OLD | NEW |
---|---|
1 /* | 1 /* |
2 * Copyright 2016 The LibYuv Project Authors. All rights reserved. | 2 * Copyright 2016 The LibYuv Project Authors. All rights reserved. |
3 * | 3 * |
4 * Use of this source code is governed by a BSD-style license | 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 | 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 | 6 * tree. An additional intellectual property rights grant can be found |
7 * in the file PATENTS. All contributing project authors may | 7 * in the file PATENTS. All contributing project authors may |
8 * be found in the AUTHORS file in the root of the source tree. | 8 * be found in the AUTHORS file in the root of the source tree. |
9 */ | 9 */ |
10 | 10 |
11 #include "libyuv/row.h" | 11 #include "libyuv/row.h" |
12 | 12 |
13 // This module is for GCC MSA | 13 // This module is for GCC MSA |
14 #if !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa) | 14 #if !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa) |
15 #include "libyuv/macros_msa.h" | 15 #include "libyuv/macros_msa.h" |
16 | 16 |
17 #ifdef __cplusplus | 17 #ifdef __cplusplus |
18 namespace libyuv { | 18 namespace libyuv { |
19 extern "C" { | 19 extern "C" { |
20 #endif | 20 #endif |
21 | 21 |
22 // Load YUV 422 pixel data | |
23 #define LOAD_I422(psrc_y, psrc_u, psrc_v, out_y, out_u, out_v) { \ | |
fbarchard1
2016/10/26 17:56:31
suggest READYUV422 as name, for consistency with r
| |
24 uint64 y_m; \ | |
25 uint32 u_m, v_m; \ | |
26 v4i32 zero_m = { 0 }; \ | |
27 y_m = LD(psrc_y); \ | |
28 u_m = LW(psrc_u); \ | |
29 v_m = LW(psrc_v); \ | |
30 out_y = (v16u8) __msa_insert_d((v2i64) zero_m, 0, (int64) y_m); \ | |
31 out_u = (v16u8) __msa_insert_w(zero_m, 0, (int32) u_m); \ | |
32 out_v = (v16u8) __msa_insert_w(zero_m, 0, (int32) v_m); \ | |
33 } | |
34 | |
22 // Convert 8 pixels of YUV 420 to RGB. | 35 // Convert 8 pixels of YUV 420 to RGB. |
23 #define I422TORGB(in0, in1, in2, ub, vr, ug, vg, \ | 36 #define I422TORGB(in_y, in_u, in_v, \ |
24 bb, bg, br, yg, out0, out1, out2) { \ | 37 ub, vr, ug, vg, bb, bg, br, yg, \ |
25 v8i16 vec0_m; \ | 38 out_b, out_g, out_r) { \ |
26 v4i32 reg0_m, reg1_m, reg2_m, reg3_m, reg4_m; \ | 39 v8i16 vec0_m; \ |
27 v4i32 reg5_m, reg6_m, reg7_m, reg8_m, reg9_m; \ | 40 v4i32 reg0_m, reg1_m, reg2_m, reg3_m, reg4_m; \ |
28 v4i32 max_val_m = __msa_ldi_w(255); \ | 41 v4i32 reg5_m, reg6_m, reg7_m, reg8_m, reg9_m; \ |
