vlc-0.8.4a: /home/huihoo/src/vlc/vlc-0.8.4a/modules/video_chroma/i420_rgb

00001 /*****************************************************************************
00002  * transforms_yuvmmx.h: MMX YUV transformation assembly
00003  *****************************************************************************
00004  * Copyright (C) 1999-2004 the VideoLAN team
00005  * $Id: i420_rgb_mmx.h 12683 2005-09-26 10:14:46Z sam $
00006  *
00007  * Authors: Olie Lho <[email protected]>
00008  *          Gaël Hendryckx <[email protected]>
00009  *          Samuel Hocevar <[email protected]>
00010  *
00011  * This program is free software; you can redistribute it and/or modify
00012  * it under the terms of the GNU General Public License as published by
00013  * the Free Software Foundation; either version 2 of the License, or
00014  * (at your option) any later version.
00015  *
00016  * This program is distributed in the hope that it will be useful,
00017  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00018  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
00019  * GNU General Public License for more details.
00020  *
00021  * You should have received a copy of the GNU General Public License
00022  * along with this program; if not, write to the Free Software
00023  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111, USA.
00024  *****************************************************************************/
00025 
00026 /* hope these constant values are cache line aligned */
00027 #if __GNUC__ > 3 || (__GNUC__ == 3 && __GNUC_MINOR__ >= 3)
00028 #define USED_U64(foo) \
00029     static const uint64_t foo __asm__ (#foo) __attribute__((used))
00030 #else
00031 #define USED_U64(foo) \
00032     static const uint64_t foo __asm__ (#foo) __attribute__((unused))
00033 #endif
00034 USED_U64(mmx_80w)     = 0x0080008000800080ULL;
00035 USED_U64(mmx_10w)     = 0x1010101010101010ULL;
00036 USED_U64(mmx_00ffw)   = 0x00ff00ff00ff00ffULL;
00037 USED_U64(mmx_Y_coeff) = 0x253f253f253f253fULL;
00038 
00039 USED_U64(mmx_U_green) = 0xf37df37df37df37dULL;
00040 USED_U64(mmx_U_blue)  = 0x4093409340934093ULL;
00041 USED_U64(mmx_V_red)   = 0x3312331233123312ULL;
00042 USED_U64(mmx_V_green) = 0xe5fce5fce5fce5fcULL;
00043 
00044 USED_U64(mmx_mask_f8) = 0xf8f8f8f8f8f8f8f8ULL;
00045 USED_U64(mmx_mask_fc) = 0xfcfcfcfcfcfcfcfcULL;
00046 #undef USED_U64
00047 
00048 /* Use RIP-relative code in PIC mode on amd64 */
00049 #if defined(__x86_64__) && defined(__PIC__)
00050 #   define G "(%%rip)"
00051 #else
00052 #   define G
00053 #endif
00054 
00055 #define MMX_INIT_16 "                                                       \n\
00056 movd      (%1), %%mm0       # Load 4 Cb       00 00 00 00 u3 u2 u1 u0       \n\
00057 movd      (%2), %%mm1       # Load 4 Cr       00 00 00 00 v3 v2 v1 v0       \n\
00058 pxor      %%mm4, %%mm4      # zero mm4                                      \n\
00059 movq      (%0), %%mm6       # Load 8 Y        Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0       \n\
00060 #movl      $0, (%3)         # cache preload for image                       \n\
00061 "
00062 
00063 #define INTRINSICS_INIT_16 \
00064     tmp64 = *(uint32_t *)p_u; \
00065     mm0 = (__m64)tmp64; \
00066     tmp64 = *(uint32_t *)p_v; \
00067     mm1 = (__m64)tmp64; \
00068     mm4 = (__m64)(uint64_t)0; \
00069     mm6 = (__m64)*(uint64_t *)p_y; \
00070     /* *(uint16_t *)p_buffer = 0; */
00071 
00072 #define MMX_INIT_16_GRAY "                                                  \n\
00073 movq      (%0), %%mm6       # Load 8 Y        Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0       \n\
