Details | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
6147 | serge | 1 | /* |
2 | * Copyright (c) 2002 Brian Foley |
||
3 | * Copyright (c) 2002 Dieter Shirley |
||
4 | * Copyright (c) 2003-2004 Romain Dolbeau |
||
5 | * |
||
6 | * This file is part of FFmpeg. |
||
7 | * |
||
8 | * FFmpeg is free software; you can redistribute it and/or |
||
9 | * modify it under the terms of the GNU Lesser General Public |
||
10 | * License as published by the Free Software Foundation; either |
||
11 | * version 2.1 of the License, or (at your option) any later version. |
||
12 | * |
||
13 | * FFmpeg is distributed in the hope that it will be useful, |
||
14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
||
15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
||
16 | * Lesser General Public License for more details. |
||
17 | * |
||
18 | * You should have received a copy of the GNU Lesser General Public |
||
19 | * License along with FFmpeg; if not, write to the Free Software |
||
20 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
||
21 | */ |
||
22 | |||
23 | #include "config.h" |
||
24 | #if HAVE_ALTIVEC_H |
||
25 | #include |
||
26 | #endif |
||
27 | |||
28 | #include "libavutil/attributes.h" |
||
29 | #include "libavutil/cpu.h" |
||
30 | #include "libavutil/ppc/cpu.h" |
||
31 | #include "libavutil/ppc/types_altivec.h" |
||
32 | #include "libavutil/ppc/util_altivec.h" |
||
33 | #include "libavcodec/avcodec.h" |
||
34 | #include "libavcodec/pixblockdsp.h" |
||
35 | |||
36 | #if HAVE_ALTIVEC |
||
37 | |||
38 | #if HAVE_VSX |
||
39 | static void get_pixels_altivec(int16_t *restrict block, const uint8_t *pixels, |
||
40 | ptrdiff_t line_size) |
||
41 | { |
||
42 | int i; |
||
43 | vector unsigned char perm = |
||
44 | (vector unsigned char) {0x00,0x10, 0x01,0x11,0x02,0x12,0x03,0x13,\ |
||
45 | 0x04,0x14,0x05,0x15,0x06,0x16,0x07,0x17}; |
||
46 | const vector unsigned char zero = |
||
47 | (const vector unsigned char) vec_splat_u8(0); |
||
48 | |||
49 | for (i = 0; i < 8; i++) { |
||
50 | /* Read potentially unaligned pixels. |
||
51 | * We're reading 16 pixels, and actually only want 8, |
||
52 | * but we simply ignore the extras. */ |
||
53 | vector unsigned char bytes = vec_vsx_ld(0, pixels); |
||
54 | |||
55 | // Convert the bytes into shorts. |
||
56 | //vector signed short shorts = (vector signed short) vec_perm(zero, bytes, perm); |
||
57 | vector signed short shorts = (vector signed short) vec_perm(bytes, zero, perm); |
||
58 | |||
59 | // Save the data to the block, we assume the block is 16-byte aligned. |
||
60 | vec_vsx_st(shorts, i * 16, (vector signed short *) block); |
||
61 | |||
62 | pixels += line_size; |
||
63 | } |
||
64 | } |
||
65 | #else |
||
66 | static void get_pixels_altivec(int16_t *restrict block, const uint8_t *pixels, |
||
67 | ptrdiff_t line_size) |
||
68 | { |
||
69 | int i; |
||
70 | vec_u8 perm = vec_lvsl(0, pixels); |
||
71 | const vec_u8 zero = (const vec_u8)vec_splat_u8(0); |
||
72 | |||
73 | for (i = 0; i < 8; i++) { |
||
74 | /* Read potentially unaligned pixels. |
||
75 | * We're reading 16 pixels, and actually only want 8, |
||
76 | * but we simply ignore the extras. */ |
||
77 | vec_u8 pixl = vec_ld(0, pixels); |
||
78 | vec_u8 pixr = vec_ld(7, pixels); |
||
79 | vec_u8 bytes = vec_perm(pixl, pixr, perm); |
||
80 | |||
81 | // Convert the bytes into shorts. |
||
82 | vec_s16 shorts = (vec_s16)vec_mergeh(zero, bytes); |
||
83 | |||
