1/*
2 * QNS functions are compiled 3 times for MMX/3DNOW/SSSE3
3 * Copyright (c) 2004 Michael Niedermayer
4 *
5 * MMX optimization by Michael Niedermayer <michaelni@gmx.at>
6 * 3DNow! and SSSE3 optimization by Zuxy Meng <zuxy.meng@gmail.com>
7 *
8 * This file is part of FFmpeg.
9 *
10 * FFmpeg is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU Lesser General Public
12 * License as published by the Free Software Foundation; either
13 * version 2.1 of the License, or (at your option) any later version.
14 *
15 * FFmpeg is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18 * Lesser General Public License for more details.
19 *
20 * You should have received a copy of the GNU Lesser General Public
21 * License along with FFmpeg; if not, write to the Free Software
22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 */
24
25#include <stdint.h>
26
27#include "libavutil/avassert.h"
28#include "libavutil/common.h"
29#include "libavutil/x86/asm.h"
30
31#include "inline_asm.h"
32
33#define MAX_ABS (512 >> (SCALE_OFFSET>0 ? SCALE_OFFSET : 0))
34
35static int DEF(try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale)
36{
37    x86_reg i=0;
38
39    av_assert2(FFABS(scale) < MAX_ABS);
40    scale<<= 16 + SCALE_OFFSET - BASIS_SHIFT + RECON_SHIFT;
41
42    SET_RND(mm6);
43    __asm__ volatile(
44        "pxor %%mm7, %%mm7              \n\t"
45        "movd  %4, %%mm5                \n\t"
46        "punpcklwd %%mm5, %%mm5         \n\t"
47        "punpcklwd %%mm5, %%mm5         \n\t"
48        ".p2align 4                     \n\t"
49        "1:                             \n\t"
50        "movq  (%1, %0), %%mm0          \n\t"
51        "movq  8(%1, %0), %%mm1         \n\t"
52        PMULHRW(%%mm0, %%mm1, %%mm5, %%mm6)
53        "paddw (%2, %0), %%mm0          \n\t"
54        "paddw 8(%2, %0), %%mm1         \n\t"
55        "psraw $6, %%mm0                \n\t"
56        "psraw $6, %%mm1                \n\t"
57        "pmullw (%3, %0), %%mm0         \n\t"
58        "pmullw 8(%3, %0), %%mm1        \n\t"
59        "pmaddwd %%mm0, %%mm0           \n\t"
60        "pmaddwd %%mm1, %%mm1           \n\t"
61        "paddd %%mm1, %%mm0             \n\t"
62        "psrld $4, %%mm0                \n\t"
63        "paddd %%mm0, %%mm7             \n\t"
64        "add $16, %0                    \n\t"
65        "cmp $128, %0                   \n\t" //FIXME optimize & bench
66        " jb 1b                         \n\t"
67        PHADDD(%%mm7, %%mm6)
68        "psrld $2, %%mm7                \n\t"
69        "movd %%mm7, %0                 \n\t"
70
71        : "+r" (i)
72        : "r"(basis), "r"(rem), "r"(weight), "g"(scale)
73    );
74    return i;
75}
76
77static void DEF(add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale)
78{
79    x86_reg i=0;
80
81    if(FFABS(scale) < MAX_ABS){
82        scale<<= 16 + SCALE_OFFSET - BASIS_SHIFT + RECON_SHIFT;
83        SET_RND(mm6);
84        __asm__ volatile(
85                "movd  %3, %%mm5        \n\t"
86                "punpcklwd %%mm5, %%mm5 \n\t"
87                "punpcklwd %%mm5, %%mm5 \n\t"
88                ".p2align 4             \n\t"
89                "1:                     \n\t"
90                "movq  (%1, %0), %%mm0  \n\t"
91                "movq  8(%1, %0), %%mm1 \n\t"
92                PMULHRW(%%mm0, %%mm1, %%mm5, %%mm6)
93                "paddw (%2, %0), %%mm0  \n\t"
94                "paddw 8(%2, %0), %%mm1 \n\t"
95                "movq %%mm0, (%2, %0)   \n\t"
96                "movq %%mm1, 8(%2, %0)  \n\t"
97                "add $16, %0            \n\t"
98                "cmp $128, %0           \n\t" // FIXME optimize & bench
99                " jb 1b                 \n\t"
100
101                : "+r" (i)
102                : "r"(basis), "r"(rem), "g"(scale)
103        );
104    }else{
105        for(i=0; i<8*8; i++){
106            rem[i] += (basis[i]*scale + (1<<(BASIS_SHIFT - RECON_SHIFT-1)))>>(BASIS_SHIFT - RECON_SHIFT);
107        }
108    }
109}
110