• 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 #include "vpx_dsp/mips/inv_txfm_msa.h"
12 
vpx_idct16_1d_rows_msa(const int16_t * input,int16_t * output)13 void vpx_idct16_1d_rows_msa(const int16_t *input, int16_t *output) {
14   v8i16 loc0, loc1, loc2, loc3;
15   v8i16 reg0, reg2, reg4, reg6, reg8, reg10, reg12, reg14;
16   v8i16 reg3, reg13, reg11, reg5, reg7, reg9, reg1, reg15;
17   v8i16 tmp5, tmp6, tmp7;
18 
19   LD_SH8(input, 16, reg0, reg1, reg2, reg3, reg4, reg5, reg6, reg7);
20   input += 8;
21   LD_SH8(input, 16, reg8, reg9, reg10, reg11, reg12, reg13, reg14, reg15);
22 
23   TRANSPOSE8x8_SH_SH(reg0, reg1, reg2, reg3, reg4, reg5, reg6, reg7,
24                      reg0, reg1, reg2, reg3, reg4, reg5, reg6, reg7);
25   TRANSPOSE8x8_SH_SH(reg8, reg9, reg10, reg11, reg12, reg13, reg14, reg15,
26                      reg8, reg9, reg10, reg11, reg12, reg13, reg14, reg15);
27   DOTP_CONST_PAIR(reg2, reg14, cospi_28_64, cospi_4_64, reg2, reg14);
28   DOTP_CONST_PAIR(reg10, reg6, cospi_12_64, cospi_20_64, reg10, reg6);
29   BUTTERFLY_4(reg2, reg14, reg6, reg10, loc0, loc1, reg14, reg2);
30   DOTP_CONST_PAIR(reg14, reg2, cospi_16_64, cospi_16_64, loc2, loc3);
31   DOTP_CONST_PAIR(reg0, reg8, cospi_16_64, cospi_16_64, reg0, reg8);
32   DOTP_CONST_PAIR(reg4, reg12, cospi_24_64, cospi_8_64, reg4, reg12);
33   BUTTERFLY_4(reg8, reg0, reg4, reg12, reg2, reg6, reg10, reg14);
34   SUB4(reg2, loc1, reg14, loc0, reg6, loc3, reg10, loc2, reg0, reg12, reg4,
35        reg8);
36   ADD4(reg2, loc1, reg14, loc0, reg6, loc3, reg10, loc2, reg2, reg14, reg6,
37        reg10);
38 
39   /* stage 2 */
40   DOTP_CONST_PAIR(reg1, reg15, cospi_30_64, cospi_2_64, reg1, reg15);
41   DOTP_CONST_PAIR(reg9, reg7, cospi_14_64, cospi_18_64, loc2, loc3);
42 
43   reg9 = reg1 - loc2;
44   reg1 = reg1 + loc2;
45   reg7 = reg15 - loc3;
46   reg15 = reg15 + loc3;
47 
48   DOTP_CONST_PAIR(reg5, reg11, cospi_22_64, cospi_10_64, reg5, reg11);
49   DOTP_CONST_PAIR(reg13, reg3, cospi_6_64, cospi_26_64, loc0, loc1);
50   BUTTERFLY_4(loc0, loc1, reg11, reg5, reg13, reg3, reg11, reg5);
51 
52   loc1 = reg15 + reg3;
53   reg3 = reg15 - reg3;
54   loc2 = reg2 + loc1;
55   reg15 = reg2 - loc1;
56 
57   loc1 = reg1 + reg13;
58   reg13 = reg1 - reg13;
59   loc0 = reg0 + loc1;
60   loc1 = reg0 - loc1;
61   tmp6 = loc0;
62   tmp7 = loc1;
63   reg0 = loc2;
64 
65   DOTP_CONST_PAIR(reg7, reg9, cospi_24_64, cospi_8_64, reg7, reg9);
66   DOTP_CONST_PAIR((-reg5), (-reg11), cospi_8_64, cospi_24_64, reg5, reg11);
67 
68   loc0 = reg9 + reg5;
69   reg5 = reg9 - reg5;
70   reg2 = reg6 + loc0;
71   reg1 = reg6 - loc0;
72 
73   loc0 = reg7 + reg11;
74   reg11 = reg7 - reg11;
75   loc1 = reg4 + loc0;
76   loc2 = reg4 - loc0;
77   tmp5 = loc1;
78 
79   DOTP_CONST_PAIR(reg5, reg11, cospi_16_64, cospi_16_64, reg5, reg11);
