@@ -2106,7 +2106,7 @@ CONFIG_EXTRA="
huffyuvencdsp
idctdsp
iirfilter
- imdct15
+ mdct15
intrax8
iso_media
ividsp
@@ -2348,7 +2348,7 @@ vc1dsp_select="h264chroma qpeldsp startcode"
rdft_select="fft"
# decoders / encoders
-aac_decoder_select="imdct15 mdct sinewin"
+aac_decoder_select="mdct15 mdct sinewin"
aac_fixed_decoder_select="mdct sinewin"
aac_encoder_select="audio_frame_queue iirfilter lpc mdct sinewin"
aac_latm_decoder_select="aac_decoder aac_latm_parser"
@@ -2490,7 +2490,7 @@ nellymoser_encoder_select="audio_frame_queue mdct sinewin"
nuv_decoder_select="idctdsp lzo"
on2avc_decoder_select="mdct"
opus_decoder_deps="swresample"
-opus_decoder_select="imdct15"
+opus_decoder_select="mdct15"
png_decoder_select="zlib"
png_encoder_select="llvidencdsp zlib"
prores_decoder_select="blockdsp idctdsp"
@@ -83,7 +83,7 @@ OBJS-$(CONFIG_HUFFYUVDSP) += huffyuvdsp.o
OBJS-$(CONFIG_HUFFYUVENCDSP) += huffyuvencdsp.o
OBJS-$(CONFIG_IDCTDSP) += idctdsp.o simple_idct.o jrevdct.o
OBJS-$(CONFIG_IIRFILTER) += iirfilter.o
-OBJS-$(CONFIG_IMDCT15) += imdct15.o
+OBJS-$(CONFIG_MDCT15) += mdct15.o
OBJS-$(CONFIG_INTRAX8) += intrax8.o intrax8dsp.o
OBJS-$(CONFIG_IVIDSP) += ivi_dsp.o
OBJS-$(CONFIG_JNI) += ffjni.o jni.o
@@ -36,7 +36,7 @@
#include "libavutil/fixed_dsp.h"
#include "avcodec.h"
#if !USE_FIXED
-#include "imdct15.h"
+#include "mdct15.h"
#endif
#include "fft.h"
#include "mpeg4audio.h"
@@ -327,7 +327,7 @@ struct AACContext {
#if USE_FIXED
AVFixedDSPContext *fdsp;
#else
- IMDCT15Context *mdct480;
+ MDCT15Context *mdct480;
AVFloatDSPContext *fdsp;
#endif /* USE_FIXED */
int random_state;
@@ -42,7 +42,7 @@
#include "internal.h"
#include "get_bits.h"
#include "fft.h"
-#include "imdct15.h"
+#include "mdct15.h"
#include "lpc.h"
#include "kbdwin.h"
#include "sinewin.h"
@@ -1185,7 +1185,7 @@ static av_cold int aac_decode_init(AVCodecContext *avctx)
AAC_RENAME_32(ff_mdct_init)(&ac->mdct_small, 8, 1, 1.0 / RANGE15(128.0));
AAC_RENAME_32(ff_mdct_init)(&ac->mdct_ltp, 11, 0, RANGE15(-2.0));
#if !USE_FIXED
- ret = ff_imdct15_init(&ac->mdct480, 5);
+ ret = ff_mdct15_init(&ac->mdct480, 1, 5, -1.0f);
if (ret < 0)
return ret;
#endif
@@ -3192,7 +3192,7 @@ static av_cold int aac_decode_close(AVCodecContext *avctx)
ff_mdct_end(&ac->mdct_ld);
ff_mdct_end(&ac->mdct_ltp);
#if !USE_FIXED
- ff_imdct15_uninit(&ac->mdct480);
+ ff_mdct15_uninit(&ac->mdct480);
#endif
av_freep(&ac->fdsp);
return 0;
new file mode 100644
@@ -0,0 +1,351 @@
+/*
+ * Copyright (c) 2013-2014 Mozilla Corporation
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+/**
+ * @file
+ * Celt non-power of 2 iMDCT
+ */
+
+#include <float.h>
+#include <math.h>
+#include <stddef.h>
+
+#include "config.h"
+
+#include "libavutil/attributes.h"
+#include "libavutil/common.h"
+
+#include "avfft.h"
+#include "mdct15.h"
+
+// complex c = a * b
+#define CMUL3(cre, cim, are, aim, bre, bim) \
+do { \
+ cre = are * bre - aim * bim; \
+ cim = are * bim + aim * bre; \
+} while (0)
+
+#define CMUL(c, a, b) CMUL3((c).re, (c).im, (a).re, (a).im, (b).re, (b).im)
