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

   1 /***
   2   This file is part of PulseAudio.
   3
   4   Copyright 2013 Peter Meerwald <pmeerw@pmeerw.net>
   5
   6   PulseAudio is free software; you can redistribute it and/or modify
   7   it under the terms of the GNU Lesser General Public License as published
   8   by the Free Software Foundation; either version 2.1 of the License,
   9   or (at your option) any later version.
  10
  11   PulseAudio is distributed in the hope that it will be useful, but
  12   WITHOUT ANY WARRANTY; without even the implied warranty of
  13   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  14   General Public License for more details.
  15 ***/
  16
  17 #ifdef HAVE_CONFIG_H
  18 #include <config.h>
  19 #endif
  20
  21 #include <pulsecore/macro.h>
  22 #include <pulsecore/endianmacros.h>
  23
  24 #include "cpu-arm.h"
  25 #include "mix.h"
  26
  27 #include <arm_neon.h>
  28
  29 static pa_do_mix_func_t fallback;
  30
  31 /* special case: mix s16ne streams, 2 channels each */
  32 static void pa_mix_ch2_s16ne_neon(pa_mix_info streams[], unsigned nstreams, uint8_t *data, unsigned length) {
  33     const unsigned mask = sizeof(int16_t) * 8 - 1;
  34     const uint8_t *end = data + (length & ~mask);
  35
  36     while (data < end) {
  37         int32x4_t sum0, sum1;
  38         unsigned i;
  39
  40         __asm__ __volatile__ (
  41             "veor.s32 %q[sum0], %q[sum0]     \n\t"
  42             "veor.s32 %q[sum1], %q[sum1]     \n\t"
  43             : [sum0] "=w" (sum0), [sum1] "=w" (sum1)
  44             :
  45             : "cc" /* clobber list */
  46         );
  47
  48         for (i = 0; i < nstreams; i++) {
  49             pa_mix_info *m = streams + i;
  50             int32_t cv0 = m->linear[0].i;
  51             int32_t cv1 = m->linear[1].i;
  52
  53             __asm__ __volatile__ (
  54                 "vld2.s16    {d0,d2}, [%[ptr]]!      \n\t"
  55                 "vmov.s32    d4[0], %[cv0]           \n\t"
  56                 "vmov.s32    d4[1], %[cv1]           \n\t"
  57                 "vshll.s16   q0, d0, #15             \n\t"
  58                 "vshll.s16   q1, d2, #15             \n\t"
  59                 "vqdmulh.s32 q0, q0, d4[0]           \n\t"
  60                 "vqdmulh.s32 q1, q1, d4[1]           \n\t"
  61                 "vqadd.s32   %q[sum0], %q[sum0], q0  \n\t"
  62                 "vqadd.s32   %q[sum1], %q[sum1], q1  \n\t"
  63                 : [ptr] "+r" (m->ptr), [sum0] "+w" (sum0), [sum1] "+w" (sum1)
  64                 : [cv0] "r" (cv0), [cv1] "r" (cv1)
  65                 : "memory", "cc", "q0", "q1", "d4" /* clobber list */
  66             );
  67         }
  68
  69         __asm__ __volatile__ (
  70             "vqmovn.s32 d0, %q[sum0]         \n\t"
  71             "vqmovn.s32 d1, %q[sum1]         \n\t"
  72             "vst2.s16   {d0,d1}, [%[data]]!  \n\t"
  73             : [data] "+r" (data)
  74             : [sum0] "w" (sum0), [sum1] "w" (sum1)
  75             : "memory", "cc", "q0" /* clobber list */
  76         );
  77     }
  78
  79     fallback(streams, nstreams, 2, data, length & mask);
  80 }
  81
  82 static void pa_mix_s16ne_neon(pa_mix_info streams[], unsigned nstreams, unsigned nchannels, void *data, unsigned length) {
  83     if (nchannels == 2)
  84         pa_mix_ch2_s16ne_neon(streams, nstreams, data, length);
  85     else
  86         fallback(streams, nstreams, nchannels, data, length);
  87 }
  88
  89 void pa_mix_func_init_neon(pa_cpu_arm_flag_t flags) {
  90     pa_log_info("Initialising ARM NEON optimized mixing functions.");
  91
  92     fallback = pa_get_mix_func(PA_SAMPLE_S16NE);
  93     pa_set_mix_func(PA_SAMPLE_S16NE, (pa_do_mix_func_t) pa_mix_s16ne_neon);
  94 }