1 /******************************************************************************
2  *
3  * Copyright (C) 2015 The Android Open Source Project
4  *
5  * Licensed under the Apache License, Version 2.0 (the "License");
6  * you may not use this file except in compliance with the License.
7  * You may obtain a copy of the License at:
8  *
9  * http://www.apache.org/licenses/LICENSE-2.0
10  *
11  * Unless required by applicable law or agreed to in writing, software
12  * distributed under the License is distributed on an "AS IS" BASIS,
13  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14  * See the License for the specific language governing permissions and
15  * limitations under the License.
16  *
17  *****************************************************************************
18  * Originally developed and contributed by Ittiam Systems Pvt. Ltd, Bangalore
19 */
20 /**
21  *******************************************************************************
22  * @file
23  *  ih264_iquant_itrans_recon_ssse3.c
24  *
25  * @brief
26  *  Contains function definitions for inverse  quantization, inverse
27  * transform and reconstruction
28  *
29  * @author
30  *  Mohit [100664]
31  *
32  * @par List of Functions:
33  *  - ih264_iquant_itrans_recon_4x4_ssse3()
34  *  - ih264_iquant_itrans_recon_8x8_ssse3()
35  *
36  * @remarks
37  *  None
38  *
39  *******************************************************************************
40  */
41 /* User include files */
42 #include "ih264_typedefs.h"
43 #include "ih264_defs.h"
44 #include "ih264_trans_macros.h"
45 #include "ih264_macros.h"
46 #include "ih264_platform_macros.h"
47 #include "ih264_trans_data.h"
48 #include "ih264_size_defs.h"
49 #include "ih264_structs.h"
50 #include "ih264_trans_quant_itrans_iquant.h"
51 #include <immintrin.h>
52 
53 /*
54  ********************************************************************************
55  *
56  * @brief This function reconstructs a 4x4 sub block from quantized resiude and
57  * prediction buffer
58  *
59  * @par Description:
60  *  The quantized residue is first inverse quantized, then inverse transformed.
61  *  This inverse transformed content is added to the prediction buffer to recon-
62  *  struct the end output
63  *
64  * @param[in] pi2_src
65  *  quantized 4x4 block
66  *
67  * @param[in] pu1_pred
68  *  prediction 4x4 block
69  *
70  * @param[out] pu1_out
71  *  reconstructed 4x4 block
72  *
73  * @param[in] src_strd
74  *  quantization buffer stride
75  *
76  * @param[in] pred_strd,
77  *  Prediction buffer stride
78  *
79  * @param[in] out_strd
80  *  recon buffer Stride
81  *
82  * @param[in] pu2_scaling_list
83  *  pointer to scaling list
84  *
85  * @param[in] pu2_norm_adjust
86  *  pointer to inverse scale matrix
87  *
88  * @param[in] u4_qp_div_6
89  *  Floor (qp/6)
90  *
91  * @param[in] pi4_tmp
92  * temporary buffer of size 1*16
93  *
94  * @returns none
95  *
96  * @remarks none
97  *
98  *******************************************************************************
99  */
ih264_iquant_itrans_recon_4x4_ssse3(WORD16 * pi2_src,UWORD8 * pu1_pred,UWORD8 * pu1_out,WORD32 pred_strd,WORD32 out_strd,const UWORD16 * pu2_iscal_mat,const UWORD16 * pu2_weigh_mat,UWORD32 u4_qp_div_6,WORD16 * pi2_tmp,WORD32 iq_start_idx,WORD16 * pi2_dc_ld_addr)100 void ih264_iquant_itrans_recon_4x4_ssse3(WORD16 *pi2_src,
101                                          UWORD8 *pu1_pred,
102                                          UWORD8 *pu1_out,
103                                          WORD32 pred_strd,
104                                          WORD32 out_strd,
105                                          const UWORD16 *pu2_iscal_mat,
106                                          const UWORD16 *pu2_weigh_mat,
107                                          UWORD32 u4_qp_div_6,
108                                          WORD16 *pi2_tmp,
109                                          WORD32 iq_start_idx,
110                                          WORD16 *pi2_dc_ld_addr)
111 {
112     UWORD32 *pu4_out = (UWORD32 *) pu1_out;
113     __m128i src_r0_r1, src_r2_r3;
114     __m128i src_r0, src_r1, src_r2, src_r3;
115     __m128i scalemat_r0_r1, scalemat_r2_r3, predload_r;
116     __m128i pred_r0, pred_r1, pred_r2, pred_r3;
117     __m128i sign_reg, dequant_r0_r1, dequant_r2_r3;
118     __m128i zero_8x16b = _mm_setzero_si128();          // all bits reset to zero
119     __m128i temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
120     __m128i resq_r0, resq_r1, resq_r2, resq_r3;
121     __m128i add_rshift = _mm_set1_epi32((1 << (3 - u4_qp_div_6)));
122     __m128i value_32 = _mm_set1_epi32(32);
123     UNUSED (pi2_tmp);
124     UNUSED (pi2_dc_ld_addr);
125 
126     /*************************************************************/
127     /* Dequantization of coefficients. Will be replaced by SIMD  */
128     /* operations on platform                                    */
129     /*************************************************************/
130     src_r0_r1 = _mm_loadu_si128((__m128i *) (pi2_src)); //a00 a01 a02 a03 a10 a11 a12 a13 -- the source matrix 0th,1st row
131     src_r2_r3 = _mm_loadu_si128((__m128i *) (pi2_src + 8)); //a20 a21 a22 a23 a30 a31 a32 a33 -- the source matrix 2nd,3rd row
132     scalemat_r0_r1 = _mm_loadu_si128((__m128i *) (pu2_iscal_mat)); //b00 b01 b02 b03 b10 b11 b12 b13 -- the scaling matrix 0th,1st row
133     scalemat_r2_r3 = _mm_loadu_si128((__m128i *) (pu2_iscal_mat + 8)); //b20 b21 b22 b23 b30 b31 b32 b33 -- the scaling matrix 2nd,3rd row
134     dequant_r0_r1 = _mm_loadu_si128((__m128i *) (pu2_weigh_mat)); //q00 q01 q02 q03 q10 q11 q12 q13 -- all 16 bits
135     dequant_r2_r3 = _mm_loadu_si128((__m128i *) (pu2_weigh_mat + 8)); //q20 q21 q22 q23 q30 q31 q32 q33 -- all 16 bits
136 
137     temp0 = _mm_mullo_epi16(scalemat_r0_r1, dequant_r0_r1); //b00*q00 b01*q01 b02*q02 b03*q03 b10*q10 b11*q11 b12*q12 b13*q13 -- 16 bit result
138     temp1 = _mm_mullo_epi16(scalemat_r2_r3, dequant_r2_r3); //b00*q00 b01*q01 b02*q02 b03*q03 b10*q10 b11*q11 b12*q12 b13*q13 -- 16 bit result
139 
140     temp4 = _mm_unpacklo_epi16(temp0, zero_8x16b); // b00*q00 0 b01*q01 0 b02*q02 0 b03*q03 0 -- 16 bit long
141     temp5 = _mm_unpackhi_epi16(temp0, zero_8x16b); // b10*q10 0 b11*q11 0 b12*q12 0 b13*q13 0 -- 16 bit long
142     temp6 = _mm_unpacklo_epi16(temp1, zero_8x16b); // b00*q00 0 b01*q01 0 b02*q02 0 b03*q03 0 -- 16 bit long
143     temp7 = _mm_unpackhi_epi16(temp1, zero_8x16b); // b10*q10 0 b11*q11 0 b12*q12 0 b13*q13 0 -- 16 bit long
144 
145     src_r0 = _mm_unpacklo_epi16(src_r0_r1, zero_8x16b); // a00 0 a01 0 a02 0 a03 0 -- 16 bit long
146     src_r1 = _mm_unpackhi_epi16(src_r0_r1, zero_8x16b); // a10 0 a11 0 a12 0 a13 0 -- 16 bit long
147     src_r2 = _mm_unpacklo_epi16(src_r2_r3, zero_8x16b); // a20 0 a21 0 a22 0 a23 0 -- 16 bit long
148     src_r3 = _mm_unpackhi_epi16(src_r2_r3, zero_8x16b); // a30 0 a31 0 a32 0 a33 0 -- 16 bit long
149 
150     temp4 = _mm_madd_epi16(src_r0, temp4); //a00*b00*q00 a10*b10*q10 a20*b20*q20 a30*b30 q30 -- 32 bits long
151     temp5 = _mm_madd_epi16(src_r1, temp5);
152     temp6 = _mm_madd_epi16(src_r2, temp6);
153     temp7 = _mm_madd_epi16(src_r3, temp7);
154 
155     if (u4_qp_div_6 >= 4) {
156         resq_r0 = _mm_slli_epi32(temp4, u4_qp_div_6 - 4);
157         resq_r1 = _mm_slli_epi32(temp5, u4_qp_div_6 - 4);
158         resq_r2 = _mm_slli_epi32(temp6, u4_qp_div_6 - 4);
159         resq_r3 = _mm_slli_epi32(temp7, u4_qp_div_6 - 4);
160     } else {
161         temp4 = _mm_add_epi32(temp4, add_rshift);
162         temp5 = _mm_add_epi32(temp5, add_rshift);
163         temp6 = _mm_add_epi32(temp6, add_rshift);
164         temp7 = _mm_add_epi32(temp7, add_rshift);