00074 #movl      $0, (%3)         # cache preload for image                       \n\
00075 "
00076 
00077 #define MMX_INIT_32 "                                                       \n\
00078 movd      (%1), %%mm0       # Load 4 Cb       00 00 00 00 u3 u2 u1 u0       \n\
00079 movl      $0, (%3)          # cache preload for image                       \n\
00080 movd      (%2), %%mm1       # Load 4 Cr       00 00 00 00 v3 v2 v1 v0       \n\
00081 pxor      %%mm4, %%mm4      # zero mm4                                      \n\
00082 movq      (%0), %%mm6       # Load 8 Y        Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0       \n\
00083 "
00084 
00085 #define INTRINSICS_INIT_32 \
00086     tmp64 = *(uint32_t *)p_u; \
00087     mm0 = (__m64)tmp64; \
00088     *(uint16_t *)p_buffer = 0; \
00089     tmp64 = *(uint32_t *)p_v; \
00090     mm1 = (__m64)tmp64; \
00091     mm4 = (__m64)(uint64_t)0; \
00092     mm6 = (__m64)*(uint64_t *)p_y;
00093 
00094 /*
00095  * Do the multiply part of the conversion for even and odd pixels,
00096  * register usage:
00097  * mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels,
00098  * mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd  pixels,
00099  * mm6 -> Y even, mm7 -> Y odd
00100  */
00101 
00102 #define MMX_YUV_MUL "                                                       \n\
00103 # convert the chroma part                                                   \n\
00104 punpcklbw %%mm4, %%mm0          # scatter 4 Cb    00 u3 00 u2 00 u1 00 u0   \n\
00105 punpcklbw %%mm4, %%mm1          # scatter 4 Cr    00 v3 00 v2 00 v1 00 v0   \n\
00106 psubsw    mmx_80w"G", %%mm0     # Cb -= 128                                 \n\
00107 psubsw    mmx_80w"G", %%mm1     # Cr -= 128                                 \n\
00108 psllw     $3, %%mm0             # Promote precision                         \n\
00109 psllw     $3, %%mm1             # Promote precision                         \n\
00110 movq      %%mm0, %%mm2          # Copy 4 Cb       00 u3 00 u2 00 u1 00 u0   \n\
00111 movq      %%mm1, %%mm3          # Copy 4 Cr       00 v3 00 v2 00 v1 00 v0   \n\
00112 pmulhw    mmx_U_green"G", %%mm2 # Mul Cb with green coeff -> Cb green       \n\
00113 pmulhw    mmx_V_green"G", %%mm3 # Mul Cr with green coeff -> Cr green       \n\
00114 pmulhw    mmx_U_blue"G", %%mm0  # Mul Cb -> Cblue 00 b3 00 b2 00 b1 00 b0   \n\
00115 pmulhw    mmx_V_red"G", %%mm1   # Mul Cr -> Cred  00 r3 00 r2 00 r1 00 r0   \n\
00116 paddsw    %%mm3, %%mm2          # Cb green + Cr green -> Cgreen             \n\
00117                                                                             \n\
00118 # convert the luma part                                                     \n\
00119 psubusb   mmx_10w"G", %%mm6     # Y -= 16                                   \n\
00120 movq      %%mm6, %%mm7          # Copy 8 Y        Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0   \n\
00121 pand      mmx_00ffw"G", %%mm6   # get Y even      00 Y6 00 Y4 00 Y2 00 Y0   \n\
00122 psrlw     $8, %%mm7             # get Y odd       00 Y7 00 Y5 00 Y3 00 Y1   \n\
00123 psllw     $3, %%mm6             # Promote precision                         \n\
00124 psllw     $3, %%mm7             # Promote precision                         \n\
00125 pmulhw    mmx_Y_coeff"G", %%mm6 # Mul 4 Y even    00 y6 00 y4 00 y2 00 y0   \n\
00126 pmulhw    mmx_Y_coeff"G", %%mm7 # Mul 4 Y odd     00 y7 00 y5 00 y3 00 y1   \n\
00127 "
00128 
00129 #define INTRINSICS_YUV_MUL \
00130     mm0 = _mm_unpacklo_pi8(mm0, mm4); \
