/src/ffmpeg/libswscale/ops_backend.h
Line | Count | 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 | | #ifndef SWSCALE_OPS_BACKEND_H |
22 | | #define SWSCALE_OPS_BACKEND_H |
23 | | |
24 | | /** |
25 | | * Helper macros for the C-based backend. |
26 | | * |
27 | | * To use these macros, the following types must be defined: |
28 | | * - PIXEL_TYPE should be one of SWS_PIXEL_* |
29 | | * - pixel_t should be the type of pixels |
30 | | * - block_t should be the type of blocks (groups of pixels) |
31 | | */ |
32 | | |
33 | | #include <assert.h> |
34 | | #include <float.h> |
35 | | #include <stdint.h> |
36 | | |
37 | | #include "libavutil/attributes.h" |
38 | | #include "libavutil/mem.h" |
39 | | |
40 | | #include "ops_chain.h" |
41 | | |
42 | | /** |
43 | | * Internal context holding per-iter execution data. The data pointers will be |
44 | | * directly incremented by the corresponding read/write functions. |
45 | | */ |
46 | | typedef struct SwsOpIter { |
47 | | const uint8_t *in[4]; |
48 | | uint8_t *out[4]; |
49 | | int x, y; |
50 | | } SwsOpIter; |
51 | | |
52 | | #ifdef __clang__ |
53 | | # define SWS_FUNC |
54 | 0 | # define SWS_LOOP AV_PRAGMA(clang loop vectorize(assume_safety)) |
55 | | #elif defined(__GNUC__) |
56 | | # define SWS_FUNC __attribute__((optimize("tree-vectorize"))) |
57 | | # define SWS_LOOP AV_PRAGMA(GCC ivdep) |
58 | | #else |
59 | | # define SWS_FUNC |
60 | | # define SWS_LOOP |
61 | | #endif |
62 | | |
63 | | /* Miscellaneous helpers */ |
64 | 0 | #define bitfn2(name, ext) name ## _ ## ext |
65 | 0 | #define bitfn(name, ext) bitfn2(name, ext) |
66 | | |
67 | | #define FN_SUFFIX AV_JOIN(FMT_CHAR, BIT_DEPTH) |
68 | 0 | #define fn(name) bitfn(name, FN_SUFFIX) |
69 | | |
70 | 0 | #define av_q2pixel(q) ((q).den ? (pixel_t) (q).num / (q).den : 0) |
71 | | |
72 | | /* Helper macros to make writing common function signatures less painful */ |
73 | | #define DECL_FUNC(NAME, ...) \ |
74 | | static av_always_inline void fn(NAME)(SwsOpIter *restrict iter, \ |
75 | | const SwsOpImpl *restrict impl, \ |
76 | | block_t x, block_t y, \ |
77 | | block_t z, block_t w, \ |
78 | | __VA_ARGS__) |
79 | | |
80 | | #define DECL_READ(NAME, ...) \ |
81 | | DECL_FUNC(NAME, const pixel_t *restrict in0, const pixel_t *restrict in1, \ |
82 | | const pixel_t *restrict in2, const pixel_t *restrict in3, \ |
83 | | __VA_ARGS__) |
84 | | |
85 | | #define DECL_WRITE(NAME, ...) \ |
86 | | DECL_FUNC(NAME, pixel_t *restrict out0, pixel_t *restrict out1, \ |
87 | | pixel_t *restrict out2, pixel_t *restrict out3, \ |
88 | | __VA_ARGS__) |
89 | | |
90 | | /* Helper macros to call into functions declared with DECL_FUNC_* */ |
91 | | #define CALL(FUNC, ...) \ |
92 | 0 | fn(FUNC)(iter, impl, x, y, z, w, __VA_ARGS__) |
93 | | |
94 | | #define CALL_READ(FUNC, ...) \ |
95 | 0 | CALL(FUNC, (const pixel_t *) iter->in[0], (const pixel_t *) iter->in[1], \ |
96 | 0 | (const pixel_t *) iter->in[2], (const pixel_t *) iter->in[3], \ |
97 | 0 | __VA_ARGS__) |
98 | | |
99 | | #define CALL_WRITE(FUNC, ...) \ |
100 | 0 | CALL(FUNC, (pixel_t *) iter->out[0], (pixel_t *) iter->out[1], \ |
101 | 0 | (pixel_t *) iter->out[2], (pixel_t *) iter->out[3], __VA_ARGS__) |
102 | | |
103 | | /* Helper macros to declare continuation functions */ |
104 | | #define DECL_IMPL(NAME) \ |
