#include "stdafx.h" #pragma hdrstop /*************************************************************************** * * INTEL Corporation Proprietary Information * * * Copyright (c) 1996 Intel Corporation. * All rights reserved. * *************************************************************************** */ /* * jfdctfst.c * * Copyright (C) 1994-1996, Thomas G. Lane. * This file is part of the Independent JPEG Group's software. * For conditions of distribution and use, see the accompanying README file. * * This file contains a fast, not so accurate integer implementation of the * forward DCT (Discrete Cosine Transform). * * A 2-D DCT can be done by 1-D DCT on each row followed by 1-D DCT * on each column. Direct algorithms are also available, but they are * much more complex and seem not to be any faster when reduced to code. * * This implementation is based on Arai, Agui, and Nakajima's algorithm for * scaled DCT. Their original paper (Trans. IEICE E-71(11):1095) is in * Japanese, but the algorithm is described in the Pennebaker & Mitchell * JPEG textbook (see REFERENCES section in file README). The following code * is based directly on figure 4-8 in P&M. * While an 8-point DCT cannot be done in less than 11 multiplies, it is * possible to arrange the computation so that many of the multiplies are * simple scalings of the final outputs. These multiplies can then be * folded into the multiplications or divisions by the JPEG quantization * table entries. The AA&N method leaves only 5 multiplies and 29 adds * to be done in the DCT itself. * The primary disadvantage of this method is that with fixed-point math, * accuracy is lost due to imprecise representation of the scaled * quantization values. The smaller the quantization table entry, the less * precise the scaled value, so this implementation does worse with high- * quality-setting files than with low-quality ones. */ #define JPEG_INTERNALS #include "jinclude.h" #include "jpeglib.h" #include "jdct.h" /* Private declarations for DCT subsystem */ #ifdef DCT_IFAST_SUPPORTED /* * This module is specialized to the case DCTSIZE = 8. */ #if DCTSIZE != 8 Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */ #endif /* Scaling decisions are generally the same as in the LL&M algorithm; * see jfdctint.c for more details. However, we choose to descale * (right shift) multiplication products as soon as they are formed, * rather than carrying additional fractional bits into subsequent additions. * This compromises accuracy slightly, but it lets us save a few shifts. * More importantly, 16-bit arithmetic is then adequate (for 8-bit samples) * everywhere except in the multiplications proper; this saves a good deal * of work on 16-bit-int machines. * * Again to save a few shifts, the intermediate results between pass 1 and * pass 2 are not upscaled, but are represented only to integral precision. * * A final compromise is to represent the multiplicative constants to only * 8 fractional bits, rather than 13. This saves some shifting work on some * machines, and may also reduce the cost of multiplication (since there * are fewer one-bits in the constants). */ #define CONST_BITS 8 /* Some C compilers fail to reduce "FIX(constant)" at compile time, thus * causing a lot of useless floating-point operations at run time. * To get around this we use the following pre-calculated constants. * If you change CONST_BITS you may want to add appropriate values. * (With a reasonable C compiler, you can just rely on the FIX() macro...) */ #if CONST_BITS == 8 #define FIX_0_382683433 98 /* FIX(0.382683433) */ #define FIX_0_541196100 139 /* FIX(0.541196100) */ #define FIX_0_707106781 181 /* FIX(0.707106781) */ #define FIX_1_306562965 334 /* FIX(1.306562965) */ #else #define FIX_0_382683433 FIX(0.382683433) #define FIX_0_541196100 FIX(0.541196100) #define FIX_0_707106781 FIX(0.707106781) #define FIX_1_306562965 FIX(1.306562965) #endif //The following constant is shifted left 8 for the pmulhw instruction const __int64 Const_FIX_0_382683433 = 0x6200620062006200; //The following constants are shifted left 7 for the pmulhw instruction const __int64 Const_FIX_0_541196100 = 0x4580458045804580; const __int64 Const_FIX_0_707106781 = 0x5a805a805a805a80; //The following constant is shifted left 6 for the pmulhw instruction const __int64 Const_FIX_1_306562965 = 0x5380538053805380; /* We can gain a little more speed, with a further compromise in accuracy, * by omitting the addition in a descaling shift. This yields an incorrectly * rounded result half the time... */ // The assembly version makes this compromise. //#ifndef USE_ACCURATE_ROUNDING //#undef DESCALE //#define DESCALE(x,n) RIGHT_SHIFT(x, n) //#endif #define DATASIZE 32 /* Multiply a DCTELEM variable by an INT32 constant, and immediately * descale to yield a DCTELEM result. */ #define MULTIPLY(var,const) ((DCTELEM) DESCALE((var) * (const), CONST_BITS)) /* * Perform the forward DCT on one block of samples. */ GLOBAL(void) mfdct8x8aan (DCTELEM * data) { __asm{ mov edi, [data] // transpose the bottom right quadrant(4X4) of the matrix // --------- --------- // | M1 | M2 | | M1'| M3'| // --------- --> --------- // | M3 | M4 | | M2'| M4'| // --------- --------- // Get the 32-bit quantities and pack into 16 bits movq mm5, [edi][DATASIZE*4+16] //| w41 | w40 | movq mm3, [edi][DATASIZE*4+24] //| w43 | w42 | movq mm6, [edi][DATASIZE*5+16] packssdw mm5, mm3 //|w43|w42|w41|w40| movq mm7, [edi][DATASIZE*5+24] movq mm4, mm5 // copy w4---0,1,3,5,6 movq mm3, [edi][DATASIZE*6+16] packssdw mm6, mm7 movq mm2, [edi][DATASIZE*6+24] punpcklwd mm5, mm6 //mm6 = w5 movq mm1, [edi][DATASIZE*7+16] packssdw mm3, mm2 movq mm0, [edi][DATASIZE*7+24] punpckhwd mm4, mm6 //---0,1,3,5,6 packssdw mm1, mm0 movq mm7, mm3 //---0,1,2,3,5,6 w6 punpcklwd mm3, mm1 //mm1 = w7 movq mm0, mm5 //---0,2,3,4,5,6,7 movq mm2, [edi][DATASIZE*4] //| w01 | w00 | punpckhdq mm0, mm3 // transposed w5---0,2,4,6,7 punpckhwd mm7, mm1 //---0,2,3,5,6,7 movq mm1, [edi][DATASIZE*5+8] movq mm6, mm4 //---0,2,3,4,6,7 movq [edi][DATASIZE*5+16], mm0 // store w5 punpckldq mm5, mm3 // transposed w4 movq mm3, [edi][DATASIZE*5] punpckldq mm4, mm7 // transposed w6 movq mm0, [edi][DATASIZE*4+8] //| w03 | w02 | punpckhdq mm6, mm7 // transposed w7---0,3,6,7 // transpose the bottom left quadrant(4X4) of the matrix and place // in the top right quadrant while doing the same for the top // right quadrant // --------- --------- // | M1 | M2 | | M1'| M3'| // --------- --> --------- // | M3 | M4 | | M2'| M4'| // --------- --------- movq [edi][DATASIZE*4+16], mm5 // store w4 packssdw mm2, mm0 //|w03|w02|w01|w00| movq mm5, [edi][DATASIZE*7] packssdw mm3, mm1 movq mm0, [edi][DATASIZE*7+8] movq [edi][DATASIZE*7+16], mm6 // store w7---5,6,7 packssdw mm5, mm0 movq mm6, [edi][DATASIZE*6] movq mm0, mm2 // copy w0---0,1,3,5,6 movq mm7, [edi][DATASIZE*6+8] punpcklwd mm2, mm3 //mm6 = w1 movq [edi][DATASIZE*6+16], mm4 // store w6---3,5,6,7 packssdw mm6, mm7 movq mm1, [edi][DATASIZE*0+24] punpckhwd mm0, mm3 //---0,1,3,5,6 movq mm7, mm6 //---0,1,2,3,5,6 w2 punpcklwd mm6, mm5 //mm1 = w3 movq mm3, [edi][DATASIZE*0+16] punpckhwd mm7, mm5 //---0,2,3,5,6,7 movq mm4, [edi][DATASIZE*2+24] packssdw