1 /*
2 * Copyright (c) 2022 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 "./vp8_rtcd.h"
12 #include "vp8/common/blockd.h"
13 #include "vpx_util/loongson_intrinsics.h"
14
15 static const int32_t cospi8sqrt2minus1 = 20091;
16 static const int32_t sinpi8sqrt2 = 35468;
17
18 #define TRANSPOSE8X4_SH_SH(in0, in1, in2, in3, out0, out1, out2, out3) \
19 do { \
20 __m128i tmp0_m, tmp1_m, tmp2_m, tmp3_m; \
21 \
22 DUP2_ARG2(__lsx_vilvl_h, in1, in0, in3, in2, tmp0_m, tmp1_m); \
23 DUP2_ARG2(__lsx_vilvh_h, in1, in0, in3, in2, tmp2_m, tmp3_m); \
24 DUP2_ARG2(__lsx_vilvl_w, tmp1_m, tmp0_m, tmp3_m, tmp2_m, out0, out2); \
25 DUP2_ARG2(__lsx_vilvh_w, tmp1_m, tmp0_m, tmp3_m, tmp2_m, out1, out3); \
26 } while (0)
27
28 #define TRANSPOSE_TWO_4x4_H(in0, in1, in2, in3, out0, out1, out2, out3) \
29 do { \
30 __m128i s4_m, s5_m, s6_m, s7_m; \
31 \
32 TRANSPOSE8X4_SH_SH(in0, in1, in2, in3, s4_m, s5_m, s6_m, s7_m); \
33 DUP2_ARG2(__lsx_vilvl_d, s6_m, s4_m, s7_m, s5_m, out0, out2); \
34 out1 = __lsx_vilvh_d(s6_m, s4_m); \
35 out3 = __lsx_vilvh_d(s7_m, s5_m); \
36 } while (0)
37
38 #define EXPAND_TO_H_MULTIPLY_SINPI8SQRT2_PCK_TO_W(in0, in1) \
39 do { \
40 __m128i zero_m = __lsx_vldi(0); \
41 __m128i tmp1_m, tmp2_m; \
42 __m128i sinpi8_sqrt2_m = __lsx_vreplgr2vr_w(sinpi8sqrt2); \
43 \
44 tmp1_m = __lsx_vilvl_h(in0, zero_m); \
45 tmp2_m = __lsx_vilvh_h(in0, zero_m); \
46 tmp1_m = __lsx_vsrai_w(tmp1_m, 16); \
47 tmp2_m = __lsx_vsrai_w(tmp2_m, 16); \
48 tmp1_m = __lsx_vmul_w(tmp1_m, sinpi8_sqrt2_m); \
49 tmp1_m = __lsx_vsrai_w(tmp1_m, 16); \
50 tmp2_m = __lsx_vmul_w(tmp2_m, sinpi8_sqrt2_m); \
51 tmp2_m = __lsx_vsrai_w(tmp2_m, 16); \
52 in1 = __lsx_vpickev_h(tmp2_m, tmp1_m); \
53 } while (0)
54
55 #define VP8_IDCT_1D_H(in0, in1, in2, in3, out0, out1, out2, out3) \
56 do { \
57 __m128i a1_m, b1_m, c1_m, d1_m; \
58 __m128i c_tmp1_m, c_tmp2_m; \
59 __m128i d_tmp1_m, d_tmp2_m; \
60 __m128i const_cospi8sqrt2minus1_m; \
61 \
62 const_cospi8sqrt2minus1_m = __lsx_vreplgr2vr_h(cospi8sqrt2minus1); \
63 a1_m = __lsx_vadd_h(in0, in2); \
64 b1_m = __lsx_vsub_h(in0, in2); \
65 EXPAND_TO_H_MULTIPLY_SINPI8SQRT2_PCK_TO_W(in1, c_tmp1_m); \
66 \
67 c_tmp2_m = __lsx_vmuh_h(in3, const_cospi8sqrt2minus1_m); \
68 c_tmp2_m = __lsx_vslli_h(c_tmp2_m, 1); \
69 c_tmp2_m = __lsx_vsrai_h(c_tmp2_m, 1); \
70 c_tmp2_m = __lsx_vadd_h(in3, c_tmp2_m); \
71 c1_m = __lsx_vsub_h(c_tmp1_m, c_tmp2_m); \
72 \
73 d_tmp1_m = __lsx_vmuh_h(in1, const_cospi8sqrt2minus1_m); \
