]>
code.delx.au - pulseaudio/blob - src/pulsecore/svolume_mmx.c
2 This file is part of PulseAudio.
4 Copyright 2004-2006 Lennart Poettering
5 Copyright 2009 Wim Taymans <wim.taymans@collabora.co.uk>
7 PulseAudio is free software; you can redistribute it and/or modify
8 it under the terms of the GNU Lesser General Public License as published
9 by the Free Software Foundation; either version 2.1 of the License,
10 or (at your option) any later version.
12 PulseAudio is distributed in the hope that it will be useful, but
13 WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 General Public License for more details.
17 You should have received a copy of the GNU Lesser General Public License
18 along with PulseAudio; if not, write to the Free Software
19 Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
27 #include <pulse/rtclock.h>
29 #include <pulsecore/random.h>
30 #include <pulsecore/macro.h>
31 #include <pulsecore/endianmacros.h>
35 #include "sample-util.h"
37 #if defined (__i386__) || defined (__amd64__)
38 /* in s: 2 int16_t samples
39 * in v: 2 int32_t volumes, fixed point 16:16
40 * out s: contains scaled and clamped int16_t samples.
42 * We calculate the high 32 bits of a 32x16 multiply which we then
43 * clamp to 16 bits. The calculation is:
47 * s = ((s * vl) >> 16) + (s * vh);
49 * For the first multiply we have to do a sign correction as we need to
50 * multiply a signed int with an unsigned int. Hacker's delight 8-3 gives a
51 * simple formula to correct the sign of the high word after the signed
54 #define VOLUME_32x16(s,v) /* .. | vh | vl | */ \
55 " pxor %%mm4, %%mm4 \n\t" /* .. | 0 | 0 | */ \
56 " punpcklwd %%mm4, "#s" \n\t" /* .. | 0 | p0 | */ \
57 " pcmpgtw "#v", %%mm4 \n\t" /* .. | 0 | s(vl) | */ \
58 " pand "#s", %%mm4 \n\t" /* .. | 0 | (p0) | (vl >> 15) & p */ \
59 " movq "#s", %%mm5 \n\t" \
60 " pmulhw "#v", "#s" \n\t" /* .. | 0 | vl*p0 | */ \
61 " paddw %%mm4, "#s" \n\t" /* .. | 0 | vl*p0 | + sign correct */ \
62 " pslld $16, "#s" \n\t" /* .. | vl*p0 | 0 | */ \
63 " psrld $16, "#v" \n\t" /* .. | 0 | vh | */ \
64 " psrad $16, "#s" \n\t" /* .. | vl*p0 | sign extend */ \
65 " pmaddwd %%mm5, "#v" \n\t" /* .. | p0 * vh | */ \
66 " paddd "#s", "#v" \n\t" /* .. | p0 * v0 | */ \
67 " packssdw "#v", "#v" \n\t" /* .. | p1*v1 | p0*v0 | */
69 /* approximately advances %3 = (%3 + a) % b. This function requires that
71 #define MOD_ADD(a,b) \
72 " add "#a", %3 \n\t" \
74 " sub "#b", %4 \n\t" \
79 " movq "#s", %%mm4 \n\t" /* .. | h l | */ \
80 " psrlw $8, %%mm4 \n\t" /* .. | 0 h | */ \
81 " psllw $8, "#s" \n\t" /* .. | l 0 | */ \
82 " por %%mm4, "#s" \n\t" /* .. | l h | */
84 /* swap 2 registers 16 bits for better pairing */
85 #define SWAP_16_2(s1,s2) \
86 " movq "#s1", %%mm4 \n\t" /* .. | h l | */ \
87 " movq "#s2", %%mm5 \n\t" \
88 " psrlw $8, %%mm4 \n\t" /* .. | 0 h | */ \
89 " psrlw $8, %%mm5 \n\t" \
90 " psllw $8, "#s1" \n\t" /* .. | l 0 | */ \
91 " psllw $8, "#s2" \n\t" \
92 " por %%mm4, "#s1" \n\t" /* .. | l h | */ \
93 " por %%mm5, "#s2" \n\t"
95 static void pa_volume_s16ne_mmx(int16_t *samples
, const int32_t *volumes
, unsigned channels
, unsigned length
) {
96 pa_reg_x86 channel
, temp
;
98 /* Channels must be at least 4, and always a multiple of the original number.
