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/attributes.h"
00099 #include "libavutil/cpu.h"
00100 #include "libavutil/pixdesc.h"
00101 #include "yuv2rgb_altivec.h"
00102 
00103 #undef PROFILE_THE_BEAST
00104 #undef INC_SCALING
00105 
00106 typedef unsigned char ubyte;
00107 typedef signed char   sbyte;
00108 
00109 /* RGB interleaver, 16 planar pels 8-bit samples per channel in
00110  * homogeneous vector registers x0,x1,x2 are interleaved with the
00111  * following technique:
00112  *
00113  *    o0 = vec_mergeh(x0, x1);
00114  *    o1 = vec_perm(o0, x2, perm_rgb_0);
00115  *    o2 = vec_perm(o0, x2, perm_rgb_1);
00116  *    o3 = vec_mergel(x0, x1);
00117  *    o4 = vec_perm(o3, o2, perm_rgb_2);
00118  *    o5 = vec_perm(o3, o2, perm_rgb_3);
00119  *
00120  * perm_rgb_0:   o0(RG).h v1(B) --> o1*
00121  *            0   1  2   3   4
00122  *           rgbr|gbrg|brgb|rgbr
00123  *           0010 0100 1001 0010
00124  *           0102 3145 2673 894A
00125  *
00126  * perm_rgb_1:   o0(RG).h v1(B) --> o2
00127  *            0   1  2   3   4
00128  *           gbrg|brgb|bbbb|bbbb
00129  *           0100 1001 1111 1111
00130  *           B5CD 6EF7 89AB CDEF
00131  *
00132  * perm_rgb_2:   o3(RG).l o2(rgbB.l) --> o4*
00133  *            0   1  2   3   4
00134  *           gbrg|brgb|rgbr|gbrg
00135  *           1111 1111 0010 0100
00136  *           89AB CDEF 0182 3945
00137  *
00138  * perm_rgb_2:   o3(RG).l o2(rgbB.l) ---> o5*
00139  *            0   1  2   3   4
00140  *           brgb|rgbr|gbrg|brgb
00141  *           1001 0010 0100 1001
00142  *           a67b 89cA BdCD eEFf
00143  *
00144  */
00145 static const vector unsigned char
00146     perm_rgb_0 = { 0x00, 0x01, 0x10, 0x02, 0x03, 0x11, 0x04, 0x05,
00147                    0x12, 0x06, 0x07, 0x13, 0x08, 0x09, 0x14, 0x0a },
00148     perm_rgb_1 = { 0x0b, 0x15, 0x0c, 0x0d, 0x16, 0x0e, 0x0f, 0x17,
00149                    0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f },
00150     perm_rgb_2 = { 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17,
00151                    0x00, 0x01, 0x18, 0x02, 0x03, 0x19, 0x04, 0x05 },
00152     perm_rgb_3 = { 0x1a, 0x06, 0x07, 0x1b, 0x08, 0x09, 0x1c, 0x0a,
00153                    0x0b, 0x1d, 0x0c, 0x0d, 0x1e, 0x0e, 0x0f, 0x1f };
00154 
00155 #define vec_merge3(x2, x1, x0, y0, y1, y2)     \
00156     do {                                       \
00157         __typeof__(x0) o0, o2, o3;             \
00158         o0 = vec_mergeh(x0, x1);               \
00159         y0 = vec_perm(o0, x2, perm_rgb_0);     \
00160         o2 = vec_perm(o0, x2, perm_rgb_1);     \
00161         o3 = vec_mergel(x0, x1);               \
00162         y1 = vec_perm(o3, o2, perm_rgb_2);     \
00163         y2 = vec_perm(o3, o2, perm_rgb_3);     \
00164     } while (0)
00165 
00166 #define vec_mstbgr24(x0, x1, x2, ptr)          \
00167     do {                                       \
00168         __typeof__(x0) _0, _1, _2;             \
00169         vec_merge3(x0, x1, x2, _0, _1, _2);    \
00170         vec_st(_0, 0, ptr++);                  \
00171         vec_st(_1, 0, ptr++);                  \
00172         vec_st(_2, 0, ptr++);                  \
00173     } while (0)
00174 
00175 #define vec_mstrgb24(x0, x1, x2, ptr)          \
00176     do {                                       \
00177         __typeof__(x0) _0, _1, _2;             \
00178         vec_merge3(x2, x1, x0, _0, _1, _2);    \
00179         vec_st(_0, 0, ptr++);                  \
00180         vec_st(_1, 0, ptr++);                  \
00181         vec_st(_2, 0, ptr++);                  \
00182     } while (0)
00183 
00184 /* pack the pixels in rgb0 format
00185  * msb R
00186  * lsb 0
00187  */
00188 #define vec_mstrgb32(T, x0, x1, x2, x3, ptr)                            \
00189     do {                                                                \
00190         T _0, _1, _2, _3;                                               \
00191         _0 = vec_mergeh(x0, x1);                                        \
00192         _1 = vec_mergeh(x2, x3);                                        \
00193         _2 = (T) vec_mergeh((vector unsigned short) _0,                 \
00194                             (vector unsigned short) _1);                \
00195         _3 = (T) vec_mergel((vector unsigned short) _0,                 \
00196                             (vector unsigned short) _1);                \
00197         vec_st(_2, 0 * 16, (T *) ptr);                                  \
00198         vec_st(_3, 1 * 16, (T *) ptr);                                  \
00199         _0 = vec_mergel(x0, x1);                                        \
00200         _1 = vec_mergel(x2, x3);                                        \
00201         _2 = (T) vec_mergeh((vector unsigned short) _0,                 \
00202                             (vector unsigned short) _1);                \
00203         _3 = (T) vec_mergel((vector unsigned short) _0,                 \
00204                             (vector unsigned short) _1);                \
00205         vec_st(_2, 2 * 16, (T *) ptr);                                  \
00206         vec_st(_3, 3 * 16, (T *) ptr);                                  \
00207         ptr += 4;                                                       \