80   BUTTERFLY_4(reg8, reg10, reg11, reg5, loc0, reg4, reg9, loc1);
81 
82   reg10 = loc0;
83   reg11 = loc1;
84 
85   DOTP_CONST_PAIR(reg3, reg13, cospi_16_64, cospi_16_64, reg3, reg13);
86   BUTTERFLY_4(reg12, reg14, reg13, reg3, reg8, reg6, reg7, reg5);
87 
88   reg13 = loc2;
89 
90   /* Transpose and store the output */
91   reg12 = tmp5;
92   reg14 = tmp6;
93   reg3 = tmp7;
94 
95   /* transpose block */
96   TRANSPOSE8x8_SH_SH(reg0, reg2, reg4, reg6, reg8, reg10, reg12, reg14,
97                      reg0, reg2, reg4, reg6, reg8, reg10, reg12, reg14);
98   ST_SH8(reg0, reg2, reg4, reg6, reg8, reg10, reg12, reg14, output, 16);
99 
100   /* transpose block */
101   TRANSPOSE8x8_SH_SH(reg3, reg13, reg11, reg5, reg7, reg9, reg1, reg15,
102                      reg3, reg13, reg11, reg5, reg7, reg9, reg1, reg15);
103   ST_SH8(reg3, reg13, reg11, reg5, reg7, reg9, reg1, reg15, (output + 8), 16);
104 }
105 
vpx_idct16_1d_columns_addblk_msa(int16_t * input,uint8_t * dst,int32_t dst_stride)106 void vpx_idct16_1d_columns_addblk_msa(int16_t *input, uint8_t *dst,
107                                       int32_t dst_stride) {
108   v8i16 loc0, loc1, loc2, loc3;
109   v8i16 reg0, reg2, reg4, reg6, reg8, reg10, reg12, reg14;
110   v8i16 reg3, reg13, reg11, reg5, reg7, reg9, reg1, reg15;
111   v8i16 tmp5, tmp6, tmp7;
112 
113   /* load up 8x8 */
114   LD_SH8(input, 16, reg0, reg1, reg2, reg3, reg4, reg5, reg6, reg7);
115   input += 8 * 16;
116   /* load bottom 8x8 */
117   LD_SH8(input, 16, reg8, reg9, reg10, reg11, reg12, reg13, reg14, reg15);
118 
119   DOTP_CONST_PAIR(reg2, reg14, cospi_28_64, cospi_4_64, reg2, reg14);
120   DOTP_CONST_PAIR(reg10, reg6, cospi_12_64, cospi_20_64, reg10, reg6);
121   BUTTERFLY_4(reg2, reg14, reg6, reg10, loc0, loc1, reg14, reg2);
122   DOTP_CONST_PAIR(reg14, reg2, cospi_16_64, cospi_16_64, loc2, loc3);
123   DOTP_CONST_PAIR(reg0, reg8, cospi_16_64, cospi_16_64, reg0, reg8);
124   DOTP_CONST_PAIR(reg4, reg12, cospi_24_64, cospi_8_64, reg4, reg12);
125   BUTTERFLY_4(reg8, reg0, reg4, reg12, reg2, reg6, reg10, reg14);
126 
127   reg0 = reg2 - loc1;
128   reg2 = reg2 + loc1;
129   reg12 = reg14 - loc0;
130   reg14 = reg14 + loc0;
131   reg4 = reg6 - loc3;
132   reg6 = reg6 + loc3;
133   reg8 = reg10 - loc2;
134   reg10 = reg10 + loc2;
135 
136   /* stage 2 */
137   DOTP_CONST_PAIR(reg1, reg15, cospi_30_64, cospi_2_64, reg1, reg15);
138   DOTP_CONST_PAIR(reg9, reg7, cospi_14_64, cospi_18_64, loc2, loc3);
139 
140   reg9 = reg1 - loc2;
141   reg1 = reg1 + loc2;
142   reg7 = reg15 - loc3;
143   reg15 = reg15 + loc3;
144 
145   DOTP_CONST_PAIR(reg5, reg11, cospi_22_64, cospi_10_64, reg5, reg11);
146   DOTP_CONST_PAIR(reg13, reg3, cospi_6_64, cospi_26_64, loc0, loc1);
147   BUTTERFLY_4(loc0, loc1, reg11, reg5, reg13, reg3, reg11, reg5);