165         resq_r0 = _mm_srai_epi32(temp4, 4 - u4_qp_div_6);
166         resq_r1 = _mm_srai_epi32(temp5, 4 - u4_qp_div_6);
167         resq_r2 = _mm_srai_epi32(temp6, 4 - u4_qp_div_6);
168         resq_r3 = _mm_srai_epi32(temp7, 4 - u4_qp_div_6);
169     }
170 
171     if (iq_start_idx == 1)
172     {
173         resq_r0 = _mm_insert_epi16(resq_r0,(WORD32)pi2_src[0],0);
174         if (pi2_src[0] >= 0)
175             resq_r0 = _mm_insert_epi16(resq_r0,0,1);
176         else
177             resq_r0 = _mm_insert_epi16(resq_r0,-1,1);
178     }
179     /* Perform Inverse transform */
180     /*-------------------------------------------------------------*/
181     /* IDCT [ Horizontal transformation ]                          */
182     /*-------------------------------------------------------------*/
183     // Matrix transpose
184     /*
185      *  a0 a1 a2 a3
186      *  b0 b1 b2 b3
187      *  c0 c1 c2 c3
188      *  d0 d1 d2 d3
189      */
190     temp1 = _mm_unpacklo_epi32(resq_r0, resq_r1);                  //a0 b0 a1 b1
191     temp3 = _mm_unpacklo_epi32(resq_r2, resq_r3);                  //c0 d0 c1 d1
192     temp2 = _mm_unpackhi_epi32(resq_r0, resq_r1);                  //a2 b2 a3 b3
193     temp4 = _mm_unpackhi_epi32(resq_r2, resq_r3);                  //c2 d2 c3 d3
194     resq_r0 = _mm_unpacklo_epi64(temp1, temp3);                    //a0 b0 c0 d0
195     resq_r1 = _mm_unpackhi_epi64(temp1, temp3);                    //a1 b1 c1 d1
196     resq_r2 = _mm_unpacklo_epi64(temp2, temp4);                    //a2 b2 c2 d2
197     resq_r3 = _mm_unpackhi_epi64(temp2, temp4);                    //a3 b3 c3 d3
198     //Transform starts -- horizontal transform
199     /*------------------------------------------------------------------*/
200     /* z0 = w0 + w2                                             */
201     temp0 = _mm_add_epi32(resq_r0, resq_r2);
202     /* z1 = w0 - w2                                             */
203     temp1 = _mm_sub_epi32(resq_r0, resq_r2);
204     /* z2 = (w1 >> 1) - w3                                      */
205     temp2 = _mm_srai_epi32(resq_r1, 1);                         //(w1>>1)
206     temp2 = _mm_sub_epi32(temp2, resq_r3);                      //(w1>>1) - w3
207     /* z3 = w1 + (w3 >> 1)                                      */
208     temp3 = _mm_srai_epi32(resq_r3, 1);                         //(w3>>1) + w1
209     temp3 = _mm_add_epi32(temp3, resq_r1);
210     /*----------------------------------------------------------*/
211     /* x0 = z0 + z3                                             */
212     resq_r0 = _mm_add_epi32(temp0, temp3);
213     /* x1 = z1 + z2                                             */
214     resq_r1 = _mm_add_epi32(temp1, temp2);
215     /* x2 = z1 - z2                                             */
216     resq_r2 = _mm_sub_epi32(temp1, temp2);
217     /* x3 = z0 - z3                                             */
218     resq_r3 = _mm_sub_epi32(temp0, temp3);
219     // Matrix transpose
220     /*
221      *  a0 b0 c0 d0
222      *  a1 b1 c1 d1
223      *  a2 b2 c2 d2
224      *  a3 b3 c3 d3
225      */
226     temp1 = _mm_unpacklo_epi32(resq_r0, resq_r1);                  //a0 a1 b0 b1
227     temp3 = _mm_unpacklo_epi32(resq_r2, resq_r3);                  //a2 a3 b2 b3
228     temp2 = _mm_unpackhi_epi32(resq_r0, resq_r1);                  //c0 c1 d0 d1
229     temp4 = _mm_unpackhi_epi32(resq_r2, resq_r3);                  //c2 c3 d2 d3
230     resq_r0 = _mm_unpacklo_epi64(temp1, temp3);                    //a0 a1 a2 a3
231     resq_r1 = _mm_unpackhi_epi64(temp1, temp3);                    //b0 b1 b2 b3
232     resq_r2 = _mm_unpacklo_epi64(temp2, temp4);                    //c0 c1 c2 c3
233     resq_r3 = _mm_unpackhi_epi64(temp2, temp4);                    //d0 d1 d2 d3
234     //Transform ends -- horizontal transform
235 
236     zero_8x16b = _mm_setzero_si128();                  // all bits reset to zero
237     //Load pred buffer
238     predload_r = _mm_loadl_epi64((__m128i *) (&pu1_pred[0])); //p00 p01 p02 p03 0 0 0 0 0 0 0 0 -- all 8 bits
239     pred_r0 = _mm_unpacklo_epi8(predload_r, zero_8x16b); //p00 p01 p02 p03 0 0 0 0 -- all 16 bits
240 
241     predload_r = _mm_loadl_epi64((__m128i *) (&pu1_pred[pred_strd])); //p10 p11 p12 p13 0 0 0 0 0 0 0 0 -- all 8 bits
242     pred_r1 = _mm_unpacklo_epi8(predload_r, zero_8x16b); //p10 p11 p12 p13 0 0 0 0 -- all 16 bits
243 
244     predload_r = _mm_loadl_epi64((__m128i *) (&pu1_pred[2 * pred_strd])); //p20 p21 p22 p23 0 0 0 0 0 0 0 0 -- all 8 bits
245     pred_r2 = _mm_unpacklo_epi8(predload_r, zero_8x16b); //p20 p21 p22 p23 0 0 0 0 -- all 16 bits
246 
247     predload_r = _mm_loadl_epi64((__m128i *) (&pu1_pred[3 * pred_strd])); //p30 p31 p32 p33 0 0 0 0 0 0 0 0 -- all 8 bits
248     pred_r3 = _mm_unpacklo_epi8(predload_r, zero_8x16b); //p30 p31 p32 p33 0 0 0 0 -- all 16 bits
249     pred_r0 = _mm_unpacklo_epi16(pred_r0, zero_8x16b); //p00 p01 p02 p03 -- 32 bits sign extended
250     pred_r1 = _mm_unpacklo_epi16(pred_r1, zero_8x16b); //p10 p11 p12 p13 -- 32 bits sign extended
251     pred_r2 = _mm_unpacklo_epi16(pred_r2, zero_8x16b); //p20 p21 p22 p23 -- 32 bits sign extended
252     pred_r3 = _mm_unpacklo_epi16(pred_r3, zero_8x16b); //p30 p31 p32 p33 -- 32 bits sign extended
253 
254     /*--------------------------------------------------------------*/
255     /* IDCT [ Vertical transformation] and Xij = (xij + 32)>>6      */
256     /*                                                              */
257     /* Add the prediction and store it back to same buffer          */
258     /*--------------------------------------------------------------*/
259     /* z0j = y0j + y2j                                                        */
260     temp0 = _mm_add_epi32(resq_r0, resq_r2);
261     /* z1j = y0j - y2j                                                        */
262     temp1 = _mm_sub_epi32(resq_r0, resq_r2);
263     /* z2j = (y1j>>1) - y3j                                                        */
264     temp2 = _mm_srai_epi32(resq_r1, 1);                             //(y1j>>1)
265     temp2 = _mm_sub_epi32(temp2, resq_r3);
266     /* z3j = y1j + (y3j>>1)                                                        */
267     temp3 = _mm_srai_epi32(resq_r3, 1);                             //(y3j>>1)
268     temp3 = _mm_add_epi32(temp3, resq_r1);
269 
270     /* x0j = z0j + z3j                                                        */
271     temp4 = _mm_add_epi32(temp0, temp3);
272     temp4 = _mm_add_epi32(temp4, value_32);
273     temp4 = _mm_srai_epi32(temp4, 6);
274     temp4 = _mm_add_epi32(temp4, pred_r0);
275     /* x1j = z1j + z2j                                                        */
276     temp5 = _mm_add_epi32(temp1, temp2);
277     temp5 = _mm_add_epi32(temp5, value_32);
278     temp5 = _mm_srai_epi32(temp5, 6);
279     temp5 = _mm_add_epi32(temp5, pred_r1);
280     /* x2j = z1j - z2j                                                        */
281     temp6 = _mm_sub_epi32(temp1, temp2);
282     temp6 = _mm_add_epi32(temp6, value_32);
283     temp6 = _mm_srai_epi32(temp6, 6);
284     temp6 = _mm_add_epi32(temp6, pred_r2);
285     /* x3j = z0j - z3j                                                        */
286     temp7 = _mm_sub_epi32(temp0, temp3);
287     temp7 = _mm_add_epi32(temp7, value_32);
288     temp7 = _mm_srai_epi32(temp7, 6);
289     temp7 = _mm_add_epi32(temp7, pred_r3);
290 
291     // 32-bit to 16-bit conversion
292     temp0 = _mm_packs_epi32(temp4, temp5);
293     temp1 = _mm_packs_epi32(temp6, temp7);
294     /*------------------------------------------------------------------*/
295     //Clipping the results to 8 bits
296     sign_reg = _mm_cmpgt_epi16(temp0, zero_8x16b);      // sign check