84 | // Save the data to the block, we assume the block is 16-byte aligned. |
||
85 | vec_st(shorts, i * 16, (vec_s16 *)block); |
||
86 | |||
87 | pixels += line_size; |
||
88 | } |
||
89 | } |
||
90 | |||
91 | #endif /* HAVE_VSX */ |
||
92 | |||
93 | #if HAVE_VSX |
||
94 | static void diff_pixels_altivec(int16_t *restrict block, const uint8_t *s1, |
||
95 | const uint8_t *s2, int stride) |
||
96 | { |
||
97 | int i; |
||
98 | const vector unsigned char zero = |
||
99 | (const vector unsigned char) vec_splat_u8(0); |
||
100 | vector signed short shorts1, shorts2; |
||
101 | |||
102 | for (i = 0; i < 4; i++) { |
||
103 | /* Read potentially unaligned pixels. |
||
104 | * We're reading 16 pixels, and actually only want 8, |
||
105 | * but we simply ignore the extras. */ |
||
106 | vector unsigned char bytes = vec_vsx_ld(0, s1); |
||
107 | |||
108 | // Convert the bytes into shorts. |
||
109 | shorts1 = (vector signed short) vec_mergeh(bytes, zero); |
||
110 | |||
111 | // Do the same for the second block of pixels. |
||
112 | bytes =vec_vsx_ld(0, s2); |
||
113 | |||
114 | // Convert the bytes into shorts. |
||
115 | shorts2 = (vector signed short) vec_mergeh(bytes, zero); |
||
116 | |||
117 | // Do the subtraction. |
||
118 | shorts1 = vec_sub(shorts1, shorts2); |
||
119 | |||
120 | // Save the data to the block, we assume the block is 16-byte aligned. |
||
121 | vec_vsx_st(shorts1, 0, (vector signed short *) block); |
||
122 | |||
123 | s1 += stride; |
||
124 | s2 += stride; |
||
125 | block += 8; |
||
126 | |||
127 | /* The code below is a copy of the code above... |
||
128 | * This is a manual unroll. */ |
||
129 | |||
130 | /* Read potentially unaligned pixels. |
||
131 | * We're reading 16 pixels, and actually only want 8, |
||
132 | * but we simply ignore the extras. */ |
||
133 | bytes = vec_vsx_ld(0, s1); |
||
134 | |||
135 | // Convert the bytes into shorts. |
||
136 | shorts1 = (vector signed short) vec_mergeh(bytes, zero); |
||
137 | |||
138 | // Do the same for the second block of pixels. |
||
139 | bytes = vec_vsx_ld(0, s2); |
||
140 | |||
141 | // Convert the bytes into shorts. |
||
142 | shorts2 = (vector signed short) vec_mergeh(bytes, zero); |
||
143 | |||
144 | // Do the subtraction. |
||
145 | shorts1 = vec_sub(shorts1, shorts2); |
||
146 | |||
147 | // Save the data to the block, we assume the block is 16-byte aligned. |
||
148 | vec_vsx_st(shorts1, 0, (vector signed short *) block); |
||
149 | |||
150 | s1 += stride; |
||
151 | s2 += stride; |
||
152 | block += 8; |
||
153 | } |
||
154 | } |
||
155 | #else |
||
156 | static void diff_pixels_altivec(int16_t *restrict block, const uint8_t *s1, |
||
157 | const uint8_t *s2, int stride) |
||
158 | { |
||
159 | int i; |
||
160 | vec_u8 perm1 = vec_lvsl(0, s1); |
||
161 | vec_u8 perm2 = vec_lvsl(0, s2); |
||
162 | const vec_u8 zero = (const vec_u8)vec_splat_u8(0); |
||
163 | vec_s16 shorts1, shorts2; |
||
164 | |||
165 | for (i = 0; i < 4; i++) { |
||
166 | /* Read potentially unaligned pixels. |
||
167 | * We're reading 16 pixels, and actually only want 8, |
||
168 | * but we simply ignore the extras. */ |
||
169 | vec_u8 pixl = vec_ld(0, s1); |
||
170 | vec_u8 pixr = vec_ld(15, s1); |
||
171 | vec_u8 bytes = vec_perm(pixl, pixr, perm1); |
||
172 | |||
173 | // Convert the bytes into shorts. |
||
174 | shorts1 = (vec_s16)vec_mergeh(zero, bytes); |
||
175 | |||
176 | // Do the same for the second block of pixels. |
||
177 | pixl = vec_ld(0, s2); |
||
