MythTV: h264_altivec.c Source File

00001 /*
00002  * Copyright (c) 2004 Romain Dolbeau <romain@dolbeau.org>
00003  *
00004  * This file is part of FFmpeg.
00005  *
00006  * FFmpeg is free software; you can redistribute it and/or
00007  * modify it under the terms of the GNU Lesser General Public
00008  * License as published by the Free Software Foundation; either
00009  * version 2.1 of the License, or (at your option) any later version.
00010  *
00011  * FFmpeg is distributed in the hope that it will be useful,
00012  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00013  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00014  * Lesser General Public License for more details.
00015  *
00016  * You should have received a copy of the GNU Lesser General Public
00017  * License along with FFmpeg; if not, write to the Free Software
00018  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00019  */
00020 
00021 #include "dsputil.h"
00022 
00023 #include "gcc_fixes.h"
00024 
00025 #include "dsputil_ppc.h"
00026 #include "util_altivec.h"
00027 #include "types_altivec.h"
00028 
00029 #define PUT_OP_U8_ALTIVEC(d, s, dst) d = s
00030 #define AVG_OP_U8_ALTIVEC(d, s, dst) d = vec_avg(dst, s)
00031 
00032 #define OP_U8_ALTIVEC                          PUT_OP_U8_ALTIVEC
00033 #define PREFIX_h264_chroma_mc8_altivec         put_h264_chroma_mc8_altivec
00034 #define PREFIX_h264_chroma_mc8_num             altivec_put_h264_chroma_mc8_num
00035 #define PREFIX_h264_qpel16_h_lowpass_altivec   put_h264_qpel16_h_lowpass_altivec
00036 #define PREFIX_h264_qpel16_h_lowpass_num       altivec_put_h264_qpel16_h_lowpass_num
00037 #define PREFIX_h264_qpel16_v_lowpass_altivec   put_h264_qpel16_v_lowpass_altivec
00038 #define PREFIX_h264_qpel16_v_lowpass_num       altivec_put_h264_qpel16_v_lowpass_num
00039 #define PREFIX_h264_qpel16_hv_lowpass_altivec  put_h264_qpel16_hv_lowpass_altivec
00040 #define PREFIX_h264_qpel16_hv_lowpass_num      altivec_put_h264_qpel16_hv_lowpass_num
00041 #include "h264_template_altivec.c"
00042 #undef OP_U8_ALTIVEC
00043 #undef PREFIX_h264_chroma_mc8_altivec
00044 #undef PREFIX_h264_chroma_mc8_num
00045 #undef PREFIX_h264_qpel16_h_lowpass_altivec
00046 #undef PREFIX_h264_qpel16_h_lowpass_num
00047 #undef PREFIX_h264_qpel16_v_lowpass_altivec
00048 #undef PREFIX_h264_qpel16_v_lowpass_num
00049 #undef PREFIX_h264_qpel16_hv_lowpass_altivec
00050 #undef PREFIX_h264_qpel16_hv_lowpass_num
00051 
00052 #define OP_U8_ALTIVEC                          AVG_OP_U8_ALTIVEC
00053 #define PREFIX_h264_chroma_mc8_altivec         avg_h264_chroma_mc8_altivec
00054 #define PREFIX_h264_chroma_mc8_num             altivec_avg_h264_chroma_mc8_num
00055 #define PREFIX_h264_qpel16_h_lowpass_altivec   avg_h264_qpel16_h_lowpass_altivec
00056 #define PREFIX_h264_qpel16_h_lowpass_num       altivec_avg_h264_qpel16_h_lowpass_num
00057 #define PREFIX_h264_qpel16_v_lowpass_altivec   avg_h264_qpel16_v_lowpass_altivec
00058 #define PREFIX_h264_qpel16_v_lowpass_num       altivec_avg_h264_qpel16_v_lowpass_num
00059 #define PREFIX_h264_qpel16_hv_lowpass_altivec  avg_h264_qpel16_hv_lowpass_altivec
00060 #define PREFIX_h264_qpel16_hv_lowpass_num      altivec_avg_h264_qpel16_hv_lowpass_num
00061 #include "h264_template_altivec.c"
00062 #undef OP_U8_ALTIVEC
00063 #undef PREFIX_h264_chroma_mc8_altivec
00064 #undef PREFIX_h264_chroma_mc8_num
00065 #undef PREFIX_h264_qpel16_h_lowpass_altivec
00066 #undef PREFIX_h264_qpel16_h_lowpass_num
00067 #undef PREFIX_h264_qpel16_v_lowpass_altivec
00068 #undef PREFIX_h264_qpel16_v_lowpass_num
00069 #undef PREFIX_h264_qpel16_hv_lowpass_altivec
00070 #undef PREFIX_h264_qpel16_hv_lowpass_num
00071 
00072 #define H264_MC(OPNAME, SIZE, CODETYPE) \
00073 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## CODETYPE (uint8_t *dst, uint8_t *src, int stride){\
00074     OPNAME ## pixels ## SIZE ## _ ## CODETYPE(dst, src, stride, SIZE);\
00075 }\
00076 \
00077 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){ \
00078     DECLARE_ALIGNED_16(uint8_t, half[SIZE*SIZE]);\
00079     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
00080     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\
00081 }\
00082 \
00083 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00084     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(dst, src, stride, stride);\
00085 }\
00086 \
00087 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00088     DECLARE_ALIGNED_16(uint8_t, half[SIZE*SIZE]);\
00089     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
00090     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+1, half, stride, stride, SIZE);\
00091 }\
00092 \
00093 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00094     DECLARE_ALIGNED_16(uint8_t, half[SIZE*SIZE]);\
00095     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
00096     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\
00097 }\
00098 \
00099 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00100     OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(dst, src, stride, stride);\
00101 }\
00102 \
00103 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00104     DECLARE_ALIGNED_16(uint8_t, half[SIZE*SIZE]);\
00105     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
00106     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+stride, half, stride, stride, SIZE);\
00107 }\
00108 \
00109 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00110     DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\
00111     DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\
00112     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
00113     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
00114     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
00115 }\
00116 \
00117 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00118     DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\
00119     DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\
