ffmpeg/libavcodec/x86/h264_qpel.c

   1 /*
   2  * Copyright (c) 2004-2005 Michael Niedermayer, Loren Merritt
   3  * Copyright (c) 2011 Daniel Kang
   4  *
   5  * This file is part of FFmpeg.
   6  *
   7  * FFmpeg is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2.1 of the License, or (at your option) any later version.
  11  *
  12  * FFmpeg is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with FFmpeg; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20  */
  21
  22 #include "libavutil/attributes.h"
  23 #include "libavutil/cpu.h"
  24 #include "libavutil/x86/asm.h"
  25 #include "libavutil/x86/cpu.h"
  26 #include "libavcodec/h264qpel.h"
  27 #include "libavcodec/mpegvideo.h"
  28 #include "libavcodec/pixels.h"
  29 #include "fpel.h"
  30
  31 #if HAVE_YASM
  32 void ff_put_pixels4_mmx(uint8_t *block, const uint8_t *pixels,
  33                         ptrdiff_t line_size, int h);
  34 void ff_avg_pixels4_mmxext(uint8_t *block, const uint8_t *pixels,
  35                            ptrdiff_t line_size, int h);
  36 void ff_put_pixels4_l2_mmxext(uint8_t *dst, const uint8_t *src1, const uint8_t *src2,
  37                               int dstStride, int src1Stride, int h);
  38 void ff_avg_pixels4_l2_mmxext(uint8_t *dst, const uint8_t *src1, const uint8_t *src2,
  39                               int dstStride, int src1Stride, int h);
  40 void ff_put_pixels8_l2_mmxext(uint8_t *dst, const uint8_t *src1, const uint8_t *src2,
  41                               int dstStride, int src1Stride, int h);
  42 void ff_avg_pixels8_l2_mmxext(uint8_t *dst, const uint8_t *src1, const uint8_t *src2,
  43                               int dstStride, int src1Stride, int h);
  44 void ff_put_pixels16_l2_mmxext(uint8_t *dst, const uint8_t *src1, const uint8_t *src2,
  45                                int dstStride, int src1Stride, int h);
  46 void ff_avg_pixels16_l2_mmxext(uint8_t *dst, const uint8_t *src1, const uint8_t *src2,
  47                                int dstStride, int src1Stride, int h);
  48 #define ff_put_pixels8_l2_sse2  ff_put_pixels8_l2_mmxext
  49 #define ff_avg_pixels8_l2_sse2  ff_avg_pixels8_l2_mmxext
  50 #define ff_put_pixels16_l2_sse2 ff_put_pixels16_l2_mmxext
  51 #define ff_avg_pixels16_l2_sse2 ff_avg_pixels16_l2_mmxext
  52 #define ff_put_pixels16_mmxext  ff_put_pixels16_mmx
  53 #define ff_put_pixels8_mmxext   ff_put_pixels8_mmx
  54 #define ff_put_pixels4_mmxext   ff_put_pixels4_mmx
  55
  56 #define DEF_QPEL(OPNAME)\
  57 void ff_ ## OPNAME ## _h264_qpel4_h_lowpass_mmxext(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride);\
  58 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_mmxext(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride);\
  59 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_ssse3(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride);\
  60 void ff_ ## OPNAME ## _h264_qpel4_h_lowpass_l2_mmxext(uint8_t *dst, const uint8_t *src, const uint8_t *src2, int dstStride, int src2Stride);\
  61 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_l2_mmxext(uint8_t *dst, const uint8_t *src, const uint8_t *src2, int dstStride, int src2Stride);\
  62 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_l2_ssse3(uint8_t *dst, const uint8_t *src, const uint8_t *src2, int dstStride, int src2Stride);\
  63 void ff_ ## OPNAME ## _h264_qpel4_v_lowpass_mmxext(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride);\
  64 void ff_ ## OPNAME ## _h264_qpel8or16_v_lowpass_op_mmxext(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h);\
  65 void ff_ ## OPNAME ## _h264_qpel8or16_v_lowpass_sse2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h);\
  66 void ff_ ## OPNAME ## _h264_qpel4_hv_lowpass_v_mmxext(const uint8_t *src, int16_t *tmp, int srcStride);\
