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 "libavcodec/dsputil.h"
00022 #include "libavcodec/h264data.h"
00023 #include "libavcodec/h264dsp.h"
00024 
00025 #include "dsputil_ppc.h"
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_no_rnd_vc1_chroma_mc8_altivec   put_no_rnd_vc1_chroma_mc8_altivec
00036 #define PREFIX_h264_chroma_mc8_num             altivec_put_h264_chroma_mc8_num
00037 #define PREFIX_h264_qpel16_h_lowpass_altivec   put_h264_qpel16_h_lowpass_altivec
00038 #define PREFIX_h264_qpel16_h_lowpass_num       altivec_put_h264_qpel16_h_lowpass_num
00039 #define PREFIX_h264_qpel16_v_lowpass_altivec   put_h264_qpel16_v_lowpass_altivec
00040 #define PREFIX_h264_qpel16_v_lowpass_num       altivec_put_h264_qpel16_v_lowpass_num
00041 #define PREFIX_h264_qpel16_hv_lowpass_altivec  put_h264_qpel16_hv_lowpass_altivec
00042 #define PREFIX_h264_qpel16_hv_lowpass_num      altivec_put_h264_qpel16_hv_lowpass_num
00043 #include "h264_template_altivec.c"
00044 #undef OP_U8_ALTIVEC
00045 #undef PREFIX_h264_chroma_mc8_altivec
00046 #undef PREFIX_no_rnd_vc1_chroma_mc8_altivec
00047 #undef PREFIX_h264_chroma_mc8_num
00048 #undef PREFIX_h264_qpel16_h_lowpass_altivec
00049 #undef PREFIX_h264_qpel16_h_lowpass_num
00050 #undef PREFIX_h264_qpel16_v_lowpass_altivec
00051 #undef PREFIX_h264_qpel16_v_lowpass_num
00052 #undef PREFIX_h264_qpel16_hv_lowpass_altivec
00053 #undef PREFIX_h264_qpel16_hv_lowpass_num
00054 
00055 #define OP_U8_ALTIVEC                          AVG_OP_U8_ALTIVEC
00056 #define PREFIX_h264_chroma_mc8_altivec         avg_h264_chroma_mc8_altivec
00057 #define PREFIX_no_rnd_vc1_chroma_mc8_altivec   avg_no_rnd_vc1_chroma_mc8_altivec
00058 #define PREFIX_h264_chroma_mc8_num             altivec_avg_h264_chroma_mc8_num
00059 #define PREFIX_h264_qpel16_h_lowpass_altivec   avg_h264_qpel16_h_lowpass_altivec
00060 #define PREFIX_h264_qpel16_h_lowpass_num       altivec_avg_h264_qpel16_h_lowpass_num
00061 #define PREFIX_h264_qpel16_v_lowpass_altivec   avg_h264_qpel16_v_lowpass_altivec
00062 #define PREFIX_h264_qpel16_v_lowpass_num       altivec_avg_h264_qpel16_v_lowpass_num
00063 #define PREFIX_h264_qpel16_hv_lowpass_altivec  avg_h264_qpel16_hv_lowpass_altivec
00064 #define PREFIX_h264_qpel16_hv_lowpass_num      altivec_avg_h264_qpel16_hv_lowpass_num
00065 #include "h264_template_altivec.c"
00066 #undef OP_U8_ALTIVEC
00067 #undef PREFIX_h264_chroma_mc8_altivec
00068 #undef PREFIX_no_rnd_vc1_chroma_mc8_altivec
00069 #undef PREFIX_h264_chroma_mc8_num
00070 #undef PREFIX_h264_qpel16_h_lowpass_altivec
00071 #undef PREFIX_h264_qpel16_h_lowpass_num
00072 #undef PREFIX_h264_qpel16_v_lowpass_altivec
00073 #undef PREFIX_h264_qpel16_v_lowpass_num
00074 #undef PREFIX_h264_qpel16_hv_lowpass_altivec
00075 #undef PREFIX_h264_qpel16_hv_lowpass_num
00076 
00077 #define H264_MC(OPNAME, SIZE, CODETYPE) \
00078 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## CODETYPE (uint8_t *dst, uint8_t *src, int stride){\
00079     OPNAME ## pixels ## SIZE ## _ ## CODETYPE(dst, src, stride, SIZE);\
00080 }\
00081 \
00082 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){ \
00083     DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
00084     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
00085     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\
00086 }\
00087 \
00088 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00089     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(dst, src, stride, stride);\
00090 }\
00091 \
00092 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00093     DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
00094     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
00095     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+1, half, stride, stride, SIZE);\
00096 }\
00097 \
00098 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00099     DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
00100     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
00101     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\
00102 }\
00103 \
00104 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00105     OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(dst, src, stride, stride);\
00106 }\
00107 \
00108 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00109     DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
00110     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
00111     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+stride, half, stride, stride, SIZE);\
00112 }\
00113 \
00114 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00115     DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
00116     DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
00117     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
00118     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
00119     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
00120 }\
00121 \
00122 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00123     DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
00124     DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
00125     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
00126     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
00127     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
00128 }\
00129 \
00130 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00131     DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
00132     DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
00133     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
00134     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
00135     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
00136 }\
00137 \
00138 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00139     DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
00140     DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
00141     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
00142     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
00143     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
00144 }\
00145 \
00146 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00147     DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
00148     OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(dst, tmp, src, stride, SIZE, stride);\
00149 }\
00150 \
