i420_rgb_mmx.h

uclinux 下的vlc播放器源代码

psrlw     $8, %%mm7                                                         \n\
psllw     $3, %%mm6                                                         \n\
psllw     $3, %%mm7                                                         \n\
pmulhw    mmx_Y_coeff"G", %%mm6                                             \n\
pmulhw    mmx_Y_coeff"G", %%mm7                                             \n\
packuswb  %%mm6, %%mm6                                                      \n\
packuswb  %%mm7, %%mm7                                                      \n\
punpcklbw %%mm7, %%mm6                                                      \n\
"

#define MMX_UNPACK_16_GRAY "                                                \n\
movq      %%mm6, %%mm5                                                      \n\
pand      mmx_mask_f8"G", %%mm6                                             \n\
pand      mmx_mask_fc"G", %%mm5                                             \n\
movq      %%mm6, %%mm7                                                      \n\
psrlw     $3, %%mm7                                                         \n\
pxor      %%mm3, %%mm3                                                      \n\
movq      %%mm7, %%mm2                                                      \n\
movq      %%mm5, %%mm0                                                      \n\
punpcklbw %%mm3, %%mm5                                                      \n\
punpcklbw %%mm6, %%mm7                                                      \n\
psllw     $3, %%mm5                                                         \n\
por       %%mm5, %%mm7                                                      \n\
movq      %%mm7, (%3)                                                       \n\
punpckhbw %%mm3, %%mm0                                                      \n\
punpckhbw %%mm6, %%mm2                                                      \n\
psllw     $3, %%mm0                                                         \n\
movq      8(%0), %%mm6                                                      \n\
por       %%mm0, %%mm2                                                      \n\
movq      %%mm2, 8(%3)                                                      \n\
"


/*
 * convert RGB plane to RGB 15 bits,
 * mm0 -> B, mm1 -> R, mm2 -> G,
 * mm4 -> GB, mm5 -> AR pixel 4-7,
 * mm6 -> GB, mm7 -> AR pixel 0-3
 */

#define MMX_UNPACK_15 "                                                     \n\
# mask unneeded bits off                                                    \n\
pand      mmx_mask_f8"G", %%mm0 # b7b6b5b4 b3______ b7b6b5b4 b3______       \n\
psrlw     $3,%%mm0              # ______b7 b6b5b4b3 ______b7 b6b5b4b3       \n\
pand      mmx_mask_f8"G", %%mm2 # g7g6g5g4 g3______ g7g6g5g4 g3______       \n\
pand      mmx_mask_f8"G", %%mm1 # r7r6r5r4 r3______ r7r6r5r4 r3______       \n\
psrlw     $1,%%mm1              # __r7r6r5 r4r3____ __r7r6r5 r4r3____       \n\
pxor      %%mm4, %%mm4          # zero mm4                                  \n\
movq      %%mm0, %%mm5          # Copy B7-B0                                \n\
movq      %%mm2, %%mm7          # Copy G7-G0                                \n\
                                                                            \n\
# convert rgb24 plane to rgb15 pack for pixel 0-3                           \n\
punpcklbw %%mm4, %%mm2          # ________ ________ g7g6g5g4 g3______       \n\
punpcklbw %%mm1, %%mm0          # r7r6r5r4 r3______ ______b7 b6b5b4b3       \n\
psllw     $2,%%mm2              # ________ ____g7g6 g5g4g3__ ________       \n\
por       %%mm2, %%mm0          # r7r6r5r4 r3__g7g6 g5g4g3b7 b6b5b4b3       \n\
movq      8(%0), %%mm6          # Load 8 Y        Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0   \n\
movq      %%mm0, (%3)           # store pixel 0-3                           \n\
                                                                            \n\
# convert rgb24 plane to rgb16 pack for pixel 0-3                           \n\
punpckhbw %%mm4, %%mm7          # ________ ________ g7g6g5g4 g3______       \n\
punpckhbw %%mm1, %%mm5          # r7r6r5r4 r3______ ______b7 b6b5b4b3       \n\
psllw     $2,%%mm7              # ________ ____g7g6 g5g4g3__ ________       \n\
movd      4(%1), %%mm0          # Load 4 Cb       __ __ __ __ u3 u2 u1 u0   \n\
por       %%mm7, %%mm5          # r7r6r5r4 r3__g7g6 g5g4g3b7 b6b5b4b3       \n\
movd      4(%2), %%mm1          # Load 4 Cr       __ __ __ __ v3 v2 v1 v0   \n\
movq      %%mm5, 8(%3)          # store pixel 4-7                           \n\
"

