1 /*
2 * Copyright (c) 2021 Loongson Technology Corporation Limited
3 * Contributed by Jin Bo <jinbo@loongson.cn>
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 #include "libavcodec/vp9dsp.h"
23 #include "libavutil/loongarch/loongson_intrinsics.h"
24 #include "vp9dsp_loongarch.h"
25 #include "libavutil/attributes.h"
26
27 #define VP9_DCT_CONST_BITS 14
28 #define ALLOC_ALIGNED(align) __attribute__ ((aligned(align)))
29 #define ROUND_POWER_OF_TWO(value, n) (((value) + (1 << ((n) - 1))) >> (n))
30
31 const int32_t cospi_1_64 = 16364;
32 const int32_t cospi_2_64 = 16305;
33 const int32_t cospi_3_64 = 16207;
34 const int32_t cospi_4_64 = 16069;
35 const int32_t cospi_5_64 = 15893;
36 const int32_t cospi_6_64 = 15679;
37 const int32_t cospi_7_64 = 15426;
38 const int32_t cospi_8_64 = 15137;
39 const int32_t cospi_9_64 = 14811;
40 const int32_t cospi_10_64 = 14449;
41 const int32_t cospi_11_64 = 14053;
42 const int32_t cospi_12_64 = 13623;
43 const int32_t cospi_13_64 = 13160;
44 const int32_t cospi_14_64 = 12665;
45 const int32_t cospi_15_64 = 12140;
46 const int32_t cospi_16_64 = 11585;
47 const int32_t cospi_17_64 = 11003;
48 const int32_t cospi_18_64 = 10394;
49 const int32_t cospi_19_64 = 9760;
50 const int32_t cospi_20_64 = 9102;
51 const int32_t cospi_21_64 = 8423;
52 const int32_t cospi_22_64 = 7723;
53 const int32_t cospi_23_64 = 7005;
54 const int32_t cospi_24_64 = 6270;
55 const int32_t cospi_25_64 = 5520;
56 const int32_t cospi_26_64 = 4756;
57 const int32_t cospi_27_64 = 3981;
58 const int32_t cospi_28_64 = 3196;
59 const int32_t cospi_29_64 = 2404;
60 const int32_t cospi_30_64 = 1606;
61 const int32_t cospi_31_64 = 804;
62
63 const int32_t sinpi_1_9 = 5283;
64 const int32_t sinpi_2_9 = 9929;
65 const int32_t sinpi_3_9 = 13377;
66 const int32_t sinpi_4_9 = 15212;
67
68 #define VP9_DOTP_CONST_PAIR(reg0, reg1, cnst0, cnst1, out0, out1) \
69 { \
70 __m128i k0_m = __lsx_vreplgr2vr_h(cnst0); \
71 __m128i s0_m, s1_m, s2_m, s3_m; \
72 \
73 s0_m = __lsx_vreplgr2vr_h(cnst1); \
74 k0_m = __lsx_vpackev_h(s0_m, k0_m); \
75 \
76 s1_m = __lsx_vilvl_h(__lsx_vneg_h(reg1), reg0); \
77 s0_m = __lsx_vilvh_h(__lsx_vneg_h(reg1), reg0); \
78 s3_m = __lsx_vilvl_h(reg0, reg1); \
79 s2_m = __lsx_vilvh_h(reg0, reg1); \
80 DUP2_ARG2(__lsx_vdp2_w_h, s1_m, k0_m, s0_m, k0_m, s1_m, s0_m); \
81 DUP2_ARG2(__lsx_vsrari_w, s1_m, VP9_DCT_CONST_BITS, \
82 s0_m, VP9_DCT_CONST_BITS, s1_m, s0_m); \
83 out0 = __lsx_vpickev_h(s0_m, s1_m); \
84 DUP2_ARG2(__lsx_vdp2_w_h, s3_m, k0_m, s2_m, k0_m, s1_m, s0_m); \
85 DUP2_ARG2(__lsx_vsrari_w, s1_m, VP9_DCT_CONST_BITS, \
86 s0_m, VP9_DCT_CONST_BITS, s1_m, s0_m); \
87 out1 = __lsx_vpickev_h(s0_m, s1_m); \
88 }
89
90 #define VP9_SET_COSPI_PAIR(c0_h, c1_h) \
91 ( { \
92 __m128i out0_m, r0_m, r1_m; \
93 \
94 r0_m = __lsx_vreplgr2vr_h(c0_h); \
95 r1_m = __lsx_vreplgr2vr_h(c1_h); \
96 out0_m = __lsx_vpackev_h(r1_m, r0_m); \
97 \
98 out0_m; \
99 } )
100
101 #define VP9_ADDBLK_ST8x4_UB(dst, dst_stride, in0, in1, in2, in3) \
102 { \
103 uint8_t *dst_m = (uint8_t *) (dst); \
104 __m128i dst0_m, dst1_m, dst2_m, dst3_m; \
105 __m128i tmp0_m, tmp1_m; \
106 __m128i res0_m, res1_m, res2_m, res3_m; \
107 __m128i zero_m = __lsx_vldi(0); \
108 DUP4_ARG2(__lsx_vld, dst_m, 0, dst_m + dst_stride, 0, \
109 dst_m + 2 * dst_stride, 0, dst_m + 3 * dst_stride, 0, \
110 dst0_m, dst1_m, dst2_m, dst3_m); \
111 DUP4_ARG2(__lsx_vilvl_b, zero_m, dst0_m, zero_m, dst1_m, zero_m, \
112 dst2_m, zero_m, dst3_m, res0_m, res1_m, res2_m, res3_m);\
113 DUP4_ARG2(__lsx_vadd_h, res0_m, in0, res1_m, in1, res2_m, in2, \
114 res3_m, in3, res0_m, res1_m, res2_m, res3_m); \
115 DUP4_ARG1(__lsx_vclip255_h, res0_m, res1_m, res2_m, res3_m, \
116 res0_m, res1_m, res2_m, res3_m); \
117 DUP2_ARG2(__lsx_vpickev_b, res1_m, res0_m, res3_m, res2_m, \
118 tmp0_m, tmp1_m); \
119 __lsx_vstelm_d(tmp0_m, dst_m, 0, 0); \
120 __lsx_vstelm_d(tmp0_m, dst_m + dst_stride, 0, 1); \
121 __lsx_vstelm_d(tmp1_m, dst_m + 2 * dst_stride, 0, 0); \
122 __lsx_vstelm_d(tmp1_m, dst_m + 3 * dst_stride, 0, 1); \
123 }
124
125 #define VP9_UNPCK_UB_SH(in, out_h, out_l) \
126 { \
127 __m128i zero = __lsx_vldi(0); \
128 out_l = __lsx_vilvl_b(zero, in); \
129 out_h = __lsx_vilvh_b(zero, in); \
130 }
131
132 #define VP9_ILVLTRANS4x8_H(in0, in1, in2, in3, in4, in5, in6, in7, \
133 out0, out1, out2, out3, out4, out5, out6, out7) \
134 { \
135 __m128i tmp0_m, tmp1_m, tmp2_m, tmp3_m; \
136 __m128i tmp0_n, tmp1_n, tmp2_n, tmp3_n; \
137 __m128i zero_m = __lsx_vldi(0); \
138 \
139 DUP4_ARG2(__lsx_vilvl_h, in1, in0, in3, in2, in5, in4, in7, in6, \
140 tmp0_n, tmp1_n, tmp2_n, tmp3_n); \
141 tmp0_m = __lsx_vilvl_w(tmp1_n, tmp0_n); \
142 tmp2_m = __lsx_vilvh_w(tmp1_n, tmp0_n); \
143 tmp1_m = __lsx_vilvl_w(tmp3_n, tmp2_n); \
144 tmp3_m = __lsx_vilvh_w(tmp3_n, tmp2_n); \
145 \
146 out0 = __lsx_vilvl_d(tmp1_m, tmp0_m); \
147 out1 = __lsx_vilvh_d(tmp1_m, tmp0_m); \
148 out2 = __lsx_vilvl_d(tmp3_m, tmp2_m); \
149 out3 = __lsx_vilvh_d(tmp3_m, tmp2_m); \
150 \
151 out4 = zero_m; \
152 out5 = zero_m; \
153 out6 = zero_m; \
154 out7 = zero_m; \
155 }
156
157 /* multiply and add macro */
158 #define VP9_MADD(inp0, inp1, inp2, inp3, cst0, cst1, cst2, cst3, \
159 out0, out1, out2, out3) \
160 { \
161 __m128i madd_s0_m, madd_s1_m, madd_s2_m, madd_s3_m; \
162 __m128i tmp0_m, tmp1_m, tmp2_m, tmp3_m; \
163 \
164 madd_s1_m = __lsx_vilvl_h(inp1, inp0); \
165 madd_s0_m = __lsx_vilvh_h(inp1, inp0); \
166 madd_s3_m = __lsx_vilvl_h(inp3, inp2); \
167 madd_s2_m = __lsx_vilvh_h(inp3, inp2); \
168 DUP4_ARG2(__lsx_vdp2_w_h, madd_s1_m, cst0, madd_s0_m, cst0, \
169 madd_s1_m, cst1, madd_s0_m, cst1, tmp0_m, tmp1_m, \
170 tmp2_m, tmp3_m); \
171 DUP4_ARG2(__lsx_vsrari_w, tmp0_m, VP9_DCT_CONST_BITS, tmp1_m, \
172 VP9_DCT_CONST_BITS, tmp2_m, VP9_DCT_CONST_BITS, tmp3_m, \
173 VP9_DCT_CONST_BITS, tmp0_m, tmp1_m, tmp2_m, tmp3_m); \
174 DUP2_ARG2(__lsx_vpickev_h, tmp1_m, tmp0_m, tmp3_m, tmp2_m, out0, out1); \
175 DUP4_ARG2(__lsx_vdp2_w_h, madd_s3_m, cst2, madd_s2_m, cst2, madd_s3_m, \
176 cst3, madd_s2_m, cst3, tmp0_m, tmp1_m, tmp2_m, tmp3_m); \
177 DUP4_ARG2(__lsx_vsrari_w, tmp0_m, VP9_DCT_CONST_BITS, \
178 tmp1_m, VP9_DCT_CONST_BITS, tmp2_m, VP9_DCT_CONST_BITS, \
179 tmp3_m, VP9_DCT_CONST_BITS, tmp0_m, tmp1_m, tmp2_m, tmp3_m); \
180 DUP2_ARG2(__lsx_vpickev_h, tmp1_m, tmp0_m, tmp3_m, tmp2_m, out2, out3); \
181 }
182
183 #define VP9_SET_CONST_PAIR(mask_h, idx1_h, idx2_h) \
184 ( { \
185 __m128i c0_m, c1_m; \
186 \
187 DUP2_ARG2(__lsx_vreplvei_h, mask_h, idx1_h, mask_h, idx2_h, c0_m, c1_m); \
188 c0_m = __lsx_vpackev_h(c1_m, c0_m); \
189 \
190 c0_m; \
191 } )
192
193 /* idct 8x8 macro */
194 #define VP9_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7, \
195 out0, out1, out2, out3, out4, out5, out6, out7) \
196 { \
197 __m128i tp0_m, tp1_m, tp2_m, tp3_m, tp4_m, tp5_m, tp6_m, tp7_m; \
