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 Libav.
   6  *
   7  * Libav 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  * Libav 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 Libav; 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 "dsputil_mmx.h"
  29
  30 #if HAVE_YASM
  31 void ff_put_pixels4_mmxext(uint8_t *block, const uint8_t *pixels, int line_size, int h);
  32 void ff_avg_pixels4_mmxext(uint8_t *block, const uint8_t *pixels, int line_size, int h);
  33 void ff_put_pixels8_mmxext(uint8_t *block, const uint8_t *pixels, int line_size, int h);
  34 static void ff_put_pixels16_mmxext(uint8_t *block, const uint8_t *pixels,
  35                                    int line_size, int h)
  36 {
  37     ff_put_pixels8_mmxext(block,     pixels,     line_size, h);
  38     ff_put_pixels8_mmxext(block + 8, pixels + 8, line_size, h);
  39 }
  40 static void ff_avg_pixels16_mmxext(uint8_t *block, const uint8_t *pixels,
  41                                    int line_size, int h)
  42 {
  43     ff_avg_pixels8_mmxext(block,     pixels,     line_size, h);
  44     ff_avg_pixels8_mmxext(block + 8, pixels + 8, line_size, h);
  45 }
  46 void ff_put_pixels4_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
  47                               int dstStride, int src1Stride, int h);
  48 void ff_avg_pixels4_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
  49                               int dstStride, int src1Stride, int h);
  50 void ff_put_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
  51                               int dstStride, int src1Stride, int h);
  52 void ff_avg_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
  53                               int dstStride, int src1Stride, int h);
  54 void ff_put_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
  55                                int dstStride, int src1Stride, int h);
  56 void ff_avg_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
  57                                int dstStride, int src1Stride, int h);
  58 void ff_put_pixels16_sse2(uint8_t *block, const uint8_t *pixels,
  59                           int line_size, int h);
  60 void ff_avg_pixels16_sse2(uint8_t *block, const uint8_t *pixels,
  61                           int line_size, int h);
  62 #define ff_put_pixels8_l2_sse2  ff_put_pixels8_l2_mmxext
  63 #define ff_avg_pixels8_l2_sse2  ff_avg_pixels8_l2_mmxext
  64 #define ff_put_pixels16_l2_sse2 ff_put_pixels16_l2_mmxext
  65 #define ff_avg_pixels16_l2_sse2 ff_avg_pixels16_l2_mmxext
  66
  67 #define DEF_QPEL(OPNAME)\
  68 void ff_ ## OPNAME ## _h264_qpel4_h_lowpass_mmxext(uint8_t *dst, uint8_t *src, int dstStride, int srcStride);\
  69 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src, int dstStride, int srcStride);\
  70 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_ssse3(uint8_t *dst, uint8_t *src, int dstStride, int srcStride);\
  71 void ff_ ## OPNAME ## _h264_qpel4_h_lowpass_l2_mmxext(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride);\
  72 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_l2_mmxext(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride);\
  73 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_l2_ssse3(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride);\
  74 void ff_ ## OPNAME ## _h264_qpel4_v_lowpass_mmxext(uint8_t *dst, uint8_t *src, int dstStride, int srcStride);\
  75 void ff_ ## OPNAME ## _h264_qpel8or16_v_lowpass_op_mmxext(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h);\
