| 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 | #include "libavutil/mem.h" | ||
| 23 | #include "libavutil/rational.h" | ||
| 24 | |||
| 25 | #include "ops_chain.h" | ||
| 26 | |||
| 27 | 99130 | SwsOpChain *ff_sws_op_chain_alloc(void) | |
| 28 | { | ||
| 29 | 99130 | return av_mallocz(sizeof(SwsOpChain)); | |
| 30 | } | ||
| 31 | |||
| 32 | 99130 | void ff_sws_op_chain_free_cb(void *ptr) | |
| 33 | { | ||
| 34 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 99130 times.
|
99130 | if (!ptr) |
| 35 | ✗ | return; | |
| 36 | |||
| 37 | 99130 | SwsOpChain *chain = ptr; | |
| 38 |
2/2✓ Branch 0 taken 571178 times.
✓ Branch 1 taken 99130 times.
|
670308 | for (int i = 0; i < chain->num_impl + 1; i++) { |
| 39 |
2/2✓ Branch 0 taken 104882 times.
✓ Branch 1 taken 466296 times.
|
571178 | if (chain->free[i]) |
| 40 | 104882 | chain->free[i](&chain->impl[i].priv); | |
| 41 | } | ||
| 42 | |||
| 43 | 99130 | av_free(chain); | |
| 44 | } | ||
| 45 | |||
| 46 | 472048 | int ff_sws_op_chain_append(SwsOpChain *chain, SwsFuncPtr func, | |
| 47 | void (*free)(SwsOpPriv *), const SwsOpPriv *priv) | ||
| 48 | { | ||
| 49 | 472048 | const int idx = chain->num_impl; | |
| 50 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 472048 times.
|
472048 | if (idx == SWS_MAX_OPS) |
| 51 | ✗ | return AVERROR(EINVAL); | |
| 52 | |||
| 53 | av_assert1(func); | ||
| 54 | 472048 | chain->impl[idx].cont = func; | |
| 55 | 472048 | chain->impl[idx + 1].priv = *priv; | |
| 56 | 472048 | chain->free[idx + 1] = free; | |
| 57 | 472048 | chain->num_impl++; | |
| 58 | 472048 | return 0; | |
| 59 | } | ||
| 60 | |||
| 61 | #define q2pixel(type, q) ((q).den ? (type) (q).num / (q).den : 0) | ||
| 62 | |||
| 63 | #if ARCH_AARCH64 | ||
| 64 | int ff_sws_setup_scale(const SwsImplParams *params, SwsImplResult *out) | ||
| 65 | { | ||
| 66 | const SwsOp *op = params->op; | ||
| 67 | const AVRational64 factor = op->scale.factor; | ||
| 68 | switch (op->type) { | ||
| 69 | case SWS_PIXEL_U8: out->priv.u8[0] = q2pixel(uint8_t, factor); break; | ||
| 70 | case SWS_PIXEL_U16: out->priv.u16[0] = q2pixel(uint16_t, factor); break; | ||
| 71 | case SWS_PIXEL_U32: out->priv.u32[0] = q2pixel(uint32_t, factor); break; | ||
| 72 | case SWS_PIXEL_F32: out->priv.f32[0] = q2pixel(float, factor); break; | ||
| 73 | default: return AVERROR(EINVAL); | ||
| 74 | } | ||
| 75 | |||
| 76 | return 0; | ||
| 77 | } | ||
| 78 | |||
| 79 | int ff_sws_setup_clamp(const SwsImplParams *params, SwsImplResult *out) | ||
| 80 | { | ||
| 81 | const SwsOp *op = params->op; | ||
| 82 | for (int i = 0; i < 4; i++) { | ||
| 83 | const AVRational64 limit = op->clamp.limit[i]; | ||
| 84 | switch (op->type) { | ||
| 85 | case SWS_PIXEL_U8: out->priv.u8[i] = q2pixel(uint8_t, limit); break; | ||
| 86 | case SWS_PIXEL_U16: out->priv.u16[i] = q2pixel(uint16_t, limit); break; | ||
| 87 | case SWS_PIXEL_U32: out->priv.u32[i] = q2pixel(uint32_t, limit); break; | ||
| 88 | case SWS_PIXEL_F32: out->priv.f32[i] = q2pixel(float, limit); break; | ||