29 v8i16 zero_m = { 0 }; \ | 42 v4i32 max_val_m = __msa_ldi_w(255); \ |
30 \ | 43 v8i16 zero_m = { 0 }; \ |
31 in1 = (v16u8) __msa_ilvr_b((v16i8) in1, (v16i8) in1); \ | 44 \ |
32 in2 = (v16u8) __msa_ilvr_b((v16i8) in2, (v16i8) in2); \ | 45 in_u = (v16u8) __msa_ilvr_b((v16i8) in_u, (v16i8) in_u); \ |
33 vec0_m = (v8i16) __msa_ilvr_b((v16i8) in0, (v16i8) in0); \ | 46 in_v = (v16u8) __msa_ilvr_b((v16i8) in_v, (v16i8) in_v); \ |
34 reg0_m = (v4i32) __msa_ilvr_h(zero_m, vec0_m); \ | 47 vec0_m = (v8i16) __msa_ilvr_b((v16i8) in_y, (v16i8) in_y); \ |
35 reg1_m = (v4i32) __msa_ilvl_h(zero_m, vec0_m); \ | 48 reg0_m = (v4i32) __msa_ilvr_h(zero_m, vec0_m); \ |
36 reg0_m *= vec_yg; \ | 49 reg1_m = (v4i32) __msa_ilvl_h(zero_m, vec0_m); \ |
37 reg1_m *= vec_yg; \ | 50 reg0_m *= vec_yg; \ |
38 reg0_m = __msa_srai_w(reg0_m, 16); \ | 51 reg1_m *= vec_yg; \ |
39 reg1_m = __msa_srai_w(reg1_m, 16); \ | 52 reg0_m = __msa_srai_w(reg0_m, 16); \ |
40 reg4_m = reg0_m + br; \ | 53 reg1_m = __msa_srai_w(reg1_m, 16); \ |
41 reg5_m = reg1_m + br; \ | 54 reg4_m = reg0_m + br; \ |
42 reg2_m = reg0_m + bg; \ | 55 reg5_m = reg1_m + br; \ |
43 reg3_m = reg1_m + bg; \ | 56 reg2_m = reg0_m + bg; \ |
44 reg0_m += bb; \ | 57 reg3_m = reg1_m + bg; \ |
45 reg1_m += bb; \ | 58 reg0_m += bb; \ |
46 vec0_m = (v8i16) __msa_ilvr_b((v16i8) zero_m, (v16i8) in1); \ | 59 reg1_m += bb; \ |
47 reg6_m = (v4i32) __msa_ilvr_h(zero_m, (v8i16) vec0_m); \ | 60 vec0_m = (v8i16) __msa_ilvr_b((v16i8) zero_m, (v16i8) in_u); \ |
48 reg7_m = (v4i32) __msa_ilvl_h(zero_m, (v8i16) vec0_m); \ | 61 reg6_m = (v4i32) __msa_ilvr_h(zero_m, (v8i16) vec0_m); \ |
49 vec0_m = (v8i16) __msa_ilvr_b((v16i8) zero_m, (v16i8) in2); \ | 62 reg7_m = (v4i32) __msa_ilvl_h(zero_m, (v8i16) vec0_m); \ |
50 reg8_m = (v4i32) __msa_ilvr_h(zero_m, (v8i16) vec0_m); \ | 63 vec0_m = (v8i16) __msa_ilvr_b((v16i8) zero_m, (v16i8) in_v); \ |
51 reg9_m = (v4i32) __msa_ilvl_h(zero_m, (v8i16) vec0_m); \ | 64 reg8_m = (v4i32) __msa_ilvr_h(zero_m, (v8i16) vec0_m); \ |
52 reg0_m -= reg6_m * ub; \ | 65 reg9_m = (v4i32) __msa_ilvl_h(zero_m, (v8i16) vec0_m); \ |
53 reg1_m -= reg7_m * ub; \ | 66 reg0_m -= reg6_m * ub; \ |
54 reg2_m -= reg6_m * ug; \ | 67 reg1_m -= reg7_m * ub; \ |
55 reg3_m -= reg7_m * ug; \ | 68 reg2_m -= reg6_m * ug; \ |
56 reg4_m -= reg8_m * vr; \ | 69 reg3_m -= reg7_m * ug; \ |
