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@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 #define BPS 32   // this is the common stride for enc/dec
28 
29 //------------------------------------------------------------------------------
30 // CPU detection
31 
32 #if defined(__GNUC__)
33 # define LOCAL_GCC_VERSION ((__GNUC__ << 8) | __GNUC_MINOR__)
34 # define LOCAL_GCC_PREREQ(maj, min) \
35     (LOCAL_GCC_VERSION >= (((maj) << 8) | (min)))
36 #else
37 # define LOCAL_GCC_VERSION 0
38 # define LOCAL_GCC_PREREQ(maj, min) 0
39 #endif
40 
41 #ifndef __has_builtin
42 # define __has_builtin(x) 0
43 #endif
44 
45 #if defined(_MSC_VER) && _MSC_VER > 1310 && \
46     (defined(_M_X64) || defined(_M_IX86))
47 #define WEBP_MSC_SSE2  // Visual C++ SSE2 targets
48 #endif
49 
50 #if defined(_MSC_VER) && _MSC_VER >= 1500 && \
51     (defined(_M_X64) || defined(_M_IX86))
52 #define WEBP_MSC_SSE41  // Visual C++ SSE4.1 targets
53 #endif
54 
55 // WEBP_HAVE_* are used to indicate the presence of the instruction set in dsp
56 // files without intrinsics, allowing the corresponding Init() to be called.
57 // Files containing intrinsics will need to be built targeting the instruction
58 // set so should succeed on one of the earlier tests.
59 #if defined(__SSE2__) || defined(WEBP_MSC_SSE2) || defined(WEBP_HAVE_SSE2)
60 #define WEBP_USE_SSE2
61 #endif
62 
63 #if defined(__SSE4_1__) || defined(WEBP_MSC_SSE41) || defined(WEBP_HAVE_SSE41)
64 #define WEBP_USE_SSE41
65 #endif
66 
67 #if defined(__AVX2__) || defined(WEBP_HAVE_AVX2)
68 #define WEBP_USE_AVX2
69 #endif
70 
71 #if defined(__ANDROID__) && defined(__ARM_ARCH_7A__)
72 #define WEBP_ANDROID_NEON  // Android targets that might support NEON
73 #endif
74 
75 // The intrinsics currently cause compiler errors with arm-nacl-gcc and the
76 // inline assembly would need to be modified for use with Native Client.
77 #if (defined(__ARM_NEON__) || \
78      defined(__aarch64__) || defined(WEBP_HAVE_NEON)) && \
79     !defined(__native_client__)
80 #define WEBP_USE_NEON
81 #endif
82 
83 #if defined(_MSC_VER) && _MSC_VER >= 1700 && defined(_M_ARM)
84 #define WEBP_USE_NEON
85 #define WEBP_USE_INTRINSICS
86 #endif
87 
88 #if defined(__mips__) && !defined(__mips64) && \
89     defined(__mips_isa_rev) && (__mips_isa_rev >= 1) && (__mips_isa_rev < 6)
90 #define WEBP_USE_MIPS32
91 #if (__mips_isa_rev >= 2)
92 #define WEBP_USE_MIPS32_R2
93 #if defined(__mips_dspr2) || (__mips_dsp_rev >= 2)
94 #define WEBP_USE_MIPS_DSP_R2
95 #endif
96 #endif
97 #endif
98 
99 #if defined(__mips_msa) && defined(__mips_isa_rev) && (__mips_isa_rev >= 5)
100 #define WEBP_USE_MSA
101 #endif
102 
103 // This macro prevents thread_sanitizer from reporting known concurrent writes.
104 #define WEBP_TSAN_IGNORE_FUNCTION
105 #if defined(__has_feature)
106 #if __has_feature(thread_sanitizer)
107 #undef WEBP_TSAN_IGNORE_FUNCTION
108 #define WEBP_TSAN_IGNORE_FUNCTION __attribute__((no_sanitize_thread))
109 #endif
110 #endif
111 
112 #define WEBP_UBSAN_IGNORE_UNDEF
113 #define WEBP_UBSAN_IGNORE_UNSIGNED_OVERFLOW
114 #if defined(__clang__) && defined(__has_attribute)
115 #if __has_attribute(no_sanitize)
116 // This macro prevents the undefined behavior sanitizer from reporting
117 // failures. This is only meant to silence unaligned loads on platforms that
118 // are known to support them.
