dsp.h revision 228b1b1f024974d7832b51a3f266e5edc9110c02
1// Copyright 2011 Google Inc. All Rights Reserved. 2// 3// This code is licensed under the same terms as WebM: 4// Software License Agreement: http://www.webmproject.org/license/software/ 5// Additional IP Rights Grant: http://www.webmproject.org/license/additional/ 6// ----------------------------------------------------------------------------- 7// 8// Speed-critical functions. 9// 10// Author: Skal (pascal.massimino@gmail.com) 11 12#ifndef WEBP_DSP_DSP_H_ 13#define WEBP_DSP_DSP_H_ 14 15#include "webp/types.h" 16 17#if defined(__cplusplus) || defined(c_plusplus) 18extern "C" { 19#endif 20 21//------------------------------------------------------------------------------ 22// CPU detection 23 24#if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86)) 25#define WEBP_MSC_SSE2 // Visual C++ SSE2 targets 26#endif 27 28#if defined(__SSE2__) || defined(WEBP_MSC_SSE2) 29#define WEBP_USE_SSE2 30#endif 31 32//#if defined(__ANDROID__) && defined(__ARM_ARCH_7A__) 33//#define WEBP_ANDROID_NEON // Android targets that might support NEON 34//#endif 35 36#if defined(__ARM_NEON__) || defined(WEBP_ANDROID_NEON) 37#define WEBP_USE_NEON 38#endif 39 40typedef enum { 41 kSSE2, 42 kSSE3, 43 kNEON 44} CPUFeature; 45// returns true if the CPU supports the feature. 46typedef int (*VP8CPUInfo)(CPUFeature feature); 47extern VP8CPUInfo VP8GetCPUInfo; 48 49//------------------------------------------------------------------------------ 50// Encoding 51 52int VP8GetAlpha(const int histo[]); 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). 57typedef void (*VP8Idct)(const uint8_t* ref, const int16_t* in, uint8_t* dst, 58 int do_two); 59typedef void (*VP8Fdct)(const uint8_t* src, const uint8_t* ref, int16_t* out); 60typedef void (*VP8WHT)(const int16_t* in, int16_t* out); 61extern VP8Idct VP8ITransform; 62extern VP8Fdct VP8FTransform; 63extern VP8WHT VP8ITransformWHT; 64extern VP8WHT VP8FTransformWHT; 65// Predictions 66// *dst is the destination block. *top and *left can be NULL. 67typedef void (*VP8IntraPreds)(uint8_t *dst, const uint8_t* left, 68 const uint8_t* top); 69typedef void (*VP8Intra4Preds)(uint8_t *dst, const uint8_t* top); 70extern VP8Intra4Preds VP8EncPredLuma4; 71extern VP8IntraPreds VP8EncPredLuma16; 72extern VP8IntraPreds VP8EncPredChroma8; 73 74typedef int (*VP8Metric)(const uint8_t* pix, const uint8_t* ref); 75extern VP8Metric VP8SSE16x16, VP8SSE16x8, VP8SSE8x8, VP8SSE4x4; 76typedef int (*VP8WMetric)(const uint8_t* pix, const uint8_t* ref, 77 const uint16_t* const weights); 78extern VP8WMetric VP8TDisto4x4, VP8TDisto16x16; 79 80typedef void (*VP8BlockCopy)(const uint8_t* src, uint8_t* dst); 81extern VP8BlockCopy VP8Copy4x4; 82// Quantization 83struct VP8Matrix; // forward declaration 84typedef int (*VP8QuantizeBlock)(int16_t in[16], int16_t out[16], 85 int n, const struct VP8Matrix* const mtx); 86extern VP8QuantizeBlock VP8EncQuantizeBlock; 87 88// Compute susceptibility based on DCT-coeff histograms: 89// the higher, the "easier" the macroblock is to compress. 90typedef int (*VP8CHisto)(const uint8_t* ref, const uint8_t* pred, 91 int start_block, int end_block); 92extern const int VP8DspScan[16 + 4 + 4]; 93extern VP8CHisto VP8CollectHistogram; 94 95void VP8EncDspInit(void); // must be called before using any of the above 96 97//------------------------------------------------------------------------------ 98// Decoding 99 100typedef void (*VP8DecIdct)(const int16_t* coeffs, uint8_t* dst); 101// when doing two transforms, coeffs is actually int16_t[2][16]. 102typedef void (*VP8DecIdct2)(const int16_t* coeffs, uint8_t* dst, int do_two); 103extern VP8DecIdct2 VP8Transform; 104extern VP8DecIdct VP8TransformUV; 105extern VP8DecIdct VP8TransformDC; 106extern VP8DecIdct VP8TransformDCUV; 107extern void (*VP8TransformWHT)(const int16_t* in, int16_t* out); 108 109// *dst is the destination block, with stride BPS. Boundary samples are 110// assumed accessible when needed. 