28590}, +{-16384, -28379}, {-16753, -28162}, +{-17119, -27940}, {-17484, -27714}, +{-17845, -27482}, {-18205, -27246}, +{-18560, -27006}, {-18911, -26760}, +{-19260, -26510}, {-19606, -26257}, +{-19947, -25997}, {-20286, -25734}, +{-20621, -25466}, {-20952, -25194}, +{-21281, -24918}, {-21605, -24637}, +{-21926, -24353}, {-22242, -24063}, +{-22555, -23770}, {-22865, -23473}, +{-23171, -23171}, {-23472, -22866}, +{-23769, -22557}, {-24063, -22244}, +{-24352, -21927}, {-24636, -21606}, +{-24917, -21282}, {-25194, -20954}, +{-25465, -20622}, {-25733, -20288}, +{-25997, -19949}, {-26255, -19607}, +{-26509, -19261}, {-26760, -18914}, +{-27004, -18561}, {-27246, -18205}, +{-27481, -17846}, {-27713, -17485}, +{-27940, -17122}, {-28162, -16755}, +{-28378, -16385}, {-28590, -16012}, +{-28797, -15636}, {-28999, -15258}, +{-29197, -14878}, {-29389, -14494}, +{-29576, -14108}, {-29757, -13720}, +{-29934, -13329}, {-30107, -12937}, +{-30274, -12540}, {-30435, -12142}, +{-30592, -11744}, {-30743, -11342}, +{-30889, -10939}, {-31030, -10534}, +{-31164, -10127}, {-31294, -9718}, +{-31418, -9307}, {-31537, -8895}, +{-31652, -8482}, {-31759, -8067}, +{-31862, -7650}, {-31960, -7233}, +{-32051, -6815}, {-32138, -6393}, +{-32219, -5973}, {-32294, -5549}, +{-32364, -5127}, {-32429, -4703}, +{-32487, -4278}, {-32541, -3852}, +{-32588, -3426}, {-32630, -2999}, +{-32667, -2572}, {-32698, -2144}, +{-32724, -1716}, {-32742, -1287}, +{-32757, -860}, {-32766, -430}, +{-32767, 0}, {-32766, 429}, +{-32757, 858}, {-32743, 1287}, +{-32724, 1715}, {-32698, 2143}, +{-32667, 2570}, {-32631, 2998}, +{-32588, 3425}, {-32541, 3851}, +{-32488, 4277}, {-32429, 4701}, +{-32364, 5125}, {-32295, 5548}, +{-32219, 5971}, {-32138, 6393}, +{-32051, 6813}, {-31960, 7231}, +{-31863, 7650}, {-31760, 8067}, +{-31652, 8481}, {-31539, 8895}, +{-31419, 9306}, {-31294, 9716}, +{-31165, 10126}, {-31030, 10532}, +{-30889, 10937}, {-30743, 11340}, +{-30592, 11741}, {-30436, 12141}, +{-30274, 12540}, {-30107, 12935}, +{-29936, 13328}, {-29758, 13718}, +{-29577, 14107}, {-29390, 14493}, +{-29197, 14875}, {-29000, 15257}, +{-28797, 15635}, {-28590, 16010}, +{-28379, 16384}, {-28162, 16753}, +{-27940, 17119}, {-27714, 17484}, +{-27482, 17845}, {-27246, 18205}, +{-27006, 18560}, {-26760, 18911}, +{-26510, 19260}, {-26257, 19606}, +{-25997, 19947}, {-25734, 20286}, +{-25466, 20621}, {-25194, 20952}, +{-24918, 21281}, {-24637, 21605}, +{-24353, 21926}, {-24063, 22242}, +{-23770, 22555}, {-23473, 22865}, +{-23171, 23171}, {-22866, 23472}, +{-22557, 23769}, {-22244, 24063}, +{-21927, 24352}, {-21606, 24636}, +{-21282, 24917}, {-20954, 25194}, +{-20622, 25465}, {-20288, 25733}, +{-19949, 25997}, {-19607, 26255}, +{-19261, 26509}, {-18914, 26760}, +{-18561, 27004}, {-18205, 27246}, +{-17846, 27481}, {-17485, 27713}, +{-17122, 27940}, {-16755, 28162}, +{-16385, 28378}, {-16012, 28590}, +{-15636, 28797}, {-15258, 28999}, +{-14878, 29197}, {-14494, 29389}, +{-14108, 29576}, {-13720, 29757}, +{-13329, 29934}, {-12937, 30107}, +{-12540, 30274}, {-12142, 30435}, +{-11744, 30592}, {-11342, 30743}, +{-10939, 30889}, {-10534, 31030}, +{-10127, 31164}, {-9718, 31294}, +{-9307, 31418}, {-8895, 31537}, +{-8482, 31652}, {-8067, 31759}, +{-7650, 31862}, {-7233, 31960}, +{-6815, 32051}, {-6393, 32138}, +{-5973, 32219}, {-5549, 32294}, +{-5127, 32364}, {-4703, 32429}, +{-4278, 32487}, {-3852, 32541}, +{-3426, 32588}, {-2999, 32630}, +{-2572, 32667}, {-2144, 32698}, +{-1716, 32724}, {-1287, 32742}, +{-860, 32757}, {-430, 32766}, +{0, 32767}, {429, 32766}, +{858, 32757}, {1287, 32743}, +{1715, 32724}, {2143, 32698}, +{2570, 32667}, {2998, 32631}, +{3425, 32588}, {3851, 32541}, +{4277, 32488}, {4701, 32429}, +{5125, 32364}, {5548, 32295}, +{5971, 32219}, {6393, 32138}, +{6813, 32051}, {7231, 31960}, +{7650, 31863}, {8067, 31760}, +{8481, 31652}, {8895, 31539}, +{9306, 31419}, {9716, 31294}, +{10126, 31165}, {10532, 31030}, +{10937, 30889}, {11340, 30743}, +{11741, 30592}, {12141, 30436}, +{12540, 30274}, {12935, 30107}, +{13328, 29936}, {13718, 29758}, +{14107, 29577}, {14493, 29390}, +{14875, 29197}, {15257, 29000}, +{15635, 28797}, {16010, 28590}, +{16384, 28379}, {16753, 28162}, +{17119, 27940}, {17484, 27714}, +{17845, 27482}, {18205, 27246}, +{18560, 27006}, {18911, 26760}, +{19260, 26510}, {19606, 26257}, +{19947, 25997}, {20286, 25734}, +{20621, 25466}, {20952, 25194}, +{21281, 24918}, {21605, 24637}, +{21926, 24353}, {22242, 24063}, +{22555, 23770}, {22865, 23473}, +{23171, 23171}, {23472, 22866}, +{23769, 22557}, {24063, 22244}, +{24352, 21927}, {24636, 21606}, +{24917, 21282}, {25194, 20954}, +{25465, 20622}, {25733, 20288}, +{25997, 19949}, {26255, 19607}, +{26509, 19261}, {26760, 18914}, +{27004, 18561}, {27246, 18205}, +{27481, 17846}, {27713, 17485}, +{27940, 17122}, {28162, 16755}, +{28378, 16385}, {28590, 16012}, +{28797, 15636}, {28999, 15258}, +{29197, 14878}, {29389, 14494}, +{29576, 14108}, {29757, 13720}, +{29934, 13329}, {30107, 12937}, +{30274, 12540}, {30435, 12142}, +{30592, 11744}, {30743, 11342}, +{30889, 10939}, {31030, 10534}, +{31164, 10127}, {31294, 9718}, +{31418, 9307}, {31537, 8895}, +{31652, 8482}, {31759, 8067}, +{31862, 7650}, {31960, 7233}, +{32051, 6815}, {32138, 6393}, +{32219, 5973}, {32294, 5549}, +{32364, 5127}, {32429, 4703}, +{32487, 4278}, {32541, 3852}, +{32588, 3426}, {32630, 2999}, +{32667, 2572}, {32698, 2144}, +{32724, 1716}, {32742, 1287}, +{32757, 860}, {32766, 430}, +}; +#ifndef FFT_BITREV480 +#define FFT_BITREV480 +static const opus_int16 fft_bitrev480[480] = { +0, 120, 240, 360, 30, 150, 270, 390, 60, 180, 300, 420, 90, 210, 330, +450, 15, 135, 255, 375, 45, 165, 285, 405, 75, 195, 315, 435, 105, 225, +345, 465, 5, 125, 245, 365, 35, 155, 275, 395, 65, 185, 305, 425, 95, +215, 335, 455, 20, 140, 260, 380, 50, 170, 290, 410, 80, 200, 320, 440, +110, 230, 350, 470, 10, 130, 250, 370, 40, 160, 280, 400, 70, 190, 310, +430, 100, 220, 340, 460, 25, 145, 265, 385, 55, 175, 295, 415, 85, 205, +325, 445, 115, 235, 355, 475, 1, 121, 241, 361, 31, 151, 271, 391, 61, +181, 301, 421, 91, 211, 331, 451, 16, 136, 256, 376, 46, 166, 286, 406, +76, 196, 316, 436, 106, 226, 346, 466, 6, 126, 246, 366, 36, 156, 276, +396, 66, 186, 306, 426, 96, 216, 336, 456, 21, 141, 261, 381, 51, 171, +291, 411, 81, 201, 321, 441, 111, 231, 351, 471, 11, 131, 251, 371, 41, +161, 281, 401, 71, 191, 311, 431, 101, 221, 341, 461, 26, 146, 266, 386, +56, 176, 296, 416, 86, 206, 326, 446, 116, 236, 356, 476, 2, 122, 242, +362, 32, 152, 272, 392, 62, 182, 302, 422, 92, 212, 332, 452, 17, 137, +257, 377, 47, 167, 287, 407, 77, 197, 317, 437, 107, 227, 347, 467, 7, +127, 247, 367, 37, 157, 277, 397, 67, 187, 307, 427, 97, 217, 337, 457, +22, 142, 262, 382, 52, 172, 292, 412, 82, 202, 322, 442, 112, 232, 352, +472, 12, 132, 252, 372, 42, 162, 282, 402, 72, 192, 312, 432, 102, 222, +342, 462, 27, 147, 267, 387, 57, 177, 297, 417, 87, 207, 327, 447, 117, +237, 357, 477, 3, 123, 243, 363, 33, 153, 273, 393, 63, 183, 303, 423, +93, 213, 333, 453, 18, 138, 258, 378, 48, 168, 288, 408, 78, 198, 318, +438, 108, 228, 348, 468, 8, 128, 248, 368, 38, 158, 278, 398, 68, 188, +308, 428, 98, 218, 338, 458, 23, 143, 263, 383, 53, 173, 293, 413, 83, +203, 323, 443, 113, 233, 353, 473, 13, 133, 253, 373, 43, 163, 283, 403, +73, 193, 313, 433, 103, 223, 343, 463, 28, 148, 268, 388, 58, 178, 298, +418, 88, 208, 328, 448, 118, 238, 358, 478, 4, 124, 244, 364, 34, 154, +274, 394, 64, 184, 304, 424, 94, 214, 334, 454, 19, 139, 259, 379, 49, +169, 289, 409, 79, 199, 319, 439, 109, 229, 349, 469, 9, 129, 249, 369, +39, 159, 279, 399, 69, 189, 309, 429, 99, 219, 339, 459, 24, 144, 264, +384, 54, 174, 294, 414, 84, 204, 324, 444, 114, 234, 354, 474, 14, 134, +254, 374, 44, 164, 284, 404, 74, 194, 314, 434, 104, 224, 344, 464, 29, +149, 269, 389, 59, 179, 299, 419, 89, 209, 329, 449, 119, 239, 359, 479, +}; +#endif + +#ifndef FFT_BITREV240 +#define FFT_BITREV240 +static const opus_int16 fft_bitrev240[240] = { +0, 60, 120, 180, 15, 75, 135, 195, 30, 90, 150, 210, 45, 105, 165, +225, 5, 65, 125, 185, 20, 80, 140, 200, 35, 95, 155, 215, 50, 110, +170, 230, 10, 70, 130, 190, 25, 85, 145, 205, 40, 100, 160, 220, 55, +115, 175, 235, 1, 61, 121, 181, 16, 76, 136, 196, 31, 91, 151, 211, +46, 106, 166, 226, 6, 66, 126, 186, 21, 81, 141, 201, 36, 96, 156, +216, 51, 111, 171, 231, 11, 71, 131, 191, 26, 86, 146, 206, 41, 101, +161, 221, 56, 116, 176, 236, 2, 62, 122, 182, 17, 77, 137, 197, 32, +92, 152, 212, 47, 107, 167, 227, 7, 67, 127, 187, 22, 82, 142, 202, +37, 97, 157, 217, 52, 112, 172, 232, 12, 72, 132, 192, 27, 87, 147, +207, 42, 102, 162, 222, 57, 117, 177, 237, 3, 63, 123, 183, 18, 78, +138, 198, 33, 93, 153, 213, 48, 108, 168, 228, 8, 68, 128, 188, 23, +83, 143, 203, 38, 98, 158, 218, 53, 113, 173, 233, 13, 73, 133, 193, +28, 88, 148, 208, 43, 103, 163, 223, 58, 118, 178, 238, 4, 64, 124, +184, 19, 79, 139, 199, 34, 94, 154, 214, 49, 109, 169, 229, 9, 69, +129, 189, 24, 84, 144, 204, 39, 99, 159, 219, 54, 114, 174, 234, 14, +74, 134, 194, 29, 89, 149, 209, 44, 104, 164, 224, 59, 119, 179, 239, +}; +#endif + +#ifndef FFT_BITREV120 +#define FFT_BITREV120 +static const opus_int16 fft_bitrev120[120] = { +0, 30, 60, 90, 15, 45, 75, 105, 5, 35, 65, 95, 20, 50, 80, +110, 10, 40, 70, 100, 25, 55, 85, 115, 1, 31, 61, 91, 16, 46, +76, 106, 6, 36, 66, 96, 21, 51, 81, 111, 11, 41, 71, 101, 26, +56, 86, 116, 2, 32, 62, 92, 17, 47, 77, 107, 7, 37, 67, 97, +22, 52, 82, 112, 12, 42, 72, 102, 27, 57, 87, 117, 3, 33, 63, +93, 18, 48, 78, 108, 8, 38, 68, 98, 23, 53, 83, 113, 13, 43, +73, 103, 28, 58, 88, 118, 4, 34, 64, 94, 19, 49, 79, 109, 9, +39, 69, 99, 24, 54, 84, 114, 14, 44, 74, 104, 29, 59, 89, 119, +}; +#endif + +#ifndef FFT_BITREV60 +#define FFT_BITREV60 +static const opus_int16 fft_bitrev60[60] = { +0, 15, 30, 45, 5, 20, 35, 50, 10, 25, 40, 55, 1, 16, 31, +46, 6, 21, 36, 51, 11, 26, 41, 56, 2, 17, 32, 47, 7, 22, +37, 52, 12, 27, 42, 57, 3, 18, 33, 48, 8, 23, 38, 53, 13, +28, 43, 58, 4, 19, 34, 49, 9, 24, 39, 54, 14, 29, 44, 59, +}; +#endif + +#ifndef FFT_STATE48000_960_0 +#define FFT_STATE48000_960_0 +static const kiss_fft_state fft_state48000_960_0 = { +480, /* nfft */ +-1, /* shift */ +{4, 120, 4, 30, 2, 15, 3, 5, 5, 1, 0, 0, 0, 0, 0, 0, }, /* factors */ +fft_bitrev480, /* bitrev */ +fft_twiddles48000_960, /* bitrev */ +}; +#endif + +#ifndef FFT_STATE48000_960_1 +#define FFT_STATE48000_960_1 +static const kiss_fft_state fft_state48000_960_1 = { +240, /* nfft */ +1, /* shift */ +{4, 60, 4, 15, 3, 5, 5, 1, 0, 0, 0, 0, 0, 0, 0, 0, }, /* factors */ +fft_bitrev240, /* bitrev */ +fft_twiddles48000_960, /* bitrev */ +}; +#endif + +#ifndef FFT_STATE48000_960_2 +#define FFT_STATE48000_960_2 +static const kiss_fft_state fft_state48000_960_2 = { +120, /* nfft */ +2, /* shift */ +{4, 30, 2, 15, 3, 5, 5, 1, 0, 0, 0, 0, 0, 0, 0, 0, }, /* factors */ +fft_bitrev120, /* bitrev */ +fft_twiddles48000_960, /* bitrev */ +}; +#endif + +#ifndef FFT_STATE48000_960_3 +#define FFT_STATE48000_960_3 +static const kiss_fft_state fft_state48000_960_3 = { +60, /* nfft */ +3, /* shift */ +{4, 15, 3, 5, 5, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, }, /* factors */ +fft_bitrev60, /* bitrev */ +fft_twiddles48000_960, /* bitrev */ +}; +#endif + +#endif + +#ifndef MDCT_TWIDDLES960 +#define MDCT_TWIDDLES960 +static const opus_val16 mdct_twiddles960[481] = { +32767, 32767, 32767, 32767, 32766, +32763, 32762, 32759, 32757, 32753, +32751, 32747, 32743, 32738, 32733, +32729, 32724, 32717, 32711, 32705, +32698, 32690, 32683, 32676, 32667, +32658, 32650, 32640, 32631, 