FFmpeg: libswscale/ppc/yuv2rgb_altivec.c Source File

00001 /*
00002  * AltiVec acceleration for colorspace conversion
00003  *
00004  * copyright (C) 2004 Marc Hoffman <marc.hoffman@analog.com>
00005  *
00006  * This file is part of FFmpeg.
00007  *
00008  * FFmpeg is free software; you can redistribute it and/or
00009  * modify it under the terms of the GNU Lesser General Public
00010  * License as published by the Free Software Foundation; either
00011  * version 2.1 of the License, or (at your option) any later version.
00012  *
00013  * FFmpeg is distributed in the hope that it will be useful,
00014  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00015  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00016  * Lesser General Public License for more details.
00017  *
00018  * You should have received a copy of the GNU Lesser General Public
00019  * License along with FFmpeg; if not, write to the Free Software
00020  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00021  */
00022 
00023 /*
00024  * Convert I420 YV12 to RGB in various formats,
00025  * it rejects images that are not in 420 formats,
00026  * it rejects images that don't have widths of multiples of 16,
00027  * it rejects images that don't have heights of multiples of 2.
00028  * Reject defers to C simulation code.
00029  *
00030  * Lots of optimizations to be done here.
00031  *
00032  * 1. Need to fix saturation code. I just couldn't get it to fly with packs
00033  * and adds, so we currently use max/min to clip.
00034  *
00035  * 2. The inefficient use of chroma loading needs a bit of brushing up.
00036  *
00037  * 3. Analysis of pipeline stalls needs to be done. Use shark to identify
00038  * pipeline stalls.
00039  *
00040  *
00041  * MODIFIED to calculate coeffs from currently selected color space.
00042  * MODIFIED core to be a macro where you specify the output format.
00043  * ADDED UYVY conversion which is never called due to some thing in swscale.
00044  * CORRECTED algorithim selection to be strict on input formats.
00045  * ADDED runtime detection of AltiVec.
00046  *
00047  * ADDED altivec_yuv2packedX vertical scl + RGB converter
00048  *
00049  * March 27,2004
00050  * PERFORMANCE ANALYSIS
00051  *
00052  * The C version uses 25% of the processor or ~250Mips for D1 video rawvideo
00053  * used as test.
00054  * The AltiVec version uses 10% of the processor or ~100Mips for D1 video
00055  * same sequence.
00056  *
00057  * 720 * 480 * 30  ~10MPS
00058  *
00059  * so we have roughly 10 clocks per pixel. This is too high, something has
00060  * to be wrong.
00061  *
00062  * OPTIMIZED clip codes to utilize vec_max and vec_packs removing the
00063  * need for vec_min.
00064  *
00065  * OPTIMIZED DST OUTPUT cache/DMA controls. We are pretty much guaranteed to
00066  * have the input video frame, it was just decompressed so it probably resides
00067  * in L1 caches. However, we are creating the output video stream. This needs
00068  * to use the DSTST instruction to optimize for the cache. We couple this with
00069  * the fact that we are not going to be visiting the input buffer again so we
00070  * mark it Least Recently Used. This shaves 25% of the processor cycles off.
00071  *
00072  * Now memcpy is the largest mips consumer in the system, probably due
00073  * to the inefficient X11 stuff.
00074  *
00075  * GL libraries seem to be very slow on this machine 1.33Ghz PB running
00076  * Jaguar, this is not the case for my 1Ghz PB.  I thought it might be
00077  * a versioning issue, however I have libGL.1.2.dylib for both
00078  * machines. (We need to figure this out now.)
00079  *
00080  * GL2 libraries work now with patch for RGB32.
00081  *
00082  * NOTE: quartz vo driver ARGB32_to_RGB24 consumes 30% of the processor.
00083  *
00084  * Integrated luma prescaling adjustment for saturation/contrast/brightness
00085  * adjustment.