  67 void ff_ ## OPNAME ## _h264_qpel4_hv_lowpass_h_mmxext(int16_t *tmp, uint8_t *dst, int dstStride);\
  68 void ff_ ## OPNAME ## _h264_qpel8or16_hv1_lowpass_op_mmxext(const uint8_t *src, int16_t *tmp, int srcStride, int size);\
  69 void ff_ ## OPNAME ## _h264_qpel8or16_hv1_lowpass_op_sse2(const uint8_t *src, int16_t *tmp, int srcStride, int size);\
  70 void ff_ ## OPNAME ## _h264_qpel8or16_hv2_lowpass_op_mmxext(uint8_t *dst, int16_t *tmp, int dstStride, int unused, int h);\
  71 void ff_ ## OPNAME ## _h264_qpel8or16_hv2_lowpass_ssse3(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size);\
  72 void ff_ ## OPNAME ## _pixels4_l2_shift5_mmxext(uint8_t *dst, const int16_t *src16, const uint8_t *src8, int dstStride, int src8Stride, int h);\
  73 void ff_ ## OPNAME ## _pixels8_l2_shift5_mmxext(uint8_t *dst, const int16_t *src16, const uint8_t *src8, int dstStride, int src8Stride, int h);
  74
  75 DEF_QPEL(avg)
  76 DEF_QPEL(put)
  77
  78 #define QPEL_H264(OPNAME, OP, MMX)\
  79 static av_always_inline void ff_ ## OPNAME ## h264_qpel4_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, const uint8_t *src, int dstStride, int tmpStride, int srcStride){\
  80     int w=3;\
  81     src -= 2*srcStride+2;\
  82     while(w--){\
  83         ff_ ## OPNAME ## h264_qpel4_hv_lowpass_v_mmxext(src, tmp, srcStride);\
  84         tmp += 4;\
  85         src += 4;\
  86     }\
  87     tmp -= 3*4;\
  88     ff_ ## OPNAME ## h264_qpel4_hv_lowpass_h_mmxext(tmp, dst, dstStride);\
  89 }\
  90 \
  91 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h){\
  92     src -= 2*srcStride;\
  93     ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_op_mmxext(dst, src, dstStride, srcStride, h);\
  94     src += 4;\
  95     dst += 4;\
  96     ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_op_mmxext(dst, src, dstStride, srcStride, h);\
  97 }\
  98 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_hv1_lowpass_ ## MMX(int16_t *tmp, const uint8_t *src, int tmpStride, int srcStride, int size){\
  99     int w = (size+8)>>2;\
 100     src -= 2*srcStride+2;\
 101     while(w--){\
 102         ff_ ## OPNAME ## h264_qpel8or16_hv1_lowpass_op_mmxext(src, tmp, srcStride, size);\
 103         tmp += 4;\
 104         src += 4;\
 105     }\
 106 }\
 107 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
 108     int w = size>>4;\
 109     do{\
 110     ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_op_mmxext(dst, tmp, dstStride, 0, size);\
 111     tmp += 8;\
 112     dst += 8;\
 113     }while(w--);\
 114 }\
 115 \
 116 static av_always_inline void ff_ ## OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\
 117     ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
 118 }\
 119 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\
 120     ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
 121     ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
 122 }\
 123 \
 124 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\
 125     ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
 126     ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
 127     src += 8*srcStride;\
 128     dst += 8*dstStride;\
 129     ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
 130     ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
 131 }\
 132 \
 133 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, const uint8_t *src, const uint8_t *src2, int dstStride, int src2Stride){\
 134     ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
 135     ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
 136     src += 8*dstStride;\
 137     dst += 8*dstStride;\
 138     src2 += 8*src2Stride;\
 139     ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
