| Line | Branch | Exec | Source |
|---|---|---|---|
| 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 <errno.h> | ||
| 20 | #include <stdbool.h> | ||
| 21 | #include <stdio.h> | ||
| 22 | #include <string.h> | ||
| 23 | |||
| 24 | #ifdef _WIN32 | ||
| 25 | #include <io.h> | ||
| 26 | #include <fcntl.h> | ||
| 27 | #endif | ||
| 28 | |||
| 29 | #include "libavutil/bprint.h" | ||
| 30 | #include "libavutil/error.h" | ||
| 31 | #include "libavutil/macros.h" | ||
| 32 | #include "libavutil/mem.h" | ||
| 33 | #include "libavutil/pixfmt.h" | ||
| 34 | #include "libavutil/thread.h" | ||
| 35 | #include "libavutil/tree.h" | ||
| 36 | #include "ops.h" | ||
| 37 | #include "ops_dispatch.h" | ||
| 38 | #include "op_list_gen_template.c" | ||
| 39 | #include "swscale.h" | ||
| 40 | #include "uops.h" | ||
| 41 | #include "uops_list.h" | ||
| 42 | |||
| 43 | static const struct { | ||
| 44 | char full[32]; | ||
| 45 | char abbr[32]; | ||
| 46 | } uop_names[SWS_UOP_TYPE_NB] = { | ||
| 47 | #define UOP_NAME(OP, ABBR) [OP] = { #OP, ABBR }, | ||
| 48 | UOPS_LIST(UOP_NAME) | ||
| 49 | }; | ||
| 50 | |||
| 51 | static const struct { | ||
| 52 | char full[16]; | ||
| 53 | char prefix[8]; | ||
| 54 | } pixel_types[SWS_PIXEL_TYPE_NB] = { | ||
| 55 | [SWS_PIXEL_NONE] = { "SWS_PIXEL_NONE", "" }, | ||
| 56 | [SWS_PIXEL_U8] = { "SWS_PIXEL_U8", "U8_" }, | ||
| 57 | [SWS_PIXEL_U16] = { "SWS_PIXEL_U16", "U16_" }, | ||
| 58 | [SWS_PIXEL_U32] = { "SWS_PIXEL_U32", "U32_" }, | ||
| 59 | [SWS_PIXEL_F32] = { "SWS_PIXEL_F32", "F32_" }, | ||
| 60 | }; | ||
| 61 | |||
| 62 | /* Print uop params either as C struct designators, or as plain values */ | ||
| 63 | 2120 | static void print_params(AVBPrint *bp, const SwsUOp *uop, bool designated) | |
| 64 | { | ||
| 65 | #define FMT(FIELD, VALUE, ...) av_bprintf(bp, designated ? ", ." FIELD " = " VALUE : ", " VALUE, __VA_ARGS__) | ||
| 66 | #define FMT_SUB(FIELD, VALUE, ...) av_bprintf(bp, designated ? ", ." FIELD " = {" VALUE "}" : ", " VALUE, __VA_ARGS__) | ||
| 67 | |||
| 68 | 2120 | const SwsUOpParams *par = &uop->par; | |
| 69 |
10/10✓ Branch 0 taken 156 times.
✓ Branch 1 taken 66 times.
✓ Branch 2 taken 105 times.
✓ Branch 3 taken 434 times.
✓ Branch 4 taken 90 times.
✓ Branch 5 taken 165 times.
✓ Branch 6 taken 219 times.
✓ Branch 7 taken 120 times.
✓ Branch 8 taken 20 times.
✓ Branch 9 taken 745 times.
|
2120 | switch (uop->uop) { |
| 70 | 156 | case SWS_UOP_READ_PLANAR_FH: | |
| 71 | case SWS_UOP_READ_PLANAR_FV: | ||
| 72 | case SWS_UOP_READ_PLANAR_FV_FMA: | ||
| 73 |
2/2✓ Branch 0 taken 60 times.
✓ Branch 1 taken 96 times.
|
156 | FMT("par.filter.type", "%s", pixel_types[par->filter.type].full); |
| 74 | 156 | break; | |
| 75 | 66 | case SWS_UOP_RW_SHUFFLE: | |
| 76 |
2/2✓ Branch 0 taken 22 times.
✓ Branch 1 taken 44 times.
|
66 | FMT("par.shuffle.clear_value", "0x%x", par->shuffle.clear_value); |
| 77 |
2/2✓ Branch 0 taken 22 times.
✓ Branch 1 taken 44 times.
|
66 | FMT("par.shuffle.read_size", "%u", par->shuffle.read_size); |
| 78 |
2/2✓ Branch 0 taken 22 times.
