h264dsp_mmx.c

tcpmp播放器的flv插件

static void h264_h_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
{
    //FIXME: could cut some load/stores by merging transpose with filter
    uint8_t trans[8*4];
    transpose4x4(trans, pix-2, 8, stride);
    transpose4x4(trans+4, pix-2+4*stride, 8, stride);
    h264_loop_filter_chroma_mmx2(trans+2*8, 8, alpha-1, beta-1, tc0);
    transpose4x4(pix-2, trans, stride, 8);
    transpose4x4(pix-2+4*stride, trans+4, stride, 8);
}

// p0 = (p0 + q1 + 2*p1 + 2) >> 2
#define H264_FILTER_CHROMA4(p0, p1, q1, one) \
    "movq    "#p0", %%mm4  \n\t"\
    "pxor    "#q1", %%mm4  \n\t"\
    "pand   "#one", %%mm4  \n\t" /* mm4 = (p0^q1)&1 */\
    "pavgb   "#q1", "#p0"  \n\t"\
    "psubusb %%mm4, "#p0"  \n\t"\
    "pavgb   "#p1", "#p0"  \n\t" /* dst = avg(p1, avg(p0,q1) - ((p0^q1)&1)) */\

static inline void h264_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha1, int beta1)
{
    asm volatile(
        "movq    (%0),    %%mm0     \n\t"
        "movq    (%0,%2), %%mm1     \n\t"
        "movq    (%1),    %%mm2     \n\t"
        "movq    (%1,%2), %%mm3     \n\t"
        H264_DEBLOCK_MASK(%3, %4)
        "movq    %%mm1,   %%mm5     \n\t"
        "movq    %%mm2,   %%mm6     \n\t"
        H264_FILTER_CHROMA4(%%mm1, %%mm0, %%mm3, %5) //p0'
        H264_FILTER_CHROMA4(%%mm2, %%mm3, %%mm0, %5) //q0'
        "psubb   %%mm5,   %%mm1     \n\t"
        "psubb   %%mm6,   %%mm2     \n\t"
        "pand    %%mm7,   %%mm1     \n\t"
        "pand    %%mm7,   %%mm2     \n\t"
        "paddb   %%mm5,   %%mm1     \n\t"
        "paddb   %%mm6,   %%mm2     \n\t"
        "movq    %%mm1,   (%0,%2)   \n\t"
        "movq    %%mm2,   (%1)      \n\t"
        :: "r"(pix-2*stride), "r"(pix), "r"((long)stride),
           "m"(alpha1), "m"(beta1), "m"(mm_bone)
    );
}

static void h264_v_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
{
    h264_loop_filter_chroma_intra_mmx2(pix, stride, alpha-1, beta-1);
}

static void h264_h_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
{
    //FIXME: could cut some load/stores by merging transpose with filter
    uint8_t trans[8*4];
    transpose4x4(trans, pix-2, 8, stride);
    transpose4x4(trans+4, pix-2+4*stride, 8, stride);
    h264_loop_filter_chroma_intra_mmx2(trans+2*8, 8, alpha-1, beta-1);
    transpose4x4(pix-2, trans, stride, 8);
    transpose4x4(pix-2+4*stride, trans+4, stride, 8);
}


/***********************************/
/* motion compensation */

#define QPEL_H264V(A,B,C,D,E,F,OP)\
        "movd (%0), "#F"		\n\t"\
        "movq "#C", %%mm6		\n\t"\
        "paddw "#D", %%mm6		\n\t"\
        "psllw $2, %%mm6		\n\t"\
        "psubw "#B", %%mm6		\n\t"\
        "psubw "#E", %%mm6		\n\t"\
        "pmullw %4, %%mm6		\n\t"\
        "add %2, %0			\n\t"\
        "punpcklbw %%mm7, "#F"		\n\t"\
        "paddw %5, "#A"			\n\t"\
        "paddw "#F", "#A"		\n\t"\
        "paddw "#A", %%mm6		\n\t"\
        "psraw $5, %%mm6		\n\t"\
        "packuswb %%mm6, %%mm6		\n\t"\
        OP(%%mm6, (%1), A, d)\
        "add %3, %1			\n\t"     

#define QPEL_H264HV(A,B,C,D,E,F,OF)\
        "movd (%0), "#F"		\n\t"\
        "movq "#C", %%mm6		\n\t"\
        "paddw "#D", %%mm6		\n\t"\
        "psllw $2, %%mm6		\n\t"\
        "psubw "#B", %%mm6		\n\t"\
        "psubw "#E", %%mm6		\n\t"\
        "pmullw %3, %%mm6		\n\t"\
        "add %2, %0			\n\t"\
        "punpcklbw %%mm7, "#F"		\n\t"\
        "paddw "#F", "#A"		\n\t"\
        "paddw "#A", %%mm6		\n\t"\
        "movq %%mm6, "#OF"(%1)		\n\t"
        