119 #undef WEBP_UBSAN_IGNORE_UNDEF
120 #define WEBP_UBSAN_IGNORE_UNDEF \
121   __attribute__((no_sanitize("undefined")))
122 
123 // This macro prevents the undefined behavior sanitizer from reporting
124 // failures related to unsigned integer overflows. This is only meant to
125 // silence cases where this well defined behavior is expected.
126 #undef WEBP_UBSAN_IGNORE_UNSIGNED_OVERFLOW
127 #define WEBP_UBSAN_IGNORE_UNSIGNED_OVERFLOW \
128   __attribute__((no_sanitize("unsigned-integer-overflow")))
129 #endif
130 #endif
131 
132 typedef enum {
133   kSSE2,
134   kSSE3,
135   kSlowSSSE3,  // special feature for slow SSSE3 architectures
136   kSSE4_1,
137   kAVX,
138   kAVX2,
139   kNEON,
140   kMIPS32,
141   kMIPSdspR2,
142   kMSA
143 } CPUFeature;
144 // returns true if the CPU supports the feature.
145 typedef int (*VP8CPUInfo)(CPUFeature feature);
146 WEBP_EXTERN(VP8CPUInfo) VP8GetCPUInfo;
147 
148 //------------------------------------------------------------------------------
149 // Init stub generator
150 
151 // Defines an init function stub to ensure each module exposes a symbol,
152 // avoiding a compiler warning.
153 #define WEBP_DSP_INIT_STUB(func) \
154   extern void func(void); \
155   WEBP_TSAN_IGNORE_FUNCTION void func(void) {}
156 
157 //------------------------------------------------------------------------------
158 // Encoding
159 
160 // Transforms
161 // VP8Idct: Does one of two inverse transforms. If do_two is set, the transforms
162 //          will be done for (ref, in, dst) and (ref + 4, in + 16, dst + 4).
163 typedef void (*VP8Idct)(const uint8_t* ref, const int16_t* in, uint8_t* dst,
164                         int do_two);
165 typedef void (*VP8Fdct)(const uint8_t* src, const uint8_t* ref, int16_t* out);
166 typedef void (*VP8WHT)(const int16_t* in, int16_t* out);
167 extern VP8Idct VP8ITransform;
168 extern VP8Fdct VP8FTransform;
169 extern VP8Fdct VP8FTransform2;   // performs two transforms at a time
170 extern VP8WHT VP8FTransformWHT;
171 // Predictions
172 // *dst is the destination block. *top and *left can be NULL.
173 typedef void (*VP8IntraPreds)(uint8_t *dst, const uint8_t* left,
174                               const uint8_t* top);
175 typedef void (*VP8Intra4Preds)(uint8_t *dst, const uint8_t* top);
176 extern VP8Intra4Preds VP8EncPredLuma4;
177 extern VP8IntraPreds VP8EncPredLuma16;
178 extern VP8IntraPreds VP8EncPredChroma8;
179 
180 typedef int (*VP8Metric)(const uint8_t* pix, const uint8_t* ref);
181 extern VP8Metric VP8SSE16x16, VP8SSE16x8, VP8SSE8x8, VP8SSE4x4;
182 typedef int (*VP8WMetric)(const uint8_t* pix, const uint8_t* ref,
183                           const uint16_t* const weights);
184 // The weights for VP8TDisto4x4 and VP8TDisto16x16 contain a row-major
185 // 4 by 4 symmetric matrix.
186 extern VP8WMetric VP8TDisto4x4, VP8TDisto16x16;
187 
188 // Compute the average (DC) of four 4x4 blocks.
189 // Each sub-4x4 block #i sum is stored in dc[i].
190 typedef void (*VP8MeanMetric)(const uint8_t* ref, uint32_t dc[4]);
191 extern VP8MeanMetric VP8Mean16x4;
192 
193 typedef void (*VP8BlockCopy)(const uint8_t* src, uint8_t* dst);
194 extern VP8BlockCopy VP8Copy4x4;
195 extern VP8BlockCopy VP8Copy16x8;
196 // Quantization
197 struct VP8Matrix;   // forward declaration
198 typedef int (*VP8QuantizeBlock)(int16_t in[16], int16_t out[16],
199                                 const struct VP8Matrix* const mtx);
200 // Same as VP8QuantizeBlock, but quantizes two consecutive blocks.
