libavcodec/ppc/h264_qpel.c

   1 /*
   2  * Copyright (c) 2004 Romain Dolbeau <romain@dolbeau.org>
   3  *
   4  * This file is part of Libav.
   5  *
   6  * Libav is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU Lesser General Public
   8  * License as published by the Free Software Foundation; either
   9  * version 2.1 of the License, or (at your option) any later version.
  10  *
  11  * Libav is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * Lesser General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU Lesser General Public
  17  * License along with Libav; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  19  */
  20
  21 #include "config.h"
  22 #include "libavutil/attributes.h"
  23 #include "libavcodec/h264qpel.h"
  24
  25 #if HAVE_ALTIVEC
  26 #include "libavutil/cpu.h"
  27 #include "libavutil/intreadwrite.h"
  28 #include "libavutil/ppc/types_altivec.h"
  29 #include "libavutil/ppc/util_altivec.h"
  30 #include "dsputil_altivec.h"
  31
  32 #define PUT_OP_U8_ALTIVEC(d, s, dst) d = s
  33 #define AVG_OP_U8_ALTIVEC(d, s, dst) d = vec_avg(dst, s)
  34
  35 #define OP_U8_ALTIVEC                          PUT_OP_U8_ALTIVEC
  36 #define PREFIX_h264_qpel16_h_lowpass_altivec   put_h264_qpel16_h_lowpass_altivec
  37 #define PREFIX_h264_qpel16_h_lowpass_num       altivec_put_h264_qpel16_h_lowpass_num
  38 #define PREFIX_h264_qpel16_v_lowpass_altivec   put_h264_qpel16_v_lowpass_altivec
  39 #define PREFIX_h264_qpel16_v_lowpass_num       altivec_put_h264_qpel16_v_lowpass_num
  40 #define PREFIX_h264_qpel16_hv_lowpass_altivec  put_h264_qpel16_hv_lowpass_altivec
  41 #define PREFIX_h264_qpel16_hv_lowpass_num      altivec_put_h264_qpel16_hv_lowpass_num
  42 #include "h264_qpel_template.c"
  43 #undef OP_U8_ALTIVEC
  44 #undef PREFIX_h264_qpel16_h_lowpass_altivec
  45 #undef PREFIX_h264_qpel16_h_lowpass_num
  46 #undef PREFIX_h264_qpel16_v_lowpass_altivec
  47 #undef PREFIX_h264_qpel16_v_lowpass_num
  48 #undef PREFIX_h264_qpel16_hv_lowpass_altivec
  49 #undef PREFIX_h264_qpel16_hv_lowpass_num
  50
  51 #define OP_U8_ALTIVEC                          AVG_OP_U8_ALTIVEC
  52 #define PREFIX_h264_qpel16_h_lowpass_altivec   avg_h264_qpel16_h_lowpass_altivec
  53 #define PREFIX_h264_qpel16_h_lowpass_num       altivec_avg_h264_qpel16_h_lowpass_num
  54 #define PREFIX_h264_qpel16_v_lowpass_altivec   avg_h264_qpel16_v_lowpass_altivec
  55 #define PREFIX_h264_qpel16_v_lowpass_num       altivec_avg_h264_qpel16_v_lowpass_num
  56 #define PREFIX_h264_qpel16_hv_lowpass_altivec  avg_h264_qpel16_hv_lowpass_altivec
  57 #define PREFIX_h264_qpel16_hv_lowpass_num      altivec_avg_h264_qpel16_hv_lowpass_num
  58 #include "h264_qpel_template.c"
  59 #undef OP_U8_ALTIVEC
  60 #undef PREFIX_h264_qpel16_h_lowpass_altivec
  61 #undef PREFIX_h264_qpel16_h_lowpass_num
  62 #undef PREFIX_h264_qpel16_v_lowpass_altivec
  63 #undef PREFIX_h264_qpel16_v_lowpass_num
  64 #undef PREFIX_h264_qpel16_hv_lowpass_altivec
  65 #undef PREFIX_h264_qpel16_hv_lowpass_num
  66
  67 #define H264_MC(OPNAME, SIZE, CODETYPE) \
  68 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## CODETYPE (uint8_t *dst, uint8_t *src, int stride){\
  69     ff_ ## OPNAME ## pixels ## SIZE ## _ ## CODETYPE(dst, src, stride, SIZE);\
  70 }\
  71 \
  72 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){ \
  73     DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
  74     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
  75     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\
  76 }\
  77 \
  78 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
  79     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(dst, src, stride, stride);\
  80 }\
  81 \
  82 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
  83     DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
  84     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
  85     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+1, half, stride, stride, SIZE);\
  86 }\
  87 \
  88 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
  89     DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
  90     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
  91     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\
  92 }\
  93 \
  94 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
  95     OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(dst, src, stride, stride);\
  96 }\
  97 \
  98 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
  99     DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
 100     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
 101     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+stride, half, stride, stride, SIZE);\
 102 }\
 103 \
 104 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
 105     DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
 106     DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
 107     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
 108     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
 109     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
 110 }\
 111 \
 112 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
 113     DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
 114     DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
 115     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
 116     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
 117     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
 118 }\
 119 \
 120 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
 121     DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
 122     DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
 123     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
 124     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
 125     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
 126 }\
 127 \
 128 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
 129     DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
 130     DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
 131     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
 132     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
 133     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
 134 }\
 135 \
 136 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
 137     DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
 138     OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(dst, tmp, src, stride, SIZE, stride);\
