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 "src/webp/config.h"
19#endif
20
21#include "src/webp/types.h"
22
23#ifdef __cplusplus
24extern "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#if defined(__clang__)
42# define LOCAL_CLANG_VERSION ((__clang_major__ << 8) | __clang_minor__)
43# define LOCAL_CLANG_PREREQ(maj, min) \
44    (LOCAL_CLANG_VERSION >= (((maj) << 8) | (min)))
45#else
46# define LOCAL_CLANG_VERSION 0
47# define LOCAL_CLANG_PREREQ(maj, min) 0
48#endif
49
50#ifndef __has_builtin
51# define __has_builtin(x) 0
52#endif
53
54// for now, none of the optimizations below are available in emscripten
55#if !defined(EMSCRIPTEN)
56
57#if defined(_MSC_VER) && _MSC_VER > 1310 && \
58    (defined(_M_X64) || defined(_M_IX86))
59#define WEBP_MSC_SSE2  // Visual C++ SSE2 targets
60#endif
61
62#if defined(_MSC_VER) && _MSC_VER >= 1500 && \
63    (defined(_M_X64) || defined(_M_IX86))
64#define WEBP_MSC_SSE41  // Visual C++ SSE4.1 targets
65#endif
66
67// WEBP_HAVE_* are used to indicate the presence of the instruction set in dsp
68// files without intrinsics, allowing the corresponding Init() to be called.
69// Files containing intrinsics will need to be built targeting the instruction
70// set so should succeed on one of the earlier tests.
71#if defined(__SSE2__) || defined(WEBP_MSC_SSE2) || defined(WEBP_HAVE_SSE2)
72#define WEBP_USE_SSE2
73#endif
74
75#if defined(__SSE4_1__) || defined(WEBP_MSC_SSE41) || defined(WEBP_HAVE_SSE41)
76#define WEBP_USE_SSE41
77#endif
78
79#if defined(__AVX2__) || defined(WEBP_HAVE_AVX2)
80#define WEBP_USE_AVX2
81#endif
82
83// The intrinsics currently cause compiler errors with arm-nacl-gcc and the
84// inline assembly would need to be modified for use with Native Client.
85#if (defined(__ARM_NEON__) || \
86     defined(__aarch64__) || defined(WEBP_HAVE_NEON)) && \
87    !defined(__native_client__)
88#define WEBP_USE_NEON
89#endif
90
91#if !defined(WEBP_USE_NEON) && defined(__ANDROID__) && \
92    defined(__ARM_ARCH_7A__) && defined(HAVE_CPU_FEATURES_H)
93#define WEBP_ANDROID_NEON  // Android targets that may have NEON
94#define WEBP_USE_NEON
95#endif
96
97#if defined(_MSC_VER) && _MSC_VER >= 1700 && defined(_M_ARM)
98#define WEBP_USE_NEON
99#define WEBP_USE_INTRINSICS
100#endif
101
102#if defined(__mips__) && !defined(__mips64) && \
103    defined(__mips_isa_rev) && (__mips_isa_rev >= 1) && (__mips_isa_rev < 6)
104#define WEBP_USE_MIPS32
105#if (__mips_isa_rev >= 2)
106#define WEBP_USE_MIPS32_R2
107#if defined(__mips_dspr2) || (defined(__mips_dsp_rev) && __mips_dsp_rev >= 2)
108#define WEBP_USE_MIPS_DSP_R2
109#endif
110#endif
111#endif
112
113#if defined(__mips_msa) && defined(__mips_isa_rev) && (__mips_isa_rev >= 5)
114#define WEBP_USE_MSA
115#endif
116
117#endif  /* EMSCRIPTEN */
118
119#ifndef WEBP_DSP_OMIT_C_CODE
120#define WEBP_DSP_OMIT_C_CODE 1
121#endif
122
123#if (defined(__aarch64__) || defined(__ARM_NEON__)) && WEBP_DSP_OMIT_C_CODE
124#define WEBP_NEON_OMIT_C_CODE 1
125#else
126#define WEBP_NEON_OMIT_C_CODE 0
127#endif
128
129#if !(LOCAL_CLANG_PREREQ(3,8) || LOCAL_GCC_PREREQ(4,8) || defined(__aarch64__))
130#define WEBP_NEON_WORK_AROUND_GCC 1
131#else
132#define WEBP_NEON_WORK_AROUND_GCC 0
133#endif
134
135// This macro prevents thread_sanitizer from reporting known concurrent writes.