99 * This is also the max amount we overread the volume array, which should
100 * have enough padding. */
101 channels
= channels
== 3 ? 6 : PA_MAX (4U, channels
);
103 __asm__
__volatile__ (
105 " sar $1, %2 \n\t" /* length /= sizeof (int16_t) */
107 " test $1, %2 \n\t" /* check for odd samples */
110 " movd (%q1, %3, 4), %%mm0 \n\t" /* | v0h | v0l | */
111 " movw (%0), %w4 \n\t" /* .. | p0 | */
112 " movd %4, %%mm1 \n\t"
113 VOLUME_32x16 (%%mm1
, %%mm0
)
114 " movd %%mm0, %4 \n\t" /* .. | p0*v0 | */
115 " movw %w4, (%0) \n\t"
120 " sar $1, %2 \n\t" /* prepare for processing 2 samples at a time */
121 " test $1, %2 \n\t" /* check for odd samples */
124 "3: \n\t" /* do samples in groups of 2 */
125 " movq (%q1, %3, 4), %%mm0 \n\t" /* | v1h | v1l | v0h | v0l | */
126 " movd (%0), %%mm1 \n\t" /* .. | p1 | p0 | */
127 VOLUME_32x16 (%%mm1
, %%mm0
)
128 " movd %%mm0, (%0) \n\t" /* .. | p1*v1 | p0*v0 | */
133 " sar $1, %2 \n\t" /* prepare for processing 4 samples at a time */
137 "5: \n\t" /* do samples in groups of 4 */
138 " movq (%q1, %3, 4), %%mm0 \n\t" /* | v1h | v1l | v0h | v0l | */
139 " movq 8(%q1, %3, 4), %%mm2 \n\t" /* | v3h | v3l | v2h | v2l | */
140 " movd (%0), %%mm1 \n\t" /* .. | p1 | p0 | */
141 " movd 4(%0), %%mm3 \n\t" /* .. | p3 | p2 | */
142 VOLUME_32x16 (%%mm1
, %%mm0
)
143 VOLUME_32x16 (%%mm3
, %%mm2
)
144 " movd %%mm0, (%0) \n\t" /* .. | p1*v1 | p0*v0 | */
145 " movd %%mm2, 4(%0) \n\t" /* .. | p3*v3 | p2*v2 | */
154 : "+r" (samples
), "+r" (volumes
), "+r" (length
), "=D" (channel
), "=&r" (temp
)
155 #if defined (__i386__)
158 : "r" ((pa_reg_x86
)channels
)
164 static void pa_volume_s16re_mmx(int16_t *samples
, const int32_t *volumes
, unsigned channels
, unsigned length
) {
165 pa_reg_x86 channel
, temp
;
167 /* Channels must be at least 4, and always a multiple of the original number.
168 * This is also the max amount we overread the volume array, which should
169 * have enough padding. */
170 channels
= channels
== 3 ? 6 : PA_MAX (4U, channels
);
172 __asm__
__volatile__ (
174 " sar $1, %2 \n\t" /* length /= sizeof (int16_t) */
175 " pcmpeqw %%mm6, %%mm6 \n\t" /* .. | ffff | ffff | */
176 " pcmpeqw %%mm7, %%mm7 \n\t" /* .. | ffff | ffff | */
177 " pslld $16, %%mm6 \n\t" /* .. | ffff | 0 | */
178 " psrld $31, %%mm7 \n\t" /* .. | 0 | 1 | */
180 " test $1, %2 \n\t" /* check for odd samples */
183 " movd (%q1, %3, 4), %%mm0 \n\t" /* | v0h | v0l | */
184 " movw (%0), %w4 \n\t" /* .. | p0 | */
186 " movd %4, %%mm1 \n\t"
187 VOLUME_32x16 (%%mm1
, %%mm0
)
188 " movd %%mm0, %4 \n\t" /* .. | p0*v0 | */
190 " movw %w4, (%0) \n\t"
195 " sar $1, %2 \n\t" /* prepare for processing 2 samples at a time */
196 " test $1, %2 \n\t" /* check for odd samples */
199 "3: \n\t" /* do samples in groups of 2 */
200 " movq (%q1, %3, 4), %%mm0 \n\t" /* | v1h | v1l | v0h | v0l | */
201 " movd (%0), %%mm1 \n\t" /* .. | p1 | p0 | */
203 VOLUME_32x16 (%%mm1
, %%mm0
)
205 " movd %%mm0, (%0) \n\t" /* .. | p1*v1 | p0*v0 | */
210 " sar $1, %2 \n\t" /* prepare for processing 4 samples at a time */
214 "5: \n\t" /* do samples in groups of 4 */
215 " movq (%q1, %3, 4), %%mm0 \n\t" /* | v1h | v1l | v0h | v0l | */
216 " movq 8(%q1, %3, 4), %%mm2 \n\t" /* | v3h | v3l | v2h | v2l | */
217 " movd (%0), %%mm1 \n\t" /* .. | p1 | p0 | */
218 " movd 4(%0), %%mm3 \n\t" /* .. | p3 | p2 | */
219 SWAP_16_2 (%%mm1
, %%mm3
)
220 VOLUME_32x16 (%%mm1
, %%mm0
)
221 VOLUME_32x16 (%%mm3
, %%mm2
)
222 SWAP_16_2 (%%mm0
, %%mm2
)
223 " movd %%mm0, (%0) \n\t" /* .. | p1*v1 | p0*v0 | */
224 " movd %%mm2, 4(%0) \n\t" /* .. | p3*v3 | p2*v2 | */
233 : "+r" (samples
), "+r" (volumes
), "+r" (length
), "=D" (channel
), "=&r" (temp
)
234 #if defined (__i386__)
237 : "r" ((pa_reg_x86
)channels
)
243 #endif /* defined (__i386__) || defined (__amd64__) */
245 void pa_volume_func_init_mmx(pa_cpu_x86_flag_t flags
) {
246 #if defined (__i386__) || defined (__amd64__)
247 if ((flags
& PA_CPU_X86_MMX
) && (flags
& PA_CPU_X86_CMOV
)) {
248 pa_log_info("Initialising MMX optimized volume functions.");
250 pa_set_volume_func(PA_SAMPLE_S16NE
, (pa_do_volume_func_t
) pa_volume_s16ne_mmx
);
251 pa_set_volume_func(PA_SAMPLE_S16RE
, (pa_do_volume_func_t
) pa_volume_s16re_mmx
);
253 #endif /* defined (__i386__) || defined (__amd64__) */