104 #undef PROFILE_THE_BEAST
107 typedef unsigned char ubyte;
108 typedef signed char sbyte;
145 static const vector
unsigned char
146 perm_rgb_0 = { 0x00, 0x01, 0x10, 0x02, 0x03, 0x11, 0x04, 0x05,
147 0x12, 0x06, 0x07, 0x13, 0x08, 0x09, 0x14, 0x0a },
148 perm_rgb_1 = { 0x0b, 0x15, 0x0c, 0x0d, 0x16, 0x0e, 0x0f, 0x17,
149 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f },
150 perm_rgb_2 = { 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17,
151 0x00, 0x01, 0x18, 0x02, 0x03, 0x19, 0x04, 0x05 },
152 perm_rgb_3 = { 0x1a, 0x06, 0x07, 0x1b, 0x08, 0x09, 0x1c, 0x0a,
153 0x0b, 0x1d, 0x0c, 0x0d, 0x1e, 0x0e, 0x0f, 0x1f };
155 #define vec_merge3(x2, x1, x0, y0, y1, y2) \
157 __typeof__(x0) o0, o2, o3; \
158 o0 = vec_mergeh(x0, x1); \
159 y0 = vec_perm(o0, x2, perm_rgb_0); \
160 o2 = vec_perm(o0, x2, perm_rgb_1); \
161 o3 = vec_mergel(x0, x1); \
162 y1 = vec_perm(o3, o2, perm_rgb_2); \
163 y2 = vec_perm(o3, o2, perm_rgb_3); \
166 #define vec_mstbgr24(x0, x1, x2, ptr) \
168 __typeof__(x0) _0, _1, _2; \
169 vec_merge3(x0, x1, x2, _0, _1, _2); \
170 vec_st(_0, 0, ptr++); \
171 vec_st(_1, 0, ptr++); \
172 vec_st(_2, 0, ptr++); \
175 #define vec_mstrgb24(x0, x1, x2, ptr) \
177 __typeof__(x0) _0, _1, _2; \
178 vec_merge3(x2, x1, x0, _0, _1, _2); \
179 vec_st(_0, 0, ptr++); \
180 vec_st(_1, 0, ptr++); \
181 vec_st(_2, 0, ptr++); \
188 #define vec_mstrgb32(T, x0, x1, x2, x3, ptr) \
191 _0 = vec_mergeh(x0, x1); \
192 _1 = vec_mergeh(x2, x3); \
193 _2 = (T) vec_mergeh((vector unsigned short) _0, \
194 (vector unsigned short) _1); \
195 _3 = (T) vec_mergel((vector unsigned short) _0, \
196 (vector unsigned short) _1); \
197 vec_st(_2, 0 * 16, (T *) ptr); \
198 vec_st(_3, 1 * 16, (T *) ptr); \
199 _0 = vec_mergel(x0, x1); \
200 _1 = vec_mergel(x2, x3); \
201 _2 = (T) vec_mergeh((vector unsigned short) _0, \
202 (vector unsigned short) _1); \
203 _3 = (T) vec_mergel((vector unsigned short) _0, \
204 (vector unsigned short) _1); \
205 vec_st(_2, 2 * 16, (T *) ptr); \
206 vec_st(_3, 3 * 16, (T *) ptr); \
225 (vector signed short) \
226 vec_perm(x, (__typeof__(x)) { 0 }, \
227 ((vector unsigned char) { \
228 0x10, 0x00, 0x10, 0x01, 0x10, 0x02, 0x10, 0x03, \
229 0x10, 0x04, 0x10, 0x05, 0x10, 0x06, 0x10, 0x07 }))
232 (vector signed short) \
233 vec_perm(x, (__typeof__(x)) { 0 }, \
234 ((vector unsigned char) { \
235 0x10, 0x08, 0x10, 0x09, 0x10, 0x0A, 0x10, 0x0B, \
236 0x10, 0x0C, 0x10, 0x0D, 0x10, 0x0E, 0x10, 0x0F }))
238 #define vec_unh(x)(vector signed short) vec_mergeh(x,(__typeof__(x)) { 0 })
239 #define vec_unl(x)(vector signed short) vec_mergel(x,(__typeof__(x)) { 0 })
242 #define vec_clip_s16(x) \
243 vec_max(vec_min(x, ((vector signed short) { \
244 235, 235, 235, 235, 235, 235, 235, 235 })), \
245 ((vector signed short) { 16, 16, 16, 16, 16, 16, 16, 16 }))
