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): Djordje Pesut    (djordje.pesut@imgtec.com)
13 //            Jovan Zelincevic (jovan.zelincevic@imgtec.com)
14 //            Slobodan Prijic  (slobodan.prijic@imgtec.com)
15 
16 #include "./dsp.h"
17 
18 #if defined(WEBP_USE_MIPS32)
19 
20 #include "./mips_macro.h"
21 #include "../enc/vp8enci.h"
22 #include "../enc/cost.h"
23 
24 static const int kC1 = 20091 + (1 << 16);
25 static const int kC2 = 35468;
26 
27 // macro for one vertical pass in ITransformOne
28 // MUL macro inlined
29 // temp0..temp15 holds tmp[0]..tmp[15]
30 // A..D - offsets in bytes to load from in buffer
31 // TEMP0..TEMP3 - registers for corresponding tmp elements
32 // TEMP4..TEMP5 - temporary registers
33 #define VERTICAL_PASS(A, B, C, D, TEMP4, TEMP0, TEMP1, TEMP2, TEMP3)        \
34   "lh      %[temp16],      " #A "(%[temp20])                 \n\t"          \
35   "lh      %[temp18],      " #B "(%[temp20])                 \n\t"          \
36   "lh      %[temp17],      " #C "(%[temp20])                 \n\t"          \
37   "lh      %[temp19],      " #D "(%[temp20])                 \n\t"          \
38   "addu    %[" #TEMP4 "],    %[temp16],      %[temp18]       \n\t"          \
39   "subu    %[temp16],      %[temp16],      %[temp18]         \n\t"          \
40   "mul     %[" #TEMP0 "],    %[temp17],      %[kC2]          \n\t"          \
41   "mul     %[temp18],      %[temp19],      %[kC1]            \n\t"          \
42   "mul     %[temp17],      %[temp17],      %[kC1]            \n\t"          \
43   "mul     %[temp19],      %[temp19],      %[kC2]            \n\t"          \
44   "sra     %[" #TEMP0 "],    %[" #TEMP0 "],    16            \n\n"          \
45   "sra     %[temp18],      %[temp18],      16                \n\n"          \
46   "sra     %[temp17],      %[temp17],      16                \n\n"          \
47   "sra     %[temp19],      %[temp19],      16                \n\n"          \
48   "subu    %[" #TEMP2 "],    %[" #TEMP0 "],    %[temp18]     \n\t"          \
49   "addu    %[" #TEMP3 "],    %[temp17],      %[temp19]       \n\t"          \
50   "addu    %[" #TEMP0 "],    %[" #TEMP4 "],    %[" #TEMP3 "] \n\t"          \
51   "addu    %[" #TEMP1 "],    %[temp16],      %[" #TEMP2 "]   \n\t"          \
52   "subu    %[" #TEMP2 "],    %[temp16],      %[" #TEMP2 "]   \n\t"          \
53   "subu    %[" #TEMP3 "],    %[" #TEMP4 "],    %[" #TEMP3 "] \n\t"
54 
55 // macro for one horizontal pass in ITransformOne
56 // MUL and STORE macros inlined
57 // a = clip_8b(a) is replaced with: a = max(a, 0); a = min(a, 255)
58 // temp0..temp15 holds tmp[0]..tmp[15]
59 // A - offset in bytes to load from ref and store to dst buffer
60 // TEMP0, TEMP4, TEMP8 and TEMP12 - registers for corresponding tmp elements
61 #define HORIZONTAL_PASS(A, TEMP0, TEMP4, TEMP8, TEMP12)                       \
62   "addiu   %[" #TEMP0 "],    %[" #TEMP0 "],    4               \n\t"          \
63   "addu    %[temp16],      %[" #TEMP0 "],    %[" #TEMP8 "]     \n\t"          \
64   "subu    %[temp17],      %[" #TEMP0 "],    %[" #TEMP8 "]     \n\t"          \
65   "mul     %[" #TEMP0 "],    %[" #TEMP4 "],    %[kC2]          \n\t"          \
66   "mul     %[" #TEMP8 "],    %[" #TEMP12 "],   %[kC1]          \n\t"          \
67   "mul     %[" #TEMP4 "],    %[" #TEMP4 "],    %[kC1]          \n\t"          \
68   "mul     %[" #TEMP12 "],   %[" #TEMP12 "],   %[kC2]          \n\t"          \
69   "sra     %[" #TEMP0 "],    %[" #TEMP0 "],    16              \n\t"          \
70   "sra     %[" #TEMP8 "],    %[" #TEMP8 "],    16              \n\t"          \
71   "sra     %[" #TEMP4 "],    %[" #TEMP4 "],    16              \n\t"          \
72   "sra     %[" #TEMP12 "],   %[" #TEMP12 "],   16              \n\t"          \
73   "subu    %[temp18],      %[" #TEMP0 "],    %[" #TEMP8 "]     \n\t"          \
74   "addu    %[temp19],      %[" #TEMP4 "],    %[" #TEMP12 "]    \n\t"          \
75   "addu    %[" #TEMP0 "],    %[temp16],      %[temp19]         \n\t"          \
76   "addu    %[" #TEMP4 "],    %[temp17],      %[temp18]         \n\t"          \
77   "subu    %[" #TEMP8 "],    %[temp17],      %[temp18]         \n\t"          \
78   "subu    %[" #TEMP12 "],   %[temp16],      %[temp19]         \n\t"          \
79   "lw      %[temp20],      0(%[args])                          \n\t"          \
80   "sra     %[" #TEMP0 "],    %[" #TEMP0 "],    3               \n\t"          \
81   "sra     %[" #TEMP4 "],    %[" #TEMP4 "],    3               \n\t"          \
82   "sra     %[" #TEMP8 "],    %[" #TEMP8 "],    3               \n\t"          \
83   "sra     %[" #TEMP12 "],   %[" #TEMP12 "],   3               \n\t"          \