00086  */
00087 
00088 #include <stdio.h>
00089 #include <stdlib.h>
00090 #include <string.h>
00091 #include <inttypes.h>
00092 #include <assert.h>
00093 
00094 #include "config.h"
00095 #include "libswscale/rgb2rgb.h"
00096 #include "libswscale/swscale.h"
00097 #include "libswscale/swscale_internal.h"
00098 #include "libavutil/cpu.h"
00099 #include "libavutil/pixdesc.h"
00100 #include "yuv2rgb_altivec.h"
00101 
00102 #undef PROFILE_THE_BEAST
00103 #undef INC_SCALING
00104 
00105 typedef unsigned char ubyte;
00106 typedef signed char   sbyte;
00107 
00108 /* RGB interleaver, 16 planar pels 8-bit samples per channel in
00109  * homogeneous vector registers x0,x1,x2 are interleaved with the
00110  * following technique:
00111  *
00112  *    o0 = vec_mergeh(x0, x1);
00113  *    o1 = vec_perm(o0, x2, perm_rgb_0);
00114  *    o2 = vec_perm(o0, x2, perm_rgb_1);
00115  *    o3 = vec_mergel(x0, x1);
00116  *    o4 = vec_perm(o3, o2, perm_rgb_2);
00117  *    o5 = vec_perm(o3, o2, perm_rgb_3);
00118  *
00119  * perm_rgb_0:   o0(RG).h v1(B) --> o1*
00120  *            0   1  2   3   4
00121  *           rgbr|gbrg|brgb|rgbr
00122  *           0010 0100 1001 0010
00123  *           0102 3145 2673 894A
00124  *
00125  * perm_rgb_1:   o0(RG).h v1(B) --> o2
00126  *            0   1  2   3   4
00127  *           gbrg|brgb|bbbb|bbbb
00128  *           0100 1001 1111 1111
00129  *           B5CD 6EF7 89AB CDEF
00130  *
00131  * perm_rgb_2:   o3(RG).l o2(rgbB.l) --> o4*
00132  *            0   1  2   3   4
00133  *           gbrg|brgb|rgbr|gbrg
00134  *           1111 1111 0010 0100
00135  *           89AB CDEF 0182 3945
00136  *
00137  * perm_rgb_2:   o3(RG).l o2(rgbB.l) ---> o5*
00138  *            0   1  2   3   4
00139  *           brgb|rgbr|gbrg|brgb
00140  *           1001 0010 0100 1001
00141  *           a67b 89cA BdCD eEFf
00142  *
00143  */
00144 static const vector unsigned char
00145     perm_rgb_0 = { 0x00, 0x01, 0x10, 0x02, 0x03, 0x11, 0x04, 0x05,
00146                    0x12, 0x06, 0x07, 0x13, 0x08, 0x09, 0x14, 0x0a },
00147     perm_rgb_1 = { 0x0b, 0x15, 0x0c, 0x0d, 0x16, 0x0e, 0x0f, 0x17,
00148                    0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f },
00149     perm_rgb_2 = { 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17,
00150                    0x00, 0x01, 0x18, 0x02, 0x03, 0x19, 0x04, 0x05 },
00151     perm_rgb_3 = { 0x1a, 0x06, 0x07, 0x1b, 0x08, 0x09, 0x1c, 0x0a,
00152                    0x0b, 0x1d, 0x0c, 0x0d, 0x1e, 0x0e, 0x0f, 0x1f };
00153 
00154 #define vec_merge3(x2, x1, x0, y0, y1, y2)     \
00155     do {                                       \
00156         __typeof__(x0) o0, o2, o3;             \
00157         o0 = vec_mergeh(x0, x1);               \
00158         y0 = vec_perm(o0, x2, perm_rgb_0);     \
00159         o2 = vec_perm(o0, x2, perm_rgb_1);     \
00160         o3 = vec_mergel(x0, x1);               \
00161         y1 = vec_perm(o3, o2, perm_rgb_2);     \
00162         y2 = vec_perm(o3, o2, perm_rgb_3);     \
00163     } while (0)
00164 
00165 #define vec_mstbgr24(x0, x1, x2, ptr)          \
00166     do {                                       \
00167         __typeof__(x0) _0, _1, _2;             \
00168         vec_merge3(x0, x1, x2, _0, _1, _2);    \
00169         vec_st(_0, 0, ptr++);                  \
00170         vec_st(_1, 0, ptr++);                  \
00171         vec_st(_2, 0, ptr++);                  \
00172     } while (0)
00173 
00174 #define vec_mstrgb24(x0, x1, x2, ptr)          \
00175     do {                                       \
00176         __typeof__(x0) _0, _1, _2;             \
00177         vec_merge3(x2, x1, x0, _0, _1, _2);    \
00178         vec_st(_0, 0, ptr++);                  \
00179         vec_st(_1, 0, ptr++);                  \
00180         vec_st(_2, 0, ptr++);                  \
00181     } while (0)
00182 
00183 /* pack the pixels in rgb0 format
00184  * msb R
00185  * lsb 0
00186  */
00187 #define vec_mstrgb32(T, x0, x1, x2, x3, ptr)                            \
00188     do {                                                                \
00189         T _0, _1, _2, _3;                                               \
00190         _0 = vec_mergeh(x0, x1);                                        \
00191         _1 = vec_mergeh(x2, x3);                                        \
00192         _2 = (T) vec_mergeh((vector unsigned short) _0,                 \
00193                             (vector unsigned short) _1);                \
00194         _3 = (T) vec_mergel((vector unsigned short) _0,                 \
00195                             (vector unsigned short) _1);                \
00196         vec_st(_2, 0 * 16, (T *) ptr);                                  \
00197         vec_st(_3, 1 * 16, (T *) ptr);                                  \
00198         _0 = vec_mergel(x0, x1);                                        \
00199         _1 = vec_mergel(x2, x3);                                        \
00200         _2 = (T) vec_mergeh((vector unsigned short) _0,                 \
00201                             (vector unsigned short) _1);                \
00202         _3 = (T) vec_mergel((vector unsigned short) _0,                 \
00203                             (vector unsigned short) _1);                \
00204         vec_st(_2, 2 * 16, (T *) ptr);                                  \
00205         vec_st(_3, 3 * 16, (T *) ptr);                                  \
00206         ptr += 4;                                                       \
00207     } while (0)
00208 
00209 /*
00210  * 1     0       1.4021   | | Y |
00211  * 1    -0.3441 -0.7142   |x| Cb|
00212  * 1     1.7718  0        | | Cr|
00213  *
00214  *
00215  * Y:      [-128 127]
00216  * Cb/Cr : [-128 127]
00217  *
00218  * typical YUV conversion works on Y: 0-255 this version has been
00219  * optimized for JPEG decoding.
