Commit | Line | Data |
---|---|---|
3719602c PC |
1 | /* Copyright (C) 2010-2020 The RetroArch team |
2 | * | |
3 | * --------------------------------------------------------------------------------------- | |
4 | * The following license statement only applies to this file (pixconv.c). | |
5 | * --------------------------------------------------------------------------------------- | |
6 | * | |
7 | * Permission is hereby granted, free of charge, | |
8 | * to any person obtaining a copy of this software and associated documentation files (the "Software"), | |
9 | * to deal in the Software without restriction, including without limitation the rights to | |
10 | * use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, | |
11 | * and to permit persons to whom the Software is furnished to do so, subject to the following conditions: | |
12 | * | |
13 | * The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. | |
14 | * | |
15 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, | |
16 | * INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
17 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. | |
18 | * IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, | |
19 | * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
20 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. | |
21 | */ | |
22 | ||
23 | #include <stdio.h> | |
24 | #include <stdint.h> | |
25 | #include <stdlib.h> | |
26 | #include <string.h> | |
27 | ||
28 | #include <retro_inline.h> | |
29 | ||
30 | #include <gfx/scaler/pixconv.h> | |
31 | ||
32 | #if _MSC_VER && _MSC_VER <= 1800 | |
33 | #define SCALER_NO_SIMD | |
34 | #endif | |
35 | ||
36 | #ifdef SCALER_NO_SIMD | |
37 | #undef __SSE2__ | |
38 | #endif | |
39 | ||
40 | #if defined(__SSE2__) | |
41 | #include <emmintrin.h> | |
42 | #elif defined(__MMX__) | |
43 | #include <mmintrin.h> | |
44 | #elif (defined(__ARM_NEON__) || defined(__ARM_NEON)) | |
45 | #include <arm_neon.h> | |
46 | #endif | |
47 | ||
48 | void conv_rgb565_0rgb1555(void *output_, const void *input_, | |
49 | int width, int height, | |
50 | int out_stride, int in_stride) | |
51 | { | |
52 | int h; | |
53 | const uint16_t *input = (const uint16_t*)input_; | |
54 | uint16_t *output = (uint16_t*)output_; | |
55 | ||
56 | #if defined(__SSE2__) | |
57 | int max_width = width - 7; | |
58 | const __m128i hi_mask = _mm_set1_epi16(0x7fe0); | |
59 | const __m128i lo_mask = _mm_set1_epi16(0x1f); | |
60 | #endif | |
61 | ||
62 | for (h = 0; h < height; | |
63 | h++, output += out_stride >> 1, input += in_stride >> 1) | |
64 | { | |
65 | int w = 0; | |
66 | #if defined(__SSE2__) | |
67 | for (; w < max_width; w += 8) | |
68 | { | |
69 | const __m128i in = _mm_loadu_si128((const __m128i*)(input + w)); | |
70 | __m128i hi = _mm_and_si128(_mm_slli_epi16(in, 1), hi_mask); | |
71 | __m128i lo = _mm_and_si128(in, lo_mask); | |
72 | _mm_storeu_si128((__m128i*)(output + w), _mm_or_si128(hi, lo)); | |
73 | } | |
74 | #endif | |
75 | ||
76 | for (; w < width; w++) | |
77 | { | |
78 | uint16_t col = input[w]; | |
79 | uint16_t hi = (col >> 1) & 0x7fe0; | |
80 | uint16_t lo = col & 0x1f; | |
81 | output[w] = hi | lo; | |
82 | } | |
83 | } | |
84 | } | |
85 | ||
86 | void conv_0rgb1555_rgb565(void *output_, const void *input_, | |
87 | int width, int height, | |
88 | int out_stride, int in_stride) | |
89 | { | |
90 | int h; | |
91 | const uint16_t *input = (const uint16_t*)input_; | |
92 | uint16_t *output = (uint16_t*)output_; | |
93 | ||
94 | #if defined(__SSE2__) | |
95 | int max_width = width - 7; | |
96 | ||
97 | const __m128i hi_mask = _mm_set1_epi16( | |
98 | (int16_t)((0x1f << 11) | (0x1f << 6))); | |
99 | const __m128i lo_mask = _mm_set1_epi16(0x1f); | |
100 | const __m128i glow_mask = _mm_set1_epi16(1 << 5); | |
101 | #endif | |
102 | ||
103 | for (h = 0; h < height; | |
104 | h++, output += out_stride >> 1, input += in_stride >> 1) | |
105 | { | |
106 | int w = 0; | |
107 | #if defined(__SSE2__) | |
108 | for (; w < max_width; w += 8) | |
109 | { | |
110 | const __m128i in = _mm_loadu_si128((const __m128i*)(input + w)); | |
111 | __m128i rg = _mm_and_si128(_mm_slli_epi16(in, 1), hi_mask); | |
112 | __m128i b = _mm_and_si128(in, lo_mask); | |
113 | __m128i glow = _mm_and_si128(_mm_srli_epi16(in, 4), glow_mask); | |
114 | _mm_storeu_si128((__m128i*)(output + w), | |
115 | _mm_or_si128(rg, _mm_or_si128(b, glow))); | |
116 | } | |
117 | #endif | |
118 | ||
119 | for (; w < width; w++) | |
120 | { | |
121 | uint16_t col = input[w]; | |
122 | uint16_t rg = (col << 1) & ((0x1f << 11) | (0x1f << 6)); | |
123 | uint16_t b = col & 0x1f; | |
124 | uint16_t glow = (col >> 4) & (1 << 5); | |
125 | output[w] = rg | b | glow; | |
126 | } | |
127 | } | |
128 | } | |
129 | ||
130 | void conv_0rgb1555_argb8888(void *output_, const void *input_, | |
131 | int width, int height, | |
132 | int out_stride, int in_stride) | |
133 | { | |
134 | int h; | |
135 | const uint16_t *input = (const uint16_t*)input_; | |
136 | uint32_t *output = (uint32_t*)output_; | |
137 | ||
138 | #ifdef __SSE2__ | |
139 | const __m128i pix_mask_r = _mm_set1_epi16(0x1f << 10); | |
140 | const __m128i pix_mask_gb = _mm_set1_epi16(0x1f << 5); | |
