• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /******************************************************************************
2  *
3  * Copyright (C) 2015 The Android Open Source Project
4  *
5  * Licensed under the Apache License, Version 2.0 (the "License");
6  * you may not use this file except in compliance with the License.
7  * You may obtain a copy of the License at:
8  *
9  * http://www.apache.org/licenses/LICENSE-2.0
10  *
11  * Unless required by applicable law or agreed to in writing, software
12  * distributed under the License is distributed on an "AS IS" BASIS,
13  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14  * See the License for the specific language governing permissions and
15  * limitations under the License.
16  *
17  *****************************************************************************
18  * Originally developed and contributed by Ittiam Systems Pvt. Ltd, Bangalore
19 */
20 /**
21  *******************************************************************************
22  * @file
23  *  ih264_iquant_itrans_recon_dc_ssse3.c
24  *
25  * @brief
26  *  Contains function definitions for inverse  quantization, inverse
27  * transform and reconstruction
28  *
29  * @author
30  *  Mohit [100664]
31  *
32  * @par List of Functions:
33  *  - ih264_iquant_itrans_recon_4x4_dc_ssse3()
34  *  - ih264_iquant_itrans_recon_8x8_dc_ssse3()
35  *
36  * @remarks
37  *  None
38  *
39  *******************************************************************************
40  */
41 /* User include files */
42 #include "ih264_typedefs.h"
43 #include "ih264_defs.h"
44 #include "ih264_trans_macros.h"
45 #include "ih264_macros.h"
46 #include "ih264_platform_macros.h"
47 #include "ih264_trans_data.h"
48 #include "ih264_size_defs.h"
49 #include "ih264_structs.h"
50 #include "ih264_trans_quant_itrans_iquant.h"
51 #include <immintrin.h>
52 
53 /*
54  ********************************************************************************
55  *
56  * @brief This function reconstructs a 4x4 sub block from quantized resiude and
57  * prediction buffer for dc input pattern only, i.e. only the (0,0) element of the input
58  * 4x4 block is non-zero. For complete function, refer ih264_iquant_itrans_recon_ssse3.c
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] pred_strd,
78  *  Prediction buffer stride
79  *
80  * @param[in] out_strd
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  */
ih264_iquant_itrans_recon_4x4_dc_ssse3(WORD16 * pi2_src,UWORD8 * pu1_pred,UWORD8 * pu1_out,WORD32 pred_strd,WORD32 out_strd,const UWORD16 * pu2_iscal_mat,const UWORD16 * pu2_weigh_mat,UWORD32 u4_qp_div_6,WORD16 * pi2_tmp,WORD32 iq_start_idx,WORD16 * pi2_dc_ld_addr)101 void ih264_iquant_itrans_recon_4x4_dc_ssse3(WORD16 *pi2_src,
102                                    UWORD8 *pu1_pred,
103                                    UWORD8 *pu1_out,
104                                    WORD32 pred_strd,
105                                    WORD32 out_strd,
106                                    const UWORD16 *pu2_iscal_mat,
107                                    const UWORD16 *pu2_weigh_mat,
108                                    UWORD32 u4_qp_div_6,
109                                    WORD16 *pi2_tmp,
110                                    WORD32 iq_start_idx,
111                                    WORD16 *pi2_dc_ld_addr)
112 {
113     UWORD32 *pu4_out = (UWORD32 *)pu1_out;
114     WORD32 q0 = pi2_src[0];
115     WORD16 i_macro, rnd_fact = (u4_qp_div_6 < 4) ? 1 << (3 - u4_qp_div_6) : 0;
116 
117     __m128i predload_r,pred_r0, pred_r1, pred_r2, pred_r3;
118     __m128i sign_reg;
