Matthieu Delahaye | 6fc3e12 | 2014-03-04 11:05:49 -0600 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) 2013 The Android Open Source Project |
| 3 | * |
| 4 | * Licensed under the Apache License, Version 2.0 (the "License"); |
| 5 | * you may not use this file except in compliance with the License. |
| 6 | * You may obtain a copy of the License at |
| 7 | * |
| 8 | * http://www.apache.org/licenses/LICENSE-2.0 |
| 9 | * |
| 10 | * Unless required by applicable law or agreed to in writing, software |
| 11 | * distributed under the License is distributed on an "AS IS" BASIS, |
| 12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 13 | * See the License for the specific language governing permissions and |
| 14 | * limitations under the License. |
| 15 | */ |
| 16 | |
| 17 | |
| 18 | #include "rsCpuIntrinsic.h" |
| 19 | #include "rsCpuIntrinsicInlines.h" |
| 20 | #include <sys/syscall.h> |
| 21 | #include "cutils/atomic.h" |
| 22 | |
| 23 | #ifdef RS_COMPATIBILITY_LIB |
| 24 | #include "rsCompatibilityLib.h" |
| 25 | #endif |
| 26 | |
| 27 | #ifndef RS_COMPATIBILITY_LIB |
| 28 | #include "hardware/gralloc.h" |
| 29 | #endif |
| 30 | |
| 31 | |
| 32 | #define INLINE inline |
| 33 | |
| 34 | #define MIN(x, y) (((x) < (y)) ? (x) : (y)) |
| 35 | #define MAX(x, y) (((x) > (y)) ? (x) : (y)) |
| 36 | |
| 37 | #define ROUND_POWER_OF_TWO(value, n) \ |
| 38 | (((value) + (1 << ((n) - 1))) >> (n)) |
| 39 | |
| 40 | |
| 41 | #define MI_SIZE_LOG2 3 |
| 42 | #define MI_BLOCK_SIZE_LOG2 (6 - MI_SIZE_LOG2) // 64 = 2^6 |
| 43 | |
| 44 | #define MI_SIZE (1 << MI_SIZE_LOG2) // pixels per mi-unit |
| 45 | #define MI_BLOCK_SIZE (1 << MI_BLOCK_SIZE_LOG2) // mi-units per max block |
| 46 | |
| 47 | #define MI_MASK (MI_BLOCK_SIZE - 1) |
| 48 | |
| 49 | #define SIMD_WIDTH 16 |
| 50 | #define MAX_LOOP_FILTER 63 |
| 51 | #define MAX_SEGMENTS 8 |
| 52 | #define MAX_REF_FRAMES 4 |
| 53 | #define MAX_MODE_LF_DELTAS 2 |
| 54 | #define MB_MODE_COUNT 14 |
| 55 | #define BLOCK_SIZES 13 |
| 56 | |
| 57 | |
| 58 | #if (defined(__GNUC__) && __GNUC__) || defined(__SUNPRO_C) |
| 59 | #define DECLARE_ALIGNED(n,typ,val) typ val __attribute__ ((aligned (n))) |
| 60 | #elif defined(_MSC_VER) |
| 61 | #define DECLARE_ALIGNED(n,typ,val) __declspec(align(n)) typ val |
| 62 | #else |
| 63 | #warning No alignment directives known for this compiler. |
| 64 | #define DECLARE_ALIGNED(n,typ,val) typ val |
| 65 | #endif |
| 66 | |
| 67 | // block transform size |
| 68 | typedef enum { |
| 69 | TX_4X4 = 0, // 4x4 transform |
| 70 | TX_8X8 = 1, // 8x8 transform |
| 71 | TX_16X16 = 2, // 16x16 transform |
| 72 | TX_32X32 = 3, // 32x32 transform |
| 73 | TX_SIZES |
| 74 | } TX_SIZE; |
| 75 | |
| 76 | typedef enum { |
| 77 | PLANE_TYPE_Y_WITH_DC, |
| 78 | PLANE_TYPE_UV, |
| 79 | } PLANE_TYPE; |
| 80 | |
| 81 | // This structure holds bit masks for all 8x8 blocks in a 64x64 region. |
| 82 | // Each 1 bit represents a position in which we want to apply the loop filter. |
| 83 | // Left_ entries refer to whether we apply a filter on the border to the |
| 84 | // left of the block. Above_ entries refer to whether or not to apply a |
| 85 | // filter on the above border. Int_ entries refer to whether or not to |
| 86 | // apply borders on the 4x4 edges within the 8x8 block that each bit |
| 87 | // represents. |
| 88 | // Since each transform is accompanied by a potentially different type of |
| 89 | // loop filter there is a different entry in the array for each transform size. |
| 90 | struct LoopFilterMask { |
| 91 | uint64_t left_y[4]; |
| 92 | uint64_t above_y[4]; |
| 93 | uint64_t int_4x4_y; |
| 94 | unsigned short left_uv[4]; |
| 95 | unsigned short above_uv[4]; |
| 96 | unsigned short int_4x4_uv; |
| 97 | unsigned char lfl_y[64]; |
| 98 | unsigned char lfl_uv[16]; |
| 99 | }; |
| 100 | |
| 101 | // Need to align this structure so when it is declared and |
| 102 | // passed it can be loaded into vector registers. |
| 103 | struct LoopFilterThresh { |
| 104 | DECLARE_ALIGNED(SIMD_WIDTH, uint8_t, mblim[SIMD_WIDTH]); |
| 105 | DECLARE_ALIGNED(SIMD_WIDTH, uint8_t, lim[SIMD_WIDTH]); |
| 106 | DECLARE_ALIGNED(SIMD_WIDTH, uint8_t, hev_thr[SIMD_WIDTH]); |
| 107 | }; |
| 108 | |
| 109 | struct LoopFilterInfoN { |
| 110 | LoopFilterThresh lfthr[MAX_LOOP_FILTER + 1]; |
| 111 | uint8_t lvl[MAX_SEGMENTS][MAX_REF_FRAMES][MAX_MODE_LF_DELTAS]; |
| 112 | uint8_t mode_lf_lut[MB_MODE_COUNT]; |
| 113 | }; |
| 114 | |
| 115 | struct BufferInfo { |
| 116 | int y_offset; |
| 117 | int u_offset; |
| 118 | int v_offset; |
| 119 | int y_stride; |
| 120 | int uv_stride; |
| 121 | }; |
| 122 | |
| 123 | #define MAX_CPU_CORES 32 |
| 124 | #define MAX_MB_PLANE 3 |
Matthieu Delahaye | adab849 | 2014-04-01 16:33:24 -0500 | [diff] [blame] | 125 | #define MAX_SB_ROW 64 |
Matthieu Delahaye | 6fc3e12 | 2014-03-04 11:05:49 -0600 | [diff] [blame] | 126 | |
| 127 | struct LoopFilterProgressChart { |
| 128 | int start; |
| 129 | int stop; |
| 130 | int num_planes; |
| 131 | int mi_rows; |
| 132 | int mi_cols; |
| 133 | BufferInfo buf_info; |
| 134 | uint8_t *buffer_alloc; |
| 135 | LoopFilterInfoN *lf_info; |
| 136 | LoopFilterMask *lfms; |
| 137 | |
| 138 | int wid; |
| 139 | int quit; |
| 140 | int doing; |
Matthieu Delahaye | adab849 | 2014-04-01 16:33:24 -0500 | [diff] [blame] | 141 | volatile int32_t chart[MAX_SB_ROW]; |
Matthieu Delahaye | 6fc3e12 | 2014-03-04 11:05:49 -0600 | [diff] [blame] | 142 | int32_t sb_row_pro; |
| 143 | pthread_t *tid; |
| 144 | pthread_mutex_t *mutex; |
| 145 | pthread_cond_t *start_cond; |
| 146 | pthread_mutex_t *hmutex; |
| 147 | pthread_cond_t *finish; |
| 148 | }; |
| 149 | |
| 150 | using namespace android; |
| 151 | using namespace android::renderscript; |
| 152 | |
| 153 | namespace android { |
| 154 | namespace renderscript { |
| 155 | |
| 156 | |
| 157 | class RsdCpuScriptIntrinsicLoopFilter : public RsdCpuScriptIntrinsic { |
| 158 | private: |
| 159 | LoopFilterProgressChart mPrch; |
| 160 | int mWorkerCount; |
| 161 | |
| 162 | public: |
| 163 | virtual void populateScript(Script *); |
| 164 | virtual void setGlobalVar(uint32_t slot, const void *data, size_t dataLength); |
| 165 | virtual void setGlobalObj(uint32_t slot, ObjectBase *data); |
| 166 | |
| 167 | virtual ~RsdCpuScriptIntrinsicLoopFilter(); |
| 168 | RsdCpuScriptIntrinsicLoopFilter(RsdCpuReferenceImpl *ctx, const Script *s, |
| 169 | const Element *e); |
| 170 | |
