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 // Image transforms and color space conversion methods for lossless decoder.
11 //
12 // Author(s):  Djordje Pesut    (djordje.pesut@imgtec.com)
13 //             Jovan Zelincevic (jovan.zelincevic@imgtec.com)
14 
15 #include "./dsp.h"
16 
17 #if defined(WEBP_USE_MIPS_DSP_R2)
18 
19 #include "./lossless.h"
20 
21 #define MAP_COLOR_FUNCS(FUNC_NAME, TYPE, GET_INDEX, GET_VALUE)                 \
22 static void FUNC_NAME(const TYPE* src,                                         \
23                       const uint32_t* const color_map,                         \
24                       TYPE* dst, int y_start, int y_end,                       \
25                       int width) {                                             \
26   int y;                                                                       \
27   for (y = y_start; y < y_end; ++y) {                                          \
28     int x;                                                                     \
29     for (x = 0; x < (width >> 2); ++x) {                                       \
30       int tmp1, tmp2, tmp3, tmp4;                                              \
31       __asm__ volatile (                                                       \
32       ".ifc        " #TYPE ",  uint8_t                  \n\t"                  \
33         "lbu       %[tmp1],  0(%[src])                  \n\t"                  \
34         "lbu       %[tmp2],  1(%[src])                  \n\t"                  \
35         "lbu       %[tmp3],  2(%[src])                  \n\t"                  \
36         "lbu       %[tmp4],  3(%[src])                  \n\t"                  \
37         "addiu     %[src],   %[src],      4             \n\t"                  \
38       ".endif                                           \n\t"                  \
39       ".ifc        " #TYPE ",  uint32_t                 \n\t"                  \
40         "lw        %[tmp1],  0(%[src])                  \n\t"                  \
41         "lw        %[tmp2],  4(%[src])                  \n\t"                  \
42         "lw        %[tmp3],  8(%[src])                  \n\t"                  \
43         "lw        %[tmp4],  12(%[src])                 \n\t"                  \
44         "ext       %[tmp1],  %[tmp1],     8,        8   \n\t"                  \
45         "ext       %[tmp2],  %[tmp2],     8,        8   \n\t"                  \
46         "ext       %[tmp3],  %[tmp3],     8,        8   \n\t"                  \
47         "ext       %[tmp4],  %[tmp4],     8,        8   \n\t"                  \
48         "addiu     %[src],   %[src],      16            \n\t"                  \
49       ".endif                                           \n\t"                  \
50         "sll       %[tmp1],  %[tmp1],     2             \n\t"                  \
51         "sll       %[tmp2],  %[tmp2],     2             \n\t"                  \
52         "sll       %[tmp3],  %[tmp3],     2             \n\t"                  \
53         "sll       %[tmp4],  %[tmp4],     2             \n\t"                  \
54         "lwx       %[tmp1],  %[tmp1](%[color_map])      \n\t"                  \
55         "lwx       %[tmp2],  %[tmp2](%[color_map])      \n\t"                  \
56         "lwx       %[tmp3],  %[tmp3](%[color_map])      \n\t"                  \
57         "lwx       %[tmp4],  %[tmp4](%[color_map])      \n\t"                  \
58       ".ifc        " #TYPE ",  uint8_t                  \n\t"                  \
59         "ext       %[tmp1],  %[tmp1],     8,        8   \n\t"                  \
60         "ext       %[tmp2],  %[tmp2],     8,        8   \n\t"                  \
61         "ext       %[tmp3],  %[tmp3],     8,        8   \n\t"                  \
62         "ext       %[tmp4],  %[tmp4],     8,        8   \n\t"                  \
63         "sb        %[tmp1],  0(%[dst])                  \n\t"                  \
64         "sb        %[tmp2],  1(%[dst])                  \n\t"                  \
65         "sb        %[tmp3],  2(%[dst])                  \n\t"                  \
66         "sb        %[tmp4],  3(%[dst])                  \n\t"                  \
67         "addiu     %[dst],   %[dst],      4             \n\t"                  \
68       ".endif                                           \n\t"                  \
69       ".ifc        " #TYPE ",  uint32_t                 \n\t"                  \
70         "sw        %[tmp1],  0(%[dst])                  \n\t"                  \
71         "sw        %[tmp2],  4(%[dst])                  \n\t"                  \
72         "sw        %[tmp3],  8(%[dst])                  \n\t"                  \
73         "sw        %[tmp4],  12(%[dst])                 \n\t"                  \
74         "addiu     %[dst],   %[dst],      16            \n\t"                  \
75       ".endif                                           \n\t"                  \
76         : [tmp1]"=&r"(tmp1), [tmp2]"=&r"(tmp2), [tmp3]"=&r"(tmp3),             \
77           [tmp4]"=&r"(tmp4), [src]"+&r"(src), [dst]"+r"(dst)                   \