00120     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
00121     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
00122     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
00123 }\
00124 \
00125 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00126     DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\
00127     DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\
00128     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
00129     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
00130     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
00131 }\
00132 \
00133 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00134     DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\
00135     DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\
00136     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
00137     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
00138     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
00139 }\
00140 \
00141 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00142     DECLARE_ALIGNED_16(int16_t, tmp[SIZE*(SIZE+8)]);\
00143     OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(dst, tmp, src, stride, SIZE, stride);\
00144 }\
00145 \
00146 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00147     DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\
00148     DECLARE_ALIGNED_16(uint8_t, halfHV[SIZE*SIZE]);\
00149     DECLARE_ALIGNED_16(int16_t, tmp[SIZE*(SIZE+8)]);\
00150     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
00151     put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
00152     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\
00153 }\
00154 \
00155 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00156     DECLARE_ALIGNED_16(uint8_t, halfH[SIZE*SIZE]);\
00157     DECLARE_ALIGNED_16(uint8_t, halfHV[SIZE*SIZE]);\
00158     DECLARE_ALIGNED_16(int16_t, tmp[SIZE*(SIZE+8)]);\
00159     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
00160     put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
00161     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\
00162 }\
00163 \
00164 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00165     DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\
00166     DECLARE_ALIGNED_16(uint8_t, halfHV[SIZE*SIZE]);\
00167     DECLARE_ALIGNED_16(int16_t, tmp[SIZE*(SIZE+8)]);\
00168     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
00169     put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
00170     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
00171 }\
00172 \
00173 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00174     DECLARE_ALIGNED_16(uint8_t, halfV[SIZE*SIZE]);\
00175     DECLARE_ALIGNED_16(uint8_t, halfHV[SIZE*SIZE]);\
00176     DECLARE_ALIGNED_16(int16_t, tmp[SIZE*(SIZE+8)]);\
00177     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
00178     put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
00179     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
00180 }\
00181 
00182 /* this code assume that stride % 16 == 0 */
00183 void put_no_rnd_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, int stride, int h, int x, int y) {
00184    DECLARE_ALIGNED_16(signed int, ABCD[4]) =
00185                         {((8 - x) * (8 - y)),
00186                              ((x) * (8 - y)),
00187                          ((8 - x) * (y)),
00188                              ((x) * (y))};
00189     register int i;
00190     vec_u8_t fperm;
00191     const vec_s32_t vABCD = vec_ld(0, ABCD);
00192     const vec_s16_t vA = vec_splat((vec_s16_t)vABCD, 1);
00193     const vec_s16_t vB = vec_splat((vec_s16_t)vABCD, 3);
00194     const vec_s16_t vC = vec_splat((vec_s16_t)vABCD, 5);
00195     const vec_s16_t vD = vec_splat((vec_s16_t)vABCD, 7);
00196     LOAD_ZERO;
00197     const vec_s16_t v28ss = vec_sub(vec_sl(vec_splat_s16(1),vec_splat_u16(5)),vec_splat_s16(4));
00198     const vec_u16_t v6us = vec_splat_u16(6);
00199     register int loadSecond     = (((unsigned long)src) % 16) <= 7 ? 0 : 1;
00200     register int reallyBadAlign = (((unsigned long)src) % 16) == 15 ? 1 : 0;
00201 
00202     vec_u8_t vsrcAuc, vsrcBuc, vsrcperm0, vsrcperm1;
00203     vec_u8_t vsrc0uc, vsrc1uc;
00204     vec_s16_t vsrc0ssH, vsrc1ssH;
00205     vec_u8_t vsrcCuc, vsrc2uc, vsrc3uc;
00206     vec_s16_t vsrc2ssH, vsrc3ssH, psum;
00207     vec_u8_t vdst, ppsum, fsum;
00208 
00209     if (((unsigned long)dst) % 16 == 0) {
00210         fperm = (vec_u8_t)AVV(0x10, 0x11, 0x12, 0x13,
00211                               0x14, 0x15, 0x16, 0x17,
00212                               0x08, 0x09, 0x0A, 0x0B,
00213                               0x0C, 0x0D, 0x0E, 0x0F);
00214     } else {
00215         fperm = (vec_u8_t)AVV(0x00, 0x01, 0x02, 0x03,
00216                               0x04, 0x05, 0x06, 0x07,
00217                               0x18, 0x19, 0x1A, 0x1B,
00218                               0x1C, 0x1D, 0x1E, 0x1F);
00219     }
00220 
00221     vsrcAuc = vec_ld(0, src);
00222 
00223     if (loadSecond)
00224         vsrcBuc = vec_ld(16, src);
00225     vsrcperm0 = vec_lvsl(0, src);
00226     vsrcperm1 = vec_lvsl(1, src);
00227 
00228     vsrc0uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm0);
00229     if (reallyBadAlign)
00230         vsrc1uc = vsrcBuc;
00231     else
00232         vsrc1uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm1);
00233 
00234     vsrc0ssH = (vec_s16_t)vec_mergeh(zero_u8v, (vec_u8_t)vsrc0uc);
00235     vsrc1ssH = (vec_s16_t)vec_mergeh(zero_u8v, (vec_u8_t)vsrc1uc);
00236 
00237     if (!loadSecond) {// -> !reallyBadAlign
00238         for (i = 0 ; i < h ; i++) {
00239 
00240 
00241             vsrcCuc = vec_ld(stride + 0, src);
00242 
00243             vsrc2uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm0);
00244             vsrc3uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm1);
00245 
00246             vsrc2ssH = (vec_s16_t)vec_mergeh(zero_u8v, (vec_u8_t)vsrc2uc);
