deblock_luma_kc.sc

deblocking 在SPI DSP平台优化好的代码,超级强

        top1_32_1 = spi_vselect8((vec uint8x4)cid_b0, top1_10, top1_32);
        top1_10_1 = spi_vselect8((vec uint8x4)cid_b0, top1_32, top1_10);
        
        top0 = (vec uint8x4)spi_vshuffleu(0x02000604, (vec uint32x1)top0_32_1, (vec uint32x1)top0_10_1);
        top1 = (vec uint8x4)spi_vshuffleu(0x02000604, (vec uint32x1)top1_32_1, (vec uint32x1)top1_10_1);
        
        // Swap tops with spi_laneid() + 8
        top0 = (vec uint8x4)spi_vperm32(perm_b3, (vec uint32x1)top0, 0);
        top1 = (vec uint8x4)spi_vperm32(perm_b3, (vec uint32x1)top1, 0);

        //Load MB from top strip
        spi_array_read(in_out_frame_top, top0_0, in_out_idx);
        spi_array_read(in_out_frame_top, top1_0, in_out_idx+1U);
        spi_array_read(in_out_frame_top, top2_0, in_out_idx+2U);
        spi_array_read(in_out_frame_top, top3_0, in_out_idx+3U);
        spi_array_read(in_out_frame_top, top0_1, in_out_idx+pitch);
        spi_array_read(in_out_frame_top, top1_1, in_out_idx+pitch+1U);
        spi_array_read(in_out_frame_top, top2_1, in_out_idx+pitch+2U);
        spi_array_read(in_out_frame_top, top3_1, in_out_idx+pitch+3U);


        //Load MB from current strip
        spi_array_read(in_frame, cur0_0, in_idx);
        spi_array_read(in_frame, cur0_1, in_idx + (vec uint32x1)1);
        spi_array_read(in_frame, cur1_0, in_idx + (vec uint32x1)2);
        spi_array_read(in_frame, cur1_1, in_idx + (vec uint32x1)3);
        spi_array_read(in_frame, cur2_0, in_idx + (vec uint32x1)4);
        spi_array_read(in_frame, cur2_1, in_idx + (vec uint32x1)5);
        spi_array_read(in_frame, cur3_0, in_idx + (vec uint32x1)6);
        spi_array_read(in_frame, cur3_1, in_idx + (vec uint32x1)7);
        
        // For clusters 0-7, current MB strip bottom pixels will 
        // be upated and for clusters 8-15, top MB strip bottom
        // pixels will be updated.
        // Update the pixels
        pud0_0 = (vec uint8x4)spi_vperm32(perm_cid_a9, (vec uint32x1)top0, 0);
        pud0_1 = (vec uint8x4)spi_vperm32(perm_cid_a9, (vec uint32x1)top1, 0);
        
        pud1_0 = (vec uint8x4)spi_vperm32(perm_cid_aB, (vec uint32x1)top0, 0);
        pud1_1 = (vec uint8x4)spi_vperm32(perm_cid_aB, (vec uint32x1)top1, 0);
        
        pud2_0 = (vec uint8x4)spi_vperm32(perm_cid_aD, (vec uint32x1)top0, 0);
        pud2_1 = (vec uint8x4)spi_vperm32(perm_cid_aD, (vec uint32x1)top1, 0);
        
        pud3_0 = top0;
        pud3_1 = top1;
        
        cur0_0 = spi_vselect8((vec uint8x4)cid_6_7, pud0_0, cur0_0);
        cur0_1 = spi_vselect8((vec uint8x4)cid_6_7, pud0_1, cur0_1);
        cur1_0 = spi_vselect8((vec uint8x4)cid_6_7, pud1_0, cur1_0);
        cur1_1 = spi_vselect8((vec uint8x4)cid_6_7, pud1_1, cur1_1);
        cur2_0 = spi_vselect8((vec uint8x4)cid_6_7, pud2_0, cur2_0);
        cur2_1 = spi_vselect8((vec uint8x4)cid_6_7, pud2_1, cur2_1);
        cur3_0 = spi_vselect8((vec uint8x4)cid_6_7, pud3_0, cur3_0);
        cur3_1 = spi_vselect8((vec uint8x4)cid_6_7, pud3_1, cur3_1);

        top0_0 = spi_vselect8((vec uint8x4)cid_14_15, pud0_0, top0_0);
        top0_1 = spi_vselect8((vec uint8x4)cid_14_15, pud0_1, top0_1);
        top1_0 = spi_vselect8((vec uint8x4)cid_14_15, pud1_0, top1_0);
        top1_1 = spi_vselect8((vec uint8x4)cid_14_15, pud1_1, top1_1);
        top2_0 = spi_vselect8((vec uint8x4)cid_14_15, pud2_0, top2_0);
        top2_1 = spi_vselect8((vec uint8x4)cid_14_15, pud2_1, top2_1);
        top3_0 = spi_vselect8((vec uint8x4)cid_14_15, pud3_0, top3_0);
        top3_1 = spi_vselect8((vec uint8x4)cid_14_15, pud3_1, top3_1);