00151 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00152     DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
00153     DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
00154     DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
00155     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
00156     put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
00157     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\
00158 }\
00159 \
00160 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00161     DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
00162     DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
00163     DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
00164     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
00165     put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
00166     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\
00167 }\
00168 \
00169 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00170     DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
00171     DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
00172     DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
00173     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
00174     put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
00175     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
00176 }\
00177 \
00178 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
00179     DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
00180     DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
00181     DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
00182     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
00183     put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
00184     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
00185 }\
00186 
00187 static inline void put_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
00188                                     const uint8_t * src2, int dst_stride,
00189                                     int src_stride1, int h)
00190 {
00191     int i;
00192     vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align;
00193 
00194     mask_ = vec_lvsl(0, src2);
00195 
00196     for (i = 0; i < h; i++) {
00197 
00198         tmp1 = vec_ld(i * src_stride1, src1);
00199         mask = vec_lvsl(i * src_stride1, src1);
00200         tmp2 = vec_ld(i * src_stride1 + 15, src1);
00201 
00202         a = vec_perm(tmp1, tmp2, mask);
00203 
00204         tmp1 = vec_ld(i * 16, src2);
00205         tmp2 = vec_ld(i * 16 + 15, src2);
00206 
00207         b = vec_perm(tmp1, tmp2, mask_);
00208 
00209         tmp1 = vec_ld(0, dst);
00210         mask = vec_lvsl(0, dst);
00211         tmp2 = vec_ld(15, dst);
00212 
00213         d = vec_avg(a, b);
00214 
00215         edges = vec_perm(tmp2, tmp1, mask);
00216 
00217         align = vec_lvsr(0, dst);
00218 
00219         tmp2 = vec_perm(d, edges, align);
00220         tmp1 = vec_perm(edges, d, align);
00221 
00222         vec_st(tmp2, 15, dst);
00223         vec_st(tmp1, 0 , dst);
00224 
00225         dst += dst_stride;
00226     }
00227 }
00228 
00229 static inline void avg_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
00230                                     const uint8_t * src2, int dst_stride,
00231                                     int src_stride1, int h)
00232 {
00233     int i;
00234     vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align;
00235 
00236     mask_ = vec_lvsl(0, src2);
00237 
00238     for (i = 0; i < h; i++) {
00239 
00240         tmp1 = vec_ld(i * src_stride1, src1);
00241         mask = vec_lvsl(i * src_stride1, src1);
00242         tmp2 = vec_ld(i * src_stride1 + 15, src1);
00243 
00244         a = vec_perm(tmp1, tmp2, mask);
00245 
00246         tmp1 = vec_ld(i * 16, src2);
00247         tmp2 = vec_ld(i * 16 + 15, src2);
00248 
00249         b = vec_perm(tmp1, tmp2, mask_);
00250 
00251         tmp1 = vec_ld(0, dst);
00252         mask = vec_lvsl(0, dst);
00253         tmp2 = vec_ld(15, dst);
00254 
00255         d = vec_avg(vec_perm(tmp1, tmp2, mask), vec_avg(a, b));
00256 
00257         edges = vec_perm(tmp2, tmp1, mask);
00258 
00259         align = vec_lvsr(0, dst);
00260 
00261         tmp2 = vec_perm(d, edges, align);
00262         tmp1 = vec_perm(edges, d, align);
00263 
00264         vec_st(tmp2, 15, dst);
00265         vec_st(tmp1, 0 , dst);
00266 
00267         dst += dst_stride;
00268     }
00269 }
00270 
00271 /* Implemented but could be faster
00272 #define put_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) put_pixels16_l2(d,s1,s2,ds,s1s,16,h)
00273 #define avg_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) avg_pixels16_l2(d,s1,s2,ds,s1s,16,h)
00274  */
00275 
00276 H264_MC(put_, 16, altivec)
00277 H264_MC(avg_, 16, altivec)
00278 
00279 
00280 /****************************************************************************
00281  * IDCT transform:
00282  ****************************************************************************/
00283 
00284 #define VEC_1D_DCT(vb0,vb1,vb2,vb3,va0,va1,va2,va3)               \
00285     /* 1st stage */                                               \
00286     vz0 = vec_add(vb0,vb2);       /* temp[0] = Y[0] + Y[2] */     \
00287     vz1 = vec_sub(vb0,vb2);       /* temp[1] = Y[0] - Y[2] */     \
00288     vz2 = vec_sra(vb1,vec_splat_u16(1));                          \
00289     vz2 = vec_sub(vz2,vb3);       /* temp[2] = Y[1].1/2 - Y[3] */ \
00290     vz3 = vec_sra(vb3,vec_splat_u16(1));                          \
00291     vz3 = vec_add(vb1,vz3);       /* temp[3] = Y[1] + Y[3].1/2 */ \
00292     /* 2nd stage: output */                                       \
00293     va0 = vec_add(vz0,vz3);       /* x[0] = temp[0] + temp[3] */  \
00294     va1 = vec_add(vz1,vz2);       /* x[1] = temp[1] + temp[2] */  \
00295     va2 = vec_sub(vz1,vz2);       /* x[2] = temp[1] - temp[2] */  \
00296     va3 = vec_sub(vz0,vz3)        /* x[3] = temp[0] - temp[3] */
00297 
00298 #define VEC_TRANSPOSE_4(a0,a1,a2,a3,b0,b1,b2,b3) \
00299     b0 = vec_mergeh( a0, a0 ); \
00300     b1 = vec_mergeh( a1, a0 ); \
00301     b2 = vec_mergeh( a2, a0 ); \
00302     b3 = vec_mergeh( a3, a0 ); \
00303     a0 = vec_mergeh( b0, b2 ); \
00304     a1 = vec_mergel( b0, b2 ); \
00305     a2 = vec_mergeh( b1, b3 ); \
00306     a3 = vec_mergel( b1, b3 ); \
