SSE optimized 32-point DCT
[libav.git] / libavcodec / x86 / dct32_sse.c
1 /*
2 * 32 point SSE-optimized DCT transform
3 * Copyright (c) 2010 Vitor Sessak
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 #include <stdint.h>
23
24 #include "libavutil/x86_cpu.h"
25 #include "libavutil/mem.h"
26 #include "libavcodec/dsputil.h"
27 #include "fft.h"
28
29 DECLARE_ALIGNED(16, static const float, b1)[] = {
30 0.500603, 0.505471, 0.515447, 0.531043,
31 0.553104, 0.582935, 0.622504, 0.674808,
32 -1.169440, -0.972568, -0.839350, -0.744536,
33 -10.190008, -3.407609, -2.057781, -1.484165,
34 0.502419, 0.522499, 0.566944, 0.646822,
35 0.788155, 1.060678, 1.722447, 5.101149,
36 0.509796, 0.601345, 0.899976, 2.562916,
37 1.000000, 1.000000, 1.306563, 0.541196,
38 1.000000, 0.707107, 1.000000, -0.707107
39 };
40
41 DECLARE_ALIGNED(16, static const int32_t, smask)[4] = {
42 0, 0, 0x80000000, 0x80000000
43 };
44
45 /* butterfly operator */
46 #define BUTTERFLY(a,b,c,tmp) \
47 "movaps %%" #a ", %%" #tmp " \n\t" \
48 "subps %%" #b ", %%" #a " \n\t" \
49 "addps %%" #tmp ", %%" #b " \n\t" \
50 "mulps " #c ", %%" #a " \n\t"
51
52 ///* Same as BUTTERFLY when vectors a and b overlap */
53 #define BUTTERFLY0(val, mask, cos, tmp, shuf) \
54 "movaps %%" #val ", %%" #tmp " \n\t" \
55 "shufps " #shuf ", %%" #val ",%%" #val " \n\t" \
56 "xorps %%" #mask ", %%" #tmp " \n\t" /* flip signs */ \
57 "addps %%" #tmp ", %%" #val " \n\t" \
58 "mulps %%" #cos ", %%" #val " \n\t"
59
60 #define BUTTERFLY2(val, mask, cos, tmp) BUTTERFLY0(val, mask, cos, tmp, $0x1b)
61 #define BUTTERFLY3(val, mask, cos, tmp) BUTTERFLY0(val, mask, cos, tmp, $0xb1)
62
63 void ff_dct32_float_sse(FFTSample *out, const FFTSample *in)
64 {
65 int32_t tmp1 = 0;
66 __asm__ volatile(
67 /* pass 1 */
68
69 "movaps (%4), %%xmm0 \n\t"
70 "movaps 112(%4), %%xmm1 \n\t"
71 "shufps $0x1b, %%xmm1, %%xmm1 \n\t"
72 BUTTERFLY(xmm0, xmm1, (%2), xmm3)
73
74 "movaps 64(%4), %%xmm7 \n\t"
75 "movaps 48(%4), %%xmm4 \n\t"
76 "shufps $0x1b, %%xmm4, %%xmm4 \n\t"
77 BUTTERFLY(xmm7, xmm4, 48(%2), xmm3)
78
79
80 /* pass 2 */
81 "movaps 64(%2), %%xmm2 \n\t"
82 BUTTERFLY(xmm1, xmm4, %%xmm2, xmm3)
83 "movaps %%xmm1, 48(%1) \n\t"
84 "movaps %%xmm4, (%1) \n\t"
85
86 /* pass 1 */
87 "movaps 16(%4), %%xmm1 \n\t"
88 "movaps 96(%4), %%xmm6 \n\t"
89 "shufps $0x1b, %%xmm6, %%xmm6 \n\t"
90 BUTTERFLY(xmm1, xmm6, 16(%2), xmm3)
91
92 "movaps 80(%4), %%xmm4 \n\t"
93 "movaps 32(%4), %%xmm5 \n\t"
94 "shufps $0x1b, %%xmm5, %%xmm5 \n\t"
95 BUTTERFLY(xmm4, xmm5, 32(%2), xmm3)
96
97 /* pass 2 */
98 BUTTERFLY(xmm0, xmm7, %%xmm2, xmm3)
99
100 "movaps 80(%2), %%xmm2 \n\t"
101 BUTTERFLY(xmm6, xmm5, %%xmm2, xmm3)
102
103 BUTTERFLY(xmm1, xmm4, %%xmm2, xmm3)
104
105 /* pass 3 */
