IFF stereo support
[libav.git] / libavcodec / i386 / fft_sse.c
CommitLineData
bb6f5690
FB
1/*
2 * FFT/MDCT transform with SSE optimizations
3 * Copyright (c) 2002 Fabrice Bellard.
4 *
b78e7197
DB
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
bb6f5690
FB
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
b78e7197 10 * version 2.1 of the License, or (at your option) any later version.
bb6f5690 11 *
b78e7197 12 * FFmpeg is distributed in the hope that it will be useful,
bb6f5690
FB
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
b78e7197 18 * License along with FFmpeg; if not, write to the Free Software
5509bffa 19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
bb6f5690 20 */
b550bfaa 21#include "dsputil.h"
40d0e665 22#include "x86_cpu.h"
bb6f5690 23
2ffb22d2
ZM
24static const int p1p1p1m1[4] __attribute__((aligned(16))) =
25 { 0, 0, 0, 1 << 31 };
bb6f5690 26
2ffb22d2
ZM
27static const int p1p1m1p1[4] __attribute__((aligned(16))) =
28 { 0, 0, 1 << 31, 0 };
e1958604 29
2ffb22d2
ZM
30static const int p1p1m1m1[4] __attribute__((aligned(16))) =
31 { 0, 0, 1 << 31, 1 << 31 };
bb6f5690 32
ebbafcb4
LM
33static const int p1m1p1m1[4] __attribute__((aligned(16))) =
34 { 0, 1 << 31, 0, 1 << 31 };
35
36static const int m1m1m1m1[4] __attribute__((aligned(16))) =
37 { 1 << 31, 1 << 31, 1 << 31, 1 << 31 };
38
bb6f5690
FB
39#if 0
40static void print_v4sf(const char *str, __m128 a)
41{
42 float *p = (float *)&a;
43 printf("%s: %f %f %f %f\n",
44 str, p[0], p[1], p[2], p[3]);
45}
46#endif
47
48/* XXX: handle reverse case */
68951ecf 49void ff_fft_calc_sse(FFTContext *s, FFTComplex *z)
bb6f5690
FB
50{
51 int ln = s->nbits;
40d0e665
RP
52 x86_reg i;
53 long j;
1e4ecf26
LM
54 long nblocks, nloops;
55 FFTComplex *p, *cptr;
56
57 asm volatile(
58 "movaps %0, %%xmm4 \n\t"
59 "movaps %1, %%xmm5 \n\t"
60 ::"m"(*p1p1m1m1),
61 "m"(*(s->inverse ? p1p1m1p1 : p1p1p1m1))
62 );
63
64 i = 8 << ln;
65 asm volatile(
66 "1: \n\t"
67 "sub $32, %0 \n\t"
68 /* do the pass 0 butterfly */
69 "movaps (%0,%1), %%xmm0 \n\t"
70 "movaps %%xmm0, %%xmm1 \n\t"
71 "shufps $0x4E, %%xmm0, %%xmm0 \n\t"
72 "xorps %%xmm4, %%xmm1 \n\t"
73 "addps %%xmm1, %%xmm0 \n\t"