| 171 | protected: |
| 172 | ObjectBaseRef<Allocation> mLfInfo; |
| 173 | ObjectBaseRef<Allocation> mLfMasks; |
| 174 | ObjectBaseRef<Allocation> mFrameBuffer; |
| 175 | |
| 176 | void doLoopFilter(); |
Chris Wailes | 80ef693 | 2014-07-08 11:22:18 -0700 | [diff] [blame] | 177 | static void kernel(const RsExpandKernelParams *p, |
Matthieu Delahaye | 6fc3e12 | 2014-03-04 11:05:49 -0600 | [diff] [blame] | 178 | uint32_t xstart, uint32_t xend, |
Chris Wailes | 9ed7910 | 2014-07-25 15:53:28 -0700 | [diff] [blame] | 179 | uint32_t outstep); |
Matthieu Delahaye | 6fc3e12 | 2014-03-04 11:05:49 -0600 | [diff] [blame] | 180 | }; |
| 181 | |
| 182 | } |
| 183 | } |
| 184 | |
Chris Wailes | 80ef693 | 2014-07-08 11:22:18 -0700 | [diff] [blame] | 185 | void RsdCpuScriptIntrinsicLoopFilter::kernel(const RsExpandKernelParams *p, |
Matthieu Delahaye | 6fc3e12 | 2014-03-04 11:05:49 -0600 | [diff] [blame] | 186 | uint32_t xstart, uint32_t xend, |
Chris Wailes | 9ed7910 | 2014-07-25 15:53:28 -0700 | [diff] [blame] | 187 | uint32_t outstep) { |
Matthieu Delahaye | 6fc3e12 | 2014-03-04 11:05:49 -0600 | [diff] [blame] | 188 | RsdCpuScriptIntrinsicLoopFilter *cp = (RsdCpuScriptIntrinsicLoopFilter*)p->usr; |
| 189 | memset((void*)&cp->mPrch.chart, 0, sizeof(cp->mPrch.chart)); |
| 190 | cp->mPrch.chart[0] = 0x0fffffff; |
| 191 | cp->mPrch.sb_row_pro = 0; |
| 192 | cp->mPrch.doing = cp->mWorkerCount; |
| 193 | |
| 194 | int i = 0; |
| 195 | for (i = 0; i < cp->mWorkerCount; ++i) { |
| 196 | pthread_cond_signal(&cp->mPrch.start_cond[i]); |
| 197 | } |
| 198 | pthread_mutex_lock(cp->mPrch.hmutex); |
| 199 | if (cp->mPrch.doing) { |
| 200 | pthread_cond_wait(cp->mPrch.finish, cp->mPrch.hmutex); |
| 201 | } |
| 202 | pthread_mutex_unlock(cp->mPrch.hmutex); |
| 203 | } |
| 204 | |
| 205 | |
| 206 | void RsdCpuScriptIntrinsicLoopFilter::setGlobalVar(uint32_t slot, |
| 207 | const void *data, |
| 208 | size_t dataLength) { |
| 209 | rsAssert(slot >= 0 && slot < 2); |
| 210 | const int *dptr = (const int *)data; |
| 211 | switch (slot) { |
| 212 | case 0: |
| 213 | rsAssert(dataLength == sizeof(int) * 5); |
| 214 | mPrch.start = dptr[0]; |
| 215 | mPrch.stop = dptr[1]; |
| 216 | mPrch.num_planes = dptr[2]; |
| 217 | mPrch.mi_rows = dptr[3]; |
| 218 | mPrch.mi_cols = dptr[4]; |
| 219 | break; |
| 220 | case 1: |
| 221 | rsAssert(dataLength == sizeof(BufferInfo)); |
| 222 | mPrch.buf_info = *((BufferInfo*)data); |
| 223 | break; |
| 224 | default: |
| 225 | ALOGE("Non-exist global value slot: %d", slot); |
| 226 | rsAssert(0); |
| 227 | } |
| 228 | } |
| 229 | |
| 230 | void RsdCpuScriptIntrinsicLoopFilter::setGlobalObj(uint32_t slot, ObjectBase *data) { |
| 231 | rsAssert(slot > 1 && slot < 5); |
| 232 | if (slot == 2) { |
| 233 | mLfInfo.set(static_cast<Allocation *>(data)); |
| 234 | mPrch.lf_info = (LoopFilterInfoN *)mLfInfo->mHal.state.userProvidedPtr; |
| 235 | } else if (slot == 3) { |
| 236 | mLfMasks.set(static_cast<Allocation *>(data)); |
| 237 | mPrch.lfms = (LoopFilterMask *)mLfMasks->mHal.state.userProvidedPtr; |
| 238 | } else { |
| 239 | mFrameBuffer.set(static_cast<Allocation *>(data)); |
| 240 | mPrch.buffer_alloc = (uint8_t *)mFrameBuffer->mHal.state.userProvidedPtr; |
| 241 | } |
| 242 | } |
| 243 | |
| 244 | RsdCpuScriptIntrinsicLoopFilter::~RsdCpuScriptIntrinsicLoopFilter() { |
| 245 | android_atomic_inc(&mPrch.quit); |
| 246 | int i = 0; |
| 247 | for (i = 0; i < mWorkerCount; ++i) { |
| 248 | pthread_cond_signal(&mPrch.start_cond[i]); |
| 249 | } |
| 250 | for (i = 0; i < mWorkerCount; ++i) { |
| 251 | pthread_join(mPrch.tid[i], NULL); |
| 252 | } |
| 253 | free(mPrch.tid); |
| 254 | } |
| 255 | |
| 256 | void RsdCpuScriptIntrinsicLoopFilter::populateScript(Script *s) { |
| 257 | s->mHal.info.exportedVariableCount = 9; |
| 258 | s->mHal.info.exportedFunctionCount = 1; |
| 259 | } |
| 260 | |
| 261 | RsdCpuScriptImpl * rsdIntrinsic_LoopFilter(RsdCpuReferenceImpl *ctx, |
| 262 | const Script *s, const Element *e) { |
| 263 | return new RsdCpuScriptIntrinsicLoopFilter(ctx, s, e); |
| 264 | } |
| 265 | |
| 266 | extern "C" void vp9_lpf_vertical_16_c(uint8_t *s, int pitch, |
| 267 | const uint8_t *blimit, |
| 268 | const uint8_t *limit, |
| 269 | const uint8_t *thresh); |
| 270 | extern "C" void vp9_lpf_vertical_16_neon(uint8_t *s, int pitch, |
| 271 | const uint8_t *blimit, |
| 272 | const uint8_t *limit, |
| 273 | const uint8_t *thresh); |
| 274 | extern "C" void vp9_lpf_vertical_16_dual_c(uint8_t *s, int pitch, |
| 275 | const uint8_t *blimit, |
| 276 | const uint8_t *limit, |
| 277 | const uint8_t *thresh); |
| 278 | extern "C" void vp9_lpf_vertical_16_dual_neon(uint8_t *s, int pitch, |
| 279 | const uint8_t *blimit, |
| 280 | const uint8_t *limit, |
| 281 | const uint8_t *thresh); |
| 282 | extern "C" void vp9_lpf_vertical_8_c(uint8_t *s, int pitch, |
| 283 | const uint8_t *blimit, |
| 284 | const uint8_t *limit, |
| 285 | const uint8_t *thresh, |
| 286 | int count); |
| 287 | extern "C" void vp9_lpf_vertical_8_neon(uint8_t *s, int pitch, |
| 288 | const uint8_t *blimit, |
| 289 | const uint8_t *limit, |
| 290 | const uint8_t *thresh, int count); |
| 291 | extern "C" void vp9_lpf_vertical_8_dual_c(uint8_t *s, int pitch, |
| 292 | const uint8_t *blimit0, |
| 293 | const uint8_t *limit0, |
| 294 | const uint8_t *thresh0, |
| 295 | const uint8_t *blimit1, |
| 296 | const uint8_t *limit1, |
| 297 | const uint8_t *thresh1); |
| 298 | extern "C" void vp9_lpf_vertical_8_dual_neon(uint8_t *s, int pitch, |
| 299 | const uint8_t *blimit0, |
| 300 | const uint8_t *limit0, |
| 301 | const uint8_t *thresh0, |
| 302 | const uint8_t *blimit1, |
| 303 | const uint8_t *limit1, |
| 304 | const uint8_t *thresh1); |
| 305 | extern "C" void vp9_lpf_vertical_4_c(uint8_t *s, int pitch, const uint8_t *blimit, |
| 306 | const uint8_t *limit, const uint8_t *thresh, |
| 307 | int count); |
| 308 | extern "C" void vp9_lpf_vertical_4_neon(uint8_t *s, int pitch, |
| 309 | const uint8_t *blimit, |
| 310 | const uint8_t *limit, |
| 311 | const uint8_t *thresh, int count); |
| 312 | extern "C" void vp9_lpf_vertical_4_dual_c(uint8_t *s, int pitch, |
| 313 | const uint8_t *blimit0, |
| 314 | const uint8_t *limit0, |
| 315 | const uint8_t *thresh0, |
| 316 | const uint8_t *blimit1, |
| 317 | const uint8_t *limit1, |
| 318 | const uint8_t *thresh1); |
| 319 | extern "C" void vp9_lpf_vertical_4_dual_neon(uint8_t *s, int pitch, |
