• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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  #ifndef VPX_DSP_MIPS_FWD_TXFM_MSA_H_
12  #define VPX_DSP_MIPS_FWD_TXFM_MSA_H_
13  
14  #include "vpx_dsp/mips/txfm_macros_msa.h"
15  #include "vpx_dsp/txfm_common.h"
16  
17  #define LD_HADD(psrc, stride) ({                                      \
18    v8i16 in0_m, in1_m, in2_m, in3_m, in4_m, in5_m, in6_m, in7_m;       \
19    v4i32 vec_w_m;                                                      \
20                                                                        \
21    LD_SH4((psrc), stride, in0_m, in1_m, in2_m, in3_m);                 \
22    ADD2(in0_m, in1_m, in2_m, in3_m, in0_m, in2_m);                     \
23    LD_SH4(((psrc) + 4 * stride), stride, in4_m, in5_m, in6_m, in7_m);  \
24    ADD4(in4_m, in5_m, in6_m, in7_m, in0_m, in2_m, in4_m, in6_m,        \
25         in4_m, in6_m, in0_m, in4_m);                                   \
26    in0_m += in4_m;                                                     \
27                                                                        \
28    vec_w_m = __msa_hadd_s_w(in0_m, in0_m);                             \
29    HADD_SW_S32(vec_w_m);                                               \
30  })
31  
32  #define VP9_FDCT4(in0, in1, in2, in3, out0, out1, out2, out3) {     \
33    v8i16 cnst0_m, cnst1_m, cnst2_m, cnst3_m;                         \
34    v8i16 vec0_m, vec1_m, vec2_m, vec3_m;                             \
35    v4i32 vec4_m, vec5_m, vec6_m, vec7_m;                             \
36    v8i16 coeff_m = { cospi_16_64, -cospi_16_64, cospi_8_64,          \
37                      cospi_24_64, -cospi_8_64, 0, 0, 0 };            \
38                                                                      \
39    BUTTERFLY_4(in0, in1, in2, in3, vec0_m, vec1_m, vec2_m, vec3_m);  \
40    ILVR_H2_SH(vec1_m, vec0_m, vec3_m, vec2_m, vec0_m, vec2_m);       \
41    SPLATI_H2_SH(coeff_m, 0, 1, cnst0_m, cnst1_m);                    \
42    cnst1_m = __msa_ilvev_h(cnst1_m, cnst0_m);                        \
43    vec5_m = __msa_dotp_s_w(vec0_m, cnst1_m);                         \
44                                                                      \
45    SPLATI_H2_SH(coeff_m, 4, 3, cnst2_m, cnst3_m);                    \
46    cnst2_m = __msa_ilvev_h(cnst3_m, cnst2_m);                        \
47    vec7_m = __msa_dotp_s_w(vec2_m, cnst2_m);                         \
48                                                                      \
49    vec4_m = __msa_dotp_s_w(vec0_m, cnst0_m);                         \
50    cnst2_m = __msa_splati_h(coeff_m, 2);                             \
51    cnst2_m = __msa_ilvev_h(cnst2_m, cnst3_m);                        \
52    vec6_m = __msa_dotp_s_w(vec2_m, cnst2_m);                         \
53                                                                      \
54    SRARI_W4_SW(vec4_m, vec5_m, vec6_m, vec7_m, DCT_CONST_BITS);      \
55    PCKEV_H4_SH(vec4_m, vec4_m, vec5_m, vec5_m, vec6_m, vec6_m,       \
56                vec7_m, vec7_m, out0, out2, out1, out3);              \
57  }
58  
59  #define SRLI_AVE_S_4V_H(in0, in1, in2, in3, in4, in5, in6, in7) {        \
60    v8i16 vec0_m, vec1_m, vec2_m, vec3_m, vec4_m, vec5_m, vec6_m, vec7_m;  \
61                                                                           \
62    SRLI_H4_SH(in0, in1, in2, in3, vec0_m, vec1_m, vec2_m, vec3_m, 15);    \
63    SRLI_H4_SH(in4, in5, in6, in7, vec4_m, vec5_m, vec6_m, vec7_m, 15);    \
64    AVE_SH4_SH(vec0_m, in0, vec1_m, in1, vec2_m, in2, vec3_m, in3,         \
65               in0, in1, in2, in3);                                        \
66    AVE_SH4_SH(vec4_m, in4, vec5_m, in5, vec6_m, in6, vec7_m, in7,         \
67               in4, in5, in6, in7);                                        \
68  }
69  
70  #define VP9_FDCT8(in0, in1, in2, in3, in4, in5, in6, in7,            \
71                    out0, out1, out2, out3, out4, out5, out6, out7) {  \
72    v8i16 s0_m, s1_m, s2_m, s3_m, s4_m, s5_m, s6_m;                    \
73    v8i16 s7_m, x0_m, x1_m, x2_m, x3_m;                                \
74    v8i16 coeff_m = { cospi_16_64, -cospi_16_64, cospi_8_64,           \