141 | const __m128i mul15_mid = _mm_set1_epi16(0x4200); | |
142 | const __m128i mul15_hi = _mm_set1_epi16(0x0210); | |
143 | const __m128i a = _mm_set1_epi16(0x00ff); | |
144 | ||
145 | int max_width = width - 7; | |
146 | #endif | |
147 | ||
148 | for (h = 0; h < height; | |
149 | h++, output += out_stride >> 2, input += in_stride >> 1) | |
150 | { | |
151 | int w = 0; | |
152 | #ifdef __SSE2__ | |
153 | for (; w < max_width; w += 8) | |
154 | { | |
155 | __m128i res_lo_bg, res_hi_bg; | |
156 | __m128i res_lo_ra, res_hi_ra; | |
157 | __m128i res_lo, res_hi; | |
158 | const __m128i in = _mm_loadu_si128((const __m128i*)(input + w)); | |
159 | __m128i r = _mm_and_si128(in, pix_mask_r); | |
160 | __m128i g = _mm_and_si128(in, pix_mask_gb); | |
161 | __m128i b = _mm_and_si128(_mm_slli_epi16(in, 5), pix_mask_gb); | |
162 | ||
163 | r = _mm_mulhi_epi16(r, mul15_hi); | |
164 | g = _mm_mulhi_epi16(g, mul15_mid); | |
165 | b = _mm_mulhi_epi16(b, mul15_mid); | |
166 | ||
167 | res_lo_bg = _mm_unpacklo_epi8(b, g); | |
168 | res_hi_bg = _mm_unpackhi_epi8(b, g); | |
169 | res_lo_ra = _mm_unpacklo_epi8(r, a); | |
170 | res_hi_ra = _mm_unpackhi_epi8(r, a); | |
171 | ||
172 | res_lo = _mm_or_si128(res_lo_bg, | |
173 | _mm_slli_si128(res_lo_ra, 2)); | |
174 | res_hi = _mm_or_si128(res_hi_bg, | |
175 | _mm_slli_si128(res_hi_ra, 2)); | |
176 | ||
177 | _mm_storeu_si128((__m128i*)(output + w + 0), res_lo); | |
178 | _mm_storeu_si128((__m128i*)(output + w + 4), res_hi); | |
179 | } | |
180 | #endif | |
181 | ||
182 | for (; w < width; w++) | |
183 | { | |
184 | uint32_t col = input[w]; | |
185 | uint32_t r = (col >> 10) & 0x1f; | |
186 | uint32_t g = (col >> 5) & 0x1f; | |
187 | uint32_t b = (col >> 0) & 0x1f; | |
188 | r = (r << 3) | (r >> 2); | |
189 | g = (g << 3) | (g >> 2); | |
190 | b = (b << 3) | (b >> 2); | |
191 | ||
192 | output[w] = (0xffu << 24) | (r << 16) | (g << 8) | (b << 0); | |
193 | } | |
194 | } | |
195 | } | |
196 | ||
197 | void conv_rgb565_argb8888(void *output_, const void *input_, | |
198 | int width, int height, | |
199 | int out_stride, int in_stride) | |
200 | { | |
201 | int h; | |
202 | const uint16_t *input = (const uint16_t*)input_; | |
203 | uint32_t *output = (uint32_t*)output_; | |
204 | ||
205 | #if defined(__SSE2__) | |
206 | const __m128i pix_mask_r = _mm_set1_epi16(0x1f << 10); | |
207 | const __m128i pix_mask_g = _mm_set1_epi16(0x3f << 5); | |
208 | const __m128i pix_mask_b = _mm_set1_epi16(0x1f << 5); | |
209 | const __m128i mul16_r = _mm_set1_epi16(0x0210); | |
210 | const __m128i mul16_g = _mm_set1_epi16(0x2080); | |
211 | const __m128i mul16_b = _mm_set1_epi16(0x4200); | |
212 | const __m128i a = _mm_set1_epi16(0x00ff); | |
213 | ||
214 | int max_width = width - 7; | |
215 | #elif defined(__MMX__) | |
216 | const __m64 pix_mask_r = _mm_set1_pi16(0x1f << 10); | |
217 | const __m64 pix_mask_g = _mm_set1_pi16(0x3f << 5); | |
218 | const __m64 pix_mask_b = _mm_set1_pi16(0x1f << 5); | |
219 | const __m64 mul16_r = _mm_set1_pi16(0x0210); | |
220 | const __m64 mul16_g = _mm_set1_pi16(0x2080); | |
221 | const __m64 mul16_b = _mm_set1_pi16(0x4200); | |
222 | const __m64 a = _mm_set1_pi16(0x00ff); | |
223 | ||
224 | int max_width = width - 3; | |
225 | #elif (defined(__ARM_NEON__) || defined(__ARM_NEON)) | |
226 | int max_width = width - 7; | |
227 | #endif | |
228 | ||
229 | for (h = 0; h < height; | |
230 | h++, output += out_stride >> 2, input += in_stride >> 1) | |
231 | { | |
232 | int w = 0; | |
233 | #if defined(__SSE2__) | |
234 | for (; w < max_width; w += 8) | |
235 | { | |
236 | __m128i res_lo, res_hi; | |
237 | __m128i res_lo_bg, res_hi_bg, res_lo_ra, res_hi_ra; | |
238 | const __m128i in = _mm_loadu_si128((const __m128i*)(input + w)); | |
239 | __m128i r = _mm_and_si128(_mm_srli_epi16(in, 1), pix_mask_r); | |
240 | __m128i g = _mm_and_si128(in, pix_mask_g); | |
241 | __m128i b = _mm_and_si128(_mm_slli_epi16(in, 5), pix_mask_b); | |
242 | ||
243 | r = _mm_mulhi_epi16(r, mul16_r); | |
244 | g = _mm_mulhi_epi16(g, mul16_g); | |
245 | b = _mm_mulhi_epi16(b, mul16_b); | |
246 | ||
247 | res_lo_bg = _mm_unpacklo_epi8(b, g); | |
248 | res_hi_bg = _mm_unpackhi_epi8(b, g); | |
249 | res_lo_ra = _mm_unpacklo_epi8(r, a); | |
250 | res_hi_ra = _mm_unpackhi_epi8(r, a); | |
251 | ||
252 | res_lo = _mm_or_si128(res_lo_bg, | |
253 | _mm_slli_si128(res_lo_ra, 2)); | |
254 | res_hi = _mm_or_si128(res_hi_bg, | |
255 | _mm_slli_si128(res_hi_ra, 2)); | |
256 | ||
257 | _mm_storeu_si128((__m128i*)(output + w + 0), res_lo); | |
258 | _mm_storeu_si128((__m128i*)(output + w + 4), res_hi); | |
259 | } | |
260 | #elif defined(__MMX__) | |
261 | for (; w < max_width; w += 4) | |
262 | { | |
263 | __m64 res_lo, res_hi; | |
264 | __m64 res_lo_bg, res_hi_bg, res_lo_ra, res_hi_ra; | |
265 | const __m64 in = *((__m64*)(input + w)); | |
266 | __m64 r = _mm_and_si64(_mm_srli_pi16(in, 1), pix_mask_r); | |
267 | __m64 g = _mm_and_si64(in, pix_mask_g); | |
268 | __m64 b = _mm_and_si64(_mm_slli_pi16(in, 5), pix_mask_b); | |
269 | ||
270 | r = _mm_mulhi_pi16(r, mul16_r); | |
271 | g = _mm_mulhi_pi16(g, mul16_g); | |
272 | b = _mm_mulhi_pi16(b, mul16_b); | |
273 | ||
274 | res_lo_bg = _mm_unpacklo_pi8(b, g); | |