78         : [color_map]"r"(color_map)                                            \
79         : "memory"                                                             \
80       );                                                                       \
81     }                                                                          \
82     for (x = 0; x < (width & 3); ++x) {                                        \
83       *dst++ = GET_VALUE(color_map[GET_INDEX(*src++)]);                        \
84     }                                                                          \
85   }                                                                            \
86 }
87 
MAP_COLOR_FUNCS(MapARGB,uint32_t,VP8GetARGBIndex,VP8GetARGBValue)88 MAP_COLOR_FUNCS(MapARGB, uint32_t, VP8GetARGBIndex, VP8GetARGBValue)
89 MAP_COLOR_FUNCS(MapAlpha, uint8_t, VP8GetAlphaIndex, VP8GetAlphaValue)
90 
91 #undef MAP_COLOR_FUNCS
92 
93 static WEBP_INLINE uint32_t ClampedAddSubtractFull(uint32_t c0, uint32_t c1,
94                                                    uint32_t c2) {
95   int temp0, temp1, temp2, temp3, temp4, temp5;
96   __asm__ volatile (
97     "preceu.ph.qbr   %[temp1],   %[c0]                 \n\t"
98     "preceu.ph.qbl   %[temp2],   %[c0]                 \n\t"
99     "preceu.ph.qbr   %[temp3],   %[c1]                 \n\t"
100     "preceu.ph.qbl   %[temp4],   %[c1]                 \n\t"
101     "preceu.ph.qbr   %[temp5],   %[c2]                 \n\t"
102     "preceu.ph.qbl   %[temp0],   %[c2]                 \n\t"
103     "subq.ph         %[temp3],   %[temp3],   %[temp5]  \n\t"
104     "subq.ph         %[temp4],   %[temp4],   %[temp0]  \n\t"
105     "addq.ph         %[temp1],   %[temp1],   %[temp3]  \n\t"
106     "addq.ph         %[temp2],   %[temp2],   %[temp4]  \n\t"
107     "shll_s.ph       %[temp1],   %[temp1],   7         \n\t"
108     "shll_s.ph       %[temp2],   %[temp2],   7         \n\t"
109     "precrqu_s.qb.ph %[temp2],   %[temp2],   %[temp1]  \n\t"
110     : [temp0]"=r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
111       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5)
112     : [c0]"r"(c0), [c1]"r"(c1), [c2]"r"(c2)
113     : "memory"
114   );
115   return temp2;
116 }
117 
ClampedAddSubtractHalf(uint32_t c0,uint32_t c1,uint32_t c2)118 static WEBP_INLINE uint32_t ClampedAddSubtractHalf(uint32_t c0, uint32_t c1,
119                                                    uint32_t c2) {
120   int temp0, temp1, temp2, temp3, temp4, temp5;
121   __asm__ volatile (
122     "adduh.qb         %[temp5],   %[c0],      %[c1]       \n\t"
123     "preceu.ph.qbr    %[temp3],   %[c2]                   \n\t"
124     "preceu.ph.qbr    %[temp1],   %[temp5]                \n\t"
125     "preceu.ph.qbl    %[temp2],   %[temp5]                \n\t"
126     "preceu.ph.qbl    %[temp4],   %[c2]                   \n\t"
127     "subq.ph          %[temp3],   %[temp1],   %[temp3]    \n\t"
128     "subq.ph          %[temp4],   %[temp2],   %[temp4]    \n\t"
129     "shrl.ph          %[temp5],   %[temp3],   15          \n\t"
130     "shrl.ph          %[temp0],   %[temp4],   15          \n\t"
131     "addq.ph          %[temp3],   %[temp3],   %[temp5]    \n\t"
132     "addq.ph          %[temp4],   %[temp0],   %[temp4]    \n\t"
133     "shra.ph          %[temp3],   %[temp3],   1           \n\t"
134     "shra.ph          %[temp4],   %[temp4],   1           \n\t"
135     "addq.ph          %[temp1],   %[temp1],   %[temp3]    \n\t"
136     "addq.ph          %[temp2],   %[temp2],   %[temp4]    \n\t"
137     "shll_s.ph        %[temp1],   %[temp1],   7           \n\t"
138     "shll_s.ph        %[temp2],   %[temp2],   7           \n\t"
139     "precrqu_s.qb.ph  %[temp1],   %[temp2],   %[temp1]    \n\t"
140     : [temp0]"=r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
141       [temp3]"=&r"(temp3), [temp4]"=r"(temp4), [temp5]"=&r"(temp5)
142     : [c0]"r"(c0), [c1]"r"(c1), [c2]"r"(c2)
143     : "memory"
144   );
145   return temp1;
146 }
147 
Select(uint32_t a,uint32_t b,uint32_t c)148 static WEBP_INLINE uint32_t Select(uint32_t a, uint32_t b, uint32_t c) {
149   int temp0, temp1, temp2, temp3, temp4, temp5;
150   __asm__ volatile (
151     "cmpgdu.lt.qb %[temp1], %[c],     %[b]             \n\t"
152     "pick.qb      %[temp1], %[b],     %[c]             \n\t"
153     "pick.qb      %[temp2], %[c],     %[b]             \n\t"
154     "cmpgdu.lt.qb %[temp4], %[c],     %[a]             \n\t"
155     "pick.qb      %[temp4], %[a],     %[c]             \n\t"
156     "pick.qb      %[temp5], %[c],     %[a]             \n\t"
157     "subu.qb      %[temp3], %[temp1], %[temp2]         \n\t"
158     "subu.qb      %[temp0], %[temp4], %[temp5]         \n\t"
159     "raddu.w.qb   %[temp3], %[temp3]                   \n\t"
160     "raddu.w.qb   %[temp0], %[temp0]                   \n\t"
161     "subu         %[temp3], %[temp3], %[temp0]         \n\t"
162     "slti         %[temp0], %[temp3], 0x1              \n\t"
163     "movz         %[a],     %[b],     %[temp0]         \n\t"
164     : [temp1]"=&r"(temp1), [temp2]"=&r"(temp2), [temp3]"=&r"(temp3),
165       [temp4]"=&r"(temp4), [temp5]"=&r"(temp5), [temp0]"=&r"(temp0),