32620, +32610, 32599, 32588, 32577, 32566, +32554, 32541, 32528, 32515, 32502, +32487, 32474, 32459, 32444, 32429, +32413, 32397, 32381, 32364, 32348, +32331, 32313, 32294, 32277, 32257, +32239, 32219, 32200, 32180, 32159, +32138, 32118, 32096, 32074, 32051, +32029, 32006, 31984, 31960, 31936, +31912, 31888, 31863, 31837, 31812, +31786, 31760, 31734, 31707, 31679, +31652, 31624, 31596, 31567, 31539, +31508, 31479, 31450, 31419, 31388, +31357, 31326, 31294, 31262, 31230, +31198, 31164, 31131, 31097, 31063, +31030, 30994, 30959, 30924, 30889, +30853, 30816, 30779, 30743, 30705, +30668, 30629, 30592, 30553, 30515, +30475, 30435, 30396, 30356, 30315, +30274, 30233, 30191, 30149, 30107, +30065, 30022, 29979, 29936, 29891, +29847, 29803, 29758, 29713, 29668, +29622, 29577, 29529, 29483, 29436, +29390, 29341, 29293, 29246, 29197, +29148, 29098, 29050, 29000, 28949, +28899, 28848, 28797, 28746, 28694, +28642, 28590, 28537, 28485, 28432, +28378, 28324, 28271, 28217, 28162, +28106, 28051, 27995, 27940, 27884, +27827, 27770, 27713, 27657, 27598, +27540, 27481, 27423, 27365, 27305, +27246, 27187, 27126, 27066, 27006, +26945, 26883, 26822, 26760, 26698, +26636, 26574, 26510, 26448, 26383, +26320, 26257, 26191, 26127, 26062, +25997, 25931, 25866, 25800, 25734, +25667, 25601, 25533, 25466, 25398, +25330, 25262, 25194, 25125, 25056, +24987, 24917, 24848, 24778, 24707, +24636, 24566, 24495, 24424, 24352, +24280, 24208, 24135, 24063, 23990, +23917, 23842, 23769, 23695, 23622, +23546, 23472, 23398, 23322, 23246, +23171, 23095, 23018, 22942, 22866, +22788, 22711, 22634, 22557, 22478, +22400, 22322, 22244, 22165, 22085, +22006, 21927, 21846, 21766, 21687, +21606, 21524, 21443, 21363, 21282, +21199, 21118, 21035, 20954, 20870, +20788, 20705, 20621, 20538, 20455, +20371, 20286, 20202, 20118, 20034, +19947, 19863, 19777, 19692, 19606, +19520, 19434, 19347, 19260, 19174, +19088, 18999, 18911, 18825, 18737, +18648, 18560, 18472, 18384, 18294, +18205, 18116, 18025, 17936, 17846, +17757, 17666, 17576, 17485, 17395, +17303, 17212, 17122, 17030, 16937, +16846, 16755, 16662, 16569, 16477, +16385, 16291, 16198, 16105, 16012, +15917, 15824, 15730, 15636, 15541, +15447, 15352, 15257, 15162, 15067, +14973, 14875, 14781, 14685, 14589, +14493, 14396, 14300, 14204, 14107, +14010, 13914, 13815, 13718, 13621, +13524, 13425, 13328, 13230, 13133, +13033, 12935, 12836, 12738, 12638, +12540, 12441, 12341, 12241, 12142, +12044, 11943, 11843, 11744, 11643, +11542, 11442, 11342, 11241, 11139, +11039, 10939, 10836, 10736, 10635, +10534, 10431, 10330, 10228, 10127, +10024, 9921, 9820, 9718, 9614, +9512, 9410, 9306, 9204, 9101, +8998, 8895, 8791, 8689, 8585, +8481, 8377, 8274, 8171, 8067, +7962, 7858, 7753, 7650, 7545, +7441, 7336, 7231, 7129, 7023, +6917, 6813, 6709, 6604, 6498, +6393, 6288, 6182, 6077, 5973, +5867, 5760, 5656, 5549, 5445, +5339, 5232, 5127, 5022, 4914, +4809, 4703, 4596, 4490, 4384, +4278, 4171, 4065, 3958, 3852, +3745, 3640, 3532, 3426, 3318, +3212, 3106, 2998, 2891, 2786, +2679, 2570, 2465, 2358, 2251, +2143, 2037, 1929, 1823, 1715, +1609, 1501, 1393, 1287, 1180, +1073, 964, 858, 751, 644, +535, 429, 322, 214, 107, +0, }; +#endif + +static const CELTMode mode48000_960_120 = { +48000, /* Fs */ +120, /* overlap */ +21, /* nbEBands */ +21, /* effEBands */ +{27853, 0, 4096, 8192, }, /* preemph */ +eband5ms, /* eBands */ +3, /* maxLM */ +8, /* nbShortMdcts */ +120, /* shortMdctSize */ +11, /* nbAllocVectors */ +band_allocation, /* allocVectors */ +logN400, /* logN */ +window120, /* window */ +{1920, 3, {&fft_state48000_960_0, &fft_state48000_960_1, &fft_state48000_960_2, &fft_state48000_960_3, }, mdct_twiddles960}, /* mdct */ +{392, cache_index50, cache_bits50, cache_caps50}, /* cache */ +}; + +/* List of all the available modes */ +#define TOTAL_MODES 1 +static const CELTMode * const static_mode_list[TOTAL_MODES] = { +&mode48000_960_120, +}; diff --git a/drivers/opus/celt/static_modes_float.h b/drivers/opus/celt/static_modes_float.h new file mode 100644 index 0000000000..fe6bb4c8a3 --- /dev/null +++ b/drivers/opus/celt/static_modes_float.h @@ -0,0 +1,599 @@ +/* The contents of this file was automatically generated by dump_modes.c + with arguments: 48000 960 + It contains static definitions for some pre-defined modes. */ +#include "opus_modes.h" +#include "rate.h" + +#ifndef DEF_WINDOW120 +#define DEF_WINDOW120 +static const opus_val16 window120[120] = { +6.7286966e-05f, 0.00060551348f, 0.0016815970f, 0.0032947962f, 0.0054439943f, +0.0081276923f, 0.011344001f, 0.015090633f, 0.019364886f, 0.024163635f, +0.029483315f, 0.035319905f, 0.041668911f, 0.048525347f, 0.055883718f, +0.063737999f, 0.072081616f, 0.080907428f, 0.090207705f, 0.099974111f, +0.11019769f, 0.12086883f, 0.13197729f, 0.14351214f, 0.15546177f, +0.16781389f, 0.18055550f, 0.19367290f, 0.20715171f, 0.22097682f, +0.23513243f, 0.24960208f, 0.26436860f, 0.27941419f, 0.29472040f, +0.31026818f, 0.32603788f, 0.34200931f, 0.35816177f, 0.37447407f, +0.39092462f, 0.40749142f, 0.42415215f, 0.44088423f, 0.45766484f, +0.47447104f, 0.49127978f, 0.50806798f, 0.52481261f, 0.54149077f, +0.55807973f, 0.57455701f, 0.59090049f, 0.60708841f, 0.62309951f, +0.63891306f, 0.65450896f, 0.66986776f, 0.68497077f, 0.69980010f, +0.71433873f, 0.72857055f, 0.74248043f, 0.75605424f, 0.76927895f, +0.78214257f, 0.79463430f, 0.80674445f, 0.81846456f, 0.82978733f, +0.84070669f, 0.85121779f, 0.86131698f, 0.87100183f, 0.88027111f, +0.88912479f, 0.89756398f, 0.90559094f, 0.91320904f, 0.92042270f, +0.92723738f, 0.93365955f, 0.93969656f, 0.94535671f, 0.95064907f, +0.95558353f, 0.96017067f, 0.96442171f, 0.96834849f, 0.97196334f, +0.97527906f, 0.97830883f, 0.98106616f, 0.98356480f, 0.98581869f, +0.98784191f, 0.98964856f, 0.99125274f, 0.99266849f, 0.99390969f, +0.99499004f, 0.99592297f, 0.99672162f, 0.99739874f, 0.99796667f, +0.99843728f, 0.99882195f, 0.99913147f, 0.99937606f, 0.99956527f, +0.99970802f, 0.99981248f, 0.99988613f, 0.99993565f, 0.99996697f, +0.99998518f, 0.99999457f, 0.99999859f, 0.99999982f, 1.0000000f, +}; +#endif + +#ifndef DEF_LOGN400 +#define DEF_LOGN400 +static const opus_int16 logN400[21] = { +0, 0, 0, 0, 0, 0, 0, 0, 8, 8, 8, 8, 16, 16, 16, 21, 21, 24, 29, 34, 36, }; +#endif + +#ifndef DEF_PULSE_CACHE50 +#define DEF_PULSE_CACHE50 +static const opus_int16 cache_index50[105] = { +-1, -1, -1, -1, -1, -1, -1, -1, 0, 0, 0, 0, 41, 41, 41, +82, 82, 123, 164, 200, 222, 0, 0, 0, 0, 0, 0, 0, 0, 41, +41, 41, 41, 123, 123, 123, 164, 164, 240, 266, 283, 295, 41, 41, 41, +41, 41, 41, 41, 41, 123, 123, 123, 123, 240, 240, 240, 266, 266, 305, +318, 328, 336, 123, 123, 123, 123, 123, 123, 123, 123, 240, 240, 240, 240, +305, 305, 305, 318, 318, 343, 351, 358, 364, 240, 240, 240, 240, 240, 240, +240, 240, 305, 305, 305, 305, 343, 343, 343, 351, 351, 370, 376, 382, 387, +}; +static const unsigned char cache_bits50[392] = { +40, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, +7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, +7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 40, 15, 23, 28, +31, 34, 36, 38, 39, 41, 42, 43, 44, 45, 46, 47, 47, 49, 50, +51, 52, 53, 54, 55, 55, 57, 58, 59, 60, 61, 62, 63, 63, 65, +66, 67, 68, 69, 70, 71, 71, 40, 20, 33, 41, 48, 53, 57, 61, +64, 66, 69, 71, 73, 75, 76, 78, 80, 82, 85, 87, 89, 91, 92, +94, 96, 98, 101, 103, 105, 107, 108, 110, 112, 114, 117, 119, 121, 123, +124, 126, 128, 40, 23, 39, 51, 60, 67, 73, 79, 83, 87, 91, 94, +97, 100, 102, 105, 107, 111, 115, 118, 121, 124, 126, 129, 131, 135, 139, +142, 145, 148, 150, 153, 155, 159, 163, 166, 169, 172, 174, 177, 179, 35, +28, 49, 65, 78, 89, 99, 107, 114, 120, 126, 132, 136, 141, 145, 149, +153, 159, 165, 171, 176, 180, 185, 189, 192, 199, 205, 211, 216, 220, 225, +229, 232, 239, 245, 251, 21, 33, 58, 79, 97, 112, 125, 137, 148, 157, +166, 174, 182, 189, 195, 201, 207, 217, 227, 235, 243, 251, 17, 35, 63, +86, 106, 123, 139, 152, 165, 177, 187, 197, 206, 214, 222, 230, 237, 250, +25, 31, 55, 75, 91, 105, 117, 128, 138, 146, 154, 161, 168, 174, 180, +185, 190, 200, 208, 215, 222, 229, 235, 240, 245, 255, 16, 36, 65, 89, +110, 128, 144, 159, 173, 185, 196, 207, 217, 226, 234, 242, 250, 11, 41, +74, 103, 128, 151, 172, 191, 209, 225, 241, 255, 9, 43, 79, 110, 138, +163, 186, 207, 227, 246, 12, 39, 71, 99, 123, 144, 164, 182, 198, 214, +228, 241, 253, 9, 44, 81, 113, 142, 168, 192, 214, 235, 255, 7, 49, +90, 127, 160, 191, 220, 247, 6, 51, 95, 134, 170, 203, 234, 7, 47, +87, 123, 155, 184, 212, 237, 6, 52, 97, 137, 174, 208, 240, 5, 57, +106, 151, 192, 231, 5, 59, 111, 158, 202, 243, 5, 55, 103, 147, 187, +224, 5, 60, 113, 161, 206, 248, 4, 65, 122, 175, 224, 4, 67, 127, +182, 234, }; +static const unsigned char cache_caps50[168] = { +224, 224, 224, 224, 224, 224, 224, 224, 160, 160, 160, 160, 185, 185, 185, +178, 178, 168, 134, 61, 37, 224, 224, 224, 224, 224, 224, 224, 224, 240, +240, 240, 240, 207, 207, 207, 198, 198, 183, 144, 66, 40, 160, 160, 160, +160, 160, 160, 160, 160, 185, 185, 185, 185, 193, 193, 193, 183, 183, 172, +138, 64, 38, 240, 240, 240, 240, 240, 240, 240, 240, 207, 207, 207, 207, +204, 204, 204, 193, 193, 180, 143, 66, 40, 185, 185, 185, 185, 185, 185, +185, 185, 193, 193, 193, 193, 193, 193, 193, 183, 183, 172, 138, 65, 39, +207, 207, 207, 207, 207, 207, 207, 207, 204, 204, 204, 204, 201, 201, 201, +188, 188, 176, 141, 66, 40, 193, 193, 193, 193, 193, 193, 193, 193, 193, +193, 193, 193, 194, 194, 194, 184, 184, 173, 139, 65, 39, 204, 204, 204, +204, 204, 204, 204, 204, 201, 201, 201, 201, 198, 198, 198, 187, 187, 175, +140, 66, 40, }; +#endif + +#ifndef FFT_TWIDDLES48000_960 +#define FFT_TWIDDLES48000_960 +static const kiss_twiddle_cpx fft_twiddles48000_960[480] = { +{1.0000000f, -0.0000000f}, {0.99991433f, -0.013089596f}, +{0.99965732f, -0.026176948f}, {0.99922904f, -0.039259816f}, +{0.99862953f, -0.052335956f}, {0.99785892f, -0.065403129f}, +{0.99691733f, -0.078459096f}, {0.99580493f, -0.091501619f}, +{0.99452190f, -0.10452846f}, {0.99306846f, -0.11753740f}, +{0.99144486f, -0.13052619f}, {0.98965139f, -0.14349262f}, +{0.98768834f, -0.15643447f}, {0.98555606f, -0.16934950f}, +{0.98325491f, -0.18223553f}, {0.98078528f, -0.19509032f}, +{0.97814760f, -0.20791169f}, {0.97534232f, -0.22069744f}, +{0.97236992f, -0.23344536f}, {0.96923091f, -0.24615329f}, +{0.96592583f, -0.25881905f}, {0.96245524f, -0.27144045f}, +{0.95881973f, -0.28401534f}, {0.95501994f, -0.29654157f}, +{0.95105652f, -0.30901699f}, {0.94693013f, -0.32143947f}, +{0.94264149f, -0.33380686f}, {0.93819134f, -0.34611706f}, +{0.93358043f, -0.35836795f}, {0.92880955f, -0.37055744f}, +{0.92387953f, -0.38268343f}, {0.91879121f, -0.39474386f}, +{0.91354546f, -0.40673664f}, {0.90814317f, -0.41865974f}, +{0.90258528f, -0.43051110f}, {0.89687274f, -0.44228869f}, +{0.89100652f, -0.45399050f}, {0.88498764f, -0.46561452f}, +{0.87881711f, -0.47715876f}, {0.87249601f, -0.48862124f}, +{0.86602540f, -0.50000000f}, {0.85940641f, -0.51129309f}, +{0.85264016f, -0.52249856f}, {0.84572782f, -0.53361452f}, +{0.83867057f, -0.54463904f}, {0.83146961f, -0.55557023f}, +{0.82412619f, -0.56640624f}, {0.81664156f, -0.57714519f}, +{0.80901699f, -0.58778525f}, {0.80125381f, -0.59832460f}, +{0.79335334f, -0.60876143f}, {0.78531693f, -0.61909395f}, +{0.77714596f, -0.62932039f}, {0.76884183f, -0.63943900f}, +{0.76040597f, -0.64944805f}, {0.75183981f, -0.65934582f}, +{0.74314483f, -0.66913061f}, {0.73432251f, -0.67880075f}, +{0.72537437f, -0.68835458f}, {0.71630194f, -0.69779046f}, +{0.70710678f, -0.70710678f}, {0.69779046f, -0.71630194f}, +{0.68835458f, -0.72537437f}, {0.67880075f, -0.73432251f}, +{0.66913