diff options
Diffstat (limited to 'libavcodec/mips')
-rw-r--r-- | libavcodec/mips/Makefile | 15 | ||||
-rw-r--r-- | libavcodec/mips/acelp_filters_mips.c | 210 | ||||
-rw-r--r-- | libavcodec/mips/acelp_vectors_mips.c | 96 | ||||
-rw-r--r-- | libavcodec/mips/amrwbdec_mips.c | 185 | ||||
-rw-r--r-- | libavcodec/mips/amrwbdec_mips.h | 62 | ||||
-rw-r--r-- | libavcodec/mips/celp_filters_mips.c | 281 | ||||
-rw-r--r-- | libavcodec/mips/celp_math_mips.c | 84 | ||||
-rw-r--r-- | libavcodec/mips/compute_antialias_fixed.h | 246 | ||||
-rw-r--r-- | libavcodec/mips/compute_antialias_float.h | 182 | ||||
-rw-r--r-- | libavcodec/mips/dsputil_mips.c | 164 | ||||
-rw-r--r-- | libavcodec/mips/fft_init_table.c | 67 | ||||
-rw-r--r-- | libavcodec/mips/fft_mips.c | 530 | ||||
-rw-r--r-- | libavcodec/mips/fft_table.h | 63 | ||||
-rw-r--r-- | libavcodec/mips/fmtconvert_mips.c | 338 | ||||
-rw-r--r-- | libavcodec/mips/lsp_mips.h | 108 | ||||
-rw-r--r-- | libavcodec/mips/mathops.h | 8 | ||||
-rw-r--r-- | libavcodec/mips/mpegaudiodsp_mips_fixed.c | 903 | ||||
-rw-r--r-- | libavcodec/mips/mpegaudiodsp_mips_float.c | 1242 |
18 files changed, 4780 insertions, 4 deletions
diff --git a/libavcodec/mips/Makefile b/libavcodec/mips/Makefile new file mode 100644 index 0000000000..2f053c99c8 --- /dev/null +++ b/libavcodec/mips/Makefile @@ -0,0 +1,15 @@ +MIPSFPU-OBJS-$(CONFIG_AMRNB_DECODER) += mips/acelp_filters_mips.o \ + mips/celp_filters_mips.o \ + mips/celp_math_mips.o \ + mips/acelp_vectors_mips.o +MIPSFPU-OBJS-$(CONFIG_AMRWB_DECODER) += mips/acelp_filters_mips.o \ + mips/celp_filters_mips.o \ + mips/amrwbdec_mips.o \ + mips/celp_math_mips.o \ + mips/acelp_vectors_mips.o +MIPSFPU-OBJS-$(CONFIG_MPEGAUDIODSP) += mips/mpegaudiodsp_mips_float.o +MIPSDSPR1-OBJS-$(CONFIG_MPEGAUDIODSP) += mips/mpegaudiodsp_mips_fixed.o +OBJS-$(CONFIG_FFT) += mips/fft_init_table.o +MIPSFPU-OBJS-$(CONFIG_FFT) += mips/fft_mips.o +MIPSFPU-OBJS-$(HAVE_INLINE_ASM) += mips/fmtconvert_mips.o +MIPSFPU-OBJS-$(HAVE_INLINE_ASM) += mips/dsputil_mips.o diff --git a/libavcodec/mips/acelp_filters_mips.c b/libavcodec/mips/acelp_filters_mips.c new file mode 100644 index 0000000000..498660b448 --- /dev/null +++ b/libavcodec/mips/acelp_filters_mips.c @@ -0,0 +1,210 @@ + /* + * Copyright (c) 2012 + * MIPS Technologies, Inc., California. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its + * contributors may be used to endorse or promote products derived from + * this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * Author: Nedeljko Babic (nbabic@mips.com) + * + * various filters for ACELP-based codecs optimized for MIPS + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +/** + * @file + * Reference: libavcodec/acelp_filters.c + */ +#include "libavutil/attributes.h" +#include "libavcodec/acelp_filters.h" + +static void ff_acelp_interpolatef_mips(float *out, const float *in, + const float *filter_coeffs, int precision, + int frac_pos, int filter_length, int length) +{ + int n, i; + int prec = precision * 4; + int fc_offset = precision - frac_pos; + float in_val_p, in_val_m, fc_val_p, fc_val_m; + + for (n = 0; n < length; n++) { + /** + * four pointers are defined in order to minimize number of + * computations done in inner loop + */ + const float *p_in_p = &in[n]; + const float *p_in_m = &in[n-1]; + const float *p_filter_coeffs_p = &filter_coeffs[frac_pos]; + const float *p_filter_coeffs_m = filter_coeffs + fc_offset; + float v = 0; + + for (i = 0; i < filter_length;i++) { + __asm__ volatile ( + "lwc1 %[in_val_p], 0(%[p_in_p]) \n\t" + "lwc1 %[fc_val_p], 0(%[p_filter_coeffs_p]) \n\t" + "lwc1 %[in_val_m], 0(%[p_in_m]) \n\t" + "lwc1 %[fc_val_m], 0(%[p_filter_coeffs_m]) \n\t" + "addiu %[p_in_p], %[p_in_p], 4 \n\t" + "madd.s %[v],%[v], %[in_val_p],%[fc_val_p] \n\t" + "addiu %[p_in_m], %[p_in_m], -4 \n\t" + "addu %[p_filter_coeffs_p], %[p_filter_coeffs_p], %[prec] \n\t" + "addu %[p_filter_coeffs_m], %[p_filter_coeffs_m], %[prec] \n\t" + "madd.s %[v],%[v],%[in_val_m], %[fc_val_m] \n\t" + + : [v] "=&f" (v),[p_in_p] "+r" (p_in_p), [p_in_m] "+r" (p_in_m), + [p_filter_coeffs_p] "+r" (p_filter_coeffs_p), + [in_val_p] "=&f" (in_val_p), [in_val_m] "=&f" (in_val_m), + [fc_val_p] "=&f" (fc_val_p), [fc_val_m] "=&f" (fc_val_m), + [p_filter_coeffs_m] "+r" (p_filter_coeffs_m) + : [prec] "r" (prec) + ); + } + out[n] = v; + } +} + +static void ff_acelp_apply_order_2_transfer_function_mips(float *out, const float *in, + const float zero_coeffs[2], + const float pole_coeffs[2], + float gain, float mem[2], int n) +{ + /** + * loop is unrolled eight times + */ + + __asm__ volatile ( + "lwc1 $f0, 0(%[mem]) \n\t" + "blez %[n], ff_acelp_apply_order_2_transfer_function_end%= \n\t" + "lwc1 $f1, 4(%[mem]) \n\t" + "lwc1 $f2, 0(%[pole_coeffs]) \n\t" + "lwc1 $f3, 4(%[pole_coeffs]) \n\t" + "lwc1 $f4, 0(%[zero_coeffs]) \n\t" + "lwc1 $f5, 4(%[zero_coeffs]) \n\t" + + "ff_acelp_apply_order_2_transfer_function_madd%=: \n\t" + + "lwc1 $f6, 0(%[in]) \n\t" + "mul.s $f9, $f3, $f1 \n\t" + "mul.s $f7, $f2, $f0 \n\t" + "msub.s $f7, $f7, %[gain], $f6 \n\t" + "sub.s $f7, $f7, $f9 \n\t" + "madd.s $f8, $f7, $f4, $f0 \n\t" + "madd.s $f8, $f8, $f5, $f1 \n\t" + "lwc1 $f11, 4(%[in]) \n\t" + "mul.s $f12, $f3, $f0 \n\t" + "mul.s $f13, $f2, $f7 \n\t" + "msub.s $f13, $f13, %[gain], $f11 \n\t" + "sub.s $f13, $f13, $f12 \n\t" + "madd.s $f14, $f13, $f4, $f7 \n\t" + "madd.s $f14, $f14, $f5, $f0 \n\t" + "swc1 $f8, 0(%[out]) \n\t" + "lwc1 $f6, 8(%[in]) \n\t" + "mul.s $f9, $f3, $f7 \n\t" + "mul.s $f15, $f2, $f13 \n\t" + "msub.s $f15, $f15, %[gain], $f6 \n\t" + "sub.s $f15, $f15, $f9 \n\t" + "madd.s $f8, $f15, $f4, $f13 \n\t" + "madd.s $f8, $f8, $f5, $f7 \n\t" + "swc1 $f14, 4(%[out]) \n\t" + "lwc1 $f11, 12(%[in]) \n\t" + "mul.s $f12, $f3, $f13 \n\t" + "mul.s $f16, $f2, $f15 \n\t" + "msub.s $f16, $f16, %[gain], $f11 \n\t" + "sub.s $f16, $f16, $f12 \n\t" + "madd.s $f14, $f16, $f4, $f15 \n\t" + "madd.s $f14, $f14, $f5, $f13 \n\t" + "swc1 $f8, 8(%[out]) \n\t" + "lwc1 $f6, 16(%[in]) \n\t" + "mul.s $f9, $f3, $f15 \n\t" + "mul.s $f7, $f2, $f16 \n\t" + "msub.s $f7, $f7, %[gain], $f6 \n\t" + "sub.s $f7, $f7, $f9 \n\t" + "madd.s $f8, $f7, $f4, $f16 \n\t" + "madd.s $f8, $f8, $f5, $f15 \n\t" + "swc1 $f14, 12(%[out]) \n\t" + "lwc1 $f11, 20(%[in]) \n\t" + "mul.s $f12, $f3, $f16 \n\t" + "mul.s $f13, $f2, $f7 \n\t" + "msub.s $f13, $f13, %[gain], $f11 \n\t" + "sub.s $f13, $f13, $f12 \n\t" + "madd.s $f14, $f13, $f4, $f7 \n\t" + "madd.s $f14, $f14, $f5, $f16 \n\t" + "swc1 $f8, 16(%[out]) \n\t" + "lwc1 $f6, 24(%[in]) \n\t" + "mul.s $f9, $f3, $f7 \n\t" + "mul.s $f15, $f2, $f13 \n\t" + "msub.s $f15, $f15, %[gain], $f6 \n\t" + "sub.s $f1, $f15, $f9 \n\t" + "madd.s $f8, $f1, $f4, $f13 \n\t" + "madd.s $f8, $f8, $f5, $f7 \n\t" + "swc1 $f14, 20(%[out]) \n\t" + "lwc1 $f11, 28(%[in]) \n\t" + "mul.s $f12, $f3, $f13 \n\t" + "mul.s $f16, $f2, $f1 \n\t" + "msub.s $f16, $f16, %[gain], $f11 \n\t" + "sub.s $f0, $f16, $f12 \n\t" + "madd.s $f14, $f0, $f4, $f1 \n\t" + "madd.s $f14, $f14, $f5, $f13 \n\t" + "swc1 $f8, 24(%[out]) \n\t" + "addiu %[out], 32 \n\t" + "addiu %[in], 32 \n\t" + "addiu %[n], -8 \n\t" + "swc1 $f14, -4(%[out]) \n\t" + "bnez %[n], ff_acelp_apply_order_2_transfer_function_madd%= \n\t" + "swc1 $f1, 4(%[mem]) \n\t" + "swc1 $f0, 0(%[mem]) \n\t" + + "ff_acelp_apply_order_2_transfer_function_end%=: \n\t" + + : [out] "+r" (out), + [in] "+r" (in), [gain] "+f" (gain), + [n] "+r" (n), [mem] "+r" (mem) + : [zero_coeffs] "r" (zero_coeffs), + [pole_coeffs] "r" (pole_coeffs) + : "$f0", "$f1", "$f2", "$f3", "$f4", "$f5", + "$f6", "$f7", "$f8", "$f9", "$f10", "$f11", + "$f12", "$f13", "$f14", "$f15", "$f16" + ); +} + +void ff_acelp_filter_init_mips(ACELPFContext *c) +{ + c->acelp_interpolatef = ff_acelp_interpolatef_mips; + c->acelp_apply_order_2_transfer_function = ff_acelp_apply_order_2_transfer_function_mips; +} diff --git a/libavcodec/mips/acelp_vectors_mips.c b/libavcodec/mips/acelp_vectors_mips.c new file mode 100644 index 0000000000..6f9390f225 --- /dev/null +++ b/libavcodec/mips/acelp_vectors_mips.c @@ -0,0 +1,96 @@ +/* + * Copyright (c) 2012 + * MIPS Technologies, Inc., California. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its + * contributors may be used to endorse or promote products derived from + * this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * Author: Nedeljko Babic (nbabic@mips.com) + * + * adaptive and fixed codebook vector operations for ACELP-based codecs + * optimized for MIPS + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +/** + * @file + * Reference: libavcodec/acelp_vectors.c + */ +#include "libavcodec/acelp_vectors.h" + +static void ff_weighted_vector_sumf_mips( + float *out, const float *in_a, const float *in_b, + float weight_coeff_a, float weight_coeff_b, int length) +{ + const float *a_end = in_a + length; + + /* loop unrolled two times */ + __asm__ volatile ( + "blez %[length], ff_weighted_vector_sumf_end%= \n\t" + + "ff_weighted_vector_sumf_madd%=: \n\t" + "lwc1 $f0, 0(%[in_a]) \n\t" + "lwc1 $f3, 4(%[in_a]) \n\t" + "lwc1 $f1, 0(%[in_b]) \n\t" + "lwc1 $f4, 4(%[in_b]) \n\t" + "mul.s $f2, %[weight_coeff_a], $f0 \n\t" + "mul.s $f5, %[weight_coeff_a], $f3 \n\t" + "madd.s $f2, $f2, %[weight_coeff_b], $f1 \n\t" + "madd.s $f5, $f5, %[weight_coeff_b], $f4 \n\t" + "addiu %[in_a], 8 \n\t" + "addiu %[in_b], 8 \n\t" + "swc1 $f2, 0(%[out]) \n\t" + "swc1 $f5, 4(%[out]) \n\t" + "addiu %[out], 8 \n\t" + "bne %[in_a], %[a_end], ff_weighted_vector_sumf_madd%= \n\t" + + "ff_weighted_vector_sumf_end%=: \n\t" + + : [out] "+r" (out), [in_a] "+r" (in_a), [in_b] "+r" (in_b) + : [weight_coeff_a] "f" (weight_coeff_a), + [weight_coeff_b] "f" (weight_coeff_b), + [length] "r" (length), [a_end]"r"(a_end) + : "$f0", "$f1", "$f2", "$f3", "$f4", "$f5" + ); +} + +void ff_acelp_vectors_init_mips(ACELPVContext *c) +{ + c->weighted_vector_sumf = ff_weighted_vector_sumf_mips; +} diff --git a/libavcodec/mips/amrwbdec_mips.c b/libavcodec/mips/amrwbdec_mips.c new file mode 100644 index 0000000000..4bfbb8c6b9 --- /dev/null +++ b/libavcodec/mips/amrwbdec_mips.c @@ -0,0 +1,185 @@ +/* + * Copyright (c) 2012 + * MIPS Technologies, Inc., California. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its + * contributors may be used to endorse or promote products derived from + * this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * Author: Nedeljko Babic (nbabic@mips.com) + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +/** + * @file + * Reference: libavcodec/amrwbdec.c + */ +#include "libavutil/avutil.h" +#include "libavcodec/amrwbdata.h" +#include "amrwbdec_mips.h" + +void hb_fir_filter_mips(float *out, const float fir_coef[HB_FIR_SIZE + 1], + float mem[HB_FIR_SIZE], const float *in) +{ + int i; + float data[AMRWB_SFR_SIZE_16k + HB_FIR_SIZE]; // past and current samples + + memcpy(data, mem, HB_FIR_SIZE * sizeof(float)); + memcpy(data + HB_FIR_SIZE, in, AMRWB_SFR_SIZE_16k * sizeof(float)); + + for (i = 0; i < AMRWB_SFR_SIZE_16k; i++) { + float output; + float * p_data = (data+i); + + /** + * inner loop is entirely unrolled and instructions are scheduled + * to minimize pipeline stall + */ + __asm__ volatile( + "mtc1 $zero, %[output] \n\t" + "lwc1 $f0, 0(%[p_data]) \n\t" + "lwc1 $f1, 0(%[fir_coef]) \n\t" + "lwc1 $f2, 4(%[p_data]) \n\t" + "madd.s %[output], %[output], $f0, $f1 \n\t" + "lwc1 $f3, 4(%[fir_coef]) \n\t" + "lwc1 $f4, 8(%[p_data]) \n\t" + "madd.s %[output], %[output], $f2, $f3 \n\t" + "lwc1 $f5, 8(%[fir_coef]) \n\t" + + "lwc1 $f0, 12(%[p_data]) \n\t" + "lwc1 $f1, 12(%[fir_coef]) \n\t" + "madd.s %[output], %[output], $f4, $f5 \n\t" + "lwc1 $f2, 16(%[p_data]) \n\t" + "madd.s %[output], %[output], $f0, $f1 \n\t" + "lwc1 $f3, 16(%[fir_coef]) \n\t" + "lwc1 $f4, 20(%[p_data]) \n\t" + "lwc1 $f5, 20(%[fir_coef]) \n\t" + "madd.s %[output], %[output], $f2, $f3 \n\t" + + "lwc1 $f0, 24(%[p_data]) \n\t" + "lwc1 $f1, 24(%[fir_coef]) \n\t" + "lwc1 $f2, 28(%[p_data]) \n\t" + "madd.s %[output], %[output], $f4, $f5 \n\t" + "lwc1 $f3, 28(%[fir_coef]) \n\t" + "madd.s %[output], %[output], $f0, $f1 \n\t" + "lwc1 $f4, 32(%[p_data]) \n\t" + "madd.s %[output], %[output], $f2, $f3 \n\t" + "lwc1 $f5, 32(%[fir_coef]) \n\t" + "madd.s %[output], %[output], $f4, $f5 \n\t" + + "lwc1 $f0, 36(%[p_data]) \n\t" + "lwc1 $f1, 36(%[fir_coef]) \n\t" + "lwc1 $f2, 40(%[p_data]) \n\t" + "lwc1 $f3, 40(%[fir_coef]) \n\t" + "madd.s %[output], %[output], $f0, $f1 \n\t" + "lwc1 $f4, 44(%[p_data]) \n\t" + "lwc1 $f5, 44(%[fir_coef]) \n\t" + "madd.s %[output], %[output], $f2, $f3 \n\t" + + "lwc1 $f0, 48(%[p_data]) \n\t" + "lwc1 $f1, 48(%[fir_coef]) \n\t" + "lwc1 $f2, 52(%[p_data]) \n\t" + "madd.s %[output], %[output], $f4, $f5 \n\t" + "lwc1 $f3, 52(%[fir_coef]) \n\t" + "lwc1 $f4, 56(%[p_data]) \n\t" + "madd.s %[output], %[output], $f0, $f1 \n\t" + "lwc1 $f5, 56(%[fir_coef]) \n\t" + "madd.s %[output], %[output], $f2, $f3 \n\t" + + "lwc1 $f0, 60(%[p_data]) \n\t" + "lwc1 $f1, 60(%[fir_coef]) \n\t" + "lwc1 $f2, 64(%[p_data]) \n\t" + "madd.s %[output], %[output], $f4, $f5 \n\t" + "lwc1 $f3, 64(%[fir_coef]) \n\t" + "madd.s %[output], %[output], $f0, $f1 \n\t" + "lwc1 $f4, 68(%[p_data]) \n\t" + "madd.s %[output], %[output], $f2, $f3 \n\t" + "lwc1 $f5, 68(%[fir_coef]) \n\t" + "madd.s %[output], %[output], $f4, $f5 \n\t" + + "lwc1 $f0, 72(%[p_data]) \n\t" + "lwc1 $f1, 72(%[fir_coef]) \n\t" + "lwc1 $f2, 76(%[p_data]) \n\t" + "lwc1 $f3, 76(%[fir_coef]) \n\t" + "madd.s %[output], %[output], $f0, $f1 \n\t" + "lwc1 $f4, 80(%[p_data]) \n\t" + "lwc1 $f5, 80(%[fir_coef]) \n\t" + "madd.s %[output], %[output], $f2, $f3 \n\t" + + "lwc1 $f0, 84(%[p_data]) \n\t" + "lwc1 $f1, 84(%[fir_coef]) \n\t" + "lwc1 $f2, 88(%[p_data]) \n\t" + "madd.s %[output], %[output], $f4, $f5 \n\t" + "lwc1 $f3, 88(%[fir_coef]) \n\t" + "lwc1 $f4, 92(%[p_data]) \n\t" + "madd.s %[output], %[output], $f0, $f1 \n\t" + "lwc1 $f5, 92(%[fir_coef]) \n\t" + "madd.s %[output], %[output], $f2, $f3 \n\t" + + "lwc1 $f0, 96(%[p_data]) \n\t" + "lwc1 $f1, 96(%[fir_coef]) \n\t" + "lwc1 $f2, 100(%[p_data]) \n\t" + "madd.s %[output], %[output], $f4, $f5 \n\t" + "lwc1 $f3, 100(%[fir_coef]) \n\t" + "lwc1 $f4, 104(%[p_data]) \n\t" + "madd.s %[output], %[output], $f0, $f1 \n\t" + "lwc1 $f5, 104(%[fir_coef]) \n\t" + "madd.s %[output], %[output], $f2, $f3 \n\t" + + "lwc1 $f0, 108(%[p_data]) \n\t" + "lwc1 $f1, 108(%[fir_coef]) \n\t" + "madd.s %[output], %[output], $f4, $f5 \n\t" + "lwc1 $f2, 112(%[p_data]) \n\t" + "lwc1 $f3, 112(%[fir_coef]) \n\t" + "madd.s %[output], %[output], $f0, $f1 \n\t" + "lwc1 $f4, 116(%[p_data]) \n\t" + "lwc1 $f5, 116(%[fir_coef]) \n\t" + "lwc1 $f0, 120(%[p_data]) \n\t" + "madd.s %[output], %[output], $f2, $f3 \n\t" + "lwc1 $f1, 120(%[fir_coef]) \n\t" + "madd.s %[output], %[output], $f4, $f5 \n\t" + "madd.s %[output], %[output], $f0, $f1 \n\t" + + : [output]"=&f"(output) + : [fir_coef]"r"(fir_coef), [p_data]"r"(p_data) + : "$f0", "$f1", "$f2", "$f3", "$f4", "$f5" + ); + out[i] = output; + } + memcpy(mem, data + AMRWB_SFR_SIZE_16k, HB_FIR_SIZE * sizeof(float)); +} diff --git a/libavcodec/mips/amrwbdec_mips.h b/libavcodec/mips/amrwbdec_mips.h new file mode 100644 index 0000000000..a469918d2c --- /dev/null +++ b/libavcodec/mips/amrwbdec_mips.h @@ -0,0 +1,62 @@ +/* + * Copyright (c) 2012 + * MIPS Technologies, Inc., California. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its + * contributors may be used to endorse or promote products derived from + * this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * Author: Nedeljko Babic (nbabic@mips.com) + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +/** + * @file + * Reference: libavcodec/amrwbdec.c + */ +#ifndef AVCODEC_AMRWBDEC_MIPS_H +#define AVCODEC_AMRWBDEC_MIPS_H +#include "config.h" + +#if HAVE_MIPSFPU && HAVE_INLINE_ASM +void hb_fir_filter_mips(float *out, const float fir_coef[], + float mem[], const float *in); +#define hb_fir_filter hb_fir_filter_mips +#endif + +#endif /* AVCODEC_AMRWBDEC_MIPS_H */ diff --git a/libavcodec/mips/celp_filters_mips.c b/libavcodec/mips/celp_filters_mips.c new file mode 100644 index 0000000000..06e0127b19 --- /dev/null +++ b/libavcodec/mips/celp_filters_mips.c @@ -0,0 +1,281 @@ +/* + * Copyright (c) 2012 + * MIPS Technologies, Inc., California. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its + * contributors may be used to endorse or promote products derived from + * this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * Author: Nedeljko Babic (nbabic@mips.com) + * + * various filters for CELP-based codecs optimized for MIPS + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +/** + * @file + * Reference: libavcodec/celp_filters.c + */ +#include "libavutil/attributes.h" +#include "libavutil/common.h" +#include "libavcodec/celp_filters.h" + +static void ff_celp_lp_synthesis_filterf_mips(float *out, + const float *filter_coeffs, + const float* in, int buffer_length, + int filter_length) +{ + int i,n; + + float out0, out1, out2, out3; + float old_out0, old_out1, old_out2, old_out3; + float a,b,c; + const float *p_filter_coeffs; + float *p_out; + + a = filter_coeffs[0]; + b = filter_coeffs[1]; + c = filter_coeffs[2]; + b -= filter_coeffs[0] * filter_coeffs[0]; + c -= filter_coeffs[1] * filter_coeffs[0]; + c -= filter_coeffs[0] * b; + + old_out0 = out[-4]; + old_out1 = out[-3]; + old_out2 = out[-2]; + old_out3 = out[-1]; + for (n = 0; n <= buffer_length - 4; n+=4) { + p_filter_coeffs = filter_coeffs; + p_out = out; + + out0 = in[0]; + out1 = in[1]; + out2 = in[2]; + out3 = in[3]; + + __asm__ volatile( + "lwc1 $f2, 8(%[filter_coeffs]) \n\t" + "lwc1 $f1, 4(%[filter_coeffs]) \n\t" + "lwc1 $f0, 0(%[filter_coeffs]) \n\t" + "nmsub.s %[out0], %[out0], $f2, %[old_out1] \n\t" + "nmsub.s %[out1], %[out1], $f2, %[old_out2] \n\t" + "nmsub.s %[out2], %[out2], $f2, %[old_out3] \n\t" + "lwc1 $f3, 12(%[filter_coeffs]) \n\t" + "nmsub.s %[out0], %[out0], $f1, %[old_out2] \n\t" + "nmsub.s %[out1], %[out1], $f1, %[old_out3] \n\t" + "nmsub.s %[out2], %[out2], $f3, %[old_out2] \n\t" + "nmsub.s %[out0], %[out0], $f0, %[old_out3] \n\t" + "nmsub.s %[out3], %[out3], $f3, %[old_out3] \n\t" + "nmsub.s %[out1], %[out1], $f3, %[old_out1] \n\t" + "nmsub.s %[out0], %[out0], $f3, %[old_out0] \n\t" + + : [out0]"+f"(out0), [out1]"+f"(out1), + [out2]"+f"(out2), [out3]"+f"(out3) + : [old_out0]"f"(old_out0), [old_out1]"f"(old_out1), + [old_out2]"f"(old_out2), [old_out3]"f"(old_out3), + [filter_coeffs]"r"(filter_coeffs) + : "$f0", "$f1", "$f2", "$f3", "$f4" + ); + + for (i = 5; i <= filter_length; i += 2) { + __asm__ volatile( + "lwc1 %[old_out3], -20(%[p_out]) \n\t" + "lwc1 $f5, 16(%[p_filter_coeffs]) \n\t" + "addiu %[p_out], -8 \n\t" + "addiu %[p_filter_coeffs], 8 \n\t" + "nmsub.s %[out1], %[out1], $f5, %[old_out0] \n\t" + "nmsub.s %[out3], %[out3], $f5, %[old_out2] \n\t" + "lwc1 $f4, 12(%[p_filter_coeffs]) \n\t" + "lwc1 %[old_out2], -16(%[p_out]) \n\t" + "nmsub.s %[out0], %[out0], $f5, %[old_out3] \n\t" + "nmsub.s %[out2], %[out2], $f5, %[old_out1] \n\t" + "nmsub.s %[out1], %[out1], $f4, %[old_out3] \n\t" + "nmsub.s %[out3], %[out3], $f4, %[old_out1] \n\t" + "mov.s %[old_out1], %[old_out3] \n\t" + "nmsub.s %[out0], %[out0], $f4, %[old_out2] \n\t" + "nmsub.s %[out2], %[out2], $f4, %[old_out0] \n\t" + + : [out0]"+f"(out0), [out1]"+f"(out1), + [out2]"+f"(out2), [out3]"+f"(out3), [old_out0]"+f"(old_out0), + [old_out1]"+f"(old_out1), [old_out2]"+f"(old_out2), + [old_out3]"+f"(old_out3),[p_filter_coeffs]"+r"(p_filter_coeffs), + [p_out]"+r"(p_out) + : + : "$f4", "$f5" + ); + FFSWAP(float, old_out0, old_out2); + } + + __asm__ volatile( + "nmsub.s %[out3], %[out3], %[a], %[out2] \n\t" + "nmsub.s %[out2], %[out2], %[a], %[out1] \n\t" + "nmsub.s %[out3], %[out3], %[b], %[out1] \n\t" + "nmsub.s %[out1], %[out1], %[a], %[out0] \n\t" + "nmsub.s %[out2], %[out2], %[b], %[out0] \n\t" + "nmsub.s %[out3], %[out3], %[c], %[out0] \n\t" + + : [out0]"+f"(out0), [out1]"+f"(out1), + [out2]"+f"(out2), [out3]"+f"(out3) + : [a]"f"(a), [b]"f"(b), [c]"f"(c) + ); + + out[0] = out0; + out[1] = out1; + out[2] = out2; + out[3] = out3; + + old_out0 = out0; + old_out1 = out1; + old_out2 = out2; + old_out3 = out3; + + out += 4; + in += 4; + } + + out -= n; + in -= n; + for (; n < buffer_length; n++) { + float out_val, out_val_i, fc_val; + p_filter_coeffs = filter_coeffs; + p_out = &out[n]; + out_val = in[n]; + for (i = 1; i <= filter_length; i++) { + __asm__ volatile( + "lwc1 %[fc_val], 0(%[p_filter_coeffs]) \n\t" + "lwc1 %[out_val_i], -4(%[p_out]) \n\t" + "addiu %[p_filter_coeffs], 4 \n\t" + "addiu %[p_out], -4 \n\t" + "nmsub.s %[out_val], %[out_val], %[fc_val], %[out_val_i] \n\t" + + : [fc_val]"=&f"(fc_val), [out_val]"+f"(out_val), + [out_val_i]"=&f"(out_val_i), [p_out]"+r"(p_out), + [p_filter_coeffs]"+r"(p_filter_coeffs) + ); + } + out[n] = out_val; + } +} + +static void ff_celp_lp_zero_synthesis_filterf_mips(float *out, + const float *filter_coeffs, + const float *in, int buffer_length, + int filter_length) +{ + int i,n; + float sum_out8, sum_out7, sum_out6, sum_out5, sum_out4, fc_val; + float sum_out3, sum_out2, sum_out1; + const float *p_filter_coeffs, *p_in; + + for (n = 0; n < buffer_length; n+=8) { + p_in = &in[n]; + p_filter_coeffs = filter_coeffs; + sum_out8 = in[n+7]; + sum_out7 = in[n+6]; + sum_out6 = in[n+5]; + sum_out5 = in[n+4]; + sum_out4 = in[n+3]; + sum_out3 = in[n+2]; + sum_out2 = in[n+1]; + sum_out1 = in[n]; + i = filter_length; + + /* i is always greater than 0 + * outer loop is unrolled eight times so there is less memory access + * inner loop is unrolled two times + */ + __asm__ volatile( + "filt_lp_inner%=: \n\t" + "lwc1 %[fc_val], 0(%[p_filter_coeffs]) \n\t" + "lwc1 $f7, 6*4(%[p_in]) \n\t" + "lwc1 $f6, 5*4(%[p_in]) \n\t" + "lwc1 $f5, 4*4(%[p_in]) \n\t" + "lwc1 $f4, 3*4(%[p_in]) \n\t" + "lwc1 $f3, 2*4(%[p_in]) \n\t" + "lwc1 $f2, 4(%[p_in]) \n\t" + "lwc1 $f1, 0(%[p_in]) \n\t" + "lwc1 $f0, -4(%[p_in]) \n\t" + "addiu %[i], -2 \n\t" + "madd.s %[sum_out8], %[sum_out8], %[fc_val], $f7 \n\t" + "madd.s %[sum_out7], %[sum_out7], %[fc_val], $f6 \n\t" + "madd.s %[sum_out6], %[sum_out6], %[fc_val], $f5 \n\t" + "madd.s %[sum_out5], %[sum_out5], %[fc_val], $f4 \n\t" + "madd.s %[sum_out4], %[sum_out4], %[fc_val], $f3 \n\t" + "madd.s %[sum_out3], %[sum_out3], %[fc_val], $f2 \n\t" + "madd.s %[sum_out2], %[sum_out2], %[fc_val], $f1 \n\t" + "madd.s %[sum_out1], %[sum_out1], %[fc_val], $f0 \n\t" + "lwc1 %[fc_val], 4(%[p_filter_coeffs]) \n\t" + "lwc1 $f7, -8(%[p_in]) \n\t" + "addiu %[p_filter_coeffs], 8 \n\t" + "addiu %[p_in], -8 \n\t" + "madd.s %[sum_out8], %[sum_out8], %[fc_val], $f6 \n\t" + "madd.s %[sum_out7], %[sum_out7], %[fc_val], $f5 \n\t" + "madd.s %[sum_out6], %[sum_out6], %[fc_val], $f4 \n\t" + "madd.s %[sum_out5], %[sum_out5], %[fc_val], $f3 \n\t" + "madd.s %[sum_out4], %[sum_out4], %[fc_val], $f2 \n\t" + "madd.s %[sum_out3], %[sum_out3], %[fc_val], $f1 \n\t" + "madd.s %[sum_out2], %[sum_out2], %[fc_val], $f0 \n\t" + "madd.s %[sum_out1], %[sum_out1], %[fc_val], $f7 \n\t" + "bgtz %[i], filt_lp_inner%= \n\t" + + : [sum_out8]"+f"(sum_out8), [sum_out7]"+f"(sum_out7), + [sum_out6]"+f"(sum_out6), [sum_out5]"+f"(sum_out5), + [sum_out4]"+f"(sum_out4), [sum_out3]"+f"(sum_out3), + [sum_out2]"+f"(sum_out2), [sum_out1]"+f"(sum_out1), + [fc_val]"=&f"(fc_val), [p_filter_coeffs]"+r"(p_filter_coeffs), + [p_in]"+r"(p_in), [i]"+r"(i) + : + : "$f0", "$f1", "$f2", "$f3", "$f4", "$f5", "$f6", "$f7" + ); + + out[n+7] = sum_out8; + out[n+6] = sum_out7; + out[n+5] = sum_out6; + out[n+4] = sum_out5; + out[n+3] = sum_out4; + out[n+2] = sum_out3; + out[n+1] = sum_out2; + out[n] = sum_out1; + } +} + +void ff_celp_filter_init_mips(CELPFContext *c) +{ + c->celp_lp_synthesis_filterf = ff_celp_lp_synthesis_filterf_mips; + c->celp_lp_zero_synthesis_filterf = ff_celp_lp_zero_synthesis_filterf_mips; +} diff --git a/libavcodec/mips/celp_math_mips.c b/libavcodec/mips/celp_math_mips.c new file mode 100644 index 0000000000..6ab1823d77 --- /dev/null +++ b/libavcodec/mips/celp_math_mips.c @@ -0,0 +1,84 @@ +/* + * Copyright (c) 2012 + * MIPS Technologies, Inc., California. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its + * contributors may be used to endorse or promote products derived from + * this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * Author: Nedeljko Babic (nbabic@mips.com) + * + * Math operations optimized for MIPS + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +/** + * @file + * Reference: libavcodec/celp_math.c + */ +#include "libavcodec/celp_math.h" + +static float ff_dot_productf_mips(const float* a, const float* b, + int length) +{ + float sum; + const float* a_end = a + length; + + __asm__ volatile ( + "mtc1 $zero, %[sum] \n\t" + "blez %[length], ff_dot_productf_end%= \n\t" + "ff_dot_productf_madd%=: \n\t" + "lwc1 $f2, 0(%[a]) \n\t" + "lwc1 $f1, 0(%[b]) \n\t" + "addiu %[a], %[a], 4 \n\t" + "addiu %[b], %[b], 4 \n\t" + "madd.s %[sum], %[sum], $f1, $f2 \n\t" + "bne %[a], %[a_end], ff_dot_productf_madd%= \n\t" + "ff_dot_productf_end%=: \n\t" + + : [sum] "=&f" (sum), [a] "+r" (a), [b] "+r" (b) + : [a_end]"r"(a_end), [length] "r" (length) + : "$f1", "$f2" + ); + return sum; +} + +void ff_celp_math_init_mips(CELPMContext *c) +{ + c->dot_productf = ff_dot_productf_mips; +} diff --git a/libavcodec/mips/compute_antialias_fixed.h b/libavcodec/mips/compute_antialias_fixed.h new file mode 100644 index 0000000000..528411fff2 --- /dev/null +++ b/libavcodec/mips/compute_antialias_fixed.h @@ -0,0 +1,246 @@ +/* + * Copyright (c) 2012 + * MIPS Technologies, Inc., California. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its + * contributors may be used to endorse or promote products derived from + * this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * Author: Bojan Zivkovic (bojan@mips.com) + * + * Compute antialias function optimised for MIPS fixed-point architecture + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +/** + * @file + * Reference: libavcodec/mpegaudiodec.c + */ + +#ifndef AVCODEC_MIPS_COMPUTE_ANTIALIAS_FIXED_H +#define AVCODEC_MIPS_COMPUTE_ANTIALIAS_FIXED_H + +static void compute_antialias_mips_fixed(MPADecodeContext *s, + GranuleDef *g) +{ + int32_t *ptr, *csa; + int n, i; + int MAX_lo = 0xffffffff; + + /* we antialias only "long" bands */ + if (g->block_type == 2) { + if (!g->switch_point) + return; + /* XXX: check this for 8000Hz case */ + n = 1; + } else { + n = SBLIMIT - 1; + } + + + ptr = g->sb_hybrid + 18; + + for(i = n;i > 0;i--) { + int tmp0, tmp1, tmp2, tmp00, tmp11; + int temp_reg1, temp_reg2, temp_reg3, temp_reg4, temp_reg5, temp_reg6; + csa = &csa_table[0][0]; + + /** + * instructions are scheduled to minimize pipeline stall. + */ + __asm__ volatile ( + "lw %[tmp0], -1*4(%[ptr]) \n\t" + "lw %[tmp1], 0*4(%[ptr]) \n\t" + "lw %[temp_reg1], 0*4(%[csa]) \n\t" + "lw %[temp_reg2], 2*4(%[csa]) \n\t" + "add %[tmp2], %[tmp0], %[tmp1] \n\t" + "lw %[temp_reg3], 3*4(%[csa]) \n\t" + "mult $ac0, %[tmp2], %[temp_reg1] \n\t" + "mult $ac1, %[tmp2], %[temp_reg1] \n\t" + "lw %[tmp00], -2*4(%[ptr]) \n\t" + "lw %[tmp11], 1*4(%[ptr]) \n\t" + "lw %[temp_reg4], 4*4(%[csa]) \n\t" + "mtlo %[MAX_lo], $ac0 \n\t" + "mtlo $zero, $ac1 \n\t" + "msub $ac0, %[tmp1], %[temp_reg2] \n\t" + "madd $ac1, %[tmp0], %[temp_reg3] \n\t" + "add %[tmp2], %[tmp00], %[tmp11] \n\t" + "lw %[temp_reg5], 6*4(%[csa]) \n\t" + "mult $ac2, %[tmp2], %[temp_reg4] \n\t" + "mult $ac3, %[tmp2], %[temp_reg4] \n\t" + "mfhi %[temp_reg1], $ac0 \n\t" + "mfhi %[temp_reg2], $ac1 \n\t" + "lw %[temp_reg6], 7*4(%[csa]) \n\t" + "mtlo %[MAX_lo], $ac2 \n\t" + "msub $ac2, %[tmp11], %[temp_reg5] \n\t" + "mtlo $zero, $ac3 \n\t" + "madd $ac3, %[tmp00], %[temp_reg6] \n\t" + "sll %[temp_reg1], %[temp_reg1], 2 \n\t" + "sw %[temp_reg1], -1*4(%[ptr]) \n\t" + "mfhi %[temp_reg4], $ac2 \n\t" + "sll %[temp_reg2], %[temp_reg2], 2 \n\t" + "mfhi %[temp_reg5], $ac3 \n\t" + "sw %[temp_reg2], 0*4(%[ptr]) \n\t" + "lw %[tmp0], -3*4(%[ptr]) \n\t" + "lw %[tmp1], 2*4(%[ptr]) \n\t" + "lw %[temp_reg1], 8*4(%[csa]) \n\t" + "sll %[temp_reg4], %[temp_reg4], 2 \n\t" + "add %[tmp2], %[tmp0], %[tmp1] \n\t" + "sll %[temp_reg5], %[temp_reg5], 2 \n\t" + "mult $ac0, %[tmp2], %[temp_reg1] \n\t" + "mult $ac1, %[tmp2], %[temp_reg1] \n\t" + "sw %[temp_reg4], -2*4(%[ptr]) \n\t" + "sw %[temp_reg5], 1*4(%[ptr]) \n\t" + "lw %[temp_reg2], 10*4(%[csa]) \n\t" + "mtlo %[MAX_lo], $ac0 \n\t" + "lw %[temp_reg3], 11*4(%[csa]) \n\t" + "msub $ac0, %[tmp1], %[temp_reg2] \n\t" + "mtlo $zero, $ac1 \n\t" + "madd $ac1, %[tmp0], %[temp_reg3] \n\t" + "lw %[tmp00], -4*4(%[ptr]) \n\t" + "lw %[tmp11], 3*4(%[ptr]) \n\t" + "mfhi %[temp_reg1], $ac0 \n\t" + "lw %[temp_reg4], 12*4(%[csa]) \n\t" + "mfhi %[temp_reg2], $ac1 \n\t" + "add %[tmp2], %[tmp00], %[tmp11] \n\t" + "mult $ac2, %[tmp2], %[temp_reg4] \n\t" + "mult $ac3, %[tmp2], %[temp_reg4] \n\t" + "lw %[temp_reg5], 14*4(%[csa]) \n\t" + "lw %[temp_reg6], 15*4(%[csa]) \n\t" + "sll %[temp_reg1], %[temp_reg1], 2 \n\t" + "mtlo %[MAX_lo], $ac2 \n\t" + "msub $ac2, %[tmp11], %[temp_reg5] \n\t" + "mtlo $zero, $ac3 \n\t" + "madd $ac3, %[tmp00], %[temp_reg6] \n\t" + "sll %[temp_reg2], %[temp_reg2], 2 \n\t" + "sw %[temp_reg1], -3*4(%[ptr]) \n\t" + "mfhi %[temp_reg4], $ac2 \n\t" + "sw %[temp_reg2], 2*4(%[ptr]) \n\t" + "mfhi %[temp_reg5], $ac3 \n\t" + "lw %[tmp0], -5*4(%[ptr]) \n\t" + "lw %[tmp1], 4*4(%[ptr]) \n\t" + "lw %[temp_reg1], 16*4(%[csa]) \n\t" + "lw %[temp_reg2], 18*4(%[csa]) \n\t" + "add %[tmp2], %[tmp0], %[tmp1] \n\t" + "lw %[temp_reg3], 19*4(%[csa]) \n\t" + "mult $ac0, %[tmp2], %[temp_reg1] \n\t" + "mult $ac1, %[tmp2], %[temp_reg1] \n\t" + "sll %[temp_reg4], %[temp_reg4], 2 \n\t" + "sll %[temp_reg5], %[temp_reg5], 2 \n\t" + "sw %[temp_reg4], -4*4(%[ptr]) \n\t" + "mtlo %[MAX_lo], $ac0 \n\t" + "msub $ac0, %[tmp1], %[temp_reg2] \n\t" + "mtlo $zero, $ac1 \n\t" + "madd $ac1, %[tmp0], %[temp_reg3] \n\t" + "sw %[temp_reg5], 3*4(%[ptr]) \n\t" + "lw %[tmp00], -6*4(%[ptr]) \n\t" + "mfhi %[temp_reg1], $ac0 \n\t" + "lw %[tmp11], 5*4(%[ptr]) \n\t" + "mfhi %[temp_reg2], $ac1 \n\t" + "lw %[temp_reg4], 20*4(%[csa]) \n\t" + "add %[tmp2], %[tmp00], %[tmp11] \n\t" + "lw %[temp_reg5], 22*4(%[csa]) \n\t" + "mult $ac2, %[tmp2], %[temp_reg4] \n\t" + "mult $ac3, %[tmp2], %[temp_reg4] \n\t" + "lw %[temp_reg6], 23*4(%[csa]) \n\t" + "sll %[temp_reg1], %[temp_reg1], 2 \n\t" + "sll %[temp_reg2], %[temp_reg2], 2 \n\t" + "mtlo %[MAX_lo], $ac2 \n\t" + "msub $ac2, %[tmp11], %[temp_reg5] \n\t" + "mtlo $zero, $ac3 \n\t" + "madd $ac3, %[tmp00], %[temp_reg6] \n\t" + "sw %[temp_reg1], -5*4(%[ptr]) \n\t" + "sw %[temp_reg2], 4*4(%[ptr]) \n\t" + "mfhi %[temp_reg4], $ac2 \n\t" + "lw %[tmp0], -7*4(%[ptr]) \n\t" + "mfhi %[temp_reg5], $ac3 \n\t" + "lw %[tmp1], 6*4(%[ptr]) \n\t" + "lw %[temp_reg1], 24*4(%[csa]) \n\t" + "lw %[temp_reg2], 26*4(%[csa]) \n\t" + "add %[tmp2], %[tmp0], %[tmp1] \n\t" + "lw %[temp_reg3], 27*4(%[csa]) \n\t" + "mult $ac0, %[tmp2], %[temp_reg1] \n\t" + "mult $ac1, %[tmp2], %[temp_reg1] \n\t" + "sll %[temp_reg4], %[temp_reg4], 2 \n\t" + "sll %[temp_reg5], %[temp_reg5], 2 \n\t" + "sw %[temp_reg4], -6*4(%[ptr]) \n\t" + "mtlo %[MAX_lo], $ac0 \n\t" + "msub $ac0, %[tmp1], %[temp_reg2] \n\t" + "mtlo $zero, $ac1 \n\t" + "madd $ac1, %[tmp0], %[temp_reg3] \n\t" + "sw %[temp_reg5], 5*4(%[ptr]) \n\t" + "lw %[tmp00], -8*4(%[ptr]) \n\t" + "mfhi %[temp_reg1], $ac0 \n\t" + "lw %[tmp11], 7*4(%[ptr]) \n\t" + "mfhi %[temp_reg2], $ac1 \n\t" + "lw %[temp_reg4], 28*4(%[csa]) \n\t" + "add %[tmp2], %[tmp00], %[tmp11] \n\t" + "lw %[temp_reg5], 30*4(%[csa]) \n\t" + "mult $ac2, %[tmp2], %[temp_reg4] \n\t" + "mult $ac3, %[tmp2], %[temp_reg4] \n\t" + "lw %[temp_reg6], 31*4(%[csa]) \n\t" + "sll %[temp_reg1], %[temp_reg1], 2 \n\t" + "sll %[temp_reg2], %[temp_reg2], 2 \n\t" + "mtlo %[MAX_lo], $ac2 \n\t" + "msub $ac2, %[tmp11], %[temp_reg5] \n\t" + "mtlo $zero, $ac3 \n\t" + "madd $ac3, %[tmp00], %[temp_reg6] \n\t" + "sw %[temp_reg1], -7*4(%[ptr]) \n\t" + "sw %[temp_reg2], 6*4(%[ptr]) \n\t" + "mfhi %[temp_reg4], $ac2 \n\t" + "mfhi %[temp_reg5], $ac3 \n\t" + "sll %[temp_reg4], %[temp_reg4], 2 \n\t" + "sll %[temp_reg5], %[temp_reg5], 2 \n\t" + "sw %[temp_reg4], -8*4(%[ptr]) \n\t" + "sw %[temp_reg5], 7*4(%[ptr]) \n\t" + + : [tmp0] "=&r" (tmp0), [tmp1] "=&r" (tmp1), [tmp2] "=&r" (tmp2), + [tmp00] "=&r" (tmp00), [tmp11] "=&r" (tmp11), + [temp_reg1] "=&r" (temp_reg1), [temp_reg2] "=&r" (temp_reg2), + [temp_reg3] "=&r" (temp_reg3), [temp_reg4] "=&r" (temp_reg4), + [temp_reg5] "=&r" (temp_reg5), [temp_reg6] "=&r" (temp_reg6) + : [csa] "r" (csa), [ptr] "r" (ptr), + [MAX_lo] "r" (MAX_lo) + ); + + ptr += 18; + } +} +#define compute_antialias compute_antialias_mips_fixed + +#endif /* AVCODEC_MIPS_COMPUTE_ANTIALIAS_FIXED_H */ diff --git a/libavcodec/mips/compute_antialias_float.h b/libavcodec/mips/compute_antialias_float.h new file mode 100644 index 0000000000..3a3f5d6dde --- /dev/null +++ b/libavcodec/mips/compute_antialias_float.h @@ -0,0 +1,182 @@ +/* + * Copyright (c) 2012 + * MIPS Technologies, Inc., California. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its + * contributors may be used to endorse or promote products derived from + * this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * Author: Bojan Zivkovic (bojan@mips.com) + * + * Compute antialias function optimised for MIPS floating-point architecture + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +/** + * @file + * Reference: libavcodec/mpegaudiodec.c + */ + +#ifndef AVCODEC_MIPS_COMPUTE_ANTIALIAS_FLOAT_H +#define AVCODEC_MIPS_COMPUTE_ANTIALIAS_FLOAT_H + +static void compute_antialias_mips_float(MPADecodeContext *s, + GranuleDef *g) +{ + float *ptr, *ptr_end; + float *csa = &csa_table[0][0]; + int n; + /* temporary variables */ + float in1, in2, in3, in4, in5, in6, in7, in8; + float out1, out2, out3, out4; + + ptr = g->sb_hybrid + 18; + /* we antialias only "long" bands */ + if (g->block_type == 2) { + if (!g->switch_point) + return; + /* XXX: check this for 8000Hz case */ + n = 1; + ptr_end = ptr + 18; + } else { + n = 31; + ptr_end = ptr + 558; + } + + /** + * instructions are scheduled to minimize pipeline stall. + */ + + __asm__ volatile ( + "compute_antialias_float_loop%=: \t\n" + "lwc1 %[in1], -1*4(%[ptr]) \t\n" + "lwc1 %[in2], 0(%[csa]) \t\n" + "lwc1 %[in3], 1*4(%[csa]) \t\n" + "lwc1 %[in4], 0(%[ptr]) \t\n" + "lwc1 %[in5], -2*4(%[ptr]) \t\n" + "lwc1 %[in6], 4*4(%[csa]) \t\n" + "mul.s %[out1], %[in1], %[in2] \t\n" + "mul.s %[out2], %[in1], %[in3] \t\n" + "lwc1 %[in7], 5*4(%[csa]) \t\n" + "lwc1 %[in8], 1*4(%[ptr]) \t\n" + "nmsub.s %[out1], %[out1], %[in3], %[in4] \t\n" + "madd.s %[out2], %[out2], %[in2], %[in4] \t\n" + "mul.s %[out3], %[in5], %[in6] \t\n" + "mul.s %[out4], %[in5], %[in7] \t\n" + "lwc1 %[in1], -3*4(%[ptr]) \t\n" + "swc1 %[out1], -1*4(%[ptr]) \t\n" + "swc1 %[out2], 0(%[ptr]) \t\n" + "nmsub.s %[out3], %[out3], %[in7], %[in8] \t\n" + "madd.s %[out4], %[out4], %[in6], %[in8] \t\n" + "lwc1 %[in2], 8*4(%[csa]) \t\n" + "swc1 %[out3], -2*4(%[ptr]) \t\n" + "swc1 %[out4], 1*4(%[ptr]) \t\n" + "lwc1 %[in3], 9*4(%[csa]) \t\n" + "lwc1 %[in4], 2*4(%[ptr]) \t\n" + "mul.s %[out1], %[in1], %[in2] \t\n" + "lwc1 %[in5], -4*4(%[ptr]) \t\n" + "lwc1 %[in6], 12*4(%[csa]) \t\n" + "mul.s %[out2], %[in1], %[in3] \t\n" + "lwc1 %[in7], 13*4(%[csa]) \t\n" + "nmsub.s %[out1], %[out1], %[in3], %[in4] \t\n" + "lwc1 %[in8], 3*4(%[ptr]) \t\n" + "mul.s %[out3], %[in5], %[in6] \t\n" + "madd.s %[out2], %[out2], %[in2], %[in4] \t\n" + "mul.s %[out4], %[in5], %[in7] \t\n" + "swc1 %[out1], -3*4(%[ptr]) \t\n" + "lwc1 %[in1], -5*4(%[ptr]) \t\n" + "nmsub.s %[out3], %[out3], %[in7], %[in8] \t\n" + "swc1 %[out2], 2*4(%[ptr]) \t\n" + "madd.s %[out4], %[out4], %[in6], %[in8] \t\n" + "lwc1 %[in2], 16*4(%[csa]) \t\n" + "lwc1 %[in3], 17*4(%[csa]) \t\n" + "swc1 %[out3], -4*4(%[ptr]) \t\n" + "lwc1 %[in4], 4*4(%[ptr]) \t\n" + "swc1 %[out4], 3*4(%[ptr]) \t\n" + "mul.s %[out1], %[in1], %[in2] \t\n" + "mul.s %[out2], %[in1], %[in3] \t\n" + "lwc1 %[in5], -6*4(%[ptr]) \t\n" + "lwc1 %[in6], 20*4(%[csa]) \t\n" + "lwc1 %[in7], 21*4(%[csa]) \t\n" + "nmsub.s %[out1], %[out1], %[in3], %[in4] \t\n" + "madd.s %[out2], %[out2], %[in2], %[in4] \t\n" + "lwc1 %[in8], 5*4(%[ptr]) \t\n" + "mul.s %[out3], %[in5], %[in6] \t\n" + "mul.s %[out4], %[in5], %[in7] \t\n" + "swc1 %[out1], -5*4(%[ptr]) \t\n" + "swc1 %[out2], 4*4(%[ptr]) \t\n" + "lwc1 %[in1], -7*4(%[ptr]) \t\n" + "nmsub.s %[out3], %[out3], %[in7], %[in8] \t\n" + "madd.s %[out4], %[out4], %[in6], %[in8] \t\n" + "lwc1 %[in2], 24*4(%[csa]) \t\n" + "lwc1 %[in3], 25*4(%[csa]) \t\n" + "lwc1 %[in4], 6*4(%[ptr]) \t\n" + "swc1 %[out3], -6*4(%[ptr]) \t\n" + "swc1 %[out4], 5*4(%[ptr]) \t\n" + "mul.s %[out1], %[in1], %[in2] \t\n" + "lwc1 %[in5], -8*4(%[ptr]) \t\n" + "mul.s %[out2], %[in1], %[in3] \t\n" + "lwc1 %[in6], 28*4(%[csa]) \t\n" + "lwc1 %[in7], 29*4(%[csa]) \t\n" + "nmsub.s %[out1], %[out1], %[in3], %[in4] \t\n" + "lwc1 %[in8], 7*4(%[ptr]) \t\n" + "madd.s %[out2], %[out2], %[in2], %[in4] \t\n" + "mul.s %[out3], %[in5], %[in6] \t\n" + "mul.s %[out4], %[in5], %[in7] \t\n" + "swc1 %[out1], -7*4(%[ptr]) \t\n" + "swc1 %[out2], 6*4(%[ptr]) \t\n" + "addiu %[ptr], %[ptr], 72 \t\n" + "nmsub.s %[out3], %[out3], %[in7], %[in8] \t\n" + "madd.s %[out4], %[out4], %[in6], %[in8] \t\n" + "swc1 %[out3], -26*4(%[ptr]) \t\n" + "swc1 %[out4], -11*4(%[ptr]) \t\n" + "bne %[ptr], %[ptr_end], compute_antialias_float_loop%= \t\n" + + : [ptr] "+r" (ptr), + [in1] "=&f" (in1), [in2] "=&f" (in2), + [in3] "=&f" (in3), [in4] "=&f" (in4), + [in5] "=&f" (in5), [in6] "=&f" (in6), + [in7] "=&f" (in7), [in8] "=&f" (in8), + [out1] "=&f" (out1), [out2] "=&f" (out2), + [out3] "=&f" (out3), [out4] "=&f" (out4) + : [csa] "r" (csa), [ptr_end] "r" (ptr_end) + ); +} +#define compute_antialias compute_antialias_mips_float + +#endif /* AVCODEC_MIPS_COMPUTE_ANTIALIAS_FLOAT_H */ diff --git a/libavcodec/mips/dsputil_mips.c b/libavcodec/mips/dsputil_mips.c new file mode 100644 index 0000000000..e46a0a949f --- /dev/null +++ b/libavcodec/mips/dsputil_mips.c @@ -0,0 +1,164 @@ +/* + * Copyright (c) 2012 + * MIPS Technologies, Inc., California. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. Neither the name of the MIPS Technologies, Inc., nor the names of is + * contributors may be used to endorse or promote products derived from + * this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * Author: Zoran Lukic (zoranl@mips.com) + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ +#include "config.h" +#include "libavcodec/dsputil.h" + +static void vector_fmul_window_mips(float *dst, const float *src0, + const float *src1, const float *win, int len) +{ + int i, j; + /* + * variables used in inline assembler + */ + float * dst_i, * dst_j, * dst_i2, * dst_j2; + float temp, temp1, temp2, temp3, temp4, temp5, temp6, temp7; + + dst += len; + win += len; + src0 += len; + + for (i = -len, j = len - 1; i < 0; i += 8, j -= 8) { + + dst_i = dst + i; + dst_j = dst + j; + + dst_i2 = dst + i + 4; + dst_j2 = dst + j - 4; + + __asm__ volatile ( + "mul.s %[temp], %[s1], %[wi] \n\t" + "mul.s %[temp1], %[s1], %[wj] \n\t" + "mul.s %[temp2], %[s11], %[wi1] \n\t" + "mul.s %[temp3], %[s11], %[wj1] \n\t" + + "msub.s %[temp], %[temp], %[s0], %[wj] \n\t" + "madd.s %[temp1], %[temp1], %[s0], %[wi] \n\t" + "msub.s %[temp2], %[temp2], %[s01], %[wj1] \n\t" + "madd.s %[temp3], %[temp3], %[s01], %[wi1] \n\t" + + "swc1 %[temp], 0(%[dst_i]) \n\t" /* dst[i] = s0*wj - s1*wi; */ + "swc1 %[temp1], 0(%[dst_j]) \n\t" /* dst[j] = s0*wi + s1*wj; */ + "swc1 %[temp2], 4(%[dst_i]) \n\t" /* dst[i+1] = s01*wj1 - s11*wi1; */ + "swc1 %[temp3], -4(%[dst_j]) \n\t" /* dst[j-1] = s01*wi1 + s11*wj1; */ + + "mul.s %[temp4], %[s12], %[wi2] \n\t" + "mul.s %[temp5], %[s12], %[wj2] \n\t" + "mul.s %[temp6], %[s13], %[wi3] \n\t" + "mul.s %[temp7], %[s13], %[wj3] \n\t" + + "msub.s %[temp4], %[temp4], %[s02], %[wj2] \n\t" + "madd.s %[temp5], %[temp5], %[s02], %[wi2] \n\t" + "msub.s %[temp6], %[temp6], %[s03], %[wj3] \n\t" + "madd.s %[temp7], %[temp7], %[s03], %[wi3] \n\t" + + "swc1 %[temp4], 8(%[dst_i]) \n\t" /* dst[i+2] = s02*wj2 - s12*wi2; */ + "swc1 %[temp5], -8(%[dst_j]) \n\t" /* dst[j-2] = s02*wi2 + s12*wj2; */ + "swc1 %[temp6], 12(%[dst_i]) \n\t" /* dst[i+2] = s03*wj3 - s13*wi3; */ + "swc1 %[temp7], -12(%[dst_j]) \n\t" /* dst[j-3] = s03*wi3 + s13*wj3; */ + : [temp]"=&f"(temp), [temp1]"=&f"(temp1), [temp2]"=&f"(temp2), + [temp3]"=&f"(temp3), [temp4]"=&f"(temp4), [temp5]"=&f"(temp5), + [temp6]"=&f"(temp6), [temp7]"=&f"(temp7) + : [dst_j]"r"(dst_j), [dst_i]"r" (dst_i), + [s0] "f"(src0[i]), [wj] "f"(win[j]), [s1] "f"(src1[j]), + [wi] "f"(win[i]), [s01]"f"(src0[i + 1]),[wj1]"f"(win[j - 1]), + [s11]"f"(src1[j - 1]), [wi1]"f"(win[i + 1]), [s02]"f"(src0[i + 