FFmpeg: libavcodec/ppc/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 "libavutil/cpu.h"
00022 #include "libavcodec/dsputil.h"
00023 #include "libavcodec/h264data.h"
00024 #include "libavcodec/h264dsp.h"
00025 
00026 #include "dsputil_altivec.h"
00027 #include "util_altivec.h"
00028 #include "types_altivec.h"
00029 
00030 #define PUT_OP_U8_ALTIVEC(d, s, dst) d = s
00031 #define AVG_OP_U8_ALTIVEC(d, s, dst) d = vec_avg(dst, s)
00032 
00033 #define OP_U8_ALTIVEC                          PUT_OP_U8_ALTIVEC
00034 #define PREFIX_h264_chroma_mc8_altivec         put_h264_chroma_mc8_altivec
00035 #define PREFIX_h264_chroma_mc8_num             altivec_put_h264_chroma_mc8_num
00036 #define PREFIX_h264_qpel16_h_lowpass_altivec   put_h264_qpel16_h_lowpass_altivec
00037 #define PREFIX_h264_qpel16_h_lowpass_num       altivec_put_h264_qpel16_h_lowpass_num
00038 #define PREFIX_h264_qpel16_v_lowpass_altivec   put_h264_qpel16_v_lowpass_altivec
00039 #define PREFIX_h264_qpel16_v_lowpass_num       altivec_put_h264_qpel16_v_lowpass_num
00040 #define PREFIX_h264_qpel16_hv_lowpass_altivec  put_h264_qpel16_hv_lowpass_altivec
00041 #define PREFIX_h264_qpel16_hv_lowpass_num      altivec_put_h264_qpel16_hv_lowpass_num
00042 #include "h264_template_altivec.c"
00043 #undef OP_U8_ALTIVEC
00044 #undef PREFIX_h264_chroma_mc8_altivec
00045 #undef PREFIX_h264_chroma_mc8_num
00046 #undef PREFIX_h264_qpel16_h_lowpass_altivec
00047 #undef PREFIX_h264_qpel16_h_lowpass_num
00048 #undef PREFIX_h264_qpel16_v_lowpass_altivec
00049 #undef PREFIX_h264_qpel16_v_lowpass_num
00050 #undef PREFIX_h264_qpel16_hv_lowpass_altivec
00051 #undef PREFIX_h264_qpel16_hv_lowpass_num
00052 
00053 #define OP_U8_ALTIVEC                          AVG_OP_U8_ALTIVEC
00054 #define PREFIX_h264_chroma_mc8_altivec         avg_h264_chroma_mc8_altivec
00055 #define PREFIX_h264_chroma_mc8_num             altivec_avg_h264_chroma_mc8_num
00056 #define PREFIX_h264_qpel16_h_lowpass_altivec   avg_h264_qpel16_h_lowpass_altivec
00057 #define PREFIX_h264_qpel16_h_lowpass_num       altivec_avg_h264_qpel16_h_lowpass_num
00058 #define PREFIX_h264_qpel16_v_lowpass_altivec   avg_h264_qpel16_v_lowpass_altivec
00059 #define PREFIX_h264_qpel16_v_lowpass_num       altivec_avg_h264_qpel16_v_lowpass_num
00060 #define PREFIX_h264_qpel16_hv_lowpass_altivec  avg_h264_qpel16_hv_lowpass_altivec
00061 #define PREFIX_h264_qpel16_hv_lowpass_num      altivec_avg_h264_qpel16_hv_lowpass_num
00062 #include "h264_template_altivec.c"
00063 #undef OP_U8_ALTIVEC
00064 #undef PREFIX_h264_chroma_mc8_altivec
00065 #undef PREFIX_h264_chroma_mc8_num
00066 #undef PREFIX_h264_qpel16_h_lowpass_altivec
00067 #undef PREFIX_h264_qpel16_h_lowpass_num
00068 #undef PREFIX_h264_qpel16_v_lowpass_altivec
00069 #undef PREFIX_h264_qpel16_v_lowpass_num
00070 #undef PREFIX_h264_qpel16_hv_lowpass_altivec
00071 #undef PREFIX_h264_qpel16_hv_lowpass_num
00072 
00073 #define H264_MC(OPNAME, SIZE, CODETYPE) \
00074 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## CODETYPE (uint8_t *dst, uint8_t *src, int stride){\
00075     OPNAME ## pixels ## SIZE ## _ ## CODETYPE(dst, src, stride, SIZE);\
00076 }\
00077 \
00078 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){ \
00079     DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
00080     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
00081     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\
00082 }\
00083 \
00084 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00085     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(dst, src, stride, stride);\
00086 }\
00087 \
00088 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00089     DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
00090     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
00091     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+1, half, stride, stride, SIZE);\
00092 }\
00093 \
00094 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00095     DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
00096     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
00097     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\
00098 }\
00099 \
00100 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00101     OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(dst, src, stride, stride);\
00102 }\
00103 \
00104 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00105     DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
00106     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
00107     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+stride, half, stride, stride, SIZE);\
00108 }\
00109 \
00110 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00111     DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
00112     DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
00113     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
00114     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
00115     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
00116 }\
00117 \
00118 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00119     DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
00120     DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
00121     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
00122     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
00123     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
00124 }\
00125 \
00126 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00127     DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
00128     DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
00129     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
00130     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
00131     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
00132 }\
00133 \
00134 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00135     DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
00136     DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
00137     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
00138     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
00139     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
00140 }\
00141 \
00142 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00143     DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
00144     OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(dst, tmp, src, stride, SIZE, stride);\