  76 void ff_ ## OPNAME ## _h264_qpel8or16_v_lowpass_sse2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h);\
  77 void ff_ ## OPNAME ## _h264_qpel4_hv_lowpass_v_mmxext(uint8_t *src, int16_t *tmp, int srcStride);\
  78 void ff_ ## OPNAME ## _h264_qpel4_hv_lowpass_h_mmxext(int16_t *tmp, uint8_t *dst, int dstStride);\
  79 void ff_ ## OPNAME ## _h264_qpel8or16_hv1_lowpass_op_mmxext(uint8_t *src, int16_t *tmp, int srcStride, int size);\
  80 void ff_ ## OPNAME ## _h264_qpel8or16_hv1_lowpass_op_sse2(uint8_t *src, int16_t *tmp, int srcStride, int size);\
  81 void ff_ ## OPNAME ## _h264_qpel8or16_hv2_lowpass_op_mmxext(uint8_t *dst, int16_t *tmp, int dstStride, int unused, int h);\
  82 void ff_ ## OPNAME ## _h264_qpel8or16_hv2_lowpass_ssse3(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size);\
  83 void ff_ ## OPNAME ## _pixels4_l2_shift5_mmxext(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h);\
  84 void ff_ ## OPNAME ## _pixels8_l2_shift5_mmxext(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h);
  85
  86 DEF_QPEL(avg)
  87 DEF_QPEL(put)
  88
  89 #define QPEL_H264(OPNAME, OP, MMX)\
  90 static av_always_inline void ff_ ## OPNAME ## h264_qpel4_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
  91     int w=3;\
  92     src -= 2*srcStride+2;\
  93     while(w--){\
  94         ff_ ## OPNAME ## h264_qpel4_hv_lowpass_v_mmxext(src, tmp, srcStride);\
  95         tmp += 4;\
  96         src += 4;\
  97     }\
  98     tmp -= 3*4;\
  99     ff_ ## OPNAME ## h264_qpel4_hv_lowpass_h_mmxext(tmp, dst, dstStride);\
 100 }\
 101 \
 102 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
 103     src -= 2*srcStride;\
 104     ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_op_mmxext(dst, src, dstStride, srcStride, h);\
 105     src += 4;\
 106     dst += 4;\
 107     ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_op_mmxext(dst, src, dstStride, srcStride, h);\
 108 }\
 109 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_hv1_lowpass_ ## MMX(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){\
 110     int w = (size+8)>>2;\
 111     src -= 2*srcStride+2;\
 112     while(w--){\
 113         ff_ ## OPNAME ## h264_qpel8or16_hv1_lowpass_op_mmxext(src, tmp, srcStride, size);\
 114         tmp += 4;\
 115         src += 4;\
 116     }\
 117 }\
 118 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
 119     int w = size>>4;\
 120     do{\
 121     ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_op_mmxext(dst, tmp, dstStride, 0, size);\
 122     tmp += 8;\
 123     dst += 8;\
 124     }while(w--);\
 125 }\
 126 \
 127 static av_always_inline void ff_ ## OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
 128     ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
 129 }\
 130 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
 131     ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
 132     ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
 133 }\
 134 \
 135 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
 136     ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
 137     ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
 138     src += 8*srcStride;\
 139     dst += 8*dstStride;\
 140     ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
 141     ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
 142 }\
 143 \
 144 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
 145     ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
 146     ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
 147     src += 8*dstStride;\
 148     dst += 8*dstStride;\
 149     src2 += 8*src2Stride;\
