virtualenv/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2024-11-26 19:01:44 +02:00
Code Issues Releases Activity
aa910a7ecd
FFmpeg/libavutil/tx_template.c

758 lines
29 KiB
C
Raw Normal View History

lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
/*
* Copyright (c) 2019 Lynne <dev@lynne.ee>
* Power of two FFT:
* Copyright (c) 2008 Loren Merritt
* Copyright (c) 2002 Fabrice Bellard
* Partly based on libdjbfft by D. J. Bernstein
*
* 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
*/
/* All costabs for a type are defined here */
COSTABLE(16);
COSTABLE(32);
COSTABLE(64);
COSTABLE(128);
COSTABLE(256);
COSTABLE(512);
COSTABLE(1024);
COSTABLE(2048);
COSTABLE(4096);
COSTABLE(8192);
COSTABLE(16384);
COSTABLE(32768);
COSTABLE(65536);
COSTABLE(131072);
DECLARE_ALIGNED(32, FFTComplex, TX_NAME(ff_cos_53))[4];
static FFTSample * const cos_tabs[18] = {
NULL,
NULL,
NULL,
NULL,
TX_NAME(ff_cos_16),
TX_NAME(ff_cos_32),
TX_NAME(ff_cos_64),
TX_NAME(ff_cos_128),
TX_NAME(ff_cos_256),
TX_NAME(ff_cos_512),
TX_NAME(ff_cos_1024),
TX_NAME(ff_cos_2048),
TX_NAME(ff_cos_4096),
TX_NAME(ff_cos_8192),
TX_NAME(ff_cos_16384),
TX_NAME(ff_cos_32768),
TX_NAME(ff_cos_65536),
TX_NAME(ff_cos_131072),
};
static av_always_inline void init_cos_tabs_idx(int index)
{
int m = 1 << index;
double freq = 2*M_PI/m;
FFTSample *tab = cos_tabs[index];
for(int i = 0; i <= m/4; i++)
lavu/tx: implement 32 bit fixed point FFT and MDCT Required minimal changes to the code so made sense to implement. FFT and MDCT tested, the output of both was properly rounded. Fun fact: the non-power-of-two fixed-point FFT and MDCT are the fastest ever non-power-of-two fixed-point FFT and MDCT written. This can replace the power of two integer MDCTs in aac and ac3 if the MIPS optimizations are ported across. Unfortunately the ac3 encoder uses a 16-bit fixed point forward transform, unlike the encoder which uses a 32bit inverse transform, so some modifications might be required there. The 3-point FFT is somewhat less accurate than it otherwise could be, having minor rounding errors with bigger transforms. However, this could be improved later, and the way its currently written is the way one would write assembly for it. Similar rounding errors can also be found throughout the power of two FFTs as well, though those are more difficult to correct. Despite this, the integer transforms are more than accurate enough.
2020-02-09 01:13:28 +02:00
tab[i] = RESCALE(cos(i*freq));
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
for(int i = 1; i < m/4; i++)
tab[m/2 - i] = tab[i];
}
#define INIT_FF_COS_TABS_FUNC(index, size) \
static av_cold void init_cos_tabs_ ## size (void) \
{ \
init_cos_tabs_idx(index); \
}
INIT_FF_COS_TABS_FUNC(4, 16)
INIT_FF_COS_TABS_FUNC(5, 32)
INIT_FF_COS_TABS_FUNC(6, 64)
INIT_FF_COS_TABS_FUNC(7, 128)
INIT_FF_COS_TABS_FUNC(8, 256)
INIT_FF_COS_TABS_FUNC(9, 512)
INIT_FF_COS_TABS_FUNC(10, 1024)
INIT_FF_COS_TABS_FUNC(11, 2048)
INIT_FF_COS_TABS_FUNC(12, 4096)
INIT_FF_COS_TABS_FUNC(13, 8192)
INIT_FF_COS_TABS_FUNC(14, 16384)
INIT_FF_COS_TABS_FUNC(15, 32768)
INIT_FF_COS_TABS_FUNC(16, 65536)
INIT_FF_COS_TABS_FUNC(17, 131072)
static av_cold void ff_init_53_tabs(void)
{
lavu/tx: implement 32 bit fixed point FFT and MDCT Required minimal changes to the code so made sense to implement. FFT and MDCT tested, the output of both was properly rounded. Fun fact: the non-power-of-two fixed-point FFT and MDCT are the fastest ever non-power-of-two fixed-point FFT and MDCT written. This can replace the power of two integer MDCTs in aac and ac3 if the MIPS optimizations are ported across. Unfortunately the ac3 encoder uses a 16-bit fixed point forward transform, unlike the encoder which uses a 32bit inverse transform, so some modifications might be required there. The 3-point FFT is somewhat less accurate than it otherwise could be, having minor rounding errors with bigger transforms. However, this could be improved later, and the way its currently written is the way one would write assembly for it. Similar rounding errors can also be found throughout the power of two FFTs as well, though those are more difficult to correct. Despite this, the integer transforms are more than accurate enough.
2020-02-09 01:13:28 +02:00
TX_NAME(ff_cos_53)[0] = (FFTComplex){ RESCALE(cos(2 * M_PI / 12)), RESCALE(cos(2 * M_PI / 12)) };
TX_NAME(ff_cos_53)[1] = (FFTComplex){ RESCALE(cos(2 * M_PI / 6)), RESCALE(cos(2 * M_PI / 6)) };
TX_NAME(ff_cos_53)[2] = (FFTComplex){ RESCALE(cos(2 * M_PI / 5)), RESCALE(sin(2 * M_PI / 5)) };
TX_NAME(ff_cos_53)[3] = (FFTComplex){ RESCALE(cos(2 * M_PI / 10)), RESCALE(sin(2 * M_PI / 10)) };
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
}
static CosTabsInitOnce cos_tabs_init_once[] = {
{ ff_init_53_tabs, AV_ONCE_INIT },
{ NULL },
{ NULL },
{ NULL },
{ init_cos_tabs_16, AV_ONCE_INIT },
{ init_cos_tabs_32, AV_ONCE_INIT },
{ init_cos_tabs_64, AV_ONCE_INIT },
{ init_cos_tabs_128, AV_ONCE_INIT },
{ init_cos_tabs_256, AV_ONCE_INIT },
{ init_cos_tabs_512, AV_ONCE_INIT },
{ init_cos_tabs_1024, AV_ONCE_INIT },
{ init_cos_tabs_2048, AV_ONCE_INIT },
{ init_cos_tabs_4096, AV_ONCE_INIT },
{ init_cos_tabs_8192, AV_ONCE_INIT },
{ init_cos_tabs_16384, AV_ONCE_INIT },
{ init_cos_tabs_32768, AV_ONCE_INIT },
{ init_cos_tabs_65536, AV_ONCE_INIT },
{ init_cos_tabs_131072, AV_ONCE_INIT },
};
static av_cold void init_cos_tabs(int index)
{
ff_thread_once(&cos_tabs_init_once[index].control,
cos_tabs_init_once[index].func);
}
static av_always_inline void fft3(FFTComplex *out, FFTComplex *in,
ptrdiff_t stride)
{
FFTComplex tmp[2];
lavu/tx: improve 3-point fixed precision There's just no reason not to when its so easy (albeit messy) and its also reducing the precision of all non-power-of-two transforms that use it.
