vf_spp.c

来自「君正早期ucos系统(只有早期的才不没有打包成库),MPLAYER,文件系统,图」· C语言 代码 · 共 625 行 · 第 1/2 页

			int temp;
			STORE(0);
			STORE(1);
			STORE(2);
			STORE(3);
			STORE(4);
			STORE(5);
			STORE(6);
			STORE(7);
		}
	}
}

#ifdef HAVE_MMX
static void store_slice_mmx(uint8_t *dst, int16_t *src, int dst_stride, int src_stride, int width, int height, int log2_scale){
	int y;

	for(y=0; y<height; y++){
		uint8_t *dst1= dst;
		int16_t *src1= src;
		asm volatile(
			"movq (%3), %%mm3	\n\t"
			"movq (%3), %%mm4	\n\t"
			"movd %4, %%mm2		\n\t"
			"pxor %%mm0, %%mm0	\n\t"
			"punpcklbw %%mm0, %%mm3	\n\t"
			"punpckhbw %%mm0, %%mm4	\n\t"
			"psraw %%mm2, %%mm3	\n\t"
			"psraw %%mm2, %%mm4	\n\t"
			"movd %5, %%mm2		\n\t"
			"1:			\n\t"
			"movq (%0), %%mm0	\n\t"
			"movq 8(%0), %%mm1	\n\t"
			"paddw %%mm3, %%mm0	\n\t"
			"paddw %%mm4, %%mm1	\n\t"
			"psraw %%mm2, %%mm0	\n\t"
			"psraw %%mm2, %%mm1	\n\t"
			"packuswb %%mm1, %%mm0	\n\t"
			"movq %%mm0, (%1) 	\n\t"
			"add $16, %0		\n\t"
			"add $8, %1		\n\t"
			"cmp %2, %1		\n\t"
			" jb 1b			\n\t"
			: "+r" (src1), "+r"(dst1)
			: "r"(dst + width), "r"(dither[y]), "g"(log2_scale), "g"(6-log2_scale)
		);
		src += src_stride;
		dst += dst_stride;
	}
//	if(width != mmxw)
//		store_slice_c(dst + mmxw, src + mmxw, dst_stride, src_stride, width - mmxw, log2_scale);
}
#endif

static void (*store_slice)(uint8_t *dst, int16_t *src, int dst_stride, int src_stride, int width, int height, int log2_scale)= store_slice_c;

static void (*requantize)(DCTELEM dst[64], DCTELEM src[64], int qp, uint8_t *permutation)= hardthresh_c;

static void filter(struct vf_priv_s *p, uint8_t *dst, uint8_t *src, int dst_stride, int src_stride, int width, int height, uint8_t *qp_store, int qp_stride, int is_luma){
	int x, y, i;
	const int count= 1<<p->log2_count;
	const int stride= is_luma ? p->temp_stride : ((width+16+15)&(~15));
	uint64_t __attribute__((aligned(16))) block_align[32];
	DCTELEM *block = (DCTELEM *)block_align;
	DCTELEM *block2= (DCTELEM *)(block_align+16);

	if (!src || !dst) return; // HACK avoid crash for Y8 colourspace
	for(y=0; y<height; y++){
		int index= 8 + 8*stride + y*stride;
		fast_memcpy(p->src + index, src + y*src_stride, width);
		for(x=0; x<8; x++){ 
			p->src[index         - x - 1]= p->src[index +         x    ];
			p->src[index + width + x    ]= p->src[index + width - x - 1];
		}
	}
	for(y=0; y<8; y++){
		fast_memcpy(p->src + (      7-y)*stride, p->src + (      y+8)*stride, stride);
		fast_memcpy(p->src + (height+8+y)*stride, p->src + (height-y+7)*stride, stride);
	}
	//FIXME (try edge emu)

	for(y=0; y<height+8; y+=8){
		memset(p->temp + (8+y)*stride, 0, 8*stride*sizeof(int16_t));
		for(x=0; x<width+8; x+=8){
			const int qps= 3 + is_luma;
			int qp;
                        
