code.delx.au - pulseaudio/blob - src/pulsecore/svolume_arm.c

   1 /***
   2   This file is part of PulseAudio.
   3
   4   Copyright 2004-2006 Lennart Poettering
   5   Copyright 2009 Wim Taymans <wim.taymans@collabora.co.uk>
   6
   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.
  11
  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.
  16
  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
  20   USA.
  21 ***/
  22
  23 #ifdef HAVE_CONFIG_H
  24 #include <config.h>
  25 #endif
  26
  27 #include <pulse/timeval.h>
  28 #include <pulsecore/random.h>
  29 #include <pulsecore/macro.h>
  30 #include <pulsecore/g711.h>
  31 #include <pulsecore/core-util.h>
  32
  33 #include "cpu-arm.h"
  34
  35 #include "sample-util.h"
  36 #include "endianmacros.h"
  37
  38 #if defined (__arm__) && defined (HAVE_ARMV6)
  39
  40 #define MOD_INC() \
  41     " subs  r0, r6, %2              \n\t" \
  42     " itt cs                        \n\t" \
  43     " addcs r0, %1                  \n\t" \
  44     " movcs r6, r0                  \n\t"
  45
  46 static void
  47 pa_volume_s16ne_arm (int16_t *samples, int32_t *volumes, unsigned channels, unsigned length)
  48 {
  49     int32_t *ve;
  50
  51     /* Channels must be at least 4, and always a multiple of the original number.
  52      * This is also the max amount we overread the volume array, which should
  53      * have enough padding. */
  54     channels = channels == 3 ? 6 : PA_MAX (4U, channels);
  55     ve = volumes + channels;
  56
  57     __asm__ __volatile__ (
  58         " mov r6, %1                      \n\t"
  59         " mov %3, %3, LSR #1              \n\t" /* length /= sizeof (int16_t) */
  60         " tst %3, #1                      \n\t" /* check for odd samples */
  61         " beq  2f                         \n\t"
  62
  63         "1:                               \n\t"
  64         " ldr  r0, [r6], #4               \n\t" /* odd samples volumes */
  65         " ldrh r2, [%0]                   \n\t"
  66
  67         " smulwb r0, r0, r2               \n\t"
  68         " ssat r0, #16, r0                \n\t"
  69
  70         " strh r0, [%0], #2               \n\t"
  71
  72         MOD_INC()
  73
  74         "2:                               \n\t"
  75         " mov %3, %3, LSR #1              \n\t"
  76         " tst %3, #1                      \n\t" /* check for odd samples */
  77         " beq  4f                         \n\t"
  78
  79         "3:                               \n\t"
  80         " ldrd r2, [r6], #8               \n\t" /* 2 samples at a time */
  81         " ldr  r0, [%0]                   \n\t"
  82
  83         " smulwt r2, r2, r0               \n\t"
  84         " smulwb r3, r3, r0               \n\t"
  85
  86         " ssat r2, #16, r2                \n\t"
  87         " ssat r3, #16, r3                \n\t"
  88
  89         " pkhbt r0, r3, r2, LSL #16       \n\t"
  90         " str  r0, [%0], #4               \n\t"
  91
  92         MOD_INC()
  93
  94         "4:                               \n\t"
  95         " movs %3, %3, LSR #1             \n\t"
  96         " beq  6f                         \n\t"
  97
  98         "5:                               \n\t"
  99         " ldrd r2, [r6], #8               \n\t" /* 4 samples at a time */
 100         " ldrd r4, [r6], #8               \n\t"
 101         " ldrd r0, [%0]                   \n\t"
 102
 103         " smulwt r2, r2, r0               \n\t"