✓ Branch 1 taken 44 times.
|
66 | FMT("par.shuffle.write_size", "%u", par->shuffle.write_size); |
| 79 | 66 | break; | |
| 80 | 105 | case SWS_UOP_LSHIFT: | |
| 81 | case SWS_UOP_RSHIFT: | ||
| 82 |
2/2✓ Branch 0 taken 42 times.
✓ Branch 1 taken 63 times.
|
105 | FMT("par.shift.amount", "%u", par->shift.amount); |
| 83 | 105 | break; | |
| 84 | 434 | case SWS_UOP_PERMUTE: | |
| 85 | case SWS_UOP_COPY: | ||
| 86 |
2/2✓ Branch 0 taken 174 times.
✓ Branch 1 taken 260 times.
|
434 | FMT("par.move.num_moves", "%d", par->move.num_moves); |
| 87 |
2/2✓ Branch 0 taken 174 times.
✓ Branch 1 taken 260 times.
|
434 | FMT_SUB("par.move.dst", "%d, %d, %d, %d, %d, %d", |
| 88 | par->move.dst[0], par->move.dst[1], par->move.dst[2], | ||
| 89 | par->move.dst[3], par->move.dst[4], par->move.dst[5]); | ||
| 90 |
2/2✓ Branch 0 taken 174 times.
✓ Branch 1 taken 260 times.
|
434 | FMT_SUB("par.move.src", "%d, %d, %d, %d, %d, %d", |
| 91 | par->move.src[0], par->move.src[1], par->move.src[2], | ||
| 92 | par->move.src[3], par->move.src[4], par->move.src[5]); | ||
| 93 | 434 | break; | |
| 94 | 90 | case SWS_UOP_PACK: | |
| 95 | case SWS_UOP_UNPACK: | ||
| 96 |
2/2✓ Branch 0 taken 36 times.
✓ Branch 1 taken 54 times.
|
90 | FMT_SUB("par.pack.pattern", "%d, %d, %d, %d", |
| 97 | par->pack.pattern[0], par->pack.pattern[1], | ||
| 98 | par->pack.pattern[2], par->pack.pattern[3]); | ||
| 99 | 90 | break; | |
| 100 | 165 | case SWS_UOP_CLEAR: | |
| 101 |
2/2✓ Branch 0 taken 66 times.
✓ Branch 1 taken 99 times.
|
165 | FMT("par.clear.one", "0x%05x", par->clear.one); |
| 102 |
2/2✓ Branch 0 taken 66 times.
✓ Branch 1 taken 99 times.
|
165 | FMT("par.clear.zero", "0x%05x", par->clear.zero); |
| 103 | 165 | break; | |
| 104 | 219 | case SWS_UOP_LINEAR: | |
| 105 | case SWS_UOP_LINEAR_FMA: | ||
| 106 |
2/2✓ Branch 0 taken 82 times.
✓ Branch 1 taken 137 times.
|
219 | FMT("par.lin.one", "0x%05x", par->lin.one); |
| 107 |
2/2✓ Branch 0 taken 82 times.
✓ Branch 1 taken 137 times.
|
219 | FMT("par.lin.zero", "0x%05x", par->lin.zero); |
| 108 |
2/2✓ Branch 0 taken 84 times.
✓ Branch 1 taken 135 times.
|
219 | if (uop->uop == SWS_UOP_LINEAR_FMA) |
| 109 |
2/2✓ Branch 0 taken 28 times.
✓ Branch 1 taken 56 times.
|
84 | FMT("par.lin.exact", "0x%05x", par->lin.exact); |
| 110 | 219 | break; | |
| 111 | 120 | case SWS_UOP_DITHER: | |
| 112 |
2/2✓ Branch 0 taken 48 times.
✓ Branch 1 taken 72 times.
|
120 | FMT_SUB("par.dither.y_offset", "%u, %u, %u, %u", |
| 113 | par->dither.y_offset[0], par->dither.y_offset[1], | ||
| 114 | par->dither.y_offset[2], par->dither.y_offset[3]); | ||
| 115 |
2/2✓ Branch 0 taken 48 times.
✓ Branch 1 taken 72 times.
|
120 | FMT("par.dither.size_log2", "%u", par->dither.size_log2); |
| 116 | 120 | break; | |
| 117 | 20 | case SWS_UOP_LUT_3D: | |
| 118 |
2/2✓ Branch 0 taken 8 times.
