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 <assert.h>
12
13 #include "./vp9_rtcd.h"
14 #include "vp9/common/vp9_enums.h"
15 #include "vp9/encoder/mips/msa/vp9_fdct_msa.h"
16 #include "vpx_dsp/mips/fwd_txfm_msa.h"
17
fadst16_cols_step1_msa(const int16_t * input,int32_t stride,const int32_t * const0,int16_t * int_buf)18 static void fadst16_cols_step1_msa(const int16_t *input, int32_t stride,
19 const int32_t *const0, int16_t *int_buf) {
20 v8i16 r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15;
21 v8i16 tp0, tp1, tp2, tp3, g0, g1, g2, g3, g8, g9, g10, g11, h0, h1, h2, h3;
22 v4i32 k0, k1, k2, k3;
23
24 /* load input data */
25 r0 = LD_SH(input);
26 r15 = LD_SH(input + 15 * stride);
27 r7 = LD_SH(input + 7 * stride);
28 r8 = LD_SH(input + 8 * stride);
29 SLLI_4V(r0, r15, r7, r8, 2);
30
31 /* stage 1 */
32 LD_SW2(const0, 4, k0, k1);
33 LD_SW2(const0 + 8, 4, k2, k3);
34 MADD_BF(r15, r0, r7, r8, k0, k1, k2, k3, g0, g1, g2, g3);
35
36 r3 = LD_SH(input + 3 * stride);
37 r4 = LD_SH(input + 4 * stride);
38 r11 = LD_SH(input + 11 * stride);
39 r12 = LD_SH(input + 12 * stride);
40 SLLI_4V(r3, r4, r11, r12, 2);
41
42 LD_SW2(const0 + 4 * 4, 4, k0, k1);
43 LD_SW2(const0 + 4 * 6, 4, k2, k3);
44 MADD_BF(r11, r4, r3, r12, k0, k1, k2, k3, g8, g9, g10, g11);
45
46 /* stage 2 */
47 BUTTERFLY_4(g0, g2, g10, g8, tp0, tp2, tp3, tp1);
48 ST_SH2(tp0, tp2, int_buf, 8);
49 ST_SH2(tp1, tp3, int_buf + 4 * 8, 8);
50
51 LD_SW2(const0 + 4 * 8, 4, k0, k1);
52 k2 = LD_SW(const0 + 4 * 10);
53 MADD_BF(g1, g3, g9, g11, k0, k1, k2, k0, h0, h1, h2, h3);
54
55 ST_SH2(h0, h1, int_buf + 8 * 8, 8);
56 ST_SH2(h3, h2, int_buf + 12 * 8, 8);
57
58 r9 = LD_SH(input + 9 * stride);
59 r6 = LD_SH(input + 6 * stride);
60 r1 = LD_SH(input + stride);
61 r14 = LD_SH(input + 14 * stride);
62 SLLI_4V(r9, r6, r1, r14, 2);
63
64 LD_SW2(const0 + 4 * 11, 4, k0, k1);
65 LD_SW2(const0 + 4 * 13, 4, k2, k3);
66 MADD_BF(r9, r6, r1, r14, k0, k1, k2, k3, g0, g1, g2, g3);
67
68 ST_SH2(g1, g3, int_buf + 3 * 8, 4 * 8);
69
70 r13 = LD_SH(input + 13 * stride);
71 r2 = LD_SH(input + 2 * stride);
72 r5 = LD_SH(input + 5 * stride);
73 r10 = LD_SH(input + 10 * stride);
74 SLLI_4V(r13, r2, r5, r10, 2);
75
76 LD_SW2(const0 + 4 * 15, 4, k0, k1);
77 LD_SW2(const0 + 4 * 17, 4, k2, k3);
78 MADD_BF(r13, r2, r5, r10, k0, k1, k2, k3, h0, h1, h2, h3);
79
80 ST_SH2(h1, h3, int_buf + 11 * 8, 4 * 8);
81
82 BUTTERFLY_4(h0, h2, g2, g0, tp0, tp1, tp2, tp3);
83 ST_SH4(tp0, tp1, tp2, tp3, int_buf + 2 * 8, 4 * 8);
84 }
85
fadst16_cols_step2_msa(int16_t * int_buf,const int32_t * const0,int16_t * out)86 static void fadst16_cols_step2_msa(int16_t *int_buf, const int32_t *const0,