148 
149   loc1 = reg15 + reg3;
150   reg3 = reg15 - reg3;
151   loc2 = reg2 + loc1;
152   reg15 = reg2 - loc1;
153 
154   loc1 = reg1 + reg13;
155   reg13 = reg1 - reg13;
156   loc0 = reg0 + loc1;
157   loc1 = reg0 - loc1;
158   tmp6 = loc0;
159   tmp7 = loc1;
160   reg0 = loc2;
161 
162   DOTP_CONST_PAIR(reg7, reg9, cospi_24_64, cospi_8_64, reg7, reg9);
163   DOTP_CONST_PAIR((-reg5), (-reg11), cospi_8_64, cospi_24_64, reg5, reg11);
164 
165   loc0 = reg9 + reg5;
166   reg5 = reg9 - reg5;
167   reg2 = reg6 + loc0;
168   reg1 = reg6 - loc0;
169 
170   loc0 = reg7 + reg11;
171   reg11 = reg7 - reg11;
172   loc1 = reg4 + loc0;
173   loc2 = reg4 - loc0;
174   tmp5 = loc1;
175 
176   DOTP_CONST_PAIR(reg5, reg11, cospi_16_64, cospi_16_64, reg5, reg11);
177   BUTTERFLY_4(reg8, reg10, reg11, reg5, loc0, reg4, reg9, loc1);
178 
179   reg10 = loc0;
180   reg11 = loc1;
181 
182   DOTP_CONST_PAIR(reg3, reg13, cospi_16_64, cospi_16_64, reg3, reg13);
183   BUTTERFLY_4(reg12, reg14, reg13, reg3, reg8, reg6, reg7, reg5);
184   reg13 = loc2;
185 
186   /* Transpose and store the output */
187   reg12 = tmp5;
188   reg14 = tmp6;
189   reg3 = tmp7;
190 
191   SRARI_H4_SH(reg0, reg2, reg4, reg6, 6);
192   VP9_ADDBLK_ST8x4_UB(dst, dst_stride, reg0, reg2, reg4, reg6);
193   dst += (4 * dst_stride);
194   SRARI_H4_SH(reg8, reg10, reg12, reg14, 6);
195   VP9_ADDBLK_ST8x4_UB(dst, dst_stride, reg8, reg10, reg12, reg14);
196   dst += (4 * dst_stride);
197   SRARI_H4_SH(reg3, reg13, reg11, reg5, 6);
198   VP9_ADDBLK_ST8x4_UB(dst, dst_stride, reg3, reg13, reg11, reg5);
199   dst += (4 * dst_stride);
200   SRARI_H4_SH(reg7, reg9, reg1, reg15, 6);
201   VP9_ADDBLK_ST8x4_UB(dst, dst_stride, reg7, reg9, reg1, reg15);
202 }
203 
vpx_idct16x16_256_add_msa(const int16_t * input,uint8_t * dst,int32_t dst_stride)204 void vpx_idct16x16_256_add_msa(const int16_t *input, uint8_t *dst,
205                                int32_t dst_stride) {
206   int32_t i;
207   DECLARE_ALIGNED(32, int16_t, out_arr[16 * 16]);
208   int16_t *out = out_arr;
209 
210   /* transform rows */
211   for (i = 0; i < 2; ++i) {
212     /* process 16 * 8 block */
213     vpx_idct16_1d_rows_msa((input + (i << 7)), (out + (i << 7)));
214   }
215 
216   /* transform columns */
217   for (i = 0; i < 2; ++i) {
218     /* process 8 * 16 block */
219     vpx_idct16_1d_columns_addblk_msa((out + (i << 3)), (dst + (i << 3)),
220                                      dst_stride);
221   }
222 }
223 
vpx_idct16x16_10_add_msa(const int16_t * input,uint8_t * dst,int32_t dst_stride)224 void vpx_idct16x16_10_add_msa(const int16_t *input, uint8_t *dst,
225                               int32_t dst_stride) {
226   uint8_t i;
227   DECLARE_ALIGNED(32, int16_t, out_arr[16 * 16]);
228   int16_t *out = out_arr;
229 
