--- /dev/null
+/*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 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_ovflw(S_MUL((a).r,(b).r) , S_MUL((a).i,(b).i)); \
+ (m).i = ADD32_ovflw(S_MUL((a).r,(b).i) , S_MUL((a).i,(b).r)); }while(0)
+
+# define C_MULC(m,a,b) \
+ do{ (m).r = ADD32_ovflw(S_MUL((a).r,(b).r) , S_MUL((a).i,(b).i)); \
+ (m).i = SUB32_ovflw(S_MUL((a).i,(b).r) , S_MUL((a).r,(b).i)); }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_ovflw((a).r,(b).r); (res).i=ADD32_ovflw((a).i,(b).i); \
+ }while(0)
+#define C_SUB( res, a,b)\
+ do {(res).r=SUB32_ovflw((a).r,(b).r); (res).i=SUB32_ovflw((a).i,(b).i); \
+ }while(0)
+#define C_ADDTO( res , a)\
+ do {(res).r = ADD32_ovflw((res).r, (a).r); (res).i = ADD32_ovflw((res).i,(a).i);\
+ }while(0)
+
+#define C_SUBFROM( res , a)\
+ do {(res).r = ADD32_ovflw((res).r,(a).r); (res).i = SUB32_ovflw((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
+#if defined(MIPSr1_ASM)
+#include "mips/kiss_fft_mipsr1.h"
+#endif
+
+#else /* not 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 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 */
--- /dev/null
+/* 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 "common.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 MIN16(a,b) ((a) < (b) ? (a) : (b)) /**< Minimum 16-bit value. */
+#define MAX16(a,b) ((a) > (b) ? (a) : (b)) /**< Maximum 16-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))
+
+/* Set this if opus_int64 is a native type of the CPU. */
+/* Assume that all LP64 architectures have fast 64-bit types; also x86_64
+ (which can be ILP32 for x32) and Win64 (which is LLP64). */
+#if defined(__x86_64__) || defined(__LP64__) || defined(_WIN64)
+#define OPUS_FAST_INT64 1
+#else
+#define OPUS_FAST_INT64 0
+#endif
+
+#define PRINT_MIPS(file)
+
+#ifdef FIXED_POINT
+
+typedef opus_int16 opus_val16;
+typedef opus_int32 opus_val32;
+typedef opus_int64 opus_val64;
+
+typedef opus_val32 celt_sig;
+typedef opus_val16 celt_norm;
+typedef opus_val32 celt_ener;
+
+#define Q15ONE 32767
+
+#define SIG_SHIFT 12
+/* Safe saturation value for 32-bit signals. Should be less than
+ 2^31*(1-0.85) to avoid blowing up on DC at deemphasis.*/
+#define SIG_SAT (300000000)
+
+#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)
+
+#define ABS16(x) ((x) < 0 ? (-(x)) : (x))
+#define ABS32(x) ((x) < 0 ? (-(x)) : (x))
+
+static OPUS_INLINE opus_int16 SAT16(opus_int32 x) {
+ return x > 32767 ? 32767 : x < -32768 ? -32768 : (opus_int16)x;
+}
+
+#ifdef FIXED_DEBUG
+#include "fixed_debug.h"
+#else
+
+#include "fixed_generic.h"
+
+#ifdef OPUS_ARM_PRESUME_AARCH64_NEON_INTR
+#include "arm/fixed_arm64.h"
+#elif 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 /* FIXED_POINT */
+
+typedef float opus_val16;
+typedef float opus_val32;
+typedef float opus_val64;
+
+typedef float celt_sig;
+typedef float celt_norm;
+typedef float celt_ener;
+
+#ifdef FLOAT_APPROX
+/* This code should reliably detect NaN/inf even when -ffast-math is used.
