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 <pulsecore/random.h>
  28 #include <pulsecore/macro.h>
  29 #include <pulsecore/endianmacros.h>
  30
  31 #include "cpu-arm.h"
  32
  33 #include "sample-util.h"
  34
  35 #if defined (__arm__) && defined (HAVE_ARMV6)
  36
  37 #define MOD_INC() \
  38     " subs  r0, r6, %2              \n\t" \
  39     " itt cs                        \n\t" \
  40     " addcs r0, %1                  \n\t" \
  41     " movcs r6, r0                  \n\t"
  42
  43 static void pa_volume_s16ne_arm(int16_t *samples, int32_t *volumes, unsigned channels, unsigned length) {
  44     int32_t *ve;
  45
  46     /* Channels must be at least 4, and always a multiple of the original number.
  47      * This is also the max amount we overread the volume array, which should
  48      * have enough padding. */
  49     channels = channels == 3 ? 6 : PA_MAX (4U, channels);
  50     ve = volumes + channels;
  51
  52     __asm__ __volatile__ (
  53         " mov r6, %1                      \n\t"
  54         " mov %3, %3, LSR #1              \n\t" /* length /= sizeof (int16_t) */
  55         " tst %3, #1                      \n\t" /* check for odd samples */
  56         " beq  2f                         \n\t"
  57
  58         "1:                               \n\t"
  59         " ldr  r0, [r6], #4               \n\t" /* odd samples volumes */
  60         " ldrh r2, [%0]                   \n\t"
  61
  62         " smulwb r0, r0, r2               \n\t"
  63         " ssat r0, #16, r0                \n\t"
  64
  65         " strh r0, [%0], #2               \n\t"
  66
  67         MOD_INC()
  68
  69         "2:                               \n\t"
  70         " mov %3, %3, LSR #1              \n\t"
  71         " tst %3, #1                      \n\t" /* check for odd samples */
  72         " beq  4f                         \n\t"
  73
  74         "3:                               \n\t"
  75         " ldrd r2, [r6], #8               \n\t" /* 2 samples at a time */
  76         " ldr  r0, [%0]                   \n\t"
  77
  78         " smulwt r2, r2, r0               \n\t"
  79         " smulwb r3, r3, r0               \n\t"
  80
  81         " ssat r2, #16, r2                \n\t"
  82         " ssat r3, #16, r3                \n\t"
  83
  84         " pkhbt r0, r3, r2, LSL #16       \n\t"
  85         " str  r0, [%0], #4               \n\t"
  86
  87         MOD_INC()
  88
  89         "4:                               \n\t"
  90         " movs %3, %3, LSR #1             \n\t"
  91         " beq  6f                         \n\t"
  92
  93         "5:                               \n\t"
  94         " ldrd r2, [r6], #8               \n\t" /* 4 samples at a time */
  95         " ldrd r4, [r6], #8               \n\t"
  96         " ldrd r0, [%0]                   \n\t"
  97
  98         " smulwt r2, r2, r0               \n\t"
  99         " smulwb r3, r3, r0               \n\t"
 100         " smulwt r4, r4, r1               \n\t"
 101         " smulwb r5, r5, r1               \n\t"
 102
 103         " ssat r2, #16, r2                \n\t"
 104         " ssat r3, #16, r3                \n\t"
 105         " ssat r4, #16, r4                \n\t"
 106         " ssat r5, #16, r5                \n\t"
 107
 108         " pkhbt r0, r3, r2, LSL #16       \n\t"
 109         " pkhbt r1, r5, r4, LSL #16       \n\t"
 110         " strd  r0, [%0], #8              \n\t"
 111
 112         MOD_INC()
 113
 114         " subs %3, %3, #1                 \n\t"
 115         " bne 5b                          \n\t"