247 #define vec_packclp(x, y) \
248 (vector unsigned char) \
249 vec_packs((vector unsigned short) \
250 vec_max(x, ((vector signed short) { 0 })), \
251 (vector unsigned short) \
252 vec_max(y, ((vector signed short) { 0 })))
254 static inline void cvtyuvtoRGB(
SwsContext *
c, vector
signed short Y,
255 vector
signed short U, vector
signed short V,
256 vector
signed short *
R, vector
signed short *
G,
257 vector
signed short *
B)
259 vector
signed short vx, ux, uvx;
261 Y = vec_mradds(
Y,
c->CY,
c->OY);
262 U = vec_sub(
U, (vector
signed short)
263 vec_splat((vector
signed short) { 128 }, 0));
264 V = vec_sub(
V, (vector
signed short)
265 vec_splat((vector
signed short) { 128 }, 0));
268 ux = vec_sl(
U,
c->CSHIFT);
269 *
B = vec_mradds(ux,
c->CBU,
Y);
272 vx = vec_sl(
V,
c->CSHIFT);
273 *
R = vec_mradds(vx,
c->CRV,
Y);
276 uvx = vec_mradds(
U,
c->CGU,
Y);
277 *
G = vec_mradds(
V,
c->CGV, uvx);
286 #define DEFCSP420_CVT(name, out_pixels) \
287 static int altivec_ ## name(SwsContext *c, const unsigned char **in, \
288 int *instrides, int srcSliceY, int srcSliceH, \
289 unsigned char **oplanes, int *outstrides) \
294 int instrides_scl[3]; \
295 vector unsigned char y0, y1; \
297 vector signed char u, v; \
299 vector signed short Y0, Y1, Y2, Y3; \
300 vector signed short U, V; \
301 vector signed short vx, ux, uvx; \
302 vector signed short vx0, ux0, uvx0; \
303 vector signed short vx1, ux1, uvx1; \
304 vector signed short R0, G0, B0; \
305 vector signed short R1, G1, B1; \
306 vector unsigned char R, G, B; \
308 vector signed short lCY = c->CY; \
309 vector signed short lOY = c->OY; \
310 vector signed short lCRV = c->CRV; \
311 vector signed short lCBU = c->CBU; \
312 vector signed short lCGU = c->CGU; \
313 vector signed short lCGV = c->CGV; \
314 vector unsigned short lCSHIFT = c->CSHIFT; \
316 const ubyte *y1i = in[0]; \
317 const ubyte *y2i = in[0] + instrides[0]; \
318 const ubyte *ui = in[1]; \
319 const ubyte *vi = in[2]; \
321 vector unsigned char *oute, *outo; \
324 instrides_scl[0] = instrides[0] * 2 - w; \
326 instrides_scl[1] = instrides[1] - w / 2; \
328 instrides_scl[2] = instrides[2] - w / 2; \
330 for (i = 0; i < h / 2; i++) { \
331 oute = (vector unsigned char *)(oplanes[0] + outstrides[0] * \
332 (srcSliceY + i * 2)); \
333 outo = oute + (outstrides[0] >> 4); \
334 vec_dstst(outo, (0x02000002 | (((w * 3 + 32) / 32) << 16)), 0); \
335 vec_dstst(oute, (0x02000002 | (((w * 3 + 32) / 32) << 16)), 1); \
337 for (j = 0; j < w / 16; j++) { \
338 y0 = vec_xl(0, y1i); \
340 y1 = vec_xl(0, y2i); \
342 u = (vector signed char) vec_xl(0, ui); \
344 v = (vector signed char) vec_xl(0, vi); \
346 u = (vector signed char) \
348 (vector signed char) \
349 vec_splat((vector signed char) { 128 }, 0)); \
350 v = (vector signed char) \