297     temp0 = _mm_and_si128(temp0, sign_reg);
298     sign_reg = _mm_cmpgt_epi16(temp1, zero_8x16b);
299     temp1 = _mm_and_si128(temp1, sign_reg);
300 
301     resq_r0 = _mm_packus_epi16(temp0, temp1);
302     resq_r1 = _mm_srli_si128(resq_r0, 4);
303     resq_r2 = _mm_srli_si128(resq_r1, 4);
304     resq_r3 = _mm_srli_si128(resq_r2, 4);
305 
306     *pu4_out = _mm_cvtsi128_si32(resq_r0);
307     pu1_out += out_strd;
308     pu4_out = (UWORD32 *) (pu1_out);
309     *(pu4_out) = _mm_cvtsi128_si32(resq_r1);
310     pu1_out += out_strd;
311     pu4_out = (UWORD32 *) (pu1_out);
312     *(pu4_out) = _mm_cvtsi128_si32(resq_r2);
313     pu1_out += out_strd;
314     pu4_out = (UWORD32 *) (pu1_out);
315     *(pu4_out) = _mm_cvtsi128_si32(resq_r3);
316 }
317 /**
318  *******************************************************************************
319  *
320  * @brief
321  *  This function performs inverse quant and Inverse transform type Ci4 for 8x8 block
322  *
323  * @par Description:
324  *  Performs inverse transform Ci8 and adds the residue to get the
325  *  reconstructed block
326  *
327  * @param[in] pi2_src
328  *  Input 8x8coefficients
329  *
330  * @param[in] pu1_pred
331  *  Prediction 8x8 block
332  *
333  * @param[out] pu1_recon
334  *  Output 8x8 block
335  *
336  * @param[in] q_div
337  *  QP/6
338  *
339  * @param[in] q_rem
340  *  QP%6
341  *
342  * @param[in] q_lev
343  *  Quantizer level
344  *
345  * @param[in] u4_src_stride
346  *  Input stride
347  *
348  * @param[in] u4_pred_stride,
349  *  Prediction stride
350  *
351  * @param[in] u4_out_stride
352  *  Output Stride
353  *
354  * @param[in] pi4_tmp
355  *  temporary buffer of size 1*64
356  *  the tmp for each block
357  *
358  * @param[in] pu4_iquant_mat
359  *  Pointer to the inverse quantization matrix
360  *
361  * @returns  Void
362  *
363  * @remarks
364  *  None
365  *
366  *******************************************************************************
367  */
368 
ih264_iquant_itrans_recon_8x8_ssse3(WORD16 * pi2_src,UWORD8 * pu1_pred,UWORD8 * pu1_out,WORD32 pred_strd,WORD32 out_strd,const UWORD16 * pu2_iscale_mat,const UWORD16 * pu2_weigh_mat,UWORD32 qp_div,WORD16 * pi2_tmp,WORD32 iq_start_idx,WORD16 * pi2_dc_ld_addr)369 void ih264_iquant_itrans_recon_8x8_ssse3(WORD16 *pi2_src,
370                                          UWORD8 *pu1_pred,
371                                          UWORD8 *pu1_out,
372                                          WORD32 pred_strd,
373                                          WORD32 out_strd,
374                                          const UWORD16 *pu2_iscale_mat,
375                                          const UWORD16 *pu2_weigh_mat,
376                                          UWORD32 qp_div,
377                                          WORD16 *pi2_tmp,
378                                          WORD32 iq_start_idx,
379                                          WORD16 *pi2_dc_ld_addr)
380 {
381     __m128i src_r0;
382     __m128i scalemat_r0;
383     __m128i zero_8x16b = _mm_setzero_si128(); // all bits reset to zero
384     // __m128i one_8x16b = _mm_set1_epi8(255); // all bits set to 1
385     // __m128i one_zero_mask = _mm_unpacklo_epi16(one_8x16b, zero_8x16b); // 1 0 1 0 1 0 1 0 --- 16 bits size
386     __m128i value_32 = _mm_set1_epi32(32);
387     __m128i add_rshift = _mm_set1_epi32((1 << (5 - qp_div)));
388     __m128i dequant_r0;
389     __m128i predload_r;
390     __m128i pred_r0_1, pred_r1_1, pred_r2_1, pred_r3_1, pred_r4_1, pred_r5_1,
391             pred_r6_1, pred_r7_1;
392     __m128i sign_reg;
393     __m128i src_r0_1, src_r0_2;
394     __m128i scalemat_r0_1, scalemat_r0_2;
395     __m128i temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8;
396     __m128i temp10, temp11, temp12, temp13, temp14, temp15, temp16, temp17,
397             temp18, temp19, temp20;
398     // To store dequantization results
399     __m128i resq_r0_1, resq_r0_2, resq_r1_1, resq_r1_2, resq_r2_1, resq_r2_2,
400             resq_r3_1, resq_r3_2, resq_r4_1, resq_r4_2, resq_r5_1, resq_r5_2,
401             resq_r6_1, resq_r6_2, resq_r7_1, resq_r7_2;
402     UNUSED (pi2_tmp);
403     UNUSED (iq_start_idx);
404     UNUSED (pi2_dc_ld_addr);
405 
406     /*************************************************************/
407     /* Dequantization of coefficients. Will be replaced by SIMD  */
408     /* operations on platform. Note : DC coeff is not scaled     */
409     /*************************************************************/
410 
411     // Row 0 processing
412     src_r0 = _mm_loadu_si128((__m128i *) (pi2_src)); //a00 a01 a02 a03 a04 a05 a06 a07 -- the source matrix 0th row
413     scalemat_r0 = _mm_loadu_si128((__m128i *) (pu2_iscale_mat)); //b00 b01 b02 b03 b04 b05 b06 b07 -- the scaling matrix 0th row
414     dequant_r0 = _mm_loadu_si128((__m128i *) (&pu2_weigh_mat[0])); //q0 q1 q2 q3 q4 q5 q6 q7 -- all 16 bits
415     src_r0_1 = _mm_unpacklo_epi16(src_r0, zero_8x16b); //a00 0 a01 0 a02 0 a03 0 -- 16 bit long
416     src_r0_2 = _mm_unpackhi_epi16(src_r0, zero_8x16b); // a04 0 a05 0 a06 0 a07 0 -- 16 bit long
417     temp10 = _mm_mullo_epi16(scalemat_r0, dequant_r0); //b00*q0 b01*q1 b02*q2 b03*q3 b04*q4 b05*q5 b06*q6 b07*q7 -- 16 bit result
418     scalemat_r0_1 = _mm_unpacklo_epi16(temp10, zero_8x16b); // b00*q0 0 b01*q1 0 b02*q2 0 b03*q3 0 -- 16 bit long
419     scalemat_r0_2 = _mm_unpackhi_epi16(temp10, zero_8x16b); // b04*q4 0 b05*q5 0 b06*q6 0 b07*q7 0 -- 16 bit long
420 
421     temp5 = _mm_madd_epi16(src_r0_1, scalemat_r0_1); // a00*b00*q0 a01*b01*q1 a02*b02*q2 a03*b03*q3 -- 32 bits long
422     temp7 = _mm_madd_epi16(src_r0_2, scalemat_r0_2); // a04*b04*q4 a05*b05*q5 a06*b06*q6 a07*b07*q7 -- 32 bits long
423 
424     if (qp_div >= 6) {
425         resq_r0_1 = _mm_slli_epi32(temp5, qp_div - 6);
426         resq_r0_2 = _mm_slli_epi32(temp7, qp_div - 6);
427     } else {
428         temp5 = _mm_add_epi32(temp5, add_rshift);
429         temp7 = _mm_add_epi32(temp7, add_rshift);
430         resq_r0_1 = _mm_srai_epi32(temp5, 6 - qp_div);
431         resq_r0_2 = _mm_srai_epi32(temp7, 6 - qp_div);
432     }
433     resq_r0_1 = _mm_packs_epi32(resq_r0_1, resq_r0_2); //a00*b00*q0 a01*b01*q1 a02*b02*q2 a03*b03*q3 a04*b04*q4 a05*b05*q5 a06*b06*q6 a07*b07*q7 -- 16 bit long
434     // Row 1 processing
435     src_r0 = _mm_loadu_si128((__m128i *) (pi2_src + 8)); //a00 a01 a02 a03 a04 a05 a06 a07 a08 -- the source matrix 1st row
436     scalemat_r0 = _mm_loadu_si128((__m128i *) (pu2_iscale_mat + 8)); //b00 b01 b02 b03 b04 b05 b06 b07 b08 -- the scaling matrix 1st row
437     dequant_r0 = _mm_loadu_si128((__m128i *) (&pu2_weigh_mat[8])); //q0 q1 q2 q3 q4 q5 q6 q7 -- all 16 bits
438     src_r0_1 = _mm_unpacklo_epi16(src_r0, zero_8x16b); //a00 0 a01 0 a02 0 a03 0 -- 16 bit long
439     src_r0_2 = _mm_unpackhi_epi16(src_r0, zero_8x16b); // a04 0 a05 0 a06 0 a07 0 -- 16 bit long
440     temp10 = _mm_mullo_epi16(scalemat_r0, dequant_r0); //b00*q0 b01*q1 b02*q2 b03*q3 b04*q4 b05*q5 b06*q6 b07*q7 -- 16 bit result
441     scalemat_r0_1 = _mm_unpacklo_epi16(temp10, zero_8x16b); // b00*q0 0 b01*q1 0 b02*q2 0 b03*q3 0 -- 16 bit long
442     scalemat_r0_2 = _mm_unpackhi_epi16(temp10, zero_8x16b); // b04*q4 0 b05*q5 0 b06*q6 0 b07*q7 0 -- 16 bit long
443     temp5 = _mm_madd_epi16(src_r0_1, scalemat_r0_1); // a00*b00*q0 a01*b01*q1 a02*b02*q2 a03*b03*q3 -- 32 bits long
444     temp7 = _mm_madd_epi16(src_r0_2, scalemat_r0_2); // a04*b04*q4 a05*b05*q5 a06*b06*q6 a07*b07*q7 -- 32 bits long
445     if (qp_div >= 6) {
446         resq_r1_1 = _mm_slli_epi32(temp5, qp_div - 6);
447         resq_r1_2 = _mm_slli_epi32(temp7, qp_div - 6);
448     } else {
449         temp5 = _mm_add_epi32(temp5, add_rshift);
