Makes it easier to work with, despite the heavy ifdeffery.
Showing the changes to libavcodec/vulkan (6 of 7 files). Show all
Unified · Split
| @@ -6,10 +6,8 @@ clean:: · 5 unmodified lines | |||
| 6 | OBJS-$(CONFIG_FFV1_VULKAN_ENCODER) += vulkan/common.o \ | 6 | OBJS-$(CONFIG_FFV1_VULKAN_ENCODER) += vulkan/common.o \ |
| 7 | vulkan/rangecoder.o vulkan/ffv1_vlc.o \ | 7 | vulkan/rangecoder.o vulkan/ffv1_vlc.o \ |
| 8 | vulkan/ffv1_common.o vulkan/ffv1_reset.o \ | 8 | vulkan/ffv1_common.o vulkan/ffv1_reset.o \ |
| 9 | vulkan/ffv1_enc_common.o \ | ||
| 10 | vulkan/ffv1_enc_rct.o vulkan/ffv1_enc_setup.o \ | 9 | vulkan/ffv1_enc_rct.o vulkan/ffv1_enc_setup.o \ |
| 11 | vulkan/ffv1_enc_vlc.o vulkan/ffv1_enc_ac.o \ | 10 | vulkan/ffv1_enc.o |
| 12 | vulkan/ffv1_enc.o vulkan/ffv1_enc_rgb.o | ||
| 13 | 11 | ||
| 14 | OBJS-$(CONFIG_FFV1_VULKAN_HWACCEL) += vulkan/common.o \ | 12 | OBJS-$(CONFIG_FFV1_VULKAN_HWACCEL) += vulkan/common.o \ |
| 15 | vulkan/rangecoder.o vulkan/ffv1_vlc.o \ | 13 | vulkan/rangecoder.o vulkan/ffv1_vlc.o \ |
| @@ -20,12 +20,186 @@ · 19 unmodified lines | |||
| 20 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | 20 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 21 | */ | 21 | */ |
| 22 | 22 | ||
| 23 | ivec2 get_diff(ivec2 pos, ivec2 off, int p, int comp, int sw, int bits) | ||
| 24 | { | ||
| 25 | const ivec2 yoff_border1 = off.x == 0 ? ivec2(1, -1) : ivec2(0, 0); | ||
| 26 | const ivec2 yoff_border2 = off.x == 1 ? ivec2(1, -1) : ivec2(0, 0); | ||
| 27 | |||
| 28 | TYPE top2 = TYPE(0); | ||
| 29 | if (off.y > 1) | ||
| 30 | top2 = TYPE(imageLoad(src[p], pos + ivec2(0, -2))[comp]); | ||
| 31 | |||
| 32 | VTYPE3 top = VTYPE3(TYPE(0), | ||
| 33 | TYPE(0), | ||
| 34 | TYPE(0)); | ||
| 35 | if (off.y > 0 && off != ivec2(0, 1)) | ||
| 36 | top[0] = TYPE(imageLoad(src[p], pos + ivec2(-1, -1) + yoff_border1)[comp]); | ||
| 37 | if (off.y > 0) { | ||
| 38 | top[1] = TYPE(imageLoad(src[p], pos + ivec2(0, -1))[comp]); | ||
| 39 | top[2] = TYPE(imageLoad(src[p], pos + ivec2(min(1, sw - off.x - 1), -1))[comp]); | ||
| 40 | } | ||
| 41 | |||
| 42 | VTYPE3 cur = VTYPE3(TYPE(0), | ||
| 43 | TYPE(0), | ||
| 44 | imageLoad(src[p], pos)[comp]); | ||
| 45 | if (off.x > 0 && off != ivec2(1, 0)) | ||
| 46 | cur[0] = TYPE(imageLoad(src[p], pos + ivec2(-2, 0) + yoff_border2)[comp]); | ||
| 47 | if (off != ivec2(0, 0)) | ||
| 48 | cur[1] = TYPE(imageLoad(src[p], pos + ivec2(-1, 0) + yoff_border1)[comp]); | ||
| 49 | |||
| 50 | /* context, diff */ | ||
| 51 | ivec2 d = ivec2(get_context(VTYPE2(cur), top, top2, context_model), | ||
| 52 | cur[2] - predict(cur[1], VTYPE2(top))); | ||
| 53 | |||
| 54 | if (d[0] < 0) | ||
| 55 | d = -d; | ||
| 56 | |||
| 57 | d[1] = fold(d[1], bits); | ||
| 58 | |||
| 59 | return d; | ||
| 60 | } | ||
| 61 | |||
| 62 | #ifndef GOLOMB | ||
| 63 | void put_rac(inout RangeCoder c, uint64_t state, bool bit) | ||
| 64 | { | ||
| 65 | put_rac_norenorm(c, state, bit); | ||
| 66 | if (c.range < 0x100) | ||
| 67 | renorm_encoder(c); | ||
| 68 | } | ||
| 69 | |||
| 70 | /* Note - only handles signed values */ | ||
