2 * DSP utils : QNS functions are compiled 3 times for mmx/3dnow/ssse3
3 * Copyright (c) 2004 Michael Niedermayer
5 * MMX optimization by Michael Niedermayer <michaelni@gmx.at>
6 * 3DNow! and SSSE3 optimization by Zuxy Meng <zuxy.meng@gmail.com>
8 * This file is part of FFmpeg.
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.
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.
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
25 #define MAX_ABS (512 >> (SCALE_OFFSET>0 ? SCALE_OFFSET : 0))
27 static int DEF(try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale)
31 av_assert2(FFABS(scale) < MAX_ABS);
32 scale<<= 16 + SCALE_OFFSET - BASIS_SHIFT + RECON_SHIFT;
36 "pxor %%mm7, %%mm7 \n\t"
38 "punpcklwd %%mm5, %%mm5 \n\t"
39 "punpcklwd %%mm5, %%mm5 \n\t"
42 "movq (%1, %0), %%mm0 \n\t"
43 "movq 8(%1, %0), %%mm1 \n\t"
44 PMULHRW(%%mm0, %%mm1, %%mm5, %%mm6)
45 "paddw (%2, %0), %%mm0 \n\t"
46 "paddw 8(%2, %0), %%mm1 \n\t"
47 "psraw $6, %%mm0 \n\t"
48 "psraw $6, %%mm1 \n\t"
49 "pmullw (%3, %0), %%mm0 \n\t"
50 "pmullw 8(%3, %0), %%mm1 \n\t"
51 "pmaddwd %%mm0, %%mm0 \n\t"
52 "pmaddwd %%mm1, %%mm1 \n\t"
53 "paddd %%mm1, %%mm0 \n\t"
54 "psrld $4, %%mm0 \n\t"
55 "paddd %%mm0, %%mm7 \n\t"
57 "cmp $128, %0 \n\t" //FIXME optimize & bench
60 "psrld $2, %%mm7 \n\t"
64 : "r"(basis), "r"(rem), "r"(weight), "g"(scale)
69 static void DEF(add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale)
73 if(FFABS(scale) < MAX_ABS){
74 scale<<= 16 + SCALE_OFFSET - BASIS_SHIFT + RECON_SHIFT;
78 "punpcklwd %%mm5, %%mm5 \n\t"
79 "punpcklwd %%mm5, %%mm5 \n\t"
82 "movq (%1, %0), %%mm0 \n\t"
83 "movq 8(%1, %0), %%mm1 \n\t"
84 PMULHRW(%%mm0, %%mm1, %%mm5, %%mm6)
85 "paddw (%2, %0), %%mm0 \n\t"
86 "paddw 8(%2, %0), %%mm1 \n\t"
87 "movq %%mm0, (%2, %0) \n\t"
88 "movq %%mm1, 8(%2, %0) \n\t"
90 "cmp $128, %0 \n\t" // FIXME optimize & bench
94 : "r"(basis), "r"(rem), "g"(scale)
98 rem[i] += (basis[i]*scale + (1<<(BASIS_SHIFT - RECON_SHIFT-1)))>>(BASIS_SHIFT - RECON_SHIFT);