git.sesse.net Git - ffmpeg/blob - libswscale/ppc/yuv2rgb_altivec.c

   1 /*
   2  * AltiVec acceleration for colorspace conversion
   3  *
   4  * copyright (C) 2004 Marc Hoffman <marc.hoffman@analog.com>
   5  *
   6  * This file is part of FFmpeg.
   7  *
   8  * FFmpeg is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU Lesser General Public
  10  * License as published by the Free Software Foundation; either
  11  * version 2.1 of the License, or (at your option) any later version.
  12  *
  13  * FFmpeg is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  16  * Lesser General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU Lesser General Public
  19  * License along with FFmpeg; if not, write to the Free Software
  20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  21  */
  22
  23 /*
  24  * Convert I420 YV12 to RGB in various formats,
  25  * it rejects images that are not in 420 formats,
  26  * it rejects images that don't have widths of multiples of 16,
  27  * it rejects images that don't have heights of multiples of 2.
  28  * Reject defers to C simulation code.
  29  *
  30  * Lots of optimizations to be done here.
  31  *
  32  * 1. Need to fix saturation code. I just couldn't get it to fly with packs
  33  * and adds, so we currently use max/min to clip.
  34  *
  35  * 2. The inefficient use of chroma loading needs a bit of brushing up.
  36  *
  37  * 3. Analysis of pipeline stalls needs to be done. Use shark to identify
  38  * pipeline stalls.
  39  *
  40  *
  41  * MODIFIED to calculate coeffs from currently selected color space.
  42  * MODIFIED core to be a macro where you specify the output format.
  43  * ADDED UYVY conversion which is never called due to some thing in swscale.
  44  * CORRECTED algorithim selection to be strict on input formats.
  45  * ADDED runtime detection of AltiVec.
  46  *
  47  * ADDED altivec_yuv2packedX vertical scl + RGB converter
  48  *
  49  * March 27,2004
  50  * PERFORMANCE ANALYSIS
  51  *
  52  * The C version uses 25% of the processor or ~250Mips for D1 video rawvideo
  53  * used as test.
  54  * The AltiVec version uses 10% of the processor or ~100Mips for D1 video
  55  * same sequence.
  56  *
  57  * 720 * 480 * 30  ~10MPS
  58  *
  59  * so we have roughly 10 clocks per pixel. This is too high, something has
  60  * to be wrong.
  61  *
  62  * OPTIMIZED clip codes to utilize vec_max and vec_packs removing the
  63  * need for vec_min.
  64  *
  65  * OPTIMIZED DST OUTPUT cache/DMA controls. We are pretty much guaranteed to
  66  * have the input video frame, it was just decompressed so it probably resides
  67  * in L1 caches. However, we are creating the output video stream. This needs
  68  * to use the DSTST instruction to optimize for the cache. We couple this with
  69  * the fact that we are not going to be visiting the input buffer again so we
  70  * mark it Least Recently Used. This shaves 25% of the processor cycles off.
  71  *
  72  * Now memcpy is the largest mips consumer in the system, probably due
  73  * to the inefficient X11 stuff.
  74  *
  75  * GL libraries seem to be very slow on this machine 1.33Ghz PB running
  76  * Jaguar, this is not the case for my 1Ghz PB.  I thought it might be
  77  * a versioning issue, however I have libGL.1.2.dylib for both
  78  * machines. (We need to figure this out now.)
  79  *
  80  * GL2 libraries work now with patch for RGB32.
  81  *
  82  * NOTE: quartz vo driver ARGB32_to_RGB24 consumes 30% of the processor.
  83  *
  84  * Integrated luma prescaling adjustment for saturation/contrast/brightness
  85  * adjustment.
