1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
|
/***
This file is part of PulseAudio.
Copyright 2004-2006 Lennart Poettering
Copyright 2009 Wim Taymans <wim.taymans@collabora.co.uk>
PulseAudio is free software; you can redistribute it and/or modify
it under the terms of the GNU Lesser General Public License as published
by the Free Software Foundation; either version 2.1 of the License,
or (at your option) any later version.
PulseAudio is distributed in the hope that it will be useful, but
WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
General Public License for more details.
You should have received a copy of the GNU Lesser General Public License
along with PulseAudio; if not, write to the Free Software
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
USA.
***/
#ifdef HAVE_CONFIG_H
#include <config.h>
#endif
#include <pulse/timeval.h>
#include <pulsecore/random.h>
#include <pulsecore/macro.h>
#include <pulsecore/g711.h>
#include <pulsecore/core-util.h>
#include "cpu-x86.h"
#include "sample-util.h"
#include "endianmacros.h"
#if defined (__i386__) || defined (__amd64__)
/* in s: 2 int16_t samples
* in v: 2 int32_t volumes, fixed point 16:16
* out s: contains scaled and clamped int16_t samples.
*
* We calculate the high 32 bits of a 32x16 multiply which we then
* clamp to 16 bits. The calulcation is:
*
* vl = (v & 0xffff)
* vh = (v >> 16)
* s = ((s * vl) >> 16) + (s * vh);
*
* For the first multiply we have to do a sign correction as we need to
* multiply a signed int with an unsigned int. Hacker's delight 8-3 gives a
* simple formula to correct the sign of the high word after the signed
* multiply.
*/
#define VOLUME_32x16(s,v) /* .. | vh | vl | */ \
" pxor %%mm4, %%mm4 \n\t" /* .. | 0 | 0 | */ \
" punpcklwd %%mm4, "#s" \n\t" /* .. | 0 | p0 | */ \
" pcmpgtw "#v", %%mm4 \n\t" /* .. | 0 | s(vl) | */ \
" pand "#s", %%mm4 \n\t" /* .. | 0 | (p0) | (vl >> 15) & p */ \
" movq "#s", %%mm5 \n\t" \
" pmulhw "#v", "#s" \n\t" /* .. | 0 | vl*p0 | */ \
" paddw %%mm4, "#s" \n\t" /* .. | 0 | vl*p0 | + sign correct */ \
" psrld $16, "#v" \n\t" /* .. | 0 | vh | */ \
" pmaddwd %%mm5, "#v" \n\t" /* .. | p0 * vh | */ \
" paddd "#s", "#v" \n\t" /* .. | p0 * v0 | */ \
" packssdw "#v", "#v" \n\t" /* .. | p1*v1 | p0*v0 | */
/* approximately advances %3 = (%3 + a) % b. This function requires that
* a <= b. */
#define MOD_ADD(a,b) \
" add "#a", %3 \n\t" \
" mov %3, %4 \n\t" \
" sub "#b", %4 \n\t" \
" cmovae %4, %3 \n\t"
/* swap 16 bits */
#define SWAP_16(s) \
" movq "#s", %%mm4 \n\t" /* .. | h l | */ \
" psrlw $8, %%mm4 \n\t" /* .. | 0 h | */ \
" psllw $8, "#s" \n\t" /* .. | l 0 | */ \
" por %%mm4, "#s" \n\t" /* .. | l h | */
/* swap 2 registers 16 bits for better pairing */
#define SWAP_16_2(s1,s2) \
" movq "#s1", %%mm4 \n\t" /* .. | h l | */ \
" movq "#s2", %%mm5 \n\t" \
" psrlw $8, %%mm4 \n\t" /* .. | 0 h | */ \
" psrlw $8, %%mm5 \n\t" \
" psllw $8, "#s1" \n\t" /* .. | l 0 | */ \
" psllw $8, "#s2" \n\t" \
" por %%mm4, "#s1" \n\t" /* .. | l h | */ \
" por %%mm5, "#s2" \n\t"
static void