#define INTRINSICS_UNPACK_15 \
    mm0 = _mm_and_si64(mm0, (__m64)mmx_mask_f8); \
    mm0 = _mm_srli_pi16(mm0, 3); \
    mm2 = _mm_and_si64(mm2, (__m64)mmx_mask_f8); \
    mm1 = _mm_and_si64(mm1, (__m64)mmx_mask_f8); \
    mm1 = _mm_srli_pi16(mm1, 1); \
    mm4 = (__m64)(uint64_t)0; \
    mm5 = mm0; \
    mm7 = mm2; \
    \
    mm2 = _mm_unpacklo_pi8(mm2, mm4); \
    mm0 = _mm_unpacklo_pi8(mm0, mm1); \
    mm2 = _mm_slli_pi16(mm2, 2); \
    mm0 = _mm_or_si64(mm0, mm2); \
    tmp64 = *(uint64_t *)(p_y + 8); \
    mm6 = (__m64)tmp64; \
    *(uint64_t *)p_buffer = (uint64_t)mm0; \
    \
    mm7 = _mm_unpackhi_pi8(mm7, mm4); \
    mm5 = _mm_unpackhi_pi8(mm5, mm1); \
    mm7 = _mm_slli_pi16(mm7, 2); \
    tmp64 = (uint64_t)*(uint32_t *)(p_u + 4); \
    mm0 = (__m64)tmp64; \
    mm5 = _mm_or_si64(mm5, mm7); \
    tmp64 = (uint64_t)*(uint32_t *)(p_v + 4); \
    mm1 = (__m64)tmp64; \
    *(uint64_t *)(p_buffer + 4) = (uint64_t)mm5;

/*
 * convert RGB plane to RGB 16 bits,
 * mm0 -> B, mm1 -> R, mm2 -> G,
 * mm4 -> GB, mm5 -> AR pixel 4-7,
 * mm6 -> GB, mm7 -> AR pixel 0-3
 */

#define MMX_UNPACK_16 "                                                     \n\
# mask unneeded bits off                                                    \n\
pand      mmx_mask_f8"G", %%mm0 # b7b6b5b4 b3______ b7b6b5b4 b3______       \n\
pand      mmx_mask_fc"G", %%mm2 # g7g6g5g4 g3g2____ g7g6g5g4 g3g2____       \n\
pand      mmx_mask_f8"G", %%mm1 # r7r6r5r4 r3______ r7r6r5r4 r3______       \n\
psrlw     $3,%%mm0              # ______b7 b6b5b4b3 ______b7 b6b5b4b3       \n\
pxor      %%mm4, %%mm4          # zero mm4                                  \n\
movq      %%mm0, %%mm5          # Copy B7-B0                                \n\
movq      %%mm2, %%mm7          # Copy G7-G0                                \n\
                                                                            \n\
# convert rgb24 plane to rgb16 pack for pixel 0-3                           \n\
punpcklbw %%mm4, %%mm2          # ________ ________ g7g6g5g4 g3g2____       \n\
punpcklbw %%mm1, %%mm0          # r7r6r5r4 r3______ ______b7 b6b5b4b3       \n\
psllw     $3,%%mm2              # ________ __g7g6g5 g4g3g2__ ________       \n\
por       %%mm2, %%mm0          # r7r6r5r4 r3g7g6g5 g4g3g2b7 b6b5b4b3       \n\
movq      8(%0), %%mm6          # Load 8 Y        Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0   \n\
movq      %%mm0, (%3)           # store pixel 0-3                           \n\
                                                                            \n\
# convert rgb24 plane to rgb16 pack for pixel 0-3                           \n\
punpckhbw %%mm4, %%mm7          # ________ ________ g7g6g5g4 g3g2____       \n\
punpckhbw %%mm1, %%mm5          # r7r6r5r4 r3______ ______b7 b6b5b4b3       \n\
psllw     $3,%%mm7              # ________ __g7g6g5 g4g3g2__ ________       \n\
movd      4(%1), %%mm0          # Load 4 Cb       __ __ __ __ u3 u2 u1 u0   \n\
por       %%mm7, %%mm5          # r7r6r5r4 r3g7g6g5 g4g3g2b7 b6b5b4b3       \n\
movd      4(%2), %%mm1          # Load 4 Cr       __ __ __ __ v3 v2 v1 v0   \n\
movq      %%mm5, 8(%3)          # store pixel 4-7                           \n\
"

#define INTRINSICS_UNPACK_16 \
    mm0 = _mm_and_si64(mm0, (__m64)mmx_mask_f8); \
    mm2 = _mm_and_si64(mm2, (__m64)mmx_mask_fc); \
    mm1 = _mm_and_si64(mm1, (__m64)mmx_mask_f8); \
    mm0 = _mm_srli_pi16(mm0, 3); \
    mm4 = (__m64)(uint64_t)0; \
    mm5 = mm0; \
    mm7 = mm2; \
    \
    mm2 = _mm_unpacklo_pi8(mm2, mm4); \
    mm0 = _mm_unpacklo_pi8(mm0, mm1); \
    mm2 = _mm_slli_pi16(mm2, 3); \
    mm0 = _mm_or_si64(mm0, mm2); \
    tmp64 = *(uint64_t *)(p_y + 8); \
    mm6 = (__m64)tmp64; \
    *(uint64_t *)p_buffer = (uint64_t)mm0; \
    \
    mm7 = _mm_unpackhi_pi8(mm7, mm4); \
    mm5 = _mm_unpackhi_pi8(mm5, mm1); \
    mm7 = _mm_slli_pi16(mm7, 3); \
    tmp64 = (uint64_t)*(uint32_t *)(p_u + 4); \
    mm0 = (__m64)tmp64; \
    mm5 = _mm_or_si64(mm5, mm7); \
    tmp64 = (uint64_t)*(uint32_t *)(p_v + 4); \
    mm1 = (__m64)tmp64; \
    *(uint64_t *)(p_buffer + 4) = (uint64_t)mm5;