136#define WEBP_TSAN_IGNORE_FUNCTION
137#if defined(__has_feature)
138#if __has_feature(thread_sanitizer)
139#undef WEBP_TSAN_IGNORE_FUNCTION
140#define WEBP_TSAN_IGNORE_FUNCTION __attribute__((no_sanitize_thread))
141#endif
142#endif
143
144#define WEBP_UBSAN_IGNORE_UNDEF
145#define WEBP_UBSAN_IGNORE_UNSIGNED_OVERFLOW
146#if defined(__clang__) && defined(__has_attribute)
147#if __has_attribute(no_sanitize)
148// This macro prevents the undefined behavior sanitizer from reporting
149// failures. This is only meant to silence unaligned loads on platforms that
150// are known to support them.
151#undef WEBP_UBSAN_IGNORE_UNDEF
152#define WEBP_UBSAN_IGNORE_UNDEF \
153  __attribute__((no_sanitize("undefined")))
154
155// This macro prevents the undefined behavior sanitizer from reporting
156// failures related to unsigned integer overflows. This is only meant to
157// silence cases where this well defined behavior is expected.
158#undef WEBP_UBSAN_IGNORE_UNSIGNED_OVERFLOW
159#define WEBP_UBSAN_IGNORE_UNSIGNED_OVERFLOW \
160  __attribute__((no_sanitize("unsigned-integer-overflow")))
161#endif
162#endif
163
164// Regularize the definition of WEBP_SWAP_16BIT_CSP (backward compatibility)
165#if !defined(WEBP_SWAP_16BIT_CSP)
166#define WEBP_SWAP_16BIT_CSP 0
167#endif
168
169typedef enum {
170  kSSE2,
171  kSSE3,
172  kSlowSSSE3,  // special feature for slow SSSE3 architectures
173  kSSE4_1,
174  kAVX,
175  kAVX2,
176  kNEON,
177  kMIPS32,
178  kMIPSdspR2,
179  kMSA
180} CPUFeature;
181// returns true if the CPU supports the feature.
182typedef int (*VP8CPUInfo)(CPUFeature feature);
183WEBP_EXTERN VP8CPUInfo VP8GetCPUInfo;
184
185//------------------------------------------------------------------------------
186// Init stub generator
187
188// Defines an init function stub to ensure each module exposes a symbol,
189// avoiding a compiler warning.
190#define WEBP_DSP_INIT_STUB(func) \
191  extern void func(void); \
192  WEBP_TSAN_IGNORE_FUNCTION void func(void) {}
193
194//------------------------------------------------------------------------------
195// Encoding
196
197// Transforms
198// VP8Idct: Does one of two inverse transforms. If do_two is set, the transforms
199//          will be done for (ref, in, dst) and (ref + 4, in + 16, dst + 4).
200typedef void (*VP8Idct)(const uint8_t* ref, const int16_t* in, uint8_t* dst,
201                        int do_two);
202typedef void (*VP8Fdct)(const uint8_t* src, const uint8_t* ref, int16_t* out);
203typedef void (*VP8WHT)(const int16_t* in, int16_t* out);
204extern VP8Idct VP8ITransform;
205extern VP8Fdct VP8FTransform;
206extern VP8Fdct VP8FTransform2;   // performs two transforms at a time
207extern VP8WHT VP8FTransformWHT;
208// Predictions
209// *dst is the destination block. *top and *left can be NULL.
210typedef void (*VP8IntraPreds)(uint8_t *dst, const uint8_t* left,
211                              const uint8_t* top);
212typedef void (*VP8Intra4Preds)(uint8_t *dst, const uint8_t* top);
213extern VP8Intra4Preds VP8EncPredLuma4;
214extern VP8IntraPreds VP8EncPredLuma16;
215extern VP8IntraPreds VP8EncPredChroma8;
216
217typedef int (*VP8Metric)(const uint8_t* pix, const uint8_t* ref);
218extern VP8Metric VP8SSE16x16, VP8SSE16x8, VP8SSE8x8, VP8SSE4x4;
219typedef int (*VP8WMetric)(const uint8_t* pix, const uint8_t* ref,
220                          const uint16_t* const weights);
221// The weights for VP8TDisto4x4 and VP8TDisto16x16 contain a row-major
222// 4 by 4 symmetric matrix.
