MrBesen
/
ffmpeg
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity
ffmpeg/libavutil/tx_priv.h

204 lines
8.5 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
/*
* 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 AVUTIL_TX_PRIV_H
#define AVUTIL_TX_PRIV_H
#include "tx.h"
#include "thread.h"
lavu/mem: move the DECLARE_ALIGNED macro family to mem_internal on next+1 bump They are not properly namespaced and not intended for public use.
2020-05-27 14:54:38 +02:00
#include "mem_internal.h"
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
#include "attributes.h"
#ifdef TX_FLOAT
#define TX_NAME(x) x ## _float
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 00:13:28 +01:00
#define SCALE_TYPE float
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
typedef float FFTSample;
typedef AVComplexFloat FFTComplex;
#elif defined(TX_DOUBLE)
#define TX_NAME(x) x ## _double
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 00:13:28 +01:00
#define SCALE_TYPE double
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
typedef double FFTSample;
typedef AVComplexDouble FFTComplex;
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 00:13:28 +01:00
#elif defined(TX_INT32)
#define TX_NAME(x) x ## _int32
#define SCALE_TYPE float
typedef int32_t FFTSample;
typedef AVComplexInt32 FFTComplex;
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
typedef void FFTComplex;
#endif
#if defined(TX_FLOAT) || defined(TX_DOUBLE)
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 00:13:28 +01:00
lavu/tx: minor code style improvements and additional comments
2021-04-10 03:45:03 +02:00
#define CMUL(dre, dim, are, aim, bre, bim) \
do { \
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
(dre) = (are) * (bre) - (aim) * (bim); \
(dim) = (are) * (bim) + (aim) * (bre); \
} while (0)
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 00:13:28 +01:00
lavu/tx: minor code style improvements and additional comments
2021-04-10 03:45:03 +02:00
#define SMUL(dre, dim, are, aim, bre, bim) \
do { \
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 00:13:28 +01:00
(dre) = (are) * (bre) - (aim) * (bim); \
(dim) = (are) * (bim) - (aim) * (bre); \
} while (0)
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 08:11:47 +01:00
#define UNSCALE(x) (x)
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 00:13:28 +01:00
#define RESCALE(x) (x)
#define FOLD(a, b) ((a) + (b))
#elif defined(TX_INT32)
/* Properly rounds the result */
lavu/tx: minor code style improvements and additional comments
2021-04-10 03:45:03 +02:00
#define CMUL(dre, dim, are, aim, bre, bim) \
do { \
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 00:13:28 +01:00
int64_t accu; \
(accu) = (int64_t)(bre) * (are); \
(accu) -= (int64_t)(bim) * (aim); \
(dre) = (int)(((accu) + 0x40000000) >> 31); \
(accu) = (int64_t)(bim) * (are); \
(accu) += (int64_t)(bre) * (aim); \
(dim) = (int)(((accu) + 0x40000000) >> 31); \
} while (0)
lavu/tx: minor code style improvements and additional comments
2021-04-10 03:45:03 +02:00
#define SMUL(dre, dim, are, aim, bre, bim) \
do { \
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 00:13:28 +01:00
int64_t accu; \
(accu) = (int64_t)(bre) * (are); \
(accu) -= (int64_t)(bim) * (aim); \
(dre) = (int)(((accu) + 0x40000000) >> 31); \
(accu) = (int64_t)(bim) * (are); \
(accu) -= (int64_t)(bre) * (aim); \
(dim) = (int)(((accu) + 0x40000000) >> 31); \
} while (0)
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 08:11:47 +01:00
#define UNSCALE(x) ((double)x/2147483648.0)
lavu/tx: clip when converting table values to fixed-point INT32_MAX (2147483647) isn't exactly representable by a floating point value, with the closest being 2147483648.0. So when rescaling a value of 1.0, this could overflow when casting the 64-bit value returned from lrintf() into 32 bits. Unfortunately the properties of integer overflows don't match up well with how a Fourier Transform operates. So clip the value before casting to a 32-bit int. Should be noted we don't have overflows with the table values we're currently using. However, converting a Kaiser-Bessel window function with a length of 256 and a parameter of 5.0 to fixed point did create overflows. So this is more of insurance to save debugging time in case something changes in the future. The macro is only used during init, so it being a little slower is not a problem.