 140     ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
 141 }\
 142 \
 143 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, const uint8_t *src, int dstStride, int tmpStride, int srcStride, int size){\
 144     ff_put_h264_qpel8or16_hv1_lowpass_ ## MMX(tmp, src, tmpStride, srcStride, size);\
 145     ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
 146 }\
 147 static av_always_inline void ff_ ## OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, const uint8_t *src, int dstStride, int tmpStride, int srcStride){\
 148     ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 8);\
 149 }\
 150 \
 151 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, const uint8_t *src, int dstStride, int tmpStride, int srcStride){\
 152     ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 16);\
 153 }\
 154 \
 155 static av_always_inline void ff_ ## OPNAME ## pixels16_l2_shift5_ ## MMX(uint8_t *dst, const int16_t *src16, const uint8_t *src8, int dstStride, int src8Stride, int h)\
 156 {\
 157     ff_ ## OPNAME ## pixels8_l2_shift5_ ## MMX(dst  , src16  , src8  , dstStride, src8Stride, h);\
 158     ff_ ## OPNAME ## pixels8_l2_shift5_ ## MMX(dst+8, src16+8, src8+8, dstStride, src8Stride, h);\
 159 }\
 160
 161
 162 #if ARCH_X86_64
 163 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
 164
 165 void ff_avg_h264_qpel16_h_lowpass_l2_ssse3(uint8_t *dst, const uint8_t *src, const uint8_t *src2, int dstStride, int src2Stride);
 166 void ff_put_h264_qpel16_h_lowpass_l2_ssse3(uint8_t *dst, const uint8_t *src, const uint8_t *src2, int dstStride, int src2Stride);
 167
 168 #else // ARCH_X86_64
 169 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
 170 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, const uint8_t *src, const uint8_t *src2, int dstStride, int src2Stride){\
 171     ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
 172     ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
 173     src += 8*dstStride;\
 174     dst += 8*dstStride;\
 175     src2 += 8*src2Stride;\
 176     ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
 177     ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
 178 }
 179 #endif // ARCH_X86_64
 180
 181 #define QPEL_H264_H_XMM(OPNAME, OP, MMX)\
 182 QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
 183 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\
 184     ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
 185     ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
 186     src += 8*srcStride;\
 187     dst += 8*dstStride;\
 188     ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
 189     ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
 190 }\
 191
 192 #define QPEL_H264_V_XMM(OPNAME, OP, MMX)\
 193 static av_always_inline void ff_ ## OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\
 194     ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
 195 }\
 196 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\
 197     ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
 198     ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
 199 }
 200
 201 static av_always_inline void put_h264_qpel8or16_hv1_lowpass_sse2(int16_t *tmp,
 202                                                                  const uint8_t *src,
 203                                                                  int tmpStride,
 204                                                                  int srcStride,
 205                                                                  int size)
 206 {
 207     int w = (size+8)>>3;
 208     src -= 2*srcStride+2;
 209     while(w--){
 210         ff_put_h264_qpel8or16_hv1_lowpass_op_sse2(src, tmp, srcStride, size);
 211         tmp += 8;
