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 #include "./lossless_common.h"
21
22 #define MAP_COLOR_FUNCS(FUNC_NAME, TYPE, GET_INDEX, GET_VALUE) \
23 static void FUNC_NAME(const TYPE* src, \
24 const uint32_t* const color_map, \
25 TYPE* dst, int y_start, int y_end, \
26 int width) { \
27 int y; \
28 for (y = y_start; y < y_end; ++y) { \
29 int x; \
30 for (x = 0; x < (width >> 2); ++x) { \
31 int tmp1, tmp2, tmp3, tmp4; \
32 __asm__ volatile ( \
33 ".ifc " #TYPE ", uint8_t \n\t" \
34 "lbu %[tmp1], 0(%[src]) \n\t" \
35 "lbu %[tmp2], 1(%[src]) \n\t" \
36 "lbu %[tmp3], 2(%[src]) \n\t" \
37 "lbu %[tmp4], 3(%[src]) \n\t" \
38 "addiu %[src], %[src], 4 \n\t" \
39 ".endif \n\t" \
40 ".ifc " #TYPE ", uint32_t \n\t" \
41 "lw %[tmp1], 0(%[src]) \n\t" \
42 "lw %[tmp2], 4(%[src]) \n\t" \
43 "lw %[tmp3], 8(%[src]) \n\t" \
44 "lw %[tmp4], 12(%[src]) \n\t" \
45 "ext %[tmp1], %[tmp1], 8, 8 \n\t" \
46 "ext %[tmp2], %[tmp2], 8, 8 \n\t" \
47 "ext %[tmp3], %[tmp3], 8, 8 \n\t" \
48 "ext %[tmp4], %[tmp4], 8, 8 \n\t" \
49 "addiu %[src], %[src], 16 \n\t" \
50 ".endif \n\t" \
51 "sll %[tmp1], %[tmp1], 2 \n\t" \
52 "sll %[tmp2], %[tmp2], 2 \n\t" \
53 "sll %[tmp3], %[tmp3], 2 \n\t" \
54 "sll %[tmp4], %[tmp4], 2 \n\t" \
55 "lwx %[tmp1], %[tmp1](%[color_map]) \n\t" \
56 "lwx %[tmp2], %[tmp2](%[color_map]) \n\t" \
57 "lwx %[tmp3], %[tmp3](%[color_map]) \n\t" \
58 "lwx %[tmp4], %[tmp4](%[color_map]) \n\t" \
59 ".ifc " #TYPE ", uint8_t \n\t" \
60 "ext %[tmp1], %[tmp1], 8, 8 \n\t" \
61 "ext %[tmp2], %[tmp2], 8, 8 \n\t" \
62 "ext %[tmp3], %[tmp3], 8, 8 \n\t" \
63 "ext %[tmp4], %[tmp4], 8, 8 \n\t" \
64 "sb %[tmp1], 0(%[dst]) \n\t" \
65 "sb %[tmp2], 1(%[dst]) \n\t" \
66 "sb %[tmp3], 2(%[dst]) \n\t" \
67 "sb %[tmp4], 3(%[dst]) \n\t" \
68 "addiu %[dst], %[dst], 4 \n\t" \
69 ".endif \n\t" \
70 ".ifc " #TYPE ", uint32_t \n\t" \
71 "sw %[tmp1], 0(%[dst]) \n\t" \
72 "sw %[tmp2], 4(%[dst]) \n\t" \
73 "sw %[tmp3], 8(%[dst]) \n\t" \
74 "sw %[tmp4], 12(%[dst]) \n\t" \
75 "addiu %[dst], %[dst], 16 \n\t" \
76 ".endif \n\t" \
77 : [tmp1]"=&r"(tmp1), [tmp2]"=&r"(tmp2), [tmp3]"=&r"(tmp3), \
78 [tmp4]"=&r"(tmp4), [src]"+&r"(src), [dst]"+r"(dst) \
79 : [color_map]"r"(color_map) \
80 : "memory" \
81 ); \
82 } \
83 for (x = 0; x < (width & 3); ++x) { \
84 *dst++ = GET_VALUE(color_map[GET_INDEX(*src++)]); \
85 } \
86 } \
87 }
88
MAP_COLOR_FUNCS(MapARGB,uint32_t,VP8GetARGBIndex,VP8GetARGBValue)89 MAP_COLOR_FUNCS(MapARGB, uint32_t, VP8GetARGBIndex, VP8GetARGBValue)
90 MAP_COLOR_FUNCS(MapAlpha, uint8_t, VP8GetAlphaIndex, VP8GetAlphaValue)
91
92 #undef MAP_COLOR_FUNCS
93
94 static WEBP_INLINE uint32_t ClampedAddSubtractFull(uint32_t c0, uint32_t c1,
95 uint32_t c2) {
96 int temp0, temp1, temp2, temp3, temp4, temp5;
97 __asm__ volatile (
98 "preceu.ph.qbr %[temp1], %[c0] \n\t"
99 "preceu.ph.qbl %[temp2], %[c0] \n\t"
100 "preceu.ph.qbr %[temp3], %[c1] \n\t"
101 "preceu.ph.qbl %[temp4], %[c1] \n\t"
102 "preceu.ph.qbr %[temp5], %[c2] \n\t"
103 "preceu.ph.qbl %[temp0], %[c2] \n\t"
104 "subq.ph %[temp3], %[temp3], %[temp5] \n\t"
105 "subq.ph %[temp4], %[temp4], %[temp0] \n\t"
106 "addq.ph %[temp1], %[temp1], %[temp3] \n\t"
107 "addq.ph %[temp2], %[temp2], %[temp4] \n\t"
108 "shll_s.ph %[temp1], %[temp1], 7 \n\t"
109 "shll_s.ph %[temp2], %[temp2], 7 \n\t"