87 int16_t *out) {
88 int16_t *out_ptr = out + 128;
89 v8i16 tp0, tp1, tp2, tp3, g5, g7, g13, g15;
90 v8i16 h0, h1, h2, h3, h4, h5, h6, h7, h10, h11;
91 v8i16 out0, out1, out2, out3, out4, out5, out6, out7;
92 v8i16 out8, out9, out10, out11, out12, out13, out14, out15;
93 v4i32 k0, k1, k2, k3;
94
95 LD_SH2(int_buf + 3 * 8, 4 * 8, g13, g15);
96 LD_SH2(int_buf + 11 * 8, 4 * 8, g5, g7);
97 LD_SW2(const0 + 4 * 19, 4, k0, k1);
98 k2 = LD_SW(const0 + 4 * 21);
99 MADD_BF(g7, g5, g15, g13, k0, k1, k2, k0, h4, h5, h6, h7);
100
101 tp0 = LD_SH(int_buf + 4 * 8);
102 tp1 = LD_SH(int_buf + 5 * 8);
103 tp3 = LD_SH(int_buf + 10 * 8);
104 tp2 = LD_SH(int_buf + 14 * 8);
105 LD_SW2(const0 + 4 * 22, 4, k0, k1);
106 k2 = LD_SW(const0 + 4 * 24);
107 MADD_BF(tp0, tp1, tp2, tp3, k0, k1, k2, k0, out4, out6, out5, out7);
108 out4 = -out4;
109 ST_SH(out4, (out + 3 * 16));
110 ST_SH(out5, (out_ptr + 4 * 16));
111
112 h1 = LD_SH(int_buf + 9 * 8);
113 h3 = LD_SH(int_buf + 12 * 8);
114 MADD_BF(h1, h3, h5, h7, k0, k1, k2, k0, out12, out14, out13, out15);
115 out13 = -out13;
116 ST_SH(out12, (out + 2 * 16));
117 ST_SH(out13, (out_ptr + 5 * 16));
118
119 tp0 = LD_SH(int_buf);
120 tp1 = LD_SH(int_buf + 8);
121 tp2 = LD_SH(int_buf + 2 * 8);
122 tp3 = LD_SH(int_buf + 6 * 8);
123
124 BUTTERFLY_4(tp0, tp1, tp3, tp2, out0, out1, h11, h10);
125 out1 = -out1;
126 ST_SH(out0, (out));
127 ST_SH(out1, (out_ptr + 7 * 16));
128
129 h0 = LD_SH(int_buf + 8 * 8);
130 h2 = LD_SH(int_buf + 13 * 8);
131
132 BUTTERFLY_4(h0, h2, h6, h4, out8, out9, out11, out10);
133 out8 = -out8;
134 ST_SH(out8, (out + 16));
135 ST_SH(out9, (out_ptr + 6 * 16));
136
137 /* stage 4 */
138 LD_SW2(const0 + 4 * 25, 4, k0, k1);
139 LD_SW2(const0 + 4 * 27, 4, k2, k3);
140 MADD_SHORT(h10, h11, k1, k2, out2, out3);
141 ST_SH(out2, (out + 7 * 16));
142 ST_SH(out3, (out_ptr));
143
144 MADD_SHORT(out6, out7, k0, k3, out6, out7);
145 ST_SH(out6, (out + 4 * 16));
146 ST_SH(out7, (out_ptr + 3 * 16));
147
148 MADD_SHORT(out10, out11, k0, k3, out10, out11);
149 ST_SH(out10, (out + 6 * 16));
150 ST_SH(out11, (out_ptr + 16));
151
152 MADD_SHORT(out14, out15, k1, k2, out14, out15);
153 ST_SH(out14, (out + 5 * 16));
154 ST_SH(out15, (out_ptr + 2 * 16));
155 }
156
fadst16_transpose_postproc_msa(int16_t * input,int16_t * out)157 static void fadst16_transpose_postproc_msa(int16_t *input, int16_t *out) {
158 v8i16 r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15;
159 v8i16 l0, l1, l2, l3, l4, l5, l6, l7, l8, l9, l10, l11, l12, l13, l14, l15;
160
161 /* load input data */
