FFmpeg coverage

Directory: ../../../ffmpeg/
File: src/libswscale/ops_tmpl_float.c
Date: 2026-04-24 19:58:39
Exec Total Coverage
Lines: 91 95 95.8%
Functions: 26 29 89.7%
Branches: 77 88 87.5%
Line Branch Exec Source
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_t float
33 # define inter_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 1150 DECL_SETUP(setup_dither, params, out)
45 {
46 1150 const SwsOp *op = params->op;
47 1150 const int size = 1 << op->dither.size_log2;
48
2/2
✓ Branch 0 taken 56 times.
✓ Branch 1 taken 1094 times.
 1150 if (size == 1) {
49 /* We special case this value */
50 av_assert1(!av_cmp_q(op->dither.matrix[0], av_make_q(1, 2)));
51 56 out->priv.ptr = NULL;
52 56 return 0;
53 }
54
55 1094 const int width = FFMAX(size, SWS_BLOCK_SIZE);
56 1094 pixel_t *matrix = out->priv.ptr = av_malloc(sizeof(pixel_t) * size * width);
57
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1094 times.
 1094 if (!matrix)
58 ✗ return AVERROR(ENOMEM);
59 1094 out->free = ff_op_priv_free;
60
61 static_assert(sizeof(out->priv.ptr) <= sizeof(uint8_t[8]),
62 ">8 byte pointers not supported");
63
64 1094 int8_t *offset = &out->priv.i8[8];
65
2/2
✓ Branch 0 taken 4376 times.
✓ Branch 1 taken 1094 times.
 5470 for (int i = 0; i < 4; i++)
66 4376 offset[i] = op->dither.y_offset[i];
67
68
2/2
✓ Branch 0 taken 38896 times.
✓ Branch 1 taken 1094 times.
 39990 for (int y = 0; y < size; y++) {
69
2/2
✓ Branch 0 taken 5058656 times.
✓ Branch 1 taken 38896 times.
 5097552 for (int x = 0; x < size; x++)
70
1/2
✓ Branch 0 taken 5058656 times.
✗ Branch 1 not taken.
 5058656 matrix[y * width + x] = av_q2pixel(op->dither.matrix[y * size + x]);
71
2/2
✓ Branch 0 taken 200096 times.
✓ Branch 1 taken 38896 times.
 238992 for (int x = size; x < width; x++) /* pad to block size */
72 200096 matrix[y * width + x] = matrix[y * width + (x % size)];
73 }
74
75 1094 return 0;
76 }
77
78 191808 DECL_FUNC(dither, const int size_log2)
79 {
80 191808 const pixel_t *restrict matrix = impl->priv.ptr;
81 191808 const int8_t *restrict offset = &impl->priv.i8[8];
82 191808 const int mask = (1 << size_log2) - 1;
83 191808 const int y_line = iter->y;
84 191808 const int size = 1 << size_log2;
85 191808 const int width = FFMAX(size, SWS_BLOCK_SIZE);
86 191808 const int base = iter->x & ~(SWS_BLOCK_SIZE - 1) & (size - 1);
87
88 #define DITHER_COMP(VAR, IDX) \
89 if (offset[IDX] >= 0) { \
90 const int row = (y_line + offset[IDX]) & mask; \
91 SWS_LOOP \
92 for (int i = 0; i < SWS_BLOCK_SIZE; i++) \
93 VAR[i] += size_log2 ? matrix[row * width + base + i] : (pixel_t) 0.5; \
94 }
95
96
6/6
✓ Branch 0 taken 166752 times.
✓ Branch 1 taken 25056 times.
✓ Branch 2 taken 5315584 times.
✓ Branch 3 taken 20480 times.
✓ Branch 4 taken 5336064 times.
✓ Branch 5 taken 166752 times.
 5527872 DITHER_COMP(x, 0)
97
6/6
✓ Branch 0 taken 161280 times.
✓ Branch 1 taken 30528 times.
✓ Branch 2 taken 5140480 times.
✓ Branch 3 taken 20480 times.
✓ Branch 4 taken 5160960 times.
✓ Branch 5 taken 161280 times.
 5352768 DITHER_COMP(y, 1)
98
6/6
✓ Branch 0 taken 157824 times.
✓ Branch 1 taken 33984 times.
✓ Branch 2 taken 5029888 times.
✓ Branch 3 taken 20480 times.
✓ Branch 4 taken 5050368 times.
✓ Branch 5 taken 157824 times.
 5242176 DITHER_COMP(z, 2)
99
6/6
✓ Branch 0 taken 64224 times.
✓ Branch 1 taken 127584 times.
✓ Branch 2 taken 2034688 times.
✓ Branch 3 taken 20480 times.
✓ Branch 4 taken 2055168 times.
✓ Branch 5 taken 64224 times.