00145 }\
00146 \
00147 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00148     DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
00149     DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
00150     DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
00151     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
00152     put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
00153     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\
00154 }\
00155 \
00156 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00157     DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
00158     DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
00159     DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
00160     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
00161     put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
00162     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\
00163 }\
00164 \
00165 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00166     DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
00167     DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
00168     DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
00169     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
00170     put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
00171     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
00172 }\
00173 \
00174 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00175     DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
00176     DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
00177     DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
00178     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
00179     put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
00180     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
00181 }\
00182 
00183 static inline void put_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
00184                                     const uint8_t * src2, int dst_stride,
00185                                     int src_stride1, int h)
00186 {
00187     int i;
00188     vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align;
00189 
00190     mask_ = vec_lvsl(0, src2);
00191 
00192     for (i = 0; i < h; i++) {
00193 
00194         tmp1 = vec_ld(i * src_stride1, src1);
00195         mask = vec_lvsl(i * src_stride1, src1);
00196         tmp2 = vec_ld(i * src_stride1 + 15, src1);
00197 
00198         a = vec_perm(tmp1, tmp2, mask);
00199 
00200         tmp1 = vec_ld(i * 16, src2);
00201         tmp2 = vec_ld(i * 16 + 15, src2);
00202 
00203         b = vec_perm(tmp1, tmp2, mask_);
00204 
00205         tmp1 = vec_ld(0, dst);
00206         mask = vec_lvsl(0, dst);
00207         tmp2 = vec_ld(15, dst);
00208 
00209         d = vec_avg(a, b);
00210 
00211         edges = vec_perm(tmp2, tmp1, mask);
00212 
00213         align = vec_lvsr(0, dst);
00214 
00215         tmp2 = vec_perm(d, edges, align);
00216         tmp1 = vec_perm(edges, d, align);
00217 
00218         vec_st(tmp2, 15, dst);
00219         vec_st(tmp1, 0 , dst);
00220 
00221         dst += dst_stride;
00222     }
00223 }
00224 
00225 static inline void avg_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
00226                                     const uint8_t * src2, int dst_stride,
00227                                     int src_stride1, int h)
00228 {
00229     int i;
00230     vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align;
00231 
00232     mask_ = vec_lvsl(0, src2);
00233 
00234     for (i = 0; i < h; i++) {
00235 
00236         tmp1 = vec_ld(i * src_stride1, src1);
00237         mask = vec_lvsl(i * src_stride1, src1);
00238         tmp2 = vec_ld(i * src_stride1 + 15, src1);
00239 
00240         a = vec_perm(tmp1, tmp2, mask);
00241 
00242         tmp1 = vec_ld(i * 16, src2);
00243         tmp2 = vec_ld(i * 16 + 15, src2);
00244 
00245         b = vec_perm(tmp1, tmp2, mask_);
00246 
00247         tmp1 = vec_ld(0, dst);
00248         mask = vec_lvsl(0, dst);
00249         tmp2 = vec_ld(15, dst);
00250 
00251         d = vec_avg(vec_perm(tmp1, tmp2, mask), vec_avg(a, b));
00252 
00253         edges = vec_perm(tmp2, tmp1, mask);
00254 
00255         align = vec_lvsr(0, dst);
00256 
00257         tmp2 = vec_perm(d, edges, align);
00258         tmp1 = vec_perm(edges, d, align);
00259 
00260         vec_st(tmp2, 15, dst);
00261         vec_st(tmp1, 0 , dst);
00262 
00263         dst += dst_stride;
00264     }
00265 }
00266 
00267 /* Implemented but could be faster
00268 #define put_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) put_pixels16_l2(d,s1,s2,ds,s1s,16,h)
00269 #define avg_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) avg_pixels16_l2(d,s1,s2,ds,s1s,16,h)
00270  */
00271 
00272 H264_MC(put_, 16, altivec)
00273 H264_MC(avg_, 16, altivec)
00274 
00275 
00276 /****************************************************************************
00277  * IDCT transform:
00278  ****************************************************************************/
00279 
00280 #define VEC_1D_DCT(vb0,vb1,vb2,vb3,va0,va1,va2,va3)               \
00281     /* 1st stage */                                               \
00282     vz0 = vec_add(vb0,vb2);       /* temp[0] = Y[0] + Y[2] */     \
00283     vz1 = vec_sub(vb0,vb2);       /* temp[1] = Y[0] - Y[2] */     \
00284     vz2 = vec_sra(vb1,vec_splat_u16(1));                          \
00285     vz2 = vec_sub(vz2,vb3);       /* temp[2] = Y[1].1/2 - Y[3] */ \
00286     vz3 = vec_sra(vb3,vec_splat_u16(1));                          \
00287     vz3 = vec_add(vb1,vz3);       /* temp[3] = Y[1] + Y[3].1/2 */ \
00288     /* 2nd stage: output */                                       \
00289     va0 = vec_add(vz0,vz3);       /* x[0] = temp[0] + temp[3] */  \
00290     va1 = vec_add(vz1,vz2);       /* x[1] = temp[1] + temp[2] */  \
00291     va2 = vec_sub(vz1,vz2);       /* x[2] = temp[1] - temp[2] */  \
00292     va3 = vec_sub(vz0,vz3)        /* x[3] = temp[0] - temp[3] */
00293 
00294 #define VEC_TRANSPOSE_4(a0,a1,a2,a3,b0,b1,b2,b3) \
00295     b0 = vec_mergeh( a0, a0 ); \
00296     b1 = vec_mergeh( a1, a0 ); \
00297     b2 = vec_mergeh( a2, a0 ); \
00298     b3 = vec_mergeh( a3, a0 ); \
00299     a0 = vec_mergeh( b0, b2 ); \
00300     a1 = vec_mergel( b0, b2 ); \
00301     a2 = vec_mergeh( b1, b3 ); \
