yading@10: /* yading@10: * DSP utils : QNS functions are compiled 3 times for mmx/3dnow/ssse3 yading@10: * Copyright (c) 2004 Michael Niedermayer yading@10: * yading@10: * MMX optimization by Michael Niedermayer yading@10: * 3DNow! and SSSE3 optimization by Zuxy Meng yading@10: * yading@10: * This file is part of FFmpeg. yading@10: * yading@10: * FFmpeg is free software; you can redistribute it and/or yading@10: * modify it under the terms of the GNU Lesser General Public yading@10: * License as published by the Free Software Foundation; either yading@10: * version 2.1 of the License, or (at your option) any later version. yading@10: * yading@10: * FFmpeg is distributed in the hope that it will be useful, yading@10: * but WITHOUT ANY WARRANTY; without even the implied warranty of yading@10: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU yading@10: * Lesser General Public License for more details. yading@10: * yading@10: * You should have received a copy of the GNU Lesser General Public yading@10: * License along with FFmpeg; if not, write to the Free Software yading@10: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA yading@10: */ yading@10: yading@10: #define MAX_ABS (512 >> (SCALE_OFFSET>0 ? SCALE_OFFSET : 0)) yading@10: yading@10: static int DEF(try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale) yading@10: { yading@10: x86_reg i=0; yading@10: yading@10: av_assert2(FFABS(scale) < MAX_ABS); yading@10: scale<<= 16 + SCALE_OFFSET - BASIS_SHIFT + RECON_SHIFT; yading@10: yading@10: SET_RND(mm6); yading@10: __asm__ volatile( yading@10: "pxor %%mm7, %%mm7 \n\t" yading@10: "movd %4, %%mm5 \n\t" yading@10: "punpcklwd %%mm5, %%mm5 \n\t" yading@10: "punpcklwd %%mm5, %%mm5 \n\t" yading@10: ".p2align 4 \n\t" yading@10: "1: \n\t" yading@10: "movq (%1, %0), %%mm0 \n\t" yading@10: "movq 8(%1, %0), %%mm1 \n\t" yading@10: PMULHRW(%%mm0, %%mm1, %%mm5, %%mm6) yading@10: "paddw (%2, %0), %%mm0 \n\t" yading@10: "paddw 8(%2, %0), %%mm1 \n\t" yading@10: "psraw $6, %%mm0 \n\t" yading@10: "psraw $6, %%mm1 \n\t" yading@10: "pmullw (%3, %0), %%mm0 \n\t" yading@10: "pmullw 8(%3, %0), %%mm1 \n\t" yading@10: "pmaddwd %%mm0, %%mm0 \n\t" yading@10: "pmaddwd %%mm1, %%mm1 \n\t" yading@10: "paddd %%mm1, %%mm0 \n\t" yading@10: "psrld $4, %%mm0 \n\t" yading@10: "paddd %%mm0, %%mm7 \n\t" yading@10: "add $16, %0 \n\t" yading@10: "cmp $128, %0 \n\t" //FIXME optimize & bench yading@10: " jb 1b \n\t" yading@10: PHADDD(%%mm7, %%mm6) yading@10: "psrld $2, %%mm7 \n\t" yading@10: "movd %%mm7, %0 \n\t" yading@10: yading@10: : "+r" (i) yading@10: : "r"(basis), "r"(rem), "r"(weight), "g"(scale) yading@10: ); yading@10: return i; yading@10: } yading@10: yading@10: static void DEF(add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale) yading@10: { yading@10: x86_reg i=0; yading@10: yading@10: if(FFABS(scale) < MAX_ABS){ yading@10: scale<<= 16 + SCALE_OFFSET - BASIS_SHIFT + RECON_SHIFT; yading@10: SET_RND(mm6); yading@10: __asm__ volatile( yading@10: "movd %3, %%mm5 \n\t" yading@10: "punpcklwd %%mm5, %%mm5 \n\t" yading@10: "punpcklwd %%mm5, %%mm5 \n\t" yading@10: ".p2align 4 \n\t" yading@10: "1: \n\t" yading@10: "movq (%1, %0), %%mm0 \n\t" yading@10: "movq 8(%1, %0), %%mm1 \n\t" yading@10: PMULHRW(%%mm0, %%mm1, %%mm5, %%mm6) yading@10: "paddw (%2, %0), %%mm0 \n\t" yading@10: "paddw 8(%2, %0), %%mm1 \n\t" yading@10: "movq %%mm0, (%2, %0) \n\t" yading@10: "movq %%mm1, 8(%2, %0) \n\t" yading@10: "add $16, %0 \n\t" yading@10: "cmp $128, %0 \n\t" // FIXME optimize & bench yading@10: " jb 1b \n\t" yading@10: yading@10: : "+r" (i) yading@10: : "r"(basis), "r"(rem), "g"(scale) yading@10: ); yading@10: }else{ yading@10: for(i=0; i<8*8; i++){ yading@10: rem[i] += (basis[i]*scale + (1<<(BASIS_SHIFT - RECON_SHIFT-1)))>>(BASIS_SHIFT - RECON_SHIFT); yading@10: } yading@10: } yading@10: }