84   "lbu     %[temp16],      0+" XSTR(BPS) "*" #A "(%[temp20])   \n\t"          \
85   "lbu     %[temp17],      1+" XSTR(BPS) "*" #A "(%[temp20])   \n\t"          \
86   "lbu     %[temp18],      2+" XSTR(BPS) "*" #A "(%[temp20])   \n\t"          \
87   "lbu     %[temp19],      3+" XSTR(BPS) "*" #A "(%[temp20])   \n\t"          \
88   "addu    %[" #TEMP0 "],    %[temp16],      %[" #TEMP0 "]     \n\t"          \
89   "addu    %[" #TEMP4 "],    %[temp17],      %[" #TEMP4 "]     \n\t"          \
90   "addu    %[" #TEMP8 "],    %[temp18],      %[" #TEMP8 "]     \n\t"          \
91   "addu    %[" #TEMP12 "],   %[temp19],      %[" #TEMP12 "]    \n\t"          \
92   "slt     %[temp16],      %[" #TEMP0 "],    $zero             \n\t"          \
93   "slt     %[temp17],      %[" #TEMP4 "],    $zero             \n\t"          \
94   "slt     %[temp18],      %[" #TEMP8 "],    $zero             \n\t"          \
95   "slt     %[temp19],      %[" #TEMP12 "],   $zero             \n\t"          \
96   "movn    %[" #TEMP0 "],    $zero,          %[temp16]         \n\t"          \
97   "movn    %[" #TEMP4 "],    $zero,          %[temp17]         \n\t"          \
98   "movn    %[" #TEMP8 "],    $zero,          %[temp18]         \n\t"          \
99   "movn    %[" #TEMP12 "],   $zero,          %[temp19]         \n\t"          \
100   "addiu   %[temp20],      $zero,          255                 \n\t"          \
101   "slt     %[temp16],      %[" #TEMP0 "],    %[temp20]         \n\t"          \
102   "slt     %[temp17],      %[" #TEMP4 "],    %[temp20]         \n\t"          \
103   "slt     %[temp18],      %[" #TEMP8 "],    %[temp20]         \n\t"          \
104   "slt     %[temp19],      %[" #TEMP12 "],   %[temp20]         \n\t"          \
105   "movz    %[" #TEMP0 "],    %[temp20],      %[temp16]         \n\t"          \
106   "movz    %[" #TEMP4 "],    %[temp20],      %[temp17]         \n\t"          \
107   "lw      %[temp16],      8(%[args])                          \n\t"          \
108   "movz    %[" #TEMP8 "],    %[temp20],      %[temp18]         \n\t"          \
109   "movz    %[" #TEMP12 "],   %[temp20],      %[temp19]         \n\t"          \
110   "sb      %[" #TEMP0 "],    0+" XSTR(BPS) "*" #A "(%[temp16]) \n\t"          \
111   "sb      %[" #TEMP4 "],    1+" XSTR(BPS) "*" #A "(%[temp16]) \n\t"          \
112   "sb      %[" #TEMP8 "],    2+" XSTR(BPS) "*" #A "(%[temp16]) \n\t"          \
113   "sb      %[" #TEMP12 "],   3+" XSTR(BPS) "*" #A "(%[temp16]) \n\t"
114 
115 // Does one or two inverse transforms.
ITransformOne(const uint8_t * ref,const int16_t * in,uint8_t * dst)116 static WEBP_INLINE void ITransformOne(const uint8_t* ref, const int16_t* in,
117                                       uint8_t* dst) {
118   int temp0, temp1, temp2, temp3, temp4, temp5, temp6;
119   int temp7, temp8, temp9, temp10, temp11, temp12, temp13;
120   int temp14, temp15, temp16, temp17, temp18, temp19, temp20;
121   const int* args[3] = {(const int*)ref, (const int*)in, (const int*)dst};
122 
123   __asm__ volatile(
124     "lw      %[temp20],      4(%[args])                      \n\t"
125     VERTICAL_PASS(0, 16,  8, 24, temp4,  temp0,  temp1,  temp2,  temp3)
126     VERTICAL_PASS(2, 18, 10, 26, temp8,  temp4,  temp5,  temp6,  temp7)
127     VERTICAL_PASS(4, 20, 12, 28, temp12, temp8,  temp9,  temp10, temp11)
128     VERTICAL_PASS(6, 22, 14, 30, temp20, temp12, temp13, temp14, temp15)
129 
130     HORIZONTAL_PASS(0, temp0, temp4, temp8,  temp12)
131     HORIZONTAL_PASS(1, temp1, temp5, temp9,  temp13)
132     HORIZONTAL_PASS(2, temp2, temp6, temp10, temp14)
133     HORIZONTAL_PASS(3, temp3, temp7, temp11, temp15)
134 
135     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
136       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
137       [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [temp8]"=&r"(temp8),
138       [temp9]"=&r"(temp9), [temp10]"=&r"(temp10), [temp11]"=&r"(temp11),
139       [temp12]"=&r"(temp12), [temp13]"=&r"(temp13), [temp14]"=&r"(temp14),
140       [temp15]"=&r"(temp15), [temp16]"=&r"(temp16), [temp17]"=&r"(temp17),
141       [temp18]"=&r"(temp18), [temp19]"=&r"(temp19), [temp20]"=&r"(temp20)
142     : [args]"r"(args), [kC1]"r"(kC1), [kC2]"r"(kC2)
143     : "memory", "hi", "lo"
144   );
145 }
146 
ITransform(const uint8_t * ref,const int16_t * in,uint8_t * dst,int do_two)147 static void ITransform(const uint8_t* ref, const int16_t* in,
148                        uint8_t* dst, int do_two) {
149   ITransformOne(ref, in, dst);
150   if (do_two) {
151     ITransformOne(ref + 4, in + 16, dst + 4);
152   }
153 }
154 
155 #undef VERTICAL_PASS
156 #undef HORIZONTAL_PASS
157 
158 // macro for one pass through for loop in QuantizeBlock
159 // QUANTDIV macro inlined