 2246976 DITHER_COMP(w, 3)
100
101 191808 CONTINUE(x, y, z, w);
102 191808 }
103
104 #define WRAP_DITHER(N) \
105 DECL_IMPL(dither, dither##N, N) \
106 \
107 DECL_ENTRY(dither##N, SWS_COMP_ALL, \
108 .op = SWS_OP_DITHER, \
109 .dither_size = N, \
110 .setup = fn(setup_dither), \
111 );
112
113 640 WRAP_DITHER(0)
114 640 WRAP_DITHER(1)
115 640 WRAP_DITHER(2)
116 640 WRAP_DITHER(3)
117 186688 WRAP_DITHER(4)
118 640 WRAP_DITHER(5)
119 640 WRAP_DITHER(6)
120 640 WRAP_DITHER(7)
121 640 WRAP_DITHER(8)
122
123 typedef struct {
124 /* Stored in split form for convenience */
125 pixel_t m[4][4];
126 pixel_t k[4];
127 } fn(LinCoeffs);
128
129 772 DECL_SETUP(setup_linear, params, out)
130 {
131 772 const SwsOp *op = params->op;
132 fn(LinCoeffs) c;
133
134
2/2
✓ Branch 0 taken 3088 times.
✓ Branch 1 taken 772 times.
 3860 for (int i = 0; i < 4; i++) {
135
2/2
✓ Branch 0 taken 12352 times.
✓ Branch 1 taken 3088 times.
 15440 for (int j = 0; j < 4; j++)
136
1/2
✓ Branch 0 taken 12352 times.
✗ Branch 1 not taken.
 12352 c.m[i][j] = av_q2pixel(op->lin.m[i][j]);
137
1/2
✓ Branch 0 taken 3088 times.
✗ Branch 1 not taken.
 3088 c.k[i] = av_q2pixel(op->lin.m[i][4]);
138 }
139
140 772 return SETUP_MEMDUP(c, out);
141 }
142
143 /**
144 * Fully general case for a 5x5 linear affine transformation. Should never be
145 * called without constant `mask`. This function will compile down to the
146 * appropriately optimized version for the required subset of operations when
147 * called with a constant mask.
148 */
149 179744 DECL_FUNC(linear_mask, const uint32_t mask)
150 {
151 179744 const fn(LinCoeffs) c = *(const fn(LinCoeffs) *) impl->priv.ptr;
152
153 SWS_LOOP
154
2/2
✓ Branch 0 taken 5751808 times.
✓ Branch 1 taken 179744 times.
 5931552 for (int i = 0; i < SWS_BLOCK_SIZE; i++) {
155 5751808 const pixel_t xx = x[i];
156 5751808 const pixel_t yy = y[i];
157 5751808 const pixel_t zz = z[i];
158 5751808 const pixel_t ww = w[i];
159
160
2/2
✓ Branch 0 taken 4435968 times.
✓ Branch 1 taken 1315840 times.
 5751808 x[i] = (mask & SWS_MASK_OFF(0)) ? c.k[0] : 0;
161
2/2
✓ Branch 0 taken 5672960 times.
✓ Branch 1 taken 78848 times.
 5751808 x[i] += (mask & SWS_MASK(0, 0)) ? c.m[0][0] * xx : xx;
162
2/2
✓ Branch 0 taken 2195456 times.
✓ Branch 1 taken 3556352 times.
 5751808 x[i] += (mask & SWS_MASK(0, 1)) ? c.m[0][1] * yy : 0;
163
2/2
✓ Branch 0 taken 4038656 times.
✓ Branch 1 taken 1713152 times.
 5751808 x[i] += (mask & SWS_MASK(0, 2)) ? c.m[0][2] * zz : 0;
164
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 5751808 times.
 5751808 x[i] += (mask & SWS_MASK(0, 3)) ? c.m[0][3] * ww : 0;
165
166
2/2
✓ Branch 0 taken 3757056 times.
✓ Branch 1 taken 1994752 times.
 5751808 y[i] = (mask & SWS_MASK_OFF(1)) ? c.k[1] : 0;
167
2/2
✓ Branch 0 taken 3696640 times.
✓ Branch 1 taken 2055168 times.
 5751808 y[i] += (mask & SWS_MASK(1, 0)) ? c.m[1][0] * xx : 0;
168
2/2
✓ Branch 0 taken 4657152 times.
✓ Branch 1 taken 1094656 times.
 5751808 y[i] += (mask & SWS_MASK(1, 1)) ? c.m[1][1] * yy : yy;
169
2/2
✓ Branch 0 taken 3696640 times.
✓ Branch 1 taken 2055168 times.
 5751808 y[i] += (mask & SWS_MASK(1, 2)) ? c.m[1][2] * zz : 0;
170
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 5751808 times.
 5751808 y[i] += (mask & SWS_MASK(1, 3)) ? c.m[1][3] * ww : 0;
171
172
2/2
✓ Branch 0 taken 3757056 times.
✓ Branch 1 taken 1994752 times.
 5751808 z[i] = (mask & SWS_MASK_OFF(2)) ? c.k[2] : 0;
173
2/2
✓ Branch 0 taken 3696640 times.
✓ Branch 1 taken 2055168 times.
 5751808 z[i] += (mask & SWS_MASK(2, 0)) ? c.m[2][0] * xx : 0;
174
2/2
✓ Branch 0 taken 3696640 times.
✓ Branch 1 taken 2055168 times.
