• 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 "./dsp.h"
16 
17 #if defined(WEBP_USE_MIPS_DSP_R2)
18 
19 #include "./mips_macro.h"
20 #include "../enc/cost_enc.h"
21 #include "../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(const uint8_t * src,const uint8_t * ref,int16_t * out)144 static void FTransform(const uint8_t* src, const uint8_t* ref, int16_t* out) {
145   const int c2217 = 2217;
146   const int c5352 = 5352;
147   int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8;
148   int temp9, temp10, temp11, temp12, temp13, temp14, temp15, temp16;
149   int temp17, temp18, temp19, temp20;
150   const int* const args[3] =
151       { (const int*)src, (const int*)ref, (const int*)out };
152 
153   __asm__ volatile (
154     HORIZONTAL_PASS(0, temp0,  temp1,  temp2,  temp3)
155     HORIZONTAL_PASS(1, temp4,  temp5,  temp6,  temp7)
156     HORIZONTAL_PASS(2, temp8,  temp9,  temp10, temp11)
157     HORIZONTAL_PASS(3, temp12, temp13, temp14, temp15)
158     "lw            %[temp20],     8(%[args])                  \n\t"
159     VERTICAL_PASS(0,  8, 16, 24, temp0, temp4, temp8,  temp12)
160     VERTICAL_PASS(2, 10, 18, 26, temp1, temp5, temp9,  temp13)
161     VERTICAL_PASS(4, 12, 20, 28, temp2, temp6, temp10, temp14)
162     VERTICAL_PASS(6, 14, 22, 30, temp3, temp7, temp11, temp15)
163     OUTPUT_EARLY_CLOBBER_REGS_18(),
164       [temp0]"=&r"(temp0), [temp19]"=&r"(temp19), [temp20]"=&r"(temp20)
165     : [args]"r"(args), [c2217]"r"(c2217), [c5352]"r"(c5352)
166     : "memory", "hi", "lo"
167   );
168 }
169 
170 #undef VERTICAL_PASS
171 #undef HORIZONTAL_PASS
172 
ITransformOne(const uint8_t * ref,const int16_t * in,uint8_t * dst)173 static WEBP_INLINE void ITransformOne(const uint8_t* ref, const int16_t* in,
174                                       uint8_t* dst) {
175   int temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8, temp9;
176   int temp10, temp11, temp12, temp13, temp14, temp15, temp16, temp17, temp18;
177 
178   __asm__ volatile (
179     "ulw              %[temp1],   0(%[in])                 \n\t"
180     "ulw              %[temp2],   16(%[in])                \n\t"
181     LOAD_IN_X2(temp5, temp6, 24, 26)
182     ADD_SUB_HALVES(temp3, temp4, temp1, temp2)
183     LOAD_IN_X2(temp1, temp2, 8, 10)
184     MUL_SHIFT_SUM(temp7, temp8, temp9, temp10, temp11, temp12, temp13, temp14,
185                   temp10, temp8, temp9, temp7, temp1, temp2, temp5, temp6,
186                   temp13, temp11, temp14, temp12)
187     INSERT_HALF_X2(temp8, temp7, temp10, temp9)
188     "ulw              %[temp17],  4(%[in])                 \n\t"
189     "ulw              %[temp18],  20(%[in])                \n\t"
190     ADD_SUB_HALVES(temp1, temp2, temp3, temp8)
191     ADD_SUB_HALVES(temp5, temp6, temp4, temp7)
192     ADD_SUB_HALVES(temp7, temp8, temp17, temp18)
193     LOAD_IN_X2(temp17, temp18, 12, 14)
194     LOAD_IN_X2(temp9, temp10, 28, 30)
195     MUL_SHIFT_SUM(temp11, temp12, temp13, temp14, temp15, temp16, temp4, temp17,
196                   temp12, temp14, temp11, temp13, temp17, temp18, temp9, temp10,
197                   temp15, temp4, temp16, temp17)
198     INSERT_HALF_X2(temp11, temp12, temp13, temp14)
199     ADD_SUB_HALVES(temp17, temp8, temp8, temp11)
200     ADD_SUB_HALVES(temp3, temp4, temp7, temp12)
201 
202     // horizontal
203     SRA_16(temp9, temp10, temp11, temp12, temp1, temp2, temp5, temp6)
204     INSERT_HALF_X2(temp1, temp6, temp5, temp2)
205     SRA_16(temp13, temp14, temp15, temp16, temp3, temp4, temp17, temp8)
206     "repl.ph          %[temp2],   0x4                      \n\t"
207     INSERT_HALF_X2(temp3, temp8, temp17, temp4)
208     "addq.ph          %[temp1],   %[temp1],  %[temp2]      \n\t"
209     "addq.ph          %[temp6],   %[temp6],  %[temp2]      \n\t"
210     ADD_SUB_HALVES(temp2, temp4, temp1, temp3)
211     ADD_SUB_HALVES(temp5, temp7, temp6, temp8)
212     MUL_SHIFT_SUM(temp1, temp3, temp6, temp8, temp9, temp13, temp17, temp18,
213                   temp3, temp13, temp1, temp9, temp9, temp13, temp11, temp15,
214                   temp6, temp17, temp8, temp18)
215     MUL_SHIFT_SUM(temp6, temp8, temp18, temp17, temp11, temp15, temp12, temp16,
216                   temp8, temp15, temp6, temp11, temp12, temp16, temp10, temp14,
217                   temp18, temp12, temp17, temp16)
218     INSERT_HALF_X2(temp1, temp3, temp9, temp13)
219     INSERT_HALF_X2(temp6, temp8, temp11, temp15)
220     SHIFT_R_SUM_X2(temp9, temp10, temp11, temp12, temp13, temp14, temp15,
221                    temp16, temp2, temp4, temp5, temp7, temp3, temp1, temp8,
222                    temp6)
223     PACK_2_HALVES_TO_WORD(temp1, temp2, temp3, temp4, temp9, temp12, temp13,
224                           temp16, temp11, temp10, temp15, temp14)
225     LOAD_WITH_OFFSET_X4(temp10, temp11, temp14, temp15, ref,
226                         0, 0, 0, 0,
227                         0, 1, 2, 3,
228                         BPS)
229     CONVERT_2_BYTES_TO_HALF(temp5, temp6, temp7, temp8, temp17, temp18, temp10,
230                             temp11, temp10, temp11, temp14, temp15)
231     STORE_SAT_SUM_X2(temp5, temp6, temp7, temp8, temp17, temp18, temp10, temp11,
232                      temp9, temp12, temp1, temp2, temp13, temp16, temp3, temp4,
233                      dst, 0, 1, 2, 3, BPS)
234 
235     OUTPUT_EARLY_CLOBBER_REGS_18()
236     : [dst]"r"(dst), [in]"r"(in), [kC1]"r"(kC1), [kC2]"r"(kC2), [ref]"r"(ref)
237     : "memory", "hi", "lo"
238   );
239 }
240 
ITransform(const uint8_t * ref,const int16_t * in,uint8_t * dst,int do_two)241 static void ITransform(const uint8_t* ref, const int16_t* in, uint8_t* dst,
242                        int do_two) {
243   ITransformOne(ref, in, dst);
244   if (do_two) {
245     ITransformOne(ref + 4, in + 16, dst + 4);
246   }
247 }
248 
Disto4x4(const uint8_t * const a,const uint8_t * const b,const uint16_t * const w)249 static int Disto4x4(const uint8_t* const a, const uint8_t* const b,
250                     const uint16_t* const w) {
251   int temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8, temp9;
252   int temp10, temp11, temp12, temp13, temp14, temp15, temp16, temp17;
253 
254   __asm__ volatile (
255     LOAD_WITH_OFFSET_X4(temp1, temp2, temp3, temp4, a,
256                         0, 0, 0, 0,
257                         0, 1, 2, 3,
258                         BPS)
259     CONVERT_2_BYTES_TO_HALF(temp5, temp6, temp7, temp8, temp9,temp10, temp11,
260                             temp12, temp1, temp2, temp3, temp4)
261     ADD_SUB_HALVES_X4(temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8,
262                       temp5, temp6, temp7, temp8, temp9, temp10, temp11, temp12)
263     PACK_2_HALVES_TO_WORD(temp9, temp10, temp11, temp12, temp1, temp3, temp5,
264                           temp7, temp2, temp4, temp6, temp8)
265     ADD_SUB_HALVES_X4(temp2, temp4, temp6, temp8, temp9, temp1, temp3, temp10,
266                       temp1, temp9, temp3, temp10, temp5, temp11, temp7, temp12)
267     ADD_SUB_HALVES_X4(temp5, temp11, temp7, temp2, temp9, temp3, temp6, temp12,
268                       temp2, temp9, temp6, temp3, temp4, temp1, temp8, temp10)
269     ADD_SUB_HALVES_X4(temp1, temp4, temp10, temp8, temp7, temp11, temp5, temp2,
270                       temp5, temp7, temp11, temp2, temp9, temp6, temp3, temp12)
271     ABS_X8(temp1, temp4, temp10, temp8, temp7, temp11, temp5, temp2)
272     LOAD_WITH_OFFSET_X4(temp3, temp6, temp9, temp12, w,
273                         0, 4, 8, 12,
274                         0, 0, 0, 0,
275                         0)
276     LOAD_WITH_OFFSET_X4(temp13, temp14, temp15, temp16, w,
277                         0, 4, 8, 12,
278                         1, 1, 1, 1,
279                         16)
280     MUL_HALF(temp17, temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8,
281              temp9, temp10, temp11, temp12, temp13, temp14, temp15, temp16)
282     LOAD_WITH_OFFSET_X4(temp1, temp2, temp3, temp4, b,
283                         0, 0, 0, 0,
284                         0, 1, 2, 3,
285                         BPS)
286     CONVERT_2_BYTES_TO_HALF(temp5,temp6, temp7, temp8, temp9,temp10, temp11,
287                             temp12, temp1, temp2, temp3, temp4)
288     ADD_SUB_HALVES_X4(temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8,
289                       temp5, temp6, temp7, temp8, temp9, temp10, temp11, temp12)
290     PACK_2_HALVES_TO_WORD(temp9, temp10, temp11, temp12, temp1, temp3, temp5,
291                           temp7, temp2, temp4, temp6, temp8)
292     ADD_SUB_HALVES_X4(temp2, temp4, temp6, temp8, temp9, temp1, temp3, temp10,
293                       temp1, temp9, temp3, temp10, temp5, temp11, temp7, temp12)
294     ADD_SUB_HALVES_X4(temp5, temp11, temp7, temp2, temp9, temp3, temp6, temp12,
295                       temp2, temp9, temp6, temp3, temp4, temp1, temp8, temp10)
296     ADD_SUB_HALVES_X4(temp1, temp4, temp10, temp8, temp7, temp11, temp5, temp2,
297                       temp5, temp7, temp11, temp2, temp9, temp6, temp3, temp12)
298     ABS_X8(temp1, temp4, temp10, temp8, temp7, temp11, temp5, temp2)
299     LOAD_WITH_OFFSET_X4(temp3, temp6, temp9, temp12, w,
300                         0, 4, 8, 12,
301                         0, 0, 0, 0,
302                         0)
303     LOAD_WITH_OFFSET_X4(temp13, temp14, temp15, temp16, w,
304                         0, 4, 8, 12,
305                         1, 1, 1, 1,
306                         16)
307     MUL_HALF(temp3, temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8,
308              temp9, temp10, temp11, temp12, temp13, temp14, temp15, temp16)
309     OUTPUT_EARLY_CLOBBER_REGS_17()
310     : [a]"r"(a), [b]"r"(b), [w]"r"(w)
311     : "memory", "hi", "lo"
312   );
313   return abs(temp3 - temp17) >> 5;
314 }
315 
Disto16x16(const uint8_t * const a,const uint8_t * const b,const uint16_t * const w)316 static int Disto16x16(const uint8_t* const a, const uint8_t* const b,
317                       const uint16_t* const w) {