 150     ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
 151     ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
 152 }\
 153 \
 154 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride, int size){\
 155     ff_put_h264_qpel8or16_hv1_lowpass_ ## MMX(tmp, src, tmpStride, srcStride, size);\
 156     ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
 157 }\
 158 static av_always_inline void ff_ ## OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
 159     ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 8);\
 160 }\
 161 \
 162 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
 163     ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 16);\
 164 }\
 165 \
 166 static av_always_inline void ff_ ## OPNAME ## pixels16_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
 167 {\
 168     ff_ ## OPNAME ## pixels8_l2_shift5_ ## MMX(dst  , src16  , src8  , dstStride, src8Stride, h);\
 169     ff_ ## OPNAME ## pixels8_l2_shift5_ ## MMX(dst+8, src16+8, src8+8, dstStride, src8Stride, h);\
 170 }\
 171
 172
 173 #if ARCH_X86_64
 174 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
 175
 176 void ff_avg_h264_qpel16_h_lowpass_l2_ssse3(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride);
 177 void ff_put_h264_qpel16_h_lowpass_l2_ssse3(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride);
 178
 179 #else // ARCH_X86_64
 180 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
 181 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
 182     ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
 183     ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
 184     src += 8*dstStride;\
 185     dst += 8*dstStride;\
 186     src2 += 8*src2Stride;\
 187     ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
 188     ff_ ## OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
 189 }
 190 #endif // ARCH_X86_64
 191
 192 #define QPEL_H264_H_XMM(OPNAME, OP, MMX)\
 193 QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
 194 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
 195     ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
 196     ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
 197     src += 8*srcStride;\
 198     dst += 8*dstStride;\
 199     ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
 200     ff_ ## OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
 201 }\
 202
 203 #define QPEL_H264_V_XMM(OPNAME, OP, MMX)\
 204 static av_always_inline void ff_ ## OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
 205     ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
 206 }\
 207 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
 208     ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
 209     ff_ ## OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
 210 }
 211
 212 static av_always_inline void ff_put_h264_qpel8or16_hv1_lowpass_sse2(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){
 213     int w = (size+8)>>3;
 214     src -= 2*srcStride+2;
 215     while(w--){
 216         ff_put_h264_qpel8or16_hv1_lowpass_op_sse2(src, tmp, srcStride, size);
 217         tmp += 8;
 218         src += 8;
 219     }
 220 }
 221
 222 #define QPEL_H264_HV_XMM(OPNAME, OP, MMX)\
 223 static av_always_inline void ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride, int size){\
 224     ff_put_h264_qpel8or16_hv1_lowpass_sse2(tmp, src, tmpStride, srcStride, size);\
 225     ff_ ## OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
 226 }\
 227 static av_always_inline void ff_ ## OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