| 71 | void put_symbol(inout RangeCoder c, uint64_t state, int v) | ||
| 72 | { | ||
| 73 | bool is_nil = (v == 0); | ||
| 74 | put_rac(c, state, is_nil); | ||
| 75 | if (is_nil) | ||
| 76 | return; | ||
| 77 | |||
| 78 | const int a = abs(v); | ||
| 79 | const int e = findMSB(a); | ||
| 80 | |||
| 81 | state += 1; | ||
| 82 | for (int i = 0; i < e; i++) | ||
| 83 | put_rac(c, state + min(i, 9), true); | ||
| 84 | put_rac(c, state + min(e, 9), false); | ||
| 85 | |||
| 86 | state += 21; | ||
| 87 | for (int i = e - 1; i >= 0; i--) | ||
| 88 | put_rac(c, state + min(i, 9), bool(bitfieldExtract(a, i, 1))); | ||
| 89 | |||
| 90 | put_rac(c, state - 11 + min(e, 10), v < 0); | ||
| 91 | } | ||
| 92 | |||
| 93 | void encode_line_pcm(inout SliceContext sc, int y, int p, int comp, | ||
| 94 | int bits) | ||
| 95 | { | ||
| 96 | ivec2 sp = sc.slice_pos; | ||
| 97 | int w = sc.slice_dim.x; | ||
| 98 | if (p > 0 && p < 3) { | ||
| 99 | w >>= chroma_shift.x; | ||
| 100 | sp >>= chroma_shift; | ||
| 101 | } | ||
| 102 | |||
| 103 | for (int x = 0; x < w; x++) { | ||
| 104 | uint v = imageLoad(src[p], (sp + ivec2(x, y)))[comp]; | ||
| 105 | for (int i = (bits - 1); i >= 0; i--) | ||
| 106 | put_rac_equi(sc.c, bool(bitfieldExtract(v, i, 1))); | ||
| 107 | } | ||
| 108 | } | ||
| 109 | |||
| 110 | void encode_line(inout SliceContext sc, uint64_t state, | ||
| 111 | int y, int p, int comp, int bits, const int run_index) | ||
| 112 | { | ||
| 113 | ivec2 sp = sc.slice_pos; | ||
| 114 | |||
| 115 | int w = sc.slice_dim.x; | ||
| 116 | if (p > 0 && p < 3) { | ||
| 117 | w >>= chroma_shift.x; | ||
| 118 | sp >>= chroma_shift; | ||
| 119 | } | ||
| 120 | |||
| 121 | for (int x = 0; x < w; x++) { | ||
| 122 | const ivec2 d = get_diff(sp + ivec2(x, y), ivec2(x, y), p, comp, w, bits); | ||
| 123 | put_symbol(sc.c, state + CONTEXT_SIZE*d[0], d[1]); | ||
| 124 | } | ||
| 125 | } | ||
| 126 | |||
| 127 | #else /* GOLOMB */ | ||
| 128 | |||
| 129 | void encode_line(inout SliceContext sc, uint64_t state, | ||
| 130 | int y, int p, int comp, int bits, inout int run_index) | ||
| 131 | { | ||
| 132 | ivec2 sp = sc.slice_pos; | ||
| 133 | |||
| 134 | int w = sc.slice_dim.x; | ||
| 135 | if (p > 0 && p < 3) { | ||
| 136 | w >>= chroma_shift.x; | ||
| 137 | sp >>= chroma_shift; | ||
| 138 | } | ||
| 139 | |||
| 140 | int run_count = 0; | ||
| 141 | bool run_mode = false; | ||
| 142 | |||
| 143 | for (int x = 0; x < w; x++) { | ||
| 144 | ivec2 d = get_diff(sp + ivec2(x, y), ivec2(x, y), p, comp, w, bits); | ||
| 145 | |||
| 146 | if (d[0] == 0) | ||
| 147 | run_mode = true; | ||
| 148 | |||
| 149 | if (run_mode) { | ||
| 150 | if (d[1] != 0) { | ||
| 151 | /* A very unlikely loop */ | ||
| 152 | while (run_count >= 1 << log2_run[run_index]) { | ||
| 153 | run_count -= 1 << log2_run[run_index]; | ||
| 154 | run_index++; | ||
| 155 | put_bits(sc.pb, 1, 1); | ||
| 156 | } | ||
| 157 | |||
| 158 | put_bits(sc.pb, 1 + log2_run[run_index], run_count); | ||
| 159 | if (run_index != 0) | ||
| 160 | run_index--; | ||
| 161 | run_count = 0; | ||
| 162 | run_mode = false; | ||
| 163 | if (d[1] > 0) | ||