  86  */
  87
  88 #include <stdio.h>
  89 #include <stdlib.h>
  90 #include <string.h>
  91 #include <inttypes.h>
  92
  93 #include "config.h"
  94 #include "libswscale/rgb2rgb.h"
  95 #include "libswscale/swscale.h"
  96 #include "libswscale/swscale_internal.h"
  97 #include "libavutil/attributes.h"
  98 #include "libavutil/cpu.h"
  99 #include "libavutil/pixdesc.h"
 100 #include "yuv2rgb_altivec.h"
 101
 102 #if HAVE_ALTIVEC
 103
 104 #undef PROFILE_THE_BEAST
 105 #undef INC_SCALING
 106
 107 typedef unsigned char ubyte;
 108 typedef signed char   sbyte;
 109
 110 /* RGB interleaver, 16 planar pels 8-bit samples per channel in
 111  * homogeneous vector registers x0,x1,x2 are interleaved with the
 112  * following technique:
 113  *
 114  *    o0 = vec_mergeh(x0, x1);
 115  *    o1 = vec_perm(o0, x2, perm_rgb_0);
 116  *    o2 = vec_perm(o0, x2, perm_rgb_1);
 117  *    o3 = vec_mergel(x0, x1);
 118  *    o4 = vec_perm(o3, o2, perm_rgb_2);
 119  *    o5 = vec_perm(o3, o2, perm_rgb_3);
 120  *
 121  * perm_rgb_0:   o0(RG).h v1(B) --> o1*
 122  *            0   1  2   3   4
 123  *           rgbr|gbrg|brgb|rgbr
 124  *           0010 0100 1001 0010
 125  *           0102 3145 2673 894A
 126  *
 127  * perm_rgb_1:   o0(RG).h v1(B) --> o2
 128  *            0   1  2   3   4
 129  *           gbrg|brgb|bbbb|bbbb
 130  *           0100 1001 1111 1111
 131  *           B5CD 6EF7 89AB CDEF
 132  *
 133  * perm_rgb_2:   o3(RG).l o2(rgbB.l) --> o4*
 134  *            0   1  2   3   4
 135  *           gbrg|brgb|rgbr|gbrg
 136  *           1111 1111 0010 0100
 137  *           89AB CDEF 0182 3945
 138  *
 139  * perm_rgb_2:   o3(RG).l o2(rgbB.l) ---> o5*
 140  *            0   1  2   3   4
 141  *           brgb|rgbr|gbrg|brgb
 142  *           1001 0010 0100 1001
 143  *           a67b 89cA BdCD eEFf
 144  */
 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 };
 154
 155 #define vec_merge3(x2, x1, x0, y0, y1, y2)     \
 156     do {                                       \
 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);     \
 164     } while (0)
 165
 166 #define vec_mstbgr24(x0, x1, x2, ptr)          \
 167     do {                                       \
 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++);                  \
 173     } while (0)
 174
 175 #define vec_mstrgb24(x0, x1, x2, ptr)          \
 176     do {                                       \
 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++);                  \
 182     } while (0)
 183
 184 /* pack the pixels in rgb0 format
 185  * msb R
 186  * lsb 0
 187  */
 188 #define vec_mstrgb32(T, x0, x1, x2, x3, ptr)                            \
 189     do {                                                                \
 190         T _0, _1, _2, _3;                                               \
 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);                                  \
 207         ptr += 4;                                                       \
 208     } while (0)
 209
 210 /*
 211  * 1     0       1.4021   | | Y |
 212  * 1    -0.3441 -0.7142   |x| Cb|
 213  * 1     1.7718  0        | | Cr|
 214  *
 215  *
 216  * Y:      [-128 127]
 217  * Cb/Cr : [-128 127]
 218  *
 219  * typical YUV conversion works on Y: 0-255 this version has been
 220  * optimized for JPEG decoding.