74 d_tmp1_m = __lsx_vslli_h(d_tmp1_m, 1); \
75 d_tmp1_m = __lsx_vsrai_h(d_tmp1_m, 1); \
76 d_tmp1_m = __lsx_vadd_h(in1, d_tmp1_m); \
77 EXPAND_TO_H_MULTIPLY_SINPI8SQRT2_PCK_TO_W(in3, d_tmp2_m); \
78 d1_m = __lsx_vadd_h(d_tmp1_m, d_tmp2_m); \
79 LSX_BUTTERFLY_4_H(a1_m, b1_m, c1_m, d1_m, out0, out1, out2, out3); \
80 } while (0)
81
82 #define VP8_IDCT_1D_W(in0, in1, in2, in3, out0, out1, out2, out3) \
83 do { \
84 __m128i a1_m, b1_m, c1_m, d1_m; \
85 __m128i c_tmp1_m, c_tmp2_m, d_tmp1_m, d_tmp2_m; \
86 __m128i const_cospi8sqrt2minus1_m, sinpi8_sqrt2_m; \
87 \
88 const_cospi8sqrt2minus1_m = __lsx_vreplgr2vr_w(cospi8sqrt2minus1); \
89 sinpi8_sqrt2_m = __lsx_vreplgr2vr_w(sinpi8sqrt2); \
90 a1_m = __lsx_vadd_w(in0, in2); \
91 b1_m = __lsx_vsub_w(in0, in2); \
92 c_tmp1_m = __lsx_vmul_w(in1, sinpi8_sqrt2_m); \
93 c_tmp1_m = __lsx_vsrai_w(c_tmp1_m, 16); \
94 c_tmp2_m = __lsx_vmul_w(in3, const_cospi8sqrt2minus1_m); \
95 c_tmp2_m = __lsx_vsrai_w(c_tmp2_m, 16); \
96 c_tmp2_m = __lsx_vadd_w(in3, c_tmp2_m); \
97 c1_m = __lsx_vsub_w(c_tmp1_m, c_tmp2_m); \
98 d_tmp1_m = __lsx_vmul_w(in1, const_cospi8sqrt2minus1_m); \
99 d_tmp1_m = __lsx_vsrai_w(d_tmp1_m, 16); \
100 d_tmp1_m = __lsx_vadd_w(in1, d_tmp1_m); \
101 d_tmp2_m = __lsx_vmul_w(in3, sinpi8_sqrt2_m); \
102 d_tmp2_m = __lsx_vsrai_w(d_tmp2_m, 16); \
103 d1_m = __lsx_vadd_w(d_tmp1_m, d_tmp2_m); \
104 LSX_BUTTERFLY_4_W(a1_m, b1_m, c1_m, d1_m, out0, out1, out2, out3); \
105 } while (0)
106
107 #define UNPCK_SH_SW(in, out0, out1) \
108 do { \
109 out0 = __lsx_vsllwil_w_h(in, 0); \
110 out1 = __lsx_vexth_w_h(in); \
111 } while (0)
112
idct4x4_addconst_lsx(int16_t in_dc,uint8_t * pred,int32_t pred_stride,uint8_t * dest,int32_t dest_stride)113 static void idct4x4_addconst_lsx(int16_t in_dc, uint8_t *pred,
114 int32_t pred_stride, uint8_t *dest,
115 int32_t dest_stride) {
116 __m128i vec, res0, res1, res2, res3, dst0, dst1;
117 __m128i pred0, pred1, pred2, pred3;
118 __m128i zero = __lsx_vldi(0);
119
120 int32_t pred_stride2 = pred_stride << 1;
121 int32_t pred_stride3 = pred_stride2 + pred_stride;
122
123 vec = __lsx_vreplgr2vr_h(in_dc);
124 vec = __lsx_vsrari_h(vec, 3);
125 pred0 = __lsx_vld(pred, 0);
126 DUP2_ARG2(__lsx_vldx, pred, pred_stride, pred, pred_stride2, pred1, pred2);
127 pred3 = __lsx_vldx(pred, pred_stride3);
128 DUP4_ARG2(__lsx_vilvl_b, zero, pred0, zero, pred1, zero, pred2, zero, pred3,
129 res0, res1, res2, res3);
