dsputil_mmx.c

Trolltech公司发布的图形界面操作系统。可在qt-embedded-2.3.10平台上编译为嵌入式图形界面操作系统。

static void OPNAME ## qpel8_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t half[8 + 9];\
    uint8_t * const halfH= ((uint8_t*)half);\
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
    put ## RND ## pixels8_l2_mmx(halfH, src+1, halfH, 8, stride, 9);\
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
}\
static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t half[9];\
    uint8_t * const halfH= ((uint8_t*)half);\
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
}\
static void OPNAME ## qpel16_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
    OPNAME ## pixels16_mmx(dst, src, stride, 16);\
}\
\
static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t temp[32];\
    uint8_t * const half= (uint8_t*)temp;\
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\
    OPNAME ## pixels16_l2_mmx(dst, src, half, stride, stride, 16);\
}\
\
static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src, stride, stride, 16);\
}\
\
static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t temp[32];\
    uint8_t * const half= (uint8_t*)temp;\
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\
    OPNAME ## pixels16_l2_mmx(dst, src+1, half, stride, stride, 16);\
}\
\
static void OPNAME ## qpel16_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t temp[32];\
    uint8_t * const half= (uint8_t*)temp;\
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\
    OPNAME ## pixels16_l2_mmx(dst, src, half, stride, stride, 16);\
}\
\
static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src, stride, stride);\
}\
\
static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t temp[32];\
    uint8_t * const half= (uint8_t*)temp;\
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\
    OPNAME ## pixels16_l2_mmx(dst, src+stride, half, stride, stride, 16);\
}\
static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t half[16*2 + 17*2];\
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
    uint8_t * const halfHV= ((uint8_t*)half);\
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
    put ## RND ## pixels16_l2_mmx(halfH, src, halfH, 16, stride, 17);\
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
    OPNAME ## pixels16_l2_mmx(dst, halfH, halfHV, stride, 16, 16);\
}\
static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t half[16*2 + 17*2];\
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
    uint8_t * const halfHV= ((uint8_t*)half);\
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
    put ## RND ## pixels16_l2_mmx(halfH, src+1, halfH, 16, stride, 17);\
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
    OPNAME ## pixels16_l2_mmx(dst, halfH, halfHV, stride, 16, 16);\
}\
static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t half[16*2 + 17*2];\
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
    uint8_t * const halfHV= ((uint8_t*)half);\
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
    put ## RND ## pixels16_l2_mmx(halfH, src, halfH, 16, stride, 17);\
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
    OPNAME ## pixels16_l2_mmx(dst, halfH+16, halfHV, stride, 16, 16);\
}\
static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t half[16*2 + 17*2];\
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
    uint8_t * const halfHV= ((uint8_t*)half);\
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
    put ## RND ## pixels16_l2_mmx(halfH, src+1, halfH, 16, stride, 17);\
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
    OPNAME ## pixels16_l2_mmx(dst, halfH+16, halfHV, stride, 16, 16);\
}\
static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t half[16*2 + 17*2];\
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
    uint8_t * const halfHV= ((uint8_t*)half);\
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
    OPNAME ## pixels16_l2_mmx(dst, halfH, halfHV, stride, 16, 16);\
}\
static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t half[16*2 + 17*2];\
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
    uint8_t * const halfHV= ((uint8_t*)half);\
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
    OPNAME ## pixels16_l2_mmx(dst, halfH+16, halfHV, stride, 16, 16);\
}\
static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t half[17*2];\
    uint8_t * const halfH= ((uint8_t*)half);\
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
    put ## RND ## pixels16_l2_mmx(halfH, src, halfH, 16, stride, 17);\
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
}\
static void OPNAME ## qpel16_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t half[17*2];\
    uint8_t * const halfH= ((uint8_t*)half);\
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
    put ## RND ## pixels16_l2_mmx(halfH, src+1, halfH, 16, stride, 17);\
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
}\
static void OPNAME ## qpel16_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t half[17*2];\
    uint8_t * const halfH= ((uint8_t*)half);\
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
}