+
+av_cold void ff_mdct15_uninit(MDCT15Context **ps)
+{
+ MDCT15Context *s = *ps;
+
+ if (!s)
+ return;
+
+ ff_fft_end(&s->ptwo_fft);
+
+ av_freep(&s->pfa_prereindex);
+ av_freep(&s->pfa_postreindex);
+ av_freep(&s->twiddle_exptab);
+ av_freep(&s->tmp);
+
+ av_freep(ps);
+}
+
+static void mdct15(MDCT15Context *s, float *dst, const float *src, ptrdiff_t stride);
+
+static void imdct15_half(MDCT15Context *s, float *dst, const float *src,
+ ptrdiff_t stride, float scale);
+
+static inline int init_pfa_reindex_tabs(MDCT15Context *s)
+{
+ int i, j;
+ const int b_ptwo = s->ptwo_fft.nbits; /* Bits for the power of two FFTs */
+ const int l_ptwo = 1 << b_ptwo; /* Total length for the power of two FFTs */
+ const int inv_1 = l_ptwo << ((4 - b_ptwo) & 3); /* (2^b_ptwo)^-1 mod 15 */
+ const int inv_2 = 0xeeeeeeef & ((1U << b_ptwo) - 1); /* 15^-1 mod 2^b_ptwo */
+
+ s->pfa_prereindex = av_malloc(15 * l_ptwo * sizeof(*s->pfa_prereindex));
+ if (!s->pfa_prereindex)
+ return 1;
+
+ s->pfa_postreindex = av_malloc(15 * l_ptwo * sizeof(*s->pfa_postreindex));
+ if (!s->pfa_postreindex)
+ return 1;
+
+ /* Pre/Post-reindex */
+ for (i = 0; i < l_ptwo; i++) {
+ for (j = 0; j < 15; j++) {
+ const int q_pre = ((l_ptwo * j)/15 + i) >> b_ptwo;
+ const int q_post = (((j*inv_1)/15) + (i*inv_2)) >> b_ptwo;
+ const int k_pre = 15*i + ((j - q_pre*15) << b_ptwo);
+ const int k_post = i*inv_2*15 + j*inv_1 - 15*q_post*l_ptwo;
+ s->pfa_prereindex[i*15 + j] = k_pre;
+ s->pfa_postreindex[k_post] = l_ptwo*j + i;
+ }
+ }
+
+ return 0;
+}
+
+av_cold int ff_mdct15_init(MDCT15Context **ps, int inverse, int N, double scale)
+{
+ MDCT15Context *s;
+ double alpha, theta;
+ int len2 = 15 * (1 << N);
+ int len = 2 * len2;
+ int i;
+
+ /* Tested and verified to work on everything in between */
+ if ((N < 2) || (N > 13))
+ return AVERROR(EINVAL);
+
+ s = av_mallocz(sizeof(*s));
+ if (!s)
+ return AVERROR(ENOMEM);
+
+ s->fft_n = N - 1;
+ s->len4 = len2 / 2;
+ s->len2 = len2;
+ s->inverse = inverse;
+ s->mdct = mdct15;
+ s->imdct_half = imdct15_half;
+
+ if (ff_fft_init(&s->ptwo_fft, N - 1, s->inverse) < 0)
+ goto fail;
+
+ if (init_pfa_reindex_tabs(s))
+ goto fail;
+
+ s->tmp = av_malloc_array(len, 2 * sizeof(*s->tmp));
+ if (!s->tmp)
+ goto fail;
+
+ s->twiddle_exptab = av_malloc_array(s->len4, sizeof(*s->twiddle_exptab));
+ if (!s->twiddle_exptab)
+ goto fail;
+
+ theta = 0.125f + (scale < 0 ? s->len4 : 0);
+ scale = sqrt(fabs(scale));
+ for (i = 0; i < s->len4; i++) {
+ alpha = 2 * M_PI * (i + theta) / len;
+ s->twiddle_exptab[i].re = cos(alpha) * scale;
+ s->twiddle_exptab[i].im = sin(alpha) * scale;
+ }
+
+ /* 15-point FFT exptab */
+ for (i = 0; i < 19; i++) {
+ if (i < 15) {
+ double theta = (2.0f * M_PI * i) / 15.0f;
+ if (!s->inverse)
+ theta *= -1;
+ s->exptab[i].re = cos(theta);
+ s->exptab[i].im = sin(theta);
+ } else { /* Wrap around to simplify fft15 */
+ s->exptab[i] = s->exptab[i - 15];
+ }
+ }
+
+ /* 5-point FFT exptab */
+ s->exptab[19].re = cos(2.0f * M_PI / 5.0f);
