35 #define vzero vec_splat_s32(0) 38 #define GET_LS(a,b,c,s) {\ 40 a = vec_vsx_ld(((b) << 1) + 16, s);\ 43 #define yuv2planeX_8(d1, d2, l1, src, x, perm, filter) do {\ 44 vector signed short ls;\ 45 vector signed int vf1, vf2, i1, i2;\ 46 GET_LS(l1, x, perm, src);\ 47 i1 = vec_mule(filter, ls);\ 48 i2 = vec_mulo(filter, ls);\ 49 vf1 = vec_mergeh(i1, i2);\ 50 vf2 = vec_mergel(i1, i2);\ 51 d1 = vec_add(d1, vf1);\ 52 d2 = vec_add(d2, vf2);\ 55 #define LOAD_FILTER(vf,f) {\ 56 vf = vec_vsx_ld(joffset, f);\ 58 #define LOAD_L1(ll1,s,p){\ 59 ll1 = vec_vsx_ld(xoffset, s);\ 68 #define GET_VF4(a, vf, f) {\ 69 vf = (vector signed short)vec_vsx_ld(a << 3, f);\ 70 vf = vec_mergeh(vf, (vector signed short)vzero);\ 72 #define FIRST_LOAD(sv, pos, s, per) {} 73 #define UPDATE_PTR(s0, d0, s1, d1) {} 74 #define LOAD_SRCV(pos, a, s, per, v0, v1, vf) {\ 75 vf = vec_vsx_ld(pos + a, s);\ 77 #define LOAD_SRCV8(pos, a, s, per, v0, v1, vf) LOAD_SRCV(pos, a, s, per, v0, v1, vf) 78 #define GET_VFD(a, b, f, vf0, vf1, per, vf, off) {\ 79 vf = vec_vsx_ld((a * 2 * filterSize) + (b * 2) + off, f);\ 82 #define FUNC(name) name ## _vsx 90 static void yuv2plane1_8_u(
const int16_t *
src,
uint8_t *dest,
int dstW,
94 for (i = start; i < dstW; i++) {
95 int val = (src[
i] + dither[(i +
offset) & 7]) >> 7;
96 dest[
i] = av_clip_uint8(val);
100 static void yuv2plane1_8_vsx(
const int16_t *src,
uint8_t *dest,
int dstW,
101 const uint8_t *dither,
int offset)
103 const int dst_u = -(uintptr_t)dest & 15;
106 const vector uint16_t
shifts = (vector uint16_t) {7, 7, 7, 7, 7, 7, 7, 7};
107 vector int16_t
vi, vileft, ditherleft, ditherright;
110 for (j = 0; j < 16; j++) {
111 val[j] = dither[(dst_u + offset + j) & 7];
114 ditherleft = vec_ld(0,
val);
115 ditherright = vec_ld(0, &
val[8]);
117 yuv2plane1_8_u(src, dest, dst_u, dither, offset, 0);
119 for (i = dst_u; i < dstW - 15; i += 16) {
121 vi = vec_vsx_ld(0, &src[i]);
122 vi = vec_adds(ditherleft, vi);
123 vileft = vec_sra(vi, shifts);
125 vi = vec_vsx_ld(0, &src[i + 8]);
126 vi = vec_adds(ditherright, vi);
127 vi = vec_sra(vi, shifts);
129 vd = vec_packsu(vileft, vi);
130 vec_st(vd, 0, &dest[i]);
133 yuv2plane1_8_u(src, dest, dstW, dither, offset, i);
138 #define output_pixel(pos, val) \ 140 AV_WB16(pos, av_clip_uintp2(val >> shift, output_bits)); \ 142 AV_WL16(pos, av_clip_uintp2(val >> shift, output_bits)); \ 145 static void yuv2plane1_nbps_u(
const int16_t *src, uint16_t *dest,
int dstW,
146 int big_endian,
int output_bits,
int start)
149 int shift = 15 - output_bits;
151 for (i = start; i < dstW; i++) {
152 int val = src[
i] + (1 << (shift - 1));
157 static void yuv2plane1_nbps_vsx(
const int16_t *src, uint16_t *dest,
int dstW,
158 int big_endian,
int output_bits)
160 const int dst_u = -(uintptr_t)dest & 7;
161 const int shift = 15 - output_bits;
162 const int add = (1 << (shift - 1));
163 const int clip = (1 << output_bits) - 1;
164 const vector uint16_t vadd = (vector uint16_t) {add, add, add, add, add, add, add, add};
165 const vector uint16_t vswap = (vector uint16_t) vec_splat_u16(big_endian ? 8 : 0);
166 const vector uint16_t vshift = (vector uint16_t) vec_splat_u16(shift);
171 yuv2plane1_nbps_u(src, dest, dst_u, big_endian, output_bits, 0);
173 for (i = dst_u; i < dstW - 7; i += 8) {
174 v = vec_vsx_ld(0, (
const uint16_t *) &src[i]);
175 v = vec_add(v, vadd);
176 v = vec_sr(v, vshift);
177 v = vec_min(v, vlargest);
178 v = vec_rl(v, vswap);
179 vec_st(v, 0, &dest[i]);
182 yuv2plane1_nbps_u(src, dest, dstW, big_endian, output_bits, i);
185 static void yuv2planeX_nbps_u(
const int16_t *
filter,
int filterSize,
186 const int16_t **src, uint16_t *dest,
int dstW,
187 int big_endian,
int output_bits,
int start)
190 int shift = 11 + 16 - output_bits;
192 for (i = start; i < dstW; i++) {
193 int val = 1 << (shift - 1);
196 for (j = 0; j < filterSize; j++)
197 val += src[j][i] * filter[j];
203 static void yuv2planeX_nbps_vsx(
const int16_t *filter,
int filterSize,
204 const int16_t **src, uint16_t *dest,
int dstW,
205 int big_endian,
int output_bits)
207 const int dst_u = -(uintptr_t)dest & 7;
208 const int shift = 11 + 16 - output_bits;
209 const int add = (1 << (shift - 1));
210 const int clip = (1 << output_bits) - 1;
211 const uint16_t swap = big_endian ? 8 : 0;
212 const vector uint32_t vadd = (vector uint32_t) {add, add, add, add};
213 const vector uint32_t vshift = (vector uint32_t) {
shift,
shift,
shift, shift};
214 const vector uint16_t vswap = (vector uint16_t) {swap, swap, swap, swap, swap, swap, swap, swap};
216 const vector int16_t vzero = vec_splat_s16(0);
217 const vector
uint8_t vperm = (vector
uint8_t) {0, 1, 8, 9, 2, 3, 10, 11, 4, 5, 12, 13, 6, 7, 14, 15};
220 vector uint32_t vleft, vright, vtmp;
223 for (i = 0; i < filterSize; i++) {
224 vfilter[
i] = (vector int16_t) {filter[
i], filter[
i], filter[
i], filter[
i],
225 filter[
i], filter[
i], filter[
i], filter[
i]};
228 yuv2planeX_nbps_u(filter, filterSize, src, dest, dst_u, big_endian, output_bits, 0);
230 for (i = dst_u; i < dstW - 7; i += 8) {
231 vleft = vright = vadd;
233 for (j = 0; j < filterSize; j++) {
234 vin = vec_vsx_ld(0, &src[j][i]);
235 vtmp = (vector uint32_t) vec_mule(vin, vfilter[j]);
236 vleft = vec_add(vleft, vtmp);
237 vtmp = (vector uint32_t) vec_mulo(vin, vfilter[j]);
238 vright = vec_add(vright, vtmp);
241 vleft = vec_sra(vleft, vshift);
242 vright = vec_sra(vright, vshift);
243 v = vec_packsu(vleft, vright);
244 v = (vector uint16_t) vec_max((vector int16_t) v, vzero);
245 v = vec_min(v, vlargest);
246 v = vec_rl(v, vswap);
247 v = vec_perm(v, v, vperm);
248 vec_st(v, 0, &dest[i]);
251 yuv2planeX_nbps_u(filter, filterSize, src, dest, dstW, big_endian, output_bits, i);
257 #define output_pixel(pos, val, bias, signedness) \ 259 AV_WB16(pos, bias + av_clip_ ## signedness ## 16(val >> shift)); \ 261 AV_WL16(pos, bias + av_clip_ ## signedness ## 16(val >> shift)); \ 264 static void yuv2plane1_16_u(
const int32_t *src, uint16_t *dest,
int dstW,
265 int big_endian,
int output_bits,
int start)
270 for (i = start; i < dstW; i++) {
271 int val = src[
i] + (1 << (shift - 1));
276 static void yuv2plane1_16_vsx(
const int32_t *src, uint16_t *dest,
int dstW,
277 int big_endian,
int output_bits)
279 const int dst_u = -(uintptr_t)dest & 7;
281 const int add = (1 << (shift - 1));
282 const vector uint32_t vadd = (vector uint32_t) {add, add, add, add};
283 const vector uint16_t vswap = (vector uint16_t) vec_splat_u16(big_endian ? 8 : 0);
284 const vector uint32_t vshift = (vector uint32_t) vec_splat_u32(shift);
285 vector uint32_t v, v2;
289 yuv2plane1_16_u(src, dest, dst_u, big_endian, output_bits, 0);
291 for (i = dst_u; i < dstW - 7; i += 8) {
292 v = vec_vsx_ld(0, (
const uint32_t *) &src[i]);
293 v = vec_add(v, vadd);
294 v = vec_sr(v, vshift);
296 v2 = vec_vsx_ld(0, (
const uint32_t *) &src[i + 4]);
297 v2 = vec_add(v2, vadd);
298 v2 = vec_sr(v2, vshift);
300 vd = vec_packsu(v, v2);
301 vd = vec_rl(vd, vswap);
303 vec_st(vd, 0, &dest[i]);
306 yuv2plane1_16_u(src, dest, dstW, big_endian, output_bits, i);
311 static void yuv2planeX_16_u(
const int16_t *filter,
int filterSize,
312 const int32_t **src, uint16_t *dest,
int dstW,
313 int big_endian,
int output_bits,
int start)
318 for (i = start; i < dstW; i++) {
319 int val = 1 << (shift - 1);
328 for (j = 0; j < filterSize; j++)
329 val += src[j][i] * (
unsigned)filter[j];
335 static void yuv2planeX_16_vsx(
const int16_t *filter,
int filterSize,
336 const int32_t **src, uint16_t *dest,
int dstW,
337 int big_endian,