106 "movaps 96(%2), %%xmm2 \n\t"
107 "shufps $0x1b, %%xmm1, %%xmm1 \n\t"
108 BUTTERFLY(xmm0, xmm1, %%xmm2, xmm3)
109 "movaps %%xmm0, 112(%1) \n\t"
110 "movaps %%xmm1, 96(%1) \n\t"
111
112 "movaps 0(%1), %%xmm0 \n\t"
113 "shufps $0x1b, %%xmm5, %%xmm5 \n\t"
114 BUTTERFLY(xmm0, xmm5, %%xmm2, xmm3)
115
116 "movaps 48(%1), %%xmm1 \n\t"
117 "shufps $0x1b, %%xmm6, %%xmm6 \n\t"
118 BUTTERFLY(xmm1, xmm6, %%xmm2, xmm3)
119 "movaps %%xmm1, 48(%1) \n\t"
120
121 "shufps $0x1b, %%xmm4, %%xmm4 \n\t"
122 BUTTERFLY(xmm7, xmm4, %%xmm2, xmm3)
123
124 /* pass 4 */
125 "movaps (%3), %%xmm3 \n\t"
126 "movaps 112(%2), %%xmm2 \n\t"
127
128 BUTTERFLY2(xmm5, xmm3, xmm2, xmm1)
129
130 BUTTERFLY2(xmm0, xmm3, xmm2, xmm1)
131 "movaps %%xmm0, 16(%1) \n\t"
132
133 BUTTERFLY2(xmm6, xmm3, xmm2, xmm1)
134 "movaps %%xmm6, 32(%1) \n\t"
135
136 "movaps 48(%1), %%xmm0 \n\t"
137 BUTTERFLY2(xmm0, xmm3, xmm2, xmm1)
138 "movaps %%xmm0, 48(%1) \n\t"
139
140 BUTTERFLY2(xmm4, xmm3, xmm2, xmm1)
141
142 BUTTERFLY2(xmm7, xmm3, xmm2, xmm1)
143
144 "movaps 96(%1), %%xmm6 \n\t"
145 BUTTERFLY2(xmm6, xmm3, xmm2, xmm1)
146
147 "movaps 112(%1), %%xmm0 \n\t"
148 BUTTERFLY2(xmm0, xmm3, xmm2, xmm1)
149
150 /* pass 5 */
151 "movaps 128(%2), %%xmm2 \n\t"
152 "shufps $0xCC, %%xmm3,%%xmm3 \n\t"
153
154 BUTTERFLY3(xmm5, xmm3, xmm2, xmm1)
155 "movaps %%xmm5, (%1) \n\t"
156
157 "movaps 16(%1), %%xmm1 \n\t"
158 BUTTERFLY3(xmm1, xmm3, xmm2, xmm5)
159 "movaps %%xmm1, 16(%1) \n\t"
160
161 BUTTERFLY3(xmm4, xmm3, xmm2, xmm5)
162 "movaps %%xmm4, 64(%1) \n\t"
163
164 BUTTERFLY3(xmm7, xmm3, xmm2, xmm5)
165 "movaps %%xmm7, 80(%1) \n\t"
166
167 "movaps 32(%1), %%xmm5 \n\t"
168 BUTTERFLY3(xmm5, xmm3, xmm2, xmm7)
169 "movaps %%xmm5, 32(%1) \n\t"
170
171 "movaps 48(%1), %%xmm4 \n\t"
172 BUTTERFLY3(xmm4, xmm3, xmm2, xmm7)
173 "movaps %%xmm4, 48(%1) \n\t"
174
175 BUTTERFLY3(xmm6, xmm3, xmm2, xmm7)
176 "movaps %%xmm6, 96(%1) \n\t"
177
178 BUTTERFLY3(xmm0, xmm3, xmm2, xmm7)
179 "movaps %%xmm0, 112(%1) \n\t"
180
181
182 /* pass 6, no SIMD... */
183 "movss 56(%1), %%xmm3 \n\t"
184 "movl 4(%1), %0 \n\t"
185 "addss 60(%1), %%xmm3 \n\t"
186 "movss 72(%1), %%xmm7 \n\t"
187 "addss %%xmm3, %%xmm4 \n\t"
188 "movss 52(%1), %%xmm2 \n\t"
189 "addss %%xmm3, %%xmm2 \n\t"
190 "movss 24(%1), %%xmm3 \n\t"
191 "addss 28(%1), %%xmm3 \n\t"
192 "addss 76(%1), %%xmm7 \n\t"
193 "addss %%xmm3, %%xmm1 \n\t"
194 "addss %%xmm4, %%xmm5 \n\t"
195 "movss %%xmm1, 16(%1) \n\t"
196 "movss 20(%1), %%xmm1 \n\t"
197 "addss %%xmm3, %%xmm1 \n\t"
198 "movss 40(%1), %%xmm3 \n\t"
199 "movss %%xmm1, 48(%1) \n\t"
200 "addss 44(%1), %%xmm3 \n\t"
201 "movss 20(%1), %%xmm1 \n\t"
202 "addss %%xmm3, %%xmm4 \n\t"
203 "addss %%xmm2, %%xmm3 \n\t"
204 "addss 28(%1), %%xmm1 \n\t"
205 "movss %%xmm3, 40(%1) \n\t"
