yading@10: /* yading@10: * Copyright (c) 2012 yading@10: * MIPS Technologies, Inc., California. yading@10: * yading@10: * Redistribution and use in source and binary forms, with or without yading@10: * modification, are permitted provided that the following conditions yading@10: * are met: yading@10: * 1. Redistributions of source code must retain the above copyright yading@10: * notice, this list of conditions and the following disclaimer. yading@10: * 2. Redistributions in binary form must reproduce the above copyright yading@10: * notice, this list of conditions and the following disclaimer in the yading@10: * documentation and/or other materials provided with the distribution. yading@10: * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its yading@10: * contributors may be used to endorse or promote products derived from yading@10: * this software without specific prior written permission. yading@10: * yading@10: * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND yading@10: * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE yading@10: * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE yading@10: * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE yading@10: * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL yading@10: * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS yading@10: * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) yading@10: * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT yading@10: * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY yading@10: * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF yading@10: * SUCH DAMAGE. yading@10: * yading@10: * Author: Bojan Zivkovic (bojan@mips.com) yading@10: * yading@10: * Compute antialias function optimised for MIPS floating-point architecture 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: /** yading@10: * @file yading@10: * Reference: libavcodec/mpegaudiodec.c yading@10: */ yading@10: yading@10: #ifndef AVCODEC_MIPS_COMPUTE_ANTIALIAS_FLOAT_H yading@10: #define AVCODEC_MIPS_COMPUTE_ANTIALIAS_FLOAT_H yading@10: yading@10: #if HAVE_INLINE_ASM yading@10: static void compute_antialias_mips_float(MPADecodeContext *s, yading@10: GranuleDef *g) yading@10: { yading@10: float *ptr, *ptr_end; yading@10: float *csa = &csa_table[0][0]; yading@10: int n; yading@10: /* temporary variables */ yading@10: float in1, in2, in3, in4, in5, in6, in7, in8; yading@10: float out1, out2, out3, out4; yading@10: yading@10: ptr = g->sb_hybrid + 18; yading@10: /* we antialias only "long" bands */ yading@10: if (g->block_type == 2) { yading@10: if (!g->switch_point) yading@10: return; yading@10: /* XXX: check this for 8000Hz case */ yading@10: n = 1; yading@10: ptr_end = ptr + 18; yading@10: } else { yading@10: n = 31; yading@10: ptr_end = ptr + 558; yading@10: } yading@10: yading@10: /** yading@10: * instructions are scheduled to minimize pipeline stall. yading@10: */ yading@10: yading@10: __asm__ volatile ( yading@10: "compute_antialias_float_loop%=: \t\n" yading@10: "lwc1 %[in1], -1*4(%[ptr]) \t\n" yading@10: "lwc1 %[in2], 0(%[csa]) \t\n" yading@10: "lwc1 %[in3], 1*4(%[csa]) \t\n" yading@10: "lwc1 %[in4], 0(%[ptr]) \t\n" yading@10: "lwc1 %[in5], -2*4(%[ptr]) \t\n" yading@10: "lwc1 %[in6], 4*4(%[csa]) \t\n" yading@10: "mul.s %[out1], %[in1], %[in2] \t\n" yading@10: "mul.s %[out2], %[in1], %[in3] \t\n" yading@10: "lwc1 %[in7], 5*4(%[csa]) \t\n" yading@10: "lwc1 %[in8], 1*4(%[ptr]) \t\n" yading@10: "nmsub.s %[out1], %[out1], %[in3], %[in4] \t\n" yading@10: "madd.s %[out2], %[out2], %[in2], %[in4] \t\n" yading@10: "mul.s %[out3], %[in5], %[in6] \t\n" yading@10: "mul.s %[out4], %[in5], %[in7] \t\n" yading@10: "lwc1 %[in1], -3*4(%[ptr]) \t\n" yading@10: "swc1 %[out1], -1*4(%[ptr]) \t\n" yading@10: "swc1 %[out2], 