mm3, mm1 movq mm1, mm2 //---0,2,3,4,5,6,7 punpckldq mm2, mm6 // transposed w4 movq mm5, [edi][DATASIZE*2+16] punpckhdq mm1, mm6 // transposed w5---0,2,4,6,7 movq [edi][DATASIZE*0+16], mm2 // store w4 packssdw mm5, mm4 movq mm4, [edi][DATASIZE*1+16] movq mm6, mm0 //---0,2,3,4,6,7 movq mm2, [edi][DATASIZE*1+24] punpckldq mm0, mm7 // transposed w6 movq [edi][DATASIZE*1+16], mm1 // store w5 punpckhdq mm6, mm7 // transposed w7---0,3,6,7 movq mm7, [edi][DATASIZE*3+24] packssdw mm4, mm2 movq [edi][DATASIZE*2+16], mm0 // store w6---3,5,6,7 movq mm1, mm3 // copy w4---0,1,3,5,6 movq mm2, [edi][DATASIZE*3+16] punpcklwd mm3, mm4 //mm6 = w5 movq [edi][DATASIZE*3+16], mm6 // store w7---5,6,7 packssdw mm2, mm7 // transpose the bottom left quadrant(4X4) of the matrix // --------- --------- // | M1 | M2 | | M1'| M3'| // --------- --> --------- // | M3 | M4 | | M2'| M4'| // --------- --------- movq mm6, [edi][DATASIZE*0] //| w01 | w00 | punpckhwd mm1, mm4 //---0,1,3,5,6 movq mm7, mm5 //---0,1,2,3,5,6 w6 punpcklwd mm5, mm2 //mm1 = w7 movq mm4, [edi][DATASIZE*0+8] //| w03 | w02 | punpckhwd mm7, mm2 //---0,2,3,5,6,7 movq mm0, mm3 //---0,2,3,4,5,6,7 packssdw mm6, mm4 //|w03|w02|w01|w00| movq mm2, [edi][DATASIZE*2+8] punpckldq mm3, mm5 // transposed w4 movq mm4, [edi][DATASIZE*1] punpckhdq mm0, mm5 // transposed w5---0,2,4,6,7 movq [edi][DATASIZE*4], mm3 // store w4 movq mm5, mm1 //---0,2,3,4,6,7 movq mm3, [edi][DATASIZE*2] punpckldq mm1, mm7 // transposed w6 movq [edi][DATASIZE*5], mm0 // store w5 punpckhdq mm5, mm7 // transposed w7---0,3,6,7 movq mm7, [edi][DATASIZE*1+8] packssdw mm3, mm2 movq [edi][DATASIZE*7], mm5 // store w7---5,6,7 movq mm5, mm6 // copy w0---0,1,3,5,6 movq [edi][DATASIZE*6], mm1 // store w6---3,5,6,7 packssdw mm4, mm7 // transpose the top left quadrant(4X4) of the matrix // --------- --------- // | M1 | M2 | | M1'| M3'| // --------- --> --------- // | M3 | M4 | | M2'| M4'| // --------- --------- // Get the 32-bit quantities and pack into 16 bits movq mm1, [edi][DATASIZE*3] punpcklwd mm6, mm4 //mm6 = w1 movq mm0, [edi][DATASIZE*3+8] punpckhwd mm5, mm4 //---0,1,3,5,6 packssdw mm1, mm0 movq mm2, mm3 //---0,1,2,3,5,6 w2 punpcklwd mm3, mm1 //mm1 = w3 movq mm0, mm6 //---0,2,3,4,5,6,7 movq mm4, [edi][DATASIZE*7] punpckhwd mm2, mm1 //---0,2,3,5,6,7 movq mm1, [edi][DATASIZE*4] punpckldq mm6, mm3 // transposed w4 punpckhdq mm0, mm3 // transposed w5---0,2,4,6,7 movq mm3, mm5 //---0,2,3,4,6,7 movq [edi][DATASIZE*0], mm6 // store w4 punpckldq mm5, mm2 // transposed w6 movq [edi][DATASIZE*1], mm0 // store w5 punpckhdq mm3, mm2 // transposed w7---0,3,6,7 movq [edi][DATASIZE*2], mm5 // store w6---3,5,6,7 paddw mm6, mm4 // tmp0 movq [edi][DATASIZE*3], mm3 // store w7---5,6,7 movq mm7, mm6 //****************************************************************************** // End of transpose. Begin row dct. //****************************************************************************** // tmp0 = dataptr[0] + dataptr[7]; // tmp7 = dataptr[0] - dataptr[7]; // tmp1 = dataptr[1] + dataptr[6]; // tmp6 = dataptr[1] - dataptr[6]; // tmp2 = dataptr[2] + dataptr[5]; // tmp5 = dataptr[2] - dataptr[5]; // tmp3 = dataptr[3] + dataptr[4]; // tmp4 = dataptr[3] - dataptr[4]; paddw mm0, [edi][DATASIZE*6] // tmp1 paddw mm3, mm1 // tmp3 paddw mm5, [edi][DATASIZE*5] // tmp2 movq mm1, mm0 // tmp10 = tmp0 + tmp3; // tmp13 = tmp0 - tmp3; // tmp11 = tmp1 + tmp2; // tmp12 = tmp1 - tmp2; psubw mm7, mm3 //tmp13 psubw mm0, mm5 //tmp12 paddw mm0, mm7 //tmp12 + tmp13 paddw mm6, mm3 //tmp10 // dataptr[0] = tmp10 + tmp11; /* phase 3 */ // dataptr[4] = tmp10 - tmp11; // z1 = MULTIPLY(tmp12 + tmp13, FIX_0_707106781); /* c4 */ //NOTE: We can't write these values out immediately. Values for tmp4 - tmp7 //haven't been calculated yet! paddw mm1, mm5 //tmp11 psllw mm0, 1 pmulhw mm0, Const_FIX_0_707106781 // z1 movq mm3, mm6 // dataptr[2] = tmp13 + z1; /* phase 5 */ // dataptr[6] = tmp13 - z1; //NOTE: We can't write these values out immediately. Values for tmp4 - tmp7 //haven't been calculated yet! movq mm5, [edi][DATASIZE*3] paddw mm6, mm1 //tmp10 + tmp11 // tmp4 = dataptr[3] - dataptr[4]// psubw mm5, [edi][DATASIZE*4] //tmp4 movq mm4, mm7 movq mm2, [edi][DATASIZE*2] psubw mm3, mm1 //tmp10 - tmp11 psubw mm2, [edi][DATASIZE*5] //tmp5 paddw mm7, mm0 //tmp13 + z1 movq mm1, [edi][DATASIZE*1] psubw mm4, mm0 //tmp13 - z1 // tmp10 = tmp4 + tmp5; /* phase 2 */ // tmp11 = tmp5 + tmp6; // tmp12 = tmp6 + tmp7; psubw mm1, [edi][DATASIZE*6] //tmp6 paddw mm5, mm2 //tmp10 movq mm0, [edi][DATASIZE*0] paddw mm2, mm1 //tmp11 // z3 = MULTIPLY(tmp11, FIX_0_707106781); /* c4 */ // z11 = tmp7 + z3; /* phase 5 */ // z13 = tmp7 - z3; psubw mm0, [edi][DATASIZE*7] //tmp7 psllw mm2, 1 movq [edi][DATASIZE*0], mm6 movq mm6, mm0 movq [edi][DATASIZE*2], mm7 movq mm7, mm5 pmulhw mm2, Const_FIX_0_707106781 //z3 paddw mm1, mm0 //tmp12 movq [edi][DATASIZE*4], mm3 psubw mm5, mm1 //tmp10 - tmp12 pmulhw mm5, Const_FIX_0_382683433 //z5 psllw mm7, 1 /* The rotator is modified from fig 4-8 to avoid extra negations. */ // z5 = MULTIPLY(tmp10 - tmp12, FIX_0_382683433); /* c6 */ // z2 = MULTIPLY(tmp10, FIX_0_541196100) + z5; /* c2-c6 */ // z4 = MULTIPLY(tmp12, FIX_1_306562965) + z5; /* c2+c6 */ pmulhw mm7, Const_FIX_0_541196100 psllw mm1, 2 pmulhw mm1, Const_FIX_1_306562965 psubw mm6, mm2 //z13 movq [edi][DATASIZE*6], mm4 paddw mm0, mm2 //z11 movq mm2, [edi][DATASIZE*3+16] paddw mm7, mm5 //z2 paddw mm2, [edi][DATASIZE*4+16] // tmp3 paddw mm1, mm5 //z4 // dataptr[5] = z13 + z2; /* phase 6 */ // dataptr[3] = z13 - z2; // dataptr[1] = z11 + z4; // dataptr[7] = z11 - z4; movq mm5, [edi][DATASIZE*0+16] movq mm3, mm6 paddw mm5, [edi][DATASIZE*7+16] //tmp0 paddw mm6, mm7 //z13 + z2 psubw mm3, mm7 //z13 - z2 movq mm7, mm5 movq [edi][DATASIZE*5], mm6 //store movq mm4, mm0 movq [edi][DATASIZE*3], mm3 //store paddw mm0, mm1 //z11 + z4 movq mm3, [edi][DATASIZE*1+16] psubw mm4, mm1 //z11 - z4 //****************************************************************************** // This completes 4x8 dct locations. Copy to do other 4x8. //****************************************************************************** // tmp0 = dataptr[0] + dataptr[7]; // tmp7 = dataptr[0] - dataptr[7]; // tmp1 = dataptr[1] + dataptr[6]; // tmp6 = dataptr[1] - dataptr[6]; // tmp2 = dataptr[2] + dataptr[5]; // tmp5 = dataptr[2] - dataptr[5]; // tmp3 = dataptr[3] + dataptr[4]; // tmp4 = dataptr[3] - dataptr[4]; paddw mm3, [edi][DATASIZE*6+16] // tmp1 paddw mm5, mm2 //tmp10 movq mm1, [edi][DATASIZE*2+16] psubw mm7, mm2 //tmp13 paddw mm1, [edi][DATASIZE*5+16] // tmp2 movq mm6, mm3 // tmp10 = tmp0 + tmp3; // tmp13 = tmp0 - tmp3; // tmp11 = tmp1 + tmp2; // tmp12 = tmp1 - tmp2; paddw mm3, mm1 //tmp11 psubw mm6, mm1 //tmp12 // dataptr[0] = tmp10 + tmp11; /* phase 3 */ // dataptr[4] = tmp10 - tmp11; // z1 = MULTIPLY(tmp12 + tmp13, FIX_0_707106781); /* c4 */ //NOTE: We can't write these values out immediately. Values for tmp4 - tmp7 //haven't been calculated yet! movq [edi][DATASIZE*1], mm0 //store paddw mm6, mm7 //tmp12 + tmp13 movq [edi][DATASIZE*7], mm4 //store psllw mm6, 1 pmulhw mm6, Const_FIX_0_707106781 // z1 movq mm1, mm5 // dataptr[2] = tmp13 + z1; /* phase 5 */ // dataptr[6] = tmp13 - z1; //NOTE: We can't write these values out immediately. Values for tmp4 - tmp7 //haven't been calculated yet! movq mm2, [edi][DATASIZE*3+16] paddw mm5, mm3 //tmp10 + tmp11 // tmp4 = dataptr[3] - dataptr[4]// psubw mm2, [edi][DATASIZE*4+16] //tmp4 movq mm4, mm7 movq mm0, [edi][DATASIZE*2+16] psubw mm1, mm3 //tmp10 - tmp11 psubw mm0, [edi][DATASIZE*5+16] //tmp5 paddw mm7, mm6 //tmp13 + z1 movq mm3, [edi][DATASIZE*1+16] psubw mm4, mm6 //tmp13 - z1 // tmp10 = tmp4 + tmp5; /* phase 2 */ // tmp11 = tmp5 + tmp6; // tmp12 = tmp6 + tmp7; psubw mm3, [edi][DATASIZE*6+16] //tmp6 paddw mm2, mm0 //tmp10 movq mm6, [edi][DATASIZE*0+16] paddw mm0, mm3 //tmp11 // z3 = MULTIPLY(tmp11, FIX_0_707106781); /* c4 */ // z11 = tmp7 + z3; /* phase 5 */ // z13 = tmp7 - z3; psubw mm6, [edi][DATASIZE*7+16] //tmp7 psllw mm0, 1 movq [edi][DATASIZE*0+16], mm5 movq mm5, mm6 movq [edi][DATASIZE*2+16], mm7 movq mm7, mm2 pmulhw mm0, Const_FIX_0_707106781 //z3 paddw mm3, mm6 //tmp12 movq [edi][DATASIZE*4+16], mm1 psubw mm2, mm3 //tmp10 - tmp12 pmulhw mm2, Const_FIX_0_382683433 //z5 psllw mm7, 1 pmulhw mm7, Const_FIX_0_541196100 paddw mm6, mm0 //z11 /* The rotator is modified from fig 4-8 to avoid extra negations. */ // z5 = MULTIPLY(tmp10 - tmp12, FIX_0_382683433); /* c6 */ // z2 = MULTIPLY(tmp10, FIX_0_541196100) + z5; /* c2-c6 */ // z4 = MULTIPLY(tmp12, FIX_1_306562965) + z5; /* c2+c6 */ movq [edi][DATASIZE*6+16], mm4 psllw mm3, 2 pmulhw mm3, Const_FIX_1_306562965 psubw mm5, mm0 //z13 paddw mm7, mm2 //z2 movq mm1, mm5 paddw mm5, mm7 //z13 + z2 psubw mm1, mm7 //z13 - z2 movq mm7, [edi][DATASIZE*4] paddw mm3, mm2 //z4 // dataptr[5] = z13 + z2; /* phase 6 */ // dataptr[3] = z13 - z2; // dataptr[1] = z11 + z4; // dataptr[7] = z11 - z4; movq [edi][DATASIZE*5+16], mm5 //store movq mm4, mm6 movq mm2, [edi][DATASIZE*7] paddw mm6, mm3 //z11 + z4 movq mm5, [edi][DATASIZE*5] psubw mm4, mm3 //z11 - z4 //****************************************************************************** //****************************************************************************** // This completes all 8x8 dct locations for the row case. // Now transpose the data for the columns. //****************************************************************************** // transpose the bottom left quadrant(4X4) of the matrix and place // in the top right quadrant while doing the same for the top // right quadrant // --------- --------- // | M1 | M2 | | M1'| M3'| // --------- --> --------- // | M3 | M4 | | M2'| M4'| // --------- --------- movq mm0, mm7 // copy w0---0,1,3,5,6 punpcklwd mm7, mm5 //mm6 = w1 movq mm3, [edi][DATASIZE*6] punpckhwd mm0, mm5 //---0,1,3,5,6 movq mm5, mm3 //---0,1,2,3,5,6 w2 punpcklwd mm3, mm2 //mm1 = w3 movq [edi][DATASIZE*7+16], mm4 //store punpckhwd mm5, mm2 //---0,2,3,5,6,7 movq mm4, mm7 //---0,2,3,4,5,6,7 punpckldq mm7, mm3 // transposed w4 movq mm2, [edi][DATASIZE*0+16] punpckhdq mm4, mm3 // transposed w5---0,2,4,6,7 movq [edi][DATASIZE*0+16], mm7 // store w4 movq mm3, mm0 //---0,2,3,4,6,7 movq [edi][DATASIZE*1+16], mm4 // store w5 punpckldq mm0, mm5 // transposed w6 movq mm7, [edi][DATASIZE*2+16] punpckhdq mm3, mm5 // transposed w7---0,3,6,7 movq mm5, mm2 // copy w4---0,1,3,5,6 punpcklwd mm2, mm6 //mm6 = w5 // transpose the top right quadrant(4X4) of the matrix // --------- --------- // | M1 | M2 | | M1'| M3'| // --------- --> --------- // | M3 | M4 | | M2'| M4'| // --------- --------- movq [edi][DATASIZE*2+16], mm0 // store w6---3,5,6,7 punpckhwd mm5, mm6 //---0,1,3,5,6 movq mm4, mm7 //---0,1,2,3,5,6 w6 punpckhwd mm7, mm1 //---0,2,3,5,6,7 movq [edi][DATASIZE*3+16], mm3 // store w7---5,6,7 movq mm0, mm2 //---0,2,3,4,5,6,7 movq mm6, [edi][DATASIZE*5+16] punpcklwd mm4, mm1 //mm1 = w7 movq mm1, [edi][DATASIZE*4+16] punpckldq mm0, mm4 // transposed w4 movq mm3, [edi][DATASIZE*6+16] punpckhdq mm2, mm4 // transposed w5---0,2,4,6,7 // transpose the bottom right quadrant(4X4) of the matrix // --------- --------- // | M1 | M2 | | M1'| M3'| // --------- --> --------- // | M3 | M4 | | M2'| M4'| // --------- --------- movq [edi][DATASIZE*4], mm0 // store w4 movq mm4, mm5 //---0,2,3,4,6,7 movq [edi][DATASIZE*5], mm2 // store w5 punpckldq mm5, mm7 // transposed w6 movq mm2, [edi][DATASIZE*7+16] punpckhdq mm4, mm7 // transposed w7---0,3,6,7 movq mm7, mm1 // copy w4---0,1,3,5,6 punpcklwd mm1, mm6 //mm6 = w5 movq [edi][DATASIZE*6], mm5 // store w6---3,5,6,7 punpckhwd mm7, mm6 //---0,1,3,5,6 movq mm5, mm3 //---0,1,2,3,5,6 w6 punpcklwd mm3, mm2 //mm1 = w7 movq [edi][DATASIZE*7], mm4 // store w7---5,6,7 punpckhwd mm5, mm2 //---0,2,3,5,6,7 movq mm0, [edi][DATASIZE*0] movq mm4, mm1 //---0,2,3,4,5,6,7 movq mm6, [edi][DATASIZE*1] punpckldq mm1, mm3 // transposed w4 punpckhdq mm4, mm3 // transposed w5---0,2,4,6,7 movq mm3, mm7 //---0,2,3,4,6,7 movq [edi][DATASIZE*4+16], mm1 // store w4 punpckldq mm7, mm5 // transposed w6 movq [edi][DATASIZE*5+16], mm4 // store w5 punpckhdq mm3, mm5 // transposed w7---0,3,6,7 // transpose the top left quadrant(4X4) of the matrix // --------- --------- // | M1 | M2 | | M1'| M3'| // --------- --> --------- // | M3 | M4 | | M2'| M4'| // --------- --------- movq mm1, [edi][DATASIZE*3] movq mm2, mm0 // copy w0---0,1,3,5,6 movq [edi][DATASIZE*7+16], mm3 // store w7---5,6,7 punpcklwd mm0, mm6 //mm6 = w1 movq mm3, [edi][DATASIZE*2] punpckhwd mm2, mm6 //---0,1,3,5,6 movq mm5, mm3 //---0,1,2,3,5,6 w2 punpcklwd mm3, mm1 //mm1 = w3 movq [edi][DATASIZE*6+16], mm7 // store w6---3,5,6,7 punpckhwd mm5, mm1 //---0,2,3,5,6,7 movq mm1, mm0 //---0,2,3,4,5,6,7 punpckldq mm0, mm3 // transposed w4 movq mm6, [edi][DATASIZE*4] punpckhdq mm1, mm3 // transposed w5---0,2,4,6,7 movq [edi][DATASIZE*0], mm0 // store w4 movq mm3, mm2 //---0,2,3,4,6,7 paddw mm0, [edi][DATASIZE*7] // tmp0 punpckhdq mm3, mm5 // transposed w7---0,3,6,7 movq [edi][DATASIZE*1], mm1 // store w5 punpckldq mm2, mm5 // transposed w6 //****************************************************************************** // This begins the column dct //****************************************************************************** // tmp0 = dataptr[0] + dataptr[7]; // tmp7 = dataptr[0] - dataptr[7]; // tmp1 = dataptr[1] + dataptr[6]; // tmp6 = dataptr[1] - dataptr[6]; // tmp2 = dataptr[2] + dataptr[5]; // tmp5 = dataptr[2] - dataptr[5]; // tmp3 = dataptr[3] + dataptr[4]; // tmp4 = dataptr[3] - dataptr[4]; movq [edi][DATASIZE*3], mm3 // store w7---5,6,7 movq mm7, mm0 paddw mm1, [edi][DATASIZE*6] // tmp1 paddw mm3, mm6 // tmp3 movq [edi][DATASIZE*2], mm2 // store w6---3,5,6,7 paddw mm0, mm3 //tmp10 paddw mm2, [edi][DATASIZE*5] // tmp2 movq mm6, mm1 // tmp10 = tmp0 + tmp3; // tmp13 = tmp0 - tmp3; // tmp11 = tmp1 + tmp2; // tmp12 = tmp1 - tmp2; psubw mm7, mm3 //tmp13 movq mm3, mm0 movq mm5, [edi][DATASIZE*2] paddw mm1, mm2 //tmp11 psubw mm3, mm1 //tmp10 - tmp11 paddw mm0, mm1 //tmp10 + tmp11 // dataptr[0] = tmp10 + tmp11; /* phase 3 */ // dataptr[4] = tmp10 - tmp11; //NOTE: We can't write these values out immediately. Values for tmp4 - tmp7 //haven't been calculated yet! movq mm1, mm3 punpcklwd mm3, mm3 psubw mm6, mm2 //tmp12 punpckhwd mm1, mm1 movq mm2, [edi][DATASIZE*3] psrad mm3, 16 // tmp4 = dataptr[3] - dataptr[4]// psubw mm2, [edi][DATASIZE*4] //tmp4 