| 320 | const uint8_t *blimit0, |
| 321 | const uint8_t *limit0, |
| 322 | const uint8_t *thresh0, |
| 323 | const uint8_t *blimit1, |
| 324 | const uint8_t *limit1, |
| 325 | const uint8_t *thresh1); |
| 326 | extern "C" void vp9_lpf_horizontal_16_c(uint8_t *s, int pitch, |
| 327 | const uint8_t *blimit, |
| 328 | const uint8_t *limit, |
| 329 | const uint8_t *thresh, int count); |
| 330 | extern "C" void vp9_lpf_horizontal_16_neon(uint8_t *s, int pitch, |
| 331 | const uint8_t *blimit, |
| 332 | const uint8_t *limit, |
| 333 | const uint8_t *thresh, int count); |
| 334 | extern "C" void vp9_lpf_horizontal_8_c(uint8_t *s, int pitch, |
| 335 | const uint8_t *blimit, |
| 336 | const uint8_t *limit, |
| 337 | const uint8_t *thresh, int count); |
| 338 | extern "C" void vp9_lpf_horizontal_8_neon(uint8_t *s, int pitch, |
| 339 | const uint8_t *blimit, |
| 340 | const uint8_t *limit, |
| 341 | const uint8_t *thresh, int count); |
| 342 | extern "C" void vp9_lpf_horizontal_8_dual_c(uint8_t *s, int pitch, |
| 343 | const uint8_t *blimit0, |
| 344 | const uint8_t *limit0, |
| 345 | const uint8_t *thresh0, |
| 346 | const uint8_t *blimit1, |
| 347 | const uint8_t *limit1, |
| 348 | const uint8_t *thresh1); |
| 349 | extern "C" void vp9_lpf_horizontal_8_dual_neon(uint8_t *s, int pitch, |
| 350 | const uint8_t *blimit0, |
| 351 | const uint8_t *limit0, |
| 352 | const uint8_t *thresh0, |
| 353 | const uint8_t *blimit1, |
| 354 | const uint8_t *limit1, |
| 355 | const uint8_t *thresh1); |
| 356 | extern "C" void vp9_lpf_horizontal_4_c(uint8_t *s, int pitch, |
| 357 | const uint8_t *blimit, |
| 358 | const uint8_t *limit, |
| 359 | const uint8_t *thresh, int count); |
| 360 | extern "C" void vp9_lpf_horizontal_4_neon(uint8_t *s, int pitch, |
| 361 | const uint8_t *blimit, |
| 362 | const uint8_t *limit, |
| 363 | const uint8_t *thresh, int count); |
| 364 | extern "C" void vp9_lpf_horizontal_4_dual_c(uint8_t *s, int pitch, |
| 365 | const uint8_t *blimit0, |
| 366 | const uint8_t *limit0, |
| 367 | const uint8_t *thresh0, |
| 368 | const uint8_t *blimit1, |
| 369 | const uint8_t *limit1, |
| 370 | const uint8_t *thresh1); |
| 371 | extern "C" void vp9_lpf_horizontal_4_dual_neon(uint8_t *s, int pitch, |
| 372 | const uint8_t *blimit0, |
| 373 | const uint8_t *limit0, |
| 374 | const uint8_t *thresh0, |
| 375 | const uint8_t *blimit1, |
| 376 | const uint8_t *limit1, |
| 377 | const uint8_t *thresh1); |
| 378 | |
| 379 | |
Jason Sams | 074424a | 2014-05-22 13:30:03 -0700 | [diff] [blame] | 380 | // remove ARM64 statement when ARM64 asm available |
| 381 | #if defined(ARCH_ARM_USE_INTRINSICS) && !defined(ARCH_ARM64_USE_INTRINSICS) |
Matthieu Delahaye | 6fc3e12 | 2014-03-04 11:05:49 -0600 | [diff] [blame] | 382 | |
| 383 | #define vp9_lpf_vertical_16 vp9_lpf_vertical_16_neon |
| 384 | #define vp9_lpf_vertical_16_dual vp9_lpf_vertical_16_dual_neon |
| 385 | #define vp9_lpf_vertical_8 vp9_lpf_vertical_8_neon |
| 386 | #define vp9_lpf_vertical_8_dual vp9_lpf_vertical_8_dual_neon |
| 387 | #define vp9_lpf_vertical_4 vp9_lpf_vertical_4_neon |
| 388 | #define vp9_lpf_vertical_4_dual vp9_lpf_vertical_4_dual_neon |
| 389 | #define vp9_lpf_horizontal_16 vp9_lpf_horizontal_16_neon |
| 390 | #define vp9_lpf_horizontal_8 vp9_lpf_horizontal_8_neon |
| 391 | #define vp9_lpf_horizontal_8_dual vp9_lpf_horizontal_8_dual_neon |
| 392 | #define vp9_lpf_horizontal_4 vp9_lpf_horizontal_4_neon |
| 393 | #define vp9_lpf_horizontal_4_dual vp9_lpf_horizontal_4_dual_neon |
| 394 | |
| 395 | void vp9_lpf_horizontal_8_dual_neon(uint8_t *s, int p /* pitch */, |
| 396 | const uint8_t *blimit0, |
| 397 | const uint8_t *limit0, |
| 398 | const uint8_t *thresh0, |
| 399 | const uint8_t *blimit1, |
| 400 | const uint8_t *limit1, |
| 401 | const uint8_t *thresh1) { |
| 402 | vp9_lpf_horizontal_8(s, p, blimit0, limit0, thresh0, 1); |
| 403 | vp9_lpf_horizontal_8(s + 8, p, blimit1, limit1, thresh1, 1); |
| 404 | } |
| 405 | |
| 406 | void vp9_lpf_vertical_4_dual_neon(uint8_t *s, int p, |
| 407 | const uint8_t *blimit0, |
| 408 | const uint8_t *limit0, |
| 409 | const uint8_t *thresh0, |
| 410 | const uint8_t *blimit1, |
| 411 | const uint8_t *limit1, |
| 412 | const uint8_t *thresh1) { |
| 413 | vp9_lpf_vertical_4_neon(s, p, blimit0, limit0, thresh0, 1); |
| 414 | vp9_lpf_vertical_4_neon(s + 8 * p, p, blimit1, limit1, thresh1, 1); |
| 415 | } |
| 416 | |
| 417 | void vp9_lpf_vertical_8_dual_neon(uint8_t *s, int p, |
| 418 | const uint8_t *blimit0, |
| 419 | const uint8_t *limit0, |
| 420 | const uint8_t *thresh0, |
| 421 | const uint8_t *blimit1, |
| 422 | const uint8_t *limit1, |
| 423 | const uint8_t *thresh1) { |
| 424 | vp9_lpf_vertical_8_neon(s, p, blimit0, limit0, thresh0, 1); |
| 425 | vp9_lpf_vertical_8_neon(s + 8 * p, p, blimit1, limit1, thresh1, 1); |
| 426 | } |
| 427 | |
| 428 | void vp9_lpf_vertical_16_dual_neon(uint8_t *s, int p, |
| 429 | const uint8_t *blimit, |
| 430 | const uint8_t *limit, |
| 431 | const uint8_t *thresh) { |
| 432 | vp9_lpf_vertical_16_neon(s, p, blimit, limit, thresh); |
| 433 | vp9_lpf_vertical_16_neon(s + 8 * p, p, blimit, limit, thresh); |
| 434 | } |
| 435 | |
| 436 | #else |
| 437 | |
| 438 | #define vp9_lpf_vertical_16 vp9_lpf_vertical_16_c |
| 439 | #define vp9_lpf_vertical_16_dual vp9_lpf_vertical_16_dual_c |
| 440 | #define vp9_lpf_vertical_8 vp9_lpf_vertical_8_c |
| 441 | #define vp9_lpf_vertical_8_dual vp9_lpf_vertical_8_dual_c |
| 442 | #define vp9_lpf_vertical_4 vp9_lpf_vertical_4_c |
| 443 | #define vp9_lpf_vertical_4_dual vp9_lpf_vertical_4_dual_c |
| 444 | #define vp9_lpf_horizontal_16 vp9_lpf_horizontal_16_c |
| 445 | #define vp9_lpf_horizontal_8 vp9_lpf_horizontal_8_c |
| 446 | #define vp9_lpf_horizontal_8_dual vp9_lpf_horizontal_8_dual_c |
| 447 | #define vp9_lpf_horizontal_4 vp9_lpf_horizontal_4_c |
| 448 | #define vp9_lpf_horizontal_4_dual vp9_lpf_horizontal_4_dual_c |
| 449 | |
Jason Sams | 074424a | 2014-05-22 13:30:03 -0700 | [diff] [blame] | 450 | #endif // ARCH_ARM_USE_INTRINSICS && !ARCH_ARM64_USE_INTRINSICS |
Matthieu Delahaye | 6fc3e12 | 2014-03-04 11:05:49 -0600 | [diff] [blame] | 451 | |
| 452 | |
| 453 | |
| 454 | |
| 455 | static INLINE int8_t signed_char_clamp(int t) { |
| 456 | return (int8_t)clamp(t, -128, 127); |
| 457 | } |
| 458 | |
| 459 | // should we apply any filter at all: 11111111 yes, 00000000 no |
| 460 | static INLINE int8_t filter_mask(uint8_t limit, uint8_t blimit, |