00247             vsrc3ssH = (vec_s16_t)vec_mergeh(zero_u8v, (vec_u8_t)vsrc3uc);
00248 
00249             psum = vec_mladd(vA, vsrc0ssH, vec_splat_s16(0));
00250             psum = vec_mladd(vB, vsrc1ssH, psum);
00251             psum = vec_mladd(vC, vsrc2ssH, psum);
00252             psum = vec_mladd(vD, vsrc3ssH, psum);
00253             psum = vec_add(v28ss, psum);
00254             psum = vec_sra(psum, v6us);
00255 
00256             vdst = vec_ld(0, dst);
00257             ppsum = (vec_u8_t)vec_packsu(psum, psum);
00258             fsum = vec_perm(vdst, ppsum, fperm);
00259 
00260             vec_st(fsum, 0, dst);
00261 
00262             vsrc0ssH = vsrc2ssH;
00263             vsrc1ssH = vsrc3ssH;
00264 
00265             dst += stride;
00266             src += stride;
00267         }
00268     } else {
00269         vec_u8_t vsrcDuc;
00270         for (i = 0 ; i < h ; i++) {
00271             vsrcCuc = vec_ld(stride + 0, src);
00272             vsrcDuc = vec_ld(stride + 16, src);
00273 
00274             vsrc2uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm0);
00275             if (reallyBadAlign)
00276                 vsrc3uc = vsrcDuc;
00277             else
00278                 vsrc3uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm1);
00279 
00280             vsrc2ssH = (vec_s16_t)vec_mergeh(zero_u8v, (vec_u8_t)vsrc2uc);
00281             vsrc3ssH = (vec_s16_t)vec_mergeh(zero_u8v, (vec_u8_t)vsrc3uc);
00282 
00283             psum = vec_mladd(vA, vsrc0ssH, vec_splat_s16(0));
00284             psum = vec_mladd(vB, vsrc1ssH, psum);
00285             psum = vec_mladd(vC, vsrc2ssH, psum);
00286             psum = vec_mladd(vD, vsrc3ssH, psum);
00287             psum = vec_add(v28ss, psum);
00288             psum = vec_sr(psum, v6us);
00289 
00290             vdst = vec_ld(0, dst);
00291             ppsum = (vec_u8_t)vec_pack(psum, psum);
00292             fsum = vec_perm(vdst, ppsum, fperm);
00293 
00294             vec_st(fsum, 0, dst);
00295 
00296             vsrc0ssH = vsrc2ssH;
00297             vsrc1ssH = vsrc3ssH;
00298 
00299             dst += stride;
00300             src += stride;
00301         }
00302     }
00303 }
00304 
00305 static inline void put_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
00306                                     const uint8_t * src2, int dst_stride,
00307                                     int src_stride1, int h)
00308 {
00309     int i;
00310     vec_u8_t a, b, d, tmp1, tmp2, mask, mask_, edges, align;
00311 
00312     mask_ = vec_lvsl(0, src2);
00313 
00314     for (i = 0; i < h; i++) {
00315 
00316         tmp1 = vec_ld(i * src_stride1, src1);
00317         mask = vec_lvsl(i * src_stride1, src1);
00318         tmp2 = vec_ld(i * src_stride1 + 15, src1);
00319 
00320         a = vec_perm(tmp1, tmp2, mask);
00321 
00322         tmp1 = vec_ld(i * 16, src2);
00323         tmp2 = vec_ld(i * 16 + 15, src2);
00324 
00325         b = vec_perm(tmp1, tmp2, mask_);
00326 
00327         tmp1 = vec_ld(0, dst);
00328         mask = vec_lvsl(0, dst);
00329         tmp2 = vec_ld(15, dst);
00330 
00331         d = vec_avg(a, b);
00332 
00333         edges = vec_perm(tmp2, tmp1, mask);
00334 
00335         align = vec_lvsr(0, dst);
00336 
00337         tmp2 = vec_perm(d, edges, align);
00338         tmp1 = vec_perm(edges, d, align);
00339 
00340         vec_st(tmp2, 15, dst);
00341         vec_st(tmp1, 0 , dst);
00342 
00343         dst += dst_stride;
00344     }
00345 }
00346 
00347 static inline void avg_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
00348                                     const uint8_t * src2, int dst_stride,
00349                                     int src_stride1, int h)
00350 {
00351     int i;
00352     vec_u8_t a, b, d, tmp1, tmp2, mask, mask_, edges, align;
00353 
00354     mask_ = vec_lvsl(0, src2);
00355 
00356     for (i = 0; i < h; i++) {
00357 
00358         tmp1 = vec_ld(i * src_stride1, src1);
00359         mask = vec_lvsl(i * src_stride1, src1);
00360         tmp2 = vec_ld(i * src_stride1 + 15, src1);
00361 
00362         a = vec_perm(tmp1, tmp2, mask);
00363 
00364         tmp1 = vec_ld(i * 16, src2);
00365         tmp2 = vec_ld(i * 16 + 15, src2);
00366 
00367         b = vec_perm(tmp1, tmp2, mask_);
00368 
00369         tmp1 = vec_ld(0, dst);
00370         mask = vec_lvsl(0, dst);
00371         tmp2 = vec_ld(15, dst);
00372 
00373         d = vec_avg(vec_perm(tmp1, tmp2, mask), vec_avg(a, b));
00374 
00375         edges = vec_perm(tmp2, tmp1, mask);
00376 
00377         align = vec_lvsr(0, dst);
00378 
00379         tmp2 = vec_perm(d, edges, align);
00380         tmp1 = vec_perm(edges, d, align);
00381 
00382         vec_st(tmp2, 15, dst);
00383         vec_st(tmp1, 0 , dst);
00384 
00385         dst += dst_stride;
00386     }
00387 }
00388 
00389 /* Implemented but could be faster
00390 #define put_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) put_pixels16_l2(d,s1,s2,ds,s1s,16,h)
00391 #define avg_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) avg_pixels16_l2(d,s1,s2,ds,s1s,16,h)
00392  */
00393 
00394   H264_MC(put_, 16, altivec)
00395   H264_MC(avg_, 16, altivec)
00396 
00397 
00398 /****************************************************************************
00399  * IDCT transform:
00400  ****************************************************************************/
00401 
00402 #define VEC_1D_DCT(vb0,vb1,vb2,vb3,va0,va1,va2,va3)               \
00403     /* 1st stage */                                               \
00404     vz0 = vec_add(vb0,vb2);       /* temp[0] = Y[0] + Y[2] */     \
00405     vz1 = vec_sub(vb0,vb2);       /* temp[1] = Y[0] - Y[2] */     \
00406     vz2 = vec_sra(vb1,vec_splat_u16(1));                          \
00407     vz2 = vec_sub(vz2,vb3);       /* temp[2] = Y[1].1/2 - Y[3] */ \
00408     vz3 = vec_sra(vb3,vec_splat_u16(1));                          \
00409     vz3 = vec_add(vb1,vz3);       /* temp[3] = Y[1] + Y[3].1/2 */ \
