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: * AAC encoder psychoacoustic model routines optimized yading@10: * 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/aacpsy.c yading@10: */ yading@10: yading@10: #ifndef AVCODEC_MIPS_AACPSY_MIPS_H yading@10: #define AVCODEC_MIPS_AACPSY_MIPS_H yading@10: yading@10: #if HAVE_INLINE_ASM && HAVE_MIPSFPU && ( PSY_LAME_FIR_LEN == 21 ) yading@10: static void calc_thr_3gpp_mips(const FFPsyWindowInfo *wi, const int num_bands, yading@10: AacPsyChannel *pch, const uint8_t *band_sizes, yading@10: const float *coefs) yading@10: { yading@10: int i, w, g; yading@10: int start = 0; yading@10: for (w = 0; w < wi->num_windows*16; w += 16) { yading@10: for (g = 0; g < num_bands; g++) { yading@10: AacPsyBand *band = &pch->band[w+g]; yading@10: yading@10: float form_factor = 0.0f; yading@10: float Temp; yading@10: band->energy = 0.0f; yading@10: for (i = 0; i < band_sizes[g]; i+=4) { yading@10: float a, b, c, d; yading@10: float ax, bx, cx, dx; yading@10: float *cf = (float *)&coefs[start+i]; yading@10: yading@10: __asm__ volatile ( yading@10: "lwc1 %[a], 0(%[cf]) \n\t" yading@10: "lwc1 %[b], 4(%[cf]) \n\t" yading@10: "lwc1 %[c], 8(%[cf]) \n\t" yading@10: "lwc1 %[d], 12(%[cf]) \n\t" yading@10: "abs.s %[a], %[a] \n\t" yading@10: "abs.s %[b], %[b] \n\t" yading@10: "abs.s %[c], %[c] \n\t" yading@10: "abs.s %[d], %[d] \n\t" yading@10: "sqrt.s %[ax], %[a] \n\t" yading@10: "sqrt.s %[bx], %[b] \n\t" yading@10: "sqrt.s %[cx], %[c] \n\t" yading@10: "sqrt.s %[dx], %[d] \n\t" yading@10: "madd.s %[e], %[e], %[a], %[a] \n\t" yading@10: "madd.s %[e], %[e], %[b], %[b] \n\t" yading@10: "madd.s %[e], %[e], %[c], %[c] \n\t" yading@10: "madd.s %[e], %[e], %[d], %[d] \n\t" yading@10: "add.s %[f], %[f], %[ax] \n\t" yading@10: "add.s %[f], %[f], %[bx] \n\t" yading@10: "add.s %[f], %[f], %[cx] \n\t" yading@10: "add.s %[f], %[f], %[dx] \n\t" yading@10: yading@10: : [a]"=&f"(a), [b]"=&f"(b), yading@10: [c]"=&f"(c), [d]"=&f"(d), yading@10: [e]"+f"(band->energy), [f]"+f"(form_factor), yading@10: [ax]"=&f"(ax), [bx]"=&f"(bx), yading@10: [cx]"=&f"(cx), [dx]"=&f"(dx) yading@10: : [cf]"r"(cf) yading@10: : "memory" yading@10: ); yading@10: } yading@10: yading@10: Temp = sqrtf((float)band_sizes[g] / band->energy); yading@10: band->thr = band->energy * 0.001258925f; yading@10: band->nz_lines = form_factor * sqrtf(Temp); yading@10: start += band_sizes[g]; yading@10: } yading@10: } yading@10: } yading@10: yading@10: static void psy_hp_filter_mips(const float *firbuf, float *hpfsmpl, const float * psy_fir_coeffs) yading@10: { yading@10: float sum1, sum2, sum3, sum4; yading@10: float *fb = (float*)firbuf; yading@10: float *fb_end = fb + AAC_BLOCK_SIZE_LONG; yading@10: float *hp = hpfsmpl; yading@10: yading@10: float coeff0 = psy_fir_coeffs[1]; yading@10: float coeff1 = psy_fir_coeffs[3]; yading@10: float coeff2 = psy_fir_coeffs[5]; yading@10: float coeff3 = psy_fir_coeffs[7]; yading@10: float coeff4 = psy_fir_coeffs[9]; yading@10: yading@10: __asm__ volatile ( yading@10: ".set push \n\t" yading@10: ".set noreorder \n\t" yading@10: yading@10: "li.s $f12, 32768 \n\t" yading@10: "1: \n\t" yading@10: "lwc1 $f0, 40(%[fb]) \n\t" yading@10: "lwc1 $f1, 4(%[fb]) \n\t" yading@10: "lwc1 $f2, 80(%[fb]) \n\t" yading@10: "lwc1 $f3, 44(%[fb]) \n\t" yading@10: "lwc1 $f4, 8(%[fb]) \n\t" yading@10: "madd.s %[sum1], $f0, $f1, %[coeff0] \n\t" yading@10: "lwc1 $f5, 84(%[fb]) \n\t" yading@10: "lwc1 $f6, 48(%[fb]) \n\t" yading@10: "madd.s %[sum2], $f3, $f4, %[coeff0] \n\t" yading@10: "lwc1 $f7, 12(%[fb]) \n\t" yading@10: "madd.s %[sum1], %[sum1], $f2, %[coeff0] \n\t" yading@10: "lwc1 $f8, 88(%[fb]) \n\t" yading@10: "lwc1 $f9, 52(%[fb]) \n\t" yading@10: "madd.s %[sum2], %[sum2], $f5, %[coeff0] \n\t" yading@10: "madd.s %[sum3], $f6, $f7, %[coeff0] \n\t" yading@10: "lwc1 $f10, 16(%[fb]) \n\t" yading@10: "lwc1 $f11, 92(%[fb]) \n\t" yading@10: "madd.s %[sum1], %[sum1], $f7, %[coeff1] \n\t" yading@10: "lwc1 $f1, 72(%[fb]) \n\t" yading@10: "madd.s %[sum3], %[sum3], $f8, %[coeff0] \n\t" yading@10: "madd.s %[sum4], $f9, $f10, %[coeff0] \n\t" yading@10: "madd.s %[sum2], %[sum2], $f10, %[coeff1] \n\t" yading@10: "madd.s %[sum1], %[sum1], $f1, %[coeff1] \n\t" yading@10: "lwc1 $f4, 76(%[fb]) \n\t" yading@10: "lwc1 $f8, 20(%[fb]) \n\t" yading@10: "madd.s %[sum4], %[sum4], $f11, %[coeff0] \n\t" yading@10: "lwc1 $f11, 24(%[fb]) \n\t" yading@10: "madd.s %[sum2], %[sum2], $f4, %[coeff1] \n\t" yading@10: "madd.s %[sum1], %[sum1], $f8, %[coeff2] \n\t" yading@10: "madd.s %[sum3], %[sum3], $f8, %[coeff1] \n\t" yading@10: "madd.s %[sum4], %[sum4], $f11, %[coeff1] \n\t" yading@10: "lwc1 $f7, 64(%[fb]) \n\t" yading@10: "madd.s %[sum2], %[sum2], $f11, %[coeff2] \n\t" yading@10: "lwc1 $f10, 68(%[fb]) \n\t" yading@10: "madd.s %[sum3], %[sum3], $f2, %[coeff1] \n\t" yading@10: "madd.s %[sum4], %[sum4], $f5, %[coeff1] \n\t" yading@10: "madd.s %[sum1], %[sum1], $f7, %[coeff2] \n\t" yading@10: "madd.s %[sum2], %[sum2], $f10, %[coeff2] \n\t" yading@10: "lwc1 $f2, 28(%[fb]) \n\t" yading@10: "lwc1 $f5, 