119     __m128i zero_8x16b = _mm_setzero_si128();          // all bits reset to zero
120     __m128i temp4, temp5, temp6, temp7;
121     __m128i value_add;
122 
123     UNUSED (pi2_tmp);
124 
125     INV_QUANT(q0, pu2_iscal_mat[0], pu2_weigh_mat[0], u4_qp_div_6, rnd_fact, 4);
126 
127     if (iq_start_idx != 0 )
128         q0 = pi2_dc_ld_addr[0];     // Restoring dc value for intra case
129 
130     i_macro = ((q0 + 32) >> 6);
131 
132     value_add = _mm_set1_epi16(i_macro);
133 
134     zero_8x16b = _mm_setzero_si128();                  // all bits reset to zero
135     //Load pred buffer
136     predload_r = _mm_loadl_epi64((__m128i *) (&pu1_pred[0])); //p00 p01 p02 p03 0 0 0 0 0 0 0 0 -- all 8 bits
137     pred_r0 = _mm_unpacklo_epi8(predload_r, zero_8x16b); //p00 p01 p02 p03 0 0 0 0 -- all 16 bits
138     predload_r = _mm_loadl_epi64((__m128i *) (&pu1_pred[pred_strd])); //p10 p11 p12 p13 0 0 0 0 0 0 0 0 -- all 8 bits
139     pred_r1 = _mm_unpacklo_epi8(predload_r, zero_8x16b); //p10 p11 p12 p13 0 0 0 0 -- all 16 bits
140     predload_r = _mm_loadl_epi64((__m128i *) (&pu1_pred[2*pred_strd])); //p20 p21 p22 p23 0 0 0 0 0 0 0 0 -- all 8 bits
141     pred_r2 = _mm_unpacklo_epi8(predload_r, zero_8x16b); //p20 p21 p22 p23 0 0 0 0 -- all 16 bits
142     predload_r = _mm_loadl_epi64((__m128i *) (&pu1_pred[3*pred_strd])); //p30 p31 p32 p33 0 0 0 0 0 0 0 0 -- all 8 bits
143     pred_r3 = _mm_unpacklo_epi8(predload_r, zero_8x16b); //p30 p31 p32 p33 0 0 0 0 -- all 16 bits
144 
145     pred_r0 = _mm_unpacklo_epi64(pred_r0, pred_r1); //p00 p01 p02 p03 p10 p11 p12 p13
146     pred_r2 = _mm_unpacklo_epi64(pred_r2, pred_r3); //p20 p21 p22p p23 p30 p31 p32 p33
147 
148     temp4 = _mm_add_epi16(value_add, pred_r0);
149     temp5 = _mm_add_epi16(value_add, pred_r2);
150     /*------------------------------------------------------------------*/
151     //Clipping the results to 8 bits
152     sign_reg = _mm_cmpgt_epi16(temp4, zero_8x16b);                 // sign check
153     temp4 = _mm_and_si128(temp4, sign_reg);
154     sign_reg = _mm_cmpgt_epi16(temp5, zero_8x16b);                 // sign check
155     temp5 = _mm_and_si128(temp5, sign_reg);
156 
157     temp4 = _mm_packus_epi16(temp4,temp5);
158     temp5 = _mm_srli_si128(temp4,4);
159     temp6 = _mm_srli_si128(temp5,4);
160     temp7 = _mm_srli_si128(temp6,4);
161 
162     *pu4_out = _mm_cvtsi128_si32(temp4);
163     pu1_out += out_strd;
164     pu4_out = (UWORD32 *)(pu1_out);
165     *(pu4_out) = _mm_cvtsi128_si32(temp5);
166     pu1_out += out_strd;
167     pu4_out = (UWORD32 *)(pu1_out);
168     *(pu4_out) = _mm_cvtsi128_si32(temp6);
169     pu1_out += out_strd;
170     pu4_out = (UWORD32 *)(pu1_out);
171     *(pu4_out) = _mm_cvtsi128_si32(temp7);
172 }
173 /**
174  *******************************************************************************
175  *
176  * @brief
177  *  This function performs inverse quant and Inverse transform type Ci4 for 8x8 block
178  *  for dc input pattern only, i.e. only the (0,0) element of the input 8x8 block is
179  *  non-zero. For complete function, refer ih264_iquant_itrans_recon_ssse3.c
180  *
181  * @par Description:
182  *  Performs inverse transform Ci8 and adds the residue to get the
183  *  reconstructed block
184  *
185  * @param[in] pi2_src
186  *  Input 8x8coefficients
187  *
188  * @param[in] pu1_pred
189  *  Prediction 8x8 block
190  *
191  * @param[out] pu1_recon
192  *  Output 8x8 block
193  *
194  * @param[in] q_div
195  *  QP/6
196  *
197  * @param[in] q_rem
198  *  QP%6
199  *
200  * @param[in] q_lev
201  *  Quantizer level
202  *
203  * @param[in] u4_src_stride
204  *  Input stride