 116         "6:                               \n\t"
 117
 118         : "+r" (samples), "+r" (volumes), "+r" (ve), "+r" (length)
 119         :
 120         : "r6", "r5", "r4", "r3", "r2", "r1", "r0", "cc"
 121     );
 122 }
 123
 124 #undef RUN_TEST
 125
 126 #ifdef RUN_TEST
 127 #define CHANNELS 2
 128 #define SAMPLES 1022
 129 #define TIMES 1000
 130 #define TIMES2 100
 131 #define PADDING 16
 132
 133 static void run_test(void) {
 134     int16_t samples[SAMPLES];
 135     int16_t samples_ref[SAMPLES];
 136     int16_t samples_orig[SAMPLES];
 137     int32_t volumes[CHANNELS + PADDING];
 138     int i, j, padding;
 139     pa_do_volume_func_t func;
 140     pa_usec_t start, stop;
 141     int k;
 142     pa_usec_t min = INT_MAX, max = 0;
 143     double s1 = 0, s2 = 0;
 144
 145     func = pa_get_volume_func(PA_SAMPLE_S16NE);
 146
 147     printf("checking ARM %zd\n", sizeof(samples));
 148
 149     pa_random(samples, sizeof(samples));
 150     memcpy(samples_ref, samples, sizeof(samples));
 151     memcpy(samples_orig, samples, sizeof(samples));
 152
 153     for (i = 0; i < CHANNELS; i++)
 154         volumes[i] = PA_CLAMP_VOLUME(rand() >> 1);
 155     for (padding = 0; padding < PADDING; padding++, i++)
 156         volumes[i] = volumes[padding];
 157
 158     func(samples_ref, volumes, CHANNELS, sizeof(samples));
 159     pa_volume_s16ne_arm(samples, volumes, CHANNELS, sizeof(samples));
 160     for (i = 0; i < SAMPLES; i++) {
 161         if (samples[i] != samples_ref[i]) {
 162             printf ("%d: %04x != %04x (%04x * %04x)\n", i, samples[i], samples_ref[i],
 163                       samples_orig[i], volumes[i % CHANNELS]);
 164         }
 165     }
 166
 167     for (k = 0; k < TIMES2; k++) {
 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
 175         if (min > (stop - start)) min = stop - start;
 176         if (max < (stop - start)) max = stop - start;
 177         s1 += stop - start;
 178         s2 += (stop - start) * (stop - start);
 179     }
 180     pa_log_info("ARM: %llu usec (min = %llu, max = %llu, stddev = %g).", (long long unsigned int)s1,
 181             (long long unsigned int)min, (long long unsigned int)max, sqrt(TIMES2 * s2 - s1 * s1) / TIMES2);
 182
 183     min = INT_MAX; max = 0;
 184     s1 = s2 = 0;
 185     for (k = 0; k < TIMES2; k++) {
 186         start = pa_rtclock_now();
 187         for (j = 0; j < TIMES; j++) {
 188             memcpy(samples_ref, samples_orig, sizeof(samples));
 189             func(samples_ref, volumes, CHANNELS, sizeof(samples));
 190         }
 191         stop = pa_rtclock_now();
 192
 193         if (min > (stop - start)) min = stop - start;
 194         if (max < (stop - start)) max = stop - start;
 195         s1 += stop - start;
 196         s2 += (stop - start) * (stop - start);
 197     }
 198     pa_log_info("ref: %llu usec (min = %llu, max = %llu, stddev = %g).", (long long unsigned int)s1,
 199             (long long unsigned int)min, (long long unsigned int)max, sqrt(TIMES2 * s2 - s1 * s1) / TIMES2);
 200
 201     pa_assert_se(memcmp(samples_ref, samples, sizeof(samples)) == 0);
 202 }
 203 #endif
 204
 205 #endif /* defined (__arm__) && defined (HAVE_ARMV6) */
 206
 207
 208 void pa_volume_func_init_arm(pa_cpu_arm_flag_t flags) {
 209 #if defined (__arm__) && defined (HAVE_ARMV6)
 210     pa_log_info("Initialising ARM optimized functions.");
 211
 212 #ifdef RUN_TEST
 213     run_test();
 214 #endif
 215
 216     pa_set_volume_func(PA_SAMPLE_S16NE, (pa_do_volume_func_t) pa_volume_s16ne_arm);
 217 #endif /* defined (__arm__) && defined (HAVE_ARMV6) */
 218 }