• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 // MIPS version of speed-critical encoding functions.
11 //
12 // Author(s): Darko Laus (darko.laus@imgtec.com)
13 //            Mirko Raus (mirko.raus@imgtec.com)
14 
15 #include "src/dsp/dsp.h"
16 
17 #if defined(WEBP_USE_MIPS_DSP_R2)
18 
19 #include "src/dsp/mips_macro.h"
20 #include "src/enc/cost_enc.h"
21 #include "src/enc/vp8i_enc.h"
22 
23 static const int kC1 = 20091 + (1 << 16);
24 static const int kC2 = 35468;
25 
26 // O - output
27 // I - input (macro doesn't change it)
28 #define ADD_SUB_HALVES_X4(O0, O1, O2, O3, O4, O5, O6, O7,                      \
29                           I0, I1, I2, I3, I4, I5, I6, I7)                      \
30   "addq.ph          %[" #O0 "],   %[" #I0 "],  %[" #I1 "]     \n\t"            \
31   "subq.ph          %[" #O1 "],   %[" #I0 "],  %[" #I1 "]     \n\t"            \
32   "addq.ph          %[" #O2 "],   %[" #I2 "],  %[" #I3 "]     \n\t"            \
33   "subq.ph          %[" #O3 "],   %[" #I2 "],  %[" #I3 "]     \n\t"            \
34   "addq.ph          %[" #O4 "],   %[" #I4 "],  %[" #I5 "]     \n\t"            \
35   "subq.ph          %[" #O5 "],   %[" #I4 "],  %[" #I5 "]     \n\t"            \
36   "addq.ph          %[" #O6 "],   %[" #I6 "],  %[" #I7 "]     \n\t"            \
37   "subq.ph          %[" #O7 "],   %[" #I6 "],  %[" #I7 "]     \n\t"
38 
39 // IO - input/output
40 #define ABS_X8(IO0, IO1, IO2, IO3, IO4, IO5, IO6, IO7)                         \
41   "absq_s.ph        %[" #IO0 "],   %[" #IO0 "]                \n\t"            \
42   "absq_s.ph        %[" #IO1 "],   %[" #IO1 "]                \n\t"            \
43   "absq_s.ph        %[" #IO2 "],   %[" #IO2 "]                \n\t"            \
44   "absq_s.ph        %[" #IO3 "],   %[" #IO3 "]                \n\t"            \
45   "absq_s.ph        %[" #IO4 "],   %[" #IO4 "]                \n\t"            \
46   "absq_s.ph        %[" #IO5 "],   %[" #IO5 "]                \n\t"            \
47   "absq_s.ph        %[" #IO6 "],   %[" #IO6 "]                \n\t"            \
48   "absq_s.ph        %[" #IO7 "],   %[" #IO7 "]                \n\t"
49 
50 // dpa.w.ph $ac0 temp0 ,temp1
51 //  $ac += temp0[31..16] * temp1[31..16] + temp0[15..0] * temp1[15..0]
52 // dpax.w.ph $ac0 temp0 ,temp1
53 //  $ac += temp0[31..16] * temp1[15..0] + temp0[15..0] * temp1[31..16]
54 // O - output
55 // I - input (macro doesn't change it)
56 #define MUL_HALF(O0, I0, I1, I2, I3, I4, I5, I6, I7,                           \
57                  I8, I9, I10, I11, I12, I13, I14, I15)                         \
58     "mult            $ac0,      $zero,     $zero              \n\t"            \
59     "dpa.w.ph        $ac0,      %[" #I2 "],  %[" #I0 "]       \n\t"            \
60     "dpax.w.ph       $ac0,      %[" #I5 "],  %[" #I6 "]       \n\t"            \
61     "dpa.w.ph        $ac0,      %[" #I8 "],  %[" #I9 "]       \n\t"            \
62     "dpax.w.ph       $ac0,      %[" #I11 "], %[" #I4 "]       \n\t"            \
63     "dpa.w.ph        $ac0,      %[" #I12 "], %[" #I7 "]       \n\t"            \
64     "dpax.w.ph       $ac0,      %[" #I13 "], %[" #I1 "]       \n\t"            \
65     "dpa.w.ph        $ac0,      %[" #I14 "], %[" #I3 "]       \n\t"            \
66     "dpax.w.ph       $ac0,      %[" #I15 "], %[" #I10 "]      \n\t"            \
67     "mflo            %[" #O0 "],  $ac0                        \n\t"
68 
69 #define OUTPUT_EARLY_CLOBBER_REGS_17()                                         \
70   OUTPUT_EARLY_CLOBBER_REGS_10(),                                              \
71   [temp11]"=&r"(temp11), [temp12]"=&r"(temp12), [temp13]"=&r"(temp13),         \
72   [temp14]"=&r"(temp14), [temp15]"=&r"(temp15), [temp16]"=&r"(temp16),         \
73   [temp17]"=&r"(temp17)
74 
75 // macro for one horizontal pass in FTransform
76 // temp0..temp15 holds tmp[0]..tmp[15]
77 // A - offset in bytes to load from src and ref buffers
78 // TEMP0..TEMP3 - registers for corresponding tmp elements
79 #define HORIZONTAL_PASS(A, TEMP0, TEMP1, TEMP2, TEMP3)                         \
80   "lw              %[" #TEMP0 "],   0(%[args])                          \n\t"  \
81   "lw              %[" #TEMP1 "],   4(%[args])                          \n\t"  \
82   "lw              %[" #TEMP2 "],   " XSTR(BPS) "*" #A "(%[" #TEMP0 "]) \n\t"  \
83   "lw              %[" #TEMP3 "],   " XSTR(BPS) "*" #A "(%[" #TEMP1 "]) \n\t"  \
84   "preceu.ph.qbl   %[" #TEMP0 "],   %[" #TEMP2 "]                       \n\t"  \
85   "preceu.ph.qbl   %[" #TEMP1 "],   %[" #TEMP3 "]                       \n\t"  \
86   "preceu.ph.qbr   %[" #TEMP2 "],   %[" #TEMP2 "]                       \n\t"  \
87   "preceu.ph.qbr   %[" #TEMP3 "],   %[" #TEMP3 "]                       \n\t"  \
88   "subq.ph         %[" #TEMP0 "],   %[" #TEMP0 "],   %[" #TEMP1 "]      \n\t"  \
89   "subq.ph         %[" #TEMP2 "],   %[" #TEMP2 "],   %[" #TEMP3 "]      \n\t"  \
90   "rotr            %[" #TEMP0 "],   %[" #TEMP0 "],   16                 \n\t"  \
91   "addq.ph         %[" #TEMP1 "],   %[" #TEMP2 "],   %[" #TEMP0 "]      \n\t"  \
92   "subq.ph         %[" #TEMP3 "],   %[" #TEMP2 "],   %[" #TEMP0 "]      \n\t"  \
93   "seh             %[" #TEMP0 "],   %[" #TEMP1 "]                       \n\t"  \
94   "sra             %[temp16],     %[" #TEMP1 "],   16                   \n\t"  \
95   "seh             %[temp19],     %[" #TEMP3 "]                         \n\t"  \
96   "sra             %[" #TEMP3 "],   %[" #TEMP3 "],   16                 \n\t"  \
97   "subu            %[" #TEMP2 "],   %[" #TEMP0 "],   %[temp16]          \n\t"  \
98   "addu            %[" #TEMP0 "],   %[" #TEMP0 "],   %[temp16]          \n\t"  \
99   "mul             %[temp17],     %[temp19],     %[c2217]               \n\t"  \
100   "mul             %[temp18],     %[" #TEMP3 "],   %[c5352]             \n\t"  \
101   "mul             %[" #TEMP1 "],   %[temp19],     %[c5352]             \n\t"  \
102   "mul             %[temp16],     %[" #TEMP3 "],   %[c2217]             \n\t"  \
103   "sll             %[" #TEMP2 "],   %[" #TEMP2 "],   3                  \n\t"  \
104   "sll             %[" #TEMP0 "],   %[" #TEMP0 "],   3                  \n\t"  \
105   "subu            %[" #TEMP3 "],   %[temp17],     %[temp18]            \n\t"  \
106   "addu            %[" #TEMP1 "],   %[temp16],     %[" #TEMP1 "]        \n\t"  \
107   "addiu           %[" #TEMP3 "],   %[" #TEMP3 "],   937                \n\t"  \
108   "addiu           %[" #TEMP1 "],   %[" #TEMP1 "],   1812               \n\t"  \
109   "sra             %[" #TEMP3 "],   %[" #TEMP3 "],   9                  \n\t"  \
110   "sra             %[" #TEMP1 "],   %[" #TEMP1 "],   9                  \n\t"
111 
112 // macro for one vertical pass in FTransform
113 // temp0..temp15 holds tmp[0]..tmp[15]
114 // A..D - offsets in bytes to store to out buffer
115 // TEMP0, TEMP4, TEMP8 and TEMP12 - registers for corresponding tmp elements
116 #define VERTICAL_PASS(A, B, C, D, TEMP0, TEMP4, TEMP8, TEMP12)                 \
117   "addu            %[temp16],     %[" #TEMP0 "],   %[" #TEMP12 "]   \n\t"      \
118   "subu            %[temp19],     %[" #TEMP0 "],   %[" #TEMP12 "]   \n\t"      \
119   "addu            %[temp17],     %[" #TEMP4 "],   %[" #TEMP8 "]    \n\t"      \
120   "subu            %[temp18],     %[" #TEMP4 "],   %[" #TEMP8 "]    \n\t"      \
121   "mul             %[" #TEMP8 "],   %[temp19],     %[c2217]         \n\t"      \
122   "mul             %[" #TEMP12 "],  %[temp18],     %[c2217]         \n\t"      \
123   "mul             %[" #TEMP4 "],   %[temp19],     %[c5352]         \n\t"      \
124   "mul             %[temp18],     %[temp18],     %[c5352]           \n\t"      \
125   "addiu           %[temp16],     %[temp16],     7                  \n\t"      \
126   "addu            %[" #TEMP0 "],   %[temp16],     %[temp17]        \n\t"      \
127   "sra             %[" #TEMP0 "],   %[" #TEMP0 "],   4              \n\t"      \
128   "addu            %[" #TEMP12 "],  %[" #TEMP12 "],  %[" #TEMP4 "]  \n\t"      \
129   "subu            %[" #TEMP4 "],   %[temp16],     %[temp17]        \n\t"      \
130   "sra             %[" #TEMP4 "],   %[" #TEMP4 "],   4              \n\t"      \
131   "addiu           %[" #TEMP8 "],   %[" #TEMP8 "],   30000          \n\t"      \
132   "addiu           %[" #TEMP12 "],  %[" #TEMP12 "],  12000          \n\t"      \
133   "addiu           %[" #TEMP8 "],   %[" #TEMP8 "],   21000          \n\t"      \
134   "subu            %[" #TEMP8 "],   %[" #TEMP8 "],   %[temp18]      \n\t"      \
135   "sra             %[" #TEMP12 "],  %[" #TEMP12 "],  16             \n\t"      \
136   "sra             %[" #TEMP8 "],   %[" #TEMP8 "],   16             \n\t"      \
137   "addiu           %[temp16],     %[" #TEMP12 "],  1                \n\t"      \
138   "movn            %[" #TEMP12 "],  %[temp16],     %[temp19]        \n\t"      \
139   "sh              %[" #TEMP0 "],   " #A "(%[temp20])               \n\t"      \
140   "sh              %[" #TEMP4 "],   " #C "(%[temp20])               \n\t"      \
141   "sh              %[" #TEMP8 "],   " #D "(%[temp20])               \n\t"      \
142   "sh              %[" #TEMP12 "],  " #B "(%[temp20])               \n\t"
143 
FTransform_MIPSdspR2(const uint8_t * src,const uint8_t * ref,int16_t * out)144 static void FTransform_MIPSdspR2(const uint8_t* src, const uint8_t* ref,
145                                  int16_t* out) {
146   const int c2217 = 2217;
147   const int c5352 = 5352;
148   int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8;
149   int temp9, temp10, temp11, temp12, temp13, temp14, temp15, temp16;
150   int temp17, temp18, temp19, temp20;
151   const int* const args[3] =
152       { (const int*)src, (const int*)ref, (const int*)out };
153 
154   __asm__ volatile (
155     HORIZONTAL_PASS(0, temp0,  temp1,  temp2,  temp3)
156     HORIZONTAL_PASS(1, temp4,  temp5,  temp6,  temp7)
157     HORIZONTAL_PASS(2, temp8,  temp9,  temp10, temp11)
158     HORIZONTAL_PASS(3, temp12, temp13, temp14, temp15)
159     "lw            %[temp20],     8(%[args])                  \n\t"
160     VERTICAL_PASS(0,  8, 16, 24, temp0, temp4, temp8,  temp12)
161     VERTICAL_PASS(2, 10, 18, 26, temp1, temp5, temp9,  temp13)
162     VERTICAL_PASS(4, 12, 20, 28, temp2, temp6, temp10, temp14)
163     VERTICAL_PASS(6, 14, 22, 30, temp3, temp7, temp11, temp15)
164     OUTPUT_EARLY_CLOBBER_REGS_18(),
165       [temp0]"=&r"(temp0), [temp19]"=&r"(temp19), [temp20]"=&r"(temp20)
166     : [args]"r"(args), [c2217]"r"(c2217), [c5352]"r"(c5352)
167     : "memory", "hi", "lo"
168   );
169 }
170 
171 #undef VERTICAL_PASS
172 #undef HORIZONTAL_PASS
173 
ITransformOne(const uint8_t * ref,const int16_t * in,uint8_t * dst)174 static WEBP_INLINE void ITransformOne(const uint8_t* ref, const int16_t* in,
175                                       uint8_t* dst) {
176   int temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8, temp9;
177   int temp10, temp11, temp12, temp13, temp14, temp15, temp16, temp17, temp18;
178 
179   __asm__ volatile (
180     "ulw              %[temp1],   0(%[in])                 \n\t"
181     "ulw              %[temp2],   16(%[in])                \n\t"
182     LOAD_IN_X2(temp5, temp6, 24, 26)
