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 #include "webp/types.h" 18 19 #if defined(__cplusplus) || defined(c_plusplus) 20 extern "C" { 21 #endif 22 23 //------------------------------------------------------------------------------ 24 // CPU detection 25 26 #if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86)) 27 #define WEBP_MSC_SSE2 // Visual C++ SSE2 targets 28 #endif 29 30 #if defined(__SSE2__) || defined(WEBP_MSC_SSE2) 31 #define WEBP_USE_SSE2 32 #endif 33 34 #if defined(__ANDROID__) && defined(__ARM_ARCH_7A__) 35 #define WEBP_ANDROID_NEON // Android targets that might support NEON 36 #endif 37 38 #if defined(__ARM_NEON__) 39 #define WEBP_USE_NEON 40 #endif 41 42 typedef enum { 43 kSSE2, 44 kSSE3, 45 kNEON 46 } CPUFeature; 47 // returns true if the CPU supports the feature. 48 typedef int (*VP8CPUInfo)(CPUFeature feature); 49 extern VP8CPUInfo VP8GetCPUInfo; 50 51 //------------------------------------------------------------------------------ 52 // Encoding 53 54 // Transforms 55 // VP8Idct: Does one of two inverse transforms. If do_two is set, the transforms 56 // will be done for (ref, in, dst) and (ref + 4, in + 16, dst + 4). 57 typedef void (*VP8Idct)(const uint8_t* ref, const int16_t* in, uint8_t* dst, 58 int do_two); 59 typedef void (*VP8Fdct)(const uint8_t* src, const uint8_t* ref, int16_t* out); 60 typedef void (*VP8WHT)(const int16_t* in, int16_t* out); 61 extern VP8Idct VP8ITransform; 62 extern VP8Fdct VP8FTransform; 63 extern VP8WHT VP8ITransformWHT; 64 extern VP8WHT VP8FTransformWHT; 65 // Predictions 66 // *dst is the destination block. *top and *left can be NULL. 67 typedef void (*VP8IntraPreds)(uint8_t *dst, const uint8_t* left, 68 const uint8_t* top); 69 typedef void (*VP8Intra4Preds)(uint8_t *dst, const uint8_t* top); 70 extern VP8Intra4Preds VP8EncPredLuma4; 71 extern VP8IntraPreds VP8EncPredLuma16; 72 extern VP8IntraPreds VP8EncPredChroma8; 73 74 typedef int (*VP8Metric)(const uint8_t* pix, const uint8_t* ref); 75 extern VP8Metric VP8SSE16x16, VP8SSE16x8, VP8SSE8x8, VP8SSE4x4; 76 typedef int (*VP8WMetric)(const uint8_t* pix, const uint8_t* ref, 77 const uint16_t* const weights); 78 extern VP8WMetric VP8TDisto4x4, VP8TDisto16x16; 79 80 typedef void (*VP8BlockCopy)(const uint8_t* src, uint8_t* dst); 81 extern VP8BlockCopy VP8Copy4x4; 82 // Quantization 83 struct VP8Matrix; // forward declaration 84 typedef int (*VP8QuantizeBlock)(int16_t in[16], int16_t out[16], 85 int n, const struct VP8Matrix* const mtx); 86 extern VP8QuantizeBlock VP8EncQuantizeBlock; 87 88 // Collect histogram for susceptibility calculation and accumulate in histo[]. 89 struct VP8Histogram; 90 typedef void (*VP8CHisto)(const uint8_t* ref, const uint8_t* pred, 91 int start_block, int end_block, 92 struct VP8Histogram* const histo); 93 extern const int VP8DspScan[16 + 4 + 4]; 94 extern VP8CHisto VP8CollectHistogram; 95 96 void VP8EncDspInit(void); // must be called before using any of the above 97 98 //------------------------------------------------------------------------------ 99 // Decoding 100 101 typedef void (*VP8DecIdct)(const int16_t* coeffs, uint8_t* dst); 102 // when doing two transforms, coeffs is actually int16_t[2][16]. 103 typedef void (*VP8DecIdct2)(const int16_t* coeffs, uint8_t* dst, int do_two); 104 extern VP8DecIdct2 VP8Transform; 105 extern VP8DecIdct VP8TransformUV; 106 extern VP8DecIdct VP8TransformDC; 107 extern VP8DecIdct VP8TransformDCUV; 108 extern VP8WHT VP8TransformWHT; 109 110 // *dst is the destination block, with stride BPS. Boundary samples are 111 // assumed accessible when needed. 