223extern VP8WMetric VP8TDisto4x4, VP8TDisto16x16;
224
225// Compute the average (DC) of four 4x4 blocks.
226// Each sub-4x4 block #i sum is stored in dc[i].
227typedef void (*VP8MeanMetric)(const uint8_t* ref, uint32_t dc[4]);
228extern VP8MeanMetric VP8Mean16x4;
229
230typedef void (*VP8BlockCopy)(const uint8_t* src, uint8_t* dst);
231extern VP8BlockCopy VP8Copy4x4;
232extern VP8BlockCopy VP8Copy16x8;
233// Quantization
234struct VP8Matrix;   // forward declaration
235typedef int (*VP8QuantizeBlock)(int16_t in[16], int16_t out[16],
236                                const struct VP8Matrix* const mtx);
237// Same as VP8QuantizeBlock, but quantizes two consecutive blocks.
238typedef int (*VP8Quantize2Blocks)(int16_t in[32], int16_t out[32],
239                                  const struct VP8Matrix* const mtx);
240
241extern VP8QuantizeBlock VP8EncQuantizeBlock;
242extern VP8Quantize2Blocks VP8EncQuantize2Blocks;
243
244// specific to 2nd transform:
245typedef int (*VP8QuantizeBlockWHT)(int16_t in[16], int16_t out[16],
246                                   const struct VP8Matrix* const mtx);
247extern VP8QuantizeBlockWHT VP8EncQuantizeBlockWHT;
248
249extern const int VP8DspScan[16 + 4 + 4];
250
251// Collect histogram for susceptibility calculation.
252#define MAX_COEFF_THRESH   31   // size of histogram used by CollectHistogram.
253typedef struct {
254  // We only need to store max_value and last_non_zero, not the distribution.
255  int max_value;
256  int last_non_zero;
257} VP8Histogram;
258typedef void (*VP8CHisto)(const uint8_t* ref, const uint8_t* pred,
259                          int start_block, int end_block,
260                          VP8Histogram* const histo);
261extern VP8CHisto VP8CollectHistogram;
262// General-purpose util function to help VP8CollectHistogram().
263void VP8SetHistogramData(const int distribution[MAX_COEFF_THRESH + 1],
264                         VP8Histogram* const histo);
265
266// must be called before using any of the above
267void VP8EncDspInit(void);
268
269//------------------------------------------------------------------------------
270// cost functions (encoding)
271
272extern const uint16_t VP8EntropyCost[256];        // 8bit fixed-point log(p)
273// approximate cost per level:
274extern const uint16_t VP8LevelFixedCosts[2047 /*MAX_LEVEL*/ + 1];
275extern const uint8_t VP8EncBands[16 + 1];
276
277struct VP8Residual;
278typedef void (*VP8SetResidualCoeffsFunc)(const int16_t* const coeffs,
279                                         struct VP8Residual* const res);
280extern VP8SetResidualCoeffsFunc VP8SetResidualCoeffs;
281
282// Cost calculation function.
283typedef int (*VP8GetResidualCostFunc)(int ctx0,
284                                      const struct VP8Residual* const res);
285extern VP8GetResidualCostFunc VP8GetResidualCost;
286
287// must be called before anything using the above
288void VP8EncDspCostInit(void);
289
290//------------------------------------------------------------------------------
291// SSIM / PSNR utils
292
293// struct for accumulating statistical moments
294typedef struct {
295  uint32_t w;              // sum(w_i) : sum of weights
296  uint32_t xm, ym;         // sum(w_i * x_i), sum(w_i * y_i)
297  uint32_t xxm, xym, yym;  // sum(w_i * x_i * x_i), etc.
298} VP8DistoStats;
299
300// Compute the final SSIM value
301// The non-clipped version assumes stats->w = (2 * VP8_SSIM_KERNEL + 1)^2.