198 __m128i k0_m, k1_m, k2_m, k3_m, res0_m, res1_m, res2_m, res3_m; \
199 __m128i tmp0_m, tmp1_m, tmp2_m, tmp3_m; \
200 v8i16 mask_m = { cospi_28_64, cospi_4_64, cospi_20_64, cospi_12_64, \
201 cospi_16_64, -cospi_4_64, -cospi_20_64, -cospi_16_64 }; \
202 \
203 k0_m = VP9_SET_CONST_PAIR(mask_m, 0, 5); \
204 k1_m = VP9_SET_CONST_PAIR(mask_m, 1, 0); \
205 k2_m = VP9_SET_CONST_PAIR(mask_m, 6, 3); \
206 k3_m = VP9_SET_CONST_PAIR(mask_m, 3, 2); \
207 VP9_MADD(in1, in7, in3, in5, k0_m, k1_m, k2_m, k3_m, in1, in7, in3, in5); \
208 DUP2_ARG2(__lsx_vsub_h, in1, in3, in7, in5, res0_m, res1_m); \
209 k0_m = VP9_SET_CONST_PAIR(mask_m, 4, 7); \
210 k1_m = __lsx_vreplvei_h(mask_m, 4); \
211 \
212 res2_m = __lsx_vilvl_h(res0_m, res1_m); \
213 res3_m = __lsx_vilvh_h(res0_m, res1_m); \
214 DUP4_ARG2(__lsx_vdp2_w_h, res2_m, k0_m, res3_m, k0_m, res2_m, k1_m, \
215 res3_m, k1_m, tmp0_m, tmp1_m, tmp2_m, tmp3_m); \
216 DUP4_ARG2(__lsx_vsrari_w, tmp0_m, VP9_DCT_CONST_BITS, \
217 tmp1_m, VP9_DCT_CONST_BITS, tmp2_m, VP9_DCT_CONST_BITS, \
218 tmp3_m, VP9_DCT_CONST_BITS, tmp0_m, tmp1_m, tmp2_m, tmp3_m); \
219 tp4_m = __lsx_vadd_h(in1, in3); \
220 DUP2_ARG2(__lsx_vpickev_h, tmp1_m, tmp0_m, tmp3_m, tmp2_m, tp5_m, tp6_m); \
221 tp7_m = __lsx_vadd_h(in7, in5); \
222 k2_m = VP9_SET_COSPI_PAIR(cospi_24_64, -cospi_8_64); \
223 k3_m = VP9_SET_COSPI_PAIR(cospi_8_64, cospi_24_64); \
224 VP9_MADD(in0, in4, in2, in6, k1_m, k0_m, k2_m, k3_m, \
225 in0, in4, in2, in6); \
226 LSX_BUTTERFLY_4_H(in0, in4, in2, in6, tp0_m, tp1_m, tp2_m, tp3_m); \
227 LSX_BUTTERFLY_8_H(tp0_m, tp1_m, tp2_m, tp3_m, tp4_m, tp5_m, tp6_m, tp7_m, \
228 out0, out1, out2, out3, out4, out5, out6, out7); \
229 }
230
231 static av_always_inline
vp9_idct8x8_1_add_lsx(int16_t *input, uint8_t *dst, int32_t dst_stride)232 void vp9_idct8x8_1_add_lsx(int16_t *input, uint8_t *dst,
233 int32_t dst_stride)
234 {
235 int16_t out;
236 int32_t val;
237 __m128i vec;
238
239 out = ROUND_POWER_OF_TWO((input[0] * cospi_16_64), VP9_DCT_CONST_BITS);
240 out = ROUND_POWER_OF_TWO((out * cospi_16_64), VP9_DCT_CONST_BITS);
241 val = ROUND_POWER_OF_TWO(out, 5);
242 vec = __lsx_vreplgr2vr_h(val);
243 input[0] = 0;
244
245 VP9_ADDBLK_ST8x4_UB(dst, dst_stride, vec, vec, vec, vec);
246 dst += (4 * dst_stride);
247 VP9_ADDBLK_ST8x4_UB(dst, dst_stride, vec, vec, vec, vec);
248 }
249
vp9_idct8x8_12_colcol_addblk_lsx(int16_t *input, uint8_t *dst, int32_t dst_stride)250 static void vp9_idct8x8_12_colcol_addblk_lsx(int16_t *input, uint8_t *dst,
251 int32_t dst_stride)
252 {
253 __m128i in0, in1, in2, in3, in4, in5, in6, in7;
254 __m128i s0, s1, s2, s3, s4, s5, s6, s7, k0, k1, k2, k3, m0, m1, m2, m3;
255 __m128i tmp0, tmp1, tmp2, tmp3;
256 __m128i zero = __lsx_vldi(0);
257
258 /* load vector elements of 8x8 block */
259 DUP4_ARG2(__lsx_vld, input, 0, input, 16, input, 32, input, 48,
260 in0, in1, in2, in3);
261 DUP4_ARG2(__lsx_vld, input, 64, input, 80, input, 96, input, 112,
262 in4, in5, in6, in7);
263 __lsx_vst(zero, input, 0);
264 __lsx_vst(zero, input, 16);
265 __lsx_vst(zero, input, 32);
266 __lsx_vst(zero, input, 48);
267 __lsx_vst(zero, input, 64);
268 __lsx_vst(zero, input, 80);
269 __lsx_vst(zero, input, 96);
270 __lsx_vst(zero, input, 112);
271 DUP4_ARG2(__lsx_vilvl_d,in1, in0, in3, in2, in5, in4, in7,
272 in6, in0, in1, in2, in3);
273
274 /* stage1 */
275 DUP2_ARG2(__lsx_vilvh_h, in3, in0, in2, in1, s0, s1);
276 k0 = VP9_SET_COSPI_PAIR(cospi_28_64, -cospi_4_64);
277 k1 = VP9_SET_COSPI_PAIR(cospi_4_64, cospi_28_64);
278 k2 = VP9_SET_COSPI_PAIR(-cospi_20_64, cospi_12_64);
279 k3 = VP9_SET_COSPI_PAIR(cospi_12_64, cospi_20_64);
280 DUP4_ARG2(__lsx_vdp2_w_h, s0, k0, s0, k1, s1, k2, s1, k3,
281 tmp0, tmp1, tmp2, tmp3);
282 DUP4_ARG2(__lsx_vsrari_w, tmp0, VP9_DCT_CONST_BITS, tmp1,
283 VP9_DCT_CONST_BITS, tmp2, VP9_DCT_CONST_BITS, tmp3,
284 VP9_DCT_CONST_BITS, tmp0, tmp1, tmp2, tmp3);
285 DUP4_ARG2(__lsx_vpickev_h, zero, tmp0, zero, tmp1, zero, tmp2, zero, tmp3,
286 s0, s1, s2, s3);
287 LSX_BUTTERFLY_4_H(s0, s1, s3, s2, s4, s7, s6, s5);
288
289 /* stage2 */
290 DUP2_ARG2(__lsx_vilvl_h, in3, in1, in2, in0, s1, s0);
291 k0 = VP9_SET_COSPI_PAIR(cospi_16_64, cospi_16_64);
292 k1 = VP9_SET_COSPI_PAIR(cospi_16_64, -cospi_16_64);
293 k2 = VP9_SET_COSPI_PAIR(cospi_24_64, -cospi_8_64);
294 k3 = VP9_SET_COSPI_PAIR(cospi_8_64, cospi_24_64);
295 DUP4_ARG2(__lsx_vdp2_w_h, s0, k0, s0, k1, s1, k2, s1, k3,
296 tmp0, tmp1, tmp2, tmp3);
297 DUP4_ARG2(__lsx_vsrari_w, tmp0, VP9_DCT_CONST_BITS, tmp1,
298 VP9_DCT_CONST_BITS, tmp2, VP9_DCT_CONST_BITS, tmp3,
299 VP9_DCT_CONST_BITS, tmp0, tmp1, tmp2, tmp3);
300 DUP4_ARG2(__lsx_vpickev_h, zero, tmp0, zero, tmp1, zero, tmp2, zero, tmp3,
301 s0, s1, s2, s3);
302 LSX_BUTTERFLY_4_H(s0, s1, s2, s3, m0, m1, m2, m3);
303
304 /* stage3 */
305 s0 = __lsx_vilvl_h(s6, s5);
306
307 k1 = VP9_SET_COSPI_PAIR(-cospi_16_64, cospi_16_64);
308 DUP2_ARG2(__lsx_vdp2_w_h, s0, k1, s0, k0, tmp0, tmp1);
309 DUP2_ARG2(__lsx_vsrari_w, tmp0, VP9_DCT_CONST_BITS, tmp1,
310 VP9_DCT_CONST_BITS, tmp0, tmp1);
311 DUP2_ARG2(__lsx_vpickev_h, zero, tmp0, zero, tmp1, s2, s3);
312
313 /* stage4 */
314 LSX_BUTTERFLY_8_H(m0, m1, m2, m3, s4, s2, s3, s7,
315 in0, in1, in2, in3, in4, in5, in6, in7);
316 VP9_ILVLTRANS4x8_H(in0, in1, in2, in3, in4, in5, in6, in7,
317 in0, in1, in2, in3, in4, in5, in6, in7);
318 VP9_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7,
319 in0, in1, in2, in3, in4, in5, in6, in7);
320
321 /* final rounding (add 2^4, divide by 2^5) and shift */
322 DUP4_ARG2(__lsx_vsrari_h, in0 , 5, in1, 5, in2, 5, in3, 5,
323 in0, in1, in2, in3);
324 DUP4_ARG2(__lsx_vsrari_h, in4 , 5, in5, 5, in6, 5, in7, 5,
325 in4, in5, in6, in7);
326
327 /* add block and store 8x8 */
328 VP9_ADDBLK_ST8x4_UB(dst, dst_stride, in0, in1, in2, in3);
329 dst += (4 * dst_stride);
330 VP9_ADDBLK_ST8x4_UB(dst, dst_stride, in4, in5, in6, in7);
331 }
332
vp9_idct8x8_colcol_addblk_lsx(int16_t *input, uint8_t *dst, int32_t dst_stride)333 static void vp9_idct8x8_colcol_addblk_lsx(int16_t *input, uint8_t *dst,
334 int32_t dst_stride)
335 {
336 __m128i in0, in1, in2, in3, in4, in5, in6, in7;
337 __m128i zero = __lsx_vldi(0);
338
339 /* load vector elements of 8x8 block */
340 DUP4_ARG2(__lsx_vld, input, 0, input, 16, input, 32, input, 48,
341 in0, in1, in2, in3);
342 DUP4_ARG2(__lsx_vld, input, 64, input, 80, input, 96, input, 112,
343 in4, in5, in6, in7);
344 __lsx_vst(zero, input, 0);
345 __lsx_vst(zero, input, 16);
346 __lsx_vst(zero, input, 32);
347 __lsx_vst(zero, input, 48);
348 __lsx_vst(zero, input, 64);
349 __lsx_vst(zero, input, 80);
350 __lsx_vst(zero, input, 96);
351 __lsx_vst(zero, input, 112);
352 /* 1D idct8x8 */
353 VP9_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7,
354 in0, in1, in2, in3, in4, in5, in6, in7);
355 /* columns transform */
356 LSX_TRANSPOSE8x8_H(in0, in1, in2, in3, in4, in5, in6, in7,