230   /* process 16 * 8 block */
231   vpx_idct16_1d_rows_msa(input, out);
232 
233   /* short case just considers top 4 rows as valid output */
234   out += 4 * 16;
235   for (i = 12; i--;) {
236     __asm__ __volatile__ (
237         "sw     $zero,   0(%[out])     \n\t"
238         "sw     $zero,   4(%[out])     \n\t"
239         "sw     $zero,   8(%[out])     \n\t"
240         "sw     $zero,  12(%[out])     \n\t"
241         "sw     $zero,  16(%[out])     \n\t"
242         "sw     $zero,  20(%[out])     \n\t"
243         "sw     $zero,  24(%[out])     \n\t"
244         "sw     $zero,  28(%[out])     \n\t"
245 
246         :
247         : [out] "r" (out)
248     );
249 
250     out += 16;
251   }
252 
253   out = out_arr;
254 
255   /* transform columns */
256   for (i = 0; i < 2; ++i) {
257     /* process 8 * 16 block */
258     vpx_idct16_1d_columns_addblk_msa((out + (i << 3)), (dst + (i << 3)),
259                                      dst_stride);
260   }
261 }
262 
vpx_idct16x16_1_add_msa(const int16_t * input,uint8_t * dst,int32_t dst_stride)263 void vpx_idct16x16_1_add_msa(const int16_t *input, uint8_t *dst,
264                              int32_t dst_stride) {
265   uint8_t i;
266   int16_t out;
267   v8i16 vec, res0, res1, res2, res3, res4, res5, res6, res7;
268   v16u8 dst0, dst1, dst2, dst3, tmp0, tmp1, tmp2, tmp3;
269 
270   out = ROUND_POWER_OF_TWO((input[0] * cospi_16_64), DCT_CONST_BITS);
271   out = ROUND_POWER_OF_TWO((out * cospi_16_64), DCT_CONST_BITS);
272   out = ROUND_POWER_OF_TWO(out, 6);
273 
274   vec = __msa_fill_h(out);
275 
276   for (i = 4; i--;) {
277     LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
278     UNPCK_UB_SH(dst0, res0, res4);
279     UNPCK_UB_SH(dst1, res1, res5);
280     UNPCK_UB_SH(dst2, res2, res6);
281     UNPCK_UB_SH(dst3, res3, res7);
282     ADD4(res0, vec, res1, vec, res2, vec, res3, vec, res0, res1, res2, res3);
283     ADD4(res4, vec, res5, vec, res6, vec, res7, vec, res4, res5, res6, res7);
284     CLIP_SH4_0_255(res0, res1, res2, res3);
285     CLIP_SH4_0_255(res4, res5, res6, res7);
286     PCKEV_B4_UB(res4, res0, res5, res1, res6, res2, res7, res3,
287                 tmp0, tmp1, tmp2, tmp3);
288     ST_UB4(tmp0, tmp1, tmp2, tmp3, dst, dst_stride);
289     dst += (4 * dst_stride);
290   }
291 }
292 
vpx_iadst16_1d_rows_msa(const int16_t * input,int16_t * output)293 void vpx_iadst16_1d_rows_msa(const int16_t *input, int16_t *output) {
294   v8i16 r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15;
295   v8i16 l0, l1, l2, l3, l4, l5, l6, l7, l8, l9, l10, l11, l12, l13, l14, l15;
296 
297   /* load input data */
298   LD_SH16(input, 8,
299           l0, l8, l1, l9, l2, l10, l3, l11, l4, l12, l5, l13, l6, l14, l7, l15);
300   TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7,