+ Assumes IEEE 754 format. */
+static OPUS_INLINE int celt_isnan(float x)
+{
+ union {float f; opus_uint32 i;} in;
+ in.f = x;
+ return ((in.i>>23)&0xFF)==0xFF && (in.i&0x007FFFFF)!=0;
+}
+#else
+#ifdef __FAST_MATH__
+#error Cannot build libopus with -ffast-math unless FLOAT_APPROX is defined. This could result in crashes on extreme (e.g. NaN) input
+#endif
+#define celt_isnan(x) ((x)!=(x))
+#endif
+
+#define Q15ONE 1.0f
+
+#define NORM_SCALING 1.f
+
+#define EPSILON 1e-15f
+#define VERY_SMALL 1e-30f
+#define VERY_LARGE16 1e15f
+#define Q15_ONE ((opus_val16)1.f)
+
+/* This appears to be the same speed as C99's fabsf() but it's more portable. */
+#define ABS16(x) ((float)fabs(x))
+#define ABS32(x) ((float)fabs(x))
+
+#define QCONST16(x,bits) (x)
+#define QCONST32(x,bits) (x)
+
+#define NEG16(x) (-(x))
+#define NEG32(x) (-(x))
+#define NEG32_ovflw(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 SROUND16(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 ADD32_ovflw(a,b) ((a)+(b))
+#define SUB32_ovflw(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 MAC16_32_Q16(c,a,b) ((c)+(a)*(b))
+
+#define MULT16_16_Q11_32(a,b) ((a)*(b))
+#define MULT16_16_Q11(a,b) ((a)*(b))
+#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))
+
+#define SIG2WORD16(x) (x)
+
+#endif /* !FIXED_POINT */
+
+#ifndef GLOBAL_STACK_SIZE
+#ifdef FIXED_POINT
+#define GLOBAL_STACK_SIZE 120000
+#else
+#define GLOBAL_STACK_SIZE 120000
+#endif
+#endif
+
+#endif /* ARCH_H */
--- /dev/null
+
+
+#ifndef COMMON_H
+#define COMMON_H
+
+#define OPUS_INLINE inline
+
+#endif
--- /dev/null
+#include <stdlib.h>
+#include <string.h>
+#include <stdio.h>
+#include "kiss_fft.h"
+
+#define FRAME_SIZE_SHIFT 2
+#define FRAME_SIZE (120<<FRAME_SIZE_SHIFT)
+#define WINDOW_SIZE (2*FRAME_SIZE)
+#define FREQ_SIZE (FRAME_SIZE + 1)
+
+#define NB_BANDS 21
+
+#define SQUARE(x) ((x)*(x))
+
+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 20k*/
+ 0, 1, 2, 3, 4, 5, 6, 7, 8, 10, 12, 14, 16, 20, 24, 28, 34, 40, 48, 60, 78, 100
+};
+
+
+typedef struct {
+ int init;
+ kiss_fft_state *kfft;
+} CommonState;
+
+typedef struct {
+ float analysis_mem[FRAME_SIZE];
+ float synthesis_mem[FRAME_SIZE];
+
+} DenoiseState;
+
+void compute_band_energy(float *bandE, const kiss_fft_cpx *X) {
+ int i;
+ for (i=0;i<NB_BANDS;i++)
+ {
+ int j;
+ opus_val32 sum = 1e-27;
+ for (j=0;j<(eband5ms[i+1]-eband5ms[i])<<FRAME_SIZE_SHIFT;j++) {
+ sum += SQUARE(X[(eband5ms[i]<<FRAME_SIZE_SHIFT) + j].r);
+ sum += SQUARE(X[(eband5ms[i]<<FRAME_SIZE_SHIFT) + j].i);
+ }
+ bandE[i] = sqrt(sum);
+ }
+}
+
+void interp_band_gain(float *g, const float *bandE) {
+ int i;
+ for (i=0;i<NB_BANDS;i++)
+ {
+ int j;
