From 2b1b4d5ae924b04192a9dfc72889e331d0721b24 Mon Sep 17 00:00:00 2001 From: Slobodan Prijic Date: Tue, 25 Feb 2014 16:22:18 +0100 Subject: [PATCH] MIPS: MIPS32r1: Add optimization for GetResidualCost + reorganize the cost-evaluation code by moving some functions to cost.h/cost.c and exposing VP8Residual Change-Id: Id976299b5d4484e65da8bed31b3d2eb9cb4c1f7d --- src/dsp/enc_mips32.c | 114 ++++++++++++++++++ src/enc/cost.c | 224 ++++++++++++++++++++++++++++++++++ src/enc/cost.h | 26 ++++ src/enc/frame.c | 277 +++++-------------------------------------- src/enc/webpenc.c | 3 +- 5 files changed, 398 insertions(+), 246 deletions(-) diff --git a/src/dsp/enc_mips32.c b/src/dsp/enc_mips32.c index 02835043..a3b92498 100755 --- a/src/dsp/enc_mips32.c +++ b/src/dsp/enc_mips32.c @@ -17,6 +17,7 @@ #if defined(WEBP_USE_MIPS32) #include "../enc/vp8enci.h" +#include "../enc/cost.h" static const int kC1 = 20091 + (1 << 16); static const int kC2 = 35468; @@ -504,6 +505,119 @@ static void FTransformMIPS32(const uint8_t* src, const uint8_t* ref, #undef VERTICAL_PASS #undef HORIZONTAL_PASS +// Forward declaration. +extern int VP8GetResidualCostMIPS32(int ctx0, const VP8Residual* const res); + +int VP8GetResidualCostMIPS32(int ctx0, const VP8Residual* const res) { + int n = res->first; + // should be prob[VP8EncBands[n]], but it's equivalent for n=0 or 1 + int p0 = res->prob[n][ctx0][0]; + const uint16_t* t = res->cost[n][ctx0]; + int cost; + const int const_2 = 2; + const int const_255 = 255; + const int const_max_level = MAX_VARIABLE_LEVEL; + int res_cost; + int res_prob; + int res_coeffs; + int res_last; + int v_reg; + int b_reg; + int ctx_reg; + int cost_add, temp_1, temp_2, temp_3; + + if (res->last < 0) { + return VP8BitCost(0, p0); + } + + cost = (ctx0 == 0) ? VP8BitCost(1, p0) : 0; + + res_cost = (int)res->cost; + res_prob = (int)res->prob; + res_coeffs = (int)res->coeffs; + res_last = (int)res->last; + + __asm__ volatile( + ".set push \n\t" + ".set noreorder \n\t" + + "sll %[temp_1], %[n], 1 \n\t" + "addu %[res_coeffs], %[res_coeffs], %[temp_1] \n\t" + "slt %[temp_2], %[n], %[res_last] \n\t" + "bnez %[temp_2], 1f \n\t" + " li %[cost_add], 0 \n\t" + "b 2f \n\t" + " nop \n\t" + "1: \n\t" + "lh %[v_reg], 0(%[res_coeffs]) \n\t" + "addu %[b_reg], %[n], %[VP8EncBands] \n\t" + "move %[temp_1], %[const_max_level] \n\t" + "addu %[cost], %[cost], %[cost_add] \n\t" + "negu %[temp_2], %[v_reg] \n\t" + "slti %[temp_3], %[v_reg], 0 \n\t" + "movn %[v_reg], %[temp_2], %[temp_3] \n\t" + "lbu %[b_reg], 1(%[b_reg]) \n\t" + "li %[cost_add], 0 \n\t" + + "sltiu %[temp_3], %[v_reg], 2 \n\t" + "move %[ctx_reg], %[v_reg] \n\t" + "movz %[ctx_reg], %[const_2], %[temp_3] \n\t" + // cost += VP8LevelCost(t, v); + "slt %[temp_3], %[v_reg], %[const_max_level] \n\t" + "movn %[temp_1], %[v_reg], %[temp_3] \n\t" + "sll %[temp_2], %[v_reg], 1 \n\t" + "addu %[temp_2], %[temp_2], %[VP8LevelFixedCosts] \n\t" + "lhu %[temp_2], 0(%[temp_2]) \n\t" + "sll %[temp_1], %[temp_1], 1 \n\t" + "addu %[temp_1], %[temp_1], %[t] \n\t" + "lhu %[temp_3], 0(%[temp_1]) \n\t" + "addu %[cost], %[cost], %[temp_2] \n\t" + + // t = res->cost[b][ctx]; + "sll %[temp_1], %[ctx_reg], 7 \n\t" + "sll %[temp_2], %[ctx_reg], 3 \n\t" + "addu %[cost], %[cost], %[temp_3] \n\t" + "addu %[temp_1], %[temp_1], %[temp_2] \n\t" + "sll %[temp_2], %[b_reg], 3 \n\t" + "sll %[temp_3], %[b_reg], 5 \n\t" + "sub %[temp_2], %[temp_3], %[temp_2] \n\t" + "sll %[temp_3], %[temp_2], 4 \n\t" + "addu %[temp_1], %[temp_1], %[temp_3] \n\t" + "addu %[temp_2], %[temp_2], %[res_cost] \n\t" + "addiu %[n], %[n], 1 \n\t" + "addu %[t], %[temp_1], %[temp_2] \n\t" + "slt %[temp_1], %[n], %[res_last] \n\t" + "bnez %[temp_1], 1b \n\t" + " addiu %[res_coeffs], %[res_coeffs], 2 \n\t" + "2: \n\t" + + ".set pop \n\t" + : [cost]"+r"(cost), [t]"+r"(t), [n]"+r"(n), [v_reg]"=&r"(v_reg), + [ctx_reg]"=&r"(ctx_reg), [b_reg]"=&r"(b_reg), [cost_add]"=&r"(cost_add), + [temp_1]"=&r"(temp_1), [temp_2]"=&r"(temp_2), [temp_3]"=&r"(temp_3) + : [const_2]"r"(const_2), [const_255]"r"(const_255), [res_last]"r"(res_last), + [VP8EntropyCost]"r"(VP8EntropyCost), [VP8EncBands]"r"(VP8EncBands), + [const_max_level]"r"(const_max_level), [res_prob]"r"(res_prob), + [VP8LevelFixedCosts]"r"(VP8LevelFixedCosts), [res_coeffs]"r"(res_coeffs), + [res_cost]"r"(res_cost) + : "memory" + ); + + // Last coefficient is always non-zero + { + const int v = abs(res->coeffs[n]); + assert(v != 0); + cost += VP8LevelCost(t, v); + if (n < 15) { + const int b = VP8EncBands[n + 1]; + const int ctx = (v == 1) ? 1 : 2; + const int last_p0 = res->prob[b][ctx][0]; + cost += VP8BitCost(0, last_p0); + } + } + return cost; +} + #endif // WEBP_USE_MIPS32 //------------------------------------------------------------------------------ diff --git a/src/enc/cost.c b/src/enc/cost.c index 3155c084..9d6a490d 100644 --- a/src/enc/cost.c +++ b/src/enc/cost.c @@ -487,4 +487,228 @@ const uint16_t VP8FixedCostsI4[NUM_BMODES][NUM_BMODES][NUM_BMODES] = { }; //------------------------------------------------------------------------------ +// Mode costs +static int GetResidualCost(int ctx0, const VP8Residual* const res) { + int n = res->first; + // should be prob[VP8EncBands[n]], but it's equivalent for n=0 or 1 + const int p0 = res->prob[n][ctx0][0]; + const uint16_t* t = res->cost[n][ctx0]; + // bit_cost(1, p0) is already incorporated in t[] tables, but only if ctx != 0 + // (as required by the syntax). For ctx0 == 0, we need to add it here or it'll + // be missing during the loop. + int cost = (ctx0 == 0) ? VP8BitCost(1, p0) : 0; + + if (res->last < 0) { + return VP8BitCost(0, p0); + } + for (; n < res->last; ++n) { + const int v = abs(res->coeffs[n]); + const int b = VP8EncBands[n + 1]; + const int ctx = (v >= 2) ? 2 : v; + cost += VP8LevelCost(t, v); + t = res->cost[b][ctx]; + } + // Last coefficient is always non-zero + { + const int v = abs(res->coeffs[n]); + assert(v != 0); + cost += VP8LevelCost(t, v); + if (n < 15) { + const int b = VP8EncBands[n + 1]; + const int ctx = (v == 1) ? 