00410     /* 2nd stage: output */                                       \
00411     va0 = vec_add(vz0,vz3);       /* x[0] = temp[0] + temp[3] */  \
00412     va1 = vec_add(vz1,vz2);       /* x[1] = temp[1] + temp[2] */  \
00413     va2 = vec_sub(vz1,vz2);       /* x[2] = temp[1] - temp[2] */  \
00414     va3 = vec_sub(vz0,vz3)        /* x[3] = temp[0] - temp[3] */
00415 
00416 #define VEC_TRANSPOSE_4(a0,a1,a2,a3,b0,b1,b2,b3) \
00417     b0 = vec_mergeh( a0, a0 ); \
00418     b1 = vec_mergeh( a1, a0 ); \
00419     b2 = vec_mergeh( a2, a0 ); \
00420     b3 = vec_mergeh( a3, a0 ); \
00421     a0 = vec_mergeh( b0, b2 ); \
00422     a1 = vec_mergel( b0, b2 ); \
00423     a2 = vec_mergeh( b1, b3 ); \
00424     a3 = vec_mergel( b1, b3 ); \
00425     b0 = vec_mergeh( a0, a2 ); \
00426     b1 = vec_mergel( a0, a2 ); \
00427     b2 = vec_mergeh( a1, a3 ); \
00428     b3 = vec_mergel( a1, a3 )
00429 
00430 #define VEC_LOAD_U8_ADD_S16_STORE_U8(va)                      \
00431     vdst_orig = vec_ld(0, dst);                               \
00432     vdst = vec_perm(vdst_orig, zero_u8v, vdst_mask);          \
00433     vdst_ss = (vec_s16_t) vec_mergeh(zero_u8v, vdst);         \
00434     va = vec_add(va, vdst_ss);                                \
00435     va_u8 = vec_packsu(va, zero_s16v);                        \
00436     va_u32 = vec_splat((vec_u32_t)va_u8, 0);                  \
00437     vec_ste(va_u32, element, (uint32_t*)dst);
00438 
00439 static void ff_h264_idct_add_altivec(uint8_t *dst, DCTELEM *block, int stride)
00440 {
00441     vec_s16_t va0, va1, va2, va3;
00442     vec_s16_t vz0, vz1, vz2, vz3;
00443     vec_s16_t vtmp0, vtmp1, vtmp2, vtmp3;
00444     vec_u8_t va_u8;
00445     vec_u32_t va_u32;
00446     vec_s16_t vdst_ss;
00447     const vec_u16_t v6us = vec_splat_u16(6);
00448     vec_u8_t vdst, vdst_orig;
00449     vec_u8_t vdst_mask = vec_lvsl(0, dst);
00450     int element = ((unsigned long)dst & 0xf) >> 2;
00451     LOAD_ZERO;
00452 
00453     block[0] += 32;  /* add 32 as a DC-level for rounding */
00454 
00455     vtmp0 = vec_ld(0,block);
00456     vtmp1 = vec_sld(vtmp0, vtmp0, 8);
00457     vtmp2 = vec_ld(16,block);
00458     vtmp3 = vec_sld(vtmp2, vtmp2, 8);
00459 
00460     VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3);
00461     VEC_TRANSPOSE_4(va0,va1,va2,va3,vtmp0,vtmp1,vtmp2,vtmp3);
00462     VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3);
00463 
00464     va0 = vec_sra(va0,v6us);
00465     va1 = vec_sra(va1,v6us);
00466     va2 = vec_sra(va2,v6us);
00467     va3 = vec_sra(va3,v6us);
00468 
00469     VEC_LOAD_U8_ADD_S16_STORE_U8(va0);
00470     dst += stride;
00471     VEC_LOAD_U8_ADD_S16_STORE_U8(va1);
00472     dst += stride;
00473     VEC_LOAD_U8_ADD_S16_STORE_U8(va2);
00474     dst += stride;
00475     VEC_LOAD_U8_ADD_S16_STORE_U8(va3);
00476 }
00477 
00478 #define IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7,  d0, d1, d2, d3, d4, d5, d6, d7) {\
00479     /*        a0  = SRC(0) + SRC(4); */ \
00480     vec_s16_t a0v = vec_add(s0, s4);    \
00481     /*        a2  = SRC(0) - SRC(4); */ \
00482     vec_s16_t a2v = vec_sub(s0, s4);    \
00483     /*        a4  =           (SRC(2)>>1) - SRC(6); */ \
00484     vec_s16_t a4v = vec_sub(vec_sra(s2, onev), s6);    \
00485     /*        a6  =           (SRC(6)>>1) + SRC(2); */ \
00486     vec_s16_t a6v = vec_add(vec_sra(s6, onev), s2);    \
00487     /*        b0  =         a0 + a6; */ \
00488     vec_s16_t b0v = vec_add(a0v, a6v);  \
00489     /*        b2  =         a2 + a4; */ \
00490     vec_s16_t b2v = vec_add(a2v, a4v);  \
00491     /*        b4  =         a2 - a4; */ \
00492     vec_s16_t b4v = vec_sub(a2v, a4v);  \
00493     /*        b6  =         a0 - a6; */ \
00494     vec_s16_t b6v = vec_sub(a0v, a6v);  \
00495     /* a1 =  SRC(5) - SRC(3) - SRC(7) - (SRC(7)>>1); */ \
00496     /*        a1 =             (SRC(5)-SRC(3)) -  (SRC(7)  +  (SRC(7)>>1)); */ \
00497     vec_s16_t a1v = vec_sub( vec_sub(s5, s3), vec_add(s7, vec_sra(s7, onev)) ); \
00498     /* a3 =  SRC(7) + SRC(1) - SRC(3) - (SRC(3)>>1); */ \
00499     /*        a3 =             (SRC(7)+SRC(1)) -  (SRC(3)  +  (SRC(3)>>1)); */ \
00500     vec_s16_t a3v = vec_sub( vec_add(s7, s1), vec_add(s3, vec_sra(s3, onev)) );\
00501     /* a5 =  SRC(7) - SRC(1) + SRC(5) + (SRC(5)>>1); */ \
00502     /*        a5 =             (SRC(7)-SRC(1)) +   SRC(5) +   (SRC(5)>>1); */ \
00503     vec_s16_t a5v = vec_add( vec_sub(s7, s1), vec_add(s5, vec_sra(s5, onev)) );\
00504     /*        a7 =                SRC(5)+SRC(3) +  SRC(1) +   (SRC(1)>>1); */ \
00505     vec_s16_t a7v = vec_add( vec_add(s5, s3), vec_add(s1, vec_sra(s1, onev)) );\
00506     /*        b1 =                  (a7>>2)  +  a1; */ \
00507     vec_s16_t b1v = vec_add( vec_sra(a7v, twov), a1v); \
00508     /*        b3 =          a3 +        (a5>>2); */ \
00509     vec_s16_t b3v = vec_add(a3v, vec_sra(a5v, twov)); \
00510     /*        b5 =                  (a3>>2)  -   a5; */ \
00511     vec_s16_t b5v = vec_sub( vec_sra(a3v, twov), a5v); \
00512     /*        b7 =           a7 -        (a1>>2); */ \
00513     vec_s16_t b7v = vec_sub( a7v, vec_sra(a1v, twov)); \
00514     /* DST(0,    b0 + b7); */ \
00515     d0 = vec_add(b0v, b7v); \
00516     /* DST(1,    b2 + b5); */ \
00517     d1 = vec_add(b2v, b5v); \
00518     /* DST(2,    b4 + b3); */ \
00519     d2 = vec_add(b4v, b3v); \
00520     /* DST(3,    b6 + b1); */ \
00521     d3 = vec_add(b6v, b1v); \
00522     /* DST(4,    b6 - b1); */ \
00523     d4 = vec_sub(b6v, b1v); \
00524     /* DST(5,    b4 - b3); */ \
00525     d5 = vec_sub(b4v, b3v); \
00526     /* DST(6,    b2 - b5); */ \
00527     d6 = vec_sub(b2v, b5v); \
00528     /* DST(7,    b0 - b7); */ \
00529     d7 = vec_sub(b0v, b7v); \
00530 }
00531 
00532 #define ALTIVEC_STORE_SUM_CLIP(dest, idctv, perm_ldv, perm_stv, sel) { \
00533     /* unaligned load */                                       \