275 | res_hi_bg = _mm_unpackhi_pi8(b, g); | |
276 | res_lo_ra = _mm_unpacklo_pi8(r, a); | |
277 | res_hi_ra = _mm_unpackhi_pi8(r, a); | |
278 | ||
279 | res_lo = _mm_or_si64(res_lo_bg, | |
280 | _mm_slli_si64(res_lo_ra, 16)); | |
281 | res_hi = _mm_or_si64(res_hi_bg, | |
282 | _mm_slli_si64(res_hi_ra, 16)); | |
283 | ||
284 | *((__m64*)(output + w + 0)) = res_lo; | |
285 | *((__m64*)(output + w + 2)) = res_hi; | |
286 | } | |
287 | ||
288 | _mm_empty(); | |
289 | #elif (defined(__ARM_NEON__) || defined(__ARM_NEON)) | |
290 | for (; w < max_width; w += 8) | |
291 | { | |
292 | uint16x8_t in = vld1q_u16(input + w); | |
293 | ||
294 | uint16x8_t r = vsriq_n_u16(in, in, 5); | |
295 | uint16x8_t b = vsliq_n_u16(in, in, 5); | |
296 | uint16x8_t g = vsriq_n_u16(b, b, 6); | |
297 | ||
298 | uint8x8x4_t res; | |
299 | res.val[3] = vdup_n_u8(0xffu); | |
300 | res.val[2] = vshrn_n_u16(r, 8); | |
301 | res.val[1] = vshrn_n_u16(g, 8); | |
302 | res.val[0] = vshrn_n_u16(b, 2); | |
303 | ||
304 | vst4_u8((uint8_t*)(output + w), res); | |
305 | } | |
306 | #endif | |
307 | ||
308 | for (; w < width; w++) | |
309 | { | |
310 | uint32_t col = input[w]; | |
311 | uint32_t r = (col >> 11) & 0x1f; | |
312 | uint32_t g = (col >> 5) & 0x3f; | |
313 | uint32_t b = (col >> 0) & 0x1f; | |
314 | r = (r << 3) | (r >> 2); | |
315 | g = (g << 2) | (g >> 4); | |
316 | b = (b << 3) | (b >> 2); | |
317 | ||
318 | output[w] = (0xffu << 24) | (r << 16) | (g << 8) | (b << 0); | |
319 | } | |
320 | } | |
321 | } | |
322 | ||
323 | void conv_rgb565_abgr8888(void *output_, const void *input_, | |
324 | int width, int height, | |
325 | int out_stride, int in_stride) | |
326 | { | |
327 | int h; | |
328 | const uint16_t *input = (const uint16_t*)input_; | |
329 | uint32_t *output = (uint32_t*)output_; | |
330 | #if defined(__SSE2__) | |
331 | const __m128i pix_mask_r = _mm_set1_epi16(0x1f << 10); | |
332 | const __m128i pix_mask_g = _mm_set1_epi16(0x3f << 5); | |
333 | const __m128i pix_mask_b = _mm_set1_epi16(0x1f << 5); | |
334 | const __m128i mul16_r = _mm_set1_epi16(0x0210); | |
335 | const __m128i mul16_g = _mm_set1_epi16(0x2080); | |
336 | const __m128i mul16_b = _mm_set1_epi16(0x4200); | |
337 | const __m128i a = _mm_set1_epi16(0x00ff); | |
338 | int max_width = width - 7; | |
339 | #elif (defined(__ARM_NEON__) || defined(__ARM_NEON)) | |
340 | int max_width = width - 7; | |
341 | #endif | |
342 | for (h = 0; h < height; | |
343 | h++, output += out_stride >> 2, input += in_stride >> 1) | |
344 | { | |
345 | int w = 0; | |
346 | #if defined(__SSE2__) | |
347 | for (; w < max_width; w += 8) | |
348 | { | |
349 | __m128i res_lo, res_hi; | |
350 | __m128i res_lo_bg, res_hi_bg, res_lo_ra, res_hi_ra; | |
351 | const __m128i in = _mm_loadu_si128((const __m128i*)(input + w)); | |
352 | __m128i r = _mm_and_si128(_mm_srli_epi16(in, 1), pix_mask_r); | |
353 | __m128i g = _mm_and_si128(in, pix_mask_g); | |
354 | __m128i b = _mm_and_si128(_mm_slli_epi16(in, 5), pix_mask_b); | |
355 | r = _mm_mulhi_epi16(r, mul16_r); | |
356 | g = _mm_mulhi_epi16(g, mul16_g); | |
357 | b = _mm_mulhi_epi16(b, mul16_b); | |
358 | res_lo_bg = _mm_unpacklo_epi8(b, g); | |
359 | res_hi_bg = _mm_unpackhi_epi8(b, g); | |
360 | res_lo_ra = _mm_unpacklo_epi8(r, a); | |
361 | res_hi_ra = _mm_unpackhi_epi8(r, a); | |
362 | res_lo = _mm_or_si128(res_lo_bg, | |
363 | _mm_slli_si128(res_lo_ra, 2)); | |
364 | res_hi = _mm_or_si128(res_hi_bg, | |
365 | _mm_slli_si128(res_hi_ra, 2)); | |
366 | _mm_storeu_si128((__m128i*)(output + w + 0), res_lo); | |
367 | _mm_storeu_si128((__m128i*)(output + w + 4), res_hi); | |
368 | } | |
369 | #elif (defined(__ARM_NEON__) || defined(__ARM_NEON)) | |
370 | for (; w < max_width; w += 8) | |
371 | { | |
372 | uint16x8_t in = vld1q_u16(input + w); | |
373 | ||
374 | uint16x8_t r = vsriq_n_u16(in, in, 5); | |
375 | uint16x8_t b = vsliq_n_u16(in, in, 5); | |
376 | uint16x8_t g = vsriq_n_u16(b, b, 6); | |
377 | ||
378 | uint8x8x4_t res; | |
379 | res.val[3] = vdup_n_u8(0xffu); | |
380 | res.val[2] = vshrn_n_u16(b, 2); | |
381 | res.val[1] = vshrn_n_u16(g, 8); | |
382 | res.val[0] = vshrn_n_u16(r, 8); | |
383 | ||
384 | vst4_u8((uint8_t*)(output + w), res); | |
385 | } | |
386 | #endif | |
387 | for (; w < width; w++) | |
388 | { | |
389 | uint32_t col = input[w]; | |
390 | uint32_t r = (col >> 11) & 0x1f; | |
391 | uint32_t g = (col >> 5) & 0x3f; | |
392 | uint32_t b = (col >> 0) & 0x1f; | |
393 | r = (r << 3) | (r >> 2); | |
394 | g = (g << 2) | (g >> 4); | |
395 | b = (b << 3) | (b >> 2); | |
396 | output[w] = (0xffu << 24) | (b << 16) | (g << 8) | (r << 0); | |
397 | } | |
398 | } | |
399 | } | |
400 | ||
401 | void conv_argb8888_rgba4444(void *output_, const void *input_, | |
402 | int width, int height, | |
403 | int out_stride, int in_stride) | |
404 | { | |
405 | int h, w; | |
406 | const uint32_t *input = (const uint32_t*)input_; | |
407 | uint16_t *output = (uint16_t*)output_; | |
408 | ||
409 | for (h = 0; h < height; | |
410 | h++, output += out_stride >> 2, input += in_stride >> 1) | |
411 | { | |
412 | for (w = 0; w < width; w++) | |
413 | { | |
414 | uint32_t col = input[w]; | |
415 | uint32_t r = (col >> 16) & 0xf; | |