160 // J - offset in bytes (kZigzag[n] * 2)
161 // K - offset in bytes (kZigzag[n] * 4)
162 // N - offset in bytes (n * 2)
163 #define QUANTIZE_ONE(J, K, N)                                               \
164   "lh           %[temp0],       " #J "(%[ppin])                     \n\t"   \
165   "lhu          %[temp1],       " #J "(%[ppsharpen])                \n\t"   \
166   "lw           %[temp2],       " #K "(%[ppzthresh])                \n\t"   \
167   "sra          %[sign],        %[temp0],           15              \n\t"   \
168   "xor          %[coeff],       %[temp0],           %[sign]         \n\t"   \
169   "subu         %[coeff],       %[coeff],           %[sign]         \n\t"   \
170   "addu         %[coeff],       %[coeff],           %[temp1]        \n\t"   \
171   "slt          %[temp4],       %[temp2],           %[coeff]        \n\t"   \
172   "addiu        %[temp5],       $zero,              0               \n\t"   \
173   "addiu        %[level],       $zero,              0               \n\t"   \
174   "beqz         %[temp4],       2f                                  \n\t"   \
175   "lhu          %[temp1],       " #J "(%[ppiq])                     \n\t"   \
176   "lw           %[temp2],       " #K "(%[ppbias])                   \n\t"   \
177   "lhu          %[temp3],       " #J "(%[ppq])                      \n\t"   \
178   "mul          %[level],       %[coeff],           %[temp1]        \n\t"   \
179   "addu         %[level],       %[level],           %[temp2]        \n\t"   \
180   "sra          %[level],       %[level],           17              \n\t"   \
181   "slt          %[temp4],       %[max_level],       %[level]        \n\t"   \
182   "movn         %[level],       %[max_level],       %[temp4]        \n\t"   \
183   "xor          %[level],       %[level],           %[sign]         \n\t"   \
184   "subu         %[level],       %[level],           %[sign]         \n\t"   \
185   "mul          %[temp5],       %[level],           %[temp3]        \n\t"   \
186 "2:                                                                 \n\t"   \
187   "sh           %[temp5],       " #J "(%[ppin])                     \n\t"   \
188   "sh           %[level],       " #N "(%[pout])                     \n\t"
189 
QuantizeBlock(int16_t in[16],int16_t out[16],const VP8Matrix * const mtx)190 static int QuantizeBlock(int16_t in[16], int16_t out[16],
191                          const VP8Matrix* const mtx) {
192   int temp0, temp1, temp2, temp3, temp4, temp5;
193   int sign, coeff, level, i;
194   int max_level = MAX_LEVEL;
195 
196   int16_t* ppin             = &in[0];
197   int16_t* pout             = &out[0];
198   const uint16_t* ppsharpen = &mtx->sharpen_[0];
199   const uint32_t* ppzthresh = &mtx->zthresh_[0];
200   const uint16_t* ppq       = &mtx->q_[0];
201   const uint16_t* ppiq      = &mtx->iq_[0];
202   const uint32_t* ppbias    = &mtx->bias_[0];
203 
204   __asm__ volatile(
205     QUANTIZE_ONE( 0,  0,  0)
206     QUANTIZE_ONE( 2,  4,  2)
207     QUANTIZE_ONE( 8, 16,  4)
208     QUANTIZE_ONE(16, 32,  6)
209     QUANTIZE_ONE(10, 20,  8)
210     QUANTIZE_ONE( 4,  8, 10)
211     QUANTIZE_ONE( 6, 12, 12)
212     QUANTIZE_ONE(12, 24, 14)
213     QUANTIZE_ONE(18, 36, 16)
214     QUANTIZE_ONE(24, 48, 18)
215     QUANTIZE_ONE(26, 52, 20)
216     QUANTIZE_ONE(20, 40, 22)
217     QUANTIZE_ONE(14, 28, 24)
218     QUANTIZE_ONE(22, 44, 26)
219     QUANTIZE_ONE(28, 56, 28)
220     QUANTIZE_ONE(30, 60, 30)
221 
222     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1),
223       [temp2]"=&r"(temp2), [temp3]"=&r"(temp3),
224       [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
225       [sign]"=&r"(sign), [coeff]"=&r"(coeff),
226       [level]"=&r"(level)
227     : [pout]"r"(pout), [ppin]"r"(ppin),
228       [ppiq]"r"(ppiq), [max_level]"r"(max_level),
229       [ppbias]"r"(ppbias), [ppzthresh]"r"(ppzthresh),
230       [ppsharpen]"r"(ppsharpen), [ppq]"r"(ppq)
231     : "memory", "hi", "lo"
232   );
233 
234   // moved out from macro to increase possibility for earlier breaking
235   for (i = 15; i >= 0; i--) {
236     if (out[i]) return 1;
237   }
238   return 0;
239 }
240 
Quantize2Blocks(int16_t in[32],int16_t out[32],const VP8Matrix * const mtx)241 static int Quantize2Blocks(int16_t in[32], int16_t out[32],
242                            const VP8Matrix* const mtx) {
243   int nz;
244   nz  = QuantizeBlock(in + 0 * 16, out + 0 * 16, mtx) << 0;
245   nz |= QuantizeBlock(in + 1 * 16, out + 1 * 16, mtx) << 1;
246   return nz;
247 }
248 
249 #undef QUANTIZE_ONE
250 
251 // macro for one horizontal pass in Disto4x4 (TTransform)
252 // two calls of function TTransform are merged into single one
253 // A - offset in bytes to load from a and b buffers