318   int D = 0;
319   int x, y;
320   for (y = 0; y < 16 * BPS; y += 4 * BPS) {
321     for (x = 0; x < 16; x += 4) {
322       D += Disto4x4(a + x + y, b + x + y, w);
323     }
324   }
325   return D;
326 }
327 
328 //------------------------------------------------------------------------------
329 // Intra predictions
330 
331 #define FILL_PART(J, SIZE)                                            \
332     "usw        %[value],  0+" #J "*" XSTR(BPS) "(%[dst])  \n\t"      \
333     "usw        %[value],  4+" #J "*" XSTR(BPS) "(%[dst])  \n\t"      \
334   ".if " #SIZE " == 16                                     \n\t"      \
335     "usw        %[value],  8+" #J "*" XSTR(BPS) "(%[dst])  \n\t"      \
336     "usw        %[value], 12+" #J "*" XSTR(BPS) "(%[dst])  \n\t"      \
337   ".endif                                                  \n\t"
338 
339 #define FILL_8_OR_16(DST, VALUE, SIZE) do {                         \
340   int value = (VALUE);                                              \
341   __asm__ volatile (                                                \
342     "replv.qb   %[value],  %[value]                      \n\t"      \
343     FILL_PART( 0, SIZE)                                             \
344     FILL_PART( 1, SIZE)                                             \
345     FILL_PART( 2, SIZE)                                             \
346     FILL_PART( 3, SIZE)                                             \
347     FILL_PART( 4, SIZE)                                             \
348     FILL_PART( 5, SIZE)                                             \
349     FILL_PART( 6, SIZE)                                             \
350     FILL_PART( 7, SIZE)                                             \
351   ".if " #SIZE " == 16                                   \n\t"      \
352     FILL_PART( 8, 16)                                               \
353     FILL_PART( 9, 16)                                               \
354     FILL_PART(10, 16)                                               \
355     FILL_PART(11, 16)                                               \
356     FILL_PART(12, 16)                                               \
357     FILL_PART(13, 16)                                               \
358     FILL_PART(14, 16)                                               \
359     FILL_PART(15, 16)                                               \
360   ".endif                                                \n\t"      \
361     : [value]"+&r"(value)                                           \
362     : [dst]"r"((DST))                                               \
363     : "memory"                                                      \
364   );                                                                \
365 } while (0)
366 
367 #define VERTICAL_PRED(DST, TOP, SIZE)                                          \
368 static WEBP_INLINE void VerticalPred##SIZE(uint8_t* (DST),                     \
369                                            const uint8_t* (TOP)) {             \
370   int j;                                                                       \
371   if ((TOP)) {                                                                 \
372     for (j = 0; j < (SIZE); ++j) memcpy((DST) + j * BPS, (TOP), (SIZE));       \
373   } else {                                                                     \
374     FILL_8_OR_16((DST), 127, (SIZE));                                          \
375   }                                                                            \
376 }
377 
378 VERTICAL_PRED(dst, top, 8)
379 VERTICAL_PRED(dst, top, 16)
380 
381 #undef VERTICAL_PRED
382 
383 #define HORIZONTAL_PRED(DST, LEFT, SIZE)                                       \
384 static WEBP_INLINE void HorizontalPred##SIZE(uint8_t* (DST),                   \
385                                              const uint8_t* (LEFT)) {          \
386   if (LEFT) {                                                                  \
387     int j;                                                                     \
388     for (j = 0; j < (SIZE); ++j) {                                             \
389       memset((DST) + j * BPS, (LEFT)[j], (SIZE));                              \
390     }                                                                          \
391   } else {                                                                     \
392     FILL_8_OR_16((DST), 129, (SIZE));                                          \
393   }                                                                            \
394 }
395 
396 HORIZONTAL_PRED(dst, left, 8)
397 HORIZONTAL_PRED(dst, left, 16)
398 
399 #undef HORIZONTAL_PRED
400 
401 #define CLIPPING()                                                             \
402   "preceu.ph.qbl   %[temp2],   %[temp0]                  \n\t"                 \
403   "preceu.ph.qbr   %[temp0],   %[temp0]                  \n\t"                 \
404   "preceu.ph.qbl   %[temp3],   %[temp1]                  \n\t"                 \
405   "preceu.ph.qbr   %[temp1],   %[temp1]                  \n\t"                 \
406   "addu.ph         %[temp2],   %[temp2],   %[leftY_1]    \n\t"                 \
407   "addu.ph         %[temp0],   %[temp0],   %[leftY_1]    \n\t"                 \
408   "addu.ph         %[temp3],   %[temp3],   %[leftY_1]    \n\t"                 \
409   "addu.ph         %[temp1],   %[temp1],   %[leftY_1]    \n\t"                 \
410   "shll_s.ph       %[temp2],   %[temp2],   7             \n\t"                 \
411   "shll_s.ph       %[temp0],   %[temp0],   7             \n\t"                 \
412   "shll_s.ph       %[temp3],   %[temp3],   7             \n\t"                 \
413   "shll_s.ph       %[temp1],   %[temp1],   7             \n\t"                 \
414   "precrqu_s.qb.ph %[temp0],   %[temp2],   %[temp0]      \n\t"                 \
415   "precrqu_s.qb.ph %[temp1],   %[temp3],   %[temp1]      \n\t"
416 
417 #define CLIP_8B_TO_DST(DST, LEFT, TOP, SIZE) do {                              \
418   int leftY_1 = ((int)(LEFT)[y] << 16) + (LEFT)[y];                            \
419   int temp0, temp1, temp2, temp3;                                              \
420   __asm__ volatile (                                                           \
421     "replv.ph        %[leftY_1], %[leftY_1]              \n\t"                 \
422     "ulw             %[temp0],   0(%[top])               \n\t"                 \
423     "ulw             %[temp1],   4(%[top])               \n\t"                 \
424     "subu.ph         %[leftY_1], %[leftY_1], %[left_1]   \n\t"                 \
425     CLIPPING()                                                                 \
426     "usw             %[temp0],   0(%[dst])               \n\t"                 \
427     "usw             %[temp1],   4(%[dst])               \n\t"                 \
428   ".if " #SIZE " == 16                                   \n\t"                 \
429     "ulw             %[temp0],   8(%[top])               \n\t"                 \
430     "ulw             %[temp1],   12(%[top])              \n\t"                 \
431     CLIPPING()                                                                 \
432     "usw             %[temp0],   8(%[dst])               \n\t"                 \
433     "usw             %[temp1],   12(%[dst])              \n\t"                 \
434   ".endif                                                \n\t"                 \
435     : [leftY_1]"+&r"(leftY_1), [temp0]"=&r"(temp0), [temp1]"=&r"(temp1),       \
436       [temp2]"=&r"(temp2), [temp3]"=&r"(temp3)                                 \
437     : [left_1]"r"(left_1), [top]"r"((TOP)), [dst]"r"((DST))                    \
438     : "memory"                                                                 \
439   );                                                                           \
440 } while (0)
441 
442 #define CLIP_TO_DST(DST, LEFT, TOP, SIZE) do {                                 \
443   int y;                                                                       \
444   const int left_1 = ((int)(LEFT)[-1] << 16) + (LEFT)[-1];                     \
445   for (y = 0; y < (SIZE); ++y) {                                               \
446     CLIP_8B_TO_DST((DST), (LEFT), (TOP), (SIZE));                              \
447     (DST) += BPS;                                                              \
448   }                                                                            \
449 } while (0)
450 
451 #define TRUE_MOTION(DST, LEFT, TOP, SIZE)                                      \
452 static WEBP_INLINE void TrueMotion##SIZE(uint8_t* (DST), const uint8_t* (LEFT),\
453                                          const uint8_t* (TOP)) {               \
454   if ((LEFT) != NULL) {                                                        \
455     if ((TOP) != NULL) {                                                       \
456       CLIP_TO_DST((DST), (LEFT), (TOP), (SIZE));                               \
457     } else {                                                                   \
458       HorizontalPred##SIZE((DST), (LEFT));                                     \
459     }                                                                          \
460   } else {                                                                     \
461     /* true motion without left samples (hence: with default 129 value)    */  \
462     /* is equivalent to VE prediction where you just copy the top samples. */  \
463     /* Note that if top samples are not available, the default value is    */  \
464     /* then 129, and not 127 as in the VerticalPred case.                  */  \
465     if ((TOP) != NULL) {                                                       \
466       VerticalPred##SIZE((DST), (TOP));                                        \
467     } else {                                                                   \
