FFmpeg
ops_tmpl_float.c
Go to the documentation of this file.
1 /**
2  * Copyright (C) 2025 Niklas Haas
3  *
4  * This file is part of FFmpeg.
5  *
6  * FFmpeg is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU Lesser General Public
8  * License as published by the Free Software Foundation; either
9  * version 2.1 of the License, or (at your option) any later version.
10  *
11  * FFmpeg is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14  * Lesser General Public License for more details.
15  *
16  * You should have received a copy of the GNU Lesser General Public
17  * License along with FFmpeg; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19  */
20 
21 #include "libavutil/avassert.h"
22 
23 #include "ops_backend.h"
24 
25 #ifndef BIT_DEPTH
26 # define BIT_DEPTH 32
27 #endif
28 
29 #if BIT_DEPTH == 32
30 # define PIXEL_TYPE SWS_PIXEL_F32
31 # define PIXEL_MAX FLT_MAX
32 # define PIXEL_MIN FLT_MIN
33 # define pixel_t float
34 # define block_t f32block_t
35 # define px f32
36 #else
37 # error Invalid BIT_DEPTH
38 #endif
39 
40 #define IS_FLOAT 1
41 #define FMT_CHAR f
42 #include "ops_tmpl_common.c"
43 
44 DECL_SETUP(setup_dither)
45 {
46  const int size = 1 << op->dither.size_log2;
47  if (size == 1) {
48  /* We special case this value */
49  av_assert1(!av_cmp_q(op->dither.matrix[0], av_make_q(1, 2)));
50  out->ptr = NULL;
51  return 0;
52  }
53 
54  const int width = FFMAX(size, SWS_BLOCK_SIZE);
55  pixel_t *matrix = out->ptr = av_malloc(sizeof(pixel_t) * size * width);
56  if (!matrix)
57  return AVERROR(ENOMEM);
58 
59  static_assert(sizeof(out->ptr) <= sizeof(uint8_t[8]), ">8 byte pointers not supported");
60  uint8_t *offset = &out->u8[8];
61  for (int i = 0; i < 4; i++)
62  offset[i] = op->dither.y_offset[i];
63 
64  for (int y = 0; y < size; y++) {
65  for (int x = 0; x < size; x++)
66  matrix[y * width + x] = av_q2pixel(op->dither.matrix[y * size + x]);
67  for (int x = size; x < width; x++) /* pad to block size */
68  matrix[y * width + x] = matrix[y * width + (x % size)];
69  }
70 
71  return 0;
72 }
73 
74 DECL_FUNC(dither, const int size_log2)
75 {
76  const pixel_t *restrict matrix = impl->priv.ptr;
77  const uint8_t *offset = &impl->priv.u8[8];
78  const int mask = (1 << size_log2) - 1;
79  const int y_line = iter->y;
80  const int row0 = (y_line + offset[0]) & mask;
81  const int row1 = (y_line + offset[1]) & mask;
82  const int row2 = (y_line + offset[2]) & mask;
83  const int row3 = (y_line + offset[3]) & mask;
84  const int size = 1 << size_log2;
85  const int width = FFMAX(size, SWS_BLOCK_SIZE);
86  const int base = iter->x & ~(SWS_BLOCK_SIZE - 1) & (size - 1);
87 
88  SWS_LOOP
89  for (int i = 0; i < SWS_BLOCK_SIZE; i++) {
90  x[i] += size_log2 ? matrix[row0 * width + base + i] : (pixel_t) 0.5;
91  y[i] += size_log2 ? matrix[row1 * width + base + i] : (pixel_t) 0.5;
92  z[i] += size_log2 ? matrix[row2 * width + base + i] : (pixel_t) 0.5;
93  w[i] += size_log2 ? matrix[row3 * width + base + i] : (pixel_t) 0.5;
94  }
95 
96  CONTINUE(block_t, x, y, z, w);
97 }
98 
99 #define WRAP_DITHER(N) \
100 DECL_IMPL(dither##N) \
101 { \
102  CALL(dither, N); \
103 } \
104  \
105 DECL_ENTRY(dither##N, \
106  .op = SWS_OP_DITHER, \
107  .dither_size = N, \
108  .setup = fn(setup_dither), \
109  .free = av_free, \
110 );
111 
112 WRAP_DITHER(0)
113 WRAP_DITHER(1)
114 WRAP_DITHER(2)
115 WRAP_DITHER(3)
116 WRAP_DITHER(4)
117 WRAP_DITHER(5)
118 WRAP_DITHER(6)
119 WRAP_DITHER(7)
120 WRAP_DITHER(8)
121 
122 typedef struct {
123  /* Stored in split form for convenience */
124  pixel_t m[4][4];
125  pixel_t k[4];
126 } fn(LinCoeffs);
127 
128 DECL_SETUP(setup_linear)
129 {
130  fn(LinCoeffs) c;
131 
132  for (int i = 0; i < 4; i++) {
133  for (int j = 0; j < 4; j++)
134  c.m[i][j] = av_q2pixel(op->lin.m[i][j]);
135  c.k[i] = av_q2pixel(op->lin.m[i][4]);
136  }
137 
138  return SETUP_MEMDUP(c);
139 }
140 
141 /**
142  * Fully general case for a 5x5 linear affine transformation. Should never be
143  * called without constant `mask`. This function will compile down to the
144  * appropriately optimized version for the required subset of operations when
145  * called with a constant mask.
