1 // Copyright 2011 Google Inc. All Rights Reserved. 2 // 3 // Use of this source code is governed by a BSD-style license 4 // that can be found in the COPYING file in the root of the source 5 // tree. An additional intellectual property rights grant can be found 6 // in the file PATENTS. All contributing project authors may 7 // be found in the AUTHORS file in the root of the source tree. 8 // ----------------------------------------------------------------------------- 9 // 10 // Speed-critical functions. 11 // 12 // Author: Skal (pascal.massimino (at) gmail.com) 13 14 #ifndef WEBP_DSP_DSP_H_ 15 #define WEBP_DSP_DSP_H_ 16 17 #ifdef HAVE_CONFIG_H 18 #include "../webp/config.h" 19 #endif 20 21 #include "../webp/types.h" 22 23 #ifdef __cplusplus 24 extern "C" { 25 #endif 26 27 //------------------------------------------------------------------------------ 28 // CPU detection 29 30 #if defined(__GNUC__) 31 # define LOCAL_GCC_VERSION ((__GNUC__ << 8) | __GNUC_MINOR__) 32 # define LOCAL_GCC_PREREQ(maj, min) \ 33 (LOCAL_GCC_VERSION >= (((maj) << 8) | (min))) 34 #else 35 # define LOCAL_GCC_PREREQ(maj, min) 0 36 #endif 37 38 #if defined(_MSC_VER) && _MSC_VER > 1310 && \ 39 (defined(_M_X64) || defined(_M_IX86)) 40 #define WEBP_MSC_SSE2 // Visual C++ SSE2 targets 41 #endif 42 43 // WEBP_HAVE_* are used to indicate the presence of the instruction set in dsp 44 // files without intrinsics, allowing the corresponding Init() to be called. 45 // Files containing intrinsics will need to be built targeting the instruction 46 // set so should succeed on one of the earlier tests. 47 #if defined(__SSE2__) || defined(WEBP_MSC_SSE2) || defined(WEBP_HAVE_SSE2) 48 #define WEBP_USE_SSE2 49 #endif 50 51 #if defined(__AVX2__) || defined(WEBP_HAVE_AVX2) 52 #define WEBP_USE_AVX2 53 #endif 54 55 #if defined(__ANDROID__) && defined(__ARM_ARCH_7A__) 56 #define WEBP_ANDROID_NEON // Android targets that might support NEON 57 #endif 58 59 #if defined(__ARM_NEON__) || defined(WEBP_ANDROID_NEON) || defined(__aarch64__) 60 #define WEBP_USE_NEON 61 #endif 62 63 #if defined(__mips__) && !defined(__mips64) && (__mips_isa_rev < 6) 64 #define WEBP_USE_MIPS32 65 #endif 66 67 typedef enum { 68 kSSE2, 69 kSSE3, 70 kAVX, 71 kAVX2, 72 kNEON, 73 kMIPS32 74 } CPUFeature; 75 // returns true if the CPU supports the feature. 76 typedef int (*VP8CPUInfo)(CPUFeature feature); 77 extern VP8CPUInfo VP8GetCPUInfo; 78 79 //------------------------------------------------------------------------------ 80 // Encoding 81 82 // Transforms 83 // VP8Idct: Does one of two inverse transforms. If do_two is set, the transforms 84 // will be done for (ref, in, dst) and (ref + 4, in + 16, dst + 4). 85 typedef void (*VP8Idct)(const uint8_t* ref, const int16_t* in, uint8_t* dst, 86 int do_two); 87 typedef void (*VP8Fdct)(const uint8_t* src, const uint8_t* ref, int16_t* out); 88 typedef void (*VP8WHT)(const int16_t* in, int16_t* out); 89 extern VP8Idct VP8ITransform; 90 extern VP8Fdct VP8FTransform; 91 extern VP8WHT VP8FTransformWHT; 92 // Predictions 93 // *dst is the destination block. *top and *left can be NULL. 94 typedef void (*VP8IntraPreds)(uint8_t *dst, const uint8_t* left, 95 const uint8_t* top); 96 typedef void (*VP8Intra4Preds)(uint8_t *dst, const uint8_t* top); 97 extern VP8Intra4Preds VP8EncPredLuma4; 98 extern VP8IntraPreds VP8EncPredLuma16; 99 extern VP8IntraPreds VP8EncPredChroma8; 100 101 typedef int (*VP8Metric)(const uint8_t* pix, const uint8_t* ref); 102 extern VP8Metric VP8SSE16x16, VP8SSE16x8, VP8SSE8x8, VP8SSE4x4; 103 typedef int (*VP8WMetric)(const uint8_t* pix, const uint8_t* ref, 104 const uint16_t* const