00220  */
00221 
00222 #define vec_unh(x)                                                      \
00223     (vector signed short)                                               \
00224         vec_perm(x, (__typeof__(x)) { 0 },                              \
00225                  ((vector unsigned char) {                              \
00226                      0x10, 0x00, 0x10, 0x01, 0x10, 0x02, 0x10, 0x03,    \
00227                      0x10, 0x04, 0x10, 0x05, 0x10, 0x06, 0x10, 0x07 }))
00228 
00229 #define vec_unl(x)                                                      \
00230     (vector signed short)                                               \
00231         vec_perm(x, (__typeof__(x)) { 0 },                              \
00232                  ((vector unsigned char) {                              \
00233                      0x10, 0x08, 0x10, 0x09, 0x10, 0x0A, 0x10, 0x0B,    \
00234                      0x10, 0x0C, 0x10, 0x0D, 0x10, 0x0E, 0x10, 0x0F }))
00235 
00236 #define vec_clip_s16(x)                                                 \
00237     vec_max(vec_min(x, ((vector signed short) {                         \
00238                     235, 235, 235, 235, 235, 235, 235, 235 })),         \
00239             ((vector signed short) { 16, 16, 16, 16, 16, 16, 16, 16 }))
00240 
00241 #define vec_packclp(x, y)                                               \
00242     (vector unsigned char)                                              \
00243         vec_packs((vector unsigned short)                               \
00244                       vec_max(x, ((vector signed short) { 0 })),        \
00245                   (vector unsigned short)                               \
00246                       vec_max(y, ((vector signed short) { 0 })))
00247 
00248 //#define out_pixels(a, b, c, ptr) vec_mstrgb32(__typeof__(a), ((__typeof__(a)) { 255 }), a, a, a, ptr)
00249 
00250 static inline void cvtyuvtoRGB(SwsContext *c, vector signed short Y,
00251                                vector signed short U, vector signed short V,
00252                                vector signed short *R, vector signed short *G,
00253                                vector signed short *B)
00254 {
00255     vector signed short vx, ux, uvx;
00256 
00257     Y = vec_mradds(Y, c->CY, c->OY);
00258     U = vec_sub(U, (vector signed short)
00259                        vec_splat((vector signed short) { 128 }, 0));
00260     V = vec_sub(V, (vector signed short)
00261                        vec_splat((vector signed short) { 128 }, 0));
00262 
00263     // ux  = (CBU * (u << c->CSHIFT) + 0x4000) >> 15;
00264     ux = vec_sl(U, c->CSHIFT);
00265     *B = vec_mradds(ux, c->CBU, Y);
00266 
00267     // vx  = (CRV * (v << c->CSHIFT) + 0x4000) >> 15;
00268     vx = vec_sl(V, c->CSHIFT);
00269     *R = vec_mradds(vx, c->CRV, Y);
00270 
00271     // uvx = ((CGU * u) + (CGV * v)) >> 15;
00272     uvx = vec_mradds(U, c->CGU, Y);
00273     *G  = vec_mradds(V, c->CGV, uvx);
00274 }
00275 
00276 /*
00277  * ------------------------------------------------------------------------------
00278  * CS converters
00279  * ------------------------------------------------------------------------------
00280  */
00281 
00282 #define DEFCSP420_CVT(name, out_pixels)                                       \
00283 static int altivec_ ## name(SwsContext *c, const unsigned char **in,          \
00284                             int *instrides, int srcSliceY, int srcSliceH,     \
00285                             unsigned char **oplanes, int *outstrides)         \
00286 {                                                                             \
00287     int w = c->srcW;                                                          \
00288     int h = srcSliceH;                                                        \
00289     int i, j;                                                                 \
00290     int instrides_scl[3];                                                     \
00291     vector unsigned char y0, y1;                                              \
00292                                                                               \
00293     vector signed char u, v;                                                  \
00294                                                                               \
00295     vector signed short Y0, Y1, Y2, Y3;                                       \
00296     vector signed short U, V;                                                 \
00297     vector signed short vx, ux, uvx;                                          \
00298     vector signed short vx0, ux0, uvx0;                                       \
00299     vector signed short vx1, ux1, uvx1;                                       \
00300     vector signed short R0, G0, B0;                                           \
00301     vector signed short R1, G1, B1;                                           \
00302     vector unsigned char R, G, B;                                             \
00303                                                                               \
00304     const vector unsigned char *y1ivP, *y2ivP, *uivP, *vivP;                  \
00305     vector unsigned char align_perm;                                          \
00306                                                                               \
00307     vector signed short lCY       = c->CY;                                    \
00308     vector signed short lOY       = c->OY;                                    \
00309     vector signed short lCRV      = c->CRV;                                   \
00310     vector signed short lCBU      = c->CBU;                                   \
00311     vector signed short lCGU      = c->CGU;                                   \
00312     vector signed short lCGV      = c->CGV;                                   \
00313     vector unsigned short lCSHIFT = c->CSHIFT;                                \
00314                                                                               \
00315     const ubyte *y1i = in[0];                                                 \
00316     const ubyte *y2i = in[0] + instrides[0];                                  \
00317     const ubyte *ui  = in[1];                                                 \
00318     const ubyte *vi  = in[2];                                                 \
00319                                                                               \
00320     vector unsigned char *oute =                                              \
00321         (vector unsigned char *)                                              \
00322             (oplanes[0] + srcSliceY * outstrides[0]);                         \
00323     vector unsigned char *outo =                                              \
00324         (vector unsigned char *)                                              \
00325             (oplanes[0] + srcSliceY * outstrides[0] + outstrides[0]);         \