352 (vector signed char) \
353 vec_splat((vector signed char) { 128 }, 0)); \
355 U = vec_unpackh(u); \
356 V = vec_unpackh(v); \
363 Y0 = vec_mradds(Y0, lCY, lOY); \
364 Y1 = vec_mradds(Y1, lCY, lOY); \
365 Y2 = vec_mradds(Y2, lCY, lOY); \
366 Y3 = vec_mradds(Y3, lCY, lOY); \
369 ux = vec_sl(U, lCSHIFT); \
370 ux = vec_mradds(ux, lCBU, (vector signed short) { 0 }); \
371 ux0 = vec_mergeh(ux, ux); \
372 ux1 = vec_mergel(ux, ux); \
375 vx = vec_sl(V, lCSHIFT); \
376 vx = vec_mradds(vx, lCRV, (vector signed short) { 0 }); \
377 vx0 = vec_mergeh(vx, vx); \
378 vx1 = vec_mergel(vx, vx); \
381 uvx = vec_mradds(U, lCGU, (vector signed short) { 0 }); \
382 uvx = vec_mradds(V, lCGV, uvx); \
383 uvx0 = vec_mergeh(uvx, uvx); \
384 uvx1 = vec_mergel(uvx, uvx); \
386 R0 = vec_add(Y0, vx0); \
387 G0 = vec_add(Y0, uvx0); \
388 B0 = vec_add(Y0, ux0); \
389 R1 = vec_add(Y1, vx1); \
390 G1 = vec_add(Y1, uvx1); \
391 B1 = vec_add(Y1, ux1); \
393 R = vec_packclp(R0, R1); \
394 G = vec_packclp(G0, G1); \
395 B = vec_packclp(B0, B1); \
397 out_pixels(R, G, B, oute); \
399 R0 = vec_add(Y2, vx0); \
400 G0 = vec_add(Y2, uvx0); \
401 B0 = vec_add(Y2, ux0); \
402 R1 = vec_add(Y3, vx1); \
403 G1 = vec_add(Y3, uvx1); \
404 B1 = vec_add(Y3, ux1); \
405 R = vec_packclp(R0, R1); \
406 G = vec_packclp(G0, G1); \
407 B = vec_packclp(B0, B1); \
410 out_pixels(R, G, B, outo); \
418 ui += instrides_scl[1]; \
419 vi += instrides_scl[2]; \
420 y1i += instrides_scl[0]; \
421 y2i += instrides_scl[0]; \
426 #define out_abgr(a, b, c, ptr) \
427 vec_mstrgb32(__typeof__(a), ((__typeof__(a)) { 255 }), c, b, a, ptr)
428 #define out_bgra(a, b, c, ptr) \
429 vec_mstrgb32(__typeof__(a), c, b, a, ((__typeof__(a)) { 255 }), ptr)
430 #define out_rgba(a, b, c, ptr) \
431 vec_mstrgb32(__typeof__(a), a, b, c, ((__typeof__(a)) { 255 }), ptr)
432 #define out_argb(a, b, c, ptr) \
433 vec_mstrgb32(__typeof__(a), ((__typeof__(a)) { 255 }), a, b, c, ptr)
434 #define out_rgb24(a, b, c, ptr) vec_mstrgb24(a, b, c, ptr)
435 #define out_bgr24(a, b, c, ptr) vec_mstbgr24(a, b, c, ptr)
437 DEFCSP420_CVT(yuv2_abgr, out_abgr)
438 DEFCSP420_CVT(yuv2_bgra, out_bgra)
439 DEFCSP420_CVT(yuv2_rgba, out_rgba)
440 DEFCSP420_CVT(yuv2_argb, out_argb)
441 DEFCSP420_CVT(yuv2_rgb24, out_rgb24)
442 DEFCSP420_CVT(yuv2_bgr24, out_bgr24)
446 static const vector
unsigned char
447 demux_u = { 0x10, 0x00, 0x10, 0x00,
448 0x10, 0x04, 0x10, 0x04,
449 0x10, 0x08, 0x10, 0x08,
450 0x10, 0x0c, 0x10, 0x0c },
451 demux_v = { 0x10, 0x02, 0x10, 0x02,
452 0x10, 0x06, 0x10, 0x06,
453 0x10, 0x0A, 0x10, 0x0A,
454 0x10, 0x0E, 0x10, 0x0E },
455 demux_y = { 0x10, 0x01, 0x10, 0x03,
456 0x10, 0x05, 0x10, 0x07,
457 0x10, 0x09, 0x10, 0x0B,
458 0x10, 0x0D, 0x10, 0x0F };
463 static int altivec_uyvy_rgb32(
SwsContext *
c,
const unsigned char **
in,
