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) {
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  for (int y = 0; y < size; y++) {
60  for (int x = 0; x < size; x++)
61  matrix[y * width + x] = av_q2pixel(op->dither.matrix[y * size + x]);
62  for (int x = size; x < width; x++) /* pad to block size */
63  matrix[y * width + x] = matrix[y * width + (x % size)];
64  }
65 
66  return 0;
67 }
68 
69 DECL_FUNC(dither, const int size_log2)
70 {
71  const pixel_t *restrict matrix = impl->priv.ptr;
72  const int mask = (1 << size_log2) - 1;
73  const int y_line = iter->y;
74  const int row0 = (y_line + 0) & mask;
75  const int row1 = (y_line + 3) & mask;
76  const int row2 = (y_line + 2) & mask;
77  const int row3 = (y_line + 5) & mask;
78  const int size = 1 << size_log2;
79  const int width = FFMAX(size, SWS_BLOCK_SIZE);
80  const int base = iter->x & ~(SWS_BLOCK_SIZE - 1) & (size - 1);
81 
82  SWS_LOOP
83  for (int i = 0; i < SWS_BLOCK_SIZE; i++) {
84  x[i] += size_log2 ? matrix[row0 * width + base + i] : (pixel_t) 0.5;
85  y[i] += size_log2 ? matrix[row1 * width + base + i] : (pixel_t) 0.5;
86  z[i] += size_log2 ? matrix[row2 * width + base + i] : (pixel_t) 0.5;
87  w[i] += size_log2 ? matrix[row3 * width + base + i] : (pixel_t) 0.5;
88  }
89 
90  CONTINUE(block_t, x, y, z, w);
91 }
92 
93 #define WRAP_DITHER(N) \
94 DECL_IMPL(dither##N) \
95 { \
96  CALL(dither, N); \
97 } \
98  \
99 DECL_ENTRY(dither##N, \
100  .op = SWS_OP_DITHER, \
101  .dither_size = N, \
102  .setup = fn(setup_dither), \
103  .free = av_free, \
104 );
105 
106 WRAP_DITHER(0)
107 WRAP_DITHER(1)
108 WRAP_DITHER(2)
109 WRAP_DITHER(3)
110 WRAP_DITHER(4)
111 WRAP_DITHER(5)
112 WRAP_DITHER(6)
113 WRAP_DITHER(7)
114 WRAP_DITHER(8)
115 
116 typedef struct {
117  /* Stored in split form for convenience */
118  pixel_t m[4][4];
119  pixel_t k[4];
120 } fn(LinCoeffs);
121 
122 DECL_SETUP(setup_linear)
123 {
124  fn(LinCoeffs) c;
125 
126  for (int i = 0; i < 4; i++) {
127  for (int j = 0; j < 4; j++)
128  c.m[i][j] = av_q2pixel(op->lin.m[i][j]);
129  c.k[i] = av_q2pixel(op->lin.m[i][4]);
130  }
131 
132  return SETUP_MEMDUP(c);
133 }
134 
135 /**
136  * Fully general case for a 5x5 linear affine transformation. Should never be
137  * called without constant `mask`. This function will compile down to the
138  * appropriately optimized version for the required subset of operations when
139  * called with a constant mask.