2020-02-14 21:55:00 +02:00
#ifdef TX_INT32
int64_t mtmp[4];
#endif
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
lavu/tx: implement 32 bit fixed point FFT and MDCT Required minimal changes to the code so made sense to implement. FFT and MDCT tested, the output of both was properly rounded. Fun fact: the non-power-of-two fixed-point FFT and MDCT are the fastest ever non-power-of-two fixed-point FFT and MDCT written. This can replace the power of two integer MDCTs in aac and ac3 if the MIPS optimizations are ported across. Unfortunately the ac3 encoder uses a 16-bit fixed point forward transform, unlike the encoder which uses a 32bit inverse transform, so some modifications might be required there. The 3-point FFT is somewhat less accurate than it otherwise could be, having minor rounding errors with bigger transforms. However, this could be improved later, and the way its currently written is the way one would write assembly for it. Similar rounding errors can also be found throughout the power of two FFTs as well, though those are more difficult to correct. Despite this, the integer transforms are more than accurate enough.
2020-02-09 01:13:28 +02:00
BF(tmp[0].re, tmp[1].im, in[1].im, in[2].im);
BF(tmp[0].im, tmp[1].re, in[1].re, in[2].re);
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
out[0*stride].re = in[0].re + tmp[1].re;
out[0*stride].im = in[0].im + tmp[1].im;
lavu/tx: improve 3-point fixed precision There's just no reason not to when its so easy (albeit messy) and its also reducing the precision of all non-power-of-two transforms that use it.
2020-02-14 21:55:00 +02:00
#ifdef TX_INT32
mtmp[0] = (int64_t)TX_NAME(ff_cos_53)[0].re * tmp[0].re;
mtmp[1] = (int64_t)TX_NAME(ff_cos_53)[0].im * tmp[0].im;
mtmp[2] = (int64_t)TX_NAME(ff_cos_53)[1].re * tmp[1].re;
mtmp[3] = (int64_t)TX_NAME(ff_cos_53)[1].re * tmp[1].im;
out[1*stride].re = in[0].re - (mtmp[2] + mtmp[0] + 0x40000000 >> 31);
out[1*stride].im = in[0].im - (mtmp[3] - mtmp[1] + 0x40000000 >> 31);
out[2*stride].re = in[0].re - (mtmp[2] - mtmp[0] + 0x40000000 >> 31);
out[2*stride].im = in[0].im - (mtmp[3] + mtmp[1] + 0x40000000 >> 31);
#else
tmp[0].re = TX_NAME(ff_cos_53)[0].re * tmp[0].re;
tmp[0].im = TX_NAME(ff_cos_53)[0].im * tmp[0].im;
tmp[1].re = TX_NAME(ff_cos_53)[1].re * tmp[1].re;
tmp[1].im = TX_NAME(ff_cos_53)[1].re * tmp[1].im;
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
out[1*stride].re = in[0].re - tmp[1].re + tmp[0].re;
out[1*stride].im = in[0].im - tmp[1].im - tmp[0].im;
out[2*stride].re = in[0].re - tmp[1].re - tmp[0].re;
out[2*stride].im = in[0].im - tmp[1].im + tmp[0].im;
lavu/tx: improve 3-point fixed precision There's just no reason not to when its so easy (albeit messy) and its also reducing the precision of all non-power-of-two transforms that use it.
2020-02-14 21:55:00 +02:00
#endif
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
}
lavu/tx: implement 32 bit fixed point FFT and MDCT Required minimal changes to the code so made sense to implement. FFT and MDCT tested, the output of both was properly rounded. Fun fact: the non-power-of-two fixed-point FFT and MDCT are the fastest ever non-power-of-two fixed-point FFT and MDCT written. This can replace the power of two integer MDCTs in aac and ac3 if the MIPS optimizations are ported across. Unfortunately the ac3 encoder uses a 16-bit fixed point forward transform, unlike the encoder which uses a 32bit inverse transform, so some modifications might be required there. The 3-point FFT is somewhat less accurate than it otherwise could be, having minor rounding errors with bigger transforms. However, this could be improved later, and the way its currently written is the way one would write assembly for it. Similar rounding errors can also be found throughout the power of two FFTs as well, though those are more difficult to correct. Despite this, the integer transforms are more than accurate enough.
2020-02-09 01:13:28 +02:00
#define DECL_FFT5(NAME, D0, D1, D2, D3, D4) \
static av_always_inline void NAME(FFTComplex *out, FFTComplex *in, \
ptrdiff_t stride) \
{ \
FFTComplex z0[4], t[6]; \
\
BF(t[1].im, t[0].re, in[1].re, in[4].re); \
BF(t[1].re, t[0].im, in[1].im, in[4].im); \
BF(t[3].im, t[2].re, in[2].re, in[3].re); \
BF(t[3].re, t[2].im, in[2].im, in[3].im); \
\
lavu/tx: slightly optimize fft15 Saves 2 additions.
2020-02-10 19:37:34 +02:00
out[D0*stride].re = in[0].re + t[0].re + t[2].re; \
out[D0*stride].im = in[0].im + t[0].im + t[2].im; \
lavu/tx: implement 32 bit fixed point FFT and MDCT Required minimal changes to the code so made sense to implement. FFT and MDCT tested, the output of both was properly rounded. Fun fact: the non-power-of-two fixed-point FFT and MDCT are the fastest ever non-power-of-two fixed-point FFT and MDCT written. This can replace the power of two integer MDCTs in aac and ac3 if the MIPS optimizations are ported across. Unfortunately the ac3 encoder uses a 16-bit fixed point forward transform, unlike the encoder which uses a 32bit inverse transform, so some modifications might be required there. The 3-point FFT is somewhat less accurate than it otherwise could be, having minor rounding errors with bigger transforms. However, this could be improved later, and the way its currently written is the way one would write assembly for it. Similar rounding errors can also be found throughout the power of two FFTs as well, though those are more difficult to correct. Despite this, the integer transforms are more than accurate enough.