450         temp7 = _mm_add_epi32(temp7, add_rshift);
451         resq_r1_1 = _mm_srai_epi32(temp5, 6 - qp_div);
452         resq_r1_2 = _mm_srai_epi32(temp7, 6 - qp_div);
453     }
454     resq_r1_1 = _mm_packs_epi32(resq_r1_1, resq_r1_2); //a00*b00*q0 a01*b01*q1 a02*b02*q2 a03*b03*q3 a04*b04*q4 a05*b05*q5 a06*b06*q6 a07*b07*q7 -- 16 bit long
455     // Row 2 processing
456     src_r0 = _mm_loadu_si128((__m128i *) (pi2_src + 16)); //a00 a01 a02 a03 a04 a05 a06 a07 a08 -- the source matrix 2nd row
457     scalemat_r0 = _mm_loadu_si128((__m128i *) (pu2_iscale_mat + 16)); //b00 b01 b02 b03 b04 b05 b06 b07 b08 -- the scaling matrix 2nd row
458     dequant_r0 = _mm_loadu_si128((__m128i *) (&pu2_weigh_mat[16])); //q0 q1 q2 q3 q4 q5 q6 q7 -- all 16 bits
459     src_r0_1 = _mm_unpacklo_epi16(src_r0, zero_8x16b); //a00 0 a01 0 a02 0 a03 0 -- 16 bit long
460     src_r0_2 = _mm_unpackhi_epi16(src_r0, zero_8x16b); // a04 0 a05 0 a06 0 a07 0 -- 16 bit long
461     temp10 = _mm_mullo_epi16(scalemat_r0, dequant_r0); //b00*q0 b01*q1 b02*q2 b03*q3 b04*q4 b05*q5 b06*q6 b07*q7 -- 16 bit result
462     scalemat_r0_1 = _mm_unpacklo_epi16(temp10, zero_8x16b); // b00*q0 0 b01*q1 0 b02*q2 0 b03*q3 0 -- 16 bit long
463     scalemat_r0_2 = _mm_unpackhi_epi16(temp10, zero_8x16b); // b04*q4 0 b05*q5 0 b06*q6 0 b07*q7 0 -- 16 bit long
464     temp5 = _mm_madd_epi16(src_r0_1, scalemat_r0_1); // a00*b00*q0 a01*b01*q1 a02*b02*q2 a03*b03*q3 -- 32 bits long
465     temp7 = _mm_madd_epi16(src_r0_2, scalemat_r0_2); // a04*b04*q4 a05*b05*q5 a06*b06*q6 a07*b07*q7 -- 32 bits long
466     if (qp_div >= 6) {
467         resq_r2_1 = _mm_slli_epi32(temp5, qp_div - 6);
468         resq_r2_2 = _mm_slli_epi32(temp7, qp_div - 6);
469     } else {
470         temp5 = _mm_add_epi32(temp5, add_rshift);
471         temp7 = _mm_add_epi32(temp7, add_rshift);
472         resq_r2_1 = _mm_srai_epi32(temp5, 6 - qp_div);
473         resq_r2_2 = _mm_srai_epi32(temp7, 6 - qp_div);
474     }
475     resq_r2_1 = _mm_packs_epi32(resq_r2_1, resq_r2_2); //a00*b00*q0 a01*b01*q1 a02*b02*q2 a03*b03*q3 a04*b04*q4 a05*b05*q5 a06*b06*q6 a07*b07*q7 -- 16 bit long
476     // Row 3 processing
477     src_r0 = _mm_loadu_si128((__m128i *) (pi2_src + 24)); //a00 a01 a02 a03 a04 a05 a06 a07 a08 -- the source matrix 3rd row
478     scalemat_r0 = _mm_loadu_si128((__m128i *) (pu2_iscale_mat + 24)); //b00 b01 b02 b03 b04 b05 b06 b07 b08 -- the scaling matrix 3rd row
479     dequant_r0 = _mm_loadu_si128((__m128i *) (&pu2_weigh_mat[24])); //q0 q1 q2 q3 q4 q5 q6 q7 -- all 16 bits
480     src_r0_1 = _mm_unpacklo_epi16(src_r0, zero_8x16b); //a00 0 a01 0 a02 0 a03 0 -- 16 bit long
481     src_r0_2 = _mm_unpackhi_epi16(src_r0, zero_8x16b); // a04 0 a05 0 a06 0 a07 0 -- 16 bit long
482     temp10 = _mm_mullo_epi16(scalemat_r0, dequant_r0); //b00*q0 b01*q1 b02*q2 b03*q3 b04*q4 b05*q5 b06*q6 b07*q7 -- 16 bit result
483     scalemat_r0_1 = _mm_unpacklo_epi16(temp10, zero_8x16b); // b00*q0 0 b01*q1 0 b02*q2 0 b03*q3 0 -- 16 bit long
484     scalemat_r0_2 = _mm_unpackhi_epi16(temp10, zero_8x16b); // b04*q4 0 b05*q5 0 b06*q6 0 b07*q7 0 -- 16 bit long
485     temp5 = _mm_madd_epi16(src_r0_1, scalemat_r0_1); // a00*b00*q0 a01*b01*q1 a02*b02*q2 a03*b03*q3 - 32 bits long
486     temp7 = _mm_madd_epi16(src_r0_2, scalemat_r0_2); // a04*b04*q4 a05*b05*q5 a06*b06*q6 a07*b07*q7 -- 32 bits long
487     if (qp_div >= 6) {
488         resq_r3_1 = _mm_slli_epi32(temp5, qp_div - 6);
489         resq_r3_2 = _mm_slli_epi32(temp7, qp_div - 6);
490     } else {
491         temp5 = _mm_add_epi32(temp5, add_rshift);
492         temp7 = _mm_add_epi32(temp7, add_rshift);
493         resq_r3_1 = _mm_srai_epi32(temp5, 6 - qp_div);
494         resq_r3_2 = _mm_srai_epi32(temp7, 6 - qp_div);
495     }
496     resq_r3_1 = _mm_packs_epi32(resq_r3_1, resq_r3_2); //a00*b00*q0 a01*b01*q1 a02*b02*q2 a03*b03*q3 a04*b04*q4 a05*b05*q5 a06*b06*q6 a07*b07*q7 -- 16 bit long
497     // Row 4 processing
498     src_r0 = _mm_loadu_si128((__m128i *) (pi2_src + 32)); //a00 a01 a02 a03 a04 a05 a06 a07 a08 -- the source matrix 4th row
499     scalemat_r0 = _mm_loadu_si128((__m128i *) (pu2_iscale_mat + 32)); //b00 b01 b02 b03 b04 b05 b06 b07 b08 -- the scaling matrix 4th row
500     dequant_r0 = _mm_loadu_si128((__m128i *) (&pu2_weigh_mat[32])); //q0 q1 q2 q3 q4 q5 q6 q7 -- all 16 bits
501     src_r0_1 = _mm_unpacklo_epi16(src_r0, zero_8x16b); //a00 0 a01 0 a02 0 a03 0 -- 16 bit long
502     src_r0_2 = _mm_unpackhi_epi16(src_r0, zero_8x16b); // a04 0 a05 0 a06 0 a07 0 -- 16 bit long
503     temp10 = _mm_mullo_epi16(scalemat_r0, dequant_r0); //b00*q0 b01*q1 b02*q2 b03*q3 b04*q4 b05*q5 b06*q6 b07*q7 -- 16 bit result
504     scalemat_r0_1 = _mm_unpacklo_epi16(temp10, zero_8x16b); // b00*q0 0 b01*q1 0 b02*q2 0 b03*q3 0 -- 16 bit long
505     scalemat_r0_2 = _mm_unpackhi_epi16(temp10, zero_8x16b); // b04*q4 0 b05*q5 0 b06*q6 0 b07*q7 0 -- 16 bit long
506     temp5 = _mm_madd_epi16(src_r0_1, scalemat_r0_1); // a00*b00*q0 a01*b01*q1 a02*b02*q2 a03*b03*q3 -- 32 bits long
507     temp7 = _mm_madd_epi16(src_r0_2, scalemat_r0_2); // a04*b04*q4 a05*b05*q5 a06*b06*q6 a07*b07*q7 -- 32 bits long
508     if (qp_div >= 6) {
509         resq_r4_1 = _mm_slli_epi32(temp5, qp_div - 6);
510         resq_r4_2 = _mm_slli_epi32(temp7, qp_div - 6);
511 
512     } else {
513         temp5 = _mm_add_epi32(temp5, add_rshift);
514         temp7 = _mm_add_epi32(temp7, add_rshift);
515         resq_r4_1 = _mm_srai_epi32(temp5, 6 - qp_div);
516         resq_r4_2 = _mm_srai_epi32(temp7, 6 - qp_div);
517     }
518     resq_r4_1 = _mm_packs_epi32(resq_r4_1, resq_r4_2); //a00*b00*q0 a01*b01*q1 a02*b02*q2 a03*b03*q3 a04*b04*q4 a05*b05*q5 a06*b06*q6 a07*b07*q7 -- 16 bit long
519     // Row 5 processing
520     src_r0 = _mm_loadu_si128((__m128i *) (pi2_src + 40)); //a00 a01 a02 a03 a04 a05 a06 a07 a08 -- the source matrix 5th row
521     scalemat_r0 = _mm_loadu_si128((__m128i *) (pu2_iscale_mat + 40)); //b00 b01 b02 b03 b04 b05 b06 b07 b08 -- the scaling matrix 5th row
522     dequant_r0 = _mm_loadu_si128((__m128i *) (&pu2_weigh_mat[40])); //q0 q1 q2 q3 q4 q5 q6 q7 -- all 16 bits
523     src_r0_1 = _mm_unpacklo_epi16(src_r0, zero_8x16b); //a00 0 a01 0 a02 0 a03 0 -- 16 bit long
524     src_r0_2 = _mm_unpackhi_epi16(src_r0, zero_8x16b); // a04 0 a05 0 a06 0 a07 0 -- 16 bit long
525     temp10 = _mm_mullo_epi16(scalemat_r0, dequant_r0); //b00*q0 b01*q1 b02*q2 b03*q3 b04*q4 b05*q5 b06*q6 b07*q7 -- 16 bit result
526     scalemat_r0_1 = _mm_unpacklo_epi16(temp10, zero_8x16b); // b00*q0 0 b01*q1 0 b02*q2 0 b03*q3 0 -- 16 bit long
527     scalemat_r0_2 = _mm_unpackhi_epi16(temp10, zero_8x16b); // b04*q4 0 b05*q5 0 b06*q6 0 b07*q7 0 -- 16 bit long
528     temp5 = _mm_madd_epi16(src_r0_1, scalemat_r0_1); // a00*b00*q0 a01*b01*q1 a02*b02*q2 a03*b03*q3 -- 32 bits long
529     temp7 = _mm_madd_epi16(src_r0_2, scalemat_r0_2); // a04*b04*q4 a05*b05*q5 a06*b06*q6 a07*b07*q7 -- 32 bits long
530     if (qp_div >= 6) {
531         resq_r5_1 = _mm_slli_epi32(temp5, qp_div - 6);
532         resq_r5_2 = _mm_slli_epi32(temp7, qp_div - 6);
533         //resq_r5_1 = _mm_and_si128(resq_r5_1,one_zero_mask);
534         //resq_r5_2 = _mm_and_si128(resq_r5_2,one_zero_mask);
535     } else {
536         temp5 = _mm_add_epi32(temp5, add_rshift);
537         temp7 = _mm_add_epi32(temp7, add_rshift);
538         resq_r5_1 = _mm_srai_epi32(temp5, 6 - qp_div);
539         resq_r5_2 = _mm_srai_epi32(temp7, 6 - qp_div);
540     }
541     resq_r5_1 = _mm_packs_epi32(resq_r5_1, resq_r5_2); //a00*b00*q0 a01*b01*q1 a02*b02*q2 a03*b03*q3 a04*b04*q4 a05*b05*q5 a06*b06*q6 a07*b07*q7 -- 16 bit long
542     // Row 6 processing
543     src_r0 = _mm_loadu_si128((__m128i *) (pi2_src + 48)); //a00 a01 a02 a03 a04 a05 a06 a07 a08 -- the source matrix 6th row