pa_volume_s16ne_mmx (int16_t *samples, int32_t *volumes, unsigned channels, unsigned length)
{
pa_reg_x86 channel, temp;
/* the max number of samples we process at a time, this is also the max amount
* we overread the volume array, which should have enough padding. */
channels = PA_MAX (4U, channels);
__asm__ __volatile__ (
" xor %3, %3 \n\t"
" sar $1, %2 \n\t" /* length /= sizeof (int16_t) */
" test $1, %2 \n\t" /* check for odd samples */
" je 2f \n\t"
" movd (%1, %3, 4), %%mm0 \n\t" /* | v0h | v0l | */
" movw (%0), %w4 \n\t" /* .. | p0 | */
" movd %4, %%mm1 \n\t"
VOLUME_32x16 (%%mm1, %%mm0)
" movd %%mm0, %4 \n\t" /* .. | p0*v0 | */
" movw %w4, (%0) \n\t"
" add $2, %0 \n\t"
MOD_ADD ($1, %5)
"2: \n\t"
" sar $1, %2 \n\t" /* prepare for processing 2 samples at a time */
" test $1, %2 \n\t" /* check for odd samples */
" je 4f \n\t"
"3: \n\t" /* do samples in groups of 2 */
" movq (%1, %3, 4), %%mm0 \n\t" /* | v1h | v1l | v0h | v0l | */
" movd (%0), %%mm1 \n\t" /* .. | p1 | p0 | */
VOLUME_32x16 (%%mm1, %%mm0)
" movd %%mm0, (%0) \n\t" /* .. | p1*v1 | p0*v0 | */
" add $4, %0 \n\t"
MOD_ADD ($2, %5)
"4: \n\t"
" sar $1, %2 \n\t" /* prepare for processing 4 samples at a time */
" cmp $0, %2 \n\t"
" je 6f \n\t"
"5: \n\t" /* do samples in groups of 4 */
" movq (%1, %3, 4), %%mm0 \n\t" /* | v1h | v1l | v0h | v0l | */
" movq 8(%1, %3, 4), %%mm2 \n\t" /* | v3h | v3l | v2h | v2l | */
" movd (%0), %%mm1 \n\t" /* .. | p1 | p0 | */
" movd 4(%0), %%mm3 \n\t" /* .. | p3 | p2 | */
VOLUME_32x16 (%%mm1, %%mm0)
VOLUME_32x16 (%%mm3, %%mm2)
" movd %%mm0, (%0) \n\t" /* .. | p1*v1 | p0*v0 | */
" movd %%mm2, 4(%0) \n\t" /* .. | p3*v3 | p2*v2 | */
" add $8, %0 \n\t"
MOD_ADD ($4, %5)
" dec %2 \n\t"
" jne 5b \n\t"
"6: \n\t"
" emms \n\t"
: "+r" (samples), "+r" (volumes), "+r" (length), "=D" ((pa_reg_x86)channel), "=&r" (temp)
: "X" ((pa_reg_x86)channels)
: "cc"
);
}
static void
pa_volume_s16re_mmx (int16_t *samples, int32_t *volumes, unsigned channels, unsigned length)
{
pa_reg_x86 channel, temp;
/* the max number of samples we process at a time, this is also the max amount
* we overread the volume array, which should have enough padding. */
channels = PA_MAX (4U, channels);
__asm__ __volatile__ (
" xor %3, %3 \n\t"
" sar $1, %2 \n\t" /* length /= sizeof (int16_t) */
" pcmpeqw %%mm6, %%mm6 \n\t" /* .. | ffff | ffff | */
" pcmpeqw %%mm7, %%mm7 \n\t" /* .. | ffff | ffff | */
" pslld $16, %%mm6 \n\t" /* .. | ffff | 0 | */
" psrld $31, %%mm7 \n\t" /* .. | 0 | 1 | */
" test $1, %2 \n\t" /* check for odd samples */
" je 2f \n\t"
" movd (%1, %3, 4), %%mm0 \n\t" /* | v0h | v0l | */
" movw (%0), %w4 \n\t" /* .. | p0 | */
" rorw $8, %w4 \n\t"
" movd %4, %%mm1 \n\t"
VOLUME_32x16 (%%mm1, %%mm0)
" movd %%mm0, %4 \n\t" /* .. | p0*v0 | */
" rorw $8, %w4 \n\t"
" movw %w4, (%0) \n\t"
" add $2, %0 \n\t"
MOD_ADD ($1, %5)
"2: \n\t"
" sar $1, %2 \n\t" /* prepare for processing 2 samples at a time */
" test $1, %2 \n\t" /* check for odd samples */
" je 4f \n\t"
"3: \n\t" /* do samples in groups of 2 */
" movq (%1, %3, 4), %%mm0 \n\t" /* | v1h | v1l | v0h | v0l | */