psrad mm1, 16 movq [edi][DATASIZE*4], mm3 movq mm3, mm0 movq [edi][DATASIZE*4+8], mm1 punpcklwd mm0, mm0 paddw mm6, mm7 //tmp12 + tmp13 punpckhwd mm3, mm3 movq mm1, [edi][DATASIZE*1] psllw mm6, 1 pmulhw mm6, Const_FIX_0_707106781 // z1 psrad mm3, 16 psubw mm5, [edi][DATASIZE*5] //tmp5 psrad mm0, 16 // z1 = MULTIPLY(tmp12 + tmp13, FIX_0_707106781); /* c4 */ // dataptr[2] = tmp13 + z1; /* phase 5 */ // dataptr[6] = tmp13 - z1; //NOTE: We can't write these values out immediately. Values for tmp4 - tmp7 //haven't been calculated yet! movq [edi][DATASIZE*0+8], mm3 movq mm4, mm7 movq mm3, [edi][DATASIZE*0] paddw mm7, mm6 //tmp13 + z1 movq [edi][DATASIZE*0], mm0 psubw mm4, mm6 //tmp13 - z1 movq mm0, mm7 punpcklwd mm7, mm7 psubw mm1, [edi][DATASIZE*6] //tmp6 punpckhwd mm0, mm0 // tmp10 = tmp4 + tmp5; /* phase 2 */ // tmp11 = tmp5 + tmp6; // tmp12 = tmp6 + tmp7; psrad mm7, 16 paddw mm2, mm5 //tmp10 psrad mm0, 16 paddw mm5, mm1 //tmp11 movq mm6, mm4 punpcklwd mm4, mm4 movq [edi][DATASIZE*2], mm7 punpckhwd mm6, mm6 psubw mm3, [edi][DATASIZE*7] //tmp7 movq mm7, mm2 // z3 = MULTIPLY(tmp11, FIX_0_707106781); /* c4 */ // z11 = tmp7 + z3; /* phase 5 */ // z13 = tmp7 - z3; movq [edi][DATASIZE*2+8], mm0 movq mm0, mm3 psllw mm5, 1 paddw mm1, mm3 //tmp12 pmulhw mm5, Const_FIX_0_707106781 //z3 psrad mm4, 16 psubw mm2, mm1 //tmp10 - tmp12 psrad mm6, 16 /* The rotator is modified from fig 4-8 to avoid extra negations. */ // z5 = MULTIPLY(tmp10 - tmp12, FIX_0_382683433); /* c6 */ // z2 = MULTIPLY(tmp10, FIX_0_541196100) + z5; /* c2-c6 */ // z4 = MULTIPLY(tmp12, FIX_1_306562965) + z5; /* c2+c6 */ pmulhw mm2, Const_FIX_0_382683433 //z5 psllw mm7, 1 pmulhw mm7, Const_FIX_0_541196100 psllw mm1, 2 pmulhw mm1, Const_FIX_1_306562965 psubw mm0, mm5 //z13 movq [edi][DATASIZE*6+8], mm6 movq mm6, mm0 movq [edi][DATASIZE*6], mm4 paddw mm7, mm2 //z2 // dataptr[5] = z13 + z2; /* phase 6 */ // dataptr[3] = z13 - z2; // dataptr[1] = z11 + z4; // dataptr[7] = z11 - z4; paddw mm0, mm7 //z13 + z2 psubw mm6, mm7 //z13 - z2 movq mm7, mm6 punpcklwd mm6, mm6 punpckhwd mm7, mm7 paddw mm3, mm5 //z11 movq mm5, mm0 punpcklwd mm0, mm0 psrad mm6, 16 movq mm4, mm3 psrad mm7, 16 paddw mm1, mm2 //z4 punpckhwd mm5, mm5 paddw mm3, mm1 //z11 + z4 psrad mm0, 16 psubw mm4, mm1 //z11 - z4 movq [edi][DATASIZE*3], mm6 //store psrad mm5, 16 movq mm6, [edi][DATASIZE*1+16] movq mm1, mm3 paddw mm6, [edi][DATASIZE*6+16] // tmp1 punpcklwd mm3, mm3 movq [edi][DATASIZE*3+8], mm7 punpckhwd mm1, mm1 movq [edi][DATASIZE*5], mm0 //store psrad mm3, 16 movq [edi][DATASIZE*5+8], mm5 psrad mm1, 16 movq mm0, [edi][DATASIZE*0+16] movq mm7, mm4 paddw mm0, [edi][DATASIZE*7+16] //tmp0 punpcklwd mm4, mm4 movq [edi][DATASIZE*1], mm3 //store punpckhwd mm7, mm7 movq [edi][DATASIZE*1+8], mm1 psrad mm4, 16 movq mm3, [edi][DATASIZE*3+16] psrad mm7, 16 //****************************************************************************** // This completes 4x8 dct locations. Copy to do other 4x8. //****************************************************************************** // tmp0 = dataptr[0] + dataptr[7]; // tmp7 = dataptr[0] - dataptr[7]; // tmp1 = dataptr[1] + dataptr[6]; // tmp6 = dataptr[1] - dataptr[6]; // tmp2 = dataptr[2] + dataptr[5]; // tmp5 = dataptr[2] - dataptr[5]; // tmp3 = dataptr[3] + dataptr[4]; // tmp4 = dataptr[3] - dataptr[4]; paddw mm3, [edi][DATASIZE*4+16] // tmp3 movq mm1, mm6 movq [edi][DATASIZE*7+8], mm7 movq mm7, mm0 movq mm2, [edi][DATASIZE*2+16] paddw mm0, mm3 //tmp10 paddw mm2, [edi][DATASIZE*5+16] // tmp2 psubw mm7, mm3 //tmp13 movq mm3, mm0 paddw mm1, mm2 //tmp11 // tmp10 = tmp0 + tmp3; // tmp13 = tmp0 - tmp3; // tmp11 = tmp1 + tmp2; // tmp12 = tmp1 - tmp2; paddw mm0, mm1 //tmp10 + tmp11 psubw mm3, mm1 //tmp10 - tmp11 // dataptr[0] = tmp10 + tmp11; /* phase 3 */ // dataptr[4] = tmp10 - tmp11; // z1 = MULTIPLY(tmp12 + tmp13, FIX_0_707106781); /* c4 */ //NOTE: We can't write these values out immediately. Values for tmp4 - tmp7 //haven't been calculated yet! movq mm1, mm3 punpcklwd mm3, mm3 punpckhwd mm1, mm1 psubw mm6, mm2 //tmp12 movq [edi][DATASIZE*7], mm4 //store psrad mm3, 16 psrad mm1, 16 paddw mm6, mm7 //tmp12 + tmp13 movq mm2, [edi][DATASIZE*3+16] psllw mm6, 1 movq mm4, mm0 punpcklwd mm0, mm0 pmulhw mm6, Const_FIX_0_707106781 // z1 punpckhwd mm4, mm4 // tmp4 = dataptr[3] - dataptr[4]// psubw mm2, [edi][DATASIZE*4+16] //tmp4 psrad mm4, 16 movq mm5, [edi][DATASIZE*2+16] psrad mm0, 16 movq [edi][DATASIZE*0+24], mm4 movq mm4, mm7 // dataptr[2] = tmp13 + z1; /* phase 5 */ // dataptr[6] = tmp13 - z1; //NOTE: We can't write these values out immediately. Values for tmp4 - tmp7 //haven't been calculated yet! psubw mm5, [edi][DATASIZE*5+16] //tmp5 paddw mm7, mm6 //tmp13 + z1 movq [edi][DATASIZE*4+16], mm3 psubw mm4, mm6 //tmp13 - z1 movq mm3, mm7 punpcklwd mm7, mm7 movq mm6, [edi][DATASIZE*0+16] punpckhwd mm3, mm3 movq [edi][DATASIZE*4+24], mm1 psrad mm7, 16 movq [edi][DATASIZE*0+16], mm0 psrad mm3, 16 movq mm1, [edi][DATASIZE*1+16] movq mm0, mm4 psubw mm1, [edi][DATASIZE*6+16] //tmp6 punpcklwd mm4, mm4 movq [edi][DATASIZE*2+16], mm7 paddw mm2, mm5 //tmp10 // tmp10 = tmp4 + tmp5; /* phase 2 */ // tmp11 = tmp5 + tmp6; // tmp12 = tmp6 + tmp7; movq mm7, mm2 paddw mm5, mm1 //tmp11 psubw mm6, [edi][DATASIZE*7+16] //tmp7 punpckhwd mm0, mm0 movq [edi][DATASIZE*2+24], mm3 psllw mm5, 1 pmulhw mm5, Const_FIX_0_707106781 //z3 psrad mm0, 16 psrad mm4, 16 paddw mm1, mm6 //tmp12 // z3 = MULTIPLY(tmp11, FIX_0_707106781); /* c4 */ // z11 = tmp7 + z3; /* phase 5 */ // z13 = tmp7 - z3; movq [edi][DATASIZE*6+24], mm0 psubw mm2, mm1 //tmp10 - tmp12 /* The rotator is modified from fig 4-8 to avoid extra negations. */ // z5 = MULTIPLY(tmp10 - tmp12, FIX_0_382683433); /* c6 */ // z2 = MULTIPLY(tmp10, FIX_0_541196100) + z5; /* c2-c6 */ // z4 = MULTIPLY(tmp12, FIX_1_306562965) + z5; /* c2+c6 */ pmulhw mm2, Const_FIX_0_382683433 //z5 psllw mm7, 1 pmulhw mm7, Const_FIX_0_541196100 psllw mm1, 2 movq [edi][DATASIZE*6+16], mm4 movq mm0, mm6 pmulhw mm1, Const_FIX_1_306562965 psubw mm0, mm5 //z13 paddw mm7, mm2 //z2 movq mm3, mm0 // dataptr[5] = z13 + z2; /* phase 6 */ // dataptr[3] = z13 - z2; // dataptr[1] = z11 + z4; // dataptr[7] = z11 - z4; paddw mm0, mm7 //z13 + z2 psubw mm3, mm7 //z13 - z2 movq mm7, mm3 punpcklwd mm3, mm3 punpckhwd mm7, mm7 paddw mm6, mm5 //z11 psrad mm3, 16 paddw mm1, mm2 //z4 psrad mm7, 16 movq mm4, mm6 movq mm5, mm0 punpcklwd mm0, mm0 punpckhwd mm5, mm5 paddw mm6, mm1 //z11 + z4 psrad mm0, 16 psubw mm4, mm1 //z11 - z4 movq [edi][DATASIZE*3+16], mm3 //store psrad mm5, 16 movq mm1, mm6 punpcklwd mm6, mm6 movq [edi][DATASIZE*3+24], mm7 punpckhwd mm1, mm1 movq [edi][DATASIZE*5+16], mm0 //store psrad mm6, 16 movq [edi][DATASIZE*5+24], mm5 psrad mm1, 16 movq mm7, mm4 punpcklwd mm4, mm4 movq [edi][DATASIZE*1+16], mm6 //store punpckhwd mm7, mm7 movq [edi][DATASIZE*1+24], mm1 psrad mm4, 16 psrad mm7, 16 movq [edi][DATASIZE*7+16], mm4 //store movq [edi][DATASIZE*7+24], mm7 //****************************************************************************** // This completes all 8x8 dct locations for the column case. //****************************************************************************** emms } } #endif /* DCT_ISLOW_SUPPORTED */