✓ Branch 1 taken 12 times.
|
20 | FMT("par.lut3d.dynamic", "%d", par->lut3d.dynamic); |
| 119 | 20 | break; | |
| 120 | } | ||
| 121 | |||
| 122 | #undef FMT | ||
| 123 | #undef FMT_SUB | ||
| 124 | 2120 | } | |
| 125 | |||
| 126 | 836 | static int generate_entry_struct(void *opaque, void *key) | |
| 127 | { | ||
| 128 | 836 | const SwsUOp *ref = opaque; | |
| 129 | 836 | const SwsUOp *uop = key; | |
| 130 | 836 | AVBPrint *bp = ref->data.opaque; | |
| 131 | char name[SWS_UOP_NAME_MAX]; | ||
| 132 | 836 | ff_sws_uop_name(uop, name); | |
| 133 | 836 | av_bprintf(bp, " \\\n MACRO(__VA_ARGS__, %-40s", name); | |
| 134 | 836 | av_bprintf(bp, ", .type = %-13s, .uop = %-24s, .mask = 0x%x", | |
| 135 | 836 | pixel_types[uop->type].full, uop_names[uop->uop].full, uop->mask); | |
| 136 | 836 | print_params(bp, uop, true); | |
| 137 | 836 | av_bprintf(bp, ")"); | |
| 138 | 836 | return 0; | |
| 139 | } | ||
| 140 | |||
| 141 | 836 | static int generate_entry_args(void *opaque, void *key) | |
| 142 | { | ||
| 143 | 836 | const SwsUOp *ref = opaque; | |
| 144 | 836 | const SwsUOp *uop = key; | |
| 145 | 836 | AVBPrint *bp = ref->data.opaque; | |
| 146 | char name[SWS_UOP_NAME_MAX]; | ||
| 147 | 836 | ff_sws_uop_name(uop, name); | |
| 148 | 836 | av_bprintf(bp, " \\\n MACRO(__VA_ARGS__, %-40s, %-13s, %-24s, 0x%x", | |
| 149 | 836 | name, pixel_types[uop->type].full, uop_names[uop->uop].full, uop->mask); | |
| 150 | 836 | print_params(bp, uop, false); | |
| 151 | 836 | av_bprintf(bp, ")"); | |
| 152 | 836 | return 0; | |
| 153 | } | ||
| 154 | |||
| 155 | 448 | static int generate_entry_nasm(void *opaque, void *key) | |
| 156 | { | ||
| 157 | 448 | const SwsUOp *ref = opaque; | |
| 158 | 448 | const SwsUOp *uop = key; | |
| 159 | 448 | AVBPrint *bp = ref->data.opaque; | |
| 160 | char name[SWS_UOP_NAME_MAX]; | ||
| 161 | 448 | ff_sws_uop_name(uop, name); | |
| 162 | 448 | av_bprintf(bp, "\n DECL_OP %%1, %-40s, %-13s, %-24s, 0x%x", | |
| 163 | 448 | name, pixel_types[uop->type].full, uop_names[uop->uop].full, uop->mask); | |
| 164 | 448 | print_params(bp, uop, false); | |
| 165 | 448 | return 0; | |
| 166 | } | ||
| 167 | |||
| 168 | struct EnumPriv { | ||
| 169 | const struct Backend *backend; | ||
| 170 | struct AVTreeNode *root; | ||
| 171 | AVMutex lock; | ||
| 172 | }; | ||
| 173 | |||
| 174 | 13106504 | static int register_uop(struct EnumPriv *s, const SwsUOp *uop) | |
| 175 | { | ||
| 176 | 13106504 | SwsUOp *key = av_memdup(uop, sizeof(*uop)); | |
| 177 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 13106504 times.
|
13106504 | if (!key) |
| 178 | ✗ | return AVERROR(ENOMEM); | |
| 179 | 13106504 | memset(&key->data, 0, sizeof(key->data)); | |
| 180 | |||
| 181 | 13106504 | struct AVTreeNode *node = av_tree_node_alloc(); | |
| 182 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 13106504 times.
|
13106504 | if (!node) { |
| 183 | ✗ | av_free(key); | |
| 184 | ✗ | return AVERROR(ENOMEM); | |
| 185 | } | ||
| 186 | |||
| 187 | 13106504 | ff_mutex_lock(&s->lock); | |
| 188 | 13106504 | av_tree_insert(&s->root, key, ff_sws_uop_cmp_v, &node); | |
| 189 | 13106504 | ff_mutex_unlock(&s->lock); | |
| 190 |
2/2✓ Branch 0 taken 13105668 times.
