1 // Copyright 2014 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 // Author: Djordje Pesut (djordje.pesut@imgtec.com)
11 
12 #include "src/dsp/dsp.h"
13 
14 #if defined(WEBP_USE_MIPS_DSP_R2)
15 
16 #include "src/enc/cost_enc.h"
17 
GetResidualCost_MIPSdspR2(int ctx0,const VP8Residual * const res)18 static int GetResidualCost_MIPSdspR2(int ctx0, const VP8Residual* const res) {
19   int temp0, temp1;
20   int v_reg, ctx_reg;
21   int n = res->first;
22   // should be prob[VP8EncBands[n]], but it's equivalent for n=0 or 1
23   int p0 = res->prob[n][ctx0][0];
24   CostArrayPtr const costs = res->costs;
25   const uint16_t* t = costs[n][ctx0];
26   // bit_cost(1, p0) is already incorporated in t[] tables, but only if ctx != 0
27   // (as required by the syntax). For ctx0 == 0, we need to add it here or it'll
28   // be missing during the loop.
29   int cost = (ctx0 == 0) ? VP8BitCost(1, p0) : 0;
30   const int16_t* res_coeffs = res->coeffs;
31   const int res_last = res->last;
32   const int const_max_level = MAX_VARIABLE_LEVEL;
33   const int const_2 = 2;
34   const uint16_t** p_costs = &costs[n][0];
35   const size_t inc_p_costs = NUM_CTX * sizeof(*p_costs);
36 
37   if (res->last < 0) {
38     return VP8BitCost(0, p0);
39   }
40 
41   __asm__ volatile (
42     ".set      push                                                     \n\t"
43     ".set      noreorder                                                \n\t"
44     "subu      %[temp1],        %[res_last],        %[n]                \n\t"
45     "blez      %[temp1],        2f                                      \n\t"
46     " nop                                                               \n\t"
47   "1:                                                                   \n\t"
48     "sll       %[temp0],        %[n],               1                   \n\t"
49     "lhx       %[v_reg],        %[temp0](%[res_coeffs])                 \n\t"
50     "addiu     %[n],            %[n],               1                   \n\t"
51     "absq_s.w  %[v_reg],        %[v_reg]                                \n\t"
52     "sltiu     %[temp0],        %[v_reg],           2                   \n\t"
53     "move      %[ctx_reg],      %[v_reg]                                \n\t"
54     "movz      %[ctx_reg],      %[const_2],         %[temp0]            \n\t"
55     "sll       %[temp1],        %[v_reg],           1                   \n\t"
56     "lhx       %[temp1],        %[temp1](%[VP8LevelFixedCosts])         \n\t"
57     "slt       %[temp0],        %[v_reg],           %[const_max_level]  \n\t"
58     "movz      %[v_reg],        %[const_max_level], %[temp0]            \n\t"
59     "addu      %[cost],         %[cost],            %[temp1]            \n\t"
60     "sll       %[v_reg],        %[v_reg],           1                   \n\t"
61     "sll       %[ctx_reg],      %[ctx_reg],         2                   \n\t"
62     "lhx       %[temp0],        %[v_reg](%[t])                          \n\t"
63     "addu      %[p_costs],      %[p_costs],         %[inc_p_costs]      \n\t"
64     "addu      %[t],            %[p_costs],         %[ctx_reg]          \n\t"
65     "addu      %[cost],         %[cost],            %[temp0]            \n\t"
66     "bne       %[n],            %[res_last],        1b                  \n\t"
67     " lw       %[t],            0(%[t])                                 \n\t"
68   "2:                                                                   \n\t"
69     ".set      pop                                                      \n\t"
70     : [cost]"+&r"(cost), [t]"+&r"(t), [n]"+&r"(n), [v_reg]"=&r"(v_reg),
71       [ctx_reg]"=&r"(ctx_reg), [p_costs]"+&r"(p_costs), [temp0]"=&r"(temp0),
72       [temp1]"=&r"(temp1)
73     : [const_2]"r"(const_2), [const_max_level]"r"(const_max_level),
74       [VP8LevelFixedCosts]"r"(VP8LevelFixedCosts), [res_last]"r"(res_last),
75       [res_coeffs]"r"(res_coeffs), [inc_p_costs]"r"(inc_p_costs)
76     : "memory"
77   );
78 
79   // Last coefficient is always non-zero
80   {
81     const int v = abs(res->coeffs[n]);
82     assert(v != 0);
83     cost += VP8LevelCost(t, v);
84     if (n < 15) {
85       const int b = VP8EncBands[n + 1];
86       const int ctx = (v == 1) ? 1 : 2;
87       const int last_p0 = res->prob[b][ctx][0];
88       cost += VP8BitCost(0, last_p0);
89     }
90   }
91   return cost;
92 }
93 
94 //------------------------------------------------------------------------------
95 // Entry point
96 
97 extern void VP8EncDspCostInitMIPSdspR2(void);
98 
VP8EncDspCostInitMIPSdspR2(void)99 WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspCostInitMIPSdspR2(void) {
100   VP8GetResidualCost = GetResidualCost_MIPSdspR2;
101 }
102 
103 #else  // !WEBP_USE_MIPS_DSP_R2
104 
105 WEBP_DSP_INIT_STUB(VP8EncDspCostInitMIPSdspR2)
106 
107 #endif  // WEBP_USE_MIPS_DSP_R2
108