#define QPEL_H264(OPNAME, OP, MMX)\
static void OPNAME ## h264_qpel4_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    int h=4;\
\
    asm volatile(\
        "pxor %%mm7, %%mm7		\n\t"\
        "movq %5, %%mm4			\n\t"\
        "movq %6, %%mm5			\n\t"\
        "1:				\n\t"\
        "movd  -1(%0), %%mm1		\n\t"\
        "movd    (%0), %%mm2		\n\t"\
        "movd   1(%0), %%mm3		\n\t"\
        "movd   2(%0), %%mm0		\n\t"\
        "punpcklbw %%mm7, %%mm1		\n\t"\
        "punpcklbw %%mm7, %%mm2		\n\t"\
        "punpcklbw %%mm7, %%mm3		\n\t"\
        "punpcklbw %%mm7, %%mm0		\n\t"\
        "paddw %%mm0, %%mm1		\n\t"\
        "paddw %%mm3, %%mm2		\n\t"\
        "movd  -2(%0), %%mm0		\n\t"\
        "movd   3(%0), %%mm3		\n\t"\
        "punpcklbw %%mm7, %%mm0		\n\t"\
        "punpcklbw %%mm7, %%mm3		\n\t"\
        "paddw %%mm3, %%mm0		\n\t"\
        "psllw $2, %%mm2		\n\t"\
        "psubw %%mm1, %%mm2		\n\t"\
        "pmullw %%mm4, %%mm2		\n\t"\
        "paddw %%mm5, %%mm0		\n\t"\
        "paddw %%mm2, %%mm0		\n\t"\
        "psraw $5, %%mm0		\n\t"\
        "packuswb %%mm0, %%mm0		\n\t"\
        OP(%%mm0, (%1),%%mm6, d)\
        "add %3, %0			\n\t"\
        "add %4, %1			\n\t"\
        "decl %2			\n\t"\
        " jnz 1b			\n\t"\
        : "+a"(src), "+c"(dst), "+m"(h)\
        : "d"((long)srcStride), "S"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
        : "memory"\
    );\
}\
static void OPNAME ## h264_qpel4_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    src -= 2*srcStride;\
    asm volatile(\
        "pxor %%mm7, %%mm7		\n\t"\
        "movd (%0), %%mm0		\n\t"\
        "add %2, %0			\n\t"\
        "movd (%0), %%mm1		\n\t"\
        "add %2, %0			\n\t"\
        "movd (%0), %%mm2		\n\t"\
        "add %2, %0			\n\t"\
        "movd (%0), %%mm3		\n\t"\
        "add %2, %0			\n\t"\
        "movd (%0), %%mm4		\n\t"\
        "add %2, %0			\n\t"\
        "punpcklbw %%mm7, %%mm0		\n\t"\
        "punpcklbw %%mm7, %%mm1		\n\t"\
        "punpcklbw %%mm7, %%mm2		\n\t"\
        "punpcklbw %%mm7, %%mm3		\n\t"\
        "punpcklbw %%mm7, %%mm4		\n\t"\
        QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
        QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
        QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
        QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
         \
        : "+a"(src), "+c"(dst)\
        : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
        : "memory"\
    );\
}\
static void OPNAME ## h264_qpel4_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
    int h=4;\
    int w=3;\
    src -= 2*srcStride+2;\
    while(w--){\
        asm volatile(\
            "pxor %%mm7, %%mm7			\n\t"\
            "movd (%0), %%mm0			\n\t"\
            "add %2, %0				\n\t"\
            "movd (%0), %%mm1			\n\t"\
            "add %2, %0				\n\t"\
            "movd (%0), %%mm2			\n\t"\
            "add %2, %0				\n\t"\
            "movd (%0), %%mm3			\n\t"\
            "add %2, %0				\n\t"\
            "movd (%0), %%mm4			\n\t"\
            "add %2, %0				\n\t"\
            "punpcklbw %%mm7, %%mm0		\n\t"\
            "punpcklbw %%mm7, %%mm1		\n\t"\
            "punpcklbw %%mm7, %%mm2		\n\t"\
            "punpcklbw %%mm7, %%mm3		\n\t"\
            "punpcklbw %%mm7, %%mm4		\n\t"\
            QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*8*3)\
            QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*8*3)\
            QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*8*3)\
            QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*8*3)\