00131     mm1 = _mm_unpacklo_pi8(mm1, mm4); \
00132     mm0 = _mm_subs_pi16(mm0, (__m64)mmx_80w); \
00133     mm1 = _mm_subs_pi16(mm1, (__m64)mmx_80w); \
00134     mm0 = _mm_slli_pi16(mm0, 3); \
00135     mm1 = _mm_slli_pi16(mm1, 3); \
00136     mm2 = mm0; \
00137     mm3 = mm1; \
00138     mm2 = _mm_mulhi_pi16(mm2, (__m64)mmx_U_green); \
00139     mm3 = _mm_mulhi_pi16(mm3, (__m64)mmx_V_green); \
00140     mm0 = _mm_mulhi_pi16(mm0, (__m64)mmx_U_blue); \
00141     mm1 = _mm_mulhi_pi16(mm1, (__m64)mmx_V_red); \
00142     mm2 = _mm_adds_pi16(mm2, mm3); \
00143     \
00144     mm6 = _mm_subs_pu8(mm6, (__m64)mmx_10w); \
00145     mm7 = mm6; \
00146     mm6 = _mm_and_si64(mm6, (__m64)mmx_00ffw); \
00147     mm7 = _mm_srli_pi16(mm7, 8); \
00148     mm6 = _mm_slli_pi16(mm6, 3); \
00149     mm7 = _mm_slli_pi16(mm7, 3); \
00150     mm6 = _mm_mulhi_pi16(mm6, (__m64)mmx_Y_coeff); \
00151     mm7 = _mm_mulhi_pi16(mm7, (__m64)mmx_Y_coeff);
00152 
00153 /*
00154  * Do the addition part of the conversion for even and odd pixels,
00155  * register usage:
00156  * mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels,
00157  * mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd  pixels,
00158  * mm6 -> Y even, mm7 -> Y odd
00159  */
00160 
00161 #define MMX_YUV_ADD "                                                       \n\
00162 # Do horizontal and vertical scaling                                        \n\
00163 movq      %%mm0, %%mm3          # Copy Cblue                                \n\
00164 movq      %%mm1, %%mm4          # Copy Cred                                 \n\
00165 movq      %%mm2, %%mm5          # Copy Cgreen                               \n\
00166 paddsw    %%mm6, %%mm0          # Y even + Cblue  00 B6 00 B4 00 B2 00 B0   \n\
00167 paddsw    %%mm7, %%mm3          # Y odd  + Cblue  00 B7 00 B5 00 B3 00 B1   \n\
00168 paddsw    %%mm6, %%mm1          # Y even + Cred   00 R6 00 R4 00 R2 00 R0   \n\
00169 paddsw    %%mm7, %%mm4          # Y odd  + Cred   00 R7 00 R5 00 R3 00 R1   \n\
00170 paddsw    %%mm6, %%mm2          # Y even + Cgreen 00 G6 00 G4 00 G2 00 G0   \n\
00171 paddsw    %%mm7, %%mm5          # Y odd  + Cgreen 00 G7 00 G5 00 G3 00 G1   \n\
00172                                                                             \n\
00173 # Limit RGB even to 0..255                                                  \n\
00174 packuswb  %%mm0, %%mm0          # B6 B4 B2 B0 / B6 B4 B2 B0                 \n\
00175 packuswb  %%mm1, %%mm1          # R6 R4 R2 R0 / R6 R4 R2 R0                 \n\
00176 packuswb  %%mm2, %%mm2          # G6 G4 G2 G0 / G6 G4 G2 G0                 \n\
00177                                                                             \n\
00178 # Limit RGB odd to 0..255                                                   \n\
00179 packuswb  %%mm3, %%mm3          # B7 B5 B3 B1 / B7 B5 B3 B1                 \n\
00180 packuswb  %%mm4, %%mm4          # R7 R5 R3 R1 / R7 R5 R3 R1                 \n\
00181 packuswb  %%mm5, %%mm5          # G7 G5 G3 G1 / G7 G5 G3 G1                 \n\
00182                                                                             \n\
00183 # Interleave RGB even and odd                                               \n\
00184 punpcklbw %%mm3, %%mm0          #                 B7 B6 B5 B4 B3 B2 B1 B0   \n\
00185 punpcklbw %%mm4, %%mm1          #                 R7 R6 R5 R4 R3 R2 R1 R0   \n\
00186 punpcklbw %%mm5, %%mm2          #                 G7 G6 G5 G4 G3 G2 G1 G0   \n\
00187 "
00188 
00189 #define INTRINSICS_YUV_ADD \
00190     mm3 = mm0; \
00191     mm4 = mm1; \