57 reg5_m -= reg9_m * vr; \ | 70 reg4_m -= reg8_m * vr; \ |
58 reg2_m -= reg8_m * vg; \ | 71 reg5_m -= reg9_m * vr; \ |
59 reg3_m -= reg9_m * vg; \ | 72 reg2_m -= reg8_m * vg; \ |
60 reg0_m = __msa_srai_w(reg0_m, 6); \ | 73 reg3_m -= reg9_m * vg; \ |
61 reg1_m = __msa_srai_w(reg1_m, 6); \ | 74 reg0_m = __msa_srai_w(reg0_m, 6); \ |
62 reg2_m = __msa_srai_w(reg2_m, 6); \ | 75 reg1_m = __msa_srai_w(reg1_m, 6); \ |
63 reg3_m = __msa_srai_w(reg3_m, 6); \ | 76 reg2_m = __msa_srai_w(reg2_m, 6); \ |
64 reg4_m = __msa_srai_w(reg4_m, 6); \ | 77 reg3_m = __msa_srai_w(reg3_m, 6); \ |
65 reg5_m = __msa_srai_w(reg5_m, 6); \ | 78 reg4_m = __msa_srai_w(reg4_m, 6); \ |
66 reg0_m = __msa_maxi_s_w(reg0_m, 0); \ | 79 reg5_m = __msa_srai_w(reg5_m, 6); \ |
67 reg1_m = __msa_maxi_s_w(reg1_m, 0); \ | 80 reg0_m = __msa_maxi_s_w(reg0_m, 0); \ |
68 reg2_m = __msa_maxi_s_w(reg2_m, 0); \ | 81 reg1_m = __msa_maxi_s_w(reg1_m, 0); \ |
69 reg3_m = __msa_maxi_s_w(reg3_m, 0); \ | 82 reg2_m = __msa_maxi_s_w(reg2_m, 0); \ |
70 reg4_m = __msa_maxi_s_w(reg4_m, 0); \ | 83 reg3_m = __msa_maxi_s_w(reg3_m, 0); \ |
71 reg5_m = __msa_maxi_s_w(reg5_m, 0); \ | 84 reg4_m = __msa_maxi_s_w(reg4_m, 0); \ |
72 reg0_m = __msa_min_s_w(reg0_m, max_val_m); \ | 85 reg5_m = __msa_maxi_s_w(reg5_m, 0); \ |
73 reg1_m = __msa_min_s_w(reg1_m, max_val_m); \ | 86 reg0_m = __msa_min_s_w(reg0_m, max_val_m); \ |
74 reg2_m = __msa_min_s_w(reg2_m, max_val_m); \ | 87 reg1_m = __msa_min_s_w(reg1_m, max_val_m); \ |
75 reg3_m = __msa_min_s_w(reg3_m, max_val_m); \ | 88 reg2_m = __msa_min_s_w(reg2_m, max_val_m); \ |
76 reg4_m = __msa_min_s_w(reg4_m, max_val_m); \ | 89 reg3_m = __msa_min_s_w(reg3_m, max_val_m); \ |
77 reg5_m = __msa_min_s_w(reg5_m, max_val_m); \ | 90 reg4_m = __msa_min_s_w(reg4_m, max_val_m); \ |
78 out0 = __msa_pckev_h((v8i16) reg1_m, (v8i16) reg0_m); \ | 91 reg5_m = __msa_min_s_w(reg5_m, max_val_m); \ |
79 out1 = __msa_pckev_h((v8i16) reg3_m, (v8i16) reg2_m); \ | 92 out_b = __msa_pckev_h((v8i16) reg1_m, (v8i16) reg0_m); \ |
80 out2 = __msa_pckev_h((v8i16) reg5_m, (v8i16) reg4_m); \ | 93 out_g = __msa_pckev_h((v8i16) reg3_m, (v8i16) reg2_m); \ |
94 out_r = __msa_pckev_h((v8i16) reg5_m, (v8i16) reg4_m); \ | |
81 } | 95 } |
82 | 96 |
83 void MirrorRow_MSA(const uint8* src, uint8* dst, int width) { | 97 void MirrorRow_MSA(const uint8* src, uint8* dst, int width) { |
84 int x; | 98 int x; |