 221  */
 222
 223 #if HAVE_BIGENDIAN
 224 #define vec_unh(x)                                                      \
 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 }))
 230
 231 #define vec_unl(x)                                                      \
 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 }))
 237 #else
 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 })
 240 #endif
 241
 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 }))
 246
 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 })))
 253
 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)
 258 {
 259     vector signed short vx, ux, uvx;
 260
 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));
 266
 267     // ux  = (CBU * (u << c->CSHIFT) + 0x4000) >> 15;
 268     ux = vec_sl(U, c->CSHIFT);
 269     *B = vec_mradds(ux, c->CBU, Y);
 270
 271     // vx  = (CRV * (v << c->CSHIFT) + 0x4000) >> 15;
 272     vx = vec_sl(V, c->CSHIFT);
 273     *R = vec_mradds(vx, c->CRV, Y);
 274
 275     // uvx = ((CGU * u) + (CGV * v)) >> 15;
 276     uvx = vec_mradds(U, c->CGU, Y);
 277     *G  = vec_mradds(V, c->CGV, uvx);
 278 }
 279
 280 /*
 281  * ------------------------------------------------------------------------------
 282  * CS converters
 283  * ------------------------------------------------------------------------------
 284  */
 285
 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)         \
 290 {                                                                             \
 291     int w = c->srcW;                                                          \
 292     int h = srcSliceH;                                                        \
 293     int i, j;                                                                 \
 294     int instrides_scl[3];                                                     \
 295     vector unsigned char y0, y1;                                              \
 296                                                                               \
 297     vector signed char u, v;                                                  \
 298                                                                               \
 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;                                             \
 307                                                                               \
 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;                                \
 315                                                                               \
 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];                                                 \
 320                                                                               \
 321     vector unsigned char *oute, *outo;                                        \
 322                                                                               \
 323     /* loop moves y{1, 2}i by w */                                            \
 324     instrides_scl[0] = instrides[0] * 2 - w;                                  \
 325     /* loop moves ui by w / 2 */                                              \
 326     instrides_scl[1] = instrides[1] - w / 2;                                  \
 327     /* loop moves vi by w / 2 */                                              \
 328     instrides_scl[2] = instrides[2] - w / 2;                                  \
 329                                                                               \
 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);       \
 336                                                                               \
 337         for (j = 0; j < w / 16; j++) {                                        \
 338             y0 = vec_xl(0, y1i);                                              \
 339                                                                               \
 340             y1 = vec_xl(0, y2i);                                              \
 341                                                                               \
 342             u = (vector signed char) vec_xl(0, ui);                           \
 343                                                                               \
 344             v = (vector signed char) vec_xl(0, vi);                           \
 345                                                                               \
 346             u = (vector signed char)                                          \
 347                     vec_sub(u,                                                \
 348                             (vector signed char)                              \
 349                                 vec_splat((vector signed char) { 128 }, 0));  \
 350             v = (vector signed char)                                          \
 351                     vec_sub(v,                                                \
 352                             (vector signed char)                              \
 353                                 vec_splat((vector signed char) { 128 }, 0));  \
 354                                                                               \
 355             U = vec_unpackh(u);                                               \
 356             V = vec_unpackh(v);                                               \
 357                                                                               \
 358             Y0 = vec_unh(y0);                                                 \
 359             Y1 = vec_unl(y0);                                                 \
 360             Y2 = vec_unh(y1);                                                 \
 361             Y3 = vec_unl(y1);                                                 \
 362                                                                               \