00192     mm5 = mm2; \
00193     mm0 = _mm_adds_pi16(mm0, mm6); \
00194     mm3 = _mm_adds_pi16(mm3, mm7); \
00195     mm1 = _mm_adds_pi16(mm1, mm6); \
00196     mm4 = _mm_adds_pi16(mm4, mm7); \
00197     mm2 = _mm_adds_pi16(mm2, mm6); \
00198     mm5 = _mm_adds_pi16(mm5, mm7); \
00199     \
00200     mm0 = _mm_packs_pu16(mm0, mm0); \
00201     mm1 = _mm_packs_pu16(mm1, mm1); \
00202     mm2 = _mm_packs_pu16(mm2, mm2); \
00203     \
00204     mm3 = _mm_packs_pu16(mm3, mm3); \
00205     mm4 = _mm_packs_pu16(mm4, mm4); \
00206     mm5 = _mm_packs_pu16(mm5, mm5); \
00207     \
00208     mm0 = _mm_unpacklo_pi8(mm0, mm3); \
00209     mm1 = _mm_unpacklo_pi8(mm1, mm4); \
00210     mm2 = _mm_unpacklo_pi8(mm2, mm5);
00211 
00212 /*
00213  * Grayscale case, only use Y
00214  */
00215 
00216 #define MMX_YUV_GRAY "                                                      \n\
00217 # convert the luma part                                                     \n\
00218 psubusb   mmx_10w"G", %%mm6                                                 \n\
00219 movq      %%mm6, %%mm7                                                      \n\
00220 pand      mmx_00ffw"G", %%mm6                                               \n\
00221 psrlw     $8, %%mm7                                                         \n\
00222 psllw     $3, %%mm6                                                         \n\
00223 psllw     $3, %%mm7                                                         \n\
00224 pmulhw    mmx_Y_coeff"G", %%mm6                                             \n\
00225 pmulhw    mmx_Y_coeff"G", %%mm7                                             \n\
00226 packuswb  %%mm6, %%mm6                                                      \n\
00227 packuswb  %%mm7, %%mm7                                                      \n\
00228 punpcklbw %%mm7, %%mm6                                                      \n\
00229 "
00230 
00231 #define MMX_UNPACK_16_GRAY "                                                \n\
00232 movq      %%mm6, %%mm5                                                      \n\
00233 pand      mmx_mask_f8"G", %%mm6                                             \n\
00234 pand      mmx_mask_fc"G", %%mm5                                             \n\
00235 movq      %%mm6, %%mm7                                                      \n\
00236 psrlw     $3, %%mm7                                                         \n\
00237 pxor      %%mm3, %%mm3                                                      \n\
00238 movq      %%mm7, %%mm2                                                      \n\
00239 movq      %%mm5, %%mm0                                                      \n\
00240 punpcklbw %%mm3, %%mm5                                                      \n\
00241 punpcklbw %%mm6, %%mm7                                                      \n\
00242 psllw     $3, %%mm5                                                         \n\
00243 por       %%mm5, %%mm7                                                      \n\
00244 movq      %%mm7, (%3)                                                       \n\
00245 punpckhbw %%mm3, %%mm0                                                      \n\
00246 punpckhbw %%mm6, %%mm2                                                      \n\
00247 psllw     $3, %%mm0                                                         \n\
00248 movq      8(%0), %%mm6                                                      \n\
00249 por       %%mm0, %%mm2                                                      \n\