| 164 | d[1]--; | ||
| 165 | } else { | ||
| 166 | run_count++; | ||
| 167 | } | ||
| 168 | } | ||
| 169 | |||
| 170 | if (!run_mode) { | ||
| 171 | VlcState sb = VlcState(state + VLC_STATE_SIZE*d[0]); | ||
| 172 | Symbol sym = get_vlc_symbol(sb, d[1], bits); | ||
| 173 | put_bits(sc.pb, sym.bits, sym.val); | ||
| 174 | } | ||
| 175 | } | ||
| 176 | |||
| 177 | if (run_mode) { | ||
| 178 | while (run_count >= (1 << log2_run[run_index])) { | ||
| 179 | run_count -= 1 << log2_run[run_index]; | ||
| 180 | run_index++; | ||
| 181 | put_bits(sc.pb, 1, 1); | ||
| 182 | } | ||
| 183 | |||
| 184 | if (run_count > 0) | ||
| 185 | put_bits(sc.pb, 1, 1); | ||
| 186 | } | ||
| 187 | } | ||
| 188 | #endif | ||
| 189 | |||
| 23 | void encode_slice(inout SliceContext sc, const uint slice_idx) | 190 | void encode_slice(inout SliceContext sc, const uint slice_idx) |
| 24 | { | 191 | { |
| 192 | #ifndef RGB | ||
| 25 | int bits = bits_per_raw_sample; | 193 | int bits = bits_per_raw_sample; |
| 194 | #else | ||
| 195 | int bits = 9; | ||
| 196 | if (bits != 8 || sc.slice_coding_mode != 0) | ||
| 197 | bits = bits_per_raw_sample + int(sc.slice_coding_mode != 1); | ||
| 198 | #endif | ||
| 26 | 199 | ||
| 27 | #ifndef GOLOMB | 200 | #ifndef GOLOMB |
| 28 | if (sc.slice_coding_mode == 1) { | 201 | if (sc.slice_coding_mode == 1) { |
| 202 | #ifndef RGB | ||
| 29 | for (int c = 0; c < components; c++) { | 203 | for (int c = 0; c < components; c++) { |
| 30 | 204 | ||
| 31 | int h = sc.slice_dim.y; | 205 | int h = sc.slice_dim.y; |
| @@ -39,12 +213,22 @@ void encode_slice(inout SliceContext sc, const uint slice_idx) · 7 unmodified lines | |||
| 39 | for (int y = 0; y < h; y++) | 213 | for (int y = 0; y < h; y++) |
| 40 | encode_line_pcm(sc, y, p, comp, bits); | 214 | encode_line_pcm(sc, y, p, comp, bits); |
| 41 | } | 215 | } |
| 216 | #else | ||
| 217 | for (int y = 0; y < sc.slice_dim.y; y++) { | ||
| 218 | encode_line_pcm(sc, y, 0, 1, bits); | ||
| 219 | encode_line_pcm(sc, y, 0, 2, bits); | ||
| 220 | encode_line_pcm(sc, y, 0, 0, bits); | ||
| 221 | if (transparency == 1) | ||
| 222 | encode_line_pcm(sc, y, 0, 3, bits); | ||
| 223 | } | ||
| 224 | #endif | ||
| 42 | } else | 225 | } else |
| 43 | #endif | 226 | #endif |
| 44 | { | 227 | { |
| 45 | uint64_t slice_state_off = uint64_t(slice_state) + | 228 | uint64_t slice_state_off = uint64_t(slice_state) + |
| 46 | slice_idx*plane_state_size*codec_planes; | 229 | slice_idx*plane_state_size*codec_planes; |
| 47 | 230 | ||
| 231 | #ifndef RGB | ||
| 48 | for (int c = 0; c < components; c++) { | 232 | for (int c = 0; c < components; c++) { |
| 49 | int run_index = 0; | 233 | int run_index = 0; |
| 50 | 234 | ||
| @@ -62,13 +246,67 @@ void encode_slice(inout SliceContext sc, const uint slice_idx) · 11 unmodified lines | |||
| 62 | if (c != 1) | 246 | if (c != 1) |
| 63 | slice_state_off += plane_state_size; | 247 | slice_state_off += plane_state_size; |
| 64 | } | 248 | } |
| 249 | #else | ||
| 250 | int run_index = 0; | ||
| 251 | for (int y = 0; y < sc.slice_dim.y; y++) { | ||
| 252 | encode_line(sc, slice_state_off + plane_state_size*0, | ||