00326                                                                               \
00327     /* loop moves y{1, 2}i by w */                                            \
00328     instrides_scl[0] = instrides[0] * 2 - w;                                  \
00329     /* loop moves ui by w / 2 */                                              \
00330     instrides_scl[1] = instrides[1] - w / 2;                                  \
00331     /* loop moves vi by w / 2 */                                              \
00332     instrides_scl[2] = instrides[2] - w / 2;                                  \
00333                                                                               \
00334     for (i = 0; i < h / 2; i++) {                                             \
00335         vec_dstst(outo, (0x02000002 | (((w * 3 + 32) / 32) << 16)), 0);       \
00336         vec_dstst(oute, (0x02000002 | (((w * 3 + 32) / 32) << 16)), 1);       \
00337                                                                               \
00338         for (j = 0; j < w / 16; j++) {                                        \
00339             y1ivP = (const vector unsigned char *) y1i;                       \
00340             y2ivP = (const vector unsigned char *) y2i;                       \
00341             uivP  = (const vector unsigned char *) ui;                        \
00342             vivP  = (const vector unsigned char *) vi;                        \
00343                                                                               \
00344             align_perm = vec_lvsl(0, y1i);                                    \
00345             y0 = (vector unsigned char)                                       \
00346                      vec_perm(y1ivP[0], y1ivP[1], align_perm);                \
00347                                                                               \
00348             align_perm = vec_lvsl(0, y2i);                                    \
00349             y1 = (vector unsigned char)                                       \
00350                      vec_perm(y2ivP[0], y2ivP[1], align_perm);                \
00351                                                                               \
00352             align_perm = vec_lvsl(0, ui);                                     \
00353             u = (vector signed char)                                          \
00354                     vec_perm(uivP[0], uivP[1], align_perm);                   \
00355                                                                               \
00356             align_perm = vec_lvsl(0, vi);                                     \
00357             v = (vector signed char)                                          \
00358                     vec_perm(vivP[0], vivP[1], align_perm);                   \
00359                                                                               \
00360             u = (vector signed char)                                          \
00361                     vec_sub(u,                                                \
00362                             (vector signed char)                              \
00363                                 vec_splat((vector signed char) { 128 }, 0));  \
00364             v = (vector signed char)                                          \
00365                     vec_sub(v,                                                \
00366                             (vector signed char)                              \
00367                                 vec_splat((vector signed char) { 128 }, 0));  \
00368                                                                               \
00369             U = vec_unpackh(u);                                               \
00370             V = vec_unpackh(v);                                               \
00371                                                                               \
00372             Y0 = vec_unh(y0);                                                 \
00373             Y1 = vec_unl(y0);                                                 \
00374             Y2 = vec_unh(y1);                                                 \
00375             Y3 = vec_unl(y1);                                                 \
00376                                                                               \
00377             Y0 = vec_mradds(Y0, lCY, lOY);                                    \
00378             Y1 = vec_mradds(Y1, lCY, lOY);                                    \
00379             Y2 = vec_mradds(Y2, lCY, lOY);                                    \
00380             Y3 = vec_mradds(Y3, lCY, lOY);                                    \
00381                                                                               \
00382             /* ux  = (CBU * (u << CSHIFT) + 0x4000) >> 15 */                  \
00383             ux  = vec_sl(U, lCSHIFT);                                         \
00384             ux  = vec_mradds(ux, lCBU, (vector signed short) { 0 });          \
00385             ux0 = vec_mergeh(ux, ux);                                         \
00386             ux1 = vec_mergel(ux, ux);                                         \
00387                                                                               \
00388             /* vx  = (CRV * (v << CSHIFT) + 0x4000) >> 15; */                 \
00389             vx  = vec_sl(V, lCSHIFT);                                         \
00390             vx  = vec_mradds(vx, lCRV, (vector signed short) { 0 });          \
00391             vx0 = vec_mergeh(vx, vx);                                         \
00392             vx1 = vec_mergel(vx, vx);                                         \
00393                                                                               \
00394             /* uvx = ((CGU * u) + (CGV * v)) >> 15 */                         \
00395             uvx  = vec_mradds(U, lCGU, (vector signed short) { 0 });          \
00396             uvx  = vec_mradds(V, lCGV, uvx);                                  \
00397             uvx0 = vec_mergeh(uvx, uvx);                                      \
00398             uvx1 = vec_mergel(uvx, uvx);                                      \
00399                                                                               \
00400             R0 = vec_add(Y0, vx0);                                            \
00401             G0 = vec_add(Y0, uvx0);                                           \
00402             B0 = vec_add(Y0, ux0);                                            \
00403             R1 = vec_add(Y1, vx1);                                            \
00404             G1 = vec_add(Y1, uvx1);                                           \
00405             B1 = vec_add(Y1, ux1);                                            \
00406                                                                               \