00250 movq      %%mm2, 8(%3)                                                      \n\
00251 "
00252 
00253 
00254 /*
00255  * convert RGB plane to RGB 15 bits,
00256  * mm0 -> B, mm1 -> R, mm2 -> G,
00257  * mm4 -> GB, mm5 -> AR pixel 4-7,
00258  * mm6 -> GB, mm7 -> AR pixel 0-3
00259  */
00260 
00261 #define MMX_UNPACK_15 "                                                     \n\
00262 # mask unneeded bits off                                                    \n\
00263 pand      mmx_mask_f8"G", %%mm0 # b7b6b5b4 b3______ b7b6b5b4 b3______       \n\
00264 psrlw     $3,%%mm0              # ______b7 b6b5b4b3 ______b7 b6b5b4b3       \n\
00265 pand      mmx_mask_f8"G", %%mm2 # g7g6g5g4 g3______ g7g6g5g4 g3______       \n\
00266 pand      mmx_mask_f8"G", %%mm1 # r7r6r5r4 r3______ r7r6r5r4 r3______       \n\
00267 psrlw     $1,%%mm1              # __r7r6r5 r4r3____ __r7r6r5 r4r3____       \n\
00268 pxor      %%mm4, %%mm4          # zero mm4                                  \n\
00269 movq      %%mm0, %%mm5          # Copy B7-B0                                \n\
00270 movq      %%mm2, %%mm7          # Copy G7-G0                                \n\
00271                                                                             \n\
00272 # convert rgb24 plane to rgb15 pack for pixel 0-3                           \n\
00273 punpcklbw %%mm4, %%mm2          # ________ ________ g7g6g5g4 g3______       \n\
00274 punpcklbw %%mm1, %%mm0          # r7r6r5r4 r3______ ______b7 b6b5b4b3       \n\
00275 psllw     $2,%%mm2              # ________ ____g7g6 g5g4g3__ ________       \n\
00276 por       %%mm2, %%mm0          # r7r6r5r4 r3__g7g6 g5g4g3b7 b6b5b4b3       \n\
00277 movq      8(%0), %%mm6          # Load 8 Y        Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0   \n\
00278 movq      %%mm0, (%3)           # store pixel 0-3                           \n\
00279                                                                             \n\
00280 # convert rgb24 plane to rgb16 pack for pixel 0-3                           \n\
00281 punpckhbw %%mm4, %%mm7          # ________ ________ g7g6g5g4 g3______       \n\
00282 punpckhbw %%mm1, %%mm5          # r7r6r5r4 r3______ ______b7 b6b5b4b3       \n\
00283 psllw     $2,%%mm7              # ________ ____g7g6 g5g4g3__ ________       \n\
00284 movd      4(%1), %%mm0          # Load 4 Cb       __ __ __ __ u3 u2 u1 u0   \n\
00285 por       %%mm7, %%mm5          # r7r6r5r4 r3__g7g6 g5g4g3b7 b6b5b4b3       \n\
00286 movd      4(%2), %%mm1          # Load 4 Cr       __ __ __ __ v3 v2 v1 v0   \n\
00287 movq      %%mm5, 8(%3)          # store pixel 4-7                           \n\
00288 "
00289 
00290 #define INTRINSICS_UNPACK_15 \
00291     mm0 = _mm_and_si64(mm0, (__m64)mmx_mask_f8); \
00292     mm0 = _mm_srli_pi16(mm0, 3); \
00293     mm2 = _mm_and_si64(mm2, (__m64)mmx_mask_f8); \
00294     mm1 = _mm_and_si64(mm1, (__m64)mmx_mask_f8); \
00295     mm1 = _mm_srli_pi16(mm1, 1); \
00296     mm4 = (__m64)(uint64_t)0; \
00297     mm5 = mm0; \
00298     mm7 = mm2; \
00299     \
00300     mm2 = _mm_unpacklo_pi8(mm2, mm4); \
00301     mm0 = _mm_unpacklo_pi8(mm0, mm1); \
00302     mm2 = _mm_slli_pi16(mm2, 2); \
00303     mm0 = _mm_or_si64(mm0, mm2); \
00304     tmp64 = *(uint64_t *)(p_y + 8); \
00305     mm6 = (__m64)tmp64; \
00306     *(uint64_t *)p_buffer = (uint64_t)mm0; \
00307     \
00308     mm7 = _mm_unpackhi_pi8(mm7, mm4); \
00309     mm5 = _mm_unpackhi_pi8(mm5, mm1); \
00310     mm7 = _mm_slli_pi16(mm7, 2); \
00311     tmp64 = (uint64_t)*(uint32_t *)(p_u + 4); \
00312     mm0 = (__m64)tmp64; \
00313     mm5 = _mm_or_si64(mm5, mm7); \
