2 * Copyright (c) 2022 The WebM project authors. All Rights Reserved.
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.
11 #include "./vpx_dsp_rtcd.h"
12 #include "vpx_util/loongson_intrinsics.h"
14 static void sub_blk_4x4_lsx(const uint8_t *src_ptr, int32_t src_stride,
15 const uint8_t *pred_ptr, int32_t pred_stride,
16 int16_t *diff_ptr, int32_t diff_stride) {
17 __m128i src0, src1, src2, src3;
18 __m128i pred0, pred1, pred2, pred3;
21 int32_t src_stride2 = src_stride << 1;
22 int32_t pred_stride2 = pred_stride << 1;
23 int32_t diff_stride2 = diff_stride << 1;
24 int32_t src_stride3 = src_stride2 + src_stride;
25 int32_t pred_stride3 = pred_stride2 + pred_stride;
26 int32_t diff_stride3 = diff_stride2 + diff_stride;
28 DUP4_ARG2(__lsx_vldrepl_w, src_ptr, 0, src_ptr + src_stride, 0,
29 src_ptr + src_stride2, 0, src_ptr + src_stride3, 0, src0, src1,
31 DUP4_ARG2(__lsx_vldrepl_w, pred_ptr, 0, pred_ptr + pred_stride, 0,
32 pred_ptr + pred_stride2, 0, pred_ptr + pred_stride3, 0, pred0,
34 DUP4_ARG2(__lsx_vilvl_w, src1, src0, src3, src2, pred1, pred0, pred3, pred2,
35 src0, src2, pred0, pred2);
36 DUP2_ARG2(__lsx_vilvl_d, src2, src0, pred2, pred0, src0, pred0);
37 reg0 = __lsx_vilvl_b(src0, pred0);
38 reg1 = __lsx_vilvh_b(src0, pred0);
39 DUP2_ARG2(__lsx_vhsubw_hu_bu, reg0, reg0, reg1, reg1, diff0, diff1);
40 __lsx_vstelm_d(diff0, diff_ptr, 0, 0);
41 __lsx_vstelm_d(diff0, diff_ptr + diff_stride, 0, 1);
42 __lsx_vstelm_d(diff1, diff_ptr + diff_stride2, 0, 0);
43 __lsx_vstelm_d(diff1, diff_ptr + diff_stride3, 0, 1);
46 static void sub_blk_8x8_lsx(const uint8_t *src_ptr, int32_t src_stride,
47 const uint8_t *pred_ptr, int32_t pred_stride,
48 int16_t *diff_ptr, int32_t diff_stride) {
49 __m128i src0, src1, src2, src3, src4, src5, src6, src7;
50 __m128i pred0, pred1, pred2, pred3, pred4, pred5, pred6, pred7;
51 __m128i reg0, reg1, reg2, reg3, reg4, reg5, reg6, reg7;
52 int32_t src_stride2 = src_stride << 1;
53 int32_t pred_stride2 = pred_stride << 1;
54 int32_t dst_stride = diff_stride << 1;
55 int32_t src_stride3 = src_stride2 + src_stride;
56 int32_t pred_stride3 = pred_stride2 + pred_stride;
57 int32_t dst_stride2 = dst_stride << 1;
58 int32_t src_stride4 = src_stride2 << 1;
59 int32_t pred_stride4 = pred_stride2 << 1;
60 int32_t dst_stride3 = dst_stride + dst_stride2;
62 DUP4_ARG2(__lsx_vldrepl_d, src_ptr, 0, src_ptr + src_stride, 0,
63 src_ptr + src_stride2, 0, src_ptr + src_stride3, 0, src0, src1,
65 DUP4_ARG2(__lsx_vldrepl_d, pred_ptr, 0, pred_ptr + pred_stride, 0,
66 pred_ptr + pred_stride2, 0, pred_ptr + pred_stride3, 0, pred0,
68 src_ptr += src_stride4;
69 pred_ptr += pred_stride4;
71 DUP4_ARG2(__lsx_vldrepl_d, src_ptr, 0, src_ptr + src_stride, 0,
72 src_ptr + src_stride2, 0, src_ptr + src_stride3, 0, src4, src5,
74 DUP4_ARG2(__lsx_vldrepl_d, pred_ptr, 0, pred_ptr + pred_stride, 0,
75 pred_ptr + pred_stride2, 0, pred_ptr + pred_stride3, 0, pred4,