			if(p->qp)
				qp= p->qp;
			else{
				qp= qp_store[ (XMIN(x, width-1)>>qps) + (XMIN(y, height-1)>>qps) * qp_stride];
				if(p->mpeg2) qp = FFMAX(1, qp>>1);
			}
			for(i=0; i<count; i++){
				const int x1= x + offset[i+count-1][0];
				const int y1= y + offset[i+count-1][1];
				const int index= x1 + y1*stride;
				p->dsp.get_pixels(block, p->src + index, stride);
				p->dsp.fdct(block);
				requantize(block2, block, qp, p->dsp.idct_permutation);
				p->dsp.idct(block2);
				add_block(p->temp + index, stride, block2);
			}
		}
		if(y)
			store_slice(dst + (y-8)*dst_stride, p->temp + 8 + y*stride, dst_stride, stride, width, XMIN(8, height+8-y), 6-p->log2_count);
	}
#if 0
	for(y=0; y<height; y++){
		for(x=0; x<width; x++){
			if((((x>>6) ^ (y>>6)) & 1) == 0)
				dst[x + y*dst_stride]= p->src[8 + 8*stride  + x + y*stride];
			if((x&63) == 0 || (y&63)==0)
				dst[x + y*dst_stride] += 128;
                }
	}
#endif
	//FIXME reorder for better caching
}

static int config(struct vf_instance_s* vf,
        int width, int height, int d_width, int d_height,
	unsigned int flags, unsigned int outfmt){
	int h= (height+16+15)&(~15);

	vf->priv->temp_stride= (width+16+15)&(~15);
        vf->priv->temp= malloc(vf->priv->temp_stride*h*sizeof(int16_t));
        vf->priv->src = malloc(vf->priv->temp_stride*h*sizeof(uint8_t));
        
	return vf_next_config(vf,width,height,d_width,d_height,flags,outfmt);
}

static void get_image(struct vf_instance_s* vf, mp_image_t *mpi){
    if(mpi->flags&MP_IMGFLAG_PRESERVE) return; // don't change
    // ok, we can do pp in-place (or pp disabled):
    vf->dmpi=vf_get_image(vf->next,mpi->imgfmt,
        mpi->type, mpi->flags | MP_IMGFLAG_READABLE, mpi->width, mpi->height);
    mpi->planes[0]=vf->dmpi->planes[0];
    mpi->stride[0]=vf->dmpi->stride[0];
    mpi->width=vf->dmpi->width;
    if(mpi->flags&MP_IMGFLAG_PLANAR){
        mpi->planes[1]=vf->dmpi->planes[1];
        mpi->planes[2]=vf->dmpi->planes[2];
	mpi->stride[1]=vf->dmpi->stride[1];
	mpi->stride[2]=vf->dmpi->stride[2];
    }
    mpi->flags|=MP_IMGFLAG_DIRECT;
}

static int put_image(struct vf_instance_s* vf, mp_image_t *mpi, double pts){
	mp_image_t *dmpi;

	if(!(mpi->flags&MP_IMGFLAG_DIRECT)){
		// no DR, so get a new image! hope we'll get DR buffer:
                dmpi=vf_get_image(vf->next,mpi->imgfmt,
                    MP_IMGTYPE_TEMP,
                    MP_IMGFLAG_ACCEPT_STRIDE|MP_IMGFLAG_PREFER_ALIGNED_STRIDE,
                    mpi->width,mpi->height);
                vf_clone_mpi_attributes(dmpi, mpi);
        }else{
           dmpi=vf->dmpi;
        }

        vf->priv->mpeg2= mpi->qscale_type;
        if(mpi->pict_type != 3 && mpi->qscale && !vf->priv->qp){
            if(!vf->priv->non_b_qp)
                vf->priv->non_b_qp= malloc(mpi->qstride * ((mpi->h + 15) >> 4));
            fast_memcpy(vf->priv->non_b_qp, mpi->qscale, mpi->qstride * ((mpi->h + 15) >> 4));
        }
	if(vf->priv->log2_count || !(mpi->flags&MP_IMGFLAG_DIRECT)){
            char *qp_tab= vf->priv->non_b_qp;
            if((vf->priv->mode&4) || !qp_tab)
                qp_tab= mpi->qscale;