2020-02-09 01:13:28 +02:00
\
SMUL(t[4].re, t[0].re, TX_NAME(ff_cos_53)[2].re, TX_NAME(ff_cos_53)[3].re, t[2].re, t[0].re); \
SMUL(t[4].im, t[0].im, TX_NAME(ff_cos_53)[2].re, TX_NAME(ff_cos_53)[3].re, t[2].im, t[0].im); \
CMUL(t[5].re, t[1].re, TX_NAME(ff_cos_53)[2].im, TX_NAME(ff_cos_53)[3].im, t[3].re, t[1].re); \
CMUL(t[5].im, t[1].im, TX_NAME(ff_cos_53)[2].im, TX_NAME(ff_cos_53)[3].im, t[3].im, t[1].im); \
\
BF(z0[0].re, z0[3].re, t[0].re, t[1].re); \
BF(z0[0].im, z0[3].im, t[0].im, t[1].im); \
BF(z0[2].re, z0[1].re, t[4].re, t[5].re); \
BF(z0[2].im, z0[1].im, t[4].im, t[5].im); \
\
out[D1*stride].re = in[0].re + z0[3].re; \
out[D1*stride].im = in[0].im + z0[0].im; \
out[D2*stride].re = in[0].re + z0[2].re; \
out[D2*stride].im = in[0].im + z0[1].im; \
out[D3*stride].re = in[0].re + z0[1].re; \
out[D3*stride].im = in[0].im + z0[2].im; \
out[D4*stride].re = in[0].re + z0[0].re; \
out[D4*stride].im = in[0].im + z0[3].im; \
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
}
DECL_FFT5(fft5, 0, 1, 2, 3, 4)
DECL_FFT5(fft5_m1, 0, 6, 12, 3, 9)
DECL_FFT5(fft5_m2, 10, 1, 7, 13, 4)
DECL_FFT5(fft5_m3, 5, 11, 2, 8, 14)
static av_always_inline void fft15(FFTComplex *out, FFTComplex *in,
ptrdiff_t stride)
{
FFTComplex tmp[15];
for (int i = 0; i < 5; i++)
fft3(tmp + i, in + i*3, 5);
fft5_m1(out, tmp + 0, stride);
fft5_m2(out, tmp + 5, stride);
fft5_m3(out, tmp + 10, stride);
}
#define BUTTERFLIES(a0,a1,a2,a3) {\
BF(t3, t5, t5, t1);\
BF(a2.re, a0.re, a0.re, t5);\
BF(a3.im, a1.im, a1.im, t3);\
BF(t4, t6, t2, t6);\
BF(a3.re, a1.re, a1.re, t4);\
BF(a2.im, a0.im, a0.im, t6);\
}
// force loading all the inputs before storing any.
// this is slightly slower for small data, but avoids store->load aliasing
// for addresses separated by large powers of 2.
#define BUTTERFLIES_BIG(a0,a1,a2,a3) {\
FFTSample r0=a0.re, i0=a0.im, r1=a1.re, i1=a1.im;\
BF(t3, t5, t5, t1);\
BF(a2.re, a0.re, r0, t5);\
BF(a3.im, a1.im, i1, t3);\
BF(t4, t6, t2, t6);\
BF(a3.re, a1.re, r1, t4);\
BF(a2.im, a0.im, i0, t6);\
}
#define TRANSFORM(a0,a1,a2,a3,wre,wim) {\
CMUL(t1, t2, a2.re, a2.im, wre, -wim);\
CMUL(t5, t6, a3.re, a3.im, wre, wim);\
BUTTERFLIES(a0,a1,a2,a3)\
}
#define TRANSFORM_ZERO(a0,a1,a2,a3) {\
t1 = a2.re;\
t2 = a2.im;\
t5 = a3.re;\
t6 = a3.im;\
BUTTERFLIES(a0,a1,a2,a3)\
}
/* z[0...8n-1], w[1...2n-1] */
#define PASS(name)\
static void name(FFTComplex *z, const FFTSample *wre, unsigned int n)\
{\
FFTSample t1, t2, t3, t4, t5, t6;\
int o1 = 2*n;\
int o2 = 4*n;\
int o3 = 6*n;\
const FFTSample *wim = wre+o1;\
n--;\
\
TRANSFORM_ZERO(z[0],z[o1],z[o2],z[o3]);\
TRANSFORM(z[1],z[o1+1],z[o2+1],z[o3+1],wre[1],wim[-1]);\
do {\
z += 2;\
wre += 2;\
wim -= 2;\
TRANSFORM(z[0],z[o1],z[o2],z[o3],wre[0],wim[0]);\
TRANSFORM(z[1],z[o1+1],z[o2+1],z[o3+1],wre[1],wim[-1]);\
} while(--n);\
}
PASS(pass)
#undef BUTTERFLIES
#define BUTTERFLIES BUTTERFLIES_BIG
PASS(pass_big)
#define DECL_FFT(n,n2,n4)\
static void fft##n(FFTComplex *z)\
{\
fft##n2(z);\
fft##n4(z+n4*2);\
fft##n4(z+n4*3);\
pass(z,TX_NAME(ff_cos_##n),n4/2);\
}
lavu/tx: add 2-point FFT transform By itself, this allows 6-point, 10-point and 30-point transforms. When the 9-point transform is added it allows for 18-point FFT, and also for a 36-point MDCT (used by MP3).
2020-02-25 08:49:26 +02:00
static void fft2(FFTComplex *z)
{
FFTComplex tmp;
BF(tmp.re, z[0].re, z[0].re, z[1].re);
BF(tmp.im, z[0].im, z[0].im, z[1].im);
z[1] = tmp;
}
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
static void fft4(FFTComplex *z)
{
FFTSample t1, t2, t3, t4, t5, t6, t7, t8;
BF(t3, t1, z[0].re, z[1].re);
BF(t8, t6, z[3].re, z[2].re);
BF(z[2].re, z[0].re, t1, t6);
BF(t4, t2, z[0].im, z[1].im);
BF(t7, t5, z[2].im, z[3].im);
BF(z[3].im, z[1].im, t4, t8);
BF(z[3].re, z[1].re, t3, t7);
BF(z[2].im, z[0].im, t2, t5);
}
static void fft8(FFTComplex *z)
{
FFTSample t1, t2, t3, t4, t5, t6;
fft4(z);
BF(t1, z[5].re, z[4].re, -z[5].re);
BF(t2, z[5].im, z[4].im, -z[5].im);
BF(t5, z[7].re, z[6].re, -z[7].re);
BF(t6, z[7].im, z[6].im, -z[7].im);
BUTTERFLIES(z[0],z[2],z[4],z[6]);
lavu/tx: implement 32 bit fixed point FFT and MDCT Required minimal changes to the code so made sense to implement. FFT and MDCT tested, the output of both was properly rounded. Fun fact: the non-power-of-two fixed-point FFT and MDCT are the fastest ever non-power-of-two fixed-point FFT and MDCT written. This can replace the power of two integer MDCTs in aac and ac3 if the MIPS optimizations are ported across. Unfortunately the ac3 encoder uses a 16-bit fixed point forward transform, unlike the encoder which uses a 32bit inverse transform, so some modifications might be required there. The 3-point FFT is somewhat less accurate than it otherwise could be, having minor rounding errors with bigger transforms. However, this could be improved later, and the way its currently written is the way one would write assembly for it. Similar rounding errors can also be found throughout the power of two FFTs as well, though those are more difficult to correct. Despite this, the integer transforms are more than accurate enough.