110 "precrqu_s.qb.ph %[temp2], %[temp2], %[temp1] \n\t"
111 : [temp0]"=r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
112 [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5)
113 : [c0]"r"(c0), [c1]"r"(c1), [c2]"r"(c2)
114 : "memory"
115 );
116 return temp2;
117 }
118
ClampedAddSubtractHalf(uint32_t c0,uint32_t c1,uint32_t c2)119 static WEBP_INLINE uint32_t ClampedAddSubtractHalf(uint32_t c0, uint32_t c1,
120 uint32_t c2) {
121 int temp0, temp1, temp2, temp3, temp4, temp5;
122 __asm__ volatile (
123 "adduh.qb %[temp5], %[c0], %[c1] \n\t"
124 "preceu.ph.qbr %[temp3], %[c2] \n\t"
125 "preceu.ph.qbr %[temp1], %[temp5] \n\t"
126 "preceu.ph.qbl %[temp2], %[temp5] \n\t"
127 "preceu.ph.qbl %[temp4], %[c2] \n\t"
128 "subq.ph %[temp3], %[temp1], %[temp3] \n\t"
129 "subq.ph %[temp4], %[temp2], %[temp4] \n\t"
130 "shrl.ph %[temp5], %[temp3], 15 \n\t"
131 "shrl.ph %[temp0], %[temp4], 15 \n\t"
132 "addq.ph %[temp3], %[temp3], %[temp5] \n\t"
133 "addq.ph %[temp4], %[temp0], %[temp4] \n\t"
134 "shra.ph %[temp3], %[temp3], 1 \n\t"
135 "shra.ph %[temp4], %[temp4], 1 \n\t"
136 "addq.ph %[temp1], %[temp1], %[temp3] \n\t"
137 "addq.ph %[temp2], %[temp2], %[temp4] \n\t"
138 "shll_s.ph %[temp1], %[temp1], 7 \n\t"
139 "shll_s.ph %[temp2], %[temp2], 7 \n\t"
140 "precrqu_s.qb.ph %[temp1], %[temp2], %[temp1] \n\t"
141 : [temp0]"=r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
142 [temp3]"=&r"(temp3), [temp4]"=r"(temp4), [temp5]"=&r"(temp5)
143 : [c0]"r"(c0), [c1]"r"(c1), [c2]"r"(c2)
144 : "memory"
145 );
146 return temp1;
147 }
148
Select(uint32_t a,uint32_t b,uint32_t c)149 static WEBP_INLINE uint32_t Select(uint32_t a, uint32_t b, uint32_t c) {
150 int temp0, temp1, temp2, temp3, temp4, temp5;
151 __asm__ volatile (
152 "cmpgdu.lt.qb %[temp1], %[c], %[b] \n\t"
153 "pick.qb %[temp1], %[b], %[c] \n\t"
154 "pick.qb %[temp2], %[c], %[b] \n\t"
155 "cmpgdu.lt.qb %[temp4], %[c], %[a] \n\t"
156 "pick.qb %[temp4], %[a], %[c] \n\t"
157 "pick.qb %[temp5], %[c], %[a] \n\t"
158 "subu.qb %[temp3], %[temp1], %[temp2] \n\t"
159 "subu.qb %[temp0], %[temp4], %[temp5] \n\t"
160 "raddu.w.qb %[temp3], %[temp3] \n\t"
161 "raddu.w.qb %[temp0], %[temp0] \n\t"
162 "subu %[temp3], %[temp3], %[temp0] \n\t"
163 "slti %[temp0], %[temp3], 0x1 \n\t"
164 "movz %[a], %[b], %[temp0] \n\t"
165 : [temp1]"=&r"(temp1), [temp2]"=&r"(temp2), [temp3]"=&r"(temp3),
166 [temp4]"=&r"(temp4), [temp5]"=&r"(temp5), [temp0]"=&r"(temp0),
167 [a]"+&r"(a)
168 : [b]"r"(b), [c]"r"(c)
169 );
170 return a;
171 }
172
Average2(uint32_t a0,uint32_t a1)173 static WEBP_INLINE uint32_t Average2(uint32_t a0, uint32_t a1) {
174 __asm__ volatile (
175 "adduh.qb %[a0], %[a0], %[a1] \n\t"
176 : [a0]"+r"(a0)
177 : [a1]"r"(a1)
178 );
179 return a0;
180 }
181
Average3(uint32_t a0,uint32_t a1,uint32_t a2)182 static WEBP_INLINE uint32_t Average3(uint32_t a0, uint32_t a1, uint32_t a2) {
183 return Average2(Average2(a0, a2), a1);
184 }
185
Average4(uint32_t a0,uint32_t a1,uint32_t a2,uint32_t a3)186 static WEBP_INLINE uint32_t Average4(uint32_t a0, uint32_t a1,
187 uint32_t a2, uint32_t a3) {
188 return Average2(Average2(a0, a1), Average2(a2, a3));
189 }
190
Predictor5(uint32_t left,const uint32_t * const top)191 static uint32_t Predictor5(uint32_t left, const uint32_t* const top) {
192 return Average3(left, top[0], top[1]);