✓ Branch 1 taken 836 times.
|
13106504 | if (node) { |
| 191 | 13105668 | av_free(node); | |
| 192 | 13105668 | av_free(key); | |
| 193 | } | ||
| 194 | 13106504 | return 0; | |
| 195 | } | ||
| 196 | |||
| 197 | 3092430 | static int register_flags(SwsContext *ctx, const SwsOpList *ops, SwsUOpFlags flags) | |
| 198 | { | ||
| 199 | 3092430 | SwsUOpList *uops = ff_sws_uop_list_alloc(); | |
| 200 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 3092430 times.
|
3092430 | if (!uops) |
| 201 | ✗ | return AVERROR(ENOMEM); | |
| 202 | |||
| 203 | 3092430 | int ret = ff_sws_ops_translate(ctx, ops, flags, uops); | |
| 204 |
2/2✓ Branch 0 taken 690624 times.
✓ Branch 1 taken 2401806 times.
|
3092430 | if (ret < 0) |
| 205 | 690624 | goto fail; | |
| 206 | |||
| 207 |
2/2✓ Branch 0 taken 13106504 times.
✓ Branch 1 taken 2401806 times.
|
15508310 | for (int i = 0; i < uops->num_ops; i++) { |
| 208 | 13106504 | ret = register_uop(ctx->opaque, &uops->ops[i]); | |
| 209 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 13106504 times.
|
13106504 | if (ret < 0) |
| 210 | ✗ | goto fail; | |
| 211 | } | ||
| 212 | |||
| 213 | 2401806 | fail: | |
| 214 | 3092430 | ff_sws_uop_list_free(&uops); | |
| 215 | 3092430 | return ret; | |
| 216 | } | ||
| 217 | |||
| 218 | enum { | ||
| 219 | COMMON_FLAGS = SWS_UOP_FLAG_EXPAND_BIT | SWS_UOP_FLAG_READ_PALETTE | | ||
| 220 | SWS_UOP_FLAG_ADD | SWS_UOP_FLAG_PUNNING, | ||
| 221 | }; | ||
| 222 | |||
| 223 | typedef struct Backend { | ||
| 224 | const char *name; | ||
| 225 | const char *guard; /* include guard for the generated C header */ | ||
| 226 | SwsUOpFlags flags[2]; /* all flag combinations used by the backend */ | ||
| 227 | } Backend; | ||
| 228 | |||
| 229 | static const Backend backends[] = { | ||
| 230 | { | ||
| 231 | .name = "c", | ||
| 232 | .guard = "SWSCALE_UOPS_MACROS_H", | ||
| 233 | .flags = { COMMON_FLAGS }, | ||
| 234 | }, { | ||
| 235 | .name = "x86", | ||
| 236 | .guard = "SWSCALE_X86_UOPS_MACROS_H", | ||
| 237 | .flags = { | ||
| 238 | COMMON_FLAGS | SWS_UOP_FLAG_PSHUFB, | ||
| 239 | COMMON_FLAGS | SWS_UOP_FLAG_PSHUFB | SWS_UOP_FLAG_FMA, /* FMA3 */ | ||
| 240 | }, | ||
| 241 | }, | ||
| 242 | }; | ||
| 243 | |||
| 244 | 1 | static const Backend *find_backend(const char *name) | |
| 245 | { | ||
| 246 |
1/2✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
|
2 | for (int i = 0; i < FF_ARRAY_ELEMS(backends); i++) { |
| 247 |
2/2✓ Branch 0 taken 1 times.
✓ Branch 1 taken 1 times.
|
2 | if (!strcmp(backends[i].name, name)) |
| 248 | 1 | return &backends[i]; | |
| 249 | } | ||
| 250 | ✗ | return NULL; | |
| 251 | } | ||
| 252 | |||
| 253 | 2291828 | static int register_uops(SwsContext *ctx, const SwsOpList *ops, | |
| 254 | SwsCompiledOp *out) | ||
| 255 | { | ||
| 256 | 2291828 | const struct EnumPriv *s = ctx->opaque; | |
| 257 |
2/2✓ Branch 0 taken 3893032 times.
✓ Branch 1 taken 800602 times.
|
4693634 | for (int i = 0; i < FF_ARRAY_ELEMS(s->backend->flags); i++) { |
| 258 |
2/2✓ Branch 0 taken 800602 times.
✓ Branch 1 taken 3092430 times.
|
3893032 | if (!s->backend->flags[i]) |
| 259 | 800602 | break; | |
| 260 | 3092430 | int ret = register_flags(ctx, ops, s->backend->flags[i]); | |
| 261 |
2/2✓ Branch 0 taken 690624 times.