301                      l0, l1, l2, l3, l4, l5, l6, l7);
302   TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15,
303                      l8, l9, l10, l11, l12, l13, l14, l15);
304 
305   /* ADST in horizontal */
306   VP9_IADST8x16_1D(l0, l1, l2, l3, l4, l5, l6, l7,
307                    l8, l9, l10, l11, l12, l13, l14, l15,
308                    r0, r1, r2, r3, r4, r5, r6, r7,
309                    r8, r9, r10, r11, r12, r13, r14, r15);
310 
311   l1 = -r8;
312   l3 = -r4;
313   l13 = -r13;
314   l15 = -r1;
315 
316   TRANSPOSE8x8_SH_SH(r0, l1, r12, l3, r6, r14, r10, r2,
317                      l0, l1, l2, l3, l4, l5, l6, l7);
318   ST_SH8(l0, l1, l2, l3, l4, l5, l6, l7, output, 16);
319   TRANSPOSE8x8_SH_SH(r3, r11, r15, r7, r5, l13, r9, l15,
320                      l8, l9, l10, l11, l12, l13, l14, l15);
321   ST_SH8(l8, l9, l10, l11, l12, l13, l14, l15, (output + 8), 16);
322 }
323 
vpx_iadst16_1d_columns_addblk_msa(int16_t * input,uint8_t * dst,int32_t dst_stride)324 void vpx_iadst16_1d_columns_addblk_msa(int16_t *input, uint8_t *dst,
325                                        int32_t dst_stride) {
326   v8i16 v0, v2, v4, v6, k0, k1, k2, k3;
327   v8i16 r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15;
328   v8i16 out0, out1, out2, out3, out4, out5, out6, out7;
329   v8i16 out8, out9, out10, out11, out12, out13, out14, out15;
330   v8i16 g0, g1, g2, g3, g4, g5, g6, g7, g8, g9, g10, g11, g12, g13, g14, g15;
331   v8i16 h0, h1, h2, h3, h4, h5, h6, h7, h8, h9, h10, h11;
332   v8i16 res0, res1, res2, res3, res4, res5, res6, res7;
333   v8i16 res8, res9, res10, res11, res12, res13, res14, res15;
334   v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
335   v16u8 dst8, dst9, dst10, dst11, dst12, dst13, dst14, dst15;
336   v16i8 zero = { 0 };
337 
338   r0 = LD_SH(input + 0 * 16);
339   r3 = LD_SH(input + 3 * 16);
340   r4 = LD_SH(input + 4 * 16);
341   r7 = LD_SH(input + 7 * 16);
342   r8 = LD_SH(input + 8 * 16);
343   r11 = LD_SH(input + 11 * 16);
344   r12 = LD_SH(input + 12 * 16);
345   r15 = LD_SH(input + 15 * 16);
346 
347   /* stage 1 */
348   k0 = VP9_SET_COSPI_PAIR(cospi_1_64, cospi_31_64);
349   k1 = VP9_SET_COSPI_PAIR(cospi_31_64, -cospi_1_64);
350   k2 = VP9_SET_COSPI_PAIR(cospi_17_64, cospi_15_64);
351   k3 = VP9_SET_COSPI_PAIR(cospi_15_64, -cospi_17_64);
352   MADD_BF(r15, r0, r7, r8, k0, k1, k2, k3, g0, g1, g2, g3);
353   k0 = VP9_SET_COSPI_PAIR(cospi_9_64, cospi_23_64);
354   k1 = VP9_SET_COSPI_PAIR(cospi_23_64, -cospi_9_64);
355   k2 = VP9_SET_COSPI_PAIR(cospi_25_64, cospi_7_64);
356   k3 = VP9_SET_COSPI_PAIR(cospi_7_64, -cospi_25_64);
357   MADD_BF(r11, r4, r3, r12, k0, k1, k2, k3, g8, g9, g10, g11);
358   BUTTERFLY_4(g0, g2, g10, g8, h8, h9, v2, v0);
359   k0 = VP9_SET_COSPI_PAIR(cospi_4_64, cospi_28_64);
