1 /*
2 * Copyright 2016 The LibYuv 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 "libyuv/rotate_row.h"
12
13 // This module is for GCC MSA
14 #if !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa)
15 #include "libyuv/macros_msa.h"
16
17 #ifdef __cplusplus
18 namespace libyuv {
19 extern "C" {
20 #endif
21
22 #define ILVRL_B(in0, in1, in2, in3, out0, out1, out2, out3) \
23 { \
24 out0 = (v16u8)__msa_ilvr_b((v16i8)in1, (v16i8)in0); \
25 out1 = (v16u8)__msa_ilvl_b((v16i8)in1, (v16i8)in0); \
26 out2 = (v16u8)__msa_ilvr_b((v16i8)in3, (v16i8)in2); \
27 out3 = (v16u8)__msa_ilvl_b((v16i8)in3, (v16i8)in2); \
28 }
29
30 #define ILVRL_H(in0, in1, in2, in3, out0, out1, out2, out3) \
31 { \
32 out0 = (v16u8)__msa_ilvr_h((v8i16)in1, (v8i16)in0); \
33 out1 = (v16u8)__msa_ilvl_h((v8i16)in1, (v8i16)in0); \
34 out2 = (v16u8)__msa_ilvr_h((v8i16)in3, (v8i16)in2); \
35 out3 = (v16u8)__msa_ilvl_h((v8i16)in3, (v8i16)in2); \
36 }
37
38 #define ILVRL_W(in0, in1, in2, in3, out0, out1, out2, out3) \
39 { \
40 out0 = (v16u8)__msa_ilvr_w((v4i32)in1, (v4i32)in0); \
41 out1 = (v16u8)__msa_ilvl_w((v4i32)in1, (v4i32)in0); \
42 out2 = (v16u8)__msa_ilvr_w((v4i32)in3, (v4i32)in2); \
43 out3 = (v16u8)__msa_ilvl_w((v4i32)in3, (v4i32)in2); \
44 }
45
46 #define ILVRL_D(in0, in1, in2, in3, out0, out1, out2, out3) \
47 { \
48 out0 = (v16u8)__msa_ilvr_d((v2i64)in1, (v2i64)in0); \
49 out1 = (v16u8)__msa_ilvl_d((v2i64)in1, (v2i64)in0); \
50 out2 = (v16u8)__msa_ilvr_d((v2i64)in3, (v2i64)in2); \
51 out3 = (v16u8)__msa_ilvl_d((v2i64)in3, (v2i64)in2); \
52 }
53
TransposeWx16_C(const uint8_t * src,int src_stride,uint8_t * dst,int dst_stride,int width)54 void TransposeWx16_C(const uint8_t* src,
55 int src_stride,
56 uint8_t* dst,
57 int dst_stride,
58 int width) {
59 TransposeWx8_C(src, src_stride, dst, dst_stride, width);
60 TransposeWx8_C((src + 8 * src_stride), src_stride, (dst + 8), dst_stride,
61 width);
62 }
63
TransposeUVWx16_C(const uint8_t * src,int src_stride,uint8_t * dst_a,int dst_stride_a,uint8_t * dst_b,int dst_stride_b,int width)64 void TransposeUVWx16_C(const uint8_t* src,
65 int src_stride,
66 uint8_t* dst_a,
67 int dst_stride_a,
68 uint8_t* dst_b,
69 int dst_stride_b,
70 int width) {
71 TransposeUVWx8_C(src, src_stride, dst_a, dst_stride_a, dst_b, dst_stride_b,
72 width);
73 TransposeUVWx8_C((src + 8 * src_stride), src_stride, (dst_a + 8),
74 dst_stride_a, (dst_b + 8), dst_stride_b, width);
75 }
76
TransposeWx16_MSA(const uint8_t * src,int src_stride,uint8_t * dst,int dst_stride,int width)77 void TransposeWx16_MSA(const uint8_t* src,
78 int src_stride,
79 uint8_t* dst,
80 int dst_stride,
81 int width) {
82 int x;
83 const uint8_t* s;
84 v16u8 src0, src1, src2, src3, dst0, dst1, dst2, dst3, vec0, vec1, vec2, vec3;
85 v16u8 reg0, reg1, reg2, reg3, reg4, reg5, reg6, reg7;
86 v16u8 res0, res1, res2, res3, res4, res5, res6, res7, res8, res9;
87
88 for (x = 0; x < width; x += 16) {
89 s = src;
90 src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
91 s += src_stride;
92 src1 = (v16u8)__msa_ld_b((v16i8*)s, 0);
93 s += src_stride;
94 src2 = (v16u8)__msa_ld_b((v16i8*)s, 0);
95 s += src_stride;
96 src3 = (v16u8)__msa_ld_b((v16i8*)s, 0);
97 s += src_stride;