85 v16u8 src0, src1, src2, src3; | 99 v16u8 src0, src1, src2, src3; |
86 v16u8 dst0, dst1, dst2, dst3; | 100 v16u8 dst0, dst1, dst2, dst3; |
87 v16i8 shuffler = { 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0 }; | 101 v16i8 shuffler = { 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0 }; |
88 src += width - 64; | 102 src += width - 64; |
89 | 103 |
90 for (x = 0; x < width; x += 64) { | 104 for (x = 0; x < width; x += 64) { |
(...skipping 68 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... | |
159 src_v += 16; | 173 src_v += 16; |
160 src_y += 32; | 174 src_y += 32; |
161 dst_uyvy += 64; | 175 dst_uyvy += 64; |
162 } | 176 } |
163 } | 177 } |
164 | 178 |
165 void I422ToARGBRow_MSA(const uint8* src_y, const uint8* src_u, | 179 void I422ToARGBRow_MSA(const uint8* src_y, const uint8* src_u, |
166 const uint8* src_v, uint8* rgb_buf, | 180 const uint8* src_v, uint8* rgb_buf, |
167 const struct YuvConstants* yuvconstants, int width) { | 181 const struct YuvConstants* yuvconstants, int width) { |
168 int x; | 182 int x; |
169 int32 data_u, data_v; | |
170 int64 data_y; | |
171 v16u8 src0, src1, src2, dst0, dst1; | 183 v16u8 src0, src1, src2, dst0, dst1; |
172 v8i16 vec0, vec1, vec2; | 184 v8i16 vec0, vec1, vec2; |
173 v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg; | 185 v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg; |
174 v16u8 const_255 = (v16u8) __msa_ldi_b(255); | 186 v16u8 const_255 = (v16u8) __msa_ldi_b(255); |
175 v4i32 zero = { 0 }; | |
176 | 187 |
177 vec_ub = __msa_fill_w(yuvconstants->kUVToB[0]); | 188 vec_ub = __msa_fill_w(yuvconstants->kUVToB[0]); |
178 vec_vr = __msa_fill_w(yuvconstants->kUVToR[1]); | 189 vec_vr = __msa_fill_w(yuvconstants->kUVToR[1]); |
179 vec_ug = __msa_fill_w(yuvconstants->kUVToG[0]); | 190 vec_ug = __msa_fill_w(yuvconstants->kUVToG[0]); |
180 vec_vg = __msa_fill_w(yuvconstants->kUVToG[1]); | 191 vec_vg = __msa_fill_w(yuvconstants->kUVToG[1]); |
181 vec_bb = __msa_fill_w(yuvconstants->kUVBiasB[0]); | 192 vec_bb = __msa_fill_w(yuvconstants->kUVBiasB[0]); |
182 vec_bg = __msa_fill_w(yuvconstants->kUVBiasG[0]); | 193 vec_bg = __msa_fill_w(yuvconstants->kUVBiasG[0]); |
183 vec_br = __msa_fill_w(yuvconstants->kUVBiasR[0]); | 194 vec_br = __msa_fill_w(yuvconstants->kUVBiasR[0]); |
184 vec_yg = __msa_fill_w(yuvconstants->kYToRgb[0]); | 195 vec_yg = __msa_fill_w(yuvconstants->kYToRgb[0]); |
185 | 196 |
