Skip to content

Commit

Permalink
Support 32-bit window and comb filter
Browse files Browse the repository at this point in the history
  • Loading branch information
jmvalin committed Oct 4, 2024
1 parent c1a161b commit 50c2292
Show file tree
Hide file tree
Showing 17 changed files with 160 additions and 81 deletions.
19 changes: 19 additions & 0 deletions celt/arch.h
Original file line number Diff line number Diff line change
Expand Up @@ -175,8 +175,20 @@ typedef opus_val16 opus_res;

#ifdef ENABLE_QEXT
typedef opus_val32 celt_coef;
#define COEF_ONE Q31ONE
#define MULT_COEF_32(a, b) MULT32_32_Q31(a,b)
#define MAC_COEF_32_ARM(c, a, b) ADD32((c), MULT32_32_Q32(a,b))
#define MULT_COEF(a, b) MULT32_32_Q31(a,b)
#define MULT_COEF_TAPS(a, b) SHL32(MULT16_16(a,b), 1)
#define COEF2VAL16(x) EXTRACT16(SHR32(x, 16))
#else
typedef opus_val16 celt_coef;
#define COEF_ONE Q15ONE
#define MULT_COEF_32(a, b) MULT16_32_Q15(a,b)
#define MAC_COEF_32_ARM(a, b, c) MAC16_32_Q16(a,b,c)
#define MULT_COEF(a, b) MULT16_16_Q15(a,b)
#define MULT_COEF_TAPS(a, b) MULT16_16_P15(a,b)
#define COEF2VAL16(x) (x)
#endif

#define celt_isnan(x) 0
Expand Down Expand Up @@ -265,6 +277,8 @@ static OPUS_INLINE int celt_isnan(float x)

#define Q15ONE 1.0f
#define Q31ONE 1.0f
#define COEF_ONE 1.0f
#define COEF2VAL16(x) (x)

#define NORM_SCALING 1.f

Expand Down Expand Up @@ -321,6 +335,7 @@ static OPUS_INLINE int celt_isnan(float x)

#define MAC16_32_Q15(c,a,b) ((c)+(a)*(b))
#define MAC16_32_Q16(c,a,b) ((c)+(a)*(b))
#define MAC_COEF_32_ARM(c,a,b) ((c)+(a)*(b))

#define MULT16_16_Q11_32(a,b) ((a)*(b))
#define MULT16_16_Q11(a,b) ((a)*(b))
Expand All @@ -332,6 +347,10 @@ static OPUS_INLINE int celt_isnan(float x)
#define MULT16_16_P14(a,b) ((a)*(b))
#define MULT16_32_P16(a,b) ((a)*(b))

#define MULT_COEF_32(a, b) ((a)*(b))
#define MULT_COEF(a, b) ((a)*(b))
#define MULT_COEF_TAPS(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))

Expand Down
79 changes: 40 additions & 39 deletions celt/celt.c
Original file line number Diff line number Diff line change
Expand Up @@ -89,14 +89,15 @@ int resampling_factor(opus_int32 rate)
return ret;
}


