FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
idctdsp_init.c
Go to the documentation of this file.
1 /*
2  * This file is part of FFmpeg.
3  *
4  * FFmpeg is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU Lesser General Public
6  * License as published by the Free Software Foundation; either
7  * version 2.1 of the License, or (at your option) any later version.
8  *
9  * FFmpeg is distributed in the hope that it will be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12  * Lesser General Public License for more details.
13  *
14  * You should have received a copy of the GNU Lesser General Public
15  * License along with FFmpeg; if not, write to the Free Software
16  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17  */
18 
19 #include "config.h"
20 #include "libavutil/attributes.h"
21 #include "libavutil/cpu.h"
22 #include "libavutil/x86/cpu.h"
23 #include "libavcodec/avcodec.h"
24 #include "libavcodec/idctdsp.h"
25 #include "idctdsp.h"
26 #include "simple_idct.h"
27 
28 /* Input permutation for the simple_idct_mmx */
29 static const uint8_t simple_mmx_permutation[64] = {
30  0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D,
31  0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D,
32  0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D,
33  0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F,
34  0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F,
35  0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D,
36  0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F,
37  0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F,
38 };
39 
40 static const uint8_t idct_sse2_row_perm[8] = { 0, 4, 1, 5, 2, 6, 3, 7 };
41 
42 av_cold int ff_init_scantable_permutation_x86(uint8_t *idct_permutation,
43  enum idct_permutation_type perm_type)
44 {
45  int i;
46 
47  switch (perm_type) {
48  case FF_IDCT_PERM_SIMPLE:
49  for (i = 0; i < 64; i++)
50  idct_permutation[i] = simple_mmx_permutation[i];
51  return 1;
52  case FF_IDCT_PERM_SSE2:
53  for (i = 0; i < 64; i++)
54  idct_permutation[i] = (i & 0x38) | idct_sse2_row_perm[i & 7];
55  return 1;
56  }
57 
58  return 0;
59 }
60 
61 av_cold void ff_idctdsp_init_x86(IDCTDSPContext *c, AVCodecContext *avctx,
62  unsigned high_bit_depth)
63 {
64  int cpu_flags = av_get_cpu_flags();
65 
66  if (EXTERNAL_MMX(cpu_flags)) {
67  c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_mmx;
68  c->put_pixels_clamped = ff_put_pixels_clamped_mmx;
69  c->add_pixels_clamped = ff_add_pixels_clamped_mmx;
70 
71  if (!high_bit_depth &&
72  avctx->lowres == 0 &&
73  (avctx->idct_algo == FF_IDCT_AUTO ||
74  avctx->idct_algo == FF_IDCT_SIMPLEAUTO ||
75  avctx->idct_algo == FF_IDCT_SIMPLEMMX)) {
76  c->idct_put = ff_simple_idct_put_mmx;
77  c->idct_add = ff_simple_idct_add_mmx;
78  c->idct = ff_simple_idct_mmx;
79  c->perm_type = FF_IDCT_PERM_SIMPLE;
80  }
81  }
82 
83  if (EXTERNAL_SSE2(cpu_flags)) {
84  c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_sse2;
85  c->put_pixels_clamped = ff_put_pixels_clamped_sse2;
86  c->add_pixels_clamped = ff_add_pixels_clamped_sse2;
87 
88  if (!high_bit_depth &&
89  avctx->lowres == 0 &&
90  (avctx->idct_algo == FF_IDCT_AUTO ||
91  avctx->idct_algo == FF_IDCT_SIMPLEAUTO ||