| 461 | uint8_t p3, uint8_t p2, |
| 462 | uint8_t p1, uint8_t p0, |
| 463 | uint8_t q0, uint8_t q1, |
| 464 | uint8_t q2, uint8_t q3) { |
| 465 | int8_t mask = 0; |
| 466 | mask |= (abs(p3 - p2) > limit) * -1; |
| 467 | mask |= (abs(p2 - p1) > limit) * -1; |
| 468 | mask |= (abs(p1 - p0) > limit) * -1; |
| 469 | mask |= (abs(q1 - q0) > limit) * -1; |
| 470 | mask |= (abs(q2 - q1) > limit) * -1; |
| 471 | mask |= (abs(q3 - q2) > limit) * -1; |
| 472 | mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2 > blimit) * -1; |
| 473 | return ~mask; |
| 474 | } |
| 475 | |
| 476 | static INLINE int8_t flat_mask4(uint8_t thresh, |
| 477 | uint8_t p3, uint8_t p2, |
| 478 | uint8_t p1, uint8_t p0, |
| 479 | uint8_t q0, uint8_t q1, |
| 480 | uint8_t q2, uint8_t q3) { |
| 481 | int8_t mask = 0; |
| 482 | mask |= (abs(p1 - p0) > thresh) * -1; |
| 483 | mask |= (abs(q1 - q0) > thresh) * -1; |
| 484 | mask |= (abs(p2 - p0) > thresh) * -1; |
| 485 | mask |= (abs(q2 - q0) > thresh) * -1; |
| 486 | mask |= (abs(p3 - p0) > thresh) * -1; |
| 487 | mask |= (abs(q3 - q0) > thresh) * -1; |
| 488 | return ~mask; |
| 489 | } |
| 490 | |
| 491 | static INLINE int8_t flat_mask5(uint8_t thresh, |
| 492 | uint8_t p4, uint8_t p3, |
| 493 | uint8_t p2, uint8_t p1, |
| 494 | uint8_t p0, uint8_t q0, |
| 495 | uint8_t q1, uint8_t q2, |
| 496 | uint8_t q3, uint8_t q4) { |
| 497 | int8_t mask = ~flat_mask4(thresh, p3, p2, p1, p0, q0, q1, q2, q3); |
| 498 | mask |= (abs(p4 - p0) > thresh) * -1; |
| 499 | mask |= (abs(q4 - q0) > thresh) * -1; |
| 500 | return ~mask; |
| 501 | } |
| 502 | |
| 503 | // is there high edge variance internal edge: 11111111 yes, 00000000 no |
| 504 | static INLINE int8_t hev_mask(uint8_t thresh, uint8_t p1, uint8_t p0, |
| 505 | uint8_t q0, uint8_t q1) { |
| 506 | int8_t hev = 0; |
| 507 | hev |= (abs(p1 - p0) > thresh) * -1; |
| 508 | hev |= (abs(q1 - q0) > thresh) * -1; |
| 509 | return hev; |
| 510 | } |
| 511 | |
| 512 | static INLINE void filter4(int8_t mask, uint8_t thresh, uint8_t *op1, |
| 513 | uint8_t *op0, uint8_t *oq0, uint8_t *oq1) { |
| 514 | int8_t filter1, filter2; |
| 515 | |
| 516 | const int8_t ps1 = (int8_t) *op1 ^ 0x80; |
| 517 | const int8_t ps0 = (int8_t) *op0 ^ 0x80; |
| 518 | const int8_t qs0 = (int8_t) *oq0 ^ 0x80; |
| 519 | const int8_t qs1 = (int8_t) *oq1 ^ 0x80; |
| 520 | const uint8_t hev = hev_mask(thresh, *op1, *op0, *oq0, *oq1); |
| 521 | |
| 522 | // add outer taps if we have high edge variance |
| 523 | int8_t filter = signed_char_clamp(ps1 - qs1) & hev; |
| 524 | |
| 525 | // inner taps |
| 526 | filter = signed_char_clamp(filter + 3 * (qs0 - ps0)) & mask; |
| 527 | |
| 528 | // save bottom 3 bits so that we round one side +4 and the other +3 |
| 529 | // if it equals 4 we'll set to adjust by -1 to account for the fact |
| 530 | // we'd round 3 the other way |
| 531 | filter1 = signed_char_clamp(filter + 4) >> 3; |
| 532 | filter2 = signed_char_clamp(filter + 3) >> 3; |
| 533 | |
| 534 | *oq0 = signed_char_clamp(qs0 - filter1) ^ 0x80; |
| 535 | *op0 = signed_char_clamp(ps0 + filter2) ^ 0x80; |
| 536 | |
| 537 | // outer tap adjustments |
| 538 | filter = ROUND_POWER_OF_TWO(filter1, 1) & ~hev; |
| 539 | |
| 540 | *oq1 = signed_char_clamp(qs1 - filter) ^ 0x80; |
| 541 | *op1 = signed_char_clamp(ps1 + filter) ^ 0x80; |
| 542 | } |
| 543 | |
| 544 | void vp9_lpf_horizontal_4_c(uint8_t *s, int p /* pitch */, |
| 545 | const uint8_t *blimit, const uint8_t *limit, |
| 546 | const uint8_t *thresh, int count) { |
| 547 | int i; |
| 548 | |
| 549 | // loop filter designed to work using chars so that we can make maximum use |
| 550 | // of 8 bit simd instructions. |
| 551 | for (i = 0; i < 8 * count; ++i) { |
| 552 | const uint8_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p]; |
| 553 | const uint8_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p]; |
| 554 | const int8_t mask = filter_mask(*limit, *blimit, |
| 555 | p3, p2, p1, p0, q0, q1, q2, q3); |
| 556 | filter4(mask, *thresh, s - 2 * p, s - 1 * p, s, s + 1 * p); |
| 557 | ++s; |
| 558 | } |
| 559 | } |
| 560 | |
| 561 | void vp9_lpf_horizontal_4_dual_c(uint8_t *s, int p, const uint8_t *blimit0, |
| 562 | const uint8_t *limit0, const uint8_t *thresh0, |
| 563 | const uint8_t *blimit1, const uint8_t *limit1, |
| 564 | const uint8_t *thresh1) { |
| 565 | vp9_lpf_horizontal_4_c(s, p, blimit0, limit0, thresh0, 1); |
| 566 | vp9_lpf_horizontal_4_c(s + 8, p, blimit1, limit1, thresh1, 1); |
| 567 | } |
| 568 | |
| 569 | void vp9_lpf_vertical_4_c(uint8_t *s, int pitch, const uint8_t *blimit, |
| 570 | const uint8_t *limit, const uint8_t *thresh, |
| 571 | int count) { |
| 572 | int i; |
| 573 | |
| 574 | // loop filter designed to work using chars so that we can make maximum use |
| 575 | // of 8 bit simd instructions. |
| 576 | for (i = 0; i < 8 * count; ++i) { |
| 577 | const uint8_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1]; |
| 578 | const uint8_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3]; |
| 579 | const int8_t mask = filter_mask(*limit, *blimit, |
| 580 | p3, p2, p1, p0, q0, q1, q2, q3); |
| 581 | filter4(mask, *thresh, s - 2, s - 1, s, s + 1); |
| 582 | s += pitch; |
| 583 | } |
| 584 | } |
| 585 | |
| 586 | void vp9_lpf_vertical_4_dual_c(uint8_t *s, int pitch, const uint8_t *blimit0, |
| 587 | const uint8_t *limit0, const uint8_t *thresh0, |
| 588 | const uint8_t *blimit1, const uint8_t *limit1, |
| 589 | const uint8_t *thresh1) { |
| 590 | vp9_lpf_vertical_4_c(s, pitch, blimit0, limit0, thresh0, 1); |
| 591 | vp9_lpf_vertical_4_c(s + 8 * pitch, pitch, blimit1, limit1, thresh1, 1); |
| 592 | } |
| 593 | |
| 594 | static INLINE void filter8(int8_t mask, uint8_t thresh, uint8_t flat, |
| 595 | uint8_t *op3, uint8_t *op2, |
| 596 | uint8_t *op1, uint8_t *op0, |
| 597 | uint8_t *oq0, uint8_t *oq1, |
| 598 | uint8_t *oq2, uint8_t *oq3) { |
| 599 | if (flat && mask) { |
| 600 | const uint8_t p3 = *op3, p2 = *op2, p1 = *op1, p0 = *op0; |
| 601 | const uint8_t q0 = *oq0, q1 = *oq1, q2 = *oq2, q3 = *oq3; |
| 602 | |
| 603 | // 7-tap filter [1, 1, 1, 2, 1, 1, 1] |
| 604 | *op2 = ROUND_POWER_OF_TWO(p3 + p3 + p3 + 2 * p2 + p1 + p0 + q0, 3); |
| 605 | *op1 = ROUND_POWER_OF_TWO(p3 + p3 + p2 + 2 * p1 + p0 + q0 + q1, 3); |
| 606 | *op0 = ROUND_POWER_OF_TWO(p3 + p2 + p1 + 2 * p0 + q0 + q1 + q2, 3); |
| 607 | *oq0 = ROUND_POWER_OF_TWO(p2 + p1 + p0 + 2 * q0 + q1 + q2 + q3, 3); |