 104         " smulwb r3, r3, r0               \n\t"
 105         " smulwt r4, r4, r1               \n\t"
 106         " smulwb r5, r5, r1               \n\t"
 107
 108         " ssat r2, #16, r2                \n\t"
 109         " ssat r3, #16, r3                \n\t"
 110         " ssat r4, #16, r4                \n\t"
 111         " ssat r5, #16, r5                \n\t"
 112
 113         " pkhbt r0, r3, r2, LSL #16       \n\t"
 114         " pkhbt r1, r5, r4, LSL #16       \n\t"
 115         " strd  r0, [%0], #8              \n\t"
 116
 117         MOD_INC()
 118
 119         " subs %3, %3, #1                 \n\t"
 120         " bne 5b                          \n\t"
 121         "6:                               \n\t"
 122
 123         : "+r" (samples), "+r" (volumes), "+r" (ve), "+r" (length)
 124         :
 125         : "r6", "r5", "r4", "r3", "r2", "r1", "r0", "cc"
 126     );
 127 }
 128
 129 #undef RUN_TEST
 130
 131 #ifdef RUN_TEST
 132 #define CHANNELS 2
 133 #define SAMPLES 1022
 134 #define TIMES 1000
 135 #define PADDING 16
 136
 137 static void run_test (void) {
 138     int16_t samples[SAMPLES];
 139     int16_t samples_ref[SAMPLES];
 140     int16_t samples_orig[SAMPLES];
 141     int32_t volumes[CHANNELS + PADDING];
 142     int i, j, padding;
 143     pa_do_volume_func_t func;
 144     pa_usec_t start, stop;
 145
 146     func = pa_get_volume_func (PA_SAMPLE_S16NE);
 147
 148     printf ("checking ARM %zd\n", sizeof (samples));
 149
 150     pa_random (samples, sizeof (samples));
 151     memcpy (samples_ref, samples, sizeof (samples));
 152     memcpy (samples_orig, samples, sizeof (samples));
 153
 154     for (i = 0; i < CHANNELS; i++)
 155         volumes[i] = PA_CLAMP_VOLUME(rand() >> 1);
 156     for (padding = 0; padding < PADDING; padding++, i++)
 157         volumes[i] = volumes[padding];
 158
 159     func (samples_ref, volumes, CHANNELS, sizeof (samples));
 160     pa_volume_s16ne_arm (samples, volumes, CHANNELS, sizeof (samples));
 161     for (i = 0; i < SAMPLES; i++) {
 162         if (samples[i] != samples_ref[i]) {
 163             printf ("%d: %04x != %04x (%04x * %04x)\n", i, samples[i], samples_ref[i],
 164                   samples_orig[i], volumes[i % CHANNELS]);
 165         }
 166     }
 167
 168     start = pa_rtclock_now();
 169     for (j = 0; j < TIMES; j++) {
 170         memcpy (samples, samples_orig, sizeof (samples));
 171         pa_volume_s16ne_arm (samples, volumes, CHANNELS, sizeof (samples));
 172     }
 173     stop = pa_rtclock_now();
 174     pa_log_info("ARM: %llu usec.", (long long unsigned int) (stop - start));
 175
 176     start = pa_rtclock_now();
 177     for (j = 0; j < TIMES; j++) {
 178         memcpy (samples_ref, samples_orig, sizeof (samples));
 179         func (samples_ref, volumes, CHANNELS, sizeof (samples));
 180     }
 181     stop = pa_rtclock_now();
 182     pa_log_info("ref: %llu usec.", (long long unsigned int) (stop - start));
 183 }
 184 #endif
 185
 186 #endif /* defined (__arm__) && defined (HAVE_ARMV6) */
 187
 188
 189 void pa_volume_func_init_arm (pa_cpu_arm_flag_t flags) {
 190 #if defined (__arm__) && defined (HAVE_ARMV6)
 191     pa_log_info("Initialising ARM optimized functions.");
 192
 193 #ifdef RUN_TEST
 194     run_test ();
 195 #endif
 196
 197     pa_set_volume_func (PA_SAMPLE_S16NE,     (pa_do_volume_func_t) pa_volume_s16ne_arm);
 198 #endif /* defined (__arm__) && defined (HAVE_ARMV6) */
 199 }