00302     a3 = vec_mergel( b1, b3 ); \
00303     b0 = vec_mergeh( a0, a2 ); \
00304     b1 = vec_mergel( a0, a2 ); \
00305     b2 = vec_mergeh( a1, a3 ); \
00306     b3 = vec_mergel( a1, a3 )
00307 
00308 #define VEC_LOAD_U8_ADD_S16_STORE_U8(va)                      \
00309     vdst_orig = vec_ld(0, dst);                               \
00310     vdst = vec_perm(vdst_orig, zero_u8v, vdst_mask);          \
00311     vdst_ss = (vec_s16) vec_mergeh(zero_u8v, vdst);         \
00312     va = vec_add(va, vdst_ss);                                \
00313     va_u8 = vec_packsu(va, zero_s16v);                        \
00314     va_u32 = vec_splat((vec_u32)va_u8, 0);                  \
00315     vec_ste(va_u32, element, (uint32_t*)dst);
00316 
00317 static void ff_h264_idct_add_altivec(uint8_t *dst, DCTELEM *block, int stride)
00318 {
00319     vec_s16 va0, va1, va2, va3;
00320     vec_s16 vz0, vz1, vz2, vz3;
00321     vec_s16 vtmp0, vtmp1, vtmp2, vtmp3;
00322     vec_u8 va_u8;
00323     vec_u32 va_u32;
00324     vec_s16 vdst_ss;
00325     const vec_u16 v6us = vec_splat_u16(6);
00326     vec_u8 vdst, vdst_orig;
00327     vec_u8 vdst_mask = vec_lvsl(0, dst);
00328     int element = ((unsigned long)dst & 0xf) >> 2;
00329     LOAD_ZERO;
00330 
00331     block[0] += 32;  /* add 32 as a DC-level for rounding */
00332 
00333     vtmp0 = vec_ld(0,block);
00334     vtmp1 = vec_sld(vtmp0, vtmp0, 8);
00335     vtmp2 = vec_ld(16,block);
00336     vtmp3 = vec_sld(vtmp2, vtmp2, 8);
00337 
00338     VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3);
00339     VEC_TRANSPOSE_4(va0,va1,va2,va3,vtmp0,vtmp1,vtmp2,vtmp3);
00340     VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3);
00341 
00342     va0 = vec_sra(va0,v6us);
00343     va1 = vec_sra(va1,v6us);
00344     va2 = vec_sra(va2,v6us);
00345     va3 = vec_sra(va3,v6us);
00346 
00347     VEC_LOAD_U8_ADD_S16_STORE_U8(va0);
00348     dst += stride;
00349     VEC_LOAD_U8_ADD_S16_STORE_U8(va1);
00350     dst += stride;
00351     VEC_LOAD_U8_ADD_S16_STORE_U8(va2);
00352     dst += stride;
00353     VEC_LOAD_U8_ADD_S16_STORE_U8(va3);
00354 }
00355 
00356 #define IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7,  d0, d1, d2, d3, d4, d5, d6, d7) {\
00357     /*        a0  = SRC(0) + SRC(4); */ \
00358     vec_s16 a0v = vec_add(s0, s4);    \
00359     /*        a2  = SRC(0) - SRC(4); */ \
00360     vec_s16 a2v = vec_sub(s0, s4);    \
00361     /*        a4  =           (SRC(2)>>1) - SRC(6); */ \
00362     vec_s16 a4v = vec_sub(vec_sra(s2, onev), s6);    \
00363     /*        a6  =           (SRC(6)>>1) + SRC(2); */ \
00364     vec_s16 a6v = vec_add(vec_sra(s6, onev), s2);    \
00365     /*        b0  =         a0 + a6; */ \
00366     vec_s16 b0v = vec_add(a0v, a6v);  \
00367     /*        b2  =         a2 + a4; */ \
00368     vec_s16 b2v = vec_add(a2v, a4v);  \
00369     /*        b4  =         a2 - a4; */ \
00370     vec_s16 b4v = vec_sub(a2v, a4v);  \
00371     /*        b6  =         a0 - a6; */ \
00372     vec_s16 b6v = vec_sub(a0v, a6v);  \
00373     /* a1 =  SRC(5) - SRC(3) - SRC(7) - (SRC(7)>>1); */ \
00374     /*        a1 =             (SRC(5)-SRC(3)) -  (SRC(7)  +  (SRC(7)>>1)); */ \
00375     vec_s16 a1v = vec_sub( vec_sub(s5, s3), vec_add(s7, vec_sra(s7, onev)) ); \
00376     /* a3 =  SRC(7) + SRC(1) - SRC(3) - (SRC(3)>>1); */ \
00377     /*        a3 =             (SRC(7)+SRC(1)) -  (SRC(3)  +  (SRC(3)>>1)); */ \
00378     vec_s16 a3v = vec_sub( vec_add(s7, s1), vec_add(s3, vec_sra(s3, onev)) );\
00379     /* a5 =  SRC(7) - SRC(1) + SRC(5) + (SRC(5)>>1); */ \
00380     /*        a5 =             (SRC(7)-SRC(1)) +   SRC(5) +   (SRC(5)>>1); */ \
00381     vec_s16 a5v = vec_add( vec_sub(s7, s1), vec_add(s5, vec_sra(s5, onev)) );\
00382     /*        a7 =                SRC(5)+SRC(3) +  SRC(1) +   (SRC(1)>>1); */ \
00383     vec_s16 a7v = vec_add( vec_add(s5, s3), vec_add(s1, vec_sra(s1, onev)) );\
00384     /*        b1 =                  (a7>>2)  +  a1; */ \
00385     vec_s16 b1v = vec_add( vec_sra(a7v, twov), a1v); \
00386     /*        b3 =          a3 +        (a5>>2); */ \
00387     vec_s16 b3v = vec_add(a3v, vec_sra(a5v, twov)); \
00388     /*        b5 =                  (a3>>2)  -   a5; */ \
00389     vec_s16 b5v = vec_sub( vec_sra(a3v, twov), a5v); \
00390     /*        b7 =           a7 -        (a1>>2); */ \
00391     vec_s16 b7v = vec_sub( a7v, vec_sra(a1v, twov)); \
00392     /* DST(0,    b0 + b7); */ \
00393     d0 = vec_add(b0v, b7v); \
00394     /* DST(1,    b2 + b5); */ \
00395     d1 = vec_add(b2v, b5v); \
00396     /* DST(2,    b4 + b3); */ \
00397     d2 = vec_add(b4v, b3v); \
00398     /* DST(3,    b6 + b1); */ \
00399     d3 = vec_add(b6v, b1v); \
00400     /* DST(4,    b6 - b1); */ \
00401     d4 = vec_sub(b6v, b1v); \
00402     /* DST(5,    b4 - b3); */ \
00403     d5 = vec_sub(b4v, b3v); \
00404     /* DST(6,    b2 - b5); */ \
00405     d6 = vec_sub(b2v, b5v); \
00406     /* DST(7,    b0 - b7); */ \
00407     d7 = vec_sub(b0v, b7v); \
00408 }
00409 
00410 #define ALTIVEC_STORE_SUM_CLIP(dest, idctv, perm_ldv, perm_stv, sel) { \
00411     /* unaligned load */                                       \
00412     vec_u8 hv = vec_ld( 0, dest );                           \
00413     vec_u8 lv = vec_ld( 7, dest );                           \
00414     vec_u8 dstv   = vec_perm( hv, lv, (vec_u8)perm_ldv );  \
00415     vec_s16 idct_sh6 = vec_sra(idctv, sixv);                 \
00416     vec_u16 dst16 = (vec_u16)vec_mergeh(zero_u8v, dstv);   \
00417     vec_s16 idstsum = vec_adds(idct_sh6, (vec_s16)dst16);  \
00418     vec_u8 idstsum8 = vec_packsu(zero_s16v, idstsum);        \
00419     vec_u8 edgehv;                                           \
00420     /* unaligned store */                                      \
00421     vec_u8 bodyv  = vec_perm( idstsum8, idstsum8, perm_stv );\
00422     vec_u8 edgelv = vec_perm( sel, zero_u8v, perm_stv );     \
00423     lv    = vec_sel( lv, bodyv, edgelv );                      \
00424     vec_st( lv, 7, dest );                                     \
00425     hv    = vec_ld( 0, dest );                                 \
00426     edgehv = vec_perm( zero_u8v, sel, perm_stv );              \
00427     hv    = vec_sel( hv, bodyv, edgehv );                      \
00428     vec_st( hv, 0, dest );                                     \
00429  }
00430 
