Commit | Line | Data |
---|---|---|
2ba45a60 DM |
1 | /* |
2 | * AltiVec-enhanced yuv2yuvX | |
3 | * | |
4 | * Copyright (C) 2004 Romain Dolbeau <romain@dolbeau.org> | |
5 | * based on the equivalent C code in swscale.c | |
6 | * | |
7 | * This file is part of FFmpeg. | |
8 | * | |
9 | * FFmpeg is free software; you can redistribute it and/or | |
10 | * modify it under the terms of the GNU Lesser General Public | |
11 | * License as published by the Free Software Foundation; either | |
12 | * version 2.1 of the License, or (at your option) any later version. | |
13 | * | |
14 | * FFmpeg is distributed in the hope that it will be useful, | |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
17 | * Lesser General Public License for more details. | |
18 | * | |
19 | * You should have received a copy of the GNU Lesser General Public | |
20 | * License along with FFmpeg; if not, write to the Free Software | |
21 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
22 | */ | |
23 | ||
24 | #include <inttypes.h> | |
25 | ||
26 | #include "config.h" | |
27 | #include "libswscale/swscale.h" | |
28 | #include "libswscale/swscale_internal.h" | |
29 | #include "libavutil/attributes.h" | |
30 | #include "libavutil/cpu.h" | |
31 | #include "yuv2rgb_altivec.h" | |
f6fa7814 | 32 | #include "libavutil/ppc/util_altivec.h" |
2ba45a60 DM |
33 | |
34 | #if HAVE_ALTIVEC | |
35 | #define vzero vec_splat_s32(0) | |
36 | ||
f6fa7814 DM |
37 | #if HAVE_BIGENDIAN |
38 | #define GET_LS(a,b,c,s) {\ | |
39 | vector signed short l2 = vec_ld(((b) << 1) + 16, s);\ | |
40 | ls = vec_perm(a, l2, c);\ | |
41 | a = l2;\ | |
42 | } | |
43 | #define GET_VF(a, b, c,d) {\ | |
44 | a = vec_mergeh(c, d);\ | |
45 | b = vec_mergel(c, d);\ | |
46 | } | |
47 | #else | |
48 | #define GET_LS(a,b,c,s) {\ | |
49 | ls = a;\ | |
50 | a = vec_vsx_ld(((b) << 1) + 16, s);\ | |
51 | } | |
52 | #define GET_VF(a, b, c, d) {\ | |
53 | a = vec_mergel(d, c);\ | |
54 | b = vec_mergeh(d, c);\ | |
55 | } | |
56 | #endif | |
57 | ||
58 | #define yuv2planeX_8(d1, d2, l1, src, x, perm, filter) do {\ | |
59 | vector signed short ls;\ | |
60 | GET_LS(l1, x, perm, src);\ | |
61 | vector signed int i1 = vec_mule(filter, ls);\ | |
62 | vector signed int i2 = vec_mulo(filter, ls);\ | |
63 | vector signed int vf1, vf2;\ | |
64 | GET_VF(vf1, vf2, i1, i2);\ | |
65 | d1 = vec_add(d1, vf1);\ | |
66 | d2 = vec_add(d2, vf2);\ | |
2ba45a60 DM |
67 | } while (0) |
68 | ||
f6fa7814 DM |
69 | #if HAVE_BIGENDIAN |
70 | #define LOAD_FILTER(vf,f) {\ | |
71 | vector unsigned char perm0 = vec_lvsl(joffset, f);\ | |
72 | vf = vec_ld(joffset, f);\ | |
73 | vf = vec_perm(vf, vf, perm0);\ | |
74 | } | |
75 | #define LOAD_L1(ll1,s,p){\ | |
76 | p = vec_lvsl(xoffset, s);\ | |
77 | ll1 = vec_ld(xoffset, s);\ | |
78 | } | |
79 | #else | |
80 | #define LOAD_FILTER(vf,f) {\ | |
81 | vf = vec_vsx_ld(joffset, f);\ | |
82 | } | |
83 | #define LOAD_L1(ll1,s,p){\ | |
84 | ll1 = vec_vsx_ld(xoffset, s);\ | |
85 | } | |
86 | #endif | |
87 | ||
2ba45a60 DM |
88 | static void yuv2planeX_16_altivec(const int16_t *filter, int filterSize, |