✓ Branch 1 taken 2401806 times.
|
3092430 | if (ret < 0) |
| 262 | 690624 | return ret; | |
| 263 | } | ||
| 264 | |||
| 265 | 1601204 | *out = (SwsCompiledOp) {0}; /* dummy value, will be immediately freed */ | |
| 266 | 1601204 | return 0; | |
| 267 | } | ||
| 268 | |||
| 269 | /* Dummy backend that just registers all seen uops */ | ||
| 270 | static const SwsOpBackend backend_uops = { | ||
| 271 | .name = "uops_gen", | ||
| 272 | .compile = register_uops, | ||
| 273 | }; | ||
| 274 | |||
| 275 | 842688 | static int register_all_uops(SwsContext *ctx, void *graph, SwsOpList *ops) | |
| 276 | { | ||
| 277 | /* ff_sws_compile_pass() takes over ownership of `ops` */ | ||
| 278 | 842688 | SwsOpList *copy = ff_sws_op_list_duplicate(ops); | |
| 279 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 842688 times.
|
842688 | if (!copy) |
| 280 | ✗ | return AVERROR(ENOMEM); | |
| 281 | |||
| 282 | 842688 | const int flags = SWS_OP_FLAG_DRY_RUN | SWS_OP_FLAG_SPLIT_MEMCPY; | |
| 283 | 842688 | return ff_sws_compile_pass(graph, &backend_uops, ©, flags, NULL, NULL); | |
| 284 | } | ||
| 285 | |||
| 286 | static const SwsFlags flags_list[] = { | ||
| 287 | 0, | ||
| 288 | SWS_ACCURATE_RND, /* may insert extra 1x1 dither ops (for accurate rounding) */ | ||
| 289 | SWS_BITEXACT, /* prevents some FMA optimizations */ | ||
| 290 | SWS_ACCURATE_RND | SWS_BITEXACT, | ||
| 291 | }; | ||
| 292 | |||
| 293 | /* Limit the range of av_tree_enumerate() to only matching uop and type */ | ||
| 294 | 8480 | static int enum_type(void *opaque, void *elem) | |
| 295 | { | ||
| 296 | 8480 | const SwsUOp *a = opaque, *b = elem; | |
| 297 |
2/2✓ Branch 0 taken 1540 times.
✓ Branch 1 taken 6940 times.
|
8480 | if (a->type != b->type) |
| 298 | 1540 | return (int) b->type - a->type; | |
| 299 |
2/2✓ Branch 0 taken 4820 times.
✓ Branch 1 taken 2120 times.
|
6940 | if (a->uop != b->uop) |
| 300 | 4820 | return (int) b->uop - a->uop; | |
| 301 | 2120 | return 0; | |
| 302 | } | ||
| 303 | |||
| 304 | 836 | static int free_uop_key(void *opaque, void *key) | |
| 305 | { | ||
| 306 | 836 | av_free(key); | |
| 307 | 836 | return 0; | |
| 308 | } | ||
| 309 | |||
| 310 | 4 | static void free_uops(struct AVTreeNode **root) | |
| 311 | { | ||
| 312 | 4 | av_tree_enumerate(*root, NULL, NULL, free_uop_key); | |
| 313 | 4 | av_tree_destroy(*root); | |
| 314 | 4 | *root = NULL; | |
| 315 | 4 | } | |
| 316 | |||
| 317 | /** | ||
| 318 | * Collect all unique SwsUOps used by `backend` into a newly allocated tree. | ||
| 319 | * | ||
| 320 | * This function can be quite slow as it iterates over every possible | ||
| 321 | * combination of pixel formats and flags. | ||
| 322 | * | ||
| 323 | * Returns 0 or a negative error code. On success, the tree is returned via | ||
| 324 | * `root`, and must be freed with free_uops() by the caller. | ||
| 325 | */ | ||
| 326 | 2 | static int collect_uops(struct AVTreeNode **root, const Backend *backend) | |
| 327 | { | ||
| 328 | 2 | struct EnumPriv s = { .backend = backend }; | |
| 329 | 2 | SwsLut3D *lut3d = NULL; | |
| 330 | 2 | int ret = ff_mutex_init(&s.lock, NULL); | |
| 331 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
|
2 | if (ret) |
| 332 | ✗ | return AVERROR(ENOSYS); | |
| 333 | |||
| 334 | /* Allocate dummy graph and context for ff_sws_compile_pass() */ | ||
| 335 | 2 | SwsGraph *graph = ff_sws_graph_alloc(); | |
| 336 | 2 | SwsContext *ctx = sws_alloc_context(); | |
| 337 |
2/4✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 2 times.