 212         src += 8;
 213     }
 214 }
 215
 216 #define QPEL_H264_HV_XMM(OPNAME, OP, MMX)\
 217 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, const uint8_t *src, int dstStride, int tmpStride, int srcStride, int size){\
 218     put_h264_qpel8or16_hv1_lowpass_sse2(tmp, src, tmpStride, srcStride, size);\
 219     ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
 220 }\
 221 static av_always_inline void ff_ ## OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, const uint8_t *src, int dstStride, int tmpStride, int srcStride){\
 222     ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 8);\
 223 }\
 224 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, const uint8_t *src, int dstStride, int tmpStride, int srcStride){\
 225     ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 16);\
 226 }\
 227
 228 #define ff_put_h264_qpel8_h_lowpass_l2_sse2  ff_put_h264_qpel8_h_lowpass_l2_mmxext
 229 #define ff_avg_h264_qpel8_h_lowpass_l2_sse2  ff_avg_h264_qpel8_h_lowpass_l2_mmxext
 230 #define ff_put_h264_qpel16_h_lowpass_l2_sse2 ff_put_h264_qpel16_h_lowpass_l2_mmxext
 231 #define ff_avg_h264_qpel16_h_lowpass_l2_sse2 ff_avg_h264_qpel16_h_lowpass_l2_mmxext
 232
 233 #define ff_put_h264_qpel8_v_lowpass_ssse3  ff_put_h264_qpel8_v_lowpass_sse2
 234 #define ff_avg_h264_qpel8_v_lowpass_ssse3  ff_avg_h264_qpel8_v_lowpass_sse2
 235 #define ff_put_h264_qpel16_v_lowpass_ssse3 ff_put_h264_qpel16_v_lowpass_sse2
 236 #define ff_avg_h264_qpel16_v_lowpass_ssse3 ff_avg_h264_qpel16_v_lowpass_sse2
 237
 238 #define ff_put_h264_qpel8or16_hv2_lowpass_sse2 ff_put_h264_qpel8or16_hv2_lowpass_mmxext
 239 #define ff_avg_h264_qpel8or16_hv2_lowpass_sse2 ff_avg_h264_qpel8or16_hv2_lowpass_mmxext
 240
 241 #define H264_MC(OPNAME, SIZE, MMX, ALIGN) \
 242 H264_MC_C(OPNAME, SIZE, MMX, ALIGN)\
 243 H264_MC_V(OPNAME, SIZE, MMX, ALIGN)\
 244 H264_MC_H(OPNAME, SIZE, MMX, ALIGN)\
 245 H264_MC_HV(OPNAME, SIZE, MMX, ALIGN)\
 246
 247 static void put_h264_qpel16_mc00_sse2 (uint8_t *dst, const uint8_t *src,
 248                                        ptrdiff_t stride)
 249 {
 250     ff_put_pixels16_sse2(dst, src, stride, 16);
 251 }
 252 static void avg_h264_qpel16_mc00_sse2 (uint8_t *dst, const uint8_t *src,
 253                                        ptrdiff_t stride)
 254 {
 255     ff_avg_pixels16_sse2(dst, src, stride, 16);
 256 }
 257 #define put_h264_qpel8_mc00_sse2 put_h264_qpel8_mc00_mmxext
 258 #define avg_h264_qpel8_mc00_sse2 avg_h264_qpel8_mc00_mmxext
 259
 260 #define H264_MC_C(OPNAME, SIZE, MMX, ALIGN) \
 261 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
 262 {\
 263     ff_ ## OPNAME ## pixels ## SIZE ## _ ## MMX(dst, src, stride, SIZE);\
 264 }\
 265
 266 #define H264_MC_H(OPNAME, SIZE, MMX, ALIGN) \
 267 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
 268 {\
 269     ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src, stride, stride);\
 270 }\
 271 \
 272 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
 273 {\
 274     ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\
 275 }\
 276 \
 277 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
 278 {\
 279     ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src+1, stride, stride);\
 280 }\
 281
 282 #define H264_MC_V(OPNAME, SIZE, MMX, ALIGN) \
 283 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
 284 {\
 285     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
 286     ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
 287     ff_ ## OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, temp, stride, stride, SIZE);\
 288 }\
 289 \
 290 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
 291 {\
 292     ff_ ## OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\
 293 }\
 294 \