98 ILVRL_B(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
99 ILVRL_H(vec0, vec2, vec1, vec3, reg0, reg1, reg2, reg3);
100 src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
101 s += src_stride;
102 src1 = (v16u8)__msa_ld_b((v16i8*)s, 0);
103 s += src_stride;
104 src2 = (v16u8)__msa_ld_b((v16i8*)s, 0);
105 s += src_stride;
106 src3 = (v16u8)__msa_ld_b((v16i8*)s, 0);
107 s += src_stride;
108 ILVRL_B(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
109 ILVRL_H(vec0, vec2, vec1, vec3, reg4, reg5, reg6, reg7);
110 ILVRL_W(reg0, reg4, reg1, reg5, res0, res1, res2, res3);
111 ILVRL_W(reg2, reg6, reg3, reg7, res4, res5, res6, res7);
112 src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
113 s += src_stride;
114 src1 = (v16u8)__msa_ld_b((v16i8*)s, 0);
115 s += src_stride;
116 src2 = (v16u8)__msa_ld_b((v16i8*)s, 0);
117 s += src_stride;
118 src3 = (v16u8)__msa_ld_b((v16i8*)s, 0);
119 s += src_stride;
120 ILVRL_B(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
121 ILVRL_H(vec0, vec2, vec1, vec3, reg0, reg1, reg2, reg3);
122 src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
123 s += src_stride;
124 src1 = (v16u8)__msa_ld_b((v16i8*)s, 0);
125 s += src_stride;
126 src2 = (v16u8)__msa_ld_b((v16i8*)s, 0);
127 s += src_stride;
128 src3 = (v16u8)__msa_ld_b((v16i8*)s, 0);
129 s += src_stride;
130 ILVRL_B(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
131 ILVRL_H(vec0, vec2, vec1, vec3, reg4, reg5, reg6, reg7);
132 res8 = (v16u8)__msa_ilvr_w((v4i32)reg4, (v4i32)reg0);
133 res9 = (v16u8)__msa_ilvl_w((v4i32)reg4, (v4i32)reg0);
134 ILVRL_D(res0, res8, res1, res9, dst0, dst1, dst2, dst3);
135 ST_UB4(dst0, dst1, dst2, dst3, dst, dst_stride);
136 dst += dst_stride * 4;
137 res8 = (v16u8)__msa_ilvr_w((v4i32)reg5, (v4i32)reg1);
138 res9 = (v16u8)__msa_ilvl_w((v4i32)reg5, (v4i32)reg1);
139 ILVRL_D(res2, res8, res3, res9, dst0, dst1, dst2, dst3);
140 ST_UB4(dst0, dst1, dst2, dst3, dst, dst_stride);
141 dst += dst_stride * 4;
142 res8 = (v16u8)__msa_ilvr_w((v4i32)reg6, (v4i32)reg2);
143 res9 = (v16u8)__msa_ilvl_w((v4i32)reg6, (v4i32)reg2);
144 ILVRL_D(res4, res8, res5, res9, dst0, dst1, dst2, dst3);
145 ST_UB4(dst0, dst1, dst2, dst3, dst, dst_stride);
146 dst += dst_stride * 4;
147 res8 = (v16u8)__msa_ilvr_w((v4i32)reg7, (v4i32)reg3);
148 res9 = (v16u8)__msa_ilvl_w((v4i32)reg7, (v4i32)reg3);
149 ILVRL_D(res6, res8, res7, res9, dst0, dst1, dst2, dst3);
150 ST_UB4(dst0, dst1, dst2, dst3, dst, dst_stride);
151 src += 16;
152 dst += dst_stride * 4;
153 }
154 }
155
TransposeUVWx16_MSA(const uint8_t * src,int src_stride,uint8_t * dst_a,int dst_stride_a,uint8_t * dst_b,int dst_stride_b,int width)156 void TransposeUVWx16_MSA(const uint8_t* src,
157 int src_stride,
158 uint8_t* dst_a,
159 int dst_stride_a,
160 uint8_t* dst_b,
161 int dst_stride_b,
162 int width) {
163 int x;
164 const uint8_t* s;
165 v16u8 src0, src1, src2, src3, dst0, dst1, dst2, dst3, vec0, vec1, vec2, vec3;
166 v16u8 reg0, reg1, reg2, reg3, reg4, reg5, reg6, reg7;
167 v16u8 res0, res1, res2, res3, res4, res5, res6, res7, res8, res9;
168
169 for (x = 0; x < width; x += 8) {
170 s = src;
171 src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