544     scalemat_r0 = _mm_loadu_si128((__m128i *) (pu2_iscale_mat + 48)); //b00 b01 b02 b03 b04 b05 b06 b07 b08 -- the scaling matrix 6th row
545     dequant_r0 = _mm_loadu_si128((__m128i *) (&pu2_weigh_mat[48])); //q0 q1 q2 q3 q4 q5 q6 q7 -- all 16 bits
546     src_r0_1 = _mm_unpacklo_epi16(src_r0, zero_8x16b); //a00 0 a01 0 a02 0 a03 0 -- 16 bit long
547     src_r0_2 = _mm_unpackhi_epi16(src_r0, zero_8x16b); // a04 0 a05 0 a06 0 a07 0 -- 16 bit long
548     temp10 = _mm_mullo_epi16(scalemat_r0, dequant_r0); //b00*q0 b01*q1 b02*q2 b03*q3 b04*q4 b05*q5 b06*q6 b07*q7 -- 16 bit result
549     scalemat_r0_1 = _mm_unpacklo_epi16(temp10, zero_8x16b); // b00*q0 0 b01*q1 0 b02*q2 0 b03*q3 0 -- 16 bit long
550     scalemat_r0_2 = _mm_unpackhi_epi16(temp10, zero_8x16b); // b04*q4 0 b05*q5 0 b06*q6 0 b07*q7 0 -- 16 bit long
551     temp5 = _mm_madd_epi16(src_r0_1, scalemat_r0_1); // a00*b00*q0 a01*b01*q1 a02*b02*q2 a03*b03*q3 -- 32 bits long
552     temp7 = _mm_madd_epi16(src_r0_2, scalemat_r0_2); // a04*b04*q4 a05*b05*q5 a06*b06*q6 a07*b07*q7 -- 32 bits long
553     if (qp_div >= 6) {
554         resq_r6_1 = _mm_slli_epi32(temp5, qp_div - 6);
555         resq_r6_2 = _mm_slli_epi32(temp7, qp_div - 6);
556         //resq_r6_1 = _mm_and_si128(resq_r6_1,one_zero_mask);
557         //resq_r6_2 = _mm_and_si128(resq_r6_2,one_zero_mask);
558     } else {
559         temp5 = _mm_add_epi32(temp5, add_rshift);
560         temp7 = _mm_add_epi32(temp7, add_rshift);
561         resq_r6_1 = _mm_srai_epi32(temp5, 6 - qp_div);
562         resq_r6_2 = _mm_srai_epi32(temp7, 6 - qp_div);
563         //resq_r6_1 = _mm_and_si128(resq_r6_1,one_zero_mask);
564         //resq_r6_2 = _mm_and_si128(resq_r6_2,one_zero_mask);
565     }
566     resq_r6_1 = _mm_packs_epi32(resq_r6_1, resq_r6_2); //a00*b00*q0 a01*b01*q1 a02*b02*q2 a03*b03*q3 a04*b04*q4 a05*b05*q5 a06*b06*q6 a07*b07*q7 -- 16 bit long
567     // Row 7 processing
568     src_r0 = _mm_loadu_si128((__m128i *) (pi2_src + 56)); //a00 a01 a02 a03 a04 a05 a06 a07 a08 -- the source matrix 7th row
569     scalemat_r0 = _mm_loadu_si128((__m128i *) (pu2_iscale_mat + 56)); //b00 b01 b02 b03 b04 b05 b06 b07 b08 -- the scaling matrix 7th row
570     dequant_r0 = _mm_loadu_si128((__m128i *) (&pu2_weigh_mat[56])); //q0 q1 q2 q3 q4 q5 q6 q7 -- all 16 bits
571     src_r0_1 = _mm_unpacklo_epi16(src_r0, zero_8x16b); //a00 0 a01 0 a02 0 a03 0 -- 16 bit long
572     src_r0_2 = _mm_unpackhi_epi16(src_r0, zero_8x16b); // a04 0 a05 0 a06 0 a07 0 -- 16 bit long
573     temp10 = _mm_mullo_epi16(scalemat_r0, dequant_r0); //b00*q0 b01*q1 b02*q2 b03*q3 b04*q4 b05*q5 b06*q6 b07*q7 -- 16 bit result
574     scalemat_r0_1 = _mm_unpacklo_epi16(temp10, zero_8x16b); // b00*q0 0 b01*q1 0 b02*q2 0 b03*q3 0 -- 16 bit long
575     scalemat_r0_2 = _mm_unpackhi_epi16(temp10, zero_8x16b); // b04*q4 0 b05*q5 0 b06*q6 0 b07*q7 0 -- 16 bit long
576     temp5 = _mm_madd_epi16(src_r0_1, scalemat_r0_1); // a00*b00*q0 a01*b01*q1 a02*b02*q2 a03*b03*q3 -- 32 bits long
577     temp7 = _mm_madd_epi16(src_r0_2, scalemat_r0_2); // a04*b04*q4 a05*b05*q5 a06*b06*q6 a07*b07*q7 -- 32 bits long
578     if (qp_div >= 6) {
579         resq_r7_1 = _mm_slli_epi32(temp5, qp_div - 6);
580         resq_r7_2 = _mm_slli_epi32(temp7, qp_div - 6);
581     } else {
582         temp5 = _mm_add_epi32(temp5, add_rshift);
583         temp7 = _mm_add_epi32(temp7, add_rshift);
584         resq_r7_1 = _mm_srai_epi32(temp5, 6 - qp_div);
585         resq_r7_2 = _mm_srai_epi32(temp7, 6 - qp_div);
586     }
587     resq_r7_1 = _mm_packs_epi32(resq_r7_1, resq_r7_2); //a00*b00*q0 a01*b01*q1 a02*b02*q2 a03*b03*q3 a04*b04*q4 a05*b05*q5 a06*b06*q6 a07*b07*q7 -- 16 bit long
588     /* Perform Inverse transform */
589     /*--------------------------------------------------------------------*/
590     /* IDCT [ Horizontal transformation ]                                 */
591     /*--------------------------------------------------------------------*/
592     // Matrix transpose
593     /*
594      *  a0 a1 a2 a3 a4 a5 a6 a7
595      *  b0 b1 b2 b3 b4 b5 b6 b7
596      *  c0 c1 c2 c3 c4 c5 c6 c7
597      *  d0 d1 d2 d3 d4 d5 d6 d7
598      */
599     temp1 = _mm_unpacklo_epi16(resq_r0_1, resq_r1_1); //a0 b0 a1 b1 a2 b2 a3 b3
600     temp3 = _mm_unpacklo_epi16(resq_r2_1, resq_r3_1); //c0 d0 c1 d1 c2 d2 c3 d3
601     temp2 = _mm_unpackhi_epi16(resq_r0_1, resq_r1_1); //a4 b4 a5 b5 a6 b6 a7 b7
602     temp4 = _mm_unpackhi_epi16(resq_r2_1, resq_r3_1); //c4 d4 c5 d5 c6 d6 c7 d7
603     resq_r0_1 = _mm_unpacklo_epi32(temp1, temp3); //a0 b0 c0 d0 a1 b1 c1 d1
604     resq_r1_1 = _mm_unpackhi_epi32(temp1, temp3); //a2 b2 c2 d2 a3 b3 c3 d3
605     resq_r2_1 = _mm_unpacklo_epi32(temp2, temp4); //a4 b4 c4 d4 a5 b5 c5 d5
606     resq_r3_1 = _mm_unpackhi_epi32(temp2, temp4); //a6 b6 c6 d6 a7 b7 c7 d7
607     /*
608      * e0 e1 e2 e3 e4 e5 e6 e7
609      * f0 f1 f2 f3 f4 f5 f6 f7
610      * g0 g1 g2 g3 g4 g5 g6 g7
611      * h0 h1 h2 h3 h4 h5 h6 h7
612      */
613     temp1 = _mm_unpacklo_epi16(resq_r4_1, resq_r5_1); //e0 f0 e1 f1 e2 f2 e2 f3
614     temp3 = _mm_unpacklo_epi16(resq_r6_1, resq_r7_1); //g0 h0 g1 h1 g2 h2 g3 h3
615     temp2 = _mm_unpackhi_epi16(resq_r4_1, resq_r5_1); //e4 f4 e5 f5 e6 f6 e7 f7
616     temp4 = _mm_unpackhi_epi16(resq_r6_1, resq_r7_1); //g4 h4 g5 h5 g6 h6 g7 h7
617     resq_r4_1 = _mm_unpacklo_epi32(temp1, temp3); //e0 f0 g0 h0 e1 f1 g1 h1
618     resq_r5_1 = _mm_unpackhi_epi32(temp1, temp3); //e2 f2 g2 h2 e3 f3 g3 h3
619     resq_r6_1 = _mm_unpacklo_epi32(temp2, temp4); //e4 f4 g4 h4 e5 f5 g5 h5
620     resq_r7_1 = _mm_unpackhi_epi32(temp2, temp4); //e6 f6 g6 h6 e7 f7 g7 h7
621     /*
622      * a0 b0 c0 d0 a1 b1 c1 d1
623      * a2 b2 c2 d2 a3 b3 c3 d3
624      * a4 b4 c4 d4 a5 b5 c5 d5
625      * a6 b6 c6 d6 a7 b7 c7 d7
626      * e0 f0 g0 h0 e1 f1 g1 h1
627      * e2 f2 g2 h2 e3 f3 g3 h3
628      * e4 f4 g4 h4 e5 f5 g5 h5
629      * e6 f6 g6 h6 e7 f7 g7 h7
630      */
631     resq_r0_2 = _mm_unpacklo_epi64(resq_r0_1, resq_r4_1); //a0 b0 c0 d0 e0 f0 g0 h0
632     resq_r1_2 = _mm_unpackhi_epi64(resq_r0_1, resq_r4_1); //a1 b1 c1 d1 e1 f1 g1 h1
633     resq_r2_2 = _mm_unpacklo_epi64(resq_r1_1, resq_r5_1); //a2 b2 c2 d2 e2 f2 g2 h2
634     resq_r3_2 = _mm_unpackhi_epi64(resq_r1_1, resq_r5_1); //a3 b3 c3 d3 e3 f3 g3 h3
635     resq_r4_2 = _mm_unpacklo_epi64(resq_r2_1, resq_r6_1); //a4 b4 c4 d4 e4 f4 g4 h4
636     resq_r5_2 = _mm_unpackhi_epi64(resq_r2_1, resq_r6_1); //a5 b5 c5 d5 e5 f5 g5 h5
637     resq_r6_2 = _mm_unpacklo_epi64(resq_r3_1, resq_r7_1); //a6 b6 c6 d6 e6 f6 g6 h6
638     resq_r7_2 = _mm_unpackhi_epi64(resq_r3_1, resq_r7_1); //a7 b7 c7 d7 e7 f7 g7 h7
639 
640     sign_reg = _mm_cmpgt_epi16(zero_8x16b, resq_r1_2);
641     resq_r1_1 = _mm_unpacklo_epi16(resq_r1_2, sign_reg); //a1 b1 c1 d1 -- 32 bit
642     resq_r1_2 = _mm_unpackhi_epi16(resq_r1_2, sign_reg); //e1 f1 g1 h1 -- 32 bit
643     sign_reg = _mm_cmpgt_epi16(zero_8x16b, resq_r3_2);
644     resq_r3_1 = _mm_unpacklo_epi16(resq_r3_2, sign_reg); //a3 b3 c3 d3 -- 32 bit
645     resq_r3_2 = _mm_unpackhi_epi16(resq_r3_2, sign_reg); //e3 f3 g3 h3 -- 32 bit
646     sign_reg = _mm_cmpgt_epi16(zero_8x16b, resq_r5_2);
647     resq_r5_1 = _mm_unpacklo_epi16(resq_r5_2, sign_reg); //a5 b5 c5 d5 -- 32 bit
648     resq_r5_2 = _mm_unpackhi_epi16(resq_r5_2, sign_reg); //e5 f5 g5 h5 -- 32 bit
649     sign_reg = _mm_cmpgt_epi16(zero_8x16b, resq_r7_2);
650     resq_r7_1 = _mm_unpacklo_epi16(resq_r7_2, sign_reg); //a7 b7 c7 d7 -- 32 bit
651     resq_r7_2 = _mm_unpackhi_epi16(resq_r7_2, sign_reg); //e7 f7 g7 h7 -- 32 bit
652     //Transform starts -- horizontal transform
653     /*------------------------------------------------------------------*/
654     /* y0 = w0 + w4                                                     */