| 253 | y, 0, 1, bits, run_index); | ||
| 254 | encode_line(sc, slice_state_off + plane_state_size*1, | ||
| 255 | y, 0, 2, bits, run_index); | ||
| 256 | encode_line(sc, slice_state_off + plane_state_size*1, | ||
| 257 | y, 0, 0, bits, run_index); | ||
| 258 | if (transparency == 1) | ||
| 259 | encode_line(sc, slice_state_off + plane_state_size*2, | ||
| 260 | y, 0, 3, bits, run_index); | ||
| 261 | } | ||
| 262 | #endif | ||
| 263 | } | ||
| 264 | } | ||
| 265 | |||
| 266 | void finalize_slice(inout SliceContext sc, const uint slice_idx) | ||
| 267 | { | ||
| 268 | #ifdef GOLOMB | ||
| 269 | uint32_t enc_len = sc.hdr_len + flush_put_bits(sc.pb); | ||
| 270 | #else | ||
| 271 | uint32_t enc_len = rac_terminate(sc.c); | ||
| 272 | #endif | ||
| 273 | |||
| 274 | u8buf bs = u8buf(sc.c.bytestream_start); | ||
| 275 | |||
| 276 | /* Append slice length */ | ||
| 277 | u8vec4 enc_len_p = unpack8(enc_len); | ||
| 278 | bs[enc_len + 0].v = enc_len_p.z; | ||
| 279 | bs[enc_len + 1].v = enc_len_p.y; | ||
| 280 | bs[enc_len + 2].v = enc_len_p.x; | ||
| 281 | enc_len += 3; | ||
| 282 | |||
| 283 | /* Calculate and write CRC */ | ||
| 284 | if (ec != 0) { | ||
| 285 | bs[enc_len].v = uint8_t(0); | ||
| 286 | enc_len++; | ||
| 287 | |||
| 288 | uint32_t crc = crcref; | ||
| 289 | for (int i = 0; i < enc_len; i++) | ||
| 290 | crc = crc_ieee[(crc & 0xFF) ^ uint32_t(bs[i].v)] ^ (crc >> 8); | ||
| 291 | |||
| 292 | if (crcref != 0x00000000) | ||
| 293 | crc ^= 0x8CD88196; | ||
| 294 | |||
| 295 | u8vec4 crc_p = unpack8(crc); | ||
| 296 | bs[enc_len + 0].v = crc_p.x; | ||
| 297 | bs[enc_len + 1].v = crc_p.y; | ||
| 298 | bs[enc_len + 2].v = crc_p.z; | ||
| 299 | bs[enc_len + 3].v = crc_p.w; | ||
| 300 | enc_len += 4; | ||
| 65 | } | 301 | } |
| 66 | 302 | ||
| 67 | finalize_slice(sc, slice_idx); | 303 | slice_results[slice_idx*2 + 0] = enc_len; |
| 304 | slice_results[slice_idx*2 + 1] = uint64_t(bs) - uint64_t(out_data); | ||
| 68 | } | 305 | } |
| 69 | 306 | ||
| 70 | void main(void) | 307 | void main(void) |
| 71 | { | 308 | { |
| 72 | const uint slice_idx = gl_WorkGroupID.y*gl_NumWorkGroups.x + gl_WorkGroupID.x; | 309 | const uint slice_idx = gl_WorkGroupID.y*gl_NumWorkGroups.x + gl_WorkGroupID.x; |
| 73 | encode_slice(slice_ctx[slice_idx], slice_idx); | 310 | encode_slice(slice_ctx[slice_idx], slice_idx); |
| 311 | finalize_slice(slice_ctx[slice_idx], slice_idx); | ||
| 74 | } | 312 | } |
| @@ -1,83 +0,0 @@ | |||
| 1 | /* | ||
| 2 | * Copyright (c) 2024 Lynne <dev@lynne.ee> | ||
| 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 | void put_rac(inout RangeCoder c, uint64_t state, bool bit) | ||
| 22 | { | ||
| 23 | put_rac_norenorm(c, state, bit); | ||
| 24 | if (c.range < 0x100) | ||
| 25 | renorm_encoder(c); | ||
| 26 | } | ||
| 27 | |||
| 28 | /* Note - only handles signed values */ | ||
| 29 | void put_symbol(inout RangeCoder c, uint64_t state, int v) | ||
| 30 | { | ||
| 31 | bool is_nil = (v == 0); | ||
| 32 | put_rac(c, state, is_nil); | ||
| 33 | if (is_nil) | ||
| 34 | return; | ||
| 35 | |||
| 36 | const int a = abs(v); | ||