00407             R = vec_packclp(R0, R1);                                          \
00408             G = vec_packclp(G0, G1);                                          \
00409             B = vec_packclp(B0, B1);                                          \
00410                                                                               \
00411             out_pixels(R, G, B, oute);                                        \
00412                                                                               \
00413             R0 = vec_add(Y2, vx0);                                            \
00414             G0 = vec_add(Y2, uvx0);                                           \
00415             B0 = vec_add(Y2, ux0);                                            \
00416             R1 = vec_add(Y3, vx1);                                            \
00417             G1 = vec_add(Y3, uvx1);                                           \
00418             B1 = vec_add(Y3, ux1);                                            \
00419             R  = vec_packclp(R0, R1);                                         \
00420             G  = vec_packclp(G0, G1);                                         \
00421             B  = vec_packclp(B0, B1);                                         \
00422                                                                               \
00423                                                                               \
00424             out_pixels(R, G, B, outo);                                        \
00425                                                                               \
00426             y1i += 16;                                                        \
00427             y2i += 16;                                                        \
00428             ui  += 8;                                                         \
00429             vi  += 8;                                                         \
00430         }                                                                     \
00431                                                                               \
00432         outo += (outstrides[0]) >> 4;                                         \
00433         oute += (outstrides[0]) >> 4;                                         \
00434                                                                               \
00435         ui  += instrides_scl[1];                                              \
00436         vi  += instrides_scl[2];                                              \
00437         y1i += instrides_scl[0];                                              \
00438         y2i += instrides_scl[0];                                              \
00439     }                                                                         \
00440     return srcSliceH;                                                         \
00441 }
00442 
00443 #define out_abgr(a, b, c, ptr)                                          \
00444     vec_mstrgb32(__typeof__(a), ((__typeof__(a)) { 255 }), c, b, a, ptr)
00445 #define out_bgra(a, b, c, ptr)                                          \
00446     vec_mstrgb32(__typeof__(a), c, b, a, ((__typeof__(a)) { 255 }), ptr)
00447 #define out_rgba(a, b, c, ptr)                                          \
00448     vec_mstrgb32(__typeof__(a), a, b, c, ((__typeof__(a)) { 255 }), ptr)
00449 #define out_argb(a, b, c, ptr)                                          \
00450     vec_mstrgb32(__typeof__(a), ((__typeof__(a)) { 255 }), a, b, c, ptr)
00451 #define out_rgb24(a, b, c, ptr) vec_mstrgb24(a, b, c, ptr)
00452 #define out_bgr24(a, b, c, ptr) vec_mstbgr24(a, b, c, ptr)
00453 
00454 DEFCSP420_CVT(yuv2_abgr,  out_abgr)
00455 DEFCSP420_CVT(yuv2_bgra,  out_bgra)
00456 DEFCSP420_CVT(yuv2_rgba,  out_rgba)
00457 DEFCSP420_CVT(yuv2_argb,  out_argb)
00458 DEFCSP420_CVT(yuv2_rgb24, out_rgb24)
00459 DEFCSP420_CVT(yuv2_bgr24, out_bgr24)
00460 
00461 // uyvy|uyvy|uyvy|uyvy
00462 // 0123 4567 89ab cdef
00463 static const vector unsigned char
00464     demux_u = { 0x10, 0x00, 0x10, 0x00,
00465                 0x10, 0x04, 0x10, 0x04,
00466                 0x10, 0x08, 0x10, 0x08,
00467                 0x10, 0x0c, 0x10, 0x0c },
00468     demux_v = { 0x10, 0x02, 0x10, 0x02,
00469                 0x10, 0x06, 0x10, 0x06,
00470                 0x10, 0x0A, 0x10, 0x0A,
00471                 0x10, 0x0E, 0x10, 0x0E },
00472     demux_y = { 0x10, 0x01, 0x10, 0x03,
00473                 0x10, 0x05, 0x10, 0x07,
00474                 0x10, 0x09, 0x10, 0x0B,
00475                 0x10, 0x0D, 0x10, 0x0F };
00476 
00477 /*
00478  * this is so I can play live CCIR raw video
00479  */
00480 static int altivec_uyvy_rgb32(SwsContext *c, const unsigned char **in,
00481                               int *instrides, int srcSliceY, int srcSliceH,
00482                               unsigned char **oplanes, int *outstrides)
00483 {
00484     int w = c->srcW;
00485     int h = srcSliceH;
00486     int i, j;
00487     vector unsigned char uyvy;
00488     vector signed short Y, U, V;
00489     vector signed short R0, G0, B0, R1, G1, B1;
00490     vector unsigned char R, G, B;
00491     vector unsigned char *out;
00492     const ubyte *img;
00493 
00494     img = in[0];
00495     out = (vector unsigned char *) (oplanes[0] + srcSliceY * outstrides[0]);
00496 
00497     for (i = 0; i < h; i++)
00498         for (j = 0; j < w / 16; j++) {
00499             uyvy = vec_ld(0, img);
00500 
00501             U = (vector signed short)
00502                     vec_perm(uyvy, (vector unsigned char) { 0 }, demux_u);
00503             V = (vector signed short)
00504                     vec_perm(uyvy, (vector unsigned char) { 0 }, demux_v);
00505             Y = (vector signed short)
00506                     vec_perm(uyvy, (vector unsigned char) { 0 }, demux_y);
00507 
00508             cvtyuvtoRGB(c, Y, U, V, &R0, &G0, &B0);
00509 
00510             uyvy = vec_ld(16, img);
00511 
00512             U = (vector signed short)
00513                     vec_perm(uyvy, (vector unsigned char) { 0 }, demux_u);
00514             V = (vector signed short)
00515                     vec_perm(uyvy, (vector unsigned char) { 0 }, demux_v);
00516             Y = (vector signed short)
00517                     vec_perm(uyvy, (vector unsigned char) { 0 }, demux_y);
00518 
00519             cvtyuvtoRGB(c, Y, U, V, &R1, &G1, &B1);
00520 
00521             R = vec_packclp(R0, R1);
00522             G = vec_packclp(G0, G1);
00523             B = vec_packclp(B0, B1);
00524 
00525             // vec_mstbgr24 (R,G,B, out);
00526             out_rgba(R, G, B, out);
00527 
00528             img += 32;
00529         }
00530     return srcSliceH;
00531 }
00532 
00533 /* Ok currently the acceleration routine only supports
00534  * inputs of widths a multiple of 16
00535  * and heights a multiple 2
00536  *
00537  * So we just fall back to the C codes for this.