655     temp1 = _mm_add_epi16(resq_r0_2, resq_r4_2);
656     /* y2 = w0 - w4                                                      */
657     temp3 = _mm_sub_epi16(resq_r0_2, resq_r4_2);
658     /* y1 = -w3 + w5 - w7 - (w7 >> 1)                                   */
659     temp2 = _mm_sub_epi32(resq_r5_1, resq_r3_1); //-w3+w5
660     temp10 = _mm_sub_epi32(resq_r5_2, resq_r3_2);
661     temp4 = _mm_sub_epi32(temp2, resq_r7_1); //-w3+w5-w7
662     temp12 = _mm_sub_epi32(temp10, resq_r7_2);
663     temp5 = _mm_srai_epi32(resq_r7_1, 1); //w7>>1
664     temp13 = _mm_srai_epi32(resq_r7_2, 1);
665     temp2 = _mm_sub_epi32(temp4, temp5); //-w3+w5-w7 -(w7>>1)
666     temp10 = _mm_sub_epi32(temp12, temp13);
667     temp2 = _mm_packs_epi32(temp2, temp10);
668     /* y3 = w1 + w7 - w3 - (w3 >> 1)                                    */
669     temp4 = _mm_add_epi32(resq_r1_1, resq_r7_1); //w1+w7
670     temp12 = _mm_add_epi32(resq_r1_2, resq_r7_2);
671     temp4 = _mm_sub_epi32(temp4, resq_r3_1); //w1+w7-w3
672     temp12 = _mm_sub_epi32(temp12, resq_r3_2);
673     temp5 = _mm_srai_epi32(resq_r3_1, 1); //w3>>1
674     temp13 = _mm_srai_epi32(resq_r3_2, 1);
675     temp4 = _mm_sub_epi32(temp4, temp5); //w1+w7-w3-(w3>>1)
676     temp12 = _mm_sub_epi32(temp12, temp13);
677     temp4 = _mm_packs_epi32(temp4, temp12);
678     /* y4 = (w2 >> 1) - w6                                              */
679     temp5 = _mm_srai_epi16(resq_r2_2, 1); //w2>>1
680     temp5 = _mm_sub_epi16(temp5, resq_r6_2); //(w2>>1)-w6
681     /* y5 = -w1 + w7 + w5 + (w5 >> 1)                                   */
682     temp6 = _mm_sub_epi32(resq_r7_1, resq_r1_1); //w7-w1
683     temp14 = _mm_sub_epi32(resq_r7_2, resq_r1_2);
684     temp6 = _mm_add_epi32(temp6, resq_r5_1); //w7-w1+w5
685     temp14 = _mm_add_epi32(temp14, resq_r5_2);
686     temp7 = _mm_srai_epi32(resq_r5_1, 1); //w5>>1
687     temp15 = _mm_srai_epi32(resq_r5_2, 1);
688     temp6 = _mm_add_epi32(temp6, temp7); //w7-w1_w5+(w5>>1)
689     temp14 = _mm_add_epi32(temp14, temp15);
690     temp6 = _mm_packs_epi32(temp6, temp14);
691     /* y6 = w2 + (w6 >> 1)                                              */
692     temp7 = _mm_srai_epi16(resq_r6_2, 1); //w6>>1
693     temp7 = _mm_add_epi16(temp7, resq_r2_2); //(w6>>1)+w2
694     /* y7 = w3 + w5 + w1 + (w1 >> 1)                                    */
695     temp8 = _mm_add_epi32(resq_r3_1, resq_r5_1); //w3+w5
696     temp16 = _mm_add_epi32(resq_r3_2, resq_r5_2);
697     temp8 = _mm_add_epi32(temp8, resq_r1_1); //w3+w5+w1
698     temp16 = _mm_add_epi32(temp16, resq_r1_2);
699     temp17 = _mm_srai_epi32(resq_r1_1, 1); //w1>>1
700     temp18 = _mm_srai_epi32(resq_r1_2, 1);
701     temp8 = _mm_add_epi32(temp8, temp17); //w3+w5+w1+(w1>>1)
702     temp16 = _mm_add_epi32(temp16, temp18);
703     temp8 = _mm_packs_epi32(temp8, temp16);
704     /*------------------------------------------------------------------*/
705     /*------------------------------------------------------------------*/
706     /* z0 = y0 + y6                                                        */
707     resq_r0_1 = _mm_add_epi16(temp1, temp7);
708     /* z1 = y1 + (y7 >> 2)                                                */
709     resq_r1_1 = _mm_srai_epi16(temp8, 2);
710     resq_r1_1 = _mm_add_epi16(resq_r1_1, temp2);
711     /* z2 = y2 + y4                                                        */
712     resq_r2_1 = _mm_add_epi16(temp3, temp5);
713     /* z3 = y3 + (y5 >> 2)                                                */
714     resq_r3_1 = _mm_srai_epi16(temp6, 2);
715     resq_r3_1 = _mm_add_epi16(resq_r3_1, temp4);
716     /* z4 = y2 - y4                                                        */
717     resq_r4_1 = _mm_sub_epi16(temp3, temp5);
718     /* z5 = (y3 >> 2) - y5                                                 */
719     resq_r5_1 = _mm_srai_epi16(temp4, 2);
720     resq_r5_1 = _mm_sub_epi16(resq_r5_1, temp6);
721     /* z6 = y0 - y6                                                     */
722     resq_r6_1 = _mm_sub_epi16(temp1, temp7);
723     /* z7 = y7 - (y1 >> 2)                                                 */
724     resq_r7_1 = _mm_srai_epi16(temp2, 2);
725     resq_r7_1 = _mm_sub_epi16(temp8, resq_r7_1);
726     /*------------------------------------------------------------------*/
727     /*------------------------------------------------------------------*/
728     /* x0 = z0 + z7                                                        */
729     temp1 = _mm_add_epi16(resq_r0_1, resq_r7_1);
730     /* x1 = z2 + z5                                                        */
731     temp2 = _mm_add_epi16(resq_r2_1, resq_r5_1);
732     /* x2 = z4 + z3                                                        */
733     temp3 = _mm_add_epi16(resq_r4_1, resq_r3_1);
734     /* x3 = z6 + z1                                                        */
735     temp4 = _mm_add_epi16(resq_r6_1, resq_r1_1);
736     /* x4 = z6 - z1                                                        */
737     temp5 = _mm_sub_epi16(resq_r6_1, resq_r1_1);
738     /* x5 = z4 - z3                                                        */
739     temp6 = _mm_sub_epi16(resq_r4_1, resq_r3_1);
740     /* x6 = z2 - z5                                                        */
741     temp7 = _mm_sub_epi16(resq_r2_1, resq_r5_1);
742     /* x7 = z0 - z7                                                        */
743     temp8 = _mm_sub_epi16(resq_r0_1, resq_r7_1);
744     /*------------------------------------------------------------------*/
745     // Matrix transpose
746     /*
747      *  a0 b0 c0 d0 e0 f0 g0 h0
748      *  a1 b1 c1 d1 e1 f1 g1 h1
749      *  a2 b2 c2 d2 e2 f2 g2 h2
750      *  a3 b3 c3 d3 e3 f3 g3 h3
751      */
752     temp17 = _mm_unpacklo_epi16(temp1, temp2); //a0 a1 b0 b1 c0 c1 d0 d1
753     temp19 = _mm_unpacklo_epi16(temp3, temp4); //a2 a3 b2 b3 c2 c3 d2 d3
754     temp18 = _mm_unpackhi_epi16(temp1, temp2); //e0 e1 f0 f1 g0 g1 h0 h1
755     temp20 = _mm_unpackhi_epi16(temp3, temp4); //e2 e3 f2 f3 g2 g3 h2 h3
756 
757     resq_r0_1 = _mm_unpacklo_epi32(temp17, temp19); //a0 a1 a2 a3 b0 b1 b2 b3
758     resq_r1_1 = _mm_unpackhi_epi32(temp17, temp19); //c0 c1 c2 c3 d0 d1 d2 d3
759     resq_r2_1 = _mm_unpacklo_epi32(temp18, temp20); //e0 e1 e2 e3 f0 f1 f2 f3
760     resq_r3_1 = _mm_unpackhi_epi32(temp18, temp20); //g0 g2 g2 g3 h0 h1 h2 h3
761     /*
762      *  a4 b4 c4 d4 e4 f4 g4 h4
763      *  a5 b5 c5 d5 e5 f5 g5 h5
764      *  a6 b6 c6 d6 e6 f6 g6 h6
765      *  a7 b7 c7 d7 e7 f7 g7 h7
766      */
767     temp17 = _mm_unpacklo_epi16(temp5, temp6); //a4 a5 b4 b5 c4 c5 d4 d5
768     temp19 = _mm_unpacklo_epi16(temp7, temp8); //a6 a7 b6 b7 c6 c7 d6 d7
769     temp18 = _mm_unpackhi_epi16(temp5, temp6); //e4 e5 f4 f5 g4 g5 h4 h5
770     temp20 = _mm_unpackhi_epi16(temp7, temp8); //e6 e7 f6 f7 g6 g7 h6 h7
771 
772     resq_r4_1 = _mm_unpacklo_epi32(temp17, temp19); //a4 a5 a6 a7 b4 b5 b6 b7
773     resq_r5_1 = _mm_unpackhi_epi32(temp17, temp19); //c4 c5 c6 c7 d4 d5 d6 d7
774     resq_r6_1 = _mm_unpacklo_epi32(temp18, temp20); //e4 e5 e6 e7 f4 f5 f6 f7
775     resq_r7_1 = _mm_unpackhi_epi32(temp18, temp20); //g4 g5 g6 g7 h4 h5 h6 h7
776     /*  a0 a1 a2 a3 b0 b1 b2 b3
777      *  c0 c1 c2 c3 d0 d1 d2 d3
778      *  e0 e1 e2 e3 f0 f1 f2 f3
779      *  g0 g2 g2 g3 h0 h1 h2 h3
780      *  a4 a5 a6 a7 b4 b5 b6 b7
781      *  c4 c5 c6 c7 d4 d5 d6 d7
782      *  e4 e5 e6 e7 f4 f5 f6 f7
783      *  g4 g5 g6 g7 h4 h5 h6 h7
784      */
785     resq_r0_2 = _mm_unpacklo_epi64(resq_r0_1, resq_r4_1); //a0 a1 a2 a3 a4 a5 a6 a7
786     resq_r1_2 = _mm_unpackhi_epi64(resq_r0_1, resq_r4_1); //b0 b1 b2 b3 b4 b5 b6 b7
787     resq_r2_2 = _mm_unpacklo_epi64(resq_r1_1, resq_r5_1); //c0 c1 c2 c3 c4 c5 c6 c7