193 }
194
Predictor6(uint32_t left,const uint32_t * const top)195 static uint32_t Predictor6(uint32_t left, const uint32_t* const top) {
196 return Average2(left, top[-1]);
197 }
198
Predictor7(uint32_t left,const uint32_t * const top)199 static uint32_t Predictor7(uint32_t left, const uint32_t* const top) {
200 return Average2(left, top[0]);
201 }
202
Predictor8(uint32_t left,const uint32_t * const top)203 static uint32_t Predictor8(uint32_t left, const uint32_t* const top) {
204 (void)left;
205 return Average2(top[-1], top[0]);
206 }
207
Predictor9(uint32_t left,const uint32_t * const top)208 static uint32_t Predictor9(uint32_t left, const uint32_t* const top) {
209 (void)left;
210 return Average2(top[0], top[1]);
211 }
212
Predictor10(uint32_t left,const uint32_t * const top)213 static uint32_t Predictor10(uint32_t left, const uint32_t* const top) {
214 return Average4(left, top[-1], top[0], top[1]);
215 }
216
Predictor11(uint32_t left,const uint32_t * const top)217 static uint32_t Predictor11(uint32_t left, const uint32_t* const top) {
218 return Select(top[0], left, top[-1]);
219 }
220
Predictor12(uint32_t left,const uint32_t * const top)221 static uint32_t Predictor12(uint32_t left, const uint32_t* const top) {
222 return ClampedAddSubtractFull(left, top[0], top[-1]);
223 }
224
Predictor13(uint32_t left,const uint32_t * const top)225 static uint32_t Predictor13(uint32_t left, const uint32_t* const top) {
226 return ClampedAddSubtractHalf(left, top[0], top[-1]);
227 }
228
229 // Add green to blue and red channels (i.e. perform the inverse transform of
230 // 'subtract green').
AddGreenToBlueAndRed(const uint32_t * src,int num_pixels,uint32_t * dst)231 static void AddGreenToBlueAndRed(const uint32_t* src, int num_pixels,
232 uint32_t* dst) {
233 uint32_t temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
234 const uint32_t* const p_loop1_end = src + (num_pixels & ~3);
235 const uint32_t* const p_loop2_end = src + num_pixels;
236 __asm__ volatile (
237 ".set push \n\t"
238 ".set noreorder \n\t"
239 "beq %[src], %[p_loop1_end], 3f \n\t"
240 " nop \n\t"
241 "0: \n\t"
242 "lw %[temp0], 0(%[src]) \n\t"
243 "lw %[temp1], 4(%[src]) \n\t"
244 "lw %[temp2], 8(%[src]) \n\t"
245 "lw %[temp3], 12(%[src]) \n\t"
246 "ext %[temp4], %[temp0], 8, 8 \n\t"
247 "ext %[temp5], %[temp1], 8, 8 \n\t"
248 "ext %[temp6], %[temp2], 8, 8 \n\t"
249 "ext %[temp7], %[temp3], 8, 8 \n\t"
250 "addiu %[src], %[src], 16 \n\t"
251 "addiu %[dst], %[dst], 16 \n\t"
252 "replv.ph %[temp4], %[temp4] \n\t"
253 "replv.ph %[temp5], %[temp5] \n\t"
254 "replv.ph %[temp6], %[temp6] \n\t"
255 "replv.ph %[temp7], %[temp7] \n\t"
256 "addu.qb %[temp0], %[temp0], %[temp4] \n\t"
257 "addu.qb %[temp1], %[temp1], %[temp5] \n\t"
258 "addu.qb %[temp2], %[temp2], %[temp6] \n\t"
259 "addu.qb %[temp3], %[temp3], %[temp7] \n\t"
260 "sw %[temp0], -16(%[dst]) \n\t"
261 "sw %[temp1], -12(%[dst]) \n\t"
262 "sw %[temp2], -8(%[dst]) \n\t"
263 "bne %[src], %[p_loop1_end], 0b \n\t"
264 " sw %[temp3], -4(%[dst]) \n\t"
265 "3: \n\t"
266 "beq %[src], %[p_loop2_end], 2f \n\t"
267 " nop \n\t"
268 "1: \n\t"
269 "lw %[temp0], 0(%[src]) \n\t"
270 "addiu %[src], %[src], 4 \n\t"
271 "addiu %[dst], %[dst], 4 \n\t"
272 "ext %[temp4], %[temp0], 8, 8 \n\t"
273 "replv.ph %[temp4], %[temp4] \n\t"