357 in0, in1, in2, in3, in4, in5, in6, in7);
358 /* 1D idct8x8 */
359 VP9_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7,
360 in0, in1, in2, in3, in4, in5, in6, in7);
361 /* final rounding (add 2^4, divide by 2^5) and shift */
362 DUP4_ARG2(__lsx_vsrari_h, in0, 5, in1, 5, in2, 5, in3, 5,
363 in0, in1, in2, in3);
364 DUP4_ARG2(__lsx_vsrari_h, in4, 5, in5, 5, in6, 5, in7, 5,
365 in4, in5, in6, in7);
366 /* add block and store 8x8 */
367 VP9_ADDBLK_ST8x4_UB(dst, dst_stride, in0, in1, in2, in3);
368 dst += (4 * dst_stride);
369 VP9_ADDBLK_ST8x4_UB(dst, dst_stride, in4, in5, in6, in7);
370 }
371
vp9_idct16_1d_columns_addblk_lsx(int16_t *input, uint8_t *dst, int32_t dst_stride)372 static void vp9_idct16_1d_columns_addblk_lsx(int16_t *input, uint8_t *dst,
373 int32_t dst_stride)
374 {
375 __m128i loc0, loc1, loc2, loc3;
376 __m128i reg0, reg2, reg4, reg6, reg8, reg10, reg12, reg14;
377 __m128i reg1, reg3, reg5, reg7, reg9, reg11, reg13, reg15;
378 __m128i tmp5, tmp6, tmp7;
379 __m128i zero = __lsx_vldi(0);
380 int32_t offset = dst_stride << 2;
381
382 DUP4_ARG2(__lsx_vld, input, 32*0, input, 32*1, input, 32*2, input, 32*3,
383 reg0, reg1, reg2, reg3);
384 DUP4_ARG2(__lsx_vld, input, 32*4, input, 32*5, input, 32*6, input, 32*7,
385 reg4, reg5, reg6, reg7);
386 DUP4_ARG2(__lsx_vld, input, 32*8, input, 32*9, input, 32*10, input, 32*11,
387 reg8, reg9, reg10, reg11);
388 DUP4_ARG2(__lsx_vld, input, 32*12, input, 32*13, input, 32*14, input,
389 32*15, reg12, reg13, reg14, reg15);
390
391 __lsx_vst(zero, input, 32*0);
392 __lsx_vst(zero, input, 32*1);
393 __lsx_vst(zero, input, 32*2);
394 __lsx_vst(zero, input, 32*3);
395 __lsx_vst(zero, input, 32*4);
396 __lsx_vst(zero, input, 32*5);
397 __lsx_vst(zero, input, 32*6);
398 __lsx_vst(zero, input, 32*7);
399 __lsx_vst(zero, input, 32*8);
400 __lsx_vst(zero, input, 32*9);
401 __lsx_vst(zero, input, 32*10);
402 __lsx_vst(zero, input, 32*11);
403 __lsx_vst(zero, input, 32*12);
404 __lsx_vst(zero, input, 32*13);
405 __lsx_vst(zero, input, 32*14);
406 __lsx_vst(zero, input, 32*15);
407
408 VP9_DOTP_CONST_PAIR(reg2, reg14, cospi_28_64, cospi_4_64, reg2, reg14);
409 VP9_DOTP_CONST_PAIR(reg10, reg6, cospi_12_64, cospi_20_64, reg10, reg6);
410 LSX_BUTTERFLY_4_H(reg2, reg14, reg6, reg10, loc0, loc1, reg14, reg2);
411 VP9_DOTP_CONST_PAIR(reg14, reg2, cospi_16_64, cospi_16_64, loc2, loc3);
412 VP9_DOTP_CONST_PAIR(reg0, reg8, cospi_16_64, cospi_16_64, reg0, reg8);
413 VP9_DOTP_CONST_PAIR(reg4, reg12, cospi_24_64, cospi_8_64, reg4, reg12);
414 LSX_BUTTERFLY_4_H(reg8, reg0, reg4, reg12, reg2, reg6, reg10, reg14);
415
416 reg0 = __lsx_vsub_h(reg2, loc1);
417 reg2 = __lsx_vadd_h(reg2, loc1);
418 reg12 = __lsx_vsub_h(reg14, loc0);
419 reg14 = __lsx_vadd_h(reg14, loc0);
420 reg4 = __lsx_vsub_h(reg6, loc3);
421 reg6 = __lsx_vadd_h(reg6, loc3);
422 reg8 = __lsx_vsub_h(reg10, loc2);
423 reg10 = __lsx_vadd_h(reg10, loc2);
424
425 /* stage2 */
426 VP9_DOTP_CONST_PAIR(reg1, reg15, cospi_30_64, cospi_2_64, reg1, reg15);
427 VP9_DOTP_CONST_PAIR(reg9, reg7, cospi_14_64, cospi_18_64, loc2, loc3);
428
429 reg9 = __lsx_vsub_h(reg1, loc2);
430 reg1 = __lsx_vadd_h(reg1, loc2);
431 reg7 = __lsx_vsub_h(reg15, loc3);
432 reg15 = __lsx_vadd_h(reg15, loc3);
433
434 VP9_DOTP_CONST_PAIR(reg5, reg11, cospi_22_64, cospi_10_64, reg5, reg11);
435 VP9_DOTP_CONST_PAIR(reg13, reg3, cospi_6_64, cospi_26_64, loc0, loc1);
436 LSX_BUTTERFLY_4_H(loc0, loc1, reg11, reg5, reg13, reg3, reg11, reg5);
437
438 loc1 = __lsx_vadd_h(reg15, reg3);
439 reg3 = __lsx_vsub_h(reg15, reg3);
440 loc2 = __lsx_vadd_h(reg2, loc1);
441 reg15 = __lsx_vsub_h(reg2, loc1);
442
443 loc1 = __lsx_vadd_h(reg1, reg13);
444 reg13 = __lsx_vsub_h(reg1, reg13);
445 loc0 = __lsx_vadd_h(reg0, loc1);
446 loc1 = __lsx_vsub_h(reg0, loc1);
447 tmp6 = loc0;
448 tmp7 = loc1;
449 reg0 = loc2;
450
451 VP9_DOTP_CONST_PAIR(reg7, reg9, cospi_24_64, cospi_8_64, reg7, reg9);
452 VP9_DOTP_CONST_PAIR(__lsx_vneg_h(reg5), __lsx_vneg_h(reg11), cospi_8_64,
453 cospi_24_64, reg5, reg11);
454
455 loc0 = __lsx_vadd_h(reg9, reg5);
456 reg5 = __lsx_vsub_h(reg9, reg5);
457 reg2 = __lsx_vadd_h(reg6, loc0);
458 reg1 = __lsx_vsub_h(reg6, loc0);
459
460 loc0 = __lsx_vadd_h(reg7, reg11);
461 reg11 = __lsx_vsub_h(reg7, reg11);
462 loc1 = __lsx_vadd_h(reg4, loc0);
463 loc2 = __lsx_vsub_h(reg4, loc0);
464 tmp5 = loc1;
465
466 VP9_DOTP_CONST_PAIR(reg5, reg11, cospi_16_64, cospi_16_64, reg5, reg11);
467 LSX_BUTTERFLY_4_H(reg8, reg10, reg11, reg5, loc0, reg4, reg9, loc1);
468
469 reg10 = loc0;
470 reg11 = loc1;
471
472 VP9_DOTP_CONST_PAIR(reg3, reg13, cospi_16_64, cospi_16_64, reg3, reg13);
473 LSX_BUTTERFLY_4_H(reg12, reg14, reg13, reg3, reg8, reg6, reg7, reg5);
474 reg13 = loc2;
475
476 /* Transpose and store the output */
477 reg12 = tmp5;
478 reg14 = tmp6;
479 reg3 = tmp7;
480
481 DUP4_ARG2(__lsx_vsrari_h, reg0, 6, reg2, 6, reg4, 6, reg6, 6,
482 reg0, reg2, reg4, reg6);
483 VP9_ADDBLK_ST8x4_UB(dst, dst_stride, reg0, reg2, reg4, reg6);
484 dst += offset;
485 DUP4_ARG2(__lsx_vsrari_h, reg8, 6, reg10, 6, reg12, 6, reg14, 6,
486 reg8, reg10, reg12, reg14);
487 VP9_ADDBLK_ST8x4_UB(dst, dst_stride, reg8, reg10, reg12, reg14);
488 dst += offset;
489 DUP4_ARG2(__lsx_vsrari_h, reg3, 6, reg5, 6, reg11, 6, reg13, 6,
490 reg3, reg5, reg11, reg13);
491 VP9_ADDBLK_ST8x4_UB(dst, dst_stride, reg3, reg13, reg11, reg5);
492 dst += offset;
493 DUP4_ARG2(__lsx_vsrari_h, reg1, 6, reg7, 6, reg9, 6, reg15, 6,
494 reg1, reg7, reg9, reg15);
495 VP9_ADDBLK_ST8x4_UB(dst, dst_stride, reg7, reg9, reg1, reg15);
496 }
497
vp9_idct16_1d_columns_lsx(int16_t *input, int16_t *output)498 static void vp9_idct16_1d_columns_lsx(int16_t *input, int16_t *output)
499 {
500 __m128i loc0, loc1, loc2, loc3;
501 __m128i reg1, reg3, reg5, reg7, reg9, reg11, reg13, reg15;
502 __m128i reg0, reg2, reg4, reg6, reg8, reg10, reg12, reg14;
503 __m128i tmp5, tmp6, tmp7;
504 __m128i zero = __lsx_vldi(0);
505 int16_t *offset;
506
507 DUP4_ARG2(__lsx_vld, input, 32*0, input, 32*1, input, 32*2, input, 32*3,
508 reg0, reg1, reg2, reg3);
509 DUP4_ARG2(__lsx_vld, input, 32*4, input, 32*5, input, 32*6, input, 32*7,
510 reg4, reg5, reg6, reg7);
511 DUP4_ARG2(__lsx_vld, input, 32*8, input, 32*9, input, 32*10, input, 32*11,
512 reg8, reg9, reg10, reg11);
513 DUP4_ARG2(__lsx_vld, input, 32*12, input, 32*13, input, 32*14, input,
514 32*15, reg12, reg13, reg14, reg15);
515
516 __lsx_vst(zero, input, 32*0);
517 __lsx_vst(zero, input, 32*1);
518 __lsx_vst(zero, input, 32*2);
519 __lsx_vst(zero, input, 32*3);
520 __lsx_vst(zero, input, 32*4);
521 __lsx_vst(zero, input, 32*5);
522 __lsx_vst(zero, input, 32*6);
523 __lsx_vst(zero, input, 32*7);
524 __lsx_vst(zero, input, 32*8);
525 __lsx_vst(zero, input, 32*9);
526 __lsx_vst(zero, input, 32*10);
527 __lsx_vst(zero, input, 32*11);
528 __lsx_vst(zero, input, 32*12);
529 __lsx_vst(zero, input, 32*13);