464 int *instrides,
int srcSliceY,
int srcSliceH,
465 unsigned char **oplanes,
int *outstrides)
470 vector
unsigned char uyvy;
471 vector
signed short Y,
U,
V;
472 vector
signed short R0, G0,
B0,
R1, G1,
B1;
473 vector
unsigned char R,
G,
B;
474 vector
unsigned char *
out;
478 out = (vector
unsigned char *) (oplanes[0] + srcSliceY * outstrides[0]);
480 for (
i = 0;
i <
h;
i++)
481 for (j = 0; j <
w / 16; j++) {
482 uyvy = vec_ld(0,
img);
485 vec_perm(uyvy, (vector
unsigned char) { 0 }, demux_u);
487 vec_perm(uyvy, (vector
unsigned char) { 0 }, demux_v);
489 vec_perm(uyvy, (vector
unsigned char) { 0 }, demux_y);
491 cvtyuvtoRGB(
c,
Y,
U,
V, &
R0, &G0, &
B0);
493 uyvy = vec_ld(16,
img);
496 vec_perm(uyvy, (vector
unsigned char) { 0 }, demux_u);
498 vec_perm(uyvy, (vector
unsigned char) { 0 }, demux_v);
500 vec_perm(uyvy, (vector
unsigned char) { 0 }, demux_y);
502 cvtyuvtoRGB(
c,
Y,
U,
V, &
R1, &G1, &
B1);
504 R = vec_packclp(
R0,
R1);
505 G = vec_packclp(G0, G1);
506 B = vec_packclp(
B0,
B1);
537 if ((
c->srcW & 0
xf) != 0)
540 switch (
c->srcFormat) {
547 if ((
c->srcH & 0x1) != 0)
550 switch (
c->dstFormat) {
553 return altivec_yuv2_rgb24;
556 return altivec_yuv2_bgr24;
559 return altivec_yuv2_argb;
562 return altivec_yuv2_abgr;
565 return altivec_yuv2_rgba;
568 return altivec_yuv2_bgra;
569 default:
return NULL;
574 switch (
c->dstFormat) {
577 return altivec_uyvy_rgb32;
578 default:
return NULL;
588 const int inv_table[4],
596 vector
signed short vec;
602 buf.tmp[0] = ((0xffffLL) * contrast >> 8) >> 9;
603 buf.tmp[1] = -256 * brightness;
604 buf.tmp[2] = (inv_table[0] >> 3) * (contrast >> 16) * (saturation >> 16);
605 buf.tmp[3] = (inv_table[1] >> 3) * (contrast >> 16) * (saturation >> 16);
606 buf.tmp[4] = -((inv_table[2] >> 1) * (contrast >> 16) * (saturation >> 16));
607 buf.tmp[5] = -((inv_table[3] >> 1) * (contrast >> 16) * (saturation >> 16));
609 c->CSHIFT = (vector
unsigned short) vec_splat_u16(2);
610 c->CY = vec_splat((vector
signed short) buf.vec, 0);
611 c->OY = vec_splat((vector
signed short) buf.vec, 1);
612 c->CRV = vec_splat((vector
signed short) buf.vec, 2);
613 c->CBU = vec_splat((vector
signed short) buf.vec, 3);
614 c->CGU = vec_splat((vector
signed short) buf.vec, 4);
615 c->CGV = vec_splat((vector
signed short) buf.vec, 5);
623 const int16_t *lumFilter,
624 const int16_t **lumSrc,
626 const int16_t *chrFilter,
627 const int16_t **chrUSrc,
628 const int16_t **chrVSrc,
630 const int16_t **alpSrc,
636 vector
signed short X, X0, X1, Y0, U0, V0, Y1, U1, V1,
U,
V;
637 vector
signed short R0, G0,
B0,
R1, G1,
B1;
639 vector
unsigned char R,
G,
B;
640 vector
unsigned char *
out, *nout;
642 vector
signed short RND = vec_splat_s16(1 << 3);
643 vector
unsigned short SCL = vec_splat_u16(4);
646 vector
signed short *YCoeffs, *CCoeffs;