416 | uint32_t g = (col >> 8) & 0xf; | |
417 | uint32_t b = (col) & 0xf; | |
418 | uint32_t a = (col >> 24) & 0xf; | |
419 | r = (r >> 4) | r; | |
420 | g = (g >> 4) | g; | |
421 | b = (b >> 4) | b; | |
422 | a = (a >> 4) | a; | |
423 | ||
424 | output[w] = (r << 12) | (g << 8) | (b << 4) | a; | |
425 | } | |
426 | } | |
427 | } | |
428 | ||
429 | void conv_rgba4444_argb8888(void *output_, const void *input_, | |
430 | int width, int height, | |
431 | int out_stride, int in_stride) | |
432 | { | |
433 | int h; | |
434 | const uint16_t *input = (const uint16_t*)input_; | |
435 | uint32_t *output = (uint32_t*)output_; | |
436 | ||
437 | #if defined(__MMX__) | |
438 | const __m64 pix_mask_r = _mm_set1_pi16(0xf << 10); | |
439 | const __m64 pix_mask_g = _mm_set1_pi16(0xf << 8); | |
440 | const __m64 pix_mask_b = _mm_set1_pi16(0xf << 8); | |
441 | const __m64 mul16_r = _mm_set1_pi16(0x0440); | |
442 | const __m64 mul16_g = _mm_set1_pi16(0x1100); | |
443 | const __m64 mul16_b = _mm_set1_pi16(0x1100); | |
444 | const __m64 a = _mm_set1_pi16(0x00ff); | |
445 | ||
446 | int max_width = width - 3; | |
447 | #endif | |
448 | ||
449 | for (h = 0; h < height; | |
450 | h++, output += out_stride >> 2, input += in_stride >> 1) | |
451 | { | |
452 | int w = 0; | |
453 | #if defined(__MMX__) | |
454 | for (; w < max_width; w += 4) | |
455 | { | |
456 | __m64 res_lo, res_hi; | |
457 | __m64 res_lo_bg, res_hi_bg, res_lo_ra, res_hi_ra; | |
458 | const __m64 in = *((__m64*)(input + w)); | |
459 | __m64 r = _mm_and_si64(_mm_srli_pi16(in, 2), pix_mask_r); | |
460 | __m64 g = _mm_and_si64(in, pix_mask_g); | |
461 | __m64 b = _mm_and_si64(_mm_slli_pi16(in, 4), pix_mask_b); | |
462 | ||
463 | r = _mm_mulhi_pi16(r, mul16_r); | |
464 | g = _mm_mulhi_pi16(g, mul16_g); | |
465 | b = _mm_mulhi_pi16(b, mul16_b); | |
466 | ||
467 | res_lo_bg = _mm_unpacklo_pi8(b, g); | |
468 | res_hi_bg = _mm_unpackhi_pi8(b, g); | |
469 | res_lo_ra = _mm_unpacklo_pi8(r, a); | |
470 | res_hi_ra = _mm_unpackhi_pi8(r, a); | |
471 | ||
472 | res_lo = _mm_or_si64(res_lo_bg, | |
473 | _mm_slli_si64(res_lo_ra, 16)); | |
474 | res_hi = _mm_or_si64(res_hi_bg, | |
475 | _mm_slli_si64(res_hi_ra, 16)); | |
476 | ||
477 | *((__m64*)(output + w + 0)) = res_lo; | |
478 | *((__m64*)(output + w + 2)) = res_hi; | |
479 | } | |
480 | ||
481 | _mm_empty(); | |
482 | #endif | |
483 | ||
484 | for (; w < width; w++) | |
485 | { | |
486 | uint32_t col = input[w]; | |
487 | uint32_t r = (col >> 12) & 0xf; | |
488 | uint32_t g = (col >> 8) & 0xf; | |
489 | uint32_t b = (col >> 4) & 0xf; | |
490 | uint32_t a = (col >> 0) & 0xf; | |
491 | r = (r << 4) | r; | |
492 | g = (g << 4) | g; | |
493 | b = (b << 4) | b; | |
494 | a = (a << 4) | a; | |
495 | ||
496 | output[w] = (a << 24) | (r << 16) | (g << 8) | (b << 0); | |
497 | } | |
498 | } | |
499 | } | |
500 | ||
501 | void conv_rgba4444_rgb565(void *output_, const void *input_, | |
502 | int width, int height, | |
503 | int out_stride, int in_stride) | |
504 | { | |
505 | int h, w; | |
506 | const uint16_t *input = (const uint16_t*)input_; | |
507 | uint16_t *output = (uint16_t*)output_; | |
508 | ||
509 | for (h = 0; h < height; | |
510 | h++, output += out_stride >> 1, input += in_stride >> 1) | |
511 | { | |
512 | for (w = 0; w < width; w++) | |
513 | { | |
514 | uint32_t col = input[w]; | |
515 | uint32_t r = (col >> 12) & 0xf; | |
516 | uint32_t g = (col >> 8) & 0xf; | |
517 | uint32_t b = (col >> 4) & 0xf; | |
518 | ||
519 | output[w] = (r << 12) | (g << 7) | (b << 1); | |
520 | } | |
521 | } | |
522 | } | |
523 | ||
524 | #if defined(__SSE2__) | |
525 | /* :( TODO: Make this saner. */ | |
526 | static INLINE void store_bgr24_sse2(void *output, __m128i a, | |
527 | __m128i b, __m128i c, __m128i d) | |
528 | { | |
529 | const __m128i mask_0 = _mm_set_epi32(0, 0, 0, 0x00ffffff); | |
530 | const __m128i mask_1 = _mm_set_epi32(0, 0, 0x00ffffff, 0); | |
531 | const __m128i mask_2 = _mm_set_epi32(0, 0x00ffffff, 0, 0); | |
532 | const __m128i mask_3 = _mm_set_epi32(0x00ffffff, 0, 0, 0); | |
533 | ||
534 | __m128i a0 = _mm_and_si128(a, mask_0); | |
535 | __m128i a1 = _mm_srli_si128(_mm_and_si128(a, mask_1), 1); | |
536 | __m128i a2 = _mm_srli_si128(_mm_and_si128(a, mask_2), 2); | |
537 | __m128i a3 = _mm_srli_si128(_mm_and_si128(a, mask_3), 3); | |
538 | __m128i a4 = _mm_slli_si128(_mm_and_si128(b, mask_0), 12); | |
539 | __m128i a5 = _mm_slli_si128(_mm_and_si128(b, mask_1), 11); | |
540 | ||
541 | __m128i b0 = _mm_srli_si128(_mm_and_si128(b, mask_1), 5); | |
542 | __m128i b1 = _mm_srli_si128(_mm_and_si128(b, mask_2), 6); | |
543 | __m128i b2 = _mm_srli_si128(_mm_and_si128(b, mask_3), 7); | |
544 | __m128i b3 = _mm_slli_si128(_mm_and_si128(c, mask_0), 8); | |
545 | __m128i b4 = _mm_slli_si128(_mm_and_si128(c, mask_1), 7); | |
546 | __m128i b5 = _mm_slli_si128(_mm_and_si128(c, mask_2), 6); | |
547 | ||
548 | __m128i c0 = _mm_srli_si128(_mm_and_si128(c, mask_2), 10); | |
549 | __m128i c1 = _mm_srli_si128(_mm_and_si128(c, mask_3), 11); | |
550 | __m128i c2 = _mm_slli_si128(_mm_and_si128(d, mask_0), 4); | |
551 | __m128i c3 = _mm_slli_si128(_mm_and_si128(d, mask_1), 3); | |
552 | __m128i c4 = _mm_slli_si128(_mm_and_si128(d, mask_2), 2); | |