1 : 2; + const int last_p0 = res->prob[b][ctx][0]; + cost += VP8BitCost(0, last_p0); + } + } + return cost; +} + +//------------------------------------------------------------------------------ +// init function + +#if defined(WEBP_USE_MIPS32) +extern int VP8GetResidualCostMIPS32(int ctx0, const VP8Residual* const res); +#endif // WEBP_USE_MIPS32 + +// TODO(skal): this, and GetResidualCost(), should probably go somewhere +// under src/dsp/ at some point. +VP8GetResidualCostFunc VP8GetResidualCost; + +void VP8GetResidualCostInit(void) { + if (VP8GetResidualCost == NULL) { + VP8GetResidualCost = GetResidualCost; + if (VP8GetCPUInfo != NULL) { +#if defined(WEBP_USE_MIPS32) + if (VP8GetCPUInfo(kMIPS32)) { + VP8GetResidualCost = VP8GetResidualCostMIPS32; + } +#endif + } + } +} + +//------------------------------------------------------------------------------ +// helper functions for residuals struct VP8Residual. + +void VP8InitResidual(int first, int coeff_type, + VP8Encoder* const enc, VP8Residual* const res) { + res->coeff_type = coeff_type; + res->prob = enc->proba_.coeffs_[coeff_type]; + res->stats = enc->proba_.stats_[coeff_type]; + res->cost = enc->proba_.level_cost_[coeff_type]; + res->first = first; +} + +void VP8SetResidualCoeffs(const int16_t* const coeffs, VP8Residual* const res) { + int n; + res->last = -1; + for (n = 15; n >= res->first; --n) { + if (coeffs[n]) { + res->last = n; + break; + } + } + res->coeffs = coeffs; +} + +//------------------------------------------------------------------------------ +// Mode costs + +int VP8GetCostLuma4(VP8EncIterator* const it, const int16_t levels[16]) { + const int x = (it->i4_ & 3), y = (it->i4_ >> 2); + VP8Residual res; + VP8Encoder* const enc = it->enc_; + int R = 0; + int ctx; + + VP8InitResidual(0, 3, enc, &res); + ctx = it->top_nz_[x] + it->left_nz_[y]; + VP8SetResidualCoeffs(levels, &res); + R += VP8GetResidualCost(ctx, &res); + return R; +} + +int VP8GetCostLuma16(VP8EncIterator* const it, const VP8ModeScore* const rd) { + VP8Residual res; + VP8Encoder* const enc = it->enc_; + int x, y; + int R = 0; + + VP8IteratorNzToBytes(it); // re-import the non-zero context + + // DC + VP8InitResidual(0, 1, enc, &res); + VP8SetResidualCoeffs(rd->y_dc_levels, &res); + R += VP8GetResidualCost(it->top_nz_[8] + it->left_nz_[8], &res); + + // AC + VP8InitResidual(1, 0, enc, &res); + for (y = 0; y < 4; ++y) { + for (x = 0; x < 4; ++x) { + const int ctx = it->top_nz_[x] + it->left_nz_[y]; + VP8SetResidualCoeffs(rd->y_ac_levels[x + y * 4], &res); + R += VP8GetResidualCost(ctx, &res); + it->top_nz_[x] = it->left_nz_[y] = (res.last >= 0); + } + } + return R; +} + +int VP8GetCostUV(VP8EncIterator* const it, const VP8ModeScore* const rd) { + VP8Residual res; + VP8Encoder* const enc = it->enc_; + int ch, x, y; + int R = 0; + + VP8IteratorNzToBytes(it); // re-import the non-zero context + + VP8InitResidual(0, 2, enc, &res); + for (ch = 0; ch <= 2; ch += 2) { + for (y = 0; y < 2; ++y) { + for (x = 0; x < 2; ++x) { + const int ctx = it->top_nz_[4 + ch + x] + it->left_nz_[4 + ch + y]; + VP8SetResidualCoeffs(rd->uv_levels[ch * 2 + x + y * 2], &res); + R += VP8GetResidualCost(ctx, &res); + it->top_nz_[4 + ch + x] = it->left_nz_[4 + ch + y] = (res.last >= 0); + } + } + } + return R; +} + + +//------------------------------------------------------------------------------ +// Recording of token probabilities. + +// Record proba context used +static int Record(int bit, proba_t* const stats) { + proba_t p = *stats; + if (p >= 0xffff0000u) { // an overflow is inbound. + p = ((p + 1u) >> 1) & 0x7fff7fffu; // -> divide the stats by 2. + } + // record bit count (lower 16 bits) and increment total count (upper 16 bits). + p += 0x00010000u + bit; + *stats = p; + return bit; +} + +// We keep the table-free variant around for reference, in case. +#define USE_LEVEL_CODE_TABLE + +// Simulate block coding, but only record statistics. +// Note: no need to record the fixed probas. +int VP8RecordCoeffs(int ctx, const VP8Residual* const res) { + int n = res->first; + // should be stats[VP8EncBands[n]], but it's equivalent for n=0 or 1 + proba_t* s = res->stats[n][ctx]; + if (res->last < 0) { + Record(0, s + 0); + return 0; + } + while (n <= res->last) { + int v; + Record(1, s + 0); // order of record doesn't matter + while ((v = res->coeffs[n++]) == 0) { + Record(0, s + 1); + s = res->stats[VP8EncBands[n]][0]; + } + Record(1, s + 1); + if (!Record(2u < (unsigned int)(v + 1), s + 2)) { // v = -1 or 1 + s = res->stats[VP8EncBands[n]][1]; + } else { + v = abs(v); +#if !defined(USE_LEVEL_CODE_TABLE) + if (!Record(v > 4, s + 3)) { + if (Record(v != 2, s + 4)) + Record(v == 4, s + 5); + } else if (!Record(v > 10, s + 6)) { + Record(v > 6, s + 7); + } else if (!Record((v >= 3 + (8 << 2)), s + 8)) { + Record((v >= 3 + (8 << 1)), s + 9); + } else { + Record((v >= 3 + (8 << 3)), s + 10); + } +#else + if (v > MAX_VARIABLE_LEVEL) { + v = MAX_VARIABLE_LEVEL; + } + + { + const int bits = VP8LevelCodes[v - 1][1]; + int pattern = VP8LevelCodes[v - 1][0]; + int i; + for (i = 0; (pattern >>= 1) != 0; ++i) { + const int mask = 2 << i; + if (pattern & 1) Record(!!(bits & mask), s + 3 + i); + } + } +#endif + s = res->stats[VP8EncBands[n]][2]; + } + } + if (n < 16) Record(0, s + 0); + return 1; +} + +//------------------------------------------------------------------------------ diff --git a/src/enc/cost.h b/src/enc/cost.h index 3cbad1ae..71fca855 100644 --- a/src/enc/cost.h +++ b/src/enc/cost.h @@ -14,12 +14,32 @@ #ifndef WEBP_ENC_COST_H_ #define WEBP_ENC_COST_H_ +#include +#include #include "./vp8enci.h" #ifdef __cplusplus extern "C" { #endif +// On-the-fly info about the current set of residuals. Handy to avoid +// passing zillions of params. +typedef struct { + int first; + int last; + const int16_t* coeffs; + + int coeff_type; + ProbaArray* prob; + StatsArray* stats; + CostArray* cost; +} VP8Residual; + +void VP8InitResidual(int first, int coeff_type, + VP8Encoder* const enc, VP8Residual* const res); +void VP8SetResidualCoeffs(const int16_t* const coeffs, VP8Residual* const res); +int VP8RecordCoeffs(int ctx, const VP8Residual* const res); + // approximate cost per level: extern const uint16_t VP8LevelFixedCosts[MAX_LEVEL + 1]; extern const uint16_t VP8EntropyCost[256]; // 8bit fixed-point log(p) @@ -29,6 +49,12 @@ static WEBP_INLINE int VP8BitCost(int bit, uint8_t proba) { return !bit ? VP8EntropyCost[proba] : VP8EntropyCost[255 - proba]; } +// Cost calculation