• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /******************************************************************************
2  *
3  * Copyright (C) 2022 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  *  isvc_iquant_itrans_recon_sse42.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  *  - isvc_iquant_itrans_recon_4x4_sse42()
34  *  - isvc_iquant_itrans_recon_chroma_4x4_sse42()
35  *
36  * @remarks
37  *  None
38  *
39  *******************************************************************************
40  */
41 #include <immintrin.h>
42 
43 #include "ih264_typedefs.h"
44 #include "ih264_debug.h"
45 #include "ih264_defs.h"
46 #include "ih264_trans_macros.h"
47 #include "ih264_macros.h"
48 #include "ih264_platform_macros.h"
49 #include "ih264_trans_data.h"
50 #include "ih264_size_defs.h"
51 #include "isvc_structs.h"
52 #include "isvc_trans_quant_itrans_iquant.h"
53 
54 /*
55  ********************************************************************************
56  *
57  * @brief This function reconstructs a 4x4 sub block from quantized resiude and
58  * prediction buffer
59  *
60  * @par Description:
61  *  The quantized residue is first inverse quantized, then inverse transformed.
62  *  This inverse transformed content is added to the prediction buffer to recon-
63  *  struct the end output
64  *
65  * @param[in] pi2_src
66  *  quantized 4x4 block
67  *
68  * @param[in] pu1_pred
69  *  prediction 4x4 block
70  *
71  * @param[out] pu1_out
72  *  reconstructed 4x4 block
73  *
74  * @param[in] src_strd
75  *  quantization buffer stride
76  *
77  * @param[in] i4_pred_stride,
78  *  Prediction buffer stride
79  *
80  * @param[in] i4_out_stride
81  *  recon buffer Stride
82  *
83  * @param[in] pu2_scaling_list
84  *  pointer to scaling list
85  *
86  * @param[in] pu2_norm_adjust
87  *  pointer to inverse scale matrix
88  *
89  * @param[in] u4_qp_div_6
90  *  Floor (qp/6)
91  *
92  * @param[in] pi4_tmp
93  * temporary buffer of size 1*16
94  *
95  * @returns none
96  *
97  * @remarks none
98  *
99  *******************************************************************************
100  */
101 
isvc_iquant_itrans_recon_4x4_sse42(buffer_container_t * ps_src,buffer_container_t * ps_pred,buffer_container_t * ps_res_pred,buffer_container_t * ps_res,buffer_container_t * ps_rec,iq_it_res_rec_constants_t * ps_iq_it_res_rec_constants,WORD16 * pi2_tmp,WORD16 * pi2_dc_src,WORD32 i4_iq_start_idx,UWORD8 u1_res_accumulate)102 void isvc_iquant_itrans_recon_4x4_sse42(buffer_container_t *ps_src, buffer_container_t *ps_pred,
103                                         buffer_container_t *ps_res_pred, buffer_container_t *ps_res,
104                                         buffer_container_t *ps_rec,
105                                         iq_it_res_rec_constants_t *ps_iq_it_res_rec_constants,
106                                         WORD16 *pi2_tmp, WORD16 *pi2_dc_src, WORD32 i4_iq_start_idx,
107                                         UWORD8 u1_res_accumulate)
108 {
109     WORD16 *pi2_src = (WORD16 *) ps_src->pv_data;
110     WORD16 *pi2_tmp_ptr = pi2_tmp;
111     UWORD8 *pu1_pred = (UWORD8 *) ps_pred->pv_data;
112     UWORD8 *pu1_out = (UWORD8 *) ps_rec->pv_data;
113     WORD32 i4_src_stride = ps_src->i4_data_stride;
114     WORD32 i4_pred_stride = ps_pred->i4_data_stride;
115     WORD32 i4_out_stride = ps_rec->i4_data_stride;
116     const UWORD16 *pu2_iscal_mat = ps_iq_it_res_rec_constants->pu2_iscal_mat;
117     const UWORD16 *pu2_weigh_mat = ps_iq_it_res_rec_constants->pu2_weigh_mat;
118     UWORD32 u4_qp_div_6 = ps_iq_it_res_rec_constants->u4_qp_div_6;
119     UWORD32 *pu4_out = (UWORD32 *) pu1_out;
120     __m128i src_r0_r1, src_r2_r3;
121     __m128i src_r0, src_r1, src_r2, src_r3;
122     __m128i scalemat_r0_r1, scalemat_r2_r3;
123     __m128i pred_r0, pred_r1, pred_r2, pred_r3;
124     __m128i sign_reg, dequant_r0_r1, dequant_r2_r3;
125     /* all bits reset to zero */
126     __m128i zero_8x16b = _mm_setzero_si128();
127     __m128i neg_255_8x16b = _mm_set1_epi16(-((WORD16) UINT8_MAX));
128     __m128i pos_255_8x16b = _mm_set1_epi16(((WORD16) UINT8_MAX));
129     __m128i temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
130     __m128i resq_r0, resq_r1, resq_r2, resq_r3;
131     __m128i add_rshift = _mm_set1_epi32((u4_qp_div_6 < 4) ? (1 << (3 - u4_qp_div_6)) : 0);
132     __m128i value_32 = _mm_set1_epi32(32);
133 
134     ASSERT(4 == i4_src_stride);
135     ASSERT(0 == u1_res_accumulate);
136 
137     UNUSED(i4_src_stride);
138     UNUSED(ps_res);
139     UNUSED(ps_res_pred);
140     UNUSED(u1_res_accumulate);
141 
142     /*************************************************************/
143     /* Dequantization of coefficients. Will be replaced by SIMD  */
144     /* operations on platform                                    */
145     /*************************************************************/
146 
147     /* a00 a01 a02 a03 a10 a11 a12 a13 -- the source
148      matrix 0th,1st row */
149     src_r0_r1 = _mm_loadu_si128((__m128i *) (pi2_src));
150 
151     /* a20 a21 a22 a23 a30 a31 a32 a33 -- the
152       source matrix 2nd,3rd row */
153     src_r2_r3 = _mm_loadu_si128((__m128i *) (pi2_src + 8));
154 
155     /* b00 b01 b02 b03 b10 b11 b12 b13 -- the
156      scaling matrix 0th,1st row */
157     scalemat_r0_r1 = _mm_loadu_si128((__m128i *) (pu2_iscal_mat));
158 
159     /* b20 b21 b22 b23 b30 b31 b32 b33 --b12 b13 -- the
160      the scaling matrix 2nd,3rd row */
161     scalemat_r2_r3 = _mm_loadu_si128((__m128i *) (pu2_iscal_mat + 8));
162 
163     /* q00 q01 q02 q03 q10 q11
164      q12 q13 -- all 16 bits */
165     dequant_r0_r1 = _mm_loadu_si128((__m128i *) (pu2_weigh_mat));
166 
167     /* q20 q21 q22 q23 q30 q31
168      q32 q33 -- all 16 bits */
169     dequant_r2_r3 = _mm_loadu_si128((__m128i *) (pu2_weigh_mat + 8));
170 
171     /* b00*q00 b01*q01 b02*q02 b03*q03 b10*q10 b11*q11
172      b12*q12 b13*q13 -- 16 bit result */
173     temp0 = _mm_mullo_epi16(scalemat_r0_r1, dequant_r0_r1);
174 
175     /* b20*q20 b21*q21 b22*q22 b23*q23 b30*q30 b31*q31
176      b32*q32 b33*q33 -- 16 bit result */
177     temp1 = _mm_mullo_epi16(scalemat_r2_r3, dequant_r2_r3);
178 
179     /* b00*q00 0 b01*q01 0 b02*q02 0 b03*q03 0 -- 16 bit long */
180     temp4 = _mm_unpacklo_epi16(temp0, zero_8x16b);
181 
182     /* b10*q10 0 b11*q11 0 b12*q12 0 b13*q13 0 -- 16 bit long */
183     temp5 = _mm_unpackhi_epi16(temp0, zero_8x16b);
184 
185     /* b00*q00 0 b01*q01 0 b02*q02 0 b03*q03 0 -- 16 bit long */
186     temp6 = _mm_unpacklo_epi16(temp1, zero_8x16b);
187 
188     /* b10*q10 0 b11*q11 0 b12*q12 0 b13*q13 0 -- 16 bit long */
189     temp7 = _mm_unpackhi_epi16(temp1, zero_8x16b);
190 
191     /* a00 0 a01 0 a02 0 a03 0 -- 16 bit long */
192     src_r0 = _mm_unpacklo_epi16(src_r0_r1, zero_8x16b);
193     /* a10 0 a11 0 a12 0 a13 0 -- 16 bit long */
194     src_r1 = _mm_unpackhi_epi16(src_r0_r1, zero_8x16b);
195     /* a20 0 a21 0 a22 0 a23 0 -- 16 bit long */
196     src_r2 = _mm_unpacklo_epi16(src_r2_r3, zero_8x16b);
197     /* a30 0 a31 0 a32 0 a33 0 -- 16 bit long */
198     src_r3 = _mm_unpackhi_epi16(src_r2_r3, zero_8x16b);
199 
200     temp4 = _mm_madd_epi16(src_r0, temp4);
201     temp5 = _mm_madd_epi16(src_r1, temp5);
202     temp6 = _mm_madd_epi16(src_r2, temp6);
203     temp7 = _mm_madd_epi16(src_r3, temp7);
204 
205     if(u4_qp_div_6 >= 4)
206     {
207         resq_r0 = _mm_slli_epi32(temp4, u4_qp_div_6 - 4);
208         resq_r1 = _mm_slli_epi32(temp5, u4_qp_div_6 - 4);
209         resq_r2 = _mm_slli_epi32(temp6, u4_qp_div_6 - 4);
210         resq_r3 = _mm_slli_epi32(temp7, u4_qp_div_6 - 4);
211     }
212     else
213     {
214         temp4 = _mm_add_epi32(temp4, add_rshift);
215         temp5 = _mm_add_epi32(temp5, add_rshift);
216         temp6 = _mm_add_epi32(temp6, add_rshift);
217         temp7 = _mm_add_epi32(temp7, add_rshift);
218         resq_r0 = _mm_srai_epi32(temp4, 4 - u4_qp_div_6);
219         resq_r1 = _mm_srai_epi32(temp5, 4 - u4_qp_div_6);
220         resq_r2 = _mm_srai_epi32(temp6, 4 - u4_qp_div_6);
221         resq_r3 = _mm_srai_epi32(temp7, 4 - u4_qp_div_6);
222     }
223 
224     if(i4_iq_start_idx == 1) resq_r0 = _mm_insert_epi32(resq_r0, (WORD32) pi2_dc_src[0], 0);
225     /* Perform Inverse transform */
226     /*-------------------------------------------------------------*/
227     /* IDCT [ Horizontal transformation ]                          */
228     /*-------------------------------------------------------------*/
229     // Matrix transpose
230     /*
231      *  a0 a1 a2 a3
232      *  b0 b1 b2 b3
233      *  c0 c1 c2 c3
234      *  d0 d1 d2 d3
235      */
236 
237     /* a0 b0 a1 b1 */
238     temp1 = _mm_unpacklo_epi32(resq_r0, resq_r1);
239     /* c0 d0 c1 d1 */
240     temp3 = _mm_unpacklo_epi32(resq_r2, resq_r3);
241     /* a2 b2 a3 b3 */
242     temp2 = _mm_unpackhi_epi32(resq_r0, resq_r1);
243     /* c2 d2 c3 d3 */
244     temp4 = _mm_unpackhi_epi32(resq_r2, resq_r3);
245     /* a0 b0 c0 d0 */
246     resq_r0 = _mm_unpacklo_epi64(temp1, temp3);
247     /* a1 b1 c1 d1 */
248     resq_r1 = _mm_unpackhi_epi64(temp1, temp3);
249     /* a2 b2 c2 d2 */
250     resq_r2 = _mm_unpacklo_epi64(temp2, temp4);
251     /* a3 b3 c3 d3 */
252     resq_r3 = _mm_unpackhi_epi64(temp2, temp4);
253     /* Transform starts -- horizontal transform */
254     /*------------------------------------------------------------------*/
255     /* z0 = w0 + w2                                             */
256     temp0 = _mm_add_epi32(resq_r0, resq_r2);
257     /* z1 = w0 - w2                                             */
258     temp1 = _mm_sub_epi32(resq_r0, resq_r2);
259     /* z2 = (w1 >> 1) - w3                                      */
260     temp2 = _mm_srai_epi32(resq_r1, 1);
261     temp2 = _mm_sub_epi32(temp2, resq_r3);
262     /* z3 = w1 + (w3 >> 1)                                      */
263     temp3 = _mm_srai_epi32(resq_r3, 1);
264     temp3 = _mm_add_epi32(temp3, resq_r1);
265     /*----------------------------------------------------------*/
266     /* x0 = z0 + z3                                             */
267     resq_r0 = _mm_add_epi32(temp0, temp3);
268     /* x1 = z1 + z2                                             */
269     resq_r1 = _mm_add_epi32(temp1, temp2);
270     /* x2 = z1 - z2                                             */
271     resq_r2 = _mm_sub_epi32(temp1, temp2);
272     /* x3 = z0 - z3                                             */
273     resq_r3 = _mm_sub_epi32(temp0, temp3);
274 
275     // Matrix transpose
276     /*
277      *  a0 b0 c0 d0
278      *  a1 b1 c1 d1
279      *  a2 b2 c2 d2
280      *  a3 b3 c3 d3
281      */
282 
283     /* a0 a1 b0 b1 */
284     temp1 = _mm_unpacklo_epi32(resq_r0, resq_r1);
285     /* a2 a3 b2 b3 */
286     temp3 = _mm_unpacklo_epi32(resq_r2, resq_r3);
287     /* c0 c1 d0 d1 */
288     temp2 = _mm_unpackhi_epi32(resq_r0, resq_r1);
289     /* c2 c3 d2 d3 */
290     temp4 = _mm_unpackhi_epi32(resq_r2, resq_r3);
291     /* a0 a1 a2 a3 */
292     resq_r0 = _mm_unpacklo_epi64(temp1, temp3);
293     /* b0 b1 b2 b3 */
294     resq_r1 = _mm_unpackhi_epi64(temp1, temp3);
295     /* c0 c1 c2 c3 */
296     resq_r2 = _mm_unpacklo_epi64(temp2, temp4);
297     /* d0 d1 d2 d3 */
298     resq_r3 = _mm_unpackhi_epi64(temp2, temp4);
299     /* Transform ends -- horizontal transform */
300 
301     temp0 = _mm_packs_epi32(resq_r0, resq_r1);
302     temp1 = _mm_packs_epi32(resq_r2, resq_r3);
303 
304     _mm_storeu_si128((__m128i *) (&pi2_tmp_ptr[0]), temp0);
305     _mm_storeu_si128((__m128i *) (&pi2_tmp_ptr[2 * 4]), temp1);
306 
307     /* Load pred buffer */
308     pred_r0 = _mm_loadl_epi64((__m128i *) (&pu1_pred[0]));
309     pred_r1 = _mm_loadl_epi64((__m128i *) (&pu1_pred[i4_pred_stride]));
310     pred_r2 = _mm_loadl_epi64((__m128i *) (&pu1_pred[2 * i4_pred_stride]));
311     pred_r3 = _mm_loadl_epi64((__m128i *) (&pu1_pred[3 * i4_pred_stride]));
312 
313     pred_r0 = _mm_cvtepu8_epi16(pred_r0);
314     pred_r1 = _mm_cvtepu8_epi16(pred_r1);
315     pred_r2 = _mm_cvtepu8_epi16(pred_r2);
316     pred_r3 = _mm_cvtepu8_epi16(pred_r3);
317 
318     pred_r0 = _mm_unpacklo_epi64(pred_r0, pred_r1);
319     pred_r1 = _mm_unpacklo_epi64(pred_r2, pred_r3);
320 
321     /*--------------------------------------------------------------*/
322     /* IDCT [ Vertical transformation] and Xij = (xij + 32)>>6      */
323     /*                                                              */
324     /* Add the prediction and store it back to same buffer          */
325     /*--------------------------------------------------------------*/
