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 fixed-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_FIXED_H yading@10: #define AVCODEC_MIPS_COMPUTE_ANTIALIAS_FIXED_H yading@10: yading@10: #if HAVE_INLINE_ASM yading@10: static void compute_antialias_mips_fixed(MPADecodeContext *s, yading@10: GranuleDef *g) yading@10: { yading@10: int32_t *ptr, *csa; yading@10: int n, i; yading@10: int MAX_lo = 0xffffffff; yading@10: 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: } else { yading@10: n = SBLIMIT - 1; yading@10: } yading@10: yading@10: yading@10: ptr = g->sb_hybrid + 18; yading@10: yading@10: for(i = n;i > 0;i--) { yading@10: int tmp0, tmp1, tmp2, tmp00, tmp11; yading@10: int temp_reg1, temp_reg2, temp_reg3, temp_reg4, temp_reg5, temp_reg6; yading@10: csa = &csa_table[0][0]; yading@10: yading@10: /** yading@10: * instructions are scheduled to minimize pipeline stall. yading@10: */ yading@10: __asm__ volatile ( yading@10: "lw %[tmp0], -1*4(%[ptr]) \n\t" yading@10: "lw %[tmp1], 0*4(%[ptr]) \n\t" yading@10: "lw %[temp_reg1], 0*4(%[csa]) \n\t" yading@10: "lw %[temp_reg2], 2*4(%[csa]) \n\t" yading@10: "add %[tmp2], %[tmp0], %[tmp1] \n\t" yading@10: "lw %[temp_reg3], 3*4(%[csa]) \n\t" yading@10: "mult $ac0, %[tmp2], %[temp_reg1] \n\t" yading@10: "mult $ac1, %[tmp2], %[temp_reg1] \n\t" yading@10: "lw %[tmp00], -2*4(%[ptr]) \n\t" yading@10: "lw %[tmp11], 1*4(%[ptr]) \n\t" yading@10: "lw %[temp_reg4], 4*4(%[csa]) \n\t" yading@10: "mtlo %[MAX_lo], $ac0 \n\t" yading@10: "mtlo $zero, $ac1 \n\t" yading@10: "msub $ac0, %[tmp1], %[temp_reg2] \n\t" yading@10: "madd $ac1, %[tmp0], %[temp_reg3] \n\t" yading@10: "add %[tmp2], %[tmp00], %[tmp11] \n\t" yading@10: "lw %[temp_reg5], 6*4(%[csa]) \n\t" yading@10: "mult $ac2, %[tmp2], %[temp_reg4] \n\t" yading@10: "mult $ac3, %[tmp2], %[temp_reg4] \n\t" yading@10: "mfhi %[temp_reg1], $ac0 \n\t" yading@10: "mfhi %[temp_reg2], $ac1 \n\t" yading@10: "lw %[temp_reg6], 7*4(%[csa]) \n\t" yading@10: "mtlo %[MAX_lo], $ac2 \n\t" yading@10: "msub $ac2, %[tmp11], %[temp_reg5] \n\t" yading@10: "mtlo $zero, $ac3 \n\t" yading@10: "madd $ac3, %[tmp00], %[temp_reg6] \n\t" yading@10: "sll %[temp_reg1], %[temp_reg1], 2 \n\t" yading@10: "sw %[temp_reg1], -1*4(%[ptr]) \n\t" yading@10: "mfhi %[temp_reg4], $ac2 \n\t" yading@10: "sll %[temp_reg2], %[temp_reg2], 2 \n\t" yading@10: "mfhi %[temp_reg5], $ac3 \n\t" yading@10: "sw %[temp_reg2], 0*4(%[ptr]) \n\t" yading@10: "lw %[tmp0], -3*4(%[ptr]) \n\t" yading@10: "lw %[tmp1], 2*4(%[ptr]) \n\t" yading@10: "lw %[temp_reg1], 8*4(%[csa]) \n\t" yading@10: "sll %[temp_reg4], %[temp_reg4], 2 \n\t" yading@10: "add %[tmp2], %[tmp0], %[tmp1] \n\t" yading@10: "sll %[temp_reg5], %[temp_reg5], 2 \n\t" yading@10: "mult $ac0, %[tmp2], %[temp_reg1] \n\t" yading@10: "mult $ac1, %[tmp2], %[temp_reg1] \n\t" yading@10: "sw %[temp_reg4], -2*4(%[ptr]) \n\t" yading@10: "sw %[temp_reg5], 1*4(%[ptr]) \n\t" yading@10: "lw %[temp_reg2], 10*4(%[csa]) \n\t" yading@10: "mtlo %[MAX_lo], $ac0 \n\t" yading@10: "lw %[temp_reg3], 11*4(%[csa]) \n\t" yading@10: "msub $ac0, %[tmp1], %[temp_reg2] \n\t" yading@10: "mtlo $zero, $ac1 \n\t" yading@10: "madd $ac1, %[tmp0], %[temp_reg3] \n\t" yading@10: "lw %[tmp00], -4*4(%[ptr]) \n\t" yading@10: "lw %[tmp11], 3*4(%[ptr]) \n\t" yading@10: "mfhi %[temp_reg1], $ac0 \n\t" yading@10: "lw %[temp_reg4], 12*4(%[csa]) \n\t" yading@10: "mfhi %[temp_reg2], $ac1 \n\t" yading@10: "add %[tmp2], %[tmp00], %[tmp11] \n\t" yading@10: "mult $ac2, %[tmp2], %[temp_reg4] \n\t" yading@10: "mult $ac3, %[tmp2], %[temp_reg4] \n\t" yading@10: "lw %[temp_reg5], 14*4(%[csa]) \n\t" yading@10: "lw %[temp_reg6], 15*4(%[csa]) \n\t" yading@10: "sll %[temp_reg1], %[temp_reg1], 2 \n\t" yading@10: "mtlo %[MAX_lo], $ac2 \n\t" yading@10: "msub $ac2, %[tmp11], %[temp_reg5] \n\t" yading@10: "mtlo $zero, $ac3 \n\t" yading@10: "madd $ac3, %[tmp00], %[temp_reg6] \n\t" yading@10: "sll %[temp_reg2], %[temp_reg2], 2 \n\t" yading@10: "sw %[temp_reg1], -3*4(%[ptr]) \n\t" yading@10: "mfhi %[temp_reg4], $ac2 \n\t" yading@10: "sw %[temp_reg2], 2*4(%[ptr]) \n\t" yading@10: "mfhi %[temp_reg5], $ac3 \n\t" yading@10: "lw %[tmp0], -5*4(%[ptr]) \n\t" yading@10: "lw %[tmp1], 4*4(%[ptr]) \n\t" yading@10: "lw %[temp_reg1], 16*4(%[csa]) \n\t" yading@10: "lw %[temp_reg2], 18*4(%[csa]) \n\t" yading@10: "add %[tmp2], %[tmp0], %[tmp1] \n\t" yading@10: "lw %[temp_reg3], 19*4(%[csa]) \n\t" yading@10: "mult $ac0, %[tmp2], %[temp_reg1] \n\t" yading@10: "mult $ac1, %[tmp2], %[temp_reg1] \n\t" yading@10: "sll %[temp_reg4], %[temp_reg4], 2 \n\t" yading@10: "sll %[temp_reg5], %[temp_reg5], 2 \n\t" yading@10: "sw %[temp_reg4], -4*4(%[ptr]) \n\t" yading@10: "mtlo %[MAX_lo], $ac0 \n\t" yading@10: "msub $ac0, %[tmp1], %[temp_reg2] \n\t" yading@10: "mtlo $zero, $ac1 \n\t" yading@10: "madd $ac1, %[tmp0], %[temp_reg3] \n\t" yading@10: "sw %[temp_reg5], 3*4(%[ptr]) \n\t" yading@10: "lw %[tmp00], -6*4(%[ptr]) \n\t" yading@10: "mfhi %[temp_reg1], $ac0 \n\t" yading@10: "lw %[tmp11], 5*4(%[ptr]) \n\t" yading@10: "mfhi %[temp_reg2], $ac1 \n\t" yading@10: "lw %[temp_reg4], 20*4(%[csa]) \n\t" yading@10: "add %[tmp2], %[tmp00], %[tmp11] \n\t" yading@10: "lw %[temp_reg5], 22*4(%[csa]) \n\t" yading@10: "mult $ac2, %[tmp2], %[temp_reg4] \n\t" yading@10: "mult $ac3, %[tmp2], %[temp_reg4] \n\t" yading@10: "lw %[temp_reg6], 23*4(%[csa]) \n\t" yading@10: "sll %[temp_reg1], %[temp_reg1], 2 \n\t" yading@10: "sll %[temp_reg2], %[temp_reg2], 2 \n\t" yading@10: "mtlo %[MAX_lo], $ac2 \n\t" yading@10: "msub $ac2, %[tmp11], %[temp_reg5] \n\t" yading@10: "mtlo $zero, $ac3 \n\t" yading@10: "madd $ac3, %[tmp00], %[temp_reg6] \n\t" yading@10: "sw %[temp_reg1], -5*4(%[ptr]) \n\t" yading@10: "sw %[temp_reg2], 4*4(%[ptr]) \n\t" yading@10: "mfhi %[temp_reg4], $ac2 \n\t" yading@10: "lw %[tmp0], -7*4(%[ptr]) \n\t" yading@10: "mfhi %[temp_reg5], $ac3 \n\t" yading@10: "lw %[tmp1], 6*4(%[ptr]) \n\t" yading@10: "lw %[temp_reg1], 24*4(%[csa]) \n\t" yading@10: "lw %[temp_reg2], 26*4(%[csa]) \n\t" yading@10: "add %[tmp2], %[tmp0], %[tmp1] \n\t" yading@10: "lw %[temp_reg3], 27*4(%[csa]) \n\t" yading@10: "mult $ac0, %[tmp2], %[temp_reg1] \n\t" yading@10: "mult $ac1, %[tmp2], %[temp_reg1] \n\t" yading@10: "sll %[temp_reg4], %[temp_reg4], 2 \n\t" yading@10: "sll %[temp_reg5], %[temp_reg5], 2 \n\t" yading@10: "sw %[temp_reg4], -6*4(%[ptr]) \n\t" yading@10: "mtlo %[MAX_lo], $ac0 \n\t" yading@10: "msub $ac0, %[tmp1], %[temp_reg2] \n\t" yading@10: "mtlo $zero, $ac1 \n\t" yading@10: "madd $ac1, %[tmp0], %[temp_reg3] \n\t" yading@10: "sw %[temp_reg5], 5*4(%[ptr]) \n\t" yading@10: "lw %[tmp00], -8*4(%[ptr]) \n\t" yading@10: "mfhi %[temp_reg1], $ac0 \n\t" yading@10: "lw %[tmp11], 7*4(%[ptr]) \n\t" yading@10: "mfhi %[temp_reg2], $ac1 \n\t" yading@10: "lw %[temp_reg4], 28*4(%[csa]) \n\t" yading@10: "add %[tmp2], %[tmp00], %[tmp11] \n\t" yading@10: "lw %[temp_reg5], 30*4(%[csa]) \n\t" yading@10: "mult $ac2, %[tmp2], %[temp_reg4] \n\t" yading@10: "mult $ac3, %[tmp2], %[temp_reg4] \n\t" yading@10: "lw %[temp_reg6], 31*4(%[csa]) \n\t" yading@10: "sll %[temp_reg1], %[temp_reg1], 2 \n\t" yading@10: "sll %[temp_reg2], %[temp_reg2], 2 \n\t" yading@10: "mtlo %[MAX_lo], $ac2 \n\t" yading@10: "msub $ac2, %[tmp11], %[temp_reg5] \n\t" yading@10: "mtlo $zero, $ac3 \n\t" yading@10: "madd $ac3, %[tmp00], %[temp_reg6] \n\t" yading@10: "sw %[temp_reg1], -7*4(%[ptr]) \n\t" yading@10: "sw %[temp_reg2], 6*4(%[ptr]) \n\t" yading@10: "mfhi %[temp_reg4], $ac2 \n\t" yading@10: "mfhi %[temp_reg5], $ac3 \n\t" yading@10: "sll %[temp_reg4], %[temp_reg4], 2 \n\t" yading@10: "sll %[temp_reg5], %[temp_reg5], 2 \n\t" yading@10: "sw %[temp_reg4], -8*4(%[ptr]) \n\t" yading@10: "sw %[temp_reg5], 7*4(%[ptr]) \n\t" yading@10: yading@10: : [tmp0] "=&r" (tmp0), [tmp1] "=&r" (tmp1), [tmp2] "=&r" (tmp2), yading@10: [tmp00] "=&r" (tmp00), [tmp11] "=&r" (tmp11), yading@10: [temp_reg1] "=&r" (temp_reg1), [temp_reg2] "=&r" (temp_reg2), yading@10: [temp_reg3] "=&r" (temp_reg3), [temp_reg4] "=&r" (temp_reg4), yading@10: [temp_reg5] "=&r" (temp_reg5), [temp_reg6] "=&r" (temp_reg6) yading@10: : [csa] "r" (csa), [ptr] "r" (ptr), yading@10: [MAX_lo] "r" (MAX_lo) yading@10: : "memory", "hi", "lo", "$ac1hi", "$ac1lo", "$ac2hi", "$ac2lo", yading@10: "$ac3hi", "$ac3lo" yading@10: ); yading@10: yading@10: ptr += 18; yading@10: } yading@10: } yading@10: #define compute_antialias compute_antialias_mips_fixed yading@10: #endif /* HAVE_INLINE_ASM */ yading@10: yading@10: #endif /* AVCODEC_MIPS_COMPUTE_ANTIALIAS_FIXED_H */