140  */
141 DECL_FUNC(linear_mask, const uint32_t mask)
142 {
143  const fn(LinCoeffs) c = *(const fn(LinCoeffs) *) impl->priv.ptr;
144 
145  SWS_LOOP
146  for (int i = 0; i < SWS_BLOCK_SIZE; i++) {
147  const pixel_t xx = x[i];
148  const pixel_t yy = y[i];
149  const pixel_t zz = z[i];
150  const pixel_t ww = w[i];
151 
152  x[i] = (mask & SWS_MASK_OFF(0)) ? c.k[0] : 0;
153  x[i] += (mask & SWS_MASK(0, 0)) ? c.m[0][0] * xx : xx;
154  x[i] += (mask & SWS_MASK(0, 1)) ? c.m[0][1] * yy : 0;
155  x[i] += (mask & SWS_MASK(0, 2)) ? c.m[0][2] * zz : 0;
156  x[i] += (mask & SWS_MASK(0, 3)) ? c.m[0][3] * ww : 0;
157 
158  y[i] = (mask & SWS_MASK_OFF(1)) ? c.k[1] : 0;
159  y[i] += (mask & SWS_MASK(1, 0)) ? c.m[1][0] * xx : 0;
160  y[i] += (mask & SWS_MASK(1, 1)) ? c.m[1][1] * yy : yy;
161  y[i] += (mask & SWS_MASK(1, 2)) ? c.m[1][2] * zz : 0;
162  y[i] += (mask & SWS_MASK(1, 3)) ? c.m[1][3] * ww : 0;
163 
164  z[i] = (mask & SWS_MASK_OFF(2)) ? c.k[2] : 0;
165  z[i] += (mask & SWS_MASK(2, 0)) ? c.m[2][0] * xx : 0;
166  z[i] += (mask & SWS_MASK(2, 1)) ? c.m[2][1] * yy : 0;
167  z[i] += (mask & SWS_MASK(2, 2)) ? c.m[2][2] * zz : zz;
168  z[i] += (mask & SWS_MASK(2, 3)) ? c.m[2][3] * ww : 0;
169 
170  w[i] = (mask & SWS_MASK_OFF(3)) ? c.k[3] : 0;
171  w[i] += (mask & SWS_MASK(3, 0)) ? c.m[3][0] * xx : 0;
172  w[i] += (mask & SWS_MASK(3, 1)) ? c.m[3][1] * yy : 0;
173  w[i] += (mask & SWS_MASK(3, 2)) ? c.m[3][2] * zz : 0;
174  w[i] += (mask & SWS_MASK(3, 3)) ? c.m[3][3] * ww : ww;
175  }
176 
177  CONTINUE(block_t, x, y, z, w);
178 }
179 
180 #define WRAP_LINEAR(NAME, MASK) \
181 DECL_IMPL(linear_##NAME) \
182 { \
183  CALL(linear_mask, MASK); \
184 } \
185  \
186 DECL_ENTRY(linear_##NAME, \
187  .op = SWS_OP_LINEAR, \
188  .setup = fn(setup_linear), \
189  .free = av_free, \
190  .linear_mask = (MASK), \
191 );
192 
193 WRAP_LINEAR(luma, SWS_MASK_LUMA)
194 WRAP_LINEAR(alpha, SWS_MASK_ALPHA)
195 WRAP_LINEAR(lumalpha, SWS_MASK_LUMA | SWS_MASK_ALPHA)
196 WRAP_LINEAR(dot3, 0x7)
197 WRAP_LINEAR(row0, SWS_MASK_ROW(0))
198 WRAP_LINEAR(row0a, SWS_MASK_ROW(0) | SWS_MASK_ALPHA)
199 WRAP_LINEAR(diag3, SWS_MASK_DIAG3)
200 WRAP_LINEAR(diag4, SWS_MASK_DIAG4)
201 WRAP_LINEAR(diagoff3, SWS_MASK_DIAG3 | SWS_MASK_OFF3)
202 WRAP_LINEAR(matrix3, SWS_MASK_MAT3)
203 WRAP_LINEAR(affine3, SWS_MASK_MAT3 | SWS_MASK_OFF3)
204 WRAP_LINEAR(affine3a, SWS_MASK_MAT3 | SWS_MASK_OFF3 | SWS_MASK_ALPHA)
205 WRAP_LINEAR(matrix4, SWS_MASK_MAT4)
206 WRAP_LINEAR(affine4, SWS_MASK_MAT4 | SWS_MASK_OFF4)
207 
208 static const SwsOpTable fn(op_table_float) = {
209  .block_size = SWS_BLOCK_SIZE,
210  .entries = {
211  REF_COMMON_PATTERNS(convert_uint8),
212  REF_COMMON_PATTERNS(convert_uint16),
213  REF_COMMON_PATTERNS(convert_uint32),
214 
215  &fn(op_clear_1110),
216  REF_COMMON_PATTERNS(min),
217  REF_COMMON_PATTERNS(max),
218  REF_COMMON_PATTERNS(scale),
219 
220  &fn(op_dither0),
221  &fn(op_dither1),
222  &fn(op_dither2),
223  &fn(op_dither3),