| 37 | const int e = findMSB(a); | ||
| 38 | |||
| 39 | state += 1; | ||
| 40 | for (int i = 0; i < e; i++) | ||
| 41 | put_rac(c, state + min(i, 9), true); | ||
| 42 | put_rac(c, state + min(e, 9), false); | ||
| 43 | |||
| 44 | state += 21; | ||
| 45 | for (int i = e - 1; i >= 0; i--) | ||
| 46 | put_rac(c, state + min(i, 9), bool(bitfieldExtract(a, i, 1))); | ||
| 47 | |||
| 48 | put_rac(c, state - 11 + min(e, 10), v < 0); | ||
| 49 | } | ||
| 50 | |||
| 51 | void encode_line_pcm(inout SliceContext sc, int y, int p, int comp, | ||
| 52 | int bits) | ||
| 53 | { | ||
| 54 | ivec2 sp = sc.slice_pos; | ||
| 55 | int w = sc.slice_dim.x; | ||
| 56 | if (p > 0 && p < 3) { | ||
| 57 | w >>= chroma_shift.x; | ||
| 58 | sp >>= chroma_shift; | ||
| 59 | } | ||
| 60 | |||
| 61 | for (int x = 0; x < w; x++) { | ||
| 62 | uint v = imageLoad(src[p], (sp + ivec2(x, y)))[comp]; | ||
| 63 | for (int i = (bits - 1); i >= 0; i--) | ||
| 64 | put_rac_equi(sc.c, bool(bitfieldExtract(v, i, 1))); | ||
| 65 | } | ||
| 66 | } | ||
| 67 | |||
| 68 | void encode_line(inout SliceContext sc, uint64_t state, | ||
| 69 | int y, int p, int comp, int bits, const int run_index) | ||
| 70 | { | ||
| 71 | ivec2 sp = sc.slice_pos; | ||
| 72 | |||
| 73 | int w = sc.slice_dim.x; | ||
| 74 | if (p > 0 && p < 3) { | ||
| 75 | w >>= chroma_shift.x; | ||
| 76 | sp >>= chroma_shift; | ||
| 77 | } | ||
| 78 | |||
| 79 | for (int x = 0; x < w; x++) { | ||
| 80 | const ivec2 d = get_diff(sp + ivec2(x, y), ivec2(x, y), p, comp, w, bits); | ||
| 81 | put_symbol(sc.c, state + CONTEXT_SIZE*d[0], d[1]); | ||
| 82 | } | ||
| 83 | } | ||
| @@ -1,101 +0,0 @@ | |||
| 1 | /* | ||
| 2 | * FFv1 codec | ||
| 3 | * | ||
| 4 | * Copyright (c) 2024 Lynne <dev@lynne.ee> | ||
| 5 | * | ||
| 6 | * This file is part of FFmpeg. | ||
| 7 | * | ||
| 8 | * FFmpeg is free software; you can redistribute it and/or | ||
| 9 | * modify it under the terms of the GNU Lesser General Public | ||
| 10 | * License as published by the Free Software Foundation; either | ||
| 11 | * version 2.1 of the License, or (at your option) any later version. | ||
| 12 | * | ||
| 13 | * FFmpeg is distributed in the hope that it will be useful, | ||
| 14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | ||
| 16 | * Lesser General Public License for more details. | ||
| 17 | * | ||
| 18 | * You should have received a copy of the GNU Lesser General Public | ||
| 19 | * License along with FFmpeg; if not, write to the Free Software | ||
| 20 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | ||
| 21 | */ | ||
| 22 | |||
| 23 | ivec2 get_diff(ivec2 pos, ivec2 off, int p, int comp, int sw, int bits) | ||
| 24 | { | ||
| 25 | const ivec2 yoff_border1 = off.x == 0 ? ivec2(1, -1) : ivec2(0, 0); | ||
| 26 | const ivec2 yoff_border2 = off.x == 1 ? ivec2(1, -1) : ivec2(0, 0); | ||
| 27 | |||
| 28 | TYPE top2 = TYPE(0); | ||
| 29 | if (off.y > 1) | ||
| 30 | top2 = TYPE(imageLoad(src[p], pos + ivec2(0, -2))[comp]); | ||
| 31 | |||
| 32 | VTYPE3 top = VTYPE3(TYPE(0), | ||
| 33 | TYPE(0), | ||
| 34 | TYPE(0)); | ||
| 35 | if (off.y > 0 && off != ivec2(0, 1)) | ||