468       FILL_8_OR_16((DST), 129, (SIZE));                                        \
469     }                                                                          \
470   }                                                                            \
471 }
472 
473 TRUE_MOTION(dst, left, top, 8)
474 TRUE_MOTION(dst, left, top, 16)
475 
476 #undef TRUE_MOTION
477 #undef CLIP_TO_DST
478 #undef CLIP_8B_TO_DST
479 #undef CLIPPING
480 
DCMode16(uint8_t * dst,const uint8_t * left,const uint8_t * top)481 static WEBP_INLINE void DCMode16(uint8_t* dst, const uint8_t* left,
482                                  const uint8_t* top) {
483   int DC, DC1;
484   int temp0, temp1, temp2, temp3;
485 
486   __asm__ volatile(
487     "beqz        %[top],   2f                  \n\t"
488     LOAD_WITH_OFFSET_X4(temp0, temp1, temp2, temp3, top,
489                         0, 4, 8, 12,
490                         0, 0, 0, 0,
491                         0)
492     "raddu.w.qb  %[temp0], %[temp0]            \n\t"
493     "raddu.w.qb  %[temp1], %[temp1]            \n\t"
494     "raddu.w.qb  %[temp2], %[temp2]            \n\t"
495     "raddu.w.qb  %[temp3], %[temp3]            \n\t"
496     "addu        %[temp0], %[temp0], %[temp1]  \n\t"
497     "addu        %[temp2], %[temp2], %[temp3]  \n\t"
498     "addu        %[DC],    %[temp0], %[temp2]  \n\t"
499     "move        %[DC1],   %[DC]               \n\t"
500     "beqz        %[left],  1f                  \n\t"
501     LOAD_WITH_OFFSET_X4(temp0, temp1, temp2, temp3, left,
502                         0, 4, 8, 12,
503                         0, 0, 0, 0,
504                         0)
505     "raddu.w.qb  %[temp0], %[temp0]            \n\t"
506     "raddu.w.qb  %[temp1], %[temp1]            \n\t"
507     "raddu.w.qb  %[temp2], %[temp2]            \n\t"
508     "raddu.w.qb  %[temp3], %[temp3]            \n\t"
509     "addu        %[temp0], %[temp0], %[temp1]  \n\t"
510     "addu        %[temp2], %[temp2], %[temp3]  \n\t"
511     "addu        %[DC1],   %[temp0], %[temp2]  \n\t"
512   "1:                                          \n\t"
513     "addu        %[DC],   %[DC],     %[DC1]    \n\t"
514     "j           3f                            \n\t"
515   "2:                                          \n\t"
516     "beqz        %[left],  4f                  \n\t"
517     LOAD_WITH_OFFSET_X4(temp0, temp1, temp2, temp3, left,
518                         0, 4, 8, 12,
519                         0, 0, 0, 0,
520                         0)
521     "raddu.w.qb  %[temp0], %[temp0]            \n\t"
522     "raddu.w.qb  %[temp1], %[temp1]            \n\t"
523     "raddu.w.qb  %[temp2], %[temp2]            \n\t"
524     "raddu.w.qb  %[temp3], %[temp3]            \n\t"
525     "addu        %[temp0], %[temp0], %[temp1]  \n\t"
526     "addu        %[temp2], %[temp2], %[temp3]  \n\t"
527     "addu        %[DC],    %[temp0], %[temp2]  \n\t"
528     "addu        %[DC],    %[DC],    %[DC]     \n\t"
529   "3:                                          \n\t"
530     "shra_r.w    %[DC],    %[DC],    5         \n\t"
531     "j           5f                            \n\t"
532   "4:                                          \n\t"
533     "li          %[DC],    0x80                \n\t"
534   "5:                                          \n\t"
535     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [DC]"=&r"(DC),
536       [temp2]"=&r"(temp2), [temp3]"=&r"(temp3), [DC1]"=&r"(DC1)
537     : [left]"r"(left), [top]"r"(top)
538     : "memory"
539   );
540 
541   FILL_8_OR_16(dst, DC, 16);
542 }
543 
DCMode8(uint8_t * dst,const uint8_t * left,const uint8_t * top)544 static WEBP_INLINE void DCMode8(uint8_t* dst, const uint8_t* left,
545                                 const uint8_t* top) {
546   int DC, DC1;
547   int temp0, temp1, temp2, temp3;
548 
549   __asm__ volatile(
550     "beqz        %[top],   2f                  \n\t"
551     "ulw         %[temp0], 0(%[top])           \n\t"
552     "ulw         %[temp1], 4(%[top])           \n\t"
553     "raddu.w.qb  %[temp0], %[temp0]            \n\t"
554     "raddu.w.qb  %[temp1], %[temp1]            \n\t"
555     "addu        %[DC],    %[temp0], %[temp1]  \n\t"
556     "move        %[DC1],   %[DC]               \n\t"
557     "beqz        %[left],  1f                  \n\t"
558     "ulw         %[temp2], 0(%[left])          \n\t"
559     "ulw         %[temp3], 4(%[left])          \n\t"
560     "raddu.w.qb  %[temp2], %[temp2]            \n\t"
561     "raddu.w.qb  %[temp3], %[temp3]            \n\t"
562     "addu        %[DC1],   %[temp2], %[temp3]  \n\t"
563   "1:                                          \n\t"
564     "addu        %[DC],    %[DC],    %[DC1]    \n\t"
565     "j           3f                            \n\t"
566   "2:                                          \n\t"
567     "beqz        %[left],  4f                  \n\t"
568     "ulw         %[temp2], 0(%[left])          \n\t"
569     "ulw         %[temp3], 4(%[left])          \n\t"
570     "raddu.w.qb  %[temp2], %[temp2]            \n\t"
571     "raddu.w.qb  %[temp3], %[temp3]            \n\t"
572     "addu        %[DC],    %[temp2], %[temp3]  \n\t"
573     "addu        %[DC],    %[DC],    %[DC]     \n\t"
574   "3:                                          \n\t"
575     "shra_r.w    %[DC], %[DC], 4               \n\t"
576     "j           5f                            \n\t"
577   "4:                                          \n\t"
578     "li          %[DC], 0x80                   \n\t"
579   "5:                                          \n\t"
580     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [DC]"=&r"(DC),
581       [temp2]"=&r"(temp2), [temp3]"=&r"(temp3), [DC1]"=&r"(DC1)
582     : [left]"r"(left), [top]"r"(top)
583     : "memory"
584   );
585 
586   FILL_8_OR_16(dst, DC, 8);
587 }
588 
DC4(uint8_t * dst,const uint8_t * top)589 static void DC4(uint8_t* dst, const uint8_t* top) {
590   int temp0, temp1;
591   __asm__ volatile(
592     "ulw          %[temp0],   0(%[top])               \n\t"
593     "ulw          %[temp1],   -5(%[top])              \n\t"
594     "raddu.w.qb   %[temp0],   %[temp0]                \n\t"
595     "raddu.w.qb   %[temp1],   %[temp1]                \n\t"
596     "addu         %[temp0],   %[temp0],    %[temp1]   \n\t"
597     "addiu        %[temp0],   %[temp0],    4          \n\t"
598     "srl          %[temp0],   %[temp0],    3          \n\t"
599     "replv.qb     %[temp0],   %[temp0]                \n\t"
600     "usw          %[temp0],   0*" XSTR(BPS) "(%[dst]) \n\t"
601     "usw          %[temp0],   1*" XSTR(BPS) "(%[dst]) \n\t"
602     "usw          %[temp0],   2*" XSTR(BPS) "(%[dst]) \n\t"
603     "usw          %[temp0],   3*" XSTR(BPS) "(%[dst]) \n\t"
604     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1)
605     : [top]"r"(top), [dst]"r"(dst)
606     : "memory"
607   );
608 }
609 
TM4(uint8_t * dst,const uint8_t * top)610 static void TM4(uint8_t* dst, const uint8_t* top) {
611   int a10, a32, temp0, temp1, temp2, temp3, temp4, temp5;
612   const int c35 = 0xff00ff;
613   __asm__ volatile (
614     "lbu              %[temp1],  0(%[top])                     \n\t"
615     "lbu              %[a10],    1(%[top])                     \n\t"
616     "lbu              %[temp2],  2(%[top])                     \n\t"
617     "lbu              %[a32],    3(%[top])                     \n\t"
618     "ulw              %[temp0],  -5(%[top])                    \n\t"
619     "lbu              %[temp4],  -1(%[top])                    \n\t"
620     "append           %[a10],    %[temp1],   16                \n\t"
621     "append           %[a32],    %[temp2],   16                \n\t"
622     "replv.ph         %[temp4],  %[temp4]                      \n\t"
623     "shrl.ph          %[temp1],  %[temp0],   8                 \n\t"
624     "and              %[temp0],  %[temp0],   %[c35]            \n\t"
625     "subu.ph          %[temp1],  %[temp1],   %[temp4]          \n\t"
626     "subu.ph          %[temp0],  %[temp0],   %[temp4]          \n\t"
627     "srl              %[temp2],  %[temp1],   16                \n\t"
628     "srl              %[temp3],  %[temp0],   16                \n\t"
629     "replv.ph         %[temp2],  %[temp2]                      \n\t"
630     "replv.ph         %[temp3],  %[temp3]                      \n\t"
631     "replv.ph         %[temp4],  %[temp1]                      \n\t"
632     "replv.ph         %[temp5],  %[temp0]                      \n\t"
633     "addu.ph          %[temp0],  %[temp3],   %[a10]            \n\t"
634     "addu.ph          %[temp1],  %[temp3],   %[a32]            \n\t"
635     "addu.ph          %[temp3],  %[temp2],   %[a10]            \n\t"
636     "addu.ph          %[temp2],  %[temp2],   %[a32]            \n\t"
637     "shll_s.ph        %[temp0],  %[temp0],   7                 \n\t"
638     "shll_s.ph        %[temp1],  %[temp1],   7                 \n\t"
639     "shll_s.ph        %[temp3],  %[temp3],   7                 \n\t"
640     "shll_s.ph        %[temp2],  %[temp2],   7                 \n\t"
641     "precrqu_s.qb.ph  %[temp0],  %[temp1],   %[temp0]          \n\t"
642     "precrqu_s.qb.ph  %[temp1],  %[temp2],   %[temp3]          \n\t"
643     "addu.ph          %[temp2],  %[temp5],   %[a10]            \n\t"
644     "addu.ph          %[temp3],  %[temp5],   %[a32]            \n\t"
645     "addu.ph          %[temp5],  %[temp4],   %[a10]            \n\t"
646     "addu.ph          %[temp4],  %[temp4],   %[a32]            \n\t"
647     "shll_s.ph        %[temp2],  %[temp2],   7                 \n\t"
648     "shll_s.ph        %[temp3],  %[temp3],   7                 \n\t"
649     "shll_s.ph        %[temp4],  %[temp4],   7                 \n\t"
650     "shll_s.ph        %[temp5],  %[temp5],   7                 \n\t"
651     "precrqu_s.qb.ph  %[temp2],  %[temp3],   %[temp2]          \n\t"
652     "precrqu_s.qb.ph  %[temp3],  %[temp4],   %[temp5]          \n\t"
653     "usw              %[temp1],  0*" XSTR(BPS) "(%[dst])       \n\t"