530 __lsx_vst(zero, input, 32*14);
531 __lsx_vst(zero, input, 32*15);
532
533 VP9_DOTP_CONST_PAIR(reg2, reg14, cospi_28_64, cospi_4_64, reg2, reg14);
534 VP9_DOTP_CONST_PAIR(reg10, reg6, cospi_12_64, cospi_20_64, reg10, reg6);
535 LSX_BUTTERFLY_4_H(reg2, reg14, reg6, reg10, loc0, loc1, reg14, reg2);
536 VP9_DOTP_CONST_PAIR(reg14, reg2, cospi_16_64, cospi_16_64, loc2, loc3);
537 VP9_DOTP_CONST_PAIR(reg0, reg8, cospi_16_64, cospi_16_64, reg0, reg8);
538 VP9_DOTP_CONST_PAIR(reg4, reg12, cospi_24_64, cospi_8_64, reg4, reg12);
539 LSX_BUTTERFLY_4_H(reg8, reg0, reg4, reg12, reg2, reg6, reg10, reg14);
540
541 reg0 = __lsx_vsub_h(reg2, loc1);
542 reg2 = __lsx_vadd_h(reg2, loc1);
543 reg12 = __lsx_vsub_h(reg14, loc0);
544 reg14 = __lsx_vadd_h(reg14, loc0);
545 reg4 = __lsx_vsub_h(reg6, loc3);
546 reg6 = __lsx_vadd_h(reg6, loc3);
547 reg8 = __lsx_vsub_h(reg10, loc2);
548 reg10 = __lsx_vadd_h(reg10, loc2);
549
550 /* stage2 */
551 VP9_DOTP_CONST_PAIR(reg1, reg15, cospi_30_64, cospi_2_64, reg1, reg15);
552 VP9_DOTP_CONST_PAIR(reg9, reg7, cospi_14_64, cospi_18_64, loc2, loc3);
553
554 reg9 = __lsx_vsub_h(reg1, loc2);
555 reg1 = __lsx_vadd_h(reg1, loc2);
556 reg7 = __lsx_vsub_h(reg15, loc3);
557 reg15 = __lsx_vadd_h(reg15, loc3);
558
559 VP9_DOTP_CONST_PAIR(reg5, reg11, cospi_22_64, cospi_10_64, reg5, reg11);
560 VP9_DOTP_CONST_PAIR(reg13, reg3, cospi_6_64, cospi_26_64, loc0, loc1);
561 LSX_BUTTERFLY_4_H(loc0, loc1, reg11, reg5, reg13, reg3, reg11, reg5);
562
563 loc1 = __lsx_vadd_h(reg15, reg3);
564 reg3 = __lsx_vsub_h(reg15, reg3);
565 loc2 = __lsx_vadd_h(reg2, loc1);
566 reg15 = __lsx_vsub_h(reg2, loc1);
567
568 loc1 = __lsx_vadd_h(reg1, reg13);
569 reg13 = __lsx_vsub_h(reg1, reg13);
570 loc0 = __lsx_vadd_h(reg0, loc1);
571 loc1 = __lsx_vsub_h(reg0, loc1);
572 tmp6 = loc0;
573 tmp7 = loc1;
574 reg0 = loc2;
575
576 VP9_DOTP_CONST_PAIR(reg7, reg9, cospi_24_64, cospi_8_64, reg7, reg9);
577 VP9_DOTP_CONST_PAIR(__lsx_vneg_h(reg5), __lsx_vneg_h(reg11), cospi_8_64,
578 cospi_24_64, reg5, reg11);
579
580 loc0 = __lsx_vadd_h(reg9, reg5);
581 reg5 = __lsx_vsub_h(reg9, reg5);
582 reg2 = __lsx_vadd_h(reg6, loc0);
583 reg1 = __lsx_vsub_h(reg6, loc0);
584
585 loc0 = __lsx_vadd_h(reg7, reg11);
586 reg11 = __lsx_vsub_h(reg7, reg11);
587 loc1 = __lsx_vadd_h(reg4, loc0);
588 loc2 = __lsx_vsub_h(reg4, loc0);
589
590 tmp5 = loc1;
591
592 VP9_DOTP_CONST_PAIR(reg5, reg11, cospi_16_64, cospi_16_64, reg5, reg11);
593 LSX_BUTTERFLY_4_H(reg8, reg10, reg11, reg5, loc0, reg4, reg9, loc1);
594
595 reg10 = loc0;
596 reg11 = loc1;
597
598 VP9_DOTP_CONST_PAIR(reg3, reg13, cospi_16_64, cospi_16_64, reg3, reg13);
599 LSX_BUTTERFLY_4_H(reg12, reg14, reg13, reg3, reg8, reg6, reg7, reg5);
600 reg13 = loc2;
601
602 /* Transpose and store the output */
603 reg12 = tmp5;
604 reg14 = tmp6;
605 reg3 = tmp7;
606
607 /* transpose block */
608 LSX_TRANSPOSE8x8_H(reg0, reg2, reg4, reg6, reg8, reg10, reg12, reg14,
609 reg0, reg2, reg4, reg6, reg8, reg10, reg12, reg14);
610
611 __lsx_vst(reg0, output, 32*0);
612 __lsx_vst(reg2, output, 32*1);
613 __lsx_vst(reg4, output, 32*2);
614 __lsx_vst(reg6, output, 32*3);
615 __lsx_vst(reg8, output, 32*4);
616 __lsx_vst(reg10, output, 32*5);
617 __lsx_vst(reg12, output, 32*6);
618 __lsx_vst(reg14, output, 32*7);
619
620 /* transpose block */
621 LSX_TRANSPOSE8x8_H(reg3, reg13, reg11, reg5, reg7, reg9, reg1, reg15,
622 reg3, reg13, reg11, reg5, reg7, reg9, reg1, reg15);
623
624 offset = output + 8;
625 __lsx_vst(reg3, offset, 32*0);
626 __lsx_vst(reg13, offset, 32*1);
627 __lsx_vst(reg11, offset, 32*2);
628 __lsx_vst(reg5, offset, 32*3);
629
630 offset = output + 8 + 4 * 16;
631 __lsx_vst(reg7, offset, 32*0);
632 __lsx_vst(reg9, offset, 32*1);
633 __lsx_vst(reg1, offset, 32*2);
634 __lsx_vst(reg15, offset, 32*3);
635 }
636
vp9_idct16x16_1_add_lsx(int16_t *input, uint8_t *dst, int32_t dst_stride)637 static void vp9_idct16x16_1_add_lsx(int16_t *input, uint8_t *dst,
638 int32_t dst_stride)
639 {
640 uint8_t i;
641 int16_t out;
642 __m128i vec, res0, res1, res2, res3, res4, res5, res6, res7;
643 __m128i dst0, dst1, dst2, dst3, tmp0, tmp1, tmp2, tmp3;
644 int32_t stride2 = dst_stride << 1;
645 int32_t stride3 = stride2 + dst_stride;
646 int32_t stride4 = stride2 << 1;
647
648 out = ROUND_POWER_OF_TWO((input[0] * cospi_16_64), VP9_DCT_CONST_BITS);
649 out = ROUND_POWER_OF_TWO((out * cospi_16_64), VP9_DCT_CONST_BITS);
650 out = ROUND_POWER_OF_TWO(out, 6);
651 input[0] = 0;
652 vec = __lsx_vreplgr2vr_h(out);
653
654 for (i = 4; i--;) {
655 dst0 = __lsx_vld(dst, 0);
656 DUP2_ARG2(__lsx_vldx, dst, dst_stride, dst, stride2, dst1, dst2);
657 dst3 = __lsx_vldx(dst, stride3);
658 VP9_UNPCK_UB_SH(dst0, res4, res0);
659 VP9_UNPCK_UB_SH(dst1, res5, res1);
660 VP9_UNPCK_UB_SH(dst2, res6, res2);
661 VP9_UNPCK_UB_SH(dst3, res7, res3);
662 DUP4_ARG2(__lsx_vadd_h, res0, vec, res1, vec, res2, vec, res3, vec,
663 res0, res1, res2, res3);
664 DUP4_ARG2(__lsx_vadd_h, res4, vec, res5, vec, res6, vec, res7, vec,
665 res4, res5, res6, res7);
666 DUP4_ARG1(__lsx_vclip255_h, res0, res1, res2, res3,
667 res0, res1, res2, res3);
668 DUP4_ARG1(__lsx_vclip255_h, res4, res5, res6, res7,
669 res4, res5, res6, res7);
670 DUP4_ARG2(__lsx_vpickev_b, res4, res0, res5, res1, res6,
671 res2, res7, res3, tmp0, tmp1, tmp2, tmp3);
672 __lsx_vst(tmp0, dst, 0);
673 __lsx_vstx(tmp1, dst, dst_stride);
674 __lsx_vstx(tmp2, dst, stride2);
675 __lsx_vstx(tmp3, dst, stride3);
676 dst += stride4;
677 }
678 }
679
vp9_idct16x16_10_colcol_addblk_lsx(int16_t *input, uint8_t *dst, int32_t dst_stride)680 static void vp9_idct16x16_10_colcol_addblk_lsx(int16_t *input, uint8_t *dst,
681 int32_t dst_stride)
682 {
683 int32_t i;
684 int16_t out_arr[16 * 16] ALLOC_ALIGNED(16);
685 int16_t *out = out_arr;
686 __m128i zero = __lsx_vldi(0);
687
688 /* transform rows */
689 vp9_idct16_1d_columns_lsx(input, out);
690
691 /* short case just considers top 4 rows as valid output */
692 out += 4 * 16;
693 for (i = 3; i--;) {
694 __lsx_vst(zero, out, 0);
695 __lsx_vst(zero, out, 16);
696 __lsx_vst(zero, out, 32);
697 __lsx_vst(zero, out, 48);
698 __lsx_vst(zero, out, 64);
699 __lsx_vst(zero, out, 80);
700 __lsx_vst(zero, out, 96);
701 __lsx_vst(zero, out, 112);
702 out += 64;
703 }
704
705 out = out_arr;
706
707 /* transform columns */
708 for (i = 0; i < 2; i++) {
709 /* process 8 * 16 block */
710 vp9_idct16_1d_columns_addblk_lsx((out + (i << 3)), (dst + (i << 3)),
711 dst_stride);
712 }
713 }
714
vp9_idct16x16_colcol_addblk_lsx(int16_t *input, uint8_t *dst, int32_t dst_stride)715 static void vp9_idct16x16_colcol_addblk_lsx(int16_t *input, uint8_t *dst,
716 int32_t dst_stride)
717 {
718 int32_t i;
719 int16_t out_arr[16 * 16] ALLOC_ALIGNED(16);
720 int16_t *out = out_arr;
721
722 /* transform rows */
723 for (i = 0; i < 2; i++) {
724 /* process 8 * 16 block */