00208     } while (0)
00209 
00210 /*
00211  * 1     0       1.4021   | | Y |
00212  * 1    -0.3441 -0.7142   |x| Cb|
00213  * 1     1.7718  0        | | Cr|
00214  *
00215  *
00216  * Y:      [-128 127]
00217  * Cb/Cr : [-128 127]
00218  *
00219  * typical YUV conversion works on Y: 0-255 this version has been
00220  * optimized for JPEG decoding.
00221  */
00222 
00223 #define vec_unh(x)                                                      \
00224     (vector signed short)                                               \
00225         vec_perm(x, (__typeof__(x)) { 0 },                              \
00226                  ((vector unsigned char) {                              \
00227                      0x10, 0x00, 0x10, 0x01, 0x10, 0x02, 0x10, 0x03,    \
00228                      0x10, 0x04, 0x10, 0x05, 0x10, 0x06, 0x10, 0x07 }))
00229 
00230 #define vec_unl(x)                                                      \
00231     (vector signed short)                                               \
00232         vec_perm(x, (__typeof__(x)) { 0 },                              \
00233                  ((vector unsigned char) {                              \
00234                      0x10, 0x08, 0x10, 0x09, 0x10, 0x0A, 0x10, 0x0B,    \
00235                      0x10, 0x0C, 0x10, 0x0D, 0x10, 0x0E, 0x10, 0x0F }))
00236 
00237 #define vec_clip_s16(x)                                                 \
00238     vec_max(vec_min(x, ((vector signed short) {                         \
00239                     235, 235, 235, 235, 235, 235, 235, 235 })),         \
00240             ((vector signed short) { 16, 16, 16, 16, 16, 16, 16, 16 }))
00241 
00242 #define vec_packclp(x, y)                                               \
00243     (vector unsigned char)                                              \
00244         vec_packs((vector unsigned short)                               \
00245                       vec_max(x, ((vector signed short) { 0 })),        \
00246                   (vector unsigned short)                               \
00247                       vec_max(y, ((vector signed short) { 0 })))
00248 
00249 //#define out_pixels(a, b, c, ptr) vec_mstrgb32(__typeof__(a), ((__typeof__(a)) { 255 }), a, a, a, ptr)
00250 
00251 static inline void cvtyuvtoRGB(SwsContext *c, vector signed short Y,
00252                                vector signed short U, vector signed short V,
00253                                vector signed short *R, vector signed short *G,
00254                                vector signed short *B)
00255 {
00256     vector signed short vx, ux, uvx;
00257 
00258     Y = vec_mradds(Y, c->CY, c->OY);
00259     U = vec_sub(U, (vector signed short)
00260                        vec_splat((vector signed short) { 128 }, 0));
00261     V = vec_sub(V, (vector signed short)
00262                        vec_splat((vector signed short) { 128 }, 0));
00263 
00264     // ux  = (CBU * (u << c->CSHIFT) + 0x4000) >> 15;
00265     ux = vec_sl(U, c->CSHIFT);
00266     *B = vec_mradds(ux, c->CBU, Y);
00267 
00268     // vx  = (CRV * (v << c->CSHIFT) + 0x4000) >> 15;
00269     vx = vec_sl(V, c->CSHIFT);
00270     *R = vec_mradds(vx, c->CRV, Y);
00271 
00272     // uvx = ((CGU * u) + (CGV * v)) >> 15;
00273     uvx = vec_mradds(U, c->CGU, Y);
00274     *G  = vec_mradds(V, c->CGV, uvx);
00275 }
00276 
00277 /*
00278  * ------------------------------------------------------------------------------
00279  * CS converters
00280  * ------------------------------------------------------------------------------
00281  */
00282 
00283 #define DEFCSP420_CVT(name, out_pixels)                                       \
00284 static int altivec_ ## name(SwsContext *c, const unsigned char **in,          \
00285                             int *instrides, int srcSliceY, int srcSliceH,     \
00286                             unsigned char **oplanes, int *outstrides)         \
00287 {                                                                             \
00288     int w = c->srcW;                                                          \
00289     int h = srcSliceH;                                                        \
00290     int i, j;                                                                 \
00291     int instrides_scl[3];                                                     \
00292     vector unsigned char y0, y1;                                              \
00293                                                                               \
00294     vector signed char u, v;                                                  \
00295                                                                               \
00296     vector signed short Y0, Y1, Y2, Y3;                                       \
00297     vector signed short U, V;                                                 \
00298     vector signed short vx, ux, uvx;                                          \
00299     vector signed short vx0, ux0, uvx0;                                       \
00300     vector signed short vx1, ux1, uvx1;                                       \
00301     vector signed short R0, G0, B0;                                           \
00302     vector signed short R1, G1, B1;                                           \
