h264dsputil.c

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,图片解码,浏览,电子书,录音,想学ucos,识货的人就下吧 russblock fmradio explore set

}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc32_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[SIZE*(SIZE+5)];\
    uint8_t * const full_mid= full + SIZE*2;\
    int16_t tmp[SIZE*(SIZE+5)];\
    uint8_t halfV[SIZE*SIZE];\
    uint8_t halfHV[SIZE*SIZE];\
    copy_block ## SIZE (full, src - stride*2 + 1, SIZE,  stride, SIZE + 5);\
    put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\
    put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\
    OPNAME ## pixels ## SIZE ## _l2(dst, halfV, halfHV, stride, SIZE, SIZE, SIZE);\
}\

#define op_avg(a, b)  a = (((a)+cm[((b) + 16)>>5]+1)>>1)
//#define op_avg2(a, b) a = (((a)*w1+cm[((b) + 16)>>5]*w2 + o + 64)>>7)
#define op_put(a, b)  a = cm[((b) + 16)>>5]
#define op2_avg(a, b)  a = (((a)+cm[((b) + 512)>>10]+1)>>1)
#define op2_put(a, b)  a = cm[((b) + 512)>>10]

#ifdef JZ4740_MXU_OPT

static void put_h264_qpel4_h_lowpass_mxu(uint8_t *dst, uint8_t *src, int dstStride, int srcStride)
{
  uint32_t src_aln1,src_aln2;
  uint32_t src_rs1,src_rs2;
  int i;
  uint32_t mul_1 = 0x14140501; //(20,20,5,1)
  uint32_t mul_2 = 0x01050105; //(1,5,1,5)
  uint32_t round = 0x100010; //(16,16)

  src_aln1 = ((uint32_t)src - 2) & 0xfffffffc;
  src_aln2 = (uint32_t)src & 0xfffffffc;
  src_rs1 = 4 - (((uint32_t)src - 2) & 3);
  src_rs2 = 4 - ((uint32_t)src & 3);

  S32I2M(xr15,mul_2);
  S32I2M(xr14,mul_1);
  S32I2M(xr13,round);

  dst -= dstStride;
  for(i=0; i<4; i++) {
    //2 pixel
    S32LDD(xr1,src_aln1,0);
    S32LDD(xr2,src_aln1,4);
    S32LDD(xr3,src_aln1,8);
    src_aln1 += srcStride;
    i_pref(0, src_aln1, 0);
    //for pixel[0]
    S32ALN(xr4,xr2,xr1,src_rs1); //xr4 <- src[-2] ~ src[1]
    S32ALN(xr5,xr3,xr2,src_rs1); //xr5 <- src[2] ~ src[5]
    //for pixel[1]
    S32ALN(xr6,xr5,xr4,3); //xr6 <- src[-1] ~ src[2]
    
    D32SLR(xr1,xr5,xr0,xr0,8);
    S32SFL(xr0,xr1,xr5,xr7,ptn3); //xr7 <- src[4],src[3],src[3],src[2]
    
    Q8MUL(xr2,xr4,xr14,xr1); //xr2 <- 20*src[1],20*src[0]  xr1 <- 5*src[-1],src[-2]
    Q8MUL(xr8,xr7,xr15,xr7); //xr8 <- src[4],5*src[3]  xr7 <- src[3],5*src[2]
    Q8MUL(xr4,xr6,xr14,xr3); //xr4 <- 20*src[2],20*src[1]  xr3 <- 5*src[0],src[-1]
    
    Q16ADD_AA_XW(xr2,xr2,xr2,xr0); //xr2 <- 20*src[1] + 20*src[0]
    Q16ADD_SS_XW(xr0,xr1,xr1,xr1); //xr1 <- src[-2] - 5*src[-1]
    Q16ADD_SS_XW(xr0,xr7,xr7,xr7); //xr7 <- src[3] - 5*src[2]
    S32SFL(xr1,xr0,xr1,xr0,ptn3); //xr1 <- src[-2] - 5*src[-1]
    
    Q16ACC_AA(xr0,xr2,xr7,xr1); //xr1 <- 20*src[1] + 20*src[0] + src[-2] - 5*src[-1] + src[3] - 5*src[2]
    
    Q16ADD_AA_XW(xr4,xr4,xr4,xr0); //xr4 <- 20*src[1] + 20*src[2]
    Q16ADD_SS_XW(xr3,xr3,xr3,xr0); //xr3 <- src[-1] - 5*src[0]
    Q16ADD_SS_XW(xr0,xr8,xr8,xr8); //xr8 <- src[4] - 5*src[3]
    S32SFL(xr3,xr0,xr3,xr0,ptn3); //xr3 <- src[-1] - 5*src[0]
    