2021-01-09 20:41:25 +01:00
#define RESCALE(x) (av_clip64(lrintf((x) * 2147483648.0), INT32_MIN, INT32_MAX))
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 00:13:28 +01:00
#define FOLD(x, y) ((int)((x) + (unsigned)(y) + 32) >> 6)
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
#endif
lavu/tx: minor code style improvements and additional comments
2021-04-10 03:45:03 +02:00
#define BF(x, y, a, b) \
do { \
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 00:13:28 +01:00
x = (a) - (b); \
y = (a) + (b); \
} while (0)
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 CMUL3(c, a, b) \
CMUL((c).re, (c).im, (a).re, (a).im, (b).re, (b).im)
lavu/tx: minor code style improvements and additional comments
2021-04-10 03:45:03 +02:00
#define COSTABLE(size) \
lavu/tx: refactor power-of-two FFT This commit refactors the power-of-two FFT, making it faster and halving the size of all tables, making the code much smaller on all systems. This removes the big/small pass split, because on modern systems the "big" pass is always faster, and even on older machines there is no measurable speed difference.
2021-04-10 03:47:18 +02:00
DECLARE_ALIGNED(32, FFTSample, TX_NAME(ff_cos_##size))[size/4 + 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
/* Used by asm, reorder with care */
struct AVTXContext {
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 17:58:22 +01:00
int n; /* Non-power-of-two part */
int m; /* Power-of-two part */
int inv; /* Is inverse */
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
int 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 17:58:22 +01:00
uint64_t flags; /* Flags */
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 08:11:47 +01:00
double scale; /* 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
FFTComplex *exptab; /* MDCT exptab */
lavu/tx: minor code style improvements and additional comments
2021-04-10 03:45:03 +02:00
FFTComplex *tmp; /* Temporary buffer needed for all compound transforms */
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
int *pfatab; /* Input/Output mapping for compound transforms */
int *revtab; /* Input mapping for power of two transforms */
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 17:58:22 +01:00
int *inplace_idx; /* Required indices to revtab for in-place transforms */
lavu/tx: add full-sized iMDCT transform flag
2021-04-10 03:55:37 +02:00
checkasm: add av_tx FFT SIMD testing code This sadly required making changes to the code itself, due to the same context needing to be reused for both versions. The lookup table had to be duplicated for both versions.
2021-04-10 03:53:38 +02:00
int *revtab_c; /* Revtab for only the C transforms, needed because
* checkasm makes us reuse the same context. */
lavu/tx: add full-sized iMDCT transform flag
2021-04-10 03:55:37 +02:00
av_tx_fn top_tx; /* Used for computing transforms derived from other
* transforms, like full-length iMDCTs and RDFTs.
* NOTE: Do NOT use this to mix assembly with C code. */
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: minor code style improvements and additional comments
2021-04-10 03:45:03 +02:00
/* Checks if type is an 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 00:13:28 +01:00
int ff_tx_type_is_mdct(enum AVTXType type);
lavu/tx: minor code style improvements and additional comments
2021-04-10 03:45:03 +02:00
/*
* Generates the PFA permutation table into AVTXContext->pfatab. The end table
* is appended to the start table.
*/
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
int ff_tx_gen_compound_mapping(AVTXContext *s);
lavu/tx: minor code style improvements and additional comments
2021-04-10 03:45:03 +02:00
/*
* Generates a standard-ish (slightly modified) Split-Radix revtab into
* AVTXContext->revtab
*/
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 04:11:04 +01:00
int ff_tx_gen_ptwo_revtab(AVTXContext *s, int invert_lookup);
lavu/tx: minor code style improvements and additional comments
2021-04-10 03:45:03 +02:00
/*
* Generates an index into AVTXContext->inplace_idx that if followed in the
* specific order, allows the revtab to be done in-place. AVTXContext->revtab
* must already exist.
*/
checkasm: add av_tx FFT SIMD testing code This sadly required making changes to the code itself, due to the same context needing to be reused for both versions. The lookup table had to be duplicated for both versions.
2021-04-10 03:53:38 +02:00
int ff_tx_gen_ptwo_inplace_revtab_idx(AVTXContext *s, int *revtab);
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: add parity revtab generator version This will be used for SIMD support.
2021-04-10 03:52:31 +02:00
/*
* This generates a parity-based revtab of length len and direction inv.
*
* Parity means even and odd complex numbers will be split, e.g. the even
* coefficients will come first, after which the odd coefficients will be
* placed. For example, a 4-point transform's coefficients after reordering:
* z[0].re, z[0].im, z[2].re, z[2].im, z[1].re, z[1].im, z[3].re, z[3].im
*
* The basis argument is the length of the largest non-composite transform
* supported, and also implies that the basis/2 transform is supported as well,
* as the split-radix algorithm requires it to be.
