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 "./vpx_dsp_rtcd.h"
12 #include "vpx_dsp/mips/fwd_txfm_msa.h"
13
vpx_fdct8x8_1_msa(const int16_t * input,tran_low_t * out,int32_t stride)14 void vpx_fdct8x8_1_msa(const int16_t *input, tran_low_t *out, int32_t stride) {
15 v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
16 v4i32 vec_w;
17
18 LD_SH8(input, stride, in0, in1, in2, in3, in4, in5, in6, in7);
19 ADD4(in0, in1, in2, in3, in4, in5, in6, in7, in0, in2, in4, in6);
20 ADD2(in0, in2, in4, in6, in0, in4);
21 vec_w = __msa_hadd_s_w(in0, in0);
22 vec_w += __msa_hadd_s_w(in4, in4);
23 out[0] = HADD_SW_S32(vec_w);
24 out[1] = 0;
25 }
26
27 #if !CONFIG_VP9_HIGHBITDEPTH
fdct8x16_1d_column(const int16_t * input,int16_t * tmp_ptr,int32_t src_stride)28 void fdct8x16_1d_column(const int16_t *input, int16_t *tmp_ptr,
29 int32_t src_stride) {
30 v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
31 v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
32 v8i16 in8, in9, in10, in11, in12, in13, in14, in15;
33 v8i16 stp21, stp22, stp23, stp24, stp25, stp26, stp30;
34 v8i16 stp31, stp32, stp33, stp34, stp35, stp36, stp37;
35 v8i16 vec0, vec1, vec2, vec3, vec4, vec5, cnst0, cnst1, cnst4, cnst5;
36 v8i16 coeff = { cospi_16_64, -cospi_16_64, cospi_8_64, cospi_24_64,
37 -cospi_8_64, -cospi_24_64, cospi_12_64, cospi_20_64 };
38 v8i16 coeff1 = { cospi_2_64, cospi_30_64, cospi_14_64, cospi_18_64,
39 cospi_10_64, cospi_22_64, cospi_6_64, cospi_26_64 };
40 v8i16 coeff2 = {
41 -cospi_2_64, -cospi_10_64, -cospi_18_64, -cospi_26_64, 0, 0, 0, 0
42 };
43
44 LD_SH16(input, src_stride, in0, in1, in2, in3, in4, in5, in6, in7, in8, in9,
45 in10, in11, in12, in13, in14, in15);
46 SLLI_4V(in0, in1, in2, in3, 2);
47 SLLI_4V(in4, in5, in6, in7, 2);
48 SLLI_4V(in8, in9, in10, in11, 2);
49 SLLI_4V(in12, in13, in14, in15, 2);
50 ADD4(in0, in15, in1, in14, in2, in13, in3, in12, tmp0, tmp1, tmp2, tmp3);
51 ADD4(in4, in11, in5, in10, in6, in9, in7, in8, tmp4, tmp5, tmp6, tmp7);
52 FDCT8x16_EVEN(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp0, tmp1,
53 tmp2, tmp3, tmp4, tmp5, tmp6, tmp7);
54 ST_SH8(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp_ptr, 32);
55 SUB4(in0, in15, in1, in14, in2, in13, in3, in12, in15, in14, in13, in12);
56 SUB4(in4, in11, in5, in10, in6, in9, in7, in8, in11, in10, in9, in8);
57
58 tmp_ptr += 16;
59
60 /* stp 1 */
61 ILVL_H2_SH(in10, in13, in11, in12, vec2, vec4);
62 ILVR_H2_SH(in10, in13, in11, in12, vec3, vec5);
63
64 cnst4 = __msa_splati_h(coeff, 0);
65 stp25 = DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst4);
66
67 cnst5 = __msa_splati_h(coeff, 1);
68 cnst5 = __msa_ilvev_h(cnst5, cnst4);
69 stp22 = DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst5);
70 stp24 = DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst4);
71 stp23 = DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst5);
72
73 /* stp2 */
74 BUTTERFLY_4(in8, in9, stp22, stp23, stp30, stp31, stp32, stp33);
75 BUTTERFLY_4(in15, in14, stp25, stp24, stp37, stp36, stp35, stp34);
76 ILVL_H2_SH(stp36, stp31, stp35, stp32, vec2, vec4);
77 ILVR_H2_SH(stp36, stp31, stp35, stp32, vec3, vec5);
78 SPLATI_H2_SH(coeff, 2, 3, cnst0, cnst1);
79 cnst0 = __msa_ilvev_h(cnst0, cnst1);
80 stp26 = DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst0);
81
82 cnst0 = __msa_splati_h(coeff, 4);
83 cnst1 = __msa_ilvev_h(cnst1, cnst0);
84 stp21 = DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst1);
85
86 BUTTERFLY_4(stp30, stp37, stp26, stp21, in8, in15, in14, in9);
87 ILVRL_H2_SH(in15, in8, vec1, vec0);
88 SPLATI_H2_SH(coeff1, 0, 1, cnst0, cnst1);