#define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b "	\n\t"
#define AVG_3DNOW_OP(a,b,temp, size) \
"mov" #size " " #b ", " #temp "	\n\t"\
"pavgusb " #temp ", " #a "	\n\t"\
"mov" #size " " #a ", " #b "	\n\t"
#define AVG_MMX2_OP(a,b,temp, size) \
"mov" #size " " #b ", " #temp "	\n\t"\
"pavgb " #temp ", " #a "	\n\t"\
"mov" #size " " #a ", " #b "	\n\t"

QPEL_BASE(put_       , ff_pw_16, _       , PUT_OP, PUT_OP)
QPEL_BASE(avg_       , ff_pw_16, _       , AVG_MMX2_OP, AVG_3DNOW_OP)
QPEL_BASE(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, PUT_OP)
QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, 3dnow)
QPEL_OP(avg_       , ff_pw_16, _       , AVG_3DNOW_OP, 3dnow)
QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, 3dnow)
QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, mmx2)
QPEL_OP(avg_       , ff_pw_16, _       , AVG_MMX2_OP, mmx2)
QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, mmx2)

#if 0
static void just_return() { return; }
#endif

#define SET_QPEL_FUNC(postfix1, postfix2) \
    c->put_ ## postfix1 = put_ ## postfix2;\
    c->put_no_rnd_ ## postfix1 = put_no_rnd_ ## postfix2;\
    c->avg_ ## postfix1 = avg_ ## postfix2;

/* external functions, from idct_mmx.c */
void ff_mmx_idct(DCTELEM *block);
void ff_mmxext_idct(DCTELEM *block);

/* XXX: those functions should be suppressed ASAP when all IDCTs are
   converted */
static void ff_libmpeg2mmx_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
{
    ff_mmx_idct (block);
    put_pixels_clamped_mmx(block, dest, line_size);
}
static void ff_libmpeg2mmx_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
{
    ff_mmx_idct (block);
    add_pixels_clamped_mmx(block, dest, line_size);
}
static void ff_libmpeg2mmx2_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
{
    ff_mmxext_idct (block);
    put_pixels_clamped_mmx(block, dest, line_size);
}
static void ff_libmpeg2mmx2_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
{
    ff_mmxext_idct (block);
    add_pixels_clamped_mmx(block, dest, line_size);
}
    
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
{
    mm_flags = mm_support();

    if (avctx->dsp_mask) {
	if (avctx->dsp_mask & FF_MM_FORCE)
	    mm_flags |= (avctx->dsp_mask & 0xffff);
	else
	    mm_flags &= ~(avctx->dsp_mask & 0xffff);
    }

#if 0
    fprintf(stderr, "libavcodec: CPU flags:");
    if (mm_flags & MM_MMX)
        fprintf(stderr, " mmx");
    if (mm_flags & MM_MMXEXT)
        fprintf(stderr, " mmxext");
    if (mm_flags & MM_3DNOW)
        fprintf(stderr, " 3dnow");
    if (mm_flags & MM_SSE)
        fprintf(stderr, " sse");
    if (mm_flags & MM_SSE2)
        fprintf(stderr, " sse2");
    fprintf(stderr, "\n");
#endif

    if (mm_flags & MM_MMX) {
        const int dct_algo = avctx->dct_algo;
        const int idct_algo= avctx->idct_algo;

#ifdef CONFIG_ENCODERS
        if(dct_algo==FF_DCT_AUTO || dct_algo==FF_DCT_MMX){
            if(mm_flags & MM_MMXEXT){
                c->fdct = ff_fdct_mmx2;
            }else{
                c->fdct = ff_fdct_mmx;
            }
        }
#endif //CONFIG_ENCODERS

        if(idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_SIMPLEMMX){
            c->idct_put= ff_simple_idct_put_mmx;
            c->idct_add= ff_simple_idct_add_mmx;
            c->idct    = ff_simple_idct_mmx;
            c->idct_permutation_type= FF_SIMPLE_IDCT_PERM;
        }else if(idct_algo==FF_IDCT_LIBMPEG2MMX){
            if(mm_flags & MM_MMXEXT){
                c->idct_put= ff_libmpeg2mmx2_idct_put;
                c->idct_add= ff_libmpeg2mmx2_idct_add;
                c->idct    = ff_mmxext_idct;
            }else{
                c->idct_put= ff_libmpeg2mmx_idct_put;
                c->idct_add= ff_libmpeg2mmx_idct_add;
                c->idct    = ff_mmx_idct;
            }
            c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;
        }
        
