rtjpegn.c

uclinux下mplayer的源码

{
 int ci=1, co=1, tmp;
 register int i;

 i=RTjpeg_ZZ[0];
 data[i]=((__u8)strm[0])*qtbl[i];

 for(co=1; co<=bt8; co++)
 {
  i=RTjpeg_ZZ[co];
  data[i]=strm[ci++]*qtbl[i];
 }
 
 for(; co<64; co++)
 {
  if(strm[ci]>63)
  {
   tmp=co+strm[ci]-63;
   for(; co<tmp; co++)data[RTjpeg_ZZ[co]]=0;
   co--;
  } else
  {
   i=RTjpeg_ZZ[co];
   data[i]=strm[ci]*qtbl[i];
  }
  ci++;
 }
 return (int)ci;
}
#endif

#if defined(MMX)
void RTjpeg_quant_init(void)
{
 int i;
 __s16 *qtbl;
 
 qtbl=(__s16 *)RTjpeg_lqt;
 for(i=0; i<64; i++)qtbl[i]=(__s16)RTjpeg_lqt[i];

 qtbl=(__s16 *)RTjpeg_cqt;
 for(i=0; i<64; i++)qtbl[i]=(__s16)RTjpeg_cqt[i];
}

static mmx_t RTjpeg_ones={0x0001000100010001LL};
static mmx_t RTjpeg_half={0x7fff7fff7fff7fffLL};

void RTjpeg_quant(__s16 *block, __s32 *qtbl)
{
 int i;
 mmx_t *bl, *ql;
 
 ql=(mmx_t *)qtbl;
 bl=(mmx_t *)block;
 
 movq_m2r(RTjpeg_ones, mm6);
 movq_m2r(RTjpeg_half, mm7);

 for(i=16; i; i--) 
 {
  movq_m2r(*(ql++), mm0); /* quant vals (4) */
  movq_m2r(*bl, mm2); /* block vals (4) */
  movq_r2r(mm0, mm1);
  movq_r2r(mm2, mm3);
  
  punpcklwd_r2r(mm6, mm0); /*           1 qb 1 qa */
  punpckhwd_r2r(mm6, mm1); /* 1 qd 1 qc */
  
  punpcklwd_r2r(mm7, mm2); /*                   32767 bb 32767 ba */
  punpckhwd_r2r(mm7, mm3); /* 32767 bd 32767 bc */
  
  pmaddwd_r2r(mm2, mm0); /*                         32767+bb*qb 32767+ba*qa */
  pmaddwd_r2r(mm3, mm1); /* 32767+bd*qd 32767+bc*qc */
  
  psrad_i2r(16, mm0);
  psrad_i2r(16, mm1);
  
  packssdw_r2r(mm1, mm0);
  
  movq_r2m(mm0, *(bl++));
  
 }
}
#else
void RTjpeg_quant_init(void)
{
}

void RTjpeg_quant(__s16 *block, __s32 *qtbl)
{
 int i;
 
 for(i=0; i<64; i++)
   block[i]=(__s16)((block[i]*qtbl[i]+32767)>>16);
}
#endif

/*
 * Perform the forward DCT on one block of samples.
 */
#ifdef MMX
static mmx_t RTjpeg_C4   ={0x2D412D412D412D41LL};
static mmx_t RTjpeg_C6   ={0x187E187E187E187ELL};
static mmx_t RTjpeg_C2mC6={0x22A322A322A322A3LL};
static mmx_t RTjpeg_C2pC6={0x539F539F539F539FLL};
static mmx_t RTjpeg_zero ={0x0000000000000000LL};

#else

#define FIX_0_382683433  ((__s32)   98)		/* FIX(0.382683433) */
#define FIX_0_541196100  ((__s32)  139)		/* FIX(0.541196100) */
#define FIX_0_707106781  ((__s32)  181)		/* FIX(0.707106781) */
#define FIX_1_306562965  ((__s32)  334)		/* FIX(1.306562965) */