162 LD_SH8(input, 16, l0, l1, l2, l3, l4, l5, l6, l7);
163 TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7, r0, r1, r2, r3, r4, r5, r6,
164 r7);
165 FDCT_POSTPROC_2V_NEG_H(r0, r1);
166 FDCT_POSTPROC_2V_NEG_H(r2, r3);
167 FDCT_POSTPROC_2V_NEG_H(r4, r5);
168 FDCT_POSTPROC_2V_NEG_H(r6, r7);
169 ST_SH8(r0, r1, r2, r3, r4, r5, r6, r7, out, 8);
170 out += 64;
171
172 LD_SH8(input + 8, 16, l8, l9, l10, l11, l12, l13, l14, l15);
173 TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15, r8, r9, r10, r11,
174 r12, r13, r14, r15);
175 FDCT_POSTPROC_2V_NEG_H(r8, r9);
176 FDCT_POSTPROC_2V_NEG_H(r10, r11);
177 FDCT_POSTPROC_2V_NEG_H(r12, r13);
178 FDCT_POSTPROC_2V_NEG_H(r14, r15);
179 ST_SH8(r8, r9, r10, r11, r12, r13, r14, r15, out, 8);
180 out += 64;
181
182 /* load input data */
183 input += 128;
184 LD_SH8(input, 16, l0, l1, l2, l3, l4, l5, l6, l7);
185 TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7, r0, r1, r2, r3, r4, r5, r6,
186 r7);
187 FDCT_POSTPROC_2V_NEG_H(r0, r1);
188 FDCT_POSTPROC_2V_NEG_H(r2, r3);
189 FDCT_POSTPROC_2V_NEG_H(r4, r5);
190 FDCT_POSTPROC_2V_NEG_H(r6, r7);
191 ST_SH8(r0, r1, r2, r3, r4, r5, r6, r7, out, 8);
192 out += 64;
193
194 LD_SH8(input + 8, 16, l8, l9, l10, l11, l12, l13, l14, l15);
195 TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15, r8, r9, r10, r11,
196 r12, r13, r14, r15);
197 FDCT_POSTPROC_2V_NEG_H(r8, r9);
198 FDCT_POSTPROC_2V_NEG_H(r10, r11);
199 FDCT_POSTPROC_2V_NEG_H(r12, r13);
200 FDCT_POSTPROC_2V_NEG_H(r14, r15);
201 ST_SH8(r8, r9, r10, r11, r12, r13, r14, r15, out, 8);
202 }
203
fadst16_rows_step1_msa(int16_t * input,const int32_t * const0,int16_t * int_buf)204 static void fadst16_rows_step1_msa(int16_t *input, const int32_t *const0,
205 int16_t *int_buf) {
206 v8i16 r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15;
207 v8i16 tp0, tp1, tp2, tp3, g0, g1, g2, g3, g8, g9, g10, g11, h0, h1, h2, h3;
208 v4i32 k0, k1, k2, k3;
209
210 /* load input data */
211 r0 = LD_SH(input);
212 r7 = LD_SH(input + 7 * 8);
213 r8 = LD_SH(input + 8 * 8);
214 r15 = LD_SH(input + 15 * 8);
215
216 /* stage 1 */
217 LD_SW2(const0, 4, k0, k1);
218 LD_SW2(const0 + 4 * 2, 4, k2, k3);
219 MADD_BF(r15, r0, r7, r8, k0, k1, k2, k3, g0, g1, g2, g3);
220
221 r3 = LD_SH(input + 3 * 8);
222 r4 = LD_SH(input + 4 * 8);
223 r11 = LD_SH(input + 11 * 8);
224 r12 = LD_SH(input + 12 * 8);
225
226 LD_SW2(const0 + 4 * 4, 4, k0, k1);
227 LD_SW2(const0 + 4 * 6, 4, k2, k3);
228 MADD_BF(r11, r4, r3, r12, k0, k1, k2, k3, g8, g9, g10, g11);
229
230 /* stage 2 */
231 BUTTERFLY_4(g0, g2, g10, g8, tp0, tp2, tp3, tp1);
232 ST_SH2(tp0, tp1, int_buf, 4 * 8);
