git.sesse.net Git - ffmpeg/blob - libavcodec/x86/mlpdsp.c

   1 /*
   2  * MLP DSP functions x86-optimized
   3  * Copyright (c) 2009 Ramiro Polla
   4  *
   5  * This file is part of Libav.
   6  *
   7  * Libav is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2.1 of the License, or (at your option) any later version.
  11  *
  12  * Libav is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with Libav; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20  */
  21
  22 #include "libavutil/attributes.h"
  23 #include "libavutil/internal.h"
  24 #include "libavutil/x86/asm.h"
  25 #include "libavcodec/mlpdsp.h"
  26 #include "libavcodec/mlp.h"
  27
  28 #if HAVE_7REGS && HAVE_INLINE_ASM
  29
  30 extern char ff_mlp_firorder_8;
  31 extern char ff_mlp_firorder_7;
  32 extern char ff_mlp_firorder_6;
  33 extern char ff_mlp_firorder_5;
  34 extern char ff_mlp_firorder_4;
  35 extern char ff_mlp_firorder_3;
  36 extern char ff_mlp_firorder_2;
  37 extern char ff_mlp_firorder_1;
  38 extern char ff_mlp_firorder_0;
  39
  40 extern char ff_mlp_iirorder_4;
  41 extern char ff_mlp_iirorder_3;
  42 extern char ff_mlp_iirorder_2;
  43 extern char ff_mlp_iirorder_1;
  44 extern char ff_mlp_iirorder_0;
  45
  46 static const void *firtable[9] = { &ff_mlp_firorder_0, &ff_mlp_firorder_1,
  47                                    &ff_mlp_firorder_2, &ff_mlp_firorder_3,
  48                                    &ff_mlp_firorder_4, &ff_mlp_firorder_5,
  49                                    &ff_mlp_firorder_6, &ff_mlp_firorder_7,
  50                                    &ff_mlp_firorder_8 };
  51 static const void *iirtable[5] = { &ff_mlp_iirorder_0, &ff_mlp_iirorder_1,
  52                                    &ff_mlp_iirorder_2, &ff_mlp_iirorder_3,
  53                                    &ff_mlp_iirorder_4 };
  54
  55 #if ARCH_X86_64
  56
  57 #define MLPMUL(label, offset, offs, offc)   \
  58     LABEL_MANGLE(label)":             \n\t" \
  59     "movslq "offset"+"offs"(%0), %%rax\n\t" \
  60     "movslq "offset"+"offc"(%1), %%rdx\n\t" \
  61     "imul                 %%rdx, %%rax\n\t" \
  62     "add                  %%rax, %%rsi\n\t"
  63
  64 #define FIRMULREG(label, offset, firc)\
  65     LABEL_MANGLE(label)":       \n\t" \
  66     "movslq "#offset"(%0), %%rax\n\t" \
  67     "imul        %"#firc", %%rax\n\t" \
  68     "add            %%rax, %%rsi\n\t"
  69
  70 #define CLEAR_ACCUM                   \
  71     "xor            %%rsi, %%rsi\n\t"
  72
  73 #define SHIFT_ACCUM                   \
  74     "shr     %%cl,         %%rsi\n\t"
  75
  76 #define ACCUM    "%%rdx"
  77 #define RESULT   "%%rsi"
  78 #define RESULT32 "%%esi"
  79
  80 #else /* if ARCH_X86_32 */
  81
  82 #define MLPMUL(label, offset, offs, offc)  \
  83     LABEL_MANGLE(label)":            \n\t" \
  84     "mov   "offset"+"offs"(%0), %%eax\n\t" \
  85     "imull "offset"+"offc"(%1)       \n\t" \
  86     "add                %%eax , %%esi\n\t" \
  87     "adc                %%edx , %%ecx\n\t"
  88
  89 #define FIRMULREG(label, offset, firc)  \
  90     MLPMUL(label, #offset, "0", "0")
  91
  92 #define CLEAR_ACCUM                  \
  93     "xor           %%esi, %%esi\n\t" \
  94     "xor           %%ecx, %%ecx\n\t"
  95
  96 #define SHIFT_ACCUM                  \
  97     "mov           %%ecx, %%edx\n\t" \
  98     "mov           %%esi, %%eax\n\t" \
  99     "movzbl        %7   , %%ecx\n\t" \