061f, -0.74314483f}, {0.65934582f, -0.75183981f}, +{0.64944805f, -0.76040597f}, {0.63943900f, -0.76884183f}, +{0.62932039f, -0.77714596f}, {0.61909395f, -0.78531693f}, +{0.60876143f, -0.79335334f}, {0.59832460f, -0.80125381f}, +{0.58778525f, -0.80901699f}, {0.57714519f, -0.81664156f}, +{0.56640624f, -0.82412619f}, {0.55557023f, -0.83146961f}, +{0.54463904f, -0.83867057f}, {0.53361452f, -0.84572782f}, +{0.52249856f, -0.85264016f}, {0.51129309f, -0.85940641f}, +{0.50000000f, -0.86602540f}, {0.48862124f, -0.87249601f}, +{0.47715876f, -0.87881711f}, {0.46561452f, -0.88498764f}, +{0.45399050f, -0.89100652f}, {0.44228869f, -0.89687274f}, +{0.43051110f, -0.90258528f}, {0.41865974f, -0.90814317f}, +{0.40673664f, -0.91354546f}, {0.39474386f, -0.91879121f}, +{0.38268343f, -0.92387953f}, {0.37055744f, -0.92880955f}, +{0.35836795f, -0.93358043f}, {0.34611706f, -0.93819134f}, +{0.33380686f, -0.94264149f}, {0.32143947f, -0.94693013f}, +{0.30901699f, -0.95105652f}, {0.29654157f, -0.95501994f}, +{0.28401534f, -0.95881973f}, {0.27144045f, -0.96245524f}, +{0.25881905f, -0.96592583f}, {0.24615329f, -0.96923091f}, +{0.23344536f, -0.97236992f}, {0.22069744f, -0.97534232f}, +{0.20791169f, -0.97814760f}, {0.19509032f, -0.98078528f}, +{0.18223553f, -0.98325491f}, {0.16934950f, -0.98555606f}, +{0.15643447f, -0.98768834f}, {0.14349262f, -0.98965139f}, +{0.13052619f, -0.99144486f}, {0.11753740f, -0.99306846f}, +{0.10452846f, -0.99452190f}, {0.091501619f, -0.99580493f}, +{0.078459096f, -0.99691733f}, {0.065403129f, -0.99785892f}, +{0.052335956f, -0.99862953f}, {0.039259816f, -0.99922904f}, +{0.026176948f, -0.99965732f}, {0.013089596f, -0.99991433f}, +{6.1230318e-17f, -1.0000000f}, {-0.013089596f, -0.99991433f}, +{-0.026176948f, -0.99965732f}, {-0.039259816f, -0.99922904f}, +{-0.052335956f, -0.99862953f}, {-0.065403129f, -0.99785892f}, +{-0.078459096f, -0.99691733f}, {-0.091501619f, -0.99580493f}, +{-0.10452846f, -0.99452190f}, {-0.11753740f, -0.99306846f}, +{-0.13052619f, -0.99144486f}, {-0.14349262f, -0.98965139f}, +{-0.15643447f, -0.98768834f}, {-0.16934950f, -0.98555606f}, +{-0.18223553f, -0.98325491f}, {-0.19509032f, -0.98078528f}, +{-0.20791169f, -0.97814760f}, {-0.22069744f, -0.97534232f}, +{-0.23344536f, -0.97236992f}, {-0.24615329f, -0.96923091f}, +{-0.25881905f, -0.96592583f}, {-0.27144045f, -0.96245524f}, +{-0.28401534f, -0.95881973f}, {-0.29654157f, -0.95501994f}, +{-0.30901699f, -0.95105652f}, {-0.32143947f, -0.94693013f}, +{-0.33380686f, -0.94264149f}, {-0.34611706f, -0.93819134f}, +{-0.35836795f, -0.93358043f}, {-0.37055744f, -0.92880955f}, +{-0.38268343f, -0.92387953f}, {-0.39474386f, -0.91879121f}, +{-0.40673664f, -0.91354546f}, {-0.41865974f, -0.90814317f}, +{-0.43051110f, -0.90258528f}, {-0.44228869f, -0.89687274f}, +{-0.45399050f, -0.89100652f}, {-0.46561452f, -0.88498764f}, +{-0.47715876f, -0.87881711f}, {-0.48862124f, -0.87249601f}, +{-0.50000000f, -0.86602540f}, {-0.51129309f, -0.85940641f}, +{-0.52249856f, -0.85264016f}, {-0.53361452f, -0.84572782f}, +{-0.54463904f, -0.83867057f}, {-0.55557023f, -0.83146961f}, +{-0.56640624f, -0.82412619f}, {-0.57714519f, -0.81664156f}, +{-0.58778525f, -0.80901699f}, {-0.59832460f, -0.80125381f}, +{-0.60876143f, -0.79335334f}, {-0.61909395f, -0.78531693f}, +{-0.62932039f, -0.77714596f}, {-0.63943900f, -0.76884183f}, +{-0.64944805f, -0.76040597f}, {-0.65934582f, -0.75183981f}, +{-0.66913061f, -0.74314483f}, {-0.67880075f, -0.73432251f}, +{-0.68835458f, -0.72537437f}, {-0.69779046f, -0.71630194f}, +{-0.70710678f, -0.70710678f}, {-0.71630194f, -0.69779046f}, +{-0.72537437f, -0.68835458f}, {-0.73432251f, -0.67880075f}, +{-0.74314483f, -0.66913061f}, {-0.75183981f, -0.65934582f}, +{-0.76040597f, -0.64944805f}, {-0.76884183f, -0.63943900f}, +{-0.77714596f, -0.62932039f}, {-0.78531693f, -0.61909395f}, +{-0.79335334f, -0.60876143f}, {-0.80125381f, -0.59832460f}, +{-0.80901699f, -0.58778525f}, {-0.81664156f, -0.57714519f}, +{-0.82412619f, -0.56640624f}, {-0.83146961f, -0.55557023f}, +{-0.83867057f, -0.54463904f}, {-0.84572782f, -0.53361452f}, +{-0.85264016f, -0.52249856f}, {-0.85940641f, -0.51129309f}, +{-0.86602540f, -0.50000000f}, {-0.87249601f, -0.48862124f}, +{-0.87881711f, -0.47715876f}, {-0.88498764f, -0.46561452f}, +{-0.89100652f, -0.45399050f}, {-0.89687274f, -0.44228869f}, +{-0.90258528f, -0.43051110f}, {-0.90814317f, -0.41865974f}, +{-0.91354546f, -0.40673664f}, {-0.91879121f, -0.39474386f}, +{-0.92387953f, -0.38268343f}, {-0.92880955f, -0.37055744f}, +{-0.93358043f, -0.35836795f}, {-0.93819134f, -0.34611706f}, +{-0.94264149f, -0.33380686f}, {-0.94693013f, -0.32143947f}, +{-0.95105652f, -0.30901699f}, {-0.95501994f, -0.29654157f}, +{-0.95881973f, -0.28401534f}, {-0.96245524f, -0.27144045f}, +{-0.96592583f, -0.25881905f}, {-0.96923091f, -0.24615329f}, +{-0.97236992f, -0.23344536f}, {-0.97534232f, -0.22069744f}, +{-0.97814760f, -0.20791169f}, {-0.98078528f, -0.19509032f}, +{-0.98325491f, -0.18223553f}, {-0.98555606f, -0.16934950f}, +{-0.98768834f, -0.15643447f}, {-0.98965139f, -0.14349262f}, +{-0.99144486f, -0.13052619f}, {-0.99306846f, -0.11753740f}, +{-0.99452190f, -0.10452846f}, {-0.99580493f, -0.091501619f}, +{-0.99691733f, -0.078459096f}, {-0.99785892f, -0.065403129f}, +{-0.99862953f, -0.052335956f}, {-0.99922904f, -0.039259816f}, +{-0.99965732f, -0.026176948f}, {-0.99991433f, -0.013089596f}, +{-1.0000000f, -1.2246064e-16f}, {-0.99991433f, 0.013089596f}, +{-0.99965732f, 0.026176948f}, {-0.99922904f, 0.039259816f}, +{-0.99862953f, 0.052335956f}, {-0.99785892f, 0.065403129f}, +{-0.99691733f, 0.078459096f}, {-0.99580493f, 0.091501619f}, +{-0.99452190f, 0.10452846f}, {-0.99306846f, 0.11753740f}, +{-0.99144486f, 0.13052619f}, {-0.98965139f, 0.14349262f}, +{-0.98768834f, 0.15643447f}, {-0.98555606f, 0.16934950f}, +{-0.98325491f, 0.18223553f}, {-0.98078528f, 0.19509032f}, +{-0.97814760f, 0.20791169f}, {-0.97534232f, 0.22069744f}, +{-0.97236992f, 0.23344536f}, {-0.96923091f, 0.24615329f}, +{-0.96592583f, 0.25881905f}, {-0.96245524f, 0.27144045f}, +{-0.95881973f, 0.28401534f}, {-0.95501994f, 0.29654157f}, +{-0.95105652f, 0.30901699f}, {-0.94693013f, 0.32143947f}, +{-0.94264149f, 0.33380686f}, {-0.93819134f, 0.34611706f}, +{-0.93358043f, 0.35836795f}, {-0.92880955f, 0.37055744f}, +{-0.92387953f, 0.38268343f}, {-0.91879121f, 0.39474386f}, +{-0.91354546f, 0.40673664f}, {-0.90814317f, 0.41865974f}, +{-0.90258528f, 0.43051110f}, {-0.89687274f, 0.44228869f}, +{-0.89100652f, 0.45399050f}, {-0.88498764f, 0.46561452f}, +{-0.87881711f, 0.47715876f}, {-0.87249601f, 0.48862124f}, +{-0.86602540f, 0.50000000f}, {-0.85940641f, 0.51129309f}, +{-0.85264016f, 0.52249856f}, {-0.84572782f, 0.53361452f}, +{-0.83867057f, 0.54463904f}, {-0.83146961f, 0.55557023f}, +{-0.82412619f, 0.56640624f}, {-0.81664156f, 0.57714519f}, +{-0.80901699f, 0.58778525f}, {-0.80125381f, 0.59832460f}, +{-0.79335334f, 0.60876143f}, {-0.78531693f, 0.61909395f}, +{-0.77714596f, 0.62932039f}, {-0.76884183f, 0.63943900f}, +{-0.76040597f, 0.64944805f}, {-0.75183981f, 0.65934582f}, +{-0.74314483f, 0.66913061f}, {-0.73432251f, 0.67880075f}, +{-0.72537437f, 0.68835458f}, {-0.71630194f, 0.69779046f}, +{-0.70710678f, 0.70710678f}, {-0.69779046f, 0.71630194f}, +{-0.68835458f, 0.72537437f}, {-0.67880075f, 0.73432251f}, +{-0.66913061f, 0.74314483f}, {-0.65934582f, 0.75183981f}, +{-0.64944805f, 0.76040597f}, {-0.63943900f, 0.76884183f}, +{-0.62932039f, 0.77714596f}, {-0.61909395f, 0.78531693f}, +{-0.60876143f, 0.79335334f}, {-0.59832460f, 0.80125381f}, +{-0.58778525f, 0.80901699f}, {-0.57714519f, 0.81664156f}, +{-0.56640624f, 0.82412619f}, {-0.55557023f, 0.83146961f}, +{-0.54463904f, 0.83867057f}, {-0.53361452f, 0.84572782f}, +{-0.52249856f, 0.85264016f}, {-0.51129309f, 0.85940641f}, +{-0.50000000f, 0.86602540f}, {-0.48862124f, 0.87249601f}, +{-0.47715876f, 0.87881711f}, {-0.46561452f, 0.88498764f}, +{-0.45399050f, 0.89100652f}, {-0.44228869f, 0.89687274f}, +{-0.43051110f, 0.90258528f}, {-0.41865974f, 0.90814317f}, +{-0.40673664f, 0.91354546f}, {-0.39474386f, 0.91879121f}, +{-0.38268343f, 0.92387953f}, {-0.37055744f, 0.92880955f}, +{-0.35836795f, 0.93358043f}, {-0.34611706f, 0.93819134f}, +{-0.33380686f, 0.94264149f}, {-0.32143947f, 0.94693013f}, +{-0.30901699f, 0.95105652f}, {-0.29654157f, 0.95501994f}, +{-0.28401534f, 0.95881973f}, {-0.27144045f, 0.96245524f}, +{-0.25881905f, 0.96592583f}, {-0.24615329f, 0.96923091f}, +{-0.23344536f, 0.97236992f}, {-0.22069744f, 0.97534232f}, +{-0.20791169f, 0.97814760f}, {-0.19509032f, 0.98078528f}, +{-0.18223553f, 0.98325491f}, {-0.16934950f, 0.98555606f}, +{-0.15643447f, 0.98768834f}, {-0.14349262f, 0.98965139f}, +{-0.13052619f, 0.99144486f}, {-0.11753740f, 0.99306846f}, +{-0.10452846f, 0.99452190f}, {-0.091501619f, 0.99580493f}, +{-0.078459096f, 0.99691733f}, {-0.065403129f, 0.99785892f}, +{-0.052335956f, 0.99862953f}, {-0.039259816f, 0.99922904f}, +{-0.026176948f, 0.99965732f}, {-0.013089596f, 0.99991433f}, +{-1.8369095e-16f, 1.0000000f}, {0.013089596f, 0.99991433f}, +{0.026176948f, 0.99965732f}, {0.039259816f, 0.99922904f}, +{0.052335956f, 0.99862953f}, {0.065403129f, 0.99785892f}, +{0.078459096f, 0.99691733f}, {0.091501619f, 0.99580493f}, +{0.10452846f, 0.99452190f}, {0.11753740f, 0.99306846f}, +{0.13052619f, 0.99144486f}, {0.14349262f, 0.98965139f}, +{0.15643447f, 0.98768834f}, {0.16934950f, 0.98555606f}, +{0.18223553f, 0.98325491f}, {0.19509032f, 0.98078528f}, +{0.20791169f, 0.97814760f}, {0.22069744f, 0.97534232f}, +{0.23344536f, 0.97236992f}, {0.24615329f, 0.96923091f}, +{0.25881905f, 0.96592583f}, {0.27144045f, 0.96245524f}, +{0.28401534f, 0.95881973f}, {0.29654157f, 0.95501994f}, +{0.30901699f, 0.95105652f}, {0.32143947f, 0.94693013f}, +{0.33380686f, 0.94264149f}, {0.34611706f, 0.93819134f}, +{0.35836795f, 0.93358043f}, {0.37055744f, 0.92880955f}, +{0.38268343f, 0.92387953f}, {0.39474386f, 0.91879121f}, +{0.40673664f, 0.91354546f}, {0.41865974f, 0.90814317f}, +{0.43051110f, 0.90258528f}, {0.44228869f, 0.89687274f}, +{0.45399050f, 0.89100652f}, {0.46561452f, 0.88498764f}, +{0.47715876f, 0.87881711f}, {0.48862124f, 0.87249601f}, +{0.50000000f, 0.86602540f}, {0.51129309f, 0.85940641f}, +{0.52249856f, 0.85264016f}, {0.53361452f, 0.84572782f}, +{0.54463904f, 0.83867057f}, {0.55557023f, 0.83146961f}, +{0.56640624f, 0.82412619f}, {0.57714519f, 0.81664156f}, +{0.58778525f, 0.80901699f}, {0.59832460f, 0.80125381f}, +{0.60876143f, 0.79335334f}, {0.61909395f, 0.78531693f}, +{0.62932039f, 0.77714596f}, {0.63943900f, 0.76884183f}, +{0.64944805f, 0.76040597f}, {0.65934582f, 0.75183981f}, +{0.66913061f, 0.74314483f}, {0.67880075f, 0.73432251f}, +{0.68835458f, 0.72537437f}, {0.69779046f, 0.71630194f}, +{0.70710678f, 0.70710678f}, {0.71630194f, 0.69779046f}, +{0.72537437f, 0.68835458f}, {0.73432251f, 0.67880075f}, +{0.74314483f, 0.66913061f}, {0.75183981f, 0.65934582f}, +{0.76040597f, 0.64944805f}, {0.76884183f, 0.63943900f}, +{0.77714596f, 0.62932039f}, {0.78531693f, 0.61909395f}, +{0.79335334f, 0.60876143f}, {0.80125381f, 0.59832460f}, +{0.80901699f, 0.58778525f}, {0.81664156f, 0.57714519f}, +{0.82412619f, 0.56640624f}, {0.83146961f, 0.55557023f}, +{0.83867057f, 0.54463904f}, {0.84572782f, 0.53361452f}, +{0.85264016f, 0.52249856f}, {0.85940641f, 0.51129309f}, +{0.86602540f, 0.50000000f}, {0.87249601f, 0.48862124f}, +{0.87881711f, 0.47715876f}, {0.88498764f, 0.46561452f}, +{0.89100652f, 0.45399050f}, {0.89687274f, 0.44228869f}, +{0.90258528f, 0.43051110f}, {0.90814317f, 0.41865974f}, +{0.91354546f, 0.40673664f}, {0.91879121f, 0.39474386f}, +{0.92387953f, 0.38268343f}, {0.92880955f, 0.37055744f}, +{0.93358043f, 0.35836795f}, {0.93819134f, 