326     /* z0j = y0j + y2j                                                        */
327     temp0 = _mm_add_epi32(resq_r0, resq_r2);
328     /* z1j = y0j - y2j                                                        */
329     temp1 = _mm_sub_epi32(resq_r0, resq_r2);
330     /* z2j = (y1j>>1) - y3j */
331     temp2 = _mm_srai_epi32(resq_r1, 1);
332     temp2 = _mm_sub_epi32(temp2, resq_r3);
333     /* z3j = y1j + (y3j>>1) */
334     temp3 = _mm_srai_epi32(resq_r3, 1);
335     temp3 = _mm_add_epi32(temp3, resq_r1);
336 
337     /* x0j = z0j + z3j                                                        */
338     temp4 = _mm_add_epi32(temp0, temp3);
339     temp4 = _mm_add_epi32(temp4, value_32);
340     temp4 = _mm_srai_epi32(temp4, 6);
341     /* x1j = z1j + z2j                                                        */
342     temp5 = _mm_add_epi32(temp1, temp2);
343     temp5 = _mm_add_epi32(temp5, value_32);
344     temp5 = _mm_srai_epi32(temp5, 6);
345     /* x2j = z1j - z2j                                                        */
346     temp6 = _mm_sub_epi32(temp1, temp2);
347     temp6 = _mm_add_epi32(temp6, value_32);
348     temp6 = _mm_srai_epi32(temp6, 6);
349     /* x3j = z0j - z3j                                                        */
350     temp7 = _mm_sub_epi32(temp0, temp3);
351     temp7 = _mm_add_epi32(temp7, value_32);
352     temp7 = _mm_srai_epi32(temp7, 6);
353 
354     /* 32-bit to 16-bit conversion */
355     temp0 = _mm_packs_epi32(temp4, temp5);
356     temp1 = _mm_packs_epi32(temp6, temp7);
357 
358     /* Saturate all values < -255 to -255 and retain the rest as it is */
359     temp4 = _mm_max_epi16(temp0, neg_255_8x16b);
360     /* Saturate all values > 255 to 255 and retain the rest as it is */
361     temp4 = _mm_min_epi16(temp4, pos_255_8x16b);
362 
363     /* Saturate all values < -255 to -255 and retain the rest as it is */
364     temp5 = _mm_max_epi16(temp1, neg_255_8x16b);
365     /* Saturate all values > 255 to 255 and retain the rest as it is */
366     temp5 = _mm_min_epi16(temp5, pos_255_8x16b);
367 
368     temp0 = _mm_add_epi16(temp4, pred_r0);
369     temp1 = _mm_add_epi16(temp5, pred_r1);
370 
371     /*------------------------------------------------------------------*/
372     /* Clipping the results to 8 bits */
373     sign_reg = _mm_cmpgt_epi16(temp0, zero_8x16b);
374     temp0 = _mm_and_si128(temp0, sign_reg);
375     sign_reg = _mm_cmpgt_epi16(temp1, zero_8x16b);
376     temp1 = _mm_and_si128(temp1, sign_reg);
377 
378     resq_r0 = _mm_packus_epi16(temp0, temp1);
379     resq_r1 = _mm_srli_si128(resq_r0, 4);
380     resq_r2 = _mm_srli_si128(resq_r1, 4);
381     resq_r3 = _mm_srli_si128(resq_r2, 4);
382 
383     *pu4_out = _mm_cvtsi128_si32(resq_r0);
384     pu1_out += i4_out_stride;
385     pu4_out = (UWORD32 *) (pu1_out);
386     *(pu4_out) = _mm_cvtsi128_si32(resq_r1);
387     pu1_out += i4_out_stride;
388     pu4_out = (UWORD32 *) (pu1_out);
389     *(pu4_out) = _mm_cvtsi128_si32(resq_r2);
390     pu1_out += i4_out_stride;
391     pu4_out = (UWORD32 *) (pu1_out);
392     *(pu4_out) = _mm_cvtsi128_si32(resq_r3);
393 }
394 
isvc_iquant_itrans_recon_res_4x4_sse42(buffer_container_t * ps_src,buffer_container_t * ps_pred,buffer_container_t * ps_res_pred,buffer_container_t * ps_res,buffer_container_t * ps_rec,iq_it_res_rec_constants_t * ps_iq_it_res_rec_constants,WORD16 * pi2_tmp,WORD16 * pi2_dc_src,WORD32 i4_iq_start_idx,UWORD8 u1_res_accumulate)395 void isvc_iquant_itrans_recon_res_4x4_sse42(buffer_container_t *ps_src, buffer_container_t *ps_pred,
396                                             buffer_container_t *ps_res_pred,
397                                             buffer_container_t *ps_res, buffer_container_t *ps_rec,
398                                             iq_it_res_rec_constants_t *ps_iq_it_res_rec_constants,
399                                             WORD16 *pi2_tmp, WORD16 *pi2_dc_src,
400                                             WORD32 i4_iq_start_idx, UWORD8 u1_res_accumulate)
401 {
402     WORD16 *pi2_src = (WORD16 *) ps_src->pv_data;
403     WORD16 *pi2_tmp_ptr = pi2_tmp;
404     WORD16 *pi2_res = (WORD16 *) ps_res->pv_data;
405     UWORD8 *pu1_pred = (UWORD8 *) ps_pred->pv_data;
406     UWORD8 *pu1_out = (UWORD8 *) ps_rec->pv_data;
407     WORD32 i4_src_stride = ps_src->i4_data_stride;
408     WORD32 i4_res_stride = ps_res->i4_data_stride;
409     WORD32 i4_pred_stride = ps_pred->i4_data_stride;
410     WORD32 i4_out_stride = ps_rec->i4_data_stride;
411     const UWORD16 *pu2_iscal_mat = ps_iq_it_res_rec_constants->pu2_iscal_mat;
412     const UWORD16 *pu2_weigh_mat = ps_iq_it_res_rec_constants->pu2_weigh_mat;
413     UWORD32 u4_qp_div_6 = ps_iq_it_res_rec_constants->u4_qp_div_6;
414     UWORD32 *pu4_out = (UWORD32 *) pu1_out;
415     __m128i src_r0_r1, src_r2_r3;
416     __m128i src_r0, src_r1, src_r2, src_r3;
417     __m128i scalemat_r0_r1, scalemat_r2_r3;
418     __m128i pred_r0, pred_r1, pred_r2, pred_r3;
419     __m128i sign_reg, dequant_r0_r1, dequant_r2_r3;
420     /* all bits reset to zero */
421     __m128i zero_8x16b = _mm_setzero_si128();
422     __m128i neg_255_8x16b = _mm_set1_epi16(-((WORD16) UINT8_MAX));
423     __m128i pos_255_8x16b = _mm_set1_epi16(((WORD16) UINT8_MAX));
424     __m128i temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
425     __m128i resq_r0, resq_r1, resq_r2, resq_r3;
426     __m128i add_rshift = _mm_set1_epi32((u4_qp_div_6 < 4) ? (1 << (3 - u4_qp_div_6)) : 0);
427     __m128i value_32 = _mm_set1_epi32(32);
428 
429     ASSERT(4 == i4_src_stride);
430     ASSERT(0 == u1_res_accumulate);
431 
432     UNUSED(i4_src_stride);
433     UNUSED(ps_res_pred);
434     UNUSED(u1_res_accumulate);
435 
436     /*************************************************************/
437     /* Dequantization of coefficients. Will be replaced by SIMD  */
438     /* operations on platform                                    */
439     /*************************************************************/
440 
441     /* a00 a01 a02 a03 a10 a11 a12 a13 -- the source
442     matrix 0th,1st row */
443     src_r0_r1 = _mm_loadu_si128((__m128i *) (pi2_src));
444 
445     /* a20 a21 a22 a23 a30 a31 a32 a33 -- the
446     source matrix 2nd,3rd row */
447     src_r2_r3 = _mm_loadu_si128((__m128i *) (pi2_src + 8));
448 
449     /* b00 b01 b02 b03 b10 b11 b12 b13 -- the
450     scaling matrix 0th,1st row */
451     scalemat_r0_r1 = _mm_loadu_si128((__m128i *) (pu2_iscal_mat));
452 
453     /* b20 b21 b22 b23 b30 b31 b32 b33 --b12 b13 -- the
454     the scaling matrix 2nd,3rd row */
455     scalemat_r2_r3 = _mm_loadu_si128((__m128i *) (pu2_iscal_mat + 8));
456 
457     /* q00 q01 q02 q03 q10 q11
458     q12 q13 -- all 16 bits */
459     dequant_r0_r1 = _mm_loadu_si128((__m128i *) (pu2_weigh_mat));
460 
461     /* q20 q21 q22 q23 q30 q31
462     q32 q33 -- all 16 bits */
463     dequant_r2_r3 = _mm_loadu_si128((__m128i *) (pu2_weigh_mat + 8));
464 
465     /* b00*q00 b01*q01 b02*q02 b03*q03 b10*q10 b11*q11
466     b12*q12 b13*q13 -- 16 bit result */
467     temp0 = _mm_mullo_epi16(scalemat_r0_r1, dequant_r0_r1);
468 
469     /* b20*q20 b21*q21 b22*q22 b23*q23 b30*q30 b31*q31
470     b32*q32 b33*q33 -- 16 bit result */
471     temp1 = _mm_mullo_epi16(scalemat_r2_r3, dequant_r2_r3);
472 
473     /* b00*q00 0 b01*q01 0 b02*q02 0 b03*q03 0 -- 16 bit long */
474     temp4 = _mm_unpacklo_epi16(temp0, zero_8x16b);
475 
476     /* b10*q10 0 b11*q11 0 b12*q12 0 b13*q13 0 -- 16 bit long */
477     temp5 = _mm_unpackhi_epi16(temp0, zero_8x16b);
478 
479     /* b00*q00 0 b01*q01 0 b02*q02 0 b03*q03 0 -- 16 bit long */
480     temp6 = _mm_unpacklo_epi16(temp1, zero_8x16b);
481 
482     /* b10*q10 0 b11*q11 0 b12*q12 0 b13*q13 0 -- 16 bit long */
483     temp7 = _mm_unpackhi_epi16(temp1, zero_8x16b);
484 
485     /* a00 0 a01 0 a02 0 a03 0 -- 16 bit long */
486     src_r0 = _mm_unpacklo_epi16(src_r0_r1, zero_8x16b);
487     /* a10 0 a11 0 a12 0 a13 0 -- 16 bit long */
488     src_r1 = _mm_unpackhi_epi16(src_r0_r1, zero_8x16b);
489     /* a20 0 a21 0 a22 0 a23 0 -- 16 bit long */
490     src_r2 = _mm_unpacklo_epi16(src_r2_r3, zero_8x16b);
491     /* a30 0 a31 0 a32 0 a33 0 -- 16 bit long */
492     src_r3 = _mm_unpackhi_epi16(src_r2_r3, zero_8x16b);
493 
494     temp4 = _mm_madd_epi16(src_r0, temp4);
495     temp5 = _mm_madd_epi16(src_r1, temp5);
496     temp6 = _mm_madd_epi16(src_r2, temp6);
497     temp7 = _mm_madd_epi16(src_r3, temp7);
498 
499     if(u4_qp_div_6 >= 4)
500     {
501         resq_r0 = _mm_slli_epi32(temp4, u4_qp_div_6 - 4);
502         resq_r1 = _mm_slli_epi32(temp5, u4_qp_div_6 - 4);
503         resq_r2 = _mm_slli_epi32(temp6, u4_qp_div_6 - 4);
504         resq_r3 = _mm_slli_epi32(temp7, u4_qp_div_6 - 4);
505     }
506     else
507     {
508         temp4 = _mm_add_epi32(temp4, add_rshift);
509         temp5 = _mm_add_epi32(temp5, add_rshift);
510         temp6 = _mm_add_epi32(temp6, add_rshift);
511         temp7 = _mm_add_epi32(temp7, add_rshift);
512         resq_r0 = _mm_srai_epi32(temp4, 4 - u4_qp_div_6);
513         resq_r1 = _mm_srai_epi32(temp5, 4 - u4_qp_div_6);
514         resq_r2 = _mm_srai_epi32(temp6, 4 - u4_qp_div_6);
515         resq_r3 = _mm_srai_epi32(temp7, 4 - u4_qp_div_6);
516     }
517 
518     if(i4_iq_start_idx == 1) resq_r0 = _mm_insert_epi32(resq_r0, (WORD32) pi2_dc_src[0], 0);
519     /* Perform Inverse transform */
520     /*-------------------------------------------------------------*/
521     /* IDCT [ Horizontal transformation ]                          */
522     /*-------------------------------------------------------------*/
523     // Matrix transpose
524     /*
525      *  a0 a1 a2 a3
526      *  b0 b1 b2 b3
527      *  c0 c1 c2 c3
528      *  d0 d1 d2 d3
529      */
530 
531     /* a0 b0 a1 b1 */
532     temp1 = _mm_unpacklo_epi32(resq_r0, resq_r1);
533     /* c0 d0 c1 d1 */
534     temp3 = _mm_unpacklo_epi32(resq_r2, resq_r3);
535     /* a2 b2 a3 b3 */
536     temp2 = _mm_unpackhi_epi32(resq_r0, resq_r1);
537     /* c2 d2 c3 d3 */
538     temp4 = _mm_unpackhi_epi32(resq_r2, resq_r3);
539     /* a0 b0 c0 d0 */
540     resq_r0 = _mm_unpacklo_epi64(temp1, temp3);
541     /* a1 b1 c1 d1 */
542     resq_r1 = _mm_unpackhi_epi64(temp1, temp3);
543     /* a2 b2 c2 d2 */
544     resq_r2 = _mm_unpacklo_epi64(temp2, temp4);
545     /* a3 b3 c3 d3 */
546     resq_r3 = _mm_unpackhi_epi64(temp2, temp4);
547     /* Transform starts -- horizontal transform */
548     /*------------------------------------------------------------------*/
549     /* z0 = w0 + w2                                             */
550     temp0 = _mm_add_epi32(resq_r0, resq_r2);
551     /* z1 = w0 - w2                                             */
552     temp1 = _mm_sub_epi32(resq_r0, resq_r2);
553     /* z2 = (w1 >> 1) - w3                                      */
554     temp2 = _mm_srai_epi32(resq_r1, 1);
555     temp2 = _mm_sub_epi32(temp2, resq_r3);
556     /* z3 = w1 + (w3 >> 1)                                      */
557     temp3 = _mm_srai_epi32(resq_r3, 1);
558     temp3 = _mm_add_epi32(temp3, resq_r1);
559     /*----------------------------------------------------------*/
560     /* x0 = z0 + z3                                             */
561     resq_r0 = _mm_add_epi32(temp0, temp3);
562     /* x1 = z1 + z2                                             */
563     resq_r1 = _mm_add_epi32(temp1, temp2);
564     /* x2 = z1 - z2                                             */
565     resq_r2 = _mm_sub_epi32(temp1, temp2);
566     /* x3 = z0 - z3                                             */
567     resq_r3 = _mm_sub_epi32(temp0, temp3);
568 
569     // Matrix transpose
570     /*
571      *  a0 b0 c0 d0
572      *  a1 b1 c1 d1
573      *  a2 b2 c2 d2
574      *  a3 b3 c3 d3
575      */
576 
577     /* a0 a1 b0 b1 */
578     temp1 = _mm_unpacklo_epi32(resq_r0, resq_r1);
579     /* a2 a3 b2 b3 */
580     temp3 = _mm_unpacklo_epi32(resq_r2, resq_r3);
581     /* c0 c1 d0 d1 */
582     temp2 = _mm_unpackhi_epi32(resq_r0, resq_r1);
583     /* c2 c3 d2 d3 */
584     temp4 = _mm_unpackhi_epi32(resq_r2, resq_r3);
585     /* a0 a1 a2 a3 */
586     resq_r0 = _mm_unpacklo_epi64(temp1, temp3);
587     /* b0 b1 b2 b3 */
588     resq_r1 = _mm_unpackhi_epi64(temp1, temp3);
589     /* c0 c1 c2 c3 */
590     resq_r2 = _mm_unpacklo_epi64(temp2, temp4);
591     /* d0 d1 d2 d3 */
592     resq_r3 = _mm_unpackhi_epi64(temp2, temp4);
593     /* Transform ends -- horizontal transform */
594 
595     temp0 = _mm_packs_epi32(resq_r0, resq_r1);
596     temp1 = _mm_packs_epi32(resq_r2, resq_r3);