        
        // Padding and Colflating
        //spi_array_read(in_out_frame_top, left_pad_0, 4);
        //spi_array_read(in_out_frame_top, left_pad_1, 4U + pitch);
        left_pad_0 = top0_0;
        left_pad_1 = top0_1;
        
        left_pad_0 = (vec uint8x4)spi_vshuffledu_lo(0x40404040, left_pad_0, left_pad_1);
        left_pad_1 = (vec uint8x4)spi_vshuffledu_hi(0x40404040, left_pad_0, left_pad_1);
        
        spi_array_write(in_out_frame_top, left_pad_0, 0);
        spi_array_write(in_out_frame_top, left_pad_0, 1);
        spi_array_write(in_out_frame_top, left_pad_0, 2);
        spi_array_write(in_out_frame_top, left_pad_0, 3);
        
        spi_array_write(in_out_frame_top, left_pad_1, pitch);
        spi_array_write(in_out_frame_top, left_pad_1, pitch+1U);
        spi_array_write(in_out_frame_top, left_pad_1, pitch+2U);
        spi_array_write(in_out_frame_top, left_pad_1, pitch+3U);
        
        //spi_array_read(out_frame, left_pad_0, 4);
        //spi_array_read(out_frame, left_pad_1, 4U + pitch);
        left_pad_0 = cur0_0;
        left_pad_1 = cur0_1;
        
        left_pad_0 = (vec uint8x4)spi_vshuffledu_lo(0x40404040, left_pad_0, left_pad_1);
        left_pad_1 = (vec uint8x4)spi_vshuffledu_hi(0x40404040, left_pad_0, left_pad_1);
        
        spi_array_write(out_frame, left_pad_0, 0);
        spi_array_write(out_frame, left_pad_0, 1);
        spi_array_write(out_frame, left_pad_0, 2);
        spi_array_write(out_frame, left_pad_0, 3);
        
        spi_array_write(out_frame, left_pad_1, pitch);
        spi_array_write(out_frame, left_pad_1, pitch+1U);
        spi_array_write(out_frame, left_pad_1, pitch+2U);
        spi_array_write(out_frame, left_pad_1, pitch+3U);
    }
    
    
    while(no_of_iter){
#pragma pipeline
        // Load top intermeidates        
        spi_array_read(in_frame/*_top_inter*/, topa_0, in_fti_idx);
        spi_array_read(in_frame/*_top_inter*/, topa_1, in_fti_idx + (vec uint32x1)1);
        spi_array_read(in_frame/*_top_inter*/, topa_2, in_fti_idx + (vec uint32x1)2);
        spi_array_read(in_frame/*_top_inter*/, topa_3, in_fti_idx + (vec uint32x1)3);
        
        // Each cluster has 4 lines of top MB (2 pixels per row), re
        // arrange the pixels such that each cluster will have 2 full
        // rows (4pixels) to write out. Swap data with neighboring 
        // clusters. 
        top0_32 = (vec uint8x4)spi_vperm32(perm_b0, (vec uint32x1)spi_vselect8((vec uint8x4)cid_b0, topa_0, topa_2), 0);
        top1_32 = (vec uint8x4)spi_vperm32(perm_b0, (vec uint32x1)spi_vselect8((vec uint8x4)cid_b0, topa_1, topa_3), 0);
        
        top0_10 = spi_vselect8((vec uint8x4)cid_b0, topa_2, topa_0);
        top1_10 = spi_vselect8((vec uint8x4)cid_b0, topa_3, topa_1);
        
        top0_32_1 = spi_vselect8((vec uint8x4)cid_b0, top0_10, top0_32);
        top0_10_1 = spi_vselect8((vec uint8x4)cid_b0, top0_32, top0_10);
        top1_32_1 = spi_vselect8((vec uint8x4)cid_b0, top1_10, top1_32);
        top1_10_1 = spi_vselect8((vec uint8x4)cid_b0, top1_32, top1_10);
        
        top0 = (vec uint8x4)spi_vshuffleu(0x02000604, (vec uint32x1)top0_32_1, (vec uint32x1)top0_10_1);
        top1 = (vec uint8x4)spi_vshuffleu(0x02000604, (vec uint32x1)top1_32_1, (vec uint32x1)top1_10_1);
        
