code.delx.au - pulseaudio/blob - src/pulsecore/remap_sse.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.com>
   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/sample.h>
  28 #include <pulse/volume.h>
  29 #include <pulsecore/log.h>
  30 #include <pulsecore/macro.h>
  31
  32 #include "cpu-x86.h"
  33 #include "remap.h"
  34
  35 #define LOAD_SAMPLES                                   \
  36                 " movdqu (%1), %%xmm0           \n\t"  \
  37                 " movdqu 16(%1), %%xmm2         \n\t"  \
  38                 " movdqu 32(%1), %%xmm4         \n\t"  \
  39                 " movdqu 48(%1), %%xmm6         \n\t"  \
  40                 " movdqa %%xmm0, %%xmm1         \n\t"  \
  41                 " movdqa %%xmm2, %%xmm3         \n\t"  \
  42                 " movdqa %%xmm4, %%xmm5         \n\t"  \
  43                 " movdqa %%xmm6, %%xmm7         \n\t"
  44
  45 #define UNPACK_SAMPLES(s)                              \
  46                 " punpckl"#s" %%xmm0, %%xmm0    \n\t"  \
  47                 " punpckh"#s" %%xmm1, %%xmm1    \n\t"  \
  48                 " punpckl"#s" %%xmm2, %%xmm2    \n\t"  \
  49                 " punpckh"#s" %%xmm3, %%xmm3    \n\t"  \
  50                 " punpckl"#s" %%xmm4, %%xmm4    \n\t"  \
  51                 " punpckh"#s" %%xmm5, %%xmm5    \n\t"  \
  52                 " punpckl"#s" %%xmm6, %%xmm6    \n\t"  \
  53                 " punpckh"#s" %%xmm7, %%xmm7    \n\t"
  54
  55 #define STORE_SAMPLES                                  \
  56                 " movdqu %%xmm0, (%0)           \n\t"  \
  57                 " movdqu %%xmm1, 16(%0)         \n\t"  \
  58                 " movdqu %%xmm2, 32(%0)         \n\t"  \
  59                 " movdqu %%xmm3, 48(%0)         \n\t"  \
  60                 " movdqu %%xmm4, 64(%0)         \n\t"  \
  61                 " movdqu %%xmm5, 80(%0)         \n\t"  \
  62                 " movdqu %%xmm6, 96(%0)         \n\t"  \
  63                 " movdqu %%xmm7, 112(%0)        \n\t"  \
  64                 " add $64, %1                   \n\t"  \
  65                 " add $128, %0                  \n\t"
  66
  67 #define HANDLE_SINGLE_dq()                             \
  68                 " movd (%1), %%xmm0             \n\t"  \
  69                 " punpckldq %%xmm0, %%xmm0      \n\t"  \
  70                 " movq %%xmm0, (%0)             \n\t"  \
  71                 " add $4, %1                    \n\t"  \
  72                 " add $8, %0                    \n\t"
  73
  74 #define HANDLE_SINGLE_wd()                             \
  75                 " movw (%1), %w3                \n\t"  \
  76                 " movd %3, %%xmm0               \n\t"  \
  77                 " punpcklwd %%xmm0, %%xmm0      \n\t"  \
  78                 " movd %%xmm0, (%0)             \n\t"  \
  79                 " add $2, %1                    \n\t"  \
  80                 " add $4, %0                    \n\t"
  81
  82 #define MONO_TO_STEREO(s,shift,mask)                   \
  83                 " mov %4, %2                    \n\t"  \
  84                 " sar $"#shift", %2             \n\t"  \
  85                 " cmp $0, %2                    \n\t"  \
  86                 " je 2f                         \n\t"  \
  87                 "1:                             \n\t"  \
  88                 LOAD_SAMPLES                           \
  89                 UNPACK_SAMPLES(s)                      \
  90                 STORE_SAMPLES                          \
  91                 " dec %2                        \n\t"  \
  92                 " jne 1b                        \n\t"  \
  93                 "2:                             \n\t"  \
  94                 " mov %4, %2                    \n\t"  \
  95                 " and $"#mask", %2              \n\t"  \
  96                 " je 4f                         \n\t"  \
  97                 "3:                             \n\t"  \
  98                 HANDLE_SINGLE_##s()                    \
  99                 " dec %2                        \n\t"  \
 100                 " jne 3b                        \n\t"  \
 101                 "4:                             \n\t"
 102
 103 #if defined (__i386__) || defined (__amd64__)
 104 static void remap_mono_to_stereo_s16ne_sse2(pa_remap_t *m, int16_t *dst, const int16_t *src, unsigned n) {
 105     pa_reg_x86 temp, temp2;
 106
 107     __asm__ __volatile__ (
 108         MONO_TO_STEREO(wd, 5, 31) /* do words to doubles */
 109         : "+r" (dst), "+r" (src), "=&r" (temp), "=&r" (temp2)
 110         : "r" ((pa_reg_x86)n)
 111         : "cc"
 112     );
 113 }
 114
 115 static void remap_mono_to_stereo_float32ne_sse2(pa_remap_t *m, float *dst, const float *src, unsigned n) {
 116     pa_reg_x86 temp, temp2;
 117
 118     __asm__ __volatile__ (
 119         MONO_TO_STEREO(dq, 4, 15) /* do doubles to quads */
 120         : "+r" (dst), "+r" (src), "=&r" (temp), "=&r" (temp2)
 121         : "r" ((pa_reg_x86)n)
 122         : "cc"
 123     );
 124 }
 125
 126 /* set the function that will execute the remapping based on the matrices */
 127 static void init_remap_sse2(pa_remap_t *m) {
 128     unsigned n_oc, n_ic;
 129
 130     n_oc = m->o_ss.channels;
 131     n_ic = m->i_ss.channels;
 132
 133     /* find some common channel remappings, fall back to full matrix operation. */
 134     if (n_ic == 1 && n_oc == 2 &&
 135             m->map_table_i[0][0] == 0x10000 && m->map_table_i[1][0] == 0x10000) {
 136
 137         pa_log_info("Using SSE2 mono to stereo remapping");
 138         pa_set_remap_func(m, (pa_do_remap_func_t) remap_mono_to_stereo_s16ne_sse2,
 139             (pa_do_remap_func_t) remap_mono_to_stereo_float32ne_sse2);
 140     }
 141 }
 142 #endif /* defined (__i386__) || defined (__amd64__) */
 143
 144 void pa_remap_func_init_sse(pa_cpu_x86_flag_t flags) {
 145 #if defined (__i386__) || defined (__amd64__)
 146
 147     if (flags & PA_CPU_X86_SSE2) {
 148         pa_log_info("Initialising SSE2 optimized remappers.");
 149         pa_set_init_remap_func ((pa_init_remap_func_t) init_remap_sse2);
 150     }
 151
 152 #endif /* defined (__i386__) || defined (__amd64__) */
 153 }