89 | const int16_t **src, uint8_t *dest, | |
90 | const uint8_t *dither, int offset, int x) | |
91 | { | |
92 | register int i, j; | |
93 | DECLARE_ALIGNED(16, int, val)[16]; | |
94 | vector signed int vo1, vo2, vo3, vo4; | |
95 | vector unsigned short vs1, vs2; | |
96 | vector unsigned char vf; | |
97 | vector unsigned int altivec_vectorShiftInt19 = | |
98 | vec_add(vec_splat_u32(10), vec_splat_u32(9)); | |
99 | ||
100 | for (i = 0; i < 16; i++) | |
101 | val[i] = dither[(x + i + offset) & 7] << 12; | |
102 | ||
103 | vo1 = vec_ld(0, val); | |
104 | vo2 = vec_ld(16, val); | |
105 | vo3 = vec_ld(32, val); | |
106 | vo4 = vec_ld(48, val); | |
107 | ||
108 | for (j = 0; j < filterSize; j++) { | |
f6fa7814 DM |
109 | unsigned int joffset=j<<1; |
110 | unsigned int xoffset=x<<1; | |
111 | vector unsigned char perm; | |
112 | vector signed short l1,vLumFilter; | |
113 | LOAD_FILTER(vLumFilter,filter); | |
114 | vLumFilter = vec_splat(vLumFilter, 0); | |
115 | LOAD_L1(l1,src[j],perm); | |
2ba45a60 DM |
116 | yuv2planeX_8(vo1, vo2, l1, src[j], x, perm, vLumFilter); |
117 | yuv2planeX_8(vo3, vo4, l1, src[j], x + 8, perm, vLumFilter); | |
118 | } | |
119 | ||
120 | vo1 = vec_sra(vo1, altivec_vectorShiftInt19); | |
121 | vo2 = vec_sra(vo2, altivec_vectorShiftInt19); | |
122 | vo3 = vec_sra(vo3, altivec_vectorShiftInt19); | |
123 | vo4 = vec_sra(vo4, altivec_vectorShiftInt19); | |
124 | vs1 = vec_packsu(vo1, vo2); | |
125 | vs2 = vec_packsu(vo3, vo4); | |
126 | vf = vec_packsu(vs1, vs2); | |
f6fa7814 | 127 | VEC_ST(vf, 0, dest); |
2ba45a60 DM |
128 | } |
129 | ||
f6fa7814 | 130 | |
2ba45a60 DM |
131 | static inline void yuv2planeX_u(const int16_t *filter, int filterSize, |
132 | const int16_t **src, uint8_t *dest, int dstW, | |
133 | const uint8_t *dither, int offset, int x) | |
134 | { | |
135 | int i, j; | |
136 | ||
137 | for (i = x; i < dstW; i++) { | |
138 | int t = dither[(i + offset) & 7] << 12; | |
139 | for (j = 0; j < filterSize; j++) | |
140 | t += src[j][i] * filter[j]; | |
141 | dest[i] = av_clip_uint8(t >> 19); | |
142 | } | |
143 | } | |
144 | ||
145 | static void yuv2planeX_altivec(const int16_t *filter, int filterSize, | |
146 | const int16_t **src, uint8_t *dest, int dstW, | |
147 | const uint8_t *dither, int offset) | |
148 | { | |
149 | int dst_u = -(uintptr_t)dest & 15; | |
150 | int i; | |
151 | ||
152 | yuv2planeX_u(filter, filterSize, src, dest, dst_u, dither, offset, 0); | |
153 | ||
154 | for (i = dst_u; i < dstW - 15; i += 16) | |
155 | yuv2planeX_16_altivec(filter, filterSize, src, dest + i, dither, | |
156 | offset, i); | |
157 | ||
158 | yuv2planeX_u(filter, filterSize, src, dest, dstW, dither, offset, i); | |
159 | } | |
160 | ||
f6fa7814 DM |
161 | #if HAVE_BIGENDIAN |
162 | // The 3 above is 2 (filterSize == 4) + 1 (sizeof(short) == 2). | |
163 | ||
164 | // The neat trick: We only care for half the elements, | |
165 | // high or low depending on (i<<3)%16 (it's 0 or 8 here), | |
166 | // and we're going to use vec_mule, so we choose | |
167 | // carefully how to "unpack" the elements into the even slots. | |