0.34611706f}, +{0.94264149f, 0.33380686f}, {0.94693013f, 0.32143947f}, +{0.95105652f, 0.30901699f}, {0.95501994f, 0.29654157f}, +{0.95881973f, 0.28401534f}, {0.96245524f, 0.27144045f}, +{0.96592583f, 0.25881905f}, {0.96923091f, 0.24615329f}, +{0.97236992f, 0.23344536f}, {0.97534232f, 0.22069744f}, +{0.97814760f, 0.20791169f}, {0.98078528f, 0.19509032f}, +{0.98325491f, 0.18223553f}, {0.98555606f, 0.16934950f}, +{0.98768834f, 0.15643447f}, {0.98965139f, 0.14349262f}, +{0.99144486f, 0.13052619f}, {0.99306846f, 0.11753740f}, +{0.99452190f, 0.10452846f}, {0.99580493f, 0.091501619f}, +{0.99691733f, 0.078459096f}, {0.99785892f, 0.065403129f}, +{0.99862953f, 0.052335956f}, {0.99922904f, 0.039259816f}, +{0.99965732f, 0.026176948f}, {0.99991433f, 0.013089596f}, +}; +#ifndef FFT_BITREV480 +#define FFT_BITREV480 +static const opus_int16 fft_bitrev480[480] = { +0, 120, 240, 360, 30, 150, 270, 390, 60, 180, 300, 420, 90, 210, 330, +450, 15, 135, 255, 375, 45, 165, 285, 405, 75, 195, 315, 435, 105, 225, +345, 465, 5, 125, 245, 365, 35, 155, 275, 395, 65, 185, 305, 425, 95, +215, 335, 455, 20, 140, 260, 380, 50, 170, 290, 410, 80, 200, 320, 440, +110, 230, 350, 470, 10, 130, 250, 370, 40, 160, 280, 400, 70, 190, 310, +430, 100, 220, 340, 460, 25, 145, 265, 385, 55, 175, 295, 415, 85, 205, +325, 445, 115, 235, 355, 475, 1, 121, 241, 361, 31, 151, 271, 391, 61, +181, 301, 421, 91, 211, 331, 451, 16, 136, 256, 376, 46, 166, 286, 406, +76, 196, 316, 436, 106, 226, 346, 466, 6, 126, 246, 366, 36, 156, 276, +396, 66, 186, 306, 426, 96, 216, 336, 456, 21, 141, 261, 381, 51, 171, +291, 411, 81, 201, 321, 441, 111, 231, 351, 471, 11, 131, 251, 371, 41, +161, 281, 401, 71, 191, 311, 431, 101, 221, 341, 461, 26, 146, 266, 386, +56, 176, 296, 416, 86, 206, 326, 446, 116, 236, 356, 476, 2, 122, 242, +362, 32, 152, 272, 392, 62, 182, 302, 422, 92, 212, 332, 452, 17, 137, +257, 377, 47, 167, 287, 407, 77, 197, 317, 437, 107, 227, 347, 467, 7, +127, 247, 367, 37, 157, 277, 397, 67, 187, 307, 427, 97, 217, 337, 457, +22, 142, 262, 382, 52, 172, 292, 412, 82, 202, 322, 442, 112, 232, 352, +472, 12, 132, 252, 372, 42, 162, 282, 402, 72, 192, 312, 432, 102, 222, +342, 462, 27, 147, 267, 387, 57, 177, 297, 417, 87, 207, 327, 447, 117, +237, 357, 477, 3, 123, 243, 363, 33, 153, 273, 393, 63, 183, 303, 423, +93, 213, 333, 453, 18, 138, 258, 378, 48, 168, 288, 408, 78, 198, 318, +438, 108, 228, 348, 468, 8, 128, 248, 368, 38, 158, 278, 398, 68, 188, +308, 428, 98, 218, 338, 458, 23, 143, 263, 383, 53, 173, 293, 413, 83, +203, 323, 443, 113, 233, 353, 473, 13, 133, 253, 373, 43, 163, 283, 403, +73, 193, 313, 433, 103, 223, 343, 463, 28, 148, 268, 388, 58, 178, 298, +418, 88, 208, 328, 448, 118, 238, 358, 478, 4, 124, 244, 364, 34, 154, +274, 394, 64, 184, 304, 424, 94, 214, 334, 454, 19, 139, 259, 379, 49, +169, 289, 409, 79, 199, 319, 439, 109, 229, 349, 469, 9, 129, 249, 369, +39, 159, 279, 399, 69, 189, 309, 429, 99, 219, 339, 459, 24, 144, 264, +384, 54, 174, 294, 414, 84, 204, 324, 444, 114, 234, 354, 474, 14, 134, +254, 374, 44, 164, 284, 404, 74, 194, 314, 434, 104, 224, 344, 464, 29, +149, 269, 389, 59, 179, 299, 419, 89, 209, 329, 449, 119, 239, 359, 479, +}; +#endif + +#ifndef FFT_BITREV240 +#define FFT_BITREV240 +static const opus_int16 fft_bitrev240[240] = { +0, 60, 120, 180, 15, 75, 135, 195, 30, 90, 150, 210, 45, 105, 165, +225, 5, 65, 125, 185, 20, 80, 140, 200, 35, 95, 155, 215, 50, 110, +170, 230, 10, 70, 130, 190, 25, 85, 145, 205, 40, 100, 160, 220, 55, +115, 175, 235, 1, 61, 121, 181, 16, 76, 136, 196, 31, 91, 151, 211, +46, 106, 166, 226, 6, 66, 126, 186, 21, 81, 141, 201, 36, 96, 156, +216, 51, 111, 171, 231, 11, 71, 131, 191, 26, 86, 146, 206, 41, 101, +161, 221, 56, 116, 176, 236, 2, 62, 122, 182, 17, 77, 137, 197, 32, +92, 152, 212, 47, 107, 167, 227, 7, 67, 127, 187, 22, 82, 142, 202, +37, 97, 157, 217, 52, 112, 172, 232, 12, 72, 132, 192, 27, 87, 147, +207, 42, 102, 162, 222, 57, 117, 177, 237, 3, 63, 123, 183, 18, 78, +138, 198, 33, 93, 153, 213, 48, 108, 168, 228, 8, 68, 128, 188, 23, +83, 143, 203, 38, 98, 158, 218, 53, 113, 173, 233, 13, 73, 133, 193, +28, 88, 148, 208, 43, 103, 163, 223, 58, 118, 178, 238, 4, 64, 124, +184, 19, 79, 139, 199, 34, 94, 154, 214, 49, 109, 169, 229, 9, 69, +129, 189, 24, 84, 144, 204, 39, 99, 159, 219, 54, 114, 174, 234, 14, +74, 134, 194, 29, 89, 149, 209, 44, 104, 164, 224, 59, 119, 179, 239, +}; +#endif + +#ifndef FFT_BITREV120 +#define FFT_BITREV120 +static const opus_int16 fft_bitrev120[120] = { +0, 30, 60, 90, 15, 45, 75, 105, 5, 35, 65, 95, 20, 50, 80, +110, 10, 40, 70, 100, 25, 55, 85, 115, 1, 31, 61, 91, 16, 46, +76, 106, 6, 36, 66, 96, 21, 51, 81, 111, 11, 41, 71, 101, 26, +56, 86, 116, 2, 32, 62, 92, 17, 47, 77, 107, 7, 37, 67, 97, +22, 52, 82, 112, 12, 42, 72, 102, 27, 57, 87, 117, 3, 33, 63, +93, 18, 48, 78, 108, 8, 38, 68, 98, 23, 53, 83, 113, 13, 43, +73, 103, 28, 58, 88, 118, 4, 34, 64, 94, 19, 49, 79, 109, 9, +39, 69, 99, 24, 54, 84, 114, 14, 44, 74, 104, 29, 59, 89, 119, +}; +#endif + +#ifndef FFT_BITREV60 +#define FFT_BITREV60 +static const opus_int16 fft_bitrev60[60] = { +0, 15, 30, 45, 5, 20, 35, 50, 10, 25, 40, 55, 1, 16, 31, +46, 6, 21, 36, 51, 11, 26, 41, 56, 2, 17, 32, 47, 7, 22, +37, 52, 12, 27, 42, 57, 3, 18, 33, 48, 8, 23, 38, 53, 13, +28, 43, 58, 4, 19, 34, 49, 9, 24, 39, 54, 14, 29, 44, 59, +}; +#endif + +#ifndef FFT_STATE48000_960_0 +#define FFT_STATE48000_960_0 +static const kiss_fft_state fft_state48000_960_0 = { +480, /* nfft */ +0.002083333f, /* scale */ +-1, /* shift */ +{4, 120, 4, 30, 2, 15, 3, 5, 5, 1, 0, 0, 0, 0, 0, 0, }, /* factors */ +fft_bitrev480, /* bitrev */ +fft_twiddles48000_960, /* bitrev */ +}; +#endif + +#ifndef FFT_STATE48000_960_1 +#define FFT_STATE48000_960_1 +static const kiss_fft_state fft_state48000_960_1 = { +240, /* nfft */ +0.004166667f, /* scale */ +1, /* shift */ +{4, 60, 4, 15, 3, 5, 5, 1, 0, 0, 0, 0, 0, 0, 0, 0, }, /* factors */ +fft_bitrev240, /* bitrev */ +fft_twiddles48000_960, /* bitrev */ +}; +#endif + +#ifndef FFT_STATE48000_960_2 +#define FFT_STATE48000_960_2 +static const kiss_fft_state fft_state48000_960_2 = { +120, /* nfft */ +0.008333333f, /* scale */ +2, /* shift */ +{4, 30, 2, 15, 3, 5, 5, 1, 0, 0, 0, 0, 0, 0, 0, 0, }, /* factors */ +fft_bitrev120, /* bitrev */ +fft_twiddles48000_960, /* bitrev */ +}; +#endif + +#ifndef FFT_STATE48000_960_3 +#define FFT_STATE48000_960_3 +static const kiss_fft_state fft_state48000_960_3 = { +60, /* nfft */ +0.016666667f, /* scale */ +3, /* shift */ +{4, 15, 3, 5, 5, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, }, /* factors */ +fft_bitrev60, /* bitrev */ +fft_twiddles48000_960, /* bitrev */ +}; +#endif + +#endif + +#ifndef MDCT_TWIDDLES960 +#define MDCT_TWIDDLES960 +static const opus_val16 mdct_twiddles960[481] = { +1.0000000f, 0.99999465f, 0.99997858f, 0.99995181f, 0.99991433f, +0.99986614f, 0.99980724f, 0.99973764f, 0.99965732f, 0.99956631f, +0.99946459f, 0.99935216f, 0.99922904f, 0.99909521f, 0.99895068f, +0.99879546f, 0.99862953f, 0.99845292f, 0.99826561f, 0.99806761f, +0.99785892f, 0.99763955f, 0.99740949f, 0.99716875f, 0.99691733f, +0.99665524f, 0.99638247f, 0.99609903f, 0.99580493f, 0.99550016f, +0.99518473f, 0.99485864f, 0.99452190f, 0.99417450f, 0.99381646f, +0.99344778f, 0.99306846f, 0.99267850f, 0.99227791f, 0.99186670f, +0.99144486f, 0.99101241f, 0.99056934f, 0.99011566f, 0.98965139f, +0.98917651f, 0.98869104f, 0.98819498f, 0.98768834f, 0.98717112f, +0.98664333f, 0.98610497f, 0.98555606f, 0.98499659f, 0.98442657f, +0.98384600f, 0.98325491f, 0.98265328f, 0.98204113f, 0.98141846f, +0.98078528f, 0.98014159f, 0.97948742f, 0.97882275f, 0.97814760f, +0.97746197f, 0.97676588f, 0.97605933f, 0.97534232f, 0.97461487f, +0.97387698f, 0.97312866f, 0.97236992f, 0.97160077f, 0.97082121f, +0.97003125f, 0.96923091f, 0.96842019f, 0.96759909f, 0.96676764f, +0.96592582f, 0.96507367f, 0.96421118f, 0.96333837f, 0.96245523f, +0.96156180f, 0.96065806f, 0.95974403f, 0.95881973f, 0.95788517f, +0.95694034f, 0.95598526f, 0.95501995f, 0.95404440f, 0.95305864f, +0.95206267f, 0.95105651f, 0.95004016f, 0.94901364f, 0.94797697f, +0.94693013f, 0.94587315f, 0.94480604f, 0.94372882f, 0.94264149f, +0.94154406f, 0.94043656f, 0.93931897f, 0.93819133f, 0.93705365f, +0.93590592f, 0.93474818f, 0.93358042f, 0.93240268f, 0.93121493f, +0.93001722f, 0.92880955f, 0.92759193f, 0.92636438f, 0.92512690f, +0.92387953f, 0.92262225f, 0.92135509f, 0.92007809f, 0.91879121f, +0.91749449f, 0.91618795f, 0.91487161f, 0.91354545f, 0.91220952f, +0.91086382f, 0.90950836f, 0.90814316f, 0.90676824f, 0.90538363f, +0.90398929f, 0.90258528f, 0.90117161f, 0.89974828f, 0.89831532f, +0.89687273f, 0.89542055f, 0.89395877f, 0.89248742f, 0.89100652f, +0.88951606f, 0.88801610f, 0.88650661f, 0.88498764f, 0.88345918f, +0.88192125f, 0.88037390f, 0.87881711f, 0.87725090f, 0.87567531f, +0.87409035f, 0.87249599f, 0.87089232f, 0.86927933f, 0.86765699f, +0.86602540f, 0.86438453f, 0.86273437f, 0.86107503f, 0.85940641f, +0.85772862f, 0.85604161f, 0.85434547f, 0.85264014f, 0.85092572f, +0.84920218f, 0.84746955f, 0.84572781f, 0.84397704f, 0.84221721f, +0.84044838f, 0.83867056f, 0.83688375f, 0.83508799f, 0.83328325f, +0.83146961f, 0.82964704f, 0.82781562f, 0.82597530f, 0.82412620f, +0.82226820f, 0.82040144f, 0.81852589f, 0.81664154f, 0.81474847f, +0.81284665f, 0.81093620f, 0.80901698f, 0.80708914f, 0.80515262f, +0.80320752f, 0.80125378f, 0.79929149f, 0.79732067f, 0.79534125f, +0.79335335f, 0.79135691f, 0.78935204f, 0.78733867f, 0.78531691f, +0.78328674f, 0.78124818f, 0.77920122f, 0.77714595f, 0.77508232f, +0.77301043f, 0.77093026f, 0.76884183f, 0.76674517f, 0.76464026f, +0.76252720f, 0.76040593f, 0.75827656f, 0.75613907f, 0.75399349f, +0.75183978f, 0.74967807f, 0.74750833f, 0.74533054f, 0.74314481f, +0.74095112f, 0.73874950f, 0.73653993f, 0.73432251f, 0.73209718f, +0.72986405f, 0.72762307f, 0.72537438f, 0.72311787f, 0.72085359f, +0.71858162f, 0.71630192f, 0.71401459f, 0.71171956f, 0.70941701f, +0.70710677f, 0.70478900f, 0.70246363f, 0.70013079f, 0.69779041f, +0.69544260f, 0.69308738f, 0.69072466f, 0.68835458f, 0.68597709f, +0.68359229f, 0.68120013f, 0.67880072f, 0.67639404f, 0.67398011f, +0.67155892f, 0.66913059f, 0.66669509f, 0.66425240f, 0.66180265f, +0.65934581f, 0.65688191f, 0.65441092f, 0.65193298f, 0.64944801f, +0.64695613f, 0.64445727f, 0.64195160f, 0.63943902f, 0.63691954f, +0.63439328f, 0.63186019f, 0.62932037f, 0.62677377f, 0.62422055f, +0.62166055f, 0.61909394f, 0.61652065f, 0.61394081f, 0.61135435f, +0.60876139f, 0.60616195f, 0.60355593f, 0.60094349f, 0.59832457f, +0.59569929f, 0.59306758f, 0.59042957f, 0.58778523f, 0.58513460f, +0.58247766f, 0.57981452f, 0.57714518f, 0.57446961f, 0.57178793f, +0.56910013f, 0.56640624f, 0.56370623f, 0.56100023f, 0.55828818f, +0.55557020f, 0.55284627f, 0.55011641f, 0.54738067f, 0.54463901f, +0.54189157f, 0.53913828f, 0.53637921f, 0.53361450f, 0.53084398f, +0.52806787f, 0.52528601f, 0.52249852f, 0.51970543f, 0.51690688f, +0.51410279f, 0.51129310f, 0.50847793f, 0.50565732f, 0.50283139f, +0.49999997f, 0.49716321f, 0.49432122f, 0.49147383f, 0.48862118f, +0.48576340f, 0.48290042f, 0.48003216f, 0.47715876f, 0.47428025f, +0.47139677f, 0.46850813f, 0.46561448f, 0.46271584f, 0.45981235f, +0.45690383f, 0.45399042f, 0.45107214f, 0.44814915f, 0.44522124f, +0.44228868f, 0.43935137f, 0.43640926f, 0.43346247f, 0.43051104f, +0.42755511f, 0.42459449f, 0.42162932f, 0.41865964f, 0.41568558f, +0.41270697f, 0.40972393f, 0.40673661f, 0.40374494f, 0.40074884f, +0.39774844f, 0.39474390f, 0.39173501f, 0.38872193f, 0.38570469f, +0.38268343f, 0.37965796f, 