00431 static void ff_h264_idct8_add_altivec( uint8_t *dst, DCTELEM *dct, int stride ) {
00432     vec_s16 s0, s1, s2, s3, s4, s5, s6, s7;
00433     vec_s16 d0, d1, d2, d3, d4, d5, d6, d7;
00434     vec_s16 idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7;
00435 
00436     vec_u8 perm_ldv = vec_lvsl(0, dst);
00437     vec_u8 perm_stv = vec_lvsr(8, dst);
00438 
00439     const vec_u16 onev = vec_splat_u16(1);
00440     const vec_u16 twov = vec_splat_u16(2);
00441     const vec_u16 sixv = vec_splat_u16(6);
00442 
00443     const vec_u8 sel = (vec_u8) {0,0,0,0,0,0,0,0,-1,-1,-1,-1,-1,-1,-1,-1};
00444     LOAD_ZERO;
00445 
00446     dct[0] += 32; // rounding for the >>6 at the end
00447 
00448     s0 = vec_ld(0x00, (int16_t*)dct);
00449     s1 = vec_ld(0x10, (int16_t*)dct);
00450     s2 = vec_ld(0x20, (int16_t*)dct);
00451     s3 = vec_ld(0x30, (int16_t*)dct);
00452     s4 = vec_ld(0x40, (int16_t*)dct);
00453     s5 = vec_ld(0x50, (int16_t*)dct);
00454     s6 = vec_ld(0x60, (int16_t*)dct);
00455     s7 = vec_ld(0x70, (int16_t*)dct);
00456 
00457     IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7,
00458                      d0, d1, d2, d3, d4, d5, d6, d7);
00459 
00460     TRANSPOSE8( d0,  d1,  d2,  d3,  d4,  d5,  d6, d7 );
00461 
00462     IDCT8_1D_ALTIVEC(d0,  d1,  d2,  d3,  d4,  d5,  d6, d7,
00463                      idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7);
00464 
00465     ALTIVEC_STORE_SUM_CLIP(&dst[0*stride], idct0, perm_ldv, perm_stv, sel);
00466     ALTIVEC_STORE_SUM_CLIP(&dst[1*stride], idct1, perm_ldv, perm_stv, sel);
00467     ALTIVEC_STORE_SUM_CLIP(&dst[2*stride], idct2, perm_ldv, perm_stv, sel);
00468     ALTIVEC_STORE_SUM_CLIP(&dst[3*stride], idct3, perm_ldv, perm_stv, sel);
00469     ALTIVEC_STORE_SUM_CLIP(&dst[4*stride], idct4, perm_ldv, perm_stv, sel);
00470     ALTIVEC_STORE_SUM_CLIP(&dst[5*stride], idct5, perm_ldv, perm_stv, sel);
00471     ALTIVEC_STORE_SUM_CLIP(&dst[6*stride], idct6, perm_ldv, perm_stv, sel);
00472     ALTIVEC_STORE_SUM_CLIP(&dst[7*stride], idct7, perm_ldv, perm_stv, sel);
00473 }
00474 
00475 static av_always_inline void h264_idct_dc_add_internal(uint8_t *dst, DCTELEM *block, int stride, int size)
00476 {
00477     vec_s16 dc16;
00478     vec_u8 dcplus, dcminus, v0, v1, v2, v3, aligner;
00479     LOAD_ZERO;
00480     DECLARE_ALIGNED(16, int, dc);
00481     int i;
00482 
00483     dc = (block[0] + 32) >> 6;
00484     dc16 = vec_splat((vec_s16) vec_lde(0, &dc), 1);
00485 
00486     if (size == 4)
00487         dc16 = vec_sld(dc16, zero_s16v, 8);
00488     dcplus = vec_packsu(dc16, zero_s16v);
00489     dcminus = vec_packsu(vec_sub(zero_s16v, dc16), zero_s16v);
00490 
00491     aligner = vec_lvsr(0, dst);
00492     dcplus = vec_perm(dcplus, dcplus, aligner);
00493     dcminus = vec_perm(dcminus, dcminus, aligner);
00494 
00495     for (i = 0; i < size; i += 4) {
00496         v0 = vec_ld(0, dst+0*stride);
00497         v1 = vec_ld(0, dst+1*stride);
00498         v2 = vec_ld(0, dst+2*stride);
00499         v3 = vec_ld(0, dst+3*stride);
00500 
00501         v0 = vec_adds(v0, dcplus);
00502         v1 = vec_adds(v1, dcplus);
00503         v2 = vec_adds(v2, dcplus);
00504         v3 = vec_adds(v3, dcplus);
00505 
00506         v0 = vec_subs(v0, dcminus);
00507         v1 = vec_subs(v1, dcminus);
00508         v2 = vec_subs(v2, dcminus);
00509         v3 = vec_subs(v3, dcminus);
00510 
00511         vec_st(v0, 0, dst+0*stride);
00512         vec_st(v1, 0, dst+1*stride);
00513         vec_st(v2, 0, dst+2*stride);
00514         vec_st(v3, 0, dst+3*stride);
00515 
00516         dst += 4*stride;
00517     }
00518 }
00519 
00520 static void h264_idct_dc_add_altivec(uint8_t *dst, DCTELEM *block, int stride)
00521 {
00522     h264_idct_dc_add_internal(dst, block, stride, 4);
00523 }
00524 
00525 static void ff_h264_idct8_dc_add_altivec(uint8_t *dst, DCTELEM *block, int stride)
00526 {
00527     h264_idct_dc_add_internal(dst, block, stride, 8);
00528 }
00529 
00530 static void ff_h264_idct_add16_altivec(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[15*8]){
00531     int i;
00532     for(i=0; i<16; i++){
00533         int nnz = nnzc[ scan8[i] ];
00534         if(nnz){
00535             if(nnz==1 && block[i*16]) h264_idct_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
00536             else                      ff_h264_idct_add_altivec(dst + block_offset[i], block + i*16, stride);
00537         }
00538     }
00539 }
00540 
00541 static void ff_h264_idct_add16intra_altivec(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[15*8]){
00542     int i;
00543     for(i=0; i<16; i++){
00544         if(nnzc[ scan8[i] ]) ff_h264_idct_add_altivec(dst + block_offset[i], block + i*16, stride);
00545         else if(block[i*16]) h264_idct_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
00546     }
00547 }
00548 
00549 static void ff_h264_idct8_add4_altivec(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[15*8]){
00550     int i;
00551     for(i=0; i<16; i+=4){
00552         int nnz = nnzc[ scan8[i] ];
00553         if(nnz){
00554             if(nnz==1 && block[i*16]) ff_h264_idct8_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
00555             else                      ff_h264_idct8_add_altivec   (dst + block_offset[i], block + i*16, stride);
00556         }
00557     }
00558 }
00559 
00560 static void ff_h264_idct_add8_altivec(uint8_t **dest, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[15*8]){
00561     int i, j;
00562     for (j = 1; j < 3; j++) {
00563         for(i = j * 16; i < j * 16 + 4; i++){
00564             if(nnzc[ scan8[i] ])
00565                 ff_h264_idct_add_altivec(dest[j-1] + block_offset[i], block + i*16, stride);
00566             else if(block[i*16])
00567                 h264_idct_dc_add_altivec(dest[j-1] + block_offset[i], block + i*16, stride);
00568         }
00569     }
00570 }
00571 
00572 #define transpose4x16(r0, r1, r2, r3) {      \
00573     register vec_u8 r4;                    \
00574     register vec_u8 r5;                    \
00575     register vec_u8 r6;                    \
00576     register vec_u8 r7;                    \
00577                                              \
00578     r4 = vec_mergeh(r0, r2);  /*0, 2 set 0*/ \
00579     r5 = vec_mergel(r0, r2);  /*0, 2 set 1*/ \
00580     r6 = vec_mergeh(r1, r3);  /*1, 3 set 0*/ \
00581     r7 = vec_mergel(r1, r3);  /*1, 3 set 1*/ \