553 | __m128i c5 = _mm_slli_si128(_mm_and_si128(d, mask_3), 1); | |
554 | ||
555 | __m128i *out = (__m128i*)output; | |
556 | ||
557 | _mm_storeu_si128(out + 0, | |
558 | _mm_or_si128(a0, _mm_or_si128(a1, _mm_or_si128(a2, | |
559 | _mm_or_si128(a3, _mm_or_si128(a4, a5)))))); | |
560 | ||
561 | _mm_storeu_si128(out + 1, | |
562 | _mm_or_si128(b0, _mm_or_si128(b1, _mm_or_si128(b2, | |
563 | _mm_or_si128(b3, _mm_or_si128(b4, b5)))))); | |
564 | ||
565 | _mm_storeu_si128(out + 2, | |
566 | _mm_or_si128(c0, _mm_or_si128(c1, _mm_or_si128(c2, | |
567 | _mm_or_si128(c3, _mm_or_si128(c4, c5)))))); | |
568 | } | |
569 | #endif | |
570 | ||
571 | void conv_0rgb1555_bgr24(void *output_, const void *input_, | |
572 | int width, int height, | |
573 | int out_stride, int in_stride) | |
574 | { | |
575 | int h; | |
576 | const uint16_t *input = (const uint16_t*)input_; | |
577 | uint8_t *output = (uint8_t*)output_; | |
578 | ||
579 | #if defined(__SSE2__) | |
580 | const __m128i pix_mask_r = _mm_set1_epi16(0x1f << 10); | |
581 | const __m128i pix_mask_gb = _mm_set1_epi16(0x1f << 5); | |
582 | const __m128i mul15_mid = _mm_set1_epi16(0x4200); | |
583 | const __m128i mul15_hi = _mm_set1_epi16(0x0210); | |
584 | const __m128i a = _mm_set1_epi16(0x00ff); | |
585 | ||
586 | int max_width = width - 15; | |
587 | #endif | |
588 | ||
589 | for (h = 0; h < height; | |
590 | h++, output += out_stride, input += in_stride >> 1) | |
591 | { | |
592 | uint8_t *out = output; | |
593 | int w = 0; | |
594 | ||
595 | #if defined(__SSE2__) | |
596 | for (; w < max_width; w += 16, out += 48) | |
597 | { | |
598 | __m128i res_lo_bg0, res_lo_bg1, res_hi_bg0, res_hi_bg1, | |
599 | res_lo_ra0, res_lo_ra1, res_hi_ra0, res_hi_ra1, | |
600 | res_lo0, res_lo1, res_hi0, res_hi1; | |
601 | const __m128i in0 = _mm_loadu_si128((const __m128i*)(input + w + 0)); | |
602 | const __m128i in1 = _mm_loadu_si128((const __m128i*)(input + w + 8)); | |
603 | __m128i r0 = _mm_and_si128(in0, pix_mask_r); | |
604 | __m128i r1 = _mm_and_si128(in1, pix_mask_r); | |
605 | __m128i g0 = _mm_and_si128(in0, pix_mask_gb); | |
606 | __m128i g1 = _mm_and_si128(in1, pix_mask_gb); | |
607 | __m128i b0 = _mm_and_si128(_mm_slli_epi16(in0, 5), pix_mask_gb); | |
608 | __m128i b1 = _mm_and_si128(_mm_slli_epi16(in1, 5), pix_mask_gb); | |
609 | ||
610 | r0 = _mm_mulhi_epi16(r0, mul15_hi); | |
611 | r1 = _mm_mulhi_epi16(r1, mul15_hi); | |
612 | g0 = _mm_mulhi_epi16(g0, mul15_mid); | |
613 | g1 = _mm_mulhi_epi16(g1, mul15_mid); | |
614 | b0 = _mm_mulhi_epi16(b0, mul15_mid); | |
615 | b1 = _mm_mulhi_epi16(b1, mul15_mid); | |
616 | ||
617 | res_lo_bg0 = _mm_unpacklo_epi8(b0, g0); | |
618 | res_lo_bg1 = _mm_unpacklo_epi8(b1, g1); | |
619 | res_hi_bg0 = _mm_unpackhi_epi8(b0, g0); | |
620 | res_hi_bg1 = _mm_unpackhi_epi8(b1, g1); | |
621 | res_lo_ra0 = _mm_unpacklo_epi8(r0, a); | |
622 | res_lo_ra1 = _mm_unpacklo_epi8(r1, a); | |
623 | res_hi_ra0 = _mm_unpackhi_epi8(r0, a); | |
624 | res_hi_ra1 = _mm_unpackhi_epi8(r1, a); | |
625 | ||
626 | res_lo0 = _mm_or_si128(res_lo_bg0, | |
627 | _mm_slli_si128(res_lo_ra0, 2)); | |
628 | res_lo1 = _mm_or_si128(res_lo_bg1, | |
629 | _mm_slli_si128(res_lo_ra1, 2)); | |
630 | res_hi0 = _mm_or_si128(res_hi_bg0, | |
631 | _mm_slli_si128(res_hi_ra0, 2)); | |
632 | res_hi1 = _mm_or_si128(res_hi_bg1, | |
633 | _mm_slli_si128(res_hi_ra1, 2)); | |
634 | ||
635 | /* Non-POT pixel sizes for the loss */ | |
636 | store_bgr24_sse2(out, res_lo0, res_hi0, res_lo1, res_hi1); | |
637 | } | |
638 | #endif | |
639 | ||
640 | for (; w < width; w++) | |
641 | { | |
642 | uint32_t col = input[w]; | |
643 | uint32_t b = (col >> 0) & 0x1f; | |
644 | uint32_t g = (col >> 5) & 0x1f; | |
645 | uint32_t r = (col >> 10) & 0x1f; | |
646 | b = (b << 3) | (b >> 2); | |
647 | g = (g << 3) | (g >> 2); | |
648 | r = (r << 3) | (r >> 2); | |
649 | ||
650 | *out++ = b; | |
651 | *out++ = g; | |
652 | *out++ = r; | |
653 | } | |
654 | } | |
655 | } | |
656 | ||
657 | void conv_rgb565_bgr24(void *output_, const void *input_, | |
658 | int width, int height, | |
659 | int out_stride, int in_stride) | |
660 | { | |
661 | int h; | |
662 | const uint16_t *input = (const uint16_t*)input_; | |
663 | uint8_t *output = (uint8_t*)output_; | |
664 | ||
665 | #if defined(__SSE2__) | |
666 | const __m128i pix_mask_r = _mm_set1_epi16(0x1f << 10); | |
667 | const __m128i pix_mask_g = _mm_set1_epi16(0x3f << 5); | |
668 | const __m128i pix_mask_b = _mm_set1_epi16(0x1f << 5); | |
669 | const __m128i mul16_r = _mm_set1_epi16(0x0210); | |
670 | const __m128i mul16_g = _mm_set1_epi16(0x2080); | |
671 | const __m128i mul16_b = _mm_set1_epi16(0x4200); | |
672 | const __m128i a = _mm_set1_epi16(0x00ff); | |
673 | ||
674 | int max_width = width - 15; | |
675 | #endif | |
676 | ||
677 | for (h = 0; h < height; h++, output += out_stride, input += in_stride >> 1) | |
678 | { | |
679 | uint8_t *out = output; | |
680 | int w = 0; | |
681 | #if defined(__SSE2__) | |
682 | for (; w < max_width; w += 16, out += 48) | |
683 | { | |
684 | __m128i res_lo_bg0, res_hi_bg0, res_lo_ra0, res_hi_ra0; | |
685 | __m128i res_lo_bg1, res_hi_bg1, res_lo_ra1, res_hi_ra1; | |