168 | #define GET_VF4(a, vf, f) {\ | |
169 | vf = vec_ld(a<< 3, f);\ | |
170 | if ((a << 3) % 16)\ | |
171 | vf = vec_mergel(vf, (vector signed short)vzero);\ | |
172 | else\ | |
173 | vf = vec_mergeh(vf, (vector signed short)vzero);\ | |
174 | } | |
175 | #define FIRST_LOAD(sv, pos, s, per) {\ | |
176 | sv = vec_ld(pos, s);\ | |
177 | per = vec_lvsl(pos, s);\ | |
178 | } | |
179 | #define UPDATE_PTR(s0, d0, s1, d1) {\ | |
180 | d0 = s0;\ | |
181 | d1 = s1;\ | |
182 | } | |
183 | #define LOAD_SRCV(pos, a, s, per, v0, v1, vf) {\ | |
184 | v1 = vec_ld(pos + a + 16, s);\ | |
185 | vf = vec_perm(v0, v1, per);\ | |
186 | } | |
187 | #define LOAD_SRCV8(pos, a, s, per, v0, v1, vf) {\ | |
188 | if ((((uintptr_t)s + pos) % 16) > 8) {\ | |
189 | v1 = vec_ld(pos + a + 16, s);\ | |
190 | }\ | |
191 | vf = vec_perm(v0, src_v1, per);\ | |
192 | } | |
193 | #define GET_VFD(a, b, f, vf0, vf1, per, vf, off) {\ | |
194 | vf1 = vec_ld((a * 2 * filterSize) + (b * 2) + 16 + off, f);\ | |
195 | vf = vec_perm(vf0, vf1, per);\ | |
196 | } | |
197 | #else /* else of #if HAVE_BIGENDIAN */ | |
198 | #define GET_VF4(a, vf, f) {\ | |
199 | vf = (vector signed short)vec_vsx_ld(a << 3, f);\ | |
200 | vf = vec_mergeh(vf, (vector signed short)vzero);\ | |
201 | } | |
202 | #define FIRST_LOAD(sv, pos, s, per) {} | |
203 | #define UPDATE_PTR(s0, d0, s1, d1) {} | |
204 | #define LOAD_SRCV(pos, a, s, per, v0, v1, vf) {\ | |
205 | vf = vec_vsx_ld(pos + a, s);\ | |
206 | } | |
207 | #define LOAD_SRCV8(pos, a, s, per, v0, v1, vf) LOAD_SRCV(pos, a, s, per, v0, v1, vf) | |
208 | #define GET_VFD(a, b, f, vf0, vf1, per, vf, off) {\ | |
209 | vf = vec_vsx_ld((a * 2 * filterSize) + (b * 2) + off, f);\ | |
210 | } | |
211 | #endif /* end of #if HAVE_BIGENDIAN */ | |
212 | ||
2ba45a60 DM |
213 | static void hScale_altivec_real(SwsContext *c, int16_t *dst, int dstW, |
214 | const uint8_t *src, const int16_t *filter, | |
215 | const int32_t *filterPos, int filterSize) | |
216 | { | |
217 | register int i; | |
218 | DECLARE_ALIGNED(16, int, tempo)[4]; | |
219 | ||
220 | if (filterSize % 4) { | |
221 | for (i = 0; i < dstW; i++) { | |
222 | register int j; | |
223 | register int srcPos = filterPos[i]; | |
224 | register int val = 0; | |
225 | for (j = 0; j < filterSize; j++) | |
226 | val += ((int)src[srcPos + j]) * filter[filterSize * i + j]; | |
227 | dst[i] = FFMIN(val >> 7, (1 << 15) - 1); | |
228 | } | |
229 | } else | |
230 | switch (filterSize) { | |
231 | case 4: | |
232 | for (i = 0; i < dstW; i++) { | |
233 | register int srcPos = filterPos[i]; | |
234 | ||
f6fa7814 | 235 | vector unsigned char src_vF = unaligned_load(srcPos, src); |
2ba45a60 DM |
236 | vector signed short src_v, filter_v; |
237 | vector signed int val_vEven, val_s; | |
2ba45a60 | 238 | src_v = // vec_unpackh sign-extends... |
f6fa7814 | 239 | (vector signed short)(VEC_MERGEH((vector unsigned char)vzero, src_vF)); |
2ba45a60 DM |
240 | // now put our elements in the even slots |
241 | src_v = vec_mergeh(src_v, (vector signed short)vzero); | |
f6fa7814 | 242 | GET_VF4(i, filter_v, filter); |