112 typedef void (*VP8PredFunc)(uint8_t* dst); 113 extern const VP8PredFunc VP8PredLuma16[/* NUM_B_DC_MODES */]; 114 extern const VP8PredFunc VP8PredChroma8[/* NUM_B_DC_MODES */]; 115 extern const VP8PredFunc VP8PredLuma4[/* NUM_BMODES */]; 116 117 // simple filter (only for luma) 118 typedef void (*VP8SimpleFilterFunc)(uint8_t* p, int stride, int thresh); 119 extern VP8SimpleFilterFunc VP8SimpleVFilter16; 120 extern VP8SimpleFilterFunc VP8SimpleHFilter16; 121 extern VP8SimpleFilterFunc VP8SimpleVFilter16i; // filter 3 inner edges 122 extern VP8SimpleFilterFunc VP8SimpleHFilter16i; 123 124 // regular filter (on both macroblock edges and inner edges) 125 typedef void (*VP8LumaFilterFunc)(uint8_t* luma, int stride, 126 int thresh, int ithresh, int hev_t); 127 typedef void (*VP8ChromaFilterFunc)(uint8_t* u, uint8_t* v, int stride, 128 int thresh, int ithresh, int hev_t); 129 // on outer edge 130 extern VP8LumaFilterFunc VP8VFilter16; 131 extern VP8LumaFilterFunc VP8HFilter16; 132 extern VP8ChromaFilterFunc VP8VFilter8; 133 extern VP8ChromaFilterFunc VP8HFilter8; 134 135 // on inner edge 136 extern VP8LumaFilterFunc VP8VFilter16i; // filtering 3 inner edges altogether 137 extern VP8LumaFilterFunc VP8HFilter16i; 138 extern VP8ChromaFilterFunc VP8VFilter8i; // filtering u and v altogether 139 extern VP8ChromaFilterFunc VP8HFilter8i; 140 141 // must be called before anything using the above 142 void VP8DspInit(void); 143 144 //------------------------------------------------------------------------------ 145 // WebP I/O 146 147 #define FANCY_UPSAMPLING // undefined to remove fancy upsampling support 148 149 typedef void (*WebPUpsampleLinePairFunc)( 150 const uint8_t* top_y, const uint8_t* bottom_y, 151 const uint8_t* top_u, const uint8_t* top_v, 152 const uint8_t* cur_u, const uint8_t* cur_v, 153 uint8_t* top_dst, uint8_t* bottom_dst, int len); 154 155 #ifdef FANCY_UPSAMPLING 156 157 // Fancy upsampling functions to convert YUV to RGB(A) modes 158 extern WebPUpsampleLinePairFunc WebPUpsamplers[/* MODE_LAST */]; 159 160 // Initializes SSE2 version of the fancy upsamplers. 161 void WebPInitUpsamplersSSE2(void); 162 163 // NEON version 164 void WebPInitUpsamplersNEON(void); 165 166 #endif // FANCY_UPSAMPLING 167 168 // Point-sampling methods. 169 typedef void (*WebPSampleLinePairFunc)( 170 const uint8_t* top_y, const uint8_t* bottom_y, 171 const uint8_t* u, const uint8_t* v, 172 uint8_t* top_dst, uint8_t* bottom_dst, int len); 173 174 extern const WebPSampleLinePairFunc WebPSamplers[/* MODE_LAST */]; 175 176 // General function for converting two lines of ARGB or RGBA. 177 // 'alpha_is_last' should be true if 0xff000000 is stored in memory as 178 // as 0x00, 0x00, 0x00, 0xff (little endian). 179 WebPUpsampleLinePairFunc WebPGetLinePairConverter(int alpha_is_last); 180 181 // YUV444->RGB converters 182 typedef void (*WebPYUV444Converter)(const uint8_t* y, 183 const uint8_t* u, const uint8_t* v, 184 uint8_t* dst, int len); 185 186 extern const WebPYUV444Converter WebPYUV444Converters[/* MODE_LAST */]; 187 188 // Main function to be called 189 void WebPInitUpsamplers(void); 190 191 //------------------------------------------------------------------------------ 192 // Pre-multiply planes with alpha values 193 194 // Apply alpha pre-multiply on an rgba, bgra or argb plane of size w * h. 195 // alpha_first should be 0 for argb, 1 for rgba or bgra (where alpha is last). 196 extern void (*WebPApplyAlphaMultiply)( 197 uint8_t* rgba, int alpha_first, int w, int h, int stride); 198 199 // Same, buf specifically for RGBA4444 format 200 extern void (*WebPApplyAlphaMultiply4444)( 201 uint8_t* rgba4444, int w, int h, int stride); 202 203 // To be called first before using the above. 204 void WebPInitPremultiply(void); 205 206 void WebPInitPremultiplySSE2(void); // should not be called directly. 207 void WebPInitPremultiplyNEON(void); 208 209 //------------------------------------------------------------------------------ 210 211 #if defined(__cplusplus) || defined(c_plusplus) 212 } // extern "C" 213 #endif 214 215 #endif /* WEBP_DSP_DSP_H_ */ 216