function. +typedef int (*VP8GetResidualCostFunc)(int ctx0, const VP8Residual* const res); +extern VP8GetResidualCostFunc VP8GetResidualCost; + +extern void VP8GetResidualCostInit(void); // must be called first + // Level cost calculations extern const uint16_t VP8LevelCodes[MAX_VARIABLE_LEVEL][2]; void VP8CalculateLevelCosts(VP8Proba* const proba); diff --git a/src/enc/frame.c b/src/enc/frame.c index e4d8433f..3ef95303 100644 --- a/src/enc/frame.c +++ b/src/enc/frame.c @@ -11,8 +11,6 @@ // // Author: Skal (pascal.massimino@gmail.com) -#include -#include #include #include @@ -23,19 +21,6 @@ #define SEGMENT_VISU 0 #define DEBUG_SEARCH 0 // useful to track search convergence -// On-the-fly info about the current set of residuals. Handy to avoid -// passing zillions of params. -typedef struct { - int first; - int last; - const int16_t* coeffs; - - int coeff_type; - ProbaArray* prob; - StatsArray* stats; - CostArray* cost; -} VP8Residual; - //------------------------------------------------------------------------------ // multi-pass convergence @@ -142,84 +127,6 @@ static int FinalizeSkipProba(VP8Encoder* const enc) { return size; } -//------------------------------------------------------------------------------ -// Recording of token probabilities. - -static void ResetTokenStats(VP8Encoder* const enc) { - VP8Proba* const proba = &enc->proba_; - memset(proba->stats_, 0, sizeof(proba->stats_)); -} - -// Record proba context used -static int Record(int bit, proba_t* const stats) { - proba_t p = *stats; - if (p >= 0xffff0000u) { // an overflow is inbound. - p = ((p + 1u) >> 1) & 0x7fff7fffu; // -> divide the stats by 2. - } - // record bit count (lower 16 bits) and increment total count (upper 16 bits). - p += 0x00010000u + bit; - *stats = p; - return bit; -} - -// We keep the table-free variant around for reference, in case. -#define USE_LEVEL_CODE_TABLE - -// Simulate block coding, but only record statistics. -// Note: no need to record the fixed probas. -static int RecordCoeffs(int ctx, const VP8Residual* const res) { - int n = res->first; - // should be stats[VP8EncBands[n]], but it's equivalent for n=0 or 1 - proba_t* s = res->stats[n][ctx]; - if (res->last < 0) { - Record(0, s + 0); - return 0; - } - while (n <= res->last) { - int v; - Record(1, s + 0); // order of record doesn't matter - while ((v = res->coeffs[n++]) == 0) { - Record(0, s + 1); - s = res->stats[VP8EncBands[n]][0]; - } - Record(1, s + 1); - if (!Record(2u < (unsigned int)(v + 1), s + 2)) { // v = -1 or 1 - s = res->stats[VP8EncBands[n]][1]; - } else { - v = abs(v); -#if !defined(USE_LEVEL_CODE_TABLE) - if (!Record(v > 4, s + 3)) { - if (Record(v != 2, s + 4)) - Record(v == 4, s + 5); - } else if (!Record(v > 10, s + 6)) { - Record(v > 6, s + 7); - } else if (!Record((v >= 3 + (8 << 2)), s + 8)) { - Record((v >= 3 + (8 << 1)), s + 9); - } else { - Record((v >= 3 + (8 << 3)), s + 10); - } -#else - if (v > MAX_VARIABLE_LEVEL) { - v = MAX_VARIABLE_LEVEL; - } - - { - const int bits = VP8LevelCodes[v - 1][1]; - int pattern = VP8LevelCodes[v - 1][0]; - int i; - for (i = 0; (pattern >>= 1) != 0; ++i) { - const int mask = 2 << i; - if (pattern & 1) Record(!!