libavcodec/x86/ttadsp.asm

   1 ;******************************************************************************
   2 ;* TTA DSP SIMD optimizations
   3 ;*
   4 ;* Copyright (C) 2014 James Almer
   5 ;*
   6 ;* This file is part of FFmpeg.
   7 ;*
   8 ;* FFmpeg is free software; you can redistribute it and/or
   9 ;* modify it under the terms of the GNU Lesser General Public
  10 ;* License as published by the Free Software Foundation; either
  11 ;* version 2.1 of the License, or (at your option) any later version.
  12 ;*
  13 ;* FFmpeg is distributed in the hope that it will be useful,
  14 ;* but WITHOUT ANY WARRANTY; without even the implied warranty of
  15 ;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  16 ;* Lesser General Public License for more details.
  17 ;*
  18 ;* You should have received a copy of the GNU Lesser General Public
  19 ;* License along with FFmpeg; if not, write to the Free Software
  20 ;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  21 ;******************************************************************************
  22
  23 %include "libavutil/x86/x86util.asm"
  24
  25 SECTION_RODATA
  26
  27 pd_n0113: dd ~0, ~1, ~1, ~3
  28 pd_1224:  dd 1, 2, 2, 4
  29
  30 SECTION .text
  31
  32 %macro TTA_FILTER 2
  33 INIT_XMM %1
  34 cglobal tta_filter_process, 5,5,%2, qm, dx, dl, error, in, shift, round
  35     mova       m2, [qmq       ]
  36     mova       m3, [qmq + 0x10]
  37     mova       m4, [dxq       ]
  38     mova       m5, [dxq + 0x10]
  39
  40     movd       m6, [errorq]         ; if (filter->error < 0) {
  41     SPLATD     m6                   ;     for (int i = 0; i < 8; i++)
  42     psignd     m0, m4, m6           ;         filter->qm[i] -= filter->dx[i];
  43     psignd     m1, m5, m6           ; } else if (filter->error > 0) {
  44     paddd      m2, m0               ;     for (int i = 0; i < 8; i++)
  45     paddd      m3, m1               ;         filter->qm[i] += filter->dx[i];
  46     mova       [qmq       ], m2     ; }
  47     mova       [qmq + 0x10], m3     ;
  48
  49     mova       m0, [dlq       ]
  50     mova       m1, [dlq + 0x10]
  51
  52 %if cpuflag(sse4)
  53     pmulld     m2, m0
  54     pmulld     m3, m1
  55 %else
  56     pshufd     m6, m0, 0xb1
  57     pshufd     m7, m2, 0xb1
  58     pmuludq    m6, m7
  59     pshufd     m6, m6, 0xd8
  60     pmuludq    m2, m0
  61     pshufd     m2, m2, 0xd8
  62     punpckldq  m2, m6
  63
  64     pshufd     m6, m1, 0xb1
  65     pshufd     m7, m3, 0xb1
  66     pmuludq    m6, m7
  67     pshufd     m6, m6, 0xd8
  68     pmuludq    m3, m1
  69     pshufd     m3, m3, 0xd8
  70     punpckldq  m3, m6
  71 %endif
  72     ; Using horizontal add (phaddd) seems to be slower than shuffling stuff around
  73     paddd      m2, m3               ; int sum = filter->round +
  74                                     ;           filter->dl[0] * filter->qm[0] +
  75     pshufd     m3, m2, 0xe          ;           filter->dl[1] * filter->qm[1] +
  76     paddd      m2, m3               ;           filter->dl[2] * filter->qm[2] +
  77                                     ;           filter->dl[3] * filter->qm[3] +
  78     movd       m6, roundm           ;           filter->dl[4] * filter->qm[4] +
  79     paddd      m6, m2               ;           filter->dl[5] * filter->qm[5] +
  80     pshufd     m2, m2, 0x1          ;           filter->dl[6] * filter->qm[6] +
  81     paddd      m6, m2               ;           filter->dl[7] * filter->qm[7];
  82
  83     palignr    m5, m4, 4            ; filter->dx[0] = filter->dx[1]; filter->dx[1] = filter->dx[2];
  84                                     ; filter->dx[2] = filter->dx[3]; filter->dx[3] = filter->dx[4];
  85
  86     palignr    m2, m1, m0, 4        ; filter->dl[0] = filter->dl[1]; filter->dl[1] = filter->dl[2];
  87                                     ; filter->dl[2] = filter->dl[3]; filter->dl[3] = filter->dl[4];
  88
  89     psrad      m4, m1, 30           ; filter->dx[4] = ((filter->dl[4] >> 30) | 1);
  90     por        m4, [pd_1224 ]       ; filter->dx[5] = ((filter->dl[5] >> 30) | 2) & ~1;
  91     pand       m4, [pd_n0113]       ; filter->dx[6] = ((filter->dl[6] >> 30) | 2) & ~1;
  92                                     ; filter->dx[7] = ((filter->dl[7] >> 30) | 4) & ~3;
  93
  94     mova       [dlq       ], m2
  95     mova       [dxq       ], m5
  96     mova       [dxq + 0x10], m4
  97     movd       m0, [inq]            ; filter->error = *in;
  98     movd       [errorq], m0         ;
  99
 100     movd       m2, shiftm           ; *in += (sum >> filter->shift);
 101     psrad      m6, m2               ;
 102     paddd      m0, m6               ;
 103     movd       [inq], m0            ;
 104
 105     psrldq     m1, 4                ;
 106     pslldq     m0, 12               ; filter->dl[4] = -filter->dl[5];
 107     pshufd     m0, m0, 0xf0         ; filter->dl[5] = -filter->dl[6];
 108     psubd      m0, m1               ; filter->dl[6] = *in - filter->dl[7];
 109     psrldq     m1, m0, 4            ; filter->dl[7] = *in;
 110     pshufd     m1, m1, 0xf4         ; filter->dl[5] += filter->dl[6];
 111     paddd      m0, m1               ; filter->dl[4] += filter->dl[5];
 112     psrldq     m1, 4                ;
 113     paddd      m0, m1               ;
 114     mova       [dlq + 0x10], m0     ;
 115     RET
 116 %endmacro
 117
 118 TTA_FILTER ssse3, 8
 119 TTA_FILTER sse4,  7