274 "addu.qb %[temp0], %[temp0], %[temp4] \n\t"
275 "bne %[src], %[p_loop2_end], 1b \n\t"
276 " sw %[temp0], -4(%[dst]) \n\t"
277 "2: \n\t"
278 ".set pop \n\t"
279 : [dst]"+&r"(dst), [src]"+&r"(src), [temp0]"=&r"(temp0),
280 [temp1]"=&r"(temp1), [temp2]"=&r"(temp2), [temp3]"=&r"(temp3),
281 [temp4]"=&r"(temp4), [temp5]"=&r"(temp5), [temp6]"=&r"(temp6),
282 [temp7]"=&r"(temp7)
283 : [p_loop1_end]"r"(p_loop1_end), [p_loop2_end]"r"(p_loop2_end)
284 : "memory"
285 );
286 }
287
TransformColorInverse(const VP8LMultipliers * const m,const uint32_t * src,int num_pixels,uint32_t * dst)288 static void TransformColorInverse(const VP8LMultipliers* const m,
289 const uint32_t* src, int num_pixels,
290 uint32_t* dst) {
291 int temp0, temp1, temp2, temp3, temp4, temp5;
292 uint32_t argb, argb1, new_red;
293 const uint32_t G_to_R = m->green_to_red_;
294 const uint32_t G_to_B = m->green_to_blue_;
295 const uint32_t R_to_B = m->red_to_blue_;
296 const uint32_t* const p_loop_end = src + (num_pixels & ~1);
297 __asm__ volatile (
298 ".set push \n\t"
299 ".set noreorder \n\t"
300 "beq %[src], %[p_loop_end], 1f \n\t"
301 " nop \n\t"
302 "replv.ph %[temp0], %[G_to_R] \n\t"
303 "replv.ph %[temp1], %[G_to_B] \n\t"
304 "replv.ph %[temp2], %[R_to_B] \n\t"
305 "shll.ph %[temp0], %[temp0], 8 \n\t"
306 "shll.ph %[temp1], %[temp1], 8 \n\t"
307 "shll.ph %[temp2], %[temp2], 8 \n\t"
308 "shra.ph %[temp0], %[temp0], 8 \n\t"
309 "shra.ph %[temp1], %[temp1], 8 \n\t"
310 "shra.ph %[temp2], %[temp2], 8 \n\t"
311 "0: \n\t"
312 "lw %[argb], 0(%[src]) \n\t"
313 "lw %[argb1], 4(%[src]) \n\t"
314 "sw %[argb], 0(%[dst]) \n\t"
315 "sw %[argb1], 4(%[dst]) \n\t"
316 "addiu %[src], %[src], 8 \n\t"
317 "addiu %[dst], %[dst], 8 \n\t"
318 "precrq.qb.ph %[temp3], %[argb], %[argb1] \n\t"
319 "preceu.ph.qbra %[temp3], %[temp3] \n\t"
320 "shll.ph %[temp3], %[temp3], 8 \n\t"
321 "shra.ph %[temp3], %[temp3], 8 \n\t"
322 "mul.ph %[temp5], %[temp3], %[temp0] \n\t"
323 "mul.ph %[temp3], %[temp3], %[temp1] \n\t"
324 "precrq.ph.w %[new_red], %[argb], %[argb1] \n\t"
325 "ins %[argb1], %[argb], 16, 16 \n\t"
326 "shra.ph %[temp5], %[temp5], 5 \n\t"
327 "shra.ph %[temp3], %[temp3], 5 \n\t"
328 "addu.ph %[new_red], %[new_red], %[temp5] \n\t"
329 "addu.ph %[argb1], %[argb1], %[temp3] \n\t"
330 "preceu.ph.qbra %[temp5], %[new_red] \n\t"
331 "shll.ph %[temp4], %[temp5], 8 \n\t"
332 "shra.ph %[temp4], %[temp4], 8 \n\t"
333 "mul.ph %[temp4], %[temp4], %[temp2] \n\t"
334 "sb %[temp5], -2(%[dst]) \n\t"
335 "sra %[temp5], %[temp5], 16 \n\t"
336 "shra.ph %[temp4], %[temp4], 5 \n\t"
337 "addu.ph %[argb1], %[argb1], %[temp4] \n\t"
338 "preceu.ph.qbra %[temp3], %[argb1] \n\t"
339 "sb %[temp5], -6(%[dst]) \n\t"
340 "sb %[temp3], -4(%[dst]) \n\t"
341 "sra %[temp3], %[temp3], 16 \n\t"
342 "bne %[src], %[p_loop_end], 0b \n\t"
343 " sb %[temp3], -8(%[dst]) \n\t"
344 "1: \n\t"
345 ".set pop \n\t"
346 : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
347 [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
348 [new_red]"=&r"(new_red), [argb]"=&r"(argb),
349 [argb1]"=&r"(argb1), [dst]"+&r"(dst), [src]"+&r"(src)
350 : [G_to_R]"r"(G_to_R), [R_to_B]"r"(R_to_B),
351 [G_to_B]"r"(G_to_B), [p_loop_end]"r"(p_loop_end)
352 : "memory", "hi", "lo"
353 );
354
355 // Fall-back to C-version for left-overs.