/*
 * convert RGB plane to RGB packed format,
 * mm0 -> B, mm1 -> R, mm2 -> G, mm3 -> 0,
 * mm4 -> GB, mm5 -> AR pixel 4-7,
 * mm6 -> GB, mm7 -> AR pixel 0-3
 */

#define MMX_UNPACK_32 "                                                     \n\
pxor      %%mm3, %%mm3  # zero mm3                                          \n\
movq      %%mm0, %%mm6  #                 B7 B6 B5 B4 B3 B2 B1 B0           \n\
movq      %%mm1, %%mm7  #                 R7 R6 R5 R4 R3 R2 R1 R0           \n\
movq      %%mm0, %%mm4  #                 B7 B6 B5 B4 B3 B2 B1 B0           \n\
movq      %%mm1, %%mm5  #                 R7 R6 R5 R4 R3 R2 R1 R0           \n\
punpcklbw %%mm2, %%mm6  #                 G3 B3 G2 B2 G1 B1 G0 B0           \n\
punpcklbw %%mm3, %%mm7  #                 00 R3 00 R2 00 R1 00 R0           \n\
punpcklwd %%mm7, %%mm6  #                 00 R1 B1 G1 00 R0 B0 G0           \n\
movq      %%mm6, (%3)   # Store ARGB1 ARGB0                                 \n\
movq      %%mm0, %%mm6  #                 B7 B6 B5 B4 B3 B2 B1 B0           \n\
punpcklbw %%mm2, %%mm6  #                 G3 B3 G2 B2 G1 B1 G0 B0           \n\
punpckhwd %%mm7, %%mm6  #                 00 R3 G3 B3 00 R2 B3 G2           \n\
movq      %%mm6, 8(%3)  # Store ARGB3 ARGB2                                 \n\
punpckhbw %%mm2, %%mm4  #                 G7 B7 G6 B6 G5 B5 G4 B4           \n\
punpckhbw %%mm3, %%mm5  #                 00 R7 00 R6 00 R5 00 R4           \n\
punpcklwd %%mm5, %%mm4  #                 00 R5 B5 G5 00 R4 B4 G4           \n\
movq      %%mm4, 16(%3) # Store ARGB5 ARGB4                                 \n\
movq      %%mm0, %%mm4  #                 B7 B6 B5 B4 B3 B2 B1 B0           \n\
punpckhbw %%mm2, %%mm4  #                 G7 B7 G6 B6 G5 B5 G4 B4           \n\
punpckhwd %%mm5, %%mm4  #                 00 R7 G7 B7 00 R6 B6 G6           \n\
movq      %%mm4, 24(%3) # Store ARGB7 ARGB6                                 \n\
                                                                            \n\
#movd      4(%1), %%mm0  # Load 4 Cb       00 00 00 00 u3 u2 u1 u0           \n\
#movd      4(%2), %%mm1  # Load 4 Cr       00 00 00 00 v3 v2 v1 v0           \n\
#pxor      %%mm4, %%mm4  # zero mm4                                          \n\
#movq      8(%0), %%mm6  # Load 8 Y        Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0           \n\
"

#define INTRINSICS_UNPACK_32 \
    mm3 = (__m64)(uint64_t)0; \
    mm6 = mm0; \
    mm7 = mm1; \
    mm4 = mm0; \
    mm5 = mm1; \
    mm6 = _mm_unpacklo_pi8(mm6, mm2); \
    mm7 = _mm_unpacklo_pi8(mm7, mm3); \
    mm6 = _mm_unpacklo_pi16(mm6, mm7); \
    *(uint64_t *)p_buffer = (uint64_t)mm6; \
    mm6 = mm0; \
    mm6 = _mm_unpacklo_pi8(mm6, mm2); \
    mm6 = _mm_unpackhi_pi16(mm6, mm7); \
    *(uint64_t *)(p_buffer + 2) = (uint64_t)mm6; \
    mm4 = _mm_unpackhi_pi8(mm4, mm2); \
    mm5 = _mm_unpackhi_pi8(mm5, mm3); \
    mm4 = _mm_unpacklo_pi16(mm4, mm5); \
    *(uint64_t *)(p_buffer + 4) = (uint64_t)mm4; \
    mm4 = mm0; \
    mm4 = _mm_unpackhi_pi8(mm4, mm2); \
    mm4 = _mm_unpackhi_pi16(mm4, mm5); \
    *(uint64_t *)(p_buffer + 6) = (uint64_t)mm4; \