254 // E..H - offsets in bytes to store first results to tmp buffer
255 // E1..H1 - offsets in bytes to store second results to tmp buffer
256 #define HORIZONTAL_PASS(A, E, F, G, H, E1, F1, G1, H1)                  \
257   "lbu    %[temp0],  0+" XSTR(BPS) "*" #A "(%[a])  \n\t"                \
258   "lbu    %[temp1],  1+" XSTR(BPS) "*" #A "(%[a])  \n\t"                \
259   "lbu    %[temp2],  2+" XSTR(BPS) "*" #A "(%[a])  \n\t"                \
260   "lbu    %[temp3],  3+" XSTR(BPS) "*" #A "(%[a])  \n\t"                \
261   "lbu    %[temp4],  0+" XSTR(BPS) "*" #A "(%[b])  \n\t"                \
262   "lbu    %[temp5],  1+" XSTR(BPS) "*" #A "(%[b])  \n\t"                \
263   "lbu    %[temp6],  2+" XSTR(BPS) "*" #A "(%[b])  \n\t"                \
264   "lbu    %[temp7],  3+" XSTR(BPS) "*" #A "(%[b])  \n\t"                \
265   "addu   %[temp8],  %[temp0],    %[temp2]         \n\t"                \
266   "subu   %[temp0],  %[temp0],    %[temp2]         \n\t"                \
267   "addu   %[temp2],  %[temp1],    %[temp3]         \n\t"                \
268   "subu   %[temp1],  %[temp1],    %[temp3]         \n\t"                \
269   "addu   %[temp3],  %[temp4],    %[temp6]         \n\t"                \
270   "subu   %[temp4],  %[temp4],    %[temp6]         \n\t"                \
271   "addu   %[temp6],  %[temp5],    %[temp7]         \n\t"                \
272   "subu   %[temp5],  %[temp5],    %[temp7]         \n\t"                \
273   "addu   %[temp7],  %[temp8],    %[temp2]         \n\t"                \
274   "subu   %[temp2],  %[temp8],    %[temp2]         \n\t"                \
275   "addu   %[temp8],  %[temp0],    %[temp1]         \n\t"                \
276   "subu   %[temp0],  %[temp0],    %[temp1]         \n\t"                \
277   "addu   %[temp1],  %[temp3],    %[temp6]         \n\t"                \
278   "subu   %[temp3],  %[temp3],    %[temp6]         \n\t"                \
279   "addu   %[temp6],  %[temp4],    %[temp5]         \n\t"                \
280   "subu   %[temp4],  %[temp4],    %[temp5]         \n\t"                \
281   "sw     %[temp7],  " #E "(%[tmp])                \n\t"                \
282   "sw     %[temp2],  " #H "(%[tmp])                \n\t"                \
283   "sw     %[temp8],  " #F "(%[tmp])                \n\t"                \
284   "sw     %[temp0],  " #G "(%[tmp])                \n\t"                \
285   "sw     %[temp1],  " #E1 "(%[tmp])               \n\t"                \
286   "sw     %[temp3],  " #H1 "(%[tmp])               \n\t"                \
287   "sw     %[temp6],  " #F1 "(%[tmp])               \n\t"                \
288   "sw     %[temp4],  " #G1 "(%[tmp])               \n\t"
289 
290 // macro for one vertical pass in Disto4x4 (TTransform)
291 // two calls of function TTransform are merged into single one
292 // since only one accu is available in mips32r1 instruction set
293 //   first is done second call of function TTransform and after
294 //   that first one.
295 //   const int sum1 = TTransform(a, w);
296 //   const int sum2 = TTransform(b, w);
297 //   return abs(sum2 - sum1) >> 5;
298 //   (sum2 - sum1) is calculated with madds (sub2) and msubs (sub1)
299 // A..D - offsets in bytes to load first results from tmp buffer
300 // A1..D1 - offsets in bytes to load second results from tmp buffer
301 // E..H - offsets in bytes to load from w buffer
302 #define VERTICAL_PASS(A, B, C, D, A1, B1, C1, D1, E, F, G, H)     \
303   "lw     %[temp0],  " #A1 "(%[tmp])         \n\t"                \
304   "lw     %[temp1],  " #C1 "(%[tmp])         \n\t"                \
305   "lw     %[temp2],  " #B1 "(%[tmp])         \n\t"                \
306   "lw     %[temp3],  " #D1 "(%[tmp])         \n\t"                \
307   "addu   %[temp8],  %[temp0],    %[temp1]   \n\t"                \
308   "subu   %[temp0],  %[temp0],    %[temp1]   \n\t"                \
309   "addu   %[temp1],  %[temp2],    %[temp3]   \n\t"                \
310   "subu   %[temp2],  %[temp2],    %[temp3]   \n\t"                \
311   "addu   %[temp3],  %[temp8],    %[temp1]   \n\t"                \
312   "subu   %[temp8],  %[temp8],    %[temp1]   \n\t"                \
313   "addu   %[temp1],  %[temp0],    %[temp2]   \n\t"                \
314   "subu   %[temp0],  %[temp0],    %[temp2]   \n\t"                \
315   "sra    %[temp4],  %[temp3],    31         \n\t"                \
316   "sra    %[temp5],  %[temp1],    31         \n\t"                \
317   "sra    %[temp6],  %[temp0],    31         \n\t"                \
318   "sra    %[temp7],  %[temp8],    31         \n\t"                \
319   "xor    %[temp3],  %[temp3],    %[temp4]   \n\t"                \