 295 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
 296 {\
 297     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
 298     ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
 299     ff_ ## OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, temp, stride, stride, SIZE);\
 300 }\
 301
 302 #define H264_MC_HV(OPNAME, SIZE, MMX, ALIGN) \
 303 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
 304 {\
 305     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
 306     ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
 307     ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
 308 }\
 309 \
 310 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
 311 {\
 312     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
 313     ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
 314     ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
 315 }\
 316 \
 317 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
 318 {\
 319     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
 320     ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
 321     ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
 322 }\
 323 \
 324 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
 325 {\
 326     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
 327     ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
 328     ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
 329 }\
 330 \
 331 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
 332 {\
 333     DECLARE_ALIGNED(ALIGN, uint16_t, temp)[SIZE*(SIZE<8?12:24)];\
 334     ff_ ## OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, temp, src, stride, SIZE, stride);\
 335 }\
 336 \
 337 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
 338 {\
 339     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
 340     uint8_t * const halfHV= temp;\
 341     int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
 342     av_assert2(((int)temp & 7) == 0);\
 343     ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
 344     ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfHV, stride, SIZE);\
 345 }\
 346 \
 347 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
 348 {\
 349     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
 350     uint8_t * const halfHV= temp;\
 351     int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
 352     av_assert2(((int)temp & 7) == 0);\
 353     ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
 354     ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfHV, stride, SIZE);\
 355 }\
 356 \
 357 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
 358 {\
 359     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
 360     uint8_t * const halfHV= temp;\
 361     int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
 362     av_assert2(((int)temp & 7) == 0);\
 363     ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
 364     ff_ ## OPNAME ## pixels ## SIZE ## _l2_shift5_mmxext(dst, halfV+2, halfHV, stride, SIZE, SIZE);\
 365 }\
 366 \
 367 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
 368 {\
 369     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
 370     uint8_t * const halfHV= temp;\
 371     int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
 372     av_assert2(((int)temp & 7) == 0);\
 373     ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
 374     ff_ ## OPNAME ## pixels ## SIZE ## _l2_shift5_mmxext(dst, halfV+3, halfHV, stride, SIZE, SIZE);\
 375 }\
 376
 377 #define H264_MC_4816(MMX)\
 378 H264_MC(put_, 4, MMX, 8)\
 379 H264_MC(put_, 8, MMX, 8)\
 380 H264_MC(put_, 16,MMX, 8)\
 381 H264_MC(avg_, 4, MMX, 8)\
 382 H264_MC(avg_, 8, MMX, 8)\
 383 H264_MC(avg_, 16,MMX, 8)\
 384