597 
598     _mm_storeu_si128((__m128i *) (&pi2_tmp_ptr[0]), temp0);
599     _mm_storeu_si128((__m128i *) (&pi2_tmp_ptr[2 * 4]), temp1);
600 
601     /* Load pred buffer */
602     pred_r0 = _mm_loadl_epi64((__m128i *) (&pu1_pred[0]));
603     pred_r1 = _mm_loadl_epi64((__m128i *) (&pu1_pred[i4_pred_stride]));
604     pred_r2 = _mm_loadl_epi64((__m128i *) (&pu1_pred[2 * i4_pred_stride]));
605     pred_r3 = _mm_loadl_epi64((__m128i *) (&pu1_pred[3 * i4_pred_stride]));
606 
607     pred_r0 = _mm_cvtepu8_epi16(pred_r0);
608     pred_r1 = _mm_cvtepu8_epi16(pred_r1);
609     pred_r2 = _mm_cvtepu8_epi16(pred_r2);
610     pred_r3 = _mm_cvtepu8_epi16(pred_r3);
611 
612     /*--------------------------------------------------------------*/
613     /* IDCT [ Vertical transformation] and Xij = (xij + 32)>>6      */
614     /*                                                              */
615     /* Add the prediction and store it back to same buffer          */
616     /*--------------------------------------------------------------*/
617     /* z0j = y0j + y2j                                                        */
618     temp0 = _mm_add_epi32(resq_r0, resq_r2);
619     /* z1j = y0j - y2j                                                        */
620     temp1 = _mm_sub_epi32(resq_r0, resq_r2);
621     /* z2j = (y1j>>1) - y3j */
622     temp2 = _mm_srai_epi32(resq_r1, 1);
623     temp2 = _mm_sub_epi32(temp2, resq_r3);
624     /* z3j = y1j + (y3j>>1) */
625     temp3 = _mm_srai_epi32(resq_r3, 1);
626     temp3 = _mm_add_epi32(temp3, resq_r1);
627 
628     /* x0j = z0j + z3j                                                        */
629     temp4 = _mm_add_epi32(temp0, temp3);
630     temp4 = _mm_add_epi32(temp4, value_32);
631     temp4 = _mm_srai_epi32(temp4, 6);
632     /* x1j = z1j + z2j                                                        */
633     temp5 = _mm_add_epi32(temp1, temp2);
634     temp5 = _mm_add_epi32(temp5, value_32);
635     temp5 = _mm_srai_epi32(temp5, 6);
636     /* x2j = z1j - z2j                                                        */
637     temp6 = _mm_sub_epi32(temp1, temp2);
638     temp6 = _mm_add_epi32(temp6, value_32);
639     temp6 = _mm_srai_epi32(temp6, 6);
640     /* x3j = z0j - z3j                                                        */
641     temp7 = _mm_sub_epi32(temp0, temp3);
642     temp7 = _mm_add_epi32(temp7, value_32);
643     temp7 = _mm_srai_epi32(temp7, 6);
644 
645     /* 32-bit to 16-bit conversion */
646     temp0 = _mm_packs_epi32(temp4, temp5);
647     temp1 = _mm_packs_epi32(temp6, temp7);
648 
649     /* Saturate all values < -255 to -255 and retain the rest as it is */
650     temp0 = _mm_max_epi16(temp0, neg_255_8x16b);
651     /* Saturate all values > 255 to 255 and retain the rest as it is */
652     temp0 = _mm_min_epi16(temp0, pos_255_8x16b);
653 
654     /* Saturate all values < -255 to -255 and retain the rest as it is */
655     temp1 = _mm_max_epi16(temp1, neg_255_8x16b);
656     /* Saturate all values > 255 to 255 and retain the rest as it is */
657     temp1 = _mm_min_epi16(temp1, pos_255_8x16b);
658 
659     _mm_storel_epi64((__m128i *) (&pi2_res[0]), temp0);
660     _mm_storel_epi64((__m128i *) (&pi2_res[2 * i4_res_stride]), temp1);
661 
662     temp4 = _mm_add_epi16(temp0, pred_r0);
663     temp0 = _mm_srli_si128(temp0, 8);
664     _mm_storel_epi64((__m128i *) (&pi2_res[i4_res_stride]), temp0);
665 
666     temp6 = _mm_add_epi16(temp1, pred_r2);
667     temp1 = _mm_srli_si128(temp1, 8);
668     _mm_storel_epi64((__m128i *) (&pi2_res[3 * i4_res_stride]), temp1);
669 
670     temp5 = _mm_add_epi16(temp0, pred_r1);
671     temp7 = _mm_add_epi16(temp1, pred_r3);
672 
673     temp4 = _mm_cvtepi16_epi32(temp4);
674     temp5 = _mm_cvtepi16_epi32(temp5);
675     temp6 = _mm_cvtepi16_epi32(temp6);
676     temp7 = _mm_cvtepi16_epi32(temp7);
677 
678     /* 32-bit to 16-bit conversion */
679     temp0 = _mm_packs_epi32(temp4, temp5);
680     temp1 = _mm_packs_epi32(temp6, temp7);
681     /*------------------------------------------------------------------*/
682     /* Clipping the results to 8 bits */
683     sign_reg = _mm_cmpgt_epi16(temp0, zero_8x16b);
684     temp0 = _mm_and_si128(temp0, sign_reg);
685     sign_reg = _mm_cmpgt_epi16(temp1, zero_8x16b);
686     temp1 = _mm_and_si128(temp1, sign_reg);
687 
688     resq_r0 = _mm_packus_epi16(temp0, temp1);
689     resq_r1 = _mm_srli_si128(resq_r0, 4);
690     resq_r2 = _mm_srli_si128(resq_r1, 4);
691     resq_r3 = _mm_srli_si128(resq_r2, 4);
692 
693     *pu4_out = _mm_cvtsi128_si32(resq_r0);
694     pu1_out += i4_out_stride;
695     pu4_out = (UWORD32 *) (pu1_out);
696     *(pu4_out) = _mm_cvtsi128_si32(resq_r1);
697     pu1_out += i4_out_stride;
698     pu4_out = (UWORD32 *) (pu1_out);
699     *(pu4_out) = _mm_cvtsi128_si32(resq_r2);
700     pu1_out += i4_out_stride;
701     pu4_out = (UWORD32 *) (pu1_out);
702     *(pu4_out) = _mm_cvtsi128_si32(resq_r3);
703 }
704 
isvc_iquant_itrans_recon_res_4x4_with_res_acc_sse42(buffer_container_t * ps_src,buffer_container_t * ps_pred,buffer_container_t * ps_res_pred,buffer_container_t * ps_res,buffer_container_t * ps_rec,iq_it_res_rec_constants_t * ps_iq_it_res_rec_constants,WORD16 * pi2_tmp,WORD16 * pi2_dc_src,WORD32 i4_iq_start_idx,UWORD8 u1_res_accumulate)705 void isvc_iquant_itrans_recon_res_4x4_with_res_acc_sse42(
706     buffer_container_t *ps_src, buffer_container_t *ps_pred, buffer_container_t *ps_res_pred,
707     buffer_container_t *ps_res, buffer_container_t *ps_rec,
708     iq_it_res_rec_constants_t *ps_iq_it_res_rec_constants, WORD16 *pi2_tmp, WORD16 *pi2_dc_src,
709     WORD32 i4_iq_start_idx, UWORD8 u1_res_accumulate)
710 {
711     WORD16 *pi2_src = (WORD16 *) ps_src->pv_data;
712     WORD16 *pi2_tmp_ptr = pi2_tmp;
713     WORD16 *pi2_res = (WORD16 *) ps_res->pv_data;
714     WORD16 *pi2_res_pred = (WORD16 *) ps_res_pred->pv_data;
715     UWORD8 *pu1_pred = (UWORD8 *) ps_pred->pv_data;
716     UWORD8 *pu1_out = (UWORD8 *) ps_rec->pv_data;
717     WORD32 i4_src_stride = ps_src->i4_data_stride;
718     WORD32 i4_res_stride = ps_res->i4_data_stride;
719     WORD32 i4_res_pred_stride = ps_res_pred->i4_data_stride;
720     WORD32 i4_pred_stride = ps_pred->i4_data_stride;
721     WORD32 i4_out_stride = ps_rec->i4_data_stride;
722     const UWORD16 *pu2_iscal_mat = ps_iq_it_res_rec_constants->pu2_iscal_mat;
723     const UWORD16 *pu2_weigh_mat = ps_iq_it_res_rec_constants->pu2_weigh_mat;
724     UWORD32 u4_qp_div_6 = ps_iq_it_res_rec_constants->u4_qp_div_6;
725     UWORD32 *pu4_out = (UWORD32 *) pu1_out;
726     __m128i src_r0_r1, src_r2_r3;
727     __m128i src_r0, src_r1, src_r2, src_r3;
728     __m128i scalemat_r0_r1, scalemat_r2_r3;
729     __m128i pred_r0, pred_r1, pred_r2, pred_r3;
730     __m128i res_pred_r0, res_pred_r1, res_pred_r2, res_pred_r3;
731     __m128i res_r0, res_r1, res_r2, res_r3;
732     __m128i sign_reg, dequant_r0_r1, dequant_r2_r3;
733     /* all bits reset to zero */
734     __m128i zero_8x16b = _mm_setzero_si128();
735     __m128i neg_255_8x16b = _mm_set1_epi16(-((WORD16) UINT8_MAX));
736     __m128i pos_255_8x16b = _mm_set1_epi16(((WORD16) UINT8_MAX));
737     __m128i temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
738     __m128i resq_r0, resq_r1, resq_r2, resq_r3;
739     __m128i add_rshift = _mm_set1_epi32((u4_qp_div_6 < 4) ? (1 << (3 - u4_qp_div_6)) : 0);
740     __m128i value_32 = _mm_set1_epi32(32);
741 
742     ASSERT(4 == i4_src_stride);
743     ASSERT(1 == u1_res_accumulate);
744 
745     UNUSED(i4_src_stride);
746     UNUSED(ps_res_pred);
747     UNUSED(u1_res_accumulate);
748 
749     /*************************************************************/
750     /* Dequantization of coefficients. Will be replaced by SIMD  */
751     /* operations on platform                                    */
752     /*************************************************************/
753 
754     /* a00 a01 a02 a03 a10 a11 a12 a13 -- the source
755      matrix 0th,1st row */
756     src_r0_r1 = _mm_loadu_si128((__m128i *) (pi2_src));
757 
758     /* a20 a21 a22 a23 a30 a31 a32 a33 -- the
759       source matrix 2nd,3rd row */
760     src_r2_r3 = _mm_loadu_si128((__m128i *) (pi2_src + 8));
761 
762     /* b00 b01 b02 b03 b10 b11 b12 b13 -- the
763      scaling matrix 0th,1st row */
764     scalemat_r0_r1 = _mm_loadu_si128((__m128i *) (pu2_iscal_mat));
765 
766     /* b20 b21 b22 b23 b30 b31 b32 b33 --b12 b13 -- the
767      the scaling matrix 2nd,3rd row */
768     scalemat_r2_r3 = _mm_loadu_si128((__m128i *) (pu2_iscal_mat + 8));
769 
770     /* q00 q01 q02 q03 q10 q11
771      q12 q13 -- all 16 bits */
772     dequant_r0_r1 = _mm_loadu_si128((__m128i *) (pu2_weigh_mat));
773 
774     /* q20 q21 q22 q23 q30 q31
775      q32 q33 -- all 16 bits */
776     dequant_r2_r3 = _mm_loadu_si128((__m128i *) (pu2_weigh_mat + 8));
777 
778     /* b00*q00 b01*q01 b02*q02 b03*q03 b10*q10 b11*q11
779      b12*q12 b13*q13 -- 16 bit result */
780     temp0 = _mm_mullo_epi16(scalemat_r0_r1, dequant_r0_r1);
781 
782     /* b20*q20 b21*q21 b22*q22 b23*q23 b30*q30 b31*q31
783      b32*q32 b33*q33 -- 16 bit result */
784     temp1 = _mm_mullo_epi16(scalemat_r2_r3, dequant_r2_r3);
785 
786     /* b00*q00 0 b01*q01 0 b02*q02 0 b03*q03 0 -- 16 bit long */
787     temp4 = _mm_unpacklo_epi16(temp0, zero_8x16b);
788 
789     /* b10*q10 0 b11*q11 0 b12*q12 0 b13*q13 0 -- 16 bit long */
790     temp5 = _mm_unpackhi_epi16(temp0, zero_8x16b);
791 
792     /* b00*q00 0 b01*q01 0 b02*q02 0 b03*q03 0 -- 16 bit long */
793     temp6 = _mm_unpacklo_epi16(temp1, zero_8x16b);
794 
795     /* b10*q10 0 b11*q11 0 b12*q12 0 b13*q13 0 -- 16 bit long */
796     temp7 = _mm_unpackhi_epi16(temp1, zero_8x16b);
797 
798     /* a00 0 a01 0 a02 0 a03 0 -- 16 bit long */
799     src_r0 = _mm_unpacklo_epi16(src_r0_r1, zero_8x16b);
800     /* a10 0 a11 0 a12 0 a13 0 -- 16 bit long */
801     src_r1 = _mm_unpackhi_epi16(src_r0_r1, zero_8x16b);
802     /* a20 0 a21 0 a22 0 a23 0 -- 16 bit long */
803     src_r2 = _mm_unpacklo_epi16(src_r2_r3, zero_8x16b);
804     /* a30 0 a31 0 a32 0 a33 0 -- 16 bit long */
805     src_r3 = _mm_unpackhi_epi16(src_r2_r3, zero_8x16b);
806 
807     temp4 = _mm_madd_epi16(src_r0, temp4);
808     temp5 = _mm_madd_epi16(src_r1, temp5);
809     temp6 = _mm_madd_epi16(src_r2, temp6);
810     temp7 = _mm_madd_epi16(src_r3, temp7);
811 
812     if(u4_qp_div_6 >= 4)
813     {
814         resq_r0 = _mm_slli_epi32(temp4, u4_qp_div_6 - 4);
815         resq_r1 = _mm_slli_epi32(temp5, u4_qp_div_6 - 4);
816         resq_r2 = _mm_slli_epi32(temp6, u4_qp_div_6 - 4);
817         resq_r3 = _mm_slli_epi32(temp7, u4_qp_div_6 - 4);
818     }
819     else
820     {
821         temp4 = _mm_add_epi32(temp4, add_rshift);
822         temp5 = _mm_add_epi32(temp5, add_rshift);
823         temp6 = _mm_add_epi32(temp6, add_rshift);
824         temp7 = _mm_add_epi32(temp7, add_rshift);
825         resq_r0 = _mm_srai_epi32(temp4, 4 - u4_qp_div_6);
826         resq_r1 = _mm_srai_epi32(temp5, 4 - u4_qp_div_6);
827         resq_r2 = _mm_srai_epi32(temp6, 4 - u4_qp_div_6);
828         resq_r3 = _mm_srai_epi32(temp7, 4 - u4_qp_div_6);
829     }
830 
831     if(i4_iq_start_idx == 1) resq_r0 = _mm_insert_epi32(resq_r0, (WORD32) pi2_dc_src[0], 0);
832     /* Perform Inverse transform */
833     /*-------------------------------------------------------------*/
834     /* IDCT [ Horizontal transformation ]                          */
835     /*-------------------------------------------------------------*/
836     // Matrix transpose
837     /*
838      *  a0 a1 a2 a3
839      *  b0 b1 b2 b3
840      *  c0 c1 c2 c3
841      *  d0 d1 d2 d3
842      */
843 
844     /* a0 b0 a1 b1 */
845     temp1 = _mm_unpacklo_epi32(resq_r0, resq_r1);
846     /* c0 d0 c1 d1 */
847     temp3 = _mm_unpacklo_epi32(resq_r2, resq_r3);
848     /* a2 b2 a3 b3 */
849     temp2 = _mm_unpackhi_epi32(resq_r0, resq_r1);
850     /* c2 d2 c3 d3 */
851     temp4 = _mm_unpackhi_epi32(resq_r2, resq_r3);
852     /* a0 b0 c0 d0 */
853     resq_r0 = _mm_unpacklo_epi64(temp1, temp3);
854     /* a1 b1 c1 d1 */
855     resq_r1 = _mm_unpackhi_epi64(temp1, temp3);
856     /* a2 b2 c2 d2 */
857     resq_r2 = _mm_unpacklo_epi64(temp2, temp4);
858     /* a3 b3 c3 d3 */
859     resq_r3 = _mm_unpackhi_epi64(temp2, temp4);
860     /* Transform starts -- horizontal transform */
861     /*------------------------------------------------------------------*/
862     /* z0 = w0 + w2                                             */
863     temp0 = _mm_add_epi32(resq_r0, resq_r2);
864     /* z1 = w0 - w2                                             */