320   "xor    %[temp1],  %[temp1],    %[temp5]   \n\t"                \
321   "xor    %[temp0],  %[temp0],    %[temp6]   \n\t"                \
322   "xor    %[temp8],  %[temp8],    %[temp7]   \n\t"                \
323   "subu   %[temp3],  %[temp3],    %[temp4]   \n\t"                \
324   "subu   %[temp1],  %[temp1],    %[temp5]   \n\t"                \
325   "subu   %[temp0],  %[temp0],    %[temp6]   \n\t"                \
326   "subu   %[temp8],  %[temp8],    %[temp7]   \n\t"                \
327   "lhu    %[temp4],  " #E "(%[w])            \n\t"                \
328   "lhu    %[temp5],  " #F "(%[w])            \n\t"                \
329   "lhu    %[temp6],  " #G "(%[w])            \n\t"                \
330   "lhu    %[temp7],  " #H "(%[w])            \n\t"                \
331   "madd   %[temp4],  %[temp3]                \n\t"                \
332   "madd   %[temp5],  %[temp1]                \n\t"                \
333   "madd   %[temp6],  %[temp0]                \n\t"                \
334   "madd   %[temp7],  %[temp8]                \n\t"                \
335   "lw     %[temp0],  " #A "(%[tmp])          \n\t"                \
336   "lw     %[temp1],  " #C "(%[tmp])          \n\t"                \
337   "lw     %[temp2],  " #B "(%[tmp])          \n\t"                \
338   "lw     %[temp3],  " #D "(%[tmp])          \n\t"                \
339   "addu   %[temp8],  %[temp0],    %[temp1]   \n\t"                \
340   "subu   %[temp0],  %[temp0],    %[temp1]   \n\t"                \
341   "addu   %[temp1],  %[temp2],    %[temp3]   \n\t"                \
342   "subu   %[temp2],  %[temp2],    %[temp3]   \n\t"                \
343   "addu   %[temp3],  %[temp8],    %[temp1]   \n\t"                \
344   "subu   %[temp1],  %[temp8],    %[temp1]   \n\t"                \
345   "addu   %[temp8],  %[temp0],    %[temp2]   \n\t"                \
346   "subu   %[temp0],  %[temp0],    %[temp2]   \n\t"                \
347   "sra    %[temp2],  %[temp3],    31         \n\t"                \
348   "xor    %[temp3],  %[temp3],    %[temp2]   \n\t"                \
349   "subu   %[temp3],  %[temp3],    %[temp2]   \n\t"                \
350   "msub   %[temp4],  %[temp3]                \n\t"                \
351   "sra    %[temp2],  %[temp8],    31         \n\t"                \
352   "sra    %[temp3],  %[temp0],    31         \n\t"                \
353   "sra    %[temp4],  %[temp1],    31         \n\t"                \
354   "xor    %[temp8],  %[temp8],    %[temp2]   \n\t"                \
355   "xor    %[temp0],  %[temp0],    %[temp3]   \n\t"                \
356   "xor    %[temp1],  %[temp1],    %[temp4]   \n\t"                \
357   "subu   %[temp8],  %[temp8],    %[temp2]   \n\t"                \
358   "subu   %[temp0],  %[temp0],    %[temp3]   \n\t"                \
359   "subu   %[temp1],  %[temp1],    %[temp4]   \n\t"                \
360   "msub   %[temp5],  %[temp8]                \n\t"                \
361   "msub   %[temp6],  %[temp0]                \n\t"                \
362   "msub   %[temp7],  %[temp1]                \n\t"
363 
Disto4x4(const uint8_t * const a,const uint8_t * const b,const uint16_t * const w)364 static int Disto4x4(const uint8_t* const a, const uint8_t* const b,
365                     const uint16_t* const w) {
366   int tmp[32];
367   int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8;
368 
369   __asm__ volatile(
370     HORIZONTAL_PASS(0,   0,  4,  8, 12,    64,  68,  72,  76)
371     HORIZONTAL_PASS(1,  16, 20, 24, 28,    80,  84,  88,  92)
372     HORIZONTAL_PASS(2,  32, 36, 40, 44,    96, 100, 104, 108)
373     HORIZONTAL_PASS(3,  48, 52, 56, 60,   112, 116, 120, 124)
374     "mthi   $zero                             \n\t"
375     "mtlo   $zero                             \n\t"
376     VERTICAL_PASS( 0, 16, 32, 48,     64, 80,  96, 112,   0,  8, 16, 24)
377     VERTICAL_PASS( 4, 20, 36, 52,     68, 84, 100, 116,   2, 10, 18, 26)
378     VERTICAL_PASS( 8, 24, 40, 56,     72, 88, 104, 120,   4, 12, 20, 28)
379     VERTICAL_PASS(12, 28, 44, 60,     76, 92, 108, 124,   6, 14, 22, 30)
380     "mflo   %[temp0]                          \n\t"
381     "sra    %[temp1],  %[temp0],  31          \n\t"
382     "xor    %[temp0],  %[temp0],  %[temp1]    \n\t"
383     "subu   %[temp0],  %[temp0],  %[temp1]    \n\t"
384     "sra    %[temp0],  %[temp0],  5           \n\t"
385 
386     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
387       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
388       [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [temp8]"=&r"(temp8)
389     : [a]"r"(a), [b]"r"(b), [w]"r"(w), [tmp]"r"(tmp)
390     : "memory", "hi", "lo"
391   );
392 
393   return temp0;
394 }
395 
396 #undef VERTICAL_PASS