788     resq_r3_2 = _mm_unpackhi_epi64(resq_r1_1, resq_r5_1); //d0 d1 d2 d3 d4 d5 d6 d7
789     resq_r4_2 = _mm_unpacklo_epi64(resq_r2_1, resq_r6_1); //e0 e1 e2 e3 e4 e5 e6 e7
790     resq_r5_2 = _mm_unpackhi_epi64(resq_r2_1, resq_r6_1); //f0 f1 f2 f3 f4 f5 f6 f7
791     resq_r6_2 = _mm_unpacklo_epi64(resq_r3_1, resq_r7_1); //g0 g1 g2 g3 g4 g5 g6 g7
792     resq_r7_2 = _mm_unpackhi_epi64(resq_r3_1, resq_r7_1); //h0 h1 h2 h3 h4 h5 h6 h7
793 
794     sign_reg = _mm_cmpgt_epi16(zero_8x16b, resq_r1_2);
795     resq_r1_1 = _mm_unpacklo_epi16(resq_r1_2, sign_reg); //a1 b1 c1 d1 -- 32 bit
796     resq_r1_2 = _mm_unpackhi_epi16(resq_r1_2, sign_reg); //e1 f1 g1 h1 -- 32 bit
797     sign_reg = _mm_cmpgt_epi16(zero_8x16b, resq_r3_2);
798     resq_r3_1 = _mm_unpacklo_epi16(resq_r3_2, sign_reg); //a3 b3 c3 d3 -- 32 bit
799     resq_r3_2 = _mm_unpackhi_epi16(resq_r3_2, sign_reg); //e3 f3 g3 h3 -- 32 bit
800     sign_reg = _mm_cmpgt_epi16(zero_8x16b, resq_r5_2);
801     resq_r5_1 = _mm_unpacklo_epi16(resq_r5_2, sign_reg); //a5 b5 c5 d5 -- 32 bit
802     resq_r5_2 = _mm_unpackhi_epi16(resq_r5_2, sign_reg); //e5 f5 g5 h5 -- 32 bit
803     sign_reg = _mm_cmpgt_epi16(zero_8x16b, resq_r7_2);
804     resq_r7_1 = _mm_unpacklo_epi16(resq_r7_2, sign_reg); //a7 b7 c7 d7 -- 32 bit
805     resq_r7_2 = _mm_unpackhi_epi16(resq_r7_2, sign_reg); //e7 f7 g7 h7 -- 32 bit
806 
807     zero_8x16b = _mm_setzero_si128(); // all bits reset to zero
808     //Load pred buffer row 0
809     predload_r = _mm_loadl_epi64((__m128i *) (&pu1_pred[0])); //p0 p1 p2 p3 p4 p5 p6 p7 0 0 0 0 0 0 0 0 -- all 8 bits
810     pred_r0_1 = _mm_unpacklo_epi8(predload_r, zero_8x16b); //p0 p1 p2 p3 p4 p5 p6 p7 -- all 16 bits
811     //Load pred buffer row 1
812     predload_r = _mm_loadl_epi64((__m128i *) (&pu1_pred[pred_strd])); //p0 p1 p2 p3 p4 p5 p6 p7 0 0 0 0 0 0 0 0 -- all 8 bits
813     pred_r1_1 = _mm_unpacklo_epi8(predload_r, zero_8x16b); //p0 p1 p2 p3 p4 p5 p6 p7 -- all 16 bits
814     //Load pred buffer row 2
815     predload_r = _mm_loadl_epi64((__m128i *) (&pu1_pred[2 * pred_strd])); //p0 p1 p2 p3 p4 p5 p6 p7 0 0 0 0 0 0 0 0 -- all 8 bits
816     pred_r2_1 = _mm_unpacklo_epi8(predload_r, zero_8x16b); //p0 p1 p2 p3 p4 p5 p6 p7 -- all 16 bits
817     //Load pred buffer row 3
818     predload_r = _mm_loadl_epi64((__m128i *) (&pu1_pred[3 * pred_strd])); //p0 p1 p2 p3 p4 p5 p6 p7 0 0 0 0 0 0 0 0 -- all 8 bits
819     pred_r3_1 = _mm_unpacklo_epi8(predload_r, zero_8x16b); //p0 p1 p2 p3 p4 p5 p6 p7 -- all 16 bits
820     //Load pred buffer row 4
821     predload_r = _mm_loadl_epi64((__m128i *) (&pu1_pred[4 * pred_strd])); //p0 p1 p2 p3 p4 p5 p6 p7 0 0 0 0 0 0 0 0 -- all 8 bits
822     pred_r4_1 = _mm_unpacklo_epi8(predload_r, zero_8x16b); //p0 p1 p2 p3 p4 p5 p6 p7 -- all 16 bits
823     //Load pred buffer row 5
824     predload_r = _mm_loadl_epi64((__m128i *) (&pu1_pred[5 * pred_strd])); //p0 p1 p2 p3 p4 p5 p6 p7 0 0 0 0 0 0 0 0 -- all 8 bit
825     pred_r5_1 = _mm_unpacklo_epi8(predload_r, zero_8x16b); //p0 p1 p2 p3 p4 p5 p6 p7 -- all 16 bits
826     //Load pred buffer row 6
827     predload_r = _mm_loadl_epi64((__m128i *) (&pu1_pred[6 * pred_strd])); //p0 p1 p2 p3 p4 p5 p6 p7 0 0 0 0 0 0 0 0 -- all 8 bits
828     pred_r6_1 = _mm_unpacklo_epi8(predload_r, zero_8x16b); //p0 p1 p2 p3 p4 p5 p6 p7 -- all 16 bits
829     //Load pred buffer row 7
830     predload_r = _mm_loadl_epi64((__m128i *) (&pu1_pred[7 * pred_strd])); //p0 p1 p2 p3 p4 p5 p6 p7 0 0 0 0 0 0 0 0 -- all 8 bits
831     pred_r7_1 = _mm_unpacklo_epi8(predload_r, zero_8x16b); //p0 p1 p2 p3 p4 p5 p6 p7 -- all 16 bits
832 
833     /*--------------------------------------------------------------------*/
834     /* IDCT [ Vertical transformation] and Xij = (xij + 32)>>6            */
835     /*                                                                    */
836     /* Add the prediction and store it back to reconstructed frame buffer */
837     /* [Prediction buffer itself in this case]                            */
838     /*--------------------------------------------------------------------*/
839 
840     /* y0j = w0j + w4j                                                     */
841     temp1 = _mm_add_epi16(resq_r0_2, resq_r4_2);
842     /* y2j = w0j - w4j                                                      */
843     temp3 = _mm_sub_epi16(resq_r0_2, resq_r4_2);
844     /* y1j = -w3j + w5j - w7j - (w7j >> 1)                                   */
845     temp2 = _mm_sub_epi32(resq_r5_1, resq_r3_1); //-w3+w5
846     temp10 = _mm_sub_epi32(resq_r5_2, resq_r3_2);
847     temp4 = _mm_sub_epi32(temp2, resq_r7_1); //-w3+w5-w7
848     temp12 = _mm_sub_epi32(temp10, resq_r7_2);
849     temp5 = _mm_srai_epi32(resq_r7_1, 1); //w7>>1
850     temp13 = _mm_srai_epi32(resq_r7_2, 1);
851     temp2 = _mm_sub_epi32(temp4, temp5); //-w3+w5-w7 -(w7>>1)
852     temp10 = _mm_sub_epi32(temp12, temp13);
853     temp2 = _mm_packs_epi32(temp2, temp10);
854     /* y3j = w1j + w7j - w3j - (w3j >> 1)                                    */
855     temp4 = _mm_add_epi32(resq_r1_1, resq_r7_1); //w1+w7
856     temp12 = _mm_add_epi32(resq_r1_2, resq_r7_2);
857     temp4 = _mm_sub_epi32(temp4, resq_r3_1); //w1+w7-w3
858     temp12 = _mm_sub_epi32(temp12, resq_r3_2);
859     temp5 = _mm_srai_epi32(resq_r3_1, 1); //w3>>1
860     temp13 = _mm_srai_epi32(resq_r3_2, 1);
861     temp4 = _mm_sub_epi32(temp4, temp5); //w1+w7-w3-(w3>>1)
862     temp12 = _mm_sub_epi32(temp12, temp13);
863     temp4 = _mm_packs_epi32(temp4, temp12);
864     /* y4j = (w2j >> 1) - w6j                                              */
865     temp5 = _mm_srai_epi16(resq_r2_2, 1); //w2>>1
866     temp5 = _mm_sub_epi16(temp5, resq_r6_2); //(w2>>1)-w6
867     /* y5j = -w1j + w7j + w5j + (w5j >> 1)                                   */
868     temp6 = _mm_sub_epi32(resq_r7_1, resq_r1_1); //w7-w1
869     temp14 = _mm_sub_epi32(resq_r7_2, resq_r1_2);
870     temp6 = _mm_add_epi32(temp6, resq_r5_1); //w7-w1+w5
871     temp14 = _mm_add_epi32(temp14, resq_r5_2);
872     temp7 = _mm_srai_epi32(resq_r5_1, 1); //w5>>1
873     temp15 = _mm_srai_epi32(resq_r5_2, 1);
874     temp6 = _mm_add_epi32(temp6, temp7); //w7-w1_w5+(w5>>1)
875     temp14 = _mm_add_epi32(temp14, temp15);
876     temp6 = _mm_packs_epi32(temp6, temp14);
877     /* y6j = w2j + (w6j >> 1)                                              */
878     temp7 = _mm_srai_epi16(resq_r6_2, 1); //w6>>1
879     temp7 = _mm_add_epi16(temp7, resq_r2_2); //(w6>>1)+w2
880     /* y7j = w3j + w5j + w1j + (w1j >> 1)                                    */
881     temp8 = _mm_add_epi32(resq_r3_1, resq_r5_1); //w3+w5
882     temp16 = _mm_add_epi32(resq_r3_2, resq_r5_2);
883     temp8 = _mm_add_epi32(temp8, resq_r1_1); //w3+w5+w1
884     temp16 = _mm_add_epi32(temp16, resq_r1_2);
885     temp17 = _mm_srai_epi32(resq_r1_1, 1); //w1>>1
886     temp18 = _mm_srai_epi32(resq_r1_2, 1);
887     temp8 = _mm_add_epi32(temp8, temp17); //w3+w5+w1+(w1>>1)
888     temp16 = _mm_add_epi32(temp16, temp18);
889     temp8 = _mm_packs_epi32(temp8, temp16);
890     /*------------------------------------------------------------------*/
891     /*------------------------------------------------------------------*/
892     /* z0j = y0j + y6j                                                        */
893     resq_r0_1 = _mm_add_epi16(temp1, temp7);
894     /* z1j = y1j + (y7j >> 2)                                                */
895     resq_r1_1 = _mm_srai_epi16(temp8, 2);
896     resq_r1_1 = _mm_add_epi16(resq_r1_1, temp2);
897     /* z2j = y2j + y4j                                                        */
898     resq_r2_1 = _mm_add_epi16(temp3, temp5);
899     /* z3j = y3j + (y5j >> 2)                                                */