233 ST_SH2(tp2, tp3, int_buf + 8, 4 * 8);
234
235 LD_SW2(const0 + 4 * 8, 4, k0, k1);
236 k2 = LD_SW(const0 + 4 * 10);
237 MADD_BF(g1, g3, g9, g11, k0, k1, k2, k0, h0, h1, h2, h3);
238 ST_SH2(h0, h3, int_buf + 8 * 8, 4 * 8);
239 ST_SH2(h1, h2, int_buf + 9 * 8, 4 * 8);
240
241 r1 = LD_SH(input + 8);
242 r6 = LD_SH(input + 6 * 8);
243 r9 = LD_SH(input + 9 * 8);
244 r14 = LD_SH(input + 14 * 8);
245
246 LD_SW2(const0 + 4 * 11, 4, k0, k1);
247 LD_SW2(const0 + 4 * 13, 4, k2, k3);
248 MADD_BF(r9, r6, r1, r14, k0, k1, k2, k3, g0, g1, g2, g3);
249 ST_SH2(g1, g3, int_buf + 3 * 8, 4 * 8);
250
251 r2 = LD_SH(input + 2 * 8);
252 r5 = LD_SH(input + 5 * 8);
253 r10 = LD_SH(input + 10 * 8);
254 r13 = LD_SH(input + 13 * 8);
255
256 LD_SW2(const0 + 4 * 15, 4, k0, k1);
257 LD_SW2(const0 + 4 * 17, 4, k2, k3);
258 MADD_BF(r13, r2, r5, r10, k0, k1, k2, k3, h0, h1, h2, h3);
259 ST_SH2(h1, h3, int_buf + 11 * 8, 4 * 8);
260 BUTTERFLY_4(h0, h2, g2, g0, tp0, tp1, tp2, tp3);
261 ST_SH4(tp0, tp1, tp2, tp3, int_buf + 2 * 8, 4 * 8);
262 }
263
fadst16_rows_step2_msa(int16_t * int_buf,const int32_t * const0,int16_t * out)264 static void fadst16_rows_step2_msa(int16_t *int_buf, const int32_t *const0,
265 int16_t *out) {
266 int16_t *out_ptr = out + 8;
267 v8i16 tp0, tp1, tp2, tp3, g5, g7, g13, g15;
268 v8i16 h0, h1, h2, h3, h4, h5, h6, h7, h10, h11;
269 v8i16 out0, out1, out2, out3, out4, out5, out6, out7;
270 v8i16 out8, out9, out10, out11, out12, out13, out14, out15;
271 v4i32 k0, k1, k2, k3;
272
273 g13 = LD_SH(int_buf + 3 * 8);
274 g15 = LD_SH(int_buf + 7 * 8);
275 g5 = LD_SH(int_buf + 11 * 8);
276 g7 = LD_SH(int_buf + 15 * 8);
277
278 LD_SW2(const0 + 4 * 19, 4, k0, k1);
279 k2 = LD_SW(const0 + 4 * 21);
280 MADD_BF(g7, g5, g15, g13, k0, k1, k2, k0, h4, h5, h6, h7);
281
282 tp0 = LD_SH(int_buf + 4 * 8);
283 tp1 = LD_SH(int_buf + 5 * 8);
284 tp3 = LD_SH(int_buf + 10 * 8);
285 tp2 = LD_SH(int_buf + 14 * 8);
286
287 LD_SW2(const0 + 4 * 22, 4, k0, k1);
288 k2 = LD_SW(const0 + 4 * 24);
289 MADD_BF(tp0, tp1, tp2, tp3, k0, k1, k2, k0, out4, out6, out5, out7);
290 out4 = -out4;
291 ST_SH(out4, (out + 3 * 16));
292 ST_SH(out5, (out_ptr + 4 * 16));
293
294 h1 = LD_SH(int_buf + 9 * 8);
295 h3 = LD_SH(int_buf + 12 * 8);
296 MADD_BF(h1, h3, h5, h7, k0, k1, k2, k0, out12, out14, out13, out15);
297 out13 = -out13;
298 ST_SH(out12, (out + 2 * 16));
299 ST_SH(out13, (out_ptr + 5 * 16));
300
301 tp0 = LD_SH(int_buf);
302 tp1 = LD_SH(int_buf + 8);
303 tp2 = LD_SH(int_buf + 2 * 8);
304 tp3 = LD_SH(int_buf + 6 * 8);
305
306 BUTTERFLY_4(tp0, tp1, tp3, tp2, out0, out1, h11, h10);
307 out1 = -out1;