2020-02-09 01:13:28 +02:00
TRANSFORM(z[1],z[3],z[5],z[7],RESCALE(M_SQRT1_2),RESCALE(M_SQRT1_2));
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
}
static void fft16(FFTComplex *z)
{
FFTSample t1, t2, t3, t4, t5, t6;
FFTSample cos_16_1 = TX_NAME(ff_cos_16)[1];
FFTSample cos_16_3 = TX_NAME(ff_cos_16)[3];
fft8(z);
fft4(z+8);
fft4(z+12);
TRANSFORM_ZERO(z[0],z[4],z[8],z[12]);
lavu/tx: implement 32 bit fixed point FFT and MDCT Required minimal changes to the code so made sense to implement. FFT and MDCT tested, the output of both was properly rounded. Fun fact: the non-power-of-two fixed-point FFT and MDCT are the fastest ever non-power-of-two fixed-point FFT and MDCT written. This can replace the power of two integer MDCTs in aac and ac3 if the MIPS optimizations are ported across. Unfortunately the ac3 encoder uses a 16-bit fixed point forward transform, unlike the encoder which uses a 32bit inverse transform, so some modifications might be required there. The 3-point FFT is somewhat less accurate than it otherwise could be, having minor rounding errors with bigger transforms. However, this could be improved later, and the way its currently written is the way one would write assembly for it. Similar rounding errors can also be found throughout the power of two FFTs as well, though those are more difficult to correct. Despite this, the integer transforms are more than accurate enough.
2020-02-09 01:13:28 +02:00
TRANSFORM(z[2],z[6],z[10],z[14],RESCALE(M_SQRT1_2),RESCALE(M_SQRT1_2));
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
TRANSFORM(z[1],z[5],z[9],z[13],cos_16_1,cos_16_3);
TRANSFORM(z[3],z[7],z[11],z[15],cos_16_3,cos_16_1);
}
DECL_FFT(32,16,8)
DECL_FFT(64,32,16)
DECL_FFT(128,64,32)
DECL_FFT(256,128,64)
DECL_FFT(512,256,128)
#define pass pass_big
DECL_FFT(1024,512,256)
DECL_FFT(2048,1024,512)
DECL_FFT(4096,2048,1024)
DECL_FFT(8192,4096,2048)
DECL_FFT(16384,8192,4096)
DECL_FFT(32768,16384,8192)
DECL_FFT(65536,32768,16384)
DECL_FFT(131072,65536,32768)
static void (* const fft_dispatch[])(FFTComplex*) = {
lavu/tx: add 2-point FFT transform By itself, this allows 6-point, 10-point and 30-point transforms. When the 9-point transform is added it allows for 18-point FFT, and also for a 36-point MDCT (used by MP3).
2020-02-25 08:49:26 +02:00
NULL, fft2, fft4, fft8, fft16, fft32, fft64, fft128, fft256, fft512,
fft1024, fft2048, fft4096, fft8192, fft16384, fft32768, fft65536, fft131072
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
};
#define DECL_COMP_FFT(N) \
static void compound_fft_##N##xM(AVTXContext *s, void *_out, \
void *_in, ptrdiff_t stride) \
{ \
const int m = s->m, *in_map = s->pfatab, *out_map = in_map + N*m; \
FFTComplex *in = _in; \
FFTComplex *out = _out; \
FFTComplex fft##N##in[N]; \
lavu/tx: add 2-point FFT transform By itself, this allows 6-point, 10-point and 30-point transforms. When the 9-point transform is added it allows for 18-point FFT, and also for a 36-point MDCT (used by MP3).
2020-02-25 08:49:26 +02:00
void (*fftp)(FFTComplex *z) = fft_dispatch[av_log2(m)]; \
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
\
for (int i = 0; i < m; i++) { \
for (int j = 0; j < N; j++) \
fft##N##in[j] = in[in_map[i*N + j]]; \
fft##N(s->tmp + s->revtab[i], fft##N##in, m); \
} \
\
for (int i = 0; i < N; i++) \
fftp(s->tmp + m*i); \
\
for (int i = 0; i < N*m; i++) \
out[i] = s->tmp[out_map[i]]; \
}
DECL_COMP_FFT(3)
DECL_COMP_FFT(5)
DECL_COMP_FFT(15)
static void monolithic_fft(AVTXContext *s, void *_out, void *_in,
ptrdiff_t stride)
{
FFTComplex *in = _in;
FFTComplex *out = _out;
lavu/tx: add 2-point FFT transform By itself, this allows 6-point, 10-point and 30-point transforms. When the 9-point transform is added it allows for 18-point FFT, and also for a 36-point MDCT (used by MP3).
2020-02-25 08:49:26 +02:00
int m = s->m, mb = av_log2(m);
lavu/tx: support in-place FFT transforms This commit adds support for in-place FFT transforms. Since our internal transforms were all in-place anyway, this only changes the permutation on the input. Unfortunately, research papers were of no help here. All focused on dry hardware implementations, where permutes are free, or on software implementations where binary bloat is of no concern so storing dozen times the transforms for each permutation and version is not considered bad practice. Still, for a pure C implementation, it's only around 28% slower than the multi-megabyte FFTW3 in unaligned mode. Unlike a closed permutation like with PFA, split-radix FFT bit-reversals contain multiple NOPs, multiple simple swaps, and a few chained swaps, so regular single-loop single-state permute loops were not possible. Instead, we filter out parts of the input indices which are redundant. This allows for a single branch, and with some clever AVX512 asm, could possibly be SIMD'd without refactoring. The inplace_idx array is guaranteed to never be larger than the revtab array, and in practice only requires around log2(len) entries. The power-of-two MDCTs can be done in-place as well. And it's possible to eliminate a copy in the compound MDCTs too, however it'll be slower than doing them out of place, and we'd need to dirty the input array.
2021-02-10 18:58:22 +02:00
if (s->flags & AV_TX_INPLACE) {
FFTComplex tmp;
int src, dst, *inplace_idx = s->inplace_idx;
src = *inplace_idx++;
do {
tmp = out[src];
dst = s->revtab[src];
do {
FFSWAP(FFTComplex, tmp, out[dst]);
dst = s->revtab[dst];
} while (dst != src); /* Can be > as well, but is less predictable */
out[dst] = tmp;
} while ((src = *inplace_idx++));
} else {
for (int i = 0; i < m; i++)
lavu/tx: invert permutation lookups out[lut[i]] = in[i] lookups were 4.04 times(!) slower than out[i] = in[lut[i]] lookups for an out-of-place FFT of length 4096. The permutes remain unchanged for anything but out-of-place monolithic FFT, as those benefit quite a lot from the current order (it means there's only 1 lookup necessary to add to an offset, rather than a full gather). The code was based around non-power-of-two FFTs, so this wasn't benchmarked early on.