00534     vec_u8_t hv = vec_ld( 0, dest );                           \
00535     vec_u8_t lv = vec_ld( 7, dest );                           \
00536     vec_u8_t dstv   = vec_perm( hv, lv, (vec_u8_t)perm_ldv );  \
00537     vec_s16_t idct_sh6 = vec_sra(idctv, sixv);                 \
00538     vec_u16_t dst16 = (vec_u16_t)vec_mergeh(zero_u8v, dstv);   \
00539     vec_s16_t idstsum = vec_adds(idct_sh6, (vec_s16_t)dst16);  \
00540     vec_u8_t idstsum8 = vec_packsu(zero_s16v, idstsum);        \
00541     vec_u8_t edgehv;                                           \
00542     /* unaligned store */                                      \
00543     vec_u8_t bodyv  = vec_perm( idstsum8, idstsum8, perm_stv );\
00544     vec_u8_t edgelv = vec_perm( sel, zero_u8v, perm_stv );     \
00545     lv    = vec_sel( lv, bodyv, edgelv );                      \
00546     vec_st( lv, 7, dest );                                     \
00547     hv    = vec_ld( 0, dest );                                 \
00548     edgehv = vec_perm( zero_u8v, sel, perm_stv );              \
00549     hv    = vec_sel( hv, bodyv, edgehv );                      \
00550     vec_st( hv, 0, dest );                                     \
00551  }
00552 
00553 void ff_h264_idct8_add_altivec( uint8_t *dst, DCTELEM *dct, int stride ) {
00554     vec_s16_t s0, s1, s2, s3, s4, s5, s6, s7;
00555     vec_s16_t d0, d1, d2, d3, d4, d5, d6, d7;
00556     vec_s16_t idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7;
00557 
00558     vec_u8_t perm_ldv = vec_lvsl(0, dst);
00559     vec_u8_t perm_stv = vec_lvsr(8, dst);
00560 
00561     const vec_u16_t onev = vec_splat_u16(1);
00562     const vec_u16_t twov = vec_splat_u16(2);
00563     const vec_u16_t sixv = vec_splat_u16(6);
00564 
00565     const vec_u8_t sel = (vec_u8_t) AVV(0,0,0,0,0,0,0,0,-1,-1,-1,-1,-1,-1,-1,-1);
00566     LOAD_ZERO;
00567 
00568     dct[0] += 32; // rounding for the >>6 at the end
00569 
00570     s0 = vec_ld(0x00, (int16_t*)dct);
00571     s1 = vec_ld(0x10, (int16_t*)dct);
00572     s2 = vec_ld(0x20, (int16_t*)dct);
00573     s3 = vec_ld(0x30, (int16_t*)dct);
00574     s4 = vec_ld(0x40, (int16_t*)dct);
00575     s5 = vec_ld(0x50, (int16_t*)dct);
00576     s6 = vec_ld(0x60, (int16_t*)dct);
00577     s7 = vec_ld(0x70, (int16_t*)dct);
00578 
00579     IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7,
00580                      d0, d1, d2, d3, d4, d5, d6, d7);
00581 
00582     TRANSPOSE8( d0,  d1,  d2,  d3,  d4,  d5,  d6, d7 );
00583 
00584     IDCT8_1D_ALTIVEC(d0,  d1,  d2,  d3,  d4,  d5,  d6, d7,
00585                      idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7);
00586 
00587     ALTIVEC_STORE_SUM_CLIP(&dst[0*stride], idct0, perm_ldv, perm_stv, sel);
00588     ALTIVEC_STORE_SUM_CLIP(&dst[1*stride], idct1, perm_ldv, perm_stv, sel);
00589     ALTIVEC_STORE_SUM_CLIP(&dst[2*stride], idct2, perm_ldv, perm_stv, sel);
00590     ALTIVEC_STORE_SUM_CLIP(&dst[3*stride], idct3, perm_ldv, perm_stv, sel);
00591     ALTIVEC_STORE_SUM_CLIP(&dst[4*stride], idct4, perm_ldv, perm_stv, sel);
00592     ALTIVEC_STORE_SUM_CLIP(&dst[5*stride], idct5, perm_ldv, perm_stv, sel);
00593     ALTIVEC_STORE_SUM_CLIP(&dst[6*stride], idct6, perm_ldv, perm_stv, sel);
00594     ALTIVEC_STORE_SUM_CLIP(&dst[7*stride], idct7, perm_ldv, perm_stv, sel);
00595 }
00596 
00597 #define transpose4x16(r0, r1, r2, r3) {      \
00598     register vec_u8_t r4;                    \
00599     register vec_u8_t r5;                    \
00600     register vec_u8_t r6;                    \
00601     register vec_u8_t r7;                    \
00602                                              \
00603     r4 = vec_mergeh(r0, r2);  /*0, 2 set 0*/ \
00604     r5 = vec_mergel(r0, r2);  /*0, 2 set 1*/ \
00605     r6 = vec_mergeh(r1, r3);  /*1, 3 set 0*/ \
00606     r7 = vec_mergel(r1, r3);  /*1, 3 set 1*/ \
00607                                              \
00608     r0 = vec_mergeh(r4, r6);  /*all set 0*/  \
00609     r1 = vec_mergel(r4, r6);  /*all set 1*/  \
00610     r2 = vec_mergeh(r5, r7);  /*all set 2*/  \
00611     r3 = vec_mergel(r5, r7);  /*all set 3*/  \
00612 }
00613 
00614 static inline void write16x4(uint8_t *dst, int dst_stride,
00615                              register vec_u8_t r0, register vec_u8_t r1,
00616                              register vec_u8_t r2, register vec_u8_t r3) {
00617     DECLARE_ALIGNED_16(unsigned char, result[64]);
00618     uint32_t *src_int = (uint32_t *)result, *dst_int = (uint32_t *)dst;
00619     int int_dst_stride = dst_stride/4;
00620 
00621     vec_st(r0, 0, result);
00622     vec_st(r1, 16, result);
00623     vec_st(r2, 32, result);
00624     vec_st(r3, 48, result);
00625     /* FIXME: there has to be a better way!!!! */
00626     *dst_int = *src_int;
00627     *(dst_int+   int_dst_stride) = *(src_int + 1);
00628     *(dst_int+ 2*int_dst_stride) = *(src_int + 2);
00629     *(dst_int+ 3*int_dst_stride) = *(src_int + 3);
00630     *(dst_int+ 4*int_dst_stride) = *(src_int + 4);
00631     *(dst_int+ 5*int_dst_stride) = *(src_int + 5);
00632     *(dst_int+ 6*int_dst_stride) = *(src_int + 6);
00633     *(dst_int+ 7*int_dst_stride) = *(src_int + 7);
00634     *(dst_int+ 8*int_dst_stride) = *(src_int + 8);
00635     *(dst_int+ 9*int_dst_stride) = *(src_int + 9);
00636     *(dst_int+10*int_dst_stride) = *(src_int + 10);
00637     *(dst_int+11*int_dst_stride) = *(src_int + 11);
00638     *(dst_int+12*int_dst_stride) = *(src_int + 12);
00639     *(dst_int+13*int_dst_stride) = *(src_int + 13);
00640     *(dst_int+14*int_dst_stride) = *(src_int + 14);
00641     *(dst_int+15*int_dst_stride) = *(src_int + 15);
00642 }
00643 
00647 #define readAndTranspose16x6(src, src_stride, r8, r9, r10, r11, r12, r13) {\
00648     register vec_u8_t r0  = unaligned_load(0,             src);            \
00649     register vec_u8_t r1  = unaligned_load(   src_stride, src);            \
00650     register vec_u8_t r2  = unaligned_load(2* src_stride, src);            \