308 ST_SH(out0, (out));
309 ST_SH(out1, (out_ptr + 7 * 16));
310
311 h0 = LD_SH(int_buf + 8 * 8);
312 h2 = LD_SH(int_buf + 13 * 8);
313 BUTTERFLY_4(h0, h2, h6, h4, out8, out9, out11, out10);
314 out8 = -out8;
315 ST_SH(out8, (out + 16));
316 ST_SH(out9, (out_ptr + 6 * 16));
317
318 /* stage 4 */
319 LD_SW2(const0 + 4 * 25, 4, k0, k1);
320 LD_SW2(const0 + 4 * 27, 4, k2, k3);
321 MADD_SHORT(h10, h11, k1, k2, out2, out3);
322 ST_SH(out2, (out + 7 * 16));
323 ST_SH(out3, (out_ptr));
324
325 MADD_SHORT(out6, out7, k0, k3, out6, out7);
326 ST_SH(out6, (out + 4 * 16));
327 ST_SH(out7, (out_ptr + 3 * 16));
328
329 MADD_SHORT(out10, out11, k0, k3, out10, out11);
330 ST_SH(out10, (out + 6 * 16));
331 ST_SH(out11, (out_ptr + 16));
332
333 MADD_SHORT(out14, out15, k1, k2, out14, out15);
334 ST_SH(out14, (out + 5 * 16));
335 ST_SH(out15, (out_ptr + 2 * 16));
336 }
337
fadst16_transpose_msa(int16_t * input,int16_t * out)338 static void fadst16_transpose_msa(int16_t *input, int16_t *out) {
339 v8i16 r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15;
340 v8i16 l0, l1, l2, l3, l4, l5, l6, l7, l8, l9, l10, l11, l12, l13, l14, l15;
341
342 /* load input data */
343 LD_SH16(input, 8, l0, l8, l1, l9, l2, l10, l3, l11, l4, l12, l5, l13, l6, l14,
344 l7, l15);
345 TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7, r0, r1, r2, r3, r4, r5, r6,
346 r7);
347 TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15, r8, r9, r10, r11,
348 r12, r13, r14, r15);
349 ST_SH8(r0, r8, r1, r9, r2, r10, r3, r11, out, 8);
350 ST_SH8(r4, r12, r5, r13, r6, r14, r7, r15, (out + 64), 8);
351 out += 16 * 8;
352
353 /* load input data */
354 input += 128;
355 LD_SH16(input, 8, l0, l8, l1, l9, l2, l10, l3, l11, l4, l12, l5, l13, l6, l14,
356 l7, l15);
357 TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7, r0, r1, r2, r3, r4, r5, r6,
358 r7);
359 TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15, r8, r9, r10, r11,
360 r12, r13, r14, r15);
361 ST_SH8(r0, r8, r1, r9, r2, r10, r3, r11, out, 8);
362 ST_SH8(r4, r12, r5, r13, r6, r14, r7, r15, (out + 64), 8);
363 }
364
postproc_fdct16x8_1d_row(int16_t * intermediate,int16_t * output)365 static void postproc_fdct16x8_1d_row(int16_t *intermediate, int16_t *output) {
366 int16_t *temp = intermediate;
367 int16_t *out = output;
368 v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
369 v8i16 in0, in1, in2, in3, in4, in5, in6, in7, in8, in9, in10, in11;
370 v8i16 in12, in13, in14, in15;
371
372 LD_SH8(temp, 16, in0, in1, in2, in3, in4, in5, in6, in7);
373 temp = intermediate + 8;
374 LD_SH8(temp, 16, in8, in9, in10, in11, in12, in13, in14, in15);