2021-02-27 05:11:04 +02:00
out[i] = in[s->revtab[i]];
lavu/tx: support in-place FFT transforms This commit adds support for in-place FFT transforms. Since our internal transforms were all in-place anyway, this only changes the permutation on the input. Unfortunately, research papers were of no help here. All focused on dry hardware implementations, where permutes are free, or on software implementations where binary bloat is of no concern so storing dozen times the transforms for each permutation and version is not considered bad practice. Still, for a pure C implementation, it's only around 28% slower than the multi-megabyte FFTW3 in unaligned mode. Unlike a closed permutation like with PFA, split-radix FFT bit-reversals contain multiple NOPs, multiple simple swaps, and a few chained swaps, so regular single-loop single-state permute loops were not possible. Instead, we filter out parts of the input indices which are redundant. This allows for a single branch, and with some clever AVX512 asm, could possibly be SIMD'd without refactoring. The inplace_idx array is guaranteed to never be larger than the revtab array, and in practice only requires around log2(len) entries. The power-of-two MDCTs can be done in-place as well. And it's possible to eliminate a copy in the compound MDCTs too, however it'll be slower than doing them out of place, and we'd need to dirty the input array.
2021-02-10 18:58:22 +02:00
}
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
fft_dispatch[mb](out);
}
lavu: support arbitrary-point FFTs and all even (i)MDCT transforms This patch adds support for arbitrary-point FFTs and all even MDCT transforms. Odd MDCTs are not supported yet as they're based on the DCT-II and DCT-III and they're very niche. With this we can now write tests.
2021-01-12 09:11:47 +02:00
static void naive_fft(AVTXContext *s, void *_out, void *_in,
ptrdiff_t stride)
{
FFTComplex *in = _in;
FFTComplex *out = _out;
const int n = s->n;
double phase = s->inv ? 2.0*M_PI/n : -2.0*M_PI/n;
for(int i = 0; i < n; i++) {
FFTComplex tmp = { 0 };
for(int j = 0; j < n; j++) {
const double factor = phase*i*j;
const FFTComplex mult = {
RESCALE(cos(factor)),
RESCALE(sin(factor)),
};
FFTComplex res;
CMUL3(res, in[j], mult);
tmp.re += res.re;
tmp.im += res.im;
}
out[i] = tmp;
}
}
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
#define DECL_COMP_IMDCT(N) \
static void compound_imdct_##N##xM(AVTXContext *s, void *_dst, void *_src, \
ptrdiff_t stride) \
{ \
FFTComplex fft##N##in[N]; \
FFTComplex *z = _dst, *exp = s->exptab; \
const int m = s->m, len8 = N*m >> 1; \
const int *in_map = s->pfatab, *out_map = in_map + N*m; \
const FFTSample *src = _src, *in1, *in2; \
lavu/tx: add 2-point FFT transform By itself, this allows 6-point, 10-point and 30-point transforms. When the 9-point transform is added it allows for 18-point FFT, and also for a 36-point MDCT (used by MP3).
2020-02-25 08:49:26 +02:00
void (*fftp)(FFTComplex *) = fft_dispatch[av_log2(m)]; \
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
\
stride /= sizeof(*src); /* To convert it from bytes */ \
in1 = src; \
in2 = src + ((N*m*2) - 1) * stride; \
\
for (int i = 0; i < m; i++) { \
for (int j = 0; j < N; j++) { \
const int k = in_map[i*N + j]; \
FFTComplex tmp = { in2[-k*stride], in1[k*stride] }; \
CMUL3(fft##N##in[j], tmp, exp[k >> 1]); \
} \
fft##N(s->tmp + s->revtab[i], fft##N##in, m); \
} \
\
for (int i = 0; i < N; i++) \
fftp(s->tmp + m*i); \
\
for (int i = 0; i < len8; i++) { \
const int i0 = len8 + i, i1 = len8 - i - 1; \
const int s0 = out_map[i0], s1 = out_map[i1]; \
FFTComplex src1 = { s->tmp[s1].im, s->tmp[s1].re }; \
FFTComplex src0 = { s->tmp[s0].im, s->tmp[s0].re }; \
\
CMUL(z[i1].re, z[i0].im, src1.re, src1.im, exp[i1].im, exp[i1].re); \
CMUL(z[i0].re, z[i1].im, src0.re, src0.im, exp[i0].im, exp[i0].re); \
} \
}
DECL_COMP_IMDCT(3)
DECL_COMP_IMDCT(5)
DECL_COMP_IMDCT(15)
#define DECL_COMP_MDCT(N) \
static void compound_mdct_##N##xM(AVTXContext *s, void *_dst, void *_src, \
ptrdiff_t stride) \
{ \
FFTSample *src = _src, *dst = _dst; \
FFTComplex *exp = s->exptab, tmp, fft##N##in[N]; \
const int m = s->m, len4 = N*m, len3 = len4 * 3, len8 = len4 >> 1; \
const int *in_map = s->pfatab, *out_map = in_map + N*m; \
lavu/tx: add 2-point FFT transform By itself, this allows 6-point, 10-point and 30-point transforms. When the 9-point transform is added it allows for 18-point FFT, and also for a 36-point MDCT (used by MP3).
2020-02-25 08:49:26 +02:00
void (*fftp)(FFTComplex *) = fft_dispatch[av_log2(m)]; \
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
\
stride /= sizeof(*dst); \
\
for (int i = 0; i < m; i++) { /* Folding and pre-reindexing */ \
for (int j = 0; j < N; j++) { \
const int k = in_map[i*N + j]; \
if (k < len4) { \
lavu/tx: implement 32 bit fixed point FFT and MDCT Required minimal changes to the code so made sense to implement. FFT and MDCT tested, the output of both was properly rounded. Fun fact: the non-power-of-two fixed-point FFT and MDCT are the fastest ever non-power-of-two fixed-point FFT and MDCT written. This can replace the power of two integer MDCTs in aac and ac3 if the MIPS optimizations are ported across. Unfortunately the ac3 encoder uses a 16-bit fixed point forward transform, unlike the encoder which uses a 32bit inverse transform, so some modifications might be required there. The 3-point FFT is somewhat less accurate than it otherwise could be, having minor rounding errors with bigger transforms. However, this could be improved later, and the way its currently written is the way one would write assembly for it. Similar rounding errors can also be found throughout the power of two FFTs as well, though those are more difficult to correct. Despite this, the integer transforms are more than accurate enough.
2020-02-09 01:13:28 +02:00
tmp.re = FOLD(-src[ len4 + k], src[1*len4 - 1 - k]); \
tmp.im = FOLD(-src[ len3 + k], -src[1*len3 - 1 - k]); \
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
} else { \
lavu/tx: implement 32 bit fixed point FFT and MDCT Required minimal changes to the code so made sense to implement. FFT and MDCT tested, the output of both was properly rounded. Fun fact: the non-power-of-two fixed-point FFT and MDCT are the fastest ever non-power-of-two fixed-point FFT and MDCT written. This can replace the power of two integer MDCTs in aac and ac3 if the MIPS optimizations are ported across. Unfortunately the ac3 encoder uses a 16-bit fixed point forward transform, unlike the encoder which uses a 32bit inverse transform, so some modifications might be required there. The 3-point FFT is somewhat less accurate than it otherwise could be, having minor rounding errors with bigger transforms. However, this could be improved later, and the way its currently written is the way one would write assembly for it. Similar rounding errors can also be found throughout the power of two FFTs as well, though those are more difficult to correct. Despite this, the integer transforms are more than accurate enough.