92  avctx->idct_algo == FF_IDCT_SIMPLEMMX)) {
93  c->idct_put = ff_simple_idct_put_sse2;
94  c->idct_add = ff_simple_idct_add_sse2;
95  c->perm_type = FF_IDCT_PERM_SIMPLE;
96  }
97 
98  if (ARCH_X86_64 &&
99  !high_bit_depth &&
100  avctx->lowres == 0 &&
101  (avctx->idct_algo == FF_IDCT_AUTO ||
102  avctx->idct_algo == FF_IDCT_SIMPLEAUTO ||
103  avctx->idct_algo == FF_IDCT_SIMPLEMMX ||
104  avctx->idct_algo == FF_IDCT_SIMPLE)) {
105  c->idct = ff_simple_idct8_sse2;
106  c->idct_put = ff_simple_idct8_put_sse2;
107  c->idct_add = ff_simple_idct8_add_sse2;
108  c->perm_type = FF_IDCT_PERM_TRANSPOSE;
109  }
110  }
111 
112  if (ARCH_X86_64 && avctx->lowres == 0) {
113  if (EXTERNAL_AVX(cpu_flags) &&
114  !high_bit_depth &&
115  (avctx->idct_algo == FF_IDCT_AUTO ||
116  avctx->idct_algo == FF_IDCT_SIMPLEAUTO ||
117  avctx->idct_algo == FF_IDCT_SIMPLEMMX ||
118  avctx->idct_algo == FF_IDCT_SIMPLE)) {
119  c->idct = ff_simple_idct8_avx;
120  c->idct_put = ff_simple_idct8_put_avx;
121  c->idct_add = ff_simple_idct8_add_avx;
122  c->perm_type = FF_IDCT_PERM_TRANSPOSE;
123  }
124 
125  if (avctx->bits_per_raw_sample == 10 &&
126  (avctx->idct_algo == FF_IDCT_AUTO ||
127  avctx->idct_algo == FF_IDCT_SIMPLEAUTO ||
128  avctx->idct_algo == FF_IDCT_SIMPLE)) {
129  if (EXTERNAL_SSE2(cpu_flags)) {
130  c->idct_put = ff_simple_idct10_put_sse2;
131  c->idct_add = NULL;
132  c->idct = ff_simple_idct10_sse2;
133  c->perm_type = FF_IDCT_PERM_TRANSPOSE;
134 
135  }
136  if (EXTERNAL_AVX(cpu_flags)) {
137  c->idct_put = ff_simple_idct10_put_avx;
138  c->idct_add = NULL;
139  c->idct = ff_simple_idct10_avx;
140  c->perm_type = FF_IDCT_PERM_TRANSPOSE;
141  }
142  }
143 
144  if (avctx->bits_per_raw_sample == 12 &&
145  (avctx->idct_algo == FF_IDCT_AUTO ||
146  avctx->idct_algo == FF_IDCT_SIMPLEMMX)) {
147  if (EXTERNAL_SSE2(cpu_flags)) {
148  c->idct_put = ff_simple_idct12_put_sse2;
149  c->idct_add = NULL;
150  c->idct = ff_simple_idct12_sse2;
151  c->perm_type = FF_IDCT_PERM_TRANSPOSE;
152  }
153  if (EXTERNAL_AVX(cpu_flags)) {
154  c->idct_put = ff_simple_idct12_put_avx;
155  c->idct_add = NULL;
156  c->idct = ff_simple_idct12_avx;
157  c->perm_type = FF_IDCT_PERM_TRANSPOSE;
158  }
159  }
160  }
161 }
ff_simple_idct12_put_sse2
void ff_simple_idct12_put_sse2(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
EXTERNAL_MMX
#define EXTERNAL_MMX(flags)
Definition: cpu.h:56
idct_permutation_type
idct_permutation_type
Definition: idctdsp.h:37
NULL
#define NULL
Definition: coverity.c:32
ff_simple_idct12_put_avx
void ff_simple_idct12_put_avx(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
simple_idct.h
FF_IDCT_SIMPLE
#define FF_IDCT_SIMPLE
Definition: avcodec.h:3124
ff_simple_idct_put_sse2
void ff_simple_idct_put_sse2(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
ff_add_pixels_clamped_mmx
void ff_add_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels, ptrdiff_t line_size)
cpu_flags
static atomic_int cpu_flags
Definition: cpu.c:50
AVCodecContext::bits_per_raw_sample
int bits_per_raw_sample
Bits per sample/pixel of internal libavcodec pixel/sample format.
Definition: avcodec.h:3164
ff_simple_idct10_sse2
void ff_simple_idct10_sse2(int16_t *block)
attributes.h
Macro definitions for various function/variable attributes.