int output_bits)
339 const int dst_u = -(uintptr_t)dest & 7;
340 const int shift = 15;
341 const int bias = 0x8000;
342 const int add = (1 << (shift - 1)) - 0x40000000;
343 const uint16_t swap = big_endian ? 8 : 0;
344 const vector uint32_t vadd = (vector uint32_t) {add, add, add, add};
345 const vector uint32_t vshift = (vector uint32_t) {
shift,
shift,
shift, shift};
346 const vector uint16_t vswap = (vector uint16_t) {swap, swap, swap, swap, swap, swap, swap, swap};
347 const vector uint16_t vbias = (vector uint16_t) {bias, bias, bias, bias, bias, bias, bias, bias};
350 vector uint32_t vleft, vright, vtmp;
354 for (i = 0; i < filterSize; i++) {
355 vfilter[
i] = (vector
int32_t) {filter[
i], filter[
i], filter[
i], filter[
i]};
358 yuv2planeX_16_u(filter, filterSize, src, dest, dst_u, big_endian, output_bits, 0);
360 for (i = dst_u; i < dstW - 7; i += 8) {
361 vleft = vright = vadd;
363 for (j = 0; j < filterSize; j++) {
364 vin32l = vec_vsx_ld(0, &src[j][i]);
365 vin32r = vec_vsx_ld(0, &src[j][i + 4]);
367 vtmp = (vector uint32_t) vec_mul(vin32l, vfilter[j]);
368 vleft = vec_add(vleft, vtmp);
369 vtmp = (vector uint32_t) vec_mul(vin32r, vfilter[j]);
370 vright = vec_add(vright, vtmp);
373 vleft = vec_sra(vleft, vshift);
374 vright = vec_sra(vright, vshift);
375 v = (vector uint16_t) vec_packs((vector
int32_t) vleft, (vector
int32_t) vright);
376 v = vec_add(v, vbias);
377 v = vec_rl(v, vswap);
378 vec_st(v, 0, &dest[i]);
381 yuv2planeX_16_u(filter, filterSize, src, dest, dstW, big_endian, output_bits, i);
386 #define yuv2NBPS(bits, BE_LE, is_be, template_size, typeX_t) \ 387 yuv2NBPS1(bits, BE_LE, is_be, template_size, typeX_t) \ 388 yuv2NBPSX(bits, BE_LE, is_be, template_size, typeX_t) 390 #define yuv2NBPS1(bits, BE_LE, is_be, template_size, typeX_t) \ 391 static void yuv2plane1_ ## bits ## BE_LE ## _vsx(const int16_t *src, \ 392 uint8_t *dest, int dstW, \ 393 const uint8_t *dither, int offset) \ 395 yuv2plane1_ ## template_size ## _vsx((const typeX_t *) src, \ 396 (uint16_t *) dest, dstW, is_be, bits); \ 399 #define yuv2NBPSX(bits, BE_LE, is_be, template_size, typeX_t) \ 400 static void yuv2planeX_ ## bits ## BE_LE ## _vsx(const int16_t *filter, int filterSize, \ 401 const int16_t **src, uint8_t *dest, int dstW, \ 402 const uint8_t *dither, int offset)\ 404 yuv2planeX_## template_size ## _vsx(filter, \ 405 filterSize, (const typeX_t **) src, \ 406 (uint16_t *) dest, dstW, is_be, bits); \ 418 yuv2NBPS1(16, BE, 1, 16,
int32_t)
419 yuv2NBPS1(16, LE, 0, 16,
int32_t)
421 yuv2NBPSX(16, BE, 1, 16,
int32_t)
422 yuv2NBPSX(16, LE, 0, 16,
int32_t)
426 R_l = vec_max(R_l, zero32); \ 427 R_r = vec_max(R_r, zero32); \ 428 G_l = vec_max(G_l, zero32); \ 429 G_r = vec_max(G_r, zero32); \ 430 B_l = vec_max(B_l, zero32); \ 431 B_r = vec_max(B_r, zero32); \ 433 R_l = vec_min(R_l, rgbclip); \ 434 R_r = vec_min(R_r, rgbclip); \ 435 G_l = vec_min(G_l, rgbclip); \ 436 G_r = vec_min(G_r, rgbclip); \ 437 B_l = vec_min(B_l, rgbclip); \ 438 B_r = vec_min(B_r, rgbclip); \ 440 R_l = vec_sr(R_l, shift22); \ 441 R_r = vec_sr(R_r, shift22); \ 442 G_l = vec_sr(G_l, shift22); \ 443 G_r = vec_sr(G_r, shift22); \ 444 B_l = vec_sr(B_l, shift22); \ 445 B_r = vec_sr(B_r, shift22); \ 447 rd16 = vec_packsu(R_l, R_r); \ 448 gd16 = vec_packsu(G_l, G_r); \ 449 bd16 = vec_packsu(B_l, B_r); \ 450 rd = vec_packsu(rd16, zero16); \ 451 gd = vec_packsu(gd16, zero16); \ 452 bd = vec_packsu(bd16, zero16); \ 455 case AV_PIX_FMT_RGB24: \ 456 out0 = vec_perm(rd, gd, perm3rg0); \ 457 out0 = vec_perm(out0, bd, perm3tb0); \ 458 out1 = vec_perm(rd, gd, perm3rg1); \ 459 out1 = vec_perm(out1, bd, perm3tb1); \ 461 vec_vsx_st(out0, 0, dest); \ 462 vec_vsx_st(out1, 16, dest); \ 466 case AV_PIX_FMT_BGR24: \ 467 out0 = vec_perm(bd, gd, perm3rg0); \ 468 out0 = vec_perm(out0, rd, perm3tb0); \ 469 out1 = vec_perm(bd, gd, perm3rg1); \ 470 out1 = vec_perm(out1, rd, perm3tb1); \ 472 vec_vsx_st(out0, 0, dest); \ 473 vec_vsx_st(out1, 16, dest); \ 477 case AV_PIX_FMT_BGRA: \ 478 out0 = vec_mergeh(bd, gd); \ 479 out1 = vec_mergeh(rd, ad); \ 481 tmp8 = (vector uint8_t) vec_mergeh((vector uint16_t) out0, (vector uint16_t) out1); \ 482 vec_vsx_st(tmp8, 0, dest); \ 483 tmp8 = (vector uint8_t) vec_mergel((vector uint16_t) out0, (vector uint16_t) out1); \ 484 vec_vsx_st(tmp8, 16, dest); \ 488 case AV_PIX_FMT_RGBA: \ 489 out0 = vec_mergeh(rd, gd); \ 490 out1 = vec_mergeh(bd, ad); \ 492 tmp8 = (vector uint8_t) vec_mergeh((vector uint16_t) out0, (vector uint16_t) out1); \ 493 vec_vsx_st(tmp8, 0, dest); \ 494 tmp8 = (vector uint8_t) vec_mergel((vector uint16_t) out0, (vector uint16_t) out1); \ 495 vec_vsx_st(tmp8, 16, dest); \ 499 case AV_PIX_FMT_ARGB: \ 500 out0 = vec_mergeh(ad, rd); \ 501 out1 = vec_mergeh(gd, bd); \ 503 tmp8 = (vector uint8_t) vec_mergeh((vector uint16_t) out0, (vector uint16_t) out1); \ 504 vec_vsx_st(tmp8, 0, dest); \ 505 tmp8 = (vector uint8_t) vec_mergel((vector uint16_t) out0, (vector uint16_t) out1); \ 506 vec_vsx_st(tmp8, 16, dest); \ 510 case AV_PIX_FMT_ABGR: \ 511 out0 = vec_mergeh(ad, bd); \ 512 out1 = vec_mergeh(gd, rd); \ 514 tmp8 = (vector uint8_t) vec_mergeh((vector uint16_t) out0, (vector uint16_t) out1); \ 515 vec_vsx_st(tmp8, 0, dest); \ 516 tmp8 = (vector uint8_t) vec_mergel((vector uint16_t) out0, (vector uint16_t) out1); \ 517 vec_vsx_st(tmp8, 16, dest); \ 524 yuv2rgb_full_X_vsx_template(
SwsContext *
c,
const int16_t *lumFilter,
525 const int16_t **lumSrc,
int lumFilterSize,
526 const int16_t *chrFilter,
const int16_t **chrUSrc,
527 const int16_t **chrVSrc,
int chrFilterSize,
528 const int16_t **alpSrc,
uint8_t *dest,
532 vector
int32_t vy32_l, vy32_r, vu32_l, vu32_r, vv32_l, vv32_r, tmp32;
533 vector
int32_t R_l, R_r, G_l, G_r, B_l, B_r;
535 vector uint16_t rd16, gd16, bd16;
536 vector
uint8_t rd, bd, gd, ad, out0, out1, tmp8;
538 const vector
int32_t ystart = vec_splats(1 << 9);
539 const vector
int32_t uvstart = vec_splats((1 << 9) - (128 << 19));
540 const vector uint16_t zero16 = vec_splat_u16(0);
543 const vector
int32_t y_add = vec_splats(1 << 21);
548 const vector
int32_t rgbclip = vec_splats(1 << 30);
549 const vector
int32_t zero32 = vec_splat_s32(0);
550 const vector uint32_t shift22 = vec_splats(22
U);
551 const vector uint32_t shift10 = vec_splat_u32(10);
574 ad = vec_splats((
uint8_t) 255);
576 for (i = 0; i < lumFilterSize; i++)
577 vlumFilter[i] = vec_splats(lumFilter[i]);
578 for (i = 0; i < chrFilterSize; i++)
579 vchrFilter[i] = vec_splats(chrFilter[i]);
581 for (i = 0; i < dstW; i += 8) {
589 for (j = 0; j < lumFilterSize; j++) {
590 vv = vec_ld(0, &lumSrc[j][i]);
591 tmp = vec_mule(vv, vlumFilter[j]);
592 tmp2 = vec_mulo(vv, vlumFilter[j]);
593 tmp3 = vec_mergeh(tmp, tmp2);
594 tmp4 = vec_mergel(tmp, tmp2);
596 vy32_l = vec_adds(vy32_l, tmp3);
597 vy32_r = vec_adds(vy32_r, tmp4);
600 for (j = 0; j < chrFilterSize; j++) {
601 vv = vec_ld(0, &chrUSrc[j][i]);
602 tmp = vec_mule(vv, vchrFilter[j]);
603 tmp2 = vec_mulo(vv, vchrFilter[j]);
604 tmp3 = vec_mergeh(tmp, tmp2);
605 tmp4 = vec_mergel(tmp, tmp2);
607 vu32_l = vec_adds(vu32_l, tmp3);
608 vu32_r = vec_adds(vu32_r, tmp4);
610 vv = vec_ld(0, &chrVSrc[j][i]);
611 tmp = vec_mule(vv, vchrFilter[j]);
612 tmp2 = vec_mulo(vv, vchrFilter[j]);
613 tmp3 = vec_mergeh(tmp, tmp2);
614 tmp4 = vec_mergel(tmp, tmp2);
616 vv32_l = vec_adds(vv32_l, tmp3);
617 vv32_r = vec_adds(vv32_r, tmp4);
620 vy32_l = vec_sra(vy32_l, shift10);
621 vy32_r = vec_sra(vy32_r, shift10);
622 vu32_l = vec_sra(vu32_l, shift10);
623 vu32_r = vec_sra(vu32_r, shift10);