397 #undef HORIZONTAL_PASS
398 
Disto16x16(const uint8_t * const a,const uint8_t * const b,const uint16_t * const w)399 static int Disto16x16(const uint8_t* const a, const uint8_t* const b,
400                       const uint16_t* const w) {
401   int D = 0;
402   int x, y;
403   for (y = 0; y < 16 * BPS; y += 4 * BPS) {
404     for (x = 0; x < 16; x += 4) {
405       D += Disto4x4(a + x + y, b + x + y, w);
406     }
407   }
408   return D;
409 }
410 
411 // macro for one horizontal pass in FTransform
412 // temp0..temp15 holds tmp[0]..tmp[15]
413 // A - offset in bytes to load from src and ref buffers
414 // TEMP0..TEMP3 - registers for corresponding tmp elements
415 #define HORIZONTAL_PASS(A, TEMP0, TEMP1, TEMP2, TEMP3)                  \
416   "lw     %[" #TEMP1 "],  0(%[args])                           \n\t"    \
417   "lw     %[" #TEMP2 "],  4(%[args])                           \n\t"    \
418   "lbu    %[temp16],    0+" XSTR(BPS) "*" #A "(%[" #TEMP1 "])  \n\t"    \
419   "lbu    %[temp17],    0+" XSTR(BPS) "*" #A "(%[" #TEMP2 "])  \n\t"    \
420   "lbu    %[temp18],    1+" XSTR(BPS) "*" #A "(%[" #TEMP1 "])  \n\t"    \
421   "lbu    %[temp19],    1+" XSTR(BPS) "*" #A "(%[" #TEMP2 "])  \n\t"    \
422   "subu   %[temp20],    %[temp16],    %[temp17]                \n\t"    \
423   "lbu    %[temp16],    2+" XSTR(BPS) "*" #A "(%[" #TEMP1 "])  \n\t"    \
424   "lbu    %[temp17],    2+" XSTR(BPS) "*" #A "(%[" #TEMP2 "])  \n\t"    \
425   "subu   %[" #TEMP0 "],  %[temp18],    %[temp19]              \n\t"    \
426   "lbu    %[temp18],    3+" XSTR(BPS) "*" #A "(%[" #TEMP1 "])  \n\t"    \
427   "lbu    %[temp19],    3+" XSTR(BPS) "*" #A "(%[" #TEMP2 "])  \n\t"    \
428   "subu   %[" #TEMP1 "],  %[temp16],    %[temp17]              \n\t"    \
429   "subu   %[" #TEMP2 "],  %[temp18],    %[temp19]              \n\t"    \
430   "addu   %[" #TEMP3 "],  %[temp20],    %[" #TEMP2 "]          \n\t"    \
431   "subu   %[" #TEMP2 "],  %[temp20],    %[" #TEMP2 "]          \n\t"    \
432   "addu   %[temp20],    %[" #TEMP0 "],  %[" #TEMP1 "]          \n\t"    \
433   "subu   %[" #TEMP0 "],  %[" #TEMP0 "],  %[" #TEMP1 "]        \n\t"    \
434   "mul    %[temp16],    %[" #TEMP2 "],  %[c5352]               \n\t"    \
435   "mul    %[temp17],    %[" #TEMP2 "],  %[c2217]               \n\t"    \
436   "mul    %[temp18],    %[" #TEMP0 "],  %[c5352]               \n\t"    \
437   "mul    %[temp19],    %[" #TEMP0 "],  %[c2217]               \n\t"    \
438   "addu   %[" #TEMP1 "],  %[" #TEMP3 "],  %[temp20]            \n\t"    \
439   "subu   %[temp20],    %[" #TEMP3 "],  %[temp20]              \n\t"    \
440   "sll    %[" #TEMP0 "],  %[" #TEMP1 "],  3                    \n\t"    \
441   "sll    %[" #TEMP2 "],  %[temp20],    3                      \n\t"    \
442   "addiu  %[temp16],    %[temp16],    1812                     \n\t"    \
443   "addiu  %[temp17],    %[temp17],    937                      \n\t"    \
444   "addu   %[temp16],    %[temp16],    %[temp19]                \n\t"    \
445   "subu   %[temp17],    %[temp17],    %[temp18]                \n\t"    \
446   "sra    %[" #TEMP1 "],  %[temp16],    9                      \n\t"    \
447   "sra    %[" #TEMP3 "],  %[temp17],    9                      \n\t"
448 
449 // macro for one vertical pass in FTransform
450 // temp0..temp15 holds tmp[0]..tmp[15]
451 // A..D - offsets in bytes to store to out buffer
452 // TEMP0, TEMP4, TEMP8 and TEMP12 - registers for corresponding tmp elements
453 #define VERTICAL_PASS(A, B, C, D, TEMP0, TEMP4, TEMP8, TEMP12)    \
454   "addu   %[temp16],    %[" #TEMP0 "],  %[" #TEMP12 "]   \n\t"    \
455   "subu   %[temp19],    %[" #TEMP0 "],  %[" #TEMP12 "]   \n\t"    \
456   "addu   %[temp17],    %[" #TEMP4 "],  %[" #TEMP8 "]    \n\t"    \
457   "subu   %[temp18],    %[" #TEMP4 "],  %[" #TEMP8 "]    \n\t"    \
458   "mul    %[" #TEMP8 "],  %[temp19],    %[c2217]         \n\t"    \
459   "mul    %[" #TEMP12 "], %[temp18],    %[c2217]         \n\t"    \
460   "mul    %[" #TEMP4 "],  %[temp19],    %[c5352]         \n\t"    \
461   "mul    %[temp18],    %[temp18],    %[c5352]           \n\t"    \
462   "addiu  %[temp16],    %[temp16],    7                  \n\t"    \
463   "addu   %[" #TEMP0 "],  %[temp16],    %[temp17]        \n\t"    \
464   "sra    %[" #TEMP0 "],  %[" #TEMP0 "],  4              \n\t"    \
465   "addu   %[" #TEMP12 "], %[" #TEMP12 "], %[" #TEMP4 "]  \n\t"    \
466   "subu   %[" #TEMP4 "],  %[temp16],    %[temp17]        \n\t"    \
467   "sra    %[" #TEMP4 "],  %[" #TEMP4 "],  4              \n\t"    \
468   "addiu  %[" #TEMP8 "],  %[" #TEMP8 "],  30000          \n\t"    \
469   "addiu  %[" #TEMP12 "], %[" #TEMP12 "], 12000          \n\t"    \