0.37662842f, 0.37359496f, 0.37055739f, +0.36751585f, 0.36447038f, 0.36142122f, 0.35836797f, 0.35531089f, +0.35225000f, 0.34918544f, 0.34611704f, 0.34304493f, 0.33996926f, +0.33688983f, 0.33380680f, 0.33072019f, 0.32763015f, 0.32453650f, +0.32143936f, 0.31833890f, 0.31523503f, 0.31212767f, 0.30901696f, +0.30590306f, 0.30278577f, 0.29966524f, 0.29654150f, 0.29341470f, +0.29028464f, 0.28715147f, 0.28401522f, 0.28087605f, 0.27773376f, +0.27458861f, 0.27144052f, 0.26828940f, 0.26513541f, 0.26197859f, +0.25881907f, 0.25565666f, 0.25249152f, 0.24932367f, 0.24615327f, +0.24298012f, 0.23980436f, 0.23662604f, 0.23344530f, 0.23026206f, +0.22707623f, 0.22388809f, 0.22069744f, 0.21750443f, 0.21430908f, +0.21111156f, 0.20791165f, 0.20470953f, 0.20150520f, 0.19829884f, +0.19509024f, 0.19187955f, 0.18866692f, 0.18545227f, 0.18223552f, +0.17901681f, 0.17579631f, 0.17257380f, 0.16934945f, 0.16612328f, +0.16289546f, 0.15966577f, 0.15643437f, 0.15320141f, 0.14996669f, +0.14673037f, 0.14349260f, 0.14025329f, 0.13701235f, 0.13376995f, +0.13052612f, 0.12728101f, 0.12403442f, 0.12078650f, 0.11753740f, +0.11428693f, 0.11103523f, 0.10778234f, 0.10452842f, 0.10127326f, +0.098017137f, 0.094759842f, 0.091501652f, 0.088242363f, 0.084982129f, +0.081721103f, 0.078459084f, 0.075196224f, 0.071932560f, 0.068668243f, +0.065403073f, 0.062137201f, 0.058870665f, 0.055603617f, 0.052335974f, +0.049067651f, 0.045798921f, 0.042529582f, 0.039259788f, 0.035989573f, +0.032719092f, 0.029448142f, 0.026176876f, 0.022905329f, 0.019633657f, +0.016361655f, 0.013089478f, 0.0098171604f, 0.0065449764f, 0.0032724839f, +-4.3711390e-08f, }; +#endif + +static const CELTMode mode48000_960_120 = { +48000, /* Fs */ +120, /* overlap */ +21, /* nbEBands */ +21, /* effEBands */ +{0.85000610f, 0.0000000f, 1.0000000f, 1.0000000f, }, /* preemph */ +eband5ms, /* eBands */ +3, /* maxLM */ +8, /* nbShortMdcts */ +120, /* shortMdctSize */ +11, /* nbAllocVectors */ +band_allocation, /* allocVectors */ +logN400, /* logN */ +window120, /* window */ +{1920, 3, {&fft_state48000_960_0, &fft_state48000_960_1, &fft_state48000_960_2, &fft_state48000_960_3, }, mdct_twiddles960}, /* mdct */ +{392, cache_index50, cache_bits50, cache_caps50}, /* cache */ +}; + +/* List of all the available modes */ +#define TOTAL_MODES 1 +static const CELTMode * const static_mode_list[TOTAL_MODES] = { +&mode48000_960_120, +}; diff --git a/drivers/opus/celt/tests/test_unit_cwrs32.c b/drivers/opus/celt/tests/test_unit_cwrs32.c new file mode 100644 index 0000000000..9cf124336a --- /dev/null +++ b/drivers/opus/celt/tests/test_unit_cwrs32.c @@ -0,0 +1,161 @@ +/* Copyright (c) 2008-2011 Xiph.Org Foundation, Mozilla Corporation, + Gregory Maxwell + Written by Jean-Marc Valin, Gregory Maxwell, and Timothy B. Terriberry */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef OPUS_HAVE_CONFIG_H +#include "opus_config.h" +#endif + +#include <stdio.h> +#include <string.h> + +#ifndef CUSTOM_MODES +#define CUSTOM_MODES +#else +#define TEST_CUSTOM_MODES +#endif + +#define CELT_C +#include "stack_alloc.h" +#include "entenc.c" +#include "entdec.c" +#include "entcode.c" +#include "cwrs.c" +#include "mathops.c" +#include "rate.h" + +#define NMAX (240) +#define KMAX (128) + +#ifdef TEST_CUSTOM_MODES + +#define NDIMS (44) +static const int pn[NDIMS]={ + 2, 3, 4, 5, 6, 7, 8, 9, 10, + 11, 12, 13, 14, 15, 16, 18, 20, 22, + 24, 26, 28, 30, 32, 36, 40, 44, 48, + 52, 56, 60, 64, 72, 80, 88, 96, 104, + 112, 120, 128, 144, 160, 176, 192, 208 +}; +static const int pkmax[NDIMS]={ + 128, 128, 128, 128, 88, 52, 36, 26, 22, + 18, 16, 15, 13, 12, 12, 11, 10, 9, + 9, 8, 8, 7, 7, 7, 7, 6, 6, + 6, 6, 6, 5, 5, 5, 5, 5, 5, + 4, 4, 4, 4, 4, 4, 4, 4 +}; + +#else /* TEST_CUSTOM_MODES */ + +#define NDIMS (22) +static const int pn[NDIMS]={ + 2, 3, 4, 6, 8, 9, 11, 12, 16, + 18, 22, 24, 32, 36, 44, 48, 64, 72, + 88, 96, 144, 176 +}; +static const int pkmax[NDIMS]={ + 128, 128, 128, 88, 36, 26, 18, 16, 12, + 11, 9, 9, 7, 7, 6, 6, 5, 5, + 5, 5, 4, 4 +}; + +#endif + +int main(void){ + int t; + int n; + ALLOC_STACK; + for(t=0;t<NDIMS;t++){ + int pseudo; + n=pn[t]; + for(pseudo=1;pseudo<41;pseudo++) + { + int k; +#if defined(SMALL_FOOTPRINT) + opus_uint32 uu[KMAX+2U]; +#endif + opus_uint32 inc; + opus_uint32 nc; + opus_uint32 i; + k=get_pulses(pseudo); + if (k>pkmax[t])break; + printf("Testing CWRS with N=%i, K=%i...\n",n,k); +#if defined(SMALL_FOOTPRINT) + nc=ncwrs_urow(n,k,uu); +#else + nc=CELT_PVQ_V(n,k); +#endif + inc=nc/20000; + if(inc<1)inc=1; + for(i=0;i<nc;i+=inc){ +#if defined(SMALL_FOOTPRINT) + opus_uint32 u[KMAX+2U]; +#endif + int y[NMAX]; + int sy; + opus_uint32 v; + opus_uint32 ii; + int j; +#if defined(SMALL_FOOTPRINT) + memcpy(u,uu,(k+2U)*sizeof(*u)); + cwrsi(n,k,i,y,u); +#else + cwrsi(n,k,i,y); +#endif + sy=0; + for(j=0;j<n;j++)sy+=ABS(y[j]); + if(sy!=k){ + fprintf(stderr,"N=%d Pulse count mismatch in cwrsi (%d!=%d).\n", + n,sy,k); + return 99; + } + /*printf("%6u of %u:",i,nc); + for(j=0;j<n;j++)printf(" %+3i",y[j]); + printf(" ->");*/ +#if defined(SMALL_FOOTPRINT) + ii=icwrs(n,k,&v,y,u); +#else + ii=icwrs(n,y); + v=CELT_PVQ_V(n,k); +#endif + if(ii!=i){ + fprintf(stderr,"Combination-index mismatch (%lu!=%lu).\n", + (long)ii,(long)i); + return 1; + } + if(v!=nc){ + fprintf(stderr,"Combination count mismatch (%lu!=%lu).\n", + (long)v,(long)nc); + return 2; + } + /*printf(" %6u\n",i);*/ + } + /*printf("\n");*/ + } + } + return 0; +} diff --git a/drivers/opus/celt/tests/test_unit_dft.c b/drivers/opus/celt/tests/test_unit_dft.c new file mode 100644 index 0000000000..4a00013b2a --- /dev/null +++ b/drivers/opus/celt/tests/test_unit_dft.c @@ -0,0 +1,164 @@ +/* Copyright (c) 2008 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef OPUS_HAVE_CONFIG_H +#include "opus_config.h" +#endif + +#define SKIP_CONFIG_H + +#ifndef CUSTOM_MODES +#define CUSTOM_MODES +#endif + +#include <stdio.h> + +#define CELT_C +#include "stack_alloc.h" +#include "kiss_fft.h" +#include "kiss_fft.c" +#include "mathops.c" +#include "entcode.c" + + +#ifndef M_PI +#define M_PI 3.141592653 +#endif + +int ret = 0; + +void check(kiss_fft_cpx * in,kiss_fft_cpx * out,int nfft,int isinverse) +{ + int bin,k; + double errpow=0,sigpow=0, snr; + + for (bin=0;bin<nfft;++bin) { + double ansr = 0; + double ansi = 0; + double difr; + double difi; + + for (k=0;k<nfft;++k) { + double phase = -2*M_PI*bin*k/nfft; + double re = cos(phase); + double im = sin(phase); + if (isinverse) + im = -im; + + if (!isinverse) + { + re /= nfft; + im /= nfft; + } + + ansr += in[k].r * re - in[k].i * im; + ansi += in[k].r * im + in[k].i * re; + } + /*printf ("%d %d ", (int)ansr, (int)ansi);*/ + difr = ansr - out[bin].r; + difi = ansi - out[bin].i; + errpow += difr*difr + difi*difi; + sigpow += ansr*ansr+ansi*ansi; + } + snr = 10*log10(sigpow/errpow); + printf("nfft=%d inverse=%d,snr = %f\n",nfft,isinverse,snr ); + if (snr<60) { + printf( "** poor snr: %f ** \n", snr); + ret = 1; + } +} + +void test1d(int nfft,int isinverse) +{ + size_t buflen = sizeof(kiss_fft_cpx)*nfft; + + kiss_fft_cpx * in = (kiss_fft_cpx*)malloc(buflen); + kiss_fft_cpx * out= (kiss_fft_cpx*)malloc(buflen); + kiss_fft_state *cfg = opus_fft_alloc(nfft,0,0); + int k; + + for (k=0;k<nfft;++k) { + in[k].r = (rand() % 32767) - 16384; + in[k].i = (rand() % 32767) - 16384; + } + + for (k=0;k<nfft;++k) { + in[k].r *= 32768; + in[k].i *= 32768; + } + + if (isinverse) + { + for (k=0;k<nfft;++k) { + in[k].r /= nfft; + in[k].i /= nfft; + } + } + + /*for (k=0;k<nfft;++k) printf("%d %d ", in[k].r, in[k].i);printf("\n");*/ + + if (isinverse) + opus_ifft(cfg,in,out); + else + opus_fft(cfg,in,out); + + /*for (k=0;k<nfft;++k) printf("%d %d ", out[k].r, out[k].i);printf("\n");*/ + + check(in,out,nfft,isinverse); + + free(in); + free(out); + free(cfg); +} + +int main(int argc,char ** argv) +{ + ALLOC_STACK; + if (argc>1) { + int k; + for (k=1;k<argc;++k) { + test1d(atoi(argv[k]),0); + test1d(atoi(argv[k]),1); + } + }else{ + test1d(32,0); + test1d(32,1); + test1d(128,0); + test1d(128,1); + test1d(256,0); + test1d(256,1); +#ifndef RADIX_TWO_ONLY + test1d(36,0); + test1d(36,1); + test1d(50,0); + test1d(50,1); + test1d(120,0); + test1d(120,1); +#endif + } + return ret; +} diff --git a/drivers/opus/celt/tests/test_unit_entropy.c b/drivers/opus/celt/tests/test_unit_entropy.c new file mode 100644 index 0000000000..62268b1564 --- /dev/null +++ b/drivers/opus/celt/tests/test_unit_entropy.c @@ -0,0 +1,382 @@ +/* Copyright (c) 2007-2011 Xiph.Org Foundation, Mozilla Corporation, + Gregory Maxwell + Written by Jean-Marc Valin, Gregory Maxwell, and Timothy B. Terriberry */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef OPUS_HAVE_CONFIG_H +#include "opus_config.h" +#endif + +#include <stdlib.h> +#include <stdio.h> +#include <math.h> +#include <time.h> +#include "entcode.h" +#include "entenc.h" +#include "entdec.h" +#include <string.h> + +#include "entenc.c" +#include "entdec.c" +#include "entcode.c" + +#ifndef M_LOG2E +# define M_LOG2E 1.4426950408889634074 +#endif +#define DATA_SIZE 10000000 +#define DATA_SIZE2 10000 + +int main(int _argc,char **_argv){ + ec_enc enc; + ec_dec dec; + long nbits; + long nbits2; + double entropy; + int ft; + int ftb; + int sz; + int i; + int ret; + unsigned int sym; + unsigned int seed; + unsigned char *ptr; + const char *env_seed; + ret=0; + entropy=0; + if (_argc > 2) { + fprintf(stderr, "Usage: %s [<seed>]\n", _argv[0]); + return 1; + } + env_seed = getenv("SEED"); + if (_argc > 1) + seed = atoi(_argv[1]); + else if (env_seed) + seed = atoi(env_seed); + else + seed = time(NULL); + /*Testing encoding of raw bit values.*/ + ptr = (unsigned char *)malloc(DATA_SIZE); + ec_enc_init(&enc,ptr, DATA_SIZE); + for(ft=2;ft<1024;ft++){ + for(i=0;i<ft;i++){ + entropy+=log(ft)*M_LOG2E; + ec_enc_uint(&enc,i,ft); + } + } + /*Testing encoding of raw bit values.*/ + for(ftb=1;ftb<16;ftb++){ + for(i=0;i<(1<<ftb);i++){ + entropy+=ftb; + nbits=ec_tell(&enc); + ec_enc_bits(&enc,i,ftb); + nbits2=ec_tell(&enc); + if(nbits2-nbits!=ftb){ + fprintf(stderr,"Used %li bits to encode %i bits directly.\n", + nbits2-nbits,ftb); + ret=-1; + } + } + } + nbits=ec_tell_frac(&enc); + ec_enc_done(&enc); + fprintf(stderr, + "Encoded %0.2lf bits of entropy to %0.2lf bits (%0.3lf%% wasted).\n", + entropy,ldexp(nbits,-3),100*(nbits-ldexp(entropy,3))/nbits); + fprintf(stderr,"Packed to %li bytes.\n",(long)ec_range_bytes(&enc)); + ec_dec_init(&dec,ptr,DATA_SIZE); + for(ft=2;ft<1024;ft++){ + for(i=0;i<ft;i++){ + sym=ec_dec_uint(&dec,ft); + if(sym!=(unsigned)i){ + fprintf(stderr,"Decoded %i instead of %i with ft of %i.\n",sym,i,ft); + ret=-1; + } + } + } + for(ftb=1;ftb<16;ftb++){ + for(i=0;i<(1<<ftb);i++){ + sym=ec_dec_bits(&dec,ftb); + if(sym!=(unsigned)i){ + fprintf(stderr,"Decoded %i instead of %i with ftb of %i.\n",sym,i,ftb); + ret=-1; + } + } + } + nbits2=ec_tell_frac(&dec); + if(nbits!=nbits2){ + fprintf(stderr, + "Reported number of bits used was %0.2lf, should be %0.2lf.\n", + ldexp(nbits2,-3),ldexp(nbits,-3)); + ret=-1; + } + /*Testing an encoder bust prefers range coder data over raw bits. + This isn't a general guarantee, will only work for data that is buffered in + the encoder state and not yet stored in the user buffer, and should never + get used in practice. + It's mostly here for code coverage completeness.*/ + /*Start with a 16-bit buffer.*/ + ec_enc_init(&enc,ptr,2); + /*Write 7 raw bits.*/ + ec_enc_bits(&enc,0x55,7); + /*Write 12.3 bits of range coder data.