#ifdef CONFIG_ENCODERS
        c->get_pixels = get_pixels_mmx;
        c->diff_pixels = diff_pixels_mmx;
#endif //CONFIG_ENCODERS
        c->put_pixels_clamped = put_pixels_clamped_mmx;
        c->add_pixels_clamped = add_pixels_clamped_mmx;
        c->clear_blocks = clear_blocks_mmx;
#ifdef CONFIG_ENCODERS
        c->pix_sum = pix_sum16_mmx;
#endif //CONFIG_ENCODERS

        c->put_pixels_tab[0][0] = put_pixels16_mmx;
        c->put_pixels_tab[0][1] = put_pixels16_x2_mmx;
        c->put_pixels_tab[0][2] = put_pixels16_y2_mmx;
        c->put_pixels_tab[0][3] = put_pixels16_xy2_mmx;

        c->put_no_rnd_pixels_tab[0][0] = put_pixels16_mmx;
        c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx;
        c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx;
        c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_mmx;

        c->avg_pixels_tab[0][0] = avg_pixels16_mmx;
        c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx;
        c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx;
        c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx;

        c->avg_no_rnd_pixels_tab[0][0] = avg_no_rnd_pixels16_mmx;
        c->avg_no_rnd_pixels_tab[0][1] = avg_no_rnd_pixels16_x2_mmx;
        c->avg_no_rnd_pixels_tab[0][2] = avg_no_rnd_pixels16_y2_mmx;
        c->avg_no_rnd_pixels_tab[0][3] = avg_no_rnd_pixels16_xy2_mmx;

        c->put_pixels_tab[1][0] = put_pixels8_mmx;
        c->put_pixels_tab[1][1] = put_pixels8_x2_mmx;
        c->put_pixels_tab[1][2] = put_pixels8_y2_mmx;
        c->put_pixels_tab[1][3] = put_pixels8_xy2_mmx;

        c->put_no_rnd_pixels_tab[1][0] = put_pixels8_mmx;
        c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx;
        c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx;
        c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_mmx;

        c->avg_pixels_tab[1][0] = avg_pixels8_mmx;
        c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx;
        c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx;
        c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx;

        c->avg_no_rnd_pixels_tab[1][0] = avg_no_rnd_pixels8_mmx;
        c->avg_no_rnd_pixels_tab[1][1] = avg_no_rnd_pixels8_x2_mmx;
        c->avg_no_rnd_pixels_tab[1][2] = avg_no_rnd_pixels8_y2_mmx;
        c->avg_no_rnd_pixels_tab[1][3] = avg_no_rnd_pixels8_xy2_mmx;
                
        c->add_bytes= add_bytes_mmx;
#ifdef CONFIG_ENCODERS
        c->diff_bytes= diff_bytes_mmx;
        
        c->hadamard8_diff[0]= hadamard8_diff16_mmx;
        c->hadamard8_diff[1]= hadamard8_diff_mmx;
        
	c->pix_norm1 = pix_norm1_mmx;
	c->sse[0] = sse16_mmx;
#endif //CONFIG_ENCODERS
        
        if (mm_flags & MM_MMXEXT) {
            c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2;
            c->put_pixels_tab[0][2] = put_pixels16_y2_mmx2;

            c->avg_pixels_tab[0][0] = avg_pixels16_mmx2;
            c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx2;
            c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx2;

            c->put_pixels_tab[1][1] = put_pixels8_x2_mmx2;
            c->put_pixels_tab[1][2] = put_pixels8_y2_mmx2;

            c->avg_pixels_tab[1][0] = avg_pixels8_mmx2;
            c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2;
            c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2;

#ifdef CONFIG_ENCODERS
            c->hadamard8_diff[0]= hadamard8_diff16_mmx2;
            c->hadamard8_diff[1]= hadamard8_diff_mmx2;
#endif //CONFIG_ENCODERS

            if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
                c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx2;
                c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx2;
                c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx2;
                c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx2;
                c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx2;
                c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2;
            }

#if 1
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_mmx2)
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_mmx2)
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_mmx2)
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_mmx2)
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_mmx2)
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_mmx2)
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_mmx2)
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_mmx2)
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_mmx2)
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_mmx2)
            SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_mmx2)
            SET_QPEL_FUNC(qpel_p