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

   1 /***
   2   This file is part of PulseAudio.
   3
   4   Copyright 2004-2006 Lennart Poettering
   5   Copyright 2006 Pierre Ossman <ossman@cendio.se> for Cendio AB
   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 <stdio.h>
  28 #include <stdlib.h>
  29
  30 #include <pulsecore/macro.h>
  31 #include <pulsecore/endianmacros.h>
  32
  33 #include "cpu-x86.h"
  34 #include "sconv.h"
  35
  36 #if (!defined(__APPLE__) && !defined(__FreeBSD__) && defined (__i386__)) || defined (__amd64__)
  37
  38 static const PA_DECLARE_ALIGNED (16, float, scale[4]) = { 0x8000, 0x8000, 0x8000, 0x8000 };
  39
  40 static void pa_sconv_s16le_from_f32ne_sse(unsigned n, const float *a, int16_t *b) {
  41     pa_reg_x86 temp, i;
  42
  43     __asm__ __volatile__ (
  44         " movaps %5, %%xmm5             \n\t"
  45         " xor %0, %0                    \n\t"
  46
  47         " mov %4, %1                    \n\t"
  48         " sar $3, %1                    \n\t" /* 8 floats at a time */
  49         " cmp $0, %1                    \n\t"
  50         " je 2f                         \n\t"
  51
  52         "1:                             \n\t"
  53         " movups (%q2, %0, 2), %%xmm0   \n\t" /* read 8 floats */
  54         " movups 16(%q2, %0, 2), %%xmm2 \n\t"
  55         " mulps  %%xmm5, %%xmm0         \n\t" /* *= 0x8000 */
  56         " mulps  %%xmm5, %%xmm2         \n\t"
  57
  58         " cvtps2pi %%xmm0, %%mm0        \n\t" /* low part to int */
  59         " cvtps2pi %%xmm2, %%mm2        \n\t"
  60         " movhlps  %%xmm0, %%xmm0       \n\t" /* bring high part in position */
  61         " movhlps  %%xmm2, %%xmm2       \n\t"
  62         " cvtps2pi %%xmm0, %%mm1        \n\t" /* high part to int */
  63         " cvtps2pi %%xmm2, %%mm3        \n\t"
  64
  65         " packssdw %%mm1, %%mm0         \n\t" /* pack parts */
  66         " packssdw %%mm3, %%mm2         \n\t"
  67         " movq     %%mm0, (%q3, %0)     \n\t"
  68         " movq     %%mm2, 8(%q3, %0)    \n\t"
  69
  70         " add $16, %0                   \n\t"
  71         " dec %1                        \n\t"
  72         " jne 1b                        \n\t"
  73
  74         "2:                             \n\t"
  75         " mov %4, %1                    \n\t" /* prepare for leftovers */
  76         " and $7, %1                    \n\t"
  77         " je 5f                         \n\t"
  78
  79         "3:                             \n\t"
  80         " movss (%q2, %0, 2), %%xmm0    \n\t"
  81         " mulss  %%xmm5, %%xmm0         \n\t"
  82         " cvtss2si %%xmm0, %4           \n\t"
  83         " add $0x8000, %4               \n\t" /* check for saturation */
  84         " and $~0xffff, %4              \n\t"
  85         " cvtss2si %%xmm0, %4           \n\t"
  86         " je 4f                         \n\t"
  87         " sar $31, %4                   \n\t"
  88         " xor $0x7fff, %4               \n\t"
  89
  90         "4:                             \n\t"
  91         " movw  %w4, (%q3, %0)          \n\t" /* store leftover */
  92         " add $2, %0                    \n\t"
  93         " dec %1                        \n\t"
  94         " jne 3b                        \n\t"
  95
  96         "5:                             \n\t"
  97         " emms                          \n\t"
