mirror of
https://git.ffmpeg.org/ffmpeg.git
synced 2025-02-22 14:57:05 +00:00
89da62f2fc
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.
165 lines
6.5 KiB
C
165 lines
6.5 KiB
C
/*
|
|
* 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"
|
|
#include "mem_internal.h"
|
|
#include "avassert.h"
|
|
#include "attributes.h"
|
|
|
|
#ifdef TX_FLOAT
|
|
#define TX_NAME(x) x ## _float
|
|
#define SCALE_TYPE float
|
|
typedef float FFTSample;
|
|
typedef AVComplexFloat FFTComplex;
|
|
#elif defined(TX_DOUBLE)
|
|
#define TX_NAME(x) x ## _double
|
|
#define SCALE_TYPE double
|
|
typedef double FFTSample;
|
|
typedef AVComplexDouble FFTComplex;
|
|
#elif defined(TX_INT32)
|
|
#define TX_NAME(x) x ## _int32
|
|
#define SCALE_TYPE float
|
|
typedef int32_t FFTSample;
|
|
typedef AVComplexInt32 FFTComplex;
|
|
#else
|
|
typedef void FFTComplex;
|
|
#endif
|
|
|
|
#if defined(TX_FLOAT) || defined(TX_DOUBLE)
|
|
|
|
#define CMUL(dre, dim, are, aim, bre, bim) \
|
|
do { \
|
|
(dre) = (are) * (bre) - (aim) * (bim); \
|
|
(dim) = (are) * (bim) + (aim) * (bre); \
|
|
} while (0)
|
|
|
|
#define SMUL(dre, dim, are, aim, bre, bim) \
|
|
do { \
|
|
(dre) = (are) * (bre) - (aim) * (bim); \
|
|
(dim) = (are) * (bim) - (aim) * (bre); \
|
|
} while (0)
|
|
|
|
#define UNSCALE(x) (x)
|
|
#define RESCALE(x) (x)
|
|
|
|
#define FOLD(a, b) ((a) + (b))
|
|
|
|
#elif defined(TX_INT32)
|
|
|
|
/* Properly rounds the result */
|
|
#define CMUL(dre, dim, are, aim, bre, bim) \
|
|
do { \
|
|
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)
|
|
|
|
#define SMUL(dre, dim, are, aim, bre, bim) \
|
|
do { \
|
|
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)
|
|
|
|
#define UNSCALE(x) ((double)x/2147483648.0)
|
|
#define RESCALE(x) (av_clip64(lrintf((x) * 2147483648.0), INT32_MIN, INT32_MAX))
|
|
|
|
#define FOLD(x, y) ((int)((x) + (unsigned)(y) + 32) >> 6)
|
|
|
|
#endif
|
|
|
|
#define BF(x, y, a, b) \
|
|
do { \
|
|
x = (a) - (b); \
|
|
y = (a) + (b); \
|
|
} while (0)
|
|
|
|
#define CMUL3(c, a, b) \
|
|
CMUL((c).re, (c).im, (a).re, (a).im, (b).re, (b).im)
|
|
|
|
#define COSTABLE(size) \
|
|
DECLARE_ALIGNED(32, FFTSample, TX_NAME(ff_cos_##size))[size/4 + 1]
|
|
|
|
/* Used by asm, reorder with care */
|
|
struct AVTXContext {
|
|
int n; /* Non-power-of-two part */
|
|
int m; /* Power-of-two part */
|
|
int inv; /* Is inverse */
|
|
int type; /* Type */
|
|
uint64_t flags; /* Flags */
|
|
double scale; /* Scale */
|
|
|
|
FFTComplex *exptab; /* MDCT exptab */
|
|
FFTComplex *tmp; /* Temporary buffer needed for all compound transforms */
|
|
int *pfatab; /* Input/Output mapping for compound transforms */
|
|
int *revtab; /* Input mapping for power of two transforms */
|
|
int *inplace_idx; /* Required indices to revtab for in-place transforms */
|
|
};
|
|
|
|
/* Checks if type is an MDCT */
|
|
int ff_tx_type_is_mdct(enum AVTXType type);
|
|
|
|
/*
|
|
* Generates the PFA permutation table into AVTXContext->pfatab. The end table
|
|
* is appended to the start table.
|
|
*/
|
|
int ff_tx_gen_compound_mapping(AVTXContext *s);
|
|
|
|
/*
|
|
* Generates a standard-ish (slightly modified) Split-Radix revtab into
|
|
* AVTXContext->revtab
|
|
*/
|
|
int ff_tx_gen_ptwo_revtab(AVTXContext *s, int invert_lookup);
|
|
|
|
/*
|
|
* 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.
|
|
*/
|
|
int ff_tx_gen_ptwo_inplace_revtab_idx(AVTXContext *s);
|
|
|
|
/* Templated init functions */
|
|
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);
|
|
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);
|
|
|
|
typedef struct CosTabsInitOnce {
|
|
void (*func)(void);
|
|
AVOnce control;
|
|
} CosTabsInitOnce;
|
|
|
|
#endif /* AVUTIL_TX_PRIV_H */
|