2ba45a60 DM |
243 | val_vEven = vec_mule(src_v, filter_v); |
244 | val_s = vec_sums(val_vEven, vzero); | |
245 | vec_st(val_s, 0, tempo); | |
246 | dst[i] = FFMIN(tempo[3] >> 7, (1 << 15) - 1); | |
247 | } | |
248 | break; | |
2ba45a60 DM |
249 | case 8: |
250 | for (i = 0; i < dstW; i++) { | |
251 | register int srcPos = filterPos[i]; | |
f6fa7814 DM |
252 | vector unsigned char src_vF, src_v0, src_v1; |
253 | vector unsigned char permS; | |
2ba45a60 DM |
254 | vector signed short src_v, filter_v; |
255 | vector signed int val_v, val_s; | |
f6fa7814 DM |
256 | FIRST_LOAD(src_v0, srcPos, src, permS); |
257 | LOAD_SRCV8(srcPos, 0, src, permS, src_v0, src_v1, src_vF); | |
2ba45a60 | 258 | src_v = // vec_unpackh sign-extends... |
f6fa7814 | 259 | (vector signed short)(VEC_MERGEH((vector unsigned char)vzero, src_vF)); |
2ba45a60 | 260 | filter_v = vec_ld(i << 4, filter); |
2ba45a60 DM |
261 | val_v = vec_msums(src_v, filter_v, (vector signed int)vzero); |
262 | val_s = vec_sums(val_v, vzero); | |
263 | vec_st(val_s, 0, tempo); | |
264 | dst[i] = FFMIN(tempo[3] >> 7, (1 << 15) - 1); | |
265 | } | |
266 | break; | |
267 | ||
268 | case 16: | |
269 | for (i = 0; i < dstW; i++) { | |
270 | register int srcPos = filterPos[i]; | |
271 | ||
f6fa7814 | 272 | vector unsigned char src_vF = unaligned_load(srcPos, src); |
2ba45a60 | 273 | vector signed short src_vA = // vec_unpackh sign-extends... |
f6fa7814 | 274 | (vector signed short)(VEC_MERGEH((vector unsigned char)vzero, src_vF)); |
2ba45a60 | 275 | vector signed short src_vB = // vec_unpackh sign-extends... |
f6fa7814 | 276 | (vector signed short)(VEC_MERGEL((vector unsigned char)vzero, src_vF)); |
2ba45a60 DM |
277 | vector signed short filter_v0 = vec_ld(i << 5, filter); |
278 | vector signed short filter_v1 = vec_ld((i << 5) + 16, filter); | |
2ba45a60 DM |
279 | |
280 | vector signed int val_acc = vec_msums(src_vA, filter_v0, (vector signed int)vzero); | |
281 | vector signed int val_v = vec_msums(src_vB, filter_v1, val_acc); | |
282 | ||
283 | vector signed int val_s = vec_sums(val_v, vzero); | |
284 | ||
f6fa7814 | 285 | VEC_ST(val_s, 0, tempo); |
2ba45a60 DM |
286 | dst[i] = FFMIN(tempo[3] >> 7, (1 << 15) - 1); |
287 | } | |
288 | break; | |
289 | ||
290 | default: | |
291 | for (i = 0; i < dstW; i++) { | |
f6fa7814 | 292 | register int j, offset = i * 2 * filterSize; |
2ba45a60 DM |
293 | register int srcPos = filterPos[i]; |
294 | ||
295 | vector signed int val_s, val_v = (vector signed int)vzero; | |
f6fa7814 DM |
296 | vector signed short filter_v0R; |
297 | vector unsigned char permF, src_v0, permS; | |
298 | FIRST_LOAD(filter_v0R, offset, filter, permF); | |
299 | FIRST_LOAD(src_v0, srcPos, src, permS); | |
2ba45a60 DM |
300 | |
301 | for (j = 0; j < filterSize - 15; j += 16) { | |
f6fa7814 DM |
302 | vector unsigned char src_v1, src_vF; |
303 | vector signed short filter_v1R, filter_v2R, filter_v0, filter_v1; | |
304 | LOAD_SRCV(srcPos, j, src, permS, src_v0, src_v1, src_vF); | |
2ba45a60 | 305 | vector signed short src_vA = // vec_unpackh sign-extends... |