725 vp9_idct16_1d_columns_lsx((input + (i << 3)), (out + (i << 7)));
726 }
727
728 /* transform columns */
729 for (i = 0; i < 2; i++) {
730 /* process 8 * 16 block */
731 vp9_idct16_1d_columns_addblk_lsx((out + (i << 3)), (dst + (i << 3)),
732 dst_stride);
733 }
734 }
735
vp9_idct_butterfly_transpose_store(int16_t *tmp_buf, int16_t *tmp_eve_buf, int16_t *tmp_odd_buf, int16_t *dst)736 static void vp9_idct_butterfly_transpose_store(int16_t *tmp_buf,
737 int16_t *tmp_eve_buf,
738 int16_t *tmp_odd_buf,
739 int16_t *dst)
740 {
741 __m128i vec0, vec1, vec2, vec3, loc0, loc1, loc2, loc3;
742 __m128i m0, m1, m2, m3, m4, m5, m6, m7, n0, n1, n2, n3, n4, n5, n6, n7;
743
744 /* FINAL BUTTERFLY : Dependency on Even & Odd */
745 vec0 = __lsx_vld(tmp_odd_buf, 0);
746 vec1 = __lsx_vld(tmp_odd_buf, 9 * 16);
747 vec2 = __lsx_vld(tmp_odd_buf, 14 * 16);
748 vec3 = __lsx_vld(tmp_odd_buf, 6 * 16);
749 loc0 = __lsx_vld(tmp_eve_buf, 0);
750 loc1 = __lsx_vld(tmp_eve_buf, 8 * 16);
751 loc2 = __lsx_vld(tmp_eve_buf, 4 * 16);
752 loc3 = __lsx_vld(tmp_eve_buf, 12 * 16);
753
754 DUP4_ARG2(__lsx_vadd_h,loc0, vec3, loc1, vec2, loc2, vec1, loc3, vec0,
755 m0, m4, m2, m6);
756
757 #define SUB(a, b) __lsx_vsub_h(a, b)
758
759 __lsx_vst(SUB(loc0, vec3), tmp_buf, 31 * 16);
760 __lsx_vst(SUB(loc1, vec2), tmp_buf, 23 * 16);
761 __lsx_vst(SUB(loc2, vec1), tmp_buf, 27 * 16);
762 __lsx_vst(SUB(loc3, vec0), tmp_buf, 19 * 16);
763
764 /* Load 8 & Store 8 */
765 vec0 = __lsx_vld(tmp_odd_buf, 4 * 16);
766 vec1 = __lsx_vld(tmp_odd_buf, 13 * 16);
767 vec2 = __lsx_vld(tmp_odd_buf, 10 * 16);
768 vec3 = __lsx_vld(tmp_odd_buf, 3 * 16);
769 loc0 = __lsx_vld(tmp_eve_buf, 2 * 16);
770 loc1 = __lsx_vld(tmp_eve_buf, 10 * 16);
771 loc2 = __lsx_vld(tmp_eve_buf, 6 * 16);
772 loc3 = __lsx_vld(tmp_eve_buf, 14 * 16);
773
774 DUP4_ARG2(__lsx_vadd_h, loc0, vec3, loc1, vec2, loc2, vec1, loc3, vec0,
775 m1, m5, m3, m7);
776
777 __lsx_vst(SUB(loc0, vec3), tmp_buf, 29 * 16);
778 __lsx_vst(SUB(loc1, vec2), tmp_buf, 21 * 16);
779 __lsx_vst(SUB(loc2, vec1), tmp_buf, 25 * 16);
780 __lsx_vst(SUB(loc3, vec0), tmp_buf, 17 * 16);
781
782 /* Load 8 & Store 8 */
783 vec0 = __lsx_vld(tmp_odd_buf, 2 * 16);
784 vec1 = __lsx_vld(tmp_odd_buf, 11 * 16);
785 vec2 = __lsx_vld(tmp_odd_buf, 12 * 16);
786 vec3 = __lsx_vld(tmp_odd_buf, 7 * 16);
787 loc0 = __lsx_vld(tmp_eve_buf, 1 * 16);
788 loc1 = __lsx_vld(tmp_eve_buf, 9 * 16);
789 loc2 = __lsx_vld(tmp_eve_buf, 5 * 16);
790 loc3 = __lsx_vld(tmp_eve_buf, 13 * 16);
791
792 DUP4_ARG2(__lsx_vadd_h, loc0, vec3, loc1, vec2, loc2, vec1, loc3, vec0,
793 n0, n4, n2, n6);
794
795 __lsx_vst(SUB(loc0, vec3), tmp_buf, 30 * 16);
796 __lsx_vst(SUB(loc1, vec2), tmp_buf, 22 * 16);
797 __lsx_vst(SUB(loc2, vec1), tmp_buf, 26 * 16);
798 __lsx_vst(SUB(loc3, vec0), tmp_buf, 18 * 16);
799
800 /* Load 8 & Store 8 */
801 vec0 = __lsx_vld(tmp_odd_buf, 5 * 16);
802 vec1 = __lsx_vld(tmp_odd_buf, 15 * 16);
803 vec2 = __lsx_vld(tmp_odd_buf, 8 * 16);
804 vec3 = __lsx_vld(tmp_odd_buf, 1 * 16);
805 loc0 = __lsx_vld(tmp_eve_buf, 3 * 16);
806 loc1 = __lsx_vld(tmp_eve_buf, 11 * 16);
807 loc2 = __lsx_vld(tmp_eve_buf, 7 * 16);
808 loc3 = __lsx_vld(tmp_eve_buf, 15 * 16);
809
810 DUP4_ARG2(__lsx_vadd_h, loc0, vec3, loc1, vec2, loc2, vec1, loc3, vec0,
811 n1, n5, n3, n7);
812
813 __lsx_vst(SUB(loc0, vec3), tmp_buf, 28 * 16);
814 __lsx_vst(SUB(loc1, vec2), tmp_buf, 20 * 16);
815 __lsx_vst(SUB(loc2, vec1), tmp_buf, 24 * 16);
816 __lsx_vst(SUB(loc3, vec0), tmp_buf, 16 * 16);
817
818 /* Transpose : 16 vectors */
819 /* 1st & 2nd 8x8 */
820 LSX_TRANSPOSE8x8_H(m0, n0, m1, n1, m2, n2, m3, n3,
821 m0, n0, m1, n1, m2, n2, m3, n3);
822 __lsx_vst(m0, dst, 0);
823 __lsx_vst(n0, dst, 32 * 2);
824 __lsx_vst(m1, dst, 32 * 4);
825 __lsx_vst(n1, dst, 32 * 6);
826 __lsx_vst(m2, dst, 32 * 8);
827 __lsx_vst(n2, dst, 32 * 10);
828 __lsx_vst(m3, dst, 32 * 12);
829 __lsx_vst(n3, dst, 32 * 14);
830
831 LSX_TRANSPOSE8x8_H(m4, n4, m5, n5, m6, n6, m7, n7,
832 m4, n4, m5, n5, m6, n6, m7, n7);
833
834 __lsx_vst(m4, dst, 16);
835 __lsx_vst(n4, dst, 16 + 32 * 2);
836 __lsx_vst(m5, dst, 16 + 32 * 4);
837 __lsx_vst(n5, dst, 16 + 32 * 6);
838 __lsx_vst(m6, dst, 16 + 32 * 8);
839 __lsx_vst(n6, dst, 16 + 32 * 10);
840 __lsx_vst(m7, dst, 16 + 32 * 12);
841 __lsx_vst(n7, dst, 16 + 32 * 14);
842
843 /* 3rd & 4th 8x8 */
844 DUP4_ARG2(__lsx_vld, tmp_buf, 16 * 16, tmp_buf, 16 * 17,
845 tmp_buf, 16 * 18, tmp_buf, 16 * 19, m0, n0, m1, n1);
846 DUP4_ARG2(__lsx_vld, tmp_buf, 16 * 20, tmp_buf, 16 * 21,
847 tmp_buf, 16 * 22, tmp_buf, 16 * 23, m2, n2, m3, n3);
848
849 DUP4_ARG2(__lsx_vld, tmp_buf, 16 * 24, tmp_buf, 16 * 25,
850 tmp_buf, 16 * 26, tmp_buf, 16 * 27, m4, n4, m5, n5);
851 DUP4_ARG2(__lsx_vld, tmp_buf, 16 * 28, tmp_buf, 16 * 29,
852 tmp_buf, 16 * 30, tmp_buf, 16 * 31, m6, n6, m7, n7);
853
854 LSX_TRANSPOSE8x8_H(m0, n0, m1, n1, m2, n2, m3, n3,
855 m0, n0, m1, n1, m2, n2, m3, n3);
856
857 __lsx_vst(m0, dst, 32);
858 __lsx_vst(n0, dst, 32 + 32 * 2);
859 __lsx_vst(m1, dst, 32 + 32 * 4);
860 __lsx_vst(n1, dst, 32 + 32 * 6);
861 __lsx_vst(m2, dst, 32 + 32 * 8);
862 __lsx_vst(n2, dst, 32 + 32 * 10);
863 __lsx_vst(m3, dst, 32 + 32 * 12);
864 __lsx_vst(n3, dst, 32 + 32 * 14);
865
866 LSX_TRANSPOSE8x8_H(m4, n4, m5, n5, m6, n6, m7, n7,
867 m4, n4, m5, n5, m6, n6, m7, n7);
868
869 __lsx_vst(m4, dst, 48);
870 __lsx_vst(n4, dst, 48 + 32 * 2);
871 __lsx_vst(m5, dst, 48 + 32 * 4);
872 __lsx_vst(n5, dst, 48 + 32 * 6);
873 __lsx_vst(m6, dst, 48 + 32 * 8);
874 __lsx_vst(n6, dst, 48 + 32 * 10);
875 __lsx_vst(m7, dst, 48 + 32 * 12);
876 __lsx_vst(n7, dst, 48 + 32 * 14);
877 }
878
vp9_idct8x32_column_even_process_store(int16_t *tmp_buf, int16_t *tmp_eve_buf)879 static void vp9_idct8x32_column_even_process_store(int16_t *tmp_buf,
880 int16_t *tmp_eve_buf)
881 {
882 __m128i vec0, vec1, vec2, vec3, loc0, loc1, loc2, loc3;
883 __m128i reg0, reg1, reg2, reg3, reg4, reg5, reg6, reg7;
884 __m128i stp0, stp1, stp2, stp3, stp4, stp5, stp6, stp7;
885 __m128i zero = __lsx_vldi(0);
886
887 /* Even stage 1 */
888 DUP4_ARG2(__lsx_vld, tmp_buf, 0, tmp_buf, 32 * 8,
889 tmp_buf, 32 * 16, tmp_buf, 32 * 24, reg0, reg1, reg2, reg3);
890 DUP4_ARG2(__lsx_vld, tmp_buf, 32 * 32, tmp_buf, 32 * 40,
891 tmp_buf, 32 * 48, tmp_buf, 32 * 56, reg4, reg5, reg6, reg7);
892
893 __lsx_vst(zero, tmp_buf, 0);
894 __lsx_vst(zero, tmp_buf, 32 * 8);
895 __lsx_vst(zero, tmp_buf, 32 * 16);
896 __lsx_vst(zero, tmp_buf, 32 * 24);
897 __lsx_vst(zero, tmp_buf, 32 * 32);
898 __lsx_vst(zero, tmp_buf, 32 * 40);
899 __lsx_vst(zero, tmp_buf, 32 * 48);
900 __lsx_vst(zero, tmp_buf, 32 * 56);
901
902 tmp_buf += (2 * 32);
903
904 VP9_DOTP_CONST_PAIR(reg1, reg7, cospi_28_64, cospi_4_64, reg1, reg7);
905 VP9_DOTP_CONST_PAIR(reg5, reg3, cospi_12_64, cospi_20_64, reg5, reg3);
906 LSX_BUTTERFLY_4_H(reg1, reg7, reg3, reg5, vec1, vec3, vec2, vec0);
907 VP9_DOTP_CONST_PAIR(vec2, vec0, cospi_16_64, cospi_16_64, loc2, loc3);