205  *
206  * @param[in] u4_pred_stride,
207  *  Prediction stride
208  *
209  * @param[in] u4_out_stride
210  *  Output Stride
211  *
212  * @param[in] pi4_tmp
213  *  temporary buffer of size 1*64
214  *  the tmp for each block
215  *
216  * @param[in] pu4_iquant_mat
217  *  Pointer to the inverse quantization matrix
218  *
219  * @returns  Void
220  *
221  * @remarks
222  *  None
223  *
224  *******************************************************************************
225  */
226 
ih264_iquant_itrans_recon_8x8_dc_ssse3(WORD16 * pi2_src,UWORD8 * pu1_pred,UWORD8 * pu1_out,WORD32 pred_strd,WORD32 out_strd,const UWORD16 * pu2_iscale_mat,const UWORD16 * pu2_weigh_mat,UWORD32 qp_div,WORD16 * pi2_tmp,WORD32 iq_start_idx,WORD16 * pi2_dc_ld_addr)227 void ih264_iquant_itrans_recon_8x8_dc_ssse3 (WORD16 *pi2_src,
228                                          UWORD8 *pu1_pred,
229                                          UWORD8 *pu1_out,
230                                          WORD32 pred_strd,
231                                          WORD32 out_strd,
232                                          const UWORD16 *pu2_iscale_mat,
233                                          const UWORD16 *pu2_weigh_mat,
234                                          UWORD32 qp_div,
235                                          WORD16 *pi2_tmp,
236                                          WORD32 iq_start_idx,
237                                          WORD16 *pi2_dc_ld_addr)
238 {
239     WORD32 q0 = pi2_src[0];
240     WORD16 i_macro, rnd_fact = (qp_div < 6) ? 1 << (5 - qp_div) : 0;
241 
242     __m128i predload_r,pred_r0, pred_r1, pred_r2, pred_r3,pred_r4,pred_r5,pred_r6,pred_r7;
243     __m128i sign_reg;
244     __m128i zero_8x16b = _mm_setzero_si128();          // all bits reset to zero
245     __m128i temp1,temp2,temp3,temp4, temp5, temp6, temp7,temp8;
246     __m128i value_add;
247 
248     UNUSED (pi2_tmp);
249     UNUSED (iq_start_idx);
250     UNUSED (pi2_dc_ld_addr);
251 
252     INV_QUANT(q0, pu2_iscale_mat[0], pu2_weigh_mat[0], qp_div, rnd_fact, 6);
253     i_macro = ((q0 + 32) >> 6);
254 
255     value_add = _mm_set1_epi16(i_macro);
256 
257     //Load pred buffer row 0
258     predload_r = _mm_loadl_epi64((__m128i *)(&pu1_pred[0])); //p0 p1 p2 p3 p4 p5 p6 p7 0 0 0 0 0 0 0 0 -- all 8 bits
259     pred_r0 = _mm_unpacklo_epi8(predload_r, zero_8x16b); //p0 p1 p2 p3 p4 p5 p6 p7 -- all 16 bits
260     //Load pred buffer row 1
261     predload_r = _mm_loadl_epi64((__m128i *)(&pu1_pred[pred_strd])); //p0 p1 p2 p3 p4 p5 p6 p7 0 0 0 0 0 0 0 0 -- all 8 bits
262     pred_r1 = _mm_unpacklo_epi8(predload_r, zero_8x16b); //p0 p1 p2 p3 p4 p5 p6 p7 -- all 16 bits
263     //Load pred buffer row 2
264     predload_r = _mm_loadl_epi64(
265                     (__m128i *)(&pu1_pred[2 * pred_strd])); //p0 p1 p2 p3 p4 p5 p6 p7 0 0 0 0 0 0 0 0 -- all 8 bits
266     pred_r2 = _mm_unpacklo_epi8(predload_r, zero_8x16b); //p0 p1 p2 p3 p4 p5 p6 p7 -- all 16 bits
267     //Load pred buffer row 3
268     predload_r = _mm_loadl_epi64(
269                     (__m128i *)(&pu1_pred[3 * pred_strd])); //p0 p1 p2 p3 p4 p5 p6 p7 0 0 0 0 0 0 0 0 -- all 8 bits
270     pred_r3 = _mm_unpacklo_epi8(predload_r, zero_8x16b); //p0 p1 p2 p3 p4 p5 p6 p7 -- all 16 bits
271     //Load pred buffer row 4
272     predload_r = _mm_loadl_epi64(
273                     (__m128i *)(&pu1_pred[4 * pred_strd])); //p0 p1 p2 p3 p4 p5 p6 p7 0 0 0 0 0 0 0 0 -- all 8 bits
274     pred_r4 = _mm_unpacklo_epi8(predload_r, zero_8x16b); //p0 p1 p2 p3 p4 p5 p6 p7 -- all 16 bits
275     //Load pred buffer row 5
276     predload_r = _mm_loadl_epi64(