 228     ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 8);\
 229 }\
 230 static av_always_inline void ff_ ## OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
 231     ff_ ## OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 16);\
 232 }\
 233
 234 #define ff_put_h264_qpel8_h_lowpass_l2_sse2  ff_put_h264_qpel8_h_lowpass_l2_mmxext
 235 #define ff_avg_h264_qpel8_h_lowpass_l2_sse2  ff_avg_h264_qpel8_h_lowpass_l2_mmxext
 236 #define ff_put_h264_qpel16_h_lowpass_l2_sse2 ff_put_h264_qpel16_h_lowpass_l2_mmxext
 237 #define ff_avg_h264_qpel16_h_lowpass_l2_sse2 ff_avg_h264_qpel16_h_lowpass_l2_mmxext
 238
 239 #define ff_put_h264_qpel8_v_lowpass_ssse3  ff_put_h264_qpel8_v_lowpass_sse2
 240 #define ff_avg_h264_qpel8_v_lowpass_ssse3  ff_avg_h264_qpel8_v_lowpass_sse2
 241 #define ff_put_h264_qpel16_v_lowpass_ssse3 ff_put_h264_qpel16_v_lowpass_sse2
 242 #define ff_avg_h264_qpel16_v_lowpass_ssse3 ff_avg_h264_qpel16_v_lowpass_sse2
 243
 244 #define ff_put_h264_qpel8or16_hv2_lowpass_sse2 ff_put_h264_qpel8or16_hv2_lowpass_mmxext
 245 #define ff_avg_h264_qpel8or16_hv2_lowpass_sse2 ff_avg_h264_qpel8or16_hv2_lowpass_mmxext
 246
 247 #define H264_MC(OPNAME, SIZE, MMX, ALIGN) \
 248 H264_MC_C(OPNAME, SIZE, MMX, ALIGN)\
 249 H264_MC_V(OPNAME, SIZE, MMX, ALIGN)\
 250 H264_MC_H(OPNAME, SIZE, MMX, ALIGN)\
 251 H264_MC_HV(OPNAME, SIZE, MMX, ALIGN)\
 252
 253 static void put_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src, int stride){
 254     ff_put_pixels16_sse2(dst, src, stride, 16);
 255 }
 256 static void avg_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src, int stride){
 257     ff_avg_pixels16_sse2(dst, src, stride, 16);
 258 }
 259 #define put_h264_qpel8_mc00_sse2 put_h264_qpel8_mc00_mmxext
 260 #define avg_h264_qpel8_mc00_sse2 avg_h264_qpel8_mc00_mmxext
 261
 262 #define H264_MC_C(OPNAME, SIZE, MMX, ALIGN) \
 263 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
 264     ff_ ## OPNAME ## pixels ## SIZE ## _ ## MMX(dst, src, stride, SIZE);\
 265 }\
 266
 267 #define H264_MC_H(OPNAME, SIZE, MMX, ALIGN) \
 268 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
 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, uint8_t *src, int stride){\
 273     ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\
 274 }\
 275 \
 276 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
 277     ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src+1, stride, stride);\
 278 }\
 279
 280 #define H264_MC_V(OPNAME, SIZE, MMX, ALIGN) \
 281 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
 282     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
 283     ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
 284     ff_ ## OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, temp, stride, stride, SIZE);\
 285 }\
 286 \
 287 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
 288     ff_ ## OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\
 289 }\
 290 \
 291 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
 292     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
 293     ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
 294     ff_ ## OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, temp, stride, stride, SIZE);\
 295 }\
 296
 297 #define H264_MC_HV(OPNAME, SIZE, MMX, ALIGN) \
 298 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
 299     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
 300     ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
 301     ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
 302 }\
 303 \