75                      cospi_24_64, cospi_4_64, cospi_28_64,            \
76                      cospi_12_64, cospi_20_64 };                      \
77                                                                       \
78    /* FDCT stage1 */                                                  \
79    BUTTERFLY_8(in0, in1, in2, in3, in4, in5, in6, in7,                \
80                s0_m, s1_m, s2_m, s3_m, s4_m, s5_m, s6_m, s7_m);       \
81    BUTTERFLY_4(s0_m, s1_m, s2_m, s3_m, x0_m, x1_m, x2_m, x3_m);       \
82    ILVL_H2_SH(x1_m, x0_m, x3_m, x2_m, s0_m, s2_m);                    \
83    ILVR_H2_SH(x1_m, x0_m, x3_m, x2_m, s1_m, s3_m);                    \
84    SPLATI_H2_SH(coeff_m, 0, 1, x0_m, x1_m);                           \
85    x1_m = __msa_ilvev_h(x1_m, x0_m);                                  \
86    out4 = DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x1_m);                    \
87                                                                       \
88    SPLATI_H2_SH(coeff_m, 2, 3, x2_m, x3_m);                           \
89    x2_m = -x2_m;                                                      \
90    x2_m = __msa_ilvev_h(x3_m, x2_m);                                  \
91    out6 = DOT_SHIFT_RIGHT_PCK_H(s2_m, s3_m, x2_m);                    \
92                                                                       \
93    out0 = DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x0_m);                    \
94    x2_m = __msa_splati_h(coeff_m, 2);                                 \
95    x2_m = __msa_ilvev_h(x2_m, x3_m);                                  \
96    out2 = DOT_SHIFT_RIGHT_PCK_H(s2_m, s3_m, x2_m);                    \
97                                                                       \
98    /* stage2 */                                                       \
99    ILVRL_H2_SH(s5_m, s6_m, s1_m, s0_m);                               \
100                                                                       \
101    s6_m = DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x0_m);                    \
102    s5_m = DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x1_m);                    \
103                                                                       \
104    /* stage3 */                                                       \
105    BUTTERFLY_4(s4_m, s7_m, s6_m, s5_m, x0_m, x3_m, x2_m, x1_m);       \
106                                                                       \
107    /* stage4 */                                                       \
108    ILVL_H2_SH(x3_m, x0_m, x2_m, x1_m, s4_m, s6_m);                    \
109    ILVR_H2_SH(x3_m, x0_m, x2_m, x1_m, s5_m, s7_m);                    \
110                                                                       \
111    SPLATI_H2_SH(coeff_m, 4, 5, x0_m, x1_m);                           \
112    x1_m = __msa_ilvev_h(x0_m, x1_m);                                  \
113    out1 = DOT_SHIFT_RIGHT_PCK_H(s4_m, s5_m, x1_m);                    \
114                                                                       \
115    SPLATI_H2_SH(coeff_m, 6, 7, x2_m, x3_m);                           \
116    x2_m = __msa_ilvev_h(x3_m, x2_m);                                  \
117    out5 = DOT_SHIFT_RIGHT_PCK_H(s6_m, s7_m, x2_m);                    \
118                                                                       \
119    x1_m = __msa_splati_h(coeff_m, 5);                                 \
120    x0_m = -x0_m;                                                      \
121    x0_m = __msa_ilvev_h(x1_m, x0_m);                                  \
122    out7 = DOT_SHIFT_RIGHT_PCK_H(s4_m, s5_m, x0_m);                    \
123                                                                       \
124    x2_m = __msa_splati_h(coeff_m, 6);                                 \
125    x3_m = -x3_m;                                                      \
126    x2_m = __msa_ilvev_h(x2_m, x3_m);                                  \
127    out3 = DOT_SHIFT_RIGHT_PCK_H(s6_m, s7_m, x2_m);                    \
128  }
129  
130  #define FDCT8x16_EVEN(in0, in1, in2, in3, in4, in5, in6, in7,                \
131                        out0, out1, out2, out3, out4, out5, out6, out7) {      \