74 "movaps 16(%0,%1), %%xmm2 \n\t"
75 "movaps %%xmm2, %%xmm3 \n\t"
76 "shufps $0x4E, %%xmm2, %%xmm2 \n\t"
77 "xorps %%xmm4, %%xmm3 \n\t"
78 "addps %%xmm3, %%xmm2 \n\t"
79 /* multiply third by -i */
80 /* by toggling the sign bit */
81 "shufps $0xB4, %%xmm2, %%xmm2 \n\t"
82 "xorps %%xmm5, %%xmm2 \n\t"
83 /* do the pass 1 butterfly */
84 "movaps %%xmm0, %%xmm1 \n\t"
85 "addps %%xmm2, %%xmm0 \n\t"
86 "subps %%xmm2, %%xmm1 \n\t"
87 "movaps %%xmm0, (%0,%1) \n\t"
88 "movaps %%xmm1, 16(%0,%1) \n\t"
89 "jg 1b \n\t"
90 :"+r"(i)
91 :"r"(z)
92 );
bb6f5690
FB
93 /* pass 2 .. ln-1 */
94
1e4ecf26 95 nblocks = 1 << (ln-3);
bb6f5690 96 nloops = 1 << 2;
1e4ecf26 97 cptr = s->exptab1;
bb6f5690
FB
98 do {
99 p = z;
bb6f5690
FB
100 j = nblocks;
101 do {
1e4ecf26
LM
102 i = nloops*8;
103 asm volatile(
104 "1: \n\t"
25e4f8aa 105 "sub $32, %0 \n\t"
1e4ecf26
LM
106 "movaps (%2,%0), %%xmm1 \n\t"
107 "movaps (%1,%0), %%xmm0 \n\t"
25e4f8aa
ZM
108 "movaps 16(%2,%0), %%xmm5 \n\t"
109 "movaps 16(%1,%0), %%xmm4 \n\t"
1e4ecf26 110 "movaps %%xmm1, %%xmm2 \n\t"
25e4f8aa 111 "movaps %%xmm5, %%xmm6 \n\t"
1e4ecf26
LM
112 "shufps $0xA0, %%xmm1, %%xmm1 \n\t"
113 "shufps $0xF5, %%xmm2, %%xmm2 \n\t"
25e4f8aa
ZM
114 "shufps $0xA0, %%xmm5, %%xmm5 \n\t"
115 "shufps $0xF5, %%xmm6, %%xmm6 \n\t"
1e4ecf26
LM
116 "mulps (%3,%0,2), %%xmm1 \n\t" // cre*re cim*re
117 "mulps 16(%3,%0,2), %%xmm2 \n\t" // -cim*im cre*im
25e4f8aa
ZM
118 "mulps 32(%3,%0,2), %%xmm5 \n\t" // cre*re cim*re
119 "mulps 48(%3,%0,2), %%xmm6 \n\t" // -cim*im cre*im
1e4ecf26 120 "addps %%xmm2, %%xmm1 \n\t"
25e4f8aa 121 "addps %%xmm6, %%xmm5 \n\t"
1e4ecf26 122 "movaps %%xmm0, %%xmm3 \n\t"
25e4f8aa 123 "movaps %%xmm4, %%xmm7 \n\t"
1e4ecf26
LM
124 "addps %%xmm1, %%xmm0 \n\t"
125 "subps %%xmm1, %%xmm3 \n\t"
25e4f8aa
ZM
126 "addps %%xmm5, %%xmm4 \n\t"
127 "subps %%xmm5, %%xmm7 \n\t"
1e4ecf26
LM
128 "movaps %%xmm0, (%1,%0) \n\t"
129 "movaps %%xmm3, (%2,%0) \n\t"
25e4f8aa
ZM
130 "movaps %%xmm4, 16(%1,%0) \n\t"
131 "movaps %%xmm7, 16(%2,%0) \n\t"
1e4ecf26
LM
132 "jg 1b \n\t"
133 :"+r"(i)
134 :"r"(p), "r"(p + nloops), "r"(cptr)
135 );
136 p += nloops*2;
bb6f5690 137 } while (--j);
1e4ecf26
LM
138 cptr += nloops*2;
139 nblocks >>= 1;