201 typedef int (*VP8Quantize2Blocks)(int16_t in[32], int16_t out[32],
202                                   const struct VP8Matrix* const mtx);
203 
204 extern VP8QuantizeBlock VP8EncQuantizeBlock;
205 extern VP8Quantize2Blocks VP8EncQuantize2Blocks;
206 
207 // specific to 2nd transform:
208 typedef int (*VP8QuantizeBlockWHT)(int16_t in[16], int16_t out[16],
209                                    const struct VP8Matrix* const mtx);
210 extern VP8QuantizeBlockWHT VP8EncQuantizeBlockWHT;
211 
212 extern const int VP8DspScan[16 + 4 + 4];
213 
214 // Collect histogram for susceptibility calculation.
215 #define MAX_COEFF_THRESH   31   // size of histogram used by CollectHistogram.
216 typedef struct {
217   // We only need to store max_value and last_non_zero, not the distribution.
218   int max_value;
219   int last_non_zero;
220 } VP8Histogram;
221 typedef void (*VP8CHisto)(const uint8_t* ref, const uint8_t* pred,
222                           int start_block, int end_block,
223                           VP8Histogram* const histo);
224 extern VP8CHisto VP8CollectHistogram;
225 // General-purpose util function to help VP8CollectHistogram().
226 void VP8SetHistogramData(const int distribution[MAX_COEFF_THRESH + 1],
227                          VP8Histogram* const histo);
228 
229 // must be called before using any of the above
230 void VP8EncDspInit(void);
231 
232 //------------------------------------------------------------------------------
233 // cost functions (encoding)
234 
235 extern const uint16_t VP8EntropyCost[256];        // 8bit fixed-point log(p)
236 // approximate cost per level:
237 extern const uint16_t VP8LevelFixedCosts[2047 /*MAX_LEVEL*/ + 1];
238 extern const uint8_t VP8EncBands[16 + 1];
239 
240 struct VP8Residual;
241 typedef void (*VP8SetResidualCoeffsFunc)(const int16_t* const coeffs,
242                                          struct VP8Residual* const res);
243 extern VP8SetResidualCoeffsFunc VP8SetResidualCoeffs;
244 
245 // Cost calculation function.
246 typedef int (*VP8GetResidualCostFunc)(int ctx0,
247                                       const struct VP8Residual* const res);
248 extern VP8GetResidualCostFunc VP8GetResidualCost;
249 
250 // must be called before anything using the above
251 void VP8EncDspCostInit(void);
252 
253 //------------------------------------------------------------------------------
254 // SSIM / PSNR utils
255 
256 // struct for accumulating statistical moments
257 typedef struct {
258   uint32_t w;              // sum(w_i) : sum of weights
259   uint32_t xm, ym;         // sum(w_i * x_i), sum(w_i * y_i)
260   uint32_t xxm, xym, yym;  // sum(w_i * x_i * x_i), etc.
261 } VP8DistoStats;
262 
263 // Compute the final SSIM value
264 // The non-clipped version assumes stats->w = (2 * VP8_SSIM_KERNEL + 1)^2.
265 double VP8SSIMFromStats(const VP8DistoStats* const stats);
266 double VP8SSIMFromStatsClipped(const VP8DistoStats* const stats);
267 
268 #define VP8_SSIM_KERNEL 3   // total size of the kernel: 2 * VP8_SSIM_KERNEL + 1
269 typedef double (*VP8SSIMGetClippedFunc)(const uint8_t* src1, int stride1,
270                                         const uint8_t* src2, int stride2,
271                                         int xo, int yo,  // center position
272                                         int W, int H);   // plane dimension
273 
274 // This version is called with the guarantee that you can load 8 bytes and
275 // 8 rows at offset src1 and src2
276 typedef double (*VP8SSIMGetFunc)(const uint8_t* src1, int stride1,
277                                  const uint8_t* src2, int stride2);
278 
279 extern VP8SSIMGetFunc VP8SSIMGet;         // unclipped / unchecked
280 extern VP8SSIMGetClippedFunc VP8SSIMGetClipped;   // with clipping
281 
282 typedef uint32_t (*VP8AccumulateSSEFunc)(const uint8_t* src1,
283                                          const uint8_t* src2, int len);
284 extern VP8AccumulateSSEFunc VP8AccumulateSSE;
285 
286 // must be called before using any of the above directly
287 void VP8SSIMDspInit(void);
288 
289 //------------------------------------------------------------------------------
290 // Decoding
291 
292 typedef void (*VP8DecIdct)(const int16_t* coeffs, uint8_t* dst);
293 // when doing two transforms, coeffs is actually int16_t[2][16].
