2 This file is part of PulseAudio.
4 Copyright 2004-2006 Lennart Poettering
5 Copyright 2006 Pierre Ossman <ossman@cendio.se> for Cendio AB
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
30 #include <pulsecore/g711.h>
31 #include <pulsecore/macro.h>
32 #include <pulsecore/endianmacros.h>
37 #if !defined(__APPLE__) && defined (__i386__) || defined (__amd64__)
39 static const PA_DECLARE_ALIGNED (16, float, one
[4]) = { 1.0, 1.0, 1.0, 1.0 };
40 static const PA_DECLARE_ALIGNED (16, float, mone
[4]) = { -1.0, -1.0, -1.0, -1.0 };
41 static const PA_DECLARE_ALIGNED (16, float, scale
[4]) = { 0x7fff, 0x7fff, 0x7fff, 0x7fff };
43 static void pa_sconv_s16le_from_f32ne_sse(unsigned n
, const float *a
, int16_t *b
) {
46 __asm__
__volatile__ (
47 " movaps %5, %%xmm5 \n\t"
48 " movaps %6, %%xmm6 \n\t"
49 " movaps %7, %%xmm7 \n\t"
53 " sar $3, %1 \n\t" /* 8 floats at a time */
58 " movups (%2, %0, 2), %%xmm0 \n\t" /* read 8 floats */
59 " movups 16(%2, %0, 2), %%xmm2 \n\t"
60 " minps %%xmm5, %%xmm0 \n\t" /* clamp to 1.0 */
61 " minps %%xmm5, %%xmm2 \n\t"
62 " maxps %%xmm6, %%xmm0 \n\t" /* clamp to -1.0 */
63 " maxps %%xmm6, %%xmm2 \n\t"
64 " mulps %%xmm7, %%xmm0 \n\t" /* *= 0x7fff */
65 " mulps %%xmm7, %%xmm2 \n\t"
67 " cvtps2pi %%xmm0, %%mm0 \n\t" /* low part to int */
68 " cvtps2pi %%xmm2, %%mm2 \n\t"
69 " movhlps %%xmm0, %%xmm0 \n\t" /* bring high part in position */
70 " movhlps %%xmm2, %%xmm2 \n\t"
71 " cvtps2pi %%xmm0, %%mm1 \n\t" /* high part to int */
72 " cvtps2pi %%xmm2, %%mm3 \n\t"
74 " packssdw %%mm1, %%mm0 \n\t" /* pack parts */
75 " packssdw %%mm3, %%mm2 \n\t"
76 " movq %%mm0, (%3, %0) \n\t"
77 " movq %%mm2, 8(%3, %0) \n\t"
84 " mov %4, %1 \n\t" /* prepare for leftovers */
89 " movss (%2, %0, 2), %%xmm0 \n\t"
90 " minss %%xmm5, %%xmm0 \n\t"
91 " maxss %%xmm6, %%xmm0 \n\t"
92 " mulss %%xmm7, %%xmm0 \n\t"
93 " cvtss2si %%xmm0, %4 \n\t"
94 " movw %w4, (%3, %0) \n\t"
102 : "=&r" (i
), "=&r" (temp
)
103 : "r" (a
), "r" (b
), "r" ((pa_reg_x86
)n
), "m" (*one
), "m" (*mone
), "m" (*scale
)
108 static void pa_sconv_s16le_from_f32ne_sse2(unsigned n
, const float *a
, int16_t *b
) {
111 __asm__
__volatile__ (
112 " movaps %5, %%xmm5 \n\t"
113 " movaps %6, %%xmm6 \n\t"
114 " movaps %7, %%xmm7 \n\t"
118 " sar $3, %1 \n\t" /* 8 floats at a time */
123 " movups (%2, %0, 2), %%xmm0 \n\t" /* read 8 floats */
124 " movups 16(%2, %0, 2), %%xmm2 \n\t"
125 " minps %%xmm5, %%xmm0 \n\t" /* clamp to 1.0 */
126 " minps %%xmm5, %%xmm2 \n\t"
127 " maxps %%xmm6, %%xmm0 \n\t" /* clamp to -1.0 */