865     temp1 = _mm_sub_epi32(resq_r0, resq_r2);
866     /* z2 = (w1 >> 1) - w3                                      */
867     temp2 = _mm_srai_epi32(resq_r1, 1);
868     temp2 = _mm_sub_epi32(temp2, resq_r3);
869     /* z3 = w1 + (w3 >> 1)                                      */
870     temp3 = _mm_srai_epi32(resq_r3, 1);
871     temp3 = _mm_add_epi32(temp3, resq_r1);
872     /*----------------------------------------------------------*/
873     /* x0 = z0 + z3                                             */
874     resq_r0 = _mm_add_epi32(temp0, temp3);
875     /* x1 = z1 + z2                                             */
876     resq_r1 = _mm_add_epi32(temp1, temp2);
877     /* x2 = z1 - z2                                             */
878     resq_r2 = _mm_sub_epi32(temp1, temp2);
879     /* x3 = z0 - z3                                             */
880     resq_r3 = _mm_sub_epi32(temp0, temp3);
881 
882     // Matrix transpose
883     /*
884      *  a0 b0 c0 d0
885      *  a1 b1 c1 d1
886      *  a2 b2 c2 d2
887      *  a3 b3 c3 d3
888      */
889 
890     /* a0 a1 b0 b1 */
891     temp1 = _mm_unpacklo_epi32(resq_r0, resq_r1);
892     /* a2 a3 b2 b3 */
893     temp3 = _mm_unpacklo_epi32(resq_r2, resq_r3);
894     /* c0 c1 d0 d1 */
895     temp2 = _mm_unpackhi_epi32(resq_r0, resq_r1);
896     /* c2 c3 d2 d3 */
897     temp4 = _mm_unpackhi_epi32(resq_r2, resq_r3);
898     /* a0 a1 a2 a3 */
899     resq_r0 = _mm_unpacklo_epi64(temp1, temp3);
900     /* b0 b1 b2 b3 */
901     resq_r1 = _mm_unpackhi_epi64(temp1, temp3);
902     /* c0 c1 c2 c3 */
903     resq_r2 = _mm_unpacklo_epi64(temp2, temp4);
904     /* d0 d1 d2 d3 */
905     resq_r3 = _mm_unpackhi_epi64(temp2, temp4);
906     /* Transform ends -- horizontal transform */
907 
908     temp0 = _mm_packs_epi32(resq_r0, resq_r1);
909     temp1 = _mm_packs_epi32(resq_r2, resq_r3);
910 
911     _mm_storeu_si128((__m128i *) (&pi2_tmp_ptr[0]), temp0);
912     _mm_storeu_si128((__m128i *) (&pi2_tmp_ptr[2 * 4]), temp1);
913 
914     /* Load pred buffer */
915     pred_r0 = _mm_loadl_epi64((__m128i *) (&pu1_pred[0]));
916     pred_r1 = _mm_loadl_epi64((__m128i *) (&pu1_pred[i4_pred_stride]));
917     pred_r2 = _mm_loadl_epi64((__m128i *) (&pu1_pred[2 * i4_pred_stride]));
918     pred_r3 = _mm_loadl_epi64((__m128i *) (&pu1_pred[3 * i4_pred_stride]));
919 
920     pred_r0 = _mm_cvtepu8_epi16(pred_r0);
921     pred_r1 = _mm_cvtepu8_epi16(pred_r1);
922     pred_r2 = _mm_cvtepu8_epi16(pred_r2);
923     pred_r3 = _mm_cvtepu8_epi16(pred_r3);
924 
925     /*--------------------------------------------------------------*/
926     /* IDCT [ Vertical transformation] and Xij = (xij + 32)>>6      */
927     /*                                                              */
928     /* Add the prediction and store it back to same buffer          */
929     /*--------------------------------------------------------------*/
930     /* z0j = y0j + y2j                                                        */
931     temp0 = _mm_add_epi32(resq_r0, resq_r2);
932     /* z1j = y0j - y2j                                                        */
933     temp1 = _mm_sub_epi32(resq_r0, resq_r2);
934     /* z2j = (y1j>>1) - y3j */
935     temp2 = _mm_srai_epi32(resq_r1, 1);
936     temp2 = _mm_sub_epi32(temp2, resq_r3);
937     /* z3j = y1j + (y3j>>1) */
938     temp3 = _mm_srai_epi32(resq_r3, 1);
939     temp3 = _mm_add_epi32(temp3, resq_r1);
940 
941     /* x0j = z0j + z3j                                                        */
942     temp4 = _mm_add_epi32(temp0, temp3);
943     temp4 = _mm_add_epi32(temp4, value_32);
944     temp4 = _mm_srai_epi32(temp4, 6);
945     res_r0 = temp4;
946     /* x1j = z1j + z2j                                                        */
947     temp5 = _mm_add_epi32(temp1, temp2);
948     temp5 = _mm_add_epi32(temp5, value_32);
949     temp5 = _mm_srai_epi32(temp5, 6);
950     res_r1 = temp5;
951     /* x2j = z1j - z2j                                                        */
952     temp6 = _mm_sub_epi32(temp1, temp2);
953     temp6 = _mm_add_epi32(temp6, value_32);
954     temp6 = _mm_srai_epi32(temp6, 6);
955     res_r2 = temp6;
956     /* x3j = z0j - z3j                                                        */
957     temp7 = _mm_sub_epi32(temp0, temp3);
958     temp7 = _mm_add_epi32(temp7, value_32);
959     temp7 = _mm_srai_epi32(temp7, 6);
960     res_r3 = temp7;
961 
962     /* Accumulating res */
963     res_pred_r0 = _mm_loadl_epi64((__m128i *) &pi2_res_pred[0]);
964     res_pred_r1 = _mm_loadl_epi64((__m128i *) &pi2_res_pred[i4_res_pred_stride]);
965     res_pred_r2 = _mm_loadl_epi64((__m128i *) &pi2_res_pred[2 * i4_res_pred_stride]);
966     res_pred_r3 = _mm_loadl_epi64((__m128i *) &pi2_res_pred[3 * i4_res_pred_stride]);
967 
968     res_pred_r0 = _mm_cvtepi16_epi32(res_pred_r0);
969     res_pred_r1 = _mm_cvtepi16_epi32(res_pred_r1);
970     res_pred_r2 = _mm_cvtepi16_epi32(res_pred_r2);
971     res_pred_r3 = _mm_cvtepi16_epi32(res_pred_r3);
972 
973     temp0 = _mm_add_epi32(res_r0, res_pred_r0);
974     temp1 = _mm_add_epi32(res_r1, res_pred_r1);
975     temp2 = _mm_add_epi32(res_r2, res_pred_r2);
976     temp3 = _mm_add_epi32(res_r3, res_pred_r3);
977 
978     temp0 = _mm_packs_epi32(temp0, temp1);
979     temp1 = _mm_packs_epi32(temp2, temp3);
980 
981     /* Saturate all values < -255 to -255 and retain the rest as it is */
982     temp0 = _mm_max_epi16(temp0, neg_255_8x16b);
983     /* Saturate all values > 255 to 255 and retain the rest as it is */
984     temp0 = _mm_min_epi16(temp0, pos_255_8x16b);
985 
986     /* Saturate all values < -255 to -255 and retain the rest as it is */
987     temp1 = _mm_max_epi16(temp1, neg_255_8x16b);
988     /* Saturate all values > 255 to 255 and retain the rest as it is */
989     temp1 = _mm_min_epi16(temp1, pos_255_8x16b);
990 
991     _mm_storel_epi64((__m128i *) (&pi2_res[0]), temp0);
992     _mm_storel_epi64((__m128i *) (&pi2_res[2 * i4_res_stride]), temp1);
993 
994     temp4 = _mm_add_epi16(temp0, pred_r0);
995     temp0 = _mm_srli_si128(temp0, 8);
996     _mm_storel_epi64((__m128i *) (&pi2_res[i4_res_stride]), temp0);
997 
998     temp6 = _mm_add_epi16(temp1, pred_r2);
999     temp1 = _mm_srli_si128(temp1, 8);
1000     _mm_storel_epi64((__m128i *) (&pi2_res[3 * i4_res_stride]), temp1);
1001 
1002     temp5 = _mm_add_epi16(temp0, pred_r1);
1003     temp7 = _mm_add_epi16(temp1, pred_r3);
1004 
1005     temp4 = _mm_cvtepi16_epi32(temp4);
1006     temp5 = _mm_cvtepi16_epi32(temp5);
1007     temp6 = _mm_cvtepi16_epi32(temp6);
1008     temp7 = _mm_cvtepi16_epi32(temp7);
1009 
1010     /* 32-bit to 16-bit conversion */
1011     temp0 = _mm_packs_epi32(temp4, temp5);
1012     temp1 = _mm_packs_epi32(temp6, temp7);
1013     /*------------------------------------------------------------------*/
1014     /* Clipping the results to 8 bits */
1015     sign_reg = _mm_cmpgt_epi16(temp0, zero_8x16b);
1016     temp0 = _mm_and_si128(temp0, sign_reg);
1017     sign_reg = _mm_cmpgt_epi16(temp1, zero_8x16b);
1018     temp1 = _mm_and_si128(temp1, sign_reg);
1019 
1020     resq_r0 = _mm_packus_epi16(temp0, temp1);
1021     resq_r1 = _mm_srli_si128(resq_r0, 4);
1022     resq_r2 = _mm_srli_si128(resq_r1, 4);
1023     resq_r3 = _mm_srli_si128(resq_r2, 4);
1024 
1025     *pu4_out = _mm_cvtsi128_si32(resq_r0);
1026     pu1_out += i4_out_stride;
1027     pu4_out = (UWORD32 *) (pu1_out);
1028     *(pu4_out) = _mm_cvtsi128_si32(resq_r1);
1029     pu1_out += i4_out_stride;
1030     pu4_out = (UWORD32 *) (pu1_out);
1031     *(pu4_out) = _mm_cvtsi128_si32(resq_r2);
1032     pu1_out += i4_out_stride;
1033     pu4_out = (UWORD32 *) (pu1_out);
1034     *(pu4_out) = _mm_cvtsi128_si32(resq_r3);
1035 }
1036 
isvc_iquant_itrans_recon_res_chroma_4x4_sse42(buffer_container_t * ps_src,buffer_container_t * ps_pred,buffer_container_t * ps_res_pred,buffer_container_t * ps_res,buffer_container_t * ps_rec,iq_it_res_rec_constants_t * ps_iq_it_res_rec_constants,WORD16 * pi2_tmp,WORD16 * pi2_dc_src,WORD32 i4_iq_start_idx,UWORD8 u1_res_accumulate)1037 void isvc_iquant_itrans_recon_res_chroma_4x4_sse42(
1038     buffer_container_t *ps_src, buffer_container_t *ps_pred, buffer_container_t *ps_res_pred,
1039     buffer_container_t *ps_res, buffer_container_t *ps_rec,
1040     iq_it_res_rec_constants_t *ps_iq_it_res_rec_constants, WORD16 *pi2_tmp, WORD16 *pi2_dc_src,
1041     WORD32 i4_iq_start_idx, UWORD8 u1_res_accumulate)
1042 {
1043     WORD16 *pi2_src = (WORD16 *) ps_src->pv_data;
1044     WORD16 *pi2_res = (WORD16 *) ps_res->pv_data;
1045     WORD16 *pi2_res_ptr = pi2_res;
1046     UWORD8 *pu1_pred = (UWORD8 *) ps_pred->pv_data;
1047     UWORD8 *pu1_out = (UWORD8 *) ps_rec->pv_data;
1048     WORD32 i4_src_stride = ps_src->i4_data_stride;
1049     WORD32 i4_res_stride = ps_res->i4_data_stride;
1050     WORD32 i4_pred_stride = ps_pred->i4_data_stride;
1051     WORD32 i4_out_stride = ps_rec->i4_data_stride;
1052     const UWORD16 *pu2_iscal_mat = ps_iq_it_res_rec_constants->pu2_iscal_mat;
1053     const UWORD16 *pu2_weigh_mat = ps_iq_it_res_rec_constants->pu2_weigh_mat;
1054     UWORD32 u4_qp_div_6 = ps_iq_it_res_rec_constants->u4_qp_div_6;
1055     __m128i src_r0_r1, src_r2_r3;
1056     __m128i src_r0, src_r1, src_r2, src_r3;
1057     __m128i scalemat_r0_r1, scalemat_r2_r3;
1058     __m128i pred_r0, pred_r1, pred_r2, pred_r3;
1059     __m128i sign_reg, dequant_r0_r1, dequant_r2_r3;
1060     /* all bits reset to zero */
1061     __m128i zero_8x16b = _mm_setzero_si128();
1062     __m128i neg_255_8x16b = _mm_set1_epi16(-((WORD16) UINT8_MAX));
1063     __m128i pos_255_8x16b = _mm_set1_epi16(((WORD16) UINT8_MAX));
1064     __m128i temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
1065     __m128i resq_r0, resq_r1, resq_r2, resq_r3;
1066     __m128i add_rshift = _mm_set1_epi32((u4_qp_div_6 < 4) ? (1 << (3 - u4_qp_div_6)) : 0);
1067     __m128i value_32 = _mm_set1_epi32(32);
1068     __m128i chroma_mask = _mm_set1_epi16(0xFF);
1069     __m128i out_r0, out_r1, out_r2, out_r3;
1070     __m128i res_r0, res_r1, res_r2, res_r3;
1071 
1072     ASSERT(4 == i4_src_stride);
1073     ASSERT(0 == u1_res_accumulate);
1074 
1075     UNUSED(i4_src_stride);
1076     UNUSED(u1_res_accumulate);
1077     UNUSED(ps_res_pred);
1078     UNUSED(i4_iq_start_idx);
1079 
1080     /*************************************************************/
1081     /* Dequantization of coefficients. Will be replaced by SIMD  */
1082     /* operations on platform                                    */
1083     /*************************************************************/
1084     /* a00 a01 a02 a03 a10 a11 a12 a13 -- the source
1085     matrix 0th,1st row */
1086     src_r0_r1 = _mm_loadu_si128((__m128i *) (pi2_src));
1087 
1088     /* a20 a21 a22 a23 a30 a31 a32 a33 -- the
1089     source matrix 2nd,3rd row */
1090     src_r2_r3 = _mm_loadu_si128((__m128i *) (pi2_src + 8));
1091 
1092     /* b00 b01 b02 b03 b10 b11 b12 b13 -- the
1093     scaling matrix 0th,1st row */
1094     scalemat_r0_r1 = _mm_loadu_si128((__m128i *) (pu2_iscal_mat));
1095 
1096     /* b20 b21 b22 b23 b30 b31 b32 b33 --b12 b13 -- the
1097     the scaling matrix 2nd,3rd row */
1098     scalemat_r2_r3 = _mm_loadu_si128((__m128i *) (pu2_iscal_mat + 8));
1099 
1100     /* q00 q01 q02 q03 q10 q11
1101     q12 q13 -- all 16 bits */
1102     dequant_r0_r1 = _mm_loadu_si128((__m128i *) (pu2_weigh_mat));
1103 
1104     /* q20 q21 q22 q23 q30 q31
1105     q32 q33 -- all 16 bits */
1106     dequant_r2_r3 = _mm_loadu_si128((__m128i *) (pu2_weigh_mat + 8));
1107 
1108     temp0 = _mm_mullo_epi16(scalemat_r0_r1,
1109                             dequant_r0_r1);  // b00*q00 b01*q01 b02*q02 b03*q03 b10*q10 b11*q11
1110                                              // b12*q12 b13*q13 -- 16 bit result
1111 
1112     temp1 = _mm_mullo_epi16(scalemat_r2_r3, dequant_r2_r3);
1113 
1114     /* b00*q00 0 b01*q01 0 b02*q02 0 b03*q03 0 -- 16 bit long */
1115     temp4 = _mm_unpacklo_epi16(temp0, zero_8x16b);
1116 
1117     /* b10*q10 0 b11*q11 0 b12*q12 0 b13*q13 0 -- 16 bit long */
1118     temp5 = _mm_unpackhi_epi16(temp0, zero_8x16b);
1119 
1120     /* b00*q00 0 b01*q01 0 b02*q02 0 b03*q03 0 -- 16 bit long */
1121     temp6 = _mm_unpacklo_epi16(temp1, zero_8x16b);
1122 
1123     /* b10*q10 0 b11*q11 0 b12*q12 0 b13*q13 0 -- 16 bit long */
1124     temp7 = _mm_unpackhi_epi16(temp1, zero_8x16b);
1125 
1126     /* a00 0 a01 0 a02 0 a03 0 -- 16 bit long */
1127     src_r0 = _mm_unpacklo_epi16(src_r0_r1, zero_8x16b);
1128     /* a10 0 a11 0 a12 0 a13 0 -- 16 bit long */
1129     src_r1 = _mm_unpackhi_epi16(src_r0_r1, zero_8x16b);