130 DUP4_ARG2(__lsx_vadd_h, res0, vec, res1, vec, res2, vec, res3, vec, res0,
131 res1, res2, res3);
132 res0 = __lsx_vclip255_h(res0);
133 res1 = __lsx_vclip255_h(res1);
134 res2 = __lsx_vclip255_h(res2);
135 res3 = __lsx_vclip255_h(res3);
136
137 DUP2_ARG2(__lsx_vpickev_b, res1, res0, res3, res2, dst0, dst1);
138 dst0 = __lsx_vpickev_w(dst1, dst0);
139 __lsx_vstelm_w(dst0, dest, 0, 0);
140 dest += dest_stride;
141 __lsx_vstelm_w(dst0, dest, 0, 1);
142 dest += dest_stride;
143 __lsx_vstelm_w(dst0, dest, 0, 2);
144 dest += dest_stride;
145 __lsx_vstelm_w(dst0, dest, 0, 3);
146 }
147
vp8_dc_only_idct_add_lsx(int16_t input_dc,uint8_t * pred_ptr,int32_t pred_stride,uint8_t * dst_ptr,int32_t dst_stride)148 void vp8_dc_only_idct_add_lsx(int16_t input_dc, uint8_t *pred_ptr,
149 int32_t pred_stride, uint8_t *dst_ptr,
150 int32_t dst_stride) {
151 idct4x4_addconst_lsx(input_dc, pred_ptr, pred_stride, dst_ptr, dst_stride);
152 }
153
dequant_idct4x4_addblk_2x_lsx(int16_t * input,int16_t * dequant_input,uint8_t * dest,int32_t dest_stride)154 static void dequant_idct4x4_addblk_2x_lsx(int16_t *input,
155 int16_t *dequant_input, uint8_t *dest,
156 int32_t dest_stride) {
157 __m128i dest0, dest1, dest2, dest3;
158 __m128i in0, in1, in2, in3, mul0, mul1, mul2, mul3, dequant_in0, dequant_in1;
159 __m128i hz0, hz1, hz2, hz3, vt0, vt1, vt2, vt3, res0, res1, res2, res3;
160 __m128i hz0l, hz1l, hz2l, hz3l, hz0r, hz1r, hz2r, hz3r;
161 __m128i vt0l, vt1l, vt2l, vt3l, vt0r, vt1r, vt2r, vt3r;
162 __m128i zero = __lsx_vldi(0);
163
164 int32_t dest_stride2 = dest_stride << 1;
165 int32_t dest_stride3 = dest_stride2 + dest_stride;
166
167 DUP4_ARG2(__lsx_vld, input, 0, input, 16, input, 32, input, 48, in0, in1, in2,
168 in3);
169 DUP2_ARG2(__lsx_vld, dequant_input, 0, dequant_input, 16, dequant_in0,
170 dequant_in1);
171
172 DUP4_ARG2(__lsx_vmul_h, in0, dequant_in0, in1, dequant_in1, in2, dequant_in0,
173 in3, dequant_in1, mul0, mul1, mul2, mul3);
174 DUP2_ARG2(__lsx_vpickev_d, mul2, mul0, mul3, mul1, in0, in2);
175 DUP2_ARG2(__lsx_vpickod_d, mul2, mul0, mul3, mul1, in1, in3);
176
177 VP8_IDCT_1D_H(in0, in1, in2, in3, hz0, hz1, hz2, hz3);
178 TRANSPOSE_TWO_4x4_H(hz0, hz1, hz2, hz3, hz0, hz1, hz2, hz3);
179 UNPCK_SH_SW(hz0, hz0r, hz0l);
180 UNPCK_SH_SW(hz1, hz1r, hz1l);
181 UNPCK_SH_SW(hz2, hz2r, hz2l);
182 UNPCK_SH_SW(hz3, hz3r, hz3l);
183 VP8_IDCT_1D_W(hz0l, hz1l, hz2l, hz3l, vt0l, vt1l, vt2l, vt3l);