#define DESCALE10(x) (__s16)( ((x)+128) >> 8)
#define DESCALE20(x)  (__s16)(((x)+32768) >> 16)
#define D_MULTIPLY(var,const)  ((__s32) ((var) * (const)))
#endif

void RTjpeg_dct_init(void)
{
 int i;
 
 for(i=0; i<64; i++)
 {
  RTjpeg_lqt[i]=(((__u64)RTjpeg_lqt[i]<<32)/RTjpeg_aan_tab[i]);
  RTjpeg_cqt[i]=(((__u64)RTjpeg_cqt[i]<<32)/RTjpeg_aan_tab[i]);
 }
}

void RTjpeg_dctY(__u8 *idata, __s16 *odata, int rskip)
{
#ifndef MMX
  __s32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
  __s32 tmp10, tmp11, tmp12, tmp13;
  __s32 z1, z2, z3, z4, z5, z11, z13;
  __u8 *idataptr;
  __s16 *odataptr;
  __s32 *wsptr;
  int ctr;

  idataptr = idata;
  wsptr = RTjpeg_ws;
  for (ctr = 7; ctr >= 0; ctr--) {
    tmp0 = idataptr[0] + idataptr[7];
    tmp7 = idataptr[0] - idataptr[7];
    tmp1 = idataptr[1] + idataptr[6];
    tmp6 = idataptr[1] - idataptr[6];
    tmp2 = idataptr[2] + idataptr[5];
    tmp5 = idataptr[2] - idataptr[5];
    tmp3 = idataptr[3] + idataptr[4];
    tmp4 = idataptr[3] - idataptr[4];
    
    tmp10 = (tmp0 + tmp3);	/* phase 2 */
    tmp13 = tmp0 - tmp3;
    tmp11 = (tmp1 + tmp2);
    tmp12 = tmp1 - tmp2;
    
    wsptr[0] = (tmp10 + tmp11)<<8; /* phase 3 */
    wsptr[4] = (tmp10 - tmp11)<<8;
    
    z1 = D_MULTIPLY(tmp12 + tmp13, FIX_0_707106781); /* c4 */
    wsptr[2] = (tmp13<<8) + z1;	/* phase 5 */
    wsptr[6] = (tmp13<<8) - z1;
    
    tmp10 = tmp4 + tmp5;	/* phase 2 */
    tmp11 = tmp5 + tmp6;
    tmp12 = tmp6 + tmp7;

    z5 = D_MULTIPLY(tmp10 - tmp12, FIX_0_382683433); /* c6 */
    z2 = D_MULTIPLY(tmp10, FIX_0_541196100) + z5; /* c2-c6 */
    z4 = D_MULTIPLY(tmp12, FIX_1_306562965) + z5; /* c2+c6 */
    z3 = D_MULTIPLY(tmp11, FIX_0_707106781); /* c4 */

    z11 = (tmp7<<8) + z3;		/* phase 5 */
    z13 = (tmp7<<8) - z3;

    wsptr[5] = z13 + z2;	/* phase 6 */
    wsptr[3] = z13 - z2;
    wsptr[1] = z11 + z4;
    wsptr[7] = z11 - z4;

    idataptr += rskip<<3;		/* advance pointer to next row */
    wsptr += 8;
  }

  wsptr = RTjpeg_ws;
  odataptr=odata;
  for (ctr = 7; ctr >= 0; ctr--) {
    tmp0 = wsptr[0] + wsptr[56];
    tmp7 = wsptr[0] - wsptr[56];
    tmp1 = wsptr[8] + wsptr[48];
    tmp6 = wsptr[8] - wsptr[48];
    tmp2 = wsptr[16] + wsptr[40];
    tmp5 = wsptr[16] - wsptr[40];
    tmp3 = wsptr[24] + wsptr[32];
    tmp4 = wsptr[24] - wsptr[32];
    
    tmp10 = tmp0 + tmp3;	/* phase 2 */
    tmp13 = tmp0 - tmp3;
    tmp11 = tmp1 + tmp2;
    tmp12 = tmp1 - tmp2;
    
    odataptr[0] = DESCALE10(tmp10 + tmp11); /* phase 3 */
    odataptr[32] = DESCALE10(tmp10 - tmp11);
    
    z1 = D_MULTIPLY(tmp12 + tmp13, FIX_0_707106781); /* c4 */
    odataptr[16] = DESCALE20((tmp13<<8) + z1); /* phase 5 */
    odataptr[48] = DESCALE20((tmp13<<8) - z1);

    tmp10 = tmp4 + tmp5;	/* phase 2 */
    tmp11 = tmp5 + tmp6;
    tmp12 = tmp6 + tmp7;

    z5 = D_MULTIPLY(tmp10 - tmp12, FIX_0_382683433); /* c6 */
    z2 = D_MULTIPLY(tmp10, FIX_0_541196100) + z5; /* c2-c6 */
    z4 = D_MULTIPLY(tmp12, FIX_1_306562965) + z5; /* c2+c6 */
    z3 = D_MULTIPLY(tmp11, FIX_0_707106781); /* c4 */