weights); 105 extern VP8WMetric VP8TDisto4x4, VP8TDisto16x16; 106 107 typedef void (*VP8BlockCopy)(const uint8_t* src, uint8_t* dst); 108 extern VP8BlockCopy VP8Copy4x4; 109 // Quantization 110 struct VP8Matrix; // forward declaration 111 typedef int (*VP8QuantizeBlock)(int16_t in[16], int16_t out[16], 112 const struct VP8Matrix* const mtx); 113 extern VP8QuantizeBlock VP8EncQuantizeBlock; 114 115 // specific to 2nd transform: 116 typedef int (*VP8QuantizeBlockWHT)(int16_t in[16], int16_t out[16], 117 const struct VP8Matrix* const mtx); 118 extern VP8QuantizeBlockWHT VP8EncQuantizeBlockWHT; 119 120 // Collect histogram for susceptibility calculation and accumulate in histo[]. 121 struct VP8Histogram; 122 typedef void (*VP8CHisto)(const uint8_t* ref, const uint8_t* pred, 123 int start_block, int end_block, 124 struct VP8Histogram* const histo); 125 extern const int VP8DspScan[16 + 4 + 4]; 126 extern VP8CHisto VP8CollectHistogram; 127 128 void VP8EncDspInit(void); // must be called before using any of the above 129 130 //------------------------------------------------------------------------------ 131 // Decoding 132 133 typedef void (*VP8DecIdct)(const int16_t* coeffs, uint8_t* dst); 134 // when doing two transforms, coeffs is actually int16_t[2][16]. 135 typedef void (*VP8DecIdct2)(const int16_t* coeffs, uint8_t* dst, int do_two); 136 extern VP8DecIdct2 VP8Transform; 137 extern VP8DecIdct VP8TransformAC3; 138 extern VP8DecIdct VP8TransformUV; 139 extern VP8DecIdct VP8TransformDC; 140 extern VP8DecIdct VP8TransformDCUV; 141 extern VP8WHT VP8TransformWHT; 142 143 // *dst is the destination block, with stride BPS. Boundary samples are 144 // assumed accessible when needed. 145 typedef void (*VP8PredFunc)(uint8_t* dst); 146 extern const VP8PredFunc VP8PredLuma16[/* NUM_B_DC_MODES */]; 147 extern const VP8PredFunc VP8PredChroma8[/* NUM_B_DC_MODES */]; 148 extern const VP8PredFunc VP8PredLuma4[/* NUM_BMODES */]; 149 150 // clipping tables (for filtering) 151 extern const int8_t* const VP8ksclip1; // clips [-1020, 1020] to [-128, 127] 152 extern const int8_t* const VP8ksclip2; // clips [-112, 112] to [-16, 15] 153 extern const uint8_t* const VP8kclip1; // clips [-255,511] to [0,255] 154 extern const uint8_t* const VP8kabs0; // abs(x) for x in [-255,255] 155 void VP8InitClipTables(void); // must be called first 156 157 // simple filter (only for luma) 158 typedef void (*VP8SimpleFilterFunc)(uint8_t* p, int stride, int thresh); 159 extern VP8SimpleFilterFunc VP8SimpleVFilter16; 160 extern VP8SimpleFilterFunc VP8SimpleHFilter16; 161 extern VP8SimpleFilterFunc VP8SimpleVFilter16i; // filter 3 inner edges 162 extern VP8SimpleFilterFunc VP8SimpleHFilter16i; 163 164 // regular filter (on both macroblock edges and inner edges) 165 typedef void (*VP8LumaFilterFunc)(uint8_t* luma, int stride, 166 int thresh, int ithresh, int hev_t); 167 typedef void (*VP8ChromaFilterFunc)(uint8_t* u, uint8_t* v, int stride, 168 int thresh, int ithresh, int hev_t); 169 // on outer edge 170 extern VP8LumaFilterFunc VP8VFilter16; 171 extern VP8LumaFilterFunc VP8HFilter16; 172 extern VP8ChromaFilterFunc VP8VFilter8; 173 extern VP8ChromaFilterFunc VP8HFilter8; 174 175 // on inner edge 176 extern VP8LumaFilterFunc VP8VFilter16i; // filtering 3 inner edges altogether 177 extern VP8LumaFilterFunc VP8HFilter16i; 178 extern VP8ChromaFilterFunc VP8VFilter8i; // filtering u and v altogether 179 extern VP8ChromaFilterFunc VP8HFilter8i; 180 181 // must be called before anything using the above 182 void VP8DspInit(void); 183 184 //------------------------------------------------------------------------------ 185 // WebP I/O 186 187 #define FANCY_UPSAMPLING // undefined to remove fancy upsampling support 188 189 // Convert a pair of y/u/v lines together to the output rgb/a colorspace. 