             \
            : "+a"(src)\
            : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5)\
            : "memory"\
        );\
        tmp += 4;\
        src += 4 - 9*srcStride;\
    }\
    tmp -= 3*4;\
    asm volatile(\
        "movq %4, %%mm6			\n\t"\
        "1:				\n\t"\
        "movq     (%0), %%mm0		\n\t"\
        "paddw  10(%0), %%mm0		\n\t"\
        "movq    2(%0), %%mm1		\n\t"\
        "paddw   8(%0), %%mm1		\n\t"\
        "movq    4(%0), %%mm2		\n\t"\
        "paddw   6(%0), %%mm2		\n\t"\
        "psubw %%mm1, %%mm0		\n\t"/*a-b   (abccba)*/\
        "psraw $2, %%mm0		\n\t"/*(a-b)/4 */\
        "psubw %%mm1, %%mm0		\n\t"/*(a-b)/4-b */\
        "paddsw %%mm2, %%mm0		\n\t"\
        "psraw $2, %%mm0		\n\t"/*((a-b)/4-b)/4 */\
        "paddw %%mm6, %%mm2		\n\t"\
        "paddw %%mm2, %%mm0		\n\t"\
        "psraw $6, %%mm0		\n\t"\
        "packuswb %%mm0, %%mm0		\n\t"\
        OP(%%mm0, (%1),%%mm7, d)\
        "add $24, %0			\n\t"\
        "add %3, %1			\n\t"\
        "decl %2			\n\t"\
        " jnz 1b			\n\t"\
        : "+a"(tmp), "+c"(dst), "+m"(h)\
        : "S"((long)dstStride), "m"(ff_pw_32)\
        : "memory"\
    );\
}\
\
static void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    int h=8;\
    asm volatile(\
        "pxor %%mm7, %%mm7		\n\t"\
        "movq %5, %%mm6			\n\t"\
        "1:				\n\t"\
        "movq    (%0), %%mm0		\n\t"\
        "movq   1(%0), %%mm2		\n\t"\
        "movq %%mm0, %%mm1		\n\t"\
        "movq %%mm2, %%mm3		\n\t"\
        "punpcklbw %%mm7, %%mm0		\n\t"\
        "punpckhbw %%mm7, %%mm1		\n\t"\
        "punpcklbw %%mm7, %%mm2		\n\t"\
        "punpckhbw %%mm7, %%mm3		\n\t"\
        "paddw %%mm2, %%mm0		\n\t"\
        "paddw %%mm3, %%mm1		\n\t"\
        "psllw $2, %%mm0		\n\t"\
        "psllw $2, %%mm1		\n\t"\
        "movq   -1(%0), %%mm2		\n\t"\
        "movq    2(%0), %%mm4		\n\t"\
        "movq %%mm2, %%mm3		\n\t"\
        "movq %%mm4, %%mm5		\n\t"\
        "punpcklbw %%mm7, %%mm2		\n\t"\
        "punpckhbw %%mm7, %%mm3		\n\t"\
        "punpcklbw %%mm7, %%mm4		\n\t"\
        "punpckhbw %%mm7, %%mm5		\n\t"\
        "paddw %%mm4, %%mm2		\n\t"\
        "paddw %%mm3, %%mm5		\n\t"\
        "psubw %%mm2, %%mm0		\n\t"\
        "psubw %%mm5, %%mm1		\n\t"\
        "pmullw %%mm6, %%mm0		\n\t"\
        "pmullw %%mm6, %%mm1		\n\t"\
        "movd   -2(%0), %%mm2		\n\t"\
        "movd    7(%0), %%mm5		\n\t"\
        "punpcklbw %%mm7, %%mm2		\n\t"\
        "punpcklbw %%mm7, %%mm5		\n\t"\
        "paddw %%mm3, %%mm2		\n\t"\
        "paddw %%mm5, %%mm4		\n\t"\
        "movq %6, %%mm5			\n\t"\
        "paddw %%mm5, %%mm2		\n\t"\
        "paddw %%mm5, %%mm4		\n\t"\
        "paddw %%mm2, %%mm0		\n\t"\
        "paddw %%mm4, %%mm1		\n\t"\
        "psraw $5, %%mm0		\n\t"\
        "psraw $5, %%mm1		\n\t"\
        "packuswb %%mm1, %%mm0		\n\t"\
        OP(%%mm0, (%1),%%mm5, q)\
        "add %3, %0			\n\t"\
        "add %4, %1			\n\t"\
        "decl %2			\n\t"\
        " jnz 1b			\n\t"\
        : "+a"(src), "+c"(dst), "+m"(h)\
        : "d"((long)srcStride), "S"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
        : "memory"\
    );\
}\
\
static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    int h= 2;\
    src -= 2*srcStride;\
    \
    while(h--){\
      asm volatile(\
        "pxor %%mm7, %%mm7		\n\t"\
        "movd (%0), %%mm0		\n\t"\
        "add %2, %0			\n\t"\
        "movd (%0), %%mm1		\n\t"\
        "add %2, %0			\n\t"\
        "movd (%0), %%mm2		\n\t"\
        "add %2, %0			\n\t"\
        "movd (%0), %%mm3		\n\t"\
        "add %2, %0			\n\t"\