0(%[ptr]) \t\n" yading@10: "nmsub.s %[out3], %[out3], %[in7], %[in8] \t\n" yading@10: "madd.s %[out4], %[out4], %[in6], %[in8] \t\n" yading@10: "lwc1 %[in2], 8*4(%[csa]) \t\n" yading@10: "swc1 %[out3], -2*4(%[ptr]) \t\n" yading@10: "swc1 %[out4], 1*4(%[ptr]) \t\n" yading@10: "lwc1 %[in3], 9*4(%[csa]) \t\n" yading@10: "lwc1 %[in4], 2*4(%[ptr]) \t\n" yading@10: "mul.s %[out1], %[in1], %[in2] \t\n" yading@10: "lwc1 %[in5], -4*4(%[ptr]) \t\n" yading@10: "lwc1 %[in6], 12*4(%[csa]) \t\n" yading@10: "mul.s %[out2], %[in1], %[in3] \t\n" yading@10: "lwc1 %[in7], 13*4(%[csa]) \t\n" yading@10: "nmsub.s %[out1], %[out1], %[in3], %[in4] \t\n" yading@10: "lwc1 %[in8], 3*4(%[ptr]) \t\n" yading@10: "mul.s %[out3], %[in5], %[in6] \t\n" yading@10: "madd.s %[out2], %[out2], %[in2], %[in4] \t\n" yading@10: "mul.s %[out4], %[in5], %[in7] \t\n" yading@10: "swc1 %[out1], -3*4(%[ptr]) \t\n" yading@10: "lwc1 %[in1], -5*4(%[ptr]) \t\n" yading@10: "nmsub.s %[out3], %[out3], %[in7], %[in8] \t\n" yading@10: "swc1 %[out2], 2*4(%[ptr]) \t\n" yading@10: "madd.s %[out4], %[out4], %[in6], %[in8] \t\n" yading@10: "lwc1 %[in2], 16*4(%[csa]) \t\n" yading@10: "lwc1 %[in3], 17*4(%[csa]) \t\n" yading@10: "swc1 %[out3], -4*4(%[ptr]) \t\n" yading@10: "lwc1 %[in4], 4*4(%[ptr]) \t\n" yading@10: "swc1 %[out4], 3*4(%[ptr]) \t\n" yading@10: "mul.s %[out1], %[in1], %[in2] \t\n" yading@10: "mul.s %[out2], %[in1], %[in3] \t\n" yading@10: "lwc1 %[in5], -6*4(%[ptr]) \t\n" yading@10: "lwc1 %[in6], 20*4(%[csa]) \t\n" yading@10: "lwc1 %[in7], 21*4(%[csa]) \t\n" yading@10: "nmsub.s %[out1], %[out1], %[in3], %[in4] \t\n" yading@10: "madd.s %[out2], %[out2], %[in2], %[in4] \t\n" yading@10: "lwc1 %[in8], 5*4(%[ptr]) \t\n" yading@10: "mul.s %[out3], %[in5], %[in6] \t\n" yading@10: "mul.s %[out4], %[in5], %[in7] \t\n" yading@10: "swc1 %[out1], -5*4(%[ptr]) \t\n" yading@10: "swc1 %[out2], 4*4(%[ptr]) \t\n" yading@10: "lwc1 %[in1], -7*4(%[ptr]) \t\n" yading@10: "nmsub.s %[out3], %[out3], %[in7], %[in8] \t\n" yading@10: "madd.s %[out4], %[out4], %[in6], %[in8] \t\n" yading@10: "lwc1 %[in2], 24*4(%[csa]) \t\n" yading@10: "lwc1 %[in3], 25*4(%[csa]) \t\n" yading@10: "lwc1 %[in4], 6*4(%[ptr]) \t\n" yading@10: "swc1 %[out3], -6*4(%[ptr]) \t\n" yading@10: "swc1 %[out4], 5*4(%[ptr]) \t\n" yading@10: "mul.s %[out1], %[in1], %[in2] \t\n" yading@10: "lwc1 %[in5], -8*4(%[ptr]) \t\n" yading@10: "mul.s %[out2], %[in1], %[in3] \t\n" yading@10: "lwc1 %[in6], 28*4(%[csa]) \t\n" yading@10: "lwc1 %[in7], 29*4(%[csa]) \t\n" yading@10: "nmsub.s %[out1], %[out1], %[in3], %[in4] \t\n" yading@10: "lwc1 %[in8], 7*4(%[ptr]) \t\n" yading@10: "madd.s %[out2], %[out2], %[in2], %[in4] \t\n" yading@10: "mul.s %[out3], %[in5], %[in6] \t\n" yading@10: "mul.s %[out4], %[in5], %[in7] \t\n" yading@10: "swc1 %[out1], -7*4(%[ptr]) \t\n" yading@10: "swc1 %[out2], 6*4(%[ptr]) \t\n" yading@10: "addiu %[ptr], %[ptr], 72 \t\n" yading@10: "nmsub.s %[out3], %[out3], %[in7], %[in8] \t\n" yading@10: "madd.s %[out4], %[out4], %[in6], %[in8] \t\n" yading@10: "swc1 %[out3], -26*4(%[ptr]) \t\n" yading@10: "swc1 %[out4], -11*4(%[ptr]) \t\n" yading@10: "bne %[ptr], %[ptr_end], compute_antialias_float_loop%= \t\n" yading@10: yading@10: : [ptr] "+r" (ptr), yading@10: [in1] "=&f" (in1), [in2] "=&f" (in2), yading@10: [in3] "=&f" (in3), [in4] "=&f" (in4), yading@10: [in5] "=&f" (in5), [in6] "=&f" (in6), yading@10: [in7] "=&f" (in7), [in8] "=&f" (in8), yading@10: [out1] "=&f" (out1), [out2] "=&f" (out2), yading@10: [out3] "=&f" (out3), [out4] "=&f" (out4) yading@10: : [csa] "r" (csa), [ptr_end] "r" (ptr_end) yading@10: : "memory" yading@10: ); yading@10: } yading@10: #define compute_antialias compute_antialias_mips_float yading@10: #endif /* HAVE_INLINE_ASM */ yading@10: yading@10: #endif /* AVCODEC_MIPS_COMPUTE_ANTIALIAS_FLOAT_H */