 304 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
 305     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
 306     ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, 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 ## _mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
 311     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
 312     ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
 313     ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
 314 }\
 315 \
 316 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
 317     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
 318     ff_put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
 319     ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
 320 }\
 321 \
 322 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
 323     DECLARE_ALIGNED(ALIGN, uint16_t, temp)[SIZE*(SIZE<8?12:24)];\
 324     ff_ ## OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, temp, src, stride, SIZE, stride);\
 325 }\
 326 \
 327 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
 328     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
 329     uint8_t * const halfHV= temp;\
 330     int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
 331     assert(((int)temp & 7) == 0);\
 332     ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
 333     ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfHV, stride, SIZE);\
 334 }\
 335 \
 336 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
 337     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
 338     uint8_t * const halfHV= temp;\
 339     int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
 340     assert(((int)temp & 7) == 0);\
 341     ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
 342     ff_ ## OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfHV, stride, SIZE);\
 343 }\
 344 \
 345 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
 346     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
 347     uint8_t * const halfHV= temp;\
 348     int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
 349     assert(((int)temp & 7) == 0);\
 350     ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
 351     ff_ ## OPNAME ## pixels ## SIZE ## _l2_shift5_mmxext(dst, halfV+2, halfHV, stride, SIZE, SIZE);\
 352 }\
 353 \
 354 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
 355     DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
 356     uint8_t * const halfHV= temp;\
 357     int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
 358     assert(((int)temp & 7) == 0);\
 359     ff_put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
 360     ff_ ## OPNAME ## pixels ## SIZE ## _l2_shift5_mmxext(dst, halfV+3, halfHV, stride, SIZE, SIZE);\
 361 }\
 362
 363 #define H264_MC_4816(MMX)\
 364 H264_MC(put_, 4, MMX, 8)\
 365 H264_MC(put_, 8, MMX, 8)\
 366 H264_MC(put_, 16,MMX, 8)\
 367 H264_MC(avg_, 4, MMX, 8)\
 368 H264_MC(avg_, 8, MMX, 8)\
 369 H264_MC(avg_, 16,MMX, 8)\
 370
 371 #define H264_MC_816(QPEL, XMM)\
 372 QPEL(put_, 8, XMM, 16)\
 373 QPEL(put_, 16,XMM, 16)\
 374 QPEL(avg_, 8, XMM, 16)\
 375 QPEL(avg_, 16,XMM, 16)\
 376
 377 #undef PAVGB
 378 #define PAVGB "pavgb"
 379 QPEL_H264(put_,        PUT_OP, mmxext)
 380 QPEL_H264(avg_, AVG_MMXEXT_OP, mmxext)
 381 QPEL_H264_V_XMM(put_,       PUT_OP, sse2)
 382 QPEL_H264_V_XMM(avg_,AVG_MMXEXT_OP, sse2)
 383 QPEL_H264_HV_XMM(put_,       PUT_OP, sse2)
 384 QPEL_H264_HV_XMM(avg_,AVG_MMXEXT_OP, sse2)
 385 QPEL_H264_H_XMM(put_,       PUT_OP, ssse3)
 386 QPEL_H264_H_XMM(avg_,AVG_MMXEXT_OP, ssse3)