78 DUP4_ARG2(__lsx_vilvl_b, src0, pred0, src1, pred1, src2, pred2, src3, pred3,
79 reg0, reg1, reg2, reg3);
80 DUP4_ARG2(__lsx_vilvl_b, src4, pred4, src5, pred5, src6, pred6, src7, pred7,
81 reg4, reg5, reg6, reg7);
82 DUP4_ARG2(__lsx_vhsubw_hu_bu, reg0, reg0, reg1, reg1, reg2, reg2, reg3, reg3,
83 src0, src1, src2, src3);
84 DUP4_ARG2(__lsx_vhsubw_hu_bu, reg4, reg4, reg5, reg5, reg6, reg6, reg7, reg7,
85 src4, src5, src6, src7);
86 __lsx_vst(src0, diff_ptr, 0);
87 __lsx_vstx(src1, diff_ptr, dst_stride);
88 __lsx_vstx(src2, diff_ptr, dst_stride2);
89 __lsx_vstx(src3, diff_ptr, dst_stride3);
90 diff_ptr += dst_stride2;
91 __lsx_vst(src4, diff_ptr, 0);
92 __lsx_vstx(src5, diff_ptr, dst_stride);
93 __lsx_vstx(src6, diff_ptr, dst_stride2);
94 __lsx_vstx(src7, diff_ptr, dst_stride3);
97 static void sub_blk_16x16_lsx(const uint8_t *src, int32_t src_stride,
98 const uint8_t *pred, int32_t pred_stride,
99 int16_t *diff, int32_t diff_stride) {
100 __m128i src0, src1, src2, src3, src4, src5, src6, src7;
101 __m128i pred0, pred1, pred2, pred3, pred4, pred5, pred6, pred7;
102 __m128i reg0, reg1, reg2, reg3, reg4, reg5, reg6, reg7;
103 __m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
104 int32_t src_stride2 = src_stride << 1;
105 int32_t pred_stride2 = pred_stride << 1;
106 int32_t dst_stride = diff_stride << 1;
107 int32_t src_stride3 = src_stride2 + src_stride;
108 int32_t pred_stride3 = pred_stride2 + pred_stride;
109 int32_t dst_stride2 = dst_stride << 1;
110 int32_t src_stride4 = src_stride2 << 1;
111 int32_t pred_stride4 = pred_stride2 << 1;
112 int32_t dst_stride3 = dst_stride + dst_stride2;
113 int16_t *diff_tmp = diff + 8;
115 DUP2_ARG2(__lsx_vld, src, 0, pred, 0, src0, pred0);
116 DUP4_ARG2(__lsx_vldx, src, src_stride, src, src_stride2, src, src_stride3,
117 src, src_stride4, src1, src2, src3, src4);
118 DUP4_ARG2(__lsx_vldx, pred, pred_stride, pred, pred_stride2, pred,
119 pred_stride3, pred, pred_stride4, pred1, pred2, pred3, pred4);
121 pred += pred_stride4;
122 DUP4_ARG2(__lsx_vldx, src, src_stride, src, src_stride2, src, src_stride3,
123 pred, pred_stride, src5, src6, src7, pred5);
124 DUP2_ARG2(__lsx_vldx, pred, pred_stride2, pred, pred_stride3, pred6, pred7);
126 pred += pred_stride4;
127 DUP4_ARG2(__lsx_vilvl_b, src0, pred0, src1, pred1, src2, pred2, src3, pred3,
128 reg0, reg2, reg4, reg6);
129 DUP4_ARG2(__lsx_vilvh_b, src0, pred0, src1, pred1, src2, pred2, src3, pred3,
130 reg1, reg3, reg5, reg7);
131 DUP4_ARG2(__lsx_vilvl_b, src4, pred4, src5, pred5, src6, pred6, src7, pred7,
132 tmp0, tmp2, tmp4, tmp6);
133 DUP4_ARG2(__lsx_vilvh_b, src4, pred4, src5, pred5, src6, pred6, src7, pred7,
134 tmp1, tmp3, tmp5, tmp7);
135 DUP4_ARG2(__lsx_vhsubw_hu_bu, reg0, reg0, reg1, reg1, reg2, reg2, reg3, reg3,
136 src0, src1, src2, src3);
137 DUP4_ARG2(__lsx_vhsubw_hu_bu, reg4, reg4, reg5, reg5, reg6, reg6, reg7, reg7,
138 src4, src5, src6, src7);
