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 <pulsecore/log.h>
  29 #include <pulsecore/macro.h>
  30
  31 #include "cpu-x86.h"
  32 #include "remap.h"
  33
  34 #define LOAD_SAMPLES                                   \
  35                 " movdqu (%1), %%xmm0           \n\t"  \
  36                 " movdqu 16(%1), %%xmm2         \n\t"  \
  37                 " movdqu 32(%1), %%xmm4         \n\t"  \
  38                 " movdqu 48(%1), %%xmm6         \n\t"  \
  39                 " movdqa %%xmm0, %%xmm1         \n\t"  \
  40                 " movdqa %%xmm2, %%xmm3         \n\t"  \
  41                 " movdqa %%xmm4, %%xmm5         \n\t"  \
  42                 " movdqa %%xmm6, %%xmm7         \n\t"
  43
  44 #define UNPACK_SAMPLES(s)                              \
  45                 " punpckl"#s" %%xmm0, %%xmm0    \n\t"  \
  46                 " punpckh"#s" %%xmm1, %%xmm1    \n\t"  \
  47                 " punpckl"#s" %%xmm2, %%xmm2    \n\t"  \
  48                 " punpckh"#s" %%xmm3, %%xmm3    \n\t"  \
  49                 " punpckl"#s" %%xmm4, %%xmm4    \n\t"  \
  50                 " punpckh"#s" %%xmm5, %%xmm5    \n\t"  \
  51                 " punpckl"#s" %%xmm6, %%xmm6    \n\t"  \
  52                 " punpckh"#s" %%xmm7, %%xmm7    \n\t"
  53
  54 #define STORE_SAMPLES                                  \
  55                 " movdqu %%xmm0, (%0)           \n\t"  \
  56                 " movdqu %%xmm1, 16(%0)         \n\t"  \
  57                 " movdqu %%xmm2, 32(%0)         \n\t"  \
  58                 " movdqu %%xmm3, 48(%0)         \n\t"  \
  59                 " movdqu %%xmm4, 64(%0)         \n\t"  \
  60                 " movdqu %%xmm5, 80(%0)         \n\t"  \
  61                 " movdqu %%xmm6, 96(%0)         \n\t"  \
  62                 " movdqu %%xmm7, 112(%0)        \n\t"  \
  63                 " add $64, %1                   \n\t"  \
  64                 " add $128, %0                  \n\t"
  65
  66 #define HANDLE_SINGLE_dq()                             \
  67                 " movd (%1), %%xmm0             \n\t"  \
  68                 " punpckldq %%xmm0, %%xmm0      \n\t"  \
  69                 " movq %%xmm0, (%0)             \n\t"  \
  70                 " add $4, %1                    \n\t"  \
  71                 " add $8, %0                    \n\t"
  72
  73 #define HANDLE_SINGLE_wd()                             \
  74                 " movw (%1), %w3                \n\t"  \
  75                 " movd %3, %%xmm0               \n\t"  \
  76                 " punpcklwd %%xmm0, %%xmm0      \n\t"  \
  77                 " movd %%xmm0, (%0)             \n\t"  \
  78                 " add $2, %1                    \n\t"  \
  79                 " add $4, %0                    \n\t"
  80
  81 #define MONO_TO_STEREO(s,shift,mask)                   \
  82                 " mov %4, %2                    \n\t"  \
  83                 " sar $"#shift", %2             \n\t"  \
  84                 " cmp $0, %2                    \n\t"  \
  85                 " je 2f                         \n\t"  \
  86                 "1:                             \n\t"  \
  87                 LOAD_SAMPLES                           \
  88                 UNPACK_SAMPLES(s)                      \
  89                 STORE_SAMPLES                          \
  90                 " dec %2                        \n\t"  \
  91                 " jne 1b                        \n\t"  \
  92                 "2:                             \n\t"  \
  93                 " mov %4, %2                    \n\t"  \
  94                 " and $"#mask", %2              \n\t"  \
  95                 " je 4f                         \n\t"  \
  96                 "3:                             \n\t"  \
  97                 HANDLE_SINGLE_##s()                    \
  98                 " dec %2                        \n\t"  \
  99                 " jne 3b                        \n\t"  \
 100                 "4:                             \n\t"
 101
 102 #if defined (__i386__) || defined (__amd64__)
 103 static void remap_mono_to_stereo_sse2(pa_remap_t *m, void *dst, const void *src, unsigned n) {
 104     pa_reg_x86 temp, temp2;
 105
 106     switch (*m->format) {
 107         case PA_SAMPLE_FLOAT32NE:
 108         {
 109             __asm__ __volatile__ (
 110                 MONO_TO_STEREO(dq, 4, 15) /* do doubles to quads */
 111                 : "+r" (dst), "+r" (src), "=&r" (temp), "=&r" (temp2)
 112                 : "r" ((pa_reg_x86)n)
 113                 : "cc"
 114             );
 115             break;
 116         }
 117         case PA_SAMPLE_S16NE:
 118         {
 119             __asm__ __volatile__ (
 120                 MONO_TO_STEREO(wd, 5, 31) /* do words to doubles */
 121                 : "+r" (dst), "+r" (src), "=&r" (temp), "=&r" (temp2)
 122                 : "r" ((pa_reg_x86)n)
 123                 : "cc"
 124             );
 125             break;
 126         }
 127         default:
 128             pa_assert_not_reached();
 129     }
 130 }
 131
 132 /* set the function that will execute the remapping based on the matrices */
 133 static void init_remap_sse2(pa_remap_t *m) {
 134     unsigned n_oc, n_ic;
 135
 136     n_oc = m->o_ss->channels;
 137     n_ic = m->i_ss->channels;
 138
 139     /* find some common channel remappings, fall back to full matrix operation. */
 140     if (n_ic == 1 && n_oc == 2 &&
 141             m->map_table_f[0][0] >= 1.0 && m->map_table_f[1][0] >= 1.0) {
 142         m->do_remap = (pa_do_remap_func_t) remap_mono_to_stereo_sse2;
 143         pa_log_info("Using SSE mono to stereo remapping");
 144     }
 145 }
 146 #endif /* defined (__i386__) || defined (__amd64__) */
 147
 148 void pa_remap_func_init_sse(pa_cpu_x86_flag_t flags) {
 149 #if defined (__i386__) || defined (__amd64__)
 150
 151     if (flags & PA_CPU_X86_SSE2) {
 152         pa_log_info("Initialising SSE2 optimized remappers.");
 153         pa_set_init_remap_func ((pa_init_remap_func_t) init_remap_sse2);
 154     }
 155
 156 #endif /* defined (__i386__) || defined (__amd64__) */
 157 }