+ for (j=0;j<(eband5ms[i+1]-eband5ms[i])<<FRAME_SIZE_SHIFT;j++)
+ g[(eband5ms[i]<<FRAME_SIZE_SHIFT) + j] = bandE[i];
+ }
+}
+
+CommonState common;
+
+static void check_init() {
+ if (common.init) return;
+ common.kfft = opus_fft_alloc_twiddles(2*FRAME_SIZE, NULL, NULL, NULL, 0);
+ common.init = 1;
+}
+
+
+void forward_transform(kiss_fft_cpx *out, const float *in) {
+ int i;
+ kiss_fft_cpx x[WINDOW_SIZE];
+ kiss_fft_cpx y[WINDOW_SIZE];
+ check_init();
+ for (i=0;i<WINDOW_SIZE;i++) {
+ x[i].r = in[i];
+ x[i].i = 0;
+ }
+ opus_fft(common.kfft, x, y, 0);
+ for (i=0;i<FREQ_SIZE;i++) {
+ out[i] = y[i];
+ }
+}
+
+void inverse_transform(float *out, const kiss_fft_cpx *in) {
+ check_init();
+}
+
+int main() {
+ int i;
+ kiss_fft_state *kfft;
+ float x[2*FRAME_SIZE];
+ kiss_fft_cpx y[2*FRAME_SIZE];
+ float bandE[NB_BANDS];
+ memset(x, 0, sizeof(x));
+ x[0] = 1;
+ x[1] = 1;
+ kfft = opus_fft_alloc_twiddles(2*FRAME_SIZE, NULL, NULL, NULL, 0);
+ //opus_fft(kfft, x, y, 0);
+ forward_transform(y, x);
+ compute_band_energy(bandE, y);
+ /*for (i=0;i<2*FRAME_SIZE;i++)
+ printf("%f %f\n", y[i].r, y[i].i);*/
+ for (i=0;i<NB_BANDS;i++)
+ printf("%f\n", bandE[i]);
+ return 0;
+}
--- /dev/null
+/*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 HAVE_CONFIG_H
+# include "config.h"
+# endif
+#endif
+
+#include "_kiss_fft_guts.h"
+#define CUSTOM_MODES
+
+/* 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,
+ int m,
+ int N
+ )
+{
+ kiss_fft_cpx * Fout2;
+ int i;
+ (void)m;
+#ifdef CUSTOM_MODES
+ if (m==1)
+ {
+ celt_assert(m==1);
+ for (i=0;i<N;i++)
+ {
+ kiss_fft_cpx t;
+ Fout2 = Fout + 1;
+ t = *Fout2;
+ C_SUB( *Fout2 , *Fout , t );
+ C_ADDTO( *Fout , t );
+ Fout += 2;
+ }
+ } else
+#endif
+ {
+ opus_val16 tw;
+ tw = QCONST16(0.7071067812f, 15);
+ /* We know that m==4 here because the radix-2 is just after a radix-4 */
+ celt_assert(m==4);
+ for (i=0;i<N;i++)
+ {
+ kiss_fft_cpx t;
+ Fout2 = Fout + 4;
+ t = Fout2[0];
+ C_SUB( Fout2[0] , Fout[0] , t );
+ C_ADDTO( Fout[0] , t );
+
+ t.r = S_MUL(ADD32_ovflw(Fout2[1].r, Fout2[1].i), tw);
+ t.i = S_MUL(SUB32_ovflw(Fout2[1].i, Fout2[1].r), tw);
+ C_SUB( Fout2[1] , Fout[1] , t );
+ C_ADDTO( Fout[1] , t );
+
+ t.r = Fout2[2].i;
+ t.i = -Fout2[2].r;
+ C_SUB( Fout2[2] , Fout[2] , t );
+ C_ADDTO( Fout[2] , t );
+
+ t.r = S_MUL(SUB32_ovflw(Fout2[3].i, Fout2[3].r), tw);
+ t.i = S_MUL(NEG32_ovflw(ADD32_ovflw(Fout2[3].i, Fout2[3].r)), tw);
+ C_SUB( Fout2[3] , Fout[3] , t );
+ C_ADDTO( Fout[3] , t );
+ Fout += 8;
+ }
+ }
+}
+
+static void kf_bfly4(
+ kiss_fft_cpx * Fout,
+ const size_t fstride,
+ const kiss_fft_state *st,
+ int m,
+ int N,
+ int mm
+ )
+{
+ int i;
+
+ if (m==1)
+ {
+ /* Degenerate case where all the twiddles are 1. */