654     "usw              %[temp0],  1*" XSTR(BPS) "(%[dst])       \n\t"
655     "usw              %[temp3],  2*" XSTR(BPS) "(%[dst])       \n\t"
656     "usw              %[temp2],  3*" XSTR(BPS) "(%[dst])       \n\t"
657     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
658       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
659       [a10]"=&r"(a10), [a32]"=&r"(a32)
660     : [c35]"r"(c35), [top]"r"(top), [dst]"r"(dst)
661     : "memory"
662   );
663 }
664 
VE4(uint8_t * dst,const uint8_t * top)665 static void VE4(uint8_t* dst, const uint8_t* top) {
666   int temp0, temp1, temp2, temp3, temp4, temp5, temp6;
667   __asm__ volatile(
668     "ulw             %[temp0],   -1(%[top])              \n\t"
669     "ulh             %[temp1],   3(%[top])               \n\t"
670     "preceu.ph.qbr   %[temp2],   %[temp0]                \n\t"
671     "preceu.ph.qbl   %[temp3],   %[temp0]                \n\t"
672     "preceu.ph.qbr   %[temp4],   %[temp1]                \n\t"
673     "packrl.ph       %[temp5],   %[temp3],    %[temp2]   \n\t"
674     "packrl.ph       %[temp6],   %[temp4],    %[temp3]   \n\t"
675     "shll.ph         %[temp5],   %[temp5],    1          \n\t"
676     "shll.ph         %[temp6],   %[temp6],    1          \n\t"
677     "addq.ph         %[temp2],   %[temp5],    %[temp2]   \n\t"
678     "addq.ph         %[temp6],   %[temp6],    %[temp4]   \n\t"
679     "addq.ph         %[temp2],   %[temp2],    %[temp3]   \n\t"
680     "addq.ph         %[temp6],   %[temp6],    %[temp3]   \n\t"
681     "shra_r.ph       %[temp2],   %[temp2],    2          \n\t"
682     "shra_r.ph       %[temp6],   %[temp6],    2          \n\t"
683     "precr.qb.ph     %[temp4],   %[temp6],    %[temp2]   \n\t"
684     "usw             %[temp4],   0*" XSTR(BPS) "(%[dst]) \n\t"
685     "usw             %[temp4],   1*" XSTR(BPS) "(%[dst]) \n\t"
686     "usw             %[temp4],   2*" XSTR(BPS) "(%[dst]) \n\t"
687     "usw             %[temp4],   3*" XSTR(BPS) "(%[dst]) \n\t"
688     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
689       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
690       [temp6]"=&r"(temp6)
691     : [top]"r"(top), [dst]"r"(dst)
692     : "memory"
693   );
694 }
695 
HE4(uint8_t * dst,const uint8_t * top)696 static void HE4(uint8_t* dst, const uint8_t* top) {
697   int temp0, temp1, temp2, temp3, temp4, temp5, temp6;
698   __asm__ volatile(
699     "ulw             %[temp0],   -4(%[top])              \n\t"
700     "lbu             %[temp1],   -5(%[top])              \n\t"
701     "preceu.ph.qbr   %[temp2],   %[temp0]                \n\t"
702     "preceu.ph.qbl   %[temp3],   %[temp0]                \n\t"
703     "replv.ph        %[temp4],   %[temp1]                \n\t"
704     "packrl.ph       %[temp5],   %[temp3],    %[temp2]   \n\t"
705     "packrl.ph       %[temp6],   %[temp2],    %[temp4]   \n\t"
706     "shll.ph         %[temp5],   %[temp5],    1          \n\t"
707     "shll.ph         %[temp6],   %[temp6],    1          \n\t"
708     "addq.ph         %[temp3],   %[temp3],    %[temp5]   \n\t"
709     "addq.ph         %[temp3],   %[temp3],    %[temp2]   \n\t"
710     "addq.ph         %[temp2],   %[temp2],    %[temp6]   \n\t"
711     "addq.ph         %[temp2],   %[temp2],    %[temp4]   \n\t"
712     "shra_r.ph       %[temp3],   %[temp3],    2          \n\t"
713     "shra_r.ph       %[temp2],   %[temp2],    2          \n\t"
714     "replv.qb        %[temp0],   %[temp3]                \n\t"
715     "replv.qb        %[temp1],   %[temp2]                \n\t"
716     "srl             %[temp3],   %[temp3],    16         \n\t"
717     "srl             %[temp2],   %[temp2],    16         \n\t"
718     "replv.qb        %[temp3],   %[temp3]                \n\t"
719     "replv.qb        %[temp2],   %[temp2]                \n\t"
720     "usw             %[temp3],   0*" XSTR(BPS) "(%[dst]) \n\t"
721     "usw             %[temp0],   1*" XSTR(BPS) "(%[dst]) \n\t"
722     "usw             %[temp2],   2*" XSTR(BPS) "(%[dst]) \n\t"
723     "usw             %[temp1],   3*" XSTR(BPS) "(%[dst]) \n\t"
724     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
725       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
726       [temp6]"=&r"(temp6)
727     : [top]"r"(top), [dst]"r"(dst)
728     : "memory"
729   );
730 }
731 
RD4(uint8_t * dst,const uint8_t * top)732 static void RD4(uint8_t* dst, const uint8_t* top) {
733   int temp0, temp1, temp2, temp3, temp4, temp5;
734   int temp6, temp7, temp8, temp9, temp10, temp11;
735   __asm__ volatile(
736     "ulw             %[temp0],    -5(%[top])               \n\t"
737     "ulw             %[temp1],    -1(%[top])               \n\t"
738     "preceu.ph.qbl   %[temp2],    %[temp0]                 \n\t"
739     "preceu.ph.qbr   %[temp3],    %[temp0]                 \n\t"
740     "preceu.ph.qbr   %[temp4],    %[temp1]                 \n\t"
741     "preceu.ph.qbl   %[temp5],    %[temp1]                 \n\t"
742     "packrl.ph       %[temp6],    %[temp2],    %[temp3]    \n\t"
743     "packrl.ph       %[temp7],    %[temp4],    %[temp2]    \n\t"
744     "packrl.ph       %[temp8],    %[temp5],    %[temp4]    \n\t"
745     "shll.ph         %[temp6],    %[temp6],    1           \n\t"
746     "addq.ph         %[temp9],    %[temp2],    %[temp6]    \n\t"
747     "shll.ph         %[temp7],    %[temp7],    1           \n\t"
748     "addq.ph         %[temp9],    %[temp9],    %[temp3]    \n\t"
749     "shll.ph         %[temp8],    %[temp8],    1           \n\t"
750     "shra_r.ph       %[temp9],    %[temp9],    2           \n\t"
751     "addq.ph         %[temp10],   %[temp4],    %[temp7]    \n\t"
752     "addq.ph         %[temp11],   %[temp5],    %[temp8]    \n\t"
753     "addq.ph         %[temp10],   %[temp10],   %[temp2]    \n\t"
754     "addq.ph         %[temp11],   %[temp11],   %[temp4]    \n\t"
755     "shra_r.ph       %[temp10],   %[temp10],   2           \n\t"
756     "shra_r.ph       %[temp11],   %[temp11],   2           \n\t"
757     "lbu             %[temp0],    3(%[top])                \n\t"
758     "lbu             %[temp1],    2(%[top])                \n\t"
759     "lbu             %[temp2],    1(%[top])                \n\t"
760     "sll             %[temp1],    %[temp1],    1           \n\t"
761     "addu            %[temp0],    %[temp0],    %[temp1]    \n\t"
762     "addu            %[temp0],    %[temp0],    %[temp2]    \n\t"
763     "precr.qb.ph     %[temp9],    %[temp10],   %[temp9]    \n\t"
764     "shra_r.w        %[temp0],    %[temp0],    2           \n\t"
765     "precr.qb.ph     %[temp10],   %[temp11],   %[temp10]   \n\t"
766     "usw             %[temp9],    3*" XSTR(BPS) "(%[dst])  \n\t"
767     "usw             %[temp10],   1*" XSTR(BPS) "(%[dst])  \n\t"
768     "prepend         %[temp9],    %[temp11],   8           \n\t"
769     "prepend         %[temp10],   %[temp0],    8           \n\t"
770     "usw             %[temp9],    2*" XSTR(BPS) "(%[dst])  \n\t"
771     "usw             %[temp10],   0*" XSTR(BPS) "(%[dst])  \n\t"
772     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
773       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
774       [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [temp8]"=&r"(temp8),
775       [temp9]"=&r"(temp9), [temp10]"=&r"(temp10), [temp11]"=&r"(temp11)
776     : [top]"r"(top), [dst]"r"(dst)
777     : "memory"
778   );
779 }
780 
VR4(uint8_t * dst,const uint8_t * top)781 static void VR4(uint8_t* dst, const uint8_t* top) {
782   int temp0, temp1, temp2, temp3, temp4;
783   int temp5, temp6, temp7, temp8, temp9;
784   __asm__ volatile (
785     "ulw              %[temp0],   -4(%[top])              \n\t"
786     "ulw              %[temp1],   0(%[top])               \n\t"
787     "preceu.ph.qbl    %[temp2],   %[temp0]                \n\t"
788     "preceu.ph.qbr    %[temp0],   %[temp0]                \n\t"
789     "preceu.ph.qbla   %[temp3],   %[temp1]                \n\t"
790     "preceu.ph.qbra   %[temp1],   %[temp1]                \n\t"
791     "packrl.ph        %[temp7],   %[temp3],    %[temp2]   \n\t"
792     "addqh_r.ph       %[temp4],   %[temp1],    %[temp3]   \n\t"
793     "move             %[temp6],   %[temp1]                \n\t"
794     "append           %[temp1],   %[temp2],    16         \n\t"
795     "shll.ph          %[temp9],   %[temp6],    1          \n\t"
796     "addqh_r.ph       %[temp5],   %[temp7],    %[temp6]   \n\t"
797     "shll.ph          %[temp8],   %[temp7],    1          \n\t"
798     "addu.ph          %[temp3],   %[temp7],    %[temp3]   \n\t"
799     "addu.ph          %[temp1],   %[temp1],    %[temp6]   \n\t"
800     "packrl.ph        %[temp7],   %[temp2],    %[temp0]   \n\t"
801     "addu.ph          %[temp6],   %[temp0],    %[temp2]   \n\t"
802     "addu.ph          %[temp3],   %[temp3],    %[temp9]   \n\t"
803     "addu.ph          %[temp1],   %[temp1],    %[temp8]   \n\t"
804     "shll.ph          %[temp7],   %[temp7],    1          \n\t"
805     "shra_r.ph        %[temp3],   %[temp3],    2          \n\t"
806     "shra_r.ph        %[temp1],   %[temp1],    2          \n\t"
807     "addu.ph          %[temp6],   %[temp6],    %[temp7]   \n\t"
808     "shra_r.ph        %[temp6],   %[temp6],    2          \n\t"
809     "precrq.ph.w      %[temp8],   %[temp4],    %[temp5]   \n\t"
810     "append           %[temp4],   %[temp5],    16         \n\t"
811     "precrq.ph.w      %[temp2],   %[temp3],    %[temp1]   \n\t"
812     "append           %[temp3],   %[temp1],    16         \n\t"
813     "precr.qb.ph      %[temp8],   %[temp8],    %[temp4]   \n\t"
814     "precr.qb.ph      %[temp3],   %[temp2],    %[temp3]   \n\t"
815     "usw              %[temp8],   0*" XSTR(BPS) "(%[dst]) \n\t"
816     "usw              %[temp3],   1*" XSTR(BPS) "(%[dst]) \n\t"
817     "append           %[temp3],   %[temp6],    8          \n\t"
818     "srl              %[temp6],   %[temp6],    16         \n\t"
819     "append           %[temp8],   %[temp6],    8          \n\t"
820     "usw              %[temp3],   3*" XSTR(BPS) "(%[dst]) \n\t"