(bits & mask), s + 3 + i); - } - } -#endif - s = res->stats[VP8EncBands[n]][2]; - } - } - if (n < 16) Record(0, s + 0); - return 1; -} - // Collect statistics and deduce probabilities for next coding pass. // Return the total bit-cost for coding the probability updates. static int CalcTokenProba(int nb, int total) { @@ -232,6 +139,11 @@ static int BranchCost(int nb, int total, int proba) { return nb * VP8BitCost(1, proba) + (total - nb) * VP8BitCost(0, proba); } +static void ResetTokenStats(VP8Encoder* const enc) { + VP8Proba* const proba = &enc->proba_; + memset(proba->stats_, 0, sizeof(proba->stats_)); +} + static int FinalizeTokenProbas(VP8Proba* const proba) { int has_changed = 0; int size = 0; @@ -309,131 +221,6 @@ static void SetSegmentProbas(VP8Encoder* const enc) { } } -//------------------------------------------------------------------------------ -// helper functions for residuals struct VP8Residual. - -static void InitResidual(int first, int coeff_type, - VP8Encoder* const enc, VP8Residual* const res) { - res->coeff_type = coeff_type; - res->prob = enc->proba_.coeffs_[coeff_type]; - res->stats = enc->proba_.stats_[coeff_type]; - res->cost = enc->proba_.level_cost_[coeff_type]; - res->first = first; -} - -static void SetResidualCoeffs(const int16_t* const coeffs, - VP8Residual* const res) { - int n; - res->last = -1; - for (n = 15; n >= res->first; --n) { - if (coeffs[n]) { - res->last = n; - break; - } - } - res->coeffs = coeffs; -} - -//------------------------------------------------------------------------------ -// Mode costs - -static int GetResidualCost(int ctx0, const VP8Residual* const res) { - int n = res->first; - // should be prob[VP8EncBands[n]], but it's equivalent for n=0 or 1 - const int p0 = res->prob[n][ctx0][0]; - const uint16_t* t = res->cost[n][ctx0]; - // bit_cost(1, p0) is already incorporated in t[] tables, but only if ctx != 0 - // (as required by the syntax). For ctx0 == 0, we need to add it here or it'll - // be missing during the loop. - int cost = (ctx0 == 0) ? VP8BitCost(1, p0) : 0; - - if (res->last < 0) { - return VP8BitCost(0, p0); - } - for (; n < res->last; ++n) { - const int v = abs(res->coeffs[n]); - const int b = VP8EncBands[n + 1]; - const int ctx = (v >= 2) ? 2 : v; - cost += VP8LevelCost(t, v); - t = res->cost[b][ctx]; - } - // Last coefficient is always non-zero - { - const int v = abs(res->coeffs[n]); - assert(v != 0); - cost += VP8LevelCost(t, v); - if (n < 15) { - const int b = VP8EncBands[n + 1]; - const int ctx = (v == 1) ? 1 : 2; - const int last_p0 = res->prob[b][ctx][0]; - cost += VP8BitCost(0, last_p0); - } - } - return cost; -} - -int VP8GetCostLuma4(VP8EncIterator* const it, const int16_t levels[16]) { - const int x = (it->i4_ & 3), y = (it->i4_ >> 2); - VP8Residual res; - VP8Encoder* const enc = it->enc_; - int R = 0; - int ctx; - - InitResidual(0, 3, enc, &res); - ctx = it->top_nz_[x] + it->left_nz_[y]; - SetResidualCoeffs(levels, &res); - R += GetResidualCost(ctx, &res); - return R; -} - -int VP8GetCostLuma16(VP8EncIterator* const it, const VP8ModeScore* const