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