375 TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
376 in4, in5, in6, in7);
377 TRANSPOSE8x8_SH_SH(in8, in9, in10, in11, in12, in13, in14, in15, in8, in9,
378 in10, in11, in12, in13, in14, in15);
379 FDCT_POSTPROC_2V_NEG_H(in0, in1);
380 FDCT_POSTPROC_2V_NEG_H(in2, in3);
381 FDCT_POSTPROC_2V_NEG_H(in4, in5);
382 FDCT_POSTPROC_2V_NEG_H(in6, in7);
383 FDCT_POSTPROC_2V_NEG_H(in8, in9);
384 FDCT_POSTPROC_2V_NEG_H(in10, in11);
385 FDCT_POSTPROC_2V_NEG_H(in12, in13);
386 FDCT_POSTPROC_2V_NEG_H(in14, in15);
387 BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7, in8, in9, in10, in11,
388 in12, in13, in14, in15, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6,
389 tmp7, in8, in9, in10, in11, in12, in13, in14, in15);
390 temp = intermediate;
391 ST_SH8(in8, in9, in10, in11, in12, in13, in14, in15, temp, 16);
392 FDCT8x16_EVEN(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp0, tmp1,
393 tmp2, tmp3, tmp4, tmp5, tmp6, tmp7);
394 temp = intermediate;
395 LD_SH8(temp, 16, in8, in9, in10, in11, in12, in13, in14, in15);
396 FDCT8x16_ODD(in8, in9, in10, in11, in12, in13, in14, in15, in0, in1, in2, in3,
397 in4, in5, in6, in7);
398 TRANSPOSE8x8_SH_SH(tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3, tmp0, in0,
399 tmp1, in1, tmp2, in2, tmp3, in3);
400 ST_SH8(tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3, out, 16);
401 TRANSPOSE8x8_SH_SH(tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7, tmp4, in4,
402 tmp5, in5, tmp6, in6, tmp7, in7);
403 out = output + 8;
404 ST_SH8(tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7, out, 16);
405 }
406
vp9_fht16x16_msa(const int16_t * input,int16_t * output,int32_t stride,int32_t tx_type)407 void vp9_fht16x16_msa(const int16_t *input, int16_t *output, int32_t stride,
408 int32_t tx_type) {
409 DECLARE_ALIGNED(32, int16_t, tmp[256]);
410 DECLARE_ALIGNED(32, int16_t, trans_buf[256]);
411 DECLARE_ALIGNED(32, int16_t, tmp_buf[128]);
412 int32_t i;
413 int16_t *ptmpbuf = &tmp_buf[0];
414 int16_t *trans = &trans_buf[0];
415 const int32_t const_arr[29 * 4] = {
416 52707308, 52707308, 52707308, 52707308, -1072430300,
417 -1072430300, -1072430300, -1072430300, 795618043, 795618043,
418 795618043, 795618043, -721080468, -721080468, -721080468,
419 -721080468, 459094491, 459094491, 459094491, 459094491,
420 -970646691, -970646691, -970646691, -970646691, 1010963856,
421 1010963856, 1010963856, 1010963856, -361743294, -361743294,
422 -361743294, -361743294, 209469125, 209469125, 209469125,
423 209469125, -1053094788, -1053094788, -1053094788, -1053094788,