277                     (__m128i *)(&pu1_pred[5 * pred_strd])); //p0 p1 p2 p3 p4 p5 p6 p7 0 0 0 0 0 0 0 0 -- all 8 bit
278     pred_r5 = _mm_unpacklo_epi8(predload_r, zero_8x16b); //p0 p1 p2 p3 p4 p5 p6 p7 -- all 16 bits
279     //Load pred buffer row 6
280     predload_r = _mm_loadl_epi64(
281                     (__m128i *)(&pu1_pred[6 * pred_strd])); //p0 p1 p2 p3 p4 p5 p6 p7 0 0 0 0 0 0 0 0 -- all 8 bits
282     pred_r6 = _mm_unpacklo_epi8(predload_r, zero_8x16b); //p0 p1 p2 p3 p4 p5 p6 p7 -- all 16 bits
283     //Load pred buffer row 7
284     predload_r = _mm_loadl_epi64(
285                     (__m128i *)(&pu1_pred[7 * pred_strd])); //p0 p1 p2 p3 p4 p5 p6 p7 0 0 0 0 0 0 0 0 -- all 8 bits
286     pred_r7 = _mm_unpacklo_epi8(predload_r, zero_8x16b); //p0 p1 p2 p3 p4 p5 p6 p7 -- all 16 bits
287 
288     temp1 = _mm_add_epi16(value_add, pred_r0);
289 
290     temp2 = _mm_add_epi16(value_add, pred_r1);
291 
292     temp3 = _mm_add_epi16(value_add, pred_r2);
293 
294     temp4 = _mm_add_epi16(value_add, pred_r3);
295 
296     temp5 = _mm_add_epi16(value_add, pred_r4);
297 
298     temp6 = _mm_add_epi16(value_add, pred_r5);
299 
300     temp7 = _mm_add_epi16(value_add, pred_r6);
301 
302     temp8 = _mm_add_epi16(value_add, pred_r7);
303     /*------------------------------------------------------------------*/
304     //Clipping the results to 8 bits
305     sign_reg = _mm_cmpgt_epi16(temp1, zero_8x16b); // sign check
306     temp1 = _mm_and_si128(temp1, sign_reg);
307     sign_reg = _mm_cmpgt_epi16(temp2, zero_8x16b); // sign check
308     temp2 = _mm_and_si128(temp2, sign_reg);
309     sign_reg = _mm_cmpgt_epi16(temp3, zero_8x16b); // sign check
310     temp3 = _mm_and_si128(temp3, sign_reg);
311     sign_reg = _mm_cmpgt_epi16(temp4, zero_8x16b); // sign check
312     temp4 = _mm_and_si128(temp4, sign_reg);
313     sign_reg = _mm_cmpgt_epi16(temp5, zero_8x16b); // sign check
314     temp5 = _mm_and_si128(temp5, sign_reg);
315     sign_reg = _mm_cmpgt_epi16(temp6, zero_8x16b); // sign check
316     temp6 = _mm_and_si128(temp6, sign_reg);
317     sign_reg = _mm_cmpgt_epi16(temp7, zero_8x16b); // sign check
318     temp7 = _mm_and_si128(temp7, sign_reg);
319     sign_reg = _mm_cmpgt_epi16(temp8, zero_8x16b); // sign check
320     temp8 = _mm_and_si128(temp8, sign_reg);
321 
322     temp1 = _mm_packus_epi16(temp1, zero_8x16b);
323     temp2 = _mm_packus_epi16(temp2, zero_8x16b);
324     temp3 = _mm_packus_epi16(temp3, zero_8x16b);
325     temp4 = _mm_packus_epi16(temp4, zero_8x16b);
326     temp5 = _mm_packus_epi16(temp5, zero_8x16b);
327     temp6 = _mm_packus_epi16(temp6, zero_8x16b);
328     temp7 = _mm_packus_epi16(temp7, zero_8x16b);
329     temp8 = _mm_packus_epi16(temp8, zero_8x16b);
330 
331     _mm_storel_epi64((__m128i *)(&pu1_out[0]), temp1);
332     _mm_storel_epi64((__m128i *)(&pu1_out[out_strd]), temp2);
333     _mm_storel_epi64((__m128i *)(&pu1_out[2 * out_strd]), temp3);
334     _mm_storel_epi64((__m128i *)(&pu1_out[3 * out_strd]), temp4);
335     _mm_storel_epi64((__m128i *)(&pu1_out[4 * out_strd]), temp5);
336     _mm_storel_epi64((__m128i *)(&pu1_out[5 * out_strd]), temp6);
337     _mm_storel_epi64((__m128i *)(&pu1_out[6 * out_strd]), temp7);
338     _mm_storel_epi64((__m128i *)(&pu1_out[7 * out_strd]), temp8);
339 }
340 
341 /*
342  ********************************************************************************
343  *
344  * @brief This function reconstructs a 4x4 sub block from quantized chroma resiude and
345  * prediction buffer
346  *
347  * @par Description:
348  *  The quantized residue is first inverse quantized, then inverse transformed.