00307     b0 = vec_mergeh( a0, a2 ); \
00308     b1 = vec_mergel( a0, a2 ); \
00309     b2 = vec_mergeh( a1, a3 ); \
00310     b3 = vec_mergel( a1, a3 )
00311 
00312 #define VEC_LOAD_U8_ADD_S16_STORE_U8(va)                      \
00313     vdst_orig = vec_ld(0, dst);                               \
00314     vdst = vec_perm(vdst_orig, zero_u8v, vdst_mask);          \
00315     vdst_ss = (vec_s16) vec_mergeh(zero_u8v, vdst);         \
00316     va = vec_add(va, vdst_ss);                                \
00317     va_u8 = vec_packsu(va, zero_s16v);                        \
00318     va_u32 = vec_splat((vec_u32)va_u8, 0);                  \
00319     vec_ste(va_u32, element, (uint32_t*)dst);
00320 
00321 static void ff_h264_idct_add_altivec(uint8_t *dst, DCTELEM *block, int stride)
00322 {
00323     vec_s16 va0, va1, va2, va3;
00324     vec_s16 vz0, vz1, vz2, vz3;
00325     vec_s16 vtmp0, vtmp1, vtmp2, vtmp3;
00326     vec_u8 va_u8;
00327     vec_u32 va_u32;
00328     vec_s16 vdst_ss;
00329     const vec_u16 v6us = vec_splat_u16(6);
00330     vec_u8 vdst, vdst_orig;
00331     vec_u8 vdst_mask = vec_lvsl(0, dst);
00332     int element = ((unsigned long)dst & 0xf) >> 2;
00333     LOAD_ZERO;
00334 
00335     block[0] += 32;  /* add 32 as a DC-level for rounding */
00336 
00337     vtmp0 = vec_ld(0,block);
00338     vtmp1 = vec_sld(vtmp0, vtmp0, 8);
00339     vtmp2 = vec_ld(16,block);
00340     vtmp3 = vec_sld(vtmp2, vtmp2, 8);
00341 
00342     VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3);
00343     VEC_TRANSPOSE_4(va0,va1,va2,va3,vtmp0,vtmp1,vtmp2,vtmp3);
00344     VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3);
00345 
00346     va0 = vec_sra(va0,v6us);
00347     va1 = vec_sra(va1,v6us);
00348     va2 = vec_sra(va2,v6us);
00349     va3 = vec_sra(va3,v6us);
00350 
00351     VEC_LOAD_U8_ADD_S16_STORE_U8(va0);
00352     dst += stride;
00353     VEC_LOAD_U8_ADD_S16_STORE_U8(va1);
00354     dst += stride;
00355     VEC_LOAD_U8_ADD_S16_STORE_U8(va2);
00356     dst += stride;
00357     VEC_LOAD_U8_ADD_S16_STORE_U8(va3);
00358 }
00359 
00360 #define IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7,  d0, d1, d2, d3, d4, d5, d6, d7) {\
00361     /*        a0  = SRC(0) + SRC(4); */ \
00362     vec_s16 a0v = vec_add(s0, s4);    \
00363     /*        a2  = SRC(0) - SRC(4); */ \
00364     vec_s16 a2v = vec_sub(s0, s4);    \
00365     /*        a4  =           (SRC(2)>>1) - SRC(6); */ \
00366     vec_s16 a4v = vec_sub(vec_sra(s2, onev), s6);    \
00367     /*        a6  =           (SRC(6)>>1) + SRC(2); */ \
00368     vec_s16 a6v = vec_add(vec_sra(s6, onev), s2);    \
00369     /*        b0  =         a0 + a6; */ \
00370     vec_s16 b0v = vec_add(a0v, a6v);  \
00371     /*        b2  =         a2 + a4; */ \
00372     vec_s16 b2v = vec_add(a2v, a4v);  \
00373     /*        b4  =         a2 - a4; */ \
00374     vec_s16 b4v = vec_sub(a2v, a4v);  \
00375     /*        b6  =         a0 - a6; */ \
00376     vec_s16 b6v = vec_sub(a0v, a6v);  \
00377     /* a1 =  SRC(5) - SRC(3) - SRC(7) - (SRC(7)>>1); */ \
00378     /*        a1 =             (SRC(5)-SRC(3)) -  (SRC(7)  +  (SRC(7)>>1)); */ \
00379     vec_s16 a1v = vec_sub( vec_sub(s5, s3), vec_add(s7, vec_sra(s7, onev)) ); \
00380     /* a3 =  SRC(7) + SRC(1) - SRC(3) - (SRC(3)>>1); */ \
00381     /*        a3 =             (SRC(7)+SRC(1)) -  (SRC(3)  +  (SRC(3)>>1)); */ \
00382     vec_s16 a3v = vec_sub( vec_add(s7, s1), vec_add(s3, vec_sra(s3, onev)) );\
00383     /* a5 =  SRC(7) - SRC(1) + SRC(5) + (SRC(5)>>1); */ \
00384     /*        a5 =             (SRC(7)-SRC(1)) +   SRC(5) +   (SRC(5)>>1); */ \
00385     vec_s16 a5v = vec_add( vec_sub(s7, s1), vec_add(s5, vec_sra(s5, onev)) );\
00386     /*        a7 =                SRC(5)+SRC(3) +  SRC(1) +   (SRC(1)>>1); */ \
00387     vec_s16 a7v = vec_add( vec_add(s5, s3), vec_add(s1, vec_sra(s1, onev)) );\
00388     /*        b1 =                  (a7>>2)  +  a1; */ \
00389     vec_s16 b1v = vec_add( vec_sra(a7v, twov), a1v); \
00390     /*        b3 =          a3 +        (a5>>2); */ \
00391     vec_s16 b3v = vec_add(a3v, vec_sra(a5v, twov)); \
00392     /*        b5 =                  (a3>>2)  -   a5; */ \
00393     vec_s16 b5v = vec_sub( vec_sra(a3v, twov), a5v); \
00394     /*        b7 =           a7 -        (a1>>2); */ \
00395     vec_s16 b7v = vec_sub( a7v, vec_sra(a1v, twov)); \
00396     /* DST(0,    b0 + b7); */ \
00397     d0 = vec_add(b0v, b7v); \
00398     /* DST(1,    b2 + b5); */ \
00399     d1 = vec_add(b2v, b5v); \
00400     /* DST(2,    b4 + b3); */ \
00401     d2 = vec_add(b4v, b3v); \
00402     /* DST(3,    b6 + b1); */ \
00403     d3 = vec_add(b6v, b1v); \
00404     /* DST(4,    b6 - b1); */ \
00405     d4 = vec_sub(b6v, b1v); \
00406     /* DST(5,    b4 - b3); */ \
00407     d5 = vec_sub(b4v, b3v); \
00408     /* DST(6,    b2 - b5); */ \
00409     d6 = vec_sub(b2v, b5v); \
00410     /* DST(7,    b0 - b7); */ \
00411     d7 = vec_sub(b0v, b7v); \
00412 }
00413 
00414 #define ALTIVEC_STORE_SUM_CLIP(dest, idctv, perm_ldv, perm_stv, sel) { \
00415     /* unaligned load */                                       \
00416     vec_u8 hv = vec_ld( 0, dest );                           \
00417     vec_u8 lv = vec_ld( 7, dest );                           \
00418     vec_u8 dstv   = vec_perm( hv, lv, (vec_u8)perm_ldv );  \
00419     vec_s16 idct_sh6 = vec_sra(idctv, sixv);                 \
00420     vec_u16 dst16 = (vec_u16)vec_mergeh(zero_u8v, dstv);   \
00421     vec_s16 idstsum = vec_adds(idct_sh6, (vec_s16)dst16);  \
00422     vec_u8 idstsum8 = vec_packsu(zero_s16v, idstsum);        \
00423     vec_u8 edgehv;                                           \
00424     /* unaligned store */                                      \
00425     vec_u8 bodyv  = vec_perm( idstsum8, idstsum8, perm_stv );\
00426     vec_u8 edgelv = vec_perm( sel, zero_u8v, perm_stv );     \
00427     lv    = vec_sel( lv, bodyv, edgelv );                      \
00428     vec_st( lv, 7, dest );                                     \
00429     hv    = vec_ld( 0, dest );                                 \
00430     edgehv = vec_perm( zero_u8v, sel, perm_stv );              \
00431     hv    = vec_sel( hv, bodyv, edgehv );                      \
00432     vec_st( hv, 0, dest );                                     \
00433  }
00434 
00435 static void ff_h264_idct8_add_altivec( uint8_t *dst, DCTELEM *dct, int stride ) {