+ s->exptab[19].im = sin(2.0f * M_PI / 5.0f);
+ s->exptab[20].re = cos(1.0f * M_PI / 5.0f);
+ s->exptab[20].im = sin(1.0f * M_PI / 5.0f);
+
+ /* Invert the phase for an inverse transform, do nothing for a forward transform */
+ if (s->inverse) {
+ s->exptab[19].im *= -1;
+ s->exptab[20].im *= -1;
+ }
+
+ *ps = s;
+
+ return 0;
+
+fail:
+ ff_mdct15_uninit(&s);
+ return AVERROR(ENOMEM);
+}
+
+/* Stride is hardcoded to 3 */
+static inline void fft5(const FFTComplex exptab[2], FFTComplex *out,
+ const FFTComplex *in)
+{
+ FFTComplex z0[4], t[6];
+
+ t[0].re = in[3].re + in[12].re;
+ t[0].im = in[3].im + in[12].im;
+ t[1].im = in[3].re - in[12].re;
+ t[1].re = in[3].im - in[12].im;
+ t[2].re = in[6].re + in[ 9].re;
+ t[2].im = in[6].im + in[ 9].im;
+ t[3].im = in[6].re - in[ 9].re;
+ t[3].re = in[6].im - in[ 9].im;
+
+ out[0].re = in[0].re + in[3].re + in[6].re + in[9].re + in[12].re;
+ out[0].im = in[0].im + in[3].im + in[6].im + in[9].im + in[12].im;
+
+ t[4].re = exptab[0].re * t[2].re - exptab[1].re * t[0].re;
+ t[4].im = exptab[0].re * t[2].im - exptab[1].re * t[0].im;
+ t[0].re = exptab[0].re * t[0].re - exptab[1].re * t[2].re;
+ t[0].im = exptab[0].re * t[0].im - exptab[1].re * t[2].im;
+ t[5].re = exptab[0].im * t[3].re - exptab[1].im * t[1].re;
+ t[5].im = exptab[0].im * t[3].im - exptab[1].im * t[1].im;
+ t[1].re = exptab[0].im * t[1].re + exptab[1].im * t[3].re;
+ t[1].im = exptab[0].im * t[1].im + exptab[1].im * t[3].im;
+
+ z0[0].re = t[0].re - t[1].re;
+ z0[0].im = t[0].im - t[1].im;
+ z0[1].re = t[4].re + t[5].re;
+ z0[1].im = t[4].im + t[5].im;
+
+ z0[2].re = t[4].re - t[5].re;
+ z0[2].im = t[4].im - t[5].im;
+ z0[3].re = t[0].re + t[1].re;
+ z0[3].im = t[0].im + t[1].im;
+
+ out[1].re = in[0].re + z0[3].re;
+ out[1].im = in[0].im + z0[0].im;
+ out[2].re = in[0].re + z0[2].re;
+ out[2].im = in[0].im + z0[1].im;
+ out[3].re = in[0].re + z0[1].re;
+ out[3].im = in[0].im + z0[2].im;
+ out[4].re = in[0].re + z0[0].re;
+ out[4].im = in[0].im + z0[3].im;
+}
+
+static __attribute__ ((noinline)) void fft15(const FFTComplex exptab[22], FFTComplex *out,
+ const FFTComplex *in, size_t stride)
+{
+ int k;
+ FFTComplex tmp1[5], tmp2[5], tmp3[5];
+
+ fft5(exptab + 19, tmp1, in + 0);
+ fft5(exptab + 19, tmp2, in + 1);
+ fft5(exptab + 19, tmp3, in + 2);
+
+ for (k = 0; k < 5; k++) {
+ FFTComplex t[2];
+
+ CMUL(t[0], tmp2[k], exptab[k]);
+ CMUL(t[1], tmp3[k], exptab[2 * k]);
+ out[stride*k].re = tmp1[k].re + t[0].re + t[1].re;
+ out[stride*k].im = tmp1[k].im + t[0].im + t[1].im;
+
+ CMUL(t[0], tmp2[k], exptab[k + 5]);
+ CMUL(t[1], tmp3[k], exptab[2 * (k + 5)]);
+ out[stride*(k + 5)].re = tmp1[k].re + t[0].re + t[1].re;
+ out[stride*(k + 5)].im = tmp1[k].im + t[0].im + t[1].im;
+
+ CMUL(t[0], tmp2[k], exptab[k + 10]);
+ CMUL(t[1], tmp3[k], exptab[2 * k + 5]);
+ out[stride*(k + 10)].re = tmp1[k].re + t[0].re + t[1].re;
+ out[stride*(k + 10)].im = tmp1[k].im + t[0].im + t[1].im;
+ }
+}
+
+int count = 0;
+