294 typedef void (*VP8DecIdct2)(const int16_t* coeffs, uint8_t* dst, int do_two);
295 extern VP8DecIdct2 VP8Transform;
296 extern VP8DecIdct VP8TransformAC3;
297 extern VP8DecIdct VP8TransformUV;
298 extern VP8DecIdct VP8TransformDC;
299 extern VP8DecIdct VP8TransformDCUV;
300 extern VP8WHT VP8TransformWHT;
301 
302 // *dst is the destination block, with stride BPS. Boundary samples are
303 // assumed accessible when needed.
304 typedef void (*VP8PredFunc)(uint8_t* dst);
305 extern VP8PredFunc VP8PredLuma16[/* NUM_B_DC_MODES */];
306 extern VP8PredFunc VP8PredChroma8[/* NUM_B_DC_MODES */];
307 extern VP8PredFunc VP8PredLuma4[/* NUM_BMODES */];
308 
309 // clipping tables (for filtering)
310 extern const int8_t* const VP8ksclip1;  // clips [-1020, 1020] to [-128, 127]
311 extern const int8_t* const VP8ksclip2;  // clips [-112, 112] to [-16, 15]
312 extern const uint8_t* const VP8kclip1;  // clips [-255,511] to [0,255]
313 extern const uint8_t* const VP8kabs0;   // abs(x) for x in [-255,255]
314 // must be called first
315 void VP8InitClipTables(void);
316 
317 // simple filter (only for luma)
318 typedef void (*VP8SimpleFilterFunc)(uint8_t* p, int stride, int thresh);
319 extern VP8SimpleFilterFunc VP8SimpleVFilter16;
320 extern VP8SimpleFilterFunc VP8SimpleHFilter16;
321 extern VP8SimpleFilterFunc VP8SimpleVFilter16i;  // filter 3 inner edges
322 extern VP8SimpleFilterFunc VP8SimpleHFilter16i;
323 
324 // regular filter (on both macroblock edges and inner edges)
325 typedef void (*VP8LumaFilterFunc)(uint8_t* luma, int stride,
326                                   int thresh, int ithresh, int hev_t);
327 typedef void (*VP8ChromaFilterFunc)(uint8_t* u, uint8_t* v, int stride,
328                                     int thresh, int ithresh, int hev_t);
329 // on outer edge
330 extern VP8LumaFilterFunc VP8VFilter16;
331 extern VP8LumaFilterFunc VP8HFilter16;
332 extern VP8ChromaFilterFunc VP8VFilter8;
333 extern VP8ChromaFilterFunc VP8HFilter8;
334 
335 // on inner edge
336 extern VP8LumaFilterFunc VP8VFilter16i;   // filtering 3 inner edges altogether
337 extern VP8LumaFilterFunc VP8HFilter16i;
338 extern VP8ChromaFilterFunc VP8VFilter8i;  // filtering u and v altogether
339 extern VP8ChromaFilterFunc VP8HFilter8i;
340 
341 // Dithering. Combines dithering values (centered around 128) with dst[],
342 // according to: dst[] = clip(dst[] + (((dither[]-128) + 8) >> 4)
343 #define VP8_DITHER_DESCALE 4
344 #define VP8_DITHER_DESCALE_ROUNDER (1 << (VP8_DITHER_DESCALE - 1))
345 #define VP8_DITHER_AMP_BITS 7
346 #define VP8_DITHER_AMP_CENTER (1 << VP8_DITHER_AMP_BITS)
347 extern void (*VP8DitherCombine8x8)(const uint8_t* dither, uint8_t* dst,
348                                    int dst_stride);
349 
350 // must be called before anything using the above
351 void VP8DspInit(void);
352 
353 //------------------------------------------------------------------------------
354 // WebP I/O
355 
356 #define FANCY_UPSAMPLING   // undefined to remove fancy upsampling support
357 
358 // Convert a pair of y/u/v lines together to the output rgb/a colorspace.