00436     vec_s16 s0, s1, s2, s3, s4, s5, s6, s7;
00437     vec_s16 d0, d1, d2, d3, d4, d5, d6, d7;
00438     vec_s16 idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7;
00439 
00440     vec_u8 perm_ldv = vec_lvsl(0, dst);
00441     vec_u8 perm_stv = vec_lvsr(8, dst);
00442 
00443     const vec_u16 onev = vec_splat_u16(1);
00444     const vec_u16 twov = vec_splat_u16(2);
00445     const vec_u16 sixv = vec_splat_u16(6);
00446 
00447     const vec_u8 sel = (vec_u8) {0,0,0,0,0,0,0,0,-1,-1,-1,-1,-1,-1,-1,-1};
00448     LOAD_ZERO;
00449 
00450     dct[0] += 32; // rounding for the >>6 at the end
00451 
00452     s0 = vec_ld(0x00, (int16_t*)dct);
00453     s1 = vec_ld(0x10, (int16_t*)dct);
00454     s2 = vec_ld(0x20, (int16_t*)dct);
00455     s3 = vec_ld(0x30, (int16_t*)dct);
00456     s4 = vec_ld(0x40, (int16_t*)dct);
00457     s5 = vec_ld(0x50, (int16_t*)dct);
00458     s6 = vec_ld(0x60, (int16_t*)dct);
00459     s7 = vec_ld(0x70, (int16_t*)dct);
00460 
00461     IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7,
00462                      d0, d1, d2, d3, d4, d5, d6, d7);
00463 
00464     TRANSPOSE8( d0,  d1,  d2,  d3,  d4,  d5,  d6, d7 );
00465 
00466     IDCT8_1D_ALTIVEC(d0,  d1,  d2,  d3,  d4,  d5,  d6, d7,
00467                      idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7);
00468 
00469     ALTIVEC_STORE_SUM_CLIP(&dst[0*stride], idct0, perm_ldv, perm_stv, sel);
00470     ALTIVEC_STORE_SUM_CLIP(&dst[1*stride], idct1, perm_ldv, perm_stv, sel);
00471     ALTIVEC_STORE_SUM_CLIP(&dst[2*stride], idct2, perm_ldv, perm_stv, sel);
00472     ALTIVEC_STORE_SUM_CLIP(&dst[3*stride], idct3, perm_ldv, perm_stv, sel);
00473     ALTIVEC_STORE_SUM_CLIP(&dst[4*stride], idct4, perm_ldv, perm_stv, sel);
00474     ALTIVEC_STORE_SUM_CLIP(&dst[5*stride], idct5, perm_ldv, perm_stv, sel);
00475     ALTIVEC_STORE_SUM_CLIP(&dst[6*stride], idct6, perm_ldv, perm_stv, sel);
00476     ALTIVEC_STORE_SUM_CLIP(&dst[7*stride], idct7, perm_ldv, perm_stv, sel);
00477 }
00478 
00479 static av_always_inline void h264_idct_dc_add_internal(uint8_t *dst, DCTELEM *block, int stride, int size)
00480 {
00481     vec_s16 dc16;
00482     vec_u8 dcplus, dcminus, v0, v1, v2, v3, aligner;
00483     LOAD_ZERO;
00484     DECLARE_ALIGNED(16, int, dc);
00485     int i;
00486 
00487     dc = (block[0] + 32) >> 6;
00488     dc16 = vec_splat((vec_s16) vec_lde(0, &dc), 1);
00489 
00490     if (size == 4)
00491         dc16 = vec_sld(dc16, zero_s16v, 8);
00492     dcplus = vec_packsu(dc16, zero_s16v);
00493     dcminus = vec_packsu(vec_sub(zero_s16v, dc16), zero_s16v);
00494 
00495     aligner = vec_lvsr(0, dst);
00496     dcplus = vec_perm(dcplus, dcplus, aligner);
00497     dcminus = vec_perm(dcminus, dcminus, aligner);
00498 
00499     for (i = 0; i < size; i += 4) {
00500         v0 = vec_ld(0, dst+0*stride);
00501         v1 = vec_ld(0, dst+1*stride);
00502         v2 = vec_ld(0, dst+2*stride);
00503         v3 = vec_ld(0, dst+3*stride);
00504 
00505         v0 = vec_adds(v0, dcplus);
00506         v1 = vec_adds(v1, dcplus);
00507         v2 = vec_adds(v2, dcplus);
00508         v3 = vec_adds(v3, dcplus);
00509 
00510         v0 = vec_subs(v0, dcminus);
00511         v1 = vec_subs(v1, dcminus);
00512         v2 = vec_subs(v2, dcminus);
00513         v3 = vec_subs(v3, dcminus);
00514 
00515         vec_st(v0, 0, dst+0*stride);
00516         vec_st(v1, 0, dst+1*stride);
00517         vec_st(v2, 0, dst+2*stride);
00518         vec_st(v3, 0, dst+3*stride);
00519 
00520         dst += 4*stride;
00521     }
00522 }
00523 
00524 static void h264_idct_dc_add_altivec(uint8_t *dst, DCTELEM *block, int stride)
00525 {
00526     h264_idct_dc_add_internal(dst, block, stride, 4);
00527 }
00528 
00529 static void ff_h264_idct8_dc_add_altivec(uint8_t *dst, DCTELEM *block, int stride)
00530 {
00531     h264_idct_dc_add_internal(dst, block, stride, 8);
00532 }
00533 
00534 static void ff_h264_idct_add16_altivec(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
00535     int i;
00536     for(i=0; i<16; i++){
00537         int nnz = nnzc[ scan8[i] ];
00538         if(nnz){
00539             if(nnz==1 && block[i*16]) h264_idct_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
00540             else                      ff_h264_idct_add_altivec(dst + block_offset[i], block + i*16, stride);
00541         }
00542     }
00543 }
00544 
00545 static void ff_h264_idct_add16intra_altivec(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
00546     int i;
00547     for(i=0; i<16; i++){
00548         if(nnzc[ scan8[i] ]) ff_h264_idct_add_altivec(dst + block_offset[i], block + i*16, stride);
00549         else if(block[i*16]) h264_idct_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
00550     }
00551 }
00552 
00553 static void ff_h264_idct8_add4_altivec(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
00554     int i;
00555     for(i=0; i<16; i+=4){
00556         int nnz = nnzc[ scan8[i] ];
00557         if(nnz){
00558             if(nnz==1 && block[i*16]) ff_h264_idct8_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
00559             else                      ff_h264_idct8_add_altivec   (dst + block_offset[i], block + i*16, stride);
00560         }
00561     }
00562 }
00563 
00564 static void ff_h264_idct_add8_altivec(uint8_t **dest, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
00565     int i;
00566     for(i=16; i<16+8; i++){
00567         if(nnzc[ scan8[i] ])
00568             ff_h264_idct_add_altivec(dest[(i&4)>>2] + block_offset[i], block + i*16, stride);
00569         else if(block[i*16])
00570             h264_idct_dc_add_altivec(dest[(i&4)>>2] + block_offset[i], block + i*16, stride);
00571     }
00572 }
00573 
00574 #define transpose4x16(r0, r1, r2, r3) {      \
00575     register vec_u8 r4;                    \
00576     register vec_u8 r5;                    \
00577     register vec_u8 r6;                    \
00578     register vec_u8 r7;                    \
00579                                              \
00580     r4 = vec_mergeh(r0, r2);  /*0, 2 set 0*/ \
00581     r5 = vec_mergel(r0, r2);  /*0, 2 set 1*/ \
00582     r6 = vec_mergeh(r1, r3);  /*1, 3 set 0*/ \
00583     r7 = vec_mergel(r1, r3);  /*1, 3 set 1*/ \
00584                                              \
00585     r0 = vec_mergeh(r4, r6);  /*all set 0*/  \
00586     r1 = vec_mergel(r4, r6);  /*all set 1*/  \