111typedef void (*VP8PredFunc)(uint8_t* dst); 112extern const VP8PredFunc VP8PredLuma16[/* NUM_B_DC_MODES */]; 113extern const VP8PredFunc VP8PredChroma8[/* NUM_B_DC_MODES */]; 114extern const VP8PredFunc VP8PredLuma4[/* NUM_BMODES */]; 115 116// simple filter (only for luma) 117typedef void (*VP8SimpleFilterFunc)(uint8_t* p, int stride, int thresh); 118extern VP8SimpleFilterFunc VP8SimpleVFilter16; 119extern VP8SimpleFilterFunc VP8SimpleHFilter16; 120extern VP8SimpleFilterFunc VP8SimpleVFilter16i; // filter 3 inner edges 121extern VP8SimpleFilterFunc VP8SimpleHFilter16i; 122 123// regular filter (on both macroblock edges and inner edges) 124typedef void (*VP8LumaFilterFunc)(uint8_t* luma, int stride, 125 int thresh, int ithresh, int hev_t); 126typedef void (*VP8ChromaFilterFunc)(uint8_t* u, uint8_t* v, int stride, 127 int thresh, int ithresh, int hev_t); 128// on outer edge 129extern VP8LumaFilterFunc VP8VFilter16; 130extern VP8LumaFilterFunc VP8HFilter16; 131extern VP8ChromaFilterFunc VP8VFilter8; 132extern VP8ChromaFilterFunc VP8HFilter8; 133 134// on inner edge 135extern VP8LumaFilterFunc VP8VFilter16i; // filtering 3 inner edges altogether 136extern VP8LumaFilterFunc VP8HFilter16i; 137extern VP8ChromaFilterFunc VP8VFilter8i; // filtering u and v altogether 138extern VP8ChromaFilterFunc VP8HFilter8i; 139 140// must be called before anything using the above 141void VP8DspInit(void); 142 143//------------------------------------------------------------------------------ 144// WebP I/O 145 146#define FANCY_UPSAMPLING // undefined to remove fancy upsampling support 147 148typedef void (*WebPUpsampleLinePairFunc)( 149 const uint8_t* top_y, const uint8_t* bottom_y, 150 const uint8_t* top_u, const uint8_t* top_v, 151 const uint8_t* cur_u, const uint8_t* cur_v, 152 uint8_t* top_dst, uint8_t* bottom_dst, int len); 153 154#ifdef FANCY_UPSAMPLING 155 156// Fancy upsampling functions to convert YUV to RGB(A) modes 157extern WebPUpsampleLinePairFunc WebPUpsamplers[/* MODE_LAST */]; 158 159// Initializes SSE2 version of the fancy upsamplers. 160void WebPInitUpsamplersSSE2(void); 161 162#endif // FANCY_UPSAMPLING 163 164// Point-sampling methods. 165typedef void (*WebPSampleLinePairFunc)( 166 const uint8_t* top_y, const uint8_t* bottom_y, 167 const uint8_t* u, const uint8_t* v, 168 uint8_t* top_dst, uint8_t* bottom_dst, int len); 169 170extern const WebPSampleLinePairFunc WebPSamplers[/* MODE_LAST */]; 171 172// General function for converting two lines of ARGB or RGBA. 173// 'alpha_is_last' should be true if 0xff000000 is stored in memory as 174// as 0x00, 0x00, 0x00, 0xff (little endian). 175WebPUpsampleLinePairFunc WebPGetLinePairConverter(int alpha_is_last); 176 177// YUV444->RGB converters 178typedef void (*WebPYUV444Converter)(const uint8_t* y, 179 const uint8_t* u, const uint8_t* v, 180 uint8_t* dst, int len); 181 182extern const WebPYUV444Converter WebPYUV444Converters[/* MODE_LAST */]; 183 184// Main function to be called 185void WebPInitUpsamplers(void); 186 187//------------------------------------------------------------------------------ 188// Pre-multiply planes with alpha values 189 190// Apply alpha pre-multiply on an rgba, bgra or argb plane of size w * h. 191// alpha_first should be 0 for argb, 1 for rgba or bgra (where alpha is last). 192extern void (*WebPApplyAlphaMultiply)( 193 uint8_t* rgba, int alpha_first, int w, int h, int stride); 194 195// Same, buf specifically for RGBA4444 format 196extern void (*WebPApplyAlphaMultiply4444)( 197 uint8_t* rgba4444, int w, int h, int stride); 198 199// To be called first before using the above. 200void WebPInitPremultiply(void); 201 202void WebPInitPremultiplySSE2(void); // should not be called directly. 203 204//------------------------------------------------------------------------------ 205 206#if defined(__cplusplus) || defined(c_plusplus) 207} // extern "C" 208#endif 209 210#endif /* WEBP_DSP_DSP_H_ */ 211