624 vv32_l = vec_sra(vv32_l, shift10);
625 vv32_r = vec_sra(vv32_r, shift10);
627 vy32_l = vec_sub(vy32_l, y_offset);
628 vy32_r = vec_sub(vy32_r, y_offset);
629 vy32_l = vec_mul(vy32_l, y_coeff);
630 vy32_r = vec_mul(vy32_r, y_coeff);
631 vy32_l = vec_add(vy32_l, y_add);
632 vy32_r = vec_add(vy32_r, y_add);
634 R_l = vec_mul(vv32_l, v2r_coeff);
635 R_l = vec_add(R_l, vy32_l);
636 R_r = vec_mul(vv32_r, v2r_coeff);
637 R_r = vec_add(R_r, vy32_r);
638 G_l = vec_mul(vv32_l, v2g_coeff);
639 tmp32 = vec_mul(vu32_l, u2g_coeff);
640 G_l = vec_add(G_l, vy32_l);
641 G_l = vec_add(G_l, tmp32);
642 G_r = vec_mul(vv32_r, v2g_coeff);
643 tmp32 = vec_mul(vu32_r, u2g_coeff);
644 G_r = vec_add(G_r, vy32_r);
645 G_r = vec_add(G_r, tmp32);
647 B_l = vec_mul(vu32_l, u2b_coeff);
648 B_l = vec_add(B_l, vy32_l);
649 B_r = vec_mul(vu32_r, u2b_coeff);
650 B_r = vec_add(B_r, vy32_r);
656 #define SETUP(x, buf0, alpha1, buf1, alpha) { \ 657 x = vec_ld(0, buf0); \ 658 tmp = vec_mule(x, alpha1); \ 659 tmp2 = vec_mulo(x, alpha1); \ 660 tmp3 = vec_mergeh(tmp, tmp2); \ 661 tmp4 = vec_mergel(tmp, tmp2); \ 663 x = vec_ld(0, buf1); \ 664 tmp = vec_mule(x, alpha); \ 665 tmp2 = vec_mulo(x, alpha); \ 666 tmp5 = vec_mergeh(tmp, tmp2); \ 667 tmp6 = vec_mergel(tmp, tmp2); \ 669 tmp3 = vec_add(tmp3, tmp5); \ 670 tmp4 = vec_add(tmp4, tmp6); \ 675 yuv2rgb_full_2_vsx_template(
SwsContext *c,
const int16_t *
buf[2],
676 const int16_t *ubuf[2],
const int16_t *vbuf[2],
677 const int16_t *abuf[2],
uint8_t *dest,
int dstW,
678 int yalpha,
int uvalpha,
int y,
681 const int16_t *buf0 = buf[0], *buf1 = buf[1],
682 *ubuf0 = ubuf[0], *ubuf1 = ubuf[1],
683 *vbuf0 = vbuf[0], *vbuf1 = vbuf[1],
684 *abuf0 = hasAlpha ? abuf[0] :
NULL,
685 *abuf1 = hasAlpha ? abuf[1] :
NULL;
686 const int16_t yalpha1 = 4096 - yalpha;
687 const int16_t uvalpha1 = 4096 - uvalpha;
688 vector int16_t vy, vu, vv,
A = vec_splat_s16(0);
689 vector
int32_t vy32_l, vy32_r, vu32_l, vu32_r, vv32_l, vv32_r, tmp32;
690 vector
int32_t R_l, R_r, G_l, G_r, B_l, B_r;
691 vector
int32_t tmp, tmp2, tmp3, tmp4, tmp5, tmp6;
692 vector uint16_t rd16, gd16, bd16;
693 vector
uint8_t rd, bd, gd, ad, out0, out1, tmp8;
694 const vector int16_t vyalpha1 = vec_splats(yalpha1);
695 const vector int16_t vuvalpha1 = vec_splats(uvalpha1);
696 const vector int16_t vyalpha = vec_splats((int16_t) yalpha);
697 const vector int16_t vuvalpha = vec_splats((int16_t) uvalpha);
698 const vector uint16_t zero16 = vec_splat_u16(0);
701 const vector
int32_t y_add = vec_splats(1 << 21);
706 const vector
int32_t rgbclip = vec_splats(1 << 30);
707 const vector
int32_t zero32 = vec_splat_s32(0);
708 const vector uint32_t shift19 = vec_splats(19
U);
709 const vector uint32_t shift22 = vec_splats(22
U);
710 const vector uint32_t shift10 = vec_splat_u32(10);
711 const vector
int32_t dec128 = vec_splats(128 << 19);
712 const vector
int32_t add18 = vec_splats(1 << 18);
738 for (i = 0; i < dstW; i += 8) {
739 SETUP(vy, &buf0[i], vyalpha1, &buf1[i], vyalpha);
740 vy32_l = vec_sra(tmp3, shift10);
741 vy32_r = vec_sra(tmp4, shift10);
743 SETUP(vu, &ubuf0[i], vuvalpha1, &ubuf1[i], vuvalpha);
744 tmp3 = vec_sub(tmp3, dec128);
745 tmp4 = vec_sub(tmp4, dec128);
746 vu32_l = vec_sra(tmp3, shift10);
747 vu32_r = vec_sra(tmp4, shift10);
749 SETUP(vv, &vbuf0[i], vuvalpha1, &vbuf1[i], vuvalpha);
750 tmp3 = vec_sub(tmp3, dec128);
751 tmp4 = vec_sub(tmp4, dec128);
752 vv32_l = vec_sra(tmp3, shift10);
753 vv32_r = vec_sra(tmp4, shift10);
756 SETUP(A, &abuf0[i], vyalpha1, &abuf1[i], vyalpha);
757 tmp3 = vec_add(tmp3, add18);
758 tmp4 = vec_add(tmp4, add18);
759 tmp3 = vec_sra(tmp3, shift19);
760 tmp4 = vec_sra(tmp4, shift19);
761 A = vec_packs(tmp3, tmp4);
762 ad = vec_packsu(A, (vector int16_t) zero16);
764 ad = vec_splats((
uint8_t) 255);
767 vy32_l = vec_sub(vy32_l, y_offset);
768 vy32_r = vec_sub(vy32_r, y_offset);
769 vy32_l = vec_mul(vy32_l, y_coeff);
770 vy32_r = vec_mul(vy32_r, y_coeff);
771 vy32_l = vec_add(vy32_l, y_add);
772 vy32_r = vec_add(vy32_r, y_add);
774 R_l = vec_mul(vv32_l, v2r_coeff);
775 R_l = vec_add(R_l, vy32_l);
776 R_r = vec_mul(vv32_r, v2r_coeff);
777 R_r = vec_add(R_r, vy32_r);
778 G_l = vec_mul(vv32_l, v2g_coeff);
779 tmp32 = vec_mul(vu32_l, u2g_coeff);
780 G_l = vec_add(G_l, vy32_l);
781 G_l = vec_add(G_l, tmp32);
782 G_r = vec_mul(vv32_r, v2g_coeff);
783 tmp32 = vec_mul(vu32_r, u2g_coeff);
784 G_r = vec_add(G_r, vy32_r);
785 G_r = vec_add(G_r, tmp32);
787 B_l = vec_mul(vu32_l, u2b_coeff);
788 B_l = vec_add(B_l, vy32_l);
789 B_r = vec_mul(vu32_r, u2b_coeff);
790 B_r = vec_add(B_r, vy32_r);
797 yuv2rgb_2_vsx_template(
SwsContext *c,
const int16_t *buf[2],
798 const int16_t *ubuf[2],
const int16_t *vbuf[2],
799 const int16_t *abuf[2],
uint8_t *dest,
int dstW,
800 int yalpha,
int uvalpha,
int y,
803 const int16_t *buf0 = buf[0], *buf1 = buf[1],
804 *ubuf0 = ubuf[0], *ubuf1 = ubuf[1],
805 *vbuf0 = vbuf[0], *vbuf1 = vbuf[1],
806 *abuf0 = hasAlpha ? abuf[0] :
NULL,
807 *abuf1 = hasAlpha ? abuf[1] :
NULL;
808 const int16_t yalpha1 = 4096 - yalpha;
809 const int16_t uvalpha1 = 4096 - uvalpha;
810 vector int16_t vy, vu, vv, A = vec_splat_s16(0);
811 vector
int32_t vy32_l, vy32_r, vu32_l, vu32_r, vv32_l, vv32_r, tmp32;
812 vector
int32_t R_l, R_r, G_l, G_r, B_l, B_r, vud32_l, vud32_r, vvd32_l, vvd32_r;
813 vector
int32_t tmp, tmp2, tmp3, tmp4, tmp5, tmp6;
814 vector uint16_t rd16, gd16, bd16;
815 vector
uint8_t rd, bd, gd, ad, out0, out1, tmp8;
816 const vector int16_t vyalpha1 = vec_splats(yalpha1);
817 const vector int16_t vuvalpha1 = vec_splats(uvalpha1);
818 const vector int16_t vyalpha = vec_splats((int16_t) yalpha);
819 const vector int16_t vuvalpha = vec_splats((int16_t) uvalpha);
820 const vector uint16_t zero16 = vec_splat_u16(0);
823 const vector
int32_t y_add = vec_splats(1 << 21);
828 const vector
int32_t rgbclip = vec_splats(1 << 30);
829 const vector
int32_t zero32 = vec_splat_s32(0);
830 const vector uint32_t shift19 = vec_splats(19
U);
831 const vector uint32_t shift22 = vec_splats(22
U);
832 const vector uint32_t shift10 = vec_splat_u32(10);
833 const vector
int32_t dec128 = vec_splats(128 << 19);
834 const vector
int32_t add18 = vec_splats(1 << 18);
842 const vector
uint8_t doubleright = (vector
uint8_t) {8, 9, 10, 11,
868 for (i = 0; i < (dstW + 1) >> 1; i += 8) {
869 SETUP(vy, &buf0[i * 2], vyalpha1, &buf1[i * 2], vyalpha);
870 vy32_l = vec_sra(tmp3, shift10);
871 vy32_r = vec_sra(tmp4, shift10);
873 SETUP(vu, &ubuf0[i], vuvalpha1, &ubuf1[i], vuvalpha);
874 tmp3 = vec_sub(tmp3, dec128);
875 tmp4 = vec_sub(tmp4, dec128);
876 vu32_l = vec_sra(tmp3, shift10);
877 vu32_r = vec_sra(tmp4, shift10);
879 SETUP(vv, &vbuf0[i], vuvalpha1, &vbuf1[i], vuvalpha);
880 tmp3 = vec_sub(tmp3, dec128);
881 tmp4 = vec_sub(tmp4, dec128);
882 vv32_l = vec_sra(tmp3, shift10);
883 vv32_r = vec_sra(tmp4, shift10);
886 SETUP(A, &abuf0[i], vyalpha1, &abuf1[i], vyalpha);
887 tmp3 = vec_add(tmp3, add18);
888 tmp4 = vec_add(tmp4, add18);
889 tmp3 = vec_sra(tmp3, shift19);
890 tmp4 = vec_sra(tmp4, shift19);
891 A = vec_packs(tmp3, tmp4);
892 ad = vec_packsu(A, (vector int16_t) zero16);
894 ad = vec_splats((
uint8_t) 255);
897 vy32_l = vec_sub(vy32_l, y_offset);
898 vy32_r = vec_sub(vy32_r, y_offset);
899 vy32_l = vec_mul(vy32_l, y_coeff);
900 vy32_r = vec_mul(vy32_r, y_coeff);
901 vy32_l = vec_add(vy32_l, y_add);
902 vy32_r = vec_add(vy32_r, y_add);
905 vud32_l = vec_perm(vu32_l, vu32_l, doubleleft);
906 vud32_r = vec_perm(vu32_l, vu32_l, doubleright);
907 vvd32_l = vec_perm(vv32_l, vv32_l, doubleleft);