00314     tmp64 = (uint64_t)*(uint32_t *)(p_v + 4); \
00315     mm1 = (__m64)tmp64; \
00316     *(uint64_t *)(p_buffer + 4) = (uint64_t)mm5;
00317 
00318 /*
00319  * convert RGB plane to RGB 16 bits,
00320  * mm0 -> B, mm1 -> R, mm2 -> G,
00321  * mm4 -> GB, mm5 -> AR pixel 4-7,
00322  * mm6 -> GB, mm7 -> AR pixel 0-3
00323  */
00324 
00325 #define MMX_UNPACK_16 "                                                     \n\
00326 # mask unneeded bits off                                                    \n\
00327 pand      mmx_mask_f8"G", %%mm0 # b7b6b5b4 b3______ b7b6b5b4 b3______       \n\
00328 pand      mmx_mask_fc"G", %%mm2 # g7g6g5g4 g3g2____ g7g6g5g4 g3g2____       \n\
00329 pand      mmx_mask_f8"G", %%mm1 # r7r6r5r4 r3______ r7r6r5r4 r3______       \n\
00330 psrlw     $3,%%mm0              # ______b7 b6b5b4b3 ______b7 b6b5b4b3       \n\
00331 pxor      %%mm4, %%mm4          # zero mm4                                  \n\
00332 movq      %%mm0, %%mm5          # Copy B7-B0                                \n\
00333 movq      %%mm2, %%mm7          # Copy G7-G0                                \n\
00334                                                                             \n\
00335 # convert rgb24 plane to rgb16 pack for pixel 0-3                           \n\
00336 punpcklbw %%mm4, %%mm2          # ________ ________ g7g6g5g4 g3g2____       \n\
00337 punpcklbw %%mm1, %%mm0          # r7r6r5r4 r3______ ______b7 b6b5b4b3       \n\
00338 psllw     $3,%%mm2              # ________ __g7g6g5 g4g3g2__ ________       \n\
00339 por       %%mm2, %%mm0          # r7r6r5r4 r3g7g6g5 g4g3g2b7 b6b5b4b3       \n\
00340 movq      8(%0), %%mm6          # Load 8 Y        Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0   \n\
00341 movq      %%mm0, (%3)           # store pixel 0-3                           \n\
00342                                                                             \n\
00343 # convert rgb24 plane to rgb16 pack for pixel 0-3                           \n\
00344 punpckhbw %%mm4, %%mm7          # ________ ________ g7g6g5g4 g3g2____       \n\
00345 punpckhbw %%mm1, %%mm5          # r7r6r5r4 r3______ ______b7 b6b5b4b3       \n\
00346 psllw     $3,%%mm7              # ________ __g7g6g5 g4g3g2__ ________       \n\
00347 movd      4(%1), %%mm0          # Load 4 Cb       __ __ __ __ u3 u2 u1 u0   \n\
00348 por       %%mm7, %%mm5          # r7r6r5r4 r3g7g6g5 g4g3g2b7 b6b5b4b3       \n\
00349 movd      4(%2), %%mm1          # Load 4 Cr       __ __ __ __ v3 v2 v1 v0   \n\
00350 movq      %%mm5, 8(%3)          # store pixel 4-7                           \n\
00351 "
00352 
00353 #define INTRINSICS_UNPACK_16 \
00354     mm0 = _mm_and_si64(mm0, (__m64)mmx_mask_f8); \
00355     mm2 = _mm_and_si64(mm2, (__m64)mmx_mask_fc); \
00356     mm1 = _mm_and_si64(mm1, (__m64)mmx_mask_f8); \
00357     mm0 = _mm_srli_pi16(mm0, 3); \
00358     mm4 = (__m64)(uint64_t)0; \
00359     mm5 = mm0; \
00360     mm7 = mm2; \
00361     \
00362     mm2 = _mm_unpacklo_pi8(mm2, mm4); \
00363     mm0 = _mm_unpacklo_pi8(mm0, mm1); \
00364     mm2 = _mm_slli_pi16(mm2, 3); \
00365     mm0 = _mm_or_si64(mm0, mm2); \
00366     tmp64 = *(uint64_t *)(p_y + 8); \
00367     mm6 = (__m64)tmp64; \
00368     *(uint64_t *)p_buffer = (uint64_t)mm0; \
00369     \
00370     mm7 = _mm_unpackhi_pi8(mm7, mm4); \
00371     mm5 = _mm_unpackhi_pi8(mm5, mm1); \
00372     mm7 = _mm_slli_pi16(mm7, 3); \
00373     tmp64 = (uint64_t)*(uint32_t *)(p_u + 4); \
00374     mm0 = (__m64)tmp64; \
00375     mm5 = _mm_or_si64(mm5, mm7); \
00376     tmp64 = (uint64_t)*(uint32_t *)(p_v + 4); \