190 // bottom_y can be NULL if only one line of output is needed (at top/bottom). 191 typedef void (*WebPUpsampleLinePairFunc)( 192 const uint8_t* top_y, const uint8_t* bottom_y, 193 const uint8_t* top_u, const uint8_t* top_v, 194 const uint8_t* cur_u, const uint8_t* cur_v, 195 uint8_t* top_dst, uint8_t* bottom_dst, int len); 196 197 #ifdef FANCY_UPSAMPLING 198 199 // Fancy upsampling functions to convert YUV to RGB(A) modes 200 extern WebPUpsampleLinePairFunc WebPUpsamplers[/* MODE_LAST */]; 201 202 #endif // FANCY_UPSAMPLING 203 204 // Per-row point-sampling methods. 205 typedef void (*WebPSamplerRowFunc)(const uint8_t* y, 206 const uint8_t* u, const uint8_t* v, 207 uint8_t* dst, int len); 208 // Generic function to apply 'WebPSamplerRowFunc' to the whole plane: 209 void WebPSamplerProcessPlane(const uint8_t* y, int y_stride, 210 const uint8_t* u, const uint8_t* v, int uv_stride, 211 uint8_t* dst, int dst_stride, 212 int width, int height, WebPSamplerRowFunc func); 213 214 // Sampling functions to convert rows of YUV to RGB(A) 215 extern WebPSamplerRowFunc WebPSamplers[/* MODE_LAST */]; 216 217 // General function for converting two lines of ARGB or RGBA. 218 // 'alpha_is_last' should be true if 0xff000000 is stored in memory as 219 // as 0x00, 0x00, 0x00, 0xff (little endian). 220 WebPUpsampleLinePairFunc WebPGetLinePairConverter(int alpha_is_last); 221 222 // YUV444->RGB converters 223 typedef void (*WebPYUV444Converter)(const uint8_t* y, 224 const uint8_t* u, const uint8_t* v, 225 uint8_t* dst, int len); 226 227 extern const WebPYUV444Converter WebPYUV444Converters[/* MODE_LAST */]; 228 229 // Must be called before using the WebPUpsamplers[] (and for premultiplied 230 // colorspaces like rgbA, rgbA4444, etc) 231 void WebPInitUpsamplers(void); 232 // Must be called before using WebPSamplers[] 233 void WebPInitSamplers(void); 234 235 //------------------------------------------------------------------------------ 236 // Utilities for processing transparent channel. 237 238 // Apply alpha pre-multiply on an rgba, bgra or argb plane of size w * h. 239 // alpha_first should be 0 for argb, 1 for rgba or bgra (where alpha is last). 240 extern void (*WebPApplyAlphaMultiply)( 241 uint8_t* rgba, int alpha_first, int w, int h, int stride); 242 243 // Same, buf specifically for RGBA4444 format 244 extern void (*WebPApplyAlphaMultiply4444)( 245 uint8_t* rgba4444, int w, int h, int stride); 246 247 // Pre-Multiply operation transforms x into x * A / 255 (where x=Y,R,G or B). 248 // Un-Multiply operation transforms x into x * 255 / A. 249 250 // Pre-Multiply or Un-Multiply (if 'inverse' is true) argb values in a row. 251 extern void (*WebPMultARGBRow)(uint32_t* const ptr, int width, int inverse); 252 253 // Same a WebPMultARGBRow(), but for several rows. 254 void WebPMultARGBRows(uint8_t* ptr, int stride, int width, int num_rows, 255 int inverse); 256 257 // Same for a row of single values, with side alpha values. 258 extern void (*WebPMultRow)(uint8_t* const ptr, const uint8_t* const alpha, 259 int width, int inverse); 260 261 // Same a WebPMultRow(), but for several 'num_rows' rows. 262 void WebPMultRows(uint8_t* ptr, int stride, 263 const uint8_t* alpha, int alpha_stride, 264 int width, int num_rows, int inverse); 265 266 // To be called first before using the above. 267 void WebPInitAlphaProcessing(void); 268 269 #ifdef __cplusplus 270 } // extern "C" 271 #endif 272 273 #endif /* WEBP_DSP_DSP_H_ */ 274