2020-02-09 01:13:28 +02:00
tmp.re = FOLD(-src[ len4 + k], -src[5*len4 - 1 - k]); \
tmp.im = FOLD( src[-len4 + k], -src[1*len3 - 1 - k]); \
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
} \
CMUL(fft##N##in[j].im, fft##N##in[j].re, tmp.re, tmp.im, \
exp[k >> 1].re, exp[k >> 1].im); \
} \
fft##N(s->tmp + s->revtab[i], fft##N##in, m); \
} \
\
for (int i = 0; i < N; i++) \
fftp(s->tmp + m*i); \
\
for (int i = 0; i < len8; i++) { \
const int i0 = len8 + i, i1 = len8 - i - 1; \
const int s0 = out_map[i0], s1 = out_map[i1]; \
FFTComplex src1 = { s->tmp[s1].re, s->tmp[s1].im }; \
FFTComplex src0 = { s->tmp[s0].re, s->tmp[s0].im }; \
\
CMUL(dst[2*i1*stride + stride], dst[2*i0*stride], src0.re, src0.im, \
exp[i0].im, exp[i0].re); \
CMUL(dst[2*i0*stride + stride], dst[2*i1*stride], src1.re, src1.im, \
exp[i1].im, exp[i1].re); \
} \
}
DECL_COMP_MDCT(3)
DECL_COMP_MDCT(5)
DECL_COMP_MDCT(15)
static void monolithic_imdct(AVTXContext *s, void *_dst, void *_src,
ptrdiff_t stride)
{
FFTComplex *z = _dst, *exp = s->exptab;
const int m = s->m, len8 = m >> 1;
const FFTSample *src = _src, *in1, *in2;
lavu/tx: add 2-point FFT transform By itself, this allows 6-point, 10-point and 30-point transforms. When the 9-point transform is added it allows for 18-point FFT, and also for a 36-point MDCT (used by MP3).
2020-02-25 08:49:26 +02:00
void (*fftp)(FFTComplex *) = fft_dispatch[av_log2(m)];
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
stride /= sizeof(*src);
in1 = src;
in2 = src + ((m*2) - 1) * stride;
for (int i = 0; i < m; i++) {
FFTComplex tmp = { in2[-2*i*stride], in1[2*i*stride] };
CMUL3(z[s->revtab[i]], tmp, exp[i]);
}
fftp(z);
for (int i = 0; i < len8; i++) {
const int i0 = len8 + i, i1 = len8 - i - 1;
FFTComplex src1 = { z[i1].im, z[i1].re };
FFTComplex src0 = { z[i0].im, z[i0].re };
CMUL(z[i1].re, z[i0].im, src1.re, src1.im, exp[i1].im, exp[i1].re);
CMUL(z[i0].re, z[i1].im, src0.re, src0.im, exp[i0].im, exp[i0].re);
}
}
static void monolithic_mdct(AVTXContext *s, void *_dst, void *_src,
ptrdiff_t stride)
{
FFTSample *src = _src, *dst = _dst;
FFTComplex *exp = s->exptab, tmp, *z = _dst;
const int m = s->m, len4 = m, len3 = len4 * 3, len8 = len4 >> 1;
lavu/tx: add 2-point FFT transform By itself, this allows 6-point, 10-point and 30-point transforms. When the 9-point transform is added it allows for 18-point FFT, and also for a 36-point MDCT (used by MP3).
2020-02-25 08:49:26 +02:00
void (*fftp)(FFTComplex *) = fft_dispatch[av_log2(m)];
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
stride /= sizeof(*dst);
for (int i = 0; i < m; i++) { /* Folding and pre-reindexing */
const int k = 2*i;
if (k < len4) {
lavu/tx: implement 32 bit fixed point FFT and MDCT Required minimal changes to the code so made sense to implement. FFT and MDCT tested, the output of both was properly rounded. Fun fact: the non-power-of-two fixed-point FFT and MDCT are the fastest ever non-power-of-two fixed-point FFT and MDCT written. This can replace the power of two integer MDCTs in aac and ac3 if the MIPS optimizations are ported across. Unfortunately the ac3 encoder uses a 16-bit fixed point forward transform, unlike the encoder which uses a 32bit inverse transform, so some modifications might be required there. The 3-point FFT is somewhat less accurate than it otherwise could be, having minor rounding errors with bigger transforms. However, this could be improved later, and the way its currently written is the way one would write assembly for it. Similar rounding errors can also be found throughout the power of two FFTs as well, though those are more difficult to correct. Despite this, the integer transforms are more than accurate enough.
2020-02-09 01:13:28 +02:00
tmp.re = FOLD(-src[ len4 + k], src[1*len4 - 1 - k]);
tmp.im = FOLD(-src[ len3 + k], -src[1*len3 - 1 - k]);
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
} else {
lavu/tx: implement 32 bit fixed point FFT and MDCT Required minimal changes to the code so made sense to implement. FFT and MDCT tested, the output of both was properly rounded. Fun fact: the non-power-of-two fixed-point FFT and MDCT are the fastest ever non-power-of-two fixed-point FFT and MDCT written. This can replace the power of two integer MDCTs in aac and ac3 if the MIPS optimizations are ported across. Unfortunately the ac3 encoder uses a 16-bit fixed point forward transform, unlike the encoder which uses a 32bit inverse transform, so some modifications might be required there. The 3-point FFT is somewhat less accurate than it otherwise could be, having minor rounding errors with bigger transforms. However, this could be improved later, and the way its currently written is the way one would write assembly for it. Similar rounding errors can also be found throughout the power of two FFTs as well, though those are more difficult to correct. Despite this, the integer transforms are more than accurate enough.
2020-02-09 01:13:28 +02:00
tmp.re = FOLD(-src[ len4 + k], -src[5*len4 - 1 - k]);
tmp.im = FOLD( src[-len4 + k], -src[1*len3 - 1 - k]);
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
}
CMUL(z[s->revtab[i]].im, z[s->revtab[i]].re, tmp.re, tmp.im,
exp[i].re, exp[i].im);
}
fftp(z);
for (int i = 0; i < len8; i++) {
const int i0 = len8 + i, i1 = len8 - i - 1;
FFTComplex src1 = { z[i1].re, z[i1].im };
FFTComplex src0 = { z[i0].re, z[i0].im };
CMUL(dst[2*i1*stride + stride], dst[2*i0*stride], src0.re, src0.im,
exp[i0].im, exp[i0].re);
CMUL(dst[2*i0*stride + stride], dst[2*i1*stride], src1.re, src1.im,
exp[i1].im, exp[i1].re);
}
}
lavu: support arbitrary-point FFTs and all even (i)MDCT transforms This patch adds support for arbitrary-point FFTs and all even MDCT transforms. Odd MDCTs are not supported yet as they're based on the DCT-II and DCT-III and they're very niche. With this we can now write tests.