00587     r2 = vec_mergeh(r5, r7);  /*all set 2*/  \
00588     r3 = vec_mergel(r5, r7);  /*all set 3*/  \
00589 }
00590 
00591 static inline void write16x4(uint8_t *dst, int dst_stride,
00592                              register vec_u8 r0, register vec_u8 r1,
00593                              register vec_u8 r2, register vec_u8 r3) {
00594     DECLARE_ALIGNED(16, unsigned char, result)[64];
00595     uint32_t *src_int = (uint32_t *)result, *dst_int = (uint32_t *)dst;
00596     int int_dst_stride = dst_stride/4;
00597 
00598     vec_st(r0, 0, result);
00599     vec_st(r1, 16, result);
00600     vec_st(r2, 32, result);
00601     vec_st(r3, 48, result);
00602     /* FIXME: there has to be a better way!!!! */
00603     *dst_int = *src_int;
00604     *(dst_int+   int_dst_stride) = *(src_int + 1);
00605     *(dst_int+ 2*int_dst_stride) = *(src_int + 2);
00606     *(dst_int+ 3*int_dst_stride) = *(src_int + 3);
00607     *(dst_int+ 4*int_dst_stride) = *(src_int + 4);
00608     *(dst_int+ 5*int_dst_stride) = *(src_int + 5);
00609     *(dst_int+ 6*int_dst_stride) = *(src_int + 6);
00610     *(dst_int+ 7*int_dst_stride) = *(src_int + 7);
00611     *(dst_int+ 8*int_dst_stride) = *(src_int + 8);
00612     *(dst_int+ 9*int_dst_stride) = *(src_int + 9);
00613     *(dst_int+10*int_dst_stride) = *(src_int + 10);
00614     *(dst_int+11*int_dst_stride) = *(src_int + 11);
00615     *(dst_int+12*int_dst_stride) = *(src_int + 12);
00616     *(dst_int+13*int_dst_stride) = *(src_int + 13);
00617     *(dst_int+14*int_dst_stride) = *(src_int + 14);
00618     *(dst_int+15*int_dst_stride) = *(src_int + 15);
00619 }
00620 
00624 #define readAndTranspose16x6(src, src_stride, r8, r9, r10, r11, r12, r13) {\
00625     register vec_u8 r0  = unaligned_load(0,             src);            \
00626     register vec_u8 r1  = unaligned_load(   src_stride, src);            \
00627     register vec_u8 r2  = unaligned_load(2* src_stride, src);            \
00628     register vec_u8 r3  = unaligned_load(3* src_stride, src);            \
00629     register vec_u8 r4  = unaligned_load(4* src_stride, src);            \
00630     register vec_u8 r5  = unaligned_load(5* src_stride, src);            \
00631     register vec_u8 r6  = unaligned_load(6* src_stride, src);            \
00632     register vec_u8 r7  = unaligned_load(7* src_stride, src);            \
00633     register vec_u8 r14 = unaligned_load(14*src_stride, src);            \
00634     register vec_u8 r15 = unaligned_load(15*src_stride, src);            \
00635                                                                            \
00636     r8  = unaligned_load( 8*src_stride, src);                              \
00637     r9  = unaligned_load( 9*src_stride, src);                              \
00638     r10 = unaligned_load(10*src_stride, src);                              \
00639     r11 = unaligned_load(11*src_stride, src);                              \
00640     r12 = unaligned_load(12*src_stride, src);                              \
00641     r13 = unaligned_load(13*src_stride, src);                              \
00642                                                                            \
00643     /*Merge first pairs*/                                                  \
00644     r0 = vec_mergeh(r0, r8);    /*0, 8*/                                   \
00645     r1 = vec_mergeh(r1, r9);    /*1, 9*/                                   \
00646     r2 = vec_mergeh(r2, r10);   /*2,10*/                                   \
00647     r3 = vec_mergeh(r3, r11);   /*3,11*/                                   \
00648     r4 = vec_mergeh(r4, r12);   /*4,12*/                                   \
00649     r5 = vec_mergeh(r5, r13);   /*5,13*/                                   \
00650     r6 = vec_mergeh(r6, r14);   /*6,14*/                                   \
00651     r7 = vec_mergeh(r7, r15);   /*7,15*/                                   \
00652                                                                            \
00653     /*Merge second pairs*/                                                 \
00654     r8  = vec_mergeh(r0, r4);   /*0,4, 8,12 set 0*/                        \
00655     r9  = vec_mergel(r0, r4);   /*0,4, 8,12 set 1*/                        \
00656     r10 = vec_mergeh(r1, r5);   /*1,5, 9,13 set 0*/                        \
00657     r11 = vec_mergel(r1, r5);   /*1,5, 9,13 set 1*/                        \
00658     r12 = vec_mergeh(r2, r6);   /*2,6,10,14 set 0*/                        \
00659     r13 = vec_mergel(r2, r6);   /*2,6,10,14 set 1*/                        \
00660     r14 = vec_mergeh(r3, r7);   /*3,7,11,15 set 0*/                        \
00661     r15 = vec_mergel(r3, r7);   /*3,7,11,15 set 1*/                        \
00662                                                                            \
00663     /*Third merge*/                                                        \
00664     r0 = vec_mergeh(r8,  r12);  /*0,2,4,6,8,10,12,14 set 0*/               \
00665     r1 = vec_mergel(r8,  r12);  /*0,2,4,6,8,10,12,14 set 1*/               \
00666     r2 = vec_mergeh(r9,  r13);  /*0,2,4,6,8,10,12,14 set 2*/               \
00667     r4 = vec_mergeh(r10, r14);  /*1,3,5,7,9,11,13,15 set 0*/               \
00668     r5 = vec_mergel(r10, r14);  /*1,3,5,7,9,11,13,15 set 1*/               \
00669     r6 = vec_mergeh(r11, r15);  /*1,3,5,7,9,11,13,15 set 2*/               \
00670     /* Don't need to compute 3 and 7*/                                     \
00671                                                                            \
00672     /*Final merge*/                                                        \
00673     r8  = vec_mergeh(r0, r4);   /*all set 0*/                              \
00674     r9  = vec_mergel(r0, r4);   /*all set 1*/                              \
00675     r10 = vec_mergeh(r1, r5);   /*all set 2*/                              \
00676     r11 = vec_mergel(r1, r5);   /*all set 3*/                              \
00677     r12 = vec_mergeh(r2, r6);   /*all set 4*/                              \
00678     r13 = vec_mergel(r2, r6);   /*all set 5*/                              \
00679     /* Don't need to compute 14 and 15*/                                   \
00680                                                                            \
00681 }
00682 
00683 // out: o = |x-y| < a