00538  */
00539 SwsFunc ff_yuv2rgb_init_altivec(SwsContext *c)
00540 {
00541     if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC))
00542         return NULL;
00543 
00544     /*
00545      * and this seems not to matter too much I tried a bunch of
00546      * videos with abnormal widths and MPlayer crashes elsewhere.
00547      * mplayer -vo x11 -rawvideo on:w=350:h=240 raw-350x240.eyuv
00548      * boom with X11 bad match.
00549      *
00550      */
00551     if ((c->srcW & 0xf) != 0)
00552         return NULL;
00553 
00554     switch (c->srcFormat) {
00555     case PIX_FMT_YUV410P:
00556     case PIX_FMT_YUV420P:
00557     /*case IMGFMT_CLPL:        ??? */
00558     case PIX_FMT_GRAY8:
00559     case PIX_FMT_NV12:
00560     case PIX_FMT_NV21:
00561         if ((c->srcH & 0x1) != 0)
00562             return NULL;
00563 
00564         switch (c->dstFormat) {
00565         case PIX_FMT_RGB24:
00566             av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space RGB24\n");
00567             return altivec_yuv2_rgb24;
00568         case PIX_FMT_BGR24:
00569             av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space BGR24\n");
00570             return altivec_yuv2_bgr24;
00571         case PIX_FMT_ARGB:
00572             av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space ARGB\n");
00573             return altivec_yuv2_argb;
00574         case PIX_FMT_ABGR:
00575             av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space ABGR\n");
00576             return altivec_yuv2_abgr;
00577         case PIX_FMT_RGBA:
00578             av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space RGBA\n");
00579             return altivec_yuv2_rgba;
00580         case PIX_FMT_BGRA:
00581             av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space BGRA\n");
00582             return altivec_yuv2_bgra;
00583         default: return NULL;
00584         }
00585         break;
00586 
00587     case PIX_FMT_UYVY422:
00588         switch (c->dstFormat) {
00589         case PIX_FMT_BGR32:
00590             av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space UYVY -> RGB32\n");
00591             return altivec_uyvy_rgb32;
00592         default: return NULL;
00593         }
00594         break;
00595     }
00596     return NULL;
00597 }
00598 
00599 void ff_yuv2rgb_init_tables_altivec(SwsContext *c, const int inv_table[4],
00600                                     int brightness, int contrast,
00601                                     int saturation)
00602 {
00603     union {
00604         DECLARE_ALIGNED(16, signed short, tmp)[8];
00605         vector signed short vec;
00606     } buf;
00607 
00608     buf.tmp[0] = ((0xffffLL) * contrast >> 8) >> 9;                               // cy
00609     buf.tmp[1] = -256 * brightness;                                               // oy
00610     buf.tmp[2] =   (inv_table[0] >> 3) * (contrast >> 16) * (saturation >> 16);   // crv
00611     buf.tmp[3] =   (inv_table[1] >> 3) * (contrast >> 16) * (saturation >> 16);   // cbu
00612     buf.tmp[4] = -((inv_table[2] >> 1) * (contrast >> 16) * (saturation >> 16));  // cgu
00613     buf.tmp[5] = -((inv_table[3] >> 1) * (contrast >> 16) * (saturation >> 16));  // cgv
00614 
00615     c->CSHIFT = (vector unsigned short) vec_splat_u16(2);
00616     c->CY     = vec_splat((vector signed short) buf.vec, 0);
00617     c->OY     = vec_splat((vector signed short) buf.vec, 1);
00618     c->CRV    = vec_splat((vector signed short) buf.vec, 2);
00619     c->CBU    = vec_splat((vector signed short) buf.vec, 3);
00620     c->CGU    = vec_splat((vector signed short) buf.vec, 4);
00621     c->CGV    = vec_splat((vector signed short) buf.vec, 5);
00622     return;
00623 }
00624 
00625 static av_always_inline void ff_yuv2packedX_altivec(SwsContext *c,
00626                                                     const int16_t *lumFilter,
00627                                                     const int16_t **lumSrc,
00628                                                     int lumFilterSize,
00629                                                     const int16_t *chrFilter,
00630                                                     const int16_t **chrUSrc,
00631                                                     const int16_t **chrVSrc,
00632                                                     int chrFilterSize,
00633                                                     const int16_t **alpSrc,