166       [a]"+&r"(a)
167     : [b]"r"(b), [c]"r"(c)
168   );
169   return a;
170 }
171 
Average2(uint32_t a0,uint32_t a1)172 static WEBP_INLINE uint32_t Average2(uint32_t a0, uint32_t a1) {
173   __asm__ volatile (
174     "adduh.qb    %[a0], %[a0], %[a1]       \n\t"
175     : [a0]"+r"(a0)
176     : [a1]"r"(a1)
177   );
178   return a0;
179 }
180 
Average3(uint32_t a0,uint32_t a1,uint32_t a2)181 static WEBP_INLINE uint32_t Average3(uint32_t a0, uint32_t a1, uint32_t a2) {
182   return Average2(Average2(a0, a2), a1);
183 }
184 
Average4(uint32_t a0,uint32_t a1,uint32_t a2,uint32_t a3)185 static WEBP_INLINE uint32_t Average4(uint32_t a0, uint32_t a1,
186                                      uint32_t a2, uint32_t a3) {
187   return Average2(Average2(a0, a1), Average2(a2, a3));
188 }
189 
Predictor5(uint32_t left,const uint32_t * const top)190 static uint32_t Predictor5(uint32_t left, const uint32_t* const top) {
191   return Average3(left, top[0], top[1]);
192 }
193 
Predictor6(uint32_t left,const uint32_t * const top)194 static uint32_t Predictor6(uint32_t left, const uint32_t* const top) {
195   return Average2(left, top[-1]);
196 }
197 
Predictor7(uint32_t left,const uint32_t * const top)198 static uint32_t Predictor7(uint32_t left, const uint32_t* const top) {
199   return Average2(left, top[0]);
200 }
201 
Predictor8(uint32_t left,const uint32_t * const top)202 static uint32_t Predictor8(uint32_t left, const uint32_t* const top) {
203   (void)left;
204   return Average2(top[-1], top[0]);
205 }
206 
Predictor9(uint32_t left,const uint32_t * const top)207 static uint32_t Predictor9(uint32_t left, const uint32_t* const top) {
208   (void)left;
209   return Average2(top[0], top[1]);
210 }
211 
Predictor10(uint32_t left,const uint32_t * const top)212 static uint32_t Predictor10(uint32_t left, const uint32_t* const top) {
213   return Average4(left, top[-1], top[0], top[1]);
214 }
215 
Predictor11(uint32_t left,const uint32_t * const top)216 static uint32_t Predictor11(uint32_t left, const uint32_t* const top) {
217   return Select(top[0], left, top[-1]);
218 }
219 
Predictor12(uint32_t left,const uint32_t * const top)220 static uint32_t Predictor12(uint32_t left, const uint32_t* const top) {
221   return ClampedAddSubtractFull(left, top[0], top[-1]);
222 }
223 
Predictor13(uint32_t left,const uint32_t * const top)224 static uint32_t Predictor13(uint32_t left, const uint32_t* const top) {
225   return ClampedAddSubtractHalf(left, top[0], top[-1]);
226 }
227 
228 // Add green to blue and red channels (i.e. perform the inverse transform of
229 // 'subtract green').
AddGreenToBlueAndRed(uint32_t * data,int num_pixels)230 static void AddGreenToBlueAndRed(uint32_t* data, int num_pixels) {
231   uint32_t temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
232   uint32_t* const p_loop1_end = data + (num_pixels & ~3);
233   uint32_t* const p_loop2_end = data + num_pixels;
234   __asm__ volatile (
235     ".set       push                                          \n\t"
236     ".set       noreorder                                     \n\t"
237     "beq        %[data],         %[p_loop1_end],     3f       \n\t"
238     " nop                                                     \n\t"
239   "0:                                                         \n\t"
240     "lw         %[temp0],        0(%[data])                   \n\t"
241     "lw         %[temp1],        4(%[data])                   \n\t"
242     "lw         %[temp2],        8(%[data])                   \n\t"
243     "lw         %[temp3],        12(%[data])                  \n\t"
244     "ext        %[temp4],        %[temp0],           8,    8  \n\t"
245     "ext        %[temp5],        %[temp1],           8,    8  \n\t"
246     "ext        %[temp6],        %[temp2],           8,    8  \n\t"
247     "ext        %[temp7],        %[temp3],           8,    8  \n\t"
248     "addiu      %[data],         %[data],            16       \n\t"
249     "replv.ph   %[temp4],        %[temp4]                     \n\t"
250     "replv.ph   %[temp5],        %[temp5]                     \n\t"
251     "replv.ph   %[temp6],        %[temp6]                     \n\t"
252     "replv.ph   %[temp7],        %[temp7]                     \n\t"
253     "addu.qb    %[temp0],        %[temp0],           %[temp4] \n\t"
254     "addu.qb    %[temp1],        %[temp1],           %[temp5] \n\t"
255     "addu.qb    %[temp2],        %[temp2],           %[temp6] \n\t"
256     "addu.qb    %[temp3],        %[temp3],           %[temp7] \n\t"
257     "sw         %[temp0],        -16(%[data])                 \n\t"
258     "sw         %[temp1],        -12(%[data])                 \n\t"
259     "sw         %[temp2],        -8(%[data])                  \n\t"
260     "bne        %[data],         %[p_loop1_end],     0b       \n\t"
261     " sw        %[temp3],        -4(%[data])                  \n\t"
262   "3:                                                         \n\t"
263     "beq        %[data],         %[p_loop2_end],     2f       \n\t"
264     " nop                                                     \n\t"
265   "1:                                                         \n\t"