 385 #define H264_MC_816(QPEL, XMM)\
 386 QPEL(put_, 8, XMM, 16)\
 387 QPEL(put_, 16,XMM, 16)\
 388 QPEL(avg_, 8, XMM, 16)\
 389 QPEL(avg_, 16,XMM, 16)\
 390
 391 QPEL_H264(put_,        PUT_OP, mmxext)
 392 QPEL_H264(avg_, AVG_MMXEXT_OP, mmxext)
 393 QPEL_H264_V_XMM(put_,       PUT_OP, sse2)
 394 QPEL_H264_V_XMM(avg_,AVG_MMXEXT_OP, sse2)
 395 QPEL_H264_HV_XMM(put_,       PUT_OP, sse2)
 396 QPEL_H264_HV_XMM(avg_,AVG_MMXEXT_OP, sse2)
 397 QPEL_H264_H_XMM(put_,       PUT_OP, ssse3)
 398 QPEL_H264_H_XMM(avg_,AVG_MMXEXT_OP, ssse3)
 399 QPEL_H264_HV_XMM(put_,       PUT_OP, ssse3)
 400 QPEL_H264_HV_XMM(avg_,AVG_MMXEXT_OP, ssse3)
 401
 402 H264_MC_4816(mmxext)
 403 H264_MC_816(H264_MC_V, sse2)
 404 H264_MC_816(H264_MC_HV, sse2)
 405 H264_MC_816(H264_MC_H, ssse3)
 406 H264_MC_816(H264_MC_HV, ssse3)
 407
 408
 409 //10bit
 410 #define LUMA_MC_OP(OP, NUM, DEPTH, TYPE, OPT) \
 411 void ff_ ## OP ## _h264_qpel ## NUM ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT \
 412     (uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
 413
 414 #define LUMA_MC_ALL(DEPTH, TYPE, OPT) \
 415     LUMA_MC_OP(put,  4, DEPTH, TYPE, OPT) \
 416     LUMA_MC_OP(avg,  4, DEPTH, TYPE, OPT) \
 417     LUMA_MC_OP(put,  8, DEPTH, TYPE, OPT) \
 418     LUMA_MC_OP(avg,  8, DEPTH, TYPE, OPT) \
 419     LUMA_MC_OP(put, 16, DEPTH, TYPE, OPT) \
 420     LUMA_MC_OP(avg, 16, DEPTH, TYPE, OPT)
 421
 422 #define LUMA_MC_816(DEPTH, TYPE, OPT) \
 423     LUMA_MC_OP(put,  8, DEPTH, TYPE, OPT) \
 424     LUMA_MC_OP(avg,  8, DEPTH, TYPE, OPT) \
 425     LUMA_MC_OP(put, 16, DEPTH, TYPE, OPT) \
 426     LUMA_MC_OP(avg, 16, DEPTH, TYPE, OPT)
 427
 428 LUMA_MC_ALL(10, mc00, mmxext)
 429 LUMA_MC_ALL(10, mc10, mmxext)
 430 LUMA_MC_ALL(10, mc20, mmxext)
 431 LUMA_MC_ALL(10, mc30, mmxext)
 432 LUMA_MC_ALL(10, mc01, mmxext)
 433 LUMA_MC_ALL(10, mc11, mmxext)
 434 LUMA_MC_ALL(10, mc21, mmxext)
 435 LUMA_MC_ALL(10, mc31, mmxext)
 436 LUMA_MC_ALL(10, mc02, mmxext)
 437 LUMA_MC_ALL(10, mc12, mmxext)
 438 LUMA_MC_ALL(10, mc22, mmxext)
 439 LUMA_MC_ALL(10, mc32, mmxext)
 440 LUMA_MC_ALL(10, mc03, mmxext)
 441 LUMA_MC_ALL(10, mc13, mmxext)
 442 LUMA_MC_ALL(10, mc23, mmxext)
 443 LUMA_MC_ALL(10, mc33, mmxext)
 444
 445 LUMA_MC_816(10, mc00, sse2)
 446 LUMA_MC_816(10, mc10, sse2)
 447 LUMA_MC_816(10, mc10, sse2_cache64)
 448 LUMA_MC_816(10, mc10, ssse3_cache64)
 449 LUMA_MC_816(10, mc20, sse2)
 450 LUMA_MC_816(10, mc20, sse2_cache64)
 451 LUMA_MC_816(10, mc20, ssse3_cache64)
 452 LUMA_MC_816(10, mc30, sse2)
 453 LUMA_MC_816(10, mc30, sse2_cache64)
 454 LUMA_MC_816(10, mc30, ssse3_cache64)
 455 LUMA_MC_816(10, mc01, sse2)
 456 LUMA_MC_816(10, mc11, sse2)
 457 LUMA_MC_816(10, mc21, sse2)
 458 LUMA_MC_816(10, mc31, sse2)
 459 LUMA_MC_816(10, mc02, sse2)
 460 LUMA_MC_816(10, mc12, sse2)
 461 LUMA_MC_816(10, mc22, sse2)
 462 LUMA_MC_816(10, mc32, sse2)
 463 LUMA_MC_816(10, mc03, sse2)
 464 LUMA_MC_816(10, mc13, sse2)
 465 LUMA_MC_816(10, mc23, sse2)
 466 LUMA_MC_816(10, mc33, sse2)
 467
 468 #define QPEL16_OPMC(OP, MC, MMX)\
 469 void ff_ ## OP ## _h264_qpel16_ ## MC ## _10_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride){\
 470     ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst   , src   , stride);\
 471     ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst+16, src+16, stride);\
 472     src += 8*stride;\
 473     dst += 8*stride;\
 474     ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst   , src   , stride);\
 475     ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst+16, src+16, stride);\
 476 }
 477
 478 #define QPEL16_OP(MC, MMX)\
 479 QPEL16_OPMC(put, MC, MMX)\
 480 QPEL16_OPMC(avg, MC, MMX)
 481
 482 #define QPEL16(MMX)\
 483 QPEL16_OP(mc00, MMX)\
 484 QPEL16_OP(mc01, MMX)\
 485 QPEL16_OP(mc02, MMX)\