00377     mm1 = (__m64)tmp64; \
00378     *(uint64_t *)(p_buffer + 4) = (uint64_t)mm5;
00379 
00380 /*
00381  * convert RGB plane to RGB packed format,
00382  * mm0 -> B, mm1 -> R, mm2 -> G, mm3 -> 0,
00383  * mm4 -> GB, mm5 -> AR pixel 4-7,
00384  * mm6 -> GB, mm7 -> AR pixel 0-3
00385  */
00386 
00387 #define MMX_UNPACK_32 "                                                     \n\
00388 pxor      %%mm3, %%mm3  # zero mm3                                          \n\
00389 movq      %%mm0, %%mm6  #                 B7 B6 B5 B4 B3 B2 B1 B0           \n\
00390 movq      %%mm1, %%mm7  #                 R7 R6 R5 R4 R3 R2 R1 R0           \n\
00391 movq      %%mm0, %%mm4  #                 B7 B6 B5 B4 B3 B2 B1 B0           \n\
00392 movq      %%mm1, %%mm5  #                 R7 R6 R5 R4 R3 R2 R1 R0           \n\
00393 punpcklbw %%mm2, %%mm6  #                 G3 B3 G2 B2 G1 B1 G0 B0           \n\
00394 punpcklbw %%mm3, %%mm7  #                 00 R3 00 R2 00 R1 00 R0           \n\
00395 punpcklwd %%mm7, %%mm6  #                 00 R1 B1 G1 00 R0 B0 G0           \n\
00396 movq      %%mm6, (%3)   # Store ARGB1 ARGB0                                 \n\
00397 movq      %%mm0, %%mm6  #                 B7 B6 B5 B4 B3 B2 B1 B0           \n\
00398 punpcklbw %%mm2, %%mm6  #                 G3 B3 G2 B2 G1 B1 G0 B0           \n\
00399 punpckhwd %%mm7, %%mm6  #                 00 R3 G3 B3 00 R2 B3 G2           \n\
00400 movq      %%mm6, 8(%3)  # Store ARGB3 ARGB2                                 \n\
00401 punpckhbw %%mm2, %%mm4  #                 G7 B7 G6 B6 G5 B5 G4 B4           \n\
00402 punpckhbw %%mm3, %%mm5  #                 00 R7 00 R6 00 R5 00 R4           \n\
00403 punpcklwd %%mm5, %%mm4  #                 00 R5 B5 G5 00 R4 B4 G4           \n\
00404 movq      %%mm4, 16(%3) # Store ARGB5 ARGB4                                 \n\
00405 movq      %%mm0, %%mm4  #                 B7 B6 B5 B4 B3 B2 B1 B0           \n\
00406 punpckhbw %%mm2, %%mm4  #                 G7 B7 G6 B6 G5 B5 G4 B4           \n\
00407 punpckhwd %%mm5, %%mm4  #                 00 R7 G7 B7 00 R6 B6 G6           \n\
00408 movq      %%mm4, 24(%3) # Store ARGB7 ARGB6                                 \n\
00409                                                                             \n\
00410 #movd      4(%1), %%mm0  # Load 4 Cb       00 00 00 00 u3 u2 u1 u0           \n\
00411 #movd      4(%2), %%mm1  # Load 4 Cr       00 00 00 00 v3 v2 v1 v0           \n\
00412 #pxor      %%mm4, %%mm4  # zero mm4                                          \n\
00413 #movq      8(%0), %%mm6  # Load 8 Y        Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0           \n\
00414 "
00415 
00416 #define INTRINSICS_UNPACK_32 \
00417     mm3 = (__m64)(uint64_t)0; \
00418     mm6 = mm0; \
00419     mm7 = mm1; \
00420     mm4 = mm0; \
00421     mm5 = mm1; \
00422     mm6 = _mm_unpacklo_pi8(mm6, mm2); \
00423     mm7 = _mm_unpacklo_pi8(mm7, mm3); \
00424     mm6 = _mm_unpacklo_pi16(mm6, mm7); \
00425     *(uint64_t *)p_buffer = (uint64_t)mm6; \
00426     mm6 = mm0; \
00427     mm6 = _mm_unpacklo_pi8(mm6, mm2); \
00428     mm6 = _mm_unpackhi_pi16(mm6, mm7); \
00429     *(uint64_t *)(p_buffer + 2) = (uint64_t)mm6; \
00430     mm4 = _mm_unpackhi_pi8(mm4, mm2); \
00431     mm5 = _mm_unpackhi_pi8(mm5, mm3); \
00432     mm4 = _mm_unpacklo_pi16(mm4, mm5); \
00433     *(uint64_t *)(p_buffer + 4) = (uint64_t)mm4; \
00434     mm4 = mm0; \
00435     mm4 = _mm_unpackhi_pi8(mm4, mm2); \
00436     mm4 = _mm_unpackhi_pi16(mm4, mm5); \
00437     *(uint64_t *)(p_buffer + 6) = (uint64_t)mm4; \
00438