| 36 | top[0] = TYPE(imageLoad(src[p], pos + ivec2(-1, -1) + yoff_border1)[comp]); | ||
| 37 | if (off.y > 0) { | ||
| 38 | top[1] = TYPE(imageLoad(src[p], pos + ivec2(0, -1))[comp]); | ||
| 39 | top[2] = TYPE(imageLoad(src[p], pos + ivec2(min(1, sw - off.x - 1), -1))[comp]); | ||
| 40 | } | ||
| 41 | |||
| 42 | VTYPE3 cur = VTYPE3(TYPE(0), | ||
| 43 | TYPE(0), | ||
| 44 | imageLoad(src[p], pos)[comp]); | ||
| 45 | if (off.x > 0 && off != ivec2(1, 0)) | ||
| 46 | cur[0] = TYPE(imageLoad(src[p], pos + ivec2(-2, 0) + yoff_border2)[comp]); | ||
| 47 | if (off != ivec2(0, 0)) | ||
| 48 | cur[1] = TYPE(imageLoad(src[p], pos + ivec2(-1, 0) + yoff_border1)[comp]); | ||
| 49 | |||
| 50 | /* context, diff */ | ||
| 51 | ivec2 d = ivec2(get_context(VTYPE2(cur), top, top2, context_model), | ||
| 52 | cur[2] - predict(cur[1], VTYPE2(top))); | ||
| 53 | |||
| 54 | if (d[0] < 0) | ||
| 55 | d = -d; | ||
| 56 | |||
| 57 | d[1] = fold(d[1], bits); | ||
| 58 | |||
| 59 | return d; | ||
| 60 | } | ||
| 61 | |||
| 62 | void finalize_slice(inout SliceContext sc, const uint slice_idx) | ||
| 63 | { | ||
| 64 | #ifdef GOLOMB | ||
| 65 | uint32_t enc_len = sc.hdr_len + flush_put_bits(sc.pb); | ||
| 66 | #else | ||
| 67 | uint32_t enc_len = rac_terminate(sc.c); | ||
| 68 | #endif | ||
| 69 | |||
| 70 | u8buf bs = u8buf(sc.c.bytestream_start); | ||
| 71 | |||
| 72 | /* Append slice length */ | ||
| 73 | u8vec4 enc_len_p = unpack8(enc_len); | ||
| 74 | bs[enc_len + 0].v = enc_len_p.z; | ||
| 75 | bs[enc_len + 1].v = enc_len_p.y; | ||
| 76 | bs[enc_len + 2].v = enc_len_p.x; | ||
| 77 | enc_len += 3; | ||
| 78 | |||
| 79 | /* Calculate and write CRC */ | ||
| 80 | if (ec != 0) { | ||
| 81 | bs[enc_len].v = uint8_t(0); | ||
| 82 | enc_len++; | ||
| 83 | |||
| 84 | uint32_t crc = crcref; | ||
| 85 | for (int i = 0; i < enc_len; i++) | ||
| 86 | crc = crc_ieee[(crc & 0xFF) ^ uint32_t(bs[i].v)] ^ (crc >> 8); | ||
| 87 | |||
| 88 | if (crcref != 0x00000000) | ||
| 89 | crc ^= 0x8CD88196; | ||
| 90 | |||
| 91 | u8vec4 crc_p = unpack8(crc); | ||
| 92 | bs[enc_len + 0].v = crc_p.x; | ||
| 93 | bs[enc_len + 1].v = crc_p.y; | ||
| 94 | bs[enc_len + 2].v = crc_p.z; | ||
| 95 | bs[enc_len + 3].v = crc_p.w; | ||
| 96 | enc_len += 4; | ||
| 97 | } | ||
| 98 | |||
| 99 | slice_results[slice_idx*2 + 0] = enc_len; | ||
| 100 | slice_results[slice_idx*2 + 1] = uint64_t(bs) - uint64_t(out_data); | ||
| 101 | } | ||
| @@ -1,83 +0,0 @@ | |||
| 1 | /* | ||
| 2 | * FFv1 codec | ||
| 3 | * | ||
| 4 | * Copyright (c) 2024 Lynne <dev@lynne.ee> | ||
| 5 | * | ||
| 6 | * This file is part of FFmpeg. | ||
| 7 | * | ||
| 8 | * FFmpeg is free software; you can redistribute it and/or | ||
| 9 | * modify it under the terms of the GNU Lesser General Public | ||
| 10 | * License as published by the Free Software Foundation; either | ||
| 11 | * version 2.1 of the License, or (at your option) any later version. | ||
| 12 | * | ||
| 13 | * FFmpeg is distributed in the hope that it will be useful, | ||
| 14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | ||
| 16 | * Lesser General Public License for more details. | ||