 486 QPEL16_OP(mc03, MMX)\
 487 QPEL16_OP(mc10, MMX)\
 488 QPEL16_OP(mc11, MMX)\
 489 QPEL16_OP(mc12, MMX)\
 490 QPEL16_OP(mc13, MMX)\
 491 QPEL16_OP(mc20, MMX)\
 492 QPEL16_OP(mc21, MMX)\
 493 QPEL16_OP(mc22, MMX)\
 494 QPEL16_OP(mc23, MMX)\
 495 QPEL16_OP(mc30, MMX)\
 496 QPEL16_OP(mc31, MMX)\
 497 QPEL16_OP(mc32, MMX)\
 498 QPEL16_OP(mc33, MMX)
 499
 500 #if ARCH_X86_32 // ARCH_X86_64 implies SSE2+
 501 QPEL16(mmxext)
 502 #endif
 503
 504 #endif /* HAVE_YASM */
 505
 506 #define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU, PREFIX)                          \
 507     do {                                                                     \
 508     c->PFX ## _pixels_tab[IDX][ 0] = PREFIX ## PFX ## SIZE ## _mc00_ ## CPU; \
 509     c->PFX ## _pixels_tab[IDX][ 1] = PREFIX ## PFX ## SIZE ## _mc10_ ## CPU; \
 510     c->PFX ## _pixels_tab[IDX][ 2] = PREFIX ## PFX ## SIZE ## _mc20_ ## CPU; \
 511     c->PFX ## _pixels_tab[IDX][ 3] = PREFIX ## PFX ## SIZE ## _mc30_ ## CPU; \
 512     c->PFX ## _pixels_tab[IDX][ 4] = PREFIX ## PFX ## SIZE ## _mc01_ ## CPU; \
 513     c->PFX ## _pixels_tab[IDX][ 5] = PREFIX ## PFX ## SIZE ## _mc11_ ## CPU; \
 514     c->PFX ## _pixels_tab[IDX][ 6] = PREFIX ## PFX ## SIZE ## _mc21_ ## CPU; \
 515     c->PFX ## _pixels_tab[IDX][ 7] = PREFIX ## PFX ## SIZE ## _mc31_ ## CPU; \
 516     c->PFX ## _pixels_tab[IDX][ 8] = PREFIX ## PFX ## SIZE ## _mc02_ ## CPU; \
 517     c->PFX ## _pixels_tab[IDX][ 9] = PREFIX ## PFX ## SIZE ## _mc12_ ## CPU; \
 518     c->PFX ## _pixels_tab[IDX][10] = PREFIX ## PFX ## SIZE ## _mc22_ ## CPU; \
 519     c->PFX ## _pixels_tab[IDX][11] = PREFIX ## PFX ## SIZE ## _mc32_ ## CPU; \
 520     c->PFX ## _pixels_tab[IDX][12] = PREFIX ## PFX ## SIZE ## _mc03_ ## CPU; \
 521     c->PFX ## _pixels_tab[IDX][13] = PREFIX ## PFX ## SIZE ## _mc13_ ## CPU; \
 522     c->PFX ## _pixels_tab[IDX][14] = PREFIX ## PFX ## SIZE ## _mc23_ ## CPU; \
 523     c->PFX ## _pixels_tab[IDX][15] = PREFIX ## PFX ## SIZE ## _mc33_ ## CPU; \
 524     } while (0)
 525
 526 #define H264_QPEL_FUNCS(x, y, CPU)                                                            \
 527     do {                                                                                      \
 528         c->put_h264_qpel_pixels_tab[0][x + y * 4] = put_h264_qpel16_mc ## x ## y ## _ ## CPU; \
 529         c->put_h264_qpel_pixels_tab[1][x + y * 4] = put_h264_qpel8_mc  ## x ## y ## _ ## CPU; \
 530         c->avg_h264_qpel_pixels_tab[0][x + y * 4] = avg_h264_qpel16_mc ## x ## y ## _ ## CPU; \
 531         c->avg_h264_qpel_pixels_tab[1][x + y * 4] = avg_h264_qpel8_mc  ## x ## y ## _ ## CPU; \
 532     } while (0)
 533
 534 #define H264_QPEL_FUNCS_10(x, y, CPU)                                                               \
 535     do {                                                                                            \
 536         c->put_h264_qpel_pixels_tab[0][x + y * 4] = ff_put_h264_qpel16_mc ## x ## y ## _10_ ## CPU; \
 537         c->put_h264_qpel_pixels_tab[1][x + y * 4] = ff_put_h264_qpel8_mc  ## x ## y ## _10_ ## CPU; \
 538         c->avg_h264_qpel_pixels_tab[0][x + y * 4] = ff_avg_h264_qpel16_mc ## x ## y ## _10_ ## CPU; \
 539         c->avg_h264_qpel_pixels_tab[1][x + y * 4] = ff_avg_h264_qpel8_mc  ## x ## y ## _10_ ## CPU; \
 540     } while (0)
 541
 542 av_cold void ff_h264qpel_init_x86(H264QpelContext *c, int bit_depth)
 543 {
 544 #if HAVE_YASM
 545     int high_bit_depth = bit_depth > 8;
 546     int cpu_flags = av_get_cpu_flags();
 547
 548     if (EXTERNAL_MMXEXT(cpu_flags)) {
 549         if (!high_bit_depth) {
 550             SET_QPEL_FUNCS(put_h264_qpel, 0, 16, mmxext, );
 551             SET_QPEL_FUNCS(put_h264_qpel, 1,  8, mmxext, );