        // Swap tops with spi_laneid() + 8
        top0 = (vec uint8x4)spi_vperm32(perm_b3, (vec uint32x1)top0, 0);
        top1 = (vec uint8x4)spi_vperm32(perm_b3, (vec uint32x1)top1, 0);
        
        //Load MB from top strip
        spi_array_read(in_out_frame_top, top0_0, in_out_idx);
        spi_array_read(in_out_frame_top, top1_0, in_out_idx+1U);
        spi_array_read(in_out_frame_top, top2_0, in_out_idx+2U);
        spi_array_read(in_out_frame_top, top3_0, in_out_idx+3U);
        spi_array_read(in_out_frame_top, top0_1, in_out_idx+pitch);
        spi_array_read(in_out_frame_top, top1_1, in_out_idx+pitch+1U);
        spi_array_read(in_out_frame_top, top2_1, in_out_idx+pitch+2U);
        spi_array_read(in_out_frame_top, top3_1, in_out_idx+pitch+3U);
        
        //Load MB from current strip
        spi_array_read(in_frame, cur0_0, in_idx);
        spi_array_read(in_frame, cur0_1, in_idx + (vec uint32x1)1);
        spi_array_read(in_frame, cur1_0, in_idx + (vec uint32x1)2);
        spi_array_read(in_frame, cur1_1, in_idx + (vec uint32x1)3);
        spi_array_read(in_frame, cur2_0, in_idx + (vec uint32x1)4);
        spi_array_read(in_frame, cur2_1, in_idx + (vec uint32x1)5);
        spi_array_read(in_frame, cur3_0, in_idx + (vec uint32x1)6);
        spi_array_read(in_frame, cur3_1, in_idx + (vec uint32x1)7);
        
        // For clusters 0-7, current MB strip bottom pixels will 
        // be upated and for clusters 8-15, top MB strip bottom
        // pixels will be updated.
        
        // Update the pixels
        pud0_0 = (vec uint8x4)spi_vperm32(spi_laneid()&0x9, (vec uint32x1)top0, 0);
        pud0_1 = (vec uint8x4)spi_vperm32(spi_laneid()&0x9, (vec uint32x1)top1, 0);
        
        pud1_0 = (vec uint8x4)spi_vperm32(spi_laneid()&0xB, (vec uint32x1)top0, 0);
        pud1_1 = (vec uint8x4)spi_vperm32(spi_laneid()&0xB, (vec uint32x1)top1, 0);
        
        pud2_0 = (vec uint8x4)spi_vperm32(spi_laneid()&0xD, (vec uint32x1)top0, 0);
        pud2_1 = (vec uint8x4)spi_vperm32(spi_laneid()&0xD, (vec uint32x1)top1, 0);
        
        pud3_0 = top0;
        pud3_1 = top1;

        cur0_0 = spi_vselect8((vec uint8x4)cid_6_7, pud0_0, cur0_0);
        cur0_1 = spi_vselect8((vec uint8x4)cid_6_7, pud0_1, cur0_1);
        cur1_0 = spi_vselect8((vec uint8x4)cid_6_7, pud1_0, cur1_0);
        cur1_1 = spi_vselect8((vec uint8x4)cid_6_7, pud1_1, cur1_1);
        cur2_0 = spi_vselect8((vec uint8x4)cid_6_7, pud2_0, cur2_0);
        cur2_1 = spi_vselect8((vec uint8x4)cid_6_7, pud2_1, cur2_1);
        cur3_0 = spi_vselect8((vec uint8x4)cid_6_7, pud3_0, cur3_0);
        cur3_1 = spi_vselect8((vec uint8x4)cid_6_7, pud3_1, cur3_1);

        top0_0 = spi_vselect8((vec uint8x4)cid_14_15, pud0_0, top0_0);
        top0_1 = spi_vselect8((vec uint8x4)cid_14_15, pud0_1, top0_1);
        top1_0 = spi_vselect8((vec uint8x4)cid_14_15, pud1_0, top1_0);
        top1_1 = spi_vselect8((vec uint8x4)cid_14_15, pud1_1, top1_1);
        top2_0 = spi_vselect8((vec uint8x4)cid_14_15, pud2_0, top2_0);
        top2_1 = spi_vselect8((vec uint8x4)cid_14_15, pud2_1, top2_1);
        top3_0 = spi_vselect8((vec uint8x4)cid_14_15, pud3_0, top3_0);
        top3_1 = spi_vselect8((vec uint8x4)cid_14_15, pud3_1, top3_1);