1130     /* a20 0 a21 0 a22 0 a23 0 -- 16 bit long */
1131     src_r2 = _mm_unpacklo_epi16(src_r2_r3, zero_8x16b);
1132     /* a30 0 a31 0 a32 0 a33 0 -- 16 bit long */
1133     src_r3 = _mm_unpackhi_epi16(src_r2_r3, zero_8x16b);
1134 
1135     temp4 = _mm_madd_epi16(src_r0, temp4);
1136     temp5 = _mm_madd_epi16(src_r1, temp5);
1137     temp6 = _mm_madd_epi16(src_r2, temp6);
1138     temp7 = _mm_madd_epi16(src_r3, temp7);
1139 
1140     if(u4_qp_div_6 >= 4)
1141     {
1142         resq_r0 = _mm_slli_epi32(temp4, u4_qp_div_6 - 4);
1143         resq_r1 = _mm_slli_epi32(temp5, u4_qp_div_6 - 4);
1144         resq_r2 = _mm_slli_epi32(temp6, u4_qp_div_6 - 4);
1145         resq_r3 = _mm_slli_epi32(temp7, u4_qp_div_6 - 4);
1146     }
1147     else
1148     {
1149         temp4 = _mm_add_epi32(temp4, add_rshift);
1150         temp5 = _mm_add_epi32(temp5, add_rshift);
1151         temp6 = _mm_add_epi32(temp6, add_rshift);
1152         temp7 = _mm_add_epi32(temp7, add_rshift);
1153         resq_r0 = _mm_srai_epi32(temp4, 4 - u4_qp_div_6);
1154         resq_r1 = _mm_srai_epi32(temp5, 4 - u4_qp_div_6);
1155         resq_r2 = _mm_srai_epi32(temp6, 4 - u4_qp_div_6);
1156         resq_r3 = _mm_srai_epi32(temp7, 4 - u4_qp_div_6);
1157     }
1158 
1159     resq_r0 = _mm_insert_epi32(resq_r0, (WORD32) pi2_dc_src[0], 0);
1160     /* Perform Inverse transform */
1161     /*-------------------------------------------------------------*/
1162     /* IDCT [ Horizontal transformation ]                          */
1163     /*-------------------------------------------------------------*/
1164     // Matrix transpose
1165     /*
1166      *  a0 a1 a2 a3
1167      *  b0 b1 b2 b3
1168      *  c0 c1 c2 c3
1169      *  d0 d1 d2 d3
1170      */
1171     /* a0 b0 a1 b1 */
1172     temp1 = _mm_unpacklo_epi32(resq_r0, resq_r1);
1173     /* c0 d0 c1 d1 */
1174     temp3 = _mm_unpacklo_epi32(resq_r2, resq_r3);
1175     /* a2 b2 a3 b3 */
1176     temp2 = _mm_unpackhi_epi32(resq_r0, resq_r1);
1177     /* c2 d2 c3 d3 */
1178     temp4 = _mm_unpackhi_epi32(resq_r2, resq_r3);
1179     /* a0 b0 c0 d0 */
1180     resq_r0 = _mm_unpacklo_epi64(temp1, temp3);
1181     /* a1 b1 c1 d1 */
1182     resq_r1 = _mm_unpackhi_epi64(temp1, temp3);
1183     /* a2 b2 c2 d2 */
1184     resq_r2 = _mm_unpacklo_epi64(temp2, temp4);
1185     /* a3 b3 c3 d3 */
1186     resq_r3 = _mm_unpackhi_epi64(temp2, temp4);
1187     /* Transform starts -- horizontal transform */
1188 
1189     /*------------------------------------------------------------------*/
1190     /* z0 = w0 + w2                                             */
1191     temp0 = _mm_add_epi32(resq_r0, resq_r2);
1192     /* z1 = w0 - w2                                             */
1193     temp1 = _mm_sub_epi32(resq_r0, resq_r2);
1194     /* z2 = (w1 >> 1) - w3                                      */
1195     temp2 = _mm_srai_epi32(resq_r1, 1);
1196     temp2 = _mm_sub_epi32(temp2, resq_r3);
1197     /* z3 = w1 + (w3 >> 1)                                      */
1198     temp3 = _mm_srai_epi32(resq_r3, 1);
1199     temp3 = _mm_add_epi32(temp3, resq_r1);
1200     /*----------------------------------------------------------*/
1201     /* x0 = z0 + z3                                             */
1202     resq_r0 = _mm_add_epi32(temp0, temp3);
1203     /* x1 = z1 + z2                                             */
1204     resq_r1 = _mm_add_epi32(temp1, temp2);
1205     /* x2 = z1 - z2                                             */
1206     resq_r2 = _mm_sub_epi32(temp1, temp2);
1207     /* x3 = z0 - z3                                             */
1208     resq_r3 = _mm_sub_epi32(temp0, temp3);
1209     // Matrix transpose
1210     /*
1211      *  a0 b0 c0 d0
1212      *  a1 b1 c1 d1
1213      *  a2 b2 c2 d2
1214      *  a3 b3 c3 d3
1215      */
1216     /* a0 a1 b0 b1 */
1217     temp1 = _mm_unpacklo_epi32(resq_r0, resq_r1);
1218     /* a2 a3 b2 b3 */
1219     temp3 = _mm_unpacklo_epi32(resq_r2, resq_r3);
1220     /* c0 c1 d0 d1 */
1221     temp2 = _mm_unpackhi_epi32(resq_r0, resq_r1);
1222     /* c2 c3 d2 d3 */
1223     temp4 = _mm_unpackhi_epi32(resq_r2, resq_r3);
1224     /* a0 a1 a2 a3 */
1225     resq_r0 = _mm_unpacklo_epi64(temp1, temp3);
1226     /* b0 b1 b2 b3 */
1227     resq_r1 = _mm_unpackhi_epi64(temp1, temp3);
1228     /* c0 c1 c2 c3 */
1229     resq_r2 = _mm_unpacklo_epi64(temp2, temp4);
1230     /* d0 d1 d2 d3 */
1231     resq_r3 = _mm_unpackhi_epi64(temp2, temp4);
1232     /* Transform ends -- horizontal transform */
1233 
1234     temp0 = _mm_packs_epi32(resq_r0, resq_r1);
1235     temp1 = _mm_packs_epi32(resq_r2, resq_r3);
1236 
1237     _mm_storeu_si128((__m128i *) (&pi2_tmp[0]), temp0);
1238     _mm_storeu_si128((__m128i *) (&pi2_tmp[2 * 4]), temp1);
1239 
1240     /* Load pred buffer */
1241     pred_r0 = _mm_loadl_epi64((__m128i *) (&pu1_pred[0]));
1242     pred_r1 = _mm_loadl_epi64((__m128i *) (&pu1_pred[i4_pred_stride]));
1243     pred_r2 = _mm_loadl_epi64((__m128i *) (&pu1_pred[2 * i4_pred_stride]));
1244     pred_r3 = _mm_loadl_epi64((__m128i *) (&pu1_pred[3 * i4_pred_stride]));
1245 
1246     pred_r0 = _mm_and_si128(pred_r0, chroma_mask);
1247     pred_r1 = _mm_and_si128(pred_r1, chroma_mask);
1248     pred_r2 = _mm_and_si128(pred_r2, chroma_mask);
1249     pred_r3 = _mm_and_si128(pred_r3, chroma_mask);
1250 
1251     pred_r0 = _mm_cvtepu16_epi32(pred_r0);
1252     pred_r1 = _mm_cvtepu16_epi32(pred_r1);
1253     pred_r2 = _mm_cvtepu16_epi32(pred_r2);
1254     pred_r3 = _mm_cvtepu16_epi32(pred_r3);
1255 
1256     /*--------------------------------------------------------------*/
1257     /* IDCT [ Vertical transformation] and Xij = (xij + 32)>>6      */
1258     /*                                                              */
1259     /* Add the prediction and store it back to same buffer          */
1260     /*--------------------------------------------------------------*/
1261     /* z0j = y0j + y2j                                         */
1262     temp0 = _mm_add_epi32(resq_r0, resq_r2);
1263     /* z1j = y0j - y2j                                                        */
1264     temp1 = _mm_sub_epi32(resq_r0, resq_r2);
1265     /* z2j = (y1j>>1) - y3j */
1266     temp2 = _mm_srai_epi32(resq_r1, 1);
1267     temp2 = _mm_sub_epi32(temp2, resq_r3);
1268     /* z3j = y1j + (y3j>>1) */
1269     temp3 = _mm_srai_epi32(resq_r3, 1);
1270     temp3 = _mm_add_epi32(temp3, resq_r1);
1271 
1272     /* x0j = z0j + z3j                                                        */
1273     temp4 = _mm_add_epi32(temp0, temp3);
1274     temp4 = _mm_add_epi32(temp4, value_32);
1275     temp4 = _mm_srai_epi32(temp4, 6);
1276     /* x1j = z1j + z2j                                                        */
1277     temp5 = _mm_add_epi32(temp1, temp2);
1278     temp5 = _mm_add_epi32(temp5, value_32);
1279     temp5 = _mm_srai_epi32(temp5, 6);
1280     /* x2j = z1j - z2j                                                        */
1281     temp6 = _mm_sub_epi32(temp1, temp2);
1282     temp6 = _mm_add_epi32(temp6, value_32);
1283     temp6 = _mm_srai_epi32(temp6, 6);
1284     /* x3j = z0j - z3j                                                        */
1285     temp7 = _mm_sub_epi32(temp0, temp3);
1286     temp7 = _mm_add_epi32(temp7, value_32);
1287     temp7 = _mm_srai_epi32(temp7, 6);
1288 
1289     /* 32-bit to 16-bit conversion */
1290     temp0 = _mm_packs_epi32(temp4, temp5);
1291     temp1 = _mm_packs_epi32(temp6, temp7);
1292 
1293     /* Saturate all values < -255 to -255 and retain the rest as it is */
1294     temp0 = _mm_max_epi16(temp0, neg_255_8x16b);
1295     /* Saturate all values > 255 to 255 and retain the rest as it is */
1296     temp0 = _mm_min_epi16(temp0, pos_255_8x16b);
1297 
1298     /* Saturate all values < -255 to -255 and retain the rest as it is */
1299     temp1 = _mm_max_epi16(temp1, neg_255_8x16b);
1300     /* Saturate all values > 255 to 255 and retain the rest as it is */
1301     temp1 = _mm_min_epi16(temp1, pos_255_8x16b);
1302 
1303     chroma_mask = _mm_set1_epi32(0xffff0000);
1304     out_r0 = _mm_loadu_si128((__m128i *) (&pi2_res_ptr[0 * i4_res_stride]));
1305     out_r1 = _mm_loadu_si128((__m128i *) (&pi2_res_ptr[1 * i4_res_stride]));
1306     out_r2 = _mm_loadu_si128((__m128i *) (&pi2_res_ptr[2 * i4_res_stride]));
1307     out_r3 = _mm_loadu_si128((__m128i *) (&pi2_res_ptr[3 * i4_res_stride]));
1308 
1309     out_r0 = _mm_and_si128(out_r0, chroma_mask);
1310     out_r1 = _mm_and_si128(out_r1, chroma_mask);
1311     out_r2 = _mm_and_si128(out_r2, chroma_mask);
1312     out_r3 = _mm_and_si128(out_r3, chroma_mask);
1313 
1314     res_r0 = _mm_cvtepu16_epi32(temp0);
1315     res_r2 = _mm_cvtepu16_epi32(temp1);
1316     res_r1 = _mm_srli_si128(temp0, 8);
1317     res_r3 = _mm_srli_si128(temp1, 8);
1318     res_r1 = _mm_cvtepu16_epi32(res_r1);
1319     res_r3 = _mm_cvtepu16_epi32(res_r3);
1320 
1321     out_r0 = _mm_add_epi16(out_r0, res_r0);
1322     out_r1 = _mm_add_epi16(out_r1, res_r1);
1323     out_r2 = _mm_add_epi16(out_r2, res_r2);
1324     out_r3 = _mm_add_epi16(out_r3, res_r3);
1325 
1326     _mm_storeu_si128((__m128i *) (&pi2_res_ptr[0 * i4_res_stride]), out_r0);
1327     _mm_storeu_si128((__m128i *) (&pi2_res_ptr[1 * i4_res_stride]), out_r1);
1328     _mm_storeu_si128((__m128i *) (&pi2_res_ptr[2 * i4_res_stride]), out_r2);
1329     _mm_storeu_si128((__m128i *) (&pi2_res_ptr[3 * i4_res_stride]), out_r3);
1330 
1331     resq_r0 = _mm_add_epi16(pred_r0, res_r0);
1332     resq_r1 = _mm_add_epi16(pred_r1, res_r1);
1333     resq_r2 = _mm_add_epi16(pred_r2, res_r2);
1334     resq_r3 = _mm_add_epi16(pred_r3, res_r3);
1335 
1336     temp0 = _mm_packus_epi32(resq_r0, resq_r1);
1337     temp1 = _mm_packus_epi32(resq_r2, resq_r3);
1338 
1339     /*------------------------------------------------------------------*/
1340     /* Clipping the results to 8 bits */
1341     sign_reg = _mm_cmpgt_epi16(temp0, zero_8x16b);
1342     temp0 = _mm_and_si128(temp0, sign_reg);
1343     sign_reg = _mm_cmpgt_epi16(temp1, zero_8x16b);
1344     temp1 = _mm_and_si128(temp1, sign_reg);
1345 
1346     resq_r0 = _mm_packus_epi16(temp0, temp1);
1347     resq_r1 = _mm_srli_si128(resq_r0, 4);
1348     resq_r2 = _mm_srli_si128(resq_r1, 4);
1349     resq_r3 = _mm_srli_si128(resq_r2, 4);
1350 
1351     resq_r0 = _mm_cvtepu8_epi16(resq_r0);
1352     resq_r1 = _mm_cvtepu8_epi16(resq_r1);
1353     resq_r2 = _mm_cvtepu8_epi16(resq_r2);
1354     resq_r3 = _mm_cvtepu8_epi16(resq_r3);
1355 
1356     chroma_mask = _mm_set1_epi16(0xff00);
1357     out_r0 = _mm_loadl_epi64((__m128i *) (&pu1_out[0]));
1358     out_r1 = _mm_loadl_epi64((__m128i *) (&pu1_out[i4_out_stride]));
1359     out_r2 = _mm_loadl_epi64((__m128i *) (&pu1_out[2 * i4_out_stride]));
1360     out_r3 = _mm_loadl_epi64((__m128i *) (&pu1_out[3 * i4_out_stride]));
1361 
1362     out_r0 = _mm_and_si128(out_r0, chroma_mask);
1363     out_r1 = _mm_and_si128(out_r1, chroma_mask);
1364     out_r2 = _mm_and_si128(out_r2, chroma_mask);
1365     out_r3 = _mm_and_si128(out_r3, chroma_mask);
1366 
1367     out_r0 = _mm_add_epi8(out_r0, resq_r0);
1368     out_r1 = _mm_add_epi8(out_r1, resq_r1);
1369     out_r2 = _mm_add_epi8(out_r2, resq_r2);
1370     out_r3 = _mm_add_epi8(out_r3, resq_r3);
1371 
1372     _mm_storel_epi64((__m128i *) (&pu1_out[0]), out_r0);
1373     _mm_storel_epi64((__m128i *) (&pu1_out[i4_out_stride]), out_r1);
1374     _mm_storel_epi64((__m128i *) (&pu1_out[2 * i4_out_stride]), out_r2);
1375     _mm_storel_epi64((__m128i *) (&pu1_out[3 * i4_out_stride]), out_r3);
1376 }
1377 
isvc_iquant_itrans_recon_res_chroma_4x4_with_res_acc_sse42(buffer_container_t * ps_src,buffer_container_t * ps_pred,buffer_container_t * ps_res_pred,buffer_container_t * ps_res,buffer_container_t * ps_rec,iq_it_res_rec_constants_t * ps_iq_it_res_rec_constants,WORD16 * pi2_tmp,WORD16 * pi2_dc_src,WORD32 i4_iq_start_idx,UWORD8 u1_res_accumulate)1378 void isvc_iquant_itrans_recon_res_chroma_4x4_with_res_acc_sse42(
1379     buffer_container_t *ps_src, buffer_container_t *ps_pred, buffer_container_t *ps_res_pred,
1380     buffer_container_t *ps_res, buffer_container_t *ps_rec,
1381     iq_it_res_rec_constants_t *ps_iq_it_res_rec_constants, WORD16 *pi2_tmp, WORD16 *pi2_dc_src,
1382     WORD32 i4_iq_start_idx, UWORD8 u1_res_accumulate)
1383 {
1384     WORD16 *pi2_src = (WORD16 *) ps_src->pv_data;
1385     WORD16 *pi2_res = (WORD16 *) ps_res->pv_data;
1386     WORD16 *pi2_res_pred = (WORD16 *) ps_res_pred->pv_data;
1387     UWORD8 *pu1_pred = (UWORD8 *) ps_pred->pv_data;
1388     UWORD8 *pu1_out = (UWORD8 *) ps_rec->pv_data;