470   "addiu  %[" #TEMP8 "],  %[" #TEMP8 "],  21000          \n\t"    \
471   "subu   %[" #TEMP8 "],  %[" #TEMP8 "],  %[temp18]      \n\t"    \
472   "sra    %[" #TEMP12 "], %[" #TEMP12 "], 16             \n\t"    \
473   "sra    %[" #TEMP8 "],  %[" #TEMP8 "],  16             \n\t"    \
474   "addiu  %[temp16],    %[" #TEMP12 "], 1                \n\t"    \
475   "movn   %[" #TEMP12 "], %[temp16],    %[temp19]        \n\t"    \
476   "sh     %[" #TEMP0 "],  " #A "(%[temp20])              \n\t"    \
477   "sh     %[" #TEMP4 "],  " #C "(%[temp20])              \n\t"    \
478   "sh     %[" #TEMP8 "],  " #D "(%[temp20])              \n\t"    \
479   "sh     %[" #TEMP12 "], " #B "(%[temp20])              \n\t"
480 
FTransform(const uint8_t * src,const uint8_t * ref,int16_t * out)481 static void FTransform(const uint8_t* src, const uint8_t* ref, int16_t* out) {
482   int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8;
483   int temp9, temp10, temp11, temp12, temp13, temp14, temp15, temp16;
484   int temp17, temp18, temp19, temp20;
485   const int c2217 = 2217;
486   const int c5352 = 5352;
487   const int* const args[3] =
488       { (const int*)src, (const int*)ref, (const int*)out };
489 
490   __asm__ volatile(
491     HORIZONTAL_PASS(0, temp0,  temp1,  temp2,  temp3)
492     HORIZONTAL_PASS(1, temp4,  temp5,  temp6,  temp7)
493     HORIZONTAL_PASS(2, temp8,  temp9,  temp10, temp11)
494     HORIZONTAL_PASS(3, temp12, temp13, temp14, temp15)
495     "lw   %[temp20],    8(%[args])                     \n\t"
496     VERTICAL_PASS(0,  8, 16, 24, temp0, temp4, temp8,  temp12)
497     VERTICAL_PASS(2, 10, 18, 26, temp1, temp5, temp9,  temp13)
498     VERTICAL_PASS(4, 12, 20, 28, temp2, temp6, temp10, temp14)
499     VERTICAL_PASS(6, 14, 22, 30, temp3, temp7, temp11, temp15)
500 
501     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
502       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
503       [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [temp8]"=&r"(temp8),
504       [temp9]"=&r"(temp9), [temp10]"=&r"(temp10), [temp11]"=&r"(temp11),
505       [temp12]"=&r"(temp12), [temp13]"=&r"(temp13), [temp14]"=&r"(temp14),
506       [temp15]"=&r"(temp15), [temp16]"=&r"(temp16), [temp17]"=&r"(temp17),
507       [temp18]"=&r"(temp18), [temp19]"=&r"(temp19), [temp20]"=&r"(temp20)
508     : [args]"r"(args), [c2217]"r"(c2217), [c5352]"r"(c5352)
509     : "memory", "hi", "lo"
510   );
511 }
512 
513 #undef VERTICAL_PASS
514 #undef HORIZONTAL_PASS
515 
516 #if !defined(WORK_AROUND_GCC)
517 
518 #define GET_SSE_INNER(A, B, C, D)                               \
519   "lbu     %[temp0],    " #A "(%[a])                 \n\t"      \
520   "lbu     %[temp1],    " #A "(%[b])                 \n\t"      \
521   "lbu     %[temp2],    " #B "(%[a])                 \n\t"      \
522   "lbu     %[temp3],    " #B "(%[b])                 \n\t"      \
523   "lbu     %[temp4],    " #C "(%[a])                 \n\t"      \
524   "lbu     %[temp5],    " #C "(%[b])                 \n\t"      \
525   "lbu     %[temp6],    " #D "(%[a])                 \n\t"      \
526   "lbu     %[temp7],    " #D "(%[b])                 \n\t"      \
527   "subu    %[temp0],    %[temp0],     %[temp1]       \n\t"      \
528   "subu    %[temp2],    %[temp2],     %[temp3]       \n\t"      \
529   "subu    %[temp4],    %[temp4],     %[temp5]       \n\t"      \
530   "subu    %[temp6],    %[temp6],     %[temp7]       \n\t"      \
531   "madd    %[temp0],    %[temp0]                     \n\t"      \
532   "madd    %[temp2],    %[temp2]                     \n\t"      \
533   "madd    %[temp4],    %[temp4]                     \n\t"      \
534   "madd    %[temp6],    %[temp6]                     \n\t"
535 
536 #define GET_SSE(A, B, C, D)               \
537   GET_SSE_INNER(A, A + 1, A + 2, A + 3)   \
538   GET_SSE_INNER(B, B + 1, B + 2, B + 3)   \
539   GET_SSE_INNER(C, C + 1, C + 2, C + 3)   \
540   GET_SSE_INNER(D, D + 1, D + 2, D + 3)
541 
SSE16x16(const uint8_t * a,const uint8_t * b)542 static int SSE16x16(const uint8_t* a, const uint8_t* b) {
543   int count;
544   int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
545 
546   __asm__ volatile(
547      "mult   $zero,    $zero                            \n\t"
548 
549      GET_SSE( 0 * BPS, 4 +  0 * BPS, 8 +  0 * BPS, 12 +  0 * BPS)
550      GET_SSE( 1 * BPS, 4 +  1 * BPS, 8 +  1 * BPS, 12 +  1 * BPS)
551      GET_SSE( 2 * BPS, 4 +  2 * BPS, 8 +  2 * BPS, 12 +  2 * BPS)
552      GET_SSE( 3 * BPS, 4 +  3 * BPS, 8 +  3 * BPS, 12 +  3 * BPS)
553      GET_SSE( 4 * BPS, 4 +  4 * BPS, 8 +  4 * BPS, 12 +  4 * BPS)
554      GET_SSE( 5 * BPS, 4 +  5 * BPS, 8 +  5 * BPS, 12 +  5 * BPS)
555      GET_SSE( 6 * BPS, 4 +  6 * BPS, 8 +  6 * BPS, 12 +  6 * BPS)