+ for (i=0;i<N;i++)
+ {
+ kiss_fft_cpx scratch0, scratch1;
+
+ C_SUB( scratch0 , *Fout, Fout[2] );
+ C_ADDTO(*Fout, Fout[2]);
+ C_ADD( scratch1 , Fout[1] , Fout[3] );
+ C_SUB( Fout[2], *Fout, scratch1 );
+ C_ADDTO( *Fout , scratch1 );
+ C_SUB( scratch1 , Fout[1] , Fout[3] );
+
+ Fout[1].r = ADD32_ovflw(scratch0.r, scratch1.i);
+ Fout[1].i = SUB32_ovflw(scratch0.i, scratch1.r);
+ Fout[3].r = SUB32_ovflw(scratch0.r, scratch1.i);
+ Fout[3].i = ADD32_ovflw(scratch0.i, scratch1.r);
+ Fout+=4;
+ }
+ } else {
+ int j;
+ kiss_fft_cpx scratch[6];
+ const kiss_twiddle_cpx *tw1,*tw2,*tw3;
+ const int m2=2*m;
+ const int m3=3*m;
+ kiss_fft_cpx * Fout_beg = Fout;
+ for (i=0;i<N;i++)
+ {
+ Fout = Fout_beg + i*mm;
+ tw3 = tw2 = tw1 = st->twiddles;
+ /* m is guaranteed to be a multiple of 4. */
+ for (j=0;j<m;j++)
+ {
+ C_MUL(scratch[0],Fout[m] , *tw1 );
+ C_MUL(scratch[1],Fout[m2] , *tw2 );
+ C_MUL(scratch[2],Fout[m3] , *tw3 );
+
+ C_SUB( scratch[5] , *Fout, scratch[1] );
+ C_ADDTO(*Fout, scratch[1]);
+ C_ADD( scratch[3] , scratch[0] , scratch[2] );
+ C_SUB( scratch[4] , scratch[0] , scratch[2] );
+ C_SUB( Fout[m2], *Fout, scratch[3] );
+ tw1 += fstride;
+ tw2 += fstride*2;
+ tw3 += fstride*3;
+ C_ADDTO( *Fout , scratch[3] );
+
+ Fout[m].r = ADD32_ovflw(scratch[5].r, scratch[4].i);
+ Fout[m].i = SUB32_ovflw(scratch[5].i, scratch[4].r);
+ Fout[m3].r = SUB32_ovflw(scratch[5].r, scratch[4].i);
+ Fout[m3].i = ADD32_ovflw(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;
+#ifdef FIXED_POINT
+ /*epi3.r = -16384;*/ /* Unused */
+ epi3.i = -28378;
+#else
+ epi3 = st->twiddles[fstride*m];
+#endif
+ for (i=0;i<N;i++)
+ {
+ Fout = Fout_beg + i*mm;
+ tw1=tw2=st->twiddles;
+ /* For non-custom modes, m is guaranteed to be a multiple of 4. */
+ k=m;
+ do {
+
+ C_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 = SUB32_ovflw(Fout->r, HALF_OF(scratch[3].r));
+ Fout[m].i = SUB32_ovflw(Fout->i, HALF_OF(scratch[3].i));
+
+ C_MULBYSCALAR( scratch[0] , epi3.i );
+
+ C_ADDTO(*Fout,scratch[3]);
+
+ Fout[m2].r = ADD32_ovflw(Fout[m].r, scratch[0].i);
+ Fout[m2].i = SUB32_ovflw(Fout[m].i, scratch[0].r);
+
+ Fout[m].r = SUB32_ovflw(Fout[m].r, scratch[0].i);
+ Fout[m].i = ADD32_ovflw(Fout[m].i, scratch[0].r);
+
+ ++Fout;
+ } while(--k);
+ }
+}
+
+
+#ifndef OVERRIDE_kf_bfly5
+static void kf_bfly5(
+ kiss_fft_cpx * Fout,
+ const size_t fstride,
+ const kiss_fft_state *st,
+ int m,
+ int N,
+ int mm
+ )
+{
+ kiss_fft_cpx *Fout0,*Fout1,*Fout2,*Fout3,*Fout4;
+ int i, u;
+ kiss_fft_cpx scratch[13];
+ const kiss_twiddle_cpx *tw;
+ kiss_twiddle_cpx ya,yb;
+ kiss_fft_cpx * Fout_beg = Fout;
+
+#ifdef FIXED_POINT
+ ya.r = 10126;
+ ya.i = -31164;
+ yb.r = -26510;