| 608 | *oq1 = ROUND_POWER_OF_TWO(p1 + p0 + q0 + 2 * q1 + q2 + q3 + q3, 3); |
| 609 | *oq2 = ROUND_POWER_OF_TWO(p0 + q0 + q1 + 2 * q2 + q3 + q3 + q3, 3); |
| 610 | } else { |
| 611 | filter4(mask, thresh, op1, op0, oq0, oq1); |
| 612 | } |
| 613 | } |
| 614 | |
| 615 | void vp9_lpf_horizontal_8_c(uint8_t *s, int p, const uint8_t *blimit, |
| 616 | const uint8_t *limit, const uint8_t *thresh, |
| 617 | int count) { |
| 618 | int i; |
| 619 | |
| 620 | // loop filter designed to work using chars so that we can make maximum use |
| 621 | // of 8 bit simd instructions. |
| 622 | for (i = 0; i < 8 * count; ++i) { |
| 623 | const uint8_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p]; |
| 624 | const uint8_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p]; |
| 625 | |
| 626 | const int8_t mask = filter_mask(*limit, *blimit, |
| 627 | p3, p2, p1, p0, q0, q1, q2, q3); |
| 628 | const int8_t flat = flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3); |
| 629 | filter8(mask, *thresh, flat, s - 4 * p, s - 3 * p, s - 2 * p, s - 1 * p, |
| 630 | s, s + 1 * p, s + 2 * p, s + 3 * p); |
| 631 | ++s; |
| 632 | } |
| 633 | } |
| 634 | |
| 635 | void vp9_lpf_horizontal_8_dual_c(uint8_t *s, int p, const uint8_t *blimit0, |
| 636 | const uint8_t *limit0, const uint8_t *thresh0, |
| 637 | const uint8_t *blimit1, const uint8_t *limit1, |
| 638 | const uint8_t *thresh1) { |
| 639 | vp9_lpf_horizontal_8_c(s, p, blimit0, limit0, thresh0, 1); |
| 640 | vp9_lpf_horizontal_8_c(s + 8, p, blimit1, limit1, thresh1, 1); |
| 641 | } |
| 642 | |
| 643 | void vp9_lpf_vertical_8_c(uint8_t *s, int pitch, const uint8_t *blimit, |
| 644 | const uint8_t *limit, const uint8_t *thresh, |
| 645 | int count) { |
| 646 | int i; |
| 647 | |
| 648 | for (i = 0; i < 8 * count; ++i) { |
| 649 | const uint8_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1]; |
| 650 | const uint8_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3]; |
| 651 | const int8_t mask = filter_mask(*limit, *blimit, |
| 652 | p3, p2, p1, p0, q0, q1, q2, q3); |
| 653 | const int8_t flat = flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3); |
| 654 | filter8(mask, *thresh, flat, s - 4, s - 3, s - 2, s - 1, |
| 655 | s, s + 1, s + 2, s + 3); |
| 656 | s += pitch; |
| 657 | } |
| 658 | } |
| 659 | |
| 660 | void vp9_lpf_vertical_8_dual_c(uint8_t *s, int pitch, const uint8_t *blimit0, |
| 661 | const uint8_t *limit0, const uint8_t *thresh0, |
| 662 | const uint8_t *blimit1, const uint8_t *limit1, |
| 663 | const uint8_t *thresh1) { |
| 664 | vp9_lpf_vertical_8_c(s, pitch, blimit0, limit0, thresh0, 1); |
| 665 | vp9_lpf_vertical_8_c(s + 8 * pitch, pitch, blimit1, limit1, thresh1, 1); |
| 666 | } |
| 667 | |
| 668 | static INLINE void filter16(int8_t mask, uint8_t thresh, |
| 669 | uint8_t flat, uint8_t flat2, |
| 670 | uint8_t *op7, uint8_t *op6, |
| 671 | uint8_t *op5, uint8_t *op4, |
| 672 | uint8_t *op3, uint8_t *op2, |
| 673 | uint8_t *op1, uint8_t *op0, |
| 674 | uint8_t *oq0, uint8_t *oq1, |
| 675 | uint8_t *oq2, uint8_t *oq3, |
| 676 | uint8_t *oq4, uint8_t *oq5, |
| 677 | uint8_t *oq6, uint8_t *oq7) { |
| 678 | if (flat2 && flat && mask) { |
| 679 | const uint8_t p7 = *op7, p6 = *op6, p5 = *op5, p4 = *op4, |
| 680 | p3 = *op3, p2 = *op2, p1 = *op1, p0 = *op0; |
| 681 | |
| 682 | const uint8_t q0 = *oq0, q1 = *oq1, q2 = *oq2, q3 = *oq3, |
| 683 | q4 = *oq4, q5 = *oq5, q6 = *oq6, q7 = *oq7; |
| 684 | |
| 685 | // 15-tap filter [1, 1, 1, 1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1] |
| 686 | *op6 = ROUND_POWER_OF_TWO(p7 * 7 + p6 * 2 + p5 + p4 + p3 + p2 + p1 + p0 + |
| 687 | q0, 4); |
| 688 | *op5 = ROUND_POWER_OF_TWO(p7 * 6 + p6 + p5 * 2 + p4 + p3 + p2 + p1 + p0 + |
| 689 | q0 + q1, 4); |
| 690 | *op4 = ROUND_POWER_OF_TWO(p7 * 5 + p6 + p5 + p4 * 2 + p3 + p2 + p1 + p0 + |
| 691 | q0 + q1 + q2, 4); |
| 692 | *op3 = ROUND_POWER_OF_TWO(p7 * 4 + p6 + p5 + p4 + p3 * 2 + p2 + p1 + p0 + |
| 693 | q0 + q1 + q2 + q3, 4); |
| 694 | *op2 = ROUND_POWER_OF_TWO(p7 * 3 + p6 + p5 + p4 + p3 + p2 * 2 + p1 + p0 + |
| 695 | q0 + q1 + q2 + q3 + q4, 4); |
| 696 | *op1 = ROUND_POWER_OF_TWO(p7 * 2 + p6 + p5 + p4 + p3 + p2 + p1 * 2 + p0 + |
| 697 | q0 + q1 + q2 + q3 + q4 + q5, 4); |
| 698 | *op0 = ROUND_POWER_OF_TWO(p7 + p6 + p5 + p4 + p3 + p2 + p1 + p0 * 2 + |
| 699 | q0 + q1 + q2 + q3 + q4 + q5 + q6, 4); |
| 700 | *oq0 = ROUND_POWER_OF_TWO(p6 + p5 + p4 + p3 + p2 + p1 + p0 + |
| 701 | q0 * 2 + q1 + q2 + q3 + q4 + q5 + q6 + q7, 4); |
| 702 | *oq1 = ROUND_POWER_OF_TWO(p5 + p4 + p3 + p2 + p1 + p0 + |
| 703 | q0 + q1 * 2 + q2 + q3 + q4 + q5 + q6 + q7 * 2, 4); |
| 704 | *oq2 = ROUND_POWER_OF_TWO(p4 + p3 + p2 + p1 + p0 + |
| 705 | q0 + q1 + q2 * 2 + q3 + q4 + q5 + q6 + q7 * 3, 4); |
| 706 | *oq3 = ROUND_POWER_OF_TWO(p3 + p2 + p1 + p0 + |
| 707 | q0 + q1 + q2 + q3 * 2 + q4 + q5 + q6 + q7 * 4, 4); |
| 708 | *oq4 = ROUND_POWER_OF_TWO(p2 + p1 + p0 + |
| 709 | q0 + q1 + q2 + q3 + q4 * 2 + q5 + q6 + q7 * 5, 4); |
| 710 | *oq5 = ROUND_POWER_OF_TWO(p1 + p0 + |
| 711 | q0 + q1 + q2 + q3 + q4 + q5 * 2 + q6 + q7 * 6, 4); |
| 712 | *oq6 = ROUND_POWER_OF_TWO(p0 + |
| 713 | q0 + q1 + q2 + q3 + q4 + q5 + q6 * 2 + q7 * 7, 4); |
| 714 | } else { |
| 715 | filter8(mask, thresh, flat, op3, op2, op1, op0, oq0, oq1, oq2, oq3); |
| 716 | } |
| 717 | } |
| 718 | |
| 719 | void vp9_lpf_horizontal_16_c(uint8_t *s, int p, const uint8_t *blimit, |
| 720 | const uint8_t *limit, const uint8_t *thresh, |
| 721 | int count) { |
| 722 | int i; |
| 723 | |
| 724 | // loop filter designed to work using chars so that we can make maximum use |
| 725 | // of 8 bit simd instructions. |
| 726 | for (i = 0; i < 8 * count; ++i) { |
| 727 | const uint8_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p]; |
| 728 | const uint8_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p]; |
| 729 | const int8_t mask = filter_mask(*limit, *blimit, |
| 730 | p3, p2, p1, p0, q0, q1, q2, q3); |
| 731 | const int8_t flat = flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3); |
| 732 | const int8_t flat2 = flat_mask5(1, s[-8 * p], s[-7 * p], s[-6 * p], s[-5 * p], p0, |
| 733 | q0, s[4 * p], s[5 * p], s[6 * p], s[7 * p]); |
| 734 | |
| 735 | filter16(mask, *thresh, flat, flat2, |
| 736 | s - 8 * p, s - 7 * p, s - 6 * p, s - 5 * p, |
| 737 | s - 4 * p, s - 3 * p, s - 2 * p, s - 1 * p, |
| 738 | s, s + 1 * p, s + 2 * p, s + 3 * p, |
| 739 | s + 4 * p, s + 5 * p, s + 6 * p, s + 7 * p); |
| 740 | ++s; |
| 741 | } |
| 742 | } |