132    v8i16 s0_m, s1_m, s2_m, s3_m, s4_m, s5_m, s6_m, s7_m;                      \
133    v8i16 x0_m, x1_m, x2_m, x3_m;                                              \
134    v8i16 coeff_m = { cospi_16_64, -cospi_16_64, cospi_8_64, cospi_24_64,      \
135                      cospi_4_64, cospi_28_64, cospi_12_64, cospi_20_64 };     \
136                                                                               \
137    /* FDCT stage1 */                                                          \
138    BUTTERFLY_8(in0, in1, in2, in3, in4, in5, in6, in7,                        \
139                s0_m, s1_m, s2_m, s3_m, s4_m, s5_m, s6_m, s7_m);               \
140    BUTTERFLY_4(s0_m, s1_m, s2_m, s3_m, x0_m, x1_m, x2_m, x3_m);               \
141    ILVL_H2_SH(x1_m, x0_m, x3_m, x2_m, s0_m, s2_m);                            \
142    ILVR_H2_SH(x1_m, x0_m, x3_m, x2_m, s1_m, s3_m);                            \
143    SPLATI_H2_SH(coeff_m, 0, 1, x0_m, x1_m);                                   \
144    x1_m = __msa_ilvev_h(x1_m, x0_m);                                          \
145    out4 = DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x1_m);                            \
146                                                                               \
147    SPLATI_H2_SH(coeff_m, 2, 3, x2_m, x3_m);                                   \
148    x2_m = -x2_m;                                                              \
149    x2_m = __msa_ilvev_h(x3_m, x2_m);                                          \
150    out6 = DOT_SHIFT_RIGHT_PCK_H(s2_m, s3_m, x2_m);                            \
151                                                                               \
152    out0 = DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x0_m);                            \
153    x2_m = __msa_splati_h(coeff_m, 2);                                         \
154    x2_m = __msa_ilvev_h(x2_m, x3_m);                                          \
155    out2 = DOT_SHIFT_RIGHT_PCK_H(s2_m, s3_m, x2_m);                            \
156                                                                               \
157    /* stage2 */                                                               \
158    ILVRL_H2_SH(s5_m, s6_m, s1_m, s0_m);                                       \
159                                                                               \
160    s6_m = DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x0_m);                            \
161    s5_m = DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x1_m);                            \
162                                                                               \
163    /* stage3 */                                                               \
164    BUTTERFLY_4(s4_m, s7_m, s6_m, s5_m, x0_m, x3_m, x2_m, x1_m);               \
165                                                                               \
166    /* stage4 */                                                               \
167    ILVL_H2_SH(x3_m, x0_m, x2_m, x1_m, s4_m, s6_m);                            \
168    ILVR_H2_SH(x3_m, x0_m, x2_m, x1_m, s5_m, s7_m);                            \
169                                                                               \
170    SPLATI_H2_SH(coeff_m, 4, 5, x0_m, x1_m);                                   \
171    x1_m = __msa_ilvev_h(x0_m, x1_m);                                          \
172    out1 = DOT_SHIFT_RIGHT_PCK_H(s4_m, s5_m, x1_m);                            \
173                                                                               \
174    SPLATI_H2_SH(coeff_m, 6, 7, x2_m, x3_m);                                   \
175    x2_m = __msa_ilvev_h(x3_m, x2_m);                                          \
176    out5 = DOT_SHIFT_RIGHT_PCK_H(s6_m, s7_m, x2_m);                            \
177                                                                               \
178    x1_m = __msa_splati_h(coeff_m, 5);                                         \
179    x0_m = -x0_m;                                                              \
180    x0_m = __msa_ilvev_h(x1_m, x0_m);                                          \