+ yb.i = -19261;
+#else
+ ya = st->twiddles[fstride*m];
+ yb = st->twiddles[fstride*2*m];
+#endif
+ tw=st->twiddles;
+
+ for (i=0;i<N;i++)
+ {
+ Fout = Fout_beg + i*mm;
+ Fout0=Fout;
+ Fout1=Fout0+m;
+ Fout2=Fout0+2*m;
+ Fout3=Fout0+3*m;
+ Fout4=Fout0+4*m;
+
+ /* For non-custom modes, m is guaranteed to be a multiple of 4. */
+ for ( u=0; u<m; ++u ) {
+ scratch[0] = *Fout0;
+
+ C_MUL(scratch[1] ,*Fout1, tw[u*fstride]);
+ C_MUL(scratch[2] ,*Fout2, tw[2*u*fstride]);
+ C_MUL(scratch[3] ,*Fout3, tw[3*u*fstride]);
+ C_MUL(scratch[4] ,*Fout4, tw[4*u*fstride]);
+
+ C_ADD( scratch[7],scratch[1],scratch[4]);
+ C_SUB( scratch[10],scratch[1],scratch[4]);
+ C_ADD( scratch[8],scratch[2],scratch[3]);
+ C_SUB( scratch[9],scratch[2],scratch[3]);
+
+ Fout0->r = ADD32_ovflw(Fout0->r, ADD32_ovflw(scratch[7].r, scratch[8].r));
+ Fout0->i = ADD32_ovflw(Fout0->i, ADD32_ovflw(scratch[7].i, scratch[8].i));
+
+ scratch[5].r = ADD32_ovflw(scratch[0].r, ADD32_ovflw(S_MUL(scratch[7].r,ya.r), S_MUL(scratch[8].r,yb.r)));
+ scratch[5].i = ADD32_ovflw(scratch[0].i, ADD32_ovflw(S_MUL(scratch[7].i,ya.r), S_MUL(scratch[8].i,yb.r)));
+
+ scratch[6].r = ADD32_ovflw(S_MUL(scratch[10].i,ya.i), S_MUL(scratch[9].i,yb.i));
+ scratch[6].i = NEG32_ovflw(ADD32_ovflw(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 = ADD32_ovflw(scratch[0].r, ADD32_ovflw(S_MUL(scratch[7].r,yb.r), S_MUL(scratch[8].r,ya.r)));
+ scratch[11].i = ADD32_ovflw(scratch[0].i, ADD32_ovflw(S_MUL(scratch[7].i,yb.r), S_MUL(scratch[8].i,ya.r)));
+ scratch[12].r = SUB32_ovflw(S_MUL(scratch[9].i,ya.i), S_MUL(scratch[10].i,yb.i));
+ scratch[12].i = SUB32_ovflw(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 /* OVERRIDE_kf_bfly5 */
+
+
+#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;
+ int i;
+ int stages=0;
+ int nbak = n;
+
+ /*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[2*stages] = p;
+ if (p==2 && stages > 1)
+ {
+ facbuf[2*stages] = 4;
+ facbuf[2] = 2;
+ }
+ stages++;
+ } while (n > 1);
+ n = nbak;
+ /* Reverse the order to get the radix 4 at the end, so we can use the
+ fast degenerate case. It turns out that reversing the order also
+ improves the noise behaviour. */
+ for (i=0;i<stages/2;i++)
+ {
+ int tmp;
+ tmp = facbuf[2*i];
+ facbuf[2*i] = facbuf[2*(stages-i-1)];
+ facbuf[2*(stages-i-1)] = tmp;
+ }
+ for (i=0;i<stages;i++)
+ {
+ n /= facbuf[2*i];
+ facbuf[2*i+1] = n;
+ }
+ return 1;
+}
+
+static void compute_twiddles(kiss_twiddle_cpx *twiddles, int nfft)
+{
+ int i;
+#ifdef 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
+}
+
+int opus_fft_alloc_arch_c(kiss_fft_state *st) {
+ (void)st;
+ return 0;
+}
+
+/*
+ *
+ * Allocates all necessary storage space for the fft and ifft.