#if !defined(OVERRIDE_COMB_FILTER_CONST) || defined(NON_STATIC_COMB_FILTER_CONST_C)
/* This version should be faster on ARM */
#ifdef OPUS_ARM_ASM
#ifndef NON_STATIC_COMB_FILTER_CONST_C
static
#endif
void comb_filter_const_c(opus_val32 *y, opus_val32 *x, int T, int N,
opus_val16 g10, opus_val16 g11, opus_val16 g12)
celt_coef g10, celt_coef g11, celt_coef g12)
{
opus_val32 x0, x1, x2, x3, x4;
int i;
Expand All @@ -108,33 +109,33 @@ void comb_filter_const_c(opus_val32 *y, opus_val32 *x, int T, int N,
{
opus_val32 t;
x0=SHL32(x[i-T+2],1);
t = MAC16_32_Q16(x[i], g10, x2);
t = MAC16_32_Q16(t, g11, ADD32(x1,x3));
t = MAC16_32_Q16(t, g12, ADD32(x0,x4));
t = MAC_COEF_32_ARM(x[i], g10, x2);
t = MAC_COEF_32_ARM(t, g11, ADD32(x1,x3));
t = MAC_COEF_32_ARM(t, g12, ADD32(x0,x4));
t = SATURATE(t, SIG_SAT);
y[i] = t;
x4=SHL32(x[i-T+3],1);
t = MAC16_32_Q16(x[i+1], g10, x1);
t = MAC16_32_Q16(t, g11, ADD32(x0,x2));
t = MAC16_32_Q16(t, g12, ADD32(x4,x3));
t = MAC_COEF_32_ARM(x[i+1], g10, x1);
t = MAC_COEF_32_ARM(t, g11, ADD32(x0,x2));
t = MAC_COEF_32_ARM(t, g12, ADD32(x4,x3));
t = SATURATE(t, SIG_SAT);
y[i+1] = t;
x3=SHL32(x[i-T+4],1);
t = MAC16_32_Q16(x[i+2], g10, x0);
t = MAC16_32_Q16(t, g11, ADD32(x4,x1));
t = MAC16_32_Q16(t, g12, ADD32(x3,x2));
t = MAC_COEF_32_ARM(x[i+2], g10, x0);
t = MAC_COEF_32_ARM(t, g11, ADD32(x4,x1));
t = MAC_COEF_32_ARM(t, g12, ADD32(x3,x2));
t = SATURATE(t, SIG_SAT);
y[i+2] = t;
x2=SHL32(x[i-T+5],1);
t = MAC16_32_Q16(x[i+3], g10, x4);
t = MAC16_32_Q16(t, g11, ADD32(x3,x0));
t = MAC16_32_Q16(t, g12, ADD32(x2,x1));
t = MAC_COEF_32_ARM(x[i+3], g10, x4);
t = MAC_COEF_32_ARM(t, g11, ADD32(x3,x0));
t = MAC_COEF_32_ARM(t, g12, ADD32(x2,x1));
t = SATURATE(t, SIG_SAT);
y[i+3] = t;
x1=SHL32(x[i-T+6],1);
t = MAC16_32_Q16(x[i+4], g10, x3);
t = MAC16_32_Q16(t, g11, ADD32(x2,x4));
t = MAC16_32_Q16(t, g12, ADD32(x1,x0));
t = MAC_COEF_32_ARM(x[i+4], g10, x3);
t = MAC_COEF_32_ARM(t, g11, ADD32(x2,x4));
t = MAC_COEF_32_ARM(t, g12, ADD32(x1,x0));
t = SATURATE(t, SIG_SAT);
y[i+4] = t;
}
Expand All @@ -143,9 +144,9 @@ void comb_filter_const_c(opus_val32 *y, opus_val32 *x, int T, int N,
{
opus_val32 t;
x0=SHL32(x[i-T+2],1);
t = MAC16_32_Q16(x[i], g10, x2);
t = MAC16_32_Q16(t, g11, ADD32(x1,x3));
t = MAC16_32_Q16(t, g12, ADD32(x0,x4));
t = MAC_COEF_32_ARM(x[i], g10, x2);
t = MAC_COEF_32_ARM(t, g11, ADD32(x1,x3));
t = MAC_COEF_32_ARM(t, g12, ADD32(x0,x4));
t = SATURATE(t, SIG_SAT);
y[i] = t;
x4=x3;
Expand All @@ -160,7 +161,7 @@ void comb_filter_const_c(opus_val32 *y, opus_val32 *x, int T, int N,
static
#endif
void comb_filter_const_c(opus_val32 *y, opus_val32 *x, int T, int N,