172 s += src_stride;
173 src1 = (v16u8)__msa_ld_b((v16i8*)s, 0);
174 s += src_stride;
175 src2 = (v16u8)__msa_ld_b((v16i8*)s, 0);
176 s += src_stride;
177 src3 = (v16u8)__msa_ld_b((v16i8*)s, 0);
178 s += src_stride;
179 ILVRL_B(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
180 ILVRL_H(vec0, vec2, vec1, vec3, reg0, reg1, reg2, reg3);
181 src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
182 s += src_stride;
183 src1 = (v16u8)__msa_ld_b((v16i8*)s, 0);
184 s += src_stride;
185 src2 = (v16u8)__msa_ld_b((v16i8*)s, 0);
186 s += src_stride;
187 src3 = (v16u8)__msa_ld_b((v16i8*)s, 0);
188 s += src_stride;
189 ILVRL_B(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
190 ILVRL_H(vec0, vec2, vec1, vec3, reg4, reg5, reg6, reg7);
191 ILVRL_W(reg0, reg4, reg1, reg5, res0, res1, res2, res3);
192 ILVRL_W(reg2, reg6, reg3, reg7, res4, res5, res6, res7);
193 src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
194 s += src_stride;
195 src1 = (v16u8)__msa_ld_b((v16i8*)s, 0);
196 s += src_stride;
197 src2 = (v16u8)__msa_ld_b((v16i8*)s, 0);
198 s += src_stride;
199 src3 = (v16u8)__msa_ld_b((v16i8*)s, 0);
200 s += src_stride;
201 ILVRL_B(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
202 ILVRL_H(vec0, vec2, vec1, vec3, reg0, reg1, reg2, reg3);
203 src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
204 s += src_stride;
205 src1 = (v16u8)__msa_ld_b((v16i8*)s, 0);
206 s += src_stride;
207 src2 = (v16u8)__msa_ld_b((v16i8*)s, 0);
208 s += src_stride;
209 src3 = (v16u8)__msa_ld_b((v16i8*)s, 0);
210 s += src_stride;
211 ILVRL_B(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
212 ILVRL_H(vec0, vec2, vec1, vec3, reg4, reg5, reg6, reg7);
213 res8 = (v16u8)__msa_ilvr_w((v4i32)reg4, (v4i32)reg0);
214 res9 = (v16u8)__msa_ilvl_w((v4i32)reg4, (v4i32)reg0);
215 ILVRL_D(res0, res8, res1, res9, dst0, dst1, dst2, dst3);
216 ST_UB2(dst0, dst2, dst_a, dst_stride_a);
217 ST_UB2(dst1, dst3, dst_b, dst_stride_b);
218 dst_a += dst_stride_a * 2;
219 dst_b += dst_stride_b * 2;
220 res8 = (v16u8)__msa_ilvr_w((v4i32)reg5, (v4i32)reg1);
221 res9 = (v16u8)__msa_ilvl_w((v4i32)reg5, (v4i32)reg1);
222 ILVRL_D(res2, res8, res3, res9, dst0, dst1, dst2, dst3);
223 ST_UB2(dst0, dst2, dst_a, dst_stride_a);
224 ST_UB2(dst1, dst3, dst_b, dst_stride_b);
225 dst_a += dst_stride_a * 2;
226 dst_b += dst_stride_b * 2;
227 res8 = (v16u8)__msa_ilvr_w((v4i32)reg6, (v4i32)reg2);
228 res9 = (v16u8)__msa_ilvl_w((v4i32)reg6, (v4i32)reg2);
229 ILVRL_D(res4, res8, res5, res9, dst0, dst1, dst2, dst3);
230 ST_UB2(dst0, dst2, dst_a, dst_stride_a);
231 ST_UB2(dst1, dst3, dst_b, dst_stride_b);
232 dst_a += dst_stride_a * 2;
233 dst_b += dst_stride_b * 2;
234 res8 = (v16u8)__msa_ilvr_w((v4i32)reg7, (v4i32)reg3);
235 res9 = (v16u8)__msa_ilvl_w((v4i32)reg7, (v4i32)reg3);
236 ILVRL_D(res6, res8, res7, res9, dst0, dst1, dst2, dst3);
237 ST_UB2(dst0, dst2, dst_a, dst_stride_a);
238 ST_UB2(dst1, dst3, dst_b, dst_stride_b);
239 src += 16;
240 dst_a += dst_stride_a * 2;
241 dst_b += dst_stride_b * 2;
242 }
243 }
244
245 #ifdef __cplusplus
246 } // extern "C"
247 } // namespace libyuv
248 #endif
249
250 #endif // !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa)
251