00582                                              \
00583     r0 = vec_mergeh(r4, r6);  /*all set 0*/  \
00584     r1 = vec_mergel(r4, r6);  /*all set 1*/  \
00585     r2 = vec_mergeh(r5, r7);  /*all set 2*/  \
00586     r3 = vec_mergel(r5, r7);  /*all set 3*/  \
00587 }
00588 
00589 static inline void write16x4(uint8_t *dst, int dst_stride,
00590                              register vec_u8 r0, register vec_u8 r1,
00591                              register vec_u8 r2, register vec_u8 r3) {
00592     DECLARE_ALIGNED(16, unsigned char, result)[64];
00593     uint32_t *src_int = (uint32_t *)result, *dst_int = (uint32_t *)dst;
00594     int int_dst_stride = dst_stride/4;
00595 
00596     vec_st(r0, 0, result);
00597     vec_st(r1, 16, result);
00598     vec_st(r2, 32, result);
00599     vec_st(r3, 48, result);
00600     /* FIXME: there has to be a better way!!!! */
00601     *dst_int = *src_int;
00602     *(dst_int+   int_dst_stride) = *(src_int + 1);
00603     *(dst_int+ 2*int_dst_stride) = *(src_int + 2);
00604     *(dst_int+ 3*int_dst_stride) = *(src_int + 3);
00605     *(dst_int+ 4*int_dst_stride) = *(src_int + 4);
00606     *(dst_int+ 5*int_dst_stride) = *(src_int + 5);
00607     *(dst_int+ 6*int_dst_stride) = *(src_int + 6);
00608     *(dst_int+ 7*int_dst_stride) = *(src_int + 7);
00609     *(dst_int+ 8*int_dst_stride) = *(src_int + 8);
00610     *(dst_int+ 9*int_dst_stride) = *(src_int + 9);
00611     *(dst_int+10*int_dst_stride) = *(src_int + 10);
00612     *(dst_int+11*int_dst_stride) = *(src_int + 11);
00613     *(dst_int+12*int_dst_stride) = *(src_int + 12);
00614     *(dst_int+13*int_dst_stride) = *(src_int + 13);
00615     *(dst_int+14*int_dst_stride) = *(src_int + 14);
00616     *(dst_int+15*int_dst_stride) = *(src_int + 15);
00617 }
00618 
00622 #define readAndTranspose16x6(src, src_stride, r8, r9, r10, r11, r12, r13) {\
00623     register vec_u8 r0  = unaligned_load(0,             src);            \
00624     register vec_u8 r1  = unaligned_load(   src_stride, src);            \
00625     register vec_u8 r2  = unaligned_load(2* src_stride, src);            \
00626     register vec_u8 r3  = unaligned_load(3* src_stride, src);            \
00627     register vec_u8 r4  = unaligned_load(4* src_stride, src);            \
00628     register vec_u8 r5  = unaligned_load(5* src_stride, src);            \
00629     register vec_u8 r6  = unaligned_load(6* src_stride, src);            \
00630     register vec_u8 r7  = unaligned_load(7* src_stride, src);            \
00631     register vec_u8 r14 = unaligned_load(14*src_stride, src);            \
00632     register vec_u8 r15 = unaligned_load(15*src_stride, src);            \
00633                                                                            \
00634     r8  = unaligned_load( 8*src_stride, src);                              \
00635     r9  = unaligned_load( 9*src_stride, src);                              \
00636     r10 = unaligned_load(10*src_stride, src);                              \
00637     r11 = unaligned_load(11*src_stride, src);                              \
00638     r12 = unaligned_load(12*src_stride, src);                              \
00639     r13 = unaligned_load(13*src_stride, src);                              \
00640                                                                            \
00641     /*Merge first pairs*/                                                  \
00642     r0 = vec_mergeh(r0, r8);    /*0, 8*/                                   \
00643     r1 = vec_mergeh(r1, r9);    /*1, 9*/                                   \
00644     r2 = vec_mergeh(r2, r10);   /*2,10*/                                   \
00645     r3 = vec_mergeh(r3, r11);   /*3,11*/                                   \
00646     r4 = vec_mergeh(r4, r12);   /*4,12*/                                   \
00647     r5 = vec_mergeh(r5, r13);   /*5,13*/                                   \
00648     r6 = vec_mergeh(r6, r14);   /*6,14*/                                   \
00649     r7 = vec_mergeh(r7, r15);   /*7,15*/                                   \
00650                                                                            \
00651     /*Merge second pairs*/                                                 \
00652     r8  = vec_mergeh(r0, r4);   /*0,4, 8,12 set 0*/                        \
00653     r9  = vec_mergel(r0, r4);   /*0,4, 8,12 set 1*/                        \
00654     r10 = vec_mergeh(r1, r5);   /*1,5, 9,13 set 0*/                        \
00655     r11 = vec_mergel(r1, r5);   /*1,5, 9,13 set 1*/                        \
00656     r12 = vec_mergeh(r2, r6);   /*2,6,10,14 set 0*/                        \
00657     r13 = vec_mergel(r2, r6);   /*2,6,10,14 set 1*/                        \
00658     r14 = vec_mergeh(r3, r7);   /*3,7,11,15 set 0*/                        \
00659     r15 = vec_mergel(r3, r7);   /*3,7,11,15 set 1*/                        \
00660                                                                            \
00661     /*Third merge*/                                                        \
00662     r0 = vec_mergeh(r8,  r12);  /*0,2,4,6,8,10,12,14 set 0*/               \
00663     r1 = vec_mergel(r8,  r12);  /*0,2,4,6,8,10,12,14 set 1*/               \
00664     r2 = vec_mergeh(r9,  r13);  /*0,2,4,6,8,10,12,14 set 2*/               \
00665     r4 = vec_mergeh(r10, r14);  /*1,3,5,7,9,11,13,15 set 0*/               \
00666     r5 = vec_mergel(r10, r14);  /*1,3,5,7,9,11,13,15 set 1*/               \
00667     r6 = vec_mergeh(r11, r15);  /*1,3,5,7,9,11,13,15 set 2*/               \
00668     /* Don't need to compute 3 and 7*/                                     \
00669                                                                            \
00670     /*Final merge*/                                                        \
00671     r8  = vec_mergeh(r0, r4);   /*all set 0*/                              \
00672     r9  = vec_mergel(r0, r4);   /*all set 1*/                              \
00673     r10 = vec_mergeh(r1, r5);   /*all set 2*/                              \
00674     r11 = vec_mergel(r1, r5);   /*all set 3*/                              \
00675     r12 = vec_mergeh(r2, r6);   /*all set 4*/                              \
00676     r13 = vec_mergel(r2, r6);   /*all set 5*/                              \
00677     /* Don't need to compute 14 and 15*/                                   \