821     "usw              %[temp8],   2*" XSTR(BPS) "(%[dst]) \n\t"
822     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
823       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
824       [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [temp8]"=&r"(temp8),
825       [temp9]"=&r"(temp9)
826     : [top]"r"(top), [dst]"r"(dst)
827     : "memory"
828   );
829 }
830 
LD4(uint8_t * dst,const uint8_t * top)831 static void LD4(uint8_t* dst, const uint8_t* top) {
832   int temp0, temp1, temp2, temp3, temp4, temp5;
833   int temp6, temp7, temp8, temp9, temp10, temp11;
834   __asm__ volatile(
835     "ulw             %[temp0],    0(%[top])               \n\t"
836     "ulw             %[temp1],    4(%[top])               \n\t"
837     "preceu.ph.qbl   %[temp2],    %[temp0]                \n\t"
838     "preceu.ph.qbr   %[temp3],    %[temp0]                \n\t"
839     "preceu.ph.qbr   %[temp4],    %[temp1]                \n\t"
840     "preceu.ph.qbl   %[temp5],    %[temp1]                \n\t"
841     "packrl.ph       %[temp6],    %[temp2],    %[temp3]   \n\t"
842     "packrl.ph       %[temp7],    %[temp4],    %[temp2]   \n\t"
843     "packrl.ph       %[temp8],    %[temp5],    %[temp4]   \n\t"
844     "shll.ph         %[temp6],    %[temp6],    1          \n\t"
845     "addq.ph         %[temp9],    %[temp2],    %[temp6]   \n\t"
846     "shll.ph         %[temp7],    %[temp7],    1          \n\t"
847     "addq.ph         %[temp9],    %[temp9],    %[temp3]   \n\t"
848     "shll.ph         %[temp8],    %[temp8],    1          \n\t"
849     "shra_r.ph       %[temp9],    %[temp9],    2          \n\t"
850     "addq.ph         %[temp10],   %[temp4],    %[temp7]   \n\t"
851     "addq.ph         %[temp11],   %[temp5],    %[temp8]   \n\t"
852     "addq.ph         %[temp10],   %[temp10],   %[temp2]   \n\t"
853     "addq.ph         %[temp11],   %[temp11],   %[temp4]   \n\t"
854     "shra_r.ph       %[temp10],   %[temp10],   2          \n\t"
855     "shra_r.ph       %[temp11],   %[temp11],   2          \n\t"
856     "srl             %[temp1],    %[temp1],    24         \n\t"
857     "sll             %[temp1],    %[temp1],    1          \n\t"
858     "raddu.w.qb      %[temp5],    %[temp5]                \n\t"
859     "precr.qb.ph     %[temp9],    %[temp10],   %[temp9]   \n\t"
860     "precr.qb.ph     %[temp10],   %[temp11],   %[temp10]  \n\t"
861     "addu            %[temp1],    %[temp1],    %[temp5]   \n\t"
862     "shra_r.w        %[temp1],    %[temp1],    2          \n\t"
863     "usw             %[temp9],    0*" XSTR(BPS) "(%[dst]) \n\t"
864     "usw             %[temp10],   2*" XSTR(BPS) "(%[dst]) \n\t"
865     "prepend         %[temp9],    %[temp11],   8          \n\t"
866     "prepend         %[temp10],   %[temp1],    8          \n\t"
867     "usw             %[temp9],    1*" XSTR(BPS) "(%[dst]) \n\t"
868     "usw             %[temp10],   3*" XSTR(BPS) "(%[dst]) \n\t"
869     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
870       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
871       [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [temp8]"=&r"(temp8),
872       [temp9]"=&r"(temp9), [temp10]"=&r"(temp10), [temp11]"=&r"(temp11)
873     : [top]"r"(top), [dst]"r"(dst)
874     : "memory"
875   );
876 }
877 
VL4(uint8_t * dst,const uint8_t * top)878 static void VL4(uint8_t* dst, const uint8_t* top) {
879   int temp0, temp1, temp2, temp3, temp4;
880   int temp5, temp6, temp7, temp8, temp9;
881   __asm__ volatile (
882     "ulw              %[temp0],   0(%[top])               \n\t"
883     "ulw              %[temp1],   4(%[top])               \n\t"
884     "preceu.ph.qbla   %[temp2],   %[temp0]                \n\t"
885     "preceu.ph.qbra   %[temp0],   %[temp0]                \n\t"
886     "preceu.ph.qbl    %[temp3],   %[temp1]                \n\t"
887     "preceu.ph.qbr    %[temp1],   %[temp1]                \n\t"
888     "addqh_r.ph       %[temp4],   %[temp0],    %[temp2]   \n\t"
889     "packrl.ph        %[temp7],   %[temp1],    %[temp0]   \n\t"
890     "precrq.ph.w      %[temp6],   %[temp1],    %[temp2]   \n\t"
891     "shll.ph          %[temp9],   %[temp2],    1          \n\t"
892     "addqh_r.ph       %[temp5],   %[temp7],    %[temp2]   \n\t"
893     "shll.ph          %[temp8],   %[temp7],    1          \n\t"
894     "addu.ph          %[temp2],   %[temp2],    %[temp6]   \n\t"
895     "addu.ph          %[temp0],   %[temp0],    %[temp7]   \n\t"
896     "packrl.ph        %[temp7],   %[temp3],    %[temp1]   \n\t"
897     "addu.ph          %[temp6],   %[temp1],    %[temp3]   \n\t"
898     "addu.ph          %[temp2],   %[temp2],    %[temp8]   \n\t"
899     "addu.ph          %[temp0],   %[temp0],    %[temp9]   \n\t"
900     "shll.ph          %[temp7],   %[temp7],    1          \n\t"
901     "shra_r.ph        %[temp2],   %[temp2],    2          \n\t"
902     "shra_r.ph        %[temp0],   %[temp0],    2          \n\t"
903     "addu.ph          %[temp6],   %[temp6],    %[temp7]   \n\t"
904     "shra_r.ph        %[temp6],   %[temp6],    2          \n\t"
905     "precrq.ph.w      %[temp8],   %[temp5],    %[temp4]   \n\t"
906     "append           %[temp5],   %[temp4],    16         \n\t"
907     "precrq.ph.w      %[temp3],   %[temp2],    %[temp0]   \n\t"
908     "append           %[temp2],   %[temp0],    16         \n\t"
909     "precr.qb.ph      %[temp8],   %[temp8],    %[temp5]   \n\t"
910     "precr.qb.ph      %[temp3],   %[temp3],    %[temp2]   \n\t"
911     "usw              %[temp8],   0*" XSTR(BPS) "(%[dst]) \n\t"
912     "prepend          %[temp8],   %[temp6],    8          \n\t"
913     "usw              %[temp3],   1*" XSTR(BPS) "(%[dst]) \n\t"
914     "srl              %[temp6],   %[temp6],    16         \n\t"
915     "prepend          %[temp3],   %[temp6],    8          \n\t"
916     "usw              %[temp8],   2*" XSTR(BPS) "(%[dst]) \n\t"
917     "usw              %[temp3],   3*" XSTR(BPS) "(%[dst]) \n\t"
918     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
919       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
920       [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [temp8]"=&r"(temp8),
921       [temp9]"=&r"(temp9)
922     : [top]"r"(top), [dst]"r"(dst)
923     : "memory"
924   );
925 }
926 
HD4(uint8_t * dst,const uint8_t * top)927 static void HD4(uint8_t* dst, const uint8_t* top) {
928   int temp0, temp1, temp2, temp3, temp4;
929   int temp5, temp6, temp7, temp8, temp9;
930   __asm__ volatile (
931     "ulw              %[temp0],   -5(%[top])              \n\t"
932     "ulw              %[temp1],   -1(%[top])              \n\t"
933     "preceu.ph.qbla   %[temp2],   %[temp0]                \n\t"
934     "preceu.ph.qbra   %[temp0],   %[temp0]                \n\t"
935     "preceu.ph.qbl    %[temp3],   %[temp1]                \n\t"
936     "preceu.ph.qbr    %[temp1],   %[temp1]                \n\t"
937     "addqh_r.ph       %[temp4],   %[temp0],    %[temp2]   \n\t"
938     "packrl.ph        %[temp7],   %[temp1],    %[temp0]   \n\t"
939     "precrq.ph.w      %[temp6],   %[temp1],    %[temp2]   \n\t"
940     "shll.ph          %[temp9],   %[temp2],    1          \n\t"
941     "addqh_r.ph       %[temp5],   %[temp7],    %[temp2]   \n\t"
942     "shll.ph          %[temp8],   %[temp7],    1          \n\t"
943     "addu.ph          %[temp2],   %[temp2],    %[temp6]   \n\t"
944     "addu.ph          %[temp0],   %[temp0],    %[temp7]   \n\t"
945     "packrl.ph        %[temp7],   %[temp3],    %[temp1]   \n\t"
946     "addu.ph          %[temp6],   %[temp1],    %[temp3]   \n\t"
947     "addu.ph          %[temp2],   %[temp2],    %[temp8]   \n\t"
948     "addu.ph          %[temp0],   %[temp0],    %[temp9]   \n\t"
949     "shll.ph          %[temp7],   %[temp7],    1          \n\t"
950     "shra_r.ph        %[temp2],   %[temp2],    2          \n\t"
951     "shra_r.ph        %[temp0],   %[temp0],    2          \n\t"
952     "addu.ph          %[temp6],   %[temp6],    %[temp7]   \n\t"
953     "shra_r.ph        %[temp6],   %[temp6],    2          \n\t"
954     "precrq.ph.w      %[temp1],   %[temp2],    %[temp5]   \n\t"
955     "precrq.ph.w      %[temp3],   %[temp0],    %[temp4]   \n\t"
956     "precr.qb.ph      %[temp7],   %[temp6],    %[temp1]   \n\t"
957     "precr.qb.ph      %[temp6],   %[temp1],    %[temp3]   \n\t"
958     "usw              %[temp7],   0*" XSTR(BPS) "(%[dst]) \n\t"
959     "usw              %[temp6],   1*" XSTR(BPS) "(%[dst]) \n\t"
960     "append           %[temp2],   %[temp5],    16         \n\t"
961     "append           %[temp0],   %[temp4],    16         \n\t"
962     "precr.qb.ph      %[temp5],   %[temp3],    %[temp2]   \n\t"
963     "precr.qb.ph      %[temp4],   %[temp2],    %[temp0]   \n\t"
964     "usw              %[temp5],   2*" XSTR(BPS) "(%[dst]) \n\t"
965     "usw              %[temp4],   3*" XSTR(BPS) "(%[dst]) \n\t"
966     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
967       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
968       [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [temp8]"=&r"(temp8),
969       [temp9]"=&r"(temp9)
970     : [top]"r"(top), [dst]"r"(dst)
971     : "memory"
972   );
973 }
974 
HU4(uint8_t * dst,const uint8_t * top)975 static void HU4(uint8_t* dst, const uint8_t* top) {
976   int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
977   __asm__ volatile (
978     "ulw             %[temp0],   -5(%[top])              \n\t"
979     "preceu.ph.qbl   %[temp1],   %[temp0]                \n\t"
980     "preceu.ph.qbr   %[temp2],   %[temp0]                \n\t"
981     "packrl.ph       %[temp3],   %[temp1],    %[temp2]   \n\t"
982     "replv.qb        %[temp7],   %[temp2]                \n\t"
983     "addqh_r.ph      %[temp4],   %[temp1],    %[temp3]   \n\t"
984     "addqh_r.ph      %[temp5],   %[temp3],    %[temp2]   \n\t"
985     "shll.ph         %[temp6],   %[temp3],    1          \n\t"
986     "addu.ph         %[temp3],   %[temp2],    %[temp3]   \n\t"
987     "addu.ph         %[temp6],   %[temp1],    %[temp6]   \n\t"