146  */
147 DECL_FUNC(linear_mask, const uint32_t mask)
148 {
149  const fn(LinCoeffs) c = *(const fn(LinCoeffs) *) impl->priv.ptr;
150 
151  SWS_LOOP
152  for (int i = 0; i < SWS_BLOCK_SIZE; i++) {
153  const pixel_t xx = x[i];
154  const pixel_t yy = y[i];
155  const pixel_t zz = z[i];
156  const pixel_t ww = w[i];
157 
158  x[i] = (mask & SWS_MASK_OFF(0)) ? c.k[0] : 0;
159  x[i] += (mask & SWS_MASK(0, 0)) ? c.m[0][0] * xx : xx;
160  x[i] += (mask & SWS_MASK(0, 1)) ? c.m[0][1] * yy : 0;
161  x[i] += (mask & SWS_MASK(0, 2)) ? c.m[0][2] * zz : 0;
162  x[i] += (mask & SWS_MASK(0, 3)) ? c.m[0][3] * ww : 0;
163 
164  y[i] = (mask & SWS_MASK_OFF(1)) ? c.k[1] : 0;
165  y[i] += (mask & SWS_MASK(1, 0)) ? c.m[1][0] * xx : 0;
166  y[i] += (mask & SWS_MASK(1, 1)) ? c.m[1][1] * yy : yy;
167  y[i] += (mask & SWS_MASK(1, 2)) ? c.m[1][2] * zz : 0;
168  y[i] += (mask & SWS_MASK(1, 3)) ? c.m[1][3] * ww : 0;
169 
170  z[i] = (mask & SWS_MASK_OFF(2)) ? c.k[2] : 0;
171  z[i] += (mask & SWS_MASK(2, 0)) ? c.m[2][0] * xx : 0;
172  z[i] += (mask & SWS_MASK(2, 1)) ? c.m[2][1] * yy : 0;
173  z[i] += (mask & SWS_MASK(2, 2)) ? c.m[2][2] * zz : zz;
174  z[i] += (mask & SWS_MASK(2, 3)) ? c.m[2][3] * ww : 0;
175 
176  w[i] = (mask & SWS_MASK_OFF(3)) ? c.k[3] : 0;
177  w[i] += (mask & SWS_MASK(3, 0)) ? c.m[3][0] * xx : 0;
178  w[i] += (mask & SWS_MASK(3, 1)) ? c.m[3][1] * yy : 0;
179  w[i] += (mask & SWS_MASK(3, 2)) ? c.m[3][2] * zz : 0;
180  w[i] += (mask & SWS_MASK(3, 3)) ? c.m[3][3] * ww : ww;
181  }
182 
183  CONTINUE(block_t, x, y, z, w);
184 }
185 
186 #define WRAP_LINEAR(NAME, MASK) \
187 DECL_IMPL(linear_##NAME) \
188 { \
189  CALL(linear_mask, MASK); \
190 } \
191  \
192 DECL_ENTRY(linear_##NAME, \
193  .op = SWS_OP_LINEAR, \
194  .setup = fn(setup_linear), \
195  .free = av_free, \
196  .linear_mask = (MASK), \
197 );
198 
199 WRAP_LINEAR(luma, SWS_MASK_LUMA)
200 WRAP_LINEAR(alpha, SWS_MASK_ALPHA)
201 WRAP_LINEAR(lumalpha, SWS_MASK_LUMA | SWS_MASK_ALPHA)
202 WRAP_LINEAR(dot3, 0x7)
203 WRAP_LINEAR(row0, SWS_MASK_ROW(0))
204 WRAP_LINEAR(row0a, SWS_MASK_ROW(0) | SWS_MASK_ALPHA)
205 WRAP_LINEAR(diag3, SWS_MASK_DIAG3)
206 WRAP_LINEAR(diag4, SWS_MASK_DIAG4)
207 WRAP_LINEAR(diagoff3, SWS_MASK_DIAG3 | SWS_MASK_OFF3)
208 WRAP_LINEAR(matrix3, SWS_MASK_MAT3)
209 WRAP_LINEAR(affine3, SWS_MASK_MAT3 | SWS_MASK_OFF3)
210 WRAP_LINEAR(affine3a, SWS_MASK_MAT3 | SWS_MASK_OFF3 | SWS_MASK_ALPHA)
211 WRAP_LINEAR(matrix4, SWS_MASK_MAT4)
212 WRAP_LINEAR(affine4, SWS_MASK_MAT4 | SWS_MASK_OFF4)
213 
214 static const SwsOpTable fn(op_table_float) = {
215  .block_size = SWS_BLOCK_SIZE,
216  .entries = {
217  REF_COMMON_PATTERNS(convert_uint8),
218  REF_COMMON_PATTERNS(convert_uint16),
219  REF_COMMON_PATTERNS(convert_uint32),
220 
221  &fn(op_clear_1110),
222  REF_COMMON_PATTERNS(min),
223  REF_COMMON_PATTERNS(max),
224  REF_COMMON_PATTERNS(scale),
225 
226  &fn(op_dither0),
227  &fn(op_dither1),
228  &fn(op_dither2),
229  &fn(op_dither3),
230  &fn(op_dither4),