00303     vector unsigned char R, G, B;                                             \
00304                                                                               \
00305     const vector unsigned char *y1ivP, *y2ivP, *uivP, *vivP;                  \
00306     vector unsigned char align_perm;                                          \
00307                                                                               \
00308     vector signed short lCY       = c->CY;                                    \
00309     vector signed short lOY       = c->OY;                                    \
00310     vector signed short lCRV      = c->CRV;                                   \
00311     vector signed short lCBU      = c->CBU;                                   \
00312     vector signed short lCGU      = c->CGU;                                   \
00313     vector signed short lCGV      = c->CGV;                                   \
00314     vector unsigned short lCSHIFT = c->CSHIFT;                                \
00315                                                                               \
00316     const ubyte *y1i = in[0];                                                 \
00317     const ubyte *y2i = in[0] + instrides[0];                                  \
00318     const ubyte *ui  = in[1];                                                 \
00319     const ubyte *vi  = in[2];                                                 \
00320                                                                               \
00321     vector unsigned char *oute, *outo;                                        \
00322                                                                               \
00323     /* loop moves y{1, 2}i by w */                                            \
00324     instrides_scl[0] = instrides[0] * 2 - w;                                  \
00325     /* loop moves ui by w / 2 */                                              \
00326     instrides_scl[1] = instrides[1] - w / 2;                                  \
00327     /* loop moves vi by w / 2 */                                              \
00328     instrides_scl[2] = instrides[2] - w / 2;                                  \
00329                                                                               \
00330     for (i = 0; i < h / 2; i++) {                                             \
00331         oute = (vector unsigned char *)(oplanes[0] + outstrides[0] *          \
00332                                         (srcSliceY + i * 2));                 \
00333         outo = oute + (outstrides[0] >> 4);                                   \
00334         vec_dstst(outo, (0x02000002 | (((w * 3 + 32) / 32) << 16)), 0);       \
00335         vec_dstst(oute, (0x02000002 | (((w * 3 + 32) / 32) << 16)), 1);       \
00336                                                                               \
00337         for (j = 0; j < w / 16; j++) {                                        \
00338             y1ivP = (const vector unsigned char *) y1i;                       \
00339             y2ivP = (const vector unsigned char *) y2i;                       \
00340             uivP  = (const vector unsigned char *) ui;                        \
00341             vivP  = (const vector unsigned char *) vi;                        \
00342                                                                               \
00343             align_perm = vec_lvsl(0, y1i);                                    \
00344             y0 = (vector unsigned char)                                       \
00345                      vec_perm(y1ivP[0], y1ivP[1], align_perm);                \
00346                                                                               \
00347             align_perm = vec_lvsl(0, y2i);                                    \
00348             y1 = (vector unsigned char)                                       \
00349                      vec_perm(y2ivP[0], y2ivP[1], align_perm);                \
00350                                                                               \
00351             align_perm = vec_lvsl(0, ui);                                     \
00352             u = (vector signed char)                                          \
00353                     vec_perm(uivP[0], uivP[1], align_perm);                   \
00354                                                                               \
00355             align_perm = vec_lvsl(0, vi);                                     \
00356             v = (vector signed char)                                          \
00357                     vec_perm(vivP[0], vivP[1], align_perm);                   \
00358                                                                               \
00359             u = (vector signed char)                                          \
00360                     vec_sub(u,                                                \
00361                             (vector signed char)                              \
00362                                 vec_splat((vector signed char) { 128 }, 0));  \
00363             v = (vector signed char)                                          \
00364                     vec_sub(v,                                                \