356 if (num_pixels & 1) VP8LTransformColorInverse_C(m, src, 1, dst);
357 }
358
ConvertBGRAToRGB(const uint32_t * src,int num_pixels,uint8_t * dst)359 static void ConvertBGRAToRGB(const uint32_t* src,
360 int num_pixels, uint8_t* dst) {
361 int temp0, temp1, temp2, temp3;
362 const uint32_t* const p_loop1_end = src + (num_pixels & ~3);
363 const uint32_t* const p_loop2_end = src + num_pixels;
364 __asm__ volatile (
365 ".set push \n\t"
366 ".set noreorder \n\t"
367 "beq %[src], %[p_loop1_end], 3f \n\t"
368 " nop \n\t"
369 "0: \n\t"
370 "lw %[temp3], 12(%[src]) \n\t"
371 "lw %[temp2], 8(%[src]) \n\t"
372 "lw %[temp1], 4(%[src]) \n\t"
373 "lw %[temp0], 0(%[src]) \n\t"
374 "ins %[temp3], %[temp2], 24, 8 \n\t"
375 "sll %[temp2], %[temp2], 8 \n\t"
376 "rotr %[temp3], %[temp3], 16 \n\t"
377 "ins %[temp2], %[temp1], 0, 16 \n\t"
378 "sll %[temp1], %[temp1], 8 \n\t"
379 "wsbh %[temp3], %[temp3] \n\t"
380 "balign %[temp0], %[temp1], 1 \n\t"
381 "wsbh %[temp2], %[temp2] \n\t"
382 "wsbh %[temp0], %[temp0] \n\t"
383 "usw %[temp3], 8(%[dst]) \n\t"
384 "rotr %[temp0], %[temp0], 16 \n\t"
385 "usw %[temp2], 4(%[dst]) \n\t"
386 "addiu %[src], %[src], 16 \n\t"
387 "usw %[temp0], 0(%[dst]) \n\t"
388 "bne %[src], %[p_loop1_end], 0b \n\t"
389 " addiu %[dst], %[dst], 12 \n\t"
390 "3: \n\t"
391 "beq %[src], %[p_loop2_end], 2f \n\t"
392 " nop \n\t"
393 "1: \n\t"
394 "lw %[temp0], 0(%[src]) \n\t"
395 "addiu %[src], %[src], 4 \n\t"
396 "wsbh %[temp1], %[temp0] \n\t"
397 "addiu %[dst], %[dst], 3 \n\t"
398 "ush %[temp1], -2(%[dst]) \n\t"
399 "sra %[temp0], %[temp0], 16 \n\t"
400 "bne %[src], %[p_loop2_end], 1b \n\t"
401 " sb %[temp0], -3(%[dst]) \n\t"
402 "2: \n\t"
403 ".set pop \n\t"
404 : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
405 [temp3]"=&r"(temp3), [dst]"+&r"(dst), [src]"+&r"(src)
406 : [p_loop1_end]"r"(p_loop1_end), [p_loop2_end]"r"(p_loop2_end)
407 : "memory"
408 );
409 }
410
ConvertBGRAToRGBA(const uint32_t * src,int num_pixels,uint8_t * dst)411 static void ConvertBGRAToRGBA(const uint32_t* src,
412 int num_pixels, uint8_t* dst) {
413 int temp0, temp1, temp2, temp3;
414 const uint32_t* const p_loop1_end = src + (num_pixels & ~3);
415 const uint32_t* const p_loop2_end = src + num_pixels;
416 __asm__ volatile (
417 ".set push \n\t"
418 ".set noreorder \n\t"
419 "beq %[src], %[p_loop1_end], 3f \n\t"
420 " nop \n\t"
421 "0: \n\t"
422 "lw %[temp0], 0(%[src]) \n\t"
423 "lw %[temp1], 4(%[src]) \n\t"
424 "lw %[temp2], 8(%[src]) \n\t"
425 "lw %[temp3], 12(%[src]) \n\t"
426 "wsbh %[temp0], %[temp0] \n\t"
427 "wsbh %[temp1], %[temp1] \n\t"
428 "wsbh %[temp2], %[temp2] \n\t"
429 "wsbh %[temp3], %[temp3] \n\t"
430 "addiu %[src], %[src], 16 \n\t"
431 "balign %[temp0], %[temp0], 1 \n\t"
432 "balign %[temp1], %[temp1], 1 \n\t"
433 "balign %[temp2], %[temp2], 1 \n\t"
434 "balign %[temp3], %[temp3], 1 \n\t"
435 "usw %[temp0], 0(%[dst]) \n\t"
436 "usw %[temp1], 4(%[dst]) \n\t"
437 "usw %[temp2], 8(%[dst]) \n\t"
