mpegvideo_mmx.c

mediastreamer2是开源的网络传输媒体流的库

                "punpckhwd %%mm1, %%mm1         \n\t"
                "punpckhdq %%mm1, %%mm1         \n\t"
                "movq %%mm1, (%0, %2)           \n\t"
                "add %1, %0                     \n\t"
                "cmp %3, %0                     \n\t"
                " jb 1b                         \n\t"
                : "+r" (ptr)
                : "r" ((long)wrap), "r" ((long)width), "r" (ptr + wrap*height)
        );
    }
    else
    {
        asm volatile(
                "1:                             \n\t"
                "movd (%0), %%mm0               \n\t"
                "punpcklbw %%mm0, %%mm0         \n\t"
                "punpcklwd %%mm0, %%mm0         \n\t"
                "punpckldq %%mm0, %%mm0         \n\t"
                "movq %%mm0, -8(%0)             \n\t"
                "movq %%mm0, -16(%0)            \n\t"
                "movq -8(%0, %2), %%mm1         \n\t"
                "punpckhbw %%mm1, %%mm1         \n\t"
                "punpckhwd %%mm1, %%mm1         \n\t"
                "punpckhdq %%mm1, %%mm1         \n\t"
                "movq %%mm1, (%0, %2)           \n\t"
                "movq %%mm1, 8(%0, %2)          \n\t"
                "add %1, %0                     \n\t"
                "cmp %3, %0                     \n\t"
                " jb 1b                         \n\t"
                : "+r" (ptr)
                : "r" ((long)wrap), "r" ((long)width), "r" (ptr + wrap*height)
        );
    }

    for(i=0;i<w;i+=4) {
        /* top and bottom (and hopefully also the corners) */
        ptr= buf - (i + 1) * wrap - w;
        asm volatile(
                "1:                             \n\t"
                "movq (%1, %0), %%mm0           \n\t"
                "movq %%mm0, (%0)               \n\t"
                "movq %%mm0, (%0, %2)           \n\t"
                "movq %%mm0, (%0, %2, 2)        \n\t"
                "movq %%mm0, (%0, %3)           \n\t"
                "add $8, %0                     \n\t"
                "cmp %4, %0                     \n\t"
                " jb 1b                         \n\t"
                : "+r" (ptr)
                : "r" ((long)buf - (long)ptr - w), "r" ((long)-wrap), "r" ((long)-wrap*3), "r" (ptr+width+2*w)
        );
        ptr= last_line + (i + 1) * wrap - w;
        asm volatile(
                "1:                             \n\t"
                "movq (%1, %0), %%mm0           \n\t"
                "movq %%mm0, (%0)               \n\t"
                "movq %%mm0, (%0, %2)           \n\t"
                "movq %%mm0, (%0, %2, 2)        \n\t"
                "movq %%mm0, (%0, %3)           \n\t"
                "add $8, %0                     \n\t"
                "cmp %4, %0                     \n\t"
                " jb 1b                         \n\t"
                : "+r" (ptr)
                : "r" ((long)last_line - (long)ptr - w), "r" ((long)wrap), "r" ((long)wrap*3), "r" (ptr+width+2*w)
        );
    }
}

static void  denoise_dct_mmx(MpegEncContext *s, DCTELEM *block){
    const int intra= s->mb_intra;
    int *sum= s->dct_error_sum[intra];
    uint16_t *offset= s->dct_offset[intra];

    s->dct_count[intra]++;

    asm volatile(
        "pxor %%mm7, %%mm7                      \n\t"
        "1:                                     \n\t"
        "pxor %%mm0, %%mm0                      \n\t"
        "pxor %%mm1, %%mm1                      \n\t"
        "movq (%0), %%mm2                       \n\t"
        "movq 8(%0), %%mm3                      \n\t"
        "pcmpgtw %%mm2, %%mm0                   \n\t"
        "pcmpgtw %%mm3, %%mm1                   \n\t"
        "pxor %%mm0, %%mm2                      \n\t"
        "pxor %%mm1, %%mm3                      \n\t"
        "psubw %%mm0, %%mm2                     \n\t"
        "psubw %%mm1, %%mm3                     \n\t"
        "movq %%mm2, %%mm4                      \n\t"
        "movq %%mm3, %%mm5                      \n\t"
        "psubusw (%2), %%mm2                    \n\t"
        "psubusw 8(%2), %%mm3                   \n\t"
        "pxor %%mm0, %%mm2                      \n\t"
        "pxor %%mm1, %%mm3                      \n\t"
        "psubw %%mm0, %%mm2                     \n\t"
        "psubw %%mm1, %%mm3                     \n\t"
        "movq %%mm2, (%0)                       \n\t"
        "movq %%mm3, 8(%0)                      \n\t"
        "movq %%mm4, %%mm2                      \n\t"
        "movq %%mm5, %%mm3                      \n\t"
        "punpcklwd %%mm7, %%mm4                 \n\t"
        "punpckhwd %%mm7, %%mm2                 \n\t"
        "punpcklwd %%mm7, %%mm5                 \n\t"
        "punpckhwd %%mm7, %%mm3                 \n\t"
        "paddd (%1), %%mm4                      \n\t"
        "paddd 8(%1), %%mm2                     \n\t"
        "paddd 16(%1), %%mm5                    \n\t"
        "paddd 24(%1), %%mm3                    \n\t"
        "movq %%mm4, (%1)                       \n\t"
        "movq %%mm2, 8(%1)                      \n\t"
        "movq %%mm5, 16(%1)                     \n\t"
        "movq %%mm3, 24(%1)                     \n\t"
        "add $16, %0                            \n\t"
        "add $32, %1                            \n\t"
        "add $16, %2                            \n\t"
        "cmp %3, %0                             \n\t"
            " jb 1b                             \n\t"
        : "+r" (block), "+r" (sum), "+r" (offset)
        : "r"(block+64)
    );
}