    Q16ACC_AA(xr0,xr4,xr8,xr3); //xr3 <- 20*src[1] + 20*src[2] + src[-1] - 5*src[0] + src[4] - 5*src[3]

    S32SFL(xr0,xr3,xr1,xr1,ptn3);
    Q16ADD_AA_WW(xr0,xr1,xr13,xr11); //xr11 <- tow new pixel


    //2 pixel
    S32LDD(xr1,src_aln2,0);
    S32LDD(xr2,src_aln2,4);
    S32LDD(xr3,src_aln2,8);
    src_aln2 += srcStride;
    i_pref(0, src_aln2, 0);
    //for pixel[0]
    S32ALN(xr4,xr2,xr1,src_rs2); //xr4 <- src[-2] ~ src[1]
    S32ALN(xr5,xr3,xr2,src_rs2); //xr5 <- src[2] ~ src[5]
    //for pixel[1]
    S32ALN(xr6,xr5,xr4,3); //xr6 <- src[-1] ~ src[2]
    
    D32SLR(xr1,xr5,xr0,xr0,8);
    S32SFL(xr0,xr1,xr5,xr7,ptn3); //xr7 <- src[4],src[3],src[3],src[2]
    
    Q8MUL(xr2,xr4,xr14,xr1); //xr2 <- 20*src[1],20*src[0]  xr1 <- 5*src[-1],src[-2]
    Q8MUL(xr8,xr7,xr15,xr7); //xr8 <- src[4],5*src[3]  xr7 <- src[3],5*src[2]
    Q8MUL(xr4,xr6,xr14,xr3); //xr4 <- 20*src[2],20*src[1]  xr3 <- 5*src[0],src[-1]
    
    Q16ADD_AA_XW(xr2,xr2,xr2,xr0); //xr2 <- 20*src[1] + 20*src[0]
    Q16ADD_SS_XW(xr0,xr1,xr1,xr1); //xr1 <- src[-2] - 5*src[-1]
    Q16ADD_SS_XW(xr0,xr7,xr7,xr7); //xr7 <- src[3] - 5*src[2]
    S32SFL(xr1,xr0,xr1,xr0,ptn3); //xr1 <- src[-2] - 5*src[-1]
    
    Q16ACC_AA(xr0,xr2,xr7,xr1); //xr1 <- 20*src[1] + 20*src[0] + src[-2] - 5*src[-1] + src[3] - 5*src[2]
    
    Q16ADD_AA_XW(xr4,xr4,xr4,xr0); //xr4 <- 20*src[1] + 20*src[2]
    Q16ADD_SS_XW(xr3,xr3,xr3,xr0); //xr3 <- src[-1] - 5*src[0]
    Q16ADD_SS_XW(xr0,xr8,xr8,xr8); //xr8 <- src[4] - 5*src[3]
    S32SFL(xr3,xr0,xr3,xr0,ptn3); //xr3 <- src[-1] - 5*src[0]
    
    Q16ACC_AA(xr0,xr4,xr8,xr3); //xr3 <- 20*src[1] + 20*src[2] + src[-1] - 5*src[0] + src[4] - 5*src[3]

    S32SFL(xr0,xr3,xr1,xr1,ptn3);
    Q16ADD_AA_WW(xr0,xr1,xr13,xr12); //xr12 <- tow new pixel
  
    // >>5 and clip to (0~255)
    Q16SAR(xr12,xr12,xr11,xr11,5);
    Q16SAT(xr1,xr12,xr11); //xr1 <- dst[3],dst[2],dst[1],dst[0]
    
    S32SDIV(xr1,dst,dstStride,0);
  }
}

static void put_h264_qpel4_v_lowpass_mxu(uint8_t *dst, uint8_t *src, int dstStride, int srcStride)
{
  uint32_t src_aln,src_aln0, src_al4;
  uint32_t src_rs;
  int i;
  uint32_t mul_20 = 0x14141414; //(20,20,20,20)
  uint32_t mul_5 = 0x05050505; //(5,5,5,5)
  uint32_t round = 0x00100010; //(0,16,0,16)

  src_aln = ((uint32_t)src - (srcStride<<1)) & 0xfffffffc;
  src_aln0 = src_aln - (srcStride<<1);
  src_rs = 4 - (((uint32_t)src - (srcStride<<1)) & 3);