 139 }\
 140 \
 141 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
 142     DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
 143     DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
 144     DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
 145     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
 146     put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
 147     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\
 148 }\
 149 \
 150 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
 151     DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
 152     DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
 153     DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
 154     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
 155     put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
 156     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\
 157 }\
 158 \
 159 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
 160     DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
 161     DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
 162     DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
 163     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
 164     put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
 165     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
 166 }\
 167 \
 168 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
 169     DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
 170     DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
 171     DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
 172     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
 173     put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
 174     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
 175 }\
 176
 177 static inline void put_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
 178                                     const uint8_t * src2, int dst_stride,
 179                                     int src_stride1, int h)
 180 {
 181     int i;
 182     vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align;
 183
 184     mask_ = vec_lvsl(0, src2);
 185
 186     for (i = 0; i < h; i++) {
 187
 188         tmp1 = vec_ld(i * src_stride1, src1);
 189         mask = vec_lvsl(i * src_stride1, src1);
 190         tmp2 = vec_ld(i * src_stride1 + 15, src1);
 191
 192         a = vec_perm(tmp1, tmp2, mask);
 193
 194         tmp1 = vec_ld(i * 16, src2);
 195         tmp2 = vec_ld(i * 16 + 15, src2);
 196
 197         b = vec_perm(tmp1, tmp2, mask_);
 198
 199         tmp1 = vec_ld(0, dst);
 200         mask = vec_lvsl(0, dst);
 201         tmp2 = vec_ld(15, dst);
 202
 203         d = vec_avg(a, b);
 204
 205         edges = vec_perm(tmp2, tmp1, mask);
 206
 207         align = vec_lvsr(0, dst);
 208
 209         tmp2 = vec_perm(d, edges, align);
 210         tmp1 = vec_perm(edges, d, align);
 211
 212         vec_st(tmp2, 15, dst);
 213         vec_st(tmp1, 0 , dst);
 214
 215         dst += dst_stride;
 216     }
 217 }
 218
 219 static inline void avg_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
 220                                     const uint8_t * src2, int dst_stride,
 221                                     int src_stride1, int h)
 222 {
 223     int i;
 224     vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align;
 225
 226     mask_ = vec_lvsl(0, src2);
 227
 228     for (i = 0; i < h; i++) {
 229
 230         tmp1 = vec_ld(i * src_stride1, src1);
 231         mask = vec_lvsl(i * src_stride1, src1);
 232         tmp2 = vec_ld(i * src_stride1 + 15, src1);
 233
 234         a = vec_perm(tmp1, tmp2, mask);
 235
 236         tmp1 = vec_ld(i * 16, src2);
 237         tmp2 = vec_ld(i * 16 + 15, src2);
 238
 239         b = vec_perm(tmp1, tmp2, mask_);
 240
 241         tmp1 = vec_ld(0, dst);
 242         mask = vec_lvsl(0, dst);
 243         tmp2 = vec_ld(15, dst);
 244
 245         d = vec_avg(vec_perm(tmp1, tmp2, mask), vec_avg(a, b));
 246
 247         edges = vec_perm(tmp2, tmp1, mask);
 248
 249         align = vec_lvsr(0, dst);
 250
 251         tmp2 = vec_perm(d, edges, align);
 252         tmp1 = vec_perm(edges, d, align);
 253
 254         vec_st(tmp2, 15, dst);
 255         vec_st(tmp1, 0 , dst);
 256
 257         dst += dst_stride;
 258     }
 259 }
 260
 261 /* Implemented but could be faster
 262 #define put_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) put_pixels16_l2(d,s1,s2,ds,s1s,16,h)
 263 #define avg_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) avg_pixels16_l2(d,s1,s2,ds,s1s,16,h)
 264  */
 265
 266 H264_MC(put_, 16, altivec)
 267 H264_MC(avg_, 16, altivec)
 268 #endif /* HAVE_ALTIVEC */
 269
 270 av_cold void ff_h264qpel_init_ppc(H264QpelContext *c, int bit_depth)
 271 {
 272 #if HAVE_ALTIVEC
 273     const int high_bit_depth = bit_depth > 8;
 274
 275     if (av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC) {
 276     if (!high_bit_depth) {
 277 #define dspfunc(PFX, IDX, NUM) \
 278         c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_altivec; \
 279         c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_altivec; \
 280         c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_altivec; \
 281         c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_altivec; \
 282         c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_altivec; \
 283         c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_altivec; \
 284         c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_altivec; \
 285         c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_altivec; \
 286         c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_altivec; \
 287         c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_altivec; \
 288         c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_altivec; \
 289         c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_altivec; \
 290         c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_altivec; \
 291         c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_altivec; \
 292         c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_altivec; \
 293         c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_altivec
 294
 295         dspfunc(put_h264_qpel, 0, 16);
 296         dspfunc(avg_h264_qpel, 0, 16);
 297 #undef dspfunc
 298     }
 299     }
 300 #endif /* HAVE_ALTIVEC */
 301 }