00678                                                                            \
00679 }
00680 
00681 // out: o = |x-y| < a
00682 static inline vec_u8 diff_lt_altivec ( register vec_u8 x,
00683                                          register vec_u8 y,
00684                                          register vec_u8 a) {
00685 
00686     register vec_u8 diff = vec_subs(x, y);
00687     register vec_u8 diffneg = vec_subs(y, x);
00688     register vec_u8 o = vec_or(diff, diffneg); /* |x-y| */
00689     o = (vec_u8)vec_cmplt(o, a);
00690     return o;
00691 }
00692 
00693 static inline vec_u8 h264_deblock_mask ( register vec_u8 p0,
00694                                            register vec_u8 p1,
00695                                            register vec_u8 q0,
00696                                            register vec_u8 q1,
00697                                            register vec_u8 alpha,
00698                                            register vec_u8 beta) {
00699 
00700     register vec_u8 mask;
00701     register vec_u8 tempmask;
00702 
00703     mask = diff_lt_altivec(p0, q0, alpha);
00704     tempmask = diff_lt_altivec(p1, p0, beta);
00705     mask = vec_and(mask, tempmask);
00706     tempmask = diff_lt_altivec(q1, q0, beta);
00707     mask = vec_and(mask, tempmask);
00708 
00709     return mask;
00710 }
00711 
00712 // out: newp1 = clip((p2 + ((p0 + q0 + 1) >> 1)) >> 1, p1-tc0, p1+tc0)
00713 static inline vec_u8 h264_deblock_q1(register vec_u8 p0,
00714                                        register vec_u8 p1,
00715                                        register vec_u8 p2,
00716                                        register vec_u8 q0,
00717                                        register vec_u8 tc0) {
00718 
00719     register vec_u8 average = vec_avg(p0, q0);
00720     register vec_u8 temp;
00721     register vec_u8 uncliped;
00722     register vec_u8 ones;
00723     register vec_u8 max;
00724     register vec_u8 min;
00725     register vec_u8 newp1;
00726 
00727     temp = vec_xor(average, p2);
00728     average = vec_avg(average, p2);     /*avg(p2, avg(p0, q0)) */
00729     ones = vec_splat_u8(1);
00730     temp = vec_and(temp, ones);         /*(p2^avg(p0, q0)) & 1 */
00731     uncliped = vec_subs(average, temp); /*(p2+((p0+q0+1)>>1))>>1 */
00732     max = vec_adds(p1, tc0);
00733     min = vec_subs(p1, tc0);
00734     newp1 = vec_max(min, uncliped);
00735     newp1 = vec_min(max, newp1);
00736     return newp1;
00737 }
00738 
00739 #define h264_deblock_p0_q0(p0, p1, q0, q1, tc0masked) {                                           \
00740                                                                                                   \
00741     const vec_u8 A0v = vec_sl(vec_splat_u8(10), vec_splat_u8(4));                               \
00742                                                                                                   \
00743     register vec_u8 pq0bit = vec_xor(p0,q0);                                                    \
00744     register vec_u8 q1minus;                                                                    \
00745     register vec_u8 p0minus;                                                                    \
00746     register vec_u8 stage1;                                                                     \
00747     register vec_u8 stage2;                                                                     \
00748     register vec_u8 vec160;                                                                     \
00749     register vec_u8 delta;                                                                      \
00750     register vec_u8 deltaneg;                                                                   \
00751                                                                                                   \
00752     q1minus = vec_nor(q1, q1);                 /* 255 - q1 */                                     \
00753     stage1 = vec_avg(p1, q1minus);             /* (p1 - q1 + 256)>>1 */                           \
00754     stage2 = vec_sr(stage1, vec_splat_u8(1));  /* (p1 - q1 + 256)>>2 = 64 + (p1 - q1) >> 2 */     \
00755     p0minus = vec_nor(p0, p0);                 /* 255 - p0 */                                     \
00756     stage1 = vec_avg(q0, p0minus);             /* (q0 - p0 + 256)>>1 */                           \
00757     pq0bit = vec_and(pq0bit, vec_splat_u8(1));                                                    \
00758     stage2 = vec_avg(stage2, pq0bit);          /* 32 + ((q0 - p0)&1 + (p1 - q1) >> 2 + 1) >> 1 */ \
00759     stage2 = vec_adds(stage2, stage1);         /* 160 + ((p0 - q0) + (p1 - q1) >> 2 + 1) >> 1 */  \
00760     vec160 = vec_ld(0, &A0v);                                                                     \
00761     deltaneg = vec_subs(vec160, stage2);       /* -d */                                           \
00762     delta = vec_subs(stage2, vec160);          /* d */                                            \
00763     deltaneg = vec_min(tc0masked, deltaneg);                                                      \
00764     delta = vec_min(tc0masked, delta);                                                            \
00765     p0 = vec_subs(p0, deltaneg);                                                                  \
00766     q0 = vec_subs(q0, delta);                                                                     \
00767     p0 = vec_adds(p0, delta);                                                                     \
00768     q0 = vec_adds(q0, deltaneg);                                                                  \
00769 }
00770 
00771 #define h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0) {            \
00772     DECLARE_ALIGNED(16, unsigned char, temp)[16];                                             \
00773     register vec_u8 alphavec;                                                              \
00774     register vec_u8 betavec;                                                               \
00775     register vec_u8 mask;                                                                  \
00776     register vec_u8 p1mask;                                                                \
00777     register vec_u8 q1mask;                                                                \
00778     register vector signed   char tc0vec;                                                    \