988     "shll.ph         %[temp0],   %[temp2],    1          \n\t"
989     "addu.ph         %[temp6],   %[temp6],    %[temp2]   \n\t"
990     "addu.ph         %[temp0],   %[temp3],    %[temp0]   \n\t"
991     "shra_r.ph       %[temp6],   %[temp6],    2          \n\t"
992     "shra_r.ph       %[temp0],   %[temp0],    2          \n\t"
993     "packrl.ph       %[temp3],   %[temp6],    %[temp5]   \n\t"
994     "precrq.ph.w     %[temp2],   %[temp6],    %[temp4]   \n\t"
995     "append          %[temp0],   %[temp5],    16         \n\t"
996     "precr.qb.ph     %[temp3],   %[temp3],    %[temp2]   \n\t"
997     "usw             %[temp3],   0*" XSTR(BPS) "(%[dst]) \n\t"
998     "precr.qb.ph     %[temp1],   %[temp7],    %[temp0]   \n\t"
999     "usw             %[temp7],   3*" XSTR(BPS) "(%[dst]) \n\t"
1000     "packrl.ph       %[temp2],   %[temp1],    %[temp3]   \n\t"
1001     "usw             %[temp1],   2*" XSTR(BPS) "(%[dst]) \n\t"
1002     "usw             %[temp2],   1*" XSTR(BPS) "(%[dst]) \n\t"
1003     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
1004       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
1005       [temp6]"=&r"(temp6), [temp7]"=&r"(temp7)
1006     : [top]"r"(top), [dst]"r"(dst)
1007     : "memory"
1008   );
1009 }
1010 
1011 //------------------------------------------------------------------------------
1012 // Chroma 8x8 prediction (paragraph 12.2)
1013 
IntraChromaPreds(uint8_t * dst,const uint8_t * left,const uint8_t * top)1014 static void IntraChromaPreds(uint8_t* dst, const uint8_t* left,
1015                              const uint8_t* top) {
1016   // U block
1017   DCMode8(C8DC8 + dst, left, top);
1018   VerticalPred8(C8VE8 + dst, top);
1019   HorizontalPred8(C8HE8 + dst, left);
1020   TrueMotion8(C8TM8 + dst, left, top);
1021   // V block
1022   dst += 8;
1023   if (top) top += 8;
1024   if (left) left += 16;
1025   DCMode8(C8DC8 + dst, left, top);
1026   VerticalPred8(C8VE8 + dst, top);
1027   HorizontalPred8(C8HE8 + dst, left);
1028   TrueMotion8(C8TM8 + dst, left, top);
1029 }
1030 
1031 //------------------------------------------------------------------------------
1032 // luma 16x16 prediction (paragraph 12.3)
1033 
Intra16Preds(uint8_t * dst,const uint8_t * left,const uint8_t * top)1034 static void Intra16Preds(uint8_t* dst,
1035                          const uint8_t* left, const uint8_t* top) {
1036   DCMode16(I16DC16 + dst, left, top);
1037   VerticalPred16(I16VE16 + dst, top);
1038   HorizontalPred16(I16HE16 + dst, left);
1039   TrueMotion16(I16TM16 + dst, left, top);
1040 }
1041 
1042 // Left samples are top[-5 .. -2], top_left is top[-1], top are
1043 // located at top[0..3], and top right is top[4..7]
Intra4Preds(uint8_t * dst,const uint8_t * top)1044 static void Intra4Preds(uint8_t* dst, const uint8_t* top) {
1045   DC4(I4DC4 + dst, top);
1046   TM4(I4TM4 + dst, top);
1047   VE4(I4VE4 + dst, top);
1048   HE4(I4HE4 + dst, top);
1049   RD4(I4RD4 + dst, top);
1050   VR4(I4VR4 + dst, top);
1051   LD4(I4LD4 + dst, top);
1052   VL4(I4VL4 + dst, top);
1053   HD4(I4HD4 + dst, top);
1054   HU4(I4HU4 + dst, top);
1055 }
1056 
1057 //------------------------------------------------------------------------------
1058 // Metric
1059 
1060 #if !defined(WORK_AROUND_GCC)
1061 
1062 #define GET_SSE_INNER(A)                                                  \
1063   "lw               %[temp0],    " #A "(%[a])                  \n\t"      \
1064   "lw               %[temp1],    " #A "(%[b])                  \n\t"      \
1065   "preceu.ph.qbr    %[temp2],    %[temp0]                      \n\t"      \
1066   "preceu.ph.qbl    %[temp0],    %[temp0]                      \n\t"      \
1067   "preceu.ph.qbr    %[temp3],    %[temp1]                      \n\t"      \
1068   "preceu.ph.qbl    %[temp1],    %[temp1]                      \n\t"      \
1069   "subq.ph          %[temp2],    %[temp2],    %[temp3]         \n\t"      \
1070   "subq.ph          %[temp0],    %[temp0],    %[temp1]         \n\t"      \
1071   "dpa.w.ph         $ac0,        %[temp2],    %[temp2]         \n\t"      \
1072   "dpa.w.ph         $ac0,        %[temp0],    %[temp0]         \n\t"
1073 
1074 #define GET_SSE(A, B, C, D)               \
1075   GET_SSE_INNER(A)                        \
1076   GET_SSE_INNER(B)                        \
1077   GET_SSE_INNER(C)                        \
1078   GET_SSE_INNER(D)
1079 
SSE16x16(const uint8_t * a,const uint8_t * b)1080 static int SSE16x16(const uint8_t* a, const uint8_t* b) {
1081   int count;
1082   int temp0, temp1, temp2, temp3;
1083   __asm__ volatile (
1084     "mult   $zero,    $zero                            \n\t"
1085     GET_SSE( 0 * BPS, 4 +  0 * BPS, 8 +  0 * BPS, 12 +  0 * BPS)
1086     GET_SSE( 1 * BPS, 4 +  1 * BPS, 8 +  1 * BPS, 12 +  1 * BPS)
1087     GET_SSE( 2 * BPS, 4 +  2 * BPS, 8 +  2 * BPS, 12 +  2 * BPS)
1088     GET_SSE( 3 * BPS, 4 +  3 * BPS, 8 +  3 * BPS, 12 +  3 * BPS)
1089     GET_SSE( 4 * BPS, 4 +  4 * BPS, 8 +  4 * BPS, 12 +  4 * BPS)
1090     GET_SSE( 5 * BPS, 4 +  5 * BPS, 8 +  5 * BPS, 12 +  5 * BPS)
1091     GET_SSE( 6 * BPS, 4 +  6 * BPS, 8 +  6 * BPS, 12 +  6 * BPS)
1092     GET_SSE( 7 * BPS, 4 +  7 * BPS, 8 +  7 * BPS, 12 +  7 * BPS)
1093     GET_SSE( 8 * BPS, 4 +  8 * BPS, 8 +  8 * BPS, 12 +  8 * BPS)
1094     GET_SSE( 9 * BPS, 4 +  9 * BPS, 8 +  9 * BPS, 12 +  9 * BPS)
1095     GET_SSE(10 * BPS, 4 + 10 * BPS, 8 + 10 * BPS, 12 + 10 * BPS)
1096     GET_SSE(11 * BPS, 4 + 11 * BPS, 8 + 11 * BPS, 12 + 11 * BPS)
1097     GET_SSE(12 * BPS, 4 + 12 * BPS, 8 + 12 * BPS, 12 + 12 * BPS)
1098     GET_SSE(13 * BPS, 4 + 13 * BPS, 8 + 13 * BPS, 12 + 13 * BPS)
1099     GET_SSE(14 * BPS, 4 + 14 * BPS, 8 + 14 * BPS, 12 + 14 * BPS)
1100     GET_SSE(15 * BPS, 4 + 15 * BPS, 8 + 15 * BPS, 12 + 15 * BPS)
1101     "mflo   %[count]                                   \n\t"
1102     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
1103       [temp3]"=&r"(temp3), [count]"=&r"(count)
1104     : [a]"r"(a), [b]"r"(b)
1105     : "memory", "hi", "lo"
1106   );
1107   return count;
1108 }
1109 
SSE16x8(const uint8_t * a,const uint8_t * b)1110 static int SSE16x8(const uint8_t* a, const uint8_t* b) {
1111   int count;
1112   int temp0, temp1, temp2, temp3;
1113   __asm__ volatile (
1114     "mult   $zero,    $zero                            \n\t"
1115     GET_SSE( 0 * BPS, 4 +  0 * BPS, 8 +  0 * BPS, 12 +  0 * BPS)
1116     GET_SSE( 1 * BPS, 4 +  1 * BPS, 8 +  1 * BPS, 12 +  1 * BPS)
1117     GET_SSE( 2 * BPS, 4 +  2 * BPS, 8 +  2 * BPS, 12 +  2 * BPS)
1118     GET_SSE( 3 * BPS, 4 +  3 * BPS, 8 +  3 * BPS, 12 +  3 * BPS)
1119     GET_SSE( 4 * BPS, 4 +  4 * BPS, 8 +  4 * BPS, 12 +  4 * BPS)
1120     GET_SSE( 5 * BPS, 4 +  5 * BPS, 8 +  5 * BPS, 12 +  5 * BPS)
1121     GET_SSE( 6 * BPS, 4 +  6 * BPS, 8 +  6 * BPS, 12 +  6 * BPS)
1122     GET_SSE( 7 * BPS, 4 +  7 * BPS, 8 +  7 * BPS, 12 +  7 * BPS)
1123     "mflo   %[count]                                   \n\t"
1124     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
1125       [temp3]"=&r"(temp3), [count]"=&r"(count)
1126     : [a]"r"(a), [b]"r"(b)
1127     : "memory", "hi", "lo"
1128   );
1129   return count;
1130 }
1131 
SSE8x8(const uint8_t * a,const uint8_t * b)1132 static int SSE8x8(const uint8_t* a, const uint8_t* b) {
1133   int count;
1134   int temp0, temp1, temp2, temp3;
1135   __asm__ volatile (
1136     "mult   $zero,    $zero                            \n\t"
1137     GET_SSE(0 * BPS, 4 + 0 * BPS, 1 * BPS, 4 + 1 * BPS)
1138     GET_SSE(2 * BPS, 4 + 2 * BPS, 3 * BPS, 4 + 3 * BPS)
1139     GET_SSE(4 * BPS, 4 + 4 * BPS, 5 * BPS, 4 + 5 * BPS)
1140     GET_SSE(6 * BPS, 4 + 6 * BPS, 7 * BPS, 4 + 7 * BPS)
1141     "mflo   %[count]                                   \n\t"
1142     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
1143       [temp3]"=&r"(temp3), [count]"=&r"(count)
1144     : [a]"r"(a), [b]"r"(b)
1145     : "memory", "hi", "lo"
1146   );
1147   return count;
1148 }
1149 
SSE4x4(const uint8_t * a,const uint8_t * b)1150 static int SSE4x4(const uint8_t* a, const uint8_t* b) {
1151   int count;
1152   int temp0, temp1, temp2, temp3;
1153   __asm__ volatile (
1154     "mult   $zero,    $zero                            \n\t"
1155     GET_SSE(0 * BPS, 1 * BPS, 2 * BPS, 3 * BPS)
1156     "mflo   %[count]                                   \n\t"
1157     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
1158       [temp3]"=&r"(temp3), [count]"=&r"(count)
1159     : [a]"r"(a), [b]"r"(b)
1160     : "memory", "hi", "lo"
1161   );
1162   return count;
1163 }
1164 
1165 #undef GET_SSE
1166 #undef GET_SSE_INNER
1167 
1168 #endif  // !WORK_AROUND_GCC
1169 
1170 #undef FILL_8_OR_16
1171 #undef FILL_PART
1172 #undef OUTPUT_EARLY_CLOBBER_REGS_17
1173 #undef MUL_HALF
1174 #undef ABS_X8
1175 #undef ADD_SUB_HALVES_X4
1176 
1177 //------------------------------------------------------------------------------
1178 // Quantization
1179 //
1180 
1181 // macro for one pass through for loop in QuantizeBlock reading 2 values at time
1182 // QUANTDIV macro inlined
1183 // J - offset in bytes (kZigzag[n] * 2)
1184 // K - offset in bytes (kZigzag[n] * 4)
1185 // N - offset in bytes (n * 2)
1186 // N1 - offset in bytes ((n + 1) * 2)
1187 #define QUANTIZE_ONE(J, K, N, N1)                                         \
1188   "ulw         %[temp1],     " #J "(%[ppin])                 \n\t"        \
1189   "ulw         %[temp2],     " #J "(%[ppsharpen])            \n\t"        \
1190   "lhu         %[temp3],     " #K "(%[ppzthresh])            \n\t"        \
1191   "lhu         %[temp6],     " #K "+4(%[ppzthresh])          \n\t"        \
1192   "absq_s.ph   %[temp4],     %[temp1]                        \n\t"        \
1193   "ins         %[temp3],     %[temp6],         16,       16  \n\t"        \
1194   "addu.ph     %[coeff],     %[temp4],         %[temp2]      \n\t"        \
1195   "shra.ph     %[sign],      %[temp1],         15            \n\t"        \
1196   "li          %[level],     0x10001                         \n\t"        \