359 // bottom_y can be NULL if only one line of output is needed (at top/bottom).
360 typedef void (*WebPUpsampleLinePairFunc)(
361     const uint8_t* top_y, const uint8_t* bottom_y,
362     const uint8_t* top_u, const uint8_t* top_v,
363     const uint8_t* cur_u, const uint8_t* cur_v,
364     uint8_t* top_dst, uint8_t* bottom_dst, int len);
365 
366 #ifdef FANCY_UPSAMPLING
367 
368 // Fancy upsampling functions to convert YUV to RGB(A) modes
369 extern WebPUpsampleLinePairFunc WebPUpsamplers[/* MODE_LAST */];
370 
371 #endif    // FANCY_UPSAMPLING
372 
373 // Per-row point-sampling methods.
374 typedef void (*WebPSamplerRowFunc)(const uint8_t* y,
375                                    const uint8_t* u, const uint8_t* v,
376                                    uint8_t* dst, int len);
377 // Generic function to apply 'WebPSamplerRowFunc' to the whole plane:
378 void WebPSamplerProcessPlane(const uint8_t* y, int y_stride,
379                              const uint8_t* u, const uint8_t* v, int uv_stride,
380                              uint8_t* dst, int dst_stride,
381                              int width, int height, WebPSamplerRowFunc func);
382 
383 // Sampling functions to convert rows of YUV to RGB(A)
384 extern WebPSamplerRowFunc WebPSamplers[/* MODE_LAST */];
385 
386 // General function for converting two lines of ARGB or RGBA.
387 // 'alpha_is_last' should be true if 0xff000000 is stored in memory as
388 // as 0x00, 0x00, 0x00, 0xff (little endian).
389 WebPUpsampleLinePairFunc WebPGetLinePairConverter(int alpha_is_last);
390 
391 // YUV444->RGB converters
392 typedef void (*WebPYUV444Converter)(const uint8_t* y,
393                                     const uint8_t* u, const uint8_t* v,
394                                     uint8_t* dst, int len);
395 
396 extern WebPYUV444Converter WebPYUV444Converters[/* MODE_LAST */];
397 
398 // Must be called before using the WebPUpsamplers[] (and for premultiplied
399 // colorspaces like rgbA, rgbA4444, etc)
400 void WebPInitUpsamplers(void);
401 // Must be called before using WebPSamplers[]
402 void WebPInitSamplers(void);
403 // Must be called before using WebPYUV444Converters[]
404 void WebPInitYUV444Converters(void);
405 
406 //------------------------------------------------------------------------------
407 // ARGB -> YUV converters
408 
409 // Convert ARGB samples to luma Y.
410 extern void (*WebPConvertARGBToY)(const uint32_t* argb, uint8_t* y, int width);
411 // Convert ARGB samples to U/V with downsampling. do_store should be '1' for
412 // even lines and '0' for odd ones. 'src_width' is the original width, not
413 // the U/V one.
414 extern void (*WebPConvertARGBToUV)(const uint32_t* argb, uint8_t* u, uint8_t* v,
415                                    int src_width, int do_store);
416 
417 // Convert a row of accumulated (four-values) of rgba32 toward U/V
418 extern void (*WebPConvertRGBA32ToUV)(const uint16_t* rgb,
419                                      uint8_t* u, uint8_t* v, int width);
420 
421 // Convert RGB or BGR to Y
422 extern void (*WebPConvertRGB24ToY)(const uint8_t* rgb, uint8_t* y, int width);
423 extern void (*WebPConvertBGR24ToY)(const uint8_t* bgr, uint8_t* y, int width);
424 
425 // used for plain-C fallback.