static void  denoise_dct_sse2(MpegEncContext *s, DCTELEM *block){
    const int intra= s->mb_intra;
    int *sum= s->dct_error_sum[intra];
    uint16_t *offset= s->dct_offset[intra];

    s->dct_count[intra]++;

    asm volatile(
        "pxor %%xmm7, %%xmm7                    \n\t"
        "1:                                     \n\t"
        "pxor %%xmm0, %%xmm0                    \n\t"
        "pxor %%xmm1, %%xmm1                    \n\t"
        "movdqa (%0), %%xmm2                    \n\t"
        "movdqa 16(%0), %%xmm3                  \n\t"
        "pcmpgtw %%xmm2, %%xmm0                 \n\t"
        "pcmpgtw %%xmm3, %%xmm1                 \n\t"
        "pxor %%xmm0, %%xmm2                    \n\t"
        "pxor %%xmm1, %%xmm3                    \n\t"
        "psubw %%xmm0, %%xmm2                   \n\t"
        "psubw %%xmm1, %%xmm3                   \n\t"
        "movdqa %%xmm2, %%xmm4                  \n\t"
        "movdqa %%xmm3, %%xmm5                  \n\t"
        "psubusw (%2), %%xmm2                   \n\t"
        "psubusw 16(%2), %%xmm3                 \n\t"
        "pxor %%xmm0, %%xmm2                    \n\t"
        "pxor %%xmm1, %%xmm3                    \n\t"
        "psubw %%xmm0, %%xmm2                   \n\t"
        "psubw %%xmm1, %%xmm3                   \n\t"
        "movdqa %%xmm2, (%0)                    \n\t"
        "movdqa %%xmm3, 16(%0)                  \n\t"
        "movdqa %%xmm4, %%xmm6                  \n\t"
        "movdqa %%xmm5, %%xmm0                  \n\t"
        "punpcklwd %%xmm7, %%xmm4               \n\t"
        "punpckhwd %%xmm7, %%xmm6               \n\t"
        "punpcklwd %%xmm7, %%xmm5               \n\t"
        "punpckhwd %%xmm7, %%xmm0               \n\t"
        "paddd (%1), %%xmm4                     \n\t"
        "paddd 16(%1), %%xmm6                   \n\t"
        "paddd 32(%1), %%xmm5                   \n\t"
        "paddd 48(%1), %%xmm0                   \n\t"
        "movdqa %%xmm4, (%1)                    \n\t"
        "movdqa %%xmm6, 16(%1)                  \n\t"
        "movdqa %%xmm5, 32(%1)                  \n\t"
        "movdqa %%xmm0, 48(%1)                  \n\t"
        "add $32, %0                            \n\t"
        "add $64, %1                            \n\t"
        "add $32, %2                            \n\t"
        "cmp %3, %0                             \n\t"
            " jb 1b                             \n\t"
        : "+r" (block), "+r" (sum), "+r" (offset)
        : "r"(block+64)
    );
}

#ifdef HAVE_SSSE3
#define HAVE_SSSE3_BAK
#endif
#undef HAVE_SSSE3

#undef HAVE_SSE2
#undef HAVE_MMX2
#define RENAME(a) a ## _MMX
#define RENAMEl(a) a ## _mmx
#include "mpegvideo_mmx_template.c"

#define HAVE_MMX2
#undef RENAME
#undef RENAMEl
#define RENAME(a) a ## _MMX2
#define RENAMEl(a) a ## _mmx2
#include "mpegvideo_mmx_template.c"

#define HAVE_SSE2
#undef RENAME
#undef RENAMEl
#define RENAME(a) a ## _SSE2
#define RENAMEl(a) a ## _sse2
#include "mpegvideo_mmx_template.c"

#ifdef HAVE_SSSE3_BAK
#define HAVE_SSSE3
#undef RENAME
#undef RENAMEl
#define RENAME(a) a ## _SSSE3
#define RENAMEl(a) a ## _sse2
#include "mpegvideo_mmx_template.c"
#endif

void MPV_common_init_mmx(MpegEncContext *s)
{
    if (mm_flags & MM_MMX) {
        const int dct_algo = s->avctx->dct_algo;

        s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_mmx;
        s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_mmx;
        s->dct_unquantize_mpeg1_intra = dct_unquantize_mpeg1_intra_mmx;
        s->dct_unquantize_mpeg1_inter = dct_unquantize_mpeg1_inter_mmx;
        if(!(s->flags & CODEC_FLAG_BITEXACT))
            s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_mmx;
        s->dct_unquantize_mpeg2_inter = dct_unquantize_mpeg2_inter_mmx;

        draw_edges = draw_edges_mmx;

        if (mm_flags & MM_SSE2) {
            s->denoise_dct= denoise_dct_sse2;
        } else {
                s->denoise_dct= denoise_dct_mmx;
        }

        if(dct_algo==FF_DCT_AUTO || dct_algo==FF_DCT_MMX){
#ifdef HAVE_SSSE3
            if(mm_flags & MM_SSSE3){
                s->dct_quantize= dct_quantize_SSSE3;
            } else
#endif
            if(mm_flags & MM_SSE2){
                s->dct_quantize= dct_quantize_SSE2;
            } else if(mm_flags & MM_MMXEXT){
                s->dct_quantize= dct_quantize_MMX2;
            } else {
                s->dct_quantize= dct_quantize_MMX;
            }
        }
    }
}