| 743 | |
| 744 | static void mb_lpf_vertical_edge_w(uint8_t *s, int p, |
| 745 | const uint8_t *blimit, |
| 746 | const uint8_t *limit, |
| 747 | const uint8_t *thresh, |
| 748 | int count) { |
| 749 | int i; |
| 750 | |
| 751 | for (i = 0; i < count; ++i) { |
| 752 | const uint8_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1]; |
| 753 | const uint8_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3]; |
| 754 | const int8_t mask = filter_mask(*limit, *blimit, |
| 755 | p3, p2, p1, p0, q0, q1, q2, q3); |
| 756 | const int8_t flat = flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3); |
| 757 | const int8_t flat2 = flat_mask5(1, s[-8], s[-7], s[-6], s[-5], p0, |
| 758 | q0, s[4], s[5], s[6], s[7]); |
| 759 | |
| 760 | filter16(mask, *thresh, flat, flat2, |
| 761 | s - 8, s - 7, s - 6, s - 5, s - 4, s - 3, s - 2, s - 1, |
| 762 | s, s + 1, s + 2, s + 3, s + 4, s + 5, s + 6, s + 7); |
| 763 | s += p; |
| 764 | } |
| 765 | } |
| 766 | |
| 767 | void vp9_lpf_vertical_16_c(uint8_t *s, int p, const uint8_t *blimit, |
| 768 | const uint8_t *limit, const uint8_t *thresh) { |
| 769 | mb_lpf_vertical_edge_w(s, p, blimit, limit, thresh, 8); |
| 770 | } |
| 771 | |
| 772 | void vp9_lpf_vertical_16_dual_c(uint8_t *s, int p, const uint8_t *blimit, |
| 773 | const uint8_t *limit, const uint8_t *thresh) { |
| 774 | mb_lpf_vertical_edge_w(s, p, blimit, limit, thresh, 16); |
| 775 | } |
| 776 | |
| 777 | |
| 778 | static void filter_selectively_vert_row2(PLANE_TYPE plane_type, |
| 779 | uint8_t *s, int pitch, |
| 780 | unsigned int mask_16x16_l, |
| 781 | unsigned int mask_8x8_l, |
| 782 | unsigned int mask_4x4_l, |
| 783 | unsigned int mask_4x4_int_l, |
| 784 | const LoopFilterInfoN *lfi_n, |
| 785 | const uint8_t *lfl) { |
| 786 | const int mask_shift = plane_type ? 4 : 8; |
| 787 | const int mask_cutoff = plane_type ? 0xf : 0xff; |
| 788 | const int lfl_forward = plane_type ? 4 : 8; |
| 789 | |
| 790 | unsigned int mask_16x16_0 = mask_16x16_l & mask_cutoff; |
| 791 | unsigned int mask_8x8_0 = mask_8x8_l & mask_cutoff; |
| 792 | unsigned int mask_4x4_0 = mask_4x4_l & mask_cutoff; |
| 793 | unsigned int mask_4x4_int_0 = mask_4x4_int_l & mask_cutoff; |
| 794 | unsigned int mask_16x16_1 = (mask_16x16_l >> mask_shift) & mask_cutoff; |
| 795 | unsigned int mask_8x8_1 = (mask_8x8_l >> mask_shift) & mask_cutoff; |
| 796 | unsigned int mask_4x4_1 = (mask_4x4_l >> mask_shift) & mask_cutoff; |
| 797 | unsigned int mask_4x4_int_1 = (mask_4x4_int_l >> mask_shift) & mask_cutoff; |
| 798 | unsigned int mask; |
| 799 | |
| 800 | for (mask = mask_16x16_0 | mask_8x8_0 | mask_4x4_0 | mask_4x4_int_0 | |
| 801 | mask_16x16_1 | mask_8x8_1 | mask_4x4_1 | mask_4x4_int_1; |
| 802 | mask; mask >>= 1) { |
| 803 | const LoopFilterThresh *lfi0 = lfi_n->lfthr + *lfl; |
| 804 | const LoopFilterThresh *lfi1 = lfi_n->lfthr + *(lfl + lfl_forward); |
| 805 | |
| 806 | // TODO(yunqingwang): count in loopfilter functions should be removed. |
| 807 | if (mask & 1) { |
| 808 | if ((mask_16x16_0 | mask_16x16_1) & 1) { |
| 809 | if ((mask_16x16_0 & mask_16x16_1) & 1) { |
| 810 | vp9_lpf_vertical_16_dual(s, pitch, lfi0->mblim, lfi0->lim, |
| 811 | lfi0->hev_thr); |
| 812 | } else if (mask_16x16_0 & 1) { |
| 813 | vp9_lpf_vertical_16(s, pitch, lfi0->mblim, lfi0->lim, |
| 814 | lfi0->hev_thr); |
| 815 | } else { |
| 816 | vp9_lpf_vertical_16(s + 8 *pitch, pitch, lfi1->mblim, |
| 817 | lfi1->lim, lfi1->hev_thr); |
| 818 | } |
| 819 | } |
| 820 | |
| 821 | if ((mask_8x8_0 | mask_8x8_1) & 1) { |
| 822 | if ((mask_8x8_0 & mask_8x8_1) & 1) { |
| 823 | vp9_lpf_vertical_8_dual(s, pitch, lfi0->mblim, lfi0->lim, |
| 824 | lfi0->hev_thr, lfi1->mblim, lfi1->lim, |
| 825 | lfi1->hev_thr); |
| 826 | } else if (mask_8x8_0 & 1) { |
| 827 | vp9_lpf_vertical_8(s, pitch, lfi0->mblim, lfi0->lim, |
| 828 | lfi0->hev_thr, 1); |
| 829 | } else { |
| 830 | vp9_lpf_vertical_8(s + 8 * pitch, pitch, lfi1->mblim, lfi1->lim, |
| 831 | lfi1->hev_thr, 1); |
| 832 | } |
| 833 | } |
| 834 | |
| 835 | if ((mask_4x4_0 | mask_4x4_1) & 1) { |
| 836 | if ((mask_4x4_0 & mask_4x4_1) & 1) { |
| 837 | vp9_lpf_vertical_4_dual(s, pitch, lfi0->mblim, lfi0->lim, |
| 838 | lfi0->hev_thr, lfi1->mblim, lfi1->lim, |
| 839 | lfi1->hev_thr); |
| 840 | } else if (mask_4x4_0 & 1) { |
| 841 | vp9_lpf_vertical_4(s, pitch, lfi0->mblim, lfi0->lim, |
| 842 | lfi0->hev_thr, 1); |
| 843 | } else { |
| 844 | vp9_lpf_vertical_4(s + 8 * pitch, pitch, lfi1->mblim, lfi1->lim, |
| 845 | lfi1->hev_thr, 1); |
| 846 | } |
| 847 | } |
| 848 | |
| 849 | if ((mask_4x4_int_0 | mask_4x4_int_1) & 1) { |
| 850 | if ((mask_4x4_int_0 & mask_4x4_int_1) & 1) { |
| 851 | vp9_lpf_vertical_4_dual(s + 4, pitch, lfi0->mblim, lfi0->lim, |
| 852 | lfi0->hev_thr, lfi1->mblim, lfi1->lim, |
| 853 | lfi1->hev_thr); |
| 854 | } else if (mask_4x4_int_0 & 1) { |
| 855 | vp9_lpf_vertical_4(s + 4, pitch, lfi0->mblim, lfi0->lim, |
| 856 | lfi0->hev_thr, 1); |
| 857 | } else { |
| 858 | vp9_lpf_vertical_4(s + 8 * pitch + 4, pitch, lfi1->mblim, |
| 859 | lfi1->lim, lfi1->hev_thr, 1); |
| 860 | } |
| 861 | } |
| 862 | } |
| 863 | |
| 864 | s += 8; |
| 865 | lfl += 1; |
| 866 | mask_16x16_0 >>= 1; |
| 867 | mask_8x8_0 >>= 1; |
| 868 | mask_4x4_0 >>= 1; |
| 869 | mask_4x4_int_0 >>= 1; |
| 870 | mask_16x16_1 >>= 1; |
| 871 | mask_8x8_1 >>= 1; |
| 872 | mask_4x4_1 >>= 1; |
| 873 | mask_4x4_int_1 >>= 1; |
| 874 | } |
| 875 | } |
| 876 | |
| 877 | static void filter_selectively_horiz(uint8_t *s, int pitch, |
| 878 | unsigned int mask_16x16, |
| 879 | unsigned int mask_8x8, |
| 880 | unsigned int mask_4x4, |
| 881 | unsigned int mask_4x4_int, |
| 882 | const LoopFilterInfoN *lfi_n, |
| 883 | const uint8_t *lfl) { |
| 884 | unsigned int mask; |
| 885 | int count; |
| 886 | |
| 887 | for (mask = mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int; |
| 888 | mask; mask >>= count) { |
| 889 | const LoopFilterThresh *lfi = lfi_n->lfthr + *lfl; |
| 890 | |
| 891 | count = 1; |
| 892 | if (mask & 1) { |
| 893 | if (mask_16x16 & 1) { |
| 894 | if ((mask_16x16 & 3) == 3) { |
| 895 | vp9_lpf_horizontal_16(s, pitch, lfi->mblim, lfi->lim, |
| 896 | lfi->hev_thr, 2); |
| 897 | count = 2; |
| 898 | } else { |
| 899 | vp9_lpf_horizontal_16(s, pitch, lfi->mblim, lfi->lim, |
| 900 | lfi->hev_thr, 1); |
| 901 | } |
| 902 | } else if (mask_8x8 & 1) { |
| 903 | if ((mask_8x8 & 3) == 3) { |
| 904 | // Next block's thresholds |