 552             SET_QPEL_FUNCS(put_h264_qpel, 2,  4, mmxext, );
 553             SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, mmxext, );
 554             SET_QPEL_FUNCS(avg_h264_qpel, 1,  8, mmxext, );
 555             SET_QPEL_FUNCS(avg_h264_qpel, 2,  4, mmxext, );
 556         } else if (bit_depth == 10) {
 557 #if ARCH_X86_32
 558             SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 10_mmxext, ff_);
 559             SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 10_mmxext, ff_);
 560             SET_QPEL_FUNCS(put_h264_qpel, 1,  8, 10_mmxext, ff_);
 561             SET_QPEL_FUNCS(avg_h264_qpel, 1,  8, 10_mmxext, ff_);
 562 #endif
 563             SET_QPEL_FUNCS(put_h264_qpel, 2, 4,  10_mmxext, ff_);
 564             SET_QPEL_FUNCS(avg_h264_qpel, 2, 4,  10_mmxext, ff_);
 565         }
 566     }
 567
 568     if (EXTERNAL_SSE2(cpu_flags)) {
 569         if (!(cpu_flags & AV_CPU_FLAG_SSE2SLOW) && !high_bit_depth) {
 570             // these functions are slower than mmx on AMD, but faster on Intel
 571             H264_QPEL_FUNCS(0, 0, sse2);
 572         }
 573
 574         if (!high_bit_depth) {
 575             H264_QPEL_FUNCS(0, 1, sse2);
 576             H264_QPEL_FUNCS(0, 2, sse2);
 577             H264_QPEL_FUNCS(0, 3, sse2);
 578             H264_QPEL_FUNCS(1, 1, sse2);
 579             H264_QPEL_FUNCS(1, 2, sse2);
 580             H264_QPEL_FUNCS(1, 3, sse2);
 581             H264_QPEL_FUNCS(2, 1, sse2);
 582             H264_QPEL_FUNCS(2, 2, sse2);
 583             H264_QPEL_FUNCS(2, 3, sse2);
 584             H264_QPEL_FUNCS(3, 1, sse2);
 585             H264_QPEL_FUNCS(3, 2, sse2);
 586             H264_QPEL_FUNCS(3, 3, sse2);
 587         }
 588
 589         if (bit_depth == 10) {
 590             SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 10_sse2, ff_);
 591             SET_QPEL_FUNCS(put_h264_qpel, 1,  8, 10_sse2, ff_);
 592             SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 10_sse2, ff_);
 593             SET_QPEL_FUNCS(avg_h264_qpel, 1,  8, 10_sse2, ff_);
 594             H264_QPEL_FUNCS_10(1, 0, sse2_cache64);
 595             H264_QPEL_FUNCS_10(2, 0, sse2_cache64);
 596             H264_QPEL_FUNCS_10(3, 0, sse2_cache64);
 597         }
 598     }
 599
 600     if (EXTERNAL_SSSE3(cpu_flags)) {
 601         if (!high_bit_depth) {
 602             H264_QPEL_FUNCS(1, 0, ssse3);
 603             H264_QPEL_FUNCS(1, 1, ssse3);
 604             H264_QPEL_FUNCS(1, 2, ssse3);
 605             H264_QPEL_FUNCS(1, 3, ssse3);
 606             H264_QPEL_FUNCS(2, 0, ssse3);
 607             H264_QPEL_FUNCS(2, 1, ssse3);
 608             H264_QPEL_FUNCS(2, 2, ssse3);
 609             H264_QPEL_FUNCS(2, 3, ssse3);
 610             H264_QPEL_FUNCS(3, 0, ssse3);
 611             H264_QPEL_FUNCS(3, 1, ssse3);
 612             H264_QPEL_FUNCS(3, 2, ssse3);
 613             H264_QPEL_FUNCS(3, 3, ssse3);
 614         }
 615
 616         if (bit_depth == 10) {
 617             H264_QPEL_FUNCS_10(1, 0, ssse3_cache64);
 618             H264_QPEL_FUNCS_10(2, 0, ssse3_cache64);
 619             H264_QPEL_FUNCS_10(3, 0, ssse3_cache64);
 620         }
 621     }
 622
 623     if (EXTERNAL_AVX(cpu_flags)) {
 624         /* AVX implies 64 byte cache lines without the need to avoid unaligned
 625          * memory accesses that cross the boundary between two cache lines.
 626          * TODO: Port X264_CPU_CACHELINE_32/64 detection from x264 to avoid
 627          * having to treat SSE2 functions with such properties as AVX. */
 628         if (bit_depth == 10) {
 629             H264_QPEL_FUNCS_10(1, 0, sse2);
 630             H264_QPEL_FUNCS_10(2, 0, sse2);
 631             H264_QPEL_FUNCS_10(3, 0, sse2);
 632         }
 633     }
 634 #endif
 635 }