139 DUP4_ARG2(__lsx_vhsubw_hu_bu, tmp0, tmp0, tmp1, tmp1, tmp2, tmp2, tmp3, tmp3,
140 pred0, pred1, pred2, pred3);
141 DUP4_ARG2(__lsx_vhsubw_hu_bu, tmp4, tmp4, tmp5, tmp5, tmp6, tmp6, tmp7, tmp7,
142 pred4, pred5, pred6, pred7);
143 __lsx_vst(src0, diff, 0);
144 __lsx_vstx(src2, diff, dst_stride);
145 __lsx_vstx(src4, diff, dst_stride2);
146 __lsx_vstx(src6, diff, dst_stride3);
147 __lsx_vst(src1, diff_tmp, 0);
148 __lsx_vstx(src3, diff_tmp, dst_stride);
149 __lsx_vstx(src5, diff_tmp, dst_stride2);
150 __lsx_vstx(src7, diff_tmp, dst_stride3);
152 diff_tmp += dst_stride2;
153 __lsx_vst(pred0, diff, 0);
154 __lsx_vstx(pred2, diff, dst_stride);
155 __lsx_vstx(pred4, diff, dst_stride2);
156 __lsx_vstx(pred6, diff, dst_stride3);
157 __lsx_vst(pred1, diff_tmp, 0);
158 __lsx_vstx(pred3, diff_tmp, dst_stride);
159 __lsx_vstx(pred5, diff_tmp, dst_stride2);
160 __lsx_vstx(pred7, diff_tmp, dst_stride3);
162 diff_tmp += dst_stride2;
163 DUP2_ARG2(__lsx_vld, src, 0, pred, 0, src0, pred0);
164 DUP4_ARG2(__lsx_vldx, src, src_stride, src, src_stride2, src, src_stride3,
165 src, src_stride4, src1, src2, src3, src4);
166 DUP4_ARG2(__lsx_vldx, pred, pred_stride, pred, pred_stride2, pred,
167 pred_stride3, pred, pred_stride4, pred1, pred2, pred3, pred4);
169 pred += pred_stride4;
170 DUP4_ARG2(__lsx_vldx, src, src_stride, src, src_stride2, src, src_stride3,
171 pred, pred_stride, src5, src6, src7, pred5);
172 DUP2_ARG2(__lsx_vldx, pred, pred_stride2, pred, pred_stride3, pred6, pred7);
173 DUP4_ARG2(__lsx_vilvl_b, src0, pred0, src1, pred1, src2, pred2, src3, pred3,
174 reg0, reg2, reg4, reg6);
175 DUP4_ARG2(__lsx_vilvh_b, src0, pred0, src1, pred1, src2, pred2, src3, pred3,
176 reg1, reg3, reg5, reg7);
177 DUP4_ARG2(__lsx_vilvl_b, src4, pred4, src5, pred5, src6, pred6, src7, pred7,
178 tmp0, tmp2, tmp4, tmp6);
179 DUP4_ARG2(__lsx_vilvh_b, src4, pred4, src5, pred5, src6, pred6, src7, pred7,
180 tmp1, tmp3, tmp5, tmp7);
181 DUP4_ARG2(__lsx_vhsubw_hu_bu, reg0, reg0, reg1, reg1, reg2, reg2, reg3, reg3,
182 src0, src1, src2, src3);
183 DUP4_ARG2(__lsx_vhsubw_hu_bu, reg4, reg4, reg5, reg5, reg6, reg6, reg7, reg7,
184 src4, src5, src6, src7);
185 DUP4_ARG2(__lsx_vhsubw_hu_bu, tmp0, tmp0, tmp1, tmp1, tmp2, tmp2, tmp3, tmp3,
186 pred0, pred1, pred2, pred3);
187 DUP4_ARG2(__lsx_vhsubw_hu_bu, tmp4, tmp4, tmp5, tmp5, tmp6, tmp6, tmp7, tmp7,
188 pred4, pred5, pred6, pred7);
189 __lsx_vst(src0, diff, 0);
190 __lsx_vstx(src2, diff, dst_stride);
191 __lsx_vstx(src4, diff, dst_stride2);
192 __lsx_vstx(src6, diff, dst_stride3);
193 __lsx_vst(src1, diff_tmp, 0);
194 __lsx_vstx(src3, diff_tmp, dst_stride);
195 __lsx_vstx(src5, diff_tmp, dst_stride2);
196 __lsx_vstx(src7, diff_tmp, dst_stride3);
198 diff_tmp += dst_stride2;
199 __lsx_vst(pred0, diff, 0);
200 __lsx_vstx(pred2, diff, dst_stride);
201 __lsx_vstx(pred4, diff, dst_stride2);
202 __lsx_vstx(pred6, diff, dst_stride3);
203 __lsx_vst(pred1, diff_tmp, 0);