349  *  This inverse transformed content is added to the prediction buffer to recon-
350  *  struct the end output
351  *
352  * @param[in] pi2_src
353  *  quantized 4x4 block
354  *
355  * @param[in] pu1_pred
356  *  prediction 4x4 block
357  *
358  * @param[out] pu1_out
359  *  reconstructed 4x4 block
360  *
361  * @param[in] src_strd
362  *  quantization buffer stride
363  *
364  * @param[in] pred_strd,
365  *  Prediction buffer stride
366  *
367  * @param[in] out_strd
368  *  recon buffer Stride
369  *
370  * @param[in] pu2_scaling_list
371  *  pointer to scaling list
372  *
373  * @param[in] pu2_norm_adjust
374  *  pointer to inverse scale matrix
375  *
376  * @param[in] u4_qp_div_6
377  *  Floor (qp/6)
378  *
379  * @param[in] pi4_tmp
380  * temporary buffer of size 1*16
381  *
382  * @returns none
383  *
384  * @remarks none
385  *
386  *******************************************************************************
387  */
ih264_iquant_itrans_recon_chroma_4x4_dc_ssse3(WORD16 * pi2_src,UWORD8 * pu1_pred,UWORD8 * pu1_out,WORD32 pred_strd,WORD32 out_strd,const UWORD16 * pu2_iscal_mat,const UWORD16 * pu2_weigh_mat,UWORD32 u4_qp_div_6,WORD16 * pi2_tmp,WORD16 * pi2_dc_src)388 void ih264_iquant_itrans_recon_chroma_4x4_dc_ssse3(WORD16 *pi2_src,
389                                    UWORD8 *pu1_pred,
390                                    UWORD8 *pu1_out,
391                                    WORD32 pred_strd,
392                                    WORD32 out_strd,
393                                    const UWORD16 *pu2_iscal_mat,
394                                    const UWORD16 *pu2_weigh_mat,
395                                    UWORD32 u4_qp_div_6,
396                                    WORD16 *pi2_tmp,
397                                    WORD16 *pi2_dc_src)
398  {
399     WORD16 q0 = pi2_dc_src[0];      // DC value won't be dequantized for chroma inverse transform
400     WORD16 i_macro = ((q0 + 32) >> 6);
401 
402     __m128i pred_r0, pred_r1, pred_r2, pred_r3, sign_reg;
403     __m128i zero_8x16b = _mm_setzero_si128();          // all bits reset to zero
404     __m128i chroma_mask = _mm_set1_epi16 (0xFF);
405     __m128i value_add = _mm_set1_epi16(i_macro);
406     __m128i out_r0, out_r1, out_r2, out_r3;
407 
408     UNUSED (pi2_src);
409     UNUSED (pu2_iscal_mat);
410     UNUSED (pu2_weigh_mat);
411     UNUSED (u4_qp_div_6);
412     UNUSED (pi2_tmp);
413 
414     //Load pred buffer
415     pred_r0 = _mm_loadl_epi64((__m128i *) (&pu1_pred[0])); //p00 p01 p02 p03 0 0 0 0 0 0 0 0 -- all 8 bits
416     pred_r1 = _mm_loadl_epi64((__m128i *) (&pu1_pred[pred_strd])); //p10 p11 p12 p13 0 0 0 0 0 0 0 0 -- all 8 bits