 387 QPEL_H264_HV_XMM(put_,       PUT_OP, ssse3)
 388 QPEL_H264_HV_XMM(avg_,AVG_MMXEXT_OP, ssse3)
 389 #undef PAVGB
 390
 391 H264_MC_4816(mmxext)
 392 H264_MC_816(H264_MC_V, sse2)
 393 H264_MC_816(H264_MC_HV, sse2)
 394 H264_MC_816(H264_MC_H, ssse3)
 395 H264_MC_816(H264_MC_HV, ssse3)
 396
 397
 398 //10bit
 399 #define LUMA_MC_OP(OP, NUM, DEPTH, TYPE, OPT) \
 400 void ff_ ## OP ## _h264_qpel ## NUM ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT \
 401     (uint8_t *dst, uint8_t *src, int stride);
 402
 403 #define LUMA_MC_ALL(DEPTH, TYPE, OPT) \
 404     LUMA_MC_OP(put,  4, DEPTH, TYPE, OPT) \
 405     LUMA_MC_OP(avg,  4, DEPTH, TYPE, OPT) \
 406     LUMA_MC_OP(put,  8, DEPTH, TYPE, OPT) \
 407     LUMA_MC_OP(avg,  8, DEPTH, TYPE, OPT) \
 408     LUMA_MC_OP(put, 16, DEPTH, TYPE, OPT) \
 409     LUMA_MC_OP(avg, 16, DEPTH, TYPE, OPT)
 410
 411 #define LUMA_MC_816(DEPTH, TYPE, OPT) \
 412     LUMA_MC_OP(put,  8, DEPTH, TYPE, OPT) \
 413     LUMA_MC_OP(avg,  8, DEPTH, TYPE, OPT) \
 414     LUMA_MC_OP(put, 16, DEPTH, TYPE, OPT) \
 415     LUMA_MC_OP(avg, 16, DEPTH, TYPE, OPT)
 416
 417 LUMA_MC_ALL(10, mc00, mmxext)
 418 LUMA_MC_ALL(10, mc10, mmxext)
 419 LUMA_MC_ALL(10, mc20, mmxext)
 420 LUMA_MC_ALL(10, mc30, mmxext)
 421 LUMA_MC_ALL(10, mc01, mmxext)
 422 LUMA_MC_ALL(10, mc11, mmxext)
 423 LUMA_MC_ALL(10, mc21, mmxext)
 424 LUMA_MC_ALL(10, mc31, mmxext)
 425 LUMA_MC_ALL(10, mc02, mmxext)
 426 LUMA_MC_ALL(10, mc12, mmxext)
 427 LUMA_MC_ALL(10, mc22, mmxext)
 428 LUMA_MC_ALL(10, mc32, mmxext)
 429 LUMA_MC_ALL(10, mc03, mmxext)
 430 LUMA_MC_ALL(10, mc13, mmxext)
 431 LUMA_MC_ALL(10, mc23, mmxext)
 432 LUMA_MC_ALL(10, mc33, mmxext)
 433
 434 LUMA_MC_816(10, mc00, sse2)
 435 LUMA_MC_816(10, mc10, sse2)
 436 LUMA_MC_816(10, mc10, sse2_cache64)
 437 LUMA_MC_816(10, mc10, ssse3_cache64)
 438 LUMA_MC_816(10, mc20, sse2)
 439 LUMA_MC_816(10, mc20, sse2_cache64)
 440 LUMA_MC_816(10, mc20, ssse3_cache64)
 441 LUMA_MC_816(10, mc30, sse2)
 442 LUMA_MC_816(10, mc30, sse2_cache64)
 443 LUMA_MC_816(10, mc30, ssse3_cache64)
 444 LUMA_MC_816(10, mc01, sse2)
 445 LUMA_MC_816(10, mc11, sse2)
 446 LUMA_MC_816(10, mc21, sse2)
 447 LUMA_MC_816(10, mc31, sse2)
 448 LUMA_MC_816(10, mc02, sse2)
 449 LUMA_MC_816(10, mc12, sse2)
 450 LUMA_MC_816(10, mc22, sse2)
 451 LUMA_MC_816(10, mc32, sse2)
 452 LUMA_MC_816(10, mc03, sse2)
 453 LUMA_MC_816(10, mc13, sse2)
 454 LUMA_MC_816(10, mc23, sse2)
 455 LUMA_MC_816(10, mc33, sse2)
 456
 457 #define QPEL16_OPMC(OP, MC, MMX)\
 458 void ff_ ## OP ## _h264_qpel16_ ## MC ## _10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
 459     ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst   , src   , stride);\
 460     ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst+16, src+16, stride);\
 461     src += 8*stride;\
 462     dst += 8*stride;\
 463     ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst   , src   , stride);\
 464     ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst+16, src+16, stride);\
 465 }
 466
 467 #define QPEL16_OP(MC, MMX)\
 468 QPEL16_OPMC(put, MC, MMX)\
 469 QPEL16_OPMC(avg, MC, MMX)
 470
 471 #define QPEL16(MMX)\
 472 QPEL16_OP(mc00, MMX)\
 473 QPEL16_OP(mc01, MMX)\
 474 QPEL16_OP(mc02, MMX)\
 475 QPEL16_OP(mc03, MMX)\
 476 QPEL16_OP(mc10, MMX)\
 477 QPEL16_OP(mc11, MMX)\
 478 QPEL16_OP(mc12, MMX)\
 479 QPEL16_OP(mc13, MMX)\
 480 QPEL16_OP(mc20, MMX)\
 481 QPEL16_OP(mc21, MMX)\
 482 QPEL16_OP(mc22, MMX)\
 483 QPEL16_OP(mc23, MMX)\
 484 QPEL16_OP(mc30, MMX)\
 485 QPEL16_OP(mc31, MMX)\
 486 QPEL16_OP(mc32, MMX)\
 487 QPEL16_OP(mc33, MMX)
 488