184 DUP4_ARG2(__lsx_vsrari_w, vt0l, 3, vt1l, 3, vt2l, 3, vt3l, 3, vt0l, vt1l,
185 vt2l, vt3l);
186 VP8_IDCT_1D_W(hz0r, hz1r, hz2r, hz3r, vt0r, vt1r, vt2r, vt3r);
187 DUP4_ARG2(__lsx_vsrari_w, vt0r, 3, vt1r, 3, vt2r, 3, vt3r, 3, vt0r, vt1r,
188 vt2r, vt3r);
189 DUP4_ARG2(__lsx_vpickev_h, vt0l, vt0r, vt1l, vt1r, vt2l, vt2r, vt3l, vt3r,
190 vt0, vt1, vt2, vt3);
191 TRANSPOSE_TWO_4x4_H(vt0, vt1, vt2, vt3, vt0, vt1, vt2, vt3);
192 dest0 = __lsx_vld(dest, 0);
193 DUP2_ARG2(__lsx_vldx, dest, dest_stride, dest, dest_stride2, dest1, dest2);
194 dest3 = __lsx_vldx(dest, dest_stride3);
195 DUP4_ARG2(__lsx_vilvl_b, zero, dest0, zero, dest1, zero, dest2, zero, dest3,
196 res0, res1, res2, res3);
197 DUP4_ARG2(__lsx_vadd_h, res0, vt0, res1, vt1, res2, vt2, res3, vt3, res0,
198 res1, res2, res3);
199
200 res0 = __lsx_vclip255_h(res0);
201 res1 = __lsx_vclip255_h(res1);
202 res2 = __lsx_vclip255_h(res2);
203 res3 = __lsx_vclip255_h(res3);
204 DUP2_ARG2(__lsx_vpickev_b, res1, res0, res3, res2, vt0l, vt1l);
205
206 __lsx_vstelm_d(vt0l, dest, 0, 0);
207 __lsx_vstelm_d(vt0l, dest + dest_stride, 0, 1);
208 __lsx_vstelm_d(vt1l, dest + dest_stride2, 0, 0);
209 __lsx_vstelm_d(vt1l, dest + dest_stride3, 0, 1);
210
211 __lsx_vst(zero, input, 0);
212 __lsx_vst(zero, input, 16);
213 __lsx_vst(zero, input, 32);
214 __lsx_vst(zero, input, 48);
215 }
216
dequant_idct_addconst_2x_lsx(int16_t * input,int16_t * dequant_input,uint8_t * dest,int32_t dest_stride)217 static void dequant_idct_addconst_2x_lsx(int16_t *input, int16_t *dequant_input,
218 uint8_t *dest, int32_t dest_stride) {
219 __m128i input_dc0, input_dc1, vec, res0, res1, res2, res3;
220 __m128i dest0, dest1, dest2, dest3;
221 __m128i zero = __lsx_vldi(0);
222 int32_t dest_stride2 = dest_stride << 1;
223 int32_t dest_stride3 = dest_stride2 + dest_stride;
224
225 input_dc0 = __lsx_vreplgr2vr_h(input[0] * dequant_input[0]);
226 input_dc1 = __lsx_vreplgr2vr_h(input[16] * dequant_input[0]);
227 DUP2_ARG2(__lsx_vsrari_h, input_dc0, 3, input_dc1, 3, input_dc0, input_dc1);
228 vec = __lsx_vpickev_d(input_dc1, input_dc0);
229 input[0] = 0;
230 input[16] = 0;
231 dest0 = __lsx_vld(dest, 0);
232 DUP2_ARG2(__lsx_vldx, dest, dest_stride, dest, dest_stride2, dest1, dest2);
233 dest3 = __lsx_vldx(dest, dest_stride3);
234 DUP4_ARG2(__lsx_vilvl_b, zero, dest0, zero, dest1, zero, dest2, zero, dest3,
235 res0, res1, res2, res3);