00365                             (vector signed char)                              \
00366                                 vec_splat((vector signed char) { 128 }, 0));  \
00367                                                                               \
00368             U = vec_unpackh(u);                                               \
00369             V = vec_unpackh(v);                                               \
00370                                                                               \
00371             Y0 = vec_unh(y0);                                                 \
00372             Y1 = vec_unl(y0);                                                 \
00373             Y2 = vec_unh(y1);                                                 \
00374             Y3 = vec_unl(y1);                                                 \
00375                                                                               \
00376             Y0 = vec_mradds(Y0, lCY, lOY);                                    \
00377             Y1 = vec_mradds(Y1, lCY, lOY);                                    \
00378             Y2 = vec_mradds(Y2, lCY, lOY);                                    \
00379             Y3 = vec_mradds(Y3, lCY, lOY);                                    \
00380                                                                               \
00381             /* ux  = (CBU * (u << CSHIFT) + 0x4000) >> 15 */                  \
00382             ux  = vec_sl(U, lCSHIFT);                                         \
00383             ux  = vec_mradds(ux, lCBU, (vector signed short) { 0 });          \
00384             ux0 = vec_mergeh(ux, ux);                                         \
00385             ux1 = vec_mergel(ux, ux);                                         \
00386                                                                               \
00387             /* vx  = (CRV * (v << CSHIFT) + 0x4000) >> 15; */                 \
00388             vx  = vec_sl(V, lCSHIFT);                                         \
00389             vx  = vec_mradds(vx, lCRV, (vector signed short) { 0 });          \
00390             vx0 = vec_mergeh(vx, vx);                                         \
00391             vx1 = vec_mergel(vx, vx);                                         \
00392                                                                               \
00393             /* uvx = ((CGU * u) + (CGV * v)) >> 15 */                         \
00394             uvx  = vec_mradds(U, lCGU, (vector signed short) { 0 });          \
00395             uvx  = vec_mradds(V, lCGV, uvx);                                  \
00396             uvx0 = vec_mergeh(uvx, uvx);                                      \
00397             uvx1 = vec_mergel(uvx, uvx);                                      \
00398                                                                               \
00399             R0 = vec_add(Y0, vx0);                                            \
00400             G0 = vec_add(Y0, uvx0);                                           \
00401             B0 = vec_add(Y0, ux0);                                            \
00402             R1 = vec_add(Y1, vx1);                                            \
00403             G1 = vec_add(Y1, uvx1);                                           \
00404             B1 = vec_add(Y1, ux1);                                            \
00405                                                                               \
00406             R = vec_packclp(R0, R1);                                          \
00407             G = vec_packclp(G0, G1);                                          \
00408             B = vec_packclp(B0, B1);                                          \
00409                                                                               \
00410             out_pixels(R, G, B, oute);                                        \
00411                                                                               \
00412             R0 = vec_add(Y2, vx0);                                            \
00413             G0 = vec_add(Y2, uvx0);                                           \
00414             B0 = vec_add(Y2, ux0);                                            \
00415             R1 = vec_add(Y3, vx1);                                            \
00416             G1 = vec_add(Y3, uvx1);                                           \
00417             B1 = vec_add(Y3, ux1);                                            \
00418             R  = vec_packclp(R0, R1);                                         \
00419             G  = vec_packclp(G0, G1);                                         \
00420             B  = vec_packclp(B0, B1);                                         \
00421                                                                               \
00422                                                                               \
00423             out_pixels(R, G, B, outo);                                        \
00424                                                                               \
00425             y1i += 16;                                                        \
00426             y2i += 16;                                                        \