00779     register vec_u8 finaltc0;                                                              \
00780     register vec_u8 tc0masked;                                                             \
00781     register vec_u8 newp1;                                                                 \
00782     register vec_u8 newq1;                                                                 \
00783                                                                                              \
00784     temp[0] = alpha;                                                                         \
00785     temp[1] = beta;                                                                          \
00786     alphavec = vec_ld(0, temp);                                                              \
00787     betavec = vec_splat(alphavec, 0x1);                                                      \
00788     alphavec = vec_splat(alphavec, 0x0);                                                     \
00789     mask = h264_deblock_mask(p0, p1, q0, q1, alphavec, betavec); /*if in block */            \
00790                                                                                              \
00791     *((int *)temp) = *((int *)tc0);                                                          \
00792     tc0vec = vec_ld(0, (signed char*)temp);                                                  \
00793     tc0vec = vec_mergeh(tc0vec, tc0vec);                                                     \
00794     tc0vec = vec_mergeh(tc0vec, tc0vec);                                                     \
00795     mask = vec_and(mask, vec_cmpgt(tc0vec, vec_splat_s8(-1)));  /* if tc0[i] >= 0 */         \
00796     finaltc0 = vec_and((vec_u8)tc0vec, mask);     /* tc = tc0 */                           \
00797                                                                                              \
00798     p1mask = diff_lt_altivec(p2, p0, betavec);                                               \
00799     p1mask = vec_and(p1mask, mask);                             /* if ( |p2 - p0| < beta) */ \
00800     tc0masked = vec_and(p1mask, (vec_u8)tc0vec);                                           \
00801     finaltc0 = vec_sub(finaltc0, p1mask);                       /* tc++ */                   \
00802     newp1 = h264_deblock_q1(p0, p1, p2, q0, tc0masked);                                      \
00803     /*end if*/                                                                               \
00804                                                                                              \
00805     q1mask = diff_lt_altivec(q2, q0, betavec);                                               \
00806     q1mask = vec_and(q1mask, mask);                             /* if ( |q2 - q0| < beta ) */\
00807     tc0masked = vec_and(q1mask, (vec_u8)tc0vec);                                           \
00808     finaltc0 = vec_sub(finaltc0, q1mask);                       /* tc++ */                   \
00809     newq1 = h264_deblock_q1(p0, q1, q2, q0, tc0masked);                                      \
00810     /*end if*/                                                                               \
00811                                                                                              \
00812     h264_deblock_p0_q0(p0, p1, q0, q1, finaltc0);                                            \
00813     p1 = newp1;                                                                              \
00814     q1 = newq1;                                                                              \
00815 }
00816 
00817 static void h264_v_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) {
00818 
00819     if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) >= 0) {
00820         register vec_u8 p2 = vec_ld(-3*stride, pix);
00821         register vec_u8 p1 = vec_ld(-2*stride, pix);
00822         register vec_u8 p0 = vec_ld(-1*stride, pix);
00823         register vec_u8 q0 = vec_ld(0, pix);
00824         register vec_u8 q1 = vec_ld(stride, pix);
00825         register vec_u8 q2 = vec_ld(2*stride, pix);
00826         h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0);
00827         vec_st(p1, -2*stride, pix);
00828         vec_st(p0, -1*stride, pix);
00829         vec_st(q0, 0, pix);
00830         vec_st(q1, stride, pix);
00831     }
00832 }
00833 
00834 static void h264_h_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) {
00835 
00836     register vec_u8 line0, line1, line2, line3, line4, line5;
00837     if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) < 0)
00838         return;
00839     readAndTranspose16x6(pix-3, stride, line0, line1, line2, line3, line4, line5);
00840     h264_loop_filter_luma_altivec(line0, line1, line2, line3, line4, line5, alpha, beta, tc0);
00841     transpose4x16(line1, line2, line3, line4);
00842     write16x4(pix-2, stride, line1, line2, line3, line4);
00843 }
00844 
00845 static av_always_inline
00846 void weight_h264_W_altivec(uint8_t *block, int stride, int height,
00847                            int log2_denom, int weight, int offset, int w)
00848 {
00849     int y, aligned;
00850     vec_u8 vblock;
00851     vec_s16 vtemp, vweight, voffset, v0, v1;
00852     vec_u16 vlog2_denom;
00853     DECLARE_ALIGNED(16, int32_t, temp)[4];
00854     LOAD_ZERO;
00855 
00856     offset <<= log2_denom;
00857     if(log2_denom) offset += 1<<(log2_denom-1);
00858     temp[0] = log2_denom;
00859     temp[1] = weight;
00860     temp[2] = offset;
00861 
00862     vtemp = (vec_s16)vec_ld(0, temp);
00863     vlog2_denom = (vec_u16)vec_splat(vtemp, 1);
00864     vweight = vec_splat(vtemp, 3);
00865     voffset = vec_splat(vtemp, 5);
00866     aligned = !((unsigned long)block & 0xf);
00867 
00868     for (y = 0; y < height; y++) {
00869         vblock = vec_ld(0, block);
00870 
00871         v0 = (vec_s16)vec_mergeh(zero_u8v, vblock);
00872         v1 = (vec_s16)vec_mergel(zero_u8v, vblock);
00873 
00874         if (w == 16 || aligned) {
00875             v0 = vec_mladd(v0, vweight, zero_s16v);
00876             v0 = vec_adds(v0, voffset);
00877             v0 = vec_sra(v0, vlog2_denom);
00878         }
00879         if (w == 16 || !aligned) {
00880             v1 = vec_mladd(v1, vweight, zero_s16v);
00881             v1 = vec_adds(v1, voffset);