140 nloops <<= 1;
bb6f5690
FB
141 } while (nblocks != 0);
142}
e1958604 143
ebbafcb4
LM
144void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output,
145 const FFTSample *input, FFTSample *tmp)
146{
40d0e665
RP
147 x86_reg k;
148 long n8, n4, n2, n;
ebbafcb4
LM
149 const uint16_t *revtab = s->fft.revtab;
150 const FFTSample *tcos = s->tcos;
151 const FFTSample *tsin = s->tsin;
152 const FFTSample *in1, *in2;
153 FFTComplex *z = (FFTComplex *)tmp;
154
155 n = 1 << s->nbits;
156 n2 = n >> 1;
157 n4 = n >> 2;
158 n8 = n >> 3;
159
25e4f8aa
ZM
160#ifdef ARCH_X86_64
161 asm volatile ("movaps %0, %%xmm8\n\t"::"m"(*p1m1p1m1));
162#define P1M1P1M1 "%%xmm8"
163#else
164#define P1M1P1M1 "%4"
165#endif
ebbafcb4
LM
166
167 /* pre rotation */
168 in1 = input;
169 in2 = input + n2 - 4;
170
25e4f8aa
ZM
171 /* Complex multiplication */
172 for (k = 0; k < n4; k += 4) {
ebbafcb4
LM
173 asm volatile (
174 "movaps %0, %%xmm0 \n\t" // xmm0 = r0 X r1 X : in2
175 "movaps %1, %%xmm3 \n\t" // xmm3 = X i1 X i0: in1
ee387b57
GB
176 "movaps -16+1*%0, %%xmm4 \n\t" // xmm4 = r0 X r1 X : in2
177 "movaps 16+1*%1, %%xmm7 \n\t" // xmm7 = X i1 X i0: in1
ebbafcb4
LM
178 "movlps %2, %%xmm1 \n\t" // xmm1 = X X R1 R0: tcos
179 "movlps %3, %%xmm2 \n\t" // xmm2 = X X I1 I0: tsin
ee387b57
GB
180 "movlps 8+1*%2, %%xmm5 \n\t" // xmm5 = X X R1 R0: tcos
181 "movlps 8+1*%3, %%xmm6 \n\t" // xmm6 = X X I1 I0: tsin
ebbafcb4
LM
182 "shufps $95, %%xmm0, %%xmm0 \n\t" // xmm0 = r1 r1 r0 r0
183 "shufps $160,%%xmm3, %%xmm3 \n\t" // xmm3 = i1 i1 i0 i0
25e4f8aa
ZM
184 "shufps $95, %%xmm4, %%xmm4 \n\t" // xmm4 = r1 r1 r0 r0
185 "shufps $160,%%xmm7, %%xmm7 \n\t" // xmm7 = i1 i1 i0 i0
ebbafcb4 186 "unpcklps %%xmm2, %%xmm1 \n\t" // xmm1 = I1 R1 I0 R0
25e4f8aa 187 "unpcklps %%xmm6, %%xmm5 \n\t" // xmm5 = I1 R1 I0 R0
ebbafcb4 188 "movaps %%xmm1, %%xmm2 \n\t" // xmm2 = I1 R1 I0 R0
25e4f8aa
ZM
189 "movaps %%xmm5, %%xmm6 \n\t" // xmm6 = I1 R1 I0 R0
190 "xorps "P1M1P1M1", %%xmm2 \n\t" // xmm2 = -I1 R1 -I0 R0
191 "xorps "P1M1P1M1", %%xmm6 \n\t" // xmm6 = -I1 R1 -I0 R0
ebbafcb4 192 "mulps %%xmm1, %%xmm0 \n\t" // xmm0 = rI rR rI rR
25e4f8aa 193 "mulps %%xmm5, %%xmm4 \n\t" // xmm4 = rI rR rI rR