266     "lw         %[temp0],        0(%[data])                   \n\t"
267     "addiu      %[data],         %[data],            4        \n\t"
268     "ext        %[temp4],        %[temp0],           8,    8  \n\t"
269     "replv.ph   %[temp4],        %[temp4]                     \n\t"
270     "addu.qb    %[temp0],        %[temp0],           %[temp4] \n\t"
271     "bne        %[data],         %[p_loop2_end],     1b       \n\t"
272     " sw        %[temp0],        -4(%[data])                  \n\t"
273   "2:                                                         \n\t"
274     ".set       pop                                           \n\t"
275     : [data]"+&r"(data), [temp0]"=&r"(temp0), [temp1]"=&r"(temp1),
276       [temp2]"=&r"(temp2), [temp3]"=&r"(temp3), [temp4]"=&r"(temp4),
277       [temp5]"=&r"(temp5), [temp6]"=&r"(temp6), [temp7]"=&r"(temp7)
278     : [p_loop1_end]"r"(p_loop1_end), [p_loop2_end]"r"(p_loop2_end)
279     : "memory"
280   );
281 }
282 
TransformColorInverse(const VP8LMultipliers * const m,uint32_t * data,int num_pixels)283 static void TransformColorInverse(const VP8LMultipliers* const m,
284                                   uint32_t* data, int num_pixels) {
285   int temp0, temp1, temp2, temp3, temp4, temp5;
286   uint32_t argb, argb1, new_red;
287   const uint32_t G_to_R = m->green_to_red_;
288   const uint32_t G_to_B = m->green_to_blue_;
289   const uint32_t R_to_B = m->red_to_blue_;
290   uint32_t* const p_loop_end = data + (num_pixels & ~1);
291   __asm__ volatile (
292     ".set            push                                    \n\t"
293     ".set            noreorder                               \n\t"
294     "beq             %[data],      %[p_loop_end],  1f        \n\t"
295     " nop                                                    \n\t"
296     "replv.ph        %[temp0],     %[G_to_R]                 \n\t"
297     "replv.ph        %[temp1],     %[G_to_B]                 \n\t"
298     "replv.ph        %[temp2],     %[R_to_B]                 \n\t"
299     "shll.ph         %[temp0],     %[temp0],       8         \n\t"
300     "shll.ph         %[temp1],     %[temp1],       8         \n\t"
301     "shll.ph         %[temp2],     %[temp2],       8         \n\t"
302     "shra.ph         %[temp0],     %[temp0],       8         \n\t"
303     "shra.ph         %[temp1],     %[temp1],       8         \n\t"
304     "shra.ph         %[temp2],     %[temp2],       8         \n\t"
305   "0:                                                        \n\t"
306     "lw              %[argb],      0(%[data])                \n\t"
307     "lw              %[argb1],     4(%[data])                \n\t"
308     "addiu           %[data],      %[data],        8         \n\t"
309     "precrq.qb.ph    %[temp3],     %[argb],        %[argb1]  \n\t"
310     "preceu.ph.qbra  %[temp3],     %[temp3]                  \n\t"
311     "shll.ph         %[temp3],     %[temp3],       8         \n\t"
312     "shra.ph         %[temp3],     %[temp3],       8         \n\t"
313     "mul.ph          %[temp5],     %[temp3],       %[temp0]  \n\t"
314     "mul.ph          %[temp3],     %[temp3],       %[temp1]  \n\t"
315     "precrq.ph.w     %[new_red],   %[argb],        %[argb1]  \n\t"
316     "ins             %[argb1],     %[argb],        16,   16  \n\t"
317     "shra.ph         %[temp5],     %[temp5],       5         \n\t"
318     "shra.ph         %[temp3],     %[temp3],       5         \n\t"
319     "addu.ph         %[new_red],   %[new_red],     %[temp5]  \n\t"
320     "addu.ph         %[argb1],     %[argb1],       %[temp3]  \n\t"
321     "preceu.ph.qbra  %[temp5],     %[new_red]                \n\t"
322     "shll.ph         %[temp4],     %[temp5],       8         \n\t"
323     "shra.ph         %[temp4],     %[temp4],       8         \n\t"
324     "mul.ph          %[temp4],     %[temp4],       %[temp2]  \n\t"
325     "sb              %[temp5],     -2(%[data])               \n\t"
326     "sra             %[temp5],     %[temp5],       16        \n\t"
327     "shra.ph         %[temp4],     %[temp4],       5         \n\t"
328     "addu.ph         %[argb1],     %[argb1],       %[temp4]  \n\t"
329     "preceu.ph.qbra  %[temp3],     %[argb1]                  \n\t"
330     "sb              %[temp5],     -6(%[data])               \n\t"
331     "sb              %[temp3],     -4(%[data])               \n\t"
332     "sra             %[temp3],     %[temp3],       16        \n\t"
333     "bne             %[data],      %[p_loop_end],  0b        \n\t"
334     " sb             %[temp3],     -8(%[data])               \n\t"
335   "1:                                                        \n\t"
336     ".set            pop                                     \n\t"
337     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
338       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
339       [new_red]"=&r"(new_red), [argb]"=&r"(argb),
340       [argb1]"=&r"(argb1), [data]"+&r"(data)
341     : [G_to_R]"r"(G_to_R), [R_to_B]"r"(R_to_B),
342       [G_to_B]"r"(G_to_B), [p_loop_end]"r"(p_loop_end)
343     : "memory", "hi", "lo"
344   );
345 
346   // Fall-back to C-version for left-overs.