186 for (x = 0; x < width; x += 8) { | 197 for (x = 0; x < width; x += 8) { |
187 data_y = LD(src_y); | 198 LOAD_I422(src_y, src_u, src_v, src0, src1, src2); |
188 data_u = LW(src_u); | |
189 data_v = LW(src_v); | |
190 src0 = (v16u8) __msa_insert_d((v2i64) zero, 0, data_y); | |
191 src1 = (v16u8) __msa_insert_w(zero, 0, data_u); | |
192 src2 = (v16u8) __msa_insert_w(zero, 0, data_v); | |
193 I422TORGB(src0, src1, src2, vec_ub, vec_vr, vec_ug, vec_vg, | 199 I422TORGB(src0, src1, src2, vec_ub, vec_vr, vec_ug, vec_vg, |
194 vec_bb, vec_bg, vec_br, vec_yg, vec0, vec1, vec2); | 200 vec_bb, vec_bg, vec_br, vec_yg, vec0, vec1, vec2); |
195 vec0 = (v8i16) __msa_ilvev_b((v16i8) vec1, (v16i8) vec0); | 201 vec0 = (v8i16) __msa_ilvev_b((v16i8) vec1, (v16i8) vec0); |
196 vec1 = (v8i16) __msa_ilvev_b((v16i8) const_255, (v16i8) vec2); | 202 vec1 = (v8i16) __msa_ilvev_b((v16i8) const_255, (v16i8) vec2); |
197 dst0 = (v16u8) __msa_ilvr_h((v8i16) vec1, (v8i16) vec0); | 203 dst0 = (v16u8) __msa_ilvr_h((v8i16) vec1, (v8i16) vec0); |
198 dst1 = (v16u8) __msa_ilvl_h((v8i16) vec1, (v8i16) vec0); | 204 dst1 = (v16u8) __msa_ilvl_h((v8i16) vec1, (v8i16) vec0); |
199 ST_UB2(dst0, dst1, rgb_buf, 16); | 205 ST_UB2(dst0, dst1, rgb_buf, 16); |
200 src_y += 8; | 206 src_y += 8; |
201 src_u += 4; | 207 src_u += 4; |
202 src_v += 4; | 208 src_v += 4; |
203 rgb_buf += 32; | 209 rgb_buf += 32; |
204 } | 210 } |
205 } | 211 } |
206 | 212 |
207 void I422ToRGBARow_MSA(const uint8* src_y, const uint8* src_u, | 213 void I422ToRGBARow_MSA(const uint8* src_y, const uint8* src_u, |
208 const uint8* src_v, uint8* rgb_buf, | 214 const uint8* src_v, uint8* rgb_buf, |
209 const struct YuvConstants* yuvconstants, int width) { | 215 const struct YuvConstants* yuvconstants, int width) { |
210 int x; | 216 int x; |
211 int64 data_y; | |
212 int32 data_u, data_v; | |
213 v16u8 src0, src1, src2, dst0, dst1; | 217 v16u8 src0, src1, src2, dst0, dst1; |
214 v8i16 vec0, vec1, vec2; | 218 v8i16 vec0, vec1, vec2; |
215 v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg; | 219 v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg; |
216 v16u8 const_255 = (v16u8) __msa_ldi_b(255); | 220 v16u8 const_255 = (v16u8) __msa_ldi_b(255); |
217 v4i32 zero = { 0 }; | |
218 | 221 |
219 vec_ub = __msa_fill_w(yuvconstants->kUVToB[0]); | 222 vec_ub = __msa_fill_w(yuvconstants->kUVToB[0]); |
