rtjpeg.c

这个库实现了录象功能

	psubw_r2r(mm2, mm1);    				// tmp10 - tmp12
	psllw_i2r(2, mm4);			// shift tmp10

	movq_m2r(RTjpeg_C2mC6, mm0);			// load C2mC6
	psllw_i2r(2, mm1);			// shift (tmp10-tmp12)

	pmulhw_m2r(RTjpeg_C6, mm1);    		// z5
	psllw_i2r(2, mm5);			// prepare for multiply 

	pmulhw_r2r(mm0, mm4);					// multiply by converted real

	/* stage 5 */

	pmulhw_m2r(RTjpeg_C4, mm5);			// z3
	psllw_i2r(2, mm2);			// prepare for multiply 

	pmulhw_m2r(RTjpeg_C2pC6, mm2);		// multiply
	movq_r2r(mm3, mm0);						// copy tmp7

	movq_m2r(*(dataptr+9), mm7);		 	// m03:m02|m01:m00 - first line (line 4)and copy into mm7
	paddw_r2r(mm1, mm4);						// z2

	paddw_r2r(mm5, mm0);						// z11
	psubw_r2r(mm5, mm3);						// z13

	/* stage 6 */

	movq_r2r(mm3, mm5);						// copy z13
	paddw_r2r(mm1, mm2);						// z4

	movq_r2r(mm0, mm6);						// copy z11
	psubw_r2r(mm4, mm5);						// y3

	paddw_r2r(mm2, mm6);						// y1
	paddw_r2r(mm4, mm3);						// y5

   movq_r2m(mm5, *(dataptr+7)); 			//save y3

   movq_r2m(mm6, *(dataptr+3)); 			//save y1
	psubw_r2r(mm2, mm0);						// y7
	
/************************************************************************************************
					Start of Transpose
************************************************************************************************/

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

	punpcklwd_r2r(mm3, mm7); 				// m11:m01|m10:m00 - interleave first and second lines
	movq_r2r(mm6, mm2);						// copy third line

	punpcklwd_r2r(mm0, mm6);  				// m31:m21|m30:m20 - interleave third and fourth lines
	movq_r2r(mm7, mm1);						// copy first intermediate result

	punpckldq_r2r(mm6, mm7);				// m30:m20|m10:m00 - interleave to produce result 1

	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);						// copy first intermediate result
	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);						// copy first line

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

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

	

	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);  // n33:n23|n32:n22 - interleave third and fourth lines
	movq_r2r(mm1, mm4);						// copy second intermediate result

	movq_r2m(mm6, *(dataptr+12));			// write result 3 out
	punpckldq_r2r(mm3, mm1);				// 

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

	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);						// copy first line

	punpcklwd_m2r(*(dataptr+2), mm0);  	// m11:m01|m10:m00 - interleave first and second lines
	movq_r2r(mm7, mm4);						// copy third line
	
	punpcklwd_m2r(*(dataptr+6), mm7);  	// m31:m21|m30:m20 - interleave third and fourth lines
	movq_r2r(mm0, mm1);						// copy first intermediate result

	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);						// save tmp07
	movq_r2r(mm2, mm5);						// write result 3

	movq_r2m(mm6, tmp6);						// save tmp06

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

	paddw_m2r(*(dataptr+10), mm2);   	// tmp02=x2+x5 /* stage 1 */
	movq_r2r(mm3, mm4);						// write result 4

/************************************************************************************************
					End of Transpose 2
************************************************************************************************/

   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

	psubw_r2r(mm2, mm1);						// tmp10 - tmp12
	psllw_i2r(2, mm4);			// m8 * 2^2

	movq_m2r(RTjpeg_C2mC6, mm0);
	psllw_i2r(2, mm1);

	pmulhw_m2r(RTjpeg_C6, mm1);			// z5
	psllw_i2r(2, mm2);

	pmulhw_r2r(mm0, mm4);					// z5

	/* stage 5 */

	pmulhw_m2r(RTjpeg_C2pC6, mm2);
	psllw_i2r(2, mm5);

	pmulhw_m2r(RTjpeg_C4, mm5);			// z3
	movq_r2r(mm3, mm0);						// copy tmp7

   movq_m2r(*(dataptr+1), mm7);
	paddw_r2r(mm1, mm4);						// z2

	paddw_r2r(mm1, mm2);						// z4

	paddw_r2r(mm5, mm0);						// z11
	psubw_r2r(mm5, mm3);						// z13

	/* stage 6 */

	movq_r2r(mm3, mm5);						// copy z13
	psubw_r2r(mm4, mm3);						// y3=z13 - z2

	paddw_r2r(mm4, mm5);						// y5=z13 + z2
	movq_r2r(mm0, mm6);						// copy z11

   movq_r2m(mm3, *(dataptr+6)); 			//save y3
	psubw_r2r(mm2, mm0);						// y7=z11 - z4

   movq_r2m(mm5, *(dataptr+10)); 		//save y5
	paddw_r2r(mm2, mm6);						// y1=z11 + z4

   movq_r2m(mm0, *(dataptr+14)); 		//save y7

	/************************************************
	 *  End of 1st 4 rows
	 ************************************************/