181    out7 = DOT_SHIFT_RIGHT_PCK_H(s4_m, s5_m, x0_m);                            \
182                                                                               \
183    x2_m = __msa_splati_h(coeff_m, 6);                                         \
184    x3_m = -x3_m;                                                              \
185    x2_m = __msa_ilvev_h(x2_m, x3_m);                                          \
186    out3 = DOT_SHIFT_RIGHT_PCK_H(s6_m, s7_m, x2_m);                            \
187  }
188  
189  #define FDCT8x16_ODD(input0, input1, input2, input3,               \
190                       input4, input5, input6, input7,               \
191                       out1, out3, out5, out7,                       \
192                       out9, out11, out13, out15) {                  \
193    v8i16 stp21_m, stp22_m, stp23_m, stp24_m, stp25_m, stp26_m;      \
194    v8i16 stp30_m, stp31_m, stp32_m, stp33_m, stp34_m, stp35_m;      \
195    v8i16 stp36_m, stp37_m, vec0_m, vec1_m;                          \
196    v8i16 vec2_m, vec3_m, vec4_m, vec5_m, vec6_m;                    \
197    v8i16 cnst0_m, cnst1_m, cnst4_m, cnst5_m;                        \
198    v8i16 coeff_m = { cospi_16_64, -cospi_16_64, cospi_8_64,         \
199                      cospi_24_64, -cospi_8_64, -cospi_24_64,        \
200                      cospi_12_64, cospi_20_64 };                    \
201    v8i16 coeff1_m = { cospi_2_64, cospi_30_64, cospi_14_64,         \
202                       cospi_18_64, cospi_10_64, cospi_22_64,        \
203                       cospi_6_64, cospi_26_64 };                    \
204    v8i16 coeff2_m = { -cospi_2_64, -cospi_10_64, -cospi_18_64,      \
205                       -cospi_26_64, 0, 0, 0, 0 };                   \
206                                                                     \
207    /* stp 1 */                                                      \
208    ILVL_H2_SH(input2, input5, input3, input4, vec2_m, vec4_m);      \
209    ILVR_H2_SH(input2, input5, input3, input4, vec3_m, vec5_m);      \
210                                                                     \
211    cnst4_m = __msa_splati_h(coeff_m, 0);                            \
212    stp25_m = DOT_SHIFT_RIGHT_PCK_H(vec2_m, vec3_m, cnst4_m);        \
213                                                                     \
214    cnst5_m = __msa_splati_h(coeff_m, 1);                            \
215    cnst5_m = __msa_ilvev_h(cnst5_m, cnst4_m);                       \
216    stp22_m = DOT_SHIFT_RIGHT_PCK_H(vec2_m, vec3_m, cnst5_m);        \
217    stp24_m = DOT_SHIFT_RIGHT_PCK_H(vec4_m, vec5_m, cnst4_m);        \
218    stp23_m = DOT_SHIFT_RIGHT_PCK_H(vec4_m, vec5_m, cnst5_m);        \
219                                                                     \
220    /* stp2 */                                                       \
221    BUTTERFLY_4(input0, input1, stp22_m, stp23_m,                    \
222                stp30_m, stp31_m, stp32_m, stp33_m);                 \
223    BUTTERFLY_4(input7, input6, stp25_m, stp24_m,                    \
224                stp37_m, stp36_m, stp35_m, stp34_m);                 \
225                                                                     \
226    ILVL_H2_SH(stp36_m, stp31_m, stp35_m, stp32_m, vec2_m, vec4_m);  \
227    ILVR_H2_SH(stp36_m, stp31_m, stp35_m, stp32_m, vec3_m, vec5_m);  \
228                                                                     \
229    SPLATI_H2_SH(coeff_m, 2, 3, cnst0_m, cnst1_m);                   \
230    cnst0_m = __msa_ilvev_h(cnst0_m, cnst1_m);                       \
231    stp26_m = DOT_SHIFT_RIGHT_PCK_H(vec2_m, vec3_m, cnst0_m);        \
232                                                                     \
233    cnst0_m = __msa_splati_h(coeff_m, 4);                            \
234    cnst1_m = __msa_ilvev_h(cnst1_m, cnst0_m);                       \
235    stp21_m = DOT_SHIFT_RIGHT_PCK_H(vec2_m, vec3_m, cnst1_m);        \
236                                                                     \
237    SPLATI_H2_SH(coeff_m, 5, 2, cnst0_m, cnst1_m);                   \
238    cnst1_m = __msa_ilvev_h(cnst0_m, cnst1_m);                       \
239    stp25_m = DOT_SHIFT_RIGHT_PCK_H(vec4_m, vec5_m, cnst1_m);        \