 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);                                    \
 367                                                                               \
 368             /* ux  = (CBU * (u << CSHIFT) + 0x4000) >> 15 */                  \
 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);                                         \
 373                                                                               \
 374             /* vx  = (CRV * (v << CSHIFT) + 0x4000) >> 15; */                 \
 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);                                         \
 379                                                                               \
 380             /* uvx = ((CGU * u) + (CGV * v)) >> 15 */                         \
 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);                                      \
 385                                                                               \
 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);                                            \
 392                                                                               \
 393             R = vec_packclp(R0, R1);                                          \
 394             G = vec_packclp(G0, G1);                                          \
 395             B = vec_packclp(B0, B1);                                          \
 396                                                                               \
 397             out_pixels(R, G, B, oute);                                        \
 398                                                                               \
 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);                                         \
 408                                                                               \
 409                                                                               \
 410             out_pixels(R, G, B, outo);                                        \
 411                                                                               \
 412             y1i += 16;                                                        \
 413             y2i += 16;                                                        \
 414             ui  += 8;                                                         \
 415             vi  += 8;                                                         \
 416         }                                                                     \
 417                                                                               \
 418         ui  += instrides_scl[1];                                              \
 419         vi  += instrides_scl[2];                                              \
 420         y1i += instrides_scl[0];                                              \
 421         y2i += instrides_scl[0];                                              \
 422     }                                                                         \
 423     return srcSliceH;                                                         \
 424 }
 425
 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)
 436
 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)
 443
 444 // uyvy|uyvy|uyvy|uyvy
 445 // 0123 4567 89ab cdef
 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 };
 459
 460 /*
 461  * this is so I can play live CCIR raw video
 462  */
 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)
 466 {
 467     int w = c->srcW;
 468     int h = srcSliceH;
 469     int i, j;
 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;
 475     const ubyte *img;
 476
 477     img = in[0];
 478     out = (vector unsigned char *) (oplanes[0] + srcSliceY * outstrides[0]);
 479
 480     for (i = 0; i < h; i++)
 481         for (j = 0; j < w / 16; j++) {
 482             uyvy = vec_ld(0, img);
 483
 484             U = (vector signed short)
 485                     vec_perm(uyvy, (vector unsigned char) { 0 }, demux_u);
 486             V = (vector signed short)
 487                     vec_perm(uyvy, (vector unsigned char) { 0 }, demux_v);
 488             Y = (vector signed short)
 489                     vec_perm(uyvy, (vector unsigned char) { 0 }, demux_y);
 490
 491             cvtyuvtoRGB(c, Y, U, V, &R0, &G0, &B0);
 492
 493             uyvy = vec_ld(16, img);
 494
 495             U = (vector signed short)
 496                     vec_perm(uyvy, (vector unsigned char) { 0 }, demux_u);
 497             V = (vector signed short)
 498                     vec_perm(uyvy, (vector unsigned char) { 0 }, demux_v);
 499             Y = (vector signed short)
 500                     vec_perm(uyvy, (vector unsigned char) { 0 }, demux_y);
 501
 502             cvtyuvtoRGB(c, Y, U, V, &R1, &G1, &B1);
 503
 504             R = vec_packclp(R0, R1);
 505             G = vec_packclp(G0, G1);
 506             B = vec_packclp(B0, B1);
 507
 508             // vec_mstbgr24 (R,G,B, out);
 509             out_rgba(R, G, B, out);
 510
 511             img += 32;
 512         }
 513     return srcSliceH;
 514 }
 515
 516 #endif /* HAVE_ALTIVEC */
 517
 518 /* Ok currently the acceleration routine only supports
 519  * inputs of widths a multiple of 16
 520  * and heights a multiple 2
 521  *
 522  * So we just fall back to the C codes for this.
 523  */
 524 av_cold SwsFunc ff_yuv2rgb_init_ppc(SwsContext *c)
 525 {
 526 #if HAVE_ALTIVEC
 527     if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC))
 528         return NULL;
 529
 530     /*
 531      * and this seems not to matter too much I tried a bunch of
 532      * videos with abnormal widths and MPlayer crashes elsewhere.
 533      * mplayer -vo x11 -rawvideo on:w=350:h=240 raw-350x240.eyuv
 534      * boom with X11 bad match.