00684 static inline vec_u8 diff_lt_altivec ( register vec_u8 x,
00685                                          register vec_u8 y,
00686                                          register vec_u8 a) {
00687 
00688     register vec_u8 diff = vec_subs(x, y);
00689     register vec_u8 diffneg = vec_subs(y, x);
00690     register vec_u8 o = vec_or(diff, diffneg); /* |x-y| */
00691     o = (vec_u8)vec_cmplt(o, a);
00692     return o;
00693 }
00694 
00695 static inline vec_u8 h264_deblock_mask ( register vec_u8 p0,
00696                                            register vec_u8 p1,
00697                                            register vec_u8 q0,
00698                                            register vec_u8 q1,
00699                                            register vec_u8 alpha,
00700                                            register vec_u8 beta) {
00701 
00702     register vec_u8 mask;
00703     register vec_u8 tempmask;
00704 
00705     mask = diff_lt_altivec(p0, q0, alpha);
00706     tempmask = diff_lt_altivec(p1, p0, beta);
00707     mask = vec_and(mask, tempmask);
00708     tempmask = diff_lt_altivec(q1, q0, beta);
00709     mask = vec_and(mask, tempmask);
00710 
00711     return mask;
00712 }
00713 
00714 // out: newp1 = clip((p2 + ((p0 + q0 + 1) >> 1)) >> 1, p1-tc0, p1+tc0)
00715 static inline vec_u8 h264_deblock_q1(register vec_u8 p0,
00716                                        register vec_u8 p1,
00717                                        register vec_u8 p2,
00718                                        register vec_u8 q0,
00719                                        register vec_u8 tc0) {
00720 
00721     register vec_u8 average = vec_avg(p0, q0);
00722     register vec_u8 temp;
00723     register vec_u8 uncliped;
00724     register vec_u8 ones;
00725     register vec_u8 max;
00726     register vec_u8 min;
00727     register vec_u8 newp1;
00728 
00729     temp = vec_xor(average, p2);
00730     average = vec_avg(average, p2);     /*avg(p2, avg(p0, q0)) */
00731     ones = vec_splat_u8(1);
00732     temp = vec_and(temp, ones);         /*(p2^avg(p0, q0)) & 1 */
00733     uncliped = vec_subs(average, temp); /*(p2+((p0+q0+1)>>1))>>1 */
00734     max = vec_adds(p1, tc0);
00735     min = vec_subs(p1, tc0);
00736     newp1 = vec_max(min, uncliped);
00737     newp1 = vec_min(max, newp1);
00738     return newp1;
00739 }
00740 
00741 #define h264_deblock_p0_q0(p0, p1, q0, q1, tc0masked) {                                           \
00742                                                                                                   \
00743     const vec_u8 A0v = vec_sl(vec_splat_u8(10), vec_splat_u8(4));                               \
00744                                                                                                   \
00745     register vec_u8 pq0bit = vec_xor(p0,q0);                                                    \
00746     register vec_u8 q1minus;                                                                    \
00747     register vec_u8 p0minus;                                                                    \
00748     register vec_u8 stage1;                                                                     \
00749     register vec_u8 stage2;                                                                     \
00750     register vec_u8 vec160;                                                                     \
00751     register vec_u8 delta;                                                                      \
00752     register vec_u8 deltaneg;                                                                   \
00753                                                                                                   \
00754     q1minus = vec_nor(q1, q1);                 /* 255 - q1 */                                     \
00755     stage1 = vec_avg(p1, q1minus);             /* (p1 - q1 + 256)>>1 */                           \
00756     stage2 = vec_sr(stage1, vec_splat_u8(1));  /* (p1 - q1 + 256)>>2 = 64 + (p1 - q1) >> 2 */     \
00757     p0minus = vec_nor(p0, p0);                 /* 255 - p0 */                                     \
00758     stage1 = vec_avg(q0, p0minus);             /* (q0 - p0 + 256)>>1 */                           \
00759     pq0bit = vec_and(pq0bit, vec_splat_u8(1));                                                    \
00760     stage2 = vec_avg(stage2, pq0bit);          /* 32 + ((q0 - p0)&1 + (p1 - q1) >> 2 + 1) >> 1 */ \
00761     stage2 = vec_adds(stage2, stage1);         /* 160 + ((p0 - q0) + (p1 - q1) >> 2 + 1) >> 1 */  \
00762     vec160 = vec_ld(0, &A0v);                                                                     \
00763     deltaneg = vec_subs(vec160, stage2);       /* -d */                                           \
00764     delta = vec_subs(stage2, vec160);          /* d */                                            \
00765     deltaneg = vec_min(tc0masked, deltaneg);                                                      \
00766     delta = vec_min(tc0masked, delta);                                                            \
00767     p0 = vec_subs(p0, deltaneg);                                                                  \
00768     q0 = vec_subs(q0, delta);                                                                     \
00769     p0 = vec_adds(p0, delta);                                                                     \
00770     q0 = vec_adds(q0, deltaneg);                                                                  \
00771 }
00772 
00773 #define h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0) {            \
00774     DECLARE_ALIGNED(16, unsigned char, temp)[16];                                             \
00775     register vec_u8 alphavec;                                                              \
00776     register vec_u8 betavec;                                                               \
00777     register vec_u8 mask;                                                                  \
00778     register vec_u8 p1mask;                                                                \
00779     register vec_u8 q1mask;                                                                \
00780     register vector signed   char tc0vec;                                                    \
00781     register vec_u8 finaltc0;                                                              \