ff_simple_idct8_sse2
void ff_simple_idct8_sse2(int16_t *block)
ff_simple_idct10_put_avx
void ff_simple_idct10_put_avx(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
ff_simple_idct10_put_sse2
void ff_simple_idct10_put_sse2(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
uint8_t
uint8_t
Definition: audio_convert.c:194
av_cold
#define av_cold
Definition: attributes.h:82
IDCTDSPContext::add_pixels_clamped
void(* add_pixels_clamped)(const int16_t *block, uint8_t *av_restrict pixels, ptrdiff_t line_size)
Definition: idctdsp.h:61
ff_add_pixels_clamped_sse2
void ff_add_pixels_clamped_sse2(const int16_t *block, uint8_t *pixels, ptrdiff_t line_size)
ff_simple_idct8_put_avx
void ff_simple_idct8_put_avx(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
ff_simple_idct8_add_avx
void ff_simple_idct8_add_avx(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
ff_simple_idct_add_sse2
void ff_simple_idct_add_sse2(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
ff_init_scantable_permutation_x86
av_cold int ff_init_scantable_permutation_x86(uint8_t *idct_permutation, enum idct_permutation_type perm_type)
Definition: idctdsp_init.c:42
AVCodecContext::lowres
int lowres
low resolution decoding, 1-> 1/2 size, 2->1/4 size
Definition: avcodec.h:3172
EXTERNAL_SSE2
#define EXTERNAL_SSE2(flags)
Definition: cpu.h:59
FF_IDCT_AUTO
#define FF_IDCT_AUTO
Definition: avcodec.h:3122
IDCTDSPContext::put_pixels_clamped
void(* put_pixels_clamped)(const int16_t *block, uint8_t *av_restrict pixels, ptrdiff_t line_size)
Definition: idctdsp.h:55
FF_IDCT_SIMPLEAUTO
#define FF_IDCT_SIMPLEAUTO
Definition: avcodec.h:3150
ff_put_signed_pixels_clamped_mmx
void ff_put_signed_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels, ptrdiff_t line_size)
ff_simple_idct12_avx
void ff_simple_idct12_avx(int16_t *block)
ff_simple_idct_add_mmx
void ff_simple_idct_add_mmx(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
AVCodecContext::idct_algo
int idct_algo
IDCT algorithm, see FF_IDCT_* below.
Definition: avcodec.h:3121
ff_put_pixels_clamped_sse2
void ff_put_pixels_clamped_sse2(const int16_t *block, uint8_t *pixels, ptrdiff_t line_size)
idct_sse2_row_perm
static const uint8_t idct_sse2_row_perm[8]
Definition: idctdsp_init.c:40
ff_put_signed_pixels_clamped_sse2
void ff_put_signed_pixels_clamped_sse2(const int16_t *block, uint8_t *pixels, ptrdiff_t line_size)
ff_simple_idct8_add_sse2
void ff_simple_idct8_add_sse2(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
avcodec.h
Libavcodec external API header.
ff_simple_idct_mmx
void ff_simple_idct_mmx(int16_t *block)
ff_simple_idct12_sse2
void ff_simple_idct12_sse2(int16_t *block)
AVCodecContext
main external API structure.
Definition: avcodec.h:1761
IDCTDSPContext::idct_add
void(* idct_add)(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
Definition: idctdsp.h:79
simple_mmx_permutation
static const uint8_t simple_mmx_permutation[64]
Definition: idctdsp_init.c:29
av_get_cpu_flags
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:92
ff_simple_idct8_put_sse2
void ff_simple_idct8_put_sse2(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
ff_idctdsp_init_x86
av_cold void ff_idctdsp_init_x86(IDCTDSPContext *c, AVCodecContext *avctx, unsigned high_bit_depth)
Definition: idctdsp_init.c:61
IDCTDSPContext::put_signed_pixels_clamped
void(* put_signed_pixels_clamped)(const int16_t *block, uint8_t *av_restrict pixels, ptrdiff_t line_size)
Definition: idctdsp.h:58
IDCTDSPContext::idct_put
void(* idct_put)(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
block -> idct -> clip to unsigned 8 bit -> dest.
Definition: idctdsp.h:72
c
static double c[64]
Definition: vsrc_mptestsrc.c:87
ff_simple_idct_put_mmx
void ff_simple_idct_put_mmx(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
FF_IDCT_SIMPLEMMX
#define FF_IDCT_SIMPLEMMX
Definition: avcodec.h:3125
ff_put_pixels_clamped_mmx
void ff_put_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels, ptrdiff_t line_size)
ff_simple_idct10_avx
void ff_simple_idct10_avx(int16_t *block)
EXTERNAL_AVX
#define EXTERNAL_AVX(flags)
Definition: cpu.h:70
IDCTDSPContext::perm_type
enum idct_permutation_type perm_type
Definition: idctdsp.h:97
IDCTDSPContext::idct
void(* idct)(int16_t *block)
Definition: idctdsp.h:65
ff_simple_idct8_avx
void ff_simple_idct8_avx(int16_t *block)
Generated on Fri Jan 12 2018 01:47:53 for FFmpeg by   doxygen 1.8.6