A Hybrid Scheme Based on Pipelining and Multitasking in Mobile Application Processors for Advanced Video Coding

Algorithm 1

(a) SIMD implementation of Zigzag scanning and (b) SIMD implementation of edge extension.

(a)

// Optimized Implementation for Zig-Zag Scanning

/

 For Input

 1 2 3 4

 5 6 7 8

 9 10 11 12

 13 14 15 16

  Output after Zig-Zag scannig is

  1 2 5 9 6 3 4 7 10 13 14 11 8 12 15 16

  /

 / load quantized coefficients to MXU registers /

  S32LDD(xr1, dct, 0x0); // xr1 = 2 1

  S32LDD(xr2, dct, 0x4); // xr2 = 4 3

  S32LDD(xr3, dct, 0x8); // xr3 = 6 5

  S32LDD(xr4, dct, 0xc); // xr4 = 8 7

  S32LDD(xr5, dct, 0x10); // xr5 = 10 9

  S32LDD(xr6, dct, 0x14); // xr6 = 12 11

  S32LDD(xr7, dct, 0x18); // xr7 = 14 13

  S32LDD(xr8, dct, 0x1c); // xr8 = 16 15

 / adjust positions of coefficients according to Zig-Zag order /

  S32SFL(xr9, xr6, xr4, xr10, 3); // xr6 = 12 11 xr4 = 8 7 xr9 = 12 8 xr10 = 11 7

  S32SFL(xr4, xr10, xr7, xr6, 3); // xr10 = 11 7 xr7 = 14 13 xr4 = 11 14 xr6 = 7 13

  S32SFL(xr7, xr6, xr2, xr10, 3); // xr6 = 7 13 xr2 = 4 3 xr7 = 7 4 xr10 = 13 3

  S32SFL(xr6, xr10, xr5, xr2, 3); // xr10 = 13 3 xr5 = 10 9 xr6 = 13 10 xr2 = 3 9

  S32SFL(xr5, xr2, xr3, xr10, 3); // xr2 = 3 9 xr6 = 6 5 xr5 = 3 6 xr10 = 9 5

 / reordered coefficients are stored to main memory from MXU registers /

  S32STD(xr1, level, 0x0); // xr1 = 2 1

  S32STD(xr10, level, 0x4); // xr10 = 9 5

  S32STD(xr5, level, 0x8); // xr5 = 3 6

  S32STD(xr7, level, 0xc); // xr7 = 7 4

  S32STD(xr6, level, 0x10); // xr6 = 13 10

  S32STD(xr4, level, 0x14); // xr4 = 11 14

  S32STD(xr9, level, 0x18); // xr9 = 12 8

  S32STD(xr8, level, 0x1c); // xr8 = 16 15

(b)

 // Optimized Implementation for top edge extension of luma component

  for (j = 0; j < (Lwidth/16); j++)

 S32LDD(xr1, ExtndDataUP_Inp, 0x0); // loading data

 S32LDD(xr2, ExtndDataUP_Inp, 0x4);

 S32LDD(xr3, ExtndDataUP_Inp, 0x8);

 S32LDD(xr4, ExtndDataUP_Inp, 0xc);

 for (i = 0; i < 16; i++)

  S32STD(xr1, ExtndDataUP_Out, 0x0); // edge extension

  S32STD(xr2, ExtndDataUP_Out, 0x4);

  S32STD(xr3, ExtndDataUP_Out, 0x8);

  S32STD(xr4, ExtndDataUP_Out, 0xc);

  ExtndDataUP_Out+ = 4;

 ExtndDataUP_Inp+ = 64;