183     ADD_SUB_HALVES(temp3, temp4, temp1, temp2)
184     LOAD_IN_X2(temp1, temp2, 8, 10)
185     MUL_SHIFT_SUM(temp7, temp8, temp9, temp10, temp11, temp12, temp13, temp14,
186                   temp10, temp8, temp9, temp7, temp1, temp2, temp5, temp6,
187                   temp13, temp11, temp14, temp12)
188     INSERT_HALF_X2(temp8, temp7, temp10, temp9)
189     "ulw              %[temp17],  4(%[in])                 \n\t"
190     "ulw              %[temp18],  20(%[in])                \n\t"
191     ADD_SUB_HALVES(temp1, temp2, temp3, temp8)
192     ADD_SUB_HALVES(temp5, temp6, temp4, temp7)
193     ADD_SUB_HALVES(temp7, temp8, temp17, temp18)
194     LOAD_IN_X2(temp17, temp18, 12, 14)
195     LOAD_IN_X2(temp9, temp10, 28, 30)
196     MUL_SHIFT_SUM(temp11, temp12, temp13, temp14, temp15, temp16, temp4, temp17,
197                   temp12, temp14, temp11, temp13, temp17, temp18, temp9, temp10,
198                   temp15, temp4, temp16, temp17)
199     INSERT_HALF_X2(temp11, temp12, temp13, temp14)
200     ADD_SUB_HALVES(temp17, temp8, temp8, temp11)
201     ADD_SUB_HALVES(temp3, temp4, temp7, temp12)
202 
203     // horizontal
204     SRA_16(temp9, temp10, temp11, temp12, temp1, temp2, temp5, temp6)
205     INSERT_HALF_X2(temp1, temp6, temp5, temp2)
206     SRA_16(temp13, temp14, temp15, temp16, temp3, temp4, temp17, temp8)
207     "repl.ph          %[temp2],   0x4                      \n\t"
208     INSERT_HALF_X2(temp3, temp8, temp17, temp4)
209     "addq.ph          %[temp1],   %[temp1],  %[temp2]      \n\t"
210     "addq.ph          %[temp6],   %[temp6],  %[temp2]      \n\t"
211     ADD_SUB_HALVES(temp2, temp4, temp1, temp3)
212     ADD_SUB_HALVES(temp5, temp7, temp6, temp8)
213     MUL_SHIFT_SUM(temp1, temp3, temp6, temp8, temp9, temp13, temp17, temp18,
214                   temp3, temp13, temp1, temp9, temp9, temp13, temp11, temp15,
215                   temp6, temp17, temp8, temp18)
216     MUL_SHIFT_SUM(temp6, temp8, temp18, temp17, temp11, temp15, temp12, temp16,
217                   temp8, temp15, temp6, temp11, temp12, temp16, temp10, temp14,
218                   temp18, temp12, temp17, temp16)
219     INSERT_HALF_X2(temp1, temp3, temp9, temp13)
220     INSERT_HALF_X2(temp6, temp8, temp11, temp15)
221     SHIFT_R_SUM_X2(temp9, temp10, temp11, temp12, temp13, temp14, temp15,
222                    temp16, temp2, temp4, temp5, temp7, temp3, temp1, temp8,
223                    temp6)
224     PACK_2_HALVES_TO_WORD(temp1, temp2, temp3, temp4, temp9, temp12, temp13,
225                           temp16, temp11, temp10, temp15, temp14)
226     LOAD_WITH_OFFSET_X4(temp10, temp11, temp14, temp15, ref,
227                         0, 0, 0, 0,
228                         0, 1, 2, 3,
229                         BPS)
230     CONVERT_2_BYTES_TO_HALF(temp5, temp6, temp7, temp8, temp17, temp18, temp10,
231                             temp11, temp10, temp11, temp14, temp15)
232     STORE_SAT_SUM_X2(temp5, temp6, temp7, temp8, temp17, temp18, temp10, temp11,
233                      temp9, temp12, temp1, temp2, temp13, temp16, temp3, temp4,
234                      dst, 0, 1, 2, 3, BPS)
235 
236     OUTPUT_EARLY_CLOBBER_REGS_18()
237     : [dst]"r"(dst), [in]"r"(in), [kC1]"r"(kC1), [kC2]"r"(kC2), [ref]"r"(ref)
238     : "memory", "hi", "lo"
239   );
240 }
241 
ITransform_MIPSdspR2(const uint8_t * ref,const int16_t * in,uint8_t * dst,int do_two)242 static void ITransform_MIPSdspR2(const uint8_t* ref, const int16_t* in,
243                                  uint8_t* dst, int do_two) {
244   ITransformOne(ref, in, dst);
245   if (do_two) {
246     ITransformOne(ref + 4, in + 16, dst + 4);
247   }
248 }
249 
Disto4x4_MIPSdspR2(const uint8_t * const a,const uint8_t * const b,const uint16_t * const w)250 static int Disto4x4_MIPSdspR2(const uint8_t* const a, const uint8_t* const b,
251                               const uint16_t* const w) {
252   int temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8, temp9;
253   int temp10, temp11, temp12, temp13, temp14, temp15, temp16, temp17;
254 
255   __asm__ volatile (
256     LOAD_WITH_OFFSET_X4(temp1, temp2, temp3, temp4, a,
257                         0, 0, 0, 0,
258                         0, 1, 2, 3,
259                         BPS)
260     CONVERT_2_BYTES_TO_HALF(temp5, temp6, temp7, temp8, temp9,temp10, temp11,
261                             temp12, temp1, temp2, temp3, temp4)
262     ADD_SUB_HALVES_X4(temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8,
263                       temp5, temp6, temp7, temp8, temp9, temp10, temp11, temp12)
264     PACK_2_HALVES_TO_WORD(temp9, temp10, temp11, temp12, temp1, temp3, temp5,
265                           temp7, temp2, temp4, temp6, temp8)
266     ADD_SUB_HALVES_X4(temp2, temp4, temp6, temp8, temp9, temp1, temp3, temp10,
267                       temp1, temp9, temp3, temp10, temp5, temp11, temp7, temp12)
268     ADD_SUB_HALVES_X4(temp5, temp11, temp7, temp2, temp9, temp3, temp6, temp12,
269                       temp2, temp9, temp6, temp3, temp4, temp1, temp8, temp10)
270     ADD_SUB_HALVES_X4(temp1, temp4, temp10, temp8, temp7, temp11, temp5, temp2,
271                       temp5, temp7, temp11, temp2, temp9, temp6, temp3, temp12)
272     ABS_X8(temp1, temp4, temp10, temp8, temp7, temp11, temp5, temp2)
273     LOAD_WITH_OFFSET_X4(temp3, temp6, temp9, temp12, w,
274                         0, 4, 8, 12,
275                         0, 0, 0, 0,
276                         0)
277     LOAD_WITH_OFFSET_X4(temp13, temp14, temp15, temp16, w,
278                         0, 4, 8, 12,
279                         1, 1, 1, 1,
280                         16)
281     MUL_HALF(temp17, temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8,
282              temp9, temp10, temp11, temp12, temp13, temp14, temp15, temp16)
283     LOAD_WITH_OFFSET_X4(temp1, temp2, temp3, temp4, b,
284                         0, 0, 0, 0,
285                         0, 1, 2, 3,
286                         BPS)
287     CONVERT_2_BYTES_TO_HALF(temp5,temp6, temp7, temp8, temp9,temp10, temp11,
288                             temp12, temp1, temp2, temp3, temp4)
289     ADD_SUB_HALVES_X4(temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8,
290                       temp5, temp6, temp7, temp8, temp9, temp10, temp11, temp12)
291     PACK_2_HALVES_TO_WORD(temp9, temp10, temp11, temp12, temp1, temp3, temp5,
292                           temp7, temp2, temp4, temp6, temp8)
293     ADD_SUB_HALVES_X4(temp2, temp4, temp6, temp8, temp9, temp1, temp3, temp10,
294                       temp1, temp9, temp3, temp10, temp5, temp11, temp7, temp12)
295     ADD_SUB_HALVES_X4(temp5, temp11, temp7, temp2, temp9, temp3, temp6, temp12,
296                       temp2, temp9, temp6, temp3, temp4, temp1, temp8, temp10)
297     ADD_SUB_HALVES_X4(temp1, temp4, temp10, temp8, temp7, temp11, temp5, temp2,
298                       temp5, temp7, temp11, temp2, temp9, temp6, temp3, temp12)
299     ABS_X8(temp1, temp4, temp10, temp8, temp7, temp11, temp5, temp2)
300     LOAD_WITH_OFFSET_X4(temp3, temp6, temp9, temp12, w,
301                         0, 4, 8, 12,
302                         0, 0, 0, 0,
303                         0)
304     LOAD_WITH_OFFSET_X4(temp13, temp14, temp15, temp16, w,
305                         0, 4, 8, 12,
306                         1, 1, 1, 1,
307                         16)
308     MUL_HALF(temp3, temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8,
309              temp9, temp10, temp11, temp12, temp13, temp14, temp15, temp16)
310     OUTPUT_EARLY_CLOBBER_REGS_17()
311     : [a]"r"(a), [b]"r"(b), [w]"r"(w)
312     : "memory", "hi", "lo"
313   );
314   return abs(temp3 - temp17) >> 5;
315 }
316 
Disto16x16_MIPSdspR2(const uint8_t * const a,const uint8_t * const b,const uint16_t * const w)317 static int Disto16x16_MIPSdspR2(const uint8_t* const a,
318                                 const uint8_t* const b,
319                                 const uint16_t* const w) {
320   int D = 0;
321   int x, y;
322   for (y = 0; y < 16 * BPS; y += 4 * BPS) {
323     for (x = 0; x < 16; x += 4) {
324       D += Disto4x4_MIPSdspR2(a + x + y, b + x + y, w);
325     }
326   }
327   return D;
328 }
329 
330 //------------------------------------------------------------------------------
331 // Intra predictions
332 
333 #define FILL_PART(J, SIZE)                                            \
334     "usw        %[value],  0+" #J "*" XSTR(BPS) "(%[dst])  \n\t"      \
335     "usw        %[value],  4+" #J "*" XSTR(BPS) "(%[dst])  \n\t"      \
336   ".if " #SIZE " == 16                                     \n\t"      \
337     "usw        %[value],  8+" #J "*" XSTR(BPS) "(%[dst])  \n\t"      \
338     "usw        %[value], 12+" #J "*" XSTR(BPS) "(%[dst])  \n\t"      \
339   ".endif                                                  \n\t"
340 
341 #define FILL_8_OR_16(DST, VALUE, SIZE) do {                         \
342   int value = (VALUE);                                              \
343   __asm__ volatile (                                                \
344     "replv.qb   %[value],  %[value]                      \n\t"      \
345     FILL_PART( 0, SIZE)                                             \
346     FILL_PART( 1, SIZE)                                             \
347     FILL_PART( 2, SIZE)                                             \
348     FILL_PART( 3, SIZE)                                             \
349     FILL_PART( 4, SIZE)                                             \
350     FILL_PART( 5, SIZE)                                             \
351     FILL_PART( 6, SIZE)                                             \
352     FILL_PART( 7, SIZE)                                             \
353   ".if " #SIZE " == 16                                   \n\t"      \
354     FILL_PART( 8, 16)                                               \
355     FILL_PART( 9, 16)                                               \
356     FILL_PART(10, 16)                                               \
357     FILL_PART(11, 16)                                               \
358     FILL_PART(12, 16)                                               \
359     FILL_PART(13, 16)                                               \
360     FILL_PART(14, 16)                                               \
361     FILL_PART(15, 16)                                               \
362   ".endif                                                \n\t"      \
363     : [value]"+&r"(value)                                           \
364     : [dst]"r"((DST))                                               \
365     : "memory"                                                      \
366   );                                                                \
367 } while (0)
368 
369 #define VERTICAL_PRED(DST, TOP, SIZE)                                          \
370 static WEBP_INLINE void VerticalPred##SIZE(uint8_t* (DST),                     \
371                                            const uint8_t* (TOP)) {             \
372   int j;                                                                       \
373   if ((TOP)) {                                                                 \
