FFmpeg
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 #if ARCH_X86_32
49  case FF_IDCT_PERM_SIMPLE:
50  for (i = 0; i < 64; i++)
51  idct_permutation[i] = simple_mmx_permutation[i];
52  return 1;
53 #endif
54  case FF_IDCT_PERM_SSE2:
55  for (i = 0; i < 64; i++)
56  idct_permutation[i] = (i & 0x38) | idct_sse2_row_perm[i & 7];
57  return 1;
58  }
59 
60  return 0;
61 }
62 
63 av_cold void ff_idctdsp_init_x86(IDCTDSPContext *c, AVCodecContext *avctx,
64  unsigned high_bit_depth)
65 {
66  int cpu_flags = av_get_cpu_flags();
67 
68 #if ARCH_X86_32
69  if (EXTERNAL_MMX(cpu_flags)) {
70  if (!high_bit_depth &&
71  avctx->lowres == 0 &&
72  (avctx->idct_algo == FF_IDCT_AUTO ||
73  avctx->idct_algo == FF_IDCT_SIMPLEAUTO ||
74  avctx->idct_algo == FF_IDCT_SIMPLEMMX)) {
75  c->idct = ff_simple_idct_mmx;
76  }
77  }
78 #endif
79 
80  if (EXTERNAL_SSE2(cpu_flags)) {
81  c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_sse2;
82  c->put_pixels_clamped = ff_put_pixels_clamped_sse2;
83  c->add_pixels_clamped = ff_add_pixels_clamped_sse2;
84 
85 #if ARCH_X86_32
86  if (!high_bit_depth &&
87  avctx->lowres == 0 &&
88  (avctx->idct_algo == FF_IDCT_AUTO ||
89  avctx->idct_algo == FF_IDCT_SIMPLEAUTO ||
90  avctx->idct_algo == FF_IDCT_SIMPLEMMX)) {
91  c->idct_put = ff_simple_idct_put_sse2;
92  c->idct_add = ff_simple_idct_add_sse2;
93  c->perm_type = FF_IDCT_PERM_SIMPLE;
94  }
95 #endif
96 
97  if (ARCH_X86_64 &&
98  !high_bit_depth &&
99  avctx->lowres == 0 &&
100  (avctx->idct_algo == FF_IDCT_AUTO ||
101  avctx->idct_algo == FF_IDCT_SIMPLEAUTO ||
102  avctx->idct_algo == FF_IDCT_SIMPLEMMX ||
103  avctx->idct_algo == FF_IDCT_SIMPLE)) {
104  c->idct = ff_simple_idct8_sse2;
105  c->idct_put = ff_simple_idct8_put_sse2;
106  c->idct_add = ff_simple_idct8_add_sse2;
107  c->perm_type = FF_IDCT_PERM_TRANSPOSE;
108  }
109  }
110 
111  if (ARCH_X86_64 && avctx->lowres == 0) {
112  if (EXTERNAL_AVX(cpu_flags) &&
113  !high_bit_depth &&
114  (avctx->idct_algo == FF_IDCT_AUTO ||
115  avctx->idct_algo == FF_IDCT_SIMPLEAUTO ||
116  avctx->idct_algo == FF_IDCT_SIMPLEMMX ||
117  avctx->idct_algo == FF_IDCT_SIMPLE)) {
118  c->idct = ff_simple_idct8_avx;
119  c->idct_put = ff_simple_idct8_put_avx;
120  c->idct_add = ff_simple_idct8_add_avx;
121  c->perm_type = FF_IDCT_PERM_TRANSPOSE;
122  }
123 
124  if (avctx->bits_per_raw_sample == 10 &&
125  avctx->codec_id != AV_CODEC_ID_MPEG4 &&
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_avx
void ff_simple_idct12_avx(int16_t *block)
cpu.h
AV_CODEC_ID_MPEG4
@ AV_CODEC_ID_MPEG4
Definition: codec_id.h:64
ff_simple_idct8_add_avx
void ff_simple_idct8_add_avx(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
ff_simple_idct10_avx
void ff_simple_idct10_avx(int16_t *block)
ff_simple_idct_put_sse2
void ff_simple_idct_put_sse2(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
av_get_cpu_flags
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:103
cpu_flags
static atomic_int cpu_flags
Definition: cpu.c:52
FF_IDCT_AUTO
#define FF_IDCT_AUTO
Definition: avcodec.h:1487
ff_simple_idct8_add_sse2
void ff_simple_idct8_add_sse2(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
FF_IDCT_SIMPLE
#define FF_IDCT_SIMPLE
Definition: avcodec.h:1489
ff_idctdsp_init_x86
av_cold void ff_idctdsp_init_x86(IDCTDSPContext *c, AVCodecContext *avctx, unsigned high_bit_depth)
Definition: idctdsp_init.c:63
FF_IDCT_SIMPLEAUTO
#define FF_IDCT_SIMPLEAUTO
Definition: avcodec.h:1503
av_cold
#define av_cold
Definition: attributes.h:90
ff_simple_idct8_put_sse2
void ff_simple_idct8_put_sse2(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
ff_simple_idct8_put_avx
void ff_simple_idct8_put_avx(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
AVCodecContext::bits_per_raw_sample
int bits_per_raw_sample
Bits per sample/pixel of internal libavcodec pixel/sample format.