105 | | static SWS_FUNC void fn(NAME)(SwsOpIter *restrict iter, \ |
106 | | const SwsOpImpl *restrict impl, \ |
107 | | block_t x, block_t y, \ |
108 | | block_t z, block_t w) |
109 | | |
110 | | /* Helper macro to call into the next continuation with a given type */ |
111 | | #define CONTINUE(TYPE, ...) \ |
112 | 0 | ((void (*)(SwsOpIter *, const SwsOpImpl *, \ |
113 | 0 | TYPE x, TYPE y, TYPE z, TYPE w)) impl->cont) \ |
114 | 0 | (iter, &impl[1], __VA_ARGS__) |
115 | | |
116 | | /* Helper macros for common op setup code */ |
117 | | #define DECL_SETUP(NAME) \ |
118 | | static int fn(NAME)(const SwsOp *op, SwsOpPriv *out) |
119 | | |
120 | 0 | #define SETUP_MEMDUP(c) ff_setup_memdup(&(c), sizeof(c), out) |
121 | | static inline int ff_setup_memdup(const void *c, size_t size, SwsOpPriv *out) |
122 | 0 | { |
123 | 0 | out->ptr = av_memdup(c, size); |
124 | 0 | return out->ptr ? 0 : AVERROR(ENOMEM); |
125 | 0 | } Unexecuted instantiation: ops_backend.c:ff_setup_memdup Unexecuted instantiation: ops_memcpy.c:ff_setup_memdup |
126 | | |
127 | | /* Helper macro for declaring op table entries */ |
128 | | #define DECL_ENTRY(NAME, ...) \ |
129 | | static const SwsOpEntry fn(op_##NAME) = { \ |
130 | | .func = (SwsFuncPtr) fn(NAME), \ |
131 | | .type = PIXEL_TYPE, \ |
132 | | __VA_ARGS__ \ |
133 | | } |
134 | | |
135 | | /* Helpers to define functions for common subsets of components */ |
136 | | #define DECL_PATTERN(NAME) \ |
137 | | DECL_FUNC(NAME, const bool X, const bool Y, const bool Z, const bool W) |
138 | | |
139 | | #define WRAP_PATTERN(FUNC, X, Y, Z, W, ...) \ |
140 | | DECL_IMPL(FUNC##_##X##Y##Z##W) \ |
141 | 0 | { \ |
142 | 0 | CALL(FUNC, X, Y, Z, W); \ |
143 | 0 | } \ Unexecuted instantiation: ops_backend.c:expand16_1000_u8 Unexecuted instantiation: ops_backend.c:expand16_1001_u8 Unexecuted instantiation: ops_backend.c:expand16_1110_u8 Unexecuted instantiation: ops_backend.c:expand16_1111_u8 Unexecuted instantiation: ops_backend.c:expand32_1000_u8 Unexecuted instantiation: ops_backend.c:expand32_1001_u8 Unexecuted instantiation: ops_backend.c:expand32_1110_u8 Unexecuted instantiation: ops_backend.c:expand32_1111_u8 Unexecuted instantiation: ops_backend.c:min_1000_u8 Unexecuted instantiation: ops_backend.c:min_1001_u8 Unexecuted instantiation: ops_backend.c:min_1110_u8 Unexecuted instantiation: ops_backend.c:min_1111_u8 Unexecuted instantiation: ops_backend.c:max_1000_u8 Unexecuted instantiation: ops_backend.c:max_1001_u8 Unexecuted instantiation: ops_backend.c:max_1110_u8 Unexecuted instantiation: ops_backend.c:max_1111_u8 Unexecuted instantiation: ops_backend.c:scale_1000_u8 Unexecuted instantiation: ops_backend.c:scale_1001_u8 Unexecuted instantiation: ops_backend.c:scale_1110_u8 Unexecuted instantiation: ops_backend.c:scale_1111_u8 Unexecuted instantiation: ops_backend.c:convert_float_1000_u8 Unexecuted instantiation: ops_backend.c:convert_float_1001_u8 Unexecuted instantiation: ops_backend.c:convert_float_1110_u8 Unexecuted instantiation: ops_backend.c:convert_float_1111_u8 Unexecuted instantiation: ops_backend.c:convert_uint16_1000_u8 Unexecuted instantiation: ops_backend.c:convert_uint16_1001_u8 Unexecuted instantiation: ops_backend.c:convert_uint16_1110_u8 Unexecuted instantiation: ops_backend.c:convert_uint16_1111_u8 Unexecuted instantiation: ops_backend.c:convert_uint32_1000_u8 Unexecuted