220 vec_vr = __msa_fill_w(yuvconstants->kUVToR[1]); | 223 vec_vr = __msa_fill_w(yuvconstants->kUVToR[1]); |
221 vec_ug = __msa_fill_w(yuvconstants->kUVToG[0]); | 224 vec_ug = __msa_fill_w(yuvconstants->kUVToG[0]); |
222 vec_vg = __msa_fill_w(yuvconstants->kUVToG[1]); | 225 vec_vg = __msa_fill_w(yuvconstants->kUVToG[1]); |
223 vec_bb = __msa_fill_w(yuvconstants->kUVBiasB[0]); | 226 vec_bb = __msa_fill_w(yuvconstants->kUVBiasB[0]); |
224 vec_bg = __msa_fill_w(yuvconstants->kUVBiasG[0]); | 227 vec_bg = __msa_fill_w(yuvconstants->kUVBiasG[0]); |
225 vec_br = __msa_fill_w(yuvconstants->kUVBiasR[0]); | 228 vec_br = __msa_fill_w(yuvconstants->kUVBiasR[0]); |
226 vec_yg = __msa_fill_w(yuvconstants->kYToRgb[0]); | 229 vec_yg = __msa_fill_w(yuvconstants->kYToRgb[0]); |
227 | 230 |
228 for (x = 0; x < width; x += 8) { | 231 for (x = 0; x < width; x += 8) { |
229 data_y = LD(src_y); | 232 LOAD_I422(src_y, src_u, src_v, src0, src1, src2); |
230 data_u = LW(src_u); | |
231 data_v = LW(src_v); | |
232 src0 = (v16u8) __msa_insert_d((v2i64) zero, 0, data_y); | |
233 src1 = (v16u8) __msa_insert_w(zero, 0, data_u); | |
234 src2 = (v16u8) __msa_insert_w(zero, 0, data_v); | |
235 I422TORGB(src0, src1, src2, vec_ub, vec_vr, vec_ug, vec_vg, | 233 I422TORGB(src0, src1, src2, vec_ub, vec_vr, vec_ug, vec_vg, |
236 vec_bb, vec_bg, vec_br, vec_yg, vec0, vec1, vec2); | 234 vec_bb, vec_bg, vec_br, vec_yg, vec0, vec1, vec2); |
237 vec0 = (v8i16) __msa_ilvev_b((v16i8) vec0, (v16i8) const_255); | 235 vec0 = (v8i16) __msa_ilvev_b((v16i8) vec0, (v16i8) const_255); |
238 vec1 = (v8i16) __msa_ilvev_b((v16i8) vec2, (v16i8) vec1); | 236 vec1 = (v8i16) __msa_ilvev_b((v16i8) vec2, (v16i8) vec1); |
239 dst0 = (v16u8) __msa_ilvr_h(vec1, vec0); | 237 dst0 = (v16u8) __msa_ilvr_h(vec1, vec0); |
240 dst1 = (v16u8) __msa_ilvl_h(vec1, vec0); | 238 dst1 = (v16u8) __msa_ilvl_h(vec1, vec0); |
241 ST_UB2(dst0, dst1, rgb_buf, 16); | 239 ST_UB2(dst0, dst1, rgb_buf, 16); |
242 src_y += 8; | 240 src_y += 8; |
243 src_u += 4; | 241 src_u += 4; |
244 src_v += 4; | 242 src_v += 4; |
245 rgb_buf += 32; | 243 rgb_buf += 32; |
246 } | 244 } |
247 } | 245 } |
248 | 246 |
247 void I422AlphaToARGBRow_MSA(const uint8* src_y, const uint8* src_u, | |
248 const uint8* src_v, const uint8* src_a, | |
249 uint8* rgb_buf, | |
250 const struct YuvConstants* yuvconstants, | |
251 int width) { | |
252 int x; | |