686 | __m128i res_lo0, res_hi0, res_lo1, res_hi1; | |
687 | const __m128i in0 = _mm_loadu_si128((const __m128i*)(input + w)); | |
688 | const __m128i in1 = _mm_loadu_si128((const __m128i*)(input + w + 8)); | |
689 | __m128i r0 = _mm_and_si128(_mm_srli_epi16(in0, 1), pix_mask_r); | |
690 | __m128i g0 = _mm_and_si128(in0, pix_mask_g); | |
691 | __m128i b0 = _mm_and_si128(_mm_slli_epi16(in0, 5), pix_mask_b); | |
692 | __m128i r1 = _mm_and_si128(_mm_srli_epi16(in1, 1), pix_mask_r); | |
693 | __m128i g1 = _mm_and_si128(in1, pix_mask_g); | |
694 | __m128i b1 = _mm_and_si128(_mm_slli_epi16(in1, 5), pix_mask_b); | |
695 | ||
696 | r0 = _mm_mulhi_epi16(r0, mul16_r); | |
697 | g0 = _mm_mulhi_epi16(g0, mul16_g); | |
698 | b0 = _mm_mulhi_epi16(b0, mul16_b); | |
699 | r1 = _mm_mulhi_epi16(r1, mul16_r); | |
700 | g1 = _mm_mulhi_epi16(g1, mul16_g); | |
701 | b1 = _mm_mulhi_epi16(b1, mul16_b); | |
702 | ||
703 | res_lo_bg0 = _mm_unpacklo_epi8(b0, g0); | |
704 | res_hi_bg0 = _mm_unpackhi_epi8(b0, g0); | |
705 | res_lo_ra0 = _mm_unpacklo_epi8(r0, a); | |
706 | res_hi_ra0 = _mm_unpackhi_epi8(r0, a); | |
707 | res_lo_bg1 = _mm_unpacklo_epi8(b1, g1); | |
708 | res_hi_bg1 = _mm_unpackhi_epi8(b1, g1); | |
709 | res_lo_ra1 = _mm_unpacklo_epi8(r1, a); | |
710 | res_hi_ra1 = _mm_unpackhi_epi8(r1, a); | |
711 | ||
712 | res_lo0 = _mm_or_si128(res_lo_bg0, | |
713 | _mm_slli_si128(res_lo_ra0, 2)); | |
714 | res_hi0 = _mm_or_si128(res_hi_bg0, | |
715 | _mm_slli_si128(res_hi_ra0, 2)); | |
716 | res_lo1 = _mm_or_si128(res_lo_bg1, | |
717 | _mm_slli_si128(res_lo_ra1, 2)); | |
718 | res_hi1 = _mm_or_si128(res_hi_bg1, | |
719 | _mm_slli_si128(res_hi_ra1, 2)); | |
720 | ||
721 | store_bgr24_sse2(out, res_lo0, res_hi0, res_lo1, res_hi1); | |
722 | } | |
723 | #endif | |
724 | ||
725 | for (; w < width; w++) | |
726 | { | |
727 | uint32_t col = input[w]; | |
728 | uint32_t r = (col >> 11) & 0x1f; | |
729 | uint32_t g = (col >> 5) & 0x3f; | |
730 | uint32_t b = (col >> 0) & 0x1f; | |
731 | r = (r << 3) | (r >> 2); | |
732 | g = (g << 2) | (g >> 4); | |
733 | b = (b << 3) | (b >> 2); | |
734 | ||
735 | *out++ = b; | |
736 | *out++ = g; | |
737 | *out++ = r; | |
738 | } | |
739 | } | |
740 | } | |
741 | ||
742 | void conv_bgr24_argb8888(void *output_, const void *input_, | |
743 | int width, int height, | |
744 | int out_stride, int in_stride) | |
745 | { | |
746 | int h, w; | |
747 | const uint8_t *input = (const uint8_t*)input_; | |
748 | uint32_t *output = (uint32_t*)output_; | |
749 | ||
750 | for (h = 0; h < height; | |
751 | h++, output += out_stride >> 2, input += in_stride) | |
752 | { | |
753 | const uint8_t *inp = input; | |
754 | for (w = 0; w < width; w++) | |
755 | { | |
756 | uint32_t b = *inp++; | |
757 | uint32_t g = *inp++; | |
758 | uint32_t r = *inp++; | |
759 | output[w] = (0xffu << 24) | (r << 16) | (g << 8) | (b << 0); | |
760 | } | |
761 | } | |
762 | } | |
763 | ||
764 | void conv_bgr24_rgb565(void *output_, const void *input_, | |
765 | int width, int height, | |
766 | int out_stride, int in_stride) | |
767 | { | |
768 | int h, w; | |
769 | const uint8_t *input = (const uint8_t*)input_; | |
770 | uint16_t *output = (uint16_t*)output_; | |
771 | for (h = 0; h < height; | |
772 | h++, output += out_stride, input += in_stride) | |
773 | { | |
774 | const uint8_t *inp = input; | |
775 | for (w = 0; w < width; w++) | |
776 | { | |
777 | uint16_t b = *inp++; | |
778 | uint16_t g = *inp++; | |
779 | uint16_t r = *inp++; | |
780 | ||
781 | output[w] = ((r & 0x00F8) << 8) | ((g&0x00FC) << 3) | ((b&0x00F8) >> 3); | |
782 | } | |
783 | } | |
784 | } | |
785 | ||
786 | void conv_argb8888_0rgb1555(void *output_, const void *input_, | |
787 | int width, int height, | |
788 | int out_stride, int in_stride) | |
789 | { | |
790 | int h, w; | |
791 | const uint32_t *input = (const uint32_t*)input_; | |
792 | uint16_t *output = (uint16_t*)output_; | |
793 | ||
794 | for (h = 0; h < height; | |
795 | h++, output += out_stride >> 1, input += in_stride >> 2) | |
796 | { | |
797 | for (w = 0; w < width; w++) | |
798 | { | |
799 | uint32_t col = input[w]; | |
800 | uint16_t r = (col >> 19) & 0x1f; | |
801 | uint16_t g = (col >> 11) & 0x1f; | |
802 | uint16_t b = (col >> 3) & 0x1f; | |
803 | output[w] = (r << 10) | (g << 5) | (b << 0); | |
804 | } | |
805 | } | |
806 | } | |
807 | ||
808 | void conv_argb8888_bgr24(void *output_, const void *input_, | |
809 | int width, int height, | |
810 | int out_stride, int in_stride) | |
811 | { | |
812 | int h; | |
813 | const uint32_t *input = (const uint32_t*)input_; | |
814 | uint8_t *output = (uint8_t*)output_; | |
815 | ||
816 | #if defined(__SSE2__) | |
817 | int max_width = width - 15; | |
818 | #endif | |
819 | ||
820 | for (h = 0; h < height; | |
821 | h++, output += out_stride, input += in_stride >> 2) | |
822 | { | |
823 | uint8_t *out = output; | |
824 | int w = 0; | |
825 | #if defined(__SSE2__) | |
826 | for (; w < max_width; w += 16, out += 48) | |
827 | { | |
828 | __m128i l0 = _mm_loadu_si128((const __m128i*)(input + w + 0)); | |
829 | __m128i l1 = _mm_loadu_si128((const __m128i*)(input + w + 4)); | |
830 | __m128i l2 = _mm_loadu_si128((const __m128i*)(input + w + 8)); | |