*/ + ec_enc_uint(&enc,1,2); + ec_enc_uint(&enc,1,3); + ec_enc_uint(&enc,1,4); + ec_enc_uint(&enc,1,5); + ec_enc_uint(&enc,2,6); + ec_enc_uint(&enc,6,7); + ec_enc_done(&enc); + ec_dec_init(&dec,ptr,2); + if(!enc.error + /*The raw bits should have been overwritten by the range coder data.*/ + ||ec_dec_bits(&dec,7)!=0x05 + /*And all the range coder data should have been encoded correctly.*/ + ||ec_dec_uint(&dec,2)!=1 + ||ec_dec_uint(&dec,3)!=1 + ||ec_dec_uint(&dec,4)!=1 + ||ec_dec_uint(&dec,5)!=1 + ||ec_dec_uint(&dec,6)!=2 + ||ec_dec_uint(&dec,7)!=6){ + fprintf(stderr,"Encoder bust overwrote range coder data with raw bits.\n"); + ret=-1; + } + srand(seed); + fprintf(stderr,"Testing random streams... Random seed: %u (%.4X)\n", seed, rand() % 65536); + for(i=0;i<409600;i++){ + unsigned *data; + unsigned *tell; + unsigned tell_bits; + int j; + int zeros; + ft=rand()/((RAND_MAX>>(rand()%11U))+1U)+10; + sz=rand()/((RAND_MAX>>(rand()%9U))+1U); + data=(unsigned *)malloc(sz*sizeof(*data)); + tell=(unsigned *)malloc((sz+1)*sizeof(*tell)); + ec_enc_init(&enc,ptr,DATA_SIZE2); + zeros = rand()%13==0; + tell[0]=ec_tell_frac(&enc); + for(j=0;j<sz;j++){ + if (zeros) + data[j]=0; + else + data[j]=rand()%ft; + ec_enc_uint(&enc,data[j],ft); + tell[j+1]=ec_tell_frac(&enc); + } + if (rand()%2==0) + while(ec_tell(&enc)%8 != 0) + ec_enc_uint(&enc, rand()%2, 2); + tell_bits = ec_tell(&enc); + ec_enc_done(&enc); + if(tell_bits!=(unsigned)ec_tell(&enc)){ + fprintf(stderr,"ec_tell() changed after ec_enc_done(): %i instead of %i (Random seed: %u)\n", + ec_tell(&enc),tell_bits,seed); + ret=-1; + } + if ((tell_bits+7)/8 < ec_range_bytes(&enc)) + { + fprintf (stderr, "ec_tell() lied, there's %i bytes instead of %d (Random seed: %u)\n", + ec_range_bytes(&enc), (tell_bits+7)/8,seed); + ret=-1; + } + ec_dec_init(&dec,ptr,DATA_SIZE2); + if(ec_tell_frac(&dec)!=tell[0]){ + fprintf(stderr, + "Tell mismatch between encoder and decoder at symbol %i: %i instead of %i (Random seed: %u).\n", + 0,ec_tell_frac(&dec),tell[0],seed); + } + for(j=0;j<sz;j++){ + sym=ec_dec_uint(&dec,ft); + if(sym!=data[j]){ + fprintf(stderr, + "Decoded %i instead of %i with ft of %i at position %i of %i (Random seed: %u).\n", + sym,data[j],ft,j,sz,seed); + ret=-1; + } + if(ec_tell_frac(&dec)!=tell[j+1]){ + fprintf(stderr, + "Tell mismatch between encoder and decoder at symbol %i: %i instead of %i (Random seed: %u).\n", + j+1,ec_tell_frac(&dec),tell[j+1],seed); + } + } + free(tell); + free(data); + } + /*Test compatibility between multiple different encode/decode routines.*/ + for(i=0;i<409600;i++){ + unsigned *logp1; + unsigned *data; + unsigned *tell; + unsigned *enc_method; + int j; + sz=rand()/((RAND_MAX>>(rand()%9U))+1U); + logp1=(unsigned *)malloc(sz*sizeof(*logp1)); + data=(unsigned *)malloc(sz*sizeof(*data)); + tell=(unsigned *)malloc((sz+1)*sizeof(*tell)); + enc_method=(unsigned *)malloc(sz*sizeof(*enc_method)); + ec_enc_init(&enc,ptr,DATA_SIZE2); + tell[0]=ec_tell_frac(&enc); + for(j=0;j<sz;j++){ + data[j]=rand()/((RAND_MAX>>1)+1); + logp1[j]=(rand()%15)+1; + enc_method[j]=rand()/((RAND_MAX>>2)+1); + switch(enc_method[j]){ + case 0:{ + ec_encode(&enc,data[j]?(1<<logp1[j])-1:0, + (1<<logp1[j])-(data[j]?0:1),1<<logp1[j]); + }break; + case 1:{ + ec_encode_bin(&enc,data[j]?(1<<logp1[j])-1:0, + (1<<logp1[j])-(data[j]?0:1),logp1[j]); + }break; + case 2:{ + ec_enc_bit_logp(&enc,data[j],logp1[j]); + }break; + case 3:{ + unsigned char icdf[2]; + icdf[0]=1; + icdf[1]=0; + ec_enc_icdf(&enc,data[j],icdf,logp1[j]); + }break; + } + tell[j+1]=ec_tell_frac(&enc); + } + ec_enc_done(&enc); + if((ec_tell(&enc)+7U)/8U<ec_range_bytes(&enc)){ + fprintf(stderr,"tell() lied, there's %i bytes instead of %d (Random seed: %u)\n", + ec_range_bytes(&enc),(ec_tell(&enc)+7)/8,seed); + ret=-1; + } + ec_dec_init(&dec,ptr,DATA_SIZE2); + if(ec_tell_frac(&dec)!=tell[0]){ + fprintf(stderr, + "Tell mismatch between encoder and decoder at symbol %i: %i instead of %i (Random seed: %u).\n", + 0,ec_tell_frac(&dec),tell[0],seed); + } + for(j=0;j<sz;j++){ + int fs; + int dec_method; + dec_method=rand()/((RAND_MAX>>2)+1); + switch(dec_method){ + case 0:{ + fs=ec_decode(&dec,1<<logp1[j]); + sym=fs>=(1<<logp1[j])-1; + ec_dec_update(&dec,sym?(1<<logp1[j])-1:0, + (1<<logp1[j])-(sym?0:1),1<<logp1[j]); + }break; + case 1:{ + fs=ec_decode_bin(&dec,logp1[j]); + sym=fs>=(1<<logp1[j])-1; + ec_dec_update(&dec,sym?(1<<logp1[j])-1:0, + (1<<logp1[j])-(sym?0:1),1<<logp1[j]); + }break; + case 2:{ + sym=ec_dec_bit_logp(&dec,logp1[j]); + }break; + case 3:{ + unsigned char icdf[2]; + icdf[0]=1; + icdf[1]=0; + sym=ec_dec_icdf(&dec,icdf,logp1[j]); + }break; + } + if(sym!=data[j]){ + fprintf(stderr, + "Decoded %i instead of %i with logp1 of %i at position %i of %i (Random seed: %u).\n", + sym,data[j],logp1[j],j,sz,seed); + fprintf(stderr,"Encoding method: %i, decoding method: %i\n", + enc_method[j],dec_method); + ret=-1; + } + if(ec_tell_frac(&dec)!=tell[j+1]){ + fprintf(stderr, + "Tell mismatch between encoder and decoder at symbol %i: %i instead of %i (Random seed: %u).\n", + j+1,ec_tell_frac(&dec),tell[j+1],seed); + } + } + free(enc_method); + free(tell); + free(data); + free(logp1); + } + ec_enc_init(&enc,ptr,DATA_SIZE2); + ec_enc_bit_logp(&enc,0,1); + ec_enc_bit_logp(&enc,0,1); + ec_enc_bit_logp(&enc,0,1); + ec_enc_bit_logp(&enc,0,1); + ec_enc_bit_logp(&enc,0,2); + ec_enc_patch_initial_bits(&enc,3,2); + if(enc.error){ + fprintf(stderr,"patch_initial_bits failed"); + ret=-1; + } + ec_enc_patch_initial_bits(&enc,0,5); + if(!enc.error){ + fprintf(stderr,"patch_initial_bits didn't fail when it should have"); + ret=-1; + } + ec_enc_done(&enc); + if(ec_range_bytes(&enc)!=1||ptr[0]!=192){ + fprintf(stderr,"Got %d when expecting 192 for patch_initial_bits",ptr[0]); + ret=-1; + } + ec_enc_init(&enc,ptr,DATA_SIZE2); + ec_enc_bit_logp(&enc,0,1); + ec_enc_bit_logp(&enc,0,1); + ec_enc_bit_logp(&enc,1,6); + ec_enc_bit_logp(&enc,0,2); + ec_enc_patch_initial_bits(&enc,0,2); + if(enc.error){ + fprintf(stderr,"patch_initial_bits failed"); + ret=-1; + } + ec_enc_done(&enc); + if(ec_range_bytes(&enc)!=2||ptr[0]!=63){ + fprintf(stderr,"Got %d when expecting 63 for patch_initial_bits",ptr[0]); + ret=-1; + } + ec_enc_init(&enc,ptr,2); + ec_enc_bit_logp(&enc,0,2); + for(i=0;i<48;i++){ + ec_enc_bits(&enc,0,1); + } + ec_enc_done(&enc); + if(!enc.error){ + fprintf(stderr,"Raw bits overfill didn't fail when it should have"); + ret=-1; + } + ec_enc_init(&enc,ptr,2); + for(i=0;i<17;i++){ + ec_enc_bits(&enc,0,1); + } + ec_enc_done(&enc); + if(!enc.error){ + fprintf(stderr,"17 raw bits encoded in two bytes"); + ret=-1; + } + free(ptr); + return ret; +} diff --git a/drivers/opus/celt/tests/test_unit_laplace.c b/drivers/opus/celt/tests/test_unit_laplace.c new file mode 100644 index 0000000000..af7d471045 --- /dev/null +++ b/drivers/opus/celt/tests/test_unit_laplace.c @@ -0,0 +1,92 @@ +/* Copyright (c) 2008-2011 Xiph.Org Foundation, Mozilla Corporation + Written by Jean-Marc Valin and Timothy B. Terriberry */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef OPUS_HAVE_CONFIG_H +#include "opus_config.h" +#endif + +#include <stdio.h> +#include <stdlib.h> +#include "laplace.h" +#define CELT_C +#include "stack_alloc.h" + +#include "entenc.c" +#include "entdec.c" +#include "entcode.c" +#include "laplace.c" + +#define DATA_SIZE 40000 + +int ec_laplace_get_start_freq(int decay) +{ + opus_uint32 ft = 32768 - LAPLACE_MINP*(2*LAPLACE_NMIN+1); + int fs = (ft*(16384-decay))/(16384+decay); + return fs+LAPLACE_MINP; +} + +int main(void) +{ + int i; + int ret = 0; + ec_enc enc; + ec_dec dec; + unsigned char *ptr; + int val[10000], decay[10000]; + ALLOC_STACK; + ptr = (unsigned char *)malloc(DATA_SIZE); + ec_enc_init(&enc,ptr,DATA_SIZE); + + val[0] = 3; decay[0] = 6000; + val[1] = 0; decay[1] = 5800; + val[2] = -1; decay[2] = 5600; + for (i=3;i<10000;i++) + { + val[i] = rand()%15-7; + decay[i] = rand()%11000+5000; + } + for (i=0;i<10000;i++) + ec_laplace_encode(&enc, &val[i], + ec_laplace_get_start_freq(decay[i]), decay[i]); + + ec_enc_done(&enc); + + ec_dec_init(&dec,ec_get_buffer(&enc),ec_range_bytes(&enc)); + + for (i=0;i<10000;i++) + { + int d = ec_laplace_decode(&dec, + ec_laplace_get_start_freq(decay[i]), decay[i]); + if (d != val[i]) + { + fprintf (stderr, "Got %d instead of %d\n", d, val[i]); + ret = 1; + } + } + + return ret; +} diff --git a/drivers/opus/celt/tests/test_unit_mathops.c b/drivers/opus/celt/tests/test_unit_mathops.c new file mode 100644 index 0000000000..36d6a4bfb4 --- /dev/null +++ b/drivers/opus/celt/tests/test_unit_mathops.c @@ -0,0 +1,275 @@ +/* Copyright (c) 2008-2011 Xiph.Org Foundation, Mozilla Corporation, + Gregory Maxwell + Written by Jean-Marc Valin, Gregory Maxwell, and Timothy B. Terriberry */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef OPUS_HAVE_CONFIG_H +#include "opus_config.h" +#endif + +#ifndef CUSTOM_MODES +#define CUSTOM_MODES +#endif + +#define CELT_C + +#include "mathops.c" +#include "entenc.c" +#include "entdec.c" +#include "entcode.c" +#include "bands.c" +#include "quant_bands.c" +#include "laplace.c" +#include "vq.c" +#include "cwrs.c" +#include <stdio.h> +#include <math.h> + +#ifdef OPUS_FIXED_POINT +#define WORD "%d" +#else +#define WORD "%f" +#endif + +int ret = 0; + +void testdiv(void) +{ + opus_int32 i; + for (i=1;i<=327670;i++) + { + double prod; + opus_val32 val; + val = celt_rcp(i); +#ifdef OPUS_FIXED_POINT + prod = (1./32768./65526.)*val*i; +#else + prod = val*i; +#endif + if (fabs(prod-1) > .00025) + { + fprintf (stderr, "div failed: 1/%d="WORD" (product = %f)\n", i, val, prod); + ret = 1; + } + } +} + +void testsqrt(void) +{ + opus_int32 i; + for (i=1;i<=1000000000;i++) + { + double ratio; + opus_val16 val; + val = celt_sqrt(i); + ratio = val/sqrt(i); + if (fabs(ratio - 1) > .0005 && fabs(val-sqrt(i)) > 2) + { + fprintf (stderr, "sqrt failed: sqrt(%d)="WORD" (ratio = %f)\n", i, val, ratio); + ret = 1; + } + i+= i>>10; + } +} + +void testbitexactcos(void) +{ + int i; + opus_int32 min_d,max_d,last,chk; + chk=max_d=0; + last=min_d=32767; + for(i=64;i<=16320;i++) + { + opus_int32 d; + opus_int32 q=bitexact_cos(i); + chk ^= q*i; + d = last - q; + if (d>max_d)max_d=d; + if (d<min_d)min_d=d; + last = q; + } + if ((chk!=89408644)||(max_d!=5)||(min_d!=0)||(bitexact_cos(64)!=32767)|| + (bitexact_cos(16320)!=200)||(bitexact_cos(8192)!=23171)) + { + fprintf (stderr, "bitexact_cos failed\n"); + ret = 1; + } +} + +void testbitexactlog2tan(void) +{ + int i,fail; + opus_int32 min_d,max_d,last,chk; + fail=chk=max_d=0; + last=min_d=15059; + for(i=64;i<8193;i++) + { + opus_int32 d; + opus_int32 mid=bitexact_cos(i); + opus_int32 side=bitexact_cos(16384-i); + opus_int32 q=bitexact_log2tan(mid,side); + chk ^= q*i; + d = last - q; + if (q!=-1*bitexact_log2tan(side,mid)) + fail = 1; + if (d>max_d)max_d=d; + if (d<min_d)min_d=d; + last = q; + } + if ((chk!=15821257)||(max_d!=61)||(min_d!=-2)||fail|| + (bitexact_log2tan(32767,200)!=15059)||(bitexact_log2tan(30274,12540)!=2611)|| + (bitexact_log2tan(23171,23171)!