opus_val16 g10, opus_val16 g11, opus_val16 g12)
celt_coef g10, celt_coef g11, celt_coef g12)
{
opus_val32 x0, x1, x2, x3, x4;
int i;
Expand All @@ -172,9 +173,9 @@ void comb_filter_const_c(opus_val32 *y, opus_val32 *x, int T, int N,
{
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));
+ MULT_COEF_32(g10,x2)
+ MULT_COEF_32(g11,ADD32(x1,x3))
+ MULT_COEF_32(g12,ADD32(x0,x4));
y[i] = SATURATE(y[i], SIG_SAT);
x4=x3;
x3=x2;
Expand All @@ -189,11 +190,11 @@ void comb_filter_const_c(opus_val32 *y, opus_val32 *x, int T, int N,
#ifndef OVERRIDE_comb_filter
void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N,
opus_val16 g0, opus_val16 g1, int tapset0, int tapset1,
const opus_val16 *window, int overlap, int arch)
const celt_coef *window, int overlap, int arch)
{
int i;
/* printf ("%d %d %f %f\n", T0, T1, g0, g1); */
opus_val16 g00, g01, g02, g10, g11, g12;
celt_coef 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)},
Expand All @@ -211,12 +212,12 @@ void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N,
to have then be at least 2 to avoid processing garbage data. */
T0 = IMAX(T0, COMBFILTER_MINPERIOD);
T1 = IMAX(T1, COMBFILTER_MINPERIOD);
g00 = MULT16_16_P15(g0, gains[tapset0][0]);
g01 = MULT16_16_P15(g0, gains[tapset0][1]);
g02 = MULT16_16_P15(g0, gains[tapset0][2]);
g10 = MULT16_16_P15(g1, gains[tapset1][0]);
g11 = MULT16_16_P15(g1, gains[tapset1][1]);
g12 = MULT16_16_P15(g1, gains[tapset1][2]);
g00 = MULT_COEF_TAPS(g0, gains[tapset0][0]);
g01 = MULT_COEF_TAPS(g0, gains[tapset0][1]);
g02 = MULT_COEF_TAPS(g0, gains[tapset0][2]);
g10 = MULT_COEF_TAPS(g1, gains[tapset1][0]);
g11 = MULT_COEF_TAPS(g1, gains[tapset1][1]);
g12 = MULT_COEF_TAPS(g1, gains[tapset1][2]);
x1 = x[-T1+1];
x2 = x[-T1 ];
x3 = x[-T1-1];
Expand All @@ -226,16 +227,16 @@ void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N,
overlap=0;
for (i=0;i<overlap;i++)
{
opus_val16 f;
celt_coef f;
x0=x[i-T1+2];
f = MULT16_16_Q15(window[i],window[i]);
f = MULT_COEF(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));
+ MULT_COEF_32(MULT_COEF((COEF_ONE-f),g00),x[i-T0])
+ MULT_COEF_32(MULT_COEF((COEF_ONE-f),g01),ADD32(x[i-T0+1],x[i-T0-1]))
+ MULT_COEF_32(MULT_COEF((COEF_ONE-f),g02),ADD32(x[i-T0+2],x[i-T0-2]))
+ MULT_COEF_32(MULT_COEF(f,g10),x2)
+ MULT_COEF_32(MULT_COEF(f,g11),ADD32(x1,x3))
+ MULT_COEF_32(MULT_COEF(f,g12),ADD32(x0,x4));
y[i] = SATURATE(y[i], SIG_SAT);
x4=x3;
x3=x2;
Expand Down
3 changes: 2 additions & 1 deletion celt/celt.h
Original file line number Diff line number Diff line change
Expand Up @@ -41,6 +41,7 @@
#include "entenc.h"
#include "entdec.h"
#include "arch.h"
#include "kiss_fft.h"