00882             v1 = vec_sra(v1, vlog2_denom);
00883         }
00884         vblock = vec_packsu(v0, v1);
00885         vec_st(vblock, 0, block);
00886 
00887         block += stride;
00888     }
00889 }
00890 
00891 static av_always_inline
00892 void biweight_h264_W_altivec(uint8_t *dst, uint8_t *src, int stride, int height,
00893                              int log2_denom, int weightd, int weights, int offset, int w)
00894 {
00895     int y, dst_aligned, src_aligned;
00896     vec_u8 vsrc, vdst;
00897     vec_s16 vtemp, vweights, vweightd, voffset, v0, v1, v2, v3;
00898     vec_u16 vlog2_denom;
00899     DECLARE_ALIGNED(16, int32_t, temp)[4];
00900     LOAD_ZERO;
00901 
00902     offset = ((offset + 1) | 1) << log2_denom;
00903     temp[0] = log2_denom+1;
00904     temp[1] = weights;
00905     temp[2] = weightd;
00906     temp[3] = offset;
00907 
00908     vtemp = (vec_s16)vec_ld(0, temp);
00909     vlog2_denom = (vec_u16)vec_splat(vtemp, 1);
00910     vweights = vec_splat(vtemp, 3);
00911     vweightd = vec_splat(vtemp, 5);
00912     voffset = vec_splat(vtemp, 7);
00913     dst_aligned = !((unsigned long)dst & 0xf);
00914     src_aligned = !((unsigned long)src & 0xf);
00915 
00916     for (y = 0; y < height; y++) {
00917         vdst = vec_ld(0, dst);
00918         vsrc = vec_ld(0, src);
00919 
00920         v0 = (vec_s16)vec_mergeh(zero_u8v, vdst);
00921         v1 = (vec_s16)vec_mergel(zero_u8v, vdst);
00922         v2 = (vec_s16)vec_mergeh(zero_u8v, vsrc);
00923         v3 = (vec_s16)vec_mergel(zero_u8v, vsrc);
00924 
00925         if (w == 8) {
00926             if (src_aligned)
00927                 v3 = v2;
00928             else
00929                 v2 = v3;
00930         }
00931 
00932         if (w == 16 || dst_aligned) {
00933             v0 = vec_mladd(v0, vweightd, zero_s16v);
00934             v2 = vec_mladd(v2, vweights, zero_s16v);
00935 
00936             v0 = vec_adds(v0, voffset);
00937             v0 = vec_adds(v0, v2);
00938             v0 = vec_sra(v0, vlog2_denom);
00939         }
00940         if (w == 16 || !dst_aligned) {
00941             v1 = vec_mladd(v1, vweightd, zero_s16v);
00942             v3 = vec_mladd(v3, vweights, zero_s16v);
00943 
00944             v1 = vec_adds(v1, voffset);
00945             v1 = vec_adds(v1, v3);
00946             v1 = vec_sra(v1, vlog2_denom);
00947         }
00948         vdst = vec_packsu(v0, v1);
00949         vec_st(vdst, 0, dst);
00950 
00951         dst += stride;
00952         src += stride;
00953     }
00954 }
00955 
00956 #define H264_WEIGHT(W) \
00957 static void ff_weight_h264_pixels ## W ## _altivec(uint8_t *block, int stride, int height, \
00958                                                    int log2_denom, int weight, int offset){ \
00959     weight_h264_W_altivec(block, stride, height, log2_denom, weight, offset, W); \
00960 }\
00961 static void ff_biweight_h264_pixels ## W ## _altivec(uint8_t *dst, uint8_t *src, int stride, int height, \
00962                                                      int log2_denom, int weightd, int weights, int offset){ \
00963     biweight_h264_W_altivec(dst, src, stride, height, log2_denom, weightd, weights, offset, W); \
00964 }
00965 
00966 H264_WEIGHT(16)
00967 H264_WEIGHT( 8)
00968 
00969 void dsputil_h264_init_ppc(DSPContext* c, AVCodecContext *avctx) {
00970     const int high_bit_depth = avctx->bits_per_raw_sample > 8;
00971 
00972     if (av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC) {
00973     if (!high_bit_depth) {
00974         c->put_h264_chroma_pixels_tab[0] = put_h264_chroma_mc8_altivec;
00975         c->avg_h264_chroma_pixels_tab[0] = avg_h264_chroma_mc8_altivec;
00976 
00977 #define dspfunc(PFX, IDX, NUM) \
00978         c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_altivec; \
00979         c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_altivec; \
00980         c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_altivec; \
00981         c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_altivec; \
00982         c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_altivec; \
00983         c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_altivec; \
00984         c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_altivec; \
00985         c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_altivec; \
00986         c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_altivec; \
00987         c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_altivec; \
00988         c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_altivec; \
00989         c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_altivec; \
00990         c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_altivec; \
00991         c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_altivec; \
00992         c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_altivec; \
00993         c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_altivec
00994 
00995         dspfunc(put_h264_qpel, 0, 16);
00996         dspfunc(avg_h264_qpel, 0, 16);
00997 #undef dspfunc
00998     }
00999     }
01000 }
01001 
01002 void ff_h264dsp_init_ppc(H264DSPContext *c, const int bit_depth, const int chroma_format_idc)
01003 {
01004     if (av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC) {
01005     if (bit_depth == 8) {
01006         c->h264_idct_add = ff_h264_idct_add_altivec;
01007         if (chroma_format_idc == 1)
01008             c->h264_idct_add8 = ff_h264_idct_add8_altivec;
01009         c->h264_idct_add16 = ff_h264_idct_add16_altivec;
01010         c->h264_idct_add16intra = ff_h264_idct_add16intra_altivec;
01011         c->h264_idct_dc_add= h264_idct_dc_add_altivec;
01012         c->h264_idct8_dc_add = ff_h264_idct8_dc_add_altivec;
01013         c->h264_idct8_add = ff_h264_idct8_add_altivec;
01014         c->h264_idct8_add4 = ff_h264_idct8_add4_altivec;
01015         c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_altivec;
01016         c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_altivec;
01017 
01018         c->weight_h264_pixels_tab[0] = ff_weight_h264_pixels16_altivec;
01019         c->weight_h264_pixels_tab[1] = ff_weight_h264_pixels8_altivec;
01020         c->biweight_h264_pixels_tab[0] = ff_biweight_h264_pixels16_altivec;
01021         c->biweight_h264_pixels_tab[1] = ff_biweight_h264_pixels8_altivec;
01022     }
01023     }
01024 }