831 | __m128i l3 = _mm_loadu_si128((const __m128i*)(input + w + 12)); | |
832 | store_bgr24_sse2(out, l0, l1, l2, l3); | |
833 | } | |
834 | #endif | |
835 | ||
836 | for (; w < width; w++) | |
837 | { | |
838 | uint32_t col = input[w]; | |
839 | *out++ = (uint8_t)(col >> 0); | |
840 | *out++ = (uint8_t)(col >> 8); | |
841 | *out++ = (uint8_t)(col >> 16); | |
842 | } | |
843 | } | |
844 | } | |
845 | ||
846 | #if defined(__SSE2__) | |
847 | static INLINE __m128i conv_shuffle_rb_epi32(__m128i c) | |
848 | { | |
849 | /* SSSE3 plz */ | |
850 | const __m128i b_mask = _mm_set1_epi32(0x000000ff); | |
851 | const __m128i g_mask = _mm_set1_epi32(0x0000ff00); | |
852 | const __m128i r_mask = _mm_set1_epi32(0x00ff0000); | |
853 | __m128i sl = _mm_and_si128(_mm_slli_epi32(c, 16), r_mask); | |
854 | __m128i sr = _mm_and_si128(_mm_srli_epi32(c, 16), b_mask); | |
855 | __m128i g = _mm_and_si128(c, g_mask); | |
856 | __m128i rb = _mm_or_si128(sl, sr); | |
857 | return _mm_or_si128(g, rb); | |
858 | } | |
859 | #endif | |
860 | ||
861 | void conv_abgr8888_bgr24(void *output_, const void *input_, | |
862 | int width, int height, | |
863 | int out_stride, int in_stride) | |
864 | { | |
865 | int h; | |
866 | const uint32_t *input = (const uint32_t*)input_; | |
867 | uint8_t *output = (uint8_t*)output_; | |
868 | ||
869 | #if defined(__SSE2__) | |
870 | int max_width = width - 15; | |
871 | #endif | |
872 | ||
873 | for (h = 0; h < height; | |
874 | h++, output += out_stride, input += in_stride >> 2) | |
875 | { | |
876 | uint8_t *out = output; | |
877 | int w = 0; | |
878 | #if defined(__SSE2__) | |
879 | for (; w < max_width; w += 16, out += 48) | |
880 | { | |
881 | __m128i a = _mm_loadu_si128((const __m128i*)(input + w + 0)); | |
882 | __m128i b = _mm_loadu_si128((const __m128i*)(input + w + 4)); | |
883 | __m128i c = _mm_loadu_si128((const __m128i*)(input + w + 8)); | |
884 | __m128i d = _mm_loadu_si128((const __m128i*)(input + w + 12)); | |
885 | a = conv_shuffle_rb_epi32(a); | |
886 | b = conv_shuffle_rb_epi32(b); | |
887 | c = conv_shuffle_rb_epi32(c); | |
888 | d = conv_shuffle_rb_epi32(d); | |
889 | store_bgr24_sse2(out, a, b, c, d); | |
890 | } | |
891 | #endif | |
892 | ||
893 | for (; w < width; w++) | |
894 | { | |
895 | uint32_t col = input[w]; | |
896 | *out++ = (uint8_t)(col >> 16); | |
897 | *out++ = (uint8_t)(col >> 8); | |
898 | *out++ = (uint8_t)(col >> 0); | |
899 | } | |
900 | } | |
901 | } | |
902 | ||
903 | void conv_argb8888_abgr8888(void *output_, const void *input_, | |
904 | int width, int height, | |
905 | int out_stride, int in_stride) | |
906 | { | |
907 | int h, w; | |
908 | const uint32_t *input = (const uint32_t*)input_; | |
909 | uint32_t *output = (uint32_t*)output_; | |
910 | ||
911 | for (h = 0; h < height; | |
912 | h++, output += out_stride >> 2, input += in_stride >> 2) | |
913 | { | |
914 | for (w = 0; w < width; w++) | |
915 | { | |
916 | uint32_t col = input[w]; | |
917 | output[w] = ((col << 16) & 0xff0000) | | |
918 | ((col >> 16) & 0xff) | (col & 0xff00ff00); | |
919 | } | |
920 | } | |
921 | } | |
922 | ||
923 | #define YUV_SHIFT 6 | |
924 | #define YUV_OFFSET (1 << (YUV_SHIFT - 1)) | |
925 | #define YUV_MAT_Y (1 << 6) | |
926 | #define YUV_MAT_U_G (-22) | |
927 | #define YUV_MAT_U_B (113) | |
928 | #define YUV_MAT_V_R (90) | |
929 | #define YUV_MAT_V_G (-46) | |
930 | ||
931 | void conv_yuyv_argb8888(void *output_, const void *input_, | |
932 | int width, int height, | |
933 | int out_stride, int in_stride) | |
934 | { | |
935 | int h; | |
936 | const uint8_t *input = (const uint8_t*)input_; | |
937 | uint32_t *output = (uint32_t*)output_; | |
938 | ||
939 | #if defined(__SSE2__) | |
940 | const __m128i mask_y = _mm_set1_epi16(0xffu); | |
941 | const __m128i mask_u = _mm_set1_epi32(0xffu << 8); | |
942 | const __m128i mask_v = _mm_set1_epi32(0xffu << 24); | |
943 | const __m128i chroma_offset = _mm_set1_epi16(128); | |
944 | const __m128i round_offset = _mm_set1_epi16(YUV_OFFSET); | |
945 | ||
946 | const __m128i yuv_mul = _mm_set1_epi16(YUV_MAT_Y); | |
947 | const __m128i u_g_mul = _mm_set1_epi16(YUV_MAT_U_G); | |
948 | const __m128i u_b_mul = _mm_set1_epi16(YUV_MAT_U_B); | |
949 | const __m128i v_r_mul = _mm_set1_epi16(YUV_MAT_V_R); | |
950 | const __m128i v_g_mul = _mm_set1_epi16(YUV_MAT_V_G); | |
951 | const __m128i a = _mm_cmpeq_epi16( | |
952 | _mm_setzero_si128(), _mm_setzero_si128()); | |
953 | #endif | |
954 | ||
955 | for (h = 0; h < height; h++, output += out_stride >> 2, input += in_stride) | |
956 | { | |
957 | const uint8_t *src = input; | |
958 | uint32_t *dst = output; | |
959 | int w = 0; | |
960 | ||
961 | #if defined(__SSE2__) | |
962 | /* Each loop processes 16 pixels. */ | |
963 | for (; w + 16 <= width; w += 16, src += 32, dst += 16) | |
964 | { | |
965 | __m128i u, v, u0_g, u1_g, u0_b, u1_b, v0_r, v1_r, v0_g, v1_g, | |
966 | r0, g0, b0, r1, g1, b1; | |
967 | __m128i res_lo_bg, res_hi_bg, res_lo_ra, res_hi_ra; | |
968 | __m128i res0, res1, res2, res3; | |
969 | __m128i yuv0 = _mm_loadu_si128((const __m128i*)(src + 0)); /* [Y0, U0, Y1, V0, Y2, U1, Y3, V1, ...] */ | |
970 | __m128i yuv1 = _mm_loadu_si128((const __m128i*)(src + 16)); /* [Y0, U0, Y1, V0, Y2, U1, Y3, V1, ...] */ | |