347   if (num_pixels & 1) VP8LTransformColorInverse_C(m, data, 1);
348 }
349 
ConvertBGRAToRGB(const uint32_t * src,int num_pixels,uint8_t * dst)350 static void ConvertBGRAToRGB(const uint32_t* src,
351                              int num_pixels, uint8_t* dst) {
352   int temp0, temp1, temp2, temp3;
353   const uint32_t* const p_loop1_end = src + (num_pixels & ~3);
354   const uint32_t* const p_loop2_end = src + num_pixels;
355   __asm__ volatile (
356     ".set       push                                       \n\t"
357     ".set       noreorder                                  \n\t"
358     "beq        %[src],      %[p_loop1_end],    3f         \n\t"
359     " nop                                                  \n\t"
360   "0:                                                      \n\t"
361     "lw         %[temp3],    12(%[src])                    \n\t"
362     "lw         %[temp2],    8(%[src])                     \n\t"
363     "lw         %[temp1],    4(%[src])                     \n\t"
364     "lw         %[temp0],    0(%[src])                     \n\t"
365     "ins        %[temp3],    %[temp2],          24,   8    \n\t"
366     "sll        %[temp2],    %[temp2],          8          \n\t"
367     "rotr       %[temp3],    %[temp3],          16         \n\t"
368     "ins        %[temp2],    %[temp1],          0,    16   \n\t"
369     "sll        %[temp1],    %[temp1],          8          \n\t"
370     "wsbh       %[temp3],    %[temp3]                      \n\t"
371     "balign     %[temp0],    %[temp1],          1          \n\t"
372     "wsbh       %[temp2],    %[temp2]                      \n\t"
373     "wsbh       %[temp0],    %[temp0]                      \n\t"
374     "usw        %[temp3],    8(%[dst])                     \n\t"
375     "rotr       %[temp0],    %[temp0],          16         \n\t"
376     "usw        %[temp2],    4(%[dst])                     \n\t"
377     "addiu      %[src],      %[src],            16         \n\t"
378     "usw        %[temp0],    0(%[dst])                     \n\t"
379     "bne        %[src],      %[p_loop1_end],    0b         \n\t"
380     " addiu     %[dst],      %[dst],            12         \n\t"
381   "3:                                                      \n\t"
382     "beq        %[src],      %[p_loop2_end],    2f         \n\t"
383     " nop                                                  \n\t"
384   "1:                                                      \n\t"
385     "lw         %[temp0],    0(%[src])                     \n\t"
386     "addiu      %[src],      %[src],            4          \n\t"
387     "wsbh       %[temp1],    %[temp0]                      \n\t"
388     "addiu      %[dst],      %[dst],            3          \n\t"
389     "ush        %[temp1],    -2(%[dst])                    \n\t"
390     "sra        %[temp0],    %[temp0],          16         \n\t"
391     "bne        %[src],      %[p_loop2_end],    1b         \n\t"
392     " sb        %[temp0],    -3(%[dst])                    \n\t"
393   "2:                                                      \n\t"
394     ".set       pop                                        \n\t"
395     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
396       [temp3]"=&r"(temp3), [dst]"+&r"(dst), [src]"+&r"(src)
397     : [p_loop1_end]"r"(p_loop1_end), [p_loop2_end]"r"(p_loop2_end)
398     : "memory"
399   );
400 }
401 
ConvertBGRAToRGBA(const uint32_t * src,int num_pixels,uint8_t * dst)402 static void ConvertBGRAToRGBA(const uint32_t* src,
403                               int num_pixels, uint8_t* dst) {
404   int temp0, temp1, temp2, temp3;
405   const uint32_t* const p_loop1_end = src + (num_pixels & ~3);
406   const uint32_t* const p_loop2_end = src + num_pixels;
407   __asm__ volatile (
408     ".set       push                                       \n\t"
409     ".set       noreorder                                  \n\t"
410     "beq        %[src],      %[p_loop1_end],    3f         \n\t"
411     " nop                                                  \n\t"
412   "0:                                                      \n\t"
413     "lw         %[temp0],    0(%[src])                     \n\t"
414     "lw         %[temp1],    4(%[src])                     \n\t"
415     "lw         %[temp2],    8(%[src])                     \n\t"
416     "lw         %[temp3],    12(%[src])                    \n\t"
417     "wsbh       %[temp0],    %[temp0]                      \n\t"
418     "wsbh       %[temp1],    %[temp1]                      \n\t"
419     "wsbh       %[temp2],    %[temp2]                      \n\t"
420     "wsbh       %[temp3],    %[temp3]                      \n\t"
421     "addiu      %[src],      %[src],            16         \n\t"
422     "balign     %[temp0],    %[temp0],          1          \n\t"
423     "balign     %[temp1],    %[temp1],          1          \n\t"
424     "balign     %[temp2],    %[temp2],          1          \n\t"
425     "balign     %[temp3],    %[temp3],          1          \n\t"
426     "usw        %[temp0],    0(%[dst])                     \n\t"