374     for (j = 0; j < (SIZE); ++j) memcpy((DST) + j * BPS, (TOP), (SIZE));       \
375   } else {                                                                     \
376     FILL_8_OR_16((DST), 127, (SIZE));                                          \
377   }                                                                            \
378 }
379 
380 VERTICAL_PRED(dst, top, 8)
381 VERTICAL_PRED(dst, top, 16)
382 
383 #undef VERTICAL_PRED
384 
385 #define HORIZONTAL_PRED(DST, LEFT, SIZE)                                       \
386 static WEBP_INLINE void HorizontalPred##SIZE(uint8_t* (DST),                   \
387                                              const uint8_t* (LEFT)) {          \
388   if (LEFT) {                                                                  \
389     int j;                                                                     \
390     for (j = 0; j < (SIZE); ++j) {                                             \
391       memset((DST) + j * BPS, (LEFT)[j], (SIZE));                              \
392     }                                                                          \
393   } else {                                                                     \
394     FILL_8_OR_16((DST), 129, (SIZE));                                          \
395   }                                                                            \
396 }
397 
398 HORIZONTAL_PRED(dst, left, 8)
399 HORIZONTAL_PRED(dst, left, 16)
400 
401 #undef HORIZONTAL_PRED
402 
403 #define CLIPPING()                                                             \
404   "preceu.ph.qbl   %[temp2],   %[temp0]                  \n\t"                 \
405   "preceu.ph.qbr   %[temp0],   %[temp0]                  \n\t"                 \
406   "preceu.ph.qbl   %[temp3],   %[temp1]                  \n\t"                 \
407   "preceu.ph.qbr   %[temp1],   %[temp1]                  \n\t"                 \
408   "addu.ph         %[temp2],   %[temp2],   %[leftY_1]    \n\t"                 \
409   "addu.ph         %[temp0],   %[temp0],   %[leftY_1]    \n\t"                 \
410   "addu.ph         %[temp3],   %[temp3],   %[leftY_1]    \n\t"                 \
411   "addu.ph         %[temp1],   %[temp1],   %[leftY_1]    \n\t"                 \
412   "shll_s.ph       %[temp2],   %[temp2],   7             \n\t"                 \
413   "shll_s.ph       %[temp0],   %[temp0],   7             \n\t"                 \
414   "shll_s.ph       %[temp3],   %[temp3],   7             \n\t"                 \
415   "shll_s.ph       %[temp1],   %[temp1],   7             \n\t"                 \
416   "precrqu_s.qb.ph %[temp0],   %[temp2],   %[temp0]      \n\t"                 \
417   "precrqu_s.qb.ph %[temp1],   %[temp3],   %[temp1]      \n\t"
418 
419 #define CLIP_8B_TO_DST(DST, LEFT, TOP, SIZE) do {                              \
420   int leftY_1 = ((int)(LEFT)[y] << 16) + (LEFT)[y];                            \
421   int temp0, temp1, temp2, temp3;                                              \
422   __asm__ volatile (                                                           \
423     "replv.ph        %[leftY_1], %[leftY_1]              \n\t"                 \
424     "ulw             %[temp0],   0(%[top])               \n\t"                 \
425     "ulw             %[temp1],   4(%[top])               \n\t"                 \
426     "subu.ph         %[leftY_1], %[leftY_1], %[left_1]   \n\t"                 \
427     CLIPPING()                                                                 \
428     "usw             %[temp0],   0(%[dst])               \n\t"                 \
429     "usw             %[temp1],   4(%[dst])               \n\t"                 \
430   ".if " #SIZE " == 16                                   \n\t"                 \
431     "ulw             %[temp0],   8(%[top])               \n\t"                 \
432     "ulw             %[temp1],   12(%[top])              \n\t"                 \
433     CLIPPING()                                                                 \
434     "usw             %[temp0],   8(%[dst])               \n\t"                 \
435     "usw             %[temp1],   12(%[dst])              \n\t"                 \
436   ".endif                                                \n\t"                 \
437     : [leftY_1]"+&r"(leftY_1), [temp0]"=&r"(temp0), [temp1]"=&r"(temp1),       \
438       [temp2]"=&r"(temp2), [temp3]"=&r"(temp3)                                 \
439     : [left_1]"r"(left_1), [top]"r"((TOP)), [dst]"r"((DST))                    \
440     : "memory"                                                                 \
441   );                                                                           \
442 } while (0)
443 
444 #define CLIP_TO_DST(DST, LEFT, TOP, SIZE) do {                                 \
445   int y;                                                                       \
446   const int left_1 = ((int)(LEFT)[-1] << 16) + (LEFT)[-1];                     \
447   for (y = 0; y < (SIZE); ++y) {                                               \
448     CLIP_8B_TO_DST((DST), (LEFT), (TOP), (SIZE));                              \
449     (DST) += BPS;                                                              \
450   }                                                                            \
451 } while (0)
452 
453 #define TRUE_MOTION(DST, LEFT, TOP, SIZE)                                      \
454 static WEBP_INLINE void TrueMotion##SIZE(uint8_t* (DST), const uint8_t* (LEFT),\
455                                          const uint8_t* (TOP)) {               \
456   if ((LEFT) != NULL) {                                                        \
457     if ((TOP) != NULL) {                                                       \
458       CLIP_TO_DST((DST), (LEFT), (TOP), (SIZE));                               \
459     } else {                                                                   \
460       HorizontalPred##SIZE((DST), (LEFT));                                     \
461     }                                                                          \
462   } else {                                                                     \
463     /* true motion without left samples (hence: with default 129 value)    */  \
464     /* is equivalent to VE prediction where you just copy the top samples. */  \
465     /* Note that if top samples are not available, the default value is    */  \
466     /* then 129, and not 127 as in the VerticalPred case.                  */  \
467     if ((TOP) != NULL) {                                                       \
468       VerticalPred##SIZE((DST), (TOP));                                        \
469     } else {                                                                   \
470       FILL_8_OR_16((DST), 129, (SIZE));                                        \
471     }                                                                          \
472   }                                                                            \
473 }
474 
475 TRUE_MOTION(dst, left, top, 8)
476 TRUE_MOTION(dst, left, top, 16)
477 
478 #undef TRUE_MOTION
479 #undef CLIP_TO_DST
480 #undef CLIP_8B_TO_DST
481 #undef CLIPPING
482 
DCMode16(uint8_t * dst,const uint8_t * left,const uint8_t * top)483 static WEBP_INLINE void DCMode16(uint8_t* dst, const uint8_t* left,
484                                  const uint8_t* top) {
485   int DC, DC1;
486   int temp0, temp1, temp2, temp3;
487 
488   __asm__ volatile(
489     "beqz        %[top],   2f                  \n\t"
490     LOAD_WITH_OFFSET_X4(temp0, temp1, temp2, temp3, top,
491                         0, 4, 8, 12,
492                         0, 0, 0, 0,
493                         0)
494     "raddu.w.qb  %[temp0], %[temp0]            \n\t"
495     "raddu.w.qb  %[temp1], %[temp1]            \n\t"
496     "raddu.w.qb  %[temp2], %[temp2]            \n\t"
497     "raddu.w.qb  %[temp3], %[temp3]            \n\t"
498     "addu        %[temp0], %[temp0], %[temp1]  \n\t"
499     "addu        %[temp2], %[temp2], %[temp3]  \n\t"
500     "addu        %[DC],    %[temp0], %[temp2]  \n\t"
501     "move        %[DC1],   %[DC]               \n\t"
502     "beqz        %[left],  1f                  \n\t"
503     LOAD_WITH_OFFSET_X4(temp0, temp1, temp2, temp3, left,
504                         0, 4, 8, 12,
505                         0, 0, 0, 0,
506                         0)
507     "raddu.w.qb  %[temp0], %[temp0]            \n\t"
508     "raddu.w.qb  %[temp1], %[temp1]            \n\t"
509     "raddu.w.qb  %[temp2], %[temp2]            \n\t"
510     "raddu.w.qb  %[temp3], %[temp3]            \n\t"
511     "addu        %[temp0], %[temp0], %[temp1]  \n\t"
512     "addu        %[temp2], %[temp2], %[temp3]  \n\t"
513     "addu        %[DC1],   %[temp0], %[temp2]  \n\t"
514   "1:                                          \n\t"
515     "addu        %[DC],   %[DC],     %[DC1]    \n\t"
516     "j           3f                            \n\t"
517   "2:                                          \n\t"
518     "beqz        %[left],  4f                  \n\t"
519     LOAD_WITH_OFFSET_X4(temp0, temp1, temp2, temp3, left,
520                         0, 4, 8, 12,
521                         0, 0, 0, 0,
522                         0)
523     "raddu.w.qb  %[temp0], %[temp0]            \n\t"
524     "raddu.w.qb  %[temp1], %[temp1]            \n\t"
525     "raddu.w.qb  %[temp2], %[temp2]            \n\t"
526     "raddu.w.qb  %[temp3], %[temp3]            \n\t"
527     "addu        %[temp0], %[temp0], %[temp1]  \n\t"
528     "addu        %[temp2], %[temp2], %[temp3]  \n\t"
529     "addu        %[DC],    %[temp0], %[temp2]  \n\t"
530     "addu        %[DC],    %[DC],    %[DC]     \n\t"
531   "3:                                          \n\t"
532     "shra_r.w    %[DC],    %[DC],    5         \n\t"
533     "j           5f                            \n\t"
534   "4:                                          \n\t"
535     "li          %[DC],    0x80                \n\t"
536   "5:                                          \n\t"
537     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [DC]"=&r"(DC),
538       [temp2]"=&r"(temp2), [temp3]"=&r"(temp3), [DC1]"=&r"(DC1)
539     : [left]"r"(left), [top]"r"(top)
540     : "memory"
541   );
542 
543   FILL_8_OR_16(dst, DC, 16);
544 }
545 
DCMode8(uint8_t * dst,const uint8_t * left,const uint8_t * top)546 static WEBP_INLINE void DCMode8(uint8_t* dst, const uint8_t* left,
547                                 const uint8_t* top) {
548   int DC, DC1;
549   int temp0, temp1, temp2, temp3;
550 
551   __asm__ volatile(
552     "beqz        %[top],   2f                  \n\t"
553     "ulw         %[temp0], 0(%[top])           \n\t"
554     "ulw         %[temp1], 4(%[top])           \n\t"
555     "raddu.w.qb  %[temp0], %[temp0]            \n\t"
556     "raddu.w.qb  %[temp1], %[temp1]            \n\t"
557     "addu        %[DC],    %[temp0], %[temp1]  \n\t"
558     "move        %[DC1],   %[DC]               \n\t"
559     "beqz        %[left],  1f                  \n\t"
560     "ulw         %[temp2], 0(%[left])          \n\t"
561     "ulw         %[temp3], 4(%[left])          \n\t"
562     "raddu.w.qb  %[temp2], %[temp2]            \n\t"