=0)) + { + fprintf (stderr, "bitexact_log2tan failed\n"); + ret = 1; + } +} + +#ifndef OPUS_FIXED_POINT +void testlog2(void) +{ + float x; + for (x=0.001;x<1677700.0;x+=(x/8.0)) + { + float error = fabs((1.442695040888963387*log(x))-celt_log2(x)); + if (error>0.0009) + { + fprintf (stderr, "celt_log2 failed: fabs((1.442695040888963387*log(x))-celt_log2(x))>0.001 (x = %f, error = %f)\n", x,error); + ret = 1; + } + } +} + +void testexp2(void) +{ + float x; + for (x=-11.0;x<24.0;x+=0.0007) + { + float error = fabs(x-(1.442695040888963387*log(celt_exp2(x)))); + if (error>0.0002) + { + fprintf (stderr, "celt_exp2 failed: fabs(x-(1.442695040888963387*log(celt_exp2(x))))>0.0005 (x = %f, error = %f)\n", x,error); + ret = 1; + } + } +} + +void testexp2log2(void) +{ + float x; + for (x=-11.0;x<24.0;x+=0.0007) + { + float error = fabs(x-(celt_log2(celt_exp2(x)))); + if (error>0.001) + { + fprintf (stderr, "celt_log2/celt_exp2 failed: fabs(x-(celt_log2(celt_exp2(x))))>0.001 (x = %f, error = %f)\n", x,error); + ret = 1; + } + } +} +#else +void testlog2(void) +{ + opus_val32 x; + for (x=8;x<1073741824;x+=(x>>3)) + { + float error = fabs((1.442695040888963387*log(x/16384.0))-celt_log2(x)/1024.0); + if (error>0.003) + { + fprintf (stderr, "celt_log2 failed: x = %ld, error = %f\n", (long)x,error); + ret = 1; + } + } +} + +void testexp2(void) +{ + opus_val16 x; + for (x=-32768;x<15360;x++) + { + float error1 = fabs(x/1024.0-(1.442695040888963387*log(celt_exp2(x)/65536.0))); + float error2 = fabs(exp(0.6931471805599453094*x/1024.0)-celt_exp2(x)/65536.0); + if (error1>0.0002&&error2>0.00004) + { + fprintf (stderr, "celt_exp2 failed: x = "WORD", error1 = %f, error2 = %f\n", x,error1,error2); + ret = 1; + } + } +} + +void testexp2log2(void) +{ + opus_val32 x; + for (x=8;x<65536;x+=(x>>3)) + { + float error = fabs(x-0.25*celt_exp2(celt_log2(x)))/16384; + if (error>0.004) + { + fprintf (stderr, "celt_log2/celt_exp2 failed: fabs(x-(celt_exp2(celt_log2(x))))>0.001 (x = %ld, error = %f)\n", (long)x,error); + ret = 1; + } + } +} + +void testilog2(void) +{ + opus_val32 x; + for (x=1;x<=268435455;x+=127) + { + opus_val32 lg; + opus_val32 y; + + lg = celt_ilog2(x); + if (lg<0 || lg>=31) + { + printf("celt_ilog2 failed: 0<=celt_ilog2(x)<31 (x = %d, celt_ilog2(x) = %d)\n",x,lg); + ret = 1; + } + y = 1<<lg; + + if (x<y || (x>>1)>=y) + { + printf("celt_ilog2 failed: 2**celt_ilog2(x)<=x<2**(celt_ilog2(x)+1) (x = %d, 2**celt_ilog2(x) = %d)\n",x,y); + ret = 1; + } + } +} +#endif + +int main(void) +{ + testbitexactcos(); + testbitexactlog2tan(); + testdiv(); + testsqrt(); + testlog2(); + testexp2(); + testexp2log2(); +#ifdef OPUS_FIXED_POINT + testilog2(); +#endif + return ret; +} diff --git a/drivers/opus/celt/tests/test_unit_mdct.c b/drivers/opus/celt/tests/test_unit_mdct.c new file mode 100644 index 0000000000..e3b5eec11c --- /dev/null +++ b/drivers/opus/celt/tests/test_unit_mdct.c @@ -0,0 +1,210 @@ +/* Copyright (c) 2008-2011 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef OPUS_HAVE_CONFIG_H +#include "opus_config.h" +#endif + +#define SKIP_CONFIG_H + +#ifndef CUSTOM_MODES +#define CUSTOM_MODES +#endif + +#include <stdio.h> + +#define CELT_C +#include "mdct.h" +#include "stack_alloc.h" + +#include "kiss_fft.c" +#include "mdct.c" +#include "mathops.c" +#include "entcode.c" + +#ifndef M_PI +#define M_PI 3.141592653 +#endif + +int ret = 0; +void check(kiss_fft_scalar * in,kiss_fft_scalar * out,int nfft,int isinverse) +{ + int bin,k; + double errpow=0,sigpow=0; + double snr; + for (bin=0;bin<nfft/2;++bin) { + double ansr = 0; + double difr; + + for (k=0;k<nfft;++k) { + double phase = 2*M_PI*(k+.5+.25*nfft)*(bin+.5)/nfft; + double re = cos(phase); + + re /= nfft/4; + + ansr += in[k] * re; + } + /*printf ("%f %f\n", ansr, out[bin]);*/ + difr = ansr - out[bin]; + errpow += difr*difr; + sigpow += ansr*ansr; + } + snr = 10*log10(sigpow/errpow); + printf("nfft=%d inverse=%d,snr = %f\n",nfft,isinverse,snr ); + if (snr<60) { + printf( "** poor snr: %f **\n", snr); + ret = 1; + } +} + +void check_inv(kiss_fft_scalar * in,kiss_fft_scalar * out,int nfft,int isinverse) +{ + int bin,k; + double errpow=0,sigpow=0; + double snr; + for (bin=0;bin<nfft;++bin) { + double ansr = 0; + double difr; + + for (k=0;k<nfft/2;++k) { + double phase = 2*M_PI*(bin+.5+.25*nfft)*(k+.5)/nfft; + double re = cos(phase); + + /*re *= 2;*/ + + ansr += in[k] * re; + } + /*printf ("%f %f\n", ansr, out[bin]);*/ + difr = ansr - out[bin]; + errpow += difr*difr; + sigpow += ansr*ansr; + } + snr = 10*log10(sigpow/errpow); + printf("nfft=%d inverse=%d,snr = %f\n",nfft,isinverse,snr ); + if (snr<60) { + printf( "** poor snr: %f **\n", snr); + ret = 1; + } +} + + +void test1d(int nfft,int isinverse) +{ + celt_mdct_lookup cfg; + size_t buflen = sizeof(kiss_fft_scalar)*nfft; + + kiss_fft_scalar * in = (kiss_fft_scalar*)malloc(buflen); + kiss_fft_scalar * in_copy = (kiss_fft_scalar*)malloc(buflen); + kiss_fft_scalar * out= (kiss_fft_scalar*)malloc(buflen); + opus_val16 * window= (opus_val16*)malloc(sizeof(opus_val16)*nfft/2); + int k; + + clt_mdct_init(&cfg, nfft, 0); + for (k=0;k<nfft;++k) { + in[k] = (rand() % 32768) - 16384; + } + + for (k=0;k<nfft/2;++k) { + window[k] = Q15ONE; + } + for (k=0;k<nfft;++k) { + in[k] *= 32768; + } + + if (isinverse) + { + for (k=0;k<nfft;++k) { + in[k] /= nfft; + } + } + + for (k=0;k<nfft;++k) + in_copy[k] = in[k]; + /*for (k=0;k<nfft;++k) printf("%d %d ", in[k].r, in[k].i);printf("\n");*/ + + if (isinverse) + { + for (k=0;k<nfft;++k) + out[k] = 0; + clt_mdct_backward(&cfg,in,out, window, nfft/2, 0, 1); + /* apply TDAC because clt_mdct_backward() no longer does that */ + for (k=0;k<nfft/4;++k) + out[nfft-k-1] = out[nfft/2+k]; + check_inv(in,out,nfft,isinverse); + } else { + clt_mdct_forward(&cfg,in,out,window, nfft/2, 0, 1); + check(in_copy,out,nfft,isinverse); + } + /*for (k=0;k<nfft;++k) printf("%d %d ", out[k].r, out[k].i);printf("\n");*/ + + + free(in); + free(out); + clt_mdct_clear(&cfg); +} + +int main(int argc,char ** argv) +{ + ALLOC_STACK; + if (argc>1) { + int k; + for (k=1;k<argc;++k) { + test1d(atoi(argv[k]),0); + test1d(atoi(argv[k]),1); + } + }else{ + test1d(32,0); + test1d(32,1); + test1d(256,0); + test1d(256,1); + test1d(512,0); + test1d(512,1); + test1d(1024,0); + test1d(1024,1); + test1d(2048,0); + test1d(2048,1); +#ifndef RADIX_TWO_ONLY + test1d(36,0); + test1d(36,1); + test1d(40,0); + test1d(40,1); + test1d(60,0); + test1d(60,1); + test1d(120,0); + test1d(120,1); + test1d(240,0); + test1d(240,1); + test1d(480,0); + test1d(480,1); + test1d(960,0); + test1d(960,1); + test1d(1920,0); + test1d(1920,1); +#endif + } + return ret; +} diff --git a/drivers/opus/celt/tests/test_unit_rotation.c b/drivers/opus/celt/tests/test_unit_rotation.c new file mode 100644 index 0000000000..c12cc3f02f --- /dev/null +++ b/drivers/opus/celt/tests/test_unit_rotation.c @@ -0,0 +1,90 @@ +/* Copyright (c) 2008-2011 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef OPUS_HAVE_CONFIG_H +#include "opus_config.h" +#endif + +#ifndef CUSTOM_MODES +#define CUSTOM_MODES +#endif + +#define CELT_C + +#include <stdio.h> +#include <stdlib.h> +#include "vq.c" +#include "cwrs.c" +#include "entcode.c" +#include "entenc.c" +#include "entdec.c" +#include "mathops.c" +#include "bands.h" +#include <math.h> +#define MAX_SIZE 100 + +int ret=0; +void test_rotation(int N, int K) +{ + int i; + double err = 0, ener = 0, snr, snr0; + opus_val16 x0[MAX_SIZE]; + opus_val16 x1[MAX_SIZE]; + for (i=0;i<N;i++) + x1[i] = x0[i] = rand()%32767-16384; + exp_rotation(x1, N, 1, 1, K, SPREAD_NORMAL); + for (i=0;i<N;i++) + { + err += (x0[i]-(double)x1[i])*(x0[i]-(double)x1[i]); + ener += x0[i]*(double)x0[i]; + } + snr0 = 20*log10(ener/err); + err = ener = 0; + exp_rotation(x1, N, -1, 1, K, SPREAD_NORMAL); + for (i=0;i<N;i++) + { + err += (x0[i]-(double)x1[i])*(x0[i]-(double)x1[i]); + ener += x0[i]*(double)x0[i]; + } + snr = 20*log10(ener/err); + printf ("SNR for size %d (%d pulses) is %f (was %f without inverse)\n", N, K, snr, snr0); + if (snr < 60 || snr0 > 20) + { + fprintf(stderr, "FAIL!\n"); + ret = 1; + } +} + +int main(void) +{ + ALLOC_STACK; + test_rotation(15, 3); + test_rotation(23, 5); + test_rotation(50, 3); + test_rotation(80, 1); + return ret; +} diff --git a/drivers/opus/celt/tests/test_unit_types.c b/drivers/opus/celt/tests/test_unit_types.c new file mode 100644 index 0000000000..29e671067f --- /dev/null +++ b/drivers/opus/celt/tests/test_unit_types.c @@ -0,0 +1,50 @@ +/* Copyright (c) 2008-2011 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef OPUS_HAVE_CONFIG_H +#include "opus_config.h" +#endif + +#include "opus_types.h" +#include <stdio.h> + +int main(void) +{ + opus_int16 i = 1; + i <<= 14; + if (i>>14 != 1) + { + fprintf(stderr, "opus_int16 isn't 16 bits\n"); + return 1; + } + if (sizeof(opus_int16)*2 != sizeof(opus_int32)) + { + fprintf(stderr, "16*2 != 32\n"); + return 1; + } + return 0; +} diff --git a/drivers/opus/celt/vq.c b/drivers/opus/celt/vq.c new file mode 100644 index 0000000000..20b0b82728 --- /dev/null +++ b/drivers/opus/celt/vq.c @@ -0,0 +1,415 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef OPUS_HAVE_CONFIG_H +#include "opus_config.h" +#endif + +#include "mathops.h" +#include "cwrs.h" +#include "vq.h" +#include "arch.h" +#include "os_support.h" +#include "bands.h" +#include "rate.h" + +static void exp_rotation1(celt_norm *X, int len, int stride, opus_val16 c, opus_val16 s) +{ + int i; + celt_norm *Xptr; + Xptr = X; + for (i=0;i<len-stride;i++) + { + celt_norm x1, x2; + x1 = Xptr[0]; + x2 = Xptr[stride]; + Xptr[stride] = EXTRACT16(SHR32(MULT16_16(c,x2) + MULT16_16(s,x1), 15)); + *Xptr++ = EXTRACT16(SHR32(MULT16_16(c,x1) - MULT16_16(s,x2), 15)); + } + Xptr = &X[len-2*stride-1]; + for (i=len-2*stride-1;i>=0;i--) + { + celt_norm x1, x2; + x1 = Xptr[0]; + x2 = Xptr[stride]; + Xptr[stride] = EXTRACT16(SHR32(MULT16_16(c,x2) + MULT16_16(s,x1), 15)); + *Xptr-- = EXTRACT16(SHR32(MULT16_16(c,x1) - MULT16_16(s,x2), 15)); + } +} + +static void exp_rotation(celt_norm *X, int len, int dir, int stride, int K, int spread) +{ + static const int SPREAD_FACTOR[3]={15,10,5}; + int i; + opus_val16 c, s; + opus_val16 gain, theta; + int stride2=0; + int factor; + + if (2*K>=len || spread==SPREAD_NONE) + return; + factor = SPREAD_FACTOR[spread-1]; + + gain = celt_div((opus_val32)MULT16_16(Q15_ONE,len),(opus_val32)(len+factor*K)); + theta = HALF16(MULT16_16_Q15(gain,gain)); + + c = celt_cos_norm(EXTEND32(theta)); + s = celt_cos_norm(EXTEND32(SUB16(Q15ONE,theta))); /* sin(theta) */ + + if (len>=8*stride) + { + stride2 = 1; + /* This is just a simple (equivalent) way of computing sqrt(len/stride) with rounding. + It's basically incrementing long as (stride2+0.5)^2 < len/stride. */ + while ((stride2*stride2+stride2)*stride + (stride>>2) < len) + stride2++; + } + /*NOTE: As a minor optimization, we could be passing around log2(B), not B, for both this and for + extract_collapse_mask().*/ + len /= stride; + for (i=0;i<stride;i++) + { + if (dir < 0) + { + if (stride2) + exp_rotation1(X+i*len, len, stride2, s, c); + exp_rotation1(X+i*len, len, 1, c, s); + } else { + exp_rotation1(X+i*len, len, 1, c, -s); + if (stride2) + exp_rotation1(X+i*len, len, stride2, s, -c); + } + } +} + +/** Takes the pitch vector and the decoded residual vector, computes the gain + that will give ||p+g*y||=1 and mixes the residual with the pitch. */ +static void normalise_residual(int * OPUS_RESTRICT iy, celt_norm * OPUS_RESTRICT X, + int N, opus_val32 Ryy, opus_val16 gain) +{ + int i; +#ifdef OPUS_FIXED_POINT + int k; +#endif + opus_val32 t; + opus_val16 g; + +#ifdef OPUS_FIXED_POINT + k = celt_ilog2(Ryy)>>1; +#endif + t = VSHR32(Ryy, 2*(k-7)); + g = MULT16_16_P15(celt_rsqrt_norm(t),gain); + + i=0; + do + X[i] = EXTRACT16(PSHR32(MULT16_16(g, iy[i]), k+1)); + while (++i < N); +} + +static unsigned extract_collapse_mask(int *iy, int N, int B) +{ + unsigned collapse_mask; + int N0; + int i; + if (B<=1) + return 1; + /*NOTE: As a minor optimization, we could be passing around log2(B), not B, for both this and for + exp_rotation().*/ + N0 = N/B; + collapse_mask = 0; + i=0; do { + int j; + j=0; do { + collapse_mask |= (iy[i*N0+j]!