427     "usw        %[temp1],    4(%[dst])                     \n\t"
428     "usw        %[temp2],    8(%[dst])                     \n\t"
429     "usw        %[temp3],    12(%[dst])                    \n\t"
430     "bne        %[src],      %[p_loop1_end],    0b         \n\t"
431     " addiu     %[dst],      %[dst],            16         \n\t"
432   "3:                                                      \n\t"
433     "beq        %[src],      %[p_loop2_end],    2f         \n\t"
434     " nop                                                  \n\t"
435   "1:                                                      \n\t"
436     "lw         %[temp0],    0(%[src])                     \n\t"
437     "wsbh       %[temp0],    %[temp0]                      \n\t"
438     "addiu      %[src],      %[src],            4          \n\t"
439     "balign     %[temp0],    %[temp0],          1          \n\t"
440     "usw        %[temp0],    0(%[dst])                     \n\t"
441     "bne        %[src],      %[p_loop2_end],    1b         \n\t"
442     " addiu     %[dst],      %[dst],            4          \n\t"
443   "2:                                                      \n\t"
444     ".set       pop                                        \n\t"
445     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
446       [temp3]"=&r"(temp3), [dst]"+&r"(dst), [src]"+&r"(src)
447     : [p_loop1_end]"r"(p_loop1_end), [p_loop2_end]"r"(p_loop2_end)
448     : "memory"
449   );
450 }
451 
ConvertBGRAToRGBA4444(const uint32_t * src,int num_pixels,uint8_t * dst)452 static void ConvertBGRAToRGBA4444(const uint32_t* src,
453                                   int num_pixels, uint8_t* dst) {
454   int temp0, temp1, temp2, temp3, temp4, temp5;
455   const uint32_t* const p_loop1_end = src + (num_pixels & ~3);
456   const uint32_t* const p_loop2_end = src + num_pixels;
457   __asm__ volatile (
458     ".set           push                                       \n\t"
459     ".set           noreorder                                  \n\t"
460     "beq            %[src],      %[p_loop1_end],    3f         \n\t"
461     " nop                                                      \n\t"
462   "0:                                                          \n\t"
463     "lw             %[temp0],    0(%[src])                     \n\t"
464     "lw             %[temp1],    4(%[src])                     \n\t"
465     "lw             %[temp2],    8(%[src])                     \n\t"
466     "lw             %[temp3],    12(%[src])                    \n\t"
467     "ext            %[temp4],    %[temp0],          28,   4    \n\t"
468     "ext            %[temp5],    %[temp0],          12,   4    \n\t"
469     "ins            %[temp0],    %[temp4],          0,    4    \n\t"
470     "ext            %[temp4],    %[temp1],          28,   4    \n\t"
471     "ins            %[temp0],    %[temp5],          16,   4    \n\t"
472     "ext            %[temp5],    %[temp1],          12,   4    \n\t"
473     "ins            %[temp1],    %[temp4],          0,    4    \n\t"
474     "ext            %[temp4],    %[temp2],          28,   4    \n\t"
475     "ins            %[temp1],    %[temp5],          16,   4    \n\t"
476     "ext            %[temp5],    %[temp2],          12,   4    \n\t"
477     "ins            %[temp2],    %[temp4],          0,    4    \n\t"
478     "ext            %[temp4],    %[temp3],          28,   4    \n\t"
479     "ins            %[temp2],    %[temp5],          16,   4    \n\t"
480     "ext            %[temp5],    %[temp3],          12,   4    \n\t"
481     "ins            %[temp3],    %[temp4],          0,    4    \n\t"
482     "precr.qb.ph    %[temp1],    %[temp1],          %[temp0]   \n\t"
483     "ins            %[temp3],    %[temp5],          16,   4    \n\t"
484     "addiu          %[src],      %[src],            16         \n\t"
485     "precr.qb.ph    %[temp3],    %[temp3],          %[temp2]   \n\t"
486 #ifdef WEBP_SWAP_16BIT_CSP
487     "usw            %[temp1],    0(%[dst])                     \n\t"
488     "usw            %[temp3],    4(%[dst])                     \n\t"
489 #else
490     "wsbh           %[temp1],    %[temp1]                      \n\t"
491     "wsbh           %[temp3],    %[temp3]                      \n\t"
492     "usw            %[temp1],    0(%[dst])                     \n\t"
493     "usw            %[temp3],    4(%[dst])                     \n\t"
494 #endif
495     "bne            %[src],      %[p_loop1_end],    0b         \n\t"
496     " addiu         %[dst],      %[dst],            8          \n\t"
497   "3:                                                          \n\t"
498     "beq            %[src],      %[p_loop2_end],    2f         \n\t"
499     " nop                                                      \n\t"
500   "1:                                                          \n\t"
501     "lw             %[temp0],    0(%[src])                     \n\t"
502     "ext            %[temp4],    %[temp0],          28,   4    \n\t"
503     "ext            %[temp5],    %[temp0],          12,   4    \n\t"
504     "ins            %[temp0],    %[temp4],          0,    4    \n\t"