00651     register vec_u8_t r3  = unaligned_load(3* src_stride, src);            \
00652     register vec_u8_t r4  = unaligned_load(4* src_stride, src);            \
00653     register vec_u8_t r5  = unaligned_load(5* src_stride, src);            \
00654     register vec_u8_t r6  = unaligned_load(6* src_stride, src);            \
00655     register vec_u8_t r7  = unaligned_load(7* src_stride, src);            \
00656     register vec_u8_t r14 = unaligned_load(14*src_stride, src);            \
00657     register vec_u8_t r15 = unaligned_load(15*src_stride, src);            \
00658                                                                            \
00659     r8  = unaligned_load( 8*src_stride, src);                              \
00660     r9  = unaligned_load( 9*src_stride, src);                              \
00661     r10 = unaligned_load(10*src_stride, src);                              \
00662     r11 = unaligned_load(11*src_stride, src);                              \
00663     r12 = unaligned_load(12*src_stride, src);                              \
00664     r13 = unaligned_load(13*src_stride, src);                              \
00665                                                                            \
00666     /*Merge first pairs*/                                                  \
00667     r0 = vec_mergeh(r0, r8);    /*0, 8*/                                   \
00668     r1 = vec_mergeh(r1, r9);    /*1, 9*/                                   \
00669     r2 = vec_mergeh(r2, r10);   /*2,10*/                                   \
00670     r3 = vec_mergeh(r3, r11);   /*3,11*/                                   \
00671     r4 = vec_mergeh(r4, r12);   /*4,12*/                                   \
00672     r5 = vec_mergeh(r5, r13);   /*5,13*/                                   \
00673     r6 = vec_mergeh(r6, r14);   /*6,14*/                                   \
00674     r7 = vec_mergeh(r7, r15);   /*7,15*/                                   \
00675                                                                            \
00676     /*Merge second pairs*/                                                 \
00677     r8  = vec_mergeh(r0, r4);   /*0,4, 8,12 set 0*/                        \
00678     r9  = vec_mergel(r0, r4);   /*0,4, 8,12 set 1*/                        \
00679     r10 = vec_mergeh(r1, r5);   /*1,5, 9,13 set 0*/                        \
00680     r11 = vec_mergel(r1, r5);   /*1,5, 9,13 set 1*/                        \
00681     r12 = vec_mergeh(r2, r6);   /*2,6,10,14 set 0*/                        \
00682     r13 = vec_mergel(r2, r6);   /*2,6,10,14 set 1*/                        \
00683     r14 = vec_mergeh(r3, r7);   /*3,7,11,15 set 0*/                        \
00684     r15 = vec_mergel(r3, r7);   /*3,7,11,15 set 1*/                        \
00685                                                                            \
00686     /*Third merge*/                                                        \
00687     r0 = vec_mergeh(r8, r12);   /*0,2,4,6,8,10,12,14 set 0*/               \
00688     r1 = vec_mergel(r8, r12);   /*0,2,4,6,8,10,12,14 set 1*/               \
00689     r2 = vec_mergeh(r9, r13);   /*0,2,4,6,8,10,12,14 set 2*/               \
00690     r4 = vec_mergeh(r10, r14);  /*1,3,5,7,9,11,13,15 set 0*/               \
00691     r5 = vec_mergel(r10, r14);  /*1,3,5,7,9,11,13,15 set 1*/               \
00692     r6 = vec_mergeh(r11, r15);  /*1,3,5,7,9,11,13,15 set 2*/               \
00693     /* Don't need to compute 3 and 7*/                                     \
00694                                                                            \
00695     /*Final merge*/                                                        \
00696     r8  = vec_mergeh(r0, r4);   /*all set 0*/                              \
00697     r9  = vec_mergel(r0, r4);   /*all set 1*/                              \
00698     r10 = vec_mergeh(r1, r5);   /*all set 2*/                              \
00699     r11 = vec_mergel(r1, r5);   /*all set 3*/                              \
00700     r12 = vec_mergeh(r2, r6);   /*all set 4*/                              \
00701     r13 = vec_mergel(r2, r6);   /*all set 5*/                              \
00702     /* Don't need to compute 14 and 15*/                                   \
00703                                                                            \
00704 }
00705 
00706 // out: o = |x-y| < a
00707 static inline vec_u8_t diff_lt_altivec ( register vec_u8_t x,
00708                                          register vec_u8_t y,
00709                                          register vec_u8_t a) {
00710 
00711     register vec_u8_t diff = vec_subs(x, y);
00712     register vec_u8_t diffneg = vec_subs(y, x);
00713     register vec_u8_t o = vec_or(diff, diffneg); /* |x-y| */
00714     o = (vec_u8_t)vec_cmplt(o, a);
00715     return o;
00716 }
00717 
00718 static inline vec_u8_t h264_deblock_mask ( register vec_u8_t p0,
00719                                            register vec_u8_t p1,
00720                                            register vec_u8_t q0,
00721                                            register vec_u8_t q1,
00722                                            register vec_u8_t alpha,
00723                                            register vec_u8_t beta) {
00724 
00725     register vec_u8_t mask;
00726     register vec_u8_t tempmask;
00727 
00728     mask = diff_lt_altivec(p0, q0, alpha);
00729     tempmask = diff_lt_altivec(p1, p0, beta);
00730     mask = vec_and(mask, tempmask);
00731     tempmask = diff_lt_altivec(q1, q0, beta);
00732     mask = vec_and(mask, tempmask);
00733 
00734     return mask;
00735 }
00736 
00737 // out: newp1 = clip((p2 + ((p0 + q0 + 1) >> 1)) >> 1, p1-tc0, p1+tc0)
00738 static inline vec_u8_t h264_deblock_q1(register vec_u8_t p0,
00739                                        register vec_u8_t p1,
00740                                        register vec_u8_t p2,