=0)<<i; + } while (++j<N0); + } while (++i<B); + return collapse_mask; +} + +unsigned alg_quant(celt_norm *X, int N, int K, int spread, int B, ec_enc *enc +#ifdef RESYNTH + , opus_val16 gain +#endif + ) +{ + VARDECL(celt_norm, y); + VARDECL(int, iy); + VARDECL(opus_val16, signx); + int i, j; + opus_val16 s; + int pulsesLeft; + opus_val32 sum; + opus_val32 xy; + opus_val16 yy; + unsigned collapse_mask; + SAVE_STACK; + + celt_assert2(K>0, "alg_quant() needs at least one pulse"); + celt_assert2(N>1, "alg_quant() needs at least two dimensions"); + + ALLOC(y, N, celt_norm); + ALLOC(iy, N, int); + ALLOC(signx, N, opus_val16); + + exp_rotation(X, N, 1, B, K, spread); + + /* Get rid of the sign */ + sum = 0; + j=0; do { + if (X[j]>0) + signx[j]=1; + else { + signx[j]=-1; + X[j]=-X[j]; + } + iy[j] = 0; + y[j] = 0; + } while (++j<N); + + xy = yy = 0; + + pulsesLeft = K; + + /* Do a pre-search by projecting on the pyramid */ + if (K > (N>>1)) + { + opus_val16 rcp; + j=0; do { + sum += X[j]; + } while (++j<N); + + /* If X is too small, just replace it with a pulse at 0 */ +#ifdef OPUS_FIXED_POINT + if (sum <= K) +#else + /* Prevents infinities and NaNs from causing too many pulses + to be allocated. 64 is an approximation of infinity here. */ + if (!(sum > EPSILON && sum < 64)) +#endif + { + X[0] = QCONST16(1.f,14); + j=1; do + X[j]=0; + while (++j<N); + sum = QCONST16(1.f,14); + } + rcp = EXTRACT16(MULT16_32_Q16(K-1, celt_rcp(sum))); + j=0; do { +#ifdef OPUS_FIXED_POINT + /* It's really important to round *towards zero* here */ + iy[j] = MULT16_16_Q15(X[j],rcp); +#else + iy[j] = (int)floor(rcp*X[j]); +#endif + y[j] = (celt_norm)iy[j]; + yy = MAC16_16(yy, y[j],y[j]); + xy = MAC16_16(xy, X[j],y[j]); + y[j] *= 2; + pulsesLeft -= iy[j]; + } while (++j<N); + } + celt_assert2(pulsesLeft>=1, "Allocated too many pulses in the quick pass"); + + /* This should never happen, but just in case it does (e.g. on silence) + we fill the first bin with pulses. */ +#ifdef OPUS_FIXED_POINT_DEBUG + celt_assert2(pulsesLeft<=N+3, "Not enough pulses in the quick pass"); +#endif + if (pulsesLeft > N+3) + { + opus_val16 tmp = (opus_val16)pulsesLeft; + yy = MAC16_16(yy, tmp, tmp); + yy = MAC16_16(yy, tmp, y[0]); + iy[0] += pulsesLeft; + pulsesLeft=0; + } + + s = 1; + for (i=0;i<pulsesLeft;i++) + { + int best_id; + opus_val32 best_num = -VERY_LARGE16; + opus_val16 best_den = 0; +#ifdef OPUS_FIXED_POINT + int rshift; +#endif +#ifdef OPUS_FIXED_POINT + rshift = 1+celt_ilog2(K-pulsesLeft+i+1); +#endif + best_id = 0; + /* The squared magnitude term gets added anyway, so we might as well + add it outside the loop */ + yy = ADD32(yy, 1); + j=0; + do { + opus_val16 Rxy, Ryy; + /* Temporary sums of the new pulse(s) */ + Rxy = EXTRACT16(SHR32(ADD32(xy, EXTEND32(X[j])),rshift)); + /* We're multiplying y[j] by two so we don't have to do it here */ + Ryy = ADD16(yy, y[j]); + + /* Approximate score: we maximise Rxy/sqrt(Ryy) (we're guaranteed that + Rxy is positive because the sign is pre-computed) */ + Rxy = MULT16_16_Q15(Rxy,Rxy); + /* The idea is to check for num/den >= best_num/best_den, but that way + we can do it without any division */ + /* OPT: Make sure to use conditional moves here */ + if (MULT16_16(best_den, Rxy) > MULT16_16(Ryy, best_num)) + { + best_den = Ryy; + best_num = Rxy; + best_id = j; + } + } while (++j<N); + + /* Updating the sums of the new pulse(s) */ + xy = ADD32(xy, EXTEND32(X[best_id])); + /* We're multiplying y[j] by two so we don't have to do it here */ + yy = ADD16(yy, y[best_id]); + + /* Only now that we've made the final choice, update y/iy */ + /* Multiplying y[j] by 2 so we don't have to do it everywhere else */ + y[best_id] += 2*s; + iy[best_id]++; + } + + /* Put the original sign back */ + j=0; + do { + X[j] = MULT16_16(signx[j],X[j]); + if (signx[j] < 0) + iy[j] = -iy[j]; + } while (++j<N); + encode_pulses(iy, N, K, enc); + +#ifdef RESYNTH + normalise_residual(iy, X, N, yy, gain); + exp_rotation(X, N, -1, B, K, spread); +#endif + + collapse_mask = extract_collapse_mask(iy, N, B); + RESTORE_STACK; + return collapse_mask; +} + +/** Decode pulse vector and combine the result with the pitch vector to produce + the final normalised signal in the current band. */ +unsigned alg_unquant(celt_norm *X, int N, int K, int spread, int B, + ec_dec *dec, opus_val16 gain) +{ + int i; + opus_val32 Ryy; + unsigned collapse_mask; + VARDECL(int, iy); + SAVE_STACK; + + celt_assert2(K>0, "alg_unquant() needs at least one pulse"); + celt_assert2(N>1, "alg_unquant() needs at least two dimensions"); + ALLOC(iy, N, int); + decode_pulses(iy, N, K, dec); + Ryy = 0; + i=0; + do { + Ryy = MAC16_16(Ryy, iy[i], iy[i]); + } while (++i < N); + normalise_residual(iy, X, N, Ryy, gain); + exp_rotation(X, N, -1, B, K, spread); + collapse_mask = extract_collapse_mask(iy, N, B); + RESTORE_STACK; + return collapse_mask; +} + +void renormalise_vector(celt_norm *X, int N, opus_val16 gain) +{ + int i; +#ifdef OPUS_FIXED_POINT + int k; +#endif + opus_val32 E = EPSILON; + opus_val16 g; + opus_val32 t; + celt_norm *xptr = X; + for (i=0;i<N;i++) + { + E = MAC16_16(E, *xptr, *xptr); + xptr++; + } +#ifdef OPUS_FIXED_POINT + k = celt_ilog2(E)>>1; +#endif + t = VSHR32(E, 2*(k-7)); + g = MULT16_16_P15(celt_rsqrt_norm(t),gain); + + xptr = X; + for (i=0;i<N;i++) + { + *xptr = EXTRACT16(PSHR32(MULT16_16(g, *xptr), k+1)); + xptr++; + } + /*return celt_sqrt(E);*/ +} + +int stereo_itheta(celt_norm *X, celt_norm *Y, int stereo, int N) +{ + int i; + int itheta; + opus_val16 mid, side; + opus_val32 Emid, Eside; + + Emid = Eside = EPSILON; + if (stereo) + { + for (i=0;i<N;i++) + { + celt_norm m, s; + m = ADD16(SHR16(X[i],1),SHR16(Y[i],1)); + s = SUB16(SHR16(X[i],1),SHR16(Y[i],1)); + Emid = MAC16_16(Emid, m, m); + Eside = MAC16_16(Eside, s, s); + } + } else { + for (i=0;i<N;i++) + { + celt_norm m, s; + m = X[i]; + s = Y[i]; + Emid = MAC16_16(Emid, m, m); + Eside = MAC16_16(Eside, s, s); + } + } + mid = celt_sqrt(Emid); + side = celt_sqrt(Eside); +#ifdef OPUS_FIXED_POINT + /* 0.63662 = 2/pi */ + itheta = MULT16_16_Q15(QCONST16(0.63662f,15),celt_atan2p(side, mid)); +#else + itheta = (int)floor(.5f+16384*0.63662f*atan2(side,mid)); +#endif + + return itheta; +} diff --git a/drivers/opus/celt/vq.h b/drivers/opus/celt/vq.h new file mode 100644 index 0000000000..8bab59c5e0 --- /dev/null +++ b/drivers/opus/celt/vq.h @@ -0,0 +1,70 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/** + @file vq.h + @brief Vector quantisation of the residual + */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef VQ_H +#define VQ_H + +#include "entenc.h" +#include "entdec.h" +#include "opus_modes.h" + +/** Algebraic pulse-vector quantiser. The signal x is replaced by the sum of + * the pitch and a combination of pulses such that its norm is still equal + * to 1. This is the function that will typically require the most CPU. + * @param X Residual signal to quantise/encode (returns quantised version) + * @param N Number of samples to encode + * @param K Number of pulses to use + * @param enc Entropy encoder state + * @ret A mask indicating which blocks in the band received pulses +*/ +unsigned alg_quant(celt_norm *X, int N, int K, int spread, int B, + ec_enc *enc +#ifdef RESYNTH + , opus_val16 gain +#endif + ); + +/** Algebraic pulse decoder + * @param X Decoded normalised spectrum (returned) + * @param N Number of samples to decode + * @param K Number of pulses to use + * @param dec Entropy decoder state + * @ret A mask indicating which blocks in the band received pulses + */ +unsigned alg_unquant(celt_norm *X, int N, int K, int spread, int B, + ec_dec *dec, opus_val16 gain); + +void renormalise_vector(celt_norm *X, int N, opus_val16 gain); + +int stereo_itheta(celt_norm *X, celt_norm *Y, int stereo, int N); + +#endif /* VQ_H */ diff --git a/drivers/opus/celt/x86/pitch_sse.h b/drivers/opus/celt/x86/pitch_sse.h new file mode 100644 index 0000000000..695122a5ad --- /dev/null +++ b/drivers/opus/celt/x86/pitch_sse.h @@ -0,0 +1,156 @@ +/* Copyright (c) 2013 Jean-Marc Valin and John Ridges */ +/** + @file pitch_sse.h + @brief Pitch analysis + */ + +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef PITCH_SSE_H +#define PITCH_SSE_H + +#include <xmmintrin.h> +#include "arch.h" + +#define OVERRIDE_XCORR_KERNEL +static OPUS_INLINE void xcorr_kernel(const opus_val16 *x, const opus_val16 *y, opus_val32 sum[4], int len) +{ + int j; + __m128 xsum1, xsum2; + xsum1 = _mm_loadu_ps(sum); + xsum2 = _mm_setzero_ps(); + + for (j = 0; j < len-3; j += 4) + { + __m128 x0 = _mm_loadu_ps(x+j); + __m128 yj = _mm_loadu_ps(y+j); + __m128 y3 = _mm_loadu_ps(y+j+3); + + xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0x00),yj)); + xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0x55), + _mm_shuffle_ps(yj,y3,0x49))); + xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0xaa), + _mm_shuffle_ps(yj,y3,0x9e))); + xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0xff),y3)); + } + if (j < len) + { + xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j))); + if (++j < len) + { + xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j))); + if (++j < len) + { + xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j))); + } + } + } + _mm_storeu_ps(sum,_mm_add_ps(xsum1,xsum2)); +} + +#define OVERRIDE_DUAL_INNER_PROD +static OPUS_INLINE void dual_inner_prod(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02, + int N, opus_val32 *xy1, opus_val32 *xy2) +{ + int i; + __m128 xsum1, xsum2; + xsum1 = _mm_setzero_ps(); + xsum2 = _mm_setzero_ps(); + for (i=0;i<N-3;i+=4) + { + __m128 xi = _mm_loadu_ps(x+i); + __m128 y1i = _mm_loadu_ps(y01+i); + __m128 y2i = _mm_loadu_ps(y02+i); + xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(xi, y1i)); + xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(xi, y2i)); + } + /* Horizontal sum */ + xsum1 = _mm_add_ps(xsum1, _mm_movehl_ps(xsum1, xsum1)); + xsum1 = _mm_add_ss(xsum1, _mm_shuffle_ps(xsum1, xsum1, 0x55)); + _mm_store_ss(xy1, xsum1); + xsum2 = _mm_add_ps(xsum2, _mm_movehl_ps(xsum2, xsum2)); + xsum2 = _mm_add_ss(xsum2, _mm_shuffle_ps(xsum2, xsum2, 0x55)); + _mm_store_ss(xy2, xsum2); + for (;i<N;i++) + { + *xy1 = MAC16_16(*xy1, x[i], y01[i]); + *xy2 = MAC16_16(*xy2, x[i], y02[i]); + } +} + +#define OVERRIDE_COMB_FILTER_CONST +static OPUS_INLINE void comb_filter_const(opus_val32 *y, opus_val32 *x, int T, int N, + opus_val16 g10, opus_val16 g11, opus_val16 g12) +{ + int i; + __m128 x0v; + __m128 g10v, g11v, g12v; + g10v = _mm_load1_ps(&g10); + g11v = _mm_load1_ps(&g11); + g12v = _mm_load1_ps(&g12); + x0v = _mm_loadu_ps(&x[-T-2]); + for (i=0;i<N-3;i+=4) + { + __m128 yi, yi2, x1v, x2v, x3v, x4v; + const opus_val32 *xp = &x[i-T-2]; + yi = _mm_loadu_ps(x+i); + x4v = _mm_loadu_ps(xp+4); +#if 0 + /* Slower version with all loads */ + x1v = _mm_loadu_ps(xp+1); + x2v = _mm_loadu_ps(xp+2); + x3v = _mm_loadu_ps(xp+3); +#else + x2v = _mm_shuffle_ps(x0v, x4v, 0x4e); + x1v = _mm_shuffle_ps(x0v, x2v, 0x99); + x3v = _mm_shuffle_ps(x2v, x4v, 0x99); +#endif + + yi = _mm_add_ps(yi, _mm_mul_ps(g10v,x2v)); +#if 0 /* Set to 1 to make it bit-exact with the non-SSE version */ + yi = _mm_add_ps(yi, _mm_mul_ps(g11v,_mm_add_ps(x3v,x1v))); + yi = _mm_add_ps(yi, _mm_mul_ps(g12v,_mm_add_ps(x4v,x0v))); +#else + /* Use partial sums */ + yi2 = _mm_add_ps(_mm_mul_ps(g11v,_mm_add_ps(x3v,x1v)), + _mm_mul_ps(g12v,_mm_add_ps(x4v,x0v))); + yi = _mm_add_ps(yi, yi2); +#endif + x0v=x4v; + _mm_storeu_ps(y+i, yi); + } +#ifdef CUSTOM_MODES + for (;i<N;i++) + { + y[i] = x[i] + + MULT16_32_Q15(g10,x[i-T]) + + MULT16_32_Q15(g11,ADD32(x[i-T+1],x[i-T-1])) + + MULT16_32_Q15(g12,ADD32(x[i-T+2],x[i-T-2])); + } +#endif +} + +#endif |