971 | ||
972 | __m128i _y0 = _mm_and_si128(yuv0, mask_y); /* [Y0, Y1, Y2, ...] (16-bit) */ | |
973 | __m128i u0 = _mm_and_si128(yuv0, mask_u); /* [0, U0, 0, 0, 0, U1, 0, 0, ...] */ | |
974 | __m128i v0 = _mm_and_si128(yuv0, mask_v); /* [0, 0, 0, V1, 0, , 0, V1, ...] */ | |
975 | __m128i _y1 = _mm_and_si128(yuv1, mask_y); /* [Y0, Y1, Y2, ...] (16-bit) */ | |
976 | __m128i u1 = _mm_and_si128(yuv1, mask_u); /* [0, U0, 0, 0, 0, U1, 0, 0, ...] */ | |
977 | __m128i v1 = _mm_and_si128(yuv1, mask_v); /* [0, 0, 0, V1, 0, , 0, V1, ...] */ | |
978 | ||
979 | /* Juggle around to get U and V in the same 16-bit format as Y. */ | |
980 | u0 = _mm_srli_si128(u0, 1); | |
981 | v0 = _mm_srli_si128(v0, 3); | |
982 | u1 = _mm_srli_si128(u1, 1); | |
983 | v1 = _mm_srli_si128(v1, 3); | |
984 | u = _mm_packs_epi32(u0, u1); | |
985 | v = _mm_packs_epi32(v0, v1); | |
986 | ||
987 | /* Apply YUV offsets (U, V) -= (-128, -128). */ | |
988 | u = _mm_sub_epi16(u, chroma_offset); | |
989 | v = _mm_sub_epi16(v, chroma_offset); | |
990 | ||
991 | /* Upscale chroma horizontally (nearest). */ | |
992 | u0 = _mm_unpacklo_epi16(u, u); | |
993 | u1 = _mm_unpackhi_epi16(u, u); | |
994 | v0 = _mm_unpacklo_epi16(v, v); | |
995 | v1 = _mm_unpackhi_epi16(v, v); | |
996 | ||
997 | /* Apply transformations. */ | |
998 | _y0 = _mm_mullo_epi16(_y0, yuv_mul); | |
999 | _y1 = _mm_mullo_epi16(_y1, yuv_mul); | |
1000 | u0_g = _mm_mullo_epi16(u0, u_g_mul); | |
1001 | u1_g = _mm_mullo_epi16(u1, u_g_mul); | |
1002 | u0_b = _mm_mullo_epi16(u0, u_b_mul); | |
1003 | u1_b = _mm_mullo_epi16(u1, u_b_mul); | |
1004 | v0_r = _mm_mullo_epi16(v0, v_r_mul); | |
1005 | v1_r = _mm_mullo_epi16(v1, v_r_mul); | |
1006 | v0_g = _mm_mullo_epi16(v0, v_g_mul); | |
1007 | v1_g = _mm_mullo_epi16(v1, v_g_mul); | |
1008 | ||
1009 | /* Add contibutions from the transformed components. */ | |
1010 | r0 = _mm_srai_epi16(_mm_adds_epi16(_mm_adds_epi16(_y0, v0_r), | |
1011 | round_offset), YUV_SHIFT); | |
1012 | g0 = _mm_srai_epi16(_mm_adds_epi16( | |
1013 | _mm_adds_epi16(_mm_adds_epi16(_y0, v0_g), u0_g), round_offset), YUV_SHIFT); | |
1014 | b0 = _mm_srai_epi16(_mm_adds_epi16( | |
1015 | _mm_adds_epi16(_y0, u0_b), round_offset), YUV_SHIFT); | |
1016 | ||
1017 | r1 = _mm_srai_epi16(_mm_adds_epi16( | |
1018 | _mm_adds_epi16(_y1, v1_r), round_offset), YUV_SHIFT); | |
1019 | g1 = _mm_srai_epi16(_mm_adds_epi16( | |
1020 | _mm_adds_epi16(_mm_adds_epi16(_y1, v1_g), u1_g), round_offset), YUV_SHIFT); | |
1021 | b1 = _mm_srai_epi16(_mm_adds_epi16( | |
1022 | _mm_adds_epi16(_y1, u1_b), round_offset), YUV_SHIFT); | |
1023 | ||
1024 | /* Saturate into 8-bit. */ | |
1025 | r0 = _mm_packus_epi16(r0, r1); | |
1026 | g0 = _mm_packus_epi16(g0, g1); | |
1027 | b0 = _mm_packus_epi16(b0, b1); | |
1028 | ||
1029 | /* Interleave into ARGB. */ | |
1030 | res_lo_bg = _mm_unpacklo_epi8(b0, g0); | |
1031 | res_hi_bg = _mm_unpackhi_epi8(b0, g0); | |
1032 | res_lo_ra = _mm_unpacklo_epi8(r0, a); | |
1033 | res_hi_ra = _mm_unpackhi_epi8(r0, a); | |
1034 | res0 = _mm_unpacklo_epi16(res_lo_bg, res_lo_ra); | |
1035 | res1 = _mm_unpackhi_epi16(res_lo_bg, res_lo_ra); | |
1036 | res2 = _mm_unpacklo_epi16(res_hi_bg, res_hi_ra); | |
1037 | res3 = _mm_unpackhi_epi16(res_hi_bg, res_hi_ra); | |
1038 | ||
1039 | _mm_storeu_si128((__m128i*)(dst + 0), res0); | |
1040 | _mm_storeu_si128((__m128i*)(dst + 4), res1); | |
1041 | _mm_storeu_si128((__m128i*)(dst + 8), res2); | |
1042 | _mm_storeu_si128((__m128i*)(dst + 12), res3); | |
1043 | } | |
1044 | #endif | |
1045 | ||
1046 | /* Finish off the rest (if any) in C. */ | |
1047 | for (; w < width; w += 2, src += 4, dst += 2) | |
1048 | { | |
1049 | int _y0 = src[0]; | |
1050 | int u = src[1] - 128; | |
1051 | int _y1 = src[2]; | |
1052 | int v = src[3] - 128; | |
1053 | ||
1054 | uint8_t r0 = clamp_8bit((YUV_MAT_Y * _y0 + YUV_MAT_V_R * v + YUV_OFFSET) >> YUV_SHIFT); | |
1055 | uint8_t g0 = clamp_8bit((YUV_MAT_Y * _y0 + YUV_MAT_U_G * u + YUV_MAT_V_G * v + YUV_OFFSET) >> YUV_SHIFT); | |
1056 | uint8_t b0 = clamp_8bit((YUV_MAT_Y * _y0 + YUV_MAT_U_B * u + YUV_OFFSET) >> YUV_SHIFT); | |
1057 | ||
1058 | uint8_t r1 = clamp_8bit((YUV_MAT_Y * _y1 + YUV_MAT_V_R * v + YUV_OFFSET) >> YUV_SHIFT); | |
1059 | uint8_t g1 = clamp_8bit((YUV_MAT_Y * _y1 + YUV_MAT_U_G * u + YUV_MAT_V_G * v + YUV_OFFSET) >> YUV_SHIFT); | |
1060 | uint8_t b1 = clamp_8bit((YUV_MAT_Y * _y1 + YUV_MAT_U_B * u + YUV_OFFSET) >> YUV_SHIFT); | |
1061 | ||
1062 | dst[0] = 0xff000000u | (r0 << 16) | (g0 << 8) | (b0 << 0); | |
1063 | dst[1] = 0xff000000u | (r1 << 16) | (g1 << 8) | (b1 << 0); | |
1064 | } | |
1065 | } | |
1066 | } | |
1067 | ||
1068 | void conv_copy(void *output_, const void *input_, | |
1069 | int width, int height, | |
1070 | int out_stride, int in_stride) | |
1071 | { | |
1072 | int h; | |
1073 | int copy_len = abs(out_stride); | |
1074 | const uint8_t *input = (const uint8_t*)input_; | |
1075 | uint8_t *output = (uint8_t*)output_; | |
1076 | ||
1077 | if (abs(in_stride) < copy_len) | |
1078 | copy_len = abs(in_stride); | |
1079 | ||
1080 | for (h = 0; h < height; | |
1081 | h++, output += out_stride, input += in_stride) | |
1082 | memcpy(output, input, copy_len); | |
1083 | } |