00741                                        register vec_u8_t q0,
00742                                        register vec_u8_t tc0) {
00743 
00744     register vec_u8_t average = vec_avg(p0, q0);
00745     register vec_u8_t temp;
00746     register vec_u8_t uncliped;
00747     register vec_u8_t ones;
00748     register vec_u8_t max;
00749     register vec_u8_t min;
00750     register vec_u8_t newp1;
00751 
00752     temp = vec_xor(average, p2);
00753     average = vec_avg(average, p2);     /*avg(p2, avg(p0, q0)) */
00754     ones = vec_splat_u8(1);
00755     temp = vec_and(temp, ones);         /*(p2^avg(p0, q0)) & 1 */
00756     uncliped = vec_subs(average, temp); /*(p2+((p0+q0+1)>>1))>>1 */
00757     max = vec_adds(p1, tc0);
00758     min = vec_subs(p1, tc0);
00759     newp1 = vec_max(min, uncliped);
00760     newp1 = vec_min(max, newp1);
00761     return newp1;
00762 }
00763 
00764 #define h264_deblock_p0_q0(p0, p1, q0, q1, tc0masked) {                                           \
00765                                                                                                   \
00766     const vec_u8_t A0v = vec_sl(vec_splat_u8(10), vec_splat_u8(4));                               \
00767                                                                                                   \
00768     register vec_u8_t pq0bit = vec_xor(p0,q0);                                                    \
00769     register vec_u8_t q1minus;                                                                    \
00770     register vec_u8_t p0minus;                                                                    \
00771     register vec_u8_t stage1;                                                                     \
00772     register vec_u8_t stage2;                                                                     \
00773     register vec_u8_t vec160;                                                                     \
00774     register vec_u8_t delta;                                                                      \
00775     register vec_u8_t deltaneg;                                                                   \
00776                                                                                                   \
00777     q1minus = vec_nor(q1, q1);                 /* 255 - q1 */                                     \
00778     stage1 = vec_avg(p1, q1minus);             /* (p1 - q1 + 256)>>1 */                           \
00779     stage2 = vec_sr(stage1, vec_splat_u8(1));  /* (p1 - q1 + 256)>>2 = 64 + (p1 - q1) >> 2 */     \
00780     p0minus = vec_nor(p0, p0);                 /* 255 - p0 */                                     \
00781     stage1 = vec_avg(q0, p0minus);             /* (q0 - p0 + 256)>>1 */                           \
00782     pq0bit = vec_and(pq0bit, vec_splat_u8(1));                                                    \
00783     stage2 = vec_avg(stage2, pq0bit);          /* 32 + ((q0 - p0)&1 + (p1 - q1) >> 2 + 1) >> 1 */ \
00784     stage2 = vec_adds(stage2, stage1);         /* 160 + ((p0 - q0) + (p1 - q1) >> 2 + 1) >> 1 */  \
00785     vec160 = vec_ld(0, &A0v);                                                                     \
00786     deltaneg = vec_subs(vec160, stage2);       /* -d */                                           \
00787     delta = vec_subs(stage2, vec160);          /* d */                                            \
00788     deltaneg = vec_min(tc0masked, deltaneg);                                                      \
00789     delta = vec_min(tc0masked, delta);                                                            \
00790     p0 = vec_subs(p0, deltaneg);                                                                  \
00791     q0 = vec_subs(q0, delta);                                                                     \
00792     p0 = vec_adds(p0, delta);                                                                     \
00793     q0 = vec_adds(q0, deltaneg);                                                                  \
00794 }
00795 
00796 #define h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0) {            \
00797     DECLARE_ALIGNED_16(unsigned char, temp[16]);                                             \
00798     register vec_u8_t alphavec;                                                              \
00799     register vec_u8_t betavec;                                                               \
00800     register vec_u8_t mask;                                                                  \
00801     register vec_u8_t p1mask;                                                                \
00802     register vec_u8_t q1mask;                                                                \
00803     register vector signed   char tc0vec;                                                    \
00804     register vec_u8_t finaltc0;                                                              \
00805     register vec_u8_t tc0masked;                                                             \
00806     register vec_u8_t newp1;                                                                 \
00807     register vec_u8_t newq1;                                                                 \
00808                                                                                              \
00809     temp[0] = alpha;                                                                         \
00810     temp[1] = beta;                                                                          \
00811     alphavec = vec_ld(0, temp);                                                              \
00812     betavec = vec_splat(alphavec, 0x1);                                                      \
00813     alphavec = vec_splat(alphavec, 0x0);                                                     \
00814     mask = h264_deblock_mask(p0, p1, q0, q1, alphavec, betavec); /*if in block */            \