563     "raddu.w.qb  %[temp3], %[temp3]            \n\t"
564     "addu        %[DC1],   %[temp2], %[temp3]  \n\t"
565   "1:                                          \n\t"
566     "addu        %[DC],    %[DC],    %[DC1]    \n\t"
567     "j           3f                            \n\t"
568   "2:                                          \n\t"
569     "beqz        %[left],  4f                  \n\t"
570     "ulw         %[temp2], 0(%[left])          \n\t"
571     "ulw         %[temp3], 4(%[left])          \n\t"
572     "raddu.w.qb  %[temp2], %[temp2]            \n\t"
573     "raddu.w.qb  %[temp3], %[temp3]            \n\t"
574     "addu        %[DC],    %[temp2], %[temp3]  \n\t"
575     "addu        %[DC],    %[DC],    %[DC]     \n\t"
576   "3:                                          \n\t"
577     "shra_r.w    %[DC], %[DC], 4               \n\t"
578     "j           5f                            \n\t"
579   "4:                                          \n\t"
580     "li          %[DC], 0x80                   \n\t"
581   "5:                                          \n\t"
582     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [DC]"=&r"(DC),
583       [temp2]"=&r"(temp2), [temp3]"=&r"(temp3), [DC1]"=&r"(DC1)
584     : [left]"r"(left), [top]"r"(top)
585     : "memory"
586   );
587 
588   FILL_8_OR_16(dst, DC, 8);
589 }
590 
DC4(uint8_t * dst,const uint8_t * top)591 static void DC4(uint8_t* dst, const uint8_t* top) {
592   int temp0, temp1;
593   __asm__ volatile(
594     "ulw          %[temp0],   0(%[top])               \n\t"
595     "ulw          %[temp1],   -5(%[top])              \n\t"
596     "raddu.w.qb   %[temp0],   %[temp0]                \n\t"
597     "raddu.w.qb   %[temp1],   %[temp1]                \n\t"
598     "addu         %[temp0],   %[temp0],    %[temp1]   \n\t"
599     "addiu        %[temp0],   %[temp0],    4          \n\t"
600     "srl          %[temp0],   %[temp0],    3          \n\t"
601     "replv.qb     %[temp0],   %[temp0]                \n\t"
602     "usw          %[temp0],   0*" XSTR(BPS) "(%[dst]) \n\t"
603     "usw          %[temp0],   1*" XSTR(BPS) "(%[dst]) \n\t"
604     "usw          %[temp0],   2*" XSTR(BPS) "(%[dst]) \n\t"
605     "usw          %[temp0],   3*" XSTR(BPS) "(%[dst]) \n\t"
606     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1)
607     : [top]"r"(top), [dst]"r"(dst)
608     : "memory"
609   );
610 }
611 
TM4(uint8_t * dst,const uint8_t * top)612 static void TM4(uint8_t* dst, const uint8_t* top) {
613   int a10, a32, temp0, temp1, temp2, temp3, temp4, temp5;
614   const int c35 = 0xff00ff;
615   __asm__ volatile (
616     "lbu              %[temp1],  0(%[top])                     \n\t"
617     "lbu              %[a10],    1(%[top])                     \n\t"
618     "lbu              %[temp2],  2(%[top])                     \n\t"
619     "lbu              %[a32],    3(%[top])                     \n\t"
620     "ulw              %[temp0],  -5(%[top])                    \n\t"
621     "lbu              %[temp4],  -1(%[top])                    \n\t"
622     "append           %[a10],    %[temp1],   16                \n\t"
623     "append           %[a32],    %[temp2],   16                \n\t"
624     "replv.ph         %[temp4],  %[temp4]                      \n\t"
625     "shrl.ph          %[temp1],  %[temp0],   8                 \n\t"
626     "and              %[temp0],  %[temp0],   %[c35]            \n\t"
627     "subu.ph          %[temp1],  %[temp1],   %[temp4]          \n\t"
628     "subu.ph          %[temp0],  %[temp0],   %[temp4]          \n\t"
629     "srl              %[temp2],  %[temp1],   16                \n\t"
630     "srl              %[temp3],  %[temp0],   16                \n\t"
631     "replv.ph         %[temp2],  %[temp2]                      \n\t"
632     "replv.ph         %[temp3],  %[temp3]                      \n\t"
633     "replv.ph         %[temp4],  %[temp1]                      \n\t"
634     "replv.ph         %[temp5],  %[temp0]                      \n\t"
635     "addu.ph          %[temp0],  %[temp3],   %[a10]            \n\t"
636     "addu.ph          %[temp1],  %[temp3],   %[a32]            \n\t"
637     "addu.ph          %[temp3],  %[temp2],   %[a10]            \n\t"
638     "addu.ph          %[temp2],  %[temp2],   %[a32]            \n\t"
639     "shll_s.ph        %[temp0],  %[temp0],   7                 \n\t"
640     "shll_s.ph        %[temp1],  %[temp1],   7                 \n\t"
641     "shll_s.ph        %[temp3],  %[temp3],   7                 \n\t"
642     "shll_s.ph        %[temp2],  %[temp2],   7                 \n\t"
643     "precrqu_s.qb.ph  %[temp0],  %[temp1],   %[temp0]          \n\t"
644     "precrqu_s.qb.ph  %[temp1],  %[temp2],   %[temp3]          \n\t"
645     "addu.ph          %[temp2],  %[temp5],   %[a10]            \n\t"
646     "addu.ph          %[temp3],  %[temp5],   %[a32]            \n\t"
647     "addu.ph          %[temp5],  %[temp4],   %[a10]            \n\t"
648     "addu.ph          %[temp4],  %[temp4],   %[a32]            \n\t"
649     "shll_s.ph        %[temp2],  %[temp2],   7                 \n\t"
650     "shll_s.ph        %[temp3],  %[temp3],   7                 \n\t"
651     "shll_s.ph        %[temp4],  %[temp4],   7                 \n\t"
652     "shll_s.ph        %[temp5],  %[temp5],   7                 \n\t"
653     "precrqu_s.qb.ph  %[temp2],  %[temp3],   %[temp2]          \n\t"
654     "precrqu_s.qb.ph  %[temp3],  %[temp4],   %[temp5]          \n\t"
655     "usw              %[temp1],  0*" XSTR(BPS) "(%[dst])       \n\t"
656     "usw              %[temp0],  1*" XSTR(BPS) "(%[dst])       \n\t"
657     "usw              %[temp3],  2*" XSTR(BPS) "(%[dst])       \n\t"
658     "usw              %[temp2],  3*" XSTR(BPS) "(%[dst])       \n\t"
659     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
660       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
661       [a10]"=&r"(a10), [a32]"=&r"(a32)
662     : [c35]"r"(c35), [top]"r"(top), [dst]"r"(dst)
663     : "memory"
664   );
665 }
666 
VE4(uint8_t * dst,const uint8_t * top)667 static void VE4(uint8_t* dst, const uint8_t* top) {
668   int temp0, temp1, temp2, temp3, temp4, temp5, temp6;
669   __asm__ volatile(
670     "ulw             %[temp0],   -1(%[top])              \n\t"
671     "ulh             %[temp1],   3(%[top])               \n\t"
672     "preceu.ph.qbr   %[temp2],   %[temp0]                \n\t"
673     "preceu.ph.qbl   %[temp3],   %[temp0]                \n\t"
674     "preceu.ph.qbr   %[temp4],   %[temp1]                \n\t"
675     "packrl.ph       %[temp5],   %[temp3],    %[temp2]   \n\t"
676     "packrl.ph       %[temp6],   %[temp4],    %[temp3]   \n\t"
677     "shll.ph         %[temp5],   %[temp5],    1          \n\t"
678     "shll.ph         %[temp6],   %[temp6],    1          \n\t"
679     "addq.ph         %[temp2],   %[temp5],    %[temp2]   \n\t"
680     "addq.ph         %[temp6],   %[temp6],    %[temp4]   \n\t"
681     "addq.ph         %[temp2],   %[temp2],    %[temp3]   \n\t"
682     "addq.ph         %[temp6],   %[temp6],    %[temp3]   \n\t"
683     "shra_r.ph       %[temp2],   %[temp2],    2          \n\t"
684     "shra_r.ph       %[temp6],   %[temp6],    2          \n\t"
685     "precr.qb.ph     %[temp4],   %[temp6],    %[temp2]   \n\t"
686     "usw             %[temp4],   0*" XSTR(BPS) "(%[dst]) \n\t"
687     "usw             %[temp4],   1*" XSTR(BPS) "(%[dst]) \n\t"
688     "usw             %[temp4],   2*" XSTR(BPS) "(%[dst]) \n\t"
689     "usw             %[temp4],   3*" XSTR(BPS) "(%[dst]) \n\t"
690     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
691       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
692       [temp6]"=&r"(temp6)
693     : [top]"r"(top), [dst]"r"(dst)
694     : "memory"
695   );
696 }
697 
HE4(uint8_t * dst,const uint8_t * top)698 static void HE4(uint8_t* dst, const uint8_t* top) {
699   int temp0, temp1, temp2, temp3, temp4, temp5, temp6;
700   __asm__ volatile(
701     "ulw             %[temp0],   -4(%[top])              \n\t"
702     "lbu             %[temp1],   -5(%[top])              \n\t"
703     "preceu.ph.qbr   %[temp2],   %[temp0]                \n\t"
704     "preceu.ph.qbl   %[temp3],   %[temp0]                \n\t"
705     "replv.ph        %[temp4],   %[temp1]                \n\t"
706     "packrl.ph       %[temp5],   %[temp3],    %[temp2]   \n\t"
707     "packrl.ph       %[temp6],   %[temp2],    %[temp4]   \n\t"
708     "shll.ph         %[temp5],   %[temp5],    1          \n\t"
709     "shll.ph         %[temp6],   %[temp6],    1          \n\t"
710     "addq.ph         %[temp3],   %[temp3],    %[temp5]   \n\t"
711     "addq.ph         %[temp3],   %[temp3],    %[temp2]   \n\t"
712     "addq.ph         %[temp2],   %[temp2],    %[temp6]   \n\t"
713     "addq.ph         %[temp2],   %[temp2],    %[temp4]   \n\t"
714     "shra_r.ph       %[temp3],   %[temp3],    2          \n\t"
715     "shra_r.ph       %[temp2],   %[temp2],    2          \n\t"
716     "replv.qb        %[temp0],   %[temp3]                \n\t"
717     "replv.qb        %[temp1],   %[temp2]                \n\t"
718     "srl             %[temp3],   %[temp3],    16         \n\t"
719     "srl             %[temp2],   %[temp2],    16         \n\t"
720     "replv.qb        %[temp3],   %[temp3]                \n\t"
721     "replv.qb        %[temp2],   %[temp2]                \n\t"
722     "usw             %[temp3],   0*" XSTR(BPS) "(%[dst]) \n\t"
723     "usw             %[temp0],   1*" XSTR(BPS) "(%[dst]) \n\t"
724     "usw             %[temp2],   2*" XSTR(BPS) "(%[dst]) \n\t"
725     "usw             %[temp1],   3*" XSTR(BPS) "(%[dst]) \n\t"
726     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
727       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
728       [temp6]"=&r"(temp6)
729     : [top]"r"(top), [dst]"r"(dst)
730     : "memory"
731   );
732 }
733 
RD4(uint8_t * dst,const uint8_t * top)734 static void RD4(uint8_t* dst, const uint8_t* top) {
735   int temp0, temp1, temp2, temp3, temp4, temp5;
736   int temp6, temp7, temp8, temp9, temp10, temp11;
737   __asm__ volatile(
738     "ulw             %[temp0],    -5(%[top])               \n\t"
739     "ulw             %[temp1],    -1(%[top])               \n\t"
740     "preceu.ph.qbl   %[temp2],    %[temp0]                 \n\t"
741     "preceu.ph.qbr   %[temp3],    %[temp0]                 \n\t"
742     "preceu.ph.qbr   %[temp4],    %[temp1]                 \n\t"
743     "preceu.ph.qbl   %[temp5],    %[temp1]                 \n\t"
744     "packrl.ph       %[temp6],    %[temp2],    %[temp3]    \n\t"
745     "packrl.ph       %[temp7],    %[temp4],    %[temp2]    \n\t"
746     "packrl.ph       %[temp8],    %[temp5],    %[temp4]    \n\t"
747     "shll.ph         %[temp6],    %[temp6],    1           \n\t"
748     "addq.ph         %[temp9],    %[temp2],    %[temp6]    \n\t"
749     "shll.ph         %[temp7],    %[temp7],    1           \n\t"
750     "addq.ph         %[temp9],    %[temp9],    %[temp3]    \n\t"
751     "shll.ph         %[temp8],    %[temp8],    1           \n\t"
752     "shra_r.ph       %[temp9],    %[temp9],    2           \n\t"
753     "addq.ph         %[temp10],   %[temp4],    %[temp7]    \n\t"
754     "addq.ph         %[temp11],   %[temp5],    %[temp8]    \n\t"