00782     register vec_u8 tc0masked;                                                             \
00783     register vec_u8 newp1;                                                                 \
00784     register vec_u8 newq1;                                                                 \
00785                                                                                              \
00786     temp[0] = alpha;                                                                         \
00787     temp[1] = beta;                                                                          \
00788     alphavec = vec_ld(0, temp);                                                              \
00789     betavec = vec_splat(alphavec, 0x1);                                                      \
00790     alphavec = vec_splat(alphavec, 0x0);                                                     \
00791     mask = h264_deblock_mask(p0, p1, q0, q1, alphavec, betavec); /*if in block */            \
00792                                                                                              \
00793     *((int *)temp) = *((int *)tc0);                                                          \
00794     tc0vec = vec_ld(0, (signed char*)temp);                                                  \
00795     tc0vec = vec_mergeh(tc0vec, tc0vec);                                                     \
00796     tc0vec = vec_mergeh(tc0vec, tc0vec);                                                     \
00797     mask = vec_and(mask, vec_cmpgt(tc0vec, vec_splat_s8(-1)));  /* if tc0[i] >= 0 */         \
00798     finaltc0 = vec_and((vec_u8)tc0vec, mask);     /* tc = tc0 */                           \
00799                                                                                              \
00800     p1mask = diff_lt_altivec(p2, p0, betavec);                                               \
00801     p1mask = vec_and(p1mask, mask);                             /* if ( |p2 - p0| < beta) */ \
00802     tc0masked = vec_and(p1mask, (vec_u8)tc0vec);                                           \
00803     finaltc0 = vec_sub(finaltc0, p1mask);                       /* tc++ */                   \
00804     newp1 = h264_deblock_q1(p0, p1, p2, q0, tc0masked);                                      \
00805     /*end if*/                                                                               \
00806                                                                                              \
00807     q1mask = diff_lt_altivec(q2, q0, betavec);                                               \
00808     q1mask = vec_and(q1mask, mask);                             /* if ( |q2 - q0| < beta ) */\
00809     tc0masked = vec_and(q1mask, (vec_u8)tc0vec);                                           \
00810     finaltc0 = vec_sub(finaltc0, q1mask);                       /* tc++ */                   \
00811     newq1 = h264_deblock_q1(p0, q1, q2, q0, tc0masked);                                      \
00812     /*end if*/                                                                               \
00813                                                                                              \
00814     h264_deblock_p0_q0(p0, p1, q0, q1, finaltc0);                                            \
00815     p1 = newp1;                                                                              \
00816     q1 = newq1;                                                                              \
00817 }
00818 
00819 static void h264_v_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) {
00820 
00821     if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) >= 0) {
00822         register vec_u8 p2 = vec_ld(-3*stride, pix);
00823         register vec_u8 p1 = vec_ld(-2*stride, pix);
00824         register vec_u8 p0 = vec_ld(-1*stride, pix);
00825         register vec_u8 q0 = vec_ld(0, pix);
00826         register vec_u8 q1 = vec_ld(stride, pix);
00827         register vec_u8 q2 = vec_ld(2*stride, pix);
00828         h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0);
00829         vec_st(p1, -2*stride, pix);
00830         vec_st(p0, -1*stride, pix);
00831         vec_st(q0, 0, pix);
00832         vec_st(q1, stride, pix);
00833     }
00834 }
00835 
00836 static void h264_h_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) {
00837 
00838     register vec_u8 line0, line1, line2, line3, line4, line5;
00839     if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) < 0)
00840         return;
00841     readAndTranspose16x6(pix-3, stride, line0, line1, line2, line3, line4, line5);
00842     h264_loop_filter_luma_altivec(line0, line1, line2, line3, line4, line5, alpha, beta, tc0);
00843     transpose4x16(line1, line2, line3, line4);
00844     write16x4(pix-2, stride, line1, line2, line3, line4);
00845 }
00846 
00847 static av_always_inline
00848 void weight_h264_WxH_altivec(uint8_t *block, int stride, int log2_denom, int weight, int offset, int w, int h)
00849 {
00850     int y, aligned;
00851     vec_u8 vblock;
00852     vec_s16 vtemp, vweight, voffset, v0, v1;
00853     vec_u16 vlog2_denom;
00854     DECLARE_ALIGNED(16, int32_t, temp)[4];
00855     LOAD_ZERO;
00856 
00857     offset <<= log2_denom;
00858     if(log2_denom) offset += 1<<(log2_denom-1);
00859     temp[0] = log2_denom;
00860     temp[1] = weight;
00861     temp[2] = offset;
00862 
00863     vtemp = (vec_s16)vec_ld(0, temp);
00864     vlog2_denom = (vec_u16)vec_splat(vtemp, 1);
00865     vweight = vec_splat(vtemp, 3);
00866     voffset = vec_splat(vtemp, 5);
00867     aligned = !((unsigned long)block & 0xf);
00868 
00869     for (y=0; y<h; y++) {
00870         vblock = vec_ld(0, block);
00871 
00872         v0 = (vec_s16)vec_mergeh(zero_u8v, vblock);
00873         v1 = (vec_s16)vec_mergel(zero_u8v, vblock);
00874 
00875         if (w == 16 || aligned) {
00876             v0 = vec_mladd(v0, vweight, zero_s16v);
00877             v0 = vec_adds(v0, voffset);
00878             v0 = vec_sra(v0, vlog2_denom);
00879         }
00880         if (w == 16 || !aligned) {
00881             v1 = vec_mladd(v1, vweight, zero_s16v);
00882             v1 = vec_adds(v1, voffset);
00883             v1 = vec_sra(v1, vlog2_denom);
00884         }
00885         vblock = vec_packsu(v0, v1);
00886         vec_st(vblock, 0, block);
00887 
00888         block += stride;