|
2 | if (!graph || !ctx) { |
| 338 | ✗ | ret = AVERROR(ENOMEM); | |
| 339 | ✗ | goto fail; | |
| 340 | } | ||
| 341 | |||
| 342 | /* Use this to plumb the enum state through all the layers of abstraction */ | ||
| 343 | 2 | graph->ctx = ctx; | |
| 344 | 2 | ctx->opaque = &s; | |
| 345 | 2 | ctx->scaler = SWS_SCALE_BILINEAR; /* cheaper to generate filter kernels */ | |
| 346 | 2 | ctx->threads = 0; /* use slice threading to speed up tree building */ | |
| 347 | |||
| 348 | /* Allocate dummy 3DLUT to force generation of SWS_UOP_LUT_3D */ | ||
| 349 | 2 | lut3d = ff_sws_lut3d_alloc(); | |
| 350 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
|
2 | if (!lut3d) { |
| 351 | ✗ | ret = AVERROR(ENOMEM); | |
| 352 | ✗ | goto fail; | |
| 353 | } | ||
| 354 | 2 | ret = ff_sws_enum_op_lists(ctx, graph, lut3d, AV_PIX_FMT_NONE, | |
| 355 | AV_PIX_FMT_NONE, register_all_uops); | ||
| 356 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
|
2 | if (ret < 0) |
| 357 | ✗ | goto fail; | |
| 358 | |||
| 359 | 2 | lut3d->dynamic = true; | |
| 360 | 2 | ret = ff_sws_enum_op_lists(ctx, graph, lut3d, AV_PIX_FMT_NONE, | |
| 361 | AV_PIX_FMT_NONE, register_all_uops); | ||
| 362 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
|
2 | if (ret < 0) |
| 363 | ✗ | goto fail; | |
| 364 | |||
| 365 | /* Register all unique uops over every relevant combination of flags */ | ||
| 366 |
2/2✓ Branch 0 taken 8 times.
✓ Branch 1 taken 2 times.
|
10 | for (int i = 0; i < FF_ARRAY_ELEMS(flags_list); i++) { |
| 367 | 8 | ctx->flags = flags_list[i]; | |
| 368 | 8 | ret = ff_sws_enum_op_lists(ctx, graph, NULL, AV_PIX_FMT_NONE, | |
| 369 | AV_PIX_FMT_NONE, register_all_uops); | ||
| 370 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 8 times.
|
8 | if (ret < 0) |
| 371 | ✗ | goto fail; | |
| 372 | } | ||
| 373 | |||
| 374 | 2 | *root = s.root; | |
| 375 | 2 | s.root = NULL; | |
| 376 | |||
| 377 | 2 | fail: | |
| 378 | 2 | av_refstruct_unref(&lut3d); | |
| 379 | 2 | free_uops(&s.root); | |
| 380 | 2 | ff_mutex_destroy(&s.lock); | |
| 381 | 2 | ff_sws_graph_free(&graph); | |
| 382 | 2 | sws_free_context(&ctx); | |
| 383 | 2 | return ret; | |
| 384 | } | ||
| 385 | |||
| 386 | #define BPRINT_STR(str) av_bprint_append_data(bp, str, strlen(str)) | ||
| 387 | |||
| 388 | /** | ||
| 389 | * Generate a set of boilerplate C preprocessor macros for describing and | ||
| 390 | * programmatically iterating over all SwsUOps in the tree at `root`. | ||
| 391 | * | ||
| 392 | * Returns 0 or a negative error code. On success, an allocated string is | ||
| 393 | * returned via `out_str`, and must be av_free()'d by the caller. | ||
| 394 | */ | ||
| 395 | 2 | static int print_macros(struct AVTreeNode *root, const Backend *backend, | |
| 396 | char **out_str) | ||
| 397 | { | ||
| 398 | 2 | AVBPrint bprint, *const bp = &bprint; | |
| 399 | 2 | av_bprint_init(bp, 0, AV_BPRINT_SIZE_UNLIMITED); | |
| 400 | |||
| 401 | 2 | BPRINT_STR( | |
| 402 | "/**\n" | ||
| 403 | " * This file is automatically generated. Do not edit manually.\n" | ||
| 404 | " * To regenerate, run: make fate-sws-uops-macros GEN=1\n" | ||
| 405 | " */\n" | ||
| 406 | "\n"); | ||
| 407 | 2 | av_bprintf(bp, "#ifndef %s\n#define %s\n\n", backend->guard, backend->guard); | |
| 408 | 2 | BPRINT_STR( | |
| 409 | "/**\n" | ||
| 410 | " * Boilerplate helper macros, for template-based backends. These will be\n" | ||
| 411 | " * instantiated like this, with parameters in struct order:\n" | ||
| 412 | " * MACRO(__VA_ARGS__, NAME, TYPE, UOP, MASK, [PARAMS,])\n" | ||
| 413 | " * The _STRUCT variants pass all arguments in C struct syntax, while the\n" | ||
| 414 | " * plain variants give them as separate C values (e.g. for use in calls)\n" | ||
| 415 | " */\n" | ||
| 416 | "#define SWS_GLUE3(x, y, z) x ## _ ## y ## _ ## z\n" | ||
| 417 | "#define SWS_FOR(TYPE, UOP, MACRO, ...) \\\n" | ||
| 418 | " SWS_GLUE3(SWS_FOR, TYPE, UOP)(MACRO, __VA_ARGS__)\n" | ||
| 419 | "#define SWS_FOR_STRUCT(TYPE, UOP, MACRO, ...) \\\n" | ||
| 420 | " SWS_GLUE3(SWS_FOR_STRUCT, TYPE, UOP)(MACRO, __VA_ARGS__)\n" | ||
| 421 | "\n"); | ||
| 422 | |||
| 423 | 2 | SwsUOp key = { .data.opaque = bp }; | |
| 424 |
2/2✓ Branch 0 taken 8 times.