128 " maxps %%xmm6, %%xmm2 \n\t"
129 " mulps %%xmm7, %%xmm0 \n\t" /* *= 0x7fff */
130 " mulps %%xmm7, %%xmm2 \n\t"
132 " cvtps2dq %%xmm0, %%xmm0 \n\t"
133 " cvtps2dq %%xmm2, %%xmm2 \n\t"
135 " packssdw %%xmm2, %%xmm0 \n\t"
136 " movdqu %%xmm0, (%3, %0) \n\t"
143 " mov %4, %1 \n\t" /* prepare for leftovers */
148 " movss (%2, %0, 2), %%xmm0 \n\t"
149 " minss %%xmm5, %%xmm0 \n\t"
150 " maxss %%xmm6, %%xmm0 \n\t"
151 " mulss %%xmm7, %%xmm0 \n\t"
152 " cvtss2si %%xmm0, %4 \n\t"
153 " movw %w4, (%3, %0) \n\t"
160 : "=&r" (i
), "=&r" (temp
)
161 : "r" (a
), "r" (b
), "r" ((pa_reg_x86
)n
), "m" (*one
), "m" (*mone
), "m" (*scale
)
172 static void run_test(void) {
173 int16_t samples
[SAMPLES
];
174 int16_t samples_ref
[SAMPLES
];
175 float floats
[SAMPLES
];
177 pa_usec_t start
, stop
;
178 pa_convert_func_t func
;
180 printf("checking SSE %zd\n", sizeof(samples
));
182 memset(samples_ref
, 0, sizeof(samples_ref
));
183 memset(samples
, 0, sizeof(samples
));
185 for (i
= 0; i
< SAMPLES
; i
++) {
186 floats
[i
] = (rand()/(RAND_MAX
+2.2)) - 1.1;
189 func
= pa_get_convert_from_float32ne_function(PA_SAMPLE_S16LE
);
190 func(SAMPLES
, floats
, samples_ref
);
191 pa_sconv_s16le_from_f32ne_sse2(SAMPLES
, floats
, samples
);
193 for (i
= 0; i
< SAMPLES
; i
++) {
194 if (samples
[i
] != samples_ref
[i
]) {
195 printf ("%d: %04x != %04x (%f)\n", i
, samples
[i
], samples_ref
[i
],
200 start
= pa_rtclock_now();
201 for (i
= 0; i
< TIMES
; i
++) {
202 pa_sconv_s16le_from_f32ne_sse2(SAMPLES
, floats
, samples
);
204 stop
= pa_rtclock_now();
205 pa_log_info("SSE: %llu usec.", (long long unsigned int)(stop
- start
));
207 start
= pa_rtclock_now();
208 for (i
= 0; i
< TIMES
; i
++) {
209 func(SAMPLES
, floats
, samples_ref
);
211 stop
= pa_rtclock_now();
212 pa_log_info("ref: %llu usec.", (long long unsigned int)(stop
- start
));
215 #endif /* defined (__i386__) || defined (__amd64__) */
218 void pa_convert_func_init_sse(pa_cpu_x86_flag_t flags
) {
219 #if !defined(__APPLE__) && defined (__i386__) || defined (__amd64__)
225 if (flags
& PA_CPU_X86_SSE2
) {
226 pa_log_info("Initialising SSE2 optimized conversions.");
227 pa_set_convert_from_float32ne_function(PA_SAMPLE_S16LE
, (pa_convert_func_t
) pa_sconv_s16le_from_f32ne_sse2
);
229 pa_log_info("Initialising SSE optimized conversions.");
230 pa_set_convert_from_float32ne_function(PA_SAMPLE_S16LE
, (pa_convert_func_t
) pa_sconv_s16le_from_f32ne_sse
);
233 #endif /* defined (__i386__) || defined (__amd64__) */