2]), + [wj2]"f"(win[j - 2]), [s12]"f"(src1[j - 2]),[wi2]"f"(win[i + 2]), + [s03]"f"(src0[i + 3]), [wj3]"f"(win[j - 3]), [s13]"f"(src1[j - 3]), + [wi3]"f"(win[i + 3]) + : "memory" + ); + + __asm__ volatile ( + "mul.s %[temp], %[s1], %[wi] \n\t" + "mul.s %[temp1], %[s1], %[wj] \n\t" + "mul.s %[temp2], %[s11], %[wi1] \n\t" + "mul.s %[temp3], %[s11], %[wj1] \n\t" + + "msub.s %[temp], %[temp], %[s0], %[wj] \n\t" + "madd.s %[temp1], %[temp1], %[s0], %[wi] \n\t" + "msub.s %[temp2], %[temp2], %[s01], %[wj1] \n\t" + "madd.s %[temp3], %[temp3], %[s01], %[wi1] \n\t" + + "swc1 %[temp], 0(%[dst_i2]) \n\t" /* dst[i] = s0*wj - s1*wi; */ + "swc1 %[temp1], 0(%[dst_j2]) \n\t" /* dst[j] = s0*wi + s1*wj; */ + "swc1 %[temp2], 4(%[dst_i2]) \n\t" /* dst[i+1] = s01*wj1 - s11*wi1; */ + "swc1 %[temp3], -4(%[dst_j2]) \n\t" /* dst[j-1] = s01*wi1 + s11*wj1; */ + + "mul.s %[temp4], %[s12], %[wi2] \n\t" + "mul.s %[temp5], %[s12], %[wj2] \n\t" + "mul.s %[temp6], %[s13], %[wi3] \n\t" + "mul.s %[temp7], %[s13], %[wj3] \n\t" + + "msub.s %[temp4], %[temp4], %[s02], %[wj2] \n\t" + "madd.s %[temp5], %[temp5], %[s02], %[wi2] \n\t" + "msub.s %[temp6], %[temp6], %[s03], %[wj3] \n\t" + "madd.s %[temp7], %[temp7], %[s03], %[wi3] \n\t" + + "swc1 %[temp4], 8(%[dst_i2]) \n\t" /* dst[i+2] = s02*wj2 - s12*wi2; */ + "swc1 %[temp5], -8(%[dst_j2]) \n\t" /* dst[j-2] = s02*wi2 + s12*wj2; */ + "swc1 %[temp6], 12(%[dst_i2]) \n\t" /* dst[i+2] = s03*wj3 - s13*wi3; */ + "swc1 %[temp7], -12(%[dst_j2]) \n\t" /* dst[j-3] = s03*wi3 + s13*wj3; */ + : [temp]"=&f"(temp), + [temp1]"=&f"(temp1), [temp2]"=&f"(temp2), [temp3]"=&f"(temp3), + [temp4]"=&f"(temp4), [temp5]"=&f"(temp5), [temp6]"=&f"(temp6), + [temp7] "=&f" (temp7) + : [dst_j2]"r"(dst_j2), [dst_i2]"r"(dst_i2), + [s0] "f"(src0[i + 4]), [wj] "f"(win[j - 4]), [s1] "f"(src1[j - 4]), + [wi] "f"(win[i + 4]), [s01]"f"(src0[i + 5]),[wj1]"f"(win[j - 5]), + [s11]"f"(src1[j - 5]), [wi1]"f"(win[i + 5]), [s02]"f"(src0[i + 6]), + [wj2]"f"(win[j - 6]), [s12]"f"(src1[j - 6]),[wi2]"f"(win[i + 6]), + [s03]"f"(src0[i + 7]), [wj3]"f"(win[j - 7]), [s13]"f"(src1[j - 7]), + [wi3]"f"(win[i + 7]) + : "memory" + ); + } +} + +av_cold void ff_dsputil_init_mips( DSPContext* c, AVCodecContext *avctx ) +{ + c->vector_fmul_window = vector_fmul_window_mips; +} diff --git a/libavcodec/mips/fft_init_table.c b/libavcodec/mips/fft_init_table.c new file mode 100644 index 0000000000..9c2e998e9c --- /dev/null +++ b/libavcodec/mips/fft_init_table.c @@ -0,0 +1,67 @@ +/* + * Copyright (c) 2012 + * MIPS Technologies, Inc., California. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its + * contributors may be used to endorse or promote products derived from + * this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * Author: Stanislav Ocovaj (socovaj@mips.com) + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +/** + * @file + * definitions and initialization of LUT table for MIPS FFT + */ +#include "fft_table.h" + +uint16_t fft_offsets_lut[0x2aab]; + +void ff_fft_lut_init(uint16_t *table, int off, int size, int *index) +{ + if (size < 16) { + table[*index] = off >> 2; + (*index)++; + } + else { + ff_fft_lut_init(table, off, size>>1, index); + ff_fft_lut_init(table, off+(size>>1), size>>2, index); + ff_fft_lut_init(table, off+3*(size>>2), size>>2, index); + } +} diff --git a/libavcodec/mips/fft_mips.c b/libavcodec/mips/fft_mips.c new file mode 100644 index 0000000000..2b1c50899c --- /dev/null +++ b/libavcodec/mips/fft_mips.c @@ -0,0 +1,530 @@ +/* + * Copyright (c) 2012 + * MIPS Technologies, Inc., California. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its + * contributors may be used to endorse or promote products derived from + * this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * Author: Stanislav Ocovaj (socovaj@mips.com) + * Author: Zoran Lukic (zoranl@mips.com) + * + * Optimized MDCT/IMDCT and FFT transforms + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ +#include "config.h" +#include "libavcodec/fft.h" +#include "fft_table.h" + +/** + * FFT transform + */ + +#if HAVE_INLINE_ASM +static void ff_fft_calc_mips(FFTContext *s, FFTComplex *z) +{ + int nbits, i, n, num_transforms, offset, step; + int n4, n2, n34; + FFTSample tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8; + FFTComplex *tmpz; + float w_re, w_im; + float *w_re_ptr, *w_im_ptr; + const int fft_size = (1 << s->nbits); + int s_n = s->nbits; + int tem1, tem2; + float pom, pom1, pom2, pom3; + float temp, temp1, temp3, temp4; + FFTComplex * tmpz_n2, * tmpz_n34, * tmpz_n4; + FFTComplex * tmpz_n2_i, * tmpz_n34_i, * tmpz_n4_i, * tmpz_i; + + /** + *num_transforms = (0x2aab >> (16 - s->nbits)) | 1; + */ + __asm__ volatile ( + "li %[tem1], 16 \n\t" + "sub %[s_n], %[tem1], %[s_n] \n\t" + "li %[tem2], 10923 \n\t" + "srav %[tem2], %[tem2], %[s_n] \n\t" + "ori %[num_t],%[tem2], 1 \n\t" + : [num_t]"=r"(num_transforms), [s_n]"+r"(s_n), + [tem1]"=&r"(tem1), [tem2]"=&r"(tem2) + ); + + + for (n=0; n<num_transforms; n++) { + offset = fft_offsets_lut[n] << 2; + tmpz = z + offset; + + tmp1 = tmpz[0].re + tmpz[1].re; + tmp5 = tmpz[2].re + tmpz[3].re; + tmp2 = tmpz[0].im + tmpz[1].im; + tmp6 = tmpz[2].im + tmpz[3].im; + tmp3 = tmpz[0].re - tmpz[1].re; + tmp8 = tmpz[2].im - tmpz[3].im; + tmp4 = tmpz[0].im - tmpz[1].im; + tmp7 = tmpz[2].re - tmpz[3].re; + + tmpz[0].re = tmp1 + tmp5; + tmpz[2].re = tmp1 - tmp5; + tmpz[0].im = tmp2 + tmp6; + tmpz[2].im = tmp2 - tmp6; + tmpz[1].re = tmp3 + tmp8; + tmpz[3].re = tmp3 - tmp8; + tmpz[1].im = tmp4 - tmp7; + tmpz[3].im = tmp4 + tmp7; + + } + + if (fft_size < 8) + return; + + num_transforms = (num_transforms >> 1) | 1; + + for (n=0; n<num_transforms; n++) { + offset = fft_offsets_lut[n] << 3; + tmpz = z + offset; + + __asm__ volatile ( + "lwc1 %[tmp1], 32(%[tmpz]) \n\t" + "lwc1 %[pom], 40(%[tmpz]) \n\t" + "lwc1 %[tmp3], 48(%[tmpz]) \n\t" + "lwc1 %[pom1], 56(%[tmpz]) \n\t" + "lwc1 %[tmp2], 36(%[tmpz]) \n\t" + "lwc1 %[pom2], 44(%[tmpz]) \n\t" + "lwc1 %[pom3], 60(%[tmpz]) \n\t" + "lwc1 %[tmp4], 52(%[tmpz]) \n\t" + "add.s %[tmp1], %[tmp1], %[pom] \n\t" // tmp1 = tmpz[4].re + tmpz[5].re; + "add.s %[tmp3], %[tmp3], %[pom1] \n\t" // tmp3 = tmpz[6].re + tmpz[7].re; + "add.s %[tmp2], %[tmp2], %[pom2] \n\t" // tmp2 = tmpz[4].im + tmpz[5].im; + "lwc1 %[pom], 40(%[tmpz]) \n\t" + "add.s %[tmp4], %[tmp4], %[pom3] \n\t" // tmp4 = tmpz[6].im + tmpz[7].im; + "add.s %[tmp5], %[tmp1], %[tmp3] \n\t" // tmp5 = tmp1 + tmp3; + "sub.s %[tmp7], %[tmp1], %[tmp3] \n\t" // tmp7 = tmp1 - tmp3; + "lwc1 %[tmp1], 32(%[tmpz]) \n\t" + "lwc1 %[pom1], 44(%[tmpz]) \n\t" + "add.s %[tmp6], %[tmp2], %[tmp4] \n\t" // tmp6 = tmp2 + tmp4; + "sub.s %[tmp8], %[tmp2], %[tmp4] \n\t" // tmp8 = tmp2 - tmp4; + "lwc1 %[tmp2], 36(%[tmpz]) \n\t" + "lwc1 %[pom2], 56(%[tmpz]) \n\t" + "lwc1 %[pom3], 60(%[tmpz]) \n\t" + "lwc1 %[tmp3], 48(%[tmpz]) \n\t" + "lwc1 %[tmp4], 52(%[tmpz]) \n\t" + "sub.s %[tmp1], %[tmp1], %[pom] \n\t" // tmp1 = tmpz[4].re - tmpz[5].re; + "lwc1 %[pom], 0(%[tmpz]) \n\t" + "sub.s %[tmp2], %[tmp2], %[pom1] \n\t" // tmp2 = tmpz[4].im - tmpz[5].im; + "sub.s %[tmp3], %[tmp3], %[pom2] \n\t" // tmp3 = tmpz[6].re - tmpz[7].re; + "lwc1 %[pom2], 4(%[tmpz]) \n\t" + "sub.s %[pom1], %[pom], %[tmp5] \n\t" + "sub.s %[tmp4], %[tmp4], %[pom3] \n\t" // tmp4 = tmpz[6].im - tmpz[7].im; + "add.s %[pom3], %[pom], %[tmp5] \n\t" + "sub.s %[pom], %[pom2], %[tmp6] \n\t" + "add.s %[pom2], %[pom2], %[tmp6] \n\t" + "swc1 %[pom1], 32(%[tmpz]) \n\t" // tmpz[4].re = tmpz[0].re - tmp5; + "swc1 %[pom3], 0(%[tmpz]) \n\t" // tmpz[0].re = tmpz[0].re + tmp5; + "swc1 %[pom], 36(%[tmpz]) \n\t" // tmpz[4].im = tmpz[0].im - tmp6; + "swc1 %[pom2], 4(%[tmpz]) \n\t" // tmpz[0].im = tmpz[0].im + tmp6; + "lwc1 %[pom1], 16(%[tmpz]) \n\t" + "lwc1 %[pom3], 20(%[tmpz]) \n\t" + "li.s %[pom], 0.7071067812 \n\t" // float pom = 0.7071067812f; + "add.s %[temp1],%[tmp1], %[tmp2] \n\t" + "sub.s %[temp], %[pom1], %[tmp8] \n\t" + "add.s %[pom2], %[pom3], %[tmp7] \n\t" + "sub.s %[temp3],%[tmp3], %[tmp4] \n\t" + "sub.s %[temp4],%[tmp2], %[tmp1] \n\t" + "swc1 %[temp], 48(%[tmpz]) \n\t" // tmpz[6].re = tmpz[2].re - tmp8; + "swc1 %[pom2], 52(%[tmpz]) \n\t" // tmpz[6].im = tmpz[2].im + tmp7; + "add.s %[pom1], %[pom1], %[tmp8] \n\t" + "sub.s %[pom3], %[pom3], %[tmp7] \n\t" + "add.s %[tmp3], %[tmp3], %[tmp4] \n\t" + "mul.s %[tmp5], %[pom], %[temp1] \n\t" // tmp5 = pom * (tmp1 + tmp2); + "mul.s %[tmp7], %[pom], %[temp3] \n\t" // tmp7 = pom * (tmp3 - tmp4); + "mul.s %[tmp6], %[pom], %[temp4] \n\t" // tmp6 = pom * (tmp2 - tmp1); + "mul.s %[tmp8], %[pom], %[tmp3] \n\t" // tmp8 = pom * (tmp3 + tmp4); + "swc1 %[pom1], 16(%[tmpz]) \n\t" // tmpz[2].re = tmpz[2].re + tmp8; + "swc1 %[pom3], 20(%[tmpz]) \n\t" // tmpz[2].im = tmpz[2].im - tmp7; + "add.s %[tmp1], %[tmp5], %[tmp7] \n\t" // tmp1 = tmp5 + tmp7; + "sub.s %[tmp3], %[tmp5], %[tmp7] \n\t" // tmp3 = tmp5 - tmp7; + "add.s %[tmp2], %[tmp6], %[tmp8] \n\t" // tmp2 = tmp6 + tmp8; + "sub.s %[tmp4], %[tmp6], %[tmp8] \n\t" // tmp4 = tmp6 - tmp8; + "lwc1 %[temp], 8(%[tmpz]) \n\t" + "lwc1 %[temp1],12(%[tmpz]) \n\t" + "lwc1 %[pom], 24(%[tmpz]) \n\t" + "lwc1 %[pom2], 28(%[tmpz]) \n\t" + "sub.s %[temp4],%[temp], %[tmp1] \n\t" + "sub.s %[temp3],%[temp1], %[tmp2] \n\t" + "add.s %[temp], %[temp], %[tmp1] \n\t" + "add.s %[temp1],%[temp1], %[tmp2] \n\t" + "sub.s %[pom1], %[pom], %[tmp4] \n\t" + "add.s %[pom3], %[pom2], %[tmp3] \n\t" + "add.s %[pom], %[pom], %[tmp4] \n\t" + "sub.s %[pom2], %[pom2], %[tmp3] \n\t" + "swc1 %[temp4],40(%[tmpz]) \n\t" // tmpz[5].re = tmpz[1].re - tmp1; + "swc1 %[temp3],44(%[tmpz]) \n\t" // tmpz[5].im = tmpz[1].im - tmp2; + "swc1 %[temp], 8(%[tmpz]) \n\t" // tmpz[1].re = tmpz[1].re + tmp1; + "swc1 %[temp1],12(%[tmpz]) \n\t" // tmpz[1].im = tmpz[1].im + tmp2; + "swc1 %[pom1], 56(%[tmpz]) \n\t" // tmpz[7].re = tmpz[3].re - tmp4; + "swc1 %[pom3], 60(%[tmpz]) \n\t" // tmpz[7].im = tmpz[3].im + tmp3; + "swc1 %[pom], 24(%[tmpz]) \n\t" // tmpz[3].re = tmpz[3].re + tmp4; + "swc1 %[pom2], 28(%[tmpz]) \n\t" // tmpz[3].im = tmpz[3].im - tmp3; + : [tmp1]"=&f"(tmp1), [pom]"=&f"(pom), [pom1]"=&f"(pom1), [pom2]"=&f"(pom2), + [tmp3]"=&f"(tmp3), [tmp2]"=&f"(tmp2), [tmp4]"=&f"(tmp4), [tmp5]"=&f"(tmp5), [tmp7]"=&f"(tmp7), + [tmp6]"=&f"(tmp6), [tmp8]"=&f"(tmp8), [pom3]"=&f"(pom3),[temp]"=&f"(temp), [temp1]"=&f"(temp1), + [temp3]"=&f"(temp3), [temp4]"=&f"(temp4) + : [tmpz]"r"(tmpz) + : "memory" + ); + } + + step = 1 << (MAX_LOG2_NFFT - 4); + n4 = 4; + + for (nbits=4; nbits<=s->nbits; nbits++) { + /* + * num_transforms = (num_transforms >> 1) | 1; + */ + __asm__ volatile ( + "sra %[num_t], %[num_t], 1 \n\t" + "ori %[num_t], %[num_t], 1 \n\t" + + : [num_t] "+r" (num_transforms) + ); + n2 = 2 * n4; + n34 = 3 * n4; + + for (n=0; n<num_transforms; n++) { + offset = fft_offsets_lut[n] << nbits; + tmpz = z + offset; + + tmpz_n2 = tmpz + n2; + tmpz_n4 = tmpz + n4; + tmpz_n34 = tmpz + n34; + + __asm__ volatile ( + "lwc1 %[pom1], 0(%[tmpz_n2]) \n\t" + "lwc1 %[pom], 0(%[tmpz_n34]) \n\t" + "lwc1 %[pom2], 4(%[tmpz_n2]) \n\t" + "lwc1 %[pom3], 4(%[tmpz_n34]) \n\t" + "lwc1 %[temp1],0(%[tmpz]) \n\t" + "lwc1 %[temp3],4(%[tmpz]) \n\t" + "add.s %[tmp5], %[pom1], %[pom] \n\t" // tmp5 = tmpz[ n2].re + tmpz[n34].re; + "sub.s %[tmp1], %[pom1], %[pom] \n\t" // tmp1 = tmpz[ n2].re - tmpz[n34].re; + "add.s %[tmp6], %[pom2], %[pom3] \n\t" // tmp6 = tmpz[ n2].im + tmpz[n34].im; + "sub.s %[tmp2], %[pom2], %[pom3] \n\t" // tmp2 = tmpz[ n2].im - tmpz[n34].im; + "sub.s %[temp], %[temp1], %[tmp5] \n\t" + "add.s %[temp1],%[temp1], %[tmp5] \n\t" + "sub.s %[temp4],%[temp3], %[tmp6] \n\t" + "add.s %[temp3],%[temp3], %[tmp6] \n\t" + "swc1 %[temp], 0(%[tmpz_n2]) \n\t" // tmpz[ n2].re = tmpz[ 0].re - tmp5; + "swc1 %[temp1],0(%[tmpz]) \n\t" // tmpz[ 0].re = tmpz[ 0].re + tmp5; + "lwc1 %[pom1], 0(%[tmpz_n4]) \n\t" + "swc1 %[temp4],4(%[tmpz_n2]) \n\t" // tmpz[ n2].im = tmpz[ 0].im - tmp6; + "lwc1 %[temp], 4(%[tmpz_n4]) \n\t" + "swc1 %[temp3],4(%[tmpz]) \n\t" // tmpz[ 0].im = tmpz[ 0].im + tmp6; + "sub.s %[pom], %[pom1], %[tmp2] \n\t" + "add.s %[pom1], %[pom1], %[tmp2] \n\t" + "add.s %[temp1],%[temp], %[tmp1] \n\t" + "sub.s %[temp], %[temp], %[tmp1] \n\t" + "swc1 %[pom], 0(%[tmpz_n34]) \n\t" // tmpz[n34].re = tmpz[n4].re - tmp2; + "swc1 %[pom1], 0(%[tmpz_n4]) \n\t" // tmpz[ n4].re = tmpz[n4].re + tmp2; + "swc1 %[temp1],4(%[tmpz_n34]) \n\t" // tmpz[n34].im = tmpz[n4].im + tmp1; + "swc1 %[temp], 4(%[tmpz_n4]) \n\t" // tmpz[ n4].im = tmpz[n4].im - tmp1; + : [tmp5]"=&f"(tmp5), + [tmp1]"=&f"(tmp1), [pom]"=&f"(pom), [pom1]"=&f"(pom1), [pom2]"=&f"(pom2), + [tmp2]"=&f"(tmp2), [tmp6]"=&f"(tmp6), [pom3]"=&f"(pom3), + [temp]"=&f"(temp), [temp1]"=&f"(temp1), [temp3]"=&f"(temp3), [temp4]"=&f"(temp4) + : [tmpz]"r"(tmpz), [tmpz_n2]"r"(tmpz_n2), [tmpz_n34]"r"(tmpz_n34), [tmpz_n4]"r"(tmpz_n4) + : "memory" + ); + + w_re_ptr = (float*)(ff_cos_65536 + step); + w_im_ptr = (float*)(ff_cos_65536 + MAX_FFT_SIZE/4 - step); + + for (i=1; i<n4; i++) { + w_re = w_re_ptr[0]; + w_im = w_im_ptr[0]; + tmpz_n2_i = tmpz_n2 + i; + tmpz_n4_i = tmpz_n4 + i; + tmpz_n34_i= tmpz_n34 + i; + tmpz_i = tmpz + i; + + __asm__ volatile ( + "lwc1 %[temp], 0(%[tmpz_n2_i]) \n\t" + "lwc1 %[temp1], 4(%[tmpz_n2_i]) \n\t" + "lwc1 %[pom], 0(%[tmpz_n34_i]) \n\t" + "lwc1 %[pom1], 4(%[tmpz_n34_i]) \n\t" + "mul.s %[temp3], %[w_im], %[temp] \n\t" + "mul.s %[temp4], %[w_im], %[temp1] \n\t" + "mul.s %[pom2], %[w_im], %[pom1] \n\t" + "mul.s %[pom3], %[w_im], %[pom] \n\t" + "msub.s %[tmp2], %[temp3], %[w_re], %[temp1] \n\t" // tmp2 = w_re * tmpz[ n2+i].im - w_im * tmpz[ n2+i].re; + "madd.s %[tmp1], %[temp4], %[w_re], %[temp] \n\t" // tmp1 = w_re * tmpz[ n2+i].re + w_im * tmpz[ n2+i].im; + "msub.s %[tmp3], %[pom2], %[w_re], %[pom] \n\t" // tmp3 = w_re * tmpz[n34+i].re - w_im * tmpz[n34+i].im; + "madd.s %[tmp4], %[pom3], %[w_re], %[pom1] \n\t" // tmp4 = w_re * tmpz[n34+i].im + w_im * tmpz[n34+i].re; + "lwc1 %[temp], 0(%[tmpz_i]) \n\t" + "lwc1 %[pom], 4(%[tmpz_i]) \n\t" + "add.s %[tmp5], %[tmp1], %[tmp3] \n\t" // tmp5 = tmp1 + tmp3; + "sub.s %[tmp1], %[tmp1], %[tmp3] \n\t" // tmp1 = tmp1 - tmp3; + "add.s %[tmp6], %[tmp2], %[tmp4] \n\t" // tmp6 = tmp2 + tmp4; + "sub.s %[tmp2], %[tmp2], %[tmp4] \n\t" // tmp2 = tmp2 - tmp4; + "sub.s %[temp1], %[temp], %[tmp5] \n\t" + "add.s %[temp], %[temp], %[tmp5] \n\t" + "sub.s %[pom1], %[pom], %[tmp6] \n\t" + "add.s %[pom], %[pom], %[tmp6] \n\t" + "lwc1 %[temp3], 0(%[tmpz_n4_i]) \n\t" + "lwc1 %[pom2], 4(%[tmpz_n4_i]) \n\t" + "swc1 %[temp1], 0(%[tmpz_n2_i]) \n\t" // tmpz[ n2+i].re = tmpz[ i].re - tmp5; + "swc1 %[temp], 0(%[tmpz_i]) \n\t" // tmpz[ i].re = tmpz[ i].re + tmp5; + "swc1 %[pom1], 4(%[tmpz_n2_i]) \n\t" // tmpz[ n2+i].im = tmpz[ i].im - tmp6; + "swc1 %[pom] , 4(%[tmpz_i]) \n\t" // tmpz[ i].im = tmpz[ i].im + tmp6; + "sub.s %[temp4], %[temp3], %[tmp2] \n\t" + "add.s %[pom3], %[pom2], %[tmp1] \n\t" + "add.s %[temp3], %[temp3], %[tmp2] \n\t" + "sub.s %[pom2], %[pom2], %[tmp1] \n\t" + "swc1 %[temp4], 0(%[tmpz_n34_i]) \n\t" // tmpz[n34+i].re = tmpz[n4+i].re - tmp2; + "swc1 %[pom3], 4(%[tmpz_n34_i]) \n\t" // tmpz[n34+i].im = tmpz[n4+i].im + tmp1; + "swc1 %[temp3], 0(%[tmpz_n4_i]) \n\t" // tmpz[ n4+i].re = tmpz[n4+i].re + tmp2; + "swc1 %[pom2], 4(%[tmpz_n4_i]) \n\t" // tmpz[ n4+i].im = tmpz[n4+i].im - tmp1; + : [tmp1]"=&f"(tmp1), [tmp2]"=&f" (tmp2), [temp]"=&f"(temp), [tmp3]"=&f"(tmp3), + [tmp4]"=&f"(tmp4), [tmp5]"=&f"(tmp5), [tmp6]"=&f"(tmp6), + [temp1]"=&f"(temp1), [temp3]"=&f"(temp3), [temp4]"=&f"(temp4), + [pom]"=&f"(pom), [pom1]"=&f"(pom1), [pom2]"=&f"(pom2), [pom3]"=&f"(pom3) + : [w_re]"f"(w_re), [w_im]"f"(w_im), + [tmpz_i]"r"(tmpz_i),[tmpz_n2_i]"r"(tmpz_n2_i), + [tmpz_n34_i]"r"(tmpz_n34_i), [tmpz_n4_i]"r"(tmpz_n4_i) + : "memory" + ); + w_re_ptr += step; + w_im_ptr -= step; + } + } + step >>= 1; + n4 <<= 1; + } +} + +/** + * MDCT/IMDCT transforms. + */ + +static void ff_imdct_half_mips(FFTContext *s, FFTSample *output, const FFTSample *input) +{ + int k, n8, n4, n2, n, j; + const uint16_t *revtab = s->revtab; + const FFTSample *tcos = s->tcos; + const FFTSample *tsin = s->tsin; + const FFTSample *in1, *in2, *in3, *in4; + FFTComplex *z = (FFTComplex *)output; + + int j1; + const float *tcos1, *tsin1, *tcos2, *tsin2; + float temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8, + temp9, temp10, temp11, temp12, temp13, temp14, temp15, temp16; + FFTComplex *z1, *z2; + + n = 1 << s->mdct_bits; + n2 = n >> 1; + n4 = n >> 2; + n8 = n >> 3; + + /* pre rotation */ + in1 = input; + in2 = input + n2 - 1; + in3 = input + 2; + in4 = input + n2 - 3; + + tcos1 = tcos; + tsin1 = tsin; + + /* n4 = 64 or 128 */ + for(k = 0; k < n4; k += 2) { + j = revtab[k ]; + j1 = revtab[k + 1]; + + __asm__ volatile ( + "lwc1 %[temp1], 0(%[in2]) \t\n" + "lwc1 %[temp2], 0(%[tcos1]) \t\n" + "lwc1 %[temp3], 0(%[tsin1]) \t\n" + "lwc1 %[temp4], 0(%[in1]) \t\n" + "lwc1 %[temp5], 0(%[in4]) \t\n" + "mul.s %[temp9], %[temp1], %[temp2] \t\n" + "mul.s %[temp10], %[temp1], %[temp3] \t\n" + "lwc1 %[temp6], 4(%[tcos1]) \t\n" + "lwc1 %[temp7], 4(%[tsin1]) \t\n" + "nmsub.s %[temp9], %[temp9], %[temp4], %[temp3] \t\n" + "madd.s %[temp10], %[temp10], %[temp4], %[temp2] \t\n" + "mul.s %[temp11], %[temp5], %[temp6] \t\n" + "mul.s %[temp12], %[temp5], %[temp7] \t\n" + "lwc1 %[temp8], 0(%[in3]) \t\n" + "addiu %[tcos1], %[tcos1], 8 \t\n" + "addiu %[tsin1], %[tsin1], 8 \t\n" + "addiu %[in1], %[in1], 16 \t\n" + "nmsub.s %[temp11], %[temp11], %[temp8], %[temp7] \t\n" + "madd.s %[temp12], %[temp12], %[temp8], %[temp6] \t\n" + "addiu %[in2], %[in2], -16 \t\n" + "addiu %[in3], %[in3], 16 \t\n" + "addiu %[in4], %[in4], -16 \t\n" + + : [temp1]"=&f"(temp1), [temp2]"=&f"(temp2), + [temp3]"=&f"(temp3), [temp4]"=&f"(temp4), + [temp5]"=&f"(temp5), [temp6]"=&f"(temp6), + [temp7]"=&f"(temp7), [temp8]"=&f"(temp8), + [temp9]"=&f"(temp9), [temp10]"=&f"(temp10), + [temp11]"=&f"(temp11), [temp12]"=&f"(temp12), + [tsin1]"+r"(tsin1), [tcos1]"+r"(tcos1), + [in1]"+r"(in1), [in2]"+r"(in2), + [in3]"+r"(in3), [in4]"+r"(in4) + ); + + z[j ].re = temp9; + z[j ].im = temp10; + z[j1].re = temp11; + z[j1].im = temp12; + } + + s->fft_calc(s, z); + + /* post rotation + reordering */ + /* n8 = 32 or 64 */ + for(k = 0; k < n8; k += 2) { + tcos1 = &tcos[n8 - k - 2]; + tsin1 = &tsin[n8 - k - 2]; + tcos2 = &tcos[n8 + k]; + tsin2 = &tsin[n8 + k]; + z1 = &z[n8 - k - 2]; + z2 = &z[n8 + k ]; + + __asm__ volatile ( + "lwc1 %[temp1], 12(%[z1]) \t\n" + "lwc1 %[temp2], 4(%[tsin1]) \t\n" + "lwc1 %[temp3], 4(%[tcos1]) \t\n" + "lwc1 %[temp4], 8(%[z1]) \t\n" + "lwc1 %[temp5], 4(%[z1]) \t\n" + "mul.s %[temp9], %[temp1], %[temp2] \t\n" + "mul.s %[temp10], %[temp1], %[temp3] \t\n" + "lwc1 %[temp6], 0(%[tsin1]) \t\n" + "lwc1 %[temp7], 0(%[tcos1]) \t\n" + "nmsub.s %[temp9], %[temp9], %[temp4], %[temp3] \t\n" + "madd.s %[temp10], %[temp10], %[temp4], %[temp2] \t\n" + "mul.s %[temp11], %[temp5], %[temp6] \t\n" + "mul.s %[temp12], %[temp5], %[temp7] \t\n" + "lwc1 %[temp8], 0(%[z1]) \t\n" + "lwc1 %[temp1], 4(%[z2]) \t\n" + "lwc1 %[temp2], 0(%[tsin2]) \t\n" + "lwc1 %[temp3], 0(%[tcos2]) \t\n" + "nmsub.s %[temp11], %[temp11], %[temp8], %[temp7] \t\n" + "madd.s %[temp12], %[temp12], %[temp8], %[temp6] \t\n" + "mul.s %[temp13], %[temp1], %[temp2] \t\n" + "mul.s %[temp14], %[temp1], %[temp3] \t\n" + "lwc1 %[temp4], 0(%[z2]) \t\n" + "lwc1 %[temp5], 12(%[z2]) \t\n" + "lwc1 %[temp6], 4(%[tsin2]) \t\n" + "lwc1 %[temp7], 4(%[tcos2]) \t\n" + "nmsub.s %[temp13], %[temp13], %[temp4], %[temp3] \t\n" + "madd.s %[temp14], %[temp14], %[temp4], %[temp2] \t\n" + "mul.s %[temp15], %[temp5], %[temp6] \t\n" + "mul.s %[temp16], %[temp5], %[temp7] \t\n" + "lwc1 %[temp8], 8(%[z2]) \t\n" + "nmsub.s %[temp15], %[temp15], %[temp8], %[temp7] \t\n" + "madd.s %[temp16], %[temp16], %[temp8], %[temp6] \t\n" + : [temp1]"=&f"(temp1), [temp2]"=&f"(temp2), + [temp3]"=&f"(temp3), [temp4]"=&f"(temp4), + [temp5]"=&f"(temp5), [temp6]"=&f"(temp6), + [temp7]"=&f"(temp7), [temp8]"=&f"(temp8), + [temp9]"=&f"(temp9), [temp10]"=&f"(temp10), + [temp11]"=&f"(temp11), [temp12]"=&f"(temp12), + [temp13]"=&f"(temp13), [temp14]"=&f"(temp14), + [temp15]"=&f"(temp15), [temp16]"=&f"(temp16) + : [z1]"r"(z1), [z2]"r"(z2), + [tsin1]"r"(tsin1), [tcos1]"r"(tcos1), + [tsin2]"r"(tsin2), [tcos2]"r"(tcos2) + ); + + z1[1].re = temp9; + z1[1].im = temp14; + z2[0].re = temp13; + z2[0].im = temp10; + + z1[0].re = temp11; + z1[0].im = temp16; + z2[1].re = temp15; + z2[1].im = temp12; + } +} +#endif /* HAVE_INLINE_ASM */ + +/** + * Compute inverse MDCT of size N = 2^nbits + * @param output N samples + * @param input N/2 samples + */ +static void ff_imdct_calc_mips(FFTContext *s, FFTSample *output, const FFTSample *input) +{ + int k; + int n = 1 << s->mdct_bits; + int n2 = n >> 1; + int n4 = n >> 2; + + ff_imdct_half_mips(s, output+n4, input); + + for(k = 0; k < n4; k+=4) { + output[k] = -output[n2-k-1]; + output[k+1] = -output[n2-k-2]; + output[k+2] = -output[n2-k-3]; + output[k+3] = -output[n2-k-4]; + + output[n-k-1] = output[n2+k]; + output[n-k-2] = output[n2+k+1]; + output[n-k-3] = output[n2+k+2]; + output[n-k-4] = output[n2+k+3]; + } +} + +av_cold void ff_fft_init_mips(FFTContext *s) +{ + int n=0; + + ff_fft_lut_init(fft_offsets_lut, 0, 1 << 16, &n); + +#if HAVE_INLINE_ASM + s->fft_calc = ff_fft_calc_mips; +#endif +#if CONFIG_MDCT + s->imdct_calc = ff_imdct_calc_mips; + s->imdct_half = ff_imdct_half_mips; +#endif +} diff --git a/libavcodec/mips/fft_table.h b/libavcodec/mips/fft_table.h new file mode 100644 index 0000000000..dd52eaf8c8 --- /dev/null +++ b/libavcodec/mips/fft_table.h @@ -0,0 +1,63 @@ +/* + * Copyright (c) 2012 + * MIPS Technologies, Inc., California. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its + * contributors may be used to endorse or promote products derived from + * this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * Author: Stanislav Ocovaj (socovaj@mips.com) + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +/** + * @file + * definitions and LUT table for MIPS FFT + */ +#ifndef AVCODEC_MIPS_FFT_TABLE_H +#define AVCODEC_MIPS_FFT_TABLE_H + +#include "libavcodec/fft.h" + +#define MAX_LOG2_NFFT 16 //!< Specifies maxiumum allowed fft size +#define MAX_FFT_SIZE (1 << MAX_LOG2_NFFT) + +extern uint16_t fft_offsets_lut[]; +void ff_fft_lut_init(uint16_t *table, int off, int size, int *index); + +#endif /* AVCODEC_MIPS_FFT_TABLE_H */ diff --git a/libavcodec/mips/fmtconvert_mips.c b/libavcodec/mips/fmtconvert_mips.c new file mode 100644 index 0000000000..f89d3b6d64 --- /dev/null +++ b/libavcodec/mips/fmtconvert_mips.c @@ -0,0 +1,338 @@ +/* + * Format Conversion Utils for MIPS + * + * Copyright (c) 2012 + * MIPS Technologies, Inc., California. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. Neither the name of the MIPS Technologies, Inc., nor the names of is + * contributors may be used to endorse or promote products derived from + * this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * Author: Zoran Lukic (zoranl@mips.com) + * Author: Nedeljko Babic (nbabic@mips.com) + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ +#include "config.h" +#include "libavcodec/avcodec.h" +#include "libavcodec/fmtconvert.h" + +#if HAVE_MIPSDSPR1 +static void float_to_int16_mips(int16_t *dst, const float *src, long len) +{ + const float *src_end = src + len; + int ret0, ret1, ret2, ret3, ret4, ret5, ret6, ret7; + float src0, src1, src2, src3, src4, src5, src6, src7; + + /* + * loop is 8 times unrolled in assembler in order to achieve better performance + */ + __asm__ volatile( + "beq %[len], $zero, fti16_end%= \n\t" + "fti16_lp%=: \n\t" + "lwc1 %[src0], 0(%[src]) \n\t" + "lwc1 %[src1], 4(%[src]) \n\t" + "lwc1 %[src2], 8(%[src]) \n\t" + "lwc1 %[src3], 12(%[src]) \n\t" + "cvt.w.s %[src0], %[src0] \n\t" + "cvt.w.s %[src1], %[src1] \n\t" + "cvt.w.s %[src2], %[src2] \n\t" + "cvt.w.s %[src3], %[src3] \n\t" + "mfc1 %[ret0], %[src0] \n\t" + "mfc1 %[ret1], %[src1] \n\t" + "mfc1 %[ret2], %[src2] \n\t" + "mfc1 %[ret3], %[src3] \n\t" + "lwc1 %[src4], 16(%[src]) \n\t" + "lwc1 %[src5], 20(%[src]) \n\t" + "lwc1 %[src6], 24(%[src]) \n\t" + "lwc1 %[src7], 28(%[src]) \n\t" + "cvt.w.s %[src4], %[src4] \n\t" + "cvt.w.s %[src5], %[src5] \n\t" + "cvt.w.s %[src6], %[src6] \n\t" + "cvt.w.s %[src7], %[src7] \n\t" + "addiu %[src], 32 \n\t" + "shll_s.w %[ret0], %[ret0], 16 \n\t" + "shll_s.w %[ret1], %[ret1], 16 \n\t" + "shll_s.w %[ret2], %[ret2], 16 \n\t" + "shll_s.w %[ret3], %[ret3], 16 \n\t" + "srl %[ret0], %[ret0], 16 \n\t" + "srl %[ret1], %[ret1], 16 \n\t" + "srl %[ret2], %[ret2], 16 \n\t" + "srl %[ret3], %[ret3], 16 \n\t" + "sh %[ret0], 0(%[dst]) \n\t" + "sh %[ret1], 2(%[dst]) \n\t" + "sh %[ret2], 4(%[dst]) \n\t" + "sh %[ret3], 6(%[dst]) \n\t" + "mfc1 %[ret4], %[src4] \n\t" + "mfc1 %[ret5], %[src5] \n\t" + "mfc1 %[ret6], %[src6] \n\t" + "mfc1 %[ret7], %[src7] \n\t" + "shll_s.w %[ret4], %[ret4], 16 \n\t" + "shll_s.w %[ret5], %[ret5], 16 \n\t" + "shll_s.w %[ret6], %[ret6], 16 \n\t" + "shll_s.w %[ret7], %[ret7], 16 \n\t" + "srl %[ret4], %[ret4], 16 \n\t" + "srl %[ret5], %[ret5], 16 \n\t" + "srl %[ret6], %[ret6], 16 \n\t" + "srl %[ret7], %[ret7], 16 \n\t" + "sh %[ret4], 8(%[dst]) \n\t" + "sh %[ret5], 10(%[dst]) \n\t" + "sh %[ret6], 12(%[dst]) \n\t" + "sh %[ret7], 14(%[dst]) \n\t" + "addiu %[dst], 16 \n\t" + "bne %[src], %[src_end], fti16_lp%= \n\t" + "fti16_end%=: \n\t" + : [ret0]"=&r"(ret0), [ret1]"=&r"(ret1), [ret2]"=&r"(ret2), [ret3]"=&r"(ret3), + [ret4]"=&r"(ret4), [ret5]"=&r"(ret5), [ret6]"=&r"(ret6), [ret7]"=&r"(ret7), + [src0]"=&f"(src0), [src1]"=&f"(src1), [src2]"=&f"(src2), [src3]"=&f"(src3), + [src4]"=&f"(src4), [src5]"=&f"(src5), [src6]"=&f"(src6), [src7]"=&f"(src7), + [src]"+r"(src), [dst]"+r"(dst) + : [src_end]"r"(src_end), [len]"r"(len) + : "memory" + ); +} + +static void float_to_int16_interleave_mips(int16_t *dst, const float **src, long len, + int channels) +{ + int c, ch2 = channels <<1; + int ret0, ret1, ret2, ret3, ret4, ret5, ret6, ret7; + float src0, src1, src2, src3, src4, src5, src6, src7; + int16_t *dst_ptr0, *dst_ptr1, *dst_ptr2, *dst_ptr3; + int16_t *dst_ptr4, *dst_ptr5, *dst_ptr6, *dst_ptr7; + const float *src_ptr, *src_ptr2, *src_end; + + if (channels == 2) { + src_ptr = &src[0][0]; + src_ptr2 = &src[1][0]; + src_end = src_ptr + len; + + __asm__ volatile ( + "fti16i2_lp%=: \n\t" + "lwc1 %[src0], 0(%[src_ptr]) \n\t" + "lwc1 %[src1], 0(%[src_ptr2]) \n\t" + "addiu %[src_ptr], 4 \n\t" + "cvt.w.s $f9, %[src0] \n\t" + "cvt.w.s $f10, %[src1] \n\t" + "mfc1 %[ret0], $f9 \n\t" + "mfc1 %[ret1], $f10 \n\t" + "shll_s.w %[ret0], %[ret0], 16 \n\t" + "shll_s.w %[ret1], %[ret1], 16 \n\t" + "addiu %[src_ptr2], 4 \n\t" + "srl %[ret0], %[ret0], 16 \n\t" + "srl %[ret1], %[ret1], 16 \n\t" + "sh %[ret0], 0(%[dst]) \n\t" + "sh %[ret1], 2(%[dst]) \n\t" + "addiu %[dst], 4 \n\t" + "bne %[src_ptr], %[src_end], fti16i2_lp%= \n\t" + : [ret0]"=&r"(ret0), [ret1]"=&r"(ret1), + [src0]"=&f"(src0), [src1]"=&f"(src1), + [src_ptr]"+r"(src_ptr), [src_ptr2]"+r"(src_ptr2), + [dst]"+r"(dst) + : [src_end]"r"(src_end) + : "memory" + ); + } else { + for (c = 0; c < channels; c++) { + src_ptr = &src[c][0]; + dst_ptr0 = &dst[c]; + src_end = src_ptr + len; + /* + * loop is 8 times unrolled in assembler in order to achieve better performance + */ + __asm__ volatile( + "fti16i_lp%=: \n\t" + "lwc1 %[src0], 0(%[src_ptr]) \n\t" + "lwc1 %[src1], 4(%[src_ptr]) \n\t" + "lwc1 %[src2], 8(%[src_ptr]) \n\t" + "lwc1 %[src3], 12(%[src_ptr]) \n\t" + "cvt.w.s %[src0], %[src0] \n\t" + "cvt.w.s %[src1], %[src1] \n\t" + "cvt.w.s %[src2], %[src2] \n\t" + "cvt.w.s %[src3], %[src3] \n\t" + "mfc1 %[ret0], %[src0] \n\t" + "mfc1 %[ret1], %[src1] \n\t" + "mfc1 %[ret2], %[src2] \n\t" + "mfc1 %[ret3], %[src3] \n\t" + "lwc1 %[src4], 16(%[src_ptr]) \n\t" + "lwc1 %[src5], 20(%[src_ptr]) \n\t" + "lwc1 %[src6], 24(%[src_ptr]) \n\t" + "lwc1 %[src7], 28(%[src_ptr]) \n\t" + "addu %[dst_ptr1], %[dst_ptr0], %[ch2] \n\t" + "addu %[dst_ptr2], %[dst_ptr1], %[ch2] \n\t" + "addu %[dst_ptr3], %[dst_ptr2], %[ch2] \n\t" + "addu %[dst_ptr4], %[dst_ptr3], %[ch2] \n\t" + "addu %[dst_ptr5], %[dst_ptr4], %[ch2] \n\t" + "addu %[dst_ptr6], %[dst_ptr5], %[ch2] \n\t" + "addu %[dst_ptr7], %[dst_ptr6], %[ch2] \n\t" + "addiu %[src_ptr], 32 \n\t" + "cvt.w.s %[src4], %[src4] \n\t" + "cvt.w.s %[src5], %[src5] \n\t" + "cvt.w.s %[src6], %[src6] \n\t" + "cvt.w.s %[src7], %[src7] \n\t" + "shll_s.w %[ret0], %[ret0], 16 \n\t" + "shll_s.w %[ret1], %[ret1], 16 \n\t" + "shll_s.w %[ret2], %[ret2], 16 \n\t" + "shll_s.w %[ret3], %[ret3], 16 \n\t" + "srl %[ret0], %[ret0], 16 \n\t" + "srl %[ret1], %[ret1], 16 \n\t" + "srl %[ret2], %[ret2], 16 \n\t" + "srl %[ret3], %[ret3], 16 \n\t" + "sh %[ret0], 0(%[dst_ptr0]) \n\t" + "sh %[ret1], 0(%[dst_ptr1]) \n\t" + "sh %[ret2], 0(%[dst_ptr2]) \n\t" + "sh %[ret3], 0(%[dst_ptr3]) \n\t" + "mfc1 %[ret4], %[src4] \n\t" + "mfc1 %[ret5], %[src5] \n\t" + "mfc1 %[ret6], %[src6] \n\t" + "mfc1 %[ret7], %[src7] \n\t" + "shll_s.w %[ret4], %[ret4], 16 \n\t" + "shll_s.w %[ret5], %[ret5], 16 \n\t" + "shll_s.w %[ret6], %[ret6], 16 \n\t" + "shll_s.w %[ret7], %[ret7], 16 \n\t" + "srl %[ret4], %[ret4], 16 \n\t" + "srl %[ret5], %[ret5], 16 \n\t" + "srl %[ret6], %[ret6], 16 \n\t" + "srl %[ret7], %[ret7], 16 \n\t" + "sh %[ret4], 0(%[dst_ptr4]) \n\t" + "sh %[ret5], 0(%[dst_ptr5]) \n\t" + "sh %[ret6], 0(%[dst_ptr6]) \n\t" + "sh %[ret7], 0(%[dst_ptr7]) \n\t" + "addu %[dst_ptr0], %[dst_ptr7], %[ch2] \n\t" + "bne %[src_ptr], %[src_end], fti16i_lp%= \n\t" + : [ret0]"=&r"(ret0), [ret1]"=&r"(ret1), [ret2]"=&r"(ret2), [ret3]"=&r"(ret3), + [ret4]"=&r"(ret4), [ret5]"=&r"(ret5), [ret6]"=&r"(ret6), [ret7]"=&r"(ret7), + [src0]"=&f"(src0), [src1]"=&f"(src1), [src2]"=&f"(src2), [src3]"=&f"(src3), + [src4]"=&f"(src4), [src5]"=&f"(src5), [src6]"=&f"(src6), [src7]"=&f"(src7), + [dst_ptr1]"=&r"(dst_ptr1), [dst_ptr2]"=&r"(dst_ptr2), [dst_ptr3]"=&r"(dst_ptr3), + [dst_ptr4]"=&r"(dst_ptr4), [dst_ptr5]"=&r"(dst_ptr5), [dst_ptr6]"=&r"(dst_ptr6), + [dst_ptr7]"=&r"(dst_ptr7), [dst_ptr0]"+r"(dst_ptr0), [src_ptr]"+r"(src_ptr) + : [ch2]"r"(ch2), [src_end]"r"(src_end) + : "memory" + ); + } + } +} +#endif /* HAVE_MIPSDSPR1 */ + +static void int32_to_float_fmul_scalar_mips(float *dst, const int *src, + float mul, int len) +{ + /* + * variables used in inline assembler + */ + float temp1, temp3, temp5, temp7, temp9, temp11, temp13, temp15; + + int rpom1, rpom2, rpom11, rpom21, rpom12, rpom22, rpom13, rpom23; + const int *src_end = src + len; + /* + * loop is 8 times unrolled in assembler in order to achieve better performance + */ + __asm__ volatile ( + "i32tf_lp%=: \n\t" + "lw %[rpom11], 