1cabdff1aSopenharmony_ci/* 2cabdff1aSopenharmony_ci * ARM NEON optimised integer operations 3cabdff1aSopenharmony_ci * Copyright (c) 2009 Kostya Shishkov 4cabdff1aSopenharmony_ci * 5cabdff1aSopenharmony_ci * This file is part of FFmpeg. 6cabdff1aSopenharmony_ci * 7cabdff1aSopenharmony_ci * FFmpeg is free software; you can redistribute it and/or 8cabdff1aSopenharmony_ci * modify it under the terms of the GNU Lesser General Public 9cabdff1aSopenharmony_ci * License as published by the Free Software Foundation; either 10cabdff1aSopenharmony_ci * version 2.1 of the License, or (at your option) any later version. 11cabdff1aSopenharmony_ci * 12cabdff1aSopenharmony_ci * FFmpeg is distributed in the hope that it will be useful, 13cabdff1aSopenharmony_ci * but WITHOUT ANY WARRANTY; without even the implied warranty of 14cabdff1aSopenharmony_ci * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 15cabdff1aSopenharmony_ci * Lesser General Public License for more details. 16cabdff1aSopenharmony_ci * 17cabdff1aSopenharmony_ci * You should have received a copy of the GNU Lesser General Public 18cabdff1aSopenharmony_ci * License along with FFmpeg; if not, write to the Free Software 19cabdff1aSopenharmony_ci * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 20cabdff1aSopenharmony_ci */ 21cabdff1aSopenharmony_ci 22cabdff1aSopenharmony_ci#include "libavutil/arm/asm.S" 23cabdff1aSopenharmony_ci 24cabdff1aSopenharmony_ci@ scalarproduct_and_madd_int16(/*aligned*/v0,v1,v2,order,mul) 25cabdff1aSopenharmony_cifunction ff_scalarproduct_and_madd_int16_neon, export=1 26cabdff1aSopenharmony_ci vld1.16 {d28[],d29[]}, [sp] 27cabdff1aSopenharmony_ci vmov.i16 q0, #0 28cabdff1aSopenharmony_ci vmov.i16 q1, #0 29cabdff1aSopenharmony_ci vmov.i16 q2, #0 30cabdff1aSopenharmony_ci vmov.i16 q3, #0 31cabdff1aSopenharmony_ci mov r12, r0 32cabdff1aSopenharmony_ci 33cabdff1aSopenharmony_ci1: vld1.16 {d16-d17}, [r0,:128]! 34cabdff1aSopenharmony_ci vld1.16 {d18-d19}, [r1]! 35cabdff1aSopenharmony_ci vld1.16 {d20-d21}, [r2]! 36cabdff1aSopenharmony_ci vld1.16 {d22-d23}, [r0,:128]! 37cabdff1aSopenharmony_ci vld1.16 {d24-d25}, [r1]! 38cabdff1aSopenharmony_ci vld1.16 {d26-d27}, [r2]! 39cabdff1aSopenharmony_ci vmul.s16 q10, q10, q14 40cabdff1aSopenharmony_ci vmul.s16 q13, q13, q14 41cabdff1aSopenharmony_ci vmlal.s16 q0, d16, d18 42cabdff1aSopenharmony_ci vmlal.s16 q1, d17, d19 43cabdff1aSopenharmony_ci vadd.s16 q10, q8, q10 44cabdff1aSopenharmony_ci vadd.s16 q13, q11, q13 45cabdff1aSopenharmony_ci vmlal.s16 q2, d22, d24 46cabdff1aSopenharmony_ci vmlal.s16 q3, d23, d25 47cabdff1aSopenharmony_ci vst1.16 {q10}, [r12,:128]! 48cabdff1aSopenharmony_ci subs r3, r3, #16 49cabdff1aSopenharmony_ci vst1.16 {q13}, [r12,:128]! 50cabdff1aSopenharmony_ci bgt 1b 51cabdff1aSopenharmony_ci 52cabdff1aSopenharmony_ci vpadd.s32 d16, d0, d1 53cabdff1aSopenharmony_ci vpadd.s32 d17, d2, d3 54cabdff1aSopenharmony_ci vpadd.s32 d18, d4, d5 55cabdff1aSopenharmony_ci vpadd.s32 d19, d6, d7 56cabdff1aSopenharmony_ci vpadd.s32 d0, d16, d17 57cabdff1aSopenharmony_ci vpadd.s32 d1, d18, d19 58cabdff1aSopenharmony_ci vpadd.s32 d2, d0, d1 59cabdff1aSopenharmony_ci vpaddl.s32 d3, d2 60cabdff1aSopenharmony_ci vmov.32 r0, d3[0] 61cabdff1aSopenharmony_ci bx lr 62cabdff1aSopenharmony_ciendfunc 63