755     "addq.ph         %[temp10],   %[temp10],   %[temp2]    \n\t"
756     "addq.ph         %[temp11],   %[temp11],   %[temp4]    \n\t"
757     "shra_r.ph       %[temp10],   %[temp10],   2           \n\t"
758     "shra_r.ph       %[temp11],   %[temp11],   2           \n\t"
759     "lbu             %[temp0],    3(%[top])                \n\t"
760     "lbu             %[temp1],    2(%[top])                \n\t"
761     "lbu             %[temp2],    1(%[top])                \n\t"
762     "sll             %[temp1],    %[temp1],    1           \n\t"
763     "addu            %[temp0],    %[temp0],    %[temp1]    \n\t"
764     "addu            %[temp0],    %[temp0],    %[temp2]    \n\t"
765     "precr.qb.ph     %[temp9],    %[temp10],   %[temp9]    \n\t"
766     "shra_r.w        %[temp0],    %[temp0],    2           \n\t"
767     "precr.qb.ph     %[temp10],   %[temp11],   %[temp10]   \n\t"
768     "usw             %[temp9],    3*" XSTR(BPS) "(%[dst])  \n\t"
769     "usw             %[temp10],   1*" XSTR(BPS) "(%[dst])  \n\t"
770     "prepend         %[temp9],    %[temp11],   8           \n\t"
771     "prepend         %[temp10],   %[temp0],    8           \n\t"
772     "usw             %[temp9],    2*" XSTR(BPS) "(%[dst])  \n\t"
773     "usw             %[temp10],   0*" XSTR(BPS) "(%[dst])  \n\t"
774     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
775       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
776       [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [temp8]"=&r"(temp8),
777       [temp9]"=&r"(temp9), [temp10]"=&r"(temp10), [temp11]"=&r"(temp11)
778     : [top]"r"(top), [dst]"r"(dst)
779     : "memory"
780   );
781 }
782 
VR4(uint8_t * dst,const uint8_t * top)783 static void VR4(uint8_t* dst, const uint8_t* top) {
784   int temp0, temp1, temp2, temp3, temp4;
785   int temp5, temp6, temp7, temp8, temp9;
786   __asm__ volatile (
787     "ulw              %[temp0],   -4(%[top])              \n\t"
788     "ulw              %[temp1],   0(%[top])               \n\t"
789     "preceu.ph.qbl    %[temp2],   %[temp0]                \n\t"
790     "preceu.ph.qbr    %[temp0],   %[temp0]                \n\t"
791     "preceu.ph.qbla   %[temp3],   %[temp1]                \n\t"
792     "preceu.ph.qbra   %[temp1],   %[temp1]                \n\t"
793     "packrl.ph        %[temp7],   %[temp3],    %[temp2]   \n\t"
794     "addqh_r.ph       %[temp4],   %[temp1],    %[temp3]   \n\t"
795     "move             %[temp6],   %[temp1]                \n\t"
796     "append           %[temp1],   %[temp2],    16         \n\t"
797     "shll.ph          %[temp9],   %[temp6],    1          \n\t"
798     "addqh_r.ph       %[temp5],   %[temp7],    %[temp6]   \n\t"
799     "shll.ph          %[temp8],   %[temp7],    1          \n\t"
800     "addu.ph          %[temp3],   %[temp7],    %[temp3]   \n\t"
801     "addu.ph          %[temp1],   %[temp1],    %[temp6]   \n\t"
802     "packrl.ph        %[temp7],   %[temp2],    %[temp0]   \n\t"
803     "addu.ph          %[temp6],   %[temp0],    %[temp2]   \n\t"
804     "addu.ph          %[temp3],   %[temp3],    %[temp9]   \n\t"
805     "addu.ph          %[temp1],   %[temp1],    %[temp8]   \n\t"
806     "shll.ph          %[temp7],   %[temp7],    1          \n\t"
807     "shra_r.ph        %[temp3],   %[temp3],    2          \n\t"
808     "shra_r.ph        %[temp1],   %[temp1],    2          \n\t"
809     "addu.ph          %[temp6],   %[temp6],    %[temp7]   \n\t"
810     "shra_r.ph        %[temp6],   %[temp6],    2          \n\t"
811     "precrq.ph.w      %[temp8],   %[temp4],    %[temp5]   \n\t"
812     "append           %[temp4],   %[temp5],    16         \n\t"
813     "precrq.ph.w      %[temp2],   %[temp3],    %[temp1]   \n\t"
814     "append           %[temp3],   %[temp1],    16         \n\t"
815     "precr.qb.ph      %[temp8],   %[temp8],    %[temp4]   \n\t"
816     "precr.qb.ph      %[temp3],   %[temp2],    %[temp3]   \n\t"
817     "usw              %[temp8],   0*" XSTR(BPS) "(%[dst]) \n\t"
818     "usw              %[temp3],   1*" XSTR(BPS) "(%[dst]) \n\t"
819     "append           %[temp3],   %[temp6],    8          \n\t"
820     "srl              %[temp6],   %[temp6],    16         \n\t"
821     "append           %[temp8],   %[temp6],    8          \n\t"
822     "usw              %[temp3],   3*" XSTR(BPS) "(%[dst]) \n\t"
823     "usw              %[temp8],   2*" XSTR(BPS) "(%[dst]) \n\t"
824     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
825       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
826       [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [temp8]"=&r"(temp8),
827       [temp9]"=&r"(temp9)
828     : [top]"r"(top), [dst]"r"(dst)
829     : "memory"
830   );
831 }
832 
LD4(uint8_t * dst,const uint8_t * top)833 static void LD4(uint8_t* dst, const uint8_t* top) {
834   int temp0, temp1, temp2, temp3, temp4, temp5;
835   int temp6, temp7, temp8, temp9, temp10, temp11;
836   __asm__ volatile(
837     "ulw             %[temp0],    0(%[top])               \n\t"
838     "ulw             %[temp1],    4(%[top])               \n\t"
839     "preceu.ph.qbl   %[temp2],    %[temp0]                \n\t"
840     "preceu.ph.qbr   %[temp3],    %[temp0]                \n\t"
841     "preceu.ph.qbr   %[temp4],    %[temp1]                \n\t"
842     "preceu.ph.qbl   %[temp5],    %[temp1]                \n\t"
843     "packrl.ph       %[temp6],    %[temp2],    %[temp3]   \n\t"
844     "packrl.ph       %[temp7],    %[temp4],    %[temp2]   \n\t"
845     "packrl.ph       %[temp8],    %[temp5],    %[temp4]   \n\t"
846     "shll.ph         %[temp6],    %[temp6],    1          \n\t"
847     "addq.ph         %[temp9],    %[temp2],    %[temp6]   \n\t"
848     "shll.ph         %[temp7],    %[temp7],    1          \n\t"
849     "addq.ph         %[temp9],    %[temp9],    %[temp3]   \n\t"
850     "shll.ph         %[temp8],    %[temp8],    1          \n\t"
851     "shra_r.ph       %[temp9],    %[temp9],    2          \n\t"
852     "addq.ph         %[temp10],   %[temp4],    %[temp7]   \n\t"
853     "addq.ph         %[temp11],   %[temp5],    %[temp8]   \n\t"
854     "addq.ph         %[temp10],   %[temp10],   %[temp2]   \n\t"
855     "addq.ph         %[temp11],   %[temp11],   %[temp4]   \n\t"
856     "shra_r.ph       %[temp10],   %[temp10],   2          \n\t"
857     "shra_r.ph       %[temp11],   %[temp11],   2          \n\t"
858     "srl             %[temp1],    %[temp1],    24         \n\t"
859     "sll             %[temp1],    %[temp1],    1          \n\t"
860     "raddu.w.qb      %[temp5],    %[temp5]                \n\t"
861     "precr.qb.ph     %[temp9],    %[temp10],   %[temp9]   \n\t"
862     "precr.qb.ph     %[temp10],   %[temp11],   %[temp10]  \n\t"
863     "addu            %[temp1],    %[temp1],    %[temp5]   \n\t"
864     "shra_r.w        %[temp1],    %[temp1],    2          \n\t"
865     "usw             %[temp9],    0*" XSTR(BPS) "(%[dst]) \n\t"
866     "usw             %[temp10],   2*" XSTR(BPS) "(%[dst]) \n\t"
867     "prepend         %[temp9],    %[temp11],   8          \n\t"
868     "prepend         %[temp10],   %[temp1],    8          \n\t"
869     "usw             %[temp9],    1*" XSTR(BPS) "(%[dst]) \n\t"
870     "usw             %[temp10],   3*" XSTR(BPS) "(%[dst]) \n\t"
871     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
872       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
873       [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [temp8]"=&r"(temp8),
874       [temp9]"=&r"(temp9), [temp10]"=&r"(temp10), [temp11]"=&r"(temp11)
875     : [top]"r"(top), [dst]"r"(dst)
876     : "memory"
877   );
878 }
879 
VL4(uint8_t * dst,const uint8_t * top)880 static void VL4(uint8_t* dst, const uint8_t* top) {
881   int temp0, temp1, temp2, temp3, temp4;
882   int temp5, temp6, temp7, temp8, temp9;
883   __asm__ volatile (
884     "ulw              %[temp0],   0(%[top])               \n\t"
885     "ulw              %[temp1],   4(%[top])               \n\t"
886     "preceu.ph.qbla   %[temp2],   %[temp0]                \n\t"
887     "preceu.ph.qbra   %[temp0],   %[temp0]                \n\t"
888     "preceu.ph.qbl    %[temp3],   %[temp1]                \n\t"
889     "preceu.ph.qbr    %[temp1],   %[temp1]                \n\t"
890     "addqh_r.ph       %[temp4],   %[temp0],    %[temp2]   \n\t"
891     "packrl.ph        %[temp7],   %[temp1],    %[temp0]   \n\t"
892     "precrq.ph.w      %[temp6],   %[temp1],    %[temp2]   \n\t"
893     "shll.ph          %[temp9],   %[temp2],    1          \n\t"
894     "addqh_r.ph       %[temp5],   %[temp7],    %[temp2]   \n\t"
895     "shll.ph          %[temp8],   %[temp7],    1          \n\t"
896     "addu.ph          %[temp2],   %[temp2],    %[temp6]   \n\t"
897     "addu.ph          %[temp0],   %[temp0],    %[temp7]   \n\t"
898     "packrl.ph        %[temp7],   %[temp3],    %[temp1]   \n\t"
899     "addu.ph          %[temp6],   %[temp1],    %[temp3]   \n\t"
900     "addu.ph          %[temp2],   %[temp2],    %[temp8]   \n\t"
901     "addu.ph          %[temp0],   %[temp0],    %[temp9]   \n\t"
902     "shll.ph          %[temp7],   %[temp7],    1          \n\t"
903     "shra_r.ph        %[temp2],   %[temp2],    2          \n\t"
904     "shra_r.ph        %[temp0],   %[temp0],    2          \n\t"
905     "addu.ph          %[temp6],   %[temp6],    %[temp7]   \n\t"
906     "shra_r.ph        %[temp6],   %[temp6],    2          \n\t"
907     "precrq.ph.w      %[temp8],   %[temp5],    %[temp4]   \n\t"
908     "append           %[temp5],   %[temp4],    16         \n\t"
909     "precrq.ph.w      %[temp3],   %[temp2],    %[temp0]   \n\t"
910     "append           %[temp2],   %[temp0],    16         \n\t"
911     "precr.qb.ph      %[temp8],   %[temp8],    %[temp5]   \n\t"
912     "precr.qb.ph      %[temp3],   %[temp3],    %[temp2]   \n\t"
913     "usw              %[temp8],   0*" XSTR(BPS) "(%[dst]) \n\t"
914     "prepend          %[temp8],   %[temp6],    8          \n\t"
915     "usw              %[temp3],   1*" XSTR(BPS) "(%[dst]) \n\t"
916     "srl              %[temp6],   %[temp6],    16         \n\t"
917     "prepend          %[temp3],   %[temp6],    8          \n\t"
918     "usw              %[temp8],   2*" XSTR(BPS) "(%[dst]) \n\t"
919     "usw              %[temp3],   3*" XSTR(BPS) "(%[dst]) \n\t"
920     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
921       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
922       [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [temp8]"=&r"(temp8),
923       [temp9]"=&r"(temp9)
924     : [top]"r"(top), [dst]"r"(dst)
925     : "memory"
926   );
927 }
928 
HD4(uint8_t * dst,const uint8_t * top)929 static void HD4(uint8_t* dst, const uint8_t* top) {
930   int temp0, temp1, temp2, temp3, temp4;
931   int temp5, temp6, temp7, temp8, temp9;
932   __asm__ volatile (
933     "ulw              %[temp0],   -5(%[top])              \n\t"
934     "ulw              %[temp1],   -1(%[top])              \n\t"
935     "preceu.ph.qbla   %[temp2],   %[temp0]                \n\t"
936     "preceu.ph.qbra   %[temp0],   %[temp0]                \n\t"
937     "preceu.ph.qbl    %[temp3],   %[temp1]                \n\t"
938     "preceu.ph.qbr    %[temp1],   %[temp1]                \n\t"