*
* The dual_stride argument indicates that both the basis, as well as the
* basis/2 transforms support doing two transforms at once, and the coefficients
* will be interleaved between each pair in a split-radix like so (stride == 2):
* tx1[0], tx1[2], tx2[0], tx2[2], tx1[1], tx1[3], tx2[1], tx2[3]
* A non-zero number switches this on, with the value indicating the stride
* (how many values of 1 transform to put first before switching to the other).
* Must be a power of two or 0. Must be less than the basis.
* Value will be clipped to the transform size, so for a basis of 16 and a
* dual_stride of 8, dual 8-point transforms will be laid out as if dual_stride
* was set to 4.
* Usually you'll set this to half the complex numbers that fit in a single
* register or 0. This allows to reuse SSE functions as dual-transform
* functions in AVX mode.
*
* If length is smaller than basis/2 this function will not do anything.
*/
void ff_tx_gen_split_radix_parity_revtab(int *revtab, int len, int inv,
int basis, int dual_stride);
lavu/tx: minor code style improvements and additional comments
2021-04-10 03:45:03 +02:00
/* Templated init functions */
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
int ff_tx_init_mdct_fft_float(AVTXContext *s, av_tx_fn *tx,
enum AVTXType type, int inv, int len,
const void *scale, uint64_t flags);
int ff_tx_init_mdct_fft_double(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 00:13:28 +01:00
int ff_tx_init_mdct_fft_int32(AVTXContext *s, av_tx_fn *tx,
enum AVTXType type, int inv, int len,
const void *scale, uint64_t 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
typedef struct CosTabsInitOnce {
void (*func)(void);
AVOnce control;
} CosTabsInitOnce;
lavu/x86: add FFT assembly This commit adds a pure x86 assembly SIMD version of the FFT in libavutil/tx. The design of this pure assembly FFT is pretty unconventional. On the lowest level, instead of splitting the complex numbers into real and imaginary parts, we keep complex numbers together but split them in terms of parity. This saves a number of shuffles in each transform, but more importantly, it splits each transform into two independent paths, which we process using separate registers in parallel. This allows us to keep all units saturated and lets us use all available registers to avoid dependencies. Moreover, it allows us to double the granularity of our per-load permutation, skipping many expensive lookups and allowing us to use just 4 loads per register, rather than 8, or in case FMA3 (and by extension, AVX2), use the vgatherdpd instruction, which is at least as fast as 4 separate loads on old hardware, and quite a bit faster on modern CPUs). Higher up, we go for a bottom-up construction of large transforms, foregoing the traditional per-transform call-return recursion chains. Instead, we always start at the bottom-most basis transform (in this case, a 32-point transform), and continue constructing larger and larger transforms until we return to the top-most transform. This way, we only touch the stack 3 times per a complete target transform: once for the 1/2 length transform and two times for the 1/4 length transform. The combination algorithm we use is a standard Split-Radix algorithm, as used in our C code. Although a version with less operations exists (Steven G. Johnson and Matteo Frigo's "A modified split-radix FFT with fewer arithmetic operations", IEEE Trans. Signal Process. 55 (1), 111–119 (2007), which is the one FFTW uses), it only has 2% less operations and requires at least 4x the binary code (due to it needing 4 different paths to do a single transform). That version also has other issues which prevent it from being implemented with SIMD code as efficiently, which makes it lose the marginal gains it offered, and cannot be performed bottom-up, requiring many recursive call-return chains, whose overhead adds up. We go through a lot of effort to minimize load/stores by keeping as much in registers in between construcring transforms. This saves us around 32 cycles, on paper, but in reality a lot more due to load/store aliasing (a load from a memory location cannot be issued while there's a store pending, and there are only so many (2 for Zen 3) load/store units in a CPU). Also, we interleave coefficients during the last stage to save on a store+load per register. Each of the smallest, basis transforms (4, 8 and 16-point in our case) has been extremely optimized. Our 8-point transform is barely 20 instructions in total, beating our old implementation 8-point transform by 1 instruction. Our 2x8-point transform is 23 instructions, beating our old implementation by 6 instruction and needing 50% less cycles. Our 16-point transform's combination code takes slightly more instructions than our old implementation, but makes up for it by requiring a lot less arithmetic operations. Overall, the transform was optimized for the timings of Zen 3, which at the time of writing has the most IPC from all documented CPUs. Shuffles were preferred over arithmetic operations due to their 1/0.5 latency/throughput. On average, this code is 30% faster than our old libavcodec implementation. It's able to trade blows with the previously-untouchable FFTW on small transforms, and due to its tiny size and better prediction, outdoes FFTW on larger transforms by 11% on the largest currently supported size.
2021-04-10 03:54:22 +02:00
void ff_tx_init_float_x86(AVTXContext *s, av_tx_fn *tx);
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
#endif /* AVUTIL_TX_PRIV_H */