+static void mdct15(MDCT15Context *s, float *dst, const float *src, ptrdiff_t stride)
+{
+ int i, j;
+ FFTComplex fft15in[15];
+
+ int len1 = s->len4 << 2;
+ int len2 = s->len4 << 1;
+ int len3 = s->len4 * 3;
+ int len4 = s->len4;
+ int len8 = s->len4 >> 1;
+
+ int l_ptwo = 1 << s->ptwo_fft.nbits;
+
+ FFTComplex *z = (FFTComplex *)dst;
+ FFTComplex x[960];
+
+ /* pre rotation */
+ for(i = 0; i < len8; i++) {
+ float re, im;
+
+ re = -src[2*i+len3] - src[len3-1-2*i];
+ im = -src[len4+2*i] + src[len4-1-2*i];
+ j = i;
+ CMUL3(x[j].re, x[j].im, re, im, s->twiddle_exptab[i].re, -s->twiddle_exptab[i].im);
+
+ re = src[2*i] - src[len2-1-2*i];
+ im = -src[len2+2*i] - src[len1-1-2*i];
+ j = len8 + i;
+ CMUL3(x[j].re, x[j].im, re, im, s->twiddle_exptab[len8 + i].re, -s->twiddle_exptab[len8 + i].im);
+ }
+
+ for (i = 0; i < l_ptwo; i++) {
+ for (j = 0; j < 15; j++) {
+ const int k = s->pfa_prereindex[i*15 + j];
+ fft15in[j] = x[k];
+ }
+ fft15(s->exptab, s->tmp + s->ptwo_fft.revtab[i], fft15in, l_ptwo);
+ }
+
+ /* Then a 15xN FFT (where N is a power of two) */
+ for (i = 0; i < 15; i++)
+ s->ptwo_fft.fft_calc(&s->ptwo_fft, s->tmp + l_ptwo*i);
+
+ /* Reindex again, apply twiddles and output */
+ for (i = 0; i < len8; i++) {
+ float re0, im0, re1, im1;
+ const int i0 = len8 + i, i1 = len8 - i - 1;
+ const int s0 = s->pfa_postreindex[i0], s1 = s->pfa_postreindex[i1];
+
+ CMUL3(im1, re0, s->tmp[s1].re, s->tmp[s1].im, s->twiddle_exptab[i1].im, s->twiddle_exptab[i1].re);
+ CMUL3(im0, re1, s->tmp[s0].re, s->tmp[s0].im, s->twiddle_exptab[i0].im, s->twiddle_exptab[i0].re);
+ z[i1].re = re0;
+ z[i1].im = im0;
+ z[i0].re = re1;
+ z[i0].im = im1;
+ }
+}
+
+static void imdct15_half(MDCT15Context *s, float *dst, const float *src,
+ ptrdiff_t stride, float scale)
+{
+ FFTComplex fft15in[15];
+ FFTComplex *z = (FFTComplex *)dst;
+ int i, j, len8 = s->len4 >> 1, l_ptwo = 1 << s->ptwo_fft.nbits;
+ const float *in1 = src, *in2 = src + (s->len2 - 1) * stride;
+
+ /* Reindex input, putting it into a buffer and doing an Nx15 FFT */
+ for (i = 0; i < l_ptwo; i++) {
+ for (j = 0; j < 15; j++) {
+ const int k = s->pfa_prereindex[i*15 + j];
+ FFTComplex tmp = { *(in2 - 2*k*stride), *(in1 + 2*k*stride) };
+ CMUL(fft15in[j], tmp, s->twiddle_exptab[k]);
+ }
+ fft15(s->exptab, s->tmp + s->ptwo_fft.revtab[i], fft15in, l_ptwo);
+ }
+
+ /* Then a 15xN FFT (where N is a power of two) */
+ for (i = 0; i < 15; i++)
+ s->ptwo_fft.fft_calc(&s->ptwo_fft, s->tmp + l_ptwo*i);
+
+ /* Reindex again, apply twiddles and output */
+ for (i = 0; i < len8; i++) {
+ float re0, im0, re1, im1;
+ const int i0 = len8 + i, i1 = len8 - i - 1;
+ const int s0 = s->pfa_postreindex[i0], s1 = s->pfa_postreindex[i1];
+
+ CMUL3(re0, im1, s->tmp[s1].im, s->tmp[s1].re, s->twiddle_exptab[i1].im, s->twiddle_exptab[i1].re);
+ CMUL3(re1, im0, s->tmp[s0].im, s->tmp[s0].re, s->twiddle_exptab[i0].im, s->twiddle_exptab[i0].re);
+ z[i1].re = scale * re0;
+ z[i1].im = scale * im0;
+ z[i0].re = scale * re1;
+ z[i0].im = scale * im1;
+ }
+}
new file mode 100644