939     "addqh_r.ph       %[temp4],   %[temp0],    %[temp2]   \n\t"
940     "packrl.ph        %[temp7],   %[temp1],    %[temp0]   \n\t"
941     "precrq.ph.w      %[temp6],   %[temp1],    %[temp2]   \n\t"
942     "shll.ph          %[temp9],   %[temp2],    1          \n\t"
943     "addqh_r.ph       %[temp5],   %[temp7],    %[temp2]   \n\t"
944     "shll.ph          %[temp8],   %[temp7],    1          \n\t"
945     "addu.ph          %[temp2],   %[temp2],    %[temp6]   \n\t"
946     "addu.ph          %[temp0],   %[temp0],    %[temp7]   \n\t"
947     "packrl.ph        %[temp7],   %[temp3],    %[temp1]   \n\t"
948     "addu.ph          %[temp6],   %[temp1],    %[temp3]   \n\t"
949     "addu.ph          %[temp2],   %[temp2],    %[temp8]   \n\t"
950     "addu.ph          %[temp0],   %[temp0],    %[temp9]   \n\t"
951     "shll.ph          %[temp7],   %[temp7],    1          \n\t"
952     "shra_r.ph        %[temp2],   %[temp2],    2          \n\t"
953     "shra_r.ph        %[temp0],   %[temp0],    2          \n\t"
954     "addu.ph          %[temp6],   %[temp6],    %[temp7]   \n\t"
955     "shra_r.ph        %[temp6],   %[temp6],    2          \n\t"
956     "precrq.ph.w      %[temp1],   %[temp2],    %[temp5]   \n\t"
957     "precrq.ph.w      %[temp3],   %[temp0],    %[temp4]   \n\t"
958     "precr.qb.ph      %[temp7],   %[temp6],    %[temp1]   \n\t"
959     "precr.qb.ph      %[temp6],   %[temp1],    %[temp3]   \n\t"
960     "usw              %[temp7],   0*" XSTR(BPS) "(%[dst]) \n\t"
961     "usw              %[temp6],   1*" XSTR(BPS) "(%[dst]) \n\t"
962     "append           %[temp2],   %[temp5],    16         \n\t"
963     "append           %[temp0],   %[temp4],    16         \n\t"
964     "precr.qb.ph      %[temp5],   %[temp3],    %[temp2]   \n\t"
965     "precr.qb.ph      %[temp4],   %[temp2],    %[temp0]   \n\t"
966     "usw              %[temp5],   2*" XSTR(BPS) "(%[dst]) \n\t"
967     "usw              %[temp4],   3*" XSTR(BPS) "(%[dst]) \n\t"
968     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
969       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
970       [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [temp8]"=&r"(temp8),
971       [temp9]"=&r"(temp9)
972     : [top]"r"(top), [dst]"r"(dst)
973     : "memory"
974   );
975 }
976 
HU4(uint8_t * dst,const uint8_t * top)977 static void HU4(uint8_t* dst, const uint8_t* top) {
978   int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
979   __asm__ volatile (
980     "ulw             %[temp0],   -5(%[top])              \n\t"
981     "preceu.ph.qbl   %[temp1],   %[temp0]                \n\t"
982     "preceu.ph.qbr   %[temp2],   %[temp0]                \n\t"
983     "packrl.ph       %[temp3],   %[temp1],    %[temp2]   \n\t"
984     "replv.qb        %[temp7],   %[temp2]                \n\t"
985     "addqh_r.ph      %[temp4],   %[temp1],    %[temp3]   \n\t"
986     "addqh_r.ph      %[temp5],   %[temp3],    %[temp2]   \n\t"
987     "shll.ph         %[temp6],   %[temp3],    1          \n\t"
988     "addu.ph         %[temp3],   %[temp2],    %[temp3]   \n\t"
989     "addu.ph         %[temp6],   %[temp1],    %[temp6]   \n\t"
990     "shll.ph         %[temp0],   %[temp2],    1          \n\t"
991     "addu.ph         %[temp6],   %[temp6],    %[temp2]   \n\t"
992     "addu.ph         %[temp0],   %[temp3],    %[temp0]   \n\t"
993     "shra_r.ph       %[temp6],   %[temp6],    2          \n\t"
994     "shra_r.ph       %[temp0],   %[temp0],    2          \n\t"
995     "packrl.ph       %[temp3],   %[temp6],    %[temp5]   \n\t"
996     "precrq.ph.w     %[temp2],   %[temp6],    %[temp4]   \n\t"
997     "append          %[temp0],   %[temp5],    16         \n\t"
998     "precr.qb.ph     %[temp3],   %[temp3],    %[temp2]   \n\t"
999     "usw             %[temp3],   0*" XSTR(BPS) "(%[dst]) \n\t"
1000     "precr.qb.ph     %[temp1],   %[temp7],    %[temp0]   \n\t"
1001     "usw             %[temp7],   3*" XSTR(BPS) "(%[dst]) \n\t"
1002     "packrl.ph       %[temp2],   %[temp1],    %[temp3]   \n\t"
1003     "usw             %[temp1],   2*" XSTR(BPS) "(%[dst]) \n\t"
1004     "usw             %[temp2],   1*" XSTR(BPS) "(%[dst]) \n\t"
1005     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
1006       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
1007       [temp6]"=&r"(temp6), [temp7]"=&r"(temp7)
1008     : [top]"r"(top), [dst]"r"(dst)
1009     : "memory"
1010   );
1011 }
1012 
1013 //------------------------------------------------------------------------------
1014 // Chroma 8x8 prediction (paragraph 12.2)
1015 
IntraChromaPreds_MIPSdspR2(uint8_t * dst,const uint8_t * left,const uint8_t * top)1016 static void IntraChromaPreds_MIPSdspR2(uint8_t* dst, const uint8_t* left,
1017                                        const uint8_t* top) {
1018   // U block
1019   DCMode8(C8DC8 + dst, left, top);
1020   VerticalPred8(C8VE8 + dst, top);
1021   HorizontalPred8(C8HE8 + dst, left);
1022   TrueMotion8(C8TM8 + dst, left, top);
1023   // V block
1024   dst += 8;
1025   if (top) top += 8;
1026   if (left) left += 16;
1027   DCMode8(C8DC8 + dst, left, top);
1028   VerticalPred8(C8VE8 + dst, top);
1029   HorizontalPred8(C8HE8 + dst, left);
1030   TrueMotion8(C8TM8 + dst, left, top);
1031 }
1032 
1033 //------------------------------------------------------------------------------
1034 // luma 16x16 prediction (paragraph 12.3)
1035 
Intra16Preds_MIPSdspR2(uint8_t * dst,const uint8_t * left,const uint8_t * top)1036 static void Intra16Preds_MIPSdspR2(uint8_t* dst,
1037                                    const uint8_t* left, const uint8_t* top) {
1038   DCMode16(I16DC16 + dst, left, top);
1039   VerticalPred16(I16VE16 + dst, top);
1040   HorizontalPred16(I16HE16 + dst, left);
1041   TrueMotion16(I16TM16 + dst, left, top);
1042 }
1043 
1044 // Left samples are top[-5 .. -2], top_left is top[-1], top are
1045 // located at top[0..3], and top right is top[4..7]
Intra4Preds_MIPSdspR2(uint8_t * dst,const uint8_t * top)1046 static void Intra4Preds_MIPSdspR2(uint8_t* dst, const uint8_t* top) {
1047   DC4(I4DC4 + dst, top);
1048   TM4(I4TM4 + dst, top);
1049   VE4(I4VE4 + dst, top);
1050   HE4(I4HE4 + dst, top);
1051   RD4(I4RD4 + dst, top);
1052   VR4(I4VR4 + dst, top);
1053   LD4(I4LD4 + dst, top);
1054   VL4(I4VL4 + dst, top);
1055   HD4(I4HD4 + dst, top);
1056   HU4(I4HU4 + dst, top);
1057 }
1058 
1059 //------------------------------------------------------------------------------
1060 // Metric
1061 
1062 #if !defined(WORK_AROUND_GCC)
1063 
1064 #define GET_SSE_INNER(A)                                                  \
1065   "lw               %[temp0],    " #A "(%[a])                  \n\t"      \
1066   "lw               %[temp1],    " #A "(%[b])                  \n\t"      \
1067   "preceu.ph.qbr    %[temp2],    %[temp0]                      \n\t"      \
1068   "preceu.ph.qbl    %[temp0],    %[temp0]                      \n\t"      \
1069   "preceu.ph.qbr    %[temp3],    %[temp1]                      \n\t"      \
1070   "preceu.ph.qbl    %[temp1],    %[temp1]                      \n\t"      \
1071   "subq.ph          %[temp2],    %[temp2],    %[temp3]         \n\t"      \
1072   "subq.ph          %[temp0],    %[temp0],    %[temp1]         \n\t"      \
1073   "dpa.w.ph         $ac0,        %[temp2],    %[temp2]         \n\t"      \
1074   "dpa.w.ph         $ac0,        %[temp0],    %[temp0]         \n\t"
1075 
1076 #define GET_SSE(A, B, C, D)               \
1077   GET_SSE_INNER(A)                        \
1078   GET_SSE_INNER(B)                        \
1079   GET_SSE_INNER(C)                        \
1080   GET_SSE_INNER(D)
1081 
SSE16x16_MIPSdspR2(const uint8_t * a,const uint8_t * b)1082 static int SSE16x16_MIPSdspR2(const uint8_t* a, const uint8_t* b) {
1083   int count;
1084   int temp0, temp1, temp2, temp3;
1085   __asm__ volatile (
1086     "mult   $zero,    $zero                            \n\t"
1087     GET_SSE( 0 * BPS, 4 +  0 * BPS, 8 +  0 * BPS, 12 +  0 * BPS)
1088     GET_SSE( 1 * BPS, 4 +  1 * BPS, 8 +  1 * BPS, 12 +  1 * BPS)
1089     GET_SSE( 2 * BPS, 4 +  2 * BPS, 8 +  2 * BPS, 12 +  2 * BPS)
1090     GET_SSE( 3 * BPS, 4 +  3 * BPS, 8 +  3 * BPS, 12 +  3 * BPS)
1091     GET_SSE( 4 * BPS, 4 +  4 * BPS, 8 +  4 * BPS, 12 +  4 * BPS)
1092     GET_SSE( 5 * BPS, 4 +  5 * BPS, 8 +  5 * BPS, 12 +  5 * BPS)
1093     GET_SSE( 6 * BPS, 4 +  6 * BPS, 8 +  6 * BPS, 12 +  6 * BPS)
1094     GET_SSE( 7 * BPS, 4 +  7 * BPS, 8 +  7 * BPS, 12 +  7 * BPS)
1095     GET_SSE( 8 * BPS, 4 +  8 * BPS, 8 +  8 * BPS, 12 +  8 * BPS)
1096     GET_SSE( 9 * BPS, 4 +  9 * BPS, 8 +  9 * BPS, 12 +  9 * BPS)
1097     GET_SSE(10 * BPS, 4 + 10 * BPS, 8 + 10 * BPS, 12 + 10 * BPS)
1098     GET_SSE(11 * BPS, 4 + 11 * BPS, 8 + 11 * BPS, 12 + 11 * BPS)
1099     GET_SSE(12 * BPS, 4 + 12 * BPS, 8 + 12 * BPS, 12 + 12 * BPS)
1100     GET_SSE(13 * BPS, 4 + 13 * BPS, 8 + 13 * BPS, 12 + 13 * BPS)
1101     GET_SSE(14 * BPS, 4 + 14 * BPS, 8 + 14 * BPS, 12 + 14 * BPS)
1102     GET_SSE(15 * BPS, 4 + 15 * BPS, 8 + 15 * BPS, 12 + 15 * BPS)
1103     "mflo   %[count]                                   \n\t"
1104     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
1105       [temp3]"=&r"(temp3), [count]"=&r"(count)
1106     : [a]"r"(a), [b]"r"(b)
1107     : "memory", "hi", "lo"
1108   );
1109   return count;
1110 }
1111 
SSE16x8_MIPSdspR2(const uint8_t * a,const uint8_t * b)1112 static int SSE16x8_MIPSdspR2(const uint8_t* a, const uint8_t* b) {
1113   int count;
1114   int temp0, temp1, temp2, temp3;
1115   __asm__ volatile (
1116     "mult   $zero,    $zero                            \n\t"
1117     GET_SSE( 0 * BPS, 4 +  0 * BPS, 8 +  0 * BPS, 12 +  0 * BPS)
1118     GET_SSE( 1 * BPS, 4 +  1 * BPS, 8 +  1 * BPS, 12 +  1 * BPS)
1119     GET_SSE( 2 * BPS, 4 +  2 * BPS, 8 +  2 * BPS, 12 +  2 * BPS)
1120     GET_SSE( 3 * BPS, 4 +  3 * BPS, 8 +  3 * BPS, 12 +  3 * BPS)
1121     GET_SSE( 4 * BPS, 4 +  4 * BPS, 8 +  4 * BPS, 12 +  4 * BPS)
1122     GET_SSE( 5 * BPS, 4 +  5 * BPS, 8 +  5 * BPS, 12 +  5 * BPS)
1123     GET_SSE( 6 * BPS, 4 +  6 * BPS, 8 +  6 * BPS, 12 +  6 * BPS)
1124     GET_SSE( 7 * BPS, 4 +  7 * BPS, 8 +  7 * BPS, 12 +  7 * BPS)
1125     "mflo   %[count]                                   \n\t"
1126     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
1127       [temp3]"=&r"(temp3), [count]"=&r"(count)
1128     : [a]"r"(a), [b]"r"(b)
1129     : "memory", "hi", "lo"
1130   );
1131   return count;
1132 }
1133 
SSE8x8_MIPSdspR2(const uint8_t * a,const uint8_t * b)1134 static int SSE8x8_MIPSdspR2(const uint8_t* a, const uint8_t* b) {
1135   int count;
1136   int temp0, temp1, temp2, temp3;
1137   __asm__ volatile (
1138     "mult   $zero,    $zero                            \n\t"
1139     GET_SSE(0 * BPS, 4 + 0 * BPS, 1 * BPS, 4 + 1 * BPS)
1140     GET_SSE(2 * BPS, 4 + 2 * BPS, 3 * BPS, 4 + 3 * BPS)
1141     GET_SSE(4 * BPS, 4 + 4 * BPS, 5 * BPS, 4 + 5 * BPS)
1142     GET_SSE(6 * BPS, 4 + 6 * BPS, 7 * BPS, 4 + 7 * BPS)
1143     "mflo   %[count]                                   \n\t"
1144     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
1145       [temp3]"=&r"(temp3), [count]"=&r"(count)
1146     : [a]"r"(a), [b]"r"(b)
1147     : "memory", "hi", "lo"
1148   );
1149   return count;