360   k1 = VP9_SET_COSPI_PAIR(cospi_28_64, -cospi_4_64);
361   k2 = VP9_SET_COSPI_PAIR(-cospi_28_64, cospi_4_64);
362   MADD_BF(g1, g3, g9, g11, k0, k1, k2, k0, h0, h1, h2, h3);
363 
364   r1 = LD_SH(input + 1 * 16);
365   r2 = LD_SH(input + 2 * 16);
366   r5 = LD_SH(input + 5 * 16);
367   r6 = LD_SH(input + 6 * 16);
368   r9 = LD_SH(input + 9 * 16);
369   r10 = LD_SH(input + 10 * 16);
370   r13 = LD_SH(input + 13 * 16);
371   r14 = LD_SH(input + 14 * 16);
372 
373   k0 = VP9_SET_COSPI_PAIR(cospi_5_64, cospi_27_64);
374   k1 = VP9_SET_COSPI_PAIR(cospi_27_64, -cospi_5_64);
375   k2 = VP9_SET_COSPI_PAIR(cospi_21_64, cospi_11_64);
376   k3 = VP9_SET_COSPI_PAIR(cospi_11_64, -cospi_21_64);
377   MADD_BF(r13, r2, r5, r10, k0, k1, k2, k3, g4, g5, g6, g7);
378   k0 = VP9_SET_COSPI_PAIR(cospi_13_64, cospi_19_64);
379   k1 = VP9_SET_COSPI_PAIR(cospi_19_64, -cospi_13_64);
380   k2 = VP9_SET_COSPI_PAIR(cospi_29_64, cospi_3_64);
381   k3 = VP9_SET_COSPI_PAIR(cospi_3_64, -cospi_29_64);
382   MADD_BF(r9, r6, r1, r14, k0, k1, k2, k3, g12, g13, g14, g15);
383   BUTTERFLY_4(g4, g6, g14, g12, h10, h11, v6, v4);
384   BUTTERFLY_4(h8, h9, h11, h10, out0, out1, h11, h10);
385   out1 = -out1;
386   SRARI_H2_SH(out0, out1, 6);
387   dst0 = LD_UB(dst + 0 * dst_stride);
388   dst1 = LD_UB(dst + 15 * dst_stride);
389   ILVR_B2_SH(zero, dst0, zero, dst1, res0, res1);
390   ADD2(res0, out0, res1, out1, res0, res1);
391   CLIP_SH2_0_255(res0, res1);
392   PCKEV_B2_SH(res0, res0, res1, res1, res0, res1);
393   ST8x1_UB(res0, dst);
394   ST8x1_UB(res1, dst + 15 * dst_stride);
395 
396   k0 = VP9_SET_COSPI_PAIR(cospi_12_64, cospi_20_64);
397   k1 = VP9_SET_COSPI_PAIR(-cospi_20_64, cospi_12_64);
398   k2 = VP9_SET_COSPI_PAIR(cospi_20_64, -cospi_12_64);
399   MADD_BF(g7, g5, g15, g13, k0, k1, k2, k0, h4, h5, h6, h7);
400   BUTTERFLY_4(h0, h2, h6, h4, out8, out9, out11, out10);
401   out8 = -out8;
402 
403   SRARI_H2_SH(out8, out9, 6);
404   dst8 = LD_UB(dst + 1 * dst_stride);
405   dst9 = LD_UB(dst + 14 * dst_stride);
406   ILVR_B2_SH(zero, dst8, zero, dst9, res8, res9);
407   ADD2(res8, out8, res9, out9, res8, res9);
408   CLIP_SH2_0_255(res8, res9);
409   PCKEV_B2_SH(res8, res8, res9, res9, res8, res9);
410   ST8x1_UB(res8, dst + dst_stride);
411   ST8x1_UB(res9, dst + 14 * dst_stride);
412 
413   k0 = VP9_SET_COSPI_PAIR(cospi_8_64, cospi_24_64);
414   k1 = VP9_SET_COSPI_PAIR(cospi_24_64, -cospi_8_64);
415   k2 = VP9_SET_COSPI_PAIR(-cospi_24_64, cospi_8_64);
416   MADD_BF(v0, v2, v4, v6, k0, k1, k2, k0, out4, out6, out5, out7);
417   out4 = -out4;
418   SRARI_H2_SH(out4, out5, 6);
419   dst4 = LD_UB(dst + 3 * dst_stride);
420   dst5 = LD_UB(dst + 12 * dst_stride);
421   ILVR_B2_SH(zero, dst4, zero, dst5, res4, res5);
422   ADD2(res4, out4, res5, out5, res4, res5);