	    if(qp_tab || vf->priv->qp){
		filter(vf->priv, dmpi->planes[0], mpi->planes[0], dmpi->stride[0], mpi->stride[0], mpi->w, mpi->h, qp_tab, mpi->qstride, 1);
		filter(vf->priv, dmpi->planes[1], mpi->planes[1], dmpi->stride[1], mpi->stride[1], mpi->w>>mpi->chroma_x_shift, mpi->h>>mpi->chroma_y_shift, qp_tab, mpi->qstride, 0);
		filter(vf->priv, dmpi->planes[2], mpi->planes[2], dmpi->stride[2], mpi->stride[2], mpi->w>>mpi->chroma_x_shift, mpi->h>>mpi->chroma_y_shift, qp_tab, mpi->qstride, 0);
	    }else{
		memcpy_pic(dmpi->planes[0], mpi->planes[0], mpi->w, mpi->h, dmpi->stride[0], mpi->stride[0]);
		memcpy_pic(dmpi->planes[1], mpi->planes[1], mpi->w>>mpi->chroma_x_shift, mpi->h>>mpi->chroma_y_shift, dmpi->stride[1], mpi->stride[1]);
		memcpy_pic(dmpi->planes[2], mpi->planes[2], mpi->w>>mpi->chroma_x_shift, mpi->h>>mpi->chroma_y_shift, dmpi->stride[2], mpi->stride[2]);
	    }
	}

#ifdef HAVE_MMX
	if(gCpuCaps.hasMMX) asm volatile ("emms\n\t");
#endif
#ifdef HAVE_MMX2
	if(gCpuCaps.hasMMX2) asm volatile ("sfence\n\t");
#endif

	return vf_next_put_image(vf,dmpi, pts);
}

static void uninit(struct vf_instance_s* vf){
	if(!vf->priv) return;

	if(vf->priv->temp) free(vf->priv->temp);
	vf->priv->temp= NULL;
	if(vf->priv->src) free(vf->priv->src);
	vf->priv->src= NULL;
        if(vf->priv->avctx) free(vf->priv->avctx);
        vf->priv->avctx= NULL;
        if(vf->priv->non_b_qp) free(vf->priv->non_b_qp);
        vf->priv->non_b_qp= NULL;
	
	free(vf->priv);
	vf->priv=NULL;
}

//===========================================================================//
static int query_format(struct vf_instance_s* vf, unsigned int fmt){
    switch(fmt){
	case IMGFMT_YVU9:
	case IMGFMT_IF09:
	case IMGFMT_YV12:
	case IMGFMT_I420:
	case IMGFMT_IYUV:
	case IMGFMT_CLPL:
	case IMGFMT_Y800:
	case IMGFMT_Y8:
	case IMGFMT_444P:
	case IMGFMT_422P:
	case IMGFMT_411P:
	    return vf_next_query_format(vf,fmt);
    }
    return 0;
}

static unsigned int fmt_list[]={
	IMGFMT_YVU9,
	IMGFMT_IF09,
	IMGFMT_YV12,
	IMGFMT_I420,
	IMGFMT_IYUV,
	IMGFMT_CLPL,
	IMGFMT_Y800,
	IMGFMT_Y8,
	IMGFMT_444P,
	IMGFMT_422P,
	IMGFMT_411P,
	0
};

static int control(struct vf_instance_s* vf, int request, void* data){
    switch(request){
    case VFCTRL_QUERY_MAX_PP_LEVEL:
	return 6;
    case VFCTRL_SET_PP_LEVEL:
	vf->priv->log2_count= *((unsigned int*)data);
	return CONTROL_TRUE;
    }
    return vf_next_control(vf,request,data);
}

static int open(vf_instance_t *vf, char* args){

    int log2c=-1;
    
    vf->config=config;
    vf->put_image=put_image;
    vf->get_image=get_image;
    vf->query_format=query_format;
    vf->uninit=uninit;
    vf->control= control;
    vf->priv=malloc(sizeof(struct vf_priv_s));
    memset(vf->priv, 0, sizeof(struct vf_priv_s));
    
    avcodec_init();

    vf->priv->avctx= avcodec_alloc_context();
    dsputil_init(&vf->priv->dsp, vf->priv->avctx);
    
    vf->priv->log2_count= 3;
    
    if (args) sscanf(args, "%d:%d:%d", &log2c, &vf->priv->qp, &vf->priv->mode);

    if( log2c >=0 && log2c <=6 )
        vf->priv->log2_count = log2c;

    if(vf->priv->qp < 0)
        vf->priv->qp = 0;

    switch(vf->priv->mode&3){
        default:
	case 0: requantize= hardthresh_c; break;
	case 1: requantize= softthresh_c; break;
    }

#ifdef HAVE_MMX
    if(gCpuCaps.hasMMX){
	store_slice= store_slice_mmx;
	switch(vf->priv->mode&3){
	    case 0: requantize= hardthresh_mmx; break;
	    case 1: requantize= softthresh_mmx; break;
	}
    }
#endif
    
    return 1;
}

vf_info_t vf_info_spp = {
    "simple postprocess",
    "spp",
    "Michael Niedermayer",
    "",
    open,
    NULL
};