#ifdef ENABLE_DEEP_PLC
#include "lpcnet.h"
Expand Down Expand Up @@ -236,7 +237,7 @@ void celt_preemphasis(const opus_res * OPUS_RESTRICT pcmp, celt_sig * OPUS_RESTR

void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N,
opus_val16 g0, opus_val16 g1, int tapset0, int tapset1,
const opus_val16 *window, int overlap, int arch);
const celt_coef *window, int overlap, int arch);

void init_caps(const CELTMode *m,int *cap,int LM,int C);

Expand Down
8 changes: 4 additions & 4 deletions celt/celt_decoder.c
Original file line number Diff line number Diff line change
Expand Up @@ -535,8 +535,8 @@ static void prefilter_and_fold(CELTDecoder * OPUS_RESTRICT st, int N)
for (i=0;i<overlap/2;i++)
{
decode_mem[c][DECODE_BUFFER_SIZE-N+i] =
MULT16_32_Q15(mode->window[i], etmp[overlap-1-i])
+ MULT16_32_Q15(mode->window[overlap-i-1], etmp[i]);
MULT16_32_Q15(COEF2VAL16(mode->window[i]), etmp[overlap-1-i])
+ MULT16_32_Q15 (COEF2VAL16(mode->window[overlap-i-1]), etmp[i]);
}
} while (++c<CC);
}
Expand Down Expand Up @@ -692,7 +692,7 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM
} else {
int exc_length;
/* Pitch-based PLC */
const opus_val16 *window;
const celt_coef *window;
opus_val16 *exc;
opus_val16 fade = Q15ONE;
int pitch_index;
Expand Down Expand Up @@ -880,7 +880,7 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM
for (i=0;i<overlap;i++)
{
opus_val16 tmp_g = Q15ONE
- MULT16_16_Q15(window[i], Q15ONE-ratio);
- MULT16_16_Q15(COEF2VAL16(window[i]), Q15ONE-ratio);
buf[DECODE_BUFFER_SIZE-N+i] =
MULT16_32_Q15(tmp_g, buf[DECODE_BUFFER_SIZE-N+i]);
}
Expand Down
7 changes: 4 additions & 3 deletions celt/celt_lpc.c
Original file line number Diff line number Diff line change
Expand Up @@ -277,7 +277,7 @@ void celt_iir(const opus_val32 *_x,
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,
const celt_coef *window,
int overlap,
int lag,
int n,
Expand All @@ -302,8 +302,9 @@ int _celt_autocorr(
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]);
opus_val16 w = COEF2VAL16(window[i]);
xx[i] = MULT16_16_Q15(x[i],w);
xx[n-i-1] = MULT16_16_Q15(x[n-i-1],w);
}
xptr = xx;
}
Expand Down
2 changes: 1 addition & 1 deletion celt/celt_lpc.h
Original file line number Diff line number Diff line change
Expand Up @@ -61,6 +61,6 @@ void celt_iir(const opus_val32 *x,
int arch);

int _celt_autocorr(const opus_val16 *x, opus_val32 *ac,
const opus_val16 *window, int overlap, int lag, int n, int arch);
const celt_coef *window, int overlap, int lag, int n, int arch);

#endif /* PLC_H */
12 changes: 11 additions & 1 deletion celt/dump_modes/dump_modes.c
Original file line number Diff line number Diff line change
Expand Up @@ -99,9 +99,19 @@ void dump_modes(FILE *file, CELTMode **modes, int nb_modes)

fprintf(file, "#ifndef DEF_WINDOW%d\n", mode->overlap);
fprintf(file, "#define DEF_WINDOW%d\n", mode->overlap);
fprintf (file, "static const opus_val16 window%d[%d] = {\n", mode->overlap, mode->overlap);
fprintf (file, "static const celt_coef window%d[%d] = {\n", mode->overlap, mode->overlap);
#if defined(FIXED_POINT) && defined(ENABLE_QEXT)
fprintf(file, "#ifdef ENABLE_QEXT\n");
for (j=0;j<mode->overlap;j++)
fprintf (file, WORD32 ",%c", mode->window[j],(j+6)%5==0?'\n':' ');
fprintf(file, "#else\n");
for (j=0;j<mode->overlap;j++)
fprintf (file, WORD16 ",%c", COEF16(mode->window[j], 16),(j+6)%5==0?'\n':' ');
fprintf(file, "#endif\n");
#else
for (j=0;j<mode->overlap;j++)
fprintf (file, WORD16 ",%c", mode->window[j],(j+6)%5==0?'\n':' ');
#endif
fprintf (file, "};\n");
fprintf(file, "#endif\n");
fprintf(file, "\n");
Expand Down
1 change: 1 addition & 0 deletions celt/fixed_debug.h
Original file line number Diff line number Diff line change
Expand Up @@ -43,6 +43,7 @@ extern opus_int64 celt_mips;