424 1053160324, 1053160324, 1053160324, 1053160324, 639644520,
425 639644520, 639644520, 639644520, -862444000, -862444000,
426 -862444000, -862444000, 1062144356, 1062144356, 1062144356,
427 1062144356, -157532337, -157532337, -157532337, -157532337,
428 260914709, 260914709, 260914709, 260914709, -1041559667,
429 -1041559667, -1041559667, -1041559667, 920985831, 920985831,
430 920985831, 920985831, -551995675, -551995675, -551995675,
431 -551995675, 596522295, 596522295, 596522295, 596522295,
432 892853362, 892853362, 892853362, 892853362, -892787826,
433 -892787826, -892787826, -892787826, 410925857, 410925857,
434 410925857, 410925857, -992012162, -992012162, -992012162,
435 -992012162, 992077698, 992077698, 992077698, 992077698,
436 759246145, 759246145, 759246145, 759246145, -759180609,
437 -759180609, -759180609, -759180609, -759222975, -759222975,
438 -759222975, -759222975, 759288511, 759288511, 759288511,
439 759288511
440 };
441
442 switch (tx_type) {
443 case DCT_DCT:
444 /* column transform */
445 for (i = 0; i < 2; ++i) {
446 fdct8x16_1d_column(input + 8 * i, tmp + 8 * i, stride);
447 }
448
449 /* row transform */
450 for (i = 0; i < 2; ++i) {
451 fdct16x8_1d_row(tmp + (128 * i), output + (128 * i));
452 }
453 break;
454 case ADST_DCT:
455 /* column transform */
456 for (i = 0; i < 2; ++i) {
457 fadst16_cols_step1_msa(input + (i << 3), stride, const_arr, ptmpbuf);
458 fadst16_cols_step2_msa(ptmpbuf, const_arr, tmp + (i << 3));
459 }
460
461 /* row transform */
462 for (i = 0; i < 2; ++i) {
463 postproc_fdct16x8_1d_row(tmp + (128 * i), output + (128 * i));
464 }
465 break;
466 case DCT_ADST:
467 /* column transform */
468 for (i = 0; i < 2; ++i) {
469 fdct8x16_1d_column(input + 8 * i, tmp + 8 * i, stride);
470 }
471
472 fadst16_transpose_postproc_msa(tmp, trans);
473
474 /* row transform */
475 for (i = 0; i < 2; ++i) {
476 fadst16_rows_step1_msa(trans + (i << 7), const_arr, ptmpbuf);
477 fadst16_rows_step2_msa(ptmpbuf, const_arr, tmp + (i << 7));
478 }
479
480 fadst16_transpose_msa(tmp, output);
481 break;
482 case ADST_ADST:
483 /* column transform */
484 for (i = 0; i < 2; ++i) {
485 fadst16_cols_step1_msa(input + (i << 3), stride, const_arr, ptmpbuf);
486 fadst16_cols_step2_msa(ptmpbuf, const_arr, tmp + (i << 3));
487 }
488
489 fadst16_transpose_postproc_msa(tmp, trans);
490
491 /* row transform */
492 for (i = 0; i < 2; ++i) {
493 fadst16_rows_step1_msa(trans + (i << 7), const_arr, ptmpbuf);
494 fadst16_rows_step2_msa(ptmpbuf, const_arr, tmp + (i << 7));
495 }
496
497 fadst16_transpose_msa(tmp, output);
498 break;
499 default: assert(0); break;
500 }
501 }
502