908
909 loc1 = vec3;
910 loc0 = vec1;
911
912 VP9_DOTP_CONST_PAIR(reg0, reg4, cospi_16_64, cospi_16_64, reg0, reg4);
913 VP9_DOTP_CONST_PAIR(reg2, reg6, cospi_24_64, cospi_8_64, reg2, reg6);
914 LSX_BUTTERFLY_4_H(reg4, reg0, reg2, reg6, vec1, vec3, vec2, vec0);
915 LSX_BUTTERFLY_4_H(vec0, vec1, loc1, loc0, stp3, stp0, stp7, stp4);
916 LSX_BUTTERFLY_4_H(vec2, vec3, loc3, loc2, stp2, stp1, stp6, stp5);
917
918 /* Even stage 2 */
919 /* Load 8 */
920 DUP4_ARG2(__lsx_vld, tmp_buf, 0, tmp_buf, 32 * 8,
921 tmp_buf, 32 * 16, tmp_buf, 32 * 24, reg0, reg1, reg2, reg3);
922 DUP4_ARG2(__lsx_vld, tmp_buf, 32 * 32, tmp_buf, 32 * 40,
923 tmp_buf, 32 * 48, tmp_buf, 32 * 56, reg4, reg5, reg6, reg7);
924
925 __lsx_vst(zero, tmp_buf, 0);
926 __lsx_vst(zero, tmp_buf, 32 * 8);
927 __lsx_vst(zero, tmp_buf, 32 * 16);
928 __lsx_vst(zero, tmp_buf, 32 * 24);
929 __lsx_vst(zero, tmp_buf, 32 * 32);
930 __lsx_vst(zero, tmp_buf, 32 * 40);
931 __lsx_vst(zero, tmp_buf, 32 * 48);
932 __lsx_vst(zero, tmp_buf, 32 * 56);
933
934 VP9_DOTP_CONST_PAIR(reg0, reg7, cospi_30_64, cospi_2_64, reg0, reg7);
935 VP9_DOTP_CONST_PAIR(reg4, reg3, cospi_14_64, cospi_18_64, reg4, reg3);
936 VP9_DOTP_CONST_PAIR(reg2, reg5, cospi_22_64, cospi_10_64, reg2, reg5);
937 VP9_DOTP_CONST_PAIR(reg6, reg1, cospi_6_64, cospi_26_64, reg6, reg1);
938
939 vec0 = __lsx_vadd_h(reg0, reg4);
940 reg0 = __lsx_vsub_h(reg0, reg4);
941 reg4 = __lsx_vadd_h(reg6, reg2);
942 reg6 = __lsx_vsub_h(reg6, reg2);
943 reg2 = __lsx_vadd_h(reg1, reg5);
944 reg1 = __lsx_vsub_h(reg1, reg5);
945 reg5 = __lsx_vadd_h(reg7, reg3);
946 reg7 = __lsx_vsub_h(reg7, reg3);
947 reg3 = vec0;
948
949 vec1 = reg2;
950 reg2 = __lsx_vadd_h(reg3, reg4);
951 reg3 = __lsx_vsub_h(reg3, reg4);
952 reg4 = __lsx_vsub_h(reg5, vec1);
953 reg5 = __lsx_vadd_h(reg5, vec1);
954
955 VP9_DOTP_CONST_PAIR(reg7, reg0, cospi_24_64, cospi_8_64, reg0, reg7);
956 VP9_DOTP_CONST_PAIR(__lsx_vneg_h(reg6), reg1, cospi_24_64, cospi_8_64,
957 reg6, reg1);
958
959 vec0 = __lsx_vsub_h(reg0, reg6);
960 reg0 = __lsx_vadd_h(reg0, reg6);
961 vec1 = __lsx_vsub_h(reg7, reg1);
962 reg7 = __lsx_vadd_h(reg7, reg1);
963
964 VP9_DOTP_CONST_PAIR(vec1, vec0, cospi_16_64, cospi_16_64, reg6, reg1);
965 VP9_DOTP_CONST_PAIR(reg4, reg3, cospi_16_64, cospi_16_64, reg3, reg4);
966
967 /* Even stage 3 : Dependency on Even stage 1 & Even stage 2 */
968 /* Store 8 */
969 LSX_BUTTERFLY_4_H(stp0, stp1, reg7, reg5, loc1, loc3, loc2, loc0);
970 __lsx_vst(loc1, tmp_eve_buf, 0);
971 __lsx_vst(loc3, tmp_eve_buf, 16);
972 __lsx_vst(loc2, tmp_eve_buf, 14 * 16);
973 __lsx_vst(loc0, tmp_eve_buf, 14 * 16 + 16);
974 LSX_BUTTERFLY_4_H(stp2, stp3, reg4, reg1, loc1, loc3, loc2, loc0);
975 __lsx_vst(loc1, tmp_eve_buf, 2 * 16);
976 __lsx_vst(loc3, tmp_eve_buf, 2 * 16 + 16);
977 __lsx_vst(loc2, tmp_eve_buf, 12 * 16);
978 __lsx_vst(loc0, tmp_eve_buf, 12 * 16 + 16);
979
980 /* Store 8 */
981 LSX_BUTTERFLY_4_H(stp4, stp5, reg6, reg3, loc1, loc3, loc2, loc0);
982 __lsx_vst(loc1, tmp_eve_buf, 4 * 16);
983 __lsx_vst(loc3, tmp_eve_buf, 4 * 16 + 16);
984 __lsx_vst(loc2, tmp_eve_buf, 10 * 16);
985 __lsx_vst(loc0, tmp_eve_buf, 10 * 16 + 16);
986
987 LSX_BUTTERFLY_4_H(stp6, stp7, reg2, reg0, loc1, loc3, loc2, loc0);
988 __lsx_vst(loc1, tmp_eve_buf, 6 * 16);
989 __lsx_vst(loc3, tmp_eve_buf, 6 * 16 + 16);
990 __lsx_vst(loc2, tmp_eve_buf, 8 * 16);
991 __lsx_vst(loc0, tmp_eve_buf, 8 * 16 + 16);
992 }
993
vp9_idct8x32_column_odd_process_store(int16_t *tmp_buf, int16_t *tmp_odd_buf)994 static void vp9_idct8x32_column_odd_process_store(int16_t *tmp_buf,
995 int16_t *tmp_odd_buf)
996 {
997 __m128i vec0, vec1, vec2, vec3, loc0, loc1, loc2, loc3;
998 __m128i reg0, reg1, reg2, reg3, reg4, reg5, reg6, reg7;
999 __m128i zero = __lsx_vldi(0);
1000
1001 /* Odd stage 1 */
1002 reg0 = __lsx_vld(tmp_buf, 64);
1003 reg1 = __lsx_vld(tmp_buf, 7 * 64);
1004 reg2 = __lsx_vld(tmp_buf, 9 * 64);
1005 reg3 = __lsx_vld(tmp_buf, 15 * 64);
1006 reg4 = __lsx_vld(tmp_buf, 17 * 64);
1007 reg5 = __lsx_vld(tmp_buf, 23 * 64);
1008 reg6 = __lsx_vld(tmp_buf, 25 * 64);
1009 reg7 = __lsx_vld(tmp_buf, 31 * 64);
1010
1011 __lsx_vst(zero, tmp_buf, 64);
1012 __lsx_vst(zero, tmp_buf, 7 * 64);
1013 __lsx_vst(zero, tmp_buf, 9 * 64);
1014 __lsx_vst(zero, tmp_buf, 15 * 64);
1015 __lsx_vst(zero, tmp_buf, 17 * 64);
1016 __lsx_vst(zero, tmp_buf, 23 * 64);
1017 __lsx_vst(zero, tmp_buf, 25 * 64);
1018 __lsx_vst(zero, tmp_buf, 31 * 64);
1019
1020 VP9_DOTP_CONST_PAIR(reg0, reg7, cospi_31_64, cospi_1_64, reg0, reg7);
1021 VP9_DOTP_CONST_PAIR(reg4, reg3, cospi_15_64, cospi_17_64, reg3, reg4);
1022 VP9_DOTP_CONST_PAIR(reg2, reg5, cospi_23_64, cospi_9_64, reg2, reg5);
1023 VP9_DOTP_CONST_PAIR(reg6, reg1, cospi_7_64, cospi_25_64, reg1, reg6);
1024
1025 vec0 = __lsx_vadd_h(reg0, reg3);
1026 reg0 = __lsx_vsub_h(reg0, reg3);
1027 reg3 = __lsx_vadd_h(reg7, reg4);
1028 reg7 = __lsx_vsub_h(reg7, reg4);
1029 reg4 = __lsx_vadd_h(reg1, reg2);
1030 reg1 = __lsx_vsub_h(reg1, reg2);
1031 reg2 = __lsx_vadd_h(reg6, reg5);
1032 reg6 = __lsx_vsub_h(reg6, reg5);
1033 reg5 = vec0;
1034
1035 /* 4 Stores */
1036 DUP2_ARG2(__lsx_vadd_h, reg5, reg4, reg3, reg2, vec0, vec1);
1037 __lsx_vst(vec0, tmp_odd_buf, 4 * 16);
1038 __lsx_vst(vec1, tmp_odd_buf, 4 * 16 + 16);
1039 DUP2_ARG2(__lsx_vsub_h, reg5, reg4, reg3, reg2, vec0, vec1);
1040 VP9_DOTP_CONST_PAIR(vec1, vec0, cospi_24_64, cospi_8_64, vec0, vec1);
1041 __lsx_vst(vec0, tmp_odd_buf, 0);
1042 __lsx_vst(vec1, tmp_odd_buf, 16);
1043
1044 /* 4 Stores */
1045 VP9_DOTP_CONST_PAIR(reg7, reg0, cospi_28_64, cospi_4_64, reg0, reg7);
1046 VP9_DOTP_CONST_PAIR(reg6, reg1, -cospi_4_64, cospi_28_64, reg1, reg6);
1047 LSX_BUTTERFLY_4_H(reg0, reg7, reg6, reg1, vec0, vec1, vec2, vec3);
1048 __lsx_vst(vec0, tmp_odd_buf, 6 * 16);
1049 __lsx_vst(vec1, tmp_odd_buf, 6 * 16 + 16);
1050 VP9_DOTP_CONST_PAIR(vec2, vec3, cospi_24_64, cospi_8_64, vec2, vec3);
1051 __lsx_vst(vec2, tmp_odd_buf, 2 * 16);
1052 __lsx_vst(vec3, tmp_odd_buf, 2 * 16 + 16);
1053
1054 /* Odd stage 2 */
1055 /* 8 loads */
1056 reg0 = __lsx_vld(tmp_buf, 3 * 64);
1057 reg1 = __lsx_vld(tmp_buf, 5 * 64);
1058 reg2 = __lsx_vld(tmp_buf, 11 * 64);
1059 reg3 = __lsx_vld(tmp_buf, 13 * 64);
1060 reg4 = __lsx_vld(tmp_buf, 19 * 64);
1061 reg5 = __lsx_vld(tmp_buf, 21 * 64);
1062 reg6 = __lsx_vld(tmp_buf, 27 * 64);
1063 reg7 = __lsx_vld(tmp_buf, 29 * 64);
1064
1065 __lsx_vst(zero, tmp_buf, 3 * 64);
1066 __lsx_vst(zero, tmp_buf, 5 * 64);
1067 __lsx_vst(zero, tmp_buf, 11 * 64);
1068 __lsx_vst(zero, tmp_buf, 13 * 64);
1069 __lsx_vst(zero, tmp_buf, 19 * 64);
1070 __lsx_vst(zero, tmp_buf, 21 * 64);
1071 __lsx_vst(zero, tmp_buf, 27 * 64);
1072 __lsx_vst(zero, tmp_buf, 29 * 64);
1073
1074 VP9_DOTP_CONST_PAIR(reg1, reg6, cospi_27_64, cospi_5_64, reg1, reg6);
1075 VP9_DOTP_CONST_PAIR(reg5, reg2, cospi_11_64, cospi_21_64, reg2, reg5);