1150 }
1151 
SSE4x4_MIPSdspR2(const uint8_t * a,const uint8_t * b)1152 static int SSE4x4_MIPSdspR2(const uint8_t* a, const uint8_t* b) {
1153   int count;
1154   int temp0, temp1, temp2, temp3;
1155   __asm__ volatile (
1156     "mult   $zero,    $zero                            \n\t"
1157     GET_SSE(0 * BPS, 1 * BPS, 2 * BPS, 3 * BPS)
1158     "mflo   %[count]                                   \n\t"
1159     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
1160       [temp3]"=&r"(temp3), [count]"=&r"(count)
1161     : [a]"r"(a), [b]"r"(b)
1162     : "memory", "hi", "lo"
1163   );
1164   return count;
1165 }
1166 
1167 #undef GET_SSE
1168 #undef GET_SSE_INNER
1169 
1170 #endif  // !WORK_AROUND_GCC
1171 
1172 #undef FILL_8_OR_16
1173 #undef FILL_PART
1174 #undef OUTPUT_EARLY_CLOBBER_REGS_17
1175 #undef MUL_HALF
1176 #undef ABS_X8
1177 #undef ADD_SUB_HALVES_X4
1178 
1179 //------------------------------------------------------------------------------
1180 // Quantization
1181 //
1182 
1183 // macro for one pass through for loop in QuantizeBlock reading 2 values at time
1184 // QUANTDIV macro inlined
1185 // J - offset in bytes (kZigzag[n] * 2)
1186 // K - offset in bytes (kZigzag[n] * 4)
1187 // N - offset in bytes (n * 2)
1188 // N1 - offset in bytes ((n + 1) * 2)
1189 #define QUANTIZE_ONE(J, K, N, N1)                                         \
1190   "ulw         %[temp1],     " #J "(%[ppin])                 \n\t"        \
1191   "ulw         %[temp2],     " #J "(%[ppsharpen])            \n\t"        \
1192   "lhu         %[temp3],     " #K "(%[ppzthresh])            \n\t"        \
1193   "lhu         %[temp6],     " #K "+4(%[ppzthresh])          \n\t"        \
1194   "absq_s.ph   %[temp4],     %[temp1]                        \n\t"        \
1195   "ins         %[temp3],     %[temp6],         16,       16  \n\t"        \
1196   "addu.ph     %[coeff],     %[temp4],         %[temp2]      \n\t"        \
1197   "shra.ph     %[sign],      %[temp1],         15            \n\t"        \
1198   "li          %[level],     0x10001                         \n\t"        \
1199   "cmp.lt.ph   %[temp3],     %[coeff]                        \n\t"        \
1200   "lhu         %[temp1],     " #J "(%[ppiq])                 \n\t"        \
1201   "pick.ph     %[temp5],     %[level],         $0            \n\t"        \
1202   "lw          %[temp2],     " #K "(%[ppbias])               \n\t"        \
1203   "beqz        %[temp5],     0f                              \n\t"        \
1204   "lhu         %[temp3],     " #J "(%[ppq])                  \n\t"        \
1205   "beq         %[temp5],     %[level],         1f            \n\t"        \
1206   "andi        %[temp5],     %[temp5],         0x1           \n\t"        \
1207   "andi        %[temp4],     %[coeff],         0xffff        \n\t"        \
1208   "beqz        %[temp5],     2f                              \n\t"        \
1209   "mul         %[level],     %[temp4],         %[temp1]      \n\t"        \
1210   "sh          $0,           " #J "+2(%[ppin])               \n\t"        \
1211   "sh          $0,           " #N1 "(%[pout])                \n\t"        \
1212   "addu        %[level],     %[level],         %[temp2]      \n\t"        \
1213   "sra         %[level],     %[level],         17            \n\t"        \
1214   "slt         %[temp4],     %[max_level],     %[level]      \n\t"        \
1215   "movn        %[level],     %[max_level],     %[temp4]      \n\t"        \
1216   "andi        %[temp6],     %[sign],          0xffff        \n\t"        \
1217   "xor         %[level],     %[level],         %[temp6]      \n\t"        \
1218   "subu        %[level],     %[level],         %[temp6]      \n\t"        \
1219   "mul         %[temp5],     %[level],         %[temp3]      \n\t"        \
1220   "or          %[ret],       %[ret],           %[level]      \n\t"        \
1221   "sh          %[level],     " #N "(%[pout])                 \n\t"        \
1222   "sh          %[temp5],     " #J "(%[ppin])                 \n\t"        \
1223   "j           3f                                            \n\t"        \
1224 "2:                                                          \n\t"        \
1225   "lhu         %[temp1],     " #J "+2(%[ppiq])               \n\t"        \
1226   "srl         %[temp5],     %[coeff],         16            \n\t"        \
1227   "mul         %[level],     %[temp5],         %[temp1]      \n\t"        \
1228   "lw          %[temp2],     " #K "+4(%[ppbias])             \n\t"        \
1229   "lhu         %[temp3],     " #J "+2(%[ppq])                \n\t"        \
1230   "addu        %[level],     %[level],         %[temp2]      \n\t"        \
1231   "sra         %[level],     %[level],         17            \n\t"        \
1232   "srl         %[temp6],     %[sign],          16            \n\t"        \
1233   "slt         %[temp4],     %[max_level],     %[level]      \n\t"        \
1234   "movn        %[level],     %[max_level],     %[temp4]      \n\t"        \
1235   "xor         %[level],     %[level],         %[temp6]      \n\t"        \
1236   "subu        %[level],     %[level],         %[temp6]      \n\t"        \
1237   "mul         %[temp5],     %[level],         %[temp3]      \n\t"        \
1238   "sh          $0,           " #J "(%[ppin])                 \n\t"        \
1239   "sh          $0,           " #N "(%[pout])                 \n\t"        \
1240   "or          %[ret],       %[ret],           %[level]      \n\t"        \
1241   "sh          %[temp5],     " #J "+2(%[ppin])               \n\t"        \
1242   "sh          %[level],     " #N1 "(%[pout])                \n\t"        \
1243   "j           3f                                            \n\t"        \
1244 "1:                                                          \n\t"        \
1245   "lhu         %[temp1],     " #J "(%[ppiq])                 \n\t"        \
1246   "lw          %[temp2],     " #K "(%[ppbias])               \n\t"        \
1247   "ulw         %[temp3],     " #J "(%[ppq])                  \n\t"        \
1248   "andi        %[temp5],     %[coeff],         0xffff        \n\t"        \
1249   "srl         %[temp0],     %[coeff],         16            \n\t"        \
1250   "lhu         %[temp6],     " #J "+2(%[ppiq])               \n\t"        \
1251   "lw          %[coeff],     " #K "+4(%[ppbias])             \n\t"        \
1252   "mul         %[level],     %[temp5],         %[temp1]      \n\t"        \
1253   "mul         %[temp4],     %[temp0],         %[temp6]      \n\t"        \
1254   "addu        %[level],     %[level],         %[temp2]      \n\t"        \
1255   "addu        %[temp4],     %[temp4],         %[coeff]      \n\t"        \
1256   "precrq.ph.w %[level],     %[temp4],         %[level]      \n\t"        \
1257   "shra.ph     %[level],     %[level],         1             \n\t"        \
1258   "cmp.lt.ph   %[max_level1],%[level]                        \n\t"        \
1259   "pick.ph     %[level],     %[max_level],     %[level]      \n\t"        \
1260   "xor         %[level],     %[level],         %[sign]       \n\t"        \
1261   "subu.ph     %[level],     %[level],         %[sign]       \n\t"        \
1262   "mul.ph      %[temp3],     %[level],         %[temp3]      \n\t"        \
1263   "or          %[ret],       %[ret],           %[level]      \n\t"        \
1264   "sh          %[level],     " #N "(%[pout])                 \n\t"        \
1265   "srl         %[level],     %[level],         16            \n\t"        \
1266   "sh          %[level],     " #N1 "(%[pout])                \n\t"        \
1267   "usw         %[temp3],     " #J "(%[ppin])                 \n\t"        \
1268   "j           3f                                            \n\t"        \
1269 "0:                                                          \n\t"        \
1270   "sh          $0,           " #N "(%[pout])                 \n\t"        \
1271   "sh          $0,           " #N1 "(%[pout])                \n\t"        \
1272   "usw         $0,           " #J "(%[ppin])                 \n\t"        \
1273 "3:                                                          \n\t"
1274 
QuantizeBlock_MIPSdspR2(int16_t in[16],int16_t out[16],const VP8Matrix * const mtx)1275 static int QuantizeBlock_MIPSdspR2(int16_t in[16], int16_t out[16],
1276                                    const VP8Matrix* const mtx) {
1277   int temp0, temp1, temp2, temp3, temp4, temp5,temp6;
1278   int sign, coeff, level;
1279   int max_level = MAX_LEVEL;
1280   int max_level1 = max_level << 16 | max_level;
1281   int ret = 0;
1282 
1283   int16_t* ppin             = &in[0];
1284   int16_t* pout             = &out[0];
1285   const uint16_t* ppsharpen = &mtx->sharpen_[0];
1286   const uint32_t* ppzthresh = &mtx->zthresh_[0];
1287   const uint16_t* ppq       = &mtx->q_[0];
1288   const uint16_t* ppiq      = &mtx->iq_[0];
1289   const uint32_t* ppbias    = &mtx->bias_[0];
1290 
1291   __asm__ volatile (
1292     QUANTIZE_ONE( 0,  0,  0,  2)
1293     QUANTIZE_ONE( 4,  8, 10, 12)
1294     QUANTIZE_ONE( 8, 16,  4,  8)
1295     QUANTIZE_ONE(12, 24, 14, 24)
1296     QUANTIZE_ONE(16, 32,  6, 16)
1297     QUANTIZE_ONE(20, 40, 22, 26)
1298     QUANTIZE_ONE(24, 48, 18, 20)
1299     QUANTIZE_ONE(28, 56, 28, 30)
1300 
1301     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1),
1302       [temp2]"=&r"(temp2), [temp3]"=&r"(temp3),
1303       [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
1304       [sign]"=&r"(sign), [coeff]"=&r"(coeff),
1305       [level]"=&r"(level), [temp6]"=&r"(temp6), [ret]"+&r"(ret)
1306     : [ppin]"r"(ppin), [pout]"r"(pout), [max_level1]"r"(max_level1),
1307       [ppiq]"r"(ppiq), [max_level]"r"(max_level),
1308       [ppbias]"r"(ppbias), [ppzthresh]"r"(ppzthresh),
1309       [ppsharpen]"r"(ppsharpen), [ppq]"r"(ppq)
1310     : "memory", "hi", "lo"
1311   );
1312 
1313   return (ret != 0);
1314 }
1315 
Quantize2Blocks_MIPSdspR2(int16_t in[32],int16_t out[32],const VP8Matrix * const mtx)1316 static int Quantize2Blocks_MIPSdspR2(int16_t in[32], int16_t out[32],
1317                                      const VP8Matrix* const mtx) {
1318   int nz;
1319   nz  = QuantizeBlock_MIPSdspR2(in + 0 * 16, out + 0 * 16, mtx) << 0;
1320   nz |= QuantizeBlock_MIPSdspR2(in + 1 * 16, out + 1 * 16, mtx) << 1;
1321   return nz;
1322 }
1323 
1324 #undef QUANTIZE_ONE
1325 
1326 // macro for one horizontal pass in FTransformWHT
1327 // temp0..temp7 holds tmp[0]..tmp[15]
1328 // A, B, C, D - offset in bytes to load from in buffer
1329 // TEMP0, TEMP1 - registers for corresponding tmp elements
1330 #define HORIZONTAL_PASS_WHT(A, B, C, D, TEMP0, TEMP1)                          \
1331   "lh              %[" #TEMP0 "],  " #A "(%[in])            \n\t"              \
1332   "lh              %[" #TEMP1 "],  " #B "(%[in])            \n\t"              \
1333   "lh              %[temp8],     " #C "(%[in])              \n\t"              \
1334   "lh              %[temp9],     " #D "(%[in])              \n\t"              \
1335   "ins             %[" #TEMP1 "],  %[" #TEMP0 "],  16,  16  \n\t"              \
1336   "ins             %[temp9],     %[temp8],     16,  16      \n\t"              \