 100     "shrd    %%cl, %%edx, %%eax\n\t" \
 101
 102 #define ACCUM    "%%edx"
 103 #define RESULT   "%%eax"
 104 #define RESULT32 "%%eax"
 105
 106 #endif /* !ARCH_X86_64 */
 107
 108 #define BINC  AV_STRINGIFY(4* MAX_CHANNELS)
 109 #define IOFFS AV_STRINGIFY(4*(MAX_FIR_ORDER + MAX_BLOCKSIZE))
 110 #define IOFFC AV_STRINGIFY(4* MAX_FIR_ORDER)
 111
 112 #define FIRMUL(label, offset) MLPMUL(label, #offset,   "0",   "0")
 113 #define IIRMUL(label, offset) MLPMUL(label, #offset, IOFFS, IOFFC)
 114
 115 static void mlp_filter_channel_x86(int32_t *state, const int32_t *coeff,
 116                                    int firorder, int iirorder,
 117                                    unsigned int filter_shift, int32_t mask,
 118                                    int blocksize, int32_t *sample_buffer)
 119 {
 120     const void *firjump = firtable[firorder];
 121     const void *iirjump = iirtable[iirorder];
 122
 123     blocksize = -blocksize;
 124
 125     __asm__ volatile(
 126         "1:                           \n\t"
 127         CLEAR_ACCUM
 128         "jmp  *%5                     \n\t"
 129         FIRMUL   (ff_mlp_firorder_8, 0x1c   )
 130         FIRMUL   (ff_mlp_firorder_7, 0x18   )
 131         FIRMUL   (ff_mlp_firorder_6, 0x14   )
 132         FIRMUL   (ff_mlp_firorder_5, 0x10   )
 133         FIRMUL   (ff_mlp_firorder_4, 0x0c   )
 134         FIRMULREG(ff_mlp_firorder_3, 0x08,10)
 135         FIRMULREG(ff_mlp_firorder_2, 0x04, 9)
 136         FIRMULREG(ff_mlp_firorder_1, 0x00, 8)
 137         LABEL_MANGLE(ff_mlp_firorder_0)":\n\t"
 138         "jmp  *%6                     \n\t"
 139         IIRMUL   (ff_mlp_iirorder_4, 0x0c   )
 140         IIRMUL   (ff_mlp_iirorder_3, 0x08   )
 141         IIRMUL   (ff_mlp_iirorder_2, 0x04   )
 142         IIRMUL   (ff_mlp_iirorder_1, 0x00   )
 143         LABEL_MANGLE(ff_mlp_iirorder_0)":\n\t"
 144         SHIFT_ACCUM
 145         "mov  "RESULT"  ,"ACCUM"      \n\t"
 146         "add  (%2)      ,"RESULT"     \n\t"
 147         "and   %4       ,"RESULT"     \n\t"
 148         "sub   $4       ,  %0         \n\t"
 149         "mov  "RESULT32", (%0)        \n\t"
 150         "mov  "RESULT32", (%2)        \n\t"
 151         "add $"BINC"    ,  %2         \n\t"
 152         "sub  "ACCUM"   ,"RESULT"     \n\t"
 153         "mov  "RESULT32","IOFFS"(%0)  \n\t"
 154         "incl              %3         \n\t"
 155         "js 1b                        \n\t"
 156         : /* 0*/"+r"(state),
 157           /* 1*/"+r"(coeff),
 158           /* 2*/"+r"(sample_buffer),
 159 #if ARCH_X86_64
 160           /* 3*/"+r"(blocksize)
 161         : /* 4*/"r"((x86_reg)mask), /* 5*/"r"(firjump),
 162           /* 6*/"r"(iirjump)      , /* 7*/"c"(filter_shift)
 163         , /* 8*/"r"((int64_t)coeff[0])
 164         , /* 9*/"r"((int64_t)coeff[1])
 165         , /*10*/"r"((int64_t)coeff[2])
 166         : "rax", "rdx", "rsi"
 167 #else /* ARCH_X86_32 */
 168           /* 3*/"+m"(blocksize)
 169         : /* 4*/"m"(         mask), /* 5*/"m"(firjump),
 170           /* 6*/"m"(iirjump)      , /* 7*/"m"(filter_shift)
 171         : "eax", "edx", "esi", "ecx"
 172 #endif /* !ARCH_X86_64 */
 173     );
 174 }
 175
 176 #endif /* HAVE_7REGS && HAVE_INLINE_ASM */
 177
 178 av_cold void ff_mlpdsp_init_x86(MLPDSPContext *c)
 179 {
 180 #if HAVE_7REGS && HAVE_INLINE_ASM
 181     c->mlp_filter_channel = mlp_filter_channel_x86;
 182 #endif
 183 }