#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))
#define MULT32_32_Q32(a,b) ADD32(ADD32(MULT16_16(SHR((a),16),SHR((b),16)), SHR(MULT16_16SU(SHR((a),16),((b)&0x0000ffff)),16)), SHR(MULT16_16SU(SHR((b),16),((a)&0x0000ffff)),16))

/** 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))
Expand Down
7 changes: 7 additions & 0 deletions celt/fixed_generic.h
Original file line number Diff line number Diff line change
Expand Up @@ -71,6 +71,13 @@
#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))
#endif

/** 32x32 multiplication, followed by a 32-bit shift right. Results fits in 32 bits */
#if OPUS_FAST_INT64
#define MULT32_32_Q32(a,b) ((opus_val32)SHR((opus_int64)(a)*(opus_int64)(b),32))
#else
#define MULT32_32_Q32(a,b) ADD32(ADD32(MULT16_16(SHR((a),16),SHR((b),16)), SHR(MULT16_16SU(SHR((a),16),((b)&0x0000ffff)),16)), SHR(MULT16_16SU(SHR((b),16),((a)&0x0000ffff)),16))
#endif

/** Compile-time conversion of float constant to 16-bit value */
#define QCONST16(x,bits) ((opus_val16)(.5+(x)*(((opus_val32)1)<<(bits))))

Expand Down
24 changes: 12 additions & 12 deletions celt/mdct.c
Original file line number Diff line number Diff line change
Expand Up @@ -120,7 +120,7 @@ void clt_mdct_clear(mdct_lookup *l, int arch)
/* Forward MDCT trashes the input array */
#ifndef OVERRIDE_clt_mdct_forward
void clt_mdct_forward_c(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out,
const opus_val16 *window, int overlap, int shift, int stride, int arch)
const celt_coef *window, int overlap, int shift, int stride, int arch)
{
int i;
int N, N2, N4;
Expand Down Expand Up @@ -159,13 +159,13 @@ void clt_mdct_forward_c(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scal
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;
const celt_coef * OPUS_RESTRICT wp1 = window+(overlap>>1);
const celt_coef * 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]);
*yp++ = S_MUL(xp1[N2], *wp2) + S_MUL(*xp2, *wp1);
*yp++ = S_MUL(*xp1, *wp1) - S_MUL(xp2[-N2], *wp2);
xp1+=2;
xp2-=2;
wp1+=2;
Expand All @@ -184,8 +184,8 @@ void clt_mdct_forward_c(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scal
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]);
*yp++ = -S_MUL(xp1[-N2], *wp1) + S_MUL(*xp2, *wp2);
*yp++ = S_MUL(*xp1, *wp2) + S_MUL(xp2[N2], *wp1);
xp1+=2;
xp2-=2;
wp1+=2;
Expand Down Expand Up @@ -258,7 +258,7 @@ void clt_mdct_forward_c(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scal

#ifndef OVERRIDE_clt_mdct_backward
void clt_mdct_backward_c(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out,
const opus_val16 * OPUS_RESTRICT window, int overlap, int shift, int stride, int arch)
const celt_coef * OPUS_RESTRICT window, int overlap, int shift, int stride, int arch)
{
int i;
int N, N2, N4;
Expand Down Expand Up @@ -346,16 +346,16 @@ void clt_mdct_backward_c(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_sca
{
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;
const celt_coef * OPUS_RESTRICT wp1 = window;
const celt_coef * OPUS_RESTRICT wp2 = window+overlap-1;

for(i = 0; i < overlap/2; i++)
{
kiss_fft_scalar x1, x2;
x1 = *xp1;
x2 = *yp1;
*yp1++ = SUB32_ovflw(MULT16_32_Q15(*wp2, x2), MULT16_32_Q15(*wp1, x1));
*xp1-- = ADD32_ovflw(MULT16_32_Q15(*wp1, x2), MULT16_32_Q15(*wp2, x1));
*yp1++ = SUB32_ovflw(S_MUL(x2, *wp2), S_MUL(x1, *wp1));
*xp1-- = ADD32_ovflw(S_MUL(x2, *wp1), S_MUL(x1, *wp2));
wp1++;
wp2--;
}
Expand Down
Loading

0 comments on commit 50c2292

Please sign in to comment.