89 cnst0 = __msa_ilvev_h(cnst0, cnst1);
90
91 in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
92 ST_SH(in8, tmp_ptr);
93
94 cnst0 = __msa_splati_h(coeff2, 0);
95 cnst0 = __msa_ilvev_h(cnst1, cnst0);
96 in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
97 ST_SH(in8, tmp_ptr + 224);
98
99 ILVRL_H2_SH(in14, in9, vec1, vec0);
100 SPLATI_H2_SH(coeff1, 2, 3, cnst0, cnst1);
101 cnst1 = __msa_ilvev_h(cnst1, cnst0);
102
103 in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst1);
104 ST_SH(in8, tmp_ptr + 128);
105
106 cnst1 = __msa_splati_h(coeff2, 2);
107 cnst0 = __msa_ilvev_h(cnst0, cnst1);
108 in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
109 ST_SH(in8, tmp_ptr + 96);
110
111 SPLATI_H2_SH(coeff, 2, 5, cnst0, cnst1);
112 cnst1 = __msa_ilvev_h(cnst1, cnst0);
113
114 stp25 = DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst1);
115
116 cnst1 = __msa_splati_h(coeff, 3);
117 cnst1 = __msa_ilvev_h(cnst0, cnst1);
118 stp22 = DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst1);
119
120 /* stp4 */
121 ADD2(stp34, stp25, stp33, stp22, in13, in10);
122
123 ILVRL_H2_SH(in13, in10, vec1, vec0);
124 SPLATI_H2_SH(coeff1, 4, 5, cnst0, cnst1);
125 cnst0 = __msa_ilvev_h(cnst0, cnst1);
126 in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
127 ST_SH(in8, tmp_ptr + 64);
128
129 cnst0 = __msa_splati_h(coeff2, 1);
130 cnst0 = __msa_ilvev_h(cnst1, cnst0);
131 in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
132 ST_SH(in8, tmp_ptr + 160);
133
134 SUB2(stp34, stp25, stp33, stp22, in12, in11);
135 ILVRL_H2_SH(in12, in11, vec1, vec0);
136 SPLATI_H2_SH(coeff1, 6, 7, cnst0, cnst1);
137 cnst1 = __msa_ilvev_h(cnst1, cnst0);
138
139 in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst1);
140 ST_SH(in8, tmp_ptr + 192);
141
142 cnst1 = __msa_splati_h(coeff2, 3);
143 cnst0 = __msa_ilvev_h(cnst0, cnst1);
144 in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
145 ST_SH(in8, tmp_ptr + 32);
146 }
147
fdct16x8_1d_row(int16_t * input,int16_t * output)148 void fdct16x8_1d_row(int16_t *input, int16_t *output) {
149 v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
150 v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
151 v8i16 in8, in9, in10, in11, in12, in13, in14, in15;
152
153 LD_SH8(input, 16, in0, in1, in2, in3, in4, in5, in6, in7);
154 LD_SH8((input + 8), 16, in8, in9, in10, in11, in12, in13, in14, in15);
155 TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
156 in4, in5, in6, in7);
157 TRANSPOSE8x8_SH_SH(in8, in9, in10, in11, in12, in13, in14, in15, in8, in9,
158 in10, in11, in12, in13, in14, in15);
159 ADD4(in0, 1, in1, 1, in2, 1, in3, 1, in0, in1, in2, in3);
160 ADD4(in4, 1, in5, 1, in6, 1, in7, 1, in4, in5, in6, in7);
161 ADD4(in8, 1, in9, 1, in10, 1, in11, 1, in8, in9, in10, in11);
162 ADD4(in12, 1, in13, 1, in14, 1, in15, 1, in12, in13, in14, in15);
163 SRA_4V(in0, in1, in2, in3, 2);
164 SRA_4V(in4, in5, in6, in7, 2);
165 SRA_4V(in8, in9, in10, in11, 2);
166 SRA_4V(in12, in13, in14, in15, 2);
167 BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7, in8, in9, in10, in11,
168 in12, in13, in14, in15, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6,
169 tmp7, in8, in9, in10, in11, in12, in13, in14, in15);
170 ST_SH8(in8, in9, in10, in11, in12, in13, in14, in15, input, 16);
171 FDCT8x16_EVEN(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp0, tmp1,
172 tmp2, tmp3, tmp4, tmp5, tmp6, tmp7);
173 LD_SH8(input, 16, in8, in9, in10, in11, in12, in13, in14, in15);
174 FDCT8x16_ODD(in8, in9, in10, in11, in12, in13, in14, in15, in0, in1, in2, in3,
175 in4, in5, in6, in7);