302double VP8SSIMFromStats(const VP8DistoStats* const stats);
303double VP8SSIMFromStatsClipped(const VP8DistoStats* const stats);
304
305#define VP8_SSIM_KERNEL 3   // total size of the kernel: 2 * VP8_SSIM_KERNEL + 1
306typedef double (*VP8SSIMGetClippedFunc)(const uint8_t* src1, int stride1,
307                                        const uint8_t* src2, int stride2,
308                                        int xo, int yo,  // center position
309                                        int W, int H);   // plane dimension
310
311#if !defined(WEBP_REDUCE_SIZE)
312// This version is called with the guarantee that you can load 8 bytes and
313// 8 rows at offset src1 and src2
314typedef double (*VP8SSIMGetFunc)(const uint8_t* src1, int stride1,
315                                 const uint8_t* src2, int stride2);
316
317extern VP8SSIMGetFunc VP8SSIMGet;         // unclipped / unchecked
318extern VP8SSIMGetClippedFunc VP8SSIMGetClipped;   // with clipping
319#endif
320
321#if !defined(WEBP_DISABLE_STATS)
322typedef uint32_t (*VP8AccumulateSSEFunc)(const uint8_t* src1,
323                                         const uint8_t* src2, int len);
324extern VP8AccumulateSSEFunc VP8AccumulateSSE;
325#endif
326
327// must be called before using any of the above directly
328void VP8SSIMDspInit(void);
329
330//------------------------------------------------------------------------------
331// Decoding
332
333typedef void (*VP8DecIdct)(const int16_t* coeffs, uint8_t* dst);
334// when doing two transforms, coeffs is actually int16_t[2][16].
335typedef void (*VP8DecIdct2)(const int16_t* coeffs, uint8_t* dst, int do_two);
336extern VP8DecIdct2 VP8Transform;
337extern VP8DecIdct VP8TransformAC3;
338extern VP8DecIdct VP8TransformUV;
339extern VP8DecIdct VP8TransformDC;
340extern VP8DecIdct VP8TransformDCUV;
341extern VP8WHT VP8TransformWHT;
342
343// *dst is the destination block, with stride BPS. Boundary samples are
344// assumed accessible when needed.
345typedef void (*VP8PredFunc)(uint8_t* dst);
346extern VP8PredFunc VP8PredLuma16[/* NUM_B_DC_MODES */];
347extern VP8PredFunc VP8PredChroma8[/* NUM_B_DC_MODES */];
348extern VP8PredFunc VP8PredLuma4[/* NUM_BMODES */];
349
350// clipping tables (for filtering)
351extern const int8_t* const VP8ksclip1;  // clips [-1020, 1020] to [-128, 127]
352extern const int8_t* const VP8ksclip2;  // clips [-112, 112] to [-16, 15]
353extern const uint8_t* const VP8kclip1;  // clips [-255,511] to [0,255]
354extern const uint8_t* const VP8kabs0;   // abs(x) for x in [-255,255]
355// must be called first
356void VP8InitClipTables(void);
357
358// simple filter (only for luma)
359typedef void (*VP8SimpleFilterFunc)(uint8_t* p, int stride, int thresh);
360extern VP8SimpleFilterFunc VP8SimpleVFilter16;
361extern VP8SimpleFilterFunc VP8SimpleHFilter16;
362extern VP8SimpleFilterFunc VP8SimpleVFilter16i;  // filter 3 inner edges
363extern VP8SimpleFilterFunc VP8SimpleHFilter16i;
364
365// regular filter (on both macroblock edges and inner edges)
366typedef void (*VP8LumaFilterFunc)(uint8_t* luma, int stride,
367                                  int thresh, int ithresh, int hev_t);
368typedef void (*VP8ChromaFilterFunc)(uint8_t* u, uint8_t* v, int stride,
369                                    int thresh, int ithresh, int hev_t);
370// on outer edge
371extern VP8LumaFilterFunc VP8VFilter16;
372extern VP8LumaFilterFunc VP8HFilter16;
373extern VP8ChromaFilterFunc VP8VFilter8;
374extern VP8ChromaFilterFunc VP8HFilter8;
375
376// on inner edge
377extern VP8LumaFilterFunc VP8VFilter16i;   // filtering 3 inner edges altogether
378extern VP8LumaFilterFunc VP8HFilter16i;
379extern VP8ChromaFilterFunc VP8VFilter8i;  // filtering u and v altogether
380extern VP8ChromaFilterFunc VP8HFilter8i;
381
382// Dithering. Combines dithering values (centered around 128) with dst[],
383// according to: dst[] = clip(dst[] + (((dither[]-128) + 8) >> 4)
384#define VP8_DITHER_DESCALE 4
385#define VP8_DITHER_DESCALE_ROUNDER (1 << (VP8_DITHER_DESCALE - 1))
386#define VP8_DITHER_AMP_BITS 7
387#define VP8_DITHER_AMP_CENTER (1 << VP8_DITHER_AMP_BITS)
388extern void (*VP8DitherCombine8x8)(const uint8_t* dither, uint8_t* dst,
389                                   int dst_stride);
390
391// must be called before anything using the above
392void VP8DspInit(void);
393
394//------------------------------------------------------------------------------
395// WebP I/O
396
397#define FANCY_UPSAMPLING   // undefined to remove fancy upsampling support
398
399// Convert a pair of y/u/v lines together to the output rgb/a colorspace.