2021-01-12 09:11:47 +02:00
static void naive_imdct(AVTXContext *s, void *_dst, void *_src,
ptrdiff_t stride)
{
int len = s->n;
int len2 = len*2;
FFTSample *src = _src;
FFTSample *dst = _dst;
double scale = s->scale;
const double phase = M_PI/(4.0*len2);
stride /= sizeof(*src);
for (int i = 0; i < len; i++) {
double sum_d = 0.0;
double sum_u = 0.0;
double i_d = phase * (4*len - 2*i - 1);
double i_u = phase * (3*len2 + 2*i + 1);
for (int j = 0; j < len2; j++) {
double a = (2 * j + 1);
double a_d = cos(a * i_d);
double a_u = cos(a * i_u);
double val = UNSCALE(src[j*stride]);
sum_d += a_d * val;
sum_u += a_u * val;
}
dst[i + 0] = RESCALE( sum_d*scale);
dst[i + len] = RESCALE(-sum_u*scale);
}
}
static void naive_mdct(AVTXContext *s, void *_dst, void *_src,
ptrdiff_t stride)
{
int len = s->n*2;
FFTSample *src = _src;
FFTSample *dst = _dst;
double scale = s->scale;
const double phase = M_PI/(4.0*len);
stride /= sizeof(*dst);
for (int i = 0; i < len; i++) {
double sum = 0.0;
for (int j = 0; j < len*2; j++) {
int a = (2*j + 1 + len) * (2*i + 1);
sum += UNSCALE(src[j]) * cos(a * phase);
}
dst[i*stride] = RESCALE(sum*scale);
}
}
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
static int gen_mdct_exptab(AVTXContext *s, int len4, double scale)
{
const double theta = (scale < 0 ? len4 : 0) + 1.0/8.0;
if (!(s->exptab = av_malloc_array(len4, sizeof(*s->exptab))))
return AVERROR(ENOMEM);
scale = sqrt(fabs(scale));
for (int i = 0; i < len4; i++) {
const double alpha = M_PI_2 * (i + theta) / len4;
lavu/tx: implement 32 bit fixed point FFT and MDCT Required minimal changes to the code so made sense to implement. FFT and MDCT tested, the output of both was properly rounded. Fun fact: the non-power-of-two fixed-point FFT and MDCT are the fastest ever non-power-of-two fixed-point FFT and MDCT written. This can replace the power of two integer MDCTs in aac and ac3 if the MIPS optimizations are ported across. Unfortunately the ac3 encoder uses a 16-bit fixed point forward transform, unlike the encoder which uses a 32bit inverse transform, so some modifications might be required there. The 3-point FFT is somewhat less accurate than it otherwise could be, having minor rounding errors with bigger transforms. However, this could be improved later, and the way its currently written is the way one would write assembly for it. Similar rounding errors can also be found throughout the power of two FFTs as well, though those are more difficult to correct. Despite this, the integer transforms are more than accurate enough.
2020-02-09 01:13:28 +02:00
s->exptab[i].re = RESCALE(cos(alpha) * scale);
s->exptab[i].im = RESCALE(sin(alpha) * scale);
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
}
return 0;
}
int TX_NAME(ff_tx_init_mdct_fft)(AVTXContext *s, av_tx_fn *tx,
enum AVTXType type, int inv, int len,
const void *scale, uint64_t flags)
{
lavu/tx: implement 32 bit fixed point FFT and MDCT Required minimal changes to the code so made sense to implement. FFT and MDCT tested, the output of both was properly rounded. Fun fact: the non-power-of-two fixed-point FFT and MDCT are the fastest ever non-power-of-two fixed-point FFT and MDCT written. This can replace the power of two integer MDCTs in aac and ac3 if the MIPS optimizations are ported across. Unfortunately the ac3 encoder uses a 16-bit fixed point forward transform, unlike the encoder which uses a 32bit inverse transform, so some modifications might be required there. The 3-point FFT is somewhat less accurate than it otherwise could be, having minor rounding errors with bigger transforms. However, this could be improved later, and the way its currently written is the way one would write assembly for it. Similar rounding errors can also be found throughout the power of two FFTs as well, though those are more difficult to correct. Despite this, the integer transforms are more than accurate enough.
2020-02-09 01:13:28 +02:00
const int is_mdct = ff_tx_type_is_mdct(type);
lavu: support arbitrary-point FFTs and all even (i)MDCT transforms This patch adds support for arbitrary-point FFTs and all even MDCT transforms. Odd MDCTs are not supported yet as they're based on the DCT-II and DCT-III and they're very niche. With this we can now write tests.
2021-01-12 09:11:47 +02:00
int err, l, n = 1, m = 1, max_ptwo = 1 << (FF_ARRAY_ELEMS(fft_dispatch) - 1);
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
if (is_mdct)
len >>= 1;
lavu: support arbitrary-point FFTs and all even (i)MDCT transforms This patch adds support for arbitrary-point FFTs and all even MDCT transforms. Odd MDCTs are not supported yet as they're based on the DCT-II and DCT-III and they're very niche. With this we can now write tests.
2021-01-12 09:11:47 +02:00
l = len;
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
#define CHECK_FACTOR(DST, FACTOR, SRC) \
if (DST == 1 && !(SRC % FACTOR)) { \
DST = FACTOR; \
SRC /= FACTOR; \
}
CHECK_FACTOR(n, 15, len)
CHECK_FACTOR(n, 5, len)
CHECK_FACTOR(n, 3, len)
lavu/tx: undef the correct macro It was renamed and no warning was given for undeffing a nonexisting one.
2020-02-10 19:36:45 +02:00
#undef CHECK_FACTOR
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
/* len must be a power of two now */
lavu/tx: add 2-point FFT transform By itself, this allows 6-point, 10-point and 30-point transforms. When the 9-point transform is added it allows for 18-point FFT, and also for a 36-point MDCT (used by MP3).
2020-02-25 08:49:26 +02:00
if (!(len & (len - 1)) && len >= 2 && len <= max_ptwo) {
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
m = len;
len = 1;
}
s->n = n;
s->m = m;
s->inv = inv;
s->type = type;
lavu/tx: support in-place FFT transforms This commit adds support for in-place FFT transforms. Since our internal transforms were all in-place anyway, this only changes the permutation on the input. Unfortunately, research papers were of no help here. All focused on dry hardware implementations, where permutes are free, or on software implementations where binary bloat is of no concern so storing dozen times the transforms for each permutation and version is not considered bad practice. Still, for a pure C implementation, it's only around 28% slower than the multi-megabyte FFTW3 in unaligned mode. Unlike a closed permutation like with PFA, split-radix FFT bit-reversals contain multiple NOPs, multiple simple swaps, and a few chained swaps, so regular single-loop single-state permute loops were not possible. Instead, we filter out parts of the input indices which are redundant. This allows for a single branch, and with some clever AVX512 asm, could possibly be SIMD'd without refactoring. The inplace_idx array is guaranteed to never be larger than the revtab array, and in practice only requires around log2(len) entries. The power-of-two MDCTs can be done in-place as well. And it's possible to eliminate a copy in the compound MDCTs too, however it'll be slower than doing them out of place, and we'd need to dirty the input array.