| 89 | default: return AVERROR(EINVAL); | ||
| 90 | } | ||
| 91 | } | ||
| 92 | |||
| 93 | return 0; | ||
| 94 | } | ||
| 95 | |||
| 96 | int ff_sws_setup_clear(const SwsImplParams *params, SwsImplResult *out) | ||
| 97 | { | ||
| 98 | const SwsOp *op = params->op; | ||
| 99 | for (int i = 0; i < 4; i++) { | ||
| 100 | const AVRational64 value = op->clear.value[i]; | ||
| 101 | if (!value.den) | ||
| 102 | continue; | ||
| 103 | switch (op->type) { | ||
| 104 | case SWS_PIXEL_U8: out->priv.u8[i] = q2pixel(uint8_t, value); break; | ||
| 105 | case SWS_PIXEL_U16: out->priv.u16[i] = q2pixel(uint16_t, value); break; | ||
| 106 | case SWS_PIXEL_U32: out->priv.u32[i] = q2pixel(uint32_t, value); break; | ||
| 107 | case SWS_PIXEL_F32: out->priv.f32[i] = q2pixel(float, value); break; | ||
| 108 | default: return AVERROR(EINVAL); | ||
| 109 | } | ||
| 110 | } | ||
| 111 | |||
| 112 | return 0; | ||
| 113 | } | ||
| 114 | #endif | ||
| 115 | |||
| 116 | 483280 | int ff_sws_uop_lookup(SwsContext *ctx, const SwsUOpTable *const tables[], | |
| 117 | int num_tables, const SwsUOp *uop, const int block_size, | ||
| 118 | SwsOpChain *chain) | ||
| 119 | { | ||
| 120 | 483280 | const unsigned cpu_flags = av_get_cpu_flags(); | |
| 121 | 483280 | const SwsUOpEntry *match = NULL; | |
| 122 | int ret; | ||
| 123 | |||
| 124 | 483280 | SwsImplParams params = { | |
| 125 | .ctx = ctx, | ||
| 126 | .uop = uop | ||
| 127 | }; | ||
| 128 | |||
| 129 |
4/4✓ Branch 0 taken 2576277 times.
✓ Branch 1 taken 472048 times.
✓ Branch 2 taken 2565045 times.
✓ Branch 3 taken 11232 times.
|
3048325 | for (int n = 0; !match && n < num_tables; n++) { |
| 130 | 2565045 | const SwsUOpTable *table = params.table = tables[n]; | |
| 131 |
3/4✓ Branch 0 taken 2565045 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 1231323 times.
✓ Branch 3 taken 1333722 times.
|
2565045 | if (table->block_size && table->block_size != block_size || |
| 132 |
2/2✓ Branch 0 taken 4284 times.
✓ Branch 1 taken 1227039 times.
|
1231323 | table->cpu_flags & ~cpu_flags) |
| 133 | 1338006 | continue; | |
| 134 | |||
| 135 |
2/2✓ Branch 0 taken 93058020 times.
✓ Branch 1 taken 754991 times.
|
93813011 | for (int i = 0; table->entries[i]; i++) { |
| 136 | 93058020 | const SwsUOpEntry *entry = table->entries[i]; | |
| 137 | 93058020 | const SwsUOp entry_uop = { | |
| 138 | 93058020 | .uop = entry->uop, | |
| 139 | 93058020 | .type = entry->type, | |
| 140 | 93058020 | .mask = entry->mask, | |
| 141 | .par = entry->par, | ||
| 142 | }; | ||
| 143 | |||
| 144 |
2/2✓ Branch 1 taken 92584282 times.
✓ Branch 2 taken 473738 times.
|
93058020 | if (ff_sws_uop_cmp(uop, &entry_uop) != 0) |
| 145 | 92585972 | continue; | |
| 146 |
4/4✓ Branch 0 taken 19294 times.
✓ Branch 1 taken 454444 times.
✓ Branch 3 taken 1690 times.
✓ Branch 4 taken 17604 times.
|
473738 | if (entry->check && !entry->check(¶ms)) |
| 147 | 1690 | continue; | |
| 148 | |||
| 149 | 472048 | match = entry; | |
| 150 | 472048 | break; | |
| 151 | } | ||
| 152 | } | ||
| 153 | |||
| 154 |
2/2✓ Branch 0 taken 11232 times.
