00001 /* 00002 * Copyright (c) 2012 00003 * MIPS Technologies, Inc., California. 00004 * 00005 * Redistribution and use in source and binary forms, with or without 00006 * modification, are permitted provided that the following conditions 00007 * are met: 00008 * 1. Redistributions of source code must retain the above copyright 00009 * notice, this list of conditions and the following disclaimer. 00010 * 2. Redistributions in binary form must reproduce the above copyright 00011 * notice, this list of conditions and the following disclaimer in the 00012 * documentation and/or other materials provided with the distribution. 00013 * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its 00014 * contributors may be used to endorse or promote products derived from 00015 * this software without specific prior written permission. 00016 * 00017 * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND 00018 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 00019 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 00020 * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE 00021 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 00022 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 00023 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 00024 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 00025 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 00026 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 00027 * SUCH DAMAGE. 00028 * 00029 * Author: Branimir Vasic (bvasic@mips.com) 00030 * 00031 * This file is part of FFmpeg. 00032 * 00033 * FFmpeg is free software; you can redistribute it and/or 00034 * modify it under the terms of the GNU Lesser General Public 00035 * License as published by the Free Software Foundation; either 00036 * version 2.1 of the License, or (at your option) any later version. 00037 * 00038 * FFmpeg is distributed in the hope that it will be useful, 00039 * but WITHOUT ANY WARRANTY; without even the implied warranty of 00040 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 00041 * Lesser General Public License for more details. 00042 * 00043 * You should have received a copy of the GNU Lesser General Public 00044 * License along with FFmpeg; if not, write to the Free Software 00045 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 00046 */ 00047 00053 #include "config.h" 00054 #include "libavutil/float_dsp.h" 00055 00056 #if HAVE_INLINE_ASM && HAVE_MIPSFPU 00057 static void vector_fmul_mips(float *dst, const float *src0, const float *src1, 00058 int len) 00059 { 00060 int i; 00061 00062 if (len & 3) { 00063 for (i = 0; i < len; i++) 00064 dst[i] = src0[i] * src1[i]; 00065 } else { 00066 float *d = (float *)dst; 00067 float *d_end = d + len; 00068 float *s0 = (float *)src0; 00069 float *s1 = (float *)src1; 00070 00071 float src0_0, src0_1, src0_2, src0_3; 00072 float src1_0, src1_1, src1_2, src1_3; 00073 00074 __asm__ volatile ( 00075 "1: \n\t" 00076 "lwc1 %[src0_0], 0(%[s0]) \n\t" 00077 "lwc1 %[src1_0], 0(%[s1]) \n\t" 00078 "lwc1 %[src0_1], 4(%[s0]) \n\t" 00079 "lwc1 %[src1_1], 4(%[s1]) \n\t" 00080 "lwc1 %[src0_2], 8(%[s0]) \n\t" 00081 "lwc1 %[src1_2], 8(%[s1]) \n\t" 00082 "lwc1 %[src0_3], 12(%[s0]) \n\t" 00083 "lwc1 %[src1_3], 12(%[s1]) \n\t" 00084 "mul.s %[src0_0], %[src0_0], %[src1_0] \n\t" 00085 "mul.s %[src0_1], %[src0_1], %[src1_1] \n\t" 00086 "mul.s %[src0_2], %[src0_2], %[src1_2] \n\t" 00087 "mul.s %[src0_3], %[src0_3], %[src1_3] \n\t" 00088 "swc1 %[src0_0], 0(%[d]) \n\t" 00089 "swc1 %[src0_1], 4(%[d]) \n\t" 00090 "swc1 %[src0_2], 8(%[d]) \n\t" 00091 "swc1 %[src0_3], 12(%[d]) \n\t" 00092 "addiu %[s0], %[s0], 16 \n\t" 00093 "addiu %[s1], %[s1], 16 \n\t" 00094 "addiu %[d], %[d], 16 \n\t" 00095 "bne %[d], %[d_end], 1b \n\t" 00096 00097 : [src0_0]"=&f"(src0_0), [src0_1]"=&f"(src0_1), 00098 [src0_2]"=&f"(src0_2), [src0_3]"=&f"(src0_3), 00099 [src1_0]"=&f"(src1_0), [src1_1]"=&f"(src1_1), 00100 [src1_2]"=&f"(src1_2), [src1_3]"=&f"(src1_3), 00101 [d]"+r"(d), [s0]"+r"(s0), [s1]"+r"(s1) 00102 : [d_end]"r"(d_end) 00103 : "memory" 00104 ); 00105 } 00106 } 00107 #endif /* HAVE_INLINE_ASM && HAVE_MIPSFPU */ 00108 00109 void ff_float_dsp_init_mips(AVFloatDSPContext *fdsp) { 00110 #if HAVE_INLINE_ASM && HAVE_MIPSFPU 00111 fdsp->vector_fmul = vector_fmul_mips; 00112 #endif /* HAVE_INLINE_ASM && HAVE_MIPSFPU */ 00113 }
1.5.8