438 "usw %[temp3], 12(%[dst]) \n\t"
439 "bne %[src], %[p_loop1_end], 0b \n\t"
440 " addiu %[dst], %[dst], 16 \n\t"
441 "3: \n\t"
442 "beq %[src], %[p_loop2_end], 2f \n\t"
443 " nop \n\t"
444 "1: \n\t"
445 "lw %[temp0], 0(%[src]) \n\t"
446 "wsbh %[temp0], %[temp0] \n\t"
447 "addiu %[src], %[src], 4 \n\t"
448 "balign %[temp0], %[temp0], 1 \n\t"
449 "usw %[temp0], 0(%[dst]) \n\t"
450 "bne %[src], %[p_loop2_end], 1b \n\t"
451 " addiu %[dst], %[dst], 4 \n\t"
452 "2: \n\t"
453 ".set pop \n\t"
454 : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
455 [temp3]"=&r"(temp3), [dst]"+&r"(dst), [src]"+&r"(src)
456 : [p_loop1_end]"r"(p_loop1_end), [p_loop2_end]"r"(p_loop2_end)
457 : "memory"
458 );
459 }
460
ConvertBGRAToRGBA4444(const uint32_t * src,int num_pixels,uint8_t * dst)461 static void ConvertBGRAToRGBA4444(const uint32_t* src,
462 int num_pixels, uint8_t* dst) {
463 int temp0, temp1, temp2, temp3, temp4, temp5;
464 const uint32_t* const p_loop1_end = src + (num_pixels & ~3);
465 const uint32_t* const p_loop2_end = src + num_pixels;
466 __asm__ volatile (
467 ".set push \n\t"
468 ".set noreorder \n\t"
469 "beq %[src], %[p_loop1_end], 3f \n\t"
470 " nop \n\t"
471 "0: \n\t"
472 "lw %[temp0], 0(%[src]) \n\t"
473 "lw %[temp1], 4(%[src]) \n\t"
474 "lw %[temp2], 8(%[src]) \n\t"
475 "lw %[temp3], 12(%[src]) \n\t"
476 "ext %[temp4], %[temp0], 28, 4 \n\t"
477 "ext %[temp5], %[temp0], 12, 4 \n\t"
478 "ins %[temp0], %[temp4], 0, 4 \n\t"
479 "ext %[temp4], %[temp1], 28, 4 \n\t"
480 "ins %[temp0], %[temp5], 16, 4 \n\t"
481 "ext %[temp5], %[temp1], 12, 4 \n\t"
482 "ins %[temp1], %[temp4], 0, 4 \n\t"
483 "ext %[temp4], %[temp2], 28, 4 \n\t"
484 "ins %[temp1], %[temp5], 16, 4 \n\t"
485 "ext %[temp5], %[temp2], 12, 4 \n\t"
486 "ins %[temp2], %[temp4], 0, 4 \n\t"
487 "ext %[temp4], %[temp3], 28, 4 \n\t"
488 "ins %[temp2], %[temp5], 16, 4 \n\t"
489 "ext %[temp5], %[temp3], 12, 4 \n\t"
490 "ins %[temp3], %[temp4], 0, 4 \n\t"
491 "precr.qb.ph %[temp1], %[temp1], %[temp0] \n\t"
492 "ins %[temp3], %[temp5], 16, 4 \n\t"
493 "addiu %[src], %[src], 16 \n\t"
494 "precr.qb.ph %[temp3], %[temp3], %[temp2] \n\t"
495 #ifdef WEBP_SWAP_16BIT_CSP
496 "usw %[temp1], 0(%[dst]) \n\t"
497 "usw %[temp3], 4(%[dst]) \n\t"
498 #else
499 "wsbh %[temp1], %[temp1] \n\t"
500 "wsbh %[temp3], %[temp3] \n\t"
501 "usw %[temp1], 0(%[dst]) \n\t"
502 "usw %[temp3], 4(%[dst]) \n\t"
503 #endif
504 "bne %[src], %[p_loop1_end], 0b \n\t"
505 " addiu %[dst], %[dst], 8 \n\t"
506 "3: \n\t"
507 "beq %[src], %[p_loop2_end], 2f \n\t"
508 " nop \n\t"
509 "1: \n\t"
510 "lw %[temp0], 0(%[src]) \n\t"
511 "ext %[temp4], %[temp0], 28, 4 \n\t"
512 "ext %[temp5], %[temp0], 12, 4 \n\t"
513 "ins %[temp0], %[temp4], 0, 4 \n\t"
514 "ins %[temp0], %[temp5], 16, 4 \n\t"
515 "addiu %[src], %[src], 4 \n\t"
516 "precr.qb.ph %[temp0], %[temp0], %[temp0] \n\t"
517 #ifdef WEBP_SWAP_16BIT_CSP
518 "ush %[temp0], 0(%[dst]) \n\t"
519 #else
520 "wsbh %[temp0], %[temp0] \n\t"