00815                                                                                              \
00816     *((int *)temp) = *((int *)tc0);                                                          \
00817     tc0vec = vec_ld(0, (signed char*)temp);                                                  \
00818     tc0vec = vec_mergeh(tc0vec, tc0vec);                                                     \
00819     tc0vec = vec_mergeh(tc0vec, tc0vec);                                                     \
00820     mask = vec_and(mask, vec_cmpgt(tc0vec, vec_splat_s8(-1)));  /* if tc0[i] >= 0 */         \
00821     finaltc0 = vec_and((vec_u8_t)tc0vec, mask);     /* tc = tc0 */                           \
00822                                                                                              \
00823     p1mask = diff_lt_altivec(p2, p0, betavec);                                               \
00824     p1mask = vec_and(p1mask, mask);                             /* if ( |p2 - p0| < beta) */ \
00825     tc0masked = vec_and(p1mask, (vec_u8_t)tc0vec);                                           \
00826     finaltc0 = vec_sub(finaltc0, p1mask);                       /* tc++ */                   \
00827     newp1 = h264_deblock_q1(p0, p1, p2, q0, tc0masked);                                      \
00828     /*end if*/                                                                               \
00829                                                                                              \
00830     q1mask = diff_lt_altivec(q2, q0, betavec);                                               \
00831     q1mask = vec_and(q1mask, mask);                             /* if ( |q2 - q0| < beta ) */\
00832     tc0masked = vec_and(q1mask, (vec_u8_t)tc0vec);                                           \
00833     finaltc0 = vec_sub(finaltc0, q1mask);                       /* tc++ */                   \
00834     newq1 = h264_deblock_q1(p0, q1, q2, q0, tc0masked);                                      \
00835     /*end if*/                                                                               \
00836                                                                                              \
00837     h264_deblock_p0_q0(p0, p1, q0, q1, finaltc0);                                            \
00838     p1 = newp1;                                                                              \
00839     q1 = newq1;                                                                              \
00840 }
00841 
00842 static void h264_v_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) {
00843 
00844     if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) >= 0) {
00845         register vec_u8_t p2 = vec_ld(-3*stride, pix);
00846         register vec_u8_t p1 = vec_ld(-2*stride, pix);
00847         register vec_u8_t p0 = vec_ld(-1*stride, pix);
00848         register vec_u8_t q0 = vec_ld(0, pix);
00849         register vec_u8_t q1 = vec_ld(stride, pix);
00850         register vec_u8_t q2 = vec_ld(2*stride, pix);
00851         h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0);
00852         vec_st(p1, -2*stride, pix);
00853         vec_st(p0, -1*stride, pix);
00854         vec_st(q0, 0, pix);
00855         vec_st(q1, stride, pix);
00856     }
00857 }
00858 
00859 static void h264_h_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) {
00860 
00861     register vec_u8_t line0, line1, line2, line3, line4, line5;
00862     if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) < 0)
00863         return;
00864     readAndTranspose16x6(pix-3, stride, line0, line1, line2, line3, line4, line5);
00865     h264_loop_filter_luma_altivec(line0, line1, line2, line3, line4, line5, alpha, beta, tc0);
00866     transpose4x16(line1, line2, line3, line4);
00867     write16x4(pix-2, stride, line1, line2, line3, line4);
00868 }
00869 
00870 void dsputil_h264_init_ppc(DSPContext* c, AVCodecContext *avctx) {
00871 
00872     if (has_altivec()) {
00873         c->put_h264_chroma_pixels_tab[0] = put_h264_chroma_mc8_altivec;
00874         c->put_no_rnd_h264_chroma_pixels_tab[0] = put_no_rnd_h264_chroma_mc8_altivec;
00875         c->avg_h264_chroma_pixels_tab[0] = avg_h264_chroma_mc8_altivec;
00876         c->h264_idct_add = ff_h264_idct_add_altivec;
00877         c->h264_idct8_add = ff_h264_idct8_add_altivec;
00878         c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_altivec;
00879         c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_altivec;
00880 
00881 #define dspfunc(PFX, IDX, NUM) \
00882         c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_altivec; \
00883         c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_altivec; \
00884         c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_altivec; \
00885         c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_altivec; \
00886         c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_altivec; \
00887         c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_altivec; \
00888         c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_altivec; \
00889         c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_altivec; \
00890         c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_altivec; \
00891         c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_altivec; \
00892         c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_altivec; \
00893         c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_altivec; \
00894         c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_altivec; \
00895         c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_altivec; \
00896         c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_altivec; \
00897         c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_altivec
00898 
00899         dspfunc(put_h264_qpel, 0, 16);
00900         dspfunc(avg_h264_qpel, 0, 16);
00901 #undef dspfunc
00902     }
00903 }