648 YCoeffs =
c->vYCoeffsBank + dstY * lumFilterSize;
649 CCoeffs =
c->vCCoeffsBank + dstY * chrFilterSize;
651 out = (vector
unsigned char *) dest;
653 for (
i = 0;
i < dstW;
i += 16) {
657 for (j = 0; j < lumFilterSize; j++) {
658 X0 = vec_ld(0, &lumSrc[j][
i]);
659 X1 = vec_ld(16, &lumSrc[j][
i]);
660 Y0 = vec_mradds(X0, YCoeffs[j], Y0);
661 Y1 = vec_mradds(X1, YCoeffs[j], Y1);
667 for (j = 0; j < chrFilterSize; j++) {
668 X = vec_ld(0, &chrUSrc[j][
i / 2]);
669 U = vec_mradds(
X, CCoeffs[j],
U);
670 X = vec_ld(0, &chrVSrc[j][
i / 2]);
671 V = vec_mradds(
X, CCoeffs[j],
V);
675 Y0 = vec_sra(Y0, SCL);
676 Y1 = vec_sra(Y1, SCL);
680 Y0 = vec_clip_s16(Y0);
681 Y1 = vec_clip_s16(Y1);
694 U0 = vec_mergeh(
U,
U);
695 V0 = vec_mergeh(
V,
V);
697 U1 = vec_mergel(
U,
U);
698 V1 = vec_mergel(
V,
V);
700 cvtyuvtoRGB(
c, Y0, U0, V0, &
R0, &G0, &
B0);
701 cvtyuvtoRGB(
c, Y1, U1, V1, &
R1, &G1, &
B1);
703 R = vec_packclp(
R0,
R1);
704 G = vec_packclp(G0, G1);
705 B = vec_packclp(
B0,
B1);
730 static int printed_error_message;
731 if (!printed_error_message) {
733 "altivec_yuv2packedX doesn't support %s output\n",
735 printed_error_message = 1;
748 for (j = 0; j < lumFilterSize; j++) {
749 X0 = vec_ld(0, &lumSrc[j][
i]);
750 X1 = vec_ld(16, &lumSrc[j][
i]);
751 Y0 = vec_mradds(X0, YCoeffs[j], Y0);
752 Y1 = vec_mradds(X1, YCoeffs[j], Y1);
758 for (j = 0; j < chrFilterSize; j++) {
759 X = vec_ld(0, &chrUSrc[j][
i / 2]);
760 U = vec_mradds(
X, CCoeffs[j],
U);
761 X = vec_ld(0, &chrVSrc[j][
i / 2]);
762 V = vec_mradds(
X, CCoeffs[j],
V);
766 Y0 = vec_sra(Y0, SCL);
767 Y1 = vec_sra(Y1, SCL);
771 Y0 = vec_clip_s16(Y0);
772 Y1 = vec_clip_s16(Y1);
785 U0 = vec_mergeh(
U,
U);
786 V0 = vec_mergeh(
V,
V);
788 U1 = vec_mergel(
U,
U);
789 V1 = vec_mergel(
V,
V);
791 cvtyuvtoRGB(
c, Y0, U0, V0, &
R0, &G0, &
B0);
792 cvtyuvtoRGB(
c, Y1, U1, V1, &
R1, &G1, &
B1);
794 R = vec_packclp(
R0,
R1);
795 G = vec_packclp(G0, G1);
796 B = vec_packclp(
B0,
B1);
798 nout = (vector
unsigned char *) scratch;
801 out_abgr(
R,
G,
B, nout);
804 out_bgra(
R,
G,
B, nout);
807 out_rgba(
R,
G,
B, nout);
810 out_argb(
R,
G,
B, nout);
813 out_rgb24(
R,
G,
B, nout);
816 out_bgr24(
R,
G,
B, nout);
821 "altivec_yuv2packedX doesn't support %s output\n",
826 memcpy(&((uint32_t *) dest)[
i], scratch, (dstW -
i) / 4);
830 #define YUV2PACKEDX_WRAPPER(suffix, pixfmt) \
831 void ff_yuv2 ## suffix ## _X_altivec(SwsContext *c, \
832 const int16_t *lumFilter, \
833 const int16_t **lumSrc, \
835 const int16_t *chrFilter, \
836 const int16_t **chrUSrc, \
837 const int16_t **chrVSrc, \
839 const int16_t **alpSrc, \
840 uint8_t *dest, int dstW, int dstY) \
842 yuv2packedX_altivec(c, lumFilter, lumSrc, lumFilterSize, \
843 chrFilter, chrUSrc, chrVSrc, \
844 chrFilterSize, alpSrc, \
845 dest, dstW, dstY, pixfmt); \