556      GET_SSE( 7 * BPS, 4 +  7 * BPS, 8 +  7 * BPS, 12 +  7 * BPS)
557      GET_SSE( 8 * BPS, 4 +  8 * BPS, 8 +  8 * BPS, 12 +  8 * BPS)
558      GET_SSE( 9 * BPS, 4 +  9 * BPS, 8 +  9 * BPS, 12 +  9 * BPS)
559      GET_SSE(10 * BPS, 4 + 10 * BPS, 8 + 10 * BPS, 12 + 10 * BPS)
560      GET_SSE(11 * BPS, 4 + 11 * BPS, 8 + 11 * BPS, 12 + 11 * BPS)
561      GET_SSE(12 * BPS, 4 + 12 * BPS, 8 + 12 * BPS, 12 + 12 * BPS)
562      GET_SSE(13 * BPS, 4 + 13 * BPS, 8 + 13 * BPS, 12 + 13 * BPS)
563      GET_SSE(14 * BPS, 4 + 14 * BPS, 8 + 14 * BPS, 12 + 14 * BPS)
564      GET_SSE(15 * BPS, 4 + 15 * BPS, 8 + 15 * BPS, 12 + 15 * BPS)
565 
566     "mflo    %[count]                                   \n\t"
567     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
568       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
569       [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [count]"=&r"(count)
570     : [a]"r"(a), [b]"r"(b)
571     : "memory", "hi", "lo"
572   );
573   return count;
574 }
575 
SSE16x8(const uint8_t * a,const uint8_t * b)576 static int SSE16x8(const uint8_t* a, const uint8_t* b) {
577   int count;
578   int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
579 
580   __asm__ volatile(
581      "mult   $zero,    $zero                            \n\t"
582 
583      GET_SSE( 0 * BPS, 4 +  0 * BPS, 8 +  0 * BPS, 12 +  0 * BPS)
584      GET_SSE( 1 * BPS, 4 +  1 * BPS, 8 +  1 * BPS, 12 +  1 * BPS)
585      GET_SSE( 2 * BPS, 4 +  2 * BPS, 8 +  2 * BPS, 12 +  2 * BPS)
586      GET_SSE( 3 * BPS, 4 +  3 * BPS, 8 +  3 * BPS, 12 +  3 * BPS)
587      GET_SSE( 4 * BPS, 4 +  4 * BPS, 8 +  4 * BPS, 12 +  4 * BPS)
588      GET_SSE( 5 * BPS, 4 +  5 * BPS, 8 +  5 * BPS, 12 +  5 * BPS)
589      GET_SSE( 6 * BPS, 4 +  6 * BPS, 8 +  6 * BPS, 12 +  6 * BPS)
590      GET_SSE( 7 * BPS, 4 +  7 * BPS, 8 +  7 * BPS, 12 +  7 * BPS)
591 
592     "mflo    %[count]                                   \n\t"
593     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
594       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
595       [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [count]"=&r"(count)
596     : [a]"r"(a), [b]"r"(b)
597     : "memory", "hi", "lo"
598   );
599   return count;
600 }
601 
SSE8x8(const uint8_t * a,const uint8_t * b)602 static int SSE8x8(const uint8_t* a, const uint8_t* b) {
603   int count;
604   int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
605 
606   __asm__ volatile(
607      "mult   $zero,    $zero                            \n\t"
608 
609      GET_SSE(0 * BPS, 4 + 0 * BPS, 1 * BPS, 4 + 1 * BPS)
610      GET_SSE(2 * BPS, 4 + 2 * BPS, 3 * BPS, 4 + 3 * BPS)
611      GET_SSE(4 * BPS, 4 + 4 * BPS, 5 * BPS, 4 + 5 * BPS)
612      GET_SSE(6 * BPS, 4 + 6 * BPS, 7 * BPS, 4 + 7 * BPS)
613 
614     "mflo    %[count]                                   \n\t"
615     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
616       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
617       [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [count]"=&r"(count)
618     : [a]"r"(a), [b]"r"(b)
619     : "memory", "hi", "lo"
620   );
621   return count;
622 }
623 
SSE4x4(const uint8_t * a,const uint8_t * b)624 static int SSE4x4(const uint8_t* a, const uint8_t* b) {
625   int count;
626   int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
627 
628   __asm__ volatile(
629      "mult   $zero,    $zero                            \n\t"
630 
631      GET_SSE(0 * BPS, 1 * BPS, 2 * BPS, 3 * BPS)
632 
633     "mflo    %[count]                                   \n\t"
634     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
635       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
636       [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [count]"=&r"(count)
637     : [a]"r"(a), [b]"r"(b)
638     : "memory", "hi", "lo"
639   );
640   return count;
641 }
642 
643 #undef GET_SSE
644 #undef GET_SSE_INNER
645 
646 #endif  // !WORK_AROUND_GCC
647 
648 //------------------------------------------------------------------------------
649 // Entry point
650 
651 extern void VP8EncDspInitMIPS32(void);
652 
VP8EncDspInitMIPS32(void)653 WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspInitMIPS32(void) {
654   VP8ITransform = ITransform;
655   VP8FTransform = FTransform;
656   VP8EncQuantizeBlock = QuantizeBlock;
657   VP8EncQuantize2Blocks = Quantize2Blocks;
658   VP8TDisto4x4 = Disto4x4;
659   VP8TDisto16x16 = Disto16x16;
660 #if !defined(WORK_AROUND_GCC)
661   VP8SSE16x16 = SSE16x16;
662   VP8SSE8x8 = SSE8x8;
663   VP8SSE16x8 = SSE16x8;
664   VP8SSE4x4 = SSE4x4;
665 #endif
666 }
667 
668 #else  // !WEBP_USE_MIPS32
669 
670 WEBP_DSP_INIT_STUB(VP8EncDspInitMIPS32)
671 
672 #endif  // WEBP_USE_MIPS32
673