521 "ush %[temp0], 0(%[dst]) \n\t"
522 #endif
523 "bne %[src], %[p_loop2_end], 1b \n\t"
524 " addiu %[dst], %[dst], 2 \n\t"
525 "2: \n\t"
526 ".set pop \n\t"
527 : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
528 [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
529 [dst]"+&r"(dst), [src]"+&r"(src)
530 : [p_loop1_end]"r"(p_loop1_end), [p_loop2_end]"r"(p_loop2_end)
531 : "memory"
532 );
533 }
534
ConvertBGRAToRGB565(const uint32_t * src,int num_pixels,uint8_t * dst)535 static void ConvertBGRAToRGB565(const uint32_t* src,
536 int num_pixels, uint8_t* dst) {
537 int temp0, temp1, temp2, temp3, temp4, temp5;
538 const uint32_t* const p_loop1_end = src + (num_pixels & ~3);
539 const uint32_t* const p_loop2_end = src + num_pixels;
540 __asm__ volatile (
541 ".set push \n\t"
542 ".set noreorder \n\t"
543 "beq %[src], %[p_loop1_end], 3f \n\t"
544 " nop \n\t"
545 "0: \n\t"
546 "lw %[temp0], 0(%[src]) \n\t"
547 "lw %[temp1], 4(%[src]) \n\t"
548 "lw %[temp2], 8(%[src]) \n\t"
549 "lw %[temp3], 12(%[src]) \n\t"
550 "ext %[temp4], %[temp0], 8, 16 \n\t"
551 "ext %[temp5], %[temp0], 5, 11 \n\t"
552 "ext %[temp0], %[temp0], 3, 5 \n\t"
553 "ins %[temp4], %[temp5], 0, 11 \n\t"
554 "ext %[temp5], %[temp1], 5, 11 \n\t"
555 "ins %[temp4], %[temp0], 0, 5 \n\t"
556 "ext %[temp0], %[temp1], 8, 16 \n\t"
557 "ext %[temp1], %[temp1], 3, 5 \n\t"
558 "ins %[temp0], %[temp5], 0, 11 \n\t"
559 "ext %[temp5], %[temp2], 5, 11 \n\t"
560 "ins %[temp0], %[temp1], 0, 5 \n\t"
561 "ext %[temp1], %[temp2], 8, 16 \n\t"
562 "ext %[temp2], %[temp2], 3, 5 \n\t"
563 "ins %[temp1], %[temp5], 0, 11 \n\t"
564 "ext %[temp5], %[temp3], 5, 11 \n\t"
565 "ins %[temp1], %[temp2], 0, 5 \n\t"
566 "ext %[temp2], %[temp3], 8, 16 \n\t"
567 "ext %[temp3], %[temp3], 3, 5 \n\t"
568 "ins %[temp2], %[temp5], 0, 11 \n\t"
569 "append %[temp0], %[temp4], 16 \n\t"
570 "ins %[temp2], %[temp3], 0, 5 \n\t"
571 "addiu %[src], %[src], 16 \n\t"
572 "append %[temp2], %[temp1], 16 \n\t"
573 #ifdef WEBP_SWAP_16BIT_CSP
574 "usw %[temp0], 0(%[dst]) \n\t"
575 "usw %[temp2], 4(%[dst]) \n\t"
576 #else
577 "wsbh %[temp0], %[temp0] \n\t"
578 "wsbh %[temp2], %[temp2] \n\t"
579 "usw %[temp0], 0(%[dst]) \n\t"
580 "usw %[temp2], 4(%[dst]) \n\t"
581 #endif
582 "bne %[src], %[p_loop1_end], 0b \n\t"
583 " addiu %[dst], %[dst], 8 \n\t"
584 "3: \n\t"
585 "beq %[src], %[p_loop2_end], 2f \n\t"
586 " nop \n\t"
587 "1: \n\t"
588 "lw %[temp0], 0(%[src]) \n\t"
589 "ext %[temp4], %[temp0], 8, 16 \n\t"
590 "ext %[temp5], %[temp0], 5, 11 \n\t"
591 "ext %[temp0], %[temp0], 3, 5 \n\t"
592 "ins %[temp4], %[temp5], 0, 11 \n\t"
593 "addiu %[src], %[src], 4 \n\t"
594 "ins %[temp4], %[temp0], 0, 5 \n\t"
595 #ifdef WEBP_SWAP_16BIT_CSP
596 "ush %[temp4], 0(%[dst]) \n\t"
597 #else
598 "wsbh %[temp4], %[temp4] \n\t"
599 "ush %[temp4], 0(%[dst]) \n\t"
600 #endif
601 "bne %[src], %[p_loop2_end], 1b \n\t"
602 " addiu %[dst], %[dst], 2 \n\t"
603 "2: \n\t"
604 ".set pop \n\t"
605 : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