426 extern void WebPConvertARGBToUV_C(const uint32_t* argb, uint8_t* u, uint8_t* v,
427                                   int src_width, int do_store);
428 extern void WebPConvertRGBA32ToUV_C(const uint16_t* rgb,
429                                     uint8_t* u, uint8_t* v, int width);
430 
431 // utilities for accurate RGB->YUV conversion
432 extern uint64_t (*WebPSharpYUVUpdateY)(const uint16_t* src, const uint16_t* ref,
433                                        uint16_t* dst, int len);
434 extern void (*WebPSharpYUVUpdateRGB)(const int16_t* src, const int16_t* ref,
435                                      int16_t* dst, int len);
436 extern void (*WebPSharpYUVFilterRow)(const int16_t* A, const int16_t* B,
437                                      int len,
438                                      const uint16_t* best_y, uint16_t* out);
439 
440 // Must be called before using the above.
441 void WebPInitConvertARGBToYUV(void);
442 
443 //------------------------------------------------------------------------------
444 // Rescaler
445 
446 struct WebPRescaler;
447 
448 // Import a row of data and save its contribution in the rescaler.
449 // 'channel' denotes the channel number to be imported. 'Expand' corresponds to
450 // the wrk->x_expand case. Otherwise, 'Shrink' is to be used.
451 typedef void (*WebPRescalerImportRowFunc)(struct WebPRescaler* const wrk,
452                                           const uint8_t* src);
453 
454 extern WebPRescalerImportRowFunc WebPRescalerImportRowExpand;
455 extern WebPRescalerImportRowFunc WebPRescalerImportRowShrink;
456 
457 // Export one row (starting at x_out position) from rescaler.
458 // 'Expand' corresponds to the wrk->y_expand case.
459 // Otherwise 'Shrink' is to be used
460 typedef void (*WebPRescalerExportRowFunc)(struct WebPRescaler* const wrk);
461 extern WebPRescalerExportRowFunc WebPRescalerExportRowExpand;
462 extern WebPRescalerExportRowFunc WebPRescalerExportRowShrink;
463 
464 // Plain-C implementation, as fall-back.
465 extern void WebPRescalerImportRowExpandC(struct WebPRescaler* const wrk,
466                                          const uint8_t* src);
467 extern void WebPRescalerImportRowShrinkC(struct WebPRescaler* const wrk,
468                                          const uint8_t* src);
469 extern void WebPRescalerExportRowExpandC(struct WebPRescaler* const wrk);
470 extern void WebPRescalerExportRowShrinkC(struct WebPRescaler* const wrk);
471 
472 // Main entry calls:
473 extern void WebPRescalerImportRow(struct WebPRescaler* const wrk,
474                                   const uint8_t* src);
475 // Export one row (starting at x_out position) from rescaler.
476 extern void WebPRescalerExportRow(struct WebPRescaler* const wrk);
477 
478 // Must be called first before using the above.
479 void WebPRescalerDspInit(void);
480 
481 //------------------------------------------------------------------------------
482 // Utilities for processing transparent channel.
483 
484 // Apply alpha pre-multiply on an rgba, bgra or argb plane of size w * h.
485 // alpha_first should be 0 for argb, 1 for rgba or bgra (where alpha is last).
486 extern void (*WebPApplyAlphaMultiply)(
487     uint8_t* rgba, int alpha_first, int w, int h, int stride);
488 
489 // Same, buf specifically for RGBA4444 format
490 extern void (*WebPApplyAlphaMultiply4444)(
491     uint8_t* rgba4444, int w, int h, int stride);
492 
493 // Dispatch the values from alpha[] plane to the ARGB destination 'dst'.
494 // Returns true if alpha[] plane has non-trivial values different from 0xff.
495 extern int (*WebPDispatchAlpha)(const uint8_t* alpha, int alpha_stride,
496                                 int width, int height,
497                                 uint8_t* dst, int dst_stride);
498 
499 // Transfer packed 8b alpha[] values to green channel in dst[], zero'ing the
500 // A/R/B values. 'dst_stride' is the stride for dst[] in uint32_t units.
501 extern void (*WebPDispatchAlphaToGreen)(const uint8_t* alpha, int alpha_stride,
502                                         int width, int height,
503                                         uint32_t* dst, int dst_stride);
504 
505 // Extract the alpha values from 32b values in argb[] and pack them into alpha[]
506 // (this is the opposite of WebPDispatchAlpha).
507 // Returns true if there's only trivial 0xff alpha values.