+ * The return value is a contiguous block of memory. As such,
+ * It can be freed with free().
+ * */
+kiss_fft_state *opus_fft_alloc_twiddles(int nfft,void * mem,size_t * lenmem,
+ const kiss_fft_state *base, int arch)
+{
+ 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;
+#ifdef FIXED_POINT
+ st->scale_shift = celt_ilog2(st->nfft);
+ if (st->nfft == 1<<st->scale_shift)
+ st->scale = Q15ONE;
+ else
+ st->scale = (1073741824+st->nfft/2)/st->nfft>>(15-st->scale_shift);
+#else
+ st->scale = 1.f/nfft;
+#endif
+ if (base != NULL)
+ {
+ st->twiddles = base->twiddles;
+ st->shift = 0;
+ while (st->shift < 32 && nfft<<st->shift != base->nfft)
+ 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);
+
+ /* Initialize architecture specific fft parameters */
+ if (opus_fft_alloc_arch(st, arch))
+ goto fail;
+ }
+ return st;
+fail:
+ opus_fft_free(st, arch);
+ return NULL;
+}
+
+kiss_fft_state *opus_fft_alloc(int nfft,void * mem,size_t * lenmem, int arch)
+{
+ return opus_fft_alloc_twiddles(nfft, mem, lenmem, NULL, arch);
+}
+
+void opus_fft_free_arch_c(kiss_fft_state *st) {
+ (void)st;
+}
+
+void opus_fft_free(const kiss_fft_state *cfg, int arch)
+{
+ if (cfg)
+ {
+ opus_fft_free_arch((kiss_fft_state *)cfg, arch);
+ opus_free((opus_int16*)cfg->bitrev);
+ if (cfg->shift < 0)
+ opus_free((kiss_twiddle_cpx*)cfg->twiddles);
+ opus_free((kiss_fft_state*)cfg);
+ }
+}
+
+#endif /* CUSTOM_MODES */
+
+void opus_fft_impl(const kiss_fft_state *st,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;
+
+ 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, m, fstride[i]);
+ 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_fft_c(const kiss_fft_state *st,const kiss_fft_cpx *fin,kiss_fft_cpx *fout)
+{
+ int i;
+ opus_val16 scale;
+#ifdef FIXED_POINT
+ /* Allows us to scale with MULT16_32_Q16(), which is faster than
+ MULT16_32_Q15() on ARM. */
+ int scale_shift = st->scale_shift-1;
+#endif
+ scale = st->scale;
+
+ celt_assert2 (fin != fout, "In-place FFT not supported");
+ /* Bit-reverse the input */
+ for (i=0;i<st->nfft;i++)
+ {
+ kiss_fft_cpx x = fin[i];
+ fout[st->bitrev[i]].r = SHR32(MULT16_32_Q16(scale, x.r), scale_shift);
+ fout[st->bitrev[i]].i = SHR32(MULT16_32_Q16(scale, x.i), scale_shift);
+ }
+ opus_fft_impl(st, fout);
+}
+
+
+void opus_ifft_c(const kiss_fft_state *st,const kiss_fft_cpx *fin,kiss_fft_cpx *fout)
+{
+ int i;
+ celt_assert2 (fin != fout, "In-place FFT not supported");
+ /* Bit-reverse the input */
+ for (i=0;i<st->nfft;i++)
+ fout[st->bitrev[i]] = fin[i];
+ for (i=0;i<st->nfft;i++)
+ fout[i].i = -fout[i].i;
+ opus_fft_impl(st, fout);
+ for (i=0;i<st->nfft;i++)
+ fout[i].i = -fout[i].i;
+}
--- /dev/null
+/*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"
+
+#include <stdlib.h>
+#define opus_alloc(x) malloc(x)
+#define opus_free(x) free(x)
+