✓ Branch 1 taken 2 times.
|
10 | for (key.type = SWS_PIXEL_NONE + 1; key.type < SWS_PIXEL_TYPE_NB; key.type++) { |
| 425 |
2/2✓ Branch 0 taken 288 times.
✓ Branch 1 taken 8 times.
|
296 | for (key.uop = SWS_UOP_INVALID + 1; key.uop < SWS_UOP_TYPE_NB; key.uop++) { |
| 426 | 288 | const char *macro = uop_names[key.uop].full + sizeof("SWS_UOP_") - 1; | |
| 427 | 288 | const char *prefix = pixel_types[key.type].prefix; | |
| 428 | 288 | av_bprintf(bp, "#define SWS_FOR_%s%s(MACRO, ...)", prefix, macro); | |
| 429 | 288 | av_tree_enumerate(root, &key, enum_type, generate_entry_args); | |
| 430 | 288 | av_bprintf(bp, "\n"); | |
| 431 | 288 | av_bprintf(bp, "#define SWS_FOR_STRUCT_%s%s(MACRO, ...)", prefix, macro); | |
| 432 | 288 | av_tree_enumerate(root, &key, enum_type, generate_entry_struct); | |
| 433 | 288 | av_bprintf(bp, "\n"); | |
| 434 | } | ||
| 435 | } | ||
| 436 | |||
| 437 | 2 | av_bprintf(bp, "\n#endif /* %s */", backend->guard); | |
| 438 | 2 | return av_bprint_finalize(bp, out_str); | |
| 439 | } | ||
| 440 | |||
| 441 | /** | ||
| 442 | * Generate the equivalent NASM macros, as one DECL_<TYPE>_<UOP> macro per | ||
| 443 | * uop group. Same return semantics as print_macros(). | ||
| 444 | */ | ||
| 445 | 1 | static int print_nasm_macros(struct AVTreeNode *root, char **out_str) | |
| 446 | { | ||
| 447 | 1 | AVBPrint bprint, *const bp = &bprint; | |
| 448 | 1 | av_bprint_init(bp, 0, AV_BPRINT_SIZE_UNLIMITED); | |
| 449 | |||
| 450 | 1 | BPRINT_STR( | |
| 451 | ";******************************************************************************\n" | ||
| 452 | ";* This file is automatically generated. Do not edit manually.\n" | ||
| 453 | ";* To regenerate, run: make fate-sws-uops-macros GEN=1\n" | ||
| 454 | ";*\n" | ||
| 455 | ";* Boilerplate helper macros, for template-based backends. Invoking\n" | ||
| 456 | ";* DECL_<TYPE>_<UOP> MACRO will instantiate, for every matching uop:\n" | ||
| 457 | ";* DECL_OP MACRO, NAME, TYPE, UOP, MASK, [PARAMS]\n" | ||
| 458 | ";* with parameters in struct order.\n" | ||
| 459 | ";******************************************************************************"); | ||
| 460 | |||
| 461 | 1 | SwsUOp key = { .data.opaque = bp }; | |
| 462 |
2/2✓ Branch 0 taken 4 times.
✓ Branch 1 taken 1 times.
|
5 | for (key.type = SWS_PIXEL_NONE + 1; key.type < SWS_PIXEL_TYPE_NB; key.type++) { |
| 463 |
2/2✓ Branch 0 taken 144 times.