instantiation: ops_backend.c:convert_uint32_1001_u8 Unexecuted instantiation: ops_backend.c:convert_uint32_1110_u8 Unexecuted instantiation: ops_backend.c:convert_uint32_1111_u8 Unexecuted instantiation: ops_backend.c:swap_bytes_1000_u16 Unexecuted instantiation: ops_backend.c:swap_bytes_1001_u16 Unexecuted instantiation: ops_backend.c:swap_bytes_1110_u16 Unexecuted instantiation: ops_backend.c:swap_bytes_1111_u16 Unexecuted instantiation: ops_backend.c:min_1000_u16 Unexecuted instantiation: ops_backend.c:min_1001_u16 Unexecuted instantiation: ops_backend.c:min_1110_u16 Unexecuted instantiation: ops_backend.c:min_1111_u16 Unexecuted instantiation: ops_backend.c:max_1000_u16 Unexecuted instantiation: ops_backend.c:max_1001_u16 Unexecuted instantiation: ops_backend.c:max_1110_u16 Unexecuted instantiation: ops_backend.c:max_1111_u16 Unexecuted instantiation: ops_backend.c:scale_1000_u16 Unexecuted instantiation: ops_backend.c:scale_1001_u16 Unexecuted instantiation: ops_backend.c:scale_1110_u16 Unexecuted instantiation: ops_backend.c:scale_1111_u16 Unexecuted instantiation: ops_backend.c:convert_float_1000_u16 Unexecuted instantiation: ops_backend.c:convert_float_1001_u16 Unexecuted instantiation: ops_backend.c:convert_float_1110_u16 Unexecuted instantiation: ops_backend.c:convert_float_1111_u16 Unexecuted instantiation: ops_backend.c:lshift_1000_u16 Unexecuted instantiation: ops_backend.c:lshift_1001_u16 Unexecuted instantiation: ops_backend.c:lshift_1110_u16 Unexecuted instantiation: ops_backend.c:lshift_1111_u16 Unexecuted instantiation: ops_backend.c:rshift_1000_u16 Unexecuted instantiation: ops_backend.c:rshift_1001_u16 Unexecuted instantiation: ops_backend.c:rshift_1110_u16 Unexecuted instantiation: ops_backend.c:rshift_1111_u16 Unexecuted instantiation: ops_backend.c:convert_uint8_1000_u16 Unexecuted instantiation: ops_backend.c:convert_uint8_1001_u16 Unexecuted instantiation: ops_backend.c:convert_uint8_1110_u16 Unexecuted instantiation: ops_backend.c:convert_uint8_1111_u16 Unexecuted instantiation: ops_backend.c:convert_uint32_1000_u16 Unexecuted instantiation: ops_backend.c:convert_uint32_1001_u16 Unexecuted instantiation: ops_backend.c:convert_uint32_1110_u16 Unexecuted instantiation: ops_backend.c:convert_uint32_1111_u16 Unexecuted instantiation: ops_backend.c:swap_bytes_1000_u32 Unexecuted instantiation: ops_backend.c:swap_bytes_1001_u32 Unexecuted instantiation: ops_backend.c:swap_bytes_1110_u32 Unexecuted instantiation: ops_backend.c:swap_bytes_1111_u32 Unexecuted instantiation: ops_backend.c:min_1000_u32 Unexecuted instantiation: ops_backend.c:min_1001_u32 Unexecuted instantiation: ops_backend.c:min_1110_u32 Unexecuted instantiation: ops_backend.c:min_1111_u32 Unexecuted instantiation: ops_backend.c:max_1000_u32 Unexecuted instantiation: ops_backend.c:max_1001_u32 Unexecuted instantiation: ops_backend.c:max_1110_u32 Unexecuted instantiation: ops_backend.c:max_1111_u32 Unexecuted instantiation: ops_backend.c:scale_1000_u32 Unexecuted instantiation: ops_backend.c:scale_1001_u32 Unexecuted instantiation: ops_backend.c:scale_1110_u32 Unexecuted instantiation: ops_backend.c:scale_1111_u32 Unexecuted instantiation: ops_backend.c:convert_float_1000_u32 Unexecuted instantiation: ops_backend.c:convert_float_1001_u32 Unexecuted instantiation: ops_backend.c:convert_float_1110_u32 