rd) { - VP8Residual res; - VP8Encoder* const enc = it->enc_; - int x, y; - int R = 0; - - VP8IteratorNzToBytes(it); // re-import the non-zero context - - // DC - InitResidual(0, 1, enc, &res); - SetResidualCoeffs(rd->y_dc_levels, &res); - R += GetResidualCost(it->top_nz_[8] + it->left_nz_[8], &res); - - // AC - InitResidual(1, 0, enc, &res); - for (y = 0; y < 4; ++y) { - for (x = 0; x < 4; ++x) { - const int ctx = it->top_nz_[x] + it->left_nz_[y]; - SetResidualCoeffs(rd->y_ac_levels[x + y * 4], &res); - R += GetResidualCost(ctx, &res); - it->top_nz_[x] = it->left_nz_[y] = (res.last >= 0); - } - } - return R; -} - -int VP8GetCostUV(VP8EncIterator* const it, const VP8ModeScore* const rd) { - VP8Residual res; - VP8Encoder* const enc = it->enc_; - int ch, x, y; - int R = 0; - - VP8IteratorNzToBytes(it); // re-import the non-zero context - - InitResidual(0, 2, enc, &res); - for (ch = 0; ch <= 2; ch += 2) { - for (y = 0; y < 2; ++y) { - for (x = 0; x < 2; ++x) { - const int ctx = it->top_nz_[4 + ch + x] + it->left_nz_[4 + ch + y]; - SetResidualCoeffs(rd->uv_levels[ch * 2 + x + y * 2], &res); - R += GetResidualCost(ctx, &res); - it->top_nz_[4 + ch + x] = it->left_nz_[4 + ch + y] = (res.last >= 0); - } - } - } - return R; -} - //------------------------------------------------------------------------------ // Coefficient coding @@ -522,32 +309,32 @@ static void CodeResiduals(VP8BitWriter* const bw, VP8EncIterator* const it, pos1 = VP8BitWriterPos(bw); if (i16) { - InitResidual(0, 1, enc, &res); - SetResidualCoeffs(rd->y_dc_levels, &res); + VP8InitResidual(0, 1, enc, &res); + VP8SetResidualCoeffs(rd->y_dc_levels, &res); it->top_nz_[8] = it->left_nz_[8] = PutCoeffs(bw, it->top_nz_[8] + it->left_nz_[8], &res); - InitResidual(1, 0, enc, &res); + VP8InitResidual(1, 0, enc, &res); } else { - InitResidual(0, 3, enc, &res); + VP8InitResidual(0, 3, enc, &res); } // luma-AC for (y = 0; y < 4; ++y) { for (x = 0; x < 4; ++x) { const int ctx = it->top_nz_[x] + it->left_nz_[y]; - SetResidualCoeffs(rd->y_ac_levels[x + y * 4], &res); + VP8SetResidualCoeffs(rd->y_ac_levels[x + y * 4], &res); it->top_nz_[x] = it->left_nz_[y] = PutCoeffs(bw, ctx, &res); } } pos2 = VP8BitWriterPos(bw); // U/V - InitResidual(0, 2, enc, &res); + VP8InitResidual(0, 2, enc, &res); for (ch = 0; ch <= 2; ch += 2) { for (y = 0; y < 2; ++y) { for (x = 0; x < 2; ++x) { const int ctx = it->top_nz_[4 + ch + x] + it->left_nz_[4 + ch + y]; - SetResidualCoeffs(rd->uv_levels[ch * 2 + x + y * 2], &res); + VP8SetResidualCoeffs(rd->uv_levels[ch * 2 + x + y * 2], &res); it->top_nz_[4 + ch + x] = it->left_nz_[4 + ch + y] = PutCoeffs(bw, ctx, &res); } @@ -572,33 +359,33 @@ static void RecordResiduals(VP8EncIterator* const it, VP8IteratorNzToBytes(it); if (it->mb_->type_ == 1) { // i16x16 - InitResidual(0, 1, enc, &res); - SetResidualCoeffs(rd->y_dc_levels, &res); + VP8InitResidual(0, 1, enc, &res); + VP8SetResidualCoeffs(rd->y_dc_levels, &res); it->top_nz_[8] = it->left_nz_[8] = - RecordCoeffs(it->top_nz_[8] + it->left_nz_[8], &res); - InitResidual(1, 0, enc, &res); + VP8RecordCoeffs(it->top_nz_[8] + it->left_nz_[8], &res); + VP8InitResidual(1, 