1389     WORD32 i4_src_stride = ps_src->i4_data_stride;
1390     WORD32 i4_res_stride = ps_res->i4_data_stride;
1391     WORD32 i4_res_pred_stride = ps_res_pred->i4_data_stride;
1392     WORD32 i4_pred_stride = ps_pred->i4_data_stride;
1393     WORD32 i4_out_stride = ps_rec->i4_data_stride;
1394     const UWORD16 *pu2_iscal_mat = ps_iq_it_res_rec_constants->pu2_iscal_mat;
1395     const UWORD16 *pu2_weigh_mat = ps_iq_it_res_rec_constants->pu2_weigh_mat;
1396     UWORD32 u4_qp_div_6 = ps_iq_it_res_rec_constants->u4_qp_div_6;
1397     __m128i src_r0_r1, src_r2_r3;
1398     __m128i src_r0, src_r1, src_r2, src_r3;
1399     __m128i scalemat_r0_r1, scalemat_r2_r3;
1400     __m128i pred_r0, pred_r1, pred_r2, pred_r3;
1401     __m128i res_pred_r0, res_pred_r1, res_pred_r2, res_pred_r3;
1402     __m128i res_r0, res_r1, res_r2, res_r3;
1403     __m128i dequant_r0_r1, dequant_r2_r3;
1404     /* all bits reset to zero */
1405     __m128i zero_8x16b = _mm_setzero_si128();
1406     __m128i reg_chroma = _mm_set1_epi32(0xFFFF);
1407     __m128i neg_255_8x16b = _mm_set1_epi16(-((WORD16) UINT8_MAX));
1408     __m128i pos_255_8x16b = _mm_set1_epi16(((WORD16) UINT8_MAX));
1409     __m128i temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
1410     __m128i resq_r0, resq_r1, resq_r2, resq_r3;
1411     __m128i add_rshift = _mm_set1_epi32((u4_qp_div_6 < 4) ? (1 << (3 - u4_qp_div_6)) : 0);
1412     __m128i value_32 = _mm_set1_epi32(32);
1413     __m128i chroma_mask = _mm_set1_epi16(0xFF);
1414     __m128i out_r0, out_r1, out_r2, out_r3;
1415     __m128i mask_r0;
1416 
1417     ASSERT(4 == i4_src_stride);
1418     ASSERT(1 == u1_res_accumulate);
1419 
1420     UNUSED(i4_src_stride);
1421     UNUSED(u1_res_accumulate);
1422     UNUSED(i4_iq_start_idx);
1423 
1424     /*************************************************************/
1425     /* Dequantization of coefficients. Will be replaced by SIMD  */
1426     /* operations on platform                                    */
1427     /*************************************************************/
1428     /* a00 a01 a02 a03 a10 a11 a12 a13 -- the source
1429     matrix 0th,1st row */
1430     src_r0_r1 = _mm_loadu_si128((__m128i *) (pi2_src));
1431 
1432     /* a20 a21 a22 a23 a30 a31 a32 a33 -- the
1433     source matrix 2nd,3rd row */
1434     src_r2_r3 = _mm_loadu_si128((__m128i *) (pi2_src + 8));
1435 
1436     /* b00 b01 b02 b03 b10 b11 b12 b13 -- the
1437     scaling matrix 0th,1st row */
1438     scalemat_r0_r1 = _mm_loadu_si128((__m128i *) (pu2_iscal_mat));
1439 
1440     /* b20 b21 b22 b23 b30 b31 b32 b33 --b12 b13 -- the
1441     the scaling matrix 2nd,3rd row */
1442     scalemat_r2_r3 = _mm_loadu_si128((__m128i *) (pu2_iscal_mat + 8));
1443 
1444     /* q00 q01 q02 q03 q10 q11
1445     q12 q13 -- all 16 bits */
1446     dequant_r0_r1 = _mm_loadu_si128((__m128i *) (pu2_weigh_mat));
1447 
1448     /* q20 q21 q22 q23 q30 q31
1449     q32 q33 -- all 16 bits */
1450     dequant_r2_r3 = _mm_loadu_si128((__m128i *) (pu2_weigh_mat + 8));
1451 
1452     temp0 = _mm_mullo_epi16(scalemat_r0_r1,
1453                             dequant_r0_r1);  // b00*q00 b01*q01 b02*q02 b03*q03 b10*q10 b11*q11
1454                                              // b12*q12 b13*q13 -- 16 bit result
1455 
1456     temp1 = _mm_mullo_epi16(scalemat_r2_r3, dequant_r2_r3);
1457 
1458     /* b00*q00 0 b01*q01 0 b02*q02 0 b03*q03 0 -- 16 bit long */
1459     temp4 = _mm_unpacklo_epi16(temp0, zero_8x16b);
1460 
1461     /* b10*q10 0 b11*q11 0 b12*q12 0 b13*q13 0 -- 16 bit long */
1462     temp5 = _mm_unpackhi_epi16(temp0, zero_8x16b);
1463 
1464     /* b00*q00 0 b01*q01 0 b02*q02 0 b03*q03 0 -- 16 bit long */
1465     temp6 = _mm_unpacklo_epi16(temp1, zero_8x16b);
1466 
1467     /* b10*q10 0 b11*q11 0 b12*q12 0 b13*q13 0 -- 16 bit long */
1468     temp7 = _mm_unpackhi_epi16(temp1, zero_8x16b);
1469 
1470     /* a00 0 a01 0 a02 0 a03 0 -- 16 bit long */
1471     src_r0 = _mm_unpacklo_epi16(src_r0_r1, zero_8x16b);
1472     /* a10 0 a11 0 a12 0 a13 0 -- 16 bit long */
1473     src_r1 = _mm_unpackhi_epi16(src_r0_r1, zero_8x16b);
1474     /* a20 0 a21 0 a22 0 a23 0 -- 16 bit long */
1475     src_r2 = _mm_unpacklo_epi16(src_r2_r3, zero_8x16b);
1476     /* a30 0 a31 0 a32 0 a33 0 -- 16 bit long */
1477     src_r3 = _mm_unpackhi_epi16(src_r2_r3, zero_8x16b);
1478 
1479     temp4 = _mm_madd_epi16(src_r0, temp4);
1480     temp5 = _mm_madd_epi16(src_r1, temp5);
1481     temp6 = _mm_madd_epi16(src_r2, temp6);
1482     temp7 = _mm_madd_epi16(src_r3, temp7);
1483 
1484     if(u4_qp_div_6 >= 4)
1485     {
1486         resq_r0 = _mm_slli_epi32(temp4, u4_qp_div_6 - 4);
1487         resq_r1 = _mm_slli_epi32(temp5, u4_qp_div_6 - 4);
1488         resq_r2 = _mm_slli_epi32(temp6, u4_qp_div_6 - 4);
1489         resq_r3 = _mm_slli_epi32(temp7, u4_qp_div_6 - 4);
1490     }
1491     else
1492     {
1493         temp4 = _mm_add_epi32(temp4, add_rshift);
1494         temp5 = _mm_add_epi32(temp5, add_rshift);
1495         temp6 = _mm_add_epi32(temp6, add_rshift);
1496         temp7 = _mm_add_epi32(temp7, add_rshift);
1497         resq_r0 = _mm_srai_epi32(temp4, 4 - u4_qp_div_6);
1498         resq_r1 = _mm_srai_epi32(temp5, 4 - u4_qp_div_6);
1499         resq_r2 = _mm_srai_epi32(temp6, 4 - u4_qp_div_6);
1500         resq_r3 = _mm_srai_epi32(temp7, 4 - u4_qp_div_6);
1501     }
1502 
1503     resq_r0 = _mm_insert_epi32(resq_r0, (WORD32) pi2_dc_src[0], 0);
1504     /* Perform Inverse transform */
1505     /*-------------------------------------------------------------*/
1506     /* IDCT [ Horizontal transformation ]                          */
1507     /*-------------------------------------------------------------*/
1508     // Matrix transpose
1509     /*
1510      *  a0 a1 a2 a3
1511      *  b0 b1 b2 b3
1512      *  c0 c1 c2 c3
1513      *  d0 d1 d2 d3
1514      */
1515     /* a0 b0 a1 b1 */
1516     temp1 = _mm_unpacklo_epi32(resq_r0, resq_r1);
1517     /* c0 d0 c1 d1 */
1518     temp3 = _mm_unpacklo_epi32(resq_r2, resq_r3);
1519     /* a2 b2 a3 b3 */
1520     temp2 = _mm_unpackhi_epi32(resq_r0, resq_r1);
1521     /* c2 d2 c3 d3 */
1522     temp4 = _mm_unpackhi_epi32(resq_r2, resq_r3);
1523     /* a0 b0 c0 d0 */
1524     resq_r0 = _mm_unpacklo_epi64(temp1, temp3);
1525     /* a1 b1 c1 d1 */
1526     resq_r1 = _mm_unpackhi_epi64(temp1, temp3);
1527     /* a2 b2 c2 d2 */
1528     resq_r2 = _mm_unpacklo_epi64(temp2, temp4);
1529     /* a3 b3 c3 d3 */
1530     resq_r3 = _mm_unpackhi_epi64(temp2, temp4);
1531     /* Transform starts -- horizontal transform */
1532 
1533     /*------------------------------------------------------------------*/
1534     /* z0 = w0 + w2                                             */
1535     temp0 = _mm_add_epi32(resq_r0, resq_r2);
1536     /* z1 = w0 - w2                                             */
1537     temp1 = _mm_sub_epi32(resq_r0, resq_r2);
1538     /* z2 = (w1 >> 1) - w3                                      */
1539     temp2 = _mm_srai_epi32(resq_r1, 1);
1540     temp2 = _mm_sub_epi32(temp2, resq_r3);
1541     /* z3 = w1 + (w3 >> 1)                                      */
1542     temp3 = _mm_srai_epi32(resq_r3, 1);  //(w3>>1) + w1
1543     temp3 = _mm_add_epi32(temp3, resq_r1);
1544     /*----------------------------------------------------------*/
1545     /* x0 = z0 + z3                                             */
1546     resq_r0 = _mm_add_epi32(temp0, temp3);
1547     /* x1 = z1 + z2                                             */
1548     resq_r1 = _mm_add_epi32(temp1, temp2);
1549     /* x2 = z1 - z2                                             */
1550     resq_r2 = _mm_sub_epi32(temp1, temp2);
1551     /* x3 = z0 - z3                                             */
1552     resq_r3 = _mm_sub_epi32(temp0, temp3);
1553     // Matrix transpose
1554     /*
1555      *  a0 b0 c0 d0
1556      *  a1 b1 c1 d1
1557      *  a2 b2 c2 d2
1558      *  a3 b3 c3 d3
1559      */
1560     /* a0 a1 b0 b1 */
1561     temp1 = _mm_unpacklo_epi32(resq_r0, resq_r1);
1562     /* a2 a3 b2 b3 */
1563     temp3 = _mm_unpacklo_epi32(resq_r2, resq_r3);
1564     /* c0 c1 d0 d1 */
1565     temp2 = _mm_unpackhi_epi32(resq_r0, resq_r1);
1566     /* c2 c3 d2 d3 */
1567     temp4 = _mm_unpackhi_epi32(resq_r2, resq_r3);
1568     /* a0 a1 a2 a3 */
1569     resq_r0 = _mm_unpacklo_epi64(temp1, temp3);
1570     /* b0 b1 b2 b3 */
1571     resq_r1 = _mm_unpackhi_epi64(temp1, temp3);
1572     /* c0 c1 c2 c3 */
1573     resq_r2 = _mm_unpacklo_epi64(temp2, temp4);
1574     /* d0 d1 d2 d3 */
1575     resq_r3 = _mm_unpackhi_epi64(temp2, temp4);
1576     /* Transform ends -- horizontal transform */
1577 
1578     temp0 = _mm_packs_epi32(resq_r0, resq_r1);
1579     temp1 = _mm_packs_epi32(resq_r2, resq_r3);
1580 
1581     _mm_storeu_si128((__m128i *) (&pi2_tmp[0]), temp0);
1582     _mm_storeu_si128((__m128i *) (&pi2_tmp[2 * 4]), temp1);
1583 
1584     /* Load pred buffer */
1585     pred_r0 = _mm_loadl_epi64((__m128i *) (&pu1_pred[0]));
1586     pred_r1 = _mm_loadl_epi64((__m128i *) (&pu1_pred[i4_pred_stride]));
1587     pred_r2 = _mm_loadl_epi64((__m128i *) (&pu1_pred[2 * i4_pred_stride]));
1588     pred_r3 = _mm_loadl_epi64((__m128i *) (&pu1_pred[3 * i4_pred_stride]));
1589 
1590     pred_r0 = _mm_and_si128(pred_r0, chroma_mask);
1591     pred_r1 = _mm_and_si128(pred_r1, chroma_mask);
1592     pred_r2 = _mm_and_si128(pred_r2, chroma_mask);
1593     pred_r3 = _mm_and_si128(pred_r3, chroma_mask);
1594 
1595     /*--------------------------------------------------------------*/
1596     /* IDCT [ Vertical transformation] and Xij = (xij + 32)>>6      */
1597     /*                                                              */
1598     /* Add the prediction and store it back to same buffer          */
1599     /*--------------------------------------------------------------*/
1600     /* z0j = y0j + y2j                                         */
1601     temp0 = _mm_add_epi32(resq_r0, resq_r2);
1602     /* z1j = y0j - y2j                                                        */
1603     temp1 = _mm_sub_epi32(resq_r0, resq_r2);
1604     /* z2j = (y1j>>1) - y3j */
1605     temp2 = _mm_srai_epi32(resq_r1, 1);
1606     temp2 = _mm_sub_epi32(temp2, resq_r3);
1607     /* z3j = y1j + (y3j>>1) */
1608     temp3 = _mm_srai_epi32(resq_r3, 1);
1609     temp3 = _mm_add_epi32(temp3, resq_r1);
1610 
1611     /* x0j = z0j + z3j                                                        */
1612     temp4 = _mm_add_epi32(temp0, temp3);
1613     temp4 = _mm_add_epi32(temp4, value_32);
1614     temp4 = _mm_srai_epi32(temp4, 6);
1615     res_r0 = temp4;
1616     /* x1j = z1j + z2j                                                        */
1617     temp5 = _mm_add_epi32(temp1, temp2);
1618     temp5 = _mm_add_epi32(temp5, value_32);
1619     temp5 = _mm_srai_epi32(temp5, 6);
1620     res_r1 = temp5;
1621     /* x2j = z1j - z2j                                                        */
1622     temp6 = _mm_sub_epi32(temp1, temp2);
1623     temp6 = _mm_add_epi32(temp6, value_32);
1624     temp6 = _mm_srai_epi32(temp6, 6);
1625     res_r2 = temp6;
1626     /* x3j = z0j - z3j                                                        */
1627     temp7 = _mm_sub_epi32(temp0, temp3);
1628     temp7 = _mm_add_epi32(temp7, value_32);
1629     temp7 = _mm_srai_epi32(temp7, 6);
1630     res_r3 = temp7;
1631 
1632     res_pred_r0 = _mm_loadu_si128((__m128i *) &pi2_res_pred[0 * i4_res_pred_stride]);
1633     res_pred_r1 = _mm_loadu_si128((__m128i *) &pi2_res_pred[1 * i4_res_pred_stride]);
1634     res_pred_r2 = _mm_loadu_si128((__m128i *) &pi2_res_pred[2 * i4_res_pred_stride]);
1635     res_pred_r3 = _mm_loadu_si128((__m128i *) &pi2_res_pred[3 * i4_res_pred_stride]);
1636 
1637     res_pred_r0 = _mm_and_si128(res_pred_r0, reg_chroma);
1638     res_pred_r1 = _mm_and_si128(res_pred_r1, reg_chroma);
1639     res_pred_r2 = _mm_and_si128(res_pred_r2, reg_chroma);
1640     res_pred_r3 = _mm_and_si128(res_pred_r3, reg_chroma);
1641 
1642     temp0 = _mm_packs_epi32(res_r0, res_r1);
1643     temp1 = _mm_packs_epi32(res_r2, res_r3);
1644 
1645     res_r0 = _mm_cvtepu16_epi32(temp0);
1646     res_r2 = _mm_cvtepu16_epi32(temp1);
1647     res_r1 = _mm_srli_si128(temp0, 8);
1648     res_r3 = _mm_srli_si128(temp1, 8);
1649     res_r1 = _mm_cvtepu16_epi32(res_r1);
1650     res_r3 = _mm_cvtepu16_epi32(res_r3);
1651 
1652     res_r0 = _mm_add_epi16(res_pred_r0, res_r0);
1653     res_r1 = _mm_add_epi16(res_pred_r1, res_r1);
1654     res_r2 = _mm_add_epi16(res_pred_r2, res_r2);
1655     res_r3 = _mm_add_epi16(res_pred_r3, res_r3);
1656 
1657     temp0 = _mm_packus_epi32(res_r0, res_r1);