1197   "cmp.lt.ph   %[temp3],     %[coeff]                        \n\t"        \
1198   "lhu         %[temp1],     " #J "(%[ppiq])                 \n\t"        \
1199   "pick.ph     %[temp5],     %[level],         $0            \n\t"        \
1200   "lw          %[temp2],     " #K "(%[ppbias])               \n\t"        \
1201   "beqz        %[temp5],     0f                              \n\t"        \
1202   "lhu         %[temp3],     " #J "(%[ppq])                  \n\t"        \
1203   "beq         %[temp5],     %[level],         1f            \n\t"        \
1204   "andi        %[temp5],     %[temp5],         0x1           \n\t"        \
1205   "andi        %[temp4],     %[coeff],         0xffff        \n\t"        \
1206   "beqz        %[temp5],     2f                              \n\t"        \
1207   "mul         %[level],     %[temp4],         %[temp1]      \n\t"        \
1208   "sh          $0,           " #J "+2(%[ppin])               \n\t"        \
1209   "sh          $0,           " #N1 "(%[pout])                \n\t"        \
1210   "addu        %[level],     %[level],         %[temp2]      \n\t"        \
1211   "sra         %[level],     %[level],         17            \n\t"        \
1212   "slt         %[temp4],     %[max_level],     %[level]      \n\t"        \
1213   "movn        %[level],     %[max_level],     %[temp4]      \n\t"        \
1214   "andi        %[temp6],     %[sign],          0xffff        \n\t"        \
1215   "xor         %[level],     %[level],         %[temp6]      \n\t"        \
1216   "subu        %[level],     %[level],         %[temp6]      \n\t"        \
1217   "mul         %[temp5],     %[level],         %[temp3]      \n\t"        \
1218   "or          %[ret],       %[ret],           %[level]      \n\t"        \
1219   "sh          %[level],     " #N "(%[pout])                 \n\t"        \
1220   "sh          %[temp5],     " #J "(%[ppin])                 \n\t"        \
1221   "j           3f                                            \n\t"        \
1222 "2:                                                          \n\t"        \
1223   "lhu         %[temp1],     " #J "+2(%[ppiq])               \n\t"        \
1224   "srl         %[temp5],     %[coeff],         16            \n\t"        \
1225   "mul         %[level],     %[temp5],         %[temp1]      \n\t"        \
1226   "lw          %[temp2],     " #K "+4(%[ppbias])             \n\t"        \
1227   "lhu         %[temp3],     " #J "+2(%[ppq])                \n\t"        \
1228   "addu        %[level],     %[level],         %[temp2]      \n\t"        \
1229   "sra         %[level],     %[level],         17            \n\t"        \
1230   "srl         %[temp6],     %[sign],          16            \n\t"        \
1231   "slt         %[temp4],     %[max_level],     %[level]      \n\t"        \
1232   "movn        %[level],     %[max_level],     %[temp4]      \n\t"        \
1233   "xor         %[level],     %[level],         %[temp6]      \n\t"        \
1234   "subu        %[level],     %[level],         %[temp6]      \n\t"        \
1235   "mul         %[temp5],     %[level],         %[temp3]      \n\t"        \
1236   "sh          $0,           " #J "(%[ppin])                 \n\t"        \
1237   "sh          $0,           " #N "(%[pout])                 \n\t"        \
1238   "or          %[ret],       %[ret],           %[level]      \n\t"        \
1239   "sh          %[temp5],     " #J "+2(%[ppin])               \n\t"        \
1240   "sh          %[level],     " #N1 "(%[pout])                \n\t"        \
1241   "j           3f                                            \n\t"        \
1242 "1:                                                          \n\t"        \
1243   "lhu         %[temp1],     " #J "(%[ppiq])                 \n\t"        \
1244   "lw          %[temp2],     " #K "(%[ppbias])               \n\t"        \
1245   "ulw         %[temp3],     " #J "(%[ppq])                  \n\t"        \
1246   "andi        %[temp5],     %[coeff],         0xffff        \n\t"        \
1247   "srl         %[temp0],     %[coeff],         16            \n\t"        \
1248   "lhu         %[temp6],     " #J "+2(%[ppiq])               \n\t"        \
1249   "lw          %[coeff],     " #K "+4(%[ppbias])             \n\t"        \
1250   "mul         %[level],     %[temp5],         %[temp1]      \n\t"        \
1251   "mul         %[temp4],     %[temp0],         %[temp6]      \n\t"        \
1252   "addu        %[level],     %[level],         %[temp2]      \n\t"        \
1253   "addu        %[temp4],     %[temp4],         %[coeff]      \n\t"        \
1254   "precrq.ph.w %[level],     %[temp4],         %[level]      \n\t"        \
1255   "shra.ph     %[level],     %[level],         1             \n\t"        \
1256   "cmp.lt.ph   %[max_level1],%[level]                        \n\t"        \
1257   "pick.ph     %[level],     %[max_level],     %[level]      \n\t"        \
1258   "xor         %[level],     %[level],         %[sign]       \n\t"        \
1259   "subu.ph     %[level],     %[level],         %[sign]       \n\t"        \
1260   "mul.ph      %[temp3],     %[level],         %[temp3]      \n\t"        \
1261   "or          %[ret],       %[ret],           %[level]      \n\t"        \
1262   "sh          %[level],     " #N "(%[pout])                 \n\t"        \
1263   "srl         %[level],     %[level],         16            \n\t"        \
1264   "sh          %[level],     " #N1 "(%[pout])                \n\t"        \
1265   "usw         %[temp3],     " #J "(%[ppin])                 \n\t"        \
1266   "j           3f                                            \n\t"        \
1267 "0:                                                          \n\t"        \
1268   "sh          $0,           " #N "(%[pout])                 \n\t"        \
1269   "sh          $0,           " #N1 "(%[pout])                \n\t"        \
1270   "usw         $0,           " #J "(%[ppin])                 \n\t"        \
1271 "3:                                                          \n\t"
1272 
QuantizeBlock(int16_t in[16],int16_t out[16],const VP8Matrix * const mtx)1273 static int QuantizeBlock(int16_t in[16], int16_t out[16],
1274                          const VP8Matrix* const mtx) {
1275   int temp0, temp1, temp2, temp3, temp4, temp5,temp6;
1276   int sign, coeff, level;
1277   int max_level = MAX_LEVEL;
1278   int max_level1 = max_level << 16 | max_level;
1279   int ret = 0;
1280 
1281   int16_t* ppin             = &in[0];
1282   int16_t* pout             = &out[0];
1283   const uint16_t* ppsharpen = &mtx->sharpen_[0];
1284   const uint32_t* ppzthresh = &mtx->zthresh_[0];
1285   const uint16_t* ppq       = &mtx->q_[0];
1286   const uint16_t* ppiq      = &mtx->iq_[0];
1287   const uint32_t* ppbias    = &mtx->bias_[0];
1288 
1289   __asm__ volatile (
1290     QUANTIZE_ONE( 0,  0,  0,  2)
1291     QUANTIZE_ONE( 4,  8, 10, 12)
1292     QUANTIZE_ONE( 8, 16,  4,  8)
1293     QUANTIZE_ONE(12, 24, 14, 24)
1294     QUANTIZE_ONE(16, 32,  6, 16)
1295     QUANTIZE_ONE(20, 40, 22, 26)
1296     QUANTIZE_ONE(24, 48, 18, 20)
1297     QUANTIZE_ONE(28, 56, 28, 30)
1298 
1299     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1),
1300       [temp2]"=&r"(temp2), [temp3]"=&r"(temp3),
1301       [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
1302       [sign]"=&r"(sign), [coeff]"=&r"(coeff),
1303       [level]"=&r"(level), [temp6]"=&r"(temp6), [ret]"+&r"(ret)
1304     : [ppin]"r"(ppin), [pout]"r"(pout), [max_level1]"r"(max_level1),
1305       [ppiq]"r"(ppiq), [max_level]"r"(max_level),
1306       [ppbias]"r"(ppbias), [ppzthresh]"r"(ppzthresh),
1307       [ppsharpen]"r"(ppsharpen), [ppq]"r"(ppq)
1308     : "memory", "hi", "lo"
1309   );
1310 
1311   return (ret != 0);
1312 }
1313 
Quantize2Blocks(int16_t in[32],int16_t out[32],const VP8Matrix * const mtx)1314 static int Quantize2Blocks(int16_t in[32], int16_t out[32],
1315                            const VP8Matrix* const mtx) {
1316   int nz;
1317   nz  = QuantizeBlock(in + 0 * 16, out + 0 * 16, mtx) << 0;
1318   nz |= QuantizeBlock(in + 1 * 16, out + 1 * 16, mtx) << 1;
1319   return nz;
1320 }
1321 
1322 #undef QUANTIZE_ONE
1323 
1324 // macro for one horizontal pass in FTransformWHT
1325 // temp0..temp7 holds tmp[0]..tmp[15]
1326 // A, B, C, D - offset in bytes to load from in buffer
1327 // TEMP0, TEMP1 - registers for corresponding tmp elements
1328 #define HORIZONTAL_PASS_WHT(A, B, C, D, TEMP0, TEMP1)                          \
1329   "lh              %[" #TEMP0 "],  " #A "(%[in])            \n\t"              \
1330   "lh              %[" #TEMP1 "],  " #B "(%[in])            \n\t"              \
1331   "lh              %[temp8],     " #C "(%[in])              \n\t"              \
1332   "lh              %[temp9],     " #D "(%[in])              \n\t"              \
1333   "ins             %[" #TEMP1 "],  %[" #TEMP0 "],  16,  16  \n\t"              \
1334   "ins             %[temp9],     %[temp8],     16,  16      \n\t"              \
1335   "subq.ph         %[temp8],     %[" #TEMP1 "],  %[temp9]   \n\t"              \
1336   "addq.ph         %[temp9],     %[" #TEMP1 "],  %[temp9]   \n\t"              \
1337   "precrq.ph.w     %[" #TEMP0 "],  %[temp8],     %[temp9]   \n\t"              \
1338   "append          %[temp8],     %[temp9],     16           \n\t"              \
1339   "subq.ph         %[" #TEMP1 "],  %[" #TEMP0 "],  %[temp8] \n\t"              \
1340   "addq.ph         %[" #TEMP0 "],  %[" #TEMP0 "],  %[temp8] \n\t"              \
1341   "rotr            %[" #TEMP1 "],  %[" #TEMP1 "],  16       \n\t"
1342 
1343 // macro for one vertical pass in FTransformWHT
1344 // temp0..temp7 holds tmp[0]..tmp[15]
1345 // A, B, C, D - offsets in bytes to store to out buffer
1346 // TEMP0, TEMP2, TEMP4 and TEMP6 - registers for corresponding tmp elements
1347 #define VERTICAL_PASS_WHT(A, B, C, D, TEMP0, TEMP2, TEMP4, TEMP6)              \
1348   "addq.ph         %[temp8],     %[" #TEMP0 "],  %[" #TEMP4 "]    \n\t"        \