✓ Branch 1 taken 4 times.
|
148 | for (key.uop = SWS_UOP_INVALID + 1; key.uop < SWS_UOP_TYPE_NB; key.uop++) { |
| 464 | 144 | const char *macro = uop_names[key.uop].full + sizeof("SWS_UOP_") - 1; | |
| 465 | 144 | const char *prefix = pixel_types[key.type].prefix; | |
| 466 | 144 | av_bprintf(bp, "\n\n%%macro DECL_%s%s 1", prefix, macro); | |
| 467 | 144 | av_tree_enumerate(root, &key, enum_type, generate_entry_nasm); | |
| 468 | 144 | av_bprintf(bp, "\n%%endmacro"); | |
| 469 | } | ||
| 470 | } | ||
| 471 | |||
| 472 | 1 | return av_bprint_finalize(bp, out_str); | |
| 473 | } | ||
| 474 | |||
| 475 | 1 | static int write_file(const char *path, const char *str) | |
| 476 | { | ||
| 477 | 1 | FILE *out = fopen(path, "wb"); | |
| 478 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
|
1 | if (!out) { |
| 479 | ✗ | int ret = AVERROR(errno); | |
| 480 | ✗ | fprintf(stderr, "Failed to open %s: %s\n", path, av_err2str(ret)); | |
| 481 | ✗ | return ret; | |
| 482 | } | ||
| 483 | |||
| 484 | 1 | int ret = 0; | |
| 485 |
2/4✓ Branch 1 taken 1 times.
✗ Branch 2 not taken.
✗ Branch 4 not taken.
✓ Branch 5 taken 1 times.
|
1 | if (fputs(str, out) < 0 || fputc('\n', out) < 0) |
| 486 | ✗ | ret = AVERROR(EIO); | |
| 487 |
1/4✗ Branch 1 not taken.
✓ Branch 2 taken 1 times.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
1 | if (fclose(out) != 0 && ret >= 0) |
| 488 | ✗ | ret = AVERROR(errno); | |
| 489 | 1 | return ret; | |
| 490 | } | ||
| 491 | |||
| 492 | ✗ | static int usage(void) | |
| 493 | { | ||
| 494 | ✗ | fprintf(stderr, "Usage: uops_macros_gen [-backend BACKEND] [-nasm FILE]\n" | |
| 495 | "Writes C macros to stdout, and optionally NASM macros to FILE\n" | ||
| 496 | "BACKEND is one of: c (default), x86\n"); | ||
| 497 | ✗ | return 1; | |
| 498 | } | ||
| 499 | |||
| 500 | 2 | int main(int argc, char **argv) | |
| 501 | { | ||
| 502 | #ifdef _WIN32 | ||
| 503 | _setmode(_fileno(stdout), _O_BINARY); | ||
| 504 | #endif | ||
| 505 | |||
| 506 | 2 | const Backend *backend = &backends[0]; | |
| 507 | 2 | const char *nasm_path = NULL; | |
| 508 |
2/2✓ Branch 0 taken 2 times.
✓ Branch 1 taken 2 times.
|
4 | for (int i = 1; i < argc; i += 2) { |
| 509 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
|
2 | if (i + 1 >= argc) |
| 510 | ✗ | return usage(); | |
| 511 |
2/2✓ Branch 0 taken 1 times.
✓ Branch 1 taken 1 times.
|
2 | if (!strcmp(argv[i], "-backend")) { |
| 512 | 1 | backend = find_backend(argv[i + 1]); | |
| 513 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
|
1 | if (!backend) |
| 514 | ✗ | return usage(); | |
| 515 |
1/2✓ Branch 0 taken 1 times.
✗ Branch 1 not taken.
|
1 | } else if (!strcmp(argv[i], "-nasm")) { |
| 516 | 1 | nasm_path = argv[i + 1]; | |
| 517 | } else { | ||
| 518 | ✗ | return usage(); | |
| 519 | } | ||
| 520 | } | ||
| 521 | |||
| 522 | 2 | struct AVTreeNode *root = NULL; | |
| 523 | 2 | int ret = collect_uops(&root, backend); | |
| 524 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
|
2 | if (ret < 0) |
| 525 | ✗ | return ret; | |
| 526 | |||
| 527 | 2 | char *macros = NULL; | |
| 528 | 2 | ret = print_macros(root, backend, ¯os); | |
| 529 |
1/2✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
|
2 | if (ret >= 0) |
| 530 | 2 | puts(macros); | |
| 531 | 2 | av_free(macros); | |
| 532 | |||
| 533 |
3/4✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 1 times.
✓ Branch 3 taken 1 times.
|
2 | if (ret >= 0 && nasm_path) { |
| 534 | 1 | ret = print_nasm_macros(root, ¯os); | |
| 535 |
1/2✓ Branch 0 taken 1 times.
✗ Branch 1 not taken.
|
1 | if (ret >= 0) |
| 536 | 1 | ret = write_file(nasm_path, macros); | |
| 537 | 1 | av_free(macros); | |
| 538 | } | ||
| 539 | |||
| 540 | 2 | free_uops(&root); | |
| 541 | 2 | return ret; | |
| 542 | } | ||
| 543 |