224  &fn(op_dither4),
225  &fn(op_dither5),
226  &fn(op_dither6),
227  &fn(op_dither7),
228  &fn(op_dither8),
229 
230  &fn(op_linear_luma),
231  &fn(op_linear_alpha),
232  &fn(op_linear_lumalpha),
233  &fn(op_linear_dot3),
234  &fn(op_linear_row0),
235  &fn(op_linear_row0a),
236  &fn(op_linear_diag3),
237  &fn(op_linear_diag4),
238  &fn(op_linear_diagoff3),
239  &fn(op_linear_matrix3),
240  &fn(op_linear_affine3),
241  &fn(op_linear_affine3a),
242  &fn(op_linear_matrix4),
243  &fn(op_linear_affine4),
244 
245  NULL
246  },
247 };
248 
249 #undef PIXEL_TYPE
250 #undef PIXEL_MAX
251 #undef PIXEL_MIN
252 #undef pixel_t
253 #undef block_t
254 #undef px
255 
256 #undef FMT_CHAR
257 #undef IS_FLOAT
SwsOpTable
Definition: ops_chain.h:118
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_ALPHA
@ SWS_MASK_ALPHA
Definition: ops.h:163
setup_linear
static int setup_linear(const SwsOp *op, SwsOpPriv *out)
Definition: ops.c:213
matrix
Definition: vc1dsp.c:43
block_t
#define block_t
Definition: ops_tmpl_float.c:34
mask
int mask
Definition: mediacodecdec_common.c:154
w
uint8_t w
Definition: llviddspenc.c:38
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
av_malloc
#define av_malloc(s)
Definition: tableprint_vlc.h:31
SWS_MASK_ROW
#define SWS_MASK_ROW(I)
Definition: ops.h:157
SWS_MASK_DIAG4
@ SWS_MASK_DIAG4
Definition: ops.h:171
av_q2pixel
#define av_q2pixel(q)
Definition: ops_backend.h:70
fn
Definition: ops_tmpl_float.c:116
avassert.h
setup_dither
static int setup_dither(const SwsOp *op, SwsOpPriv *out)
Definition: ops.c:189
SWS_MASK_MAT4
@ SWS_MASK_MAT4
Definition: ops.h:173
DECL_SETUP
DECL_SETUP(setup_dither)
Definition: ops_tmpl_float.c:44
SETUP_MEMDUP
#define SETUP_MEMDUP(c)
Definition: ops_backend.h:125
CONTINUE
#define CONTINUE(TYPE,...)
Definition: ops_backend.h:116
SWS_MASK_OFF4
@ SWS_MASK_OFF4
Definition: ops.h:172
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_MASK_MAT3
@ SWS_MASK_MAT3
Definition: ops.h:167
SWS_MASK_DIAG3
@ SWS_MASK_DIAG3
Definition: ops.h:165
SWS_BLOCK_SIZE
#define SWS_BLOCK_SIZE
Copyright (C) 2025 Niklas Haas.
Definition: ops_backend.c:30
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:69
for
for(k=2;k<=8;++k)
Definition: h264pred_template.c:424
SWS_MASK_OFF3
@ SWS_MASK_OFF3
Definition: ops.h:166
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
SWS_MASK_LUMA
@ SWS_MASK_LUMA
Definition: ops.h:162
pixel_t
#define pixel_t
Definition: ops_tmpl_float.c:33
WRAP_LINEAR
#define WRAP_LINEAR(NAME, MASK)
Definition: ops_tmpl_float.c:180
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:161
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_OFF
#define SWS_MASK_OFF(I)
Definition: ops.h:156
WRAP_DITHER
#define WRAP_DITHER(N)
Definition: ops_tmpl_float.c:93
ops_tmpl_common.c
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
dither
static const uint8_t dither[8][8]
Definition: vf_fspp.c:62
Generated on Wed Sep 3 2025 19:26:19 for FFmpeg by   doxygen 1.8.17