00634                                                     uint8_t *dest,
00635                                                     int dstW, int dstY,
00636                                                     enum PixelFormat target)
00637 {
00638     int i, j;
00639     vector signed short X, X0, X1, Y0, U0, V0, Y1, U1, V1, U, V;
00640     vector signed short R0, G0, B0, R1, G1, B1;
00641 
00642     vector unsigned char R, G, B;
00643     vector unsigned char *out, *nout;
00644 
00645     vector signed short RND   = vec_splat_s16(1 << 3);
00646     vector unsigned short SCL = vec_splat_u16(4);
00647     DECLARE_ALIGNED(16, unsigned int, scratch)[16];
00648 
00649     vector signed short *YCoeffs, *CCoeffs;
00650 
00651     YCoeffs = c->vYCoeffsBank + dstY * lumFilterSize;
00652     CCoeffs = c->vCCoeffsBank + dstY * chrFilterSize;
00653 
00654     out = (vector unsigned char *) dest;
00655 
00656     for (i = 0; i < dstW; i += 16) {
00657         Y0 = RND;
00658         Y1 = RND;
00659         /* extract 16 coeffs from lumSrc */
00660         for (j = 0; j < lumFilterSize; j++) {
00661             X0 = vec_ld(0, &lumSrc[j][i]);
00662             X1 = vec_ld(16, &lumSrc[j][i]);
00663             Y0 = vec_mradds(X0, YCoeffs[j], Y0);
00664             Y1 = vec_mradds(X1, YCoeffs[j], Y1);
00665         }
00666 
00667         U = RND;
00668         V = RND;
00669         /* extract 8 coeffs from U,V */
00670         for (j = 0; j < chrFilterSize; j++) {
00671             X = vec_ld(0, &chrUSrc[j][i / 2]);
00672             U = vec_mradds(X, CCoeffs[j], U);
00673             X = vec_ld(0, &chrVSrc[j][i / 2]);
00674             V = vec_mradds(X, CCoeffs[j], V);
00675         }
00676 
00677         /* scale and clip signals */
00678         Y0 = vec_sra(Y0, SCL);
00679         Y1 = vec_sra(Y1, SCL);
00680         U  = vec_sra(U, SCL);
00681         V  = vec_sra(V, SCL);
00682 
00683         Y0 = vec_clip_s16(Y0);
00684         Y1 = vec_clip_s16(Y1);
00685         U  = vec_clip_s16(U);
00686         V  = vec_clip_s16(V);
00687 
00688         /* now we have
00689          * Y0 = y0 y1 y2 y3 y4 y5 y6 y7    Y1 = y8 y9 y10 y11 y12 y13 y14 y15
00690          * U  = u0 u1 u2 u3 u4 u5 u6 u7    V  = v0 v1 v2 v3 v4 v5 v6 v7
00691          *
00692          * Y0 = y0 y1 y2 y3 y4 y5 y6 y7    Y1 = y8 y9 y10 y11 y12 y13 y14 y15
00693          * U0 = u0 u0 u1 u1 u2 u2 u3 u3    U1 = u4 u4 u5 u5 u6 u6 u7 u7
00694          * V0 = v0 v0 v1 v1 v2 v2 v3 v3    V1 = v4 v4 v5 v5 v6 v6 v7 v7
00695          */
00696 
00697         U0 = vec_mergeh(U, U);
00698         V0 = vec_mergeh(V, V);
00699 
00700         U1 = vec_mergel(U, U);
00701         V1 = vec_mergel(V, V);
00702 
00703         cvtyuvtoRGB(c, Y0, U0, V0, &R0, &G0, &B0);
00704         cvtyuvtoRGB(c, Y1, U1, V1, &R1, &G1, &B1);
00705 
00706         R = vec_packclp(R0, R1);
00707         G = vec_packclp(G0, G1);
00708         B = vec_packclp(B0, B1);
00709 
00710         switch (target) {
00711         case PIX_FMT_ABGR:
00712             out_abgr(R, G, B, out);
00713             break;
00714         case PIX_FMT_BGRA:
00715             out_bgra(R, G, B, out);
00716             break;
00717         case PIX_FMT_RGBA:
00718             out_rgba(R, G, B, out);
00719             break;
00720         case PIX_FMT_ARGB:
00721             out_argb(R, G, B, out);
00722             break;
00723         case PIX_FMT_RGB24:
00724             out_rgb24(R, G, B, out);
00725             break;
00726         case PIX_FMT_BGR24:
00727             out_bgr24(R, G, B, out);
00728             break;
00729         default:
00730         {
00731             /* If this is reached, the caller should have called yuv2packedXinC
00732              * instead. */
00733             static int printed_error_message;
00734             if (!printed_error_message) {
00735                 av_log(c, AV_LOG_ERROR,
00736                        "altivec_yuv2packedX doesn't support %s output\n",
00737                        av_get_pix_fmt_name(c->dstFormat));
00738                 printed_error_message = 1;
00739             }
00740             return;
00741         }
00742         }
00743     }
00744 
00745     if (i < dstW) {
00746         i -= 16;
00747 
00748         Y0 = RND;
00749         Y1 = RND;
00750         /* extract 16 coeffs from lumSrc */
00751         for (j = 0; j < lumFilterSize; j++) {
00752             X0 = vec_ld(0, &lumSrc[j][i]);