400// bottom_y can be NULL if only one line of output is needed (at top/bottom).
401typedef void (*WebPUpsampleLinePairFunc)(
402    const uint8_t* top_y, const uint8_t* bottom_y,
403    const uint8_t* top_u, const uint8_t* top_v,
404    const uint8_t* cur_u, const uint8_t* cur_v,
405    uint8_t* top_dst, uint8_t* bottom_dst, int len);
406
407#ifdef FANCY_UPSAMPLING
408
409// Fancy upsampling functions to convert YUV to RGB(A) modes
410extern WebPUpsampleLinePairFunc WebPUpsamplers[/* MODE_LAST */];
411
412#endif    // FANCY_UPSAMPLING
413
414// Per-row point-sampling methods.
415typedef void (*WebPSamplerRowFunc)(const uint8_t* y,
416                                   const uint8_t* u, const uint8_t* v,
417                                   uint8_t* dst, int len);
418// Generic function to apply 'WebPSamplerRowFunc' to the whole plane:
419void WebPSamplerProcessPlane(const uint8_t* y, int y_stride,
420                             const uint8_t* u, const uint8_t* v, int uv_stride,
421                             uint8_t* dst, int dst_stride,
422                             int width, int height, WebPSamplerRowFunc func);
423
424// Sampling functions to convert rows of YUV to RGB(A)
425extern WebPSamplerRowFunc WebPSamplers[/* MODE_LAST */];
426
427// General function for converting two lines of ARGB or RGBA.
428// 'alpha_is_last' should be true if 0xff000000 is stored in memory as
429// as 0x00, 0x00, 0x00, 0xff (little endian).
430WebPUpsampleLinePairFunc WebPGetLinePairConverter(int alpha_is_last);
431
432// YUV444->RGB converters
433typedef void (*WebPYUV444Converter)(const uint8_t* y,
434                                    const uint8_t* u, const uint8_t* v,
435                                    uint8_t* dst, int len);
436
437extern WebPYUV444Converter WebPYUV444Converters[/* MODE_LAST */];
438
439// Must be called before using the WebPUpsamplers[] (and for premultiplied
440// colorspaces like rgbA, rgbA4444, etc)
441void WebPInitUpsamplers(void);
442// Must be called before using WebPSamplers[]
443void WebPInitSamplers(void);
444// Must be called before using WebPYUV444Converters[]
445void WebPInitYUV444Converters(void);
446
447//------------------------------------------------------------------------------
448// ARGB -> YUV converters
449
450// Convert ARGB samples to luma Y.
451extern void (*WebPConvertARGBToY)(const uint32_t* argb, uint8_t* y, int width);
452// Convert ARGB samples to U/V with downsampling. do_store should be '1' for
453// even lines and '0' for odd ones. 'src_width' is the original width, not
454// the U/V one.
455extern void (*WebPConvertARGBToUV)(const uint32_t* argb, uint8_t* u, uint8_t* v,
456                                   int src_width, int do_store);
457
458// Convert a row of accumulated (four-values) of rgba32 toward U/V
459extern void (*WebPConvertRGBA32ToUV)(const uint16_t* rgb,
460                                     uint8_t* u, uint8_t* v, int width);
461
462// Convert RGB or BGR to Y
463extern void (*WebPConvertRGB24ToY)(const uint8_t* rgb, uint8_t* y, int width);
464extern void (*WebPConvertBGR24ToY)(const uint8_t* bgr, uint8_t* y, int width);
465
466// used for plain-C fallback.