00889     }
00890 }
00891 
00892 static av_always_inline
00893 void biweight_h264_WxH_altivec(uint8_t *dst, uint8_t *src, int stride, int log2_denom,
00894                                int weightd, int weights, int offset, int w, int h)
00895 {
00896     int y, dst_aligned, src_aligned;
00897     vec_u8 vsrc, vdst;
00898     vec_s16 vtemp, vweights, vweightd, voffset, v0, v1, v2, v3;
00899     vec_u16 vlog2_denom;
00900     DECLARE_ALIGNED(16, int32_t, temp)[4];
00901     LOAD_ZERO;
00902 
00903     offset = ((offset + 1) | 1) << log2_denom;
00904     temp[0] = log2_denom+1;
00905     temp[1] = weights;
00906     temp[2] = weightd;
00907     temp[3] = offset;
00908 
00909     vtemp = (vec_s16)vec_ld(0, temp);
00910     vlog2_denom = (vec_u16)vec_splat(vtemp, 1);
00911     vweights = vec_splat(vtemp, 3);
00912     vweightd = vec_splat(vtemp, 5);
00913     voffset = vec_splat(vtemp, 7);
00914     dst_aligned = !((unsigned long)dst & 0xf);
00915     src_aligned = !((unsigned long)src & 0xf);
00916 
00917     for (y=0; y<h; y++) {
00918         vdst = vec_ld(0, dst);
00919         vsrc = vec_ld(0, src);
00920 
00921         v0 = (vec_s16)vec_mergeh(zero_u8v, vdst);
00922         v1 = (vec_s16)vec_mergel(zero_u8v, vdst);
00923         v2 = (vec_s16)vec_mergeh(zero_u8v, vsrc);
00924         v3 = (vec_s16)vec_mergel(zero_u8v, vsrc);
00925 
00926         if (w == 8) {
00927             if (src_aligned)
00928                 v3 = v2;
00929             else
00930                 v2 = v3;
00931         }
00932 
00933         if (w == 16 || dst_aligned) {
00934             v0 = vec_mladd(v0, vweightd, zero_s16v);
00935             v2 = vec_mladd(v2, vweights, zero_s16v);
00936 
00937             v0 = vec_adds(v0, voffset);
00938             v0 = vec_adds(v0, v2);
00939             v0 = vec_sra(v0, vlog2_denom);
00940         }
00941         if (w == 16 || !dst_aligned) {
00942             v1 = vec_mladd(v1, vweightd, zero_s16v);
00943             v3 = vec_mladd(v3, vweights, zero_s16v);
00944 
00945             v1 = vec_adds(v1, voffset);
00946             v1 = vec_adds(v1, v3);
00947             v1 = vec_sra(v1, vlog2_denom);
00948         }
00949         vdst = vec_packsu(v0, v1);
00950         vec_st(vdst, 0, dst);
00951 
00952         dst += stride;
00953         src += stride;
00954     }
00955 }
00956 
00957 #define H264_WEIGHT(W,H) \
00958 static void ff_weight_h264_pixels ## W ## x ## H ## _altivec(uint8_t *block, int stride, int log2_denom, int weight, int offset){ \
00959     weight_h264_WxH_altivec(block, stride, log2_denom, weight, offset, W, H); \
00960 }\
00961 static void ff_biweight_h264_pixels ## W ## x ## H ## _altivec(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset){ \
00962     biweight_h264_WxH_altivec(dst, src, stride, log2_denom, weightd, weights, offset, W, H); \
00963 }
00964 
00965 H264_WEIGHT(16,16)
00966 H264_WEIGHT(16, 8)
00967 H264_WEIGHT( 8,16)
00968 H264_WEIGHT( 8, 8)
00969 H264_WEIGHT( 8, 4)
00970 
00971 void dsputil_h264_init_ppc(DSPContext* c, AVCodecContext *avctx) {
00972 
00973     if (has_altivec()) {
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         c->put_no_rnd_vc1_chroma_pixels_tab[0] = put_no_rnd_vc1_chroma_mc8_altivec;
00977         c->avg_no_rnd_vc1_chroma_pixels_tab[0] = avg_no_rnd_vc1_chroma_mc8_altivec;
00978 
00979 #define dspfunc(PFX, IDX, NUM) \
00980         c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_altivec; \
00981         c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_altivec; \
00982         c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_altivec; \
00983         c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_altivec; \
00984         c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_altivec; \
00985         c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_altivec; \
00986         c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_altivec; \
00987         c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_altivec; \
00988         c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_altivec; \
00989         c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_altivec; \
00990         c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_altivec; \
00991         c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_altivec; \
00992         c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_altivec; \
00993         c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_altivec; \
00994         c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_altivec; \
00995         c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_altivec
00996 
00997         dspfunc(put_h264_qpel, 0, 16);
00998         dspfunc(avg_h264_qpel, 0, 16);
00999 #undef dspfunc
01000     }
01001 }
01002 
01003 void ff_h264dsp_init_ppc(H264DSPContext *c)
01004 {
01005     if (has_altivec()) {
01006         c->h264_idct_add = ff_h264_idct_add_altivec;
01007         c->h264_idct_add8 = ff_h264_idct_add8_altivec;
01008         c->h264_idct_add16 = ff_h264_idct_add16_altivec;
01009         c->h264_idct_add16intra = ff_h264_idct_add16intra_altivec;
01010         c->h264_idct_dc_add= h264_idct_dc_add_altivec;
01011         c->h264_idct8_dc_add = ff_h264_idct8_dc_add_altivec;
01012         c->h264_idct8_add = ff_h264_idct8_add_altivec;
01013         c->h264_idct8_add4 = ff_h264_idct8_add4_altivec;
01014         c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_altivec;
01015         c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_altivec;
01016 
01017         c->weight_h264_pixels_tab[0] = ff_weight_h264_pixels16x16_altivec;
01018         c->weight_h264_pixels_tab[1] = ff_weight_h264_pixels16x8_altivec;
01019         c->weight_h264_pixels_tab[2] = ff_weight_h264_pixels8x16_altivec;
01020         c->weight_h264_pixels_tab[3] = ff_weight_h264_pixels8x8_altivec;
01021         c->weight_h264_pixels_tab[4] = ff_weight_h264_pixels8x4_altivec;
01022         c->biweight_h264_pixels_tab[0] = ff_biweight_h264_pixels16x16_altivec;
01023         c->biweight_h264_pixels_tab[1] = ff_biweight_h264_pixels16x8_altivec;
01024         c->biweight_h264_pixels_tab[2] = ff_biweight_h264_pixels8x16_altivec;
01025         c->biweight_h264_pixels_tab[3] = ff_biweight_h264_pixels8x8_altivec;
01026         c->biweight_h264_pixels_tab[4] = ff_biweight_h264_pixels8x4_altivec;
01027     }
01028 }