+#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 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 arch_fft_state{
+ int is_supported;
+ void *priv;
+} arch_fft_state;
+
+typedef struct kiss_fft_state{
+ int nfft;
+ opus_val16 scale;
+#ifdef FIXED_POINT
+ int scale_shift;
+#endif
+ int shift;
+ opus_int16 factors[2*MAXFACTORS];
+ const opus_int16 *bitrev;
+ const kiss_twiddle_cpx *twiddles;
+ arch_fft_state *arch_fft;
+} kiss_fft_state;
+
+#if defined(HAVE_ARM_NE10)
+#include "arm/fft_arm.h"
+#endif
+
+/*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, int arch);
+
+kiss_fft_state *opus_fft_alloc(int nfft,void * mem,size_t * lenmem, int arch);
+
+/**
+ * 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_c(const kiss_fft_state *cfg,const kiss_fft_cpx *fin,kiss_fft_cpx *fout);
+void opus_ifft_c(const kiss_fft_state *cfg,const kiss_fft_cpx *fin,kiss_fft_cpx *fout);
+
+void opus_fft_impl(const kiss_fft_state *st,kiss_fft_cpx *fout);
+void opus_ifft_impl(const kiss_fft_state *st,kiss_fft_cpx *fout);
+
+void opus_fft_free(const kiss_fft_state *cfg, int arch);
+
+
+void opus_fft_free_arch_c(kiss_fft_state *st);
+int opus_fft_alloc_arch_c(kiss_fft_state *st);
+
+#if !defined(OVERRIDE_OPUS_FFT)
+/* Is run-time CPU detection enabled on this platform? */
+#if defined(OPUS_HAVE_RTCD) && (defined(HAVE_ARM_NE10))
+
+extern int (*const OPUS_FFT_ALLOC_ARCH_IMPL[OPUS_ARCHMASK+1])(
+ kiss_fft_state *st);
+
+#define opus_fft_alloc_arch(_st, arch) \
+ ((*OPUS_FFT_ALLOC_ARCH_IMPL[(arch)&OPUS_ARCHMASK])(_st))
+
+extern void (*const OPUS_FFT_FREE_ARCH_IMPL[OPUS_ARCHMASK+1])(
+ kiss_fft_state *st);
+#define opus_fft_free_arch(_st, arch) \
+ ((*OPUS_FFT_FREE_ARCH_IMPL[(arch)&OPUS_ARCHMASK])(_st))
+
+extern void (*const OPUS_FFT[OPUS_ARCHMASK+1])(const kiss_fft_state *cfg,
+ const kiss_fft_cpx *fin, kiss_fft_cpx *fout);
+#define opus_fft(_cfg, _fin, _fout, arch) \
+ ((*OPUS_FFT[(arch)&OPUS_ARCHMASK])(_cfg, _fin, _fout))
+
+extern void (*const OPUS_IFFT[OPUS_ARCHMASK+1])(const kiss_fft_state *cfg,
+ const kiss_fft_cpx *fin, kiss_fft_cpx *fout);
+#define opus_ifft(_cfg, _fin, _fout, arch) \
+ ((*OPUS_IFFT[(arch)&OPUS_ARCHMASK])(_cfg, _fin, _fout))
+
+#else /* else for if defined(OPUS_HAVE_RTCD) && (defined(HAVE_ARM_NE10)) */
+
+#define opus_fft_alloc_arch(_st, arch) \
+ ((void)(arch), opus_fft_alloc_arch_c(_st))
+
+#define opus_fft_free_arch(_st, arch) \
+ ((void)(arch), opus_fft_free_arch_c(_st))
+
+#define opus_fft(_cfg, _fin, _fout, arch) \
+ ((void)(arch), opus_fft_c(_cfg, _fin, _fout))
+
+#define opus_ifft(_cfg, _fin, _fout, arch) \
+ ((void)(arch), opus_ifft_c(_cfg, _fin, _fout))
+
+#endif /* end if defined(OPUS_HAVE_RTCD) && (defined(HAVE_ARM_NE10)) */
+#endif /* end if !defined(OVERRIDE_OPUS_FFT) */
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif
--- /dev/null
+/* (C) COPYRIGHT 1994-2002 Xiph.Org Foundation */
+/* Modified 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.