206 "addss 36(%1), %%xmm2 \n\t"
207 "movss 8(%1), %%xmm3 \n\t"
208 "movss %%xmm2, 56(%1) \n\t"
209 "addss 12(%1), %%xmm3 \n\t"
210 "movss %%xmm5, 8(%1) \n\t"
211 "movss %%xmm3, 32(%1) \n\t"
212 "movss 52(%1), %%xmm2 \n\t"
213 "movss 80(%1), %%xmm3 \n\t"
214 "movss 120(%1), %%xmm5 \n\t"
215 "movss %%xmm1, 80(%1) \n\t"
216 "movss %%xmm4, 24(%1) \n\t"
217 "addss 124(%1), %%xmm5 \n\t"
218 "movss 64(%1), %%xmm1 \n\t"
219 "addss 60(%1), %%xmm2 \n\t"
220 "addss %%xmm5, %%xmm0 \n\t"
221 "addss 116(%1), %%xmm5 \n\t"
222 "movl %0, 64(%1) \n\t"
223 "addss %%xmm0, %%xmm6 \n\t"
224 "addss %%xmm6, %%xmm1 \n\t"
225 "movl 12(%1), %0 \n\t"
226 "movss %%xmm1, 4(%1) \n\t"
227 "movss 88(%1), %%xmm1 \n\t"
228 "movl %0, 96(%1) \n\t"
229 "addss 92(%1), %%xmm1 \n\t"
230 "movss 104(%1), %%xmm4 \n\t"
231 "movl 28(%1), %0 \n\t"
232 "addss 108(%1), %%xmm4 \n\t"
233 "addss %%xmm4, %%xmm0 \n\t"
234 "addss %%xmm1, %%xmm3 \n\t"
235 "addss 84(%1), %%xmm1 \n\t"
236 "addss %%xmm5, %%xmm4 \n\t"
237 "addss %%xmm3, %%xmm6 \n\t"
238 "addss %%xmm0, %%xmm3 \n\t"
239 "addss %%xmm7, %%xmm0 \n\t"
240 "addss 100(%1), %%xmm5 \n\t"
241 "addss %%xmm4, %%xmm7 \n\t"
242 "movl %0, 112(%1) \n\t"
243 "movss %%xmm0, 28(%1) \n\t"
244 "movss 36(%1), %%xmm0 \n\t"
245 "movss %%xmm7, 36(%1) \n\t"
246 "addss %%xmm1, %%xmm4 \n\t"
247 "movss 116(%1), %%xmm7 \n\t"
248 "addss %%xmm2, %%xmm0 \n\t"
249 "addss 124(%1), %%xmm7 \n\t"
250 "movss %%xmm0, 72(%1) \n\t"
251 "movss 44(%1), %%xmm0 \n\t"
252 "movss %%xmm6, 12(%1) \n\t"
253 "movss %%xmm3, 20(%1) \n\t"
254 "addss %%xmm0, %%xmm2 \n\t"
255 "movss %%xmm4, 44(%1) \n\t"
256 "movss %%xmm2, 88(%1) \n\t"
257 "addss 60(%1), %%xmm0 \n\t"
258 "movl 60(%1), %0 \n\t"
259 "movl %0, 120(%1) \n\t"
260 "movss %%xmm0, 104(%1) \n\t"
261 "addss %%xmm5, %%xmm1 \n\t"
262 "addss 68(%1), %%xmm5 \n\t"
263 "movss %%xmm1, 52(%1) \n\t"
264 "movss %%xmm5, 60(%1) \n\t"
265 "movss 68(%1), %%xmm1 \n\t"
266 "movss 100(%1), %%xmm5 \n\t"
267 "addss %%xmm7, %%xmm5 \n\t"
268 "addss 108(%1), %%xmm7 \n\t"
269 "addss %%xmm5, %%xmm1 \n\t"
270 "movss 84(%1), %%xmm2 \n\t"
271 "addss 92(%1), %%xmm2 \n\t"
272 "addss %%xmm2, %%xmm5 \n\t"
273 "movss %%xmm1, 68(%1) \n\t"
274 "addss %%xmm7, %%xmm2 \n\t"
275 "movss 76(%1), %%xmm1 \n\t"
276 "movss %%xmm2, 84(%1) \n\t"
277 "movss %%xmm5, 76(%1) \n\t"
278 "movss 108(%1), %%xmm2 \n\t"
279 "addss %%xmm1, %%xmm7 \n\t"
280 "addss 124(%1), %%xmm2 \n\t"
281 "addss %%xmm2, %%xmm1 \n\t"
282 "addss 92(%1), %%xmm2 \n\t"
283 "movss %%xmm1, 100(%1) \n\t"
284 "movss %%xmm2, 108(%1) \n\t"
285 "movss 92(%1), %%xmm2 \n\t"
286 "movss %%xmm7, 92(%1) \n\t"
287 "addss 124(%1), %%xmm2 \n\t"
288 "movss %%xmm2, 116(%1) \n\t"
289 :"+&r"(tmp1)
290 :"r"(out), "r"(b1), "r"(smask), "r"(in)
291 :"memory"
292 );
293 }
294