240                                                                     \
241    cnst0_m = __msa_splati_h(coeff_m, 3);                            \
242    cnst1_m = __msa_ilvev_h(cnst1_m, cnst0_m);                       \
243    stp22_m = DOT_SHIFT_RIGHT_PCK_H(vec4_m, vec5_m, cnst1_m);        \
244                                                                     \
245    /* stp4 */                                                       \
246    BUTTERFLY_4(stp30_m, stp37_m, stp26_m, stp21_m,                  \
247                vec6_m, vec2_m, vec4_m, vec5_m);                     \
248    BUTTERFLY_4(stp33_m, stp34_m, stp25_m, stp22_m,                  \
249                stp21_m, stp23_m, stp24_m, stp31_m);                 \
250                                                                     \
251    ILVRL_H2_SH(vec2_m, vec6_m, vec1_m, vec0_m);                     \
252    SPLATI_H2_SH(coeff1_m, 0, 1, cnst0_m, cnst1_m);                  \
253    cnst0_m = __msa_ilvev_h(cnst0_m, cnst1_m);                       \
254                                                                     \
255    out1 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_m);           \
256                                                                     \
257    cnst0_m = __msa_splati_h(coeff2_m, 0);                           \
258    cnst0_m = __msa_ilvev_h(cnst1_m, cnst0_m);                       \
259    out15 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_m);          \
260                                                                     \
261    ILVRL_H2_SH(vec4_m, vec5_m, vec1_m, vec0_m);                     \
262    SPLATI_H2_SH(coeff1_m, 2, 3, cnst0_m, cnst1_m);                  \
263    cnst1_m = __msa_ilvev_h(cnst1_m, cnst0_m);                       \
264                                                                     \
265    out9 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst1_m);           \
266                                                                     \
267    cnst1_m = __msa_splati_h(coeff2_m, 2);                           \
268    cnst0_m = __msa_ilvev_h(cnst0_m, cnst1_m);                       \
269    out7 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_m);           \
270                                                                     \
271    ILVRL_H2_SH(stp23_m, stp21_m, vec1_m, vec0_m);                   \
272    SPLATI_H2_SH(coeff1_m, 4, 5, cnst0_m, cnst1_m);                  \
273    cnst0_m = __msa_ilvev_h(cnst0_m, cnst1_m);                       \
274    out5 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_m);           \
275                                                                     \
276    cnst0_m = __msa_splati_h(coeff2_m, 1);                           \
277    cnst0_m = __msa_ilvev_h(cnst1_m, cnst0_m);                       \
278    out11 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_m);          \
279                                                                     \
280    ILVRL_H2_SH(stp24_m, stp31_m, vec1_m, vec0_m);                   \
281    SPLATI_H2_SH(coeff1_m, 6, 7, cnst0_m, cnst1_m);                  \
282    cnst1_m = __msa_ilvev_h(cnst1_m, cnst0_m);                       \
283                                                                     \
284    out13 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst1_m);          \
285                                                                     \
286    cnst1_m = __msa_splati_h(coeff2_m, 3);                           \
287    cnst0_m = __msa_ilvev_h(cnst0_m, cnst1_m);                       \
288    out3 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_m);           \
289  }
290  
291  #define FDCT_POSTPROC_2V_NEG_H(vec0, vec1) {      \
292    v8i16 tp0_m, tp1_m;                             \
293    v8i16 one_m = __msa_ldi_h(1);                   \
294                                                    \
295    tp0_m = __msa_clti_s_h(vec0, 0);                \
296    tp1_m = __msa_clti_s_h(vec1, 0);                \
297    vec0 += 1;                                      \
298    vec1 += 1;                                      \
299    tp0_m = one_m & tp0_m;                          \