900     resq_r3_1 = _mm_srai_epi16(temp6, 2);
901     resq_r3_1 = _mm_add_epi16(resq_r3_1, temp4);
902     /* z4j = y2j - y4j                                                        */
903     resq_r4_1 = _mm_sub_epi16(temp3, temp5);
904     /* z5j = (y3j >> 2) - y5j                                                 */
905     resq_r5_1 = _mm_srai_epi16(temp4, 2);
906     resq_r5_1 = _mm_sub_epi16(resq_r5_1, temp6);
907     /* z6j = y0j - y6j                                                     */
908     resq_r6_1 = _mm_sub_epi16(temp1, temp7);
909     /* z7j = y7j - (y1j >> 2)                                                 */
910     resq_r7_1 = _mm_srai_epi16(temp2, 2);
911     resq_r7_1 = _mm_sub_epi16(temp8, resq_r7_1);
912     /*------------------------------------------------------------------*/
913 
914     /*------------------------------------------------------------------*/
915     /* x0j = z0j + z7j                                                        */
916     temp1 = _mm_add_epi16(resq_r0_1, resq_r7_1);
917     sign_reg = _mm_cmpgt_epi16(zero_8x16b, temp1);
918     temp10 = _mm_unpacklo_epi16(temp1, sign_reg);
919     temp11 = _mm_unpackhi_epi16(temp1, sign_reg);
920     temp10 = _mm_add_epi32(temp10, value_32);
921     temp11 = _mm_add_epi32(temp11, value_32);
922     temp10 = _mm_srai_epi32(temp10, 6);
923     temp11 = _mm_srai_epi32(temp11, 6);
924     temp10 = _mm_packs_epi32(temp10, temp11);
925     temp1 = _mm_add_epi16(temp10, pred_r0_1);
926     /* x1j = z2j + z5j                                                        */
927     temp2 = _mm_add_epi16(resq_r2_1, resq_r5_1);
928     sign_reg = _mm_cmpgt_epi16(zero_8x16b, temp2);
929     temp10 = _mm_unpacklo_epi16(temp2, sign_reg);
930     temp11 = _mm_unpackhi_epi16(temp2, sign_reg);
931     temp10 = _mm_add_epi32(temp10, value_32);
932     temp11 = _mm_add_epi32(temp11, value_32);
933     temp10 = _mm_srai_epi32(temp10, 6);
934     temp11 = _mm_srai_epi32(temp11, 6);
935     temp10 = _mm_packs_epi32(temp10, temp11);
936     temp2 = _mm_add_epi16(temp10, pred_r1_1);
937     /* x2j = z4j + z3j                                                        */
938     temp3 = _mm_add_epi16(resq_r4_1, resq_r3_1);
939     sign_reg = _mm_cmpgt_epi16(zero_8x16b, temp3);
940     temp10 = _mm_unpacklo_epi16(temp3, sign_reg);
941     temp11 = _mm_unpackhi_epi16(temp3, sign_reg);
942     temp10 = _mm_add_epi32(temp10, value_32);
943     temp11 = _mm_add_epi32(temp11, value_32);
944     temp10 = _mm_srai_epi32(temp10, 6);
945     temp11 = _mm_srai_epi32(temp11, 6);
946     temp10 = _mm_packs_epi32(temp10, temp11);
947     temp3 = _mm_add_epi16(temp10, pred_r2_1);
948     /* x3j = z6j + z1j                                                        */
949     temp4 = _mm_add_epi16(resq_r6_1, resq_r1_1);
950     sign_reg = _mm_cmpgt_epi16(zero_8x16b, temp4);
951     temp10 = _mm_unpacklo_epi16(temp4, sign_reg);
952     temp11 = _mm_unpackhi_epi16(temp4, sign_reg);
953     temp10 = _mm_add_epi32(temp10, value_32);
954     temp11 = _mm_add_epi32(temp11, value_32);
955     temp10 = _mm_srai_epi32(temp10, 6);
956     temp11 = _mm_srai_epi32(temp11, 6);
957     temp10 = _mm_packs_epi32(temp10, temp11);
958     temp4 = _mm_add_epi16(temp10, pred_r3_1);
959     /* x4j = z6j - z1j                                                        */
960     temp5 = _mm_sub_epi16(resq_r6_1, resq_r1_1);
961     sign_reg = _mm_cmpgt_epi16(zero_8x16b, temp5);
962     temp10 = _mm_unpacklo_epi16(temp5, sign_reg);
963     temp11 = _mm_unpackhi_epi16(temp5, sign_reg);
964     temp10 = _mm_add_epi32(temp10, value_32);
965     temp11 = _mm_add_epi32(temp11, value_32);
966     temp10 = _mm_srai_epi32(temp10, 6);
967     temp11 = _mm_srai_epi32(temp11, 6);
968     temp10 = _mm_packs_epi32(temp10, temp11);
969     temp5 = _mm_add_epi16(temp10, pred_r4_1);
970     /* x5j = z4j - z3j                                                        */
971     temp6 = _mm_sub_epi16(resq_r4_1, resq_r3_1);
972     sign_reg = _mm_cmpgt_epi16(zero_8x16b, temp6);
973     temp10 = _mm_unpacklo_epi16(temp6, sign_reg);
974     temp11 = _mm_unpackhi_epi16(temp6, sign_reg);
975     temp10 = _mm_add_epi32(temp10, value_32);
976     temp11 = _mm_add_epi32(temp11, value_32);
977     temp10 = _mm_srai_epi32(temp10, 6);
978     temp11 = _mm_srai_epi32(temp11, 6);
979     temp10 = _mm_packs_epi32(temp10, temp11);
980     temp6 = _mm_add_epi16(temp10, pred_r5_1);
981     /* x6j = z2j - z5j                                                        */
982     temp7 = _mm_sub_epi16(resq_r2_1, resq_r5_1);
983     sign_reg = _mm_cmpgt_epi16(zero_8x16b, temp7);
984     temp10 = _mm_unpacklo_epi16(temp7, sign_reg);
985     temp11 = _mm_unpackhi_epi16(temp7, sign_reg);
986     temp10 = _mm_add_epi32(temp10, value_32);
987     temp11 = _mm_add_epi32(temp11, value_32);
988     temp10 = _mm_srai_epi32(temp10, 6);
989     temp11 = _mm_srai_epi32(temp11, 6);
990     temp10 = _mm_packs_epi32(temp10, temp11);
991     temp7 = _mm_add_epi16(temp10, pred_r6_1);
992     /* x7j = z0j - z7j                                                        */
993     temp8 = _mm_sub_epi16(resq_r0_1, resq_r7_1);
994     sign_reg = _mm_cmpgt_epi16(zero_8x16b, temp8);
995     temp10 = _mm_unpacklo_epi16(temp8, sign_reg);
996     temp11 = _mm_unpackhi_epi16(temp8, sign_reg);
997     temp10 = _mm_add_epi32(temp10, value_32);
998     temp11 = _mm_add_epi32(temp11, value_32);
999     temp10 = _mm_srai_epi32(temp10, 6);
1000     temp11 = _mm_srai_epi32(temp11, 6);
1001     temp10 = _mm_packs_epi32(temp10, temp11);
1002     temp8 = _mm_add_epi16(temp10, pred_r7_1);
1003     /*------------------------------------------------------------------*/
1004     //Clipping the results to 8 bits
1005     sign_reg = _mm_cmpgt_epi16(temp1, zero_8x16b); // sign check
1006     temp1 = _mm_and_si128(temp1, sign_reg);
1007     sign_reg = _mm_cmpgt_epi16(temp2, zero_8x16b); // sign check
1008     temp2 = _mm_and_si128(temp2, sign_reg);
1009     sign_reg = _mm_cmpgt_epi16(temp3, zero_8x16b); // sign check
1010     temp3 = _mm_and_si128(temp3, sign_reg);
1011     sign_reg = _mm_cmpgt_epi16(temp4, zero_8x16b); // sign check
1012     temp4 = _mm_and_si128(temp4, sign_reg);
1013     sign_reg = _mm_cmpgt_epi16(temp5, zero_8x16b); // sign check
1014     temp5 = _mm_and_si128(temp5, sign_reg);
1015     sign_reg = _mm_cmpgt_epi16(temp6, zero_8x16b); // sign check
1016     temp6 = _mm_and_si128(temp6, sign_reg);
1017     sign_reg = _mm_cmpgt_epi16(temp7, zero_8x16b); // sign check
1018     temp7 = _mm_and_si128(temp7, sign_reg);
1019     sign_reg = _mm_cmpgt_epi16(temp8, zero_8x16b); // sign check
1020     temp8 = _mm_and_si128(temp8, sign_reg);
1021 
1022     resq_r0_2 = _mm_packus_epi16(temp1, zero_8x16b);
1023     resq_r1_2 = _mm_packus_epi16(temp2, zero_8x16b);
1024     resq_r2_2 = _mm_packus_epi16(temp3, zero_8x16b);
1025     resq_r3_2 = _mm_packus_epi16(temp4, zero_8x16b);
1026     resq_r4_2 = _mm_packus_epi16(temp5, zero_8x16b);
1027     resq_r5_2 = _mm_packus_epi16(temp6, zero_8x16b);
1028     resq_r6_2 = _mm_packus_epi16(temp7, zero_8x16b);
1029     resq_r7_2 = _mm_packus_epi16(temp8, zero_8x16b);
1030 
1031     _mm_storel_epi64((__m128i *) (&pu1_out[0]), resq_r0_2);
1032     _mm_storel_epi64((__m128i *) (&pu1_out[out_strd]), resq_r1_2);
1033     _mm_storel_epi64((__m128i *) (&pu1_out[2 * out_strd]), resq_r2_2);
1034     _mm_storel_epi64((__m128i *) (&pu1_out[3 * out_strd]), resq_r3_2);
1035     _mm_storel_epi64((__m128i *) (&pu1_out[4 * out_strd]), resq_r4_2);
1036     _mm_storel_epi64((__m128i *) (&pu1_out[5 * out_strd]), resq_r5_2);
1037     _mm_storel_epi64((__m128i *) (&pu1_out[6 * out_strd]), resq_r6_2);
1038     _mm_storel_epi64((__m128i *) (&pu1_out[7 * out_strd]), resq_r7_2);
1039 }
1040 
1041