253 int64 data_a; | |
254 v16u8 src0, src1, src2, src3, dst0, dst1; | |
255 v8i16 vec0, vec1, vec2; | |
256 v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg; | |
257 v4i32 zero = { 0 }; | |
258 | |
259 vec_ub = __msa_fill_w(yuvconstants->kUVToB[0]); | |
260 vec_vr = __msa_fill_w(yuvconstants->kUVToR[1]); | |
261 vec_ug = __msa_fill_w(yuvconstants->kUVToG[0]); | |
262 vec_vg = __msa_fill_w(yuvconstants->kUVToG[1]); | |
263 vec_bb = __msa_fill_w(yuvconstants->kUVBiasB[0]); | |
264 vec_bg = __msa_fill_w(yuvconstants->kUVBiasG[0]); | |
265 vec_br = __msa_fill_w(yuvconstants->kUVBiasR[0]); | |
266 vec_yg = __msa_fill_w(yuvconstants->kYToRgb[0]); | |
267 | |
268 for (x = 0; x < width; x += 8) { | |
269 data_a = LD(src_a); | |
270 LOAD_I422(src_y, src_u, src_v, src0, src1, src2); | |
271 src3 = (v16u8) __msa_insert_d((v2i64) zero, 0, data_a); | |
272 I422TORGB(src0, src1, src2, vec_ub, vec_vr, vec_ug, vec_vg, | |
273 vec_bb, vec_bg, vec_br, vec_yg, vec0, vec1, vec2); | |
274 src3 = (v16u8) __msa_ilvr_b((v16i8) src3, (v16i8) src3); | |
fbarchard1
2016/10/26 17:56:31
consider a macro for STOREARGB since it will come
| |
275 vec0 = (v8i16) __msa_ilvev_b((v16i8) vec1, (v16i8) vec0); | |
276 vec1 = (v8i16) __msa_ilvev_b((v16i8) src3, (v16i8) vec2); | |
277 dst0 = (v16u8) __msa_ilvr_h((v8i16) vec1, (v8i16) vec0); | |
278 dst1 = (v16u8) __msa_ilvl_h((v8i16) vec1, (v8i16) vec0); | |
279 ST_UB2(dst0, dst1, rgb_buf, 16); | |
280 src_y += 8; | |
281 src_u += 4; | |
282 src_v += 4; | |
283 src_a += 8; | |
284 rgb_buf += 32; | |
285 } | |
286 } | |
287 | |
288 void I422ToRGB24Row_MSA(const uint8* src_y, const uint8* src_u, | |
289 const uint8* src_v, uint8* rgb_buf, | |
290 const struct YuvConstants* yuvconstants, int32 width) { | |
291 int x; | |
292 int64 data_u, data_v; | |
293 v16u8 src0, src1, src2, src3, src4, src5, dst0, dst1, dst2; | |
294 v8i16 vec0, vec1, vec2, vec3, vec4, vec5; | |
295 v4i32 vec_ub, vec_vr, vec_ug, vec_vg, vec_bb, vec_bg, vec_br, vec_yg; | |
296 v16u8 reg0, reg1, reg2, reg3; | |
297 v2i64 zero = { 0 }; | |
298 v16i8 shuffler0 = { 0, 1, 16, 2, 3, 17, 4, 5, 18, 6, 7, 19, 8, 9, 20, 10 }; | |
299 v16i8 shuffler1 = { 0, 21, 1, 2, 22, 3, 4, 23, 5, 6, 24, 7, 8, 25, 9, 10 }; | |
300 v16i8 shuffler2 = | |
301 { 26, 6, 7, 27, 8, 9, 28, 10, 11, 29, 12, 13, 30, 14, 15, 31 }; | |
fbarchard1
2016/10/26 17:56:31
indent should be 4 from line above.
