1 /*
2  *  Copyright (c) 2015 The WebM project authors. All Rights Reserved.
3  *
4  *  Use of this source code is governed by a BSD-style license
5  *  that can be found in the LICENSE file in the root of the source
6  *  tree. An additional intellectual property rights grant can be found
7  *  in the file PATENTS.  All contributing project authors may
8  *  be found in the AUTHORS file in the root of the source tree.
9  */
10 
11 #include "./vp9_rtcd.h"
12 #include "vpx_dsp/mips/macros_msa.h"
13 
14 #define BLOCK_ERROR_BLOCKSIZE_MSA(BSize)                                   \
15 static int64_t block_error_##BSize##size_msa(const int16_t *coeff_ptr,     \
16                                              const int16_t *dq_coeff_ptr,  \
17                                              int64_t *ssz) {               \
18   int64_t err = 0;                                                         \
19   uint32_t loop_cnt;                                                       \
20   v8i16 coeff, dq_coeff, coeff_r_h, coeff_l_h;                             \
21   v4i32 diff_r, diff_l, coeff_r_w, coeff_l_w;                              \
22   v2i64 sq_coeff_r, sq_coeff_l;                                            \
23   v2i64 err0, err_dup0, err1, err_dup1;                                    \
24                                                                            \
25   coeff = LD_SH(coeff_ptr);                                                \
26   dq_coeff = LD_SH(dq_coeff_ptr);                                          \
27   UNPCK_SH_SW(coeff, coeff_r_w, coeff_l_w);                                \
28   ILVRL_H2_SH(coeff, dq_coeff, coeff_r_h, coeff_l_h);                      \
29   HSUB_UH2_SW(coeff_r_h, coeff_l_h, diff_r, diff_l);                       \
30   DOTP_SW2_SD(coeff_r_w, coeff_l_w, coeff_r_w, coeff_l_w,                  \
31               sq_coeff_r, sq_coeff_l);                                     \
32   DOTP_SW2_SD(diff_r, diff_l, diff_r, diff_l, err0, err1);                 \
33                                                                            \
34   coeff = LD_SH(coeff_ptr + 8);                                            \
35   dq_coeff = LD_SH(dq_coeff_ptr + 8);                                      \
36   UNPCK_SH_SW(coeff, coeff_r_w, coeff_l_w);                                \
37   ILVRL_H2_SH(coeff, dq_coeff, coeff_r_h, coeff_l_h);                      \
38   HSUB_UH2_SW(coeff_r_h, coeff_l_h, diff_r, diff_l);                       \
39   DPADD_SD2_SD(coeff_r_w, coeff_l_w, sq_coeff_r, sq_coeff_l);              \
40   DPADD_SD2_SD(diff_r, diff_l, err0, err1);                                \
41                                                                            \
42   coeff_ptr += 16;                                                         \
43   dq_coeff_ptr += 16;                                                      \
44                                                                            \
45   for (loop_cnt = ((BSize >> 4) - 1); loop_cnt--;) {                       \
46     coeff = LD_SH(coeff_ptr);                                              \
47     dq_coeff = LD_SH(dq_coeff_ptr);                                        \
48     UNPCK_SH_SW(coeff, coeff_r_w, coeff_l_w);                              \
49     ILVRL_H2_SH(coeff, dq_coeff, coeff_r_h, coeff_l_h);                    \
50     HSUB_UH2_SW(coeff_r_h, coeff_l_h, diff_r, diff_l);                     \
51     DPADD_SD2_SD(coeff_r_w, coeff_l_w, sq_coeff_r, sq_coeff_l);            \
52     DPADD_SD2_SD(diff_r, diff_l, err0, err1);                              \
53                                                                            \
54     coeff = LD_SH(coeff_ptr + 8);                                          \
55     dq_coeff = LD_SH(dq_coeff_ptr + 8);                                    \
56     UNPCK_SH_SW(coeff, coeff_r_w, coeff_l_w);                              \
57     ILVRL_H2_SH(coeff, dq_coeff, coeff_r_h, coeff_l_h);                    \
58     HSUB_UH2_SW(coeff_r_h, coeff_l_h, diff_r, diff_l);                     \
59     DPADD_SD2_SD(coeff_r_w, coeff_l_w, sq_coeff_r, sq_coeff_l);            \
60     DPADD_SD2_SD(diff_r, diff_l, err0, err1);                              \
61                                                                            \
62     coeff_ptr += 16;                                                       \
63     dq_coeff_ptr += 16;                                                    \
64   }                                                                        \
65                                                                            \
66   err_dup0 = __msa_splati_d(sq_coeff_r, 1);                                \
67   err_dup1 = __msa_splati_d(sq_coeff_l, 1);                                \
68   sq_coeff_r += err_dup0;                                                  \
69   sq_coeff_l += err_dup1;                                                  \
70   *ssz = __msa_copy_s_d(sq_coeff_r, 0);                                    \
71   *ssz += __msa_copy_s_d(sq_coeff_l, 0);                                   \
72                                                                            \
73   err_dup0 = __msa_splati_d(err0, 1);                                      \
74   err_dup1 = __msa_splati_d(err1, 1);                                      \
75   err0 += err_dup0;                                                        \
76   err1 += err_dup1;                                                        \
77   err = __msa_copy_s_d(err0, 0);                                           \
78   err += __msa_copy_s_d(err1, 0);                                          \
79                                                                            \
80   return err;                                                              \
81 }
82 
83 BLOCK_ERROR_BLOCKSIZE_MSA(16);
84 BLOCK_ERROR_BLOCKSIZE_MSA(64);
85 BLOCK_ERROR_BLOCKSIZE_MSA(256);
86 BLOCK_ERROR_BLOCKSIZE_MSA(1024);
87 
vp9_block_error_msa(const tran_low_t * coeff_ptr,const tran_low_t * dq_coeff_ptr,intptr_t blk_size,int64_t * ssz)88 int64_t vp9_block_error_msa(const tran_low_t *coeff_ptr,
89                             const tran_low_t *dq_coeff_ptr,
90                             intptr_t blk_size, int64_t *ssz) {
91   int64_t err;
92   const int16_t *coeff = (const int16_t *)coeff_ptr;
93   const int16_t *dq_coeff = (const int16_t *)dq_coeff_ptr;
94 
95   switch (blk_size) {
96     case 16:
97       err = block_error_16size_msa(coeff, dq_coeff, ssz);
98       break;
99     case 64:
100       err = block_error_64size_msa(coeff, dq_coeff, ssz);
101       break;
102     case 256:
103       err = block_error_256size_msa(coeff, dq_coeff, ssz);
104       break;
105     case 1024:
106       err = block_error_1024size_msa(coeff, dq_coeff, ssz);
107       break;
108     default:
109       err = vp9_block_error_c(coeff_ptr, dq_coeff_ptr, blk_size, ssz);
110       break;
111   }
112 
113   return err;
114 }
115