300    tp1_m = one_m & tp1_m;                          \
301    vec0 += tp0_m;                                  \
302    vec1 += tp1_m;                                  \
303    vec0 >>= 2;                                     \
304    vec1 >>= 2;                                     \
305  }
306  
307  #define FDCT32_POSTPROC_NEG_W(vec) {      \
308    v4i32 temp_m;                           \
309    v4i32 one_m = __msa_ldi_w(1);           \
310                                            \
311    temp_m = __msa_clti_s_w(vec, 0);        \
312    vec += 1;                               \
313    temp_m = one_m & temp_m;                \
314    vec += temp_m;                          \
315    vec >>= 2;                              \
316  }
317  
318  #define FDCT32_POSTPROC_2V_POS_H(vec0, vec1) {      \
319    v8i16 tp0_m, tp1_m;                               \
320    v8i16 one = __msa_ldi_h(1);                       \
321                                                      \
322    tp0_m = __msa_clei_s_h(vec0, 0);                  \
323    tp1_m = __msa_clei_s_h(vec1, 0);                  \
324    tp0_m = (v8i16)__msa_xori_b((v16u8)tp0_m, 255);   \
325    tp1_m = (v8i16)__msa_xori_b((v16u8)tp1_m, 255);   \
326    vec0 += 1;                                        \
327    vec1 += 1;                                        \
328    tp0_m = one & tp0_m;                              \
329    tp1_m = one & tp1_m;                              \
330    vec0 += tp0_m;                                    \
331    vec1 += tp1_m;                                    \
332    vec0 >>= 2;                                       \
333    vec1 >>= 2;                                       \
334  }
335  
336  #define DOTP_CONST_PAIR_W(reg0_left, reg1_left, reg0_right,      \
337                            reg1_right, const0, const1,            \
338                            out0, out1, out2, out3) {              \
339    v4i32 s0_m, s1_m, s2_m, s3_m, s4_m, s5_m, s6_m, s7_m;          \
340    v2i64 tp0_m, tp1_m, tp2_m, tp3_m;                              \
341    v4i32 k0_m = __msa_fill_w((int32_t) const0);                   \
342                                                                   \
343    s0_m = __msa_fill_w((int32_t) const1);                         \
344    k0_m = __msa_ilvev_w(s0_m, k0_m);                              \
345                                                                   \
346    ILVRL_W2_SW(-reg1_left, reg0_left, s1_m, s0_m);                \
347    ILVRL_W2_SW(reg0_left, reg1_left, s3_m, s2_m);                 \
348    ILVRL_W2_SW(-reg1_right, reg0_right, s5_m, s4_m);              \
349    ILVRL_W2_SW(reg0_right, reg1_right, s7_m, s6_m);               \
350                                                                   \
351    DOTP_SW2_SD(s0_m, s1_m, k0_m, k0_m, tp0_m, tp1_m);             \
352    DOTP_SW2_SD(s4_m, s5_m, k0_m, k0_m, tp2_m, tp3_m);             \
353    tp0_m = __msa_srari_d(tp0_m, DCT_CONST_BITS);                  \
354    tp1_m = __msa_srari_d(tp1_m, DCT_CONST_BITS);                  \
355    tp2_m = __msa_srari_d(tp2_m, DCT_CONST_BITS);                  \
356    tp3_m = __msa_srari_d(tp3_m, DCT_CONST_BITS);                  \
357    out0 = __msa_pckev_w((v4i32)tp0_m, (v4i32)tp1_m);              \
358    out1 = __msa_pckev_w((v4i32)tp2_m, (v4i32)tp3_m);              \
359                                                                   \
360    DOTP_SW2_SD(s2_m, s3_m, k0_m, k0_m, tp0_m, tp1_m);             \
361    DOTP_SW2_SD(s6_m, s7_m, k0_m, k0_m, tp2_m, tp3_m);             \
362    tp0_m = __msa_srari_d(tp0_m, DCT_CONST_BITS);                  \
363    tp1_m = __msa_srari_d(tp1_m, DCT_CONST_BITS);                  \
364    tp2_m = __msa_srari_d(tp2_m, DCT_CONST_BITS);                  \
365    tp3_m = __msa_srari_d(tp3_m, DCT_CONST_BITS);                  \
366    out2 = __msa_pckev_w((v4i32)tp0_m, (v4i32)tp1_m);              \
367    out3 = __msa_pckev_w((v4i32)tp2_m, (v4i32)tp3_m);              \
368  }
369  
370  void fdct8x16_1d_column(const int16_t *input, int16_t *tmp_ptr,
371                          int32_t src_stride);
372  void fdct16x8_1d_row(int16_t *input, int16_t *output);
373  #endif  // VPX_DSP_MIPS_FWD_TXFM_MSA_H_
374