606 [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
607 [dst]"+&r"(dst), [src]"+&r"(src)
608 : [p_loop1_end]"r"(p_loop1_end), [p_loop2_end]"r"(p_loop2_end)
609 : "memory"
610 );
611 }
612
ConvertBGRAToBGR(const uint32_t * src,int num_pixels,uint8_t * dst)613 static void ConvertBGRAToBGR(const uint32_t* src,
614 int num_pixels, uint8_t* dst) {
615 int temp0, temp1, temp2, temp3;
616 const uint32_t* const p_loop1_end = src + (num_pixels & ~3);
617 const uint32_t* const p_loop2_end = src + num_pixels;
618 __asm__ volatile (
619 ".set push \n\t"
620 ".set noreorder \n\t"
621 "beq %[src], %[p_loop1_end], 3f \n\t"
622 " nop \n\t"
623 "0: \n\t"
624 "lw %[temp0], 0(%[src]) \n\t"
625 "lw %[temp1], 4(%[src]) \n\t"
626 "lw %[temp2], 8(%[src]) \n\t"
627 "lw %[temp3], 12(%[src]) \n\t"
628 "ins %[temp0], %[temp1], 24, 8 \n\t"
629 "sra %[temp1], %[temp1], 8 \n\t"
630 "ins %[temp1], %[temp2], 16, 16 \n\t"
631 "sll %[temp2], %[temp2], 8 \n\t"
632 "balign %[temp3], %[temp2], 1 \n\t"
633 "addiu %[src], %[src], 16 \n\t"
634 "usw %[temp0], 0(%[dst]) \n\t"
635 "usw %[temp1], 4(%[dst]) \n\t"
636 "usw %[temp3], 8(%[dst]) \n\t"
637 "bne %[src], %[p_loop1_end], 0b \n\t"
638 " addiu %[dst], %[dst], 12 \n\t"
639 "3: \n\t"
640 "beq %[src], %[p_loop2_end], 2f \n\t"
641 " nop \n\t"
642 "1: \n\t"
643 "lw %[temp0], 0(%[src]) \n\t"
644 "addiu %[src], %[src], 4 \n\t"
645 "addiu %[dst], %[dst], 3 \n\t"
646 "ush %[temp0], -3(%[dst]) \n\t"
647 "sra %[temp0], %[temp0], 16 \n\t"
648 "bne %[src], %[p_loop2_end], 1b \n\t"
649 " sb %[temp0], -1(%[dst]) \n\t"
650 "2: \n\t"
651 ".set pop \n\t"
652 : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),
653 [temp3]"=&r"(temp3), [dst]"+&r"(dst), [src]"+&r"(src)
654 : [p_loop1_end]"r"(p_loop1_end), [p_loop2_end]"r"(p_loop2_end)
655 : "memory"
656 );
657 }
658
659 //------------------------------------------------------------------------------
660 // Entry point
661
662 extern void VP8LDspInitMIPSdspR2(void);
663
VP8LDspInitMIPSdspR2(void)664 WEBP_TSAN_IGNORE_FUNCTION void VP8LDspInitMIPSdspR2(void) {
665 VP8LMapColor32b = MapARGB;
666 VP8LMapColor8b = MapAlpha;
667 VP8LPredictors[5] = Predictor5;
668 VP8LPredictors[6] = Predictor6;
669 VP8LPredictors[7] = Predictor7;
670 VP8LPredictors[8] = Predictor8;
671 VP8LPredictors[9] = Predictor9;
672 VP8LPredictors[10] = Predictor10;
673 VP8LPredictors[11] = Predictor11;
674 VP8LPredictors[12] = Predictor12;
675 VP8LPredictors[13] = Predictor13;
676 VP8LAddGreenToBlueAndRed = AddGreenToBlueAndRed;
677 VP8LTransformColorInverse = TransformColorInverse;
678 VP8LConvertBGRAToRGB = ConvertBGRAToRGB;
679 VP8LConvertBGRAToRGBA = ConvertBGRAToRGBA;
680 VP8LConvertBGRAToRGBA4444 = ConvertBGRAToRGBA4444;
681 VP8LConvertBGRAToRGB565 = ConvertBGRAToRGB565;
682 VP8LConvertBGRAToBGR = ConvertBGRAToBGR;
683 }
684
685 #else // !WEBP_USE_MIPS_DSP_R2
686
687 WEBP_DSP_INIT_STUB(VP8LDspInitMIPSdspR2)
688
689 #endif // WEBP_USE_MIPS_DSP_R2
690