908 vvd32_r = vec_perm(vv32_l, vv32_l, doubleright);
910 R_l = vec_mul(vvd32_l, v2r_coeff);
911 R_l = vec_add(R_l, vy32_l);
912 R_r = vec_mul(vvd32_r, v2r_coeff);
913 R_r = vec_add(R_r, vy32_r);
914 G_l = vec_mul(vvd32_l, v2g_coeff);
915 tmp32 = vec_mul(vud32_l, u2g_coeff);
916 G_l = vec_add(G_l, vy32_l);
917 G_l = vec_add(G_l, tmp32);
918 G_r = vec_mul(vvd32_r, v2g_coeff);
919 tmp32 = vec_mul(vud32_r, u2g_coeff);
920 G_r = vec_add(G_r, vy32_r);
921 G_r = vec_add(G_r, tmp32);
923 B_l = vec_mul(vud32_l, u2b_coeff);
924 B_l = vec_add(B_l, vy32_l);
925 B_r = vec_mul(vud32_r, u2b_coeff);
926 B_r = vec_add(B_r, vy32_r);
931 SETUP(vy, &buf0[i * 2 + 8], vyalpha1, &buf1[i * 2 + 8], vyalpha);
932 vy32_l = vec_sra(tmp3, shift10);
933 vy32_r = vec_sra(tmp4, shift10);
935 vy32_l = vec_sub(vy32_l, y_offset);
936 vy32_r = vec_sub(vy32_r, y_offset);
937 vy32_l = vec_mul(vy32_l, y_coeff);
938 vy32_r = vec_mul(vy32_r, y_coeff);
939 vy32_l = vec_add(vy32_l, y_add);
940 vy32_r = vec_add(vy32_r, y_add);
943 vud32_l = vec_perm(vu32_r, vu32_r, doubleleft);
944 vud32_r = vec_perm(vu32_r, vu32_r, doubleright);
945 vvd32_l = vec_perm(vv32_r, vv32_r, doubleleft);
946 vvd32_r = vec_perm(vv32_r, vv32_r, doubleright);
948 R_l = vec_mul(vvd32_l, v2r_coeff);
949 R_l = vec_add(R_l, vy32_l);
950 R_r = vec_mul(vvd32_r, v2r_coeff);
951 R_r = vec_add(R_r, vy32_r);
952 G_l = vec_mul(vvd32_l, v2g_coeff);
953 tmp32 = vec_mul(vud32_l, u2g_coeff);
954 G_l = vec_add(G_l, vy32_l);
955 G_l = vec_add(G_l, tmp32);
956 G_r = vec_mul(vvd32_r, v2g_coeff);
957 tmp32 = vec_mul(vud32_r, u2g_coeff);
958 G_r = vec_add(G_r, vy32_r);
959 G_r = vec_add(G_r, tmp32);
961 B_l = vec_mul(vud32_l, u2b_coeff);
962 B_l = vec_add(B_l, vy32_l);
963 B_r = vec_mul(vud32_r, u2b_coeff);
964 B_r = vec_add(B_r, vy32_r);
973 yuv2rgb_full_1_vsx_template(
SwsContext *c,
const int16_t *buf0,
974 const int16_t *ubuf[2],
const int16_t *vbuf[2],
975 const int16_t *abuf0,
uint8_t *dest,
int dstW,
979 const int16_t *ubuf0 = ubuf[0], *vbuf0 = vbuf[0];
980 const int16_t *ubuf1 = ubuf[1], *vbuf1 = vbuf[1];
981 vector int16_t vy, vu, vv, A = vec_splat_s16(0), tmp16;
982 vector
int32_t vy32_l, vy32_r, vu32_l, vu32_r, vv32_l, vv32_r, tmp32, tmp32_2;
983 vector
int32_t R_l, R_r, G_l, G_r, B_l, B_r;
984 vector uint16_t rd16, gd16, bd16;
985 vector
uint8_t rd, bd, gd, ad, out0, out1, tmp8;
986 const vector uint16_t zero16 = vec_splat_u16(0);
989 const vector
int32_t y_add = vec_splats(1 << 21);
994 const vector
int32_t rgbclip = vec_splats(1 << 30);
995 const vector
int32_t zero32 = vec_splat_s32(0);
996 const vector uint32_t
shift2 = vec_splat_u32(2);
997 const vector uint32_t shift22 = vec_splats(22
U);
998 const vector uint16_t sub7 = vec_splats((uint16_t) (128 << 7));
999 const vector uint16_t sub8 = vec_splats((uint16_t) (128 << 8));
1000 const vector int16_t mul4 = vec_splat_s16(4);
1001 const vector int16_t
mul8 = vec_splat_s16(8);
1002 const vector int16_t add64 = vec_splat_s16(64);
1003 const vector uint16_t shift7 = vec_splat_u16(7);
1004 const vector int16_t max255 = vec_splat_s16(255);
1017 const vector
uint8_t perm3tb0 = (vector
uint8_t) {0x0, 0x1, 0x10,
1027 for (i = 0; i < dstW; i += 8) {
1028 vy = vec_ld(0, &buf0[i]);
1029 vy32_l = vec_unpackh(vy);
1030 vy32_r = vec_unpackl(vy);
1031 vy32_l = vec_sl(vy32_l, shift2);
1032 vy32_r = vec_sl(vy32_r, shift2);
1034 vu = vec_ld(0, &ubuf0[i]);
1035 vv = vec_ld(0, &vbuf0[i]);
1036 if (uvalpha < 2048) {
1037 vu = (vector int16_t) vec_sub((vector uint16_t) vu, sub7);
1038 vv = (vector int16_t) vec_sub((vector uint16_t) vv, sub7);
1040 tmp32 = vec_mule(vu, mul4);
1041 tmp32_2 = vec_mulo(vu, mul4);
1042 vu32_l = vec_mergeh(tmp32, tmp32_2);
1043 vu32_r = vec_mergel(tmp32, tmp32_2);
1044 tmp32 = vec_mule(vv, mul4);
1045 tmp32_2 = vec_mulo(vv, mul4);
1046 vv32_l = vec_mergeh(tmp32, tmp32_2);
1047 vv32_r = vec_mergel(tmp32, tmp32_2);
1049 tmp16 = vec_ld(0, &ubuf1[i]);
1050 vu = vec_add(vu, tmp16);
1051 vu = (vector int16_t) vec_sub((vector uint16_t) vu, sub8);
1052 tmp16 = vec_ld(0, &vbuf1[i]);
1053 vv = vec_add(vv, tmp16);
1054 vv = (vector int16_t) vec_sub((vector uint16_t) vv, sub8);
1056 vu32_l = vec_mule(vu, mul8);
1057 vu32_r = vec_mulo(vu, mul8);
1058 vv32_l = vec_mule(vv, mul8);
1059 vv32_r = vec_mulo(vv, mul8);
1063 A = vec_ld(0, &abuf0[i]);
1064 A = vec_add(A, add64);
1065 A = vec_sr(A, shift7);
1066 A = vec_max(A, max255);
1067 ad = vec_packsu(A, (vector int16_t) zero16);
1069 ad = vec_splats((
uint8_t) 255);
1072 vy32_l = vec_sub(vy32_l, y_offset);
1073 vy32_r = vec_sub(vy32_r, y_offset);
1074 vy32_l = vec_mul(vy32_l, y_coeff);
1075 vy32_r = vec_mul(vy32_r, y_coeff);
1076 vy32_l = vec_add(vy32_l, y_add);
1077 vy32_r = vec_add(vy32_r, y_add);
1079 R_l = vec_mul(vv32_l, v2r_coeff);
1080 R_l = vec_add(R_l, vy32_l);
1081 R_r = vec_mul(vv32_r, v2r_coeff);
1082 R_r = vec_add(R_r, vy32_r);
1083 G_l = vec_mul(vv32_l, v2g_coeff);
1084 tmp32 = vec_mul(vu32_l, u2g_coeff);
1085 G_l = vec_add(G_l, vy32_l);
1086 G_l = vec_add(G_l, tmp32);
1087 G_r = vec_mul(vv32_r, v2g_coeff);
1088 tmp32 = vec_mul(vu32_r, u2g_coeff);
1089 G_r = vec_add(G_r, vy32_r);
1090 G_r = vec_add(G_r, tmp32);
1092 B_l = vec_mul(vu32_l, u2b_coeff);
1093 B_l = vec_add(B_l, vy32_l);
1094 B_r = vec_mul(vu32_r, u2b_coeff);
1095 B_r = vec_add(B_r, vy32_r);
1102 yuv2rgb_1_vsx_template(
SwsContext *c,
const int16_t *buf0,
1103 const int16_t *ubuf[2],
const int16_t *vbuf[2],
1104 const int16_t *abuf0,
uint8_t *dest,
int dstW,
1108 const int16_t *ubuf0 = ubuf[0], *vbuf0 = vbuf[0];
1109 const int16_t *ubuf1 = ubuf[1], *vbuf1 = vbuf[1];
1110 vector int16_t vy, vu, vv, A = vec_splat_s16(0), tmp16;
1111 vector
int32_t vy32_l, vy32_r, vu32_l, vu32_r, vv32_l, vv32_r, tmp32, tmp32_2;
1112 vector
int32_t vud32_l, vud32_r, vvd32_l, vvd32_r;
1113 vector
int32_t R_l, R_r, G_l, G_r, B_l, B_r;
1114 vector uint16_t rd16, gd16, bd16;
1115 vector
uint8_t rd, bd, gd, ad, out0, out1, tmp8;
1116 const vector uint16_t zero16 = vec_splat_u16(0);
1119 const vector
int32_t y_add = vec_splats(1 << 21);
1124 const vector
int32_t rgbclip = vec_splats(1 << 30);
1125 const vector
int32_t zero32 = vec_splat_s32(0);
1126 const vector uint32_t shift2 = vec_splat_u32(2);
1127 const vector uint32_t shift22 = vec_splats(22
U);
1128 const vector uint16_t sub7 = vec_splats((uint16_t) (128 << 7));
1129 const vector uint16_t sub8 = vec_splats((uint16_t) (128 << 8));
1130 const vector int16_t mul4 = vec_splat_s16(4);
1131 const vector int16_t mul8 = vec_splat_s16(8);
1132 const vector int16_t add64 = vec_splat_s16(64);
1133 const vector uint16_t shift7 = vec_splat_u16(7);
1134 const vector int16_t max255 = vec_splat_s16(255);
1142 const vector
uint8_t doubleright = (vector
uint8_t) {8, 9, 10, 11,
1155 const vector
uint8_t perm3tb0 = (vector
uint8_t) {0x0, 0x1, 0x10,
1165 for (i = 0; i < (dstW + 1) >> 1; i += 8) {
1166 vy = vec_ld(0, &buf0[i * 2]);
1167 vy32_l = vec_unpackh(vy);
1168 vy32_r = vec_unpackl(vy);
1169 vy32_l = vec_sl(vy32_l, shift2);
1170 vy32_r = vec_sl(vy32_r, shift2);
1172 vu = vec_ld(0, &ubuf0[i]);
1173 vv = vec_ld(0, &vbuf0[i]);
1174 if (uvalpha < 2048) {
1175 vu = (vector int16_t) vec_sub((vector uint16_t) vu, sub7);
1176 vv = (vector int16_t) vec_sub((vector uint16_t) vv, sub7);
1178 tmp32 = vec_mule(vu, mul4);
1179 tmp32_2 = vec_mulo(vu, mul4);
1180 vu32_l = vec_mergeh(tmp32, tmp32_2);
1181 vu32_r = vec_mergel(tmp32, tmp32_2);
1182 tmp32 = vec_mule(vv, mul4);
1183 tmp32_2 = vec_mulo(vv, mul4);
1184 vv32_l = vec_mergeh(tmp32, tmp32_2);