| 905 | const LoopFilterThresh *lfin = lfi_n->lfthr + *(lfl + 1); |
| 906 | |
| 907 | vp9_lpf_horizontal_8_dual(s, pitch, lfi->mblim, lfi->lim, |
| 908 | lfi->hev_thr, lfin->mblim, lfin->lim, |
| 909 | lfin->hev_thr); |
| 910 | |
| 911 | if ((mask_4x4_int & 3) == 3) { |
| 912 | vp9_lpf_horizontal_4_dual(s + 4 * pitch, pitch, lfi->mblim, |
| 913 | lfi->lim, lfi->hev_thr, lfin->mblim, |
| 914 | lfin->lim, lfin->hev_thr); |
| 915 | } else { |
| 916 | if (mask_4x4_int & 1) |
| 917 | vp9_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, |
| 918 | lfi->lim, lfi->hev_thr, 1); |
| 919 | else if (mask_4x4_int & 2) |
| 920 | vp9_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim, |
| 921 | lfin->lim, lfin->hev_thr, 1); |
| 922 | } |
| 923 | count = 2; |
| 924 | } else { |
| 925 | vp9_lpf_horizontal_8(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr, 1); |
| 926 | |
| 927 | if (mask_4x4_int & 1) |
| 928 | vp9_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, |
| 929 | lfi->lim, lfi->hev_thr, 1); |
| 930 | } |
| 931 | } else if (mask_4x4 & 1) { |
| 932 | if ((mask_4x4 & 3) == 3) { |
| 933 | // Next block's thresholds |
| 934 | const LoopFilterThresh *lfin = lfi_n->lfthr + *(lfl + 1); |
| 935 | |
| 936 | vp9_lpf_horizontal_4_dual(s, pitch, lfi->mblim, lfi->lim, |
| 937 | lfi->hev_thr, lfin->mblim, lfin->lim, |
| 938 | lfin->hev_thr); |
| 939 | if ((mask_4x4_int & 3) == 3) { |
| 940 | vp9_lpf_horizontal_4_dual(s + 4 * pitch, pitch, lfi->mblim, |
| 941 | lfi->lim, lfi->hev_thr, lfin->mblim, |
| 942 | lfin->lim, lfin->hev_thr); |
| 943 | } else { |
| 944 | if (mask_4x4_int & 1) |
| 945 | vp9_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, |
| 946 | lfi->lim, lfi->hev_thr, 1); |
| 947 | else if (mask_4x4_int & 2) |
| 948 | vp9_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim, |
| 949 | lfin->lim, lfin->hev_thr, 1); |
| 950 | } |
| 951 | count = 2; |
| 952 | } else { |
| 953 | vp9_lpf_horizontal_4(s, pitch, lfi->mblim, lfi->lim, lfi->hev_thr, 1); |
| 954 | |
| 955 | if (mask_4x4_int & 1) |
| 956 | vp9_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim, |
| 957 | lfi->hev_thr, 1); |
| 958 | } |
| 959 | } else if (mask_4x4_int & 1) { |
| 960 | vp9_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim, |
| 961 | lfi->hev_thr, 1); |
| 962 | } |
| 963 | } |
| 964 | s += 8 * count; |
| 965 | lfl += count; |
| 966 | mask_16x16 >>= count; |
| 967 | mask_8x8 >>= count; |
| 968 | mask_4x4 >>= count; |
| 969 | mask_4x4_int >>= count; |
| 970 | } |
| 971 | } |
| 972 | |
| 973 | static void filter_block_plane_y(LoopFilterInfoN *lf_info, |
| 974 | LoopFilterMask *lfm, |
| 975 | int stride, |
| 976 | uint8_t *buf, |
| 977 | int mi_rows, |
| 978 | int mi_row) { |
| 979 | uint8_t* dst0 = buf; |
| 980 | int r; //, c; |
| 981 | |
| 982 | uint64_t mask_16x16 = lfm->left_y[TX_16X16]; |
| 983 | uint64_t mask_8x8 = lfm->left_y[TX_8X8]; |
| 984 | uint64_t mask_4x4 = lfm->left_y[TX_4X4]; |
| 985 | uint64_t mask_4x4_int = lfm->int_4x4_y; |
| 986 | |
| 987 | // Vertical pass: do 2 rows at one time |
| 988 | for (r = 0; r < MI_BLOCK_SIZE && mi_row + r < mi_rows; r += 2) { |
| 989 | unsigned int mask_16x16_l = mask_16x16 & 0xffff; |
| 990 | unsigned int mask_8x8_l = mask_8x8 & 0xffff; |
| 991 | unsigned int mask_4x4_l = mask_4x4 & 0xffff; |
| 992 | unsigned int mask_4x4_int_l = mask_4x4_int & 0xffff; |
| 993 | |
| 994 | // Disable filtering on the leftmost column |
| 995 | filter_selectively_vert_row2(PLANE_TYPE_Y_WITH_DC, buf, stride, |
| 996 | mask_16x16_l, mask_8x8_l, mask_4x4_l, mask_4x4_int_l, lf_info, |
| 997 | &lfm->lfl_y[r << 3]); |
| 998 | |
| 999 | buf += 16 * stride; |
| 1000 | mask_16x16 >>= 16; |
| 1001 | mask_8x8 >>= 16; |
| 1002 | mask_4x4 >>= 16; |
| 1003 | mask_4x4_int >>= 16; |
| 1004 | } |
| 1005 | |
| 1006 | // Horizontal pass |
| 1007 | buf = dst0; |
| 1008 | mask_16x16 = lfm->above_y[TX_16X16]; |
| 1009 | mask_8x8 = lfm->above_y[TX_8X8]; |
| 1010 | mask_4x4 = lfm->above_y[TX_4X4]; |
| 1011 | mask_4x4_int = lfm->int_4x4_y; |
| 1012 | |
| 1013 | for (r = 0; r < MI_BLOCK_SIZE && mi_row + r < mi_rows; r++) { |
| 1014 | unsigned int mask_16x16_r; |
| 1015 | unsigned int mask_8x8_r; |
| 1016 | unsigned int mask_4x4_r; |
| 1017 | |
| 1018 | if (mi_row + r == 0) { |
| 1019 | mask_16x16_r = 0; |
| 1020 | mask_8x8_r = 0; |
| 1021 | mask_4x4_r = 0; |
| 1022 | } else { |
| 1023 | mask_16x16_r = mask_16x16 & 0xff; |
| 1024 | mask_8x8_r = mask_8x8 & 0xff; |
| 1025 | mask_4x4_r = mask_4x4 & 0xff; |
| 1026 | } |
| 1027 | |
| 1028 | filter_selectively_horiz(buf, stride, mask_16x16_r, mask_8x8_r, |
| 1029 | mask_4x4_r, mask_4x4_int & 0xff, lf_info, &lfm->lfl_y[r << 3]); |
| 1030 | |
| 1031 | buf += 8 * stride; |
| 1032 | mask_16x16 >>= 8; |
| 1033 | mask_8x8 >>= 8; |
| 1034 | mask_4x4 >>= 8; |
| 1035 | mask_4x4_int >>= 8; |
| 1036 | } |
| 1037 | } |
| 1038 | |
| 1039 | static void filter_block_plane_uv(LoopFilterInfoN *lf_info, |
| 1040 | LoopFilterMask *lfm, |
| 1041 | int stride, |
| 1042 | uint8_t *buf, |
| 1043 | int mi_rows, |
| 1044 | int mi_row) { |
| 1045 | uint8_t* dst0 = buf; |
| 1046 | int r, c; |
| 1047 | |
| 1048 | uint16_t mask_16x16 = lfm->left_uv[TX_16X16]; |
| 1049 | uint16_t mask_8x8 = lfm->left_uv[TX_8X8]; |
| 1050 | uint16_t mask_4x4 = lfm->left_uv[TX_4X4]; |
| 1051 | uint16_t mask_4x4_int = lfm->int_4x4_uv; |
| 1052 | |
| 1053 | // Vertical pass: do 2 rows at one time |
| 1054 | for (r = 0; r < MI_BLOCK_SIZE && mi_row + r < mi_rows; r += 4) { |
| 1055 | |
| 1056 | for (c = 0; c < (MI_BLOCK_SIZE >> 1); c++) { |
| 1057 | lfm->lfl_uv[(r << 1) + c] = lfm->lfl_y[(r << 3) + (c << 1)]; |
| 1058 | lfm->lfl_uv[((r + 2) << 1) + c] = lfm->lfl_y[((r + 2) << 3) + (c << 1)]; |
| 1059 | } |
| 1060 | |
| 1061 | { |
| 1062 | unsigned int mask_16x16_l = mask_16x16 & 0xff; |
| 1063 | unsigned int mask_8x8_l = mask_8x8 & 0xff; |
| 1064 | unsigned int mask_4x4_l = mask_4x4 & 0xff; |
| 1065 | unsigned int mask_4x4_int_l = mask_4x4_int & 0xff; |
| 1066 | |
| 1067 | // Disable filtering on the leftmost column |
| 1068 | filter_selectively_vert_row2(PLANE_TYPE_UV, buf, stride, |
| 1069 | mask_16x16_l, mask_8x8_l, mask_4x4_l, mask_4x4_int_l, |
| 1070 | lf_info, &lfm->lfl_uv[r << 1]); |
| 1071 | |
| 1072 | buf += 16 * stride; |
| 1073 | mask_16x16 >>= 8; |
| 1074 | mask_8x8 >>= 8; |
| 1075 | mask_4x4 >>= 8; |
| 1076 | mask_4x4_int >>= 8; |
| 1077 | } |
| 1078 | } |
| 1079 | |
| 1080 | // Horizontal pass |