    z11 = (tmp7<<8) + z3;		/* phase 5 */
    z13 = (tmp7<<8) - z3;

    odataptr[40] = DESCALE20(z13 + z2); /* phase 6 */
    odataptr[24] = DESCALE20(z13 - z2);
    odataptr[8] = DESCALE20(z11 + z4);
    odataptr[56] = DESCALE20(z11 - z4);

    odataptr++;			/* advance pointer to next column */
    wsptr++;
  }
#else
  volatile mmx_t tmp6, tmp7;
  register mmx_t *dataptr = (mmx_t *)odata;
  mmx_t *idata2 = (mmx_t *)idata;

   // first copy the input 8 bit to the destination 16 bits

   movq_m2r(RTjpeg_zero, mm2);


	movq_m2r(*idata2, mm0);		 
	movq_r2r(mm0, mm1);		 			

	punpcklbw_r2r(mm2, mm0);
	movq_r2m(mm0, *(dataptr));

	punpckhbw_r2r(mm2, mm1);
	movq_r2m(mm1, *(dataptr+1));
	
	idata2 += rskip;

	movq_m2r(*idata2, mm0);		 
	movq_r2r(mm0, mm1);		 			

	punpcklbw_r2r(mm2, mm0);
	movq_r2m(mm0, *(dataptr+2));

	punpckhbw_r2r(mm2, mm1);
	movq_r2m(mm1, *(dataptr+3));
	
	idata2 += rskip;

	movq_m2r(*idata2, mm0);		 
	movq_r2r(mm0, mm1);		 			

	punpcklbw_r2r(mm2, mm0);
	movq_r2m(mm0, *(dataptr+4));

	punpckhbw_r2r(mm2, mm1);
	movq_r2m(mm1, *(dataptr+5));
	
	idata2 += rskip;

	movq_m2r(*idata2, mm0);		 
	movq_r2r(mm0, mm1);		 			

	punpcklbw_r2r(mm2, mm0);
	movq_r2m(mm0, *(dataptr+6));

	punpckhbw_r2r(mm2, mm1);
	movq_r2m(mm1, *(dataptr+7));
	
	idata2 += rskip;

	movq_m2r(*idata2, mm0);		 
	movq_r2r(mm0, mm1);		 			

	punpcklbw_r2r(mm2, mm0);
	movq_r2m(mm0, *(dataptr+8));

	punpckhbw_r2r(mm2, mm1);
	movq_r2m(mm1, *(dataptr+9));
	
	idata2 += rskip;

	movq_m2r(*idata2, mm0);		 
	movq_r2r(mm0, mm1);		 			

	punpcklbw_r2r(mm2, mm0);
	movq_r2m(mm0, *(dataptr+10));

	punpckhbw_r2r(mm2, mm1);
	movq_r2m(mm1, *(dataptr+11));
	
	idata2 += rskip;

	movq_m2r(*idata2, mm0);		 
	movq_r2r(mm0, mm1);		 			

	punpcklbw_r2r(mm2, mm0);
	movq_r2m(mm0, *(dataptr+12));

	punpckhbw_r2r(mm2, mm1);
	movq_r2m(mm1, *(dataptr+13));
	
	idata2 += rskip;

	movq_m2r(*idata2, mm0);		 
	movq_r2r(mm0, mm1);		 			

	punpcklbw_r2r(mm2, mm0);
	movq_r2m(mm0, *(dataptr+14));

	punpckhbw_r2r(mm2, mm1);
	movq_r2m(mm1, *(dataptr+15));

/*  Start Transpose to do calculations on rows */

	movq_m2r(*(dataptr+9), mm7);		 	// m03:m02|m01:m00 - first line (line 4)and copy into m5

	movq_m2r(*(dataptr+13), mm6);	    	// m23:m22|m21:m20 - third line (line 6)and copy into m2
	movq_r2r(mm7, mm5);		 			

	punpcklwd_m2r(*(dataptr+11), mm7); 	// m11:m01|m10:m00 - interleave first and second lines
	movq_r2r(mm6, mm2);						 

	punpcklwd_m2r(*(dataptr+15), mm6);  // m31:m21|m30:m20 - interleave third and fourth lines
	movq_r2r(mm7, mm1);

	movq_m2r(*(dataptr+11), mm3);	      // m13:m13|m11:m10 - second line	 
	punpckldq_r2r(mm6, mm7);				// m30:m20|m10:m00 - interleave to produce result 1