505     "ins            %[temp0],    %[temp5],          16,   4    \n\t"
506     "addiu          %[src],      %[src],            4          \n\t"
507     "precr.qb.ph    %[temp0],    %[temp0],          %[temp0]   \n\t"
508 #ifdef WEBP_SWAP_16BIT_CSP
509     "ush            %[temp0],    0(%[dst])                     \n\t"
510 #else
511     "wsbh           %[temp0],    %[temp0]                      \n\t"
512     "ush            %[temp0],    0(%[dst])                     \n\t"
513 #endif
514     "bne            %[src],      %[p_loop2_end],    1b         \n\t"
515     " addiu         %[dst],      %[dst],            2          \n\t"
516   "2:                                                          \n\t"
517     ".set           pop                                        \n\t"
518     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
519       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
520       [dst]"+&r"(dst), [src]"+&r"(src)
521     : [p_loop1_end]"r"(p_loop1_end), [p_loop2_end]"r"(p_loop2_end)
522     : "memory"
523   );
524 }
525 
ConvertBGRAToRGB565(const uint32_t * src,int num_pixels,uint8_t * dst)526 static void ConvertBGRAToRGB565(const uint32_t* src,
527                                 int num_pixels, uint8_t* dst) {
528   int temp0, temp1, temp2, temp3, temp4, temp5;
529   const uint32_t* const p_loop1_end = src + (num_pixels & ~3);
530   const uint32_t* const p_loop2_end = src + num_pixels;
531   __asm__ volatile (
532     ".set           push                                       \n\t"
533     ".set           noreorder                                  \n\t"
534     "beq            %[src],      %[p_loop1_end],    3f         \n\t"
535     " nop                                                      \n\t"
536   "0:                                                          \n\t"
537     "lw             %[temp0],    0(%[src])                     \n\t"
538     "lw             %[temp1],    4(%[src])                     \n\t"
539     "lw             %[temp2],    8(%[src])                     \n\t"
540     "lw             %[temp3],    12(%[src])                    \n\t"
541     "ext            %[temp4],    %[temp0],          8,    16   \n\t"
542     "ext            %[temp5],    %[temp0],          5,    11   \n\t"
543     "ext            %[temp0],    %[temp0],          3,    5    \n\t"
544     "ins            %[temp4],    %[temp5],          0,    11   \n\t"
545     "ext            %[temp5],    %[temp1],          5,    11   \n\t"
546     "ins            %[temp4],    %[temp0],          0,    5    \n\t"
547     "ext            %[temp0],    %[temp1],          8,    16   \n\t"
548     "ext            %[temp1],    %[temp1],          3,    5    \n\t"
549     "ins            %[temp0],    %[temp5],          0,    11   \n\t"
550     "ext            %[temp5],    %[temp2],          5,    11   \n\t"
551     "ins            %[temp0],    %[temp1],          0,    5    \n\t"
552     "ext            %[temp1],    %[temp2],          8,    16   \n\t"
553     "ext            %[temp2],    %[temp2],          3,    5    \n\t"
554     "ins            %[temp1],    %[temp5],          0,    11   \n\t"
555     "ext            %[temp5],    %[temp3],          5,    11   \n\t"
556     "ins            %[temp1],    %[temp2],          0,    5    \n\t"
557     "ext            %[temp2],    %[temp3],          8,    16   \n\t"
558     "ext            %[temp3],    %[temp3],          3,    5    \n\t"
559     "ins            %[temp2],    %[temp5],          0,    11   \n\t"
560     "append         %[temp0],    %[temp4],          16         \n\t"
561     "ins            %[temp2],    %[temp3],          0,    5    \n\t"
562     "addiu          %[src],      %[src],            16         \n\t"
563     "append         %[temp2],    %[temp1],          16         \n\t"
564 #ifdef WEBP_SWAP_16BIT_CSP
565     "usw            %[temp0],    0(%[dst])                     \n\t"
566     "usw            %[temp2],    4(%[dst])                     \n\t"
567 #else
568     "wsbh           %[temp0],    %[temp0]                      \n\t"
569     "wsbh           %[temp2],    %[temp2]                      \n\t"
570     "usw            %[temp0],    0(%[dst])                     \n\t"
571     "usw            %[temp2],    4(%[dst])                     \n\t"
572 #endif
573     "bne            %[src],      %[p_loop1_end],    0b         \n\t"
574     " addiu         %[dst],      %[dst],            8          \n\t"
575   "3:                                                          \n\t"
576     "beq            %[src],      %[p_loop2_end],    2f         \n\t"
577     " nop                                                      \n\t"
578   "1:                                                          \n\t"
579     "lw             %[temp0],    0(%[src])                     \n\t"
580     "ext            %[temp4],    %[temp0],          8,    16   \n\t"
581     "ext            %[temp5],    %[temp0],          5,    11   \n\t"
582     "ext            %[temp0],    %[temp0],          3,    5    \n\t"
583     "ins            %[temp4],    %[temp5],          0,    11   \n\t"