00427             ui  += 8;                                                         \
00428             vi  += 8;                                                         \
00429         }                                                                     \
00430                                                                               \
00431         ui  += instrides_scl[1];                                              \
00432         vi  += instrides_scl[2];                                              \
00433         y1i += instrides_scl[0];                                              \
00434         y2i += instrides_scl[0];                                              \
00435     }                                                                         \
00436     return srcSliceH;                                                         \
00437 }
00438 
00439 #define out_abgr(a, b, c, ptr)                                          \
00440     vec_mstrgb32(__typeof__(a), ((__typeof__(a)) { 255 }), c, b, a, ptr)
00441 #define out_bgra(a, b, c, ptr)                                          \
00442     vec_mstrgb32(__typeof__(a), c, b, a, ((__typeof__(a)) { 255 }), ptr)
00443 #define out_rgba(a, b, c, ptr)                                          \
00444     vec_mstrgb32(__typeof__(a), a, b, c, ((__typeof__(a)) { 255 }), ptr)
00445 #define out_argb(a, b, c, ptr)                                          \
00446     vec_mstrgb32(__typeof__(a), ((__typeof__(a)) { 255 }), a, b, c, ptr)
00447 #define out_rgb24(a, b, c, ptr) vec_mstrgb24(a, b, c, ptr)
00448 #define out_bgr24(a, b, c, ptr) vec_mstbgr24(a, b, c, ptr)
00449 
00450 DEFCSP420_CVT(yuv2_abgr,  out_abgr)
00451 DEFCSP420_CVT(yuv2_bgra,  out_bgra)
00452 DEFCSP420_CVT(yuv2_rgba,  out_rgba)
00453 DEFCSP420_CVT(yuv2_argb,  out_argb)
00454 DEFCSP420_CVT(yuv2_rgb24, out_rgb24)
00455 DEFCSP420_CVT(yuv2_bgr24, out_bgr24)
00456 
00457 // uyvy|uyvy|uyvy|uyvy
00458 // 0123 4567 89ab cdef
00459 static const vector unsigned char
00460     demux_u = { 0x10, 0x00, 0x10, 0x00,
00461                 0x10, 0x04, 0x10, 0x04,
00462                 0x10, 0x08, 0x10, 0x08,
00463                 0x10, 0x0c, 0x10, 0x0c },
00464     demux_v = { 0x10, 0x02, 0x10, 0x02,
00465                 0x10, 0x06, 0x10, 0x06,
00466                 0x10, 0x0A, 0x10, 0x0A,
00467                 0x10, 0x0E, 0x10, 0x0E },
00468     demux_y = { 0x10, 0x01, 0x10, 0x03,
00469                 0x10, 0x05, 0x10, 0x07,
00470                 0x10, 0x09, 0x10, 0x0B,
00471                 0x10, 0x0D, 0x10, 0x0F };
00472 
00473 /*
00474  * this is so I can play live CCIR raw video
00475  */
00476 static int altivec_uyvy_rgb32(SwsContext *c, const unsigned char **in,
00477                               int *instrides, int srcSliceY, int srcSliceH,
00478                               unsigned char **oplanes, int *outstrides)
00479 {
00480     int w = c->srcW;
00481     int h = srcSliceH;
00482     int i, j;
00483     vector unsigned char uyvy;
00484     vector signed short Y, U, V;
00485     vector signed short R0, G0, B0, R1, G1, B1;
00486     vector unsigned char R, G, B;
00487     vector unsigned char *out;
00488     const ubyte *img;
00489 
00490     img = in[0];
00491     out = (vector unsigned char *) (oplanes[0] + srcSliceY * outstrides[0]);
00492 
00493     for (i = 0; i < h; i++)
00494         for (j = 0; j < w / 16; j++) {
00495             uyvy = vec_ld(0, img);
00496 
00497             U = (vector signed short)
00498                     vec_perm(uyvy, (vector unsigned char) { 0 }, demux_u);
00499             V = (vector signed short)
00500                     vec_perm(uyvy, (vector unsigned char) { 0 }, demux_v);
00501             Y = (vector signed short)
00502                     vec_perm(uyvy, (vector unsigned char) { 0 }, demux_y);
00503 
00504             cvtyuvtoRGB(c, Y, U, V, &R0, &G0, &B0);
00505 
00506             uyvy = vec_ld(16, img);
00507 
00508             U = (vector signed short)
00509                     vec_perm(uyvy, (vector unsigned char) { 0 }, demux_u);
00510             V = (vector signed short)
00511                     vec_perm(uyvy, (vector unsigned char) { 0 }, demux_v);
00512             Y = (vector signed short)
00513                     vec_perm(uyvy, (vector unsigned char) { 0 }, demux_y);
00514 
00515             cvtyuvtoRGB(c, Y, U, V, &R1, &G1, &B1);
00516 
00517             R = vec_packclp(R0, R1);
00518             G = vec_packclp(G0, G1);
00519             B = vec_packclp(B0, B1);
00520 
00521             // vec_mstbgr24 (R,G,B, out);
00522             out_rgba(R, G, B, out);
00523 
00524             img += 32;
00525         }
00526     return srcSliceH;
00527 }
00528 
00529 /* Ok currently the acceleration routine only supports
00530  * inputs of widths a multiple of 16
00531  * and heights a multiple 2
00532  *
00533  * So we just fall back to the C codes for this.
00534  */
00535 av_cold SwsFunc ff_yuv2rgb_init_altivec(SwsContext *c)
00536 {
00537     if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC))
00538         return NULL;
00539 
00540     /*
00541      * and this seems not to matter too much I tried a bunch of
00542      * videos with abnormal widths and MPlayer crashes elsewhere.
00543      * mplayer -vo x11 -rawvideo on:w=350:h=240 raw-350x240.eyuv
00544      * boom with X11 bad match.