 489 #if ARCH_X86_32 && HAVE_YASM && CONFIG_H264QPEL // ARCH_X86_64 implies SSE2+
 490 QPEL16(mmxext)
 491 #endif
 492
 493 #endif /* HAVE_YASM */
 494
 495 #define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU, PREFIX)                          \
 496     do {                                                                     \
 497     c->PFX ## _pixels_tab[IDX][ 0] = PREFIX ## PFX ## SIZE ## _mc00_ ## CPU; \
 498     c->PFX ## _pixels_tab[IDX][ 1] = PREFIX ## PFX ## SIZE ## _mc10_ ## CPU; \
 499     c->PFX ## _pixels_tab[IDX][ 2] = PREFIX ## PFX ## SIZE ## _mc20_ ## CPU; \
 500     c->PFX ## _pixels_tab[IDX][ 3] = PREFIX ## PFX ## SIZE ## _mc30_ ## CPU; \
 501     c->PFX ## _pixels_tab[IDX][ 4] = PREFIX ## PFX ## SIZE ## _mc01_ ## CPU; \
 502     c->PFX ## _pixels_tab[IDX][ 5] = PREFIX ## PFX ## SIZE ## _mc11_ ## CPU; \
 503     c->PFX ## _pixels_tab[IDX][ 6] = PREFIX ## PFX ## SIZE ## _mc21_ ## CPU; \
 504     c->PFX ## _pixels_tab[IDX][ 7] = PREFIX ## PFX ## SIZE ## _mc31_ ## CPU; \
 505     c->PFX ## _pixels_tab[IDX][ 8] = PREFIX ## PFX ## SIZE ## _mc02_ ## CPU; \
 506     c->PFX ## _pixels_tab[IDX][ 9] = PREFIX ## PFX ## SIZE ## _mc12_ ## CPU; \
 507     c->PFX ## _pixels_tab[IDX][10] = PREFIX ## PFX ## SIZE ## _mc22_ ## CPU; \
 508     c->PFX ## _pixels_tab[IDX][11] = PREFIX ## PFX ## SIZE ## _mc32_ ## CPU; \
 509     c->PFX ## _pixels_tab[IDX][12] = PREFIX ## PFX ## SIZE ## _mc03_ ## CPU; \
 510     c->PFX ## _pixels_tab[IDX][13] = PREFIX ## PFX ## SIZE ## _mc13_ ## CPU; \
 511     c->PFX ## _pixels_tab[IDX][14] = PREFIX ## PFX ## SIZE ## _mc23_ ## CPU; \
 512     c->PFX ## _pixels_tab[IDX][15] = PREFIX ## PFX ## SIZE ## _mc33_ ## CPU; \
 513     } while (0)
 514
 515 #define H264_QPEL_FUNCS(x, y, CPU)                                                            \
 516     do {                                                                                      \
 517         c->put_h264_qpel_pixels_tab[0][x + y * 4] = put_h264_qpel16_mc ## x ## y ## _ ## CPU; \
 518         c->put_h264_qpel_pixels_tab[1][x + y * 4] = put_h264_qpel8_mc  ## x ## y ## _ ## CPU; \
 519         c->avg_h264_qpel_pixels_tab[0][x + y * 4] = avg_h264_qpel16_mc ## x ## y ## _ ## CPU; \
 520         c->avg_h264_qpel_pixels_tab[1][x + y * 4] = avg_h264_qpel8_mc  ## x ## y ## _ ## CPU; \
 521     } while (0)
 522
 523 #define H264_QPEL_FUNCS_10(x, y, CPU)                                                               \
 524     do {                                                                                            \
 525         c->put_h264_qpel_pixels_tab[0][x + y * 4] = ff_put_h264_qpel16_mc ## x ## y ## _10_ ## CPU; \
 526         c->put_h264_qpel_pixels_tab[1][x + y * 4] = ff_put_h264_qpel8_mc  ## x ## y ## _10_ ## CPU; \
 527         c->avg_h264_qpel_pixels_tab[0][x + y * 4] = ff_avg_h264_qpel16_mc ## x ## y ## _10_ ## CPU; \
 528         c->avg_h264_qpel_pixels_tab[1][x + y * 4] = ff_avg_h264_qpel8_mc  ## x ## y ## _10_ ## CPU; \
 529     } while (0)
 530
 531 av_cold void ff_h264qpel_init_x86(H264QpelContext *c, int bit_depth)
 532 {
 533 #if HAVE_YASM
 534     int high_bit_depth = bit_depth > 8;
 535     int mm_flags = av_get_cpu_flags();
 536
 537     if (EXTERNAL_MMXEXT(mm_flags)) {
 538         if (!high_bit_depth) {
 539             SET_QPEL_FUNCS(put_h264_qpel, 0, 16, mmxext, );
 540             SET_QPEL_FUNCS(put_h264_qpel, 1,  8, mmxext, );
 541             SET_QPEL_FUNCS(put_h264_qpel, 2,  4, mmxext, );
 542             SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, mmxext, );