 5751808 z[i] += (mask & SWS_MASK(2, 1)) ? c.m[2][1] * yy : 0;
175
2/2
✓ Branch 0 taken 4657152 times.
✓ Branch 1 taken 1094656 times.
 5751808 z[i] += (mask & SWS_MASK(2, 2)) ? c.m[2][2] * zz : zz;
176
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 5751808 times.
 5751808 z[i] += (mask & SWS_MASK(2, 3)) ? c.m[2][3] * ww : 0;
177
178
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 5751808 times.
 5751808 w[i] = (mask & SWS_MASK_OFF(3)) ? c.k[3] : 0;
179
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 5751808 times.
 5751808 w[i] += (mask & SWS_MASK(3, 0)) ? c.m[3][0] * xx : 0;
180
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 5751808 times.
 5751808 w[i] += (mask & SWS_MASK(3, 1)) ? c.m[3][1] * yy : 0;
181
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 5751808 times.
 5751808 w[i] += (mask & SWS_MASK(3, 2)) ? c.m[3][2] * zz : 0;
182
2/2
✓ Branch 0 taken 1278976 times.
✓ Branch 1 taken 4472832 times.
 5751808 w[i] += (mask & SWS_MASK(3, 3)) ? c.m[3][3] * ww : ww;
183 }
184
185 179744 CONTINUE(x, y, z, w);
186 179744 }
187
188 #define WRAP_LINEAR(NAME, MASK) \
189 DECL_IMPL(linear_mask, linear_##NAME, MASK) \
190 \
191 DECL_ENTRY(linear_##NAME, SWS_COMP_ALL, \
192 .op = SWS_OP_LINEAR, \
193 .setup = fn(setup_linear), \
194 .linear_mask = (MASK), \
195 );
196
197 19168 WRAP_LINEAR(luma, SWS_MASK_LUMA)
198 2464 WRAP_LINEAR(alpha, SWS_MASK_ALPHA)
199 1888 WRAP_LINEAR(lumalpha, SWS_MASK_LUMA | SWS_MASK_ALPHA)
200 ✗ WRAP_LINEAR(yalpha, SWS_MASK(1, 1)) /* ya alpha */
201 9952 WRAP_LINEAR(dot3, 0x7)
202 ✗ WRAP_LINEAR(dot3a, 0x7 | SWS_MASK_ALPHA)
203 736 WRAP_LINEAR(row0, SWS_MASK_ROW(0) ^ SWS_MASK(0, 3)) /* row0 sans alpha */
204 10528 WRAP_LINEAR(diag3, SWS_MASK_DIAG3)
205 17600 WRAP_LINEAR(diag4, SWS_MASK_DIAG4)
206 1888 WRAP_LINEAR(diagoff3, SWS_MASK_DIAG3 | SWS_MASK_OFF3)
207 47392 WRAP_LINEAR(affine3, SWS_MASK_MAT3 | SWS_MASK_OFF3)
208 576 WRAP_LINEAR(affine3uv, SWS_MASK_MAT3 | SWS_MASK_OFF(1) | SWS_MASK_OFF(2))
209 49536 WRAP_LINEAR(affine3x, SWS_MASK_MAT3 ^ SWS_MASK(0, 1) | SWS_MASK_OFF3)
210 8064 WRAP_LINEAR(affine3xa, SWS_MASK_MAT3 ^ SWS_MASK(0, 1) | SWS_MASK_OFF3 | SWS_MASK_ALPHA)
211 ✗ WRAP_LINEAR(affine3xy, SWS_MASK_MAT3 ^ SWS_MASK(0, 0) ^ SWS_MASK(0, 1) | SWS_MASK_OFF3)
212 9952 WRAP_LINEAR(affine3a, SWS_MASK_MAT3 | SWS_MASK_OFF3 | SWS_MASK_ALPHA)
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_0001),
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_clear_0001),
237 &fn(op_clear_1000),
238 &fn(op_clear_1100),
239
240 &fn(op_linear_luma),
241 &fn(op_linear_alpha),
242 &fn(op_linear_lumalpha),
243 &fn(op_linear_yalpha),
244 &fn(op_linear_dot3),
245 &fn(op_linear_dot3a),
246 &fn(op_linear_row0),
247 &fn(op_linear_diag3),
248 &fn(op_linear_diag4),
249 &fn(op_linear_diagoff3),
250 &fn(op_linear_affine3),
251 &fn(op_linear_affine3uv),
252 &fn(op_linear_affine3x),
253 &fn(op_linear_affine3xa),
254 &fn(op_linear_affine3xy),
255 &fn(op_linear_affine3a),
256
257 &fn(op_filter1_v),
258 &fn(op_filter2_v),
259 &fn(op_filter3_v),
260 &fn(op_filter4_v),
261
262 &fn(op_filter1_h),
263 &fn(op_filter2_h),
264 &fn(op_filter3_h),
265 &fn(op_filter4_h),
266
267 NULL
268 },
269 };
270
271 #undef PIXEL_TYPE
272 #undef PIXEL_MAX
273 #undef pixel_t
274 #undef inter_t
275 #undef block_t
276 #undef px
277
278 #undef FMT_CHAR
279 #undef IS_FLOAT
280