204 __lsx_vstx(pred3, diff_tmp, dst_stride);
205 __lsx_vstx(pred5, diff_tmp, dst_stride2);
206 __lsx_vstx(pred7, diff_tmp, dst_stride3);
209 static void sub_blk_32x32_lsx(const uint8_t *src, int32_t src_stride,
210 const uint8_t *pred, int32_t pred_stride,
211 int16_t *diff, int32_t diff_stride) {
212 __m128i src0, src1, src2, src3, src4, src5, src6, src7;
213 __m128i pred0, pred1, pred2, pred3, pred4, pred5, pred6, pred7;
214 __m128i reg0, reg1, reg2, reg3, reg4, reg5, reg6, reg7;
215 __m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
217 int32_t src_stride2 = src_stride << 1;
218 int32_t pred_stride2 = pred_stride << 1;
219 int32_t src_stride3 = src_stride2 + src_stride;
220 int32_t pred_stride3 = pred_stride2 + pred_stride;
221 int32_t src_stride4 = src_stride2 << 1;
222 int32_t pred_stride4 = pred_stride2 << 1;
224 for (loop_cnt = 8; loop_cnt--;) {
225 const uint8_t *src_tmp = src + 16;
226 const uint8_t *pred_tmp = pred + 16;
227 DUP4_ARG2(__lsx_vld, src, 0, src_tmp, 0, pred, 0, pred_tmp, 0, src0, src1,
229 DUP4_ARG2(__lsx_vldx, src, src_stride, src_tmp, src_stride, src,
230 src_stride2, src_tmp, src_stride2, src2, src3, src4, src5);
231 DUP4_ARG2(__lsx_vldx, src, src_stride3, src_tmp, src_stride3, pred,
232 pred_stride, pred_tmp, pred_stride, src6, src7, pred2, pred3);
233 DUP4_ARG2(__lsx_vldx, pred, pred_stride2, pred_tmp, pred_stride2, pred,
234 pred_stride3, pred_tmp, pred_stride3, pred4, pred5, pred6, pred7);
235 DUP4_ARG2(__lsx_vilvl_b, src0, pred0, src1, pred1, src2, pred2, src3, pred3,
236 reg0, reg2, reg4, reg6);
237 DUP4_ARG2(__lsx_vilvh_b, src0, pred0, src1, pred1, src2, pred2, src3, pred3,
238 reg1, reg3, reg5, reg7);
239 DUP4_ARG2(__lsx_vilvl_b, src4, pred4, src5, pred5, src6, pred6, src7, pred7,
240 tmp0, tmp2, tmp4, tmp6);
241 DUP4_ARG2(__lsx_vilvh_b, src4, pred4, src5, pred5, src6, pred6, src7, pred7,
242 tmp1, tmp3, tmp5, tmp7);
243 DUP4_ARG2(__lsx_vhsubw_hu_bu, reg0, reg0, reg1, reg1, reg2, reg2, reg3,
244 reg3, src0, src1, src2, src3);
245 DUP4_ARG2(__lsx_vhsubw_hu_bu, reg4, reg4, reg5, reg5, reg6, reg6, reg7,
246 reg7, src4, src5, src6, src7);
247 DUP4_ARG2(__lsx_vhsubw_hu_bu, tmp0, tmp0, tmp1, tmp1, tmp2, tmp2, tmp3,
248 tmp3, pred0, pred1, pred2, pred3);
249 DUP4_ARG2(__lsx_vhsubw_hu_bu, tmp4, tmp4, tmp5, tmp5, tmp6, tmp6, tmp7,
250 tmp7, pred4, pred5, pred6, pred7);
252 pred += pred_stride4;
253 __lsx_vst(src0, diff, 0);
254 __lsx_vst(src1, diff, 16);
255 __lsx_vst(src2, diff, 32);
256 __lsx_vst(src3, diff, 48);
258 __lsx_vst(src4, diff, 0);
259 __lsx_vst(src5, diff, 16);
260 __lsx_vst(src6, diff, 32);
261 __lsx_vst(src7, diff, 48);
263 __lsx_vst(pred0, diff, 0);
264 __lsx_vst(pred1, diff, 16);
265 __lsx_vst(pred2, diff, 32);
266 __lsx_vst(pred3, diff, 48);
268 __lsx_vst(pred4, diff, 0);
269 __lsx_vst(pred5, diff, 16);
270 __lsx_vst(pred6, diff, 32);
271 __lsx_vst(pred7, diff, 48);
276 static void sub_blk_64x64_lsx(const uint8_t *src, int32_t src_stride,