1185 vv32_r = vec_mergel(tmp32, tmp32_2);
1187 tmp16 = vec_ld(0, &ubuf1[i]);
1188 vu = vec_add(vu, tmp16);
1189 vu = (vector int16_t) vec_sub((vector uint16_t) vu, sub8);
1190 tmp16 = vec_ld(0, &vbuf1[i]);
1191 vv = vec_add(vv, tmp16);
1192 vv = (vector int16_t) vec_sub((vector uint16_t) vv, sub8);
1194 vu32_l = vec_mule(vu, mul8);
1195 vu32_r = vec_mulo(vu, mul8);
1196 vv32_l = vec_mule(vv, mul8);
1197 vv32_r = vec_mulo(vv, mul8);
1201 A = vec_ld(0, &abuf0[i]);
1202 A = vec_add(A, add64);
1203 A = vec_sr(A, shift7);
1204 A = vec_max(A, max255);
1205 ad = vec_packsu(A, (vector int16_t) zero16);
1207 ad = vec_splats((
uint8_t) 255);
1210 vy32_l = vec_sub(vy32_l, y_offset);
1211 vy32_r = vec_sub(vy32_r, y_offset);
1212 vy32_l = vec_mul(vy32_l, y_coeff);
1213 vy32_r = vec_mul(vy32_r, y_coeff);
1214 vy32_l = vec_add(vy32_l, y_add);
1215 vy32_r = vec_add(vy32_r, y_add);
1218 vud32_l = vec_perm(vu32_l, vu32_l, doubleleft);
1219 vud32_r = vec_perm(vu32_l, vu32_l, doubleright);
1220 vvd32_l = vec_perm(vv32_l, vv32_l, doubleleft);
1221 vvd32_r = vec_perm(vv32_l, vv32_l, doubleright);
1223 R_l = vec_mul(vvd32_l, v2r_coeff);
1224 R_l = vec_add(R_l, vy32_l);
1225 R_r = vec_mul(vvd32_r, v2r_coeff);
1226 R_r = vec_add(R_r, vy32_r);
1227 G_l = vec_mul(vvd32_l, v2g_coeff);
1228 tmp32 = vec_mul(vud32_l, u2g_coeff);
1229 G_l = vec_add(G_l, vy32_l);
1230 G_l = vec_add(G_l, tmp32);
1231 G_r = vec_mul(vvd32_r, v2g_coeff);
1232 tmp32 = vec_mul(vud32_r, u2g_coeff);
1233 G_r = vec_add(G_r, vy32_r);
1234 G_r = vec_add(G_r, tmp32);
1236 B_l = vec_mul(vud32_l, u2b_coeff);
1237 B_l = vec_add(B_l, vy32_l);
1238 B_r = vec_mul(vud32_r, u2b_coeff);
1239 B_r = vec_add(B_r, vy32_r);
1244 vy = vec_ld(16, &buf0[i * 2]);
1245 vy32_l = vec_unpackh(vy);
1246 vy32_r = vec_unpackl(vy);
1247 vy32_l = vec_sl(vy32_l, shift2);
1248 vy32_r = vec_sl(vy32_r, shift2);
1250 vy32_l = vec_sub(vy32_l, y_offset);
1251 vy32_r = vec_sub(vy32_r, y_offset);
1252 vy32_l = vec_mul(vy32_l, y_coeff);
1253 vy32_r = vec_mul(vy32_r, y_coeff);
1254 vy32_l = vec_add(vy32_l, y_add);
1255 vy32_r = vec_add(vy32_r, y_add);
1258 vud32_l = vec_perm(vu32_r, vu32_r, doubleleft);
1259 vud32_r = vec_perm(vu32_r, vu32_r, doubleright);
1260 vvd32_l = vec_perm(vv32_r, vv32_r, doubleleft);
1261 vvd32_r = vec_perm(vv32_r, vv32_r, doubleright);
1263 R_l = vec_mul(vvd32_l, v2r_coeff);
1264 R_l = vec_add(R_l, vy32_l);
1265 R_r = vec_mul(vvd32_r, v2r_coeff);
1266 R_r = vec_add(R_r, vy32_r);
1267 G_l = vec_mul(vvd32_l, v2g_coeff);
1268 tmp32 = vec_mul(vud32_l, u2g_coeff);
1269 G_l = vec_add(G_l, vy32_l);
1270 G_l = vec_add(G_l, tmp32);
1271 G_r = vec_mul(vvd32_r, v2g_coeff);
1272 tmp32 = vec_mul(vud32_r, u2g_coeff);
1273 G_r = vec_add(G_r, vy32_r);
1274 G_r = vec_add(G_r, tmp32);
1276 B_l = vec_mul(vud32_l, u2b_coeff);
1277 B_l = vec_add(B_l, vy32_l);
1278 B_r = vec_mul(vud32_r, u2b_coeff);
1279 B_r = vec_add(B_r, vy32_r);
1287 #define YUV2RGBWRAPPERX(name, base, ext, fmt, hasAlpha) \ 1288 static void name ## ext ## _X_vsx(SwsContext *c, const int16_t *lumFilter, \ 1289 const int16_t **lumSrc, int lumFilterSize, \ 1290 const int16_t *chrFilter, const int16_t **chrUSrc, \ 1291 const int16_t **chrVSrc, int chrFilterSize, \ 1292 const int16_t **alpSrc, uint8_t *dest, int dstW, \ 1295 name ## base ## _X_vsx_template(c, lumFilter, lumSrc, lumFilterSize, \ 1296 chrFilter, chrUSrc, chrVSrc, chrFilterSize, \ 1297 alpSrc, dest, dstW, y, fmt, hasAlpha); \ 1300 #define YUV2RGBWRAPPERX2(name, base, ext, fmt, hasAlpha) \ 1301 static void name ## ext ## _2_vsx(SwsContext *c, const int16_t *buf[2], \ 1302 const int16_t *ubuf[2], const int16_t *vbuf[2], \ 1303 const int16_t *abuf[2], uint8_t *dest, int dstW, \ 1304 int yalpha, int uvalpha, int y) \ 1306 name ## base ## _2_vsx_template(c, buf, ubuf, vbuf, abuf, \ 1307 dest, dstW, yalpha, uvalpha, y, fmt, hasAlpha); \ 1310 #define YUV2RGBWRAPPER(name, base, ext, fmt, hasAlpha) \ 1311 static void name ## ext ## _1_vsx(SwsContext *c, const int16_t *buf0, \ 1312 const int16_t *ubuf[2], const int16_t *vbuf[2], \ 1313 const int16_t *abuf0, uint8_t *dest, int dstW, \ 1314 int uvalpha, int y) \ 1316 name ## base ## _1_vsx_template(c, buf0, ubuf, vbuf, abuf0, dest, \ 1317 dstW, uvalpha, y, fmt, hasAlpha); \ 1361 write422(const vector int16_t vy1, const vector int16_t vy2,
1362 const vector int16_t vu, const vector int16_t vv,
1367 0x0, 0x10, 0x1, 0x18,
1368 0x2, 0x11, 0x3, 0x19,
1369 0x4, 0x12, 0x5, 0x1a,
1370 0x6, 0x13, 0x7, 0x1b };
1372 0x8, 0x14, 0x9, 0x1c,
1373 0xa, 0x15, 0xb, 0x1d,
1374 0xc, 0x16, 0xd, 0x1e,
1375 0xe, 0x17, 0xf, 0x1f };
1377 0x0, 0x18, 0x1, 0x10,
1378 0x2, 0x19, 0x3, 0x11,
1379 0x4, 0x1a, 0x5, 0x12,
1380 0x6, 0x1b, 0x7, 0x13 };
1382 0x8, 0x1c, 0x9, 0x14,
1383 0xa, 0x1d, 0xb, 0x15,
1384 0xc, 0x1e, 0xd, 0x16,
1385 0xe, 0x1f, 0xf, 0x17 };
1387 0x10, 0x0, 0x18, 0x1,
1388 0x11, 0x2, 0x19, 0x3,
1389 0x12, 0x4, 0x1a, 0x5,
1390 0x13, 0x6, 0x1b, 0x7 };
1392 0x14, 0x8, 0x1c, 0x9,
1393 0x15, 0xa, 0x1d, 0xb,
1394 0x16, 0xc, 0x1e, 0xd,
1395 0x17, 0xe, 0x1f, 0xf };
1397 vd1 = vec_packsu(vy1, vy2);
1398 vd2 = vec_packsu(vu, vv);
1402 tmp = vec_perm(vd1, vd2, yuyv1);
1403 vec_st(tmp, 0, dest);
1404 tmp = vec_perm(vd1, vd2, yuyv2);
1405 vec_st(tmp, 16, dest);
1408 tmp = vec_perm(vd1, vd2, yvyu1);
1409 vec_st(tmp, 0, dest);
1410 tmp = vec_perm(vd1, vd2, yvyu2);
1411 vec_st(tmp, 16, dest);
1414 tmp = vec_perm(vd1, vd2, uyvy1);
1415 vec_st(tmp, 0, dest);
1416 tmp = vec_perm(vd1, vd2, uyvy2);
1417 vec_st(tmp, 16, dest);
1423 yuv2422_X_vsx_template(
SwsContext *c,
const int16_t *lumFilter,
1424 const int16_t **lumSrc,
int lumFilterSize,
1425 const int16_t *chrFilter,
const int16_t **chrUSrc,
1426 const int16_t **chrVSrc,
int chrFilterSize,
1427 const int16_t **alpSrc,
uint8_t *dest,
int dstW,
1431 vector int16_t vy1, vy2, vu, vv;
1432 vector
int32_t vy32[4], vu32[2], vv32[2],
tmp, tmp2, tmp3, tmp4;
1434 const vector
int32_t start = vec_splats(1 << 18);
1435 const vector uint32_t shift19 = vec_splats(19
U);
1437 for (i = 0; i < lumFilterSize; i++)
1438 vlumFilter[i] = vec_splats(lumFilter[i]);
1439 for (i = 0; i < chrFilterSize; i++)
1440 vchrFilter[i] = vec_splats(chrFilter[i]);
1442 for (i = 0; i < ((dstW + 1) >> 1); i += 8) {
1452 for (j = 0; j < lumFilterSize; j++) {
1453 vv = vec_ld(0, &lumSrc[j][i * 2]);
1454 tmp = vec_mule(vv, vlumFilter[j]);
1455 tmp2 = vec_mulo(vv, vlumFilter[j]);
1456 tmp3 = vec_mergeh(tmp, tmp2);
1457 tmp4 = vec_mergel(tmp, tmp2);
1459 vy32[0] = vec_adds(vy32[0], tmp3);
1460 vy32[1] = vec_adds(vy32[1], tmp4);
1462 vv = vec_ld(0, &lumSrc[j][(i + 4) * 2]);
1463 tmp = vec_mule(vv, vlumFilter[j]);
1464 tmp2 = vec_mulo(vv, vlumFilter[j]);
1465 tmp3 = vec_mergeh(tmp, tmp2);
1466 tmp4 = vec_mergel(tmp, tmp2);
1468 vy32[2] = vec_adds(vy32[2], tmp3);
1469 vy32[3] = vec_adds(vy32[3], tmp4);
1472 for (j = 0; j < chrFilterSize; j++) {
1473 vv = vec_ld(0, &chrUSrc[j][i]);
1474 tmp = vec_mule(vv, vchrFilter[j]);
1475 tmp2 = vec_mulo(vv, vchrFilter[j]);
1476 tmp3 = vec_mergeh(tmp, tmp2);
1477 tmp4 = vec_mergel(tmp, tmp2);
1479 vu32[0] = vec_adds(vu32[0], tmp3);
1480 vu32[1] = vec_adds(vu32[1], tmp4);
1482 vv = vec_ld(0, &chrVSrc[j][i]);
1483 tmp = vec_mule(vv, vchrFilter[j]);
1484 tmp2 = vec_mulo(vv, vchrFilter[j]);
1485 tmp3 = vec_mergeh(tmp, tmp2);
1486 tmp4 = vec_mergel(tmp, tmp2);
1488 vv32[0] = vec_adds(vv32[0], tmp3);
1489 vv32[1] = vec_adds(vv32[1], tmp4);
1492 for (j = 0; j < 4; j++) {
1493 vy32[j] = vec_sra(vy32[j], shift19);
1495 for (j = 0; j < 2; j++) {
1496 vu32[j] = vec_sra(vu32[j], shift19);