        // Write out the updated pixels of top strip.
        spi_array_write(in_out_frame_top, top0_0, in_out_idx);
        spi_array_write(in_out_frame_top, top1_0, in_out_idx+1U);
        spi_array_write(in_out_frame_top, top2_0, in_out_idx+2U);
        spi_array_write(in_out_frame_top, top3_0, in_out_idx+3U);
        spi_array_write(in_out_frame_top, top0_1, in_out_idx+pitch);
        spi_array_write(in_out_frame_top, top1_1, in_out_idx+pitch+1U);
        spi_array_write(in_out_frame_top, top2_1, in_out_idx+pitch+2U);
        spi_array_write(in_out_frame_top, top3_1, in_out_idx+pitch+3U);
        
        //Save MB from current strip
        spi_array_write(out_frame, cur0_0, in_out_idx);
        spi_array_write(out_frame, cur1_0, in_out_idx+1U);
        spi_array_write(out_frame, cur2_0, in_out_idx+2U);
        spi_array_write(out_frame, cur3_0, in_out_idx+3U);
        spi_array_write(out_frame, cur0_1, in_out_idx+pitch);
        spi_array_write(out_frame, cur1_1, in_out_idx+pitch+1U);
        spi_array_write(out_frame, cur2_1, in_out_idx+pitch+2U);
        spi_array_write(out_frame, cur3_1, in_out_idx+pitch+3U);
                
        in_out_idx = in_out_idx + 4U;
        in_fti_idx = in_fti_idx + 4U;
        in_idx = in_idx + 8U;
        no_of_iter = (int32x1)no_of_iter - 1;
    }

    if (s_filter_last_mb_horz_edges != 0){
        spi_array_read(in_out_frame_top, right_pad_0, pitch - 5U);
        spi_array_read(in_out_frame_top, right_pad_1, spi_vshift32(pitch, 1) - 5U);
        
        right_pad_0 = (vec uint8x4)spi_vshuffledu_lo(0x73737373, right_pad_0, right_pad_1);
        right_pad_1 = (vec uint8x4)spi_vshuffledu_hi(0x73737373, right_pad_0, right_pad_1);
        
        spi_array_write(in_out_frame_top, right_pad_0, pitch - 4U);
        spi_array_write(in_out_frame_top, right_pad_0, pitch - 3U);
        spi_array_write(in_out_frame_top, right_pad_0, pitch - 2U);
        spi_array_write(in_out_frame_top, right_pad_0, pitch - 1U);
        
        spi_array_write(in_out_frame_top, right_pad_1, spi_vshift32(pitch, 1) - 4U);
        spi_array_write(in_out_frame_top, right_pad_1, spi_vshift32(pitch, 1) - 3U);
        spi_array_write(in_out_frame_top, right_pad_1, spi_vshift32(pitch, 1) - 2U);
        spi_array_write(in_out_frame_top, right_pad_1, spi_vshift32(pitch, 1) - 1U);
        
        spi_array_read(out_frame, right_pad_0, pitch - 5U);
        spi_array_read(out_frame, right_pad_1, spi_vshift32(pitch, 1) - 5U);
        
        right_pad_0 = (vec uint8x4)spi_vshuffledu_lo(0x73737373, right_pad_0, right_pad_1);
        right_pad_1 = (vec uint8x4)spi_vshuffledu_hi(0x73737373, right_pad_0, right_pad_1);
        
        spi_array_write(out_frame, right_pad_0, pitch - 4U);
        spi_array_write(out_frame, right_pad_0, pitch - 3U);
        spi_array_write(out_frame, right_pad_0, pitch - 2U);
        spi_array_write(out_frame, right_pad_0, pitch - 1U);
        
        spi_array_write(out_frame, right_pad_1, spi_vshift32(pitch, 1) - 4U);
        spi_array_write(out_frame, right_pad_1, spi_vshift32(pitch, 1) - 3U);
        spi_array_write(out_frame, right_pad_1, spi_vshift32(pitch, 1) - 2U);
        spi_array_write(out_frame, right_pad_1, spi_vshift32(pitch, 1) - 1U);

    }
}