1076 VP9_DOTP_CONST_PAIR(reg3, reg4, cospi_19_64, cospi_13_64, reg3, reg4);
1077 VP9_DOTP_CONST_PAIR(reg7, reg0, cospi_3_64, cospi_29_64, reg0, reg7);
1078
1079 /* 4 Stores */
1080 DUP4_ARG2(__lsx_vsub_h,reg1, reg2, reg6, reg5, reg0, reg3, reg7, reg4,
1081 vec0, vec1, vec2, vec3);
1082 VP9_DOTP_CONST_PAIR(vec1, vec0, cospi_12_64, cospi_20_64, loc0, loc1);
1083 VP9_DOTP_CONST_PAIR(vec3, vec2, -cospi_20_64, cospi_12_64, loc2, loc3);
1084 LSX_BUTTERFLY_4_H(loc2, loc3, loc1, loc0, vec0, vec1, vec3, vec2);
1085 __lsx_vst(vec0, tmp_odd_buf, 12 * 16);
1086 __lsx_vst(vec1, tmp_odd_buf, 12 * 16 + 3 * 16);
1087 VP9_DOTP_CONST_PAIR(vec3, vec2, -cospi_8_64, cospi_24_64, vec0, vec1);
1088 __lsx_vst(vec0, tmp_odd_buf, 10 * 16);
1089 __lsx_vst(vec1, tmp_odd_buf, 10 * 16 + 16);
1090
1091 /* 4 Stores */
1092 DUP4_ARG2(__lsx_vadd_h, reg0, reg3, reg1, reg2, reg5, reg6, reg4, reg7,
1093 vec0, vec1, vec2, vec3);
1094 LSX_BUTTERFLY_4_H(vec0, vec3, vec2, vec1, reg0, reg1, reg3, reg2);
1095 __lsx_vst(reg0, tmp_odd_buf, 13 * 16);
1096 __lsx_vst(reg1, tmp_odd_buf, 13 * 16 + 16);
1097 VP9_DOTP_CONST_PAIR(reg3, reg2, -cospi_8_64, cospi_24_64,
1098 reg0, reg1);
1099 __lsx_vst(reg0, tmp_odd_buf, 8 * 16);
1100 __lsx_vst(reg1, tmp_odd_buf, 8 * 16 + 16);
1101
1102 /* Odd stage 3 : Dependency on Odd stage 1 & Odd stage 2 */
1103 /* Load 8 & Store 8 */
1104 DUP4_ARG2(__lsx_vld, tmp_odd_buf, 0, tmp_odd_buf, 16,
1105 tmp_odd_buf, 32, tmp_odd_buf, 48, reg0, reg1, reg2, reg3);
1106 DUP4_ARG2(__lsx_vld, tmp_odd_buf, 8 * 16, tmp_odd_buf, 8 * 16 + 16,
1107 tmp_odd_buf, 8 * 16 + 32, tmp_odd_buf, 8 * 16 + 48,
1108 reg4, reg5, reg6, reg7);
1109
1110 DUP4_ARG2(__lsx_vadd_h, reg0, reg4, reg1, reg5, reg2, reg6, reg3, reg7,
1111 loc0, loc1, loc2, loc3);
1112 __lsx_vst(loc0, tmp_odd_buf, 0);
1113 __lsx_vst(loc1, tmp_odd_buf, 16);
1114 __lsx_vst(loc2, tmp_odd_buf, 32);
1115 __lsx_vst(loc3, tmp_odd_buf, 48);
1116 DUP2_ARG2(__lsx_vsub_h, reg0, reg4, reg1, reg5, vec0, vec1);
1117 VP9_DOTP_CONST_PAIR(vec1, vec0, cospi_16_64, cospi_16_64, loc0, loc1);
1118
1119 DUP2_ARG2(__lsx_vsub_h, reg2, reg6, reg3, reg7, vec0, vec1);
1120 VP9_DOTP_CONST_PAIR(vec1, vec0, cospi_16_64, cospi_16_64, loc2, loc3);
1121 __lsx_vst(loc0, tmp_odd_buf, 8 * 16);
1122 __lsx_vst(loc1, tmp_odd_buf, 8 * 16 + 16);
1123 __lsx_vst(loc2, tmp_odd_buf, 8 * 16 + 32);
1124 __lsx_vst(loc3, tmp_odd_buf, 8 * 16 + 48);
1125
1126 /* Load 8 & Store 8 */
1127 DUP4_ARG2(__lsx_vld, tmp_odd_buf, 4 * 16, tmp_odd_buf, 4 * 16 + 16,
1128 tmp_odd_buf, 4 * 16 + 32, tmp_odd_buf, 4 * 16 + 48,
1129 reg1, reg2, reg0, reg3);
1130 DUP4_ARG2(__lsx_vld, tmp_odd_buf, 12 * 16, tmp_odd_buf, 12 * 16 + 16,
1131 tmp_odd_buf, 12 * 16 + 32, tmp_odd_buf, 12 * 16 + 48,
1132 reg4, reg5, reg6, reg7);
1133
1134 DUP4_ARG2(__lsx_vadd_h, reg0, reg4, reg1, reg5, reg2, reg6, reg3, reg7,
1135 loc0, loc1, loc2, loc3);
1136 __lsx_vst(loc0, tmp_odd_buf, 4 * 16);
1137 __lsx_vst(loc1, tmp_odd_buf, 4 * 16 + 16);
1138 __lsx_vst(loc2, tmp_odd_buf, 4 * 16 + 32);
1139 __lsx_vst(loc3, tmp_odd_buf, 4 * 16 + 48);
1140
1141 DUP2_ARG2(__lsx_vsub_h, reg0, reg4, reg3, reg7, vec0, vec1);
1142 VP9_DOTP_CONST_PAIR(vec1, vec0, cospi_16_64, cospi_16_64, loc0, loc1);
1143
1144 DUP2_ARG2(__lsx_vsub_h, reg1, reg5, reg2, reg6, vec0, vec1);
1145 VP9_DOTP_CONST_PAIR(vec1, vec0, cospi_16_64, cospi_16_64, loc2, loc3);
1146 __lsx_vst(loc0, tmp_odd_buf, 12 * 16);
1147 __lsx_vst(loc1, tmp_odd_buf, 12 * 16 + 16);
1148 __lsx_vst(loc2, tmp_odd_buf, 12 * 16 + 32);
1149 __lsx_vst(loc3, tmp_odd_buf, 12 * 16 + 48);
1150 }
1151
vp9_idct8x32_column_butterfly_addblk(int16_t *tmp_eve_buf, int16_t *tmp_odd_buf, uint8_t *dst, int32_t dst_stride)1152 static void vp9_idct8x32_column_butterfly_addblk(int16_t *tmp_eve_buf,
1153 int16_t *tmp_odd_buf,
1154 uint8_t *dst,
1155 int32_t dst_stride)
1156 {
1157 __m128i vec0, vec1, vec2, vec3, loc0, loc1, loc2, loc3;
1158 __m128i m0, m1, m2, m3, m4, m5, m6, m7, n0, n1, n2, n3, n4, n5, n6, n7;
1159
1160 /* FINAL BUTTERFLY : Dependency on Even & Odd */
1161 vec0 = __lsx_vld(tmp_odd_buf, 0);
1162 vec1 = __lsx_vld(tmp_odd_buf, 9 * 16);
1163 vec2 = __lsx_vld(tmp_odd_buf, 14 * 16);
1164 vec3 = __lsx_vld(tmp_odd_buf, 6 * 16);
1165 loc0 = __lsx_vld(tmp_eve_buf, 0);
1166 loc1 = __lsx_vld(tmp_eve_buf, 8 * 16);
1167 loc2 = __lsx_vld(tmp_eve_buf, 4 * 16);
1168 loc3 = __lsx_vld(tmp_eve_buf, 12 * 16);
1169
1170 DUP4_ARG2(__lsx_vadd_h, loc0, vec3, loc1, vec2, loc2, vec1, loc3, vec0,
1171 m0, m4, m2, m6);
1172 DUP4_ARG2(__lsx_vsrari_h, m0, 6, m2, 6, m4, 6, m6, 6, m0, m2, m4, m6);
1173 VP9_ADDBLK_ST8x4_UB(dst, (4 * dst_stride), m0, m2, m4, m6);
1174
1175 DUP4_ARG2(__lsx_vsub_h, loc0, vec3, loc1, vec2, loc2, vec1, loc3, vec0,
1176 m6, m2, m4, m0);
1177 DUP4_ARG2(__lsx_vsrari_h, m0, 6, m2, 6, m4, 6, m6, 6, m0, m2, m4, m6);
1178 VP9_ADDBLK_ST8x4_UB((dst + 19 * dst_stride), (4 * dst_stride),
1179 m0, m2, m4, m6);
1180
1181 /* Load 8 & Store 8 */
1182 vec0 = __lsx_vld(tmp_odd_buf, 4 * 16);
1183 vec1 = __lsx_vld(tmp_odd_buf, 13 * 16);
1184 vec2 = __lsx_vld(tmp_odd_buf, 10 * 16);
1185 vec3 = __lsx_vld(tmp_odd_buf, 3 * 16);
1186 loc0 = __lsx_vld(tmp_eve_buf, 2 * 16);
1187 loc1 = __lsx_vld(tmp_eve_buf, 10 * 16);
1188 loc2 = __lsx_vld(tmp_eve_buf, 6 * 16);
1189 loc3 = __lsx_vld(tmp_eve_buf, 14 * 16);
1190
1191 DUP4_ARG2(__lsx_vadd_h, loc0, vec3, loc1, vec2, loc2, vec1, loc3, vec0,
1192 m1, m5, m3, m7);
1193 DUP4_ARG2(__lsx_vsrari_h, m1, 6, m3, 6, m5, 6, m7, 6, m1, m3, m5, m7);
1194 VP9_ADDBLK_ST8x4_UB((dst + 2 * dst_stride), (4 * dst_stride),
1195 m1, m3, m5, m7);
1196
1197 DUP4_ARG2(__lsx_vsub_h, loc0, vec3, loc1, vec2, loc2, vec1, loc3, vec0,
1198 m7, m3, m5, m1);
1199 DUP4_ARG2(__lsx_vsrari_h, m1, 6, m3, 6, m5, 6, m7, 6, m1, m3, m5, m7);
1200 VP9_ADDBLK_ST8x4_UB((dst + 17 * dst_stride), (4 * dst_stride),
1201 m1, m3, m5, m7);
1202
1203 /* Load 8 & Store 8 */
1204 vec0 = __lsx_vld(tmp_odd_buf, 2 * 16);
1205 vec1 = __lsx_vld(tmp_odd_buf, 11 * 16);
1206 vec2 = __lsx_vld(tmp_odd_buf, 12 * 16);
1207 vec3 = __lsx_vld(tmp_odd_buf, 7 * 16);
1208 loc0 = __lsx_vld(tmp_eve_buf, 1 * 16);
1209 loc1 = __lsx_vld(tmp_eve_buf, 9 * 16);
1210 loc2 = __lsx_vld(tmp_eve_buf, 5 * 16);
1211 loc3 = __lsx_vld(tmp_eve_buf, 13 * 16);
1212
1213 DUP4_ARG2(__lsx_vadd_h, loc0, vec3, loc1, vec2, loc2, vec1, loc3, vec0,
1214 n0, n4, n2, n6);
1215 DUP4_ARG2(__lsx_vsrari_h, n0, 6, n2, 6, n4, 6, n6, 6, n0, n2, n4, n6);
1216 VP9_ADDBLK_ST8x4_UB((dst + 1 * dst_stride), (4 * dst_stride),
1217 n0, n2, n4, n6);
1218 DUP4_ARG2(__lsx_vsub_h, loc0, vec3, loc1, vec2, loc2, vec1, loc3, vec0,
1219 n6, n2, n4, n0);
1220 DUP4_ARG2(__lsx_vsrari_h, n0, 6, n2, 6, n4, 6, n6, 6, n0, n2, n4, n6);
1221 VP9_ADDBLK_ST8x4_UB((dst + 18 * dst_stride), (4 * dst_stride),
1222 n0, n2, n4, n6);
1223
1224 /* Load 8 & Store 8 */
1225 vec0 = __lsx_vld(tmp_odd_buf, 5 * 16);
1226 vec1 = __lsx_vld(tmp_odd_buf, 15 * 16);