417     pred_r2 = _mm_loadl_epi64((__m128i *) (&pu1_pred[2 * pred_strd])); //p20 p21 p22 p23 0 0 0 0 0 0 0 0 -- all 8 bits
418     pred_r3 = _mm_loadl_epi64((__m128i *) (&pu1_pred[3 * pred_strd])); //p30 p31 p32 p33 0 0 0 0 0 0 0 0 -- all 8 bits
419 
420     pred_r0 = _mm_and_si128(pred_r0, chroma_mask);
421     pred_r1 = _mm_and_si128(pred_r1, chroma_mask);
422     pred_r2 = _mm_and_si128(pred_r2, chroma_mask);
423     pred_r3 = _mm_and_si128(pred_r3, chroma_mask);
424 
425     pred_r0 = _mm_unpacklo_epi64(pred_r0, pred_r1); //p00 p01 p02 p03 p10 p11 p12 p13
426     pred_r2 = _mm_unpacklo_epi64(pred_r2, pred_r3); //p20 p21 p22p p23 p30 p31 p32 p33
427 
428     pred_r0 = _mm_add_epi16(value_add, pred_r0);
429     pred_r2 = _mm_add_epi16(value_add, pred_r2);
430 
431     /*------------------------------------------------------------------*/
432     //Clipping the results to 8 bits
433     sign_reg = _mm_cmpgt_epi16(pred_r0, zero_8x16b);        // sign check
434     pred_r0 = _mm_and_si128(pred_r0, sign_reg);
435     sign_reg = _mm_cmpgt_epi16(pred_r2, zero_8x16b);
436     pred_r2 = _mm_and_si128(pred_r2, sign_reg);
437 
438     pred_r0 = _mm_packus_epi16(pred_r0, pred_r2);
439     pred_r1 = _mm_srli_si128(pred_r0, 4);
440     pred_r2 = _mm_srli_si128(pred_r1, 4);
441     pred_r3 = _mm_srli_si128(pred_r2, 4);
442 
443     pred_r0 = _mm_unpacklo_epi8(pred_r0, zero_8x16b); //p00 p01 p02 p03 -- all 16 bits
444     pred_r1 = _mm_unpacklo_epi8(pred_r1, zero_8x16b); //p10 p11 p12 p13 -- all 16 bits
445     pred_r2 = _mm_unpacklo_epi8(pred_r2, zero_8x16b); //p20 p21 p22 p23 -- all 16 bits
446     pred_r3 = _mm_unpacklo_epi8(pred_r3, zero_8x16b); //p30 p31 p32 p33 -- all 16 bits
447 
448     chroma_mask = _mm_set1_epi16 (0xFF00);
449     out_r0 = _mm_loadl_epi64((__m128i *) (&pu1_out[0]));
450     out_r1 = _mm_loadl_epi64((__m128i *) (&pu1_out[out_strd]));
451     out_r2 = _mm_loadl_epi64((__m128i *) (&pu1_out[2 * out_strd]));
452     out_r3 = _mm_loadl_epi64((__m128i *) (&pu1_out[3 * out_strd]));
453 
454     out_r0 = _mm_and_si128(out_r0, chroma_mask);
455     out_r1 = _mm_and_si128(out_r1, chroma_mask);
456     out_r2 = _mm_and_si128(out_r2, chroma_mask);
457     out_r3 = _mm_and_si128(out_r3, chroma_mask);
458 
459     out_r0 = _mm_add_epi8(out_r0, pred_r0);
460     out_r1 = _mm_add_epi8(out_r1, pred_r1);
461     out_r2 = _mm_add_epi8(out_r2, pred_r2);
462     out_r3 = _mm_add_epi8(out_r3, pred_r3);
463 
464     _mm_storel_epi64((__m128i *)(&pu1_out[0]), out_r0);
465     _mm_storel_epi64((__m128i *)(&pu1_out[out_strd]), out_r1);
466     _mm_storel_epi64((__m128i *)(&pu1_out[2 * out_strd]), out_r2);
467     _mm_storel_epi64((__m128i *)(&pu1_out[3 * out_strd]), out_r3);
468 }
469 
470 
471