@@ -0,0 +1,70 @@
+/*
+ * Copyright (c) 2016 Rostislav Pehlivanov <atomnuker@gmail.com>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef AVCODEC_MDCT15_H
+#define AVCODEC_MDCT15_H
+
+#include <stddef.h>
+
+#include "fft.h"
+
+typedef struct MDCT15Context {
+ int fft_n;
+ int len2;
+ int len4;
+ int inverse;
+ int *pfa_prereindex;
+ int *pfa_postreindex;
+
+ FFTContext ptwo_fft;
+
+ FFTComplex *tmp;
+
+ FFTComplex *twiddle_exptab;
+
+ /* 0 - 18: fft15 twiddles, 19 - 20: fft5 twiddles */
+ FFTComplex exptab[21];
+
+ /**
+ * Calculate a full 2N -> N MDCT
+ */
+ void (*mdct)(struct MDCT15Context *s, float *dst, const float *src, ptrdiff_t stride);
+
+ /**
+ * Calculate the middle half of the iMDCT
+ */
+ void (*imdct_half)(struct MDCT15Context *s, float *dst, const float *src,
+ ptrdiff_t src_stride, float scale);
+} MDCT15Context;
+
+/**
+ * Init an (i)MDCT of the length 2 * 15 * (2^N)
+ */
+int ff_mdct15_init(MDCT15Context **ps, int inverse, int N, double scale);
+
+/**
+ * Frees a context
+ */
+void ff_mdct15_uninit(MDCT15Context **ps);
+
+
+void ff_mdct15_init_aarch64(MDCT15Context *s);
+
+#endif /* AVCODEC_MDCT15_H */
@@ -29,7 +29,7 @@
#include "libavutil/float_dsp.h"
#include "libavutil/libm.h"
-#include "imdct15.h"
+#include "mdct15.h"
#include "opus.h"
#include "opustab.h"
@@ -63,7 +63,7 @@ typedef struct CeltFrame {
struct CeltContext {
// constant values that do not change during context lifetime
AVCodecContext *avctx;
- IMDCT15Context *imdct[4];
+ MDCT15Context *imdct[4];
AVFloatDSPContext *dsp;
int output_channels;
@@ -1596,7 +1596,7 @@ int ff_celt_decode_frame(CeltContext *s, OpusRangeCoder *rc,
int silence = 0;
int transient = 0;
int anticollapse = 0;
- IMDCT15Context *imdct;
+ MDCT15Context *imdct;
float imdct_scale = 1.0;
if (coded_channels != 1 && coded_channels != 2) {
@@ -1792,7 +1792,7 @@ void ff_celt_free(CeltContext **ps)
return;
for (i = 0; i < FF_ARRAY_ELEMS(s->imdct); i++)
- ff_imdct15_uninit(&s->imdct[i]);
+ ff_mdct15_uninit(&s->imdct[i]);
av_freep(&s->dsp);
av_freep(ps);
@@ -1817,7 +1817,7 @@ int ff_celt_init(AVCodecContext *avctx, CeltContext **ps, int output_channels)
s->output_channels = output_channels;
for (i = 0; i < FF_ARRAY_ELEMS(s->imdct); i++) {
- ret = ff_imdct15_init(&s->imdct[i], i + 3);
+ ret = ff_mdct15_init(&s->imdct[i], 1, i + 3, -1.0f);
if (ret < 0)
goto fail;
}
The stride parameter does not currently do anything. I can't figure out how to stride since the damn post-rotation happens from the middle and goes both up and down at the same time. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com> --- configure | 6 +- libavcodec/Makefile | 2 +- libavcodec/aac.h | 4 +- libavcodec/aacdec.c | 2 +- libavcodec/aacdec_template.c | 4 +- libavcodec/mdct15.c | 351 +++++++++++++++++++++++++++++++++++++++++++ libavcodec/mdct15.h | 70 +++++++++ libavcodec/opus_celt.c | 10 +- 8 files changed, 435 insertions(+), 14 deletions(-) create mode 100644 libavcodec/mdct15.c create mode 100644 libavcodec/mdct15.h