277 const uint8_t *pred, int32_t pred_stride,
278 int16_t *diff, int32_t diff_stride) {
279 __m128i src0, src1, src2, src3, src4, src5, src6, src7;
280 __m128i pred0, pred1, pred2, pred3, pred4, pred5, pred6, pred7;
281 __m128i reg0, reg1, reg2, reg3, reg4, reg5, reg6, reg7;
282 __m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
285 for (loop_cnt = 32; loop_cnt--;) {
286 DUP4_ARG2(__lsx_vld, src, 0, src, 16, src, 32, src, 48, src0, src1, src2,
288 DUP4_ARG2(__lsx_vld, pred, 0, pred, 16, pred, 32, pred, 48, pred0, pred1,
292 DUP4_ARG2(__lsx_vld, src, 0, src, 16, src, 32, src, 48, src4, src5, src6,
294 DUP4_ARG2(__lsx_vld, pred, 0, pred, 16, pred, 32, pred, 48, pred4, pred5,
299 DUP4_ARG2(__lsx_vilvl_b, src0, pred0, src1, pred1, src2, pred2, src3, pred3,
300 reg0, reg2, reg4, reg6);
301 DUP4_ARG2(__lsx_vilvh_b, src0, pred0, src1, pred1, src2, pred2, src3, pred3,
302 reg1, reg3, reg5, reg7);
303 DUP4_ARG2(__lsx_vilvl_b, src4, pred4, src5, pred5, src6, pred6, src7, pred7,
304 tmp0, tmp2, tmp4, tmp6);
305 DUP4_ARG2(__lsx_vilvh_b, src4, pred4, src5, pred5, src6, pred6, src7, pred7,
306 tmp1, tmp3, tmp5, tmp7);
307 DUP4_ARG2(__lsx_vhsubw_hu_bu, reg0, reg0, reg1, reg1, reg2, reg2, reg3,
308 reg3, src0, src1, src2, src3);
309 DUP4_ARG2(__lsx_vhsubw_hu_bu, reg4, reg4, reg5, reg5, reg6, reg6, reg7,
310 reg7, src4, src5, src6, src7);
311 DUP4_ARG2(__lsx_vhsubw_hu_bu, tmp0, tmp0, tmp1, tmp1, tmp2, tmp2, tmp3,
312 tmp3, pred0, pred1, pred2, pred3);
313 DUP4_ARG2(__lsx_vhsubw_hu_bu, tmp4, tmp4, tmp5, tmp5, tmp6, tmp6, tmp7,
314 tmp7, pred4, pred5, pred6, pred7);
315 __lsx_vst(src0, diff, 0);
316 __lsx_vst(src1, diff, 16);
317 __lsx_vst(src2, diff, 32);
318 __lsx_vst(src3, diff, 48);
319 __lsx_vst(src4, diff, 64);
320 __lsx_vst(src5, diff, 80);
321 __lsx_vst(src6, diff, 96);
322 __lsx_vst(src7, diff, 112);
324 __lsx_vst(pred0, diff, 0);
325 __lsx_vst(pred1, diff, 16);
326 __lsx_vst(pred2, diff, 32);
327 __lsx_vst(pred3, diff, 48);
328 __lsx_vst(pred4, diff, 64);
329 __lsx_vst(pred5, diff, 80);
330 __lsx_vst(pred6, diff, 96);
331 __lsx_vst(pred7, diff, 112);
336 void vpx_subtract_block_lsx(int32_t rows, int32_t cols, int16_t *diff_ptr,
337 ptrdiff_t diff_stride, const uint8_t *src_ptr,
338 ptrdiff_t src_stride, const uint8_t *pred_ptr,
339 ptrdiff_t pred_stride) {
343 sub_blk_4x4_lsx(src_ptr, src_stride, pred_ptr, pred_stride, diff_ptr,
347 sub_blk_8x8_lsx(src_ptr, src_stride, pred_ptr, pred_stride, diff_ptr,
351 sub_blk_16x16_lsx(src_ptr, src_stride, pred_ptr, pred_stride, diff_ptr,
355 sub_blk_32x32_lsx(src_ptr, src_stride, pred_ptr, pred_stride, diff_ptr,
359 sub_blk_64x64_lsx(src_ptr, src_stride, pred_ptr, pred_stride, diff_ptr,
363 vpx_subtract_block_c(rows, cols, diff_ptr, diff_stride, src_ptr,
364 src_stride, pred_ptr, pred_stride);
368 vpx_subtract_block_c(rows, cols, diff_ptr, diff_stride, src_ptr, src_stride,
369 pred_ptr, pred_stride);