508 extern int (*WebPExtractAlpha)(const uint8_t* argb, int argb_stride,
509                                int width, int height,
510                                uint8_t* alpha, int alpha_stride);
511 
512 // Extract the green values from 32b values in argb[] and pack them into alpha[]
513 // (this is the opposite of WebPDispatchAlphaToGreen).
514 extern void (*WebPExtractGreen)(const uint32_t* argb, uint8_t* alpha, int size);
515 
516 // Pre-Multiply operation transforms x into x * A / 255  (where x=Y,R,G or B).
517 // Un-Multiply operation transforms x into x * 255 / A.
518 
519 // Pre-Multiply or Un-Multiply (if 'inverse' is true) argb values in a row.
520 extern void (*WebPMultARGBRow)(uint32_t* const ptr, int width, int inverse);
521 
522 // Same a WebPMultARGBRow(), but for several rows.
523 void WebPMultARGBRows(uint8_t* ptr, int stride, int width, int num_rows,
524                       int inverse);
525 
526 // Same for a row of single values, with side alpha values.
527 extern void (*WebPMultRow)(uint8_t* const ptr, const uint8_t* const alpha,
528                            int width, int inverse);
529 
530 // Same a WebPMultRow(), but for several 'num_rows' rows.
531 void WebPMultRows(uint8_t* ptr, int stride,
532                   const uint8_t* alpha, int alpha_stride,
533                   int width, int num_rows, int inverse);
534 
535 // Plain-C versions, used as fallback by some implementations.
536 void WebPMultRowC(uint8_t* const ptr, const uint8_t* const alpha,
537                   int width, int inverse);
538 void WebPMultARGBRowC(uint32_t* const ptr, int width, int inverse);
539 
540 // To be called first before using the above.
541 void WebPInitAlphaProcessing(void);
542 
543 // ARGB packing function: a/r/g/b input is rgba or bgra order.
544 extern void (*VP8PackARGB)(const uint8_t* a, const uint8_t* r,
545                            const uint8_t* g, const uint8_t* b, int len,
546                            uint32_t* out);
547 
548 // RGB packing function. 'step' can be 3 or 4. r/g/b input is rgb or bgr order.
549 extern void (*VP8PackRGB)(const uint8_t* r, const uint8_t* g, const uint8_t* b,
550                           int len, int step, uint32_t* out);
551 
552 // To be called first before using the above.
553 void VP8EncDspARGBInit(void);
554 
555 //------------------------------------------------------------------------------
556 // Filter functions
557 
558 typedef enum {     // Filter types.
559   WEBP_FILTER_NONE = 0,
560   WEBP_FILTER_HORIZONTAL,
561   WEBP_FILTER_VERTICAL,
562   WEBP_FILTER_GRADIENT,
563   WEBP_FILTER_LAST = WEBP_FILTER_GRADIENT + 1,  // end marker
564   WEBP_FILTER_BEST,    // meta-types
565   WEBP_FILTER_FAST
566 } WEBP_FILTER_TYPE;
567 
568 typedef void (*WebPFilterFunc)(const uint8_t* in, int width, int height,
569                                int stride, uint8_t* out);
570 // In-place un-filtering.
571 // Warning! 'prev_line' pointer can be equal to 'cur_line' or 'preds'.
572 typedef void (*WebPUnfilterFunc)(const uint8_t* prev_line, const uint8_t* preds,
573                                  uint8_t* cur_line, int width);
574 
575 // Filter the given data using the given predictor.
576 // 'in' corresponds to a 2-dimensional pixel array of size (stride * height)
577 // in raster order.
578 // 'stride' is number of bytes per scan line (with possible padding).
579 // 'out' should be pre-allocated.
580 extern WebPFilterFunc WebPFilters[WEBP_FILTER_LAST];
581 
582 // In-place reconstruct the original data from the given filtered data.
583 // The reconstruction will be done for 'num_rows' rows starting from 'row'
584 // (assuming rows upto 'row - 1' are already reconstructed).
585 extern WebPUnfilterFunc WebPUnfilters[WEBP_FILTER_LAST];
586 
587 // To be called first before using the above.
588 void VP8FiltersInit(void);
589 
590 #ifdef __cplusplus
591 }    // extern "C"
592 #endif
593 
594 #endif  /* WEBP_DSP_DSP_H_ */
595