584     "addiu          %[src],      %[src],            4          \n\t"
585     "ins            %[temp4],    %[temp0],          0,    5    \n\t"
586 #ifdef WEBP_SWAP_16BIT_CSP
587     "ush            %[temp4],    0(%[dst])                     \n\t"
588 #else
589     "wsbh           %[temp4],    %[temp4]                      \n\t"
590     "ush            %[temp4],    0(%[dst])                     \n\t"
591 #endif
592     "bne            %[src],      %[p_loop2_end],    1b         \n\t"
593     " addiu         %[dst],      %[dst],            2          \n\t"
594   "2:                                                          \n\t"
595     ".set           pop                                        \n\t"
596     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
597       [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
598       [dst]"+&r"(dst), [src]"+&r"(src)
599     : [p_loop1_end]"r"(p_loop1_end), [p_loop2_end]"r"(p_loop2_end)
600     : "memory"
601   );
602 }
603 
ConvertBGRAToBGR(const uint32_t * src,int num_pixels,uint8_t * dst)604 static void ConvertBGRAToBGR(const uint32_t* src,
605                              int num_pixels, uint8_t* dst) {
606   int temp0, temp1, temp2, temp3;
607   const uint32_t* const p_loop1_end = src + (num_pixels & ~3);
608   const uint32_t* const p_loop2_end = src + num_pixels;
609   __asm__ volatile (
610     ".set       push                                         \n\t"
611     ".set       noreorder                                    \n\t"
612     "beq        %[src],      %[p_loop1_end],    3f           \n\t"
613     " nop                                                    \n\t"
614   "0:                                                        \n\t"
615     "lw         %[temp0],    0(%[src])                       \n\t"
616     "lw         %[temp1],    4(%[src])                       \n\t"
617     "lw         %[temp2],    8(%[src])                       \n\t"
618     "lw         %[temp3],    12(%[src])                      \n\t"
619     "ins        %[temp0],    %[temp1],          24,    8     \n\t"
620     "sra        %[temp1],    %[temp1],          8            \n\t"
621     "ins        %[temp1],    %[temp2],          16,    16    \n\t"
622     "sll        %[temp2],    %[temp2],          8            \n\t"
623     "balign     %[temp3],    %[temp2],          1            \n\t"
624     "addiu      %[src],      %[src],            16           \n\t"
625     "usw        %[temp0],    0(%[dst])                       \n\t"
626     "usw        %[temp1],    4(%[dst])                       \n\t"
627     "usw        %[temp3],    8(%[dst])                       \n\t"
628     "bne        %[src],      %[p_loop1_end],    0b           \n\t"
629     " addiu     %[dst],      %[dst],            12           \n\t"
630   "3:                                                        \n\t"
631     "beq        %[src],      %[p_loop2_end],    2f           \n\t"
632     " nop                                                    \n\t"
633   "1:                                                        \n\t"
634     "lw         %[temp0],    0(%[src])                       \n\t"
635     "addiu      %[src],      %[src],            4            \n\t"
636     "addiu      %[dst],      %[dst],            3            \n\t"
637     "ush        %[temp0],    -3(%[dst])                      \n\t"
638     "sra        %[temp0],    %[temp0],          16           \n\t"
639     "bne        %[src],      %[p_loop2_end],    1b           \n\t"
640     " sb        %[temp0],    -1(%[dst])                      \n\t"
641   "2:                                                        \n\t"
642     ".set       pop                                          \n\t"
643     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
644       [temp3]"=&r"(temp3), [dst]"+&r"(dst), [src]"+&r"(src)
645     : [p_loop1_end]"r"(p_loop1_end), [p_loop2_end]"r"(p_loop2_end)
646     : "memory"
647   );
648 }
649 
650 //------------------------------------------------------------------------------
651 // Entry point
652 
653 extern void VP8LDspInitMIPSdspR2(void);
654 
VP8LDspInitMIPSdspR2(void)655 WEBP_TSAN_IGNORE_FUNCTION void VP8LDspInitMIPSdspR2(void) {
656   VP8LMapColor32b = MapARGB;
657   VP8LMapColor8b = MapAlpha;
658   VP8LPredictors[5] = Predictor5;
659   VP8LPredictors[6] = Predictor6;
660   VP8LPredictors[7] = Predictor7;
661   VP8LPredictors[8] = Predictor8;
662   VP8LPredictors[9] = Predictor9;
663   VP8LPredictors[10] = Predictor10;
664   VP8LPredictors[11] = Predictor11;
665   VP8LPredictors[12] = Predictor12;
666   VP8LPredictors[13] = Predictor13;
667   VP8LAddGreenToBlueAndRed = AddGreenToBlueAndRed;
668   VP8LTransformColorInverse = TransformColorInverse;
669   VP8LConvertBGRAToRGB = ConvertBGRAToRGB;
670   VP8LConvertBGRAToRGBA = ConvertBGRAToRGBA;
671   VP8LConvertBGRAToRGBA4444 = ConvertBGRAToRGBA4444;
672   VP8LConvertBGRAToRGB565 = ConvertBGRAToRGB565;
673   VP8LConvertBGRAToBGR = ConvertBGRAToBGR;
674 }
675 
676 #else  // !WEBP_USE_MIPS_DSP_R2
677 
678 WEBP_DSP_INIT_STUB(VP8LDspInitMIPSdspR2)
679 
680 #endif  // WEBP_USE_MIPS_DSP_R2
681