467extern void WebPConvertARGBToUV_C(const uint32_t* argb, uint8_t* u, uint8_t* v,
468                                  int src_width, int do_store);
469extern void WebPConvertRGBA32ToUV_C(const uint16_t* rgb,
470                                    uint8_t* u, uint8_t* v, int width);
471
472// utilities for accurate RGB->YUV conversion
473extern uint64_t (*WebPSharpYUVUpdateY)(const uint16_t* src, const uint16_t* ref,
474                                       uint16_t* dst, int len);
475extern void (*WebPSharpYUVUpdateRGB)(const int16_t* src, const int16_t* ref,
476                                     int16_t* dst, int len);
477extern void (*WebPSharpYUVFilterRow)(const int16_t* A, const int16_t* B,
478                                     int len,
479                                     const uint16_t* best_y, uint16_t* out);
480
481// Must be called before using the above.
482void WebPInitConvertARGBToYUV(void);
483
484//------------------------------------------------------------------------------
485// Rescaler
486
487struct WebPRescaler;
488
489// Import a row of data and save its contribution in the rescaler.
490// 'channel' denotes the channel number to be imported. 'Expand' corresponds to
491// the wrk->x_expand case. Otherwise, 'Shrink' is to be used.
492typedef void (*WebPRescalerImportRowFunc)(struct WebPRescaler* const wrk,
493                                          const uint8_t* src);
494
495extern WebPRescalerImportRowFunc WebPRescalerImportRowExpand;
496extern WebPRescalerImportRowFunc WebPRescalerImportRowShrink;
497
498// Export one row (starting at x_out position) from rescaler.
499// 'Expand' corresponds to the wrk->y_expand case.
500// Otherwise 'Shrink' is to be used
501typedef void (*WebPRescalerExportRowFunc)(struct WebPRescaler* const wrk);
502extern WebPRescalerExportRowFunc WebPRescalerExportRowExpand;
503extern WebPRescalerExportRowFunc WebPRescalerExportRowShrink;
504
505// Plain-C implementation, as fall-back.
506extern void WebPRescalerImportRowExpand_C(struct WebPRescaler* const wrk,
507                                          const uint8_t* src);
508extern void WebPRescalerImportRowShrink_C(struct WebPRescaler* const wrk,
509                                          const uint8_t* src);
510extern void WebPRescalerExportRowExpand_C(struct WebPRescaler* const wrk);
511extern void WebPRescalerExportRowShrink_C(struct WebPRescaler* const wrk);
512
513// Main entry calls:
514extern void WebPRescalerImportRow(struct WebPRescaler* const wrk,
515                                  const uint8_t* src);
516// Export one row (starting at x_out position) from rescaler.
517extern void WebPRescalerExportRow(struct WebPRescaler* const wrk);
518
519// Must be called first before using the above.
520void WebPRescalerDspInit(void);
521
522//------------------------------------------------------------------------------
523// Utilities for processing transparent channel.
524
525// Apply alpha pre-multiply on an rgba, bgra or argb plane of size w * h.
526// alpha_first should be 0 for argb, 1 for rgba or bgra (where alpha is last).
527extern void (*WebPApplyAlphaMultiply)(
528    uint8_t* rgba, int alpha_first, int w, int h, int stride);
529
530// Same, buf specifically for RGBA4444 format
531extern void (*WebPApplyAlphaMultiply4444)(
532    uint8_t* rgba4444, int w, int h, int stride);
533
534// Dispatch the values from alpha[] plane to the ARGB destination 'dst'.
535// Returns true if alpha[] plane has non-trivial values different from 0xff.
536extern int (*WebPDispatchAlpha)(const uint8_t* alpha, int alpha_stride,
537                                int width, int height,
538                                uint8_t* dst, int dst_stride);
539
540// Transfer packed 8b alpha[] values to green channel in dst[], zero'ing the
541// A/R/B values. 'dst_stride' is the stride for dst[] in uint32_t units.
542extern void (*WebPDispatchAlphaToGreen)(const uint8_t* alpha, int alpha_stride,
543                                        int width, int height,
544                                        uint32_t* dst, int dst_stride);
545
546// Extract the alpha values from 32b values in argb[] and pack them into alpha[]
547// (this is the opposite of WebPDispatchAlpha).