1497 vv32[j] = vec_sra(vv32[j], shift19);
1500 vy1 = vec_packs(vy32[0], vy32[1]);
1501 vy2 = vec_packs(vy32[2], vy32[3]);
1502 vu = vec_packs(vu32[0], vu32[1]);
1503 vv = vec_packs(vv32[0], vv32[1]);
1505 write422(vy1, vy2, vu, vv, &dest[i * 4], target);
1509 #define SETUP(x, buf0, buf1, alpha) { \ 1510 x = vec_ld(0, buf0); \ 1511 tmp = vec_mule(x, alpha); \ 1512 tmp2 = vec_mulo(x, alpha); \ 1513 tmp3 = vec_mergeh(tmp, tmp2); \ 1514 tmp4 = vec_mergel(tmp, tmp2); \ 1516 x = vec_ld(0, buf1); \ 1517 tmp = vec_mule(x, alpha); \ 1518 tmp2 = vec_mulo(x, alpha); \ 1519 tmp5 = vec_mergeh(tmp, tmp2); \ 1520 tmp6 = vec_mergel(tmp, tmp2); \ 1522 tmp3 = vec_add(tmp3, tmp5); \ 1523 tmp4 = vec_add(tmp4, tmp6); \ 1525 tmp3 = vec_sra(tmp3, shift19); \ 1526 tmp4 = vec_sra(tmp4, shift19); \ 1527 x = vec_packs(tmp3, tmp4); \ 1531 yuv2422_2_vsx_template(
SwsContext *c,
const int16_t *buf[2],
1532 const int16_t *ubuf[2],
const int16_t *vbuf[2],
1533 const int16_t *abuf[2],
uint8_t *dest,
int dstW,
1534 int yalpha,
int uvalpha,
int y,
1537 const int16_t *buf0 = buf[0], *buf1 = buf[1],
1538 *ubuf0 = ubuf[0], *ubuf1 = ubuf[1],
1539 *vbuf0 = vbuf[0], *vbuf1 = vbuf[1];
1540 const int16_t yalpha1 = 4096 - yalpha;
1541 const int16_t uvalpha1 = 4096 - uvalpha;
1542 vector int16_t vy1, vy2, vu, vv;
1543 vector
int32_t tmp, tmp2, tmp3, tmp4, tmp5, tmp6;
1544 const vector int16_t vyalpha1 = vec_splats(yalpha1);
1545 const vector int16_t vuvalpha1 = vec_splats(uvalpha1);
1546 const vector uint32_t shift19 = vec_splats(19
U);
1551 for (i = 0; i < ((dstW + 1) >> 1); i += 8) {
1553 SETUP(vy1, &buf0[i * 2], &buf1[i * 2], vyalpha1)
1554 SETUP(vy2, &buf0[(i + 4) * 2], &buf1[(i + 4) * 2], vyalpha1)
1555 SETUP(vu, &ubuf0[i], &ubuf1[i], vuvalpha1)
1556 SETUP(vv, &vbuf0[i], &vbuf1[i], vuvalpha1)
1558 write422(vy1, vy2, vu, vv, &dest[i * 4], target);
1565 yuv2422_1_vsx_template(
SwsContext *c,
const int16_t *buf0,
1566 const int16_t *ubuf[2],
const int16_t *vbuf[2],
1567 const int16_t *abuf0,
uint8_t *dest,
int dstW,
1570 const int16_t *ubuf0 = ubuf[0], *vbuf0 = vbuf[0];
1571 vector int16_t vy1, vy2, vu, vv,
tmp;
1572 const vector int16_t add64 = vec_splats((int16_t) 64);
1573 const vector int16_t add128 = vec_splats((int16_t) 128);
1574 const vector uint16_t shift7 = vec_splat_u16(7);
1575 const vector uint16_t shift8 = vec_splat_u16(8);
1578 if (uvalpha < 2048) {
1579 for (i = 0; i < ((dstW + 1) >> 1); i += 8) {
1580 vy1 = vec_ld(0, &buf0[i * 2]);
1581 vy2 = vec_ld(0, &buf0[(i + 4) * 2]);
1582 vu = vec_ld(0, &ubuf0[i]);
1583 vv = vec_ld(0, &vbuf0[i]);
1585 vy1 = vec_add(vy1, add64);
1586 vy2 = vec_add(vy2, add64);
1587 vu = vec_add(vu, add64);
1588 vv = vec_add(vv, add64);
1590 vy1 = vec_sra(vy1, shift7);
1591 vy2 = vec_sra(vy2, shift7);
1592 vu = vec_sra(vu, shift7);
1593 vv = vec_sra(vv, shift7);
1595 write422(vy1, vy2, vu, vv, &dest[i * 4], target);
1598 const int16_t *ubuf1 = ubuf[1], *vbuf1 = vbuf[1];
1599 for (i = 0; i < ((dstW + 1) >> 1); i += 8) {
1600 vy1 = vec_ld(0, &buf0[i * 2]);
1601 vy2 = vec_ld(0, &buf0[(i + 4) * 2]);
1602 vu = vec_ld(0, &ubuf0[i]);
1603 tmp = vec_ld(0, &ubuf1[i]);
1604 vu = vec_adds(vu, tmp);
1605 vv = vec_ld(0, &vbuf0[i]);
1606 tmp = vec_ld(0, &vbuf1[i]);
1607 vv = vec_adds(vv, tmp);
1609 vy1 = vec_add(vy1, add64);
1610 vy2 = vec_add(vy2, add64);
1611 vu = vec_adds(vu, add128);
1612 vv = vec_adds(vv, add128);
1614 vy1 = vec_sra(vy1, shift7);
1615 vy2 = vec_sra(vy2, shift7);
1616 vu = vec_sra(vu, shift8);
1617 vv = vec_sra(vv, shift8);
1619 write422(vy1, vy2, vu, vv, &dest[i * 4], target);
1624 #define YUV2PACKEDWRAPPERX(name, base, ext, fmt) \ 1625 static void name ## ext ## _X_vsx(SwsContext *c, const int16_t *lumFilter, \ 1626 const int16_t **lumSrc, int lumFilterSize, \ 1627 const int16_t *chrFilter, const int16_t **chrUSrc, \ 1628 const int16_t **chrVSrc, int chrFilterSize, \ 1629 const int16_t **alpSrc, uint8_t *dest, int dstW, \ 1632 name ## base ## _X_vsx_template(c, lumFilter, lumSrc, lumFilterSize, \ 1633 chrFilter, chrUSrc, chrVSrc, chrFilterSize, \ 1634 alpSrc, dest, dstW, y, fmt); \ 1637 #define YUV2PACKEDWRAPPER2(name, base, ext, fmt) \ 1638 YUV2PACKEDWRAPPERX(name, base, ext, fmt) \ 1639 static void name ## ext ## _2_vsx(SwsContext *c, const int16_t *buf[2], \ 1640 const int16_t *ubuf[2], const int16_t *vbuf[2], \ 1641 const int16_t *abuf[2], uint8_t *dest, int dstW, \ 1642 int yalpha, int uvalpha, int y) \ 1644 name ## base ## _2_vsx_template(c, buf, ubuf, vbuf, abuf, \ 1645 dest, dstW, yalpha, uvalpha, y, fmt); \ 1648 #define YUV2PACKEDWRAPPER(name, base, ext, fmt) \ 1649 YUV2PACKEDWRAPPER2(name, base, ext, fmt) \ 1650 static void name ## ext ## _1_vsx(SwsContext *c, const int16_t *buf0, \ 1651 const int16_t *ubuf[2], const int16_t *vbuf[2], \ 1652 const int16_t *abuf0, uint8_t *dest, int dstW, \ 1653 int uvalpha, int y) \ 1655 name ## base ## _1_vsx_template(c, buf0, ubuf, vbuf, \ 1656 abuf0, dest, dstW, uvalpha, \ 1664 static void hyscale_fast_vsx(
SwsContext *c, int16_t *dst,
int dstWidth,
1665 const uint8_t *src,
int srcW,
int xInc)
1668 unsigned int xpos = 0, xx;
1669 vector
uint8_t vin, vin2, vperm;
1670 vector int8_t vmul, valpha;
1671 vector int16_t vtmp, vtmp2, vtmp3, vtmp4;
1672 vector uint16_t vd_l, vd_r, vcoord16[2];
1673 vector uint32_t vcoord[4];
1674 const vector uint32_t vadd = (vector uint32_t) {
1680 const vector uint16_t vadd16 = (vector uint16_t) {
1690 const vector uint32_t vshift16 = vec_splats((uint32_t) 16);
1691 const vector uint16_t vshift9 = vec_splat_u16(9);
1692 const vector
uint8_t vzero = vec_splat_u8(0);
1693 const vector uint16_t vshift = vec_splat_u16(7);
1695 for (i = 0; i < dstWidth; i += 16) {
1696 vcoord16[0] = vec_splats((uint16_t) xpos);
1697 vcoord16[1] = vec_splats((uint16_t) (xpos + xInc * 8));
1699 vcoord16[0] = vec_add(vcoord16[0], vadd16);
1700 vcoord16[1] = vec_add(vcoord16[1], vadd16);
1702 vcoord16[0] = vec_sr(vcoord16[0], vshift9);
1703 vcoord16[1] = vec_sr(vcoord16[1], vshift9);
1704 valpha = (vector int8_t) vec_pack(vcoord16[0], vcoord16[1]);
1707 vin = vec_vsx_ld(0, &src[xx]);
1709 vcoord[0] = vec_splats(xpos & 0xffff);
1710 vcoord[1] = vec_splats((xpos & 0xffff) + xInc * 4);
1711 vcoord[2] = vec_splats((xpos & 0xffff) + xInc * 8);
1712 vcoord[3] = vec_splats((xpos & 0xffff) + xInc * 12);
1714 vcoord[0] = vec_add(vcoord[0], vadd);
1715 vcoord[1] = vec_add(vcoord[1], vadd);
1716 vcoord[2] = vec_add(vcoord[2], vadd);
1717 vcoord[3] = vec_add(vcoord[3], vadd);
1719 vcoord[0] = vec_sr(vcoord[0], vshift16);
1720 vcoord[1] = vec_sr(vcoord[1], vshift16);
1721 vcoord[2] = vec_sr(vcoord[2], vshift16);
1722 vcoord[3] = vec_sr(vcoord[3], vshift16);
1724 vcoord16[0] = vec_pack(vcoord[0], vcoord[1]);
1725 vcoord16[1] = vec_pack(vcoord[2], vcoord[3]);
1726 vperm = vec_pack(vcoord16[0], vcoord16[1]);
1728 vin = vec_perm(vin, vin, vperm);
1730 vin2 = vec_vsx_ld(1, &src[xx]);
1731 vin2 = vec_perm(vin2, vin2, vperm);
1733 vmul = (vector int8_t) vec_sub(vin2, vin);
1734 vtmp = vec_mule(vmul, valpha);
1735 vtmp2 = vec_mulo(vmul, valpha);
1736 vtmp3 = vec_mergeh(vtmp, vtmp2);
1737 vtmp4 = vec_mergel(vtmp, vtmp2);
1739 vd_l = (vector uint16_t) vec_mergeh(vin, vzero);
1740 vd_r = (vector uint16_t) vec_mergel(vin, vzero);
1741 vd_l = vec_sl(vd_l, vshift);
1742 vd_r = vec_sl(vd_r, vshift);
1744 vd_l = vec_add(vd_l, (vector uint16_t) vtmp3);
1745 vd_r = vec_add(vd_r, (vector uint16_t) vtmp4);
1747 vec_st((vector int16_t) vd_l, 0, &dst[i]);
1748 vec_st((vector int16_t) vd_r, 0, &dst[i + 8]);
1752 for (i=dstWidth-1; (i*xInc)>>16 >=srcW-1; i--)