+*/
+/* opus_types.h based on ogg_types.h from libogg */
+
+/**
+ @file opus_types.h
+ @brief Opus reference implementation types
+*/
+#ifndef OPUS_TYPES_H
+#define OPUS_TYPES_H
+
+/* Use the real stdint.h if it's there (taken from Paul Hsieh's pstdint.h) */
+#if (defined(__STDC__) && __STDC__ && defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L) || (defined(__GNUC__) && (defined(_STDINT_H) || defined(_STDINT_H_)) || defined (HAVE_STDINT_H))
+#include <stdint.h>
+
+ typedef int16_t opus_int16;
+ typedef uint16_t opus_uint16;
+ typedef int32_t opus_int32;
+ typedef uint32_t opus_uint32;
+#elif defined(_WIN32)
+
+# if defined(__CYGWIN__)
+# include <_G_config.h>
+ typedef _G_int32_t opus_int32;
+ typedef _G_uint32_t opus_uint32;
+ typedef _G_int16 opus_int16;
+ typedef _G_uint16 opus_uint16;
+# elif defined(__MINGW32__)
+ typedef short opus_int16;
+ typedef unsigned short opus_uint16;
+ typedef int opus_int32;
+ typedef unsigned int opus_uint32;
+# elif defined(__MWERKS__)
+ typedef int opus_int32;
+ typedef unsigned int opus_uint32;
+ typedef short opus_int16;
+ typedef unsigned short opus_uint16;
+# else
+ /* MSVC/Borland */
+ typedef __int32 opus_int32;
+ typedef unsigned __int32 opus_uint32;
+ typedef __int16 opus_int16;
+ typedef unsigned __int16 opus_uint16;
+# endif
+
+#elif defined(__MACOS__)
+
+# include <sys/types.h>
+ typedef SInt16 opus_int16;
+ typedef UInt16 opus_uint16;
+ typedef SInt32 opus_int32;
+ typedef UInt32 opus_uint32;
+
+#elif (defined(__APPLE__) && defined(__MACH__)) /* MacOS X Framework build */
+
+# include <sys/types.h>
+ typedef int16_t opus_int16;
+ typedef u_int16_t opus_uint16;
+ typedef int32_t opus_int32;
+ typedef u_int32_t opus_uint32;
+
+#elif defined(__BEOS__)
+
+ /* Be */
+# include <inttypes.h>
+ typedef int16 opus_int16;
+ typedef u_int16 opus_uint16;
+ typedef int32_t opus_int32;
+ typedef u_int32_t opus_uint32;
+
+#elif defined (__EMX__)
+
+ /* OS/2 GCC */
+ typedef short opus_int16;
+ typedef unsigned short opus_uint16;
+ typedef int opus_int32;
+ typedef unsigned int opus_uint32;
+
+#elif defined (DJGPP)
+
+ /* DJGPP */
+ typedef short opus_int16;
+ typedef unsigned short opus_uint16;
+ typedef int opus_int32;
+ typedef unsigned int opus_uint32;
+
+#elif defined(R5900)
+
+ /* PS2 EE */
+ typedef int opus_int32;
+ typedef unsigned opus_uint32;
+ typedef short opus_int16;
+ typedef unsigned short opus_uint16;
+
+#elif defined(__SYMBIAN32__)
+
+ /* Symbian GCC */
+ typedef signed short opus_int16;
+ typedef unsigned short opus_uint16;
+ typedef signed int opus_int32;
+ typedef unsigned int opus_uint32;
+
+#elif defined(CONFIG_TI_C54X) || defined (CONFIG_TI_C55X)
+
+ typedef short opus_int16;
+ typedef unsigned short opus_uint16;
+ typedef long opus_int32;
+ typedef unsigned long opus_uint32;
+
+#elif defined(CONFIG_TI_C6X)
+
+ typedef short opus_int16;
+ typedef unsigned short opus_uint16;
+ typedef int opus_int32;
+ typedef unsigned int opus_uint32;
+
+#else
+
+ /* Give up, take a reasonable guess */
+ typedef short opus_int16;
+ typedef unsigned short opus_uint16;
+ typedef int opus_int32;
+ typedef unsigned int opus_uint32;
+
+#endif
+
+#define opus_int int /* used for counters etc; at least 16 bits */
+#define opus_int64 long long
+#define opus_int8 signed char
+
+#define opus_uint unsigned int /* used for counters etc; at least 16 bits */
+#define opus_uint64 unsigned long long
+#define opus_uint8 unsigned char
+
+#endif /* OPUS_TYPES_H */