2021-02-10 18:58:22 +02:00
s->flags = flags;
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
lavu: support arbitrary-point FFTs and all even (i)MDCT transforms This patch adds support for arbitrary-point FFTs and all even MDCT transforms. Odd MDCTs are not supported yet as they're based on the DCT-II and DCT-III and they're very niche. With this we can now write tests.
2021-01-12 09:11:47 +02:00
/* If we weren't able to split the length into factors we can handle,
* resort to using the naive and slow FT. This also filters out
* direct 3, 5 and 15 transforms as they're too niche. */
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
if (len > 1 || m == 1) {
lavu: support arbitrary-point FFTs and all even (i)MDCT transforms This patch adds support for arbitrary-point FFTs and all even MDCT transforms. Odd MDCTs are not supported yet as they're based on the DCT-II and DCT-III and they're very niche. With this we can now write tests.
2021-01-12 09:11:47 +02:00
if (is_mdct && (l & 1)) /* Odd (i)MDCTs are not supported yet */
avutil/tx: use ENOSYS instead of ENOTSUP It's the standard error code used across the codebase to signal unimplemented or unsupported features. Signed-off-by: James Almer <jamrial@gmail.com>
2021-01-14 04:02:19 +02:00
return AVERROR(ENOSYS);
lavu/tx: support in-place FFT transforms This commit adds support for in-place FFT transforms. Since our internal transforms were all in-place anyway, this only changes the permutation on the input. Unfortunately, research papers were of no help here. All focused on dry hardware implementations, where permutes are free, or on software implementations where binary bloat is of no concern so storing dozen times the transforms for each permutation and version is not considered bad practice. Still, for a pure C implementation, it's only around 28% slower than the multi-megabyte FFTW3 in unaligned mode. Unlike a closed permutation like with PFA, split-radix FFT bit-reversals contain multiple NOPs, multiple simple swaps, and a few chained swaps, so regular single-loop single-state permute loops were not possible. Instead, we filter out parts of the input indices which are redundant. This allows for a single branch, and with some clever AVX512 asm, could possibly be SIMD'd without refactoring. The inplace_idx array is guaranteed to never be larger than the revtab array, and in practice only requires around log2(len) entries. The power-of-two MDCTs can be done in-place as well. And it's possible to eliminate a copy in the compound MDCTs too, however it'll be slower than doing them out of place, and we'd need to dirty the input array.
2021-02-10 18:58:22 +02:00
if (flags & AV_TX_INPLACE) /* Neither are in-place naive transforms */
return AVERROR(ENOSYS);
lavu: support arbitrary-point FFTs and all even (i)MDCT transforms This patch adds support for arbitrary-point FFTs and all even MDCT transforms. Odd MDCTs are not supported yet as they're based on the DCT-II and DCT-III and they're very niche. With this we can now write tests.
2021-01-12 09:11:47 +02:00
s->n = l;
s->m = 1;
*tx = naive_fft;
if (is_mdct) {
s->scale = *((SCALE_TYPE *)scale);
*tx = inv ? naive_imdct : naive_mdct;
}
return 0;
}
if (n > 1 && m > 1) { /* 2D transform case */
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
if ((err = ff_tx_gen_compound_mapping(s)))
return err;
if (!(s->tmp = av_malloc(n*m*sizeof(*s->tmp))))
return AVERROR(ENOMEM);
*tx = n == 3 ? compound_fft_3xM :
n == 5 ? compound_fft_5xM :
compound_fft_15xM;
if (is_mdct)
*tx = n == 3 ? inv ? compound_imdct_3xM : compound_mdct_3xM :
n == 5 ? inv ? compound_imdct_5xM : compound_mdct_5xM :
inv ? compound_imdct_15xM : compound_mdct_15xM;
} else { /* Direct transform case */
*tx = monolithic_fft;
if (is_mdct)
*tx = inv ? monolithic_imdct : monolithic_mdct;
}
if (n != 1)
init_cos_tabs(0);
if (m != 1) {
lavu/tx: do not invert permutes on MDCTs
2021-02-27 05:59:21 +02:00
if ((err = ff_tx_gen_ptwo_revtab(s, n == 1 && !is_mdct && !(flags & AV_TX_INPLACE))))
lavu/tx: support in-place FFT transforms This commit adds support for in-place FFT transforms. Since our internal transforms were all in-place anyway, this only changes the permutation on the input. Unfortunately, research papers were of no help here. All focused on dry hardware implementations, where permutes are free, or on software implementations where binary bloat is of no concern so storing dozen times the transforms for each permutation and version is not considered bad practice. Still, for a pure C implementation, it's only around 28% slower than the multi-megabyte FFTW3 in unaligned mode. Unlike a closed permutation like with PFA, split-radix FFT bit-reversals contain multiple NOPs, multiple simple swaps, and a few chained swaps, so regular single-loop single-state permute loops were not possible. Instead, we filter out parts of the input indices which are redundant. This allows for a single branch, and with some clever AVX512 asm, could possibly be SIMD'd without refactoring. The inplace_idx array is guaranteed to never be larger than the revtab array, and in practice only requires around log2(len) entries. The power-of-two MDCTs can be done in-place as well. And it's possible to eliminate a copy in the compound MDCTs too, however it'll be slower than doing them out of place, and we'd need to dirty the input array.
2021-02-10 18:58:22 +02:00
return err;
if (flags & AV_TX_INPLACE) {
if (is_mdct) /* In-place MDCTs are not supported yet */
return AVERROR(ENOSYS);
if ((err = ff_tx_gen_ptwo_inplace_revtab_idx(s)))
return err;
}
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
for (int i = 4; i <= av_log2(m); i++)
init_cos_tabs(i);
}
if (is_mdct)
lavu/tx: implement 32 bit fixed point FFT and MDCT Required minimal changes to the code so made sense to implement. FFT and MDCT tested, the output of both was properly rounded. Fun fact: the non-power-of-two fixed-point FFT and MDCT are the fastest ever non-power-of-two fixed-point FFT and MDCT written. This can replace the power of two integer MDCTs in aac and ac3 if the MIPS optimizations are ported across. Unfortunately the ac3 encoder uses a 16-bit fixed point forward transform, unlike the encoder which uses a 32bit inverse transform, so some modifications might be required there. The 3-point FFT is somewhat less accurate than it otherwise could be, having minor rounding errors with bigger transforms. However, this could be improved later, and the way its currently written is the way one would write assembly for it. Similar rounding errors can also be found throughout the power of two FFTs as well, though those are more difficult to correct. Despite this, the integer transforms are more than accurate enough.
2020-02-09 01:13:28 +02:00
return gen_mdct_exptab(s, n*m, *((SCALE_TYPE *)scale));
lavu/tx: add support for double precision FFT and MDCT Simply moves and templates the actual transforms to support an additional data type. Unlike the float version, which is equal or better than libfftw3f, double precision output is bit identical with libfftw3.
2019-07-27 19:54:20 +02:00
return 0;
}
Reference in New Issue Copy Permalink