✓ Branch 1 taken 472048 times.
|
483280 | if (!match) { |
| 155 | char name[64]; | ||
| 156 | 11232 | ff_sws_uop_name(uop, name); | |
| 157 | 11232 | av_log(ctx, AV_LOG_DEBUG, "No implementation found for: %s\n", name); | |
| 158 | 11232 | return AVERROR(ENOTSUP); | |
| 159 | } | ||
| 160 | |||
| 161 | 472048 | SwsImplResult res = {0}; | |
| 162 |
2/2✓ Branch 0 taken 211232 times.
✓ Branch 1 taken 260816 times.
|
472048 | if (match->setup) { |
| 163 | 211232 | ret = match->setup(¶ms, &res); | |
| 164 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 211232 times.
|
211232 | if (ret < 0) |
| 165 | ✗ | return ret; | |
| 166 | } | ||
| 167 | |||
| 168 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 472048 times.
|
472048 | ret = ff_sws_op_chain_append(chain, res.func ? res.func : match->func, |
| 169 | res.free, &res.priv); | ||
| 170 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 472048 times.
|
472048 | if (ret < 0) { |
| 171 | ✗ | if (res.free) | |
| 172 | ✗ | res.free(&res.priv); | |
| 173 | ✗ | return ret; | |
| 174 | } | ||
| 175 | |||
| 176 |
2/2✓ Branch 0 taken 1888192 times.
✓ Branch 1 taken 472048 times.
|
2360240 | for (int i = 0; i < 4; i++) { |
| 177 | 1888192 | chain->over_read[i] = FFMAX(chain->over_read[i], res.over_read[i]); | |
| 178 | 1888192 | chain->over_write[i] = FFMAX(chain->over_write[i], res.over_write[i]); | |
| 179 | } | ||
| 180 | |||
| 181 | 472048 | chain->cpu_flags |= params.table->cpu_flags; | |
| 182 | 472048 | return 0; | |
| 183 | } | ||
| 184 | |||
| 185 | 1571 | int ff_sws_setup_scalar(const SwsImplParams *params, SwsImplResult *out) | |
| 186 | { | ||
| 187 | 1571 | const SwsUOp *uop = params->uop; | |
| 188 | 1571 | const SwsPixel scalar = uop->data.scalar; | |
| 189 |
4/5✓ Branch 0 taken 14 times.
✓ Branch 1 taken 28 times.
✓ Branch 2 taken 53 times.
✓ Branch 3 taken 1476 times.
✗ Branch 4 not taken.
|
1571 | switch (uop->type) { |
| 190 | 14 | case SWS_PIXEL_U8: out->priv.u8[0] = scalar.u8; break; | |
| 191 | 28 | case SWS_PIXEL_U16: out->priv.u16[0] = scalar.u16; break; | |
| 192 | 53 | case SWS_PIXEL_U32: out->priv.u32[0] = scalar.u32; break; | |
| 193 | 1476 | case SWS_PIXEL_F32: out->priv.f32[0] = scalar.f32; break; | |
| 194 | ✗ | default: return AVERROR(EINVAL); | |
| 195 | } | ||
| 196 | |||
| 197 | 1571 | return 0; | |
| 198 | } | ||
| 199 | |||
| 200 | 64290 | int ff_sws_setup_vec4(const SwsImplParams *params, SwsImplResult *out) | |
| 201 | { | ||
| 202 | 64290 | const SwsUOp *uop = params->uop; | |
| 203 |
2/2✓ Branch 0 taken 257160 times.
✓ Branch 1 taken 64290 times.
|
321450 | for (int i = 0; i < 4; i++) { |
| 204 | 257160 | const SwsPixel vi = uop->data.vec4[i]; | |
| 205 |
4/5✓ Branch 0 taken 1336 times.
✓ Branch 1 taken 11772 times.
✓ Branch 2 taken 12224 times.
✓ Branch 3 taken 231828 times.
✗ Branch 4 not taken.
|
257160 | switch (uop->type) { |
| 206 | 1336 | case SWS_PIXEL_U8: out->priv.u8[i] = vi.u8; break; | |
| 207 | 11772 | case SWS_PIXEL_U16: out->priv.u16[i] = vi.u16; break; | |
| 208 | 12224 | case SWS_PIXEL_U32: out->priv.u32[i] = vi.u32; break; | |
| 209 | 231828 | case SWS_PIXEL_F32: out->priv.f32[i] = vi.f32; break; | |
| 210 | ✗ | default: return AVERROR(EINVAL); | |
| 211 | } | ||
| 212 | } | ||
| 213 | |||
| 214 | 64290 | return 0; | |
| 215 | } | ||
| 216 |