f6fa7814 | 306 | (vector signed short)(VEC_MERGEH((vector unsigned char)vzero, src_vF)); |
2ba45a60 | 307 | vector signed short src_vB = // vec_unpackh sign-extends... |
f6fa7814 DM |
308 | (vector signed short)(VEC_MERGEL((vector unsigned char)vzero, src_vF)); |
309 | GET_VFD(i, j, filter, filter_v0R, filter_v1R, permF, filter_v0, 0); | |
310 | GET_VFD(i, j, filter, filter_v1R, filter_v2R, permF, filter_v1, 16); | |
2ba45a60 DM |
311 | |
312 | vector signed int val_acc = vec_msums(src_vA, filter_v0, val_v); | |
313 | val_v = vec_msums(src_vB, filter_v1, val_acc); | |
f6fa7814 | 314 | UPDATE_PTR(filter_v2R, filter_v0R, src_v1, src_v0); |
2ba45a60 DM |
315 | } |
316 | ||
317 | if (j < filterSize - 7) { | |
318 | // loading src_v0 is useless, it's already done above | |
2ba45a60 DM |
319 | vector unsigned char src_v1, src_vF; |
320 | vector signed short src_v, filter_v1R, filter_v; | |
f6fa7814 | 321 | LOAD_SRCV8(srcPos, j, src, permS, src_v0, src_v1, src_vF); |
2ba45a60 | 322 | src_v = // vec_unpackh sign-extends... |
f6fa7814 DM |
323 | (vector signed short)(VEC_MERGEH((vector unsigned char)vzero, src_vF)); |
324 | GET_VFD(i, j, filter, filter_v0R, filter_v1R, permF, filter_v, 0); | |
2ba45a60 DM |
325 | val_v = vec_msums(src_v, filter_v, val_v); |
326 | } | |
2ba45a60 DM |
327 | val_s = vec_sums(val_v, vzero); |
328 | ||
f6fa7814 | 329 | VEC_ST(val_s, 0, tempo); |
2ba45a60 DM |
330 | dst[i] = FFMIN(tempo[3] >> 7, (1 << 15) - 1); |
331 | } | |
332 | } | |
333 | } | |
334 | #endif /* HAVE_ALTIVEC */ | |
335 | ||
336 | av_cold void ff_sws_init_swscale_ppc(SwsContext *c) | |
337 | { | |
338 | #if HAVE_ALTIVEC | |
339 | enum AVPixelFormat dstFormat = c->dstFormat; | |
340 | ||
341 | if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)) | |
342 | return; | |
343 | ||
344 | if (c->srcBpc == 8 && c->dstBpc <= 14) { | |
345 | c->hyScale = c->hcScale = hScale_altivec_real; | |
346 | } | |
347 | if (!is16BPS(dstFormat) && !is9_OR_10BPS(dstFormat) && | |
348 | dstFormat != AV_PIX_FMT_NV12 && dstFormat != AV_PIX_FMT_NV21 && | |
349 | !c->alpPixBuf) { | |
350 | c->yuv2planeX = yuv2planeX_altivec; | |
351 | } | |
352 | ||
353 | /* The following list of supported dstFormat values should | |
354 | * match what's found in the body of ff_yuv2packedX_altivec() */ | |
355 | if (!(c->flags & (SWS_BITEXACT | SWS_FULL_CHR_H_INT)) && !c->alpPixBuf) { | |
356 | switch (c->dstFormat) { | |
357 | case AV_PIX_FMT_ABGR: | |
358 | c->yuv2packedX = ff_yuv2abgr_X_altivec; | |
359 | break; | |
360 | case AV_PIX_FMT_BGRA: | |
361 | c->yuv2packedX = ff_yuv2bgra_X_altivec; | |
362 | break; | |
363 | case AV_PIX_FMT_ARGB: | |
364 | c->yuv2packedX = ff_yuv2argb_X_altivec; | |
365 | break; | |
366 | case AV_PIX_FMT_RGBA: | |
367 | c->yuv2packedX = ff_yuv2rgba_X_altivec; | |
368 | break; | |
369 | case AV_PIX_FMT_BGR24: | |
370 | c->yuv2packedX = ff_yuv2bgr24_X_altivec; | |
371 | break; | |
372 | case AV_PIX_FMT_RGB24: | |
373 | c->yuv2packedX = ff_yuv2rgb24_X_altivec; | |
374 | break; | |
375 | } | |
376 | } | |
377 | #endif /* HAVE_ALTIVEC */ | |
378 | } |