00545      *
00546      */
00547     if ((c->srcW & 0xf) != 0)
00548         return NULL;
00549 
00550     switch (c->srcFormat) {
00551     case PIX_FMT_YUV410P:
00552     case PIX_FMT_YUV420P:
00553     /*case IMGFMT_CLPL:        ??? */
00554     case PIX_FMT_GRAY8:
00555     case PIX_FMT_NV12:
00556     case PIX_FMT_NV21:
00557         if ((c->srcH & 0x1) != 0)
00558             return NULL;
00559 
00560         switch (c->dstFormat) {
00561         case PIX_FMT_RGB24:
00562             av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space RGB24\n");
00563             return altivec_yuv2_rgb24;
00564         case PIX_FMT_BGR24:
00565             av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space BGR24\n");
00566             return altivec_yuv2_bgr24;
00567         case PIX_FMT_ARGB:
00568             av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space ARGB\n");
00569             return altivec_yuv2_argb;
00570         case PIX_FMT_ABGR:
00571             av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space ABGR\n");
00572             return altivec_yuv2_abgr;
00573         case PIX_FMT_RGBA:
00574             av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space RGBA\n");
00575             return altivec_yuv2_rgba;
00576         case PIX_FMT_BGRA:
00577             av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space BGRA\n");
00578             return altivec_yuv2_bgra;
00579         default: return NULL;
00580         }
00581         break;
00582 
00583     case PIX_FMT_UYVY422:
00584         switch (c->dstFormat) {
00585         case PIX_FMT_BGR32:
00586             av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space UYVY -> RGB32\n");
00587             return altivec_uyvy_rgb32;
00588         default: return NULL;
00589         }
00590         break;
00591     }
00592     return NULL;
00593 }
00594 
00595 av_cold void ff_yuv2rgb_init_tables_altivec(SwsContext *c,
00596                                             const int inv_table[4],
00597                                             int brightness,
00598                                             int contrast,
00599                                             int saturation)
00600 {
00601     union {
00602         DECLARE_ALIGNED(16, signed short, tmp)[8];
00603         vector signed short vec;
00604     } buf;
00605 
00606     buf.tmp[0] = ((0xffffLL) * contrast >> 8) >> 9;                               // cy
00607     buf.tmp[1] = -256 * brightness;                                               // oy
00608     buf.tmp[2] =   (inv_table[0] >> 3) * (contrast >> 16) * (saturation >> 16);   // crv
00609     buf.tmp[3] =   (inv_table[1] >> 3) * (contrast >> 16) * (saturation >> 16);   // cbu
00610     buf.tmp[4] = -((inv_table[2] >> 1) * (contrast >> 16) * (saturation >> 16));  // cgu
00611     buf.tmp[5] = -((inv_table[3] >> 1) * (contrast >> 16) * (saturation >> 16));  // cgv
00612 
00613     c->CSHIFT = (vector unsigned short) vec_splat_u16(2);
00614     c->CY     = vec_splat((vector signed short) buf.vec, 0);
00615     c->OY     = vec_splat((vector signed short) buf.vec, 1);
00616     c->CRV    = vec_splat((vector signed short) buf.vec, 2);
00617     c->CBU    = vec_splat((vector signed short) buf.vec, 3);
00618     c->CGU    = vec_splat((vector signed short) buf.vec, 4);
00619     c->CGV    = vec_splat((vector signed short) buf.vec, 5);
00620     return;
00621 }
00622 
00623 static av_always_inline void ff_yuv2packedX_altivec(SwsContext *c,
00624                                                     const int16_t *lumFilter,
00625                                                     const int16_t **lumSrc,
00626                                                     int lumFilterSize,
00627                                                     const int16_t *chrFilter,
00628                                                     const int16_t **chrUSrc,
00629                                                     const int16_t **chrVSrc,
00630                                                     int chrFilterSize,
00631                                                     const int16_t **alpSrc,
00632                                                     uint8_t *dest,
00633                                                     int dstW, int dstY,