1337   "subq.ph         %[temp8],     %[" #TEMP1 "],  %[temp9]   \n\t"              \
1338   "addq.ph         %[temp9],     %[" #TEMP1 "],  %[temp9]   \n\t"              \
1339   "precrq.ph.w     %[" #TEMP0 "],  %[temp8],     %[temp9]   \n\t"              \
1340   "append          %[temp8],     %[temp9],     16           \n\t"              \
1341   "subq.ph         %[" #TEMP1 "],  %[" #TEMP0 "],  %[temp8] \n\t"              \
1342   "addq.ph         %[" #TEMP0 "],  %[" #TEMP0 "],  %[temp8] \n\t"              \
1343   "rotr            %[" #TEMP1 "],  %[" #TEMP1 "],  16       \n\t"
1344 
1345 // macro for one vertical pass in FTransformWHT
1346 // temp0..temp7 holds tmp[0]..tmp[15]
1347 // A, B, C, D - offsets in bytes to store to out buffer
1348 // TEMP0, TEMP2, TEMP4 and TEMP6 - registers for corresponding tmp elements
1349 #define VERTICAL_PASS_WHT(A, B, C, D, TEMP0, TEMP2, TEMP4, TEMP6)              \
1350   "addq.ph         %[temp8],     %[" #TEMP0 "],  %[" #TEMP4 "]    \n\t"        \
1351   "addq.ph         %[temp9],     %[" #TEMP2 "],  %[" #TEMP6 "]    \n\t"        \
1352   "subq.ph         %[" #TEMP2 "],  %[" #TEMP2 "],  %[" #TEMP6 "]  \n\t"        \
1353   "subq.ph         %[" #TEMP6 "],  %[" #TEMP0 "],  %[" #TEMP4 "]  \n\t"        \
1354   "addqh.ph        %[" #TEMP0 "],  %[temp8],     %[temp9]         \n\t"        \
1355   "subqh.ph        %[" #TEMP4 "],  %[" #TEMP6 "],  %[" #TEMP2 "]  \n\t"        \
1356   "addqh.ph        %[" #TEMP2 "],  %[" #TEMP2 "],  %[" #TEMP6 "]  \n\t"        \
1357   "subqh.ph        %[" #TEMP6 "],  %[temp8],     %[temp9]         \n\t"        \
1358   "usw             %[" #TEMP0 "],  " #A "(%[out])                 \n\t"        \
1359   "usw             %[" #TEMP2 "],  " #B "(%[out])                 \n\t"        \
1360   "usw             %[" #TEMP4 "],  " #C "(%[out])                 \n\t"        \
1361   "usw             %[" #TEMP6 "],  " #D "(%[out])                 \n\t"
1362 
FTransformWHT_MIPSdspR2(const int16_t * in,int16_t * out)1363 static void FTransformWHT_MIPSdspR2(const int16_t* in, int16_t* out) {
1364   int temp0, temp1, temp2, temp3, temp4;
1365   int temp5, temp6, temp7, temp8, temp9;
1366 
1367   __asm__ volatile (
1368     HORIZONTAL_PASS_WHT(  0,  32,  64,  96, temp0, temp1)
1369     HORIZONTAL_PASS_WHT(128, 160, 192, 224, temp2, temp3)
1370     HORIZONTAL_PASS_WHT(256, 288, 320, 352, temp4, temp5)
1371     HORIZONTAL_PASS_WHT(384, 416, 448, 480, temp6, temp7)
1372     VERTICAL_PASS_WHT(0,  8, 16, 24, temp0, temp2, temp4, temp6)
1373     VERTICAL_PASS_WHT(4, 12, 20, 28, temp1, temp3, temp5, temp7)
1374     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
1375       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
1376       [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [temp8]"=&r"(temp8),
1377       [temp9]"=&r"(temp9)
1378     : [in]"r"(in), [out]"r"(out)
1379     : "memory"
1380   );
1381 }
1382 
1383 #undef VERTICAL_PASS_WHT
1384 #undef HORIZONTAL_PASS_WHT
1385 
1386 // macro for converting coefficients to bin
1387 // convert 8 coeffs at time
1388 // A, B, C, D - offsets in bytes to load from out buffer
1389 #define CONVERT_COEFFS_TO_BIN(A, B, C, D)                                      \
1390   "ulw        %[temp0],  " #A "(%[out])                \n\t"                   \
1391   "ulw        %[temp1],  " #B "(%[out])                \n\t"                   \
1392   "ulw        %[temp2],  " #C "(%[out])                \n\t"                   \
1393   "ulw        %[temp3],  " #D "(%[out])                \n\t"                   \
1394   "absq_s.ph  %[temp0],  %[temp0]                      \n\t"                   \
1395   "absq_s.ph  %[temp1],  %[temp1]                      \n\t"                   \
1396   "absq_s.ph  %[temp2],  %[temp2]                      \n\t"                   \
1397   "absq_s.ph  %[temp3],  %[temp3]                      \n\t"                   \
1398   "shra.ph    %[temp0],  %[temp0],    3                \n\t"                   \
1399   "shra.ph    %[temp1],  %[temp1],    3                \n\t"                   \
1400   "shra.ph    %[temp2],  %[temp2],    3                \n\t"                   \
1401   "shra.ph    %[temp3],  %[temp3],    3                \n\t"                   \
1402   "shll_s.ph  %[temp0],  %[temp0],    10               \n\t"                   \
1403   "shll_s.ph  %[temp1],  %[temp1],    10               \n\t"                   \
1404   "shll_s.ph  %[temp2],  %[temp2],    10               \n\t"                   \
1405   "shll_s.ph  %[temp3],  %[temp3],    10               \n\t"                   \
1406   "shrl.ph    %[temp0],  %[temp0],    10               \n\t"                   \
1407   "shrl.ph    %[temp1],  %[temp1],    10               \n\t"                   \
1408   "shrl.ph    %[temp2],  %[temp2],    10               \n\t"                   \
1409   "shrl.ph    %[temp3],  %[temp3],    10               \n\t"                   \
1410   "shll.ph    %[temp0],  %[temp0],    2                \n\t"                   \
1411   "shll.ph    %[temp1],  %[temp1],    2                \n\t"                   \
1412   "shll.ph    %[temp2],  %[temp2],    2                \n\t"                   \
1413   "shll.ph    %[temp3],  %[temp3],    2                \n\t"                   \
1414   "ext        %[temp4],  %[temp0],    0,       16      \n\t"                   \
1415   "ext        %[temp0],  %[temp0],    16,      16      \n\t"                   \
1416   "addu       %[temp4],  %[temp4],    %[dist]          \n\t"                   \
1417   "addu       %[temp0],  %[temp0],    %[dist]          \n\t"                   \
1418   "ext        %[temp5],  %[temp1],    0,       16      \n\t"                   \
1419   "lw         %[temp8],  0(%[temp4])                   \n\t"                   \
1420   "ext        %[temp1],  %[temp1],    16,      16      \n\t"                   \
1421   "addu       %[temp5],  %[temp5],    %[dist]          \n\t"                   \
1422   "addiu      %[temp8],  %[temp8],    1                \n\t"                   \
1423   "sw         %[temp8],  0(%[temp4])                   \n\t"                   \
1424   "lw         %[temp8],  0(%[temp0])                   \n\t"                   \
1425   "addu       %[temp1],  %[temp1],    %[dist]          \n\t"                   \
1426   "ext        %[temp6],  %[temp2],    0,       16      \n\t"                   \
1427   "addiu      %[temp8],  %[temp8],    1                \n\t"                   \
1428   "sw         %[temp8],  0(%[temp0])                   \n\t"                   \
1429   "lw         %[temp8],  0(%[temp5])                   \n\t"                   \
1430   "ext        %[temp2],  %[temp2],    16,      16      \n\t"                   \
1431   "addu       %[temp6],  %[temp6],    %[dist]          \n\t"                   \
1432   "addiu      %[temp8],  %[temp8],    1                \n\t"                   \
1433   "sw         %[temp8],  0(%[temp5])                   \n\t"                   \
1434   "lw         %[temp8],  0(%[temp1])                   \n\t"                   \
1435   "addu       %[temp2],  %[temp2],    %[dist]          \n\t"                   \
1436   "ext        %[temp7],  %[temp3],    0,       16      \n\t"                   \
1437   "addiu      %[temp8],  %[temp8],    1                \n\t"                   \
1438   "sw         %[temp8],  0(%[temp1])                   \n\t"                   \
1439   "lw         %[temp8],  0(%[temp6])                   \n\t"                   \
1440   "ext        %[temp3],  %[temp3],    16,      16      \n\t"                   \
1441   "addu       %[temp7],  %[temp7],    %[dist]          \n\t"                   \
1442   "addiu      %[temp8],  %[temp8],    1                \n\t"                   \
1443   "sw         %[temp8],  0(%[temp6])                   \n\t"                   \
1444   "lw         %[temp8],  0(%[temp2])                   \n\t"                   \
1445   "addu       %[temp3],  %[temp3],    %[dist]          \n\t"                   \
1446   "addiu      %[temp8],  %[temp8],    1                \n\t"                   \
1447   "sw         %[temp8],  0(%[temp2])                   \n\t"                   \
1448   "lw         %[temp8],  0(%[temp7])                   \n\t"                   \
1449   "addiu      %[temp8],  %[temp8],    1                \n\t"                   \
1450   "sw         %[temp8],  0(%[temp7])                   \n\t"                   \
1451   "lw         %[temp8],  0(%[temp3])                   \n\t"                   \
1452   "addiu      %[temp8],  %[temp8],    1                \n\t"                   \
1453   "sw         %[temp8],  0(%[temp3])                   \n\t"
1454 
CollectHistogram_MIPSdspR2(const uint8_t * ref,const uint8_t * pred,int start_block,int end_block,VP8Histogram * const histo)1455 static void CollectHistogram_MIPSdspR2(const uint8_t* ref, const uint8_t* pred,
1456                                        int start_block, int end_block,
1457                                        VP8Histogram* const histo) {
1458   int j;
1459   int distribution[MAX_COEFF_THRESH + 1] = { 0 };
1460   const int max_coeff = (MAX_COEFF_THRESH << 16) + MAX_COEFF_THRESH;
1461   for (j = start_block; j < end_block; ++j) {
1462     int16_t out[16];
1463     int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8;
1464 
1465     VP8FTransform(ref + VP8DspScan[j], pred + VP8DspScan[j], out);
1466 
1467     // Convert coefficients to bin.
1468     __asm__ volatile (
1469       CONVERT_COEFFS_TO_BIN( 0,  4,  8, 12)
1470       CONVERT_COEFFS_TO_BIN(16, 20, 24, 28)
1471       : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
1472         [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
1473         [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [temp8]"=&r"(temp8)
1474       : [dist]"r"(distribution), [out]"r"(out), [max_coeff]"r"(max_coeff)
1475       : "memory"
1476     );
1477   }
1478   VP8SetHistogramData(distribution, histo);
1479 }
1480 
1481 #undef CONVERT_COEFFS_TO_BIN
1482 
1483 //------------------------------------------------------------------------------
1484 // Entry point
1485 
1486 extern void VP8EncDspInitMIPSdspR2(void);
1487 
VP8EncDspInitMIPSdspR2(void)1488 WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspInitMIPSdspR2(void) {
1489   VP8FTransform = FTransform_MIPSdspR2;
1490   VP8FTransformWHT = FTransformWHT_MIPSdspR2;
1491   VP8ITransform = ITransform_MIPSdspR2;
1492 
1493   VP8TDisto4x4 = Disto4x4_MIPSdspR2;
1494   VP8TDisto16x16 = Disto16x16_MIPSdspR2;
1495 
1496   VP8EncPredLuma16 = Intra16Preds_MIPSdspR2;
1497   VP8EncPredChroma8 = IntraChromaPreds_MIPSdspR2;
1498   VP8EncPredLuma4 = Intra4Preds_MIPSdspR2;
1499 
1500 #if !defined(WORK_AROUND_GCC)
1501   VP8SSE16x16 = SSE16x16_MIPSdspR2;
1502   VP8SSE8x8 = SSE8x8_MIPSdspR2;
1503   VP8SSE16x8 = SSE16x8_MIPSdspR2;
1504   VP8SSE4x4 = SSE4x4_MIPSdspR2;
1505 #endif
1506 
1507   VP8EncQuantizeBlock = QuantizeBlock_MIPSdspR2;
1508   VP8EncQuantize2Blocks = Quantize2Blocks_MIPSdspR2;
1509 
1510   VP8CollectHistogram = CollectHistogram_MIPSdspR2;
1511 }
1512 
1513 #else  // !WEBP_USE_MIPS_DSP_R2
1514 
1515 WEBP_DSP_INIT_STUB(VP8EncDspInitMIPSdspR2)
1516 
1517 #endif  // WEBP_USE_MIPS_DSP_R2
1518