236 DUP4_ARG2(__lsx_vadd_h, res0, vec, res1, vec, res2, vec, res3, vec, res0,
237 res1, res2, res3);
238 res0 = __lsx_vclip255_h(res0);
239 res1 = __lsx_vclip255_h(res1);
240 res2 = __lsx_vclip255_h(res2);
241 res3 = __lsx_vclip255_h(res3);
242
243 DUP2_ARG2(__lsx_vpickev_b, res1, res0, res3, res2, res0, res1);
244 __lsx_vstelm_d(res0, dest, 0, 0);
245 __lsx_vstelm_d(res0, dest + dest_stride, 0, 1);
246 __lsx_vstelm_d(res1, dest + dest_stride2, 0, 0);
247 __lsx_vstelm_d(res1, dest + dest_stride3, 0, 1);
248 }
249
vp8_dequant_idct_add_y_block_lsx(int16_t * q,int16_t * dq,uint8_t * dst,int32_t stride,char * eobs)250 void vp8_dequant_idct_add_y_block_lsx(int16_t *q, int16_t *dq, uint8_t *dst,
251 int32_t stride, char *eobs) {
252 int16_t *eobs_h = (int16_t *)eobs;
253 uint8_t i;
254
255 for (i = 4; i--;) {
256 if (eobs_h[0]) {
257 if (eobs_h[0] & 0xfefe) {
258 dequant_idct4x4_addblk_2x_lsx(q, dq, dst, stride);
259 } else {
260 dequant_idct_addconst_2x_lsx(q, dq, dst, stride);
261 }
262 }
263
264 q += 32;
265
266 if (eobs_h[1]) {
267 if (eobs_h[1] & 0xfefe) {
268 dequant_idct4x4_addblk_2x_lsx(q, dq, dst + 8, stride);
269 } else {
270 dequant_idct_addconst_2x_lsx(q, dq, dst + 8, stride);
271 }
272 }
273
274 q += 32;
275 dst += (4 * stride);
276 eobs_h += 2;
277 }
278 }
279
vp8_dequant_idct_add_uv_block_lsx(int16_t * q,int16_t * dq,uint8_t * dst_u,uint8_t * dst_v,int32_t stride,char * eobs)280 void vp8_dequant_idct_add_uv_block_lsx(int16_t *q, int16_t *dq, uint8_t *dst_u,
281 uint8_t *dst_v, int32_t stride,
282 char *eobs) {
283 int16_t *eobs_h = (int16_t *)eobs;
284 if (eobs_h[0]) {
285 if (eobs_h[0] & 0xfefe) {
286 dequant_idct4x4_addblk_2x_lsx(q, dq, dst_u, stride);
287 } else {
288 dequant_idct_addconst_2x_lsx(q, dq, dst_u, stride);
289 }
290 }
291
292 q += 32;
293 dst_u += (stride * 4);
294
295 if (eobs_h[1]) {
296 if (eobs_h[1] & 0xfefe) {
297 dequant_idct4x4_addblk_2x_lsx(q, dq, dst_u, stride);
298 } else {
299 dequant_idct_addconst_2x_lsx(q, dq, dst_u, stride);
300 }
301 }
302
303 q += 32;
304
305 if (eobs_h[2]) {
306 if (eobs_h[2] & 0xfefe) {
307 dequant_idct4x4_addblk_2x_lsx(q, dq, dst_v, stride);
308 } else {
309 dequant_idct_addconst_2x_lsx(q, dq, dst_v, stride);
310 }
311 }
312 q += 32;
313 dst_v += (stride * 4);
314
315 if (eobs_h[3]) {
316 if (eobs_h[3] & 0xfefe) {
317 dequant_idct4x4_addblk_2x_lsx(q, dq, dst_v, stride);
318 } else {
319 dequant_idct_addconst_2x_lsx(q, dq, dst_v, stride);
320 }
321 }
322 }
323