| |
302 | |
303 vec_ub = __msa_fill_w(yuvconstants->kUVToB[0]); | |
304 vec_vr = __msa_fill_w(yuvconstants->kUVToR[1]); | |
305 vec_ug = __msa_fill_w(yuvconstants->kUVToG[0]); | |
306 vec_vg = __msa_fill_w(yuvconstants->kUVToG[1]); | |
307 vec_bb = __msa_fill_w(yuvconstants->kUVBiasB[0]); | |
308 vec_bg = __msa_fill_w(yuvconstants->kUVBiasG[0]); | |
309 vec_br = __msa_fill_w(yuvconstants->kUVBiasR[0]); | |
310 vec_yg = __msa_fill_w(yuvconstants->kYToRgb[0]); | |
311 | |
312 for (x = 0; x < width; x += 16) { | |
313 src0 = (v16u8) __msa_ld_b((v16u8*) src_y, 0); | |
314 data_u = LD(src_u); | |
315 data_v = LD(src_v); | |
316 src1 = (v16u8) __msa_insert_d(zero, 0, data_u); | |
317 src2 = (v16u8) __msa_insert_d(zero, 0, data_v); | |
318 src3 = (v16u8) __msa_sldi_b((v16i8) src0, (v16i8) src0, 8); | |
319 src4 = (v16u8) __msa_sldi_b((v16i8) src1, (v16i8) src1, 4); | |
320 src5 = (v16u8) __msa_sldi_b((v16i8) src2, (v16i8) src2, 4); | |
321 I422TORGB(src0, src1, src2, vec_ub, vec_vr, vec_ug, vec_vg, | |
322 vec_bb, vec_bg, vec_br, vec_yg, vec0, vec1, vec2); | |
323 I422TORGB(src3, src4, src5, vec_ub, vec_vr, vec_ug, vec_vg, | |
324 vec_bb, vec_bg, vec_br, vec_yg, vec3, vec4, vec5); | |
325 reg0 = (v16u8) __msa_ilvev_b((v16i8) vec1, (v16i8) vec0); | |
326 reg2 = (v16u8) __msa_ilvev_b((v16i8) vec4, (v16i8) vec3); | |
327 reg3 = (v16u8) __msa_pckev_b((v16i8) vec5, (v16i8) vec2); | |
328 reg1 = (v16u8) __msa_sldi_b((v16i8) reg2, (v16i8) reg0, 11); | |
329 dst0 = (v16u8) __msa_vshf_b(shuffler0, (v16i8) reg3, (v16i8) reg0); | |
330 dst1 = (v16u8) __msa_vshf_b(shuffler1, (v16i8) reg3, (v16i8) reg1); | |
331 dst2 = (v16u8) __msa_vshf_b(shuffler2, (v16i8) reg3, (v16i8) reg2); | |
332 ST_UB2(dst0, dst1, rgb_buf, 16); | |
333 ST_UB(dst2, (rgb_buf + 32)); | |
334 src_y += 16; | |
335 src_u += 8; | |
336 src_v += 8; | |
337 rgb_buf += 48; | |
338 } | |
339 } | |
340 | |
249 void YUY2ToYRow_MSA(const uint8* src_yuy2, uint8* dst_y, int width) { | 341 void YUY2ToYRow_MSA(const uint8* src_yuy2, uint8* dst_y, int width) { |
250 int x; | 342 int x; |
251 v16u8 src0, src1, src2, src3, dst0, dst1; | 343 v16u8 src0, src1, src2, src3, dst0, dst1; |
252 | 344 |
253 for (x = 0; x < width; x += 32) { | 345 for (x = 0; x < width; x += 32) { |
254 LD_UB4(src_yuy2, 16, src0, src1, src2, src3); | 346 LD_UB4(src_yuy2, 16, src0, src1, src2, src3); |
255 dst0 = (v16u8) __msa_pckev_b((v16i8) src1, (v16i8) src0); | 347 dst0 = (v16u8) __msa_pckev_b((v16i8) src1, (v16i8) src0); |
256 dst1 = (v16u8) __msa_pckev_b((v16i8) src3, (v16i8) src2); | 348 dst1 = (v16u8) __msa_pckev_b((v16i8) src3, (v16i8) src2); |
257 ST_UB2(dst0, dst1, dst_y, 16); | 349 ST_UB2(dst0, dst1, dst_y, 16); |
258 src_yuy2 += 64; | 350 src_yuy2 += 64; |
(...skipping 295 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... | |
554 dst_argb += 64; | 646 dst_argb += 64; |
555 } | 647 } |
556 } | 648 } |
557 | 649 |
558 #ifdef __cplusplus | 650 #ifdef __cplusplus |
559 } // extern "C" | 651 } // extern "C" |
560 } // namespace libyuv | 652 } // namespace libyuv |
561 #endif | 653 #endif |
562 | 654 |
563 #endif // !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa) | 655 #endif // !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa) |
OLD | NEW |