1658     temp1 = _mm_packus_epi32(res_r2, res_r3);
1659 
1660     /* Saturate all values < -255 to -255 and retain the rest as it is */
1661     temp0 = _mm_max_epi16(temp0, neg_255_8x16b);
1662     /* Saturate all values > 255 to 255 and retain the rest as it is */
1663     temp0 = _mm_min_epi16(temp0, pos_255_8x16b);
1664 
1665     /* Saturate all values < -255 to -255 and retain the rest as it is */
1666     temp1 = _mm_max_epi16(temp1, neg_255_8x16b);
1667     /* Saturate all values > 255 to 255 and retain the rest as it is */
1668     temp1 = _mm_min_epi16(temp1, pos_255_8x16b);
1669 
1670     res_r0 = _mm_cvtepu16_epi32(temp0);
1671     res_r1 = _mm_srli_si128(temp0, 8);
1672     res_r1 = _mm_cvtepu16_epi32(res_r1);
1673 
1674     res_r2 = _mm_cvtepu16_epi32(temp1);
1675     res_r3 = _mm_srli_si128(temp1, 8);
1676     res_r3 = _mm_cvtepu16_epi32(res_r3);
1677 
1678     chroma_mask = _mm_set1_epi32(0xffff0000);
1679     out_r0 = _mm_loadu_si128((__m128i *) (&pi2_res[0 * i4_res_stride]));
1680     out_r1 = _mm_loadu_si128((__m128i *) (&pi2_res[1 * i4_res_stride]));
1681     out_r2 = _mm_loadu_si128((__m128i *) (&pi2_res[2 * i4_res_stride]));
1682     out_r3 = _mm_loadu_si128((__m128i *) (&pi2_res[3 * i4_res_stride]));
1683 
1684     out_r0 = _mm_and_si128(out_r0, chroma_mask);
1685     out_r1 = _mm_and_si128(out_r1, chroma_mask);
1686     out_r2 = _mm_and_si128(out_r2, chroma_mask);
1687     out_r3 = _mm_and_si128(out_r3, chroma_mask);
1688 
1689     out_r0 = _mm_add_epi16(out_r0, res_r0);
1690     out_r1 = _mm_add_epi16(out_r1, res_r1);
1691     out_r2 = _mm_add_epi16(out_r2, res_r2);
1692     out_r3 = _mm_add_epi16(out_r3, res_r3);
1693 
1694     _mm_storeu_si128((__m128i *) (&pi2_res[0 * i4_res_stride]), out_r0);
1695     _mm_storeu_si128((__m128i *) (&pi2_res[1 * i4_res_stride]), out_r1);
1696     _mm_storeu_si128((__m128i *) (&pi2_res[2 * i4_res_stride]), out_r2);
1697     _mm_storeu_si128((__m128i *) (&pi2_res[3 * i4_res_stride]), out_r3);
1698 
1699     pred_r0 = _mm_cvtepu16_epi32(pred_r0);
1700     pred_r1 = _mm_cvtepu16_epi32(pred_r1);
1701     pred_r2 = _mm_cvtepu16_epi32(pred_r2);
1702     pred_r3 = _mm_cvtepu16_epi32(pred_r3);
1703 
1704     resq_r0 = _mm_add_epi16(pred_r0, res_r0);
1705     resq_r1 = _mm_add_epi16(pred_r1, res_r1);
1706     resq_r2 = _mm_add_epi16(pred_r2, res_r2);
1707     resq_r3 = _mm_add_epi16(pred_r3, res_r3);
1708 
1709     temp0 = _mm_packus_epi32(resq_r0, resq_r1);
1710     temp1 = _mm_packus_epi32(resq_r2, resq_r3);
1711 
1712     /* Clipping the results to 8 bits */
1713     mask_r0 = _mm_cmpgt_epi16(temp0, zero_8x16b);
1714     temp0 = _mm_and_si128(temp0, mask_r0);
1715     mask_r0 = _mm_cmpgt_epi16(temp1, zero_8x16b);
1716     temp1 = _mm_and_si128(temp1, mask_r0);
1717 
1718     resq_r0 = _mm_packus_epi16(temp0, temp1);
1719     resq_r1 = _mm_srli_si128(resq_r0, 4);
1720     resq_r2 = _mm_srli_si128(resq_r1, 4);
1721     resq_r3 = _mm_srli_si128(resq_r2, 4);
1722 
1723     resq_r0 = _mm_cvtepu8_epi16(resq_r0);
1724     resq_r1 = _mm_cvtepu8_epi16(resq_r1);
1725     resq_r2 = _mm_cvtepu8_epi16(resq_r2);
1726     resq_r3 = _mm_cvtepu8_epi16(resq_r3);
1727 
1728     chroma_mask = _mm_set1_epi16(0xFF00);
1729     out_r0 = _mm_loadl_epi64((__m128i *) (&pu1_out[0 * i4_out_stride]));
1730     out_r1 = _mm_loadl_epi64((__m128i *) (&pu1_out[1 * i4_out_stride]));
1731     out_r2 = _mm_loadl_epi64((__m128i *) (&pu1_out[2 * i4_out_stride]));
1732     out_r3 = _mm_loadl_epi64((__m128i *) (&pu1_out[3 * i4_out_stride]));
1733 
1734     out_r0 = _mm_and_si128(out_r0, chroma_mask);
1735     out_r1 = _mm_and_si128(out_r1, chroma_mask);
1736     out_r2 = _mm_and_si128(out_r2, chroma_mask);
1737     out_r3 = _mm_and_si128(out_r3, chroma_mask);
1738 
1739     out_r0 = _mm_add_epi8(out_r0, resq_r0);
1740     out_r1 = _mm_add_epi8(out_r1, resq_r1);
1741     out_r2 = _mm_add_epi8(out_r2, resq_r2);
1742     out_r3 = _mm_add_epi8(out_r3, resq_r3);
1743 
1744     _mm_storel_epi64((__m128i *) (&pu1_out[0 * i4_out_stride]), out_r0);
1745     _mm_storel_epi64((__m128i *) (&pu1_out[1 * i4_out_stride]), out_r1);
1746     _mm_storel_epi64((__m128i *) (&pu1_out[2 * i4_out_stride]), out_r2);
1747     _mm_storel_epi64((__m128i *) (&pu1_out[3 * i4_out_stride]), out_r3);
1748 }
1749 
isvc_iquant_itrans_recon_dc_4x4_sse42(buffer_container_t * ps_src,buffer_container_t * ps_pred,buffer_container_t * ps_res_pred,buffer_container_t * ps_res,buffer_container_t * ps_rec,iq_it_res_rec_constants_t * ps_iq_it_res_rec_constants,WORD16 * pi2_tmp,WORD16 * pi2_dc_src,WORD32 i4_iq_start_idx,UWORD8 u1_res_accumulate)1750 void isvc_iquant_itrans_recon_dc_4x4_sse42(buffer_container_t *ps_src, buffer_container_t *ps_pred,
1751                                            buffer_container_t *ps_res_pred,
1752                                            buffer_container_t *ps_res, buffer_container_t *ps_rec,
1753                                            iq_it_res_rec_constants_t *ps_iq_it_res_rec_constants,
1754                                            WORD16 *pi2_tmp, WORD16 *pi2_dc_src,
1755                                            WORD32 i4_iq_start_idx, UWORD8 u1_res_accumulate)
1756 {
1757     UWORD8 *pu1_pred = (UWORD8 *) ps_pred->pv_data;
1758     UWORD8 *pu1_out = (UWORD8 *) ps_rec->pv_data;
1759     WORD32 i4_pred_stride = ps_pred->i4_data_stride;
1760     WORD32 i4_out_stride = ps_rec->i4_data_stride;
1761     const UWORD16 *pu2_iscal_mat = ps_iq_it_res_rec_constants->pu2_iscal_mat;
1762     const UWORD16 *pu2_weigh_mat = ps_iq_it_res_rec_constants->pu2_weigh_mat;
1763     UWORD32 u4_qp_div_6 = ps_iq_it_res_rec_constants->u4_qp_div_6;
1764     UWORD32 *pu4_out = (UWORD32 *) pu1_out;
1765     WORD32 q0 = ((WORD16 *) (ps_src->pv_data))[0];
1766     WORD16 i_macro, rnd_fact = (u4_qp_div_6 < 4) ? 1 << (3 - u4_qp_div_6) : 0;
1767 
1768     __m128i pred_r0, pred_r1, pred_r2, pred_r3;
1769     __m128i sign_reg;
1770     /* all bits reset to zero */
1771     __m128i zero_8x16b = _mm_setzero_si128();
1772     __m128i temp4, temp5, temp6, temp7;
1773     __m128i value_add;
1774 
1775     ASSERT(0 == u1_res_accumulate);
1776 
1777     UNUSED(pi2_tmp);
1778     UNUSED(ps_res);
1779     UNUSED(ps_res_pred);
1780     UNUSED(u1_res_accumulate);
1781 
1782     INV_QUANT(q0, pu2_iscal_mat[0], pu2_weigh_mat[0], u4_qp_div_6, rnd_fact, 4);
1783 
1784     /* Restoring dc value for intra case */
1785     if(i4_iq_start_idx != 0)
1786     {
1787         q0 = pi2_dc_src[0];
1788     }
1789 
1790     i_macro = ((q0 + 32) >> 6);
1791 
1792     value_add = _mm_set1_epi16(i_macro);
1793 
1794     zero_8x16b = _mm_setzero_si128();
1795 
1796     /* Load pred buffer */
1797 
1798     /* p00 p01 p02 p03 0 0 0 0 -- all 8 bits */
1799     pred_r0 = _mm_loadl_epi64((__m128i *) (&pu1_pred[0]));
1800 
1801     /* p10 p11 p12 p13 0 0 0 0 -- all 8 bits */
1802     pred_r1 = _mm_loadl_epi64((__m128i *) (&pu1_pred[i4_pred_stride]));
1803 
1804     /* p20 p21 p22 p23 0 0 0 0 -- all 8 bits */
1805     pred_r2 = _mm_loadl_epi64((__m128i *) (&pu1_pred[2 * i4_pred_stride]));
1806 
1807     /* p30 p31 p32 p33 0 0 0 0 -- all 8 bits */
1808     pred_r3 = _mm_loadl_epi64((__m128i *) (&pu1_pred[3 * i4_pred_stride]));
1809 
1810     pred_r0 = _mm_cvtepu8_epi16(pred_r0);
1811     pred_r1 = _mm_cvtepu8_epi16(pred_r1);
1812     pred_r2 = _mm_cvtepu8_epi16(pred_r2);
1813     pred_r3 = _mm_cvtepu8_epi16(pred_r3);
1814 
1815     pred_r0 = _mm_unpacklo_epi64(pred_r0, pred_r1);
1816     pred_r2 = _mm_unpacklo_epi64(pred_r2, pred_r3);
1817 
1818     temp4 = _mm_add_epi16(value_add, pred_r0);
1819     temp5 = _mm_add_epi16(value_add, pred_r2);
1820     /*------------------------------------------------------------------*/
1821     /* Clipping the results to 8 bits */
1822     sign_reg = _mm_cmpgt_epi16(temp4, zero_8x16b);
1823     temp4 = _mm_and_si128(temp4, sign_reg);
1824     sign_reg = _mm_cmpgt_epi16(temp5, zero_8x16b);
1825     temp5 = _mm_and_si128(temp5, sign_reg);
1826 
1827     temp4 = _mm_packus_epi16(temp4, temp5);
1828     temp5 = _mm_srli_si128(temp4, 4);
1829     temp6 = _mm_srli_si128(temp5, 4);
1830     temp7 = _mm_srli_si128(temp6, 4);
1831 
1832     *pu4_out = _mm_cvtsi128_si32(temp4);
1833     pu1_out += i4_out_stride;
1834     pu4_out = (UWORD32 *) (pu1_out);
1835     *(pu4_out) = _mm_cvtsi128_si32(temp5);
1836     pu1_out += i4_out_stride;
1837     pu4_out = (UWORD32 *) (pu1_out);
1838     *(pu4_out) = _mm_cvtsi128_si32(temp6);
1839     pu1_out += i4_out_stride;
1840     pu4_out = (UWORD32 *) (pu1_out);
1841     *(pu4_out) = _mm_cvtsi128_si32(temp7);
1842 }
1843 
isvc_iquant_itrans_recon_res_chroma_4x4_dc_sse42(buffer_container_t * ps_src,buffer_container_t * ps_pred,buffer_container_t * ps_res_pred,buffer_container_t * ps_res,buffer_container_t * ps_rec,iq_it_res_rec_constants_t * ps_iq_it_res_rec_constants,WORD16 * pi2_tmp,WORD16 * pi2_dc_src,WORD32 i4_iq_start_idx,UWORD8 u1_res_accumulate)1844 void isvc_iquant_itrans_recon_res_chroma_4x4_dc_sse42(
1845     buffer_container_t *ps_src, buffer_container_t *ps_pred, buffer_container_t *ps_res_pred,
1846     buffer_container_t *ps_res, buffer_container_t *ps_rec,
1847     iq_it_res_rec_constants_t *ps_iq_it_res_rec_constants, WORD16 *pi2_tmp, WORD16 *pi2_dc_src,
1848     WORD32 i4_iq_start_idx, UWORD8 u1_res_accumulate)
1849 {
1850     WORD16 *pi2_src = (WORD16 *) ps_src->pv_data;
1851     WORD16 *pi2_res = (WORD16 *) ps_res->pv_data;
1852     WORD16 *pi2_res_ptr = pi2_res;
1853     UWORD8 *pu1_pred = (UWORD8 *) ps_pred->pv_data;
1854     UWORD8 *pu1_out = (UWORD8 *) ps_rec->pv_data;
1855     WORD32 i4_res_stride = ps_res->i4_data_stride;
1856     WORD32 i4_pred_stride = ps_pred->i4_data_stride;
1857     WORD32 i4_out_stride = ps_rec->i4_data_stride;
1858     const UWORD16 *pu2_iscal_mat = ps_iq_it_res_rec_constants->pu2_iscal_mat;
1859     const UWORD16 *pu2_weigh_mat = ps_iq_it_res_rec_constants->pu2_weigh_mat;
1860     UWORD32 u4_qp_div_6 = ps_iq_it_res_rec_constants->u4_qp_div_6;
1861     /* DC value won't be dequantized for chroma
1862     inverse transform */
1863     WORD16 q0 = pi2_dc_src[0];
1864     WORD16 i_macro = ((q0 + 32) >> 6);
1865 
1866     __m128i pred_r0, pred_r1, pred_r2, pred_r3, sign_reg;
1867     /* all bits reset to zero */
1868     __m128i zero_8x16b = _mm_setzero_si128();
1869     __m128i chroma_mask = _mm_set1_epi16(0xFF);
1870     __m128i value_add = _mm_set1_epi16(isvc_get_residue(i_macro, 0, 0));
1871     __m128i out_r0, out_r1, out_r2, out_r3;
1872 
1873     ASSERT(0 == u1_res_accumulate);
1874 
1875     UNUSED(pi2_src);
1876     UNUSED(pu2_iscal_mat);
1877     UNUSED(pu2_weigh_mat);
1878     UNUSED(u4_qp_div_6);
1879     UNUSED(pi2_tmp);
1880     UNUSED(ps_res_pred);
1881     UNUSED(i4_iq_start_idx);
1882     UNUSED(u1_res_accumulate);
1883 
1884     /* Load pred buffer */
1885     pred_r0 = _mm_loadl_epi64((__m128i *) (&pu1_pred[0]));
1886 
1887     pred_r1 = _mm_loadl_epi64((__m128i *) (&pu1_pred[i4_pred_stride]));
1888 
1889     pred_r2 = _mm_loadl_epi64((__m128i *) (&pu1_pred[2 * i4_pred_stride]));
1890 
1891     pred_r3 = _mm_loadl_epi64((__m128i *) (&pu1_pred[3 * i4_pred_stride]));
1892 
1893     /* Mask alternate pred values from the interleaved pred buf */
1894     pred_r0 = _mm_and_si128(pred_r0, chroma_mask);
1895     pred_r1 = _mm_and_si128(pred_r1, chroma_mask);
1896     pred_r2 = _mm_and_si128(pred_r2, chroma_mask);
1897     pred_r3 = _mm_and_si128(pred_r3, chroma_mask);
1898 
1899     /* Pack the first four 16 bit values of 2 regs into a single reg*/
1900     pred_r0 = _mm_unpacklo_epi64(pred_r0, pred_r1);
1901     pred_r2 = _mm_unpacklo_epi64(pred_r2, pred_r3);
1902 
1903     /* Compute out pixel by adding res to pred */
1904     pred_r0 = _mm_add_epi16(value_add, pred_r0);
1905     pred_r2 = _mm_add_epi16(value_add, pred_r2);
1906 
1907     /* Convert res from 16 bits to 32 bits  */
1908     value_add = _mm_cvtepu16_epi32(value_add);
1909 
1910     out_r0 = _mm_loadu_si128((__m128i *) (&pi2_res_ptr[0 * i4_res_stride]));
1911     out_r1 = _mm_loadu_si128((__m128i *) (&pi2_res_ptr[1 * i4_res_stride]));
1912     out_r2 = _mm_loadu_si128((__m128i *) (&pi2_res_ptr[2 * i4_res_stride]));
1913     out_r3 = _mm_loadu_si128((__m128i *) (&pi2_res_ptr[3 * i4_res_stride]));
1914 
1915     /* Mask the loaded res in order to save the U/V res data computed in
1916     this function call without thrashing the U/V res data that was saved
1917     during an earlier function call */
1918     chroma_mask = _mm_set1_epi32(0xffff0000);
1919     out_r0 = _mm_and_si128(out_r0, chroma_mask);
1920     out_r1 = _mm_and_si128(out_r1, chroma_mask);