| 17 | * | ||
| 18 | * You should have received a copy of the GNU Lesser General Public | ||
| 19 | * License along with FFmpeg; if not, write to the Free Software | ||
| 20 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | ||
| 21 | */ | ||
| 22 | |||
| 23 | void encode_slice_rgb(inout SliceContext sc, const uint slice_idx) | ||
| 24 | { | ||
| 25 | int bits = 9; | ||
| 26 | if (bits != 8 || sc.slice_coding_mode != 0) | ||
| 27 | bits = bits_per_raw_sample + int(sc.slice_coding_mode != 1); | ||
| 28 | |||
| 29 | int run_index = 0; | ||
| 30 | |||
| 31 | #ifndef GOLOMB | ||
| 32 | if (sc.slice_coding_mode == 1) { | ||
| 33 | if (transparency == 1) { | ||
| 34 | for (int y = 0; y < sc.slice_dim.y; y++) { | ||
| 35 | encode_line_pcm(sc, y, 0, 1, bits); | ||
| 36 | encode_line_pcm(sc, y, 0, 2, bits); | ||
| 37 | encode_line_pcm(sc, y, 0, 0, bits); | ||
| 38 | encode_line_pcm(sc, y, 0, 3, bits); | ||
| 39 | } | ||
| 40 | } else { | ||
| 41 | for (int y = 0; y < sc.slice_dim.y; y++) { | ||
| 42 | encode_line_pcm(sc, y, 0, 1, bits); | ||
| 43 | encode_line_pcm(sc, y, 0, 2, bits); | ||
| 44 | encode_line_pcm(sc, y, 0, 0, bits); | ||
| 45 | } | ||
| 46 | } | ||
| 47 | } else | ||
| 48 | #endif | ||
| 49 | { | ||
| 50 | uint64_t slice_state_off = uint64_t(slice_state) + | ||
| 51 | slice_idx*plane_state_size*codec_planes; | ||
| 52 | |||
| 53 | if (transparency == 1) { | ||
| 54 | for (int y = 0; y < sc.slice_dim.y; y++) { | ||
| 55 | encode_line(sc, slice_state_off + plane_state_size*0, | ||
| 56 | y, 0, 1, bits, run_index); | ||
| 57 | encode_line(sc, slice_state_off + plane_state_size*1, | ||
| 58 | y, 0, 2, bits, run_index); | ||
| 59 | encode_line(sc, slice_state_off + plane_state_size*1, | ||
| 60 | y, 0, 0, bits, run_index); | ||
| 61 | encode_line(sc, slice_state_off + plane_state_size*2, | ||
| 62 | y, 0, 3, bits, run_index); | ||
| 63 | } | ||
| 64 | } else { | ||
| 65 | for (int y = 0; y < sc.slice_dim.y; y++) { | ||
| 66 | encode_line(sc, slice_state_off + plane_state_size*0, | ||
| 67 | y, 0, 1, bits, run_index); | ||
| 68 | encode_line(sc, slice_state_off + plane_state_size*1, | ||
| 69 | y, 0, 2, bits, run_index); | ||
| 70 | encode_line(sc, slice_state_off + plane_state_size*1, | ||
| 71 | y, 0, 0, bits, run_index); | ||
| 72 | } | ||
| 73 | } | ||
| 74 | } | ||
| 75 | |||
| 76 | finalize_slice(sc, slice_idx); | ||
| 77 | } | ||
| 78 | |||
| 79 | void main(void) | ||
| 80 | { | ||
| 81 | const uint slice_idx = gl_WorkGroupID.y*gl_NumWorkGroups.x + gl_WorkGroupID.x; | ||
| 82 | encode_slice_rgb(slice_ctx[slice_idx], slice_idx); | ||
| 83 | } | ||
| @@ -1,112 +0,0 @@ | |||
| 1 | /* | ||
| 2 | * FFv1 codec | ||
| 3 | * | ||
| 4 | * Copyright (c) 2024 Lynne <dev@lynne.ee> | ||
| 5 | * | ||
| 6 | * This file is part of FFmpeg. | ||
| 7 | * | ||
| 8 | * FFmpeg is free software; you can redistribute it and/or | ||
| 9 | * modify it under the terms of the GNU Lesser General Public | ||
| 10 | * License as published by the Free Software Foundation; either | ||
| 11 | * version 2.1 of the License, or (at your option) any later version. | ||