ebbafcb4 194 "shufps $177,%%xmm2, %%xmm2 \n\t" // xmm2 = R1 -I1 R0 -I0
25e4f8aa 195 "shufps $177,%%xmm6, %%xmm6 \n\t" // xmm6 = R1 -I1 R0 -I0
ebbafcb4 196 "mulps %%xmm2, %%xmm3 \n\t" // xmm3 = Ri -Ii Ri -Ii
25e4f8aa 197 "mulps %%xmm6, %%xmm7 \n\t" // xmm7 = Ri -Ii Ri -Ii
ebbafcb4 198 "addps %%xmm3, %%xmm0 \n\t" // xmm0 = result
25e4f8aa 199 "addps %%xmm7, %%xmm4 \n\t" // xmm4 = result
ebbafcb4
LM
200 ::"m"(in2[-2*k]), "m"(in1[2*k]),
201 "m"(tcos[k]), "m"(tsin[k])
25e4f8aa
ZM
202#ifndef ARCH_X86_64
203 ,"m"(*p1m1p1m1)
204#endif
ebbafcb4
LM
205 );
206 /* Should be in the same block, hack for gcc2.95 & gcc3 */
207 asm (
208 "movlps %%xmm0, %0 \n\t"
209 "movhps %%xmm0, %1 \n\t"
25e4f8aa
ZM
210 "movlps %%xmm4, %2 \n\t"
211 "movhps %%xmm4, %3 \n\t"
212 :"=m"(z[revtab[k]]), "=m"(z[revtab[k + 1]]),
213 "=m"(z[revtab[k + 2]]), "=m"(z[revtab[k + 3]])
ebbafcb4
LM
214 );
215 }
216
217 ff_fft_calc_sse(&s->fft, z);
218
25e4f8aa
ZM
219#ifndef ARCH_X86_64
220#undef P1M1P1M1
221#define P1M1P1M1 "%3"
222#endif
ebbafcb4
LM
223
224 /* post rotation + reordering */
25e4f8aa 225 for (k = 0; k < n4; k += 4) {
ebbafcb4
LM
226 asm (
227 "movaps %0, %%xmm0 \n\t" // xmm0 = i1 r1 i0 r0: z
ee387b57 228 "movaps 16+1*%0, %%xmm4 \n\t" // xmm4 = i1 r1 i0 r0: z
ebbafcb4 229 "movlps %1, %%xmm1 \n\t" // xmm1 = X X R1 R0: tcos
ee387b57 230 "movlps 8+1*%1, %%xmm5 \n\t" // xmm5 = X X R1 R0: tcos
ebbafcb4 231 "movaps %%xmm0, %%xmm3 \n\t" // xmm3 = i1 r1 i0 r0
25e4f8aa 232 "movaps %%xmm4, %%xmm7 \n\t" // xmm7 = i1 r1 i0 r0
ebbafcb4 233 "movlps %2, %%xmm2 \n\t" // xmm2 = X X I1 I0: tsin
ee387b57 234 "movlps 8+1*%2, %%xmm6 \n\t" // xmm6 = X X I1 I0: tsin
ebbafcb4
LM
235 "shufps $160,%%xmm0, %%xmm0 \n\t" // xmm0 = r1 r1 r0 r0
236 "shufps $245,%%xmm3, %%xmm3 \n\t" // xmm3 = i1 i1 i0 i0
25e4f8aa
ZM
237 "shufps $160,%%xmm4, %%xmm4 \n\t" // xmm4 = r1 r1 r0 r0
238 "shufps $245,%%xmm7, %%xmm7 \n\t" // xmm7 = i1 i1 i0 i0
ebbafcb4 239 "unpcklps %%xmm2, %%xmm1 \n\t" // xmm1 = I1 R1 I0 R0
25e4f8aa 240 "unpcklps %%xmm6, %%xmm5 \n\t" // xmm5 = I1 R1 I0 R0
ebbafcb4 241 "movaps %%xmm1, %%xmm2 \n\t" // xmm2 = I1 R1 I0 R0
25e4f8aa
ZM
242 "movaps %%xmm5, %%xmm6 \n\t" // xmm6 = I1 R1 I0 R0