	movq_m2r(*(dataptr+15), mm0);	      // m13:m13|m11:m10 - fourth line
	punpckhdq_r2r(mm6, mm1);				// m31:m21|m11:m01 - interleave to produce result 2

	movq_r2m(mm7,*(dataptr+9));			// write result 1
	punpckhwd_r2r(mm3, mm5);				// m13:m03|m12:m02 - interleave first and second lines
	
	movq_r2m(mm1,*(dataptr+11));			// write result 2
	punpckhwd_r2r(mm0, mm2);				// m33:m23|m32:m22 - interleave third and fourth lines

	movq_r2r(mm5, mm1);
	punpckldq_r2r(mm2, mm5);				// m32:m22|m12:m02 - interleave to produce result 3

	movq_m2r(*(dataptr+1), mm0);			// m03:m02|m01:m00 - first line, 4x4
	punpckhdq_r2r(mm2, mm1);				// m33:m23|m13:m03 - interleave to produce result 4

	movq_r2m(mm5,*(dataptr+13));			// write result 3

	// last 4x4 done

	movq_r2m(mm1, *(dataptr+15));			// write result 4, last 4x4

	movq_m2r(*(dataptr+5), mm2);			// m23:m22|m21:m20 - third line
	movq_r2r(mm0, mm6);

	punpcklwd_m2r(*(dataptr+3), mm0);  	// m11:m01|m10:m00 - interleave first and second lines
	movq_r2r(mm2, mm7);

	punpcklwd_m2r(*(dataptr+7), mm2);  	// m31:m21|m30:m20 - interleave third and fourth lines
	movq_r2r(mm0, mm4);

	//
	movq_m2r(*(dataptr+8), mm1);			// n03:n02|n01:n00 - first line 
	punpckldq_r2r(mm2, mm0);				// m30:m20|m10:m00 - interleave to produce first result

	movq_m2r(*(dataptr+12), mm3);			// n23:n22|n21:n20 - third line
	punpckhdq_r2r(mm2, mm4);				// m31:m21|m11:m01 - interleave to produce second result

	punpckhwd_m2r(*(dataptr+3), mm6);  	// m13:m03|m12:m02 - interleave first and second lines
	movq_r2r(mm1, mm2);               	// copy first line

	punpckhwd_m2r(*(dataptr+7), mm7);  	// m33:m23|m32:m22 - interleave third and fourth lines
	movq_r2r(mm6, mm5);						// copy first intermediate result

	movq_r2m(mm0, *(dataptr+8));			// write result 1
	punpckhdq_r2r(mm7, mm5);				// m33:m23|m13:m03 - produce third result

	punpcklwd_m2r(*(dataptr+10), mm1);  // n11:n01|n10:n00 - interleave first and second lines
	movq_r2r(mm3, mm0);						// copy third line

	punpckhwd_m2r(*(dataptr+10), mm2);  // n13:n03|n12:n02 - interleave first and second lines

	movq_r2m(mm4, *(dataptr+10));			// write result 2 out
	punpckldq_r2r(mm7, mm6);				// m32:m22|m12:m02 - produce fourth result

	punpcklwd_m2r(*(dataptr+14), mm3);  // n31:n21|n30:n20 - interleave third and fourth lines
	movq_r2r(mm1, mm4);

	movq_r2m(mm6, *(dataptr+12));			// write result 3 out
	punpckldq_r2r(mm3, mm1);				// n30:n20|n10:n00 - produce first result

	punpckhwd_m2r(*(dataptr+14), mm0);  // n33:n23|n32:n22 - interleave third and fourth lines
	movq_r2r(mm2, mm6);

	movq_r2m(mm5, *(dataptr+14));			// write result 4 out
	punpckhdq_r2r(mm3, mm4);				// n31:n21|n11:n01- produce second result

	movq_r2m(mm1, *(dataptr+1));			// write result 5 out - (first result for other 4 x 4 block)
	punpckldq_r2r(mm0, mm2);				// n32:n22|n12:n02- produce third result

	movq_r2m(mm4, *(dataptr+3));			// write result 6 out
	punpckhdq_r2r(mm0, mm6);				// n33:n23|n13:n03 - produce fourth result

	movq_r2m(mm2, *(dataptr+5));			// write result 7 out

	movq_m2r(*dataptr, mm0);				// m03:m02|m01:m00 - first line, first 4x4

	movq_r2m(mm6, *(dataptr+7));			// write result 8 out


// Do first 4x4 quadrant, which is used in the beginning of the DCT:

	movq_m2r(*(dataptr+4), mm7);			// m23:m22|m21:m20 - third line
	movq_r2r(mm0, mm2);

	punpcklwd_m2r(*(dataptr+2), mm0);  	// m11:m01|m10:m00 - interleave first and second lines
	movq_r2r(mm7, mm4);

	punpcklwd_m2r(*(dataptr+6), mm7);  	// m31:m21|m30:m20 - interleave third and fourth lines
	movq_r2r(mm0, mm1);

	movq_m2r(*(dataptr+2), mm6);			// m13:m12|m11:m10 - second line
	punpckldq_r2r(mm7, mm0);				// m30:m20|m10:m00 - interleave to produce result 1

	movq_m2r(*(dataptr+6), mm5);			// m33:m32|m31:m30 - fourth line
	punpckhdq_r2r(mm7, mm1);				// m31:m21|m11:m01 - interleave to produce result 2

	movq_r2r(mm0, mm7);						// write result 1
	punpckhwd_r2r(mm6, mm2);				// m13:m03|m12:m02 - interleave first and second lines

	psubw_m2r(*(dataptr+14), mm7);		// tmp07=x0-x7	/* Stage 1 */
	movq_r2r(mm1, mm6);						// write result 2

	paddw_m2r(*(dataptr+14), mm0);		// tmp00=x0+x7	/* Stage 1 */
	punpckhwd_r2r(mm5, mm4);   			// m33:m23|m32:m22 - interleave third and fourth lines

	paddw_m2r(*(dataptr+12), mm1);		// tmp01=x1+x6	/* Stage 1 */
	movq_r2r(mm2, mm3);						// copy first intermediate result

	psubw_m2r(*(dataptr+12), mm6);		// tmp06=x1-x6	/* Stage 1 */
	punpckldq_r2r(mm4, mm2);				// m32:m22|m12:m02 - interleave to produce result 3

   movq_r2m(mm7, tmp7);
	movq_r2r(mm2, mm5);						// write result 3

   movq_r2m(mm6, tmp6);
	punpckhdq_r2r(mm4, mm3);				// m33:m23|m13:m03 - interleave to produce result 4

	paddw_m2r(*(dataptr+10), mm2);     	// tmp02=x2+5 /* Stage 1 */
	movq_r2r(mm3, mm4);						// write result 4

/************************************************************************************************
					End of Transpose
************************************************************************************************/


   paddw_m2r(*(dataptr+8), mm3);    	// tmp03=x3+x4 /* stage 1*/
   movq_r2r(mm0, mm7);

   psubw_m2r(*(dataptr+8), mm4);    	// tmp04=x3-x4 /* stage 1*/
   movq_r2r(mm1, mm6);

	paddw_r2r(mm3, mm0);  					// tmp10 = tmp00 + tmp03 /* even 2 */
	psubw_r2r(mm3, mm7);  					// tmp13 = tmp00 - tmp03 /* even 2 */

	psubw_r2r(mm2, mm6);  					// tmp12 = tmp01 - tmp02 /* even 2 */
	paddw_r2r(mm2, mm1);  					// tmp11 = tmp01 + tmp02 /* even 2 */

   psubw_m2r(*(dataptr+10), mm5);    	// tmp05=x2-x5 /* stage 1*/
	paddw_r2r(mm7, mm6);						// tmp12 + tmp13

	/* stage 3 */

   movq_m2r(tmp6, mm2);
   movq_r2r(mm0, mm3);

	psllw_i2r(2, mm6);			// m8 * 2^2
	paddw_r2r(mm1, mm0);		

	pmulhw_m2r(RTjpeg_C4, mm6);			// z1
	psubw_r2r(mm1, mm3);		

   movq_r2m(mm0, *dataptr);
   movq_r2r(mm7, mm0);
   
    /* Odd part */
   movq_r2m(mm3, *(dataptr+8));
	paddw_r2r(mm5, mm4);						// tmp10

   movq_m2r(tmp7, mm3);
	paddw_r2r(mm6, mm0);						// tmp32

	paddw_r2r(mm2, mm5);						// tmp11
	psubw_r2r(mm6, mm7);						// tmp33

   movq_r2m(mm0, *(dataptr+4));
	paddw_r2r(mm3, mm2);						// tmp12

	/* stage 4 */

   movq_r2m(mm7, *(dataptr+12));
	movq_r2r(mm4, mm1);						// copy of tmp10