0, enc, &res); } else { - InitResidual(0, 3, enc, &res); + VP8InitResidual(0, 3, enc, &res); } // luma-AC for (y = 0; y < 4; ++y) { for (x = 0; x < 4; ++x) { const int ctx = it->top_nz_[x] + it->left_nz_[y]; - SetResidualCoeffs(rd->y_ac_levels[x + y * 4], &res); - it->top_nz_[x] = it->left_nz_[y] = RecordCoeffs(ctx, &res); + VP8SetResidualCoeffs(rd->y_ac_levels[x + y * 4], &res); + it->top_nz_[x] = it->left_nz_[y] = VP8RecordCoeffs(ctx, &res); } } // U/V - InitResidual(0, 2, enc, &res); + VP8InitResidual(0, 2, enc, &res); for (ch = 0; ch <= 2; ch += 2) { for (y = 0; y < 2; ++y) { for (x = 0; x < 2; ++x) { const int ctx = it->top_nz_[4 + ch + x] + it->left_nz_[4 + ch + y]; - SetResidualCoeffs(rd->uv_levels[ch * 2 + x + y * 2], &res); + VP8SetResidualCoeffs(rd->uv_levels[ch * 2 + x + y * 2], &res); it->top_nz_[4 + ch + x] = it->left_nz_[4 + ch + y] = - RecordCoeffs(ctx, &res); + VP8RecordCoeffs(ctx, &res); } } } @@ -620,40 +407,40 @@ static void RecordTokens(VP8EncIterator* const it, const VP8ModeScore* const rd, VP8IteratorNzToBytes(it); if (it->mb_->type_ == 1) { // i16x16 const int ctx = it->top_nz_[8] + it->left_nz_[8]; - InitResidual(0, 1, enc, &res); - SetResidualCoeffs(rd->y_dc_levels, &res); + VP8InitResidual(0, 1, enc, &res); + VP8SetResidualCoeffs(rd->y_dc_levels, &res); it->top_nz_[8] = it->left_nz_[8] = VP8RecordCoeffTokens(ctx, 1, res.first, res.last, res.coeffs, tokens); - RecordCoeffs(ctx, &res); - InitResidual(1, 0, enc, &res); + VP8RecordCoeffs(ctx, &res); + VP8InitResidual(1, 0, enc, &res); } else { - InitResidual(0, 3, enc, &res); + VP8InitResidual(0, 3, enc, &res); } // luma-AC for (y = 0; y < 4; ++y) { for (x = 0; x < 4; ++x) { const int ctx = it->top_nz_[x] + it->left_nz_[y]; - SetResidualCoeffs(rd->y_ac_levels[x + y * 4], &res); + VP8SetResidualCoeffs(rd->y_ac_levels[x + y * 4], &res); it->top_nz_[x] = it->left_nz_[y] = VP8RecordCoeffTokens(ctx, res.coeff_type, res.first, res.last, res.coeffs, tokens); - RecordCoeffs(ctx, &res); + VP8RecordCoeffs(ctx, &res); } } // U/V - InitResidual(0, 2, enc, &res); + VP8InitResidual(0, 2, enc, &res); for (ch = 0; ch <= 2; ch += 2) { for (y = 0; y < 2; ++y) { for (x = 0; x < 2; ++x) { const int ctx = it->top_nz_[4 + ch + x] + it->left_nz_[4 + ch + y]; - SetResidualCoeffs(rd->uv_levels[ch * 2 + x + y * 2], &res); + VP8SetResidualCoeffs(rd->uv_levels[ch * 2 + x + y * 2], &res); it->top_nz_[4 + ch + x] = it->left_nz_[4 + ch + y] = VP8RecordCoeffTokens(ctx, 2, res.first, res.last, res.coeffs, tokens); - RecordCoeffs(ctx, &res); + VP8RecordCoeffs(ctx, &res); } } } diff --git a/src/enc/webpenc.c b/src/enc/webpenc.c index 79b03420..177a6e28 100644 --- a/src/enc/webpenc.c +++ b/src/enc/webpenc.c @@ -18,6 +18,7 @@ #include "./vp8enci.h" #include "./vp8li.h" +#include "./cost.h" #include "../utils/utils.h" // #define PRINT_MEMORY_INFO @@ -251,7 +252,7 @@ static VP8Encoder* InitVP8Encoder(const WebPConfig* const config, ResetSegmentHeader(enc); ResetFilterHeader(enc); ResetBoundaryPredictions(enc); - + VP8GetResidualCostInit(); VP8EncInitAlpha(enc); #ifdef WEBP_EXPERIMENTAL_FEATURES VP8EncInitLayer(enc);