176 TRANSPOSE8x8_SH_SH(tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3, tmp0, in0,
177 tmp1, in1, tmp2, in2, tmp3, in3);
178 ST_SH8(tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3, output, 16);
179 TRANSPOSE8x8_SH_SH(tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7, tmp4, in4,
180 tmp5, in5, tmp6, in6, tmp7, in7);
181 ST_SH8(tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7, output + 8, 16);
182 }
183
vpx_fdct4x4_msa(const int16_t * input,int16_t * output,int32_t src_stride)184 void vpx_fdct4x4_msa(const int16_t *input, int16_t *output,
185 int32_t src_stride) {
186 v8i16 in0, in1, in2, in3;
187
188 LD_SH4(input, src_stride, in0, in1, in2, in3);
189
190 /* fdct4 pre-process */
191 {
192 v8i16 vec, mask;
193 v16i8 zero = { 0 };
194 v16i8 one = __msa_ldi_b(1);
195
196 mask = (v8i16)__msa_sldi_b(zero, one, 15);
197 SLLI_4V(in0, in1, in2, in3, 4);
198 vec = __msa_ceqi_h(in0, 0);
199 vec = vec ^ 255;
200 vec = mask & vec;
201 in0 += vec;
202 }
203
204 VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
205 TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
206 VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
207 TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
208 ADD4(in0, 1, in1, 1, in2, 1, in3, 1, in0, in1, in2, in3);
209 SRA_4V(in0, in1, in2, in3, 2);
210 PCKEV_D2_SH(in1, in0, in3, in2, in0, in2);
211 ST_SH2(in0, in2, output, 8);
212 }
213
vpx_fdct8x8_msa(const int16_t * input,int16_t * output,int32_t src_stride)214 void vpx_fdct8x8_msa(const int16_t *input, int16_t *output,
215 int32_t src_stride) {
216 v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
217
218 LD_SH8(input, src_stride, in0, in1, in2, in3, in4, in5, in6, in7);
219 SLLI_4V(in0, in1, in2, in3, 2);
220 SLLI_4V(in4, in5, in6, in7, 2);
221 VP9_FDCT8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, in4,
222 in5, in6, in7);
223 TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
224 in4, in5, in6, in7);
225 VP9_FDCT8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, in4,
226 in5, in6, in7);
227 TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
228 in4, in5, in6, in7);
229 SRLI_AVE_S_4V_H(in0, in1, in2, in3, in4, in5, in6, in7);
230 ST_SH8(in0, in1, in2, in3, in4, in5, in6, in7, output, 8);
231 }
232
vpx_fdct16x16_msa(const int16_t * input,int16_t * output,int32_t src_stride)233 void vpx_fdct16x16_msa(const int16_t *input, int16_t *output,
234 int32_t src_stride) {
235 int32_t i;
236 DECLARE_ALIGNED(32, int16_t, tmp_buf[16 * 16]);
237
238 /* column transform */
239 for (i = 0; i < 2; ++i) {
240 fdct8x16_1d_column((input + 8 * i), (&tmp_buf[0] + 8 * i), src_stride);
241 }
242
243 /* row transform */
244 for (i = 0; i < 2; ++i) {
245 fdct16x8_1d_row((&tmp_buf[0] + (128 * i)), (output + (128 * i)));
246 }
247 }
248
vpx_fdct16x16_1_msa(const int16_t * input,int16_t * out,int32_t stride)249 void vpx_fdct16x16_1_msa(const int16_t *input, int16_t *out, int32_t stride) {
250 int sum, i;
251 v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
252 v4i32 vec_w = { 0 };
253
254 for (i = 0; i < 4; ++i) {
255 LD_SH2(input, 8, in0, in1);
256 input += stride;
257 LD_SH2(input, 8, in2, in3);
258 input += stride;
259 LD_SH2(input, 8, in4, in5);
260 input += stride;
261 LD_SH2(input, 8, in6, in7);
262 input += stride;
263 ADD4(in0, in1, in2, in3, in4, in5, in6, in7, in0, in2, in4, in6);
264 ADD2(in0, in2, in4, in6, in0, in4);
265 vec_w += __msa_hadd_s_w(in0, in0);
266 vec_w += __msa_hadd_s_w(in4, in4);
267 }
268
269 sum = HADD_SW_S32(vec_w);
270 out[0] = (int16_t)(sum >> 1);
271 }
272 #endif // !CONFIG_VP9_HIGHBITDEPTH
273