 535      *
 536      */
 537     if ((c->srcW & 0xf) != 0)
 538         return NULL;
 539
 540     switch (c->srcFormat) {
 541     case AV_PIX_FMT_YUV410P:
 542     case AV_PIX_FMT_YUV420P:
 543     /*case IMGFMT_CLPL:        ??? */
 544     case AV_PIX_FMT_GRAY8:
 545     case AV_PIX_FMT_NV12:
 546     case AV_PIX_FMT_NV21:
 547         if ((c->srcH & 0x1) != 0)
 548             return NULL;
 549
 550         switch (c->dstFormat) {
 551         case AV_PIX_FMT_RGB24:
 552             av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space RGB24\n");
 553             return altivec_yuv2_rgb24;
 554         case AV_PIX_FMT_BGR24:
 555             av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space BGR24\n");
 556             return altivec_yuv2_bgr24;
 557         case AV_PIX_FMT_ARGB:
 558             av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space ARGB\n");
 559             return altivec_yuv2_argb;
 560         case AV_PIX_FMT_ABGR:
 561             av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space ABGR\n");
 562             return altivec_yuv2_abgr;
 563         case AV_PIX_FMT_RGBA:
 564             av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space RGBA\n");
 565             return altivec_yuv2_rgba;
 566         case AV_PIX_FMT_BGRA:
 567             av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space BGRA\n");
 568             return altivec_yuv2_bgra;
 569         default: return NULL;
 570         }
 571         break;
 572
 573     case AV_PIX_FMT_UYVY422:
 574         switch (c->dstFormat) {
 575         case AV_PIX_FMT_BGR32:
 576             av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space UYVY -> RGB32\n");
 577             return altivec_uyvy_rgb32;
 578         default: return NULL;
 579         }
 580         break;
 581     }
 582 #endif /* HAVE_ALTIVEC */
 583
 584     return NULL;
 585 }
 586
 587 av_cold void ff_yuv2rgb_init_tables_ppc(SwsContext *c,
 588                                         const int inv_table[4],
 589                                         int brightness,
 590                                         int contrast,
 591                                         int saturation)
 592 {
 593 #if HAVE_ALTIVEC
 594     union {
 595         DECLARE_ALIGNED(16, signed short, tmp)[8];
 596         vector signed short vec;
 597     } buf;
 598
 599     if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC))
 600         return;
 601
 602     buf.tmp[0] = ((0xffffLL) * contrast >> 8) >> 9;                               // cy
 603     buf.tmp[1] = -256 * brightness;                                               // oy
 604     buf.tmp[2] =   (inv_table[0] >> 3) * (contrast >> 16) * (saturation >> 16);   // crv
 605     buf.tmp[3] =   (inv_table[1] >> 3) * (contrast >> 16) * (saturation >> 16);   // cbu
 606     buf.tmp[4] = -((inv_table[2] >> 1) * (contrast >> 16) * (saturation >> 16));  // cgu
 607     buf.tmp[5] = -((inv_table[3] >> 1) * (contrast >> 16) * (saturation >> 16));  // cgv
 608
 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);
 616     return;
 617 #endif /* HAVE_ALTIVEC */
 618 }
 619
 620 #if HAVE_ALTIVEC
 621
 622 static av_always_inline void yuv2packedX_altivec(SwsContext *c,
 623                                                  const int16_t *lumFilter,
 624                                                  const int16_t **lumSrc,
 625                                                  int lumFilterSize,
 626                                                  const int16_t *chrFilter,
 627                                                  const int16_t **chrUSrc,
 628                                                  const int16_t **chrVSrc,
 629                                                  int chrFilterSize,