   movq_m2r(*(dataptr+3), mm1); 			// load x1   /* stage 1 */
	movq_r2r(mm7, mm0);						// copy x0

   movq_r2m(mm6, *(dataptr+2)); 			//save y1

   movq_m2r(*(dataptr+5), mm2); 			// load x2   /* stage 1 */
	movq_r2r(mm1, mm6);						// copy x1

   paddw_m2r(*(dataptr+15), mm0);  		// tmp00 = x0 + x7

   movq_m2r(*(dataptr+7), mm3); 			// load x3   /* stage 1 */
	movq_r2r(mm2, mm5);						// copy x2

   psubw_m2r(*(dataptr+15), mm7);  		// tmp07 = x0 - x7
	movq_r2r(mm3, mm4);						// copy x3

   paddw_m2r(*(dataptr+13), mm1);  		// tmp01 = x1 + x6

	movq_r2m(mm7, tmp7);						// save tmp07
	movq_r2r(mm0, mm7);						// copy tmp00

   psubw_m2r(*(dataptr+13), mm6);  		// tmp06 = x1 - x6

   /* stage 2, Even Part */

   paddw_m2r(*(dataptr+9), mm3);  		// tmp03 = x3 + x4

	movq_r2m(mm6, tmp6);						// save tmp07
	movq_r2r(mm1, mm6);						// copy tmp01

   paddw_m2r(*(dataptr+11), mm2);  		// tmp02 = x2 + x5
	paddw_r2r(mm3, mm0);              	// tmp10 = tmp00 + tmp03

	psubw_r2r(mm3, mm7);              	// tmp13 = tmp00 - tmp03

   psubw_m2r(*(dataptr+9), mm4);  		// tmp04 = x3 - x4
	psubw_r2r(mm2, mm6);              	// tmp12 = tmp01 - tmp02

	paddw_r2r(mm2, mm1);              	// tmp11 = tmp01 + tmp02

   psubw_m2r(*(dataptr+11), mm5);  		// tmp05 = x2 - x5
	paddw_r2r(mm7, mm6);              	//  tmp12 + tmp13

   /* stage 3, Even and stage 4 & 5 even */

	movq_m2r(tmp6, mm2);            		// load tmp6
	movq_r2r(mm0, mm3);						// copy tmp10

	psllw_i2r(2, mm6);			// shift z1
	paddw_r2r(mm1, mm0);    				// y0=tmp10 + tmp11

	pmulhw_m2r(RTjpeg_C4, mm6);    		// z1
	psubw_r2r(mm1, mm3);    				// y4=tmp10 - tmp11

   movq_r2m(mm0, *(dataptr+1)); 			//save y0
	movq_r2r(mm7, mm0);						// copy tmp13
  
	/* odd part */

   movq_r2m(mm3, *(dataptr+9)); 			//save y4
	paddw_r2r(mm5, mm4);              	// tmp10 = tmp4 + tmp5

	movq_m2r(tmp7, mm3);            		// load tmp7
	paddw_r2r(mm6, mm0);              	// tmp32 = tmp13 + z1

	paddw_r2r(mm2, mm5);              	// tmp11 = tmp5 + tmp6
	psubw_r2r(mm6, mm7);              	// tmp33 = tmp13 - z1

   movq_r2m(mm0, *(dataptr+5)); 			//save y2
	paddw_r2r(mm3, mm2);              	// tmp12 = tmp6 + tmp7

	/* stage 4 */

   movq_r2m(mm7, *(dataptr+13)); 		//save y6
	movq_r2r(mm4, mm1);						// copy tmp10

	psubw_r2r(mm2, mm1);    				// tmp10 - tmp12
	psllw_i2r(2, mm4);			// shift tmp10

	movq_m2r(RTjpeg_C2mC6, mm0);			// load C2mC6
	psllw_i2r(2, mm1);			// shift (tmp10-tmp12)

	pmulhw_m2r(RTjpeg_C6, mm1);    		// z5
	psllw_i2r(2, mm5);			// prepare for multiply 

	pmulhw_r2r(mm0, mm4);					// multiply by converted real

	/* stage 5 */

	pmulhw_m2r(RTjpeg_C4, mm5);			// z3
	psllw_i2r(2, mm2);			// prepare for multiply 

	pmulhw_m2r(RTjpeg_C2pC6, mm2);		// multiply
	movq_r2r(mm3, mm0);						// copy tmp7

	movq_m2r(*(dataptr+9), mm7);		 	// m03:m02|m01:m00 - first line (line 4)and copy into mm7
	paddw_r2r(mm1, mm4);						// z2

	paddw_r2r(mm5, mm0);						// z11
	psubw_r2r(mm5, mm3);						// z13

	/* stage 6 */

	movq_r2r(mm3, mm5);						// copy z13
	paddw_r2r(mm1, mm2);						// z4

	movq_r2r(mm0, mm6);						// copy z11
	psubw_r2r(mm4, mm5);						// y3

	paddw_r2r(mm2, mm6);						// y1
	paddw_r2r(mm4, mm3);						// y5

   movq_r2m(mm5, *(dataptr+7)); 			//save y3
	psubw_r2r(mm2, mm0);						// y