243 "xorps "P1M1P1M1", %%xmm2 \n\t" // xmm2 = -I1 R1 -I0 R0
ebbafcb4 244 "mulps %%xmm1, %%xmm0 \n\t" // xmm0 = rI rR rI rR
25e4f8aa
ZM
245 "xorps "P1M1P1M1", %%xmm6 \n\t" // xmm6 = -I1 R1 -I0 R0
246 "mulps %%xmm5, %%xmm4 \n\t" // xmm4 = rI rR rI rR
ebbafcb4 247 "shufps $177,%%xmm2, %%xmm2 \n\t" // xmm2 = R1 -I1 R0 -I0
25e4f8aa 248 "shufps $177,%%xmm6, %%xmm6 \n\t" // xmm6 = R1 -I1 R0 -I0
ebbafcb4 249 "mulps %%xmm2, %%xmm3 \n\t" // xmm3 = Ri -Ii Ri -Ii
25e4f8aa 250 "mulps %%xmm6, %%xmm7 \n\t" // xmm7 = Ri -Ii Ri -Ii
ebbafcb4 251 "addps %%xmm3, %%xmm0 \n\t" // xmm0 = result
25e4f8aa 252 "addps %%xmm7, %%xmm4 \n\t" // xmm4 = result
ebbafcb4 253 "movaps %%xmm0, %0 \n\t"
ee387b57 254 "movaps %%xmm4, 16+1*%0\n\t"
ebbafcb4
LM
255 :"+m"(z[k])
256 :"m"(tcos[k]), "m"(tsin[k])
25e4f8aa
ZM
257#ifndef ARCH_X86_64
258 ,"m"(*p1m1p1m1)
259#endif
ebbafcb4
LM
260 );
261 }
262
263 /*
264 Mnemonics:
265 0 = z[k].re
266 1 = z[k].im
267 2 = z[k + 1].re
268 3 = z[k + 1].im
269 4 = z[-k - 2].re
270 5 = z[-k - 2].im
271 6 = z[-k - 1].re
272 7 = z[-k - 1].im
273 */
274 k = 16-n;
275 asm volatile("movaps %0, %%xmm7 \n\t"::"m"(*m1m1m1m1));
276 asm volatile(
277 "1: \n\t"
278 "movaps -16(%4,%0), %%xmm1 \n\t" // xmm1 = 4 5 6 7 = z[-2-k]
279 "neg %0 \n\t"
280 "movaps (%4,%0), %%xmm0 \n\t" // xmm0 = 0 1 2 3 = z[k]
281 "xorps %%xmm7, %%xmm0 \n\t" // xmm0 = -0 -1 -2 -3
282 "movaps %%xmm0, %%xmm2 \n\t" // xmm2 = -0 -1 -2 -3
283 "shufps $141,%%xmm1, %%xmm0 \n\t" // xmm0 = -1 -3 4 6
284 "shufps $216,%%xmm1, %%xmm2 \n\t" // xmm2 = -0 -2 5 7
285 "shufps $156,%%xmm0, %%xmm0 \n\t" // xmm0 = -1 6 -3 4 !
286 "shufps $156,%%xmm2, %%xmm2 \n\t" // xmm2 = -0 7 -2 5 !
287 "movaps %%xmm0, (%1,%0) \n\t" // output[2*k]
288 "movaps %%xmm2, (%2,%0) \n\t" // output[n2+2*k]
289 "neg %0 \n\t"
290 "shufps $27, %%xmm0, %%xmm0 \n\t" // xmm0 = 4 -3 6 -1
291 "xorps %%xmm7, %%xmm0 \n\t" // xmm0 = -4 3 -6 1 !
292 "shufps $27, %%xmm2, %%xmm2 \n\t" // xmm2 = 5 -2 7 -0 !
293 "movaps %%xmm0, -16(%2,%0) \n\t" // output[n2-4-2*k]
294 "movaps %%xmm2, -16(%3,%0) \n\t" // output[n-4-2*k]
295 "add $16, %0 \n\t"
296 "jle 1b \n\t"
297 :"+r"(k)
298 :"r"(output), "r"(output+n2), "r"(output+n), "r"(z+n8)
299 :"memory"
300 );
301}
302