 630                                                  const int16_t **alpSrc,
 631                                                  uint8_t *dest,
 632                                                  int dstW, int dstY,
 633                                                  enum AVPixelFormat target)
 634 {
 635     int i, j;
 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;
 638
 639     vector unsigned char R, G, B;
 640     vector unsigned char *out, *nout;
 641
 642     vector signed short RND   = vec_splat_s16(1 << 3);
 643     vector unsigned short SCL = vec_splat_u16(4);
 644     DECLARE_ALIGNED(16, unsigned int, scratch)[16];
 645
 646     vector signed short *YCoeffs, *CCoeffs;
 647
 648     YCoeffs = c->vYCoeffsBank + dstY * lumFilterSize;
 649     CCoeffs = c->vCCoeffsBank + dstY * chrFilterSize;
 650
 651     out = (vector unsigned char *) dest;
 652
 653     for (i = 0; i < dstW; i += 16) {
 654         Y0 = RND;
 655         Y1 = RND;
 656         /* extract 16 coeffs from lumSrc */
 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);
 662         }
 663
 664         U = RND;
 665         V = RND;
 666         /* extract 8 coeffs from U,V */
 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);
 672         }
 673
 674         /* scale and clip signals */
 675         Y0 = vec_sra(Y0, SCL);
 676         Y1 = vec_sra(Y1, SCL);
 677         U  = vec_sra(U, SCL);
 678         V  = vec_sra(V, SCL);
 679
 680         Y0 = vec_clip_s16(Y0);
 681         Y1 = vec_clip_s16(Y1);
 682         U  = vec_clip_s16(U);
 683         V  = vec_clip_s16(V);
 684
 685         /* now we have
 686          * Y0 = y0 y1 y2 y3 y4 y5 y6 y7    Y1 = y8 y9 y10 y11 y12 y13 y14 y15
 687          * U  = u0 u1 u2 u3 u4 u5 u6 u7    V  = v0 v1 v2 v3 v4 v5 v6 v7
 688          *
 689          * Y0 = y0 y1 y2 y3 y4 y5 y6 y7    Y1 = y8 y9 y10 y11 y12 y13 y14 y15
 690          * U0 = u0 u0 u1 u1 u2 u2 u3 u3    U1 = u4 u4 u5 u5 u6 u6 u7 u7
 691          * V0 = v0 v0 v1 v1 v2 v2 v3 v3    V1 = v4 v4 v5 v5 v6 v6 v7 v7
 692          */
 693
 694         U0 = vec_mergeh(U, U);
 695         V0 = vec_mergeh(V, V);
 696
 697         U1 = vec_mergel(U, U);
 698         V1 = vec_mergel(V, V);
 699
 700         cvtyuvtoRGB(c, Y0, U0, V0, &R0, &G0, &B0);
 701         cvtyuvtoRGB(c, Y1, U1, V1, &R1, &G1, &B1);
 702
 703         R = vec_packclp(R0, R1);
 704         G = vec_packclp(G0, G1);
 705         B = vec_packclp(B0, B1);
 706
 707         switch (target) {
 708         case AV_PIX_FMT_ABGR:
 709             out_abgr(R, G, B, out);
 710             break;
 711         case AV_PIX_FMT_BGRA:
 712             out_bgra(R, G, B, out);
 713             break;
 714         case AV_PIX_FMT_RGBA:
 715             out_rgba(R, G, B, out);
 716             break;
 717         case AV_PIX_FMT_ARGB:
 718             out_argb(R, G, B, out);
 719             break;
 720         case AV_PIX_FMT_RGB24:
 721             out_rgb24(R, G, B, out);
 722             break;
 723         case AV_PIX_FMT_BGR24:
 724             out_bgr24(R, G, B, out);
 725             break;
 726         default:
 727         {
 728             /* If this is reached, the caller should have called yuv2packedXinC
 729              * instead. */
 730             static int printed_error_message;
 731             if (!printed_error_message) {
 732                 av_log(c, AV_LOG_ERROR,
 733                        "altivec_yuv2packedX doesn't support %s output\n",
 734                        av_get_pix_fmt_name(c->dstFormat));
 735                 printed_error_message = 1;
 736             }
 737             return;
 738         }
 739         }
 740     }
 741
 742     if (i < dstW) {
 743         i -= 16;
 744
 745         Y0 = RND;
 746         Y1 = RND;
 747         /* extract 16 coeffs from lumSrc */
 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);