178 | pixr = vec_ld(15, s2); |
||
179 | bytes = vec_perm(pixl, pixr, perm2); |
||
180 | |||
181 | // Convert the bytes into shorts. |
||
182 | shorts2 = (vec_s16)vec_mergeh(zero, bytes); |
||
183 | |||
184 | // Do the subtraction. |
||
185 | shorts1 = vec_sub(shorts1, shorts2); |
||
186 | |||
187 | // Save the data to the block, we assume the block is 16-byte aligned. |
||
188 | vec_st(shorts1, 0, (vec_s16 *)block); |
||
189 | |||
190 | s1 += stride; |
||
191 | s2 += stride; |
||
192 | block += 8; |
||
193 | |||
194 | /* The code below is a copy of the code above... |
||
195 | * This is a manual unroll. */ |
||
196 | |||
197 | /* Read potentially unaligned pixels. |
||
198 | * We're reading 16 pixels, and actually only want 8, |
||
199 | * but we simply ignore the extras. */ |
||
200 | pixl = vec_ld(0, s1); |
||
201 | pixr = vec_ld(15, s1); |
||
202 | bytes = vec_perm(pixl, pixr, perm1); |
||
203 | |||
204 | // Convert the bytes into shorts. |
||
205 | shorts1 = (vec_s16)vec_mergeh(zero, bytes); |
||
206 | |||
207 | // Do the same for the second block of pixels. |
||
208 | pixl = vec_ld(0, s2); |
||
209 | pixr = vec_ld(15, s2); |
||
210 | bytes = vec_perm(pixl, pixr, perm2); |
||
211 | |||
212 | // Convert the bytes into shorts. |
||
213 | shorts2 = (vec_s16)vec_mergeh(zero, bytes); |
||
214 | |||
215 | // Do the subtraction. |
||
216 | shorts1 = vec_sub(shorts1, shorts2); |
||
217 | |||
218 | // Save the data to the block, we assume the block is 16-byte aligned. |
||
219 | vec_st(shorts1, 0, (vec_s16 *)block); |
||
220 | |||
221 | s1 += stride; |
||
222 | s2 += stride; |
||
223 | block += 8; |
||
224 | } |
||
225 | } |
||
226 | |||
227 | #endif /* HAVE_VSX */ |
||
228 | |||
229 | #endif /* HAVE_ALTIVEC */ |
||
230 | |||
231 | #if HAVE_VSX |
||
232 | static void get_pixels_vsx(int16_t *restrict block, const uint8_t *pixels, |
||
233 | ptrdiff_t line_size) |
||
234 | { |
||
235 | int i; |
||
236 | for (i = 0; i < 8; i++) { |
||
237 | vec_s16 shorts = vsx_ld_u8_s16(0, pixels); |
||
238 | |||
239 | vec_vsx_st(shorts, i * 16, block); |
||
240 | |||
241 | pixels += line_size; |
||
242 | } |
||
243 | } |
||
244 | |||
245 | static void diff_pixels_vsx(int16_t *restrict block, const uint8_t *s1, |
||
246 | const uint8_t *s2, int stride) |
||
247 | { |
||
248 | int i; |
||
249 | vec_s16 shorts1, shorts2; |
||
250 | for (i = 0; i < 8; i++) { |
||
251 | shorts1 = vsx_ld_u8_s16(0, s1); |
||
252 | shorts2 = vsx_ld_u8_s16(0, s2); |
||
253 | |||
254 | shorts1 = vec_sub(shorts1, shorts2); |
||
255 | |||
256 | vec_vsx_st(shorts1, 0, block); |
||
257 | |||
258 | s1 += stride; |
||
259 | s2 += stride; |
||
260 | block += 8; |
||
261 | } |
||
262 | } |
||
263 | #endif /* HAVE_VSX */ |
||
264 | |||
265 | av_cold void ff_pixblockdsp_init_ppc(PixblockDSPContext *c, |
||
266 | AVCodecContext *avctx, |
||
267 | unsigned high_bit_depth) |
||
268 | { |
||
269 | #if HAVE_ALTIVEC |
||
270 | if (!PPC_ALTIVEC(av_get_cpu_flags())) |
||
271 | return; |
||
272 | |||
273 | c->diff_pixels = diff_pixels_altivec; |
||
274 | |||
275 | if (!high_bit_depth) { |
||
276 | c->get_pixels = get_pixels_altivec; |
||
277 | } |
||
278 | #endif /* HAVE_ALTIVEC */ |
||
279 | |||
280 | #if HAVE_VSX |
||
281 | if (!PPC_VSX(av_get_cpu_flags())) |
||
282 | return; |
||
283 | |||
284 | c->diff_pixels = diff_pixels_vsx; |
||
285 | |||
286 | if (!high_bit_depth) |
||
287 | c->get_pixels = get_pixels_vsx; |
||
288 | #endif /* HAVE_VSX */ |
||
289 | }>>>>>> |