1227 vec2 = __lsx_vld(tmp_odd_buf, 8 * 16);
1228 vec3 = __lsx_vld(tmp_odd_buf, 1 * 16);
1229 loc0 = __lsx_vld(tmp_eve_buf, 3 * 16);
1230 loc1 = __lsx_vld(tmp_eve_buf, 11 * 16);
1231 loc2 = __lsx_vld(tmp_eve_buf, 7 * 16);
1232 loc3 = __lsx_vld(tmp_eve_buf, 15 * 16);
1233
1234 DUP4_ARG2(__lsx_vadd_h, loc0, vec3, loc1, vec2, loc2, vec1, loc3, vec0,
1235 n1, n5, n3, n7);
1236 DUP4_ARG2(__lsx_vsrari_h, n1, 6, n3, 6, n5, 6, n7, 6, n1, n3, n5, n7);
1237 VP9_ADDBLK_ST8x4_UB((dst + 3 * dst_stride), (4 * dst_stride),
1238 n1, n3, n5, n7);
1239 DUP4_ARG2(__lsx_vsub_h, loc0, vec3, loc1, vec2, loc2, vec1, loc3, vec0,
1240 n7, n3, n5, n1);
1241 DUP4_ARG2(__lsx_vsrari_h, n1, 6, n3, 6, n5, 6, n7, 6, n1, n3, n5, n7);
1242 VP9_ADDBLK_ST8x4_UB((dst + 16 * dst_stride), (4 * dst_stride),
1243 n1, n3, n5, n7);
1244 }
1245
vp9_idct8x32_1d_columns_addblk_lsx(int16_t *input, uint8_t *dst, int32_t dst_stride)1246 static void vp9_idct8x32_1d_columns_addblk_lsx(int16_t *input, uint8_t *dst,
1247 int32_t dst_stride)
1248 {
1249 int16_t tmp_odd_buf[16 * 8] ALLOC_ALIGNED(16);
1250 int16_t tmp_eve_buf[16 * 8] ALLOC_ALIGNED(16);
1251
1252 vp9_idct8x32_column_even_process_store(input, &tmp_eve_buf[0]);
1253 vp9_idct8x32_column_odd_process_store(input, &tmp_odd_buf[0]);
1254 vp9_idct8x32_column_butterfly_addblk(&tmp_eve_buf[0], &tmp_odd_buf[0],
1255 dst, dst_stride);
1256 }
1257
vp9_idct8x32_1d_columns_lsx(int16_t *input, int16_t *output, int16_t *tmp_buf)1258 static void vp9_idct8x32_1d_columns_lsx(int16_t *input, int16_t *output,
1259 int16_t *tmp_buf)
1260 {
1261 int16_t tmp_odd_buf[16 * 8] ALLOC_ALIGNED(16);
1262 int16_t tmp_eve_buf[16 * 8] ALLOC_ALIGNED(16);
1263
1264 vp9_idct8x32_column_even_process_store(input, &tmp_eve_buf[0]);
1265 vp9_idct8x32_column_odd_process_store(input, &tmp_odd_buf[0]);
1266 vp9_idct_butterfly_transpose_store(tmp_buf, &tmp_eve_buf[0],
1267 &tmp_odd_buf[0], output);
1268 }
1269
vp9_idct32x32_1_add_lsx(int16_t *input, uint8_t *dst, int32_t dst_stride)1270 static void vp9_idct32x32_1_add_lsx(int16_t *input, uint8_t *dst,
1271 int32_t dst_stride)
1272 {
1273 int32_t i;
1274 int16_t out;
1275 uint8_t *dst_tmp = dst + dst_stride;
1276 __m128i zero = __lsx_vldi(0);
1277 __m128i dst0, dst1, dst2, dst3, tmp0, tmp1, tmp2, tmp3;
1278 __m128i res0, res1, res2, res3, res4, res5, res6, res7, vec;
1279
1280 out = ROUND_POWER_OF_TWO((input[0] * cospi_16_64), VP9_DCT_CONST_BITS);
1281 out = ROUND_POWER_OF_TWO((out * cospi_16_64), VP9_DCT_CONST_BITS);
1282 out = ROUND_POWER_OF_TWO(out, 6);
1283 input[0] = 0;
1284
1285 vec = __lsx_vreplgr2vr_h(out);
1286
1287 for (i = 16; i--;) {
1288 DUP2_ARG2(__lsx_vld, dst, 0, dst, 16, dst0, dst1);
1289 DUP2_ARG2(__lsx_vld, dst_tmp, 0, dst_tmp, 16, dst2, dst3);
1290
1291 DUP4_ARG2(__lsx_vilvl_b, zero, dst0, zero, dst1, zero, dst2, zero, dst3,
1292 res0, res1, res2, res3);
1293 DUP4_ARG2(__lsx_vilvh_b, zero, dst0, zero, dst1, zero, dst2, zero, dst3,
1294 res4, res5, res6, res7);
1295 DUP4_ARG2(__lsx_vadd_h, res0, vec, res1, vec, res2, vec, res3, vec,
1296 res0, res1, res2, res3);
1297 DUP4_ARG2(__lsx_vadd_h, res4, vec, res5, vec, res6, vec, res7, vec,
1298 res4, res5, res6, res7);
1299 DUP4_ARG1(__lsx_vclip255_h, res0, res1, res2, res3, res0, res1, res2, res3);
1300 DUP4_ARG1(__lsx_vclip255_h, res4, res5, res6, res7, res4, res5, res6, res7);
1301 DUP4_ARG2(__lsx_vpickev_b, res4, res0, res5, res1, res6, res2, res7, res3,
1302 tmp0, tmp1, tmp2, tmp3);
1303
1304 __lsx_vst(tmp0, dst, 0);
1305 __lsx_vst(tmp1, dst, 16);
1306 __lsx_vst(tmp2, dst_tmp, 0);
1307 __lsx_vst(tmp3, dst_tmp, 16);
1308 dst = dst_tmp + dst_stride;
1309 dst_tmp = dst + dst_stride;
1310 }
1311 }
1312
vp9_idct32x32_34_colcol_addblk_lsx(int16_t *input, uint8_t *dst, int32_t dst_stride)1313 static void vp9_idct32x32_34_colcol_addblk_lsx(int16_t *input, uint8_t *dst,
1314 int32_t dst_stride)
1315 {
1316 int32_t i;
1317 int16_t out_arr[32 * 32] ALLOC_ALIGNED(16);
1318 int16_t *out_ptr = out_arr;
1319 int16_t tmp_buf[8 * 32] ALLOC_ALIGNED(16);
1320 __m128i zero = __lsx_vldi(0);
1321
1322 for (i = 16; i--;) {
1323 __lsx_vst(zero, out_ptr, 0);
1324 __lsx_vst(zero, out_ptr, 16);
1325 __lsx_vst(zero, out_ptr, 32);
1326 __lsx_vst(zero, out_ptr, 48);
1327 __lsx_vst(zero, out_ptr, 64);
1328 __lsx_vst(zero, out_ptr, 80);
1329 __lsx_vst(zero, out_ptr, 96);
1330 __lsx_vst(zero, out_ptr, 112);
1331 out_ptr += 64;
1332 }
1333
1334 out_ptr = out_arr;
1335
1336 /* process 8*32 block */
1337 vp9_idct8x32_1d_columns_lsx(input, out_ptr, &tmp_buf[0]);
1338
1339 /* transform columns */
1340 for (i = 0; i < 4; i++) {
1341 /* process 8*32 block */
1342 vp9_idct8x32_1d_columns_addblk_lsx((out_ptr + (i << 3)),
1343 (dst + (i << 3)), dst_stride);
1344 }
1345 }
1346
vp9_idct32x32_colcol_addblk_lsx(int16_t *input, uint8_t *dst, int32_t dst_stride)1347 static void vp9_idct32x32_colcol_addblk_lsx(int16_t *input, uint8_t *dst,
1348 int32_t dst_stride)
1349 {
1350 int32_t i;
1351 int16_t out_arr[32 * 32] ALLOC_ALIGNED(16);
1352 int16_t *out_ptr = out_arr;
1353 int16_t tmp_buf[8 * 32] ALLOC_ALIGNED(16);
1354
1355 /* transform rows */
1356 for (i = 0; i < 4; i++) {
1357 /* process 8*32 block */
1358 vp9_idct8x32_1d_columns_lsx((input + (i << 3)), (out_ptr + (i << 8)),
1359 &tmp_buf[0]);
1360 }
1361
1362 /* transform columns */
1363 for (i = 0; i < 4; i++) {
1364 /* process 8*32 block */
1365 vp9_idct8x32_1d_columns_addblk_lsx((out_ptr + (i << 3)),
1366 (dst + (i << 3)), dst_stride);
1367 }
1368 }
1369
ff_idct_idct_8x8_add_lsx(uint8_t *dst, ptrdiff_t stride, int16_t *block, int eob)1370 void ff_idct_idct_8x8_add_lsx(uint8_t *dst, ptrdiff_t stride,
1371 int16_t *block, int eob)
1372 {
1373 if (eob == 1) {
1374 vp9_idct8x8_1_add_lsx(block, dst, stride);
1375 }
1376 else if (eob <= 12) {
1377 vp9_idct8x8_12_colcol_addblk_lsx(block, dst, stride);
1378 }
1379 else {
1380 vp9_idct8x8_colcol_addblk_lsx(block, dst, stride);
1381 }
1382 }
1383
ff_idct_idct_16x16_add_lsx(uint8_t *dst, ptrdiff_t stride, int16_t *block, int eob)1384 void ff_idct_idct_16x16_add_lsx(uint8_t *dst, ptrdiff_t stride,
1385 int16_t *block, int eob)
1386 {
1387 if (eob == 1) {
1388 /* DC only DCT coefficient. */
1389 vp9_idct16x16_1_add_lsx(block, dst, stride);
1390 }
1391 else if (eob <= 10) {
1392 vp9_idct16x16_10_colcol_addblk_lsx(block, dst, stride);
1393 }
1394 else {
1395 vp9_idct16x16_colcol_addblk_lsx(block, dst, stride);
1396 }
1397 }
1398
ff_idct_idct_32x32_add_lsx(uint8_t *dst, ptrdiff_t stride, int16_t *block, int eob)1399 void ff_idct_idct_32x32_add_lsx(uint8_t *dst, ptrdiff_t stride,
1400 int16_t *block, int eob)
1401 {
1402 if (eob == 1) {
1403 vp9_idct32x32_1_add_lsx(block, dst, stride);
1404 }
1405 else if (eob <= 34) {
1406 vp9_idct32x32_34_colcol_addblk_lsx(block, dst, stride);
1407 }
1408 else {
1409 vp9_idct32x32_colcol_addblk_lsx(block, dst, stride);
1410 }
1411 }
1412