" movd (%0), %%mm1 \n\t" /* .. | p1 | p0 | */
SWAP_16 (%%mm1)
VOLUME_32x16 (%%mm1, %%mm0)
SWAP_16 (%%mm0)
" movd %%mm0, (%0) \n\t" /* .. | p1*v1 | p0*v0 | */
" add $4, %0 \n\t"
MOD_ADD ($2, %5)
"4: \n\t"
" sar $1, %2 \n\t" /* prepare for processing 4 samples at a time */
" cmp $0, %2 \n\t"
" je 6f \n\t"
"5: \n\t" /* do samples in groups of 4 */
" movq (%1, %3, 4), %%mm0 \n\t" /* | v1h | v1l | v0h | v0l | */
" movq 8(%1, %3, 4), %%mm2 \n\t" /* | v3h | v3l | v2h | v2l | */
" movd (%0), %%mm1 \n\t" /* .. | p1 | p0 | */
" movd 4(%0), %%mm3 \n\t" /* .. | p3 | p2 | */
SWAP_16_2 (%%mm1, %%mm3)
VOLUME_32x16 (%%mm1, %%mm0)
VOLUME_32x16 (%%mm3, %%mm2)
SWAP_16_2 (%%mm0, %%mm2)
" movd %%mm0, (%0) \n\t" /* .. | p1*v1 | p0*v0 | */
" movd %%mm2, 4(%0) \n\t" /* .. | p3*v3 | p2*v2 | */
" add $8, %0 \n\t"
MOD_ADD ($4, %5)
" dec %2 \n\t"
" jne 5b \n\t"
"6: \n\t"
" emms \n\t"
: "+r" (samples), "+r" (volumes), "+r" (length), "=D" ((pa_reg_x86)channel), "=&r" (temp)
: "X" ((pa_reg_x86)channels)
: "cc"
);
}
#undef RUN_TEST
#ifdef RUN_TEST
#define CHANNELS 2
#define SAMPLES 1021
#define TIMES 1000
#define PADDING 16
static void run_test (void) {
int16_t samples[SAMPLES];
int16_t samples_ref[SAMPLES];
int16_t samples_orig[SAMPLES];
int32_t volumes[CHANNELS + PADDING];
int i, j, padding;
pa_do_volume_func_t func;
pa_usec_t start, stop;
func = pa_get_volume_func (PA_SAMPLE_S16NE);
printf ("checking MMX %zd\n", sizeof (samples));
pa_random (samples, sizeof (samples));
memcpy (samples_ref, samples, sizeof (samples));
memcpy (samples_orig, samples, sizeof (samples));
for (i = 0; i < CHANNELS; i++)
volumes[i] = rand() >> 1;
for (padding = 0; padding < PADDING; padding++, i++)
volumes[i] = volumes[padding];
func (samples_ref, volumes, CHANNELS, sizeof (samples));
pa_volume_s16ne_mmx (samples, volumes, CHANNELS, sizeof (samples));
for (i = 0; i < SAMPLES; i++) {
if (samples[i] != samples_ref[i]) {
printf ("%d: %04x != %04x (%04x * %04x)\n", i, samples[i], samples_ref[i],
samples_orig[i], volumes[i % CHANNELS]);
}
}
start = pa_rtclock_now();
for (j = 0; j < TIMES; j++) {
memcpy (samples, samples_orig, sizeof (samples));
pa_volume_s16ne_mmx (samples, volumes, CHANNELS, sizeof (samples));
}
stop = pa_rtclock_now();
pa_log_info("MMX: %llu usec.", (long long unsigned int)(stop - start));
start = pa_rtclock_now();
for (j = 0; j < TIMES; j++) {
memcpy (samples_ref, samples_orig, sizeof (samples));
func (samples_ref, volumes, CHANNELS, sizeof (samples));
}
stop = pa_rtclock_now();
pa_log_info("ref: %llu usec.", (long long unsigned int)(stop - start));
}
#endif
#endif /* defined (__i386__) || defined (__amd64__) */
void pa_volume_func_init_mmx (pa_cpu_x86_flag_t flags) {
#if defined (__i386__) || defined (__amd64__)
#ifdef RUN_TEST
run_test ();
#endif
if (flags & PA_CPU_X86_MMX) {
pa_log_info("Initialising MMX optimized functions.");
pa_set_volume_func (PA_SAMPLE_S16NE, (pa_do_volume_func_t) pa_volume_s16ne_mmx);
pa_set_volume_func (PA_SAMPLE_S16RE, (pa_do_volume_func_t) pa_volume_s16re_mmx);
}
#endif /* defined (__i386__) || defined (__amd64__) */
}
|