1349   "addq.ph         %[temp9],     %[" #TEMP2 "],  %[" #TEMP6 "]    \n\t"        \
1350   "subq.ph         %[" #TEMP2 "],  %[" #TEMP2 "],  %[" #TEMP6 "]  \n\t"        \
1351   "subq.ph         %[" #TEMP6 "],  %[" #TEMP0 "],  %[" #TEMP4 "]  \n\t"        \
1352   "addqh.ph        %[" #TEMP0 "],  %[temp8],     %[temp9]         \n\t"        \
1353   "subqh.ph        %[" #TEMP4 "],  %[" #TEMP6 "],  %[" #TEMP2 "]  \n\t"        \
1354   "addqh.ph        %[" #TEMP2 "],  %[" #TEMP2 "],  %[" #TEMP6 "]  \n\t"        \
1355   "subqh.ph        %[" #TEMP6 "],  %[temp8],     %[temp9]         \n\t"        \
1356   "usw             %[" #TEMP0 "],  " #A "(%[out])                 \n\t"        \
1357   "usw             %[" #TEMP2 "],  " #B "(%[out])                 \n\t"        \
1358   "usw             %[" #TEMP4 "],  " #C "(%[out])                 \n\t"        \
1359   "usw             %[" #TEMP6 "],  " #D "(%[out])                 \n\t"
1360 
FTransformWHT(const int16_t * in,int16_t * out)1361 static void FTransformWHT(const int16_t* in, int16_t* out) {
1362   int temp0, temp1, temp2, temp3, temp4;
1363   int temp5, temp6, temp7, temp8, temp9;
1364 
1365   __asm__ volatile (
1366     HORIZONTAL_PASS_WHT(  0,  32,  64,  96, temp0, temp1)
1367     HORIZONTAL_PASS_WHT(128, 160, 192, 224, temp2, temp3)
1368     HORIZONTAL_PASS_WHT(256, 288, 320, 352, temp4, temp5)
1369     HORIZONTAL_PASS_WHT(384, 416, 448, 480, temp6, temp7)
1370     VERTICAL_PASS_WHT(0,  8, 16, 24, temp0, temp2, temp4, temp6)
1371     VERTICAL_PASS_WHT(4, 12, 20, 28, temp1, temp3, temp5, temp7)
1372     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
1373       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
1374       [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [temp8]"=&r"(temp8),
1375       [temp9]"=&r"(temp9)
1376     : [in]"r"(in), [out]"r"(out)
1377     : "memory"
1378   );
1379 }
1380 
1381 #undef VERTICAL_PASS_WHT
1382 #undef HORIZONTAL_PASS_WHT
1383 
1384 // macro for converting coefficients to bin
1385 // convert 8 coeffs at time
1386 // A, B, C, D - offsets in bytes to load from out buffer
1387 #define CONVERT_COEFFS_TO_BIN(A, B, C, D)                                      \
1388   "ulw        %[temp0],  " #A "(%[out])                \n\t"                   \
1389   "ulw        %[temp1],  " #B "(%[out])                \n\t"                   \
1390   "ulw        %[temp2],  " #C "(%[out])                \n\t"                   \
1391   "ulw        %[temp3],  " #D "(%[out])                \n\t"                   \
1392   "absq_s.ph  %[temp0],  %[temp0]                      \n\t"                   \
1393   "absq_s.ph  %[temp1],  %[temp1]                      \n\t"                   \
1394   "absq_s.ph  %[temp2],  %[temp2]                      \n\t"                   \
1395   "absq_s.ph  %[temp3],  %[temp3]                      \n\t"                   \
1396   "shra.ph    %[temp0],  %[temp0],    3                \n\t"                   \
1397   "shra.ph    %[temp1],  %[temp1],    3                \n\t"                   \
1398   "shra.ph    %[temp2],  %[temp2],    3                \n\t"                   \
1399   "shra.ph    %[temp3],  %[temp3],    3                \n\t"                   \
1400   "shll_s.ph  %[temp0],  %[temp0],    10               \n\t"                   \
1401   "shll_s.ph  %[temp1],  %[temp1],    10               \n\t"                   \
1402   "shll_s.ph  %[temp2],  %[temp2],    10               \n\t"                   \
1403   "shll_s.ph  %[temp3],  %[temp3],    10               \n\t"                   \
1404   "shrl.ph    %[temp0],  %[temp0],    10               \n\t"                   \
1405   "shrl.ph    %[temp1],  %[temp1],    10               \n\t"                   \
1406   "shrl.ph    %[temp2],  %[temp2],    10               \n\t"                   \
1407   "shrl.ph    %[temp3],  %[temp3],    10               \n\t"                   \
1408   "shll.ph    %[temp0],  %[temp0],    2                \n\t"                   \
1409   "shll.ph    %[temp1],  %[temp1],    2                \n\t"                   \
1410   "shll.ph    %[temp2],  %[temp2],    2                \n\t"                   \
1411   "shll.ph    %[temp3],  %[temp3],    2                \n\t"                   \
1412   "ext        %[temp4],  %[temp0],    0,       16      \n\t"                   \
1413   "ext        %[temp0],  %[temp0],    16,      16      \n\t"                   \
1414   "addu       %[temp4],  %[temp4],    %[dist]          \n\t"                   \
1415   "addu       %[temp0],  %[temp0],    %[dist]          \n\t"                   \
1416   "ext        %[temp5],  %[temp1],    0,       16      \n\t"                   \
1417   "lw         %[temp8],  0(%[temp4])                   \n\t"                   \
1418   "ext        %[temp1],  %[temp1],    16,      16      \n\t"                   \
1419   "addu       %[temp5],  %[temp5],    %[dist]          \n\t"                   \
1420   "addiu      %[temp8],  %[temp8],    1                \n\t"                   \
1421   "sw         %[temp8],  0(%[temp4])                   \n\t"                   \
1422   "lw         %[temp8],  0(%[temp0])                   \n\t"                   \
1423   "addu       %[temp1],  %[temp1],    %[dist]          \n\t"                   \
1424   "ext        %[temp6],  %[temp2],    0,       16      \n\t"                   \
1425   "addiu      %[temp8],  %[temp8],    1                \n\t"                   \
1426   "sw         %[temp8],  0(%[temp0])                   \n\t"                   \
1427   "lw         %[temp8],  0(%[temp5])                   \n\t"                   \
1428   "ext        %[temp2],  %[temp2],    16,      16      \n\t"                   \
1429   "addu       %[temp6],  %[temp6],    %[dist]          \n\t"                   \
1430   "addiu      %[temp8],  %[temp8],    1                \n\t"                   \
1431   "sw         %[temp8],  0(%[temp5])                   \n\t"                   \
1432   "lw         %[temp8],  0(%[temp1])                   \n\t"                   \
1433   "addu       %[temp2],  %[temp2],    %[dist]          \n\t"                   \
1434   "ext        %[temp7],  %[temp3],    0,       16      \n\t"                   \
1435   "addiu      %[temp8],  %[temp8],    1                \n\t"                   \
1436   "sw         %[temp8],  0(%[temp1])                   \n\t"                   \
1437   "lw         %[temp8],  0(%[temp6])                   \n\t"                   \
1438   "ext        %[temp3],  %[temp3],    16,      16      \n\t"                   \
1439   "addu       %[temp7],  %[temp7],    %[dist]          \n\t"                   \
1440   "addiu      %[temp8],  %[temp8],    1                \n\t"                   \
1441   "sw         %[temp8],  0(%[temp6])                   \n\t"                   \
1442   "lw         %[temp8],  0(%[temp2])                   \n\t"                   \
1443   "addu       %[temp3],  %[temp3],    %[dist]          \n\t"                   \
1444   "addiu      %[temp8],  %[temp8],    1                \n\t"                   \
1445   "sw         %[temp8],  0(%[temp2])                   \n\t"                   \
1446   "lw         %[temp8],  0(%[temp7])                   \n\t"                   \
1447   "addiu      %[temp8],  %[temp8],    1                \n\t"                   \
1448   "sw         %[temp8],  0(%[temp7])                   \n\t"                   \
1449   "lw         %[temp8],  0(%[temp3])                   \n\t"                   \
1450   "addiu      %[temp8],  %[temp8],    1                \n\t"                   \
1451   "sw         %[temp8],  0(%[temp3])                   \n\t"
1452 
CollectHistogram(const uint8_t * ref,const uint8_t * pred,int start_block,int end_block,VP8Histogram * const histo)1453 static void CollectHistogram(const uint8_t* ref, const uint8_t* pred,
1454                              int start_block, int end_block,
1455                              VP8Histogram* const histo) {
1456   int j;
1457   int distribution[MAX_COEFF_THRESH + 1] = { 0 };
1458   const int max_coeff = (MAX_COEFF_THRESH << 16) + MAX_COEFF_THRESH;
1459   for (j = start_block; j < end_block; ++j) {
1460     int16_t out[16];
1461     int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8;
1462 
1463     VP8FTransform(ref + VP8DspScan[j], pred + VP8DspScan[j], out);
1464 
1465     // Convert coefficients to bin.
1466     __asm__ volatile (
1467       CONVERT_COEFFS_TO_BIN( 0,  4,  8, 12)
1468       CONVERT_COEFFS_TO_BIN(16, 20, 24, 28)
1469       : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
1470         [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
1471         [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [temp8]"=&r"(temp8)
1472       : [dist]"r"(distribution), [out]"r"(out), [max_coeff]"r"(max_coeff)
1473       : "memory"
1474     );
1475   }
1476   VP8SetHistogramData(distribution, histo);
1477 }
1478 
1479 #undef CONVERT_COEFFS_TO_BIN
1480 
1481 //------------------------------------------------------------------------------
1482 // Entry point
1483 
1484 extern void VP8EncDspInitMIPSdspR2(void);
1485 
VP8EncDspInitMIPSdspR2(void)1486 WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspInitMIPSdspR2(void) {
1487   VP8FTransform = FTransform;
1488   VP8ITransform = ITransform;
1489   VP8TDisto4x4 = Disto4x4;
1490   VP8TDisto16x16 = Disto16x16;
1491   VP8EncPredLuma16 = Intra16Preds;
1492   VP8EncPredChroma8 = IntraChromaPreds;
1493   VP8EncPredLuma4 = Intra4Preds;
1494 #if !defined(WORK_AROUND_GCC)
1495   VP8SSE16x16 = SSE16x16;
1496   VP8SSE8x8 = SSE8x8;
1497   VP8SSE16x8 = SSE16x8;
1498   VP8SSE4x4 = SSE4x4;
1499 #endif
1500   VP8EncQuantizeBlock = QuantizeBlock;
1501   VP8EncQuantize2Blocks = Quantize2Blocks;
1502   VP8FTransformWHT = FTransformWHT;
1503   VP8CollectHistogram = CollectHistogram;
1504 }
1505 
1506 #else  // !WEBP_USE_MIPS_DSP_R2
1507 
1508 WEBP_DSP_INIT_STUB(VP8EncDspInitMIPSdspR2)
1509 
1510 #endif  // WEBP_USE_MIPS_DSP_R2
1511