00634                                                     enum PixelFormat target)
00635 {
00636     int i, j;
00637     vector signed short X, X0, X1, Y0, U0, V0, Y1, U1, V1, U, V;
00638     vector signed short R0, G0, B0, R1, G1, B1;
00639 
00640     vector unsigned char R, G, B;
00641     vector unsigned char *out, *nout;
00642 
00643     vector signed short RND   = vec_splat_s16(1 << 3);
00644     vector unsigned short SCL = vec_splat_u16(4);
00645     DECLARE_ALIGNED(16, unsigned int, scratch)[16];
00646 
00647     vector signed short *YCoeffs, *CCoeffs;
00648 
00649     YCoeffs = c->vYCoeffsBank + dstY * lumFilterSize;
00650     CCoeffs = c->vCCoeffsBank + dstY * chrFilterSize;
00651 
00652     out = (vector unsigned char *) dest;
00653 
00654     for (i = 0; i < dstW; i += 16) {
00655         Y0 = RND;
00656         Y1 = RND;
00657         /* extract 16 coeffs from lumSrc */
00658         for (j = 0; j < lumFilterSize; j++) {
00659             X0 = vec_ld(0, &lumSrc[j][i]);
00660             X1 = vec_ld(16, &lumSrc[j][i]);
00661             Y0 = vec_mradds(X0, YCoeffs[j], Y0);
00662             Y1 = vec_mradds(X1, YCoeffs[j], Y1);
00663         }
00664 
00665         U = RND;
00666         V = RND;
00667         /* extract 8 coeffs from U,V */
00668         for (j = 0; j < chrFilterSize; j++) {
00669             X = vec_ld(0, &chrUSrc[j][i / 2]);
00670             U = vec_mradds(X, CCoeffs[j], U);
00671             X = vec_ld(0, &chrVSrc[j][i / 2]);
00672             V = vec_mradds(X, CCoeffs[j], V);
00673         }
00674 
00675         /* scale and clip signals */
00676         Y0 = vec_sra(Y0, SCL);
00677         Y1 = vec_sra(Y1, SCL);
00678         U  = vec_sra(U, SCL);
00679         V  = vec_sra(V, SCL);
00680 
00681         Y0 = vec_clip_s16(Y0);
00682         Y1 = vec_clip_s16(Y1);
00683         U  = vec_clip_s16(U);
00684         V  = vec_clip_s16(V);
00685 
00686         /* now we have
00687          * Y0 = y0 y1 y2 y3 y4 y5 y6 y7    Y1 = y8 y9 y10 y11 y12 y13 y14 y15
00688          * U  = u0 u1 u2 u3 u4 u5 u6 u7    V  = v0 v1 v2 v3 v4 v5 v6 v7
00689          *
00690          * Y0 = y0 y1 y2 y3 y4 y5 y6 y7    Y1 = y8 y9 y10 y11 y12 y13 y14 y15
00691          * U0 = u0 u0 u1 u1 u2 u2 u3 u3    U1 = u4 u4 u5 u5 u6 u6 u7 u7
00692          * V0 = v0 v0 v1 v1 v2 v2 v3 v3    V1 = v4 v4 v5 v5 v6 v6 v7 v7
00693          */
00694 
00695         U0 = vec_mergeh(U, U);
00696         V0 = vec_mergeh(V, V);
00697 
00698         U1 = vec_mergel(U, U);
00699         V1 = vec_mergel(V, V);
00700 
00701         cvtyuvtoRGB(c, Y0, U0, V0, &R0, &G0, &B0);
00702         cvtyuvtoRGB(c, Y1, U1, V1, &R1, &G1, &B1);
00703 
00704         R = vec_packclp(R0, R1);
00705         G = vec_packclp(G0, G1);
00706         B = vec_packclp(B0, B1);
00707 
00708         switch (target) {
00709         case PIX_FMT_ABGR:
00710             out_abgr(R, G, B, out);
00711             break;
00712         case PIX_FMT_BGRA:
00713             out_bgra(R, G, B, out);
00714             break;
00715         case PIX_FMT_RGBA:
00716             out_rgba(R, G, B, out);
00717             break;
00718         case PIX_FMT_ARGB:
00719             out_argb(R, G, B, out);
00720             break;
00721         case PIX_FMT_RGB24:
00722             out_rgb24(R, G, B, out);
00723             break;
00724         case PIX_FMT_BGR24:
00725             out_bgr24(R, G, B, out);
00726             break;
00727         default:
00728         {
00729             /* If this is reached, the caller should have called yuv2packedXinC
00730              * instead. */
00731             static int printed_error_message;
00732             if (!printed_error_message) {
00733                 av_log(c, AV_LOG_ERROR,
00734                        "altivec_yuv2packedX doesn't support %s output\n",
00735                        av_get_pix_fmt_name(c->dstFormat));
00736                 printed_error_message = 1;
00737             }
00738             return;
00739         }
00740         }
00741     }
00742 
00743     if (i < dstW) {
00744         i -= 16;
00745 
00746         Y0 = RND;
00747         Y1 = RND;
00748         /* extract 16 coeffs from lumSrc */
00749         for (j = 0; j < lumFilterSize; j++) {
00750             X0 = vec_ld(0, &lumSrc[j][i]);
00751             X1 = vec_ld(16, &lumSrc[j][i]);
00752             Y0 = vec_mradds(X0, YCoeffs[j], Y0);