 753         }
 754
 755         U = RND;
 756         V = RND;
 757         /* extract 8 coeffs from U,V */
 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);
 763         }
 764
 765         /* scale and clip signals */
 766         Y0 = vec_sra(Y0, SCL);
 767         Y1 = vec_sra(Y1, SCL);
 768         U  = vec_sra(U, SCL);
 769         V  = vec_sra(V, SCL);
 770
 771         Y0 = vec_clip_s16(Y0);
 772         Y1 = vec_clip_s16(Y1);
 773         U  = vec_clip_s16(U);
 774         V  = vec_clip_s16(V);
 775
 776         /* now we have
 777          * Y0 = y0 y1 y2 y3 y4 y5 y6 y7    Y1 = y8 y9 y10 y11 y12 y13 y14 y15
 778          * U  = u0 u1 u2 u3 u4 u5 u6 u7    V  = v0 v1 v2 v3 v4 v5 v6 v7
 779          *
 780          * Y0 = y0 y1 y2 y3 y4 y5 y6 y7    Y1 = y8 y9 y10 y11 y12 y13 y14 y15
 781          * U0 = u0 u0 u1 u1 u2 u2 u3 u3    U1 = u4 u4 u5 u5 u6 u6 u7 u7
 782          * V0 = v0 v0 v1 v1 v2 v2 v3 v3    V1 = v4 v4 v5 v5 v6 v6 v7 v7
 783          */
 784
 785         U0 = vec_mergeh(U, U);
 786         V0 = vec_mergeh(V, V);
 787
 788         U1 = vec_mergel(U, U);
 789         V1 = vec_mergel(V, V);
 790
 791         cvtyuvtoRGB(c, Y0, U0, V0, &R0, &G0, &B0);
 792         cvtyuvtoRGB(c, Y1, U1, V1, &R1, &G1, &B1);
 793
 794         R = vec_packclp(R0, R1);
 795         G = vec_packclp(G0, G1);
 796         B = vec_packclp(B0, B1);
 797
 798         nout = (vector unsigned char *) scratch;
 799         switch (target) {
 800         case AV_PIX_FMT_ABGR:
 801             out_abgr(R, G, B, nout);
 802             break;
 803         case AV_PIX_FMT_BGRA:
 804             out_bgra(R, G, B, nout);
 805             break;
 806         case AV_PIX_FMT_RGBA:
 807             out_rgba(R, G, B, nout);
 808             break;
 809         case AV_PIX_FMT_ARGB:
 810             out_argb(R, G, B, nout);
 811             break;
 812         case AV_PIX_FMT_RGB24:
 813             out_rgb24(R, G, B, nout);
 814             break;
 815         case AV_PIX_FMT_BGR24:
 816             out_bgr24(R, G, B, nout);
 817             break;
 818         default:
 819             /* Unreachable, I think. */
 820             av_log(c, AV_LOG_ERROR,
 821                    "altivec_yuv2packedX doesn't support %s output\n",
 822                    av_get_pix_fmt_name(c->dstFormat));
 823             return;
 824         }
 825
 826         memcpy(&((uint32_t *) dest)[i], scratch, (dstW - i) / 4);
 827     }
 828 }
 829
 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,            \
 834                                      int lumFilterSize,                 \
 835                                      const int16_t *chrFilter,          \
 836                                      const int16_t **chrUSrc,           \
 837                                      const int16_t **chrVSrc,           \
 838                                      int chrFilterSize,                 \
 839                                      const int16_t **alpSrc,            \
 840                                      uint8_t *dest, int dstW, int dstY) \
 841 {                                                                       \
 842     yuv2packedX_altivec(c, lumFilter, lumSrc, lumFilterSize,            \
 843                         chrFilter, chrUSrc, chrVSrc,                    \
 844                         chrFilterSize, alpSrc,                          \
 845                         dest, dstW, dstY, pixfmt);                      \
 846 }
 847
 848 YUV2PACKEDX_WRAPPER(abgr,  AV_PIX_FMT_ABGR);
 849 YUV2PACKEDX_WRAPPER(bgra,  AV_PIX_FMT_BGRA);
 850 YUV2PACKEDX_WRAPPER(argb,  AV_PIX_FMT_ARGB);
 851 YUV2PACKEDX_WRAPPER(rgba,  AV_PIX_FMT_RGBA);
 852 YUV2PACKEDX_WRAPPER(rgb24, AV_PIX_FMT_RGB24);
 853 YUV2PACKEDX_WRAPPER(bgr24, AV_PIX_FMT_BGR24);
 854
 855 #endif /* HAVE_ALTIVEC */