32(%[fb]) \n\t" yading@10: "lwc1 $f8, 56(%[fb]) \n\t" yading@10: "lwc1 $f11, 60(%[fb]) \n\t" yading@10: "madd.s %[sum3], %[sum3], $f2, %[coeff2] \n\t" yading@10: "madd.s %[sum4], %[sum4], $f5, %[coeff2] \n\t" yading@10: "madd.s %[sum1], %[sum1], $f2, %[coeff3] \n\t" yading@10: "madd.s %[sum2], %[sum2], $f5, %[coeff3] \n\t" yading@10: "madd.s %[sum3], %[sum3], $f1, %[coeff2] \n\t" yading@10: "madd.s %[sum4], %[sum4], $f4, %[coeff2] \n\t" yading@10: "madd.s %[sum1], %[sum1], $f8, %[coeff3] \n\t" yading@10: "madd.s %[sum2], %[sum2], $f11, %[coeff3] \n\t" yading@10: "lwc1 $f1, 36(%[fb]) \n\t" yading@10: "addiu %[fb], %[fb], 16 \n\t" yading@10: "madd.s %[sum4], %[sum4], $f0, %[coeff3] \n\t" yading@10: "madd.s %[sum3], %[sum3], $f1, %[coeff3] \n\t" yading@10: "madd.s %[sum1], %[sum1], $f1, %[coeff4] \n\t" yading@10: "madd.s %[sum2], %[sum2], $f0, %[coeff4] \n\t" yading@10: "madd.s %[sum4], %[sum4], $f10, %[coeff3] \n\t" yading@10: "madd.s %[sum3], %[sum3], $f7, %[coeff3] \n\t" yading@10: "madd.s %[sum1], %[sum1], $f6, %[coeff4] \n\t" yading@10: "madd.s %[sum2], %[sum2], $f9, %[coeff4] \n\t" yading@10: "madd.s %[sum4], %[sum4], $f6, %[coeff4] \n\t" yading@10: "madd.s %[sum3], %[sum3], $f3, %[coeff4] \n\t" yading@10: "mul.s %[sum1], %[sum1], $f12 \n\t" yading@10: "mul.s %[sum2], %[sum2], $f12 \n\t" yading@10: "madd.s %[sum4], %[sum4], $f11, %[coeff4] \n\t" yading@10: "madd.s %[sum3], %[sum3], $f8, %[coeff4] \n\t" yading@10: "swc1 %[sum1], 0(%[hp]) \n\t" yading@10: "swc1 %[sum2], 4(%[hp]) \n\t" yading@10: "mul.s %[sum4], %[sum4], $f12 \n\t" yading@10: "mul.s %[sum3], %[sum3], $f12 \n\t" yading@10: "swc1 %[sum4], 12(%[hp]) \n\t" yading@10: "swc1 %[sum3], 8(%[hp]) \n\t" yading@10: "bne %[fb], %[fb_end], 1b \n\t" yading@10: " addiu %[hp], %[hp], 16 \n\t" yading@10: yading@10: ".set pop \n\t" yading@10: yading@10: : [sum1]"=&f"(sum1), [sum2]"=&f"(sum2), yading@10: [sum3]"=&f"(sum3), [sum4]"=&f"(sum4), yading@10: [fb]"+r"(fb), [hp]"+r"(hp) yading@10: : [coeff0]"f"(coeff0), [coeff1]"f"(coeff1), yading@10: [coeff2]"f"(coeff2), [coeff3]"f"(coeff3), yading@10: [coeff4]"f"(coeff4), [fb_end]"r"(fb_end) yading@10: : "$f0", "$f1", "$f2", "$f3", "$f4", "$f5", "$f6", yading@10: "$f7", "$f8", "$f9", "$f10", "$f11", "$f12", yading@10: "memory" yading@10: ); yading@10: } yading@10: yading@10: #define calc_thr_3gpp calc_thr_3gpp_mips yading@10: #define psy_hp_filter psy_hp_filter_mips yading@10: yading@10: #endif /* HAVE_INLINE_ASM && HAVE_MIPSFPU */ yading@10: #endif /* AVCODEC_MIPS_AACPSY_MIPS_H */