1753 dst[i] = src[srcW-1]*128;
1756 #define HCSCALE(in, out) \ 1757 vin = vec_vsx_ld(0, &in[xx]); \ 1758 vin = vec_perm(vin, vin, vperm); \ 1760 vin2 = vec_vsx_ld(1, &in[xx]); \ 1761 vin2 = vec_perm(vin2, vin2, vperm); \ 1763 vtmp = vec_mule(vin, valphaxor); \ 1764 vtmp2 = vec_mulo(vin, valphaxor); \ 1765 vtmp3 = vec_mergeh(vtmp, vtmp2); \ 1766 vtmp4 = vec_mergel(vtmp, vtmp2); \ 1768 vtmp = vec_mule(vin2, valpha); \ 1769 vtmp2 = vec_mulo(vin2, valpha); \ 1770 vd_l = vec_mergeh(vtmp, vtmp2); \ 1771 vd_r = vec_mergel(vtmp, vtmp2); \ 1773 vd_l = vec_add(vd_l, vtmp3); \ 1774 vd_r = vec_add(vd_r, vtmp4); \ 1776 vec_st((vector int16_t) vd_l, 0, &out[i]); \ 1777 vec_st((vector int16_t) vd_r, 0, &out[i + 8]) 1779 static void hcscale_fast_vsx(
SwsContext *c, int16_t *dst1, int16_t *dst2,
1781 const uint8_t *src2,
int srcW,
int xInc)
1784 unsigned int xpos = 0, xx;
1785 vector
uint8_t vin, vin2, vperm;
1786 vector
uint8_t valpha, valphaxor;
1787 vector uint16_t vtmp, vtmp2, vtmp3, vtmp4;
1788 vector uint16_t vd_l, vd_r, vcoord16[2];
1789 vector uint32_t vcoord[4];
1791 const vector uint32_t vadd = (vector uint32_t) {
1797 const vector uint16_t vadd16 = (vector uint16_t) {
1807 const vector uint32_t vshift16 = vec_splats((uint32_t) 16);
1808 const vector uint16_t vshift9 = vec_splat_u16(9);
1810 for (i = 0; i < dstWidth; i += 16) {
1811 vcoord16[0] = vec_splats((uint16_t) xpos);
1812 vcoord16[1] = vec_splats((uint16_t) (xpos + xInc * 8));
1814 vcoord16[0] = vec_add(vcoord16[0], vadd16);
1815 vcoord16[1] = vec_add(vcoord16[1], vadd16);
1817 vcoord16[0] = vec_sr(vcoord16[0], vshift9);
1818 vcoord16[1] = vec_sr(vcoord16[1], vshift9);
1819 valpha = vec_pack(vcoord16[0], vcoord16[1]);
1820 valphaxor = vec_xor(valpha, vxor);
1824 vcoord[0] = vec_splats(xpos & 0xffff);
1825 vcoord[1] = vec_splats((xpos & 0xffff) + xInc * 4);
1826 vcoord[2] = vec_splats((xpos & 0xffff) + xInc * 8);
1827 vcoord[3] = vec_splats((xpos & 0xffff) + xInc * 12);
1829 vcoord[0] = vec_add(vcoord[0], vadd);
1830 vcoord[1] = vec_add(vcoord[1], vadd);
1831 vcoord[2] = vec_add(vcoord[2], vadd);
1832 vcoord[3] = vec_add(vcoord[3], vadd);
1834 vcoord[0] = vec_sr(vcoord[0], vshift16);
1835 vcoord[1] = vec_sr(vcoord[1], vshift16);
1836 vcoord[2] = vec_sr(vcoord[2], vshift16);
1837 vcoord[3] = vec_sr(vcoord[3], vshift16);
1839 vcoord16[0] = vec_pack(vcoord[0], vcoord[1]);
1840 vcoord16[1] = vec_pack(vcoord[2], vcoord[3]);
1841 vperm = vec_pack(vcoord16[0], vcoord16[1]);
1843 HCSCALE(src1, dst1);
1844 HCSCALE(src2, dst2);
1848 for (i=dstWidth-1; (i*xInc)>>16 >=srcW-1; i--) {
1849 dst1[
i] = src1[srcW-1]*128;
1850 dst2[
i] = src2[srcW-1]*128;
1856 static void hScale8To19_vsx(
SwsContext *c, int16_t *_dst,
int dstW,
1857 const uint8_t *src,
const int16_t *filter,
1858 const int32_t *filterPos,
int filterSize)
1862 vector int16_t vfilter, vin;
1865 const vector
uint8_t vzero = vec_splat_u8(0);
1866 const vector
uint8_t vunusedtab[8] = {
1867 (vector
uint8_t) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
1868 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf},
1869 (vector
uint8_t) {0x0, 0x1, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10,
1870 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
1871 (vector
uint8_t) {0x0, 0x1, 0x2, 0x3, 0x10, 0x10, 0x10, 0x10,
1872 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
1873 (vector
uint8_t) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x10, 0x10,
1874 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
1875 (vector
uint8_t) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
1876 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
1877 (vector
uint8_t) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
1878 0x8, 0x9, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
1879 (vector
uint8_t) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
1880 0x8, 0x9, 0xa, 0xb, 0x10, 0x10, 0x10, 0x10},
1881 (vector
uint8_t) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
1882 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0x10, 0x10},
1884 const vector
uint8_t vunused = vunusedtab[filterSize % 8];
1886 if (filterSize == 1) {
1887 for (i = 0; i < dstW; i++) {
1888 int srcPos = filterPos[
i];
1890 for (j = 0; j < filterSize; j++) {
1891 val += ((
int)src[srcPos + j]) * filter[filterSize * i + j];
1893 dst[
i] =
FFMIN(val >> 3, (1 << 19) - 1);
1896 for (i = 0; i < dstW; i++) {
1897 const int srcPos = filterPos[
i];
1898 vout = vec_splat_s32(0);
1899 for (j = 0; j < filterSize; j += 8) {
1900 vin8 = vec_vsx_ld(0, &src[srcPos + j]);
1901 vin = (vector int16_t) vec_mergeh(vin8, vzero);
1902 if (j + 8 > filterSize)
1903 vin = vec_perm(vin, (vector int16_t) vzero, vunused);
1905 vfilter = vec_vsx_ld(0, &filter[filterSize * i + j]);
1906 vout = vec_msums(vin, vfilter, vout);
1908 vout = vec_sums(vout, (vector
int32_t) vzero);
1909 dst[
i] =
FFMIN(vout[3] >> 3, (1 << 19) - 1);
1914 static void hScale16To19_vsx(
SwsContext *c, int16_t *_dst,
int dstW,
1915 const uint8_t *_src,
const int16_t *filter,
1916 const int32_t *filterPos,
int filterSize)
1921 const uint16_t *src = (
const uint16_t *) _src;
1924 vector int16_t vfilter, vin;
1925 vector
int32_t vout, vtmp, vtmp2, vfilter32_l, vfilter32_r;
1926 const vector
uint8_t vzero = vec_splat_u8(0);
1927 const vector
uint8_t vunusedtab[8] = {
1928 (vector
uint8_t) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
1929 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf},
1930 (vector
uint8_t) {0x0, 0x1, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10,
1931 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
1932 (vector
uint8_t) {0x0, 0x1, 0x2, 0x3, 0x10, 0x10, 0x10, 0x10,
1933 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
1934 (vector
uint8_t) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x10, 0x10,
1935 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
1936 (vector
uint8_t) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
1937 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
1938 (vector
uint8_t) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
1939 0x8, 0x9, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
1940 (vector
uint8_t) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
1941 0x8, 0x9, 0xa, 0xb, 0x10, 0x10, 0x10, 0x10},
1942 (vector
uint8_t) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
1943 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0x10, 0x10},
1945 const vector
uint8_t vunused = vunusedtab[filterSize % 8];
1953 if (filterSize == 1) {
1954 for (i = 0; i < dstW; i++) {
1955 int srcPos = filterPos[
i];
1958 for (j = 0; j < filterSize; j++) {
1959 val += src[srcPos + j] * filter[filterSize * i + j];
1962 dst[
i] =
FFMIN(val >> sh, (1 << 19) - 1);
1965 for (i = 0; i < dstW; i++) {
1966 const int srcPos = filterPos[
i];
1967 vout = vec_splat_s32(0);
1968 for (j = 0; j < filterSize; j += 8) {
1969 vin = (vector int16_t) vec_vsx_ld(0, &src[srcPos + j]);
1970 if (j + 8 > filterSize)
1971 vin = vec_perm(vin, (vector int16_t) vzero, vunused);
1973 vfilter = vec_vsx_ld(0, &filter[filterSize * i + j]);
1974 vfilter32_l = vec_unpackh(vfilter);
1975 vfilter32_r = vec_unpackl(vfilter);
1977 vtmp = (vector
int32_t) vec_mergeh(vin, (vector int16_t) vzero);
1978 vtmp2 = (vector
int32_t) vec_mergel(vin, (vector int16_t) vzero);
1980 vtmp = vec_mul(vtmp, vfilter32_l);
1981 vtmp2 = vec_mul(vtmp2, vfilter32_r);
1983 vout = vec_adds(vout, vtmp);
1984 vout = vec_adds(vout, vtmp2);
1986 vout = vec_sums(vout, (vector