| 1081 | buf = dst0; |
| 1082 | mask_16x16 = lfm->above_uv[TX_16X16]; |
| 1083 | mask_8x8 = lfm->above_uv[TX_8X8]; |
| 1084 | mask_4x4 = lfm->above_uv[TX_4X4]; |
| 1085 | mask_4x4_int = lfm->int_4x4_uv; |
| 1086 | |
| 1087 | for (r = 0; r < MI_BLOCK_SIZE && mi_row + r < mi_rows; r += 2) { |
| 1088 | int skip_border_4x4_r = mi_row + r == mi_rows - 1; |
| 1089 | unsigned int mask_4x4_int_r = skip_border_4x4_r ? 0 : (mask_4x4_int & 0xf); |
| 1090 | unsigned int mask_16x16_r; |
| 1091 | unsigned int mask_8x8_r; |
| 1092 | unsigned int mask_4x4_r; |
| 1093 | |
| 1094 | if (mi_row + r == 0) { |
| 1095 | mask_16x16_r = 0; |
| 1096 | mask_8x8_r = 0; |
| 1097 | mask_4x4_r = 0; |
| 1098 | } else { |
| 1099 | mask_16x16_r = mask_16x16 & 0xf; |
| 1100 | mask_8x8_r = mask_8x8 & 0xf; |
| 1101 | mask_4x4_r = mask_4x4 & 0xf; |
| 1102 | } |
| 1103 | |
| 1104 | filter_selectively_horiz(buf, stride, mask_16x16_r, mask_8x8_r, |
| 1105 | mask_4x4_r, mask_4x4_int_r, lf_info, &lfm->lfl_uv[r << 1]); |
| 1106 | |
| 1107 | buf += 8 * stride; |
| 1108 | mask_16x16 >>= 4; |
| 1109 | mask_8x8 >>= 4; |
| 1110 | mask_4x4 >>= 4; |
| 1111 | mask_4x4_int >>= 4; |
| 1112 | } |
| 1113 | } |
| 1114 | |
| 1115 | static void *vp9_loop_filter_rows_work_proc(void *data) { |
| 1116 | LoopFilterProgressChart *param = (LoopFilterProgressChart *)data; |
| 1117 | int wid = android_atomic_inc(¶m->wid); |
| 1118 | int sb_row; |
| 1119 | int mi_row, mi_col; |
| 1120 | int lfm_idx; |
| 1121 | uint8_t *buf_start[MAX_MB_PLANE]; |
| 1122 | uint8_t *buf[MAX_MB_PLANE]; |
| 1123 | BufferInfo *buf_info = ¶m->buf_info; |
| 1124 | |
| 1125 | while (!android_atomic_release_load(¶m->quit)) { |
| 1126 | pthread_mutex_lock(¶m->mutex[wid]); |
| 1127 | pthread_cond_wait(¶m->start_cond[wid], ¶m->mutex[wid]); |
| 1128 | pthread_mutex_unlock(¶m->mutex[wid]); |
| 1129 | |
| 1130 | if (android_atomic_release_load(¶m->quit)) return NULL; |
| 1131 | |
| 1132 | buf_start[0] = param->buffer_alloc + buf_info->y_offset; |
| 1133 | buf_start[1] = param->buffer_alloc + buf_info->u_offset; |
| 1134 | buf_start[2] = param->buffer_alloc + buf_info->v_offset; |
| 1135 | sb_row = android_atomic_inc(¶m->sb_row_pro); |
| 1136 | mi_row = (sb_row * MI_BLOCK_SIZE) + param->start; |
| 1137 | |
| 1138 | while (mi_row < param->stop) { |
| 1139 | buf[0] = buf_start[0] + (mi_row * buf_info->y_stride << 3); |
| 1140 | buf[1] = buf_start[1] + (mi_row * buf_info->uv_stride << 2); |
| 1141 | buf[2] = buf_start[2] + (mi_row * buf_info->uv_stride << 2); |
| 1142 | lfm_idx = sb_row * ((param->mi_cols + 7) >> 3); |
| 1143 | for (mi_col = 0; mi_col < param->mi_cols; mi_col += MI_BLOCK_SIZE) { |
| 1144 | |
| 1145 | while (param->chart[sb_row+1] + 2 > android_atomic_release_load(¶m->chart[sb_row])) { |
| 1146 | usleep(1); |
| 1147 | } |
| 1148 | |
| 1149 | filter_block_plane_y(param->lf_info, param->lfms + lfm_idx, |
| 1150 | buf_info->y_stride, buf[0], param->mi_rows, |
| 1151 | mi_row); |
| 1152 | mi_col += MI_BLOCK_SIZE; |
| 1153 | if (mi_col < param->mi_cols) { |
| 1154 | lfm_idx++; |
| 1155 | buf[0] += MI_BLOCK_SIZE * MI_BLOCK_SIZE; |
| 1156 | filter_block_plane_y(param->lf_info, param->lfms + lfm_idx, |
| 1157 | buf_info->y_stride, buf[0], |
| 1158 | param->mi_rows, mi_row); |
| 1159 | } |
| 1160 | buf[0] += MI_BLOCK_SIZE * MI_BLOCK_SIZE; |
| 1161 | if (param->num_planes > 1) { |
| 1162 | lfm_idx--; |
| 1163 | filter_block_plane_uv(param->lf_info, param->lfms + lfm_idx, |
| 1164 | buf_info->uv_stride, buf[1], |
| 1165 | param->mi_rows, mi_row); |
| 1166 | filter_block_plane_uv(param->lf_info, param->lfms + lfm_idx, |
| 1167 | buf_info->uv_stride, buf[2], |
| 1168 | param->mi_rows, mi_row); |
| 1169 | if (mi_col < param->mi_cols) { |
| 1170 | lfm_idx++; |
| 1171 | buf[1] += MI_BLOCK_SIZE * MI_BLOCK_SIZE >> 1; |
| 1172 | buf[2] += MI_BLOCK_SIZE * MI_BLOCK_SIZE >> 1; |
| 1173 | filter_block_plane_uv(param->lf_info, |
| 1174 | param->lfms + lfm_idx, |
| 1175 | buf_info->uv_stride, buf[1], |
| 1176 | param->mi_rows, mi_row); |
| 1177 | filter_block_plane_uv(param->lf_info, |
| 1178 | param->lfms + lfm_idx, |
| 1179 | buf_info->uv_stride, buf[2], |
| 1180 | param->mi_rows, mi_row); |
| 1181 | } |
| 1182 | buf[1] += MI_BLOCK_SIZE * MI_BLOCK_SIZE >> 1; |
| 1183 | buf[2] += MI_BLOCK_SIZE * MI_BLOCK_SIZE >> 1; |
| 1184 | } |
| 1185 | lfm_idx++; |
| 1186 | android_atomic_inc(¶m->chart[sb_row+1]); |
| 1187 | } |
| 1188 | android_atomic_inc(¶m->chart[sb_row+1]); |
| 1189 | sb_row = android_atomic_inc(¶m->sb_row_pro); |
| 1190 | mi_row = (sb_row << 3) + param->start; |
| 1191 | } |
| 1192 | |
| 1193 | pthread_mutex_lock(param->hmutex); |
| 1194 | if ((--param->doing) == 0) |
| 1195 | pthread_cond_signal(param->finish); |
| 1196 | pthread_mutex_unlock(param->hmutex); |
| 1197 | } |
| 1198 | |
| 1199 | return NULL; |
| 1200 | } |
| 1201 | |
| 1202 | RsdCpuScriptIntrinsicLoopFilter::RsdCpuScriptIntrinsicLoopFilter( |
| 1203 | RsdCpuReferenceImpl *ctx, const Script *s, const Element *e) |
| 1204 | : RsdCpuScriptIntrinsic(ctx, s, e, RS_SCRIPT_INTRINSIC_ID_YUV_TO_RGB) { |
| 1205 | mRootPtr = &kernel; |
| 1206 | mWorkerCount = sysconf(_SC_NPROCESSORS_ONLN); |
| 1207 | mPrch.quit = 0; |
| 1208 | mPrch.wid = 0; |
| 1209 | mPrch.sb_row_pro = 0; |
| 1210 | mPrch.doing = mWorkerCount; |
| 1211 | int size = mWorkerCount * sizeof(pthread_t) + |
| 1212 | mWorkerCount * sizeof(pthread_mutex_t) + |
| 1213 | mWorkerCount * sizeof(pthread_cond_t) + |
| 1214 | sizeof(pthread_mutex_t) + sizeof(pthread_cond_t); |
| 1215 | uint8_t *ptr = (uint8_t *)malloc(size); |
| 1216 | rsAssert(ptr); |
| 1217 | mPrch.tid = (pthread_t *)ptr; |
| 1218 | mPrch.mutex = (pthread_mutex_t *) (mPrch.tid + mWorkerCount); |
| 1219 | mPrch.start_cond = (pthread_cond_t *) (mPrch.mutex + mWorkerCount); |
| 1220 | mPrch.hmutex = (pthread_mutex_t *) (mPrch.start_cond + mWorkerCount); |
| 1221 | mPrch.finish = (pthread_cond_t *) (mPrch.hmutex + 1); |
| 1222 | int i = 0; |
| 1223 | int rv = 0; |
| 1224 | pthread_mutex_init(mPrch.hmutex, NULL); |
| 1225 | pthread_cond_init(mPrch.finish, NULL); |
| 1226 | for (i = 0; i < mWorkerCount; ++i) { |
| 1227 | pthread_mutex_init(&mPrch.mutex[i], NULL); |
| 1228 | pthread_cond_init(&mPrch.start_cond[i], NULL); |
| 1229 | } |
| 1230 | for (i = 0; i < mWorkerCount; ++i) { |
| 1231 | rv = pthread_create(&mPrch.tid[i], NULL, &vp9_loop_filter_rows_work_proc, &mPrch); |
| 1232 | rsAssert(rv == 0); |
| 1233 | } |
| 1234 | } |