1921     out_r2 = _mm_and_si128(out_r2, chroma_mask);
1922     out_r3 = _mm_and_si128(out_r3, chroma_mask);
1923 
1924     /* Save the res in alternate locations */
1925     out_r0 = _mm_add_epi16(out_r0, value_add);
1926     out_r1 = _mm_add_epi16(out_r1, value_add);
1927     out_r2 = _mm_add_epi16(out_r2, value_add);
1928     out_r3 = _mm_add_epi16(out_r3, value_add);
1929 
1930     _mm_storeu_si128((__m128i *) (&pi2_res_ptr[0 * i4_res_stride]), out_r0);
1931     _mm_storeu_si128((__m128i *) (&pi2_res_ptr[1 * i4_res_stride]), out_r1);
1932     _mm_storeu_si128((__m128i *) (&pi2_res_ptr[2 * i4_res_stride]), out_r2);
1933     _mm_storeu_si128((__m128i *) (&pi2_res_ptr[3 * i4_res_stride]), out_r3);
1934     /*------------------------------------------------------------------*/
1935     /* Clipping the results to 8 bits */
1936     sign_reg = _mm_cmpgt_epi16(pred_r0, zero_8x16b);
1937     pred_r0 = _mm_and_si128(pred_r0, sign_reg);
1938     sign_reg = _mm_cmpgt_epi16(pred_r2, zero_8x16b);
1939     pred_r2 = _mm_and_si128(pred_r2, sign_reg);
1940 
1941     pred_r0 = _mm_packus_epi16(pred_r0, pred_r2);
1942     pred_r1 = _mm_srli_si128(pred_r0, 4);
1943     pred_r2 = _mm_srli_si128(pred_r1, 4);
1944     pred_r3 = _mm_srli_si128(pred_r2, 4);
1945 
1946     /* p00 p01 p02 p03 -- all 16 bits */
1947     pred_r0 = _mm_unpacklo_epi8(pred_r0, zero_8x16b);
1948     /* p10 p11 p12 p13 -- all 16 bits */
1949     pred_r1 = _mm_unpacklo_epi8(pred_r1, zero_8x16b);
1950     /* p20 p21 p22 p23 -- all 16 bits */
1951     pred_r2 = _mm_unpacklo_epi8(pred_r2, zero_8x16b);
1952     /* p30 p31 p32 p33 -- all 16 bits */
1953     pred_r3 = _mm_unpacklo_epi8(pred_r3, zero_8x16b);
1954 
1955     /* Load interleaved out buffer */
1956     out_r0 = _mm_loadl_epi64((__m128i *) (&pu1_out[0]));
1957     out_r1 = _mm_loadl_epi64((__m128i *) (&pu1_out[i4_out_stride]));
1958     out_r2 = _mm_loadl_epi64((__m128i *) (&pu1_out[2 * i4_out_stride]));
1959     out_r3 = _mm_loadl_epi64((__m128i *) (&pu1_out[3 * i4_out_stride]));
1960 
1961     /* Mask the interleaved out buf in order to save the U/V out pixel computed in
1962     this function call without thrashing the U/V out pixel that was saved
1963     during an earlier function call */
1964     chroma_mask = _mm_set1_epi16(0xFF00);
1965 
1966     out_r0 = _mm_and_si128(out_r0, chroma_mask);
1967     out_r1 = _mm_and_si128(out_r1, chroma_mask);
1968     out_r2 = _mm_and_si128(out_r2, chroma_mask);
1969     out_r3 = _mm_and_si128(out_r3, chroma_mask);
1970 
1971     /* Save the out pixels in alternate locations */
1972     out_r0 = _mm_add_epi8(out_r0, pred_r0);
1973     out_r1 = _mm_add_epi8(out_r1, pred_r1);
1974     out_r2 = _mm_add_epi8(out_r2, pred_r2);
1975     out_r3 = _mm_add_epi8(out_r3, pred_r3);
1976 
1977     _mm_storel_epi64((__m128i *) (&pu1_out[0]), out_r0);
1978     _mm_storel_epi64((__m128i *) (&pu1_out[i4_out_stride]), out_r1);
1979     _mm_storel_epi64((__m128i *) (&pu1_out[2 * i4_out_stride]), out_r2);
1980     _mm_storel_epi64((__m128i *) (&pu1_out[3 * i4_out_stride]), out_r3);
1981 }
1982 
isvc_iquant_itrans_recon_res_chroma_4x4_dc_with_res_acc_sse42(buffer_container_t * ps_src,buffer_container_t * ps_pred,buffer_container_t * ps_res_pred,buffer_container_t * ps_res,buffer_container_t * ps_rec,iq_it_res_rec_constants_t * ps_iq_it_res_rec_constants,WORD16 * pi2_tmp,WORD16 * pi2_dc_src,WORD32 i4_iq_start_idx,UWORD8 u1_res_accumulate)1983 void isvc_iquant_itrans_recon_res_chroma_4x4_dc_with_res_acc_sse42(
1984     buffer_container_t *ps_src, buffer_container_t *ps_pred, buffer_container_t *ps_res_pred,
1985     buffer_container_t *ps_res, buffer_container_t *ps_rec,
1986     iq_it_res_rec_constants_t *ps_iq_it_res_rec_constants, WORD16 *pi2_tmp, WORD16 *pi2_dc_src,
1987     WORD32 i4_iq_start_idx, UWORD8 u1_res_accumulate)
1988 {
1989     WORD16 *pi2_src = (WORD16 *) ps_src->pv_data;
1990     WORD16 *pi2_res = (WORD16 *) ps_res->pv_data;
1991     WORD16 *pi2_res_pred = (WORD16 *) ps_res_pred->pv_data;
1992     UWORD8 *pu1_pred = (UWORD8 *) ps_pred->pv_data;
1993     UWORD8 *pu1_out = (UWORD8 *) ps_rec->pv_data;
1994     WORD32 i4_res_stride = ps_res->i4_data_stride;
1995     WORD32 i4_res_pred_stride = ps_res_pred->i4_data_stride;
1996     WORD32 i4_pred_stride = ps_pred->i4_data_stride;
1997     WORD32 i4_out_stride = ps_rec->i4_data_stride;
1998     const UWORD16 *pu2_iscal_mat = ps_iq_it_res_rec_constants->pu2_iscal_mat;
1999     const UWORD16 *pu2_weigh_mat = ps_iq_it_res_rec_constants->pu2_weigh_mat;
2000     UWORD32 u4_qp_div_6 = ps_iq_it_res_rec_constants->u4_qp_div_6;
2001     /* DC value won't be dequantized for chroma
2002     inverse transform */
2003     WORD16 q0 = pi2_dc_src[0];
2004     WORD16 i_macro = ((q0 + 32) >> 6);
2005 
2006     __m128i pred_r0, pred_r1, pred_r2, pred_r3;
2007     /* all bits reset to zero */
2008     __m128i zero_8x16b = _mm_setzero_si128();
2009     __m128i chroma_mask = _mm_set1_epi16(0xFF);
2010     __m128i reg_chroma = _mm_set_epi16(0, 0xFFFF, 0, 0xFFFF, 0, 0xFFFF, 0, 0xFFFF);
2011     __m128i value_add = _mm_set1_epi16(i_macro);
2012     __m128i out_r0, out_r1, out_r2, out_r3;
2013     __m128i res_r0, res_r1, res_r2, res_r3;
2014     __m128i res_pred_r0, res_pred_r1, res_pred_r2, res_pred_r3;
2015     __m128i temp0, temp1;
2016     __m128i neg_255_8x16b = _mm_set1_epi16(-((WORD16) UINT8_MAX));
2017     __m128i pos_255_8x16b = _mm_set1_epi16(((WORD16) UINT8_MAX));
2018 
2019     ASSERT(1 == u1_res_accumulate);
2020 
2021     UNUSED(pi2_src);
2022     UNUSED(pu2_iscal_mat);
2023     UNUSED(pu2_weigh_mat);
2024     UNUSED(u4_qp_div_6);
2025     UNUSED(pi2_tmp);
2026     UNUSED(i4_iq_start_idx);
2027     UNUSED(u1_res_accumulate);
2028 
2029     /* Load pred buffer */
2030     pred_r0 = _mm_loadl_epi64((__m128i *) (&pu1_pred[0]));
2031 
2032     pred_r1 = _mm_loadl_epi64((__m128i *) (&pu1_pred[i4_pred_stride]));
2033 
2034     pred_r2 = _mm_loadl_epi64((__m128i *) (&pu1_pred[2 * i4_pred_stride]));
2035 
2036     pred_r3 = _mm_loadl_epi64((__m128i *) (&pu1_pred[3 * i4_pred_stride]));
2037     /* Mask alternate pred values from the interleaved pred buf */
2038     pred_r0 = _mm_and_si128(pred_r0, chroma_mask);
2039     pred_r1 = _mm_and_si128(pred_r1, chroma_mask);
2040     pred_r2 = _mm_and_si128(pred_r2, chroma_mask);
2041     pred_r3 = _mm_and_si128(pred_r3, chroma_mask);
2042 
2043     /* Pack the first four 16 bit values of 2 regs into a single reg*/
2044     pred_r0 = _mm_unpacklo_epi64(pred_r0, pred_r1);
2045     pred_r2 = _mm_unpacklo_epi64(pred_r2, pred_r3);
2046 
2047     /* Accumulating res */
2048 
2049     /* load res pred buffer */
2050     res_pred_r0 = _mm_loadu_si128((__m128i *) &pi2_res_pred[0 * i4_res_pred_stride]);
2051     res_pred_r1 = _mm_loadu_si128((__m128i *) &pi2_res_pred[1 * i4_res_pred_stride]);
2052     res_pred_r2 = _mm_loadu_si128((__m128i *) &pi2_res_pred[2 * i4_res_pred_stride]);
2053     res_pred_r3 = _mm_loadu_si128((__m128i *) &pi2_res_pred[3 * i4_res_pred_stride]);
2054 
2055     /* Mask res pred and retain alternate values */
2056     res_pred_r0 = _mm_and_si128(res_pred_r0, reg_chroma);
2057     res_pred_r1 = _mm_and_si128(res_pred_r1, reg_chroma);
2058     res_pred_r2 = _mm_and_si128(res_pred_r2, reg_chroma);
2059     res_pred_r3 = _mm_and_si128(res_pred_r3, reg_chroma);
2060 
2061     /* Convert to 32 bits */
2062     res_r0 = _mm_cvtepu16_epi32(value_add);
2063     res_r2 = _mm_cvtepu16_epi32(value_add);
2064     res_r1 = _mm_cvtepu16_epi32(value_add);
2065     res_r3 = _mm_cvtepu16_epi32(value_add);
2066 
2067     /* Add res pred to the res obtained from inv transform */
2068     res_r0 = _mm_add_epi16(res_pred_r0, res_r0);
2069     res_r1 = _mm_add_epi16(res_pred_r1, res_r1);
2070     res_r2 = _mm_add_epi16(res_pred_r2, res_r2);
2071     res_r3 = _mm_add_epi16(res_pred_r3, res_r3);
2072 
2073     /* Convert 32 bit res of the format [a0 0 a1 0 a2 0 a3 0] to
2074     16 bits of the format [a0 a1 a2 a3] using hadd [ao + 0,
2075     a1 + 0, a2 + 0, a3 + 0] To be optimized */
2076     temp0 = _mm_hadd_epi16(res_r0, res_r1);
2077     temp1 = _mm_hadd_epi16(res_r2, res_r3);
2078 
2079     /* Saturate all values < -255 to -255 and retain the rest as it is */
2080     temp0 = _mm_max_epi16(temp0, neg_255_8x16b);
2081     /* Saturate all values > 255 to 255 and retain the rest as it is */
2082     temp0 = _mm_min_epi16(temp0, pos_255_8x16b);
2083 
2084     /* Saturate all values < -255 to -255 and retain the rest as it is */
2085     temp1 = _mm_max_epi16(temp1, neg_255_8x16b);
2086     /* Saturate all values > 255 to 255 and retain the rest as it is */
2087     temp1 = _mm_min_epi16(temp1, pos_255_8x16b);
2088 
2089     /* Compute out pixel by adding res to pred */
2090     pred_r0 = _mm_add_epi16(temp0, pred_r0);
2091     pred_r2 = _mm_add_epi16(temp1, pred_r2);
2092 
2093     res_r0 = _mm_cvtepu16_epi32(temp0);
2094     res_r2 = _mm_cvtepu16_epi32(temp1);
2095     res_r1 = _mm_srli_si128(temp0, 8);
2096     res_r3 = _mm_srli_si128(temp1, 8);
2097     res_r1 = _mm_cvtepu16_epi32(res_r1);
2098     res_r3 = _mm_cvtepu16_epi32(res_r3);
2099 
2100     /* Load res buffer */
2101     out_r0 = _mm_loadu_si128((__m128i *) (&pi2_res[0 * i4_res_stride]));
2102     out_r1 = _mm_loadu_si128((__m128i *) (&pi2_res[1 * i4_res_stride]));
2103     out_r2 = _mm_loadu_si128((__m128i *) (&pi2_res[2 * i4_res_stride]));
2104     out_r3 = _mm_loadu_si128((__m128i *) (&pi2_res[3 * i4_res_stride]));
2105 
2106     /* Mask the loaded res in order to save the U/V res data computed in
2107     this function call without thrashing the U/V res data that was saved
2108     during an earlier function call */
2109     chroma_mask = _mm_set1_epi32(0xffff0000);
2110 
2111     out_r0 = _mm_and_si128(out_r0, chroma_mask);
2112     out_r1 = _mm_and_si128(out_r1, chroma_mask);
2113     out_r2 = _mm_and_si128(out_r2, chroma_mask);
2114     out_r3 = _mm_and_si128(out_r3, chroma_mask);
2115 
2116     /* Save the res in alternate locations */
2117     out_r0 = _mm_add_epi16(out_r0, res_r0);
2118     out_r1 = _mm_add_epi16(out_r1, res_r1);
2119     out_r2 = _mm_add_epi16(out_r2, res_r2);
2120     out_r3 = _mm_add_epi16(out_r3, res_r3);
2121 
2122     _mm_storeu_si128((__m128i *) (&pi2_res[0 * i4_res_stride]), out_r0);
2123     _mm_storeu_si128((__m128i *) (&pi2_res[1 * i4_res_stride]), out_r1);
2124     _mm_storeu_si128((__m128i *) (&pi2_res[2 * i4_res_stride]), out_r2);
2125     _mm_storeu_si128((__m128i *) (&pi2_res[3 * i4_res_stride]), out_r3);
2126     /*------------------------------------------------------------------*/
2127     /* Clipping the results to 8 bits */
2128     pred_r0 = _mm_packus_epi16(pred_r0, pred_r2);
2129     pred_r1 = _mm_srli_si128(pred_r0, 4);
2130     pred_r2 = _mm_srli_si128(pred_r1, 4);
2131     pred_r3 = _mm_srli_si128(pred_r2, 4);
2132 
2133     /* p00 p01 p02 p03 -- all 16 bits */
2134     pred_r0 = _mm_unpacklo_epi8(pred_r0, zero_8x16b);
2135     /* p10 p11 p12 p13 -- all 16 bits */
2136     pred_r1 = _mm_unpacklo_epi8(pred_r1, zero_8x16b);
2137     /* p20 p21 p22 p23 -- all 16 bits */
2138     pred_r2 = _mm_unpacklo_epi8(pred_r2, zero_8x16b);
2139     /* p30 p31 p32 p33 -- all 16 bits */
2140     pred_r3 = _mm_unpacklo_epi8(pred_r3, zero_8x16b);
2141 
2142     /* Load interleaved out buffer */
2143     out_r0 = _mm_loadl_epi64((__m128i *) (&pu1_out[0]));
2144     out_r1 = _mm_loadl_epi64((__m128i *) (&pu1_out[i4_out_stride]));
2145     out_r2 = _mm_loadl_epi64((__m128i *) (&pu1_out[2 * i4_out_stride]));
2146     out_r3 = _mm_loadl_epi64((__m128i *) (&pu1_out[3 * i4_out_stride]));
2147 
2148     /* Mask the interleaved out buf in order to save the U/V out pixel computed in
2149     this function call without thrashing the U/V out pixel that was saved
2150     during an earlier function call */
2151     chroma_mask = _mm_set1_epi16(0xFF00);
2152 
2153     out_r0 = _mm_and_si128(out_r0, chroma_mask);
2154     out_r1 = _mm_and_si128(out_r1, chroma_mask);
2155     out_r2 = _mm_and_si128(out_r2, chroma_mask);
2156     out_r3 = _mm_and_si128(out_r3, chroma_mask);
2157 
2158     /* Save the out pixels in alternate locations */
2159     out_r0 = _mm_add_epi8(out_r0, pred_r0);
2160     out_r1 = _mm_add_epi8(out_r1, pred_r1);
2161     out_r2 = _mm_add_epi8(out_r2, pred_r2);
2162     out_r3 = _mm_add_epi8(out_r3, pred_r3);
2163 
2164     _mm_storel_epi64((__m128i *) (&pu1_out[0]), out_r0);
2165     _mm_storel_epi64((__m128i *) (&pu1_out[i4_out_stride]), out_r1);
2166     _mm_storel_epi64((__m128i *) (&pu1_out[2 * i4_out_stride]), out_r2);
2167     _mm_storel_epi64((__m128i *) (&pu1_out[3 * i4_out_stride]), out_r3);
2168 }
2169