code.delx.au - pulseaudio/blob - src/pulsecore/remap_mmx.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 <string.h>
  28
  29 #include <pulse/sample.h>
  30 #include <pulsecore/log.h>
  31 #include <pulsecore/macro.h>
  32
  33 #include "cpu-x86.h"
  34 #include "remap.h"
  35
  36 #define LOAD_SAMPLES                                   \
  37                 " movq (%1), %%mm0              \n\t"  \
  38                 " movq 8(%1), %%mm2             \n\t"  \
  39                 " movq 16(%1), %%mm4            \n\t"  \
  40                 " movq 24(%1), %%mm6            \n\t"  \
  41                 " movq %%mm0, %%mm1             \n\t"  \
  42                 " movq %%mm2, %%mm3             \n\t"  \
  43                 " movq %%mm4, %%mm5             \n\t"  \
  44                 " movq %%mm6, %%mm7             \n\t"
  45
  46 #define UNPACK_SAMPLES(s)                              \
  47                 " punpckl"#s" %%mm0, %%mm0      \n\t"  \
  48                 " punpckh"#s" %%mm1, %%mm1      \n\t"  \
  49                 " punpckl"#s" %%mm2, %%mm2      \n\t"  \
  50                 " punpckh"#s" %%mm3, %%mm3      \n\t"  \
  51                 " punpckl"#s" %%mm4, %%mm4      \n\t"  \
  52                 " punpckh"#s" %%mm5, %%mm5      \n\t"  \
  53                 " punpckl"#s" %%mm6, %%mm6      \n\t"  \
  54                 " punpckh"#s" %%mm7, %%mm7      \n\t"
  55
  56 #define STORE_SAMPLES                                  \
  57                 " movq %%mm0, (%0)              \n\t"  \
  58                 " movq %%mm1, 8(%0)             \n\t"  \
  59                 " movq %%mm2, 16(%0)            \n\t"  \
  60                 " movq %%mm3, 24(%0)            \n\t"  \
  61                 " movq %%mm4, 32(%0)            \n\t"  \
  62                 " movq %%mm5, 40(%0)            \n\t"  \
  63                 " movq %%mm6, 48(%0)            \n\t"  \
  64                 " movq %%mm7, 56(%0)            \n\t"  \
  65                 " add $32, %1                   \n\t"  \
  66                 " add $64, %0                   \n\t"
  67
  68 #define HANDLE_SINGLE_dq()                            \
  69                 " movd (%1), %%mm0              \n\t"  \
  70                 " punpckldq %%mm0, %%mm0        \n\t"  \
  71                 " movq %%mm0, (%0)              \n\t"  \
  72                 " add $4, %1                    \n\t"  \
  73                 " add $8, %0                    \n\t"
  74
  75 #define HANDLE_SINGLE_wd()                             \
  76                 " movw (%1), %w3                \n\t"  \
  77                 " movd %3,  %%mm0               \n\t"  \
  78                 " punpcklwd %%mm0, %%mm0        \n\t"  \
  79                 " movd %%mm0, (%0)              \n\t"  \
  80                 " add $2, %1                    \n\t"  \
  81                 " add $4, %0                    \n\t"
  82
  83 #define MONO_TO_STEREO(s,shift,mask)                   \
  84                 " mov %4, %2                    \n\t"  \
  85                 " sar $"#shift", %2             \n\t"  \
  86                 " cmp $0, %2                    \n\t"  \
  87                 " je 2f                         \n\t"  \
  88                 "1:                             \n\t"  \
  89                 LOAD_SAMPLES                           \
  90                 UNPACK_SAMPLES(s)                      \
  91                 STORE_SAMPLES                          \
  92                 " dec %2                        \n\t"  \
  93                 " jne 1b                        \n\t"  \
  94                 "2:                             \n\t"  \
  95                 " mov %4, %2                    \n\t"  \
  96                 " and $"#mask", %2              \n\t"  \
  97                 " je 4f                         \n\t"  \
  98                 "3:                             \n\t"  \
  99                 HANDLE_SINGLE_##s()                    \
 100                 " dec %2                        \n\t"  \
 101                 " jne 3b                        \n\t"  \
 102                 "4:                             \n\t"  \
 103                 " emms                          \n\t"
 104
 105 #if defined (__i386__) || defined (__amd64__)
 106 static void remap_mono_to_stereo_mmx (pa_remap_t *m, void *dst, const void *src, unsigned n) {
 107     pa_reg_x86 temp, temp2;
 108
 109     switch (*m->format) {
 110         case PA_SAMPLE_FLOAT32NE:
 111         {
 112             __asm__ __volatile__ (
 113                 MONO_TO_STEREO(dq,3,7) /* do doubles to quads */
 114                 : "+r" (dst), "+r" (src), "=&r" (temp), "=&r" (temp2)
 115                 : "r" ((pa_reg_x86)n)
 116                 : "cc"
 117             );
 118             break;
 119         }
 120         case PA_SAMPLE_S16NE:
 121         {
 122             __asm__ __volatile__ (
 123                 MONO_TO_STEREO(wd,4,15) /* do words to doubles */
 124                 : "+r" (dst), "+r" (src), "=&r" (temp), "=&r" (temp2)
 125                 : "r" ((pa_reg_x86)n)
 126                 : "cc"
 127             );
 128             break;
 129         }
 130         default:
 131             pa_assert_not_reached();
 132     }
 133 }
 134
 135 /* set the function that will execute the remapping based on the matrices */
 136 static void init_remap_mmx (pa_remap_t *m) {
 137     unsigned n_oc, n_ic;
 138
 139     n_oc = m->o_ss->channels;
 140     n_ic = m->i_ss->channels;
 141
 142     /* find some common channel remappings, fall back to full matrix operation. */
 143     if (n_ic == 1 && n_oc == 2 &&
 144             m->map_table_f[0][0] >= 1.0 && m->map_table_f[1][0] >= 1.0) {
 145         m->do_remap = (pa_do_remap_func_t) remap_mono_to_stereo_mmx;
 146         pa_log_info("Using MMX mono to stereo remapping");
 147     }
 148 }
 149 #endif /* defined (__i386__) || defined (__amd64__) */
 150
 151 void pa_remap_func_init_mmx (pa_cpu_x86_flag_t flags) {
 152 #if defined (__i386__) || defined (__amd64__)
 153
 154     if (flags & PA_CPU_X86_MMX) {
 155         pa_log_info("Initialising MMX optimized remappers.");
 156
 157         pa_set_init_remap_func ((pa_init_remap_func_t) init_remap_mmx);
 158     }
 159
 160 #endif /* defined (__i386__) || defined (__amd64__) */
 161 }