mpegvideo_mmx.c

wince下著名的视频播放器源码

		"movq 8(%0, %%"REG_a"), %%mm1	\n\t"
		"movq (%1, %%"REG_a"), %%mm4	\n\t"
		"movq 8(%1, %%"REG_a"), %%mm5	\n\t"
		"pmullw %%mm6, %%mm4		\n\t" // q=qscale*quant_matrix[i]
		"pmullw %%mm6, %%mm5		\n\t" // q=qscale*quant_matrix[i]
		"pxor %%mm2, %%mm2		\n\t"
		"pxor %%mm3, %%mm3		\n\t"
		"pcmpgtw %%mm0, %%mm2		\n\t" // block[i] < 0 ? -1 : 0
		"pcmpgtw %%mm1, %%mm3		\n\t" // block[i] < 0 ? -1 : 0
		"pxor %%mm2, %%mm0		\n\t"
		"pxor %%mm3, %%mm1		\n\t"
		"psubw %%mm2, %%mm0		\n\t" // abs(block[i])
		"psubw %%mm3, %%mm1		\n\t" // abs(block[i])
		"pmullw %%mm4, %%mm0		\n\t" // abs(block[i])*q
		"pmullw %%mm5, %%mm1		\n\t" // abs(block[i])*q
		"pxor %%mm4, %%mm4		\n\t"
		"pxor %%mm5, %%mm5		\n\t" // FIXME slow
		"pcmpeqw (%0, %%"REG_a"), %%mm4	\n\t" // block[i] == 0 ? -1 : 0
		"pcmpeqw 8(%0, %%"REG_a"), %%mm5\n\t" // block[i] == 0 ? -1 : 0
		"psraw $3, %%mm0		\n\t"
		"psraw $3, %%mm1		\n\t"
		"pxor %%mm2, %%mm0		\n\t"
		"pxor %%mm3, %%mm1		\n\t"
		"psubw %%mm2, %%mm0		\n\t"
		"psubw %%mm3, %%mm1		\n\t"
		"pandn %%mm0, %%mm4		\n\t"
		"pandn %%mm1, %%mm5		\n\t"
		"movq %%mm4, (%0, %%"REG_a")	\n\t"
		"movq %%mm5, 8(%0, %%"REG_a")	\n\t"

		"add $16, %%"REG_a"		\n\t"
		"jng 1b				\n\t"
		::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "g" (qscale), "g" (-2*nCoeffs)
		: "%"REG_a, "memory"
	);    
    block[0]= block0;
        //Note, we dont do mismatch control for intra as errors cannot accumulate
}

static void dct_unquantize_mpeg2_inter_mmx(MpegEncContext *s,
                                     DCTELEM *block, int n, int qscale)
{
    long nCoeffs;
    const uint16_t *quant_matrix;
    
    assert(s->block_last_index[n]>=0);

    if(s->alternate_scan) nCoeffs= 63; //FIXME
    else nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];

        quant_matrix = s->inter_matrix;
asm volatile(
		"pcmpeqw %%mm7, %%mm7		\n\t"
                "psrlq $48, %%mm7		\n\t"
		"movd %2, %%mm6			\n\t"
		"packssdw %%mm6, %%mm6		\n\t"
		"packssdw %%mm6, %%mm6		\n\t"
		"mov %3, %%"REG_a"		\n\t"
		".balign 16\n\t"
		"1:				\n\t"
		"movq (%0, %%"REG_a"), %%mm0	\n\t"
		"movq 8(%0, %%"REG_a"), %%mm1	\n\t"
		"movq (%1, %%"REG_a"), %%mm4	\n\t"
		"movq 8(%1, %%"REG_a"), %%mm5	\n\t"
		"pmullw %%mm6, %%mm4		\n\t" // q=qscale*quant_matrix[i]
		"pmullw %%mm6, %%mm5		\n\t" // q=qscale*quant_matrix[i]
		"pxor %%mm2, %%mm2		\n\t"
		"pxor %%mm3, %%mm3		\n\t"
		"pcmpgtw %%mm0, %%mm2		\n\t" // block[i] < 0 ? -1 : 0
		"pcmpgtw %%mm1, %%mm3		\n\t" // block[i] < 0 ? -1 : 0
		"pxor %%mm2, %%mm0		\n\t"
		"pxor %%mm3, %%mm1		\n\t"
		"psubw %%mm2, %%mm0		\n\t" // abs(block[i])
		"psubw %%mm3, %%mm1		\n\t" // abs(block[i])
		"paddw %%mm0, %%mm0		\n\t" // abs(block[i])*2
		"paddw %%mm1, %%mm1		\n\t" // abs(block[i])*2
		"pmullw %%mm4, %%mm0		\n\t" // abs(block[i])*2*q
		"pmullw %%mm5, %%mm1		\n\t" // abs(block[i])*2*q
		"paddw %%mm4, %%mm0		\n\t" // (abs(block[i])*2 + 1)*q
		"paddw %%mm5, %%mm1		\n\t" // (abs(block[i])*2 + 1)*q
		"pxor %%mm4, %%mm4		\n\t"
		"pxor %%mm5, %%mm5		\n\t" // FIXME slow
		"pcmpeqw (%0, %%"REG_a"), %%mm4	\n\t" // block[i] == 0 ? -1 : 0
		"pcmpeqw 8(%0, %%"REG_a"), %%mm5\n\t" // block[i] == 0 ? -1 : 0
		"psrlw $4, %%mm0		\n\t"
		"psrlw $4, %%mm1		\n\t"
		"pxor %%mm2, %%mm0		\n\t"
		"pxor %%mm3, %%mm1		\n\t"
		"psubw %%mm2, %%mm0		\n\t"
		"psubw %%mm3, %%mm1		\n\t"
		"pandn %%mm0, %%mm4		\n\t"
		"pandn %%mm1, %%mm5		\n\t"
                "pxor %%mm4, %%mm7		\n\t"
                "pxor %%mm5, %%mm7		\n\t"
		"movq %%mm4, (%0, %%"REG_a")	\n\t"
		"movq %%mm5, 8(%0, %%"REG_a")	\n\t"

		"add $16, %%"REG_a"		\n\t"
		"jng 1b				\n\t"
                "movd 124(%0, %3), %%mm0	\n\t"
                "movq %%mm7, %%mm6		\n\t"
                "psrlq $32, %%mm7		\n\t"
                "pxor %%mm6, %%mm7		\n\t"
                "movq %%mm7, %%mm6		\n\t"
                "psrlq $16, %%mm7		\n\t"
                "pxor %%mm6, %%mm7		\n\t"
                "pslld $31, %%mm7		\n\t"
                "psrlq $15, %%mm7		\n\t"
                "pxor %%mm7, %%mm0		\n\t"
                "movd %%mm0, 124(%0, %3)	\n\t"
                
		::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "g" (qscale), "r" (-2*nCoeffs)
		: "%"REG_a, "memory"
	);
}

/* draw the edges of width 'w' of an image of size width, height 
   this mmx version can only handle w==8 || w==16 */
static void draw_edges_mmx(uint8_t *buf, int wrap, int width, int height, int w)
{
    uint8_t *ptr, *last_line;
    int i;

    last_line = buf + (height - 1) * wrap;
    /* left and right */
    ptr = buf;
    if(w==8)
    {
	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 -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"
		"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)
    );
}

#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"

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

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;
        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){
            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;
            }
        }
    }
}