231  &fn(op_dither5),
232  &fn(op_dither6),
233  &fn(op_dither7),
234  &fn(op_dither8),
235 
236  &fn(op_linear_luma),
237  &fn(op_linear_alpha),
238  &fn(op_linear_lumalpha),
239  &fn(op_linear_dot3),
240  &fn(op_linear_row0),
241  &fn(op_linear_row0a),
242  &fn(op_linear_diag3),
243  &fn(op_linear_diag4),
244  &fn(op_linear_diagoff3),
245  &fn(op_linear_matrix3),
246  &fn(op_linear_affine3),
247  &fn(op_linear_affine3a),
248  &fn(op_linear_matrix4),
249  &fn(op_linear_affine4),
250 
251  NULL
252  },
253 };
254 
255 #undef PIXEL_TYPE
256 #undef PIXEL_MAX
257 #undef PIXEL_MIN
258 #undef pixel_t
259 #undef block_t
260 #undef px
261 
262 #undef FMT_CHAR
263 #undef IS_FLOAT
SwsOpTable
Definition: ops_chain.h:122
AVERROR
Filter the word “frame” indicates either a video frame or a group of audio as stored in an AVFrame structure Format for each input and each output the list of supported formats For video that means pixel format For audio that means channel sample they are references to shared objects When the negotiation mechanism computes the intersection of the formats supported at each end of a all references to both lists are replaced with a reference to the intersection And when a single format is eventually chosen for a link amongst the remaining all references to the list are updated That means that if a filter requires that its input and output have the same format amongst a supported all it has to do is use a reference to the same list of formats query_formats can leave some formats unset and return AVERROR(EAGAIN) to cause the negotiation mechanism toagain later. That can be used by filters with complex requirements to use the format negotiated on one link to set the formats supported on another. Frame references ownership and permissions
ops_backend.h
out
FILE * out
Definition: movenc.c:55
SWS_MASK_DIAG3
@ SWS_MASK_DIAG3
Definition: ops.h:165
setup_linear
static int setup_linear(const SwsOp *op, SwsOpPriv *out)
Definition: ops.c:236
matrix
Definition: vc1dsp.c:43
block_t
#define block_t
Definition: ops_tmpl_float.c:34
mask
int mask
Definition: mediacodecdec_common.c:154
SWS_MASK_OFF3
@ SWS_MASK_OFF3
Definition: ops.h:166
SWS_MASK_LUMA
@ SWS_MASK_LUMA
Definition: ops.h:162
SWS_MASK_ALPHA
@ SWS_MASK_ALPHA
Definition: ops.h:163
base
uint8_t base
Definition: vp3data.h:128
max
#define max(a, b)
Definition: cuda_runtime.h:33
FFMAX
#define FFMAX(a, b)
Definition: macros.h:47
SWS_MASK_ROW
#define SWS_MASK_ROW(I)
Definition: ops.h:157
SWS_MASK_OFF4
@ SWS_MASK_OFF4
Definition: ops.h:172
av_q2pixel
#define av_q2pixel(q)
Definition: ops_backend.h:70
fn
Definition: ops_tmpl_float.c:122
avassert.h
setup_dither
static int setup_dither(const SwsOp *op, SwsOpPriv *out)
Definition: ops.c:187
DECL_SETUP
DECL_SETUP(setup_dither)
Definition: ops_tmpl_float.c:44
SETUP_MEMDUP
#define SETUP_MEMDUP(c)
Definition: ops_backend.h:129
dither
static const uint16_t dither[8][8]
Definition: vf_gradfun.c:46
CONTINUE
#define CONTINUE(TYPE,...)
