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 <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 pa_do_volume_func_t _volume_ref;
  44
  45 static void pa_volume_s16ne_arm(int16_t *samples, const int32_t *volumes, unsigned channels, unsigned length) {
  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     const int32_t *ve = volumes + (channels == 3 ? 6 : PA_MAX (4U, channels));
  50     unsigned rem = PA_ALIGN((size_t) samples) - (size_t) samples;
  51
  52     /* Make sure we're word-aligned, else performance _really_ sucks */
  53     if (rem) {
  54         _volume_ref(samples, volumes, channels, rem < length ? rem : length);
  55
  56         if (rem < length) {
  57             length -= rem;
  58             samples += rem / sizeof(*samples);
  59         } else
  60             return; /* we're done */
  61     }
  62
  63     __asm__ __volatile__ (
  64         " mov r6, %4                      \n\t" /* r6 = volumes + rem */
  65         " mov %3, %3, LSR #1              \n\t" /* length /= sizeof (int16_t) */
  66
  67         " cmp %3, #4                      \n\t" /* check for 4+ samples */
  68         " blt 2f                          \n\t"
  69
  70         /* See final case for how the multiplication works */
  71
  72         "1:                               \n\t"
  73         " ldrd r2, [r6], #8               \n\t" /* 4 samples at a time */
  74         " ldrd r4, [r6], #8               \n\t"
  75         " ldrd r0, [%0]                   \n\t"
  76
  77 #ifdef WORDS_BIGENDIAN
  78         " smulwt r2, r2, r0               \n\t"
  79         " smulwb r3, r3, r0               \n\t"
  80         " smulwt r4, r4, r1               \n\t"
  81         " smulwb r5, r5, r1               \n\t"
  82 #else
  83         " smulwb r2, r2, r0               \n\t"
  84         " smulwt r3, r3, r0               \n\t"
  85         " smulwb r4, r4, r1               \n\t"
  86         " smulwt r5, r5, r1               \n\t"
  87 #endif
  88
  89         " ssat r2, #16, r2                \n\t"
  90         " ssat r3, #16, r3                \n\t"
  91         " ssat r4, #16, r4                \n\t"
  92         " ssat r5, #16, r5                \n\t"
  93
  94 #ifdef WORDS_BIGENDIAN
  95         " pkhbt r0, r3, r2, LSL #16       \n\t"
  96         " pkhbt r1, r5, r4, LSL #16       \n\t"
  97 #else
  98         " pkhbt r0, r2, r3, LSL #16       \n\t"
  99         " pkhbt r1, r4, r5, LSL #16       \n\t"
 100 #endif
 101         " strd  r0, [%0], #8              \n\t"
 102
 103         MOD_INC()
 104
 105         " subs %3, %3, #4                 \n\t"
 106         " cmp %3, #4                      \n\t"
 107         " bge 1b                          \n\t"
 108
 109         "2:                               \n\t"
 110         " cmp %3, #2                      \n\t"
 111         " blt 3f                          \n\t"
 112
 113         " ldrd r2, [r6], #8               \n\t" /* 2  samples at a time */
 114         " ldr  r0, [%0]                   \n\t"
 115
 116 #ifdef WORDS_BIGENDIAN
 117         " smulwt r2, r2, r0               \n\t"
 118         " smulwb r3, r3, r0               \n\t"
 119 #else
 120         " smulwb r2, r2, r0               \n\t"
 121         " smulwt r3, r3, r0               \n\t"
 122 #endif
 123
 124         " ssat r2, #16, r2                \n\t"
 125         " ssat r3, #16, r3                \n\t"
 126
 127 #ifdef WORDS_BIGENDIAN
 128         " pkhbt r0, r3, r2, LSL #16       \n\t"
 129 #else
 130         " pkhbt r0, r2, r3, LSL #16       \n\t"
 131 #endif
 132         " str  r0, [%0], #4               \n\t"
 133
 134         MOD_INC()
 135
 136         " subs %3, %3, #2                 \n\t"
 137
 138         "3:                               \n\t" /* check for odd # of samples */
 139         " cmp %3, #1                      \n\t"
 140         " bne 4f                          \n\t"
 141
 142         " ldr  r0, [r6], #4               \n\t" /* r0 = volume */
 143         " ldrh r2, [%0]                   \n\t" /* r2 = sample */
 144
 145         " smulwb r0, r0, r2               \n\t" /* r0 = (r0 * r2) >> 16 */
 146         " ssat r0, #16, r0                \n\t" /* r0 = PA_CLAMP(r0, 0x7FFF) */
 147
 148         " strh r0, [%0], #2               \n\t" /* sample = r0 */
 149
 150         "4:                               \n\t"
 151
 152         : "+r" (samples), "+r" (volumes), "+r" (ve), "+r" (length)
 153         : "r" (volumes + ((rem / sizeof(*samples)) % channels))
 154         : "r6", "r5", "r4", "r3", "r2", "r1", "r0", "cc"
 155     );
 156 }
 157
 158 #endif /* defined (__arm__) && defined (HAVE_ARMV6) */
 159
 160 void pa_volume_func_init_arm(pa_cpu_arm_flag_t flags) {
 161 #if defined (__arm__) && defined (HAVE_ARMV6)
 162     pa_log_info("Initialising ARM optimized volume functions.");
 163
 164     if (!_volume_ref)
 165         _volume_ref = pa_get_volume_func(PA_SAMPLE_S16NE);
 166
 167     pa_set_volume_func(PA_SAMPLE_S16NE, (pa_do_volume_func_t) pa_volume_s16ne_arm);
 168 #endif /* defined (__arm__) && defined (HAVE_ARMV6) */
 169 }