00753             Y1 = vec_mradds(X1, YCoeffs[j], Y1);
00754         }
00755 
00756         U = RND;
00757         V = RND;
00758         /* extract 8 coeffs from U,V */
00759         for (j = 0; j < chrFilterSize; j++) {
00760             X = vec_ld(0, &chrUSrc[j][i / 2]);
00761             U = vec_mradds(X, CCoeffs[j], U);
00762             X = vec_ld(0, &chrVSrc[j][i / 2]);
00763             V = vec_mradds(X, CCoeffs[j], V);
00764         }
00765 
00766         /* scale and clip signals */
00767         Y0 = vec_sra(Y0, SCL);
00768         Y1 = vec_sra(Y1, SCL);
00769         U  = vec_sra(U, SCL);
00770         V  = vec_sra(V, SCL);
00771 
00772         Y0 = vec_clip_s16(Y0);
00773         Y1 = vec_clip_s16(Y1);
00774         U  = vec_clip_s16(U);
00775         V  = vec_clip_s16(V);
00776 
00777         /* now we have
00778          * Y0 = y0 y1 y2 y3 y4 y5 y6 y7    Y1 = y8 y9 y10 y11 y12 y13 y14 y15
00779          * U  = u0 u1 u2 u3 u4 u5 u6 u7    V  = v0 v1 v2 v3 v4 v5 v6 v7
00780          *
00781          * Y0 = y0 y1 y2 y3 y4 y5 y6 y7    Y1 = y8 y9 y10 y11 y12 y13 y14 y15
00782          * U0 = u0 u0 u1 u1 u2 u2 u3 u3    U1 = u4 u4 u5 u5 u6 u6 u7 u7
00783          * V0 = v0 v0 v1 v1 v2 v2 v3 v3    V1 = v4 v4 v5 v5 v6 v6 v7 v7
00784          */
00785 
00786         U0 = vec_mergeh(U, U);
00787         V0 = vec_mergeh(V, V);
00788 
00789         U1 = vec_mergel(U, U);
00790         V1 = vec_mergel(V, V);
00791 
00792         cvtyuvtoRGB(c, Y0, U0, V0, &R0, &G0, &B0);
00793         cvtyuvtoRGB(c, Y1, U1, V1, &R1, &G1, &B1);
00794 
00795         R = vec_packclp(R0, R1);
00796         G = vec_packclp(G0, G1);
00797         B = vec_packclp(B0, B1);
00798 
00799         nout = (vector unsigned char *) scratch;
00800         switch (target) {
00801         case PIX_FMT_ABGR:
00802             out_abgr(R, G, B, nout);
00803             break;
00804         case PIX_FMT_BGRA:
00805             out_bgra(R, G, B, nout);
00806             break;
00807         case PIX_FMT_RGBA:
00808             out_rgba(R, G, B, nout);
00809             break;
00810         case PIX_FMT_ARGB:
00811             out_argb(R, G, B, nout);
00812             break;
00813         case PIX_FMT_RGB24:
00814             out_rgb24(R, G, B, nout);
00815             break;
00816         case PIX_FMT_BGR24:
00817             out_bgr24(R, G, B, nout);
00818             break;
00819         default:
00820             /* Unreachable, I think. */
00821             av_log(c, AV_LOG_ERROR,
00822                    "altivec_yuv2packedX doesn't support %s output\n",
00823                    av_get_pix_fmt_name(c->dstFormat));
00824             return;
00825         }
00826 
00827         memcpy(&((uint32_t *) dest)[i], scratch, (dstW - i) / 4);
00828     }
00829 }
00830 
00831 #define YUV2PACKEDX_WRAPPER(suffix, pixfmt)                             \
00832 void ff_yuv2 ## suffix ## _X_altivec(SwsContext *c,                     \
00833                                      const int16_t *lumFilter,          \
00834                                      const int16_t **lumSrc,            \
00835                                      int lumFilterSize,                 \
00836                                      const int16_t *chrFilter,          \
00837                                      const int16_t **chrUSrc,           \
00838                                      const int16_t **chrVSrc,           \
00839                                      int chrFilterSize,                 \
00840                                      const int16_t **alpSrc,            \
00841                                      uint8_t *dest, int dstW, int dstY) \
00842 {                                                                       \
00843     ff_yuv2packedX_altivec(c, lumFilter, lumSrc, lumFilterSize,         \
00844                            chrFilter, chrUSrc, chrVSrc,                 \
00845                            chrFilterSize, alpSrc,                       \
00846                            dest, dstW, dstY, pixfmt);                   \
00847 }
00848 
00849 YUV2PACKEDX_WRAPPER(abgr,  PIX_FMT_ABGR);
00850 YUV2PACKEDX_WRAPPER(bgra,  PIX_FMT_BGRA);
00851 YUV2PACKEDX_WRAPPER(argb,  PIX_FMT_ARGB);
00852 YUV2PACKEDX_WRAPPER(rgba,  PIX_FMT_RGBA);
00853 YUV2PACKEDX_WRAPPER(rgb24, PIX_FMT_RGB24);
00854 YUV2PACKEDX_WRAPPER(bgr24, PIX_FMT_BGR24);