423   CLIP_SH2_0_255(res4, res5);
424   PCKEV_B2_SH(res4, res4, res5, res5, res4, res5);
425   ST8x1_UB(res4, dst + 3 * dst_stride);
426   ST8x1_UB(res5, dst + 12 * dst_stride);
427 
428   MADD_BF(h1, h3, h5, h7, k0, k1, k2, k0, out12, out14, out13, out15);
429   out13 = -out13;
430   SRARI_H2_SH(out12, out13, 6);
431   dst12 = LD_UB(dst + 2 * dst_stride);
432   dst13 = LD_UB(dst + 13 * dst_stride);
433   ILVR_B2_SH(zero, dst12, zero, dst13, res12, res13);
434   ADD2(res12, out12, res13, out13, res12, res13);
435   CLIP_SH2_0_255(res12, res13);
436   PCKEV_B2_SH(res12, res12, res13, res13, res12, res13);
437   ST8x1_UB(res12, dst + 2 * dst_stride);
438   ST8x1_UB(res13, dst + 13 * dst_stride);
439 
440   k0 = VP9_SET_COSPI_PAIR(cospi_16_64, cospi_16_64);
441   k3 = VP9_SET_COSPI_PAIR(-cospi_16_64, cospi_16_64);
442   MADD_SHORT(out6, out7, k0, k3, out6, out7);
443   SRARI_H2_SH(out6, out7, 6);
444   dst6 = LD_UB(dst + 4 * dst_stride);
445   dst7 = LD_UB(dst + 11 * dst_stride);
446   ILVR_B2_SH(zero, dst6, zero, dst7, res6, res7);
447   ADD2(res6, out6, res7, out7, res6, res7);
448   CLIP_SH2_0_255(res6, res7);
449   PCKEV_B2_SH(res6, res6, res7, res7, res6, res7);
450   ST8x1_UB(res6, dst + 4 * dst_stride);
451   ST8x1_UB(res7, dst + 11 * dst_stride);
452 
453   MADD_SHORT(out10, out11, k0, k3, out10, out11);
454   SRARI_H2_SH(out10, out11, 6);
455   dst10 = LD_UB(dst + 6 * dst_stride);
456   dst11 = LD_UB(dst + 9 * dst_stride);
457   ILVR_B2_SH(zero, dst10, zero, dst11, res10, res11);
458   ADD2(res10, out10, res11, out11, res10, res11);
459   CLIP_SH2_0_255(res10, res11);
460   PCKEV_B2_SH(res10, res10, res11, res11, res10, res11);
461   ST8x1_UB(res10, dst + 6 * dst_stride);
462   ST8x1_UB(res11, dst + 9 * dst_stride);
463 
464   k1 = VP9_SET_COSPI_PAIR(-cospi_16_64, -cospi_16_64);
465   k2 = VP9_SET_COSPI_PAIR(cospi_16_64, -cospi_16_64);
466   MADD_SHORT(h10, h11, k1, k2, out2, out3);
467   SRARI_H2_SH(out2, out3, 6);
468   dst2 = LD_UB(dst + 7 * dst_stride);
469   dst3 = LD_UB(dst + 8 * dst_stride);
470   ILVR_B2_SH(zero, dst2, zero, dst3, res2, res3);
471   ADD2(res2, out2, res3, out3, res2, res3);
472   CLIP_SH2_0_255(res2, res3);
473   PCKEV_B2_SH(res2, res2, res3, res3, res2, res3);
474   ST8x1_UB(res2, dst + 7 * dst_stride);
475   ST8x1_UB(res3, dst + 8 * dst_stride);
476 
477   MADD_SHORT(out14, out15, k1, k2, out14, out15);
478   SRARI_H2_SH(out14, out15, 6);
479   dst14 = LD_UB(dst + 5 * dst_stride);
480   dst15 = LD_UB(dst + 10 * dst_stride);
481   ILVR_B2_SH(zero, dst14, zero, dst15, res14, res15);
482   ADD2(res14, out14, res15, out15, res14, res15);
483   CLIP_SH2_0_255(res14, res15);
484   PCKEV_B2_SH(res14, res14, res15, res15, res14, res15);
485   ST8x1_UB(res14, dst + 5 * dst_stride);
486   ST8x1_UB(res15, dst + 10 * dst_stride);
487 }
488