Definition: ops_backend.h:120
op
static int op(uint8_t **dst, const uint8_t *dst_end, GetByteContext *gb, int pixel, int count, int *x, int width, int linesize)
Perform decode operation.
Definition: anm.c:76
SWS_LOOP
#define SWS_LOOP
Definition: ops_backend.h:57
SWS_BLOCK_SIZE
#define SWS_BLOCK_SIZE
Copyright (C) 2025 Niklas Haas.
Definition: ops_backend.c:30
SWS_MASK_DIAG4
@ SWS_MASK_DIAG4
Definition: ops.h:171
NULL
#define NULL
Definition: coverity.c:32
SWS_MASK
#define SWS_MASK(I, J)
Definition: ops.h:155
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
DECL_FUNC
DECL_FUNC(dither, const int size_log2)
Definition: ops_tmpl_float.c:74
for
for(k=2;k<=8;++k)
Definition: h264pred_template.c:424
size
int size
Definition: twinvq_data.h:10344
av_make_q
static AVRational av_make_q(int num, int den)
Create an AVRational.
Definition: rational.h:71
pixel_t
#define pixel_t
Definition: ops_tmpl_float.c:33
WRAP_LINEAR
#define WRAP_LINEAR(NAME, MASK)
Definition: ops_tmpl_float.c:186
offset
it s the only field you need to keep assuming you have a context There is some magic you don t need to care about around this just let it vf offset
Definition: writing_filters.txt:86
i
#define i(width, name, range_min, range_max)
Definition: cbs_h2645.c:256
REF_COMMON_PATTERNS
#define REF_COMMON_PATTERNS(NAME)
Definition: ops_backend.h:165
av_assert1
#define av_assert1(cond)
assert() equivalent, that does not lie in speed critical code.
Definition: avassert.h:57
av_cmp_q
static int av_cmp_q(AVRational a, AVRational b)
Compare two rationals.
Definition: rational.h:89
SWS_MASK_MAT4
@ SWS_MASK_MAT4
Definition: ops.h:173
av_malloc
void * av_malloc(size_t size)
Allocate a memory block with alignment suitable for all memory accesses (including vectors if availab...
Definition: mem.c:98
SWS_MASK_MAT3
@ SWS_MASK_MAT3
Definition: ops.h:167
SWS_MASK_OFF
#define SWS_MASK_OFF(I)
Definition: ops.h:156
WRAP_DITHER
#define WRAP_DITHER(N)
Definition: ops_tmpl_float.c:99
ops_tmpl_common.c
w
uint8_t w
Definition: llvidencdsp.c:39
fn
static const SwsOpTable fn(op_table_float)
scale
static void scale(int *out, const int *in, const int w, const int h, const int shift)
Definition: intra.c:273
alpha
static const int16_t alpha[]
Definition: ilbcdata.h:55
width
#define width
Definition: dsp.h:89
min
float min
Definition: vorbis_enc_data.h:429
Generated on Mon Dec 15 2025 19:23:32 for FFmpeg by   doxygen 1.8.17