00753             X1 = vec_ld(16, &lumSrc[j][i]);
00754             Y0 = vec_mradds(X0, YCoeffs[j], Y0);
00755             Y1 = vec_mradds(X1, YCoeffs[j], Y1);
00756         }
00757 
00758         U = RND;
00759         V = RND;
00760         /* extract 8 coeffs from U,V */
00761         for (j = 0; j < chrFilterSize; j++) {
00762             X = vec_ld(0, &chrUSrc[j][i / 2]);
00763             U = vec_mradds(X, CCoeffs[j], U);
00764             X = vec_ld(0, &chrVSrc[j][i / 2]);
00765             V = vec_mradds(X, CCoeffs[j], V);
00766         }
00767 
00768         /* scale and clip signals */
00769         Y0 = vec_sra(Y0, SCL);
00770         Y1 = vec_sra(Y1, SCL);
00771         U  = vec_sra(U, SCL);
00772         V  = vec_sra(V, SCL);
00773 
00774         Y0 = vec_clip_s16(Y0);
00775         Y1 = vec_clip_s16(Y1);
00776         U  = vec_clip_s16(U);
00777         V  = vec_clip_s16(V);
00778 
00779         /* now we have
00780          * Y0 = y0 y1 y2 y3 y4 y5 y6 y7    Y1 = y8 y9 y10 y11 y12 y13 y14 y15
00781          * U  = u0 u1 u2 u3 u4 u5 u6 u7    V  = v0 v1 v2 v3 v4 v5 v6 v7
00782          *
00783          * Y0 = y0 y1 y2 y3 y4 y5 y6 y7    Y1 = y8 y9 y10 y11 y12 y13 y14 y15
00784          * U0 = u0 u0 u1 u1 u2 u2 u3 u3    U1 = u4 u4 u5 u5 u6 u6 u7 u7
00785          * V0 = v0 v0 v1 v1 v2 v2 v3 v3    V1 = v4 v4 v5 v5 v6 v6 v7 v7
00786          */
00787 
00788         U0 = vec_mergeh(U, U);
00789         V0 = vec_mergeh(V, V);
00790 
00791         U1 = vec_mergel(U, U);
00792         V1 = vec_mergel(V, V);
00793 
00794         cvtyuvtoRGB(c, Y0, U0, V0, &R0, &G0, &B0);
00795         cvtyuvtoRGB(c, Y1, U1, V1, &R1, &G1, &B1);
00796 
00797         R = vec_packclp(R0, R1);
00798         G = vec_packclp(G0, G1);
00799         B = vec_packclp(B0, B1);
00800 
00801         nout = (vector unsigned char *) scratch;
00802         switch (target) {
00803         case PIX_FMT_ABGR:
00804             out_abgr(R, G, B, nout);
00805             break;
00806         case PIX_FMT_BGRA:
00807             out_bgra(R, G, B, nout);
00808             break;
00809         case PIX_FMT_RGBA:
00810             out_rgba(R, G, B, nout);
00811             break;
00812         case PIX_FMT_ARGB:
00813             out_argb(R, G, B, nout);
00814             break;
00815         case PIX_FMT_RGB24:
00816             out_rgb24(R, G, B, nout);
00817             break;
00818         case PIX_FMT_BGR24:
00819             out_bgr24(R, G, B, nout);
00820             break;
00821         default:
00822             /* Unreachable, I think. */
00823             av_log(c, AV_LOG_ERROR,
00824                    "altivec_yuv2packedX doesn't support %s output\n",
00825                    av_get_pix_fmt_name(c->dstFormat));
00826             return;
00827         }
00828 
00829         memcpy(&((uint32_t *) dest)[i], scratch, (dstW - i) / 4);
00830     }
00831 }
00832 
00833 #define YUV2PACKEDX_WRAPPER(suffix, pixfmt)                             \
00834 void ff_yuv2 ## suffix ## _X_altivec(SwsContext *c,                     \
00835                                      const int16_t *lumFilter,          \
00836                                      const int16_t **lumSrc,            \
00837                                      int lumFilterSize,                 \
00838                                      const int16_t *chrFilter,          \
00839                                      const int16_t **chrUSrc,           \
00840                                      const int16_t **chrVSrc,           \
00841                                      int chrFilterSize,                 \
00842                                      const int16_t **alpSrc,            \
00843                                      uint8_t *dest, int dstW, int dstY) \
00844 {                                                                       \
00845     ff_yuv2packedX_altivec(c, lumFilter, lumSrc, lumFilterSize,         \
00846                            chrFilter, chrUSrc, chrVSrc,                 \
00847                            chrFilterSize, alpSrc,                       \
00848                            dest, dstW, dstY, pixfmt);                   \
00849 }
00850 
00851 YUV2PACKEDX_WRAPPER(abgr,  PIX_FMT_ABGR);
00852 YUV2PACKEDX_WRAPPER(bgra,  PIX_FMT_BGRA);
00853 YUV2PACKEDX_WRAPPER(argb,  PIX_FMT_ARGB);
00854 YUV2PACKEDX_WRAPPER(rgba,  PIX_FMT_RGBA);
00855 YUV2PACKEDX_WRAPPER(rgb24, PIX_FMT_RGB24);
00856 YUV2PACKEDX_WRAPPER(bgr24, PIX_FMT_BGR24);