  S32I2M(xr15,mul_20);
  S32I2M(xr14,mul_5);
  S32I2M(xr13,round);
  dst -= dstStride;
  for(i=0; i<4; i++) {
    src_aln0 += srcStride;
    src_aln = src_aln0;
    src_al4 = src_aln0 + 4;
    //first line
    S32LDIV(xr1, src_aln, srcStride, 0);
    S32LDIV(xr2, src_al4, srcStride, 0);
    //second line
    S32LDIV(xr3, src_aln, srcStride, 0);
    S32LDIV(xr4, src_al4, srcStride, 0);
    //ALN 1st and 2nd line 4 pixel
    S32ALN(xr1,xr2,xr1,src_rs); 
    S32ALN(xr2,xr4,xr3,src_rs);
    //third line
    S32LDIV(xr3, src_aln, srcStride, 0);
    S32LDIV(xr4, src_al4, srcStride, 0);
    Q8MUL(xr10,xr2,xr14,xr9); //2nd 4 pixel mul 5 <--here
    //ALN 3rd line 4 pixel
    S32ALN(xr3,xr4,xr3,src_rs);

    Q8MUL(xr8,xr3,xr15,xr7); //3rd 4 pixel mul 20
    //fourth line
    S32LDIV(xr3, src_aln, srcStride, 0);
    S32LDIV(xr4, src_al4, srcStride, 0);
    S32LDIV(xr2, src_aln, srcStride, 0); // <--here
    S32LDIV(xr5, src_al4, srcStride, 0);
    //ALN 4th line 4 pixel
    S32ALN(xr4,xr4,xr3,src_rs);

    //fiveth line
    //ALN 5th line 4 pixel
    S32ALN(xr5,xr5,xr2,src_rs);

    S32LDIV(xr11, src_aln, srcStride, 0); // <--here
    S32LDIV(xr12, src_al4, srcStride, 0);
    Q8MUL(xr6,xr5,xr14,xr5); //5th 4 pixel mul 5
    Q8MUL(xr4,xr4,xr15,xr3); //4th 4 pixel mul 20
    
    //sixth line
    //ALN 6th line 4 pixel
    S32ALN(xr2,xr12,xr11,src_rs);
    
    //1st line + 6th line
    Q8ADDE_AA(xr2,xr1,xr2,xr1);
    
    //1st line + 6th line + 3rd line - 2nd line
    Q16ACC_SS(xr2,xr8,xr10,xr0);
    Q16ACC_SS(xr1,xr7,xr9,xr0);
    
    //1st line + 6th line + 3rd line - 2nd line + 4th line - 5th line
    Q16ACC_SS(xr2,xr4,xr6,xr0);
    Q16ACC_SS(xr1,xr3,xr5,xr0);

    //+ round
    Q16ADD_AA_WW(xr2,xr2,xr13,xr0);
    Q16ADD_AA_WW(xr1,xr1,xr13,xr0);

    // >>5 and clip to (0~255)
    Q16SAR(xr2,xr2,xr1,xr1,5);
    Q16SAT(xr1,xr2,xr1); //xr1 <- dst[3],dst[2],dst[1],dst[0]
    
    S32SDIV(xr1, dst, dstStride, 0);
  }
}

static void put_h264_qpel4_hv_lowpass_mxu(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride)
{
  uint32_t src_aln[2],tmp_h,tmp_v;
  uint32_t src_rs[2];
  int i,j;
  uint32_t mul_1 = 0x14140501; //(20,20,5,1)
  uint32_t mul_2 = 0x01050105; //(1,5,1,5)
  uint32_t mul_20 = 0x00140014; //(0,20,0,20)
  uint32_t mul_5 = 0x00050005; //(0,5,0,5)
  uint32_t round = 0x200; //512

  tmp_h = (uint32_t)tmp;
  src_aln[0] = ((uint32_t)src - 2*srcStride - 2) & 0xfffffffc;
  src_aln[1] = ((uint32_t)src - 2*srcStride) & 0xfffffffc;
  src_rs[0] = 4 - (((uint32_t)src - 2) & 3);
  src_rs[1] = 4 - ((uint32_t)src & 3);

  S32I2M(xr15,mul_2);
  S32I2M(xr14,mul_1);
  tmp_h -= 4;
  for(i=0; i<9; i++) {
    for(j=0; j<2; j++) {
      //2 pixel
      S32LDD(xr1,src_aln[j],0);
      S32LDD(xr2,src_aln[j],4);
      S32LDD(xr3,src_aln[j],8);
      src_aln[j] += srcStride;
      i_pref(0, src_aln[j], 0);
      //for pixel[0]
      S32ALN(xr4,xr2,xr1,src_rs[j]); //xr4 <- src[-2] ~ src[1]
      S32ALN(xr5,xr3,xr2,src_rs[j]); //xr5 <- src[2] ~ src[5]
      //for pixel[1]
      S32ALN(xr6,xr5,xr4,3); //xr6 <- src[-1] ~ src[2]
    
      D32SLR(xr1,xr5,xr0,xr0,8);
      S32SFL(xr0,xr1,xr5,xr7,ptn3); //xr7 <- src[4],src[3],src[3],src[2]
    