Definition: avcodec.h:1517
ff_simple_idct8_avx
void ff_simple_idct8_avx(int16_t *block)
AVCodecContext::codec_id
enum AVCodecID codec_id
Definition: avcodec.h:451
ff_simple_idct12_put_avx
void ff_simple_idct12_put_avx(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
NULL
#define NULL
Definition: coverity.c:32
ff_simple_idct12_put_sse2
void ff_simple_idct12_put_sse2(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
ff_add_pixels_clamped_sse2
void ff_add_pixels_clamped_sse2(const int16_t *block, uint8_t *pixels, ptrdiff_t line_size)
ff_put_pixels_clamped_sse2
void ff_put_pixels_clamped_sse2(const int16_t *block, uint8_t *pixels, ptrdiff_t line_size)
c
Undefined Behavior In the C some operations are like signed integer dereferencing freed accessing outside allocated Undefined Behavior must not occur in a C it is not safe even if the output of undefined operations is unused The unsafety may seem nit picking but Optimizing compilers have in fact optimized code on the assumption that no undefined Behavior occurs Optimizing code based on wrong assumptions can and has in some cases lead to effects beyond the output of computations The signed integer overflow problem in speed critical code Code which is highly optimized and works with signed integers sometimes has the problem that often the output of the computation does not c
Definition: undefined.txt:32
AVCodecContext::lowres
int lowres
low resolution decoding, 1-> 1/2 size, 2->1/4 size
Definition: avcodec.h:1524
cpu.h
ff_put_signed_pixels_clamped_sse2
void ff_put_signed_pixels_clamped_sse2(const int16_t *block, uint8_t *pixels, ptrdiff_t line_size)
attributes.h
ff_simple_idct12_sse2
void ff_simple_idct12_sse2(int16_t *block)
EXTERNAL_SSE2
#define EXTERNAL_SSE2(flags)
Definition: cpu.h:59
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
i
#define i(width, name, range_min, range_max)
Definition: cbs_h2645.c:255
AVCodecContext::idct_algo
int idct_algo
IDCT algorithm, see FF_IDCT_* below.
Definition: avcodec.h:1486
idctdsp.h
avcodec.h
ff_simple_idct10_put_avx
void ff_simple_idct10_put_avx(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
EXTERNAL_AVX
#define EXTERNAL_AVX(flags)
Definition: cpu.h:70
FF_IDCT_PERM_TRANSPOSE
@ FF_IDCT_PERM_TRANSPOSE
Definition: idctdsp.h:33
IDCTDSPContext
Definition: idctdsp.h:45
simple_idct.h
FF_IDCT_PERM_SSE2
@ FF_IDCT_PERM_SSE2
Definition: idctdsp.h:35
AVCodecContext
main external API structure.
Definition: avcodec.h:441
ff_simple_idct_mmx
void ff_simple_idct_mmx(int16_t *block)
ff_simple_idct_add_sse2
void ff_simple_idct_add_sse2(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
idct_sse2_row_perm
static const uint8_t idct_sse2_row_perm[8]
Definition: idctdsp_init.c:40
idct_permutation_type
idct_permutation_type
Definition: idctdsp.h:29
ff_simple_idct10_put_sse2
void ff_simple_idct10_put_sse2(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
ff_simple_idct10_sse2
void ff_simple_idct10_sse2(int16_t *block)
FF_IDCT_SIMPLEMMX
#define FF_IDCT_SIMPLEMMX
Definition: avcodec.h:1490
ff_simple_idct8_sse2
void ff_simple_idct8_sse2(int16_t *block)
FF_IDCT_PERM_SIMPLE
@ FF_IDCT_PERM_SIMPLE
Definition: idctdsp.h:32
EXTERNAL_MMX
#define EXTERNAL_MMX(flags)
Definition: cpu.h:56
simple_mmx_permutation
static const uint8_t simple_mmx_permutation[64]
Definition: idctdsp_init.c:29
Generated on Thu Apr 18 2024 22:42:08 for FFmpeg by   doxygen 1.8.17