int32_t) vzero);
1987 dst[
i] =
FFMIN(vout[3] >> sh, (1 << 19) - 1);
1992 static void hScale16To15_vsx(
SwsContext *c, int16_t *dst,
int dstW,
1993 const uint8_t *_src,
const int16_t *filter,
1994 const int32_t *filterPos,
int filterSize)
1998 const uint16_t *src = (
const uint16_t *) _src;
2000 vector int16_t vfilter, vin;
2001 vector
int32_t vout, vtmp, vtmp2, vfilter32_l, vfilter32_r;
2002 const vector
uint8_t vzero = vec_splat_u8(0);
2003 const vector
uint8_t vunusedtab[8] = {
2004 (vector
uint8_t) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
2005 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf},
2006 (vector
uint8_t) {0x0, 0x1, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10,
2007 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
2008 (vector
uint8_t) {0x0, 0x1, 0x2, 0x3, 0x10, 0x10, 0x10, 0x10,
2009 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
2010 (vector
uint8_t) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x10, 0x10,
2011 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
2012 (vector
uint8_t) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
2013 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
2014 (vector
uint8_t) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
2015 0x8, 0x9, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
2016 (vector
uint8_t) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
2017 0x8, 0x9, 0xa, 0xb, 0x10, 0x10, 0x10, 0x10},
2018 (vector
uint8_t) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
2019 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0x10, 0x10},
2021 const vector
uint8_t vunused = vunusedtab[filterSize % 8];
2029 if (filterSize == 1) {
2030 for (i = 0; i < dstW; i++) {
2031 int srcPos = filterPos[
i];
2034 for (j = 0; j < filterSize; j++) {
2035 val += src[srcPos + j] * filter[filterSize * i + j];
2038 dst[
i] =
FFMIN(val >> sh, (1 << 15) - 1);
2041 for (i = 0; i < dstW; i++) {
2042 const int srcPos = filterPos[
i];
2043 vout = vec_splat_s32(0);
2044 for (j = 0; j < filterSize; j += 8) {
2045 vin = (vector int16_t) vec_vsx_ld(0, &src[srcPos + j]);
2046 if (j + 8 > filterSize)
2047 vin = vec_perm(vin, (vector int16_t) vzero, vunused);
2049 vfilter = vec_vsx_ld(0, &filter[filterSize * i + j]);
2050 vfilter32_l = vec_unpackh(vfilter);
2051 vfilter32_r = vec_unpackl(vfilter);
2053 vtmp = (vector
int32_t) vec_mergeh(vin, (vector int16_t) vzero);
2054 vtmp2 = (vector
int32_t) vec_mergel(vin, (vector int16_t) vzero);
2056 vtmp = vec_mul(vtmp, vfilter32_l);
2057 vtmp2 = vec_mul(vtmp2, vfilter32_r);
2059 vout = vec_adds(vout, vtmp);
2060 vout = vec_adds(vout, vtmp2);
2062 vout = vec_sums(vout, (vector
int32_t) vzero);
2063 dst[
i] =
FFMIN(vout[3] >> sh, (1 << 15) - 1);
2113 c->
yuv2plane1 =
isBE(dstFormat) ? yuv2plane1_9BE_vsx : yuv2plane1_9LE_vsx;
2114 c->
yuv2planeX =
isBE(dstFormat) ? yuv2planeX_9BE_vsx : yuv2planeX_9LE_vsx;
2117 c->
yuv2plane1 =
isBE(dstFormat) ? yuv2plane1_10BE_vsx : yuv2plane1_10LE_vsx;
2118 c->
yuv2planeX =
isBE(dstFormat) ? yuv2planeX_10BE_vsx : yuv2planeX_10LE_vsx;
2121 c->
yuv2plane1 =
isBE(dstFormat) ? yuv2plane1_12BE_vsx : yuv2plane1_12LE_vsx;
2122 c->
yuv2planeX =
isBE(dstFormat) ? yuv2planeX_12BE_vsx : yuv2planeX_12LE_vsx;
2125 c->
yuv2plane1 =
isBE(dstFormat) ? yuv2plane1_14BE_vsx : yuv2plane1_14LE_vsx;
2126 c->
yuv2planeX =
isBE(dstFormat) ? yuv2planeX_14BE_vsx : yuv2planeX_14LE_vsx;
2129 c->
yuv2plane1 =
isBE(dstFormat) ? yuv2plane1_16BE_vsx : yuv2plane1_16LE_vsx;
2131 if (cpu_flags & AV_CPU_FLAG_POWER8) {
2132 c->
yuv2planeX =
isBE(dstFormat) ? yuv2planeX_16BE_vsx : yuv2planeX_16LE_vsx;
2145 switch (dstFormat) {
2198 switch (dstFormat) {
packed YUV 4:2:2, 16bpp, Cb Y0 Cr Y1
IEEE-754 single precision Y, 32bpp, big-endian.
void(* hcScale)(struct SwsContext *c, int16_t *dst, int dstW, const uint8_t *src, const int16_t *filter, const int32_t *filterPos, int filterSize)
const char const char void * val
#define YUV2RGBWRAPPERX2(name, base, ext, fmt, hasAlpha)
static av_always_inline int isAnyRGB(enum AVPixelFormat pix_fmt)
static int shift(int a, int b)
const AVPixFmtDescriptor * av_pix_fmt_desc_get(enum AVPixelFormat pix_fmt)
packed RGB 8:8:8, 24bpp, RGBRGB...
#define YUV2RGBWRAPPERX(name, base, ext, fmt, hasAlpha)
static av_always_inline int is16BPS(enum AVPixelFormat pix_fmt)
void(* hyScale)(struct SwsContext *c, int16_t *dst, int dstW, const uint8_t *src, const int16_t *filter, const int32_t *filterPos, int filterSize)
Scale one horizontal line of input data using a filter over the input lines, to produce one (differen...
static atomic_int cpu_flags
void(* hyscale_fast)(struct SwsContext *c, int16_t *dst, int dstWidth, const uint8_t *src, int srcW, int xInc)
Scale one horizontal line of input data using a bilinear filter to produce one line of output data...
Macro definitions for various function/variable attributes.
static void filter(int16_t *output, ptrdiff_t out_stride, int16_t *low, ptrdiff_t low_stride, int16_t *high, ptrdiff_t high_stride, int len, int clip)
AVComponentDescriptor comp[4]
Parameters that describe how pixels are packed.
#define av_assert2(cond)
assert() equivalent, that does lie in speed critical code.
8 bits with AV_PIX_FMT_RGB32 palette
#define AV_PIX_FMT_FLAG_FLOAT
The pixel format contains IEEE-754 floating point values.
#define SWS_FULL_CHR_H_INT
packed ABGR 8:8:8:8, 32bpp, ABGRABGR...
#define SWS_FAST_BILINEAR
enum AVPixelFormat dstFormat
Destination pixel format.
yuv2packedX_fn yuv2packedX
#define i(width, name, range_min, range_max)
packed BGRA 8:8:8:8, 32bpp, BGRABGRA...
static const uint8_t dither[8][8]
yuv2packed1_fn yuv2packed1
static const uint8_t offset[127][2]
static av_always_inline int isSemiPlanarYUV(enum AVPixelFormat pix_fmt)
void(* hcscale_fast)(struct SwsContext *c, int16_t *dst1, int16_t *dst2, int dstWidth, const uint8_t *src1, const uint8_t *src2, int srcW, int xInc)
packed ARGB 8:8:8:8, 32bpp, ARGBARGB...
int chrDstW
Width of destination chroma planes.
packed RGBA 8:8:8:8, 32bpp, RGBARGBA...
uint64_t flags
Combination of AV_PIX_FMT_FLAG_...
yuv2planar1_fn yuv2plane1
packed YUV 4:2:2, 16bpp, Y0 Cr Y1 Cb
packed RGB 8:8:8, 24bpp, BGRBGR...
int dstW
Width of destination luma/alpha planes.
static av_always_inline int isBE(enum AVPixelFormat pix_fmt)
Descriptor that unambiguously describes how the bits of a pixel are stored in the up to 4 data planes...
yuv2planarX_fn yuv2planeX
packed YUV 4:2:2, 16bpp, Y0 Cb Y1 Cr
#define AV_CPU_FLAG_VSX
ISA 2.06.
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
#define YUV2RGBWRAPPER(name, base, ext, fmt, hasAlpha)
Contains misc utility macros and inline functions.
#define AV_CPU_FLAG_POWER8
ISA 2.07.
yuv2NBPS(yuv2NBPS(9, yuv2NBPS(BE, yuv2NBPS(1, yuv2NBPS(10, int16_t)
yuv2packed2_fn yuv2packed2
#define LOCAL_ALIGNED(a, t, v,...)
static double clip(void *opaque, double val)
Clip value val in the minval - maxval range.
static const int shift2[6]
static const uint8_t shifts[2][12]
enum AVPixelFormat srcFormat
Source pixel format.
#define output_pixel(pos, val, bias, signedness)
av_cold void ff_sws_init_swscale_vsx(SwsContext *c)
int chrSrcW
Width of source chroma planes.
int depth
Number of bits in the component.
IEEE-754 single precision Y, 32bpp, little-endian.
#define YUV2PACKEDWRAPPER(name, base, ext, fmt)
int srcW
Width of source luma/alpha planes.
int flags
Flags passed by the user to select scaler algorithm, optimizations, subsampling, etc...
AVPixelFormat
Pixel format.
static av_always_inline int isNBPS(enum AVPixelFormat pix_fmt)