  98
  99         : "=&r" (i), "=&r" (temp)
 100         : "r" (a), "r" (b), "r" ((pa_reg_x86)n), "m" (*scale)
 101         : "cc", "memory"
 102     );
 103 }
 104
 105 static void pa_sconv_s16le_from_f32ne_sse2(unsigned n, const float *a, int16_t *b) {
 106     pa_reg_x86 temp, i;
 107
 108     __asm__ __volatile__ (
 109         " movaps %5, %%xmm5             \n\t"
 110         " xor %0, %0                    \n\t"
 111
 112         " mov %4, %1                    \n\t"
 113         " sar $3, %1                    \n\t" /* 8 floats at a time */
 114         " cmp $0, %1                    \n\t"
 115         " je 2f                         \n\t"
 116
 117         "1:                             \n\t"
 118         " movups (%q2, %0, 2), %%xmm0   \n\t" /* read 8 floats */
 119         " movups 16(%q2, %0, 2), %%xmm2 \n\t"
 120         " mulps  %%xmm5, %%xmm0         \n\t" /* *= 0x8000 */
 121         " mulps  %%xmm5, %%xmm2         \n\t"
 122
 123         " cvtps2dq %%xmm0, %%xmm0       \n\t"
 124         " cvtps2dq %%xmm2, %%xmm2       \n\t"
 125
 126         " packssdw %%xmm2, %%xmm0       \n\t"
 127         " movdqu   %%xmm0, (%q3, %0)    \n\t"
 128
 129         " add $16, %0                   \n\t"
 130         " dec %1                        \n\t"
 131         " jne 1b                        \n\t"
 132
 133         "2:                             \n\t"
 134         " mov %4, %1                    \n\t" /* prepare for leftovers */
 135         " and $7, %1                    \n\t"
 136         " je 5f                         \n\t"
 137
 138         "3:                             \n\t"
 139         " movss (%q2, %0, 2), %%xmm0    \n\t"
 140         " mulss  %%xmm5, %%xmm0         \n\t"
 141         " cvtss2si %%xmm0, %4           \n\t"
 142         " add $0x8000, %4               \n\t"
 143         " and $~0xffff, %4              \n\t" /* check for saturation */
 144         " cvtss2si %%xmm0, %4           \n\t"
 145         " je 4f                         \n\t"
 146         " sar $31, %4                   \n\t"
 147         " xor $0x7fff, %4               \n\t"
 148
 149         "4:                             \n\t"
 150         " movw  %w4, (%q3, %0)          \n\t" /* store leftover */
 151         " add $2, %0                    \n\t"
 152         " dec %1                        \n\t"
 153         " jne 3b                        \n\t"
 154
 155         "5:                             \n\t"
 156
 157         : "=&r" (i), "=&r" (temp)
 158         : "r" (a), "r" (b), "r" ((pa_reg_x86)n), "m" (*scale)
 159         : "cc", "memory"
 160     );
 161 }
 162
 163 #endif /* defined (__i386__) || defined (__amd64__) */
 164
 165 void pa_convert_func_init_sse(pa_cpu_x86_flag_t flags) {
 166 #if (!defined(__APPLE__) && !defined(__FreeBSD__) && defined (__i386__)) || defined (__amd64__)
 167
 168     if (flags & PA_CPU_X86_SSE2) {
 169         pa_log_info("Initialising SSE2 optimized conversions.");
 170         pa_set_convert_from_float32ne_function(PA_SAMPLE_S16LE, (pa_convert_func_t) pa_sconv_s16le_from_f32ne_sse2);
 171
 172     } else if (flags & PA_CPU_X86_SSE) {
 173         pa_log_info("Initialising SSE optimized conversions.");
 174         pa_set_convert_from_float32ne_function(PA_SAMPLE_S16LE, (pa_convert_func_t) pa_sconv_s16le_from_f32ne_sse);
 175     }
 176
 177 #endif /* defined (__i386__) || defined (__amd64__) */
 178 }