548// Returns true if there's only trivial 0xff alpha values.
549extern int (*WebPExtractAlpha)(const uint8_t* argb, int argb_stride,
550                               int width, int height,
551                               uint8_t* alpha, int alpha_stride);
552
553// Extract the green values from 32b values in argb[] and pack them into alpha[]
554// (this is the opposite of WebPDispatchAlphaToGreen).
555extern void (*WebPExtractGreen)(const uint32_t* argb, uint8_t* alpha, int size);
556
557// Pre-Multiply operation transforms x into x * A / 255  (where x=Y,R,G or B).
558// Un-Multiply operation transforms x into x * 255 / A.
559
560// Pre-Multiply or Un-Multiply (if 'inverse' is true) argb values in a row.
561extern void (*WebPMultARGBRow)(uint32_t* const ptr, int width, int inverse);
562
563// Same a WebPMultARGBRow(), but for several rows.
564void WebPMultARGBRows(uint8_t* ptr, int stride, int width, int num_rows,
565                      int inverse);
566
567// Same for a row of single values, with side alpha values.
568extern void (*WebPMultRow)(uint8_t* const ptr, const uint8_t* const alpha,
569                           int width, int inverse);
570
571// Same a WebPMultRow(), but for several 'num_rows' rows.
572void WebPMultRows(uint8_t* ptr, int stride,
573                  const uint8_t* alpha, int alpha_stride,
574                  int width, int num_rows, int inverse);
575
576// Plain-C versions, used as fallback by some implementations.
577void WebPMultRow_C(uint8_t* const ptr, const uint8_t* const alpha,
578                   int width, int inverse);
579void WebPMultARGBRow_C(uint32_t* const ptr, int width, int inverse);
580
581// RGB packing function. 'step' can be 3 or 4. r/g/b input is rgb or bgr order.
582extern void (*WebPPackRGB)(const uint8_t* r, const uint8_t* g, const uint8_t* b,
583                           int len, int step, uint32_t* out);
584
585// This function returns true if src[i] contains a value different from 0xff.
586extern int (*WebPHasAlpha8b)(const uint8_t* src, int length);
587// This function returns true if src[4*i] contains a value different from 0xff.
588extern int (*WebPHasAlpha32b)(const uint8_t* src, int length);
589
590// To be called first before using the above.
591void WebPInitAlphaProcessing(void);
592
593//------------------------------------------------------------------------------
594// Filter functions
595
596typedef enum {     // Filter types.
597  WEBP_FILTER_NONE = 0,
598  WEBP_FILTER_HORIZONTAL,
599  WEBP_FILTER_VERTICAL,
600  WEBP_FILTER_GRADIENT,
601  WEBP_FILTER_LAST = WEBP_FILTER_GRADIENT + 1,  // end marker
602  WEBP_FILTER_BEST,    // meta-types
603  WEBP_FILTER_FAST
604} WEBP_FILTER_TYPE;
605
606typedef void (*WebPFilterFunc)(const uint8_t* in, int width, int height,
607                               int stride, uint8_t* out);
608// In-place un-filtering.
609// Warning! 'prev_line' pointer can be equal to 'cur_line' or 'preds'.
610typedef void (*WebPUnfilterFunc)(const uint8_t* prev_line, const uint8_t* preds,
611                                 uint8_t* cur_line, int width);
612
613// Filter the given data using the given predictor.
614// 'in' corresponds to a 2-dimensional pixel array of size (stride * height)
615// in raster order.
616// 'stride' is number of bytes per scan line (with possible padding).
617// 'out' should be pre-allocated.
618extern WebPFilterFunc WebPFilters[WEBP_FILTER_LAST];
619
620// In-place reconstruct the original data from the given filtered data.
621// The reconstruction will be done for 'num_rows' rows starting from 'row'
622// (assuming rows upto 'row - 1' are already reconstructed).
623extern WebPUnfilterFunc WebPUnfilters[WEBP_FILTER_LAST];
624
625// To be called first before using the above.
626void VP8FiltersInit(void);
627
628#ifdef __cplusplus
629}    // extern "C"
630#endif
631
632#endif  /* WEBP_DSP_DSP_H_ */
633