| 12 | * | ||
| 13 | * FFmpeg is distributed in the hope that it will be useful, | ||
| 14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | ||
| 16 | * Lesser General Public License for more details. | ||
| 17 | * | ||
| 18 | * You should have received a copy of the GNU Lesser General Public | ||
| 19 | * License along with FFmpeg; if not, write to the Free Software | ||
| 20 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | ||
| 21 | */ | ||
| 22 | |||
| 23 | struct RLEState { | ||
| 24 | int count; | ||
| 25 | int diff; | ||
| 26 | int index; | ||
| 27 | bool mode; | ||
| 28 | }; | ||
| 29 | |||
| 30 | void calc_new_state(inout RLEState state, int context) | ||
| 31 | { | ||
| 32 | if (context == 0) | ||
| 33 | state.mode = false; | ||
| 34 | |||
| 35 | if (!state.mode) | ||
| 36 | return; | ||
| 37 | |||
| 38 | if (state.diff > 0) { | ||
| 39 | while (state.count >= (1 << log2_run[state.index])) { | ||
| 40 | state.count -= 1 << log2_run[state.index]; | ||
| 41 | state.index++; | ||
| 42 | } | ||
| 43 | if (state.index > 0) | ||
| 44 | state.index--; | ||
| 45 | state.count = 0; | ||
| 46 | state.mode = false; | ||
| 47 | if (state.diff > 0) | ||
| 48 | state.diff--; | ||
| 49 | } else { | ||
| 50 | state.count++; | ||
| 51 | } | ||
| 52 | } | ||
| 53 | |||
| 54 | void encode_line(inout SliceContext sc, uint64_t state, | ||
| 55 | int y, int p, int comp, int bits, inout int run_index) | ||
| 56 | { | ||
| 57 | ivec2 sp = sc.slice_pos; | ||
| 58 | |||
| 59 | int w = sc.slice_dim.x; | ||
| 60 | if (p > 0 && p < 3) { | ||
| 61 | w >>= chroma_shift.x; | ||
| 62 | sp >>= chroma_shift; | ||
| 63 | } | ||
| 64 | |||
| 65 | int run_count = 0; | ||
| 66 | bool run_mode = false; | ||
| 67 | |||
| 68 | for (int x = 0; x < w; x++) { | ||
| 69 | ivec2 d = get_diff(sp + ivec2(x, y), ivec2(x, y), p, comp, w, bits); | ||
| 70 | |||
| 71 | if (d[0] == 0) | ||
| 72 | run_mode = true; | ||
| 73 | |||
| 74 | if (run_mode) { | ||
| 75 | if (d[1] != 0) { | ||
| 76 | /* A very unlikely loop */ | ||
| 77 | while (run_count >= 1 << log2_run[run_index]) { | ||
| 78 | run_count -= 1 << log2_run[run_index]; | ||
| 79 | run_index++; | ||
| 80 | put_bits(sc.pb, 1, 1); | ||
| 81 | } | ||
| 82 | |||
| 83 | put_bits(sc.pb, 1 + log2_run[run_index], run_count); | ||
| 84 | if (run_index != 0) | ||
| 85 | run_index--; | ||
| 86 | run_count = 0; | ||
| 87 | run_mode = false; | ||
| 88 | if (d[1] > 0) | ||
| 89 | d[1]--; | ||
| 90 | } else { | ||
| 91 | run_count++; | ||
| 92 | } | ||
| 93 | } | ||
| 94 | |||
| 95 | if (!run_mode) { | ||
| 96 | VlcState sb = VlcState(state + VLC_STATE_SIZE*d[0]); | ||
| 97 | Symbol sym = get_vlc_symbol(sb, d[1], bits); | ||
| 98 | put_bits(sc.pb, sym.bits, sym.val); | ||
| 99 | } | ||
| 100 | } | ||
| 101 | |||
| 102 | if (run_mode) { | ||
| 103 | while (run_count >= (1 << log2_run[run_index])) { | ||
| 104 | run_count -= 1 << log2_run[run_index]; | ||
| 105 | run_index++; | ||
| 106 | put_bits(sc.pb, 1, 1); | ||
| 107 | } | ||
| 108 | |||
| 109 | if (run_count > 0) | ||
| 110 | put_bits(sc.pb, 1, 1); | ||
| 111 | } | ||
| 112 | } | ||