Unexecuted instantiation: ops_backend.c:convert_float_1111_u32 Unexecuted instantiation: ops_backend.c:lshift_1000_u32 Unexecuted instantiation: ops_backend.c:lshift_1001_u32 Unexecuted instantiation: ops_backend.c:lshift_1110_u32 Unexecuted instantiation: ops_backend.c:lshift_1111_u32 Unexecuted instantiation: ops_backend.c:rshift_1000_u32 Unexecuted instantiation: ops_backend.c:rshift_1001_u32 Unexecuted instantiation: ops_backend.c:rshift_1110_u32 Unexecuted instantiation: ops_backend.c:rshift_1111_u32 Unexecuted instantiation: ops_backend.c:convert_uint8_1000_u32 Unexecuted instantiation: ops_backend.c:convert_uint8_1001_u32 Unexecuted instantiation: ops_backend.c:convert_uint8_1110_u32 Unexecuted instantiation: ops_backend.c:convert_uint8_1111_u32 Unexecuted instantiation: ops_backend.c:convert_uint16_1000_u32 Unexecuted instantiation: ops_backend.c:convert_uint16_1001_u32 Unexecuted instantiation: ops_backend.c:convert_uint16_1110_u32 Unexecuted instantiation: ops_backend.c:convert_uint16_1111_u32 Unexecuted instantiation: ops_backend.c:convert_uint8_1000_f32 Unexecuted instantiation: ops_backend.c:convert_uint8_1001_f32 Unexecuted instantiation: ops_backend.c:convert_uint8_1110_f32 Unexecuted instantiation: ops_backend.c:convert_uint8_1111_f32 Unexecuted instantiation: ops_backend.c:convert_uint16_1000_f32 Unexecuted instantiation: ops_backend.c:convert_uint16_1001_f32 Unexecuted instantiation: ops_backend.c:convert_uint16_1110_f32 Unexecuted instantiation: ops_backend.c:convert_uint16_1111_f32 Unexecuted instantiation: ops_backend.c:convert_uint32_1000_f32 Unexecuted instantiation: ops_backend.c:convert_uint32_1001_f32 Unexecuted instantiation: ops_backend.c:convert_uint32_1110_f32 Unexecuted instantiation: ops_backend.c:convert_uint32_1111_f32 Unexecuted instantiation: ops_backend.c:min_1000_f32 Unexecuted instantiation: ops_backend.c:min_1001_f32 Unexecuted instantiation: ops_backend.c:min_1110_f32 Unexecuted instantiation: ops_backend.c:min_1111_f32 Unexecuted instantiation: ops_backend.c:max_1000_f32 Unexecuted instantiation: ops_backend.c:max_1001_f32 Unexecuted instantiation: ops_backend.c:max_1110_f32 Unexecuted instantiation: ops_backend.c:max_1111_f32 Unexecuted instantiation: ops_backend.c:scale_1000_f32 Unexecuted instantiation: ops_backend.c:scale_1001_f32 Unexecuted instantiation: ops_backend.c:scale_1110_f32 Unexecuted instantiation: ops_backend.c:scale_1111_f32 Unexecuted instantiation: ops_backend.c:convert_uint32_1000_u32 Unexecuted instantiation: ops_backend.c:convert_uint32_1001_u32 Unexecuted instantiation: ops_backend.c:convert_uint32_1110_u32 Unexecuted instantiation: ops_backend.c:convert_uint32_1111_u32 |
144 | | \ |
145 | | DECL_ENTRY(FUNC##_##X##Y##Z##W, \ |
146 | | .unused = { !X, !Y, !Z, !W }, \ |
147 | | __VA_ARGS__ \ |
148 | | ) |
149 | | |
150 | | #define WRAP_COMMON_PATTERNS(FUNC, ...) \ |
151 | | WRAP_PATTERN(FUNC, 1, 0, 0, 0, __VA_ARGS__); \ |
152 | | WRAP_PATTERN(FUNC, 1, 0, 0, 1, __VA_ARGS__); \ |
153 | | WRAP_PATTERN(FUNC, 1, 1, 1, 0, __VA_ARGS__); \ |
154 | | WRAP_PATTERN(FUNC, 1, 1, 1, 1, __VA_ARGS__) |
155 | | |
156 | | #define REF_COMMON_PATTERNS(NAME) \ |
157 | | &fn(op_##NAME##_1000), \ |
158 | | &fn(op_##NAME##_1001), \ |
159 | | &fn(op_##NAME##_1110), \ |
160 | | &fn(op_##NAME##_1111) |
161 | | |
162 | | #endif |