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 <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                 " movdqu (%1), %%xmm0           \n\t"  \
  38                 " movdqu 16(%1), %%xmm2         \n\t"  \
  39                 " movdqu 32(%1), %%xmm4         \n\t"  \
  40                 " movdqu 48(%1), %%xmm6         \n\t"  \
  41                 " movdqa %%xmm0, %%xmm1         \n\t"  \
  42                 " movdqa %%xmm2, %%xmm3         \n\t"  \
  43                 " movdqa %%xmm4, %%xmm5         \n\t"  \
  44                 " movdqa %%xmm6, %%xmm7         \n\t"
  45
  46 #define UNPACK_SAMPLES(s)                              \
  47                 " punpckl"#s" %%xmm0, %%xmm0    \n\t"  \
  48                 " punpckh"#s" %%xmm1, %%xmm1    \n\t"  \
  49                 " punpckl"#s" %%xmm2, %%xmm2    \n\t"  \
  50                 " punpckh"#s" %%xmm3, %%xmm3    \n\t"  \
  51                 " punpckl"#s" %%xmm4, %%xmm4    \n\t"  \
  52                 " punpckh"#s" %%xmm5, %%xmm5    \n\t"  \
  53                 " punpckl"#s" %%xmm6, %%xmm6    \n\t"  \
  54                 " punpckh"#s" %%xmm7, %%xmm7    \n\t"
  55
  56 #define STORE_SAMPLES                                  \
  57                 " movdqu %%xmm0, (%0)           \n\t"  \
  58                 " movdqu %%xmm1, 16(%0)         \n\t"  \
  59                 " movdqu %%xmm2, 32(%0)         \n\t"  \
  60                 " movdqu %%xmm3, 48(%0)         \n\t"  \
  61                 " movdqu %%xmm4, 64(%0)         \n\t"  \
  62                 " movdqu %%xmm5, 80(%0)         \n\t"  \
  63                 " movdqu %%xmm6, 96(%0)         \n\t"  \
  64                 " movdqu %%xmm7, 112(%0)        \n\t"  \
  65                 " add $64, %1                   \n\t"  \
  66                 " add $128, %0                  \n\t"
  67
  68 #define HANDLE_SINGLE_dq()                             \
  69                 " movd (%1), %%xmm0             \n\t"  \
  70                 " punpckldq %%xmm0, %%xmm0      \n\t"  \
  71                 " movq %%xmm0, (%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, %%xmm0               \n\t"  \
  78                 " punpcklwd %%xmm0, %%xmm0      \n\t"  \
  79                 " movd %%xmm0, (%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
 104 #if defined (__i386__) || defined (__amd64__)
 105 static void remap_mono_to_stereo_sse2 (pa_remap_t *m, void *dst, const void *src, unsigned n) {
 106     pa_reg_x86 temp, temp2;
 107
 108     switch (*m->format) {
 109         case PA_SAMPLE_FLOAT32NE:
 110         {
 111             __asm__ __volatile__ (
 112                 MONO_TO_STEREO(dq, 4, 15) /* do doubles to quads */
 113                 : "+r" (dst), "+r" (src), "=&r" (temp), "=&r" (temp2)
 114                 : "r" ((pa_reg_x86)n)
 115                 : "cc"
 116             );
 117             break;
 118         }
 119         case PA_SAMPLE_S16NE:
 120         {
 121             __asm__ __volatile__ (
 122                 MONO_TO_STEREO(wd, 5, 31) /* do words to doubles */
 123                 : "+r" (dst), "+r" (src), "=&r" (temp), "=&r" (temp2)
 124                 : "r" ((pa_reg_x86)n)
 125                 : "cc"
 126             );
 127             break;
 128         }
 129         default:
 130             pa_assert_not_reached();
 131     }
 132 }
 133
 134 /* set the function that will execute the remapping based on the matrices */
 135 static void init_remap_sse2 (pa_remap_t *m) {
 136     unsigned n_oc, n_ic;
 137
 138     n_oc = m->o_ss->channels;
 139     n_ic = m->i_ss->channels;
 140
 141     /* find some common channel remappings, fall back to full matrix operation. */
 142     if (n_ic == 1 && n_oc == 2 &&
 143             m->map_table_f[0][0] >= 1.0 && m->map_table_f[1][0] >= 1.0) {
 144         m->do_remap = (pa_do_remap_func_t) remap_mono_to_stereo_sse2;
 145         pa_log_info("Using SSE mono to stereo remapping");
 146     }
 147 }
 148 #endif /* defined (__i386__) || defined (__amd64__) */
 149
 150 void pa_remap_func_init_sse (pa_cpu_x86_flag_t flags) {
 151 #if defined (__i386__) || defined (__amd64__)
 152
 153     if (flags & PA_CPU_X86_SSE2) {
 154         pa_log_info("Initialising SSE2 optimized remappers.");
 155         pa_set_init_remap_func ((pa_init_remap_func_t) init_remap_sse2);
 156     }
 157
 158 #endif /* defined (__i386__) || defined (__amd64__) */
 159 }