 543             SET_QPEL_FUNCS(avg_h264_qpel, 1,  8, mmxext, );
 544             SET_QPEL_FUNCS(avg_h264_qpel, 2,  4, mmxext, );
 545         } else if (bit_depth == 10) {
 546 #if ARCH_X86_32
 547             SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 10_mmxext, ff_);
 548             SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 10_mmxext, ff_);
 549             SET_QPEL_FUNCS(put_h264_qpel, 1,  8, 10_mmxext, ff_);
 550             SET_QPEL_FUNCS(avg_h264_qpel, 1,  8, 10_mmxext, ff_);
 551 #endif
 552             SET_QPEL_FUNCS(put_h264_qpel, 2, 4,  10_mmxext, ff_);
 553             SET_QPEL_FUNCS(avg_h264_qpel, 2, 4,  10_mmxext, ff_);
 554         }
 555     }
 556
 557     if (EXTERNAL_SSE2(mm_flags)) {
 558         if (!(mm_flags & AV_CPU_FLAG_SSE2SLOW) && !high_bit_depth) {
 559             // these functions are slower than mmx on AMD, but faster on Intel
 560             H264_QPEL_FUNCS(0, 0, sse2);
 561         }
 562
 563         if (!high_bit_depth) {
 564             H264_QPEL_FUNCS(0, 1, sse2);
 565             H264_QPEL_FUNCS(0, 2, sse2);
 566             H264_QPEL_FUNCS(0, 3, sse2);
 567             H264_QPEL_FUNCS(1, 1, sse2);
 568             H264_QPEL_FUNCS(1, 2, sse2);
 569             H264_QPEL_FUNCS(1, 3, sse2);
 570             H264_QPEL_FUNCS(2, 1, sse2);
 571             H264_QPEL_FUNCS(2, 2, sse2);
 572             H264_QPEL_FUNCS(2, 3, sse2);
 573             H264_QPEL_FUNCS(3, 1, sse2);
 574             H264_QPEL_FUNCS(3, 2, sse2);
 575             H264_QPEL_FUNCS(3, 3, sse2);
 576         }
 577
 578         if (bit_depth == 10) {
 579             SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 10_sse2, ff_);
 580             SET_QPEL_FUNCS(put_h264_qpel, 1,  8, 10_sse2, ff_);
 581             SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 10_sse2, ff_);
 582             SET_QPEL_FUNCS(avg_h264_qpel, 1,  8, 10_sse2, ff_);
 583             H264_QPEL_FUNCS_10(1, 0, sse2_cache64);
 584             H264_QPEL_FUNCS_10(2, 0, sse2_cache64);
 585             H264_QPEL_FUNCS_10(3, 0, sse2_cache64);
 586         }
 587     }
 588
 589     if (EXTERNAL_SSSE3(mm_flags)) {
 590         if (!high_bit_depth) {
 591             H264_QPEL_FUNCS(1, 0, ssse3);
 592             H264_QPEL_FUNCS(1, 1, ssse3);
 593             H264_QPEL_FUNCS(1, 2, ssse3);
 594             H264_QPEL_FUNCS(1, 3, ssse3);
 595             H264_QPEL_FUNCS(2, 0, ssse3);
 596             H264_QPEL_FUNCS(2, 1, ssse3);
 597             H264_QPEL_FUNCS(2, 2, ssse3);
 598             H264_QPEL_FUNCS(2, 3, ssse3);
 599             H264_QPEL_FUNCS(3, 0, ssse3);
 600             H264_QPEL_FUNCS(3, 1, ssse3);
 601             H264_QPEL_FUNCS(3, 2, ssse3);
 602             H264_QPEL_FUNCS(3, 3, ssse3);
 603         }
 604
 605         if (bit_depth == 10) {
 606             H264_QPEL_FUNCS_10(1, 0, ssse3_cache64);
 607             H264_QPEL_FUNCS_10(2, 0, ssse3_cache64);
 608             H264_QPEL_FUNCS_10(3, 0, ssse3_cache64);
 609         }
 610     }
 611
 612     if (EXTERNAL_AVX(mm_flags)) {
 613         /* AVX implies 64 byte cache lines without the need to avoid unaligned
 614          * memory accesses that cross the boundary between two cache lines.
 615          * TODO: Port X264_CPU_CACHELINE_32/64 detection from x264 to avoid
 616          * having to treat SSE2 functions with such properties as AVX. */
 617         if (bit_depth == 10) {
 618             H264_QPEL_FUNCS_10(1, 0, sse2);
 619             H264_QPEL_FUNCS_10(2, 0, sse2);
 620             H264_QPEL_FUNCS_10(3, 0, sse2);
 621         }
 622     }
 623 #endif
 624 }