      Q8MUL(xr2,xr4,xr14,xr1); //xr2 <- 20*src[1],20*src[0]  xr1 <- 5*src[-1],src[-2]
      Q8MUL(xr8,xr7,xr15,xr7); //xr8 <- src[4],5*src[3]  xr7 <- src[3],5*src[2]
      Q8MUL(xr4,xr6,xr14,xr3); //xr4 <- 20*src[2],20*src[1]  xr3 <- 5*src[0],src[-1]
      
      Q16ADD_AA_XW(xr2,xr2,xr2,xr0); //xr2 <- 20*src[1] + 20*src[0]
      Q16ADD_SS_XW(xr0,xr1,xr1,xr1); //xr1 <- src[-2] - 5*src[-1]
      Q16ADD_SS_XW(xr0,xr7,xr7,xr7); //xr7 <- src[3] - 5*src[2]
      S32SFL(xr1,xr0,xr1,xr0,ptn3); //xr1 <- src[-2] - 5*src[-1]
    
      Q16ACC_AA(xr0,xr2,xr7,xr1); //xr1 <- 20*src[1] + 20*src[0] + src[-2] - 5*src[-1] + src[3] - 5*src[2]
    
      Q16ADD_AA_XW(xr4,xr4,xr4,xr0); //xr4 <- 20*src[1] + 20*src[2]
      Q16ADD_SS_XW(xr3,xr3,xr3,xr0); //xr3 <- src[-1] - 5*src[0]
      Q16ADD_SS_XW(xr0,xr8,xr8,xr8); //xr8 <- src[4] - 5*src[3]
      S32SFL(xr3,xr0,xr3,xr0,ptn3); //xr3 <- src[-1] - 5*src[0]
    
      Q16ACC_AA(xr0,xr4,xr8,xr3); //xr3 <- 20*src[1] + 20*src[2] + src[-1] - 5*src[0] + src[4] - 5*src[3]

      S32SFL(xr0,xr3,xr1,xr1,ptn3);
      S32SDI(xr1, tmp_h, 4);
    }
    tmp_h += tmpStride - 8;
  }


  S32I2M(xr15,mul_20);
  S32I2M(xr14,mul_5);
  S32I2M(xr13,round);
  S32I2M(xr11,0x10001);
  tmp_v = (uint32_t)tmp - (tmpStride << 1);
  for(i=0; i<4; i++) {
    tmp_v += tmpStride;
    for(j=0; j<2; j++) {
      tmp_h = tmp_v + (j<<2);
      //1st line 2 pixel
      S32LDIV(xr1, tmp_h, tmpStride, 0);
      //2nd line 2 pixel
      S32LDIV(xr2, tmp_h, tmpStride, 0);
      //3rd line 2 pixel
      S32LDIV(xr3, tmp_h, tmpStride, 0);
      //2line*5; 3line*20
      D16MUL_WW(xr10,xr2,xr14,xr9); //2nd 2 pixel mul 5
      D16MUL_WW(xr8,xr3,xr15,xr7); //3rd 2 pixel mul 20

      //4th line 2 pixel
      S32LDIV(xr4, tmp_h, tmpStride, 0);
      //5th line 2 pixel
      S32LDIV(xr5, tmp_h, tmpStride, 0);
      S32LDDV(xr12, tmp_h, tmpStride, 0);
      //5line*5; 4line*20
      D16MUL_WW(xr3,xr5,xr14,xr2); //5th 2 pixel mul 5
      D16MUL_WW(xr5,xr4,xr15,xr4); //4th 2 pixel mul 20
      //6st line 2 pixel

      //1st line + 6th line
      Q16ADD_AA_WW(xr12,xr1,xr12,xr0);
      D16MUL_WW(xr12,xr12,xr11,xr1);

      //1st line + 6th line + 3rd line - 2nd line
      D32ACC_SS(xr12,xr8,xr10,xr0);
      D32ACC_SS(xr1,xr7,xr9,xr0);
    
      //1st line + 6th line + 3rd line - 2nd line + 4th line - 5th line
      D32ACC_SS(xr12,xr5,xr3,xr0);
      D32ACC_SS(xr1,xr4,xr2,xr0);

      //+ round
      D32ADD_AA(xr12,xr12,xr13,xr0);
      D32ADD_AA(xr1,xr1,xr13,xr0);
      
      //right shift
      D32SAR(xr2,xr12,xr1,xr1,10);
      S32SFL(xr0, xr2, xr1, xr1, ptn3);
      Q16SAT(xr1, xr0, xr1);
      
      *((uint16_t*)dst) = S32M2I(xr1);
      dst += 2;
    }
    dst