0(%[src]) \n\t" + "lw %[rpom21], 4(%[src]) \n\t" + "lw %[rpom1], 8(%[src]) \n\t" + "lw %[rpom2], 12(%[src]) \n\t" + "mtc1 %[rpom11], %[temp1] \n\t" + "mtc1 %[rpom21], %[temp3] \n\t" + "mtc1 %[rpom1], %[temp5] \n\t" + "mtc1 %[rpom2], %[temp7] \n\t" + + "lw %[rpom13], 16(%[src]) \n\t" + "lw %[rpom23], 20(%[src]) \n\t" + "lw %[rpom12], 24(%[src]) \n\t" + "lw %[rpom22], 28(%[src]) \n\t" + "mtc1 %[rpom13], %[temp9] \n\t" + "mtc1 %[rpom23], %[temp11] \n\t" + "mtc1 %[rpom12], %[temp13] \n\t" + "mtc1 %[rpom22], %[temp15] \n\t" + + "addiu %[src], 32 \n\t" + "cvt.s.w %[temp1], %[temp1] \n\t" + "cvt.s.w %[temp3], %[temp3] \n\t" + "cvt.s.w %[temp5], %[temp5] \n\t" + "cvt.s.w %[temp7], %[temp7] \n\t" + + "cvt.s.w %[temp9], %[temp9] \n\t" + "cvt.s.w %[temp11], %[temp11] \n\t" + "cvt.s.w %[temp13], %[temp13] \n\t" + "cvt.s.w %[temp15], %[temp15] \n\t" + + "mul.s %[temp1], %[temp1], %[mul] \n\t" + "mul.s %[temp3], %[temp3], %[mul] \n\t" + "mul.s %[temp5], %[temp5], %[mul] \n\t" + "mul.s %[temp7], %[temp7], %[mul] \n\t" + + "mul.s %[temp9], %[temp9], %[mul] \n\t" + "mul.s %[temp11], %[temp11], %[mul] \n\t" + "mul.s %[temp13], %[temp13], %[mul] \n\t" + "mul.s %[temp15], %[temp15], %[mul] \n\t" + + "swc1 %[temp1], 0(%[dst]) \n\t" /*dst[i] = src[i] * mul; */ + "swc1 %[temp3], 4(%[dst]) \n\t" /*dst[i+1] = src[i+1] * mul;*/ + "swc1 %[temp5], 8(%[dst]) \n\t" /*dst[i+2] = src[i+2] * mul;*/ + "swc1 %[temp7], 12(%[dst]) \n\t" /*dst[i+3] = src[i+3] * mul;*/ + + "swc1 %[temp9], 16(%[dst]) \n\t" /*dst[i+4] = src[i+4] * mul;*/ + "swc1 %[temp11], 20(%[dst]) \n\t" /*dst[i+5] = src[i+5] * mul;*/ + "swc1 %[temp13], 24(%[dst]) \n\t" /*dst[i+6] = src[i+6] * mul;*/ + "swc1 %[temp15], 28(%[dst]) \n\t" /*dst[i+7] = src[i+7] * mul;*/ + "addiu %[dst], 32 \n\t" + "bne %[src], %[src_end], i32tf_lp%= \n\t" + : [temp1]"=&f"(temp1), [temp11]"=&f"(temp11), + [temp13]"=&f"(temp13), [temp15]"=&f"(temp15), + [temp3]"=&f"(temp3), [temp5]"=&f"(temp5), + [temp7]"=&f"(temp7), [temp9]"=&f"(temp9), + [rpom1]"=&r"(rpom1), [rpom2]"=&r"(rpom2), + [rpom11]"=&r"(rpom11), [rpom21]"=&r"(rpom21), + [rpom12]"=&r"(rpom12), [rpom22]"=&r"(rpom22), + [rpom13]"=&r"(rpom13), [rpom23]"=&r"(rpom23), + [dst]"+r"(dst), [src]"+r"(src) + : [mul]"f"(mul), [src_end]"r"(src_end) + : "memory" + ); +} + +av_cold void ff_fmt_convert_init_mips(FmtConvertContext *c) +{ +#if HAVE_MIPSDSPR1 + c->float_to_int16_interleave = float_to_int16_interleave_mips; + c->float_to_int16 = float_to_int16_mips; +#endif + c->int32_to_float_fmul_scalar = int32_to_float_fmul_scalar_mips; +} diff --git a/libavcodec/mips/lsp_mips.h b/libavcodec/mips/lsp_mips.h new file mode 100644 index 0000000000..749770625c --- /dev/null +++ b/libavcodec/mips/lsp_mips.h @@ -0,0 +1,108 @@ +/* + * Copyright (c) 2012 + * MIPS Technologies, Inc., California. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its + * contributors may be used to endorse or promote products derived from + * this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * Author: Nedeljko Babic (nbabic@mips.com) + * + * LSP routines for ACELP-based codecs optimized for MIPS + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +/** + * @file + * Reference: libavcodec/lsp.c + */ +#ifndef AVCODEC_LSP_MIPS_H +#define AVCODEC_LSP_MIPS_H + +#if HAVE_MIPSFPU && HAVE_INLINE_ASM +static av_always_inline void ff_lsp2polyf_mips(const double *lsp, double *f, int lp_half_order) +{ + int i, j = 0; + double * p_fi = f; + double * p_f = 0; + + f[0] = 1.0; + f[1] = -2 * lsp[0]; + lsp -= 2; + + for(i=2; i<=lp_half_order; i++) + { + double tmp, f_j_2, f_j_1, f_j; + double val = lsp[2*i]; + + __asm__ volatile( + "move %[p_f], %[p_fi] \n\t" + "add.d %[val], %[val], %[val] \n\t" + "addiu %[p_fi], 8 \n\t" + "ldc1 %[f_j_1], 0(%[p_f]) \n\t" + "ldc1 %[f_j], 8(%[p_f]) \n\t" + "neg.d %[val], %[val] \n\t" + "add.d %[tmp], %[f_j_1], %[f_j_1] \n\t" + "madd.d %[tmp], %[tmp], %[f_j], %[val] \n\t" + "addiu %[j], %[i], -2 \n\t" + "ldc1 %[f_j_2], -8(%[p_f]) \n\t" + "sdc1 %[tmp], 16(%[p_f]) \n\t" + "beqz %[j], ff_lsp2polyf_lp_j_end%= \n\t" + "ff_lsp2polyf_lp_j%=: \n\t" + "add.d %[tmp], %[f_j], %[f_j_2] \n\t" + "madd.d %[tmp], %[tmp], %[f_j_1], %[val] \n\t" + "mov.d %[f_j], %[f_j_1] \n\t" + "addiu %[j], -1 \n\t" + "mov.d %[f_j_1], %[f_j_2] \n\t" + "ldc1 %[f_j_2], -16(%[p_f]) \n\t" + "sdc1 %[tmp], 8(%[p_f]) \n\t" + "addiu %[p_f], -8 \n\t" + "bgtz %[j], ff_lsp2polyf_lp_j%= \n\t" + "ff_lsp2polyf_lp_j_end%=: \n\t" + + : [f_j_2]"=&f"(f_j_2), [f_j_1]"=&f"(f_j_1), [val]"+f"(val), + [tmp]"=&f"(tmp), [f_j]"=&f"(f_j), [p_f]"+r"(p_f), + [j]"+r"(j), [p_fi]"+r"(p_fi) + : [i]"r"(i) + ); + f[1] += val; + } +} +#define ff_lsp2polyf ff_lsp2polyf_mips +#endif /* HAVE_MIPSFPU && HAVE_INLINE_ASM */ +#endif /* AVCODEC_LSP_MIPS_H */ diff --git a/libavcodec/mips/mathops.h b/libavcodec/mips/mathops.h index dd80f68072..368290ac5b 100644 --- a/libavcodec/mips/mathops.h +++ b/libavcodec/mips/mathops.h @@ -1,20 +1,20 @@ /* * Copyright (c) 2009 Mans Rullgard <mans@mansr.com> * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * FFmpeg is free software; you can redistribute it and/or * modify it under the terms of the GNU Lesser General Public * License as published by the Free Software Foundation; either * version 2.1 of the License, or (at your option) any later version. * - * Libav is distributed in the hope that it will be useful, + * FFmpeg is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * Lesser General Public License for more details. * * You should have received a copy of the GNU Lesser General Public - * License along with Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ diff --git a/libavcodec/mips/mpegaudiodsp_mips_fixed.c b/libavcodec/mips/mpegaudiodsp_mips_fixed.c new file mode 100644 index 0000000000..d1b02775d4 --- /dev/null +++ b/libavcodec/mips/mpegaudiodsp_mips_fixed.c @@ -0,0 +1,903 @@ +/* + * Copyright (c) 2012 + * MIPS Technologies, Inc., California. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its + * contributors may be used to endorse or promote products derived from + * this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * Author: Bojan Zivkovic (bojan@mips.com) + * + * MPEG Audio decoder optimized for MIPS fixed-point architecture + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +/** + * @file + * Reference: libavcodec/mpegaudiodsp_template.c + */ + +#include <string.h> + +#include "libavcodec/mpegaudiodsp.h" + +static void ff_mpadsp_apply_window_mips_fixed(int32_t *synth_buf, int32_t *window, + int *dither_state, int16_t *samples, int incr) +{ + register const int32_t *w, *w2, *p; + int j; + int16_t *samples2; + int w_asm, p_asm, w_asm1, p_asm1, w_asm2, p_asm2; + int w2_asm, w2_asm1, *p_temp1, *p_temp2; + int sum1 = 0; + int const min_asm = -32768, max_asm = 32767; + int temp1, temp2 = 0, temp3 = 0; + int64_t sum; + + /* copy to avoid wrap */ + memcpy(synth_buf + 512, synth_buf, 32 * sizeof(*synth_buf)); + samples2 = samples + 31 * incr; + w = window; + w2 = window + 31; + sum = *dither_state; + p = synth_buf + 16; + p_temp1 = synth_buf + 16; + p_temp2 = synth_buf + 48; + temp1 = sum; + + /** + * use of round_sample function from the original code is eliminated, + * changed with appropriate assembly instructions. + */ + __asm__ volatile ( + "mthi $zero \n\t" + "mtlo %[temp1] \n\t" + "lw %[w_asm], 0(%[w]) \n\t" + "lw %[p_asm], 0(%[p]) \n\t" + "lw %[w_asm1], 64*4(%[w]) \n\t" + "lw %[p_asm1], 64*4(%[p]) \n\t" + "lw %[w_asm2], 128*4(%[w]) \n\t" + "lw %[p_asm2], 128*4(%[p]) \n\t" + "madd %[w_asm], %[p_asm] \n\t" + "madd %[w_asm1], %[p_asm1] \n\t" + "madd %[w_asm2], %[p_asm2] \n\t" + "lw %[w_asm], 192*4(%[w]) \n\t" + "lw %[p_asm], 192*4(%[p]) \n\t" + "lw %[w_asm1], 256*4(%[w]) \n\t" + "lw %[p_asm1], 256*4(%[p]) \n\t" + "lw %[w_asm2], 320*4(%[w]) \n\t" + "lw %[p_asm2], 320*4(%[p]) \n\t" + "madd %[w_asm], %[p_asm] \n\t" + "madd %[w_asm1], %[p_asm1] \n\t" + "madd %[w_asm2], %[p_asm2] \n\t" + "lw %[w_asm], 384*4(%[w]) \n\t" + "lw %[p_asm], 384*4(%[p]) \n\t" + "lw %[w_asm1], 448*4(%[w]) \n\t" + "lw %[p_asm1], 448*4(%[p]) \n\t" + "lw %[w_asm2], 32*4(%[w]) \n\t" + "lw %[p_asm2], 32*4(%[p]) \n\t" + "madd %[w_asm], %[p_asm] \n\t" + "madd %[w_asm1], %[p_asm1] \n\t" + "msub %[w_asm2], %[p_asm2] \n\t" + "lw %[w_asm], 96*4(%[w]) \n\t" + "lw %[p_asm], 96*4(%[p]) \n\t" + "lw %[w_asm1], 160*4(%[w]) \n\t" + "lw %[p_asm1], 160*4(%[p]) \n\t" + "lw %[w_asm2], 224*4(%[w]) \n\t" + "lw %[p_asm2], 224*4(%[p]) \n\t" + "msub %[w_asm], %[p_asm] \n\t" + "msub %[w_asm1], %[p_asm1] \n\t" + "msub %[w_asm2], %[p_asm2] \n\t" + "lw %[w_asm], 288*4(%[w]) \n\t" + "lw %[p_asm], 288*4(%[p]) \n\t" + "lw %[w_asm1], 352*4(%[w]) \n\t" + "lw %[p_asm1], 352*4(%[p]) \n\t" + "msub %[w_asm], %[p_asm] \n\t" + "lw %[w_asm], 480*4(%[w]) \n\t" + "lw %[p_asm], 480*4(%[p]) \n\t" + "lw %[w_asm2], 416*4(%[w]) \n\t" + "lw %[p_asm2], 416*4(%[p]) \n\t" + "msub %[w_asm], %[p_asm] \n\t" + "msub %[w_asm1], %[p_asm1] \n\t" + "msub %[w_asm2], %[p_asm2] \n\t" + + /*round_sample function from the original code is eliminated, + * changed with appropriate assembly instructions + * code example: + + "extr.w %[sum1],$ac0,24 \n\t" + "mflo %[temp3], $ac0 \n\t" + "and %[temp1], %[temp3], 0x00ffffff \n\t" + "slt %[temp2], %[sum1], %[min_asm] \n\t" + "movn %[sum1], %[min_asm],%[temp2] \n\t" + "slt %[temp2], %[max_asm],%[sum1] \n\t" + "movn %[sum1], %[max_asm],%[temp2] \n\t" + "sh %[sum1], 0(%[samples]) \n\t" + */ + + "extr.w %[sum1], $ac0, 24 \n\t" + "mflo %[temp3] \n\t" + "addi %[w], %[w], 4 \n\t" + "and %[temp1], %[temp3], 0x00ffffff \n\t" + "slt %[temp2], %[sum1], %[min_asm] \n\t" + "movn %[sum1], %[min_asm], %[temp2] \n\t" + "slt %[temp2], %[max_asm], %[sum1] \n\t" + "movn %[sum1], %[max_asm], %[temp2] \n\t" + "sh %[sum1], 0(%[samples]) \n\t" + + : [w_asm] "=&r" (w_asm), [p_asm] "=&r" (p_asm), [w_asm1] "=&r" (w_asm1), + [p_asm1] "=&r" (p_asm1), [temp1] "+r" (temp1), [temp2] "+r" (temp2), + [w_asm2] "=&r" (w_asm2), [p_asm2] "=&r" (p_asm2), + [sum1] "+r" (sum1), [w] "+r" (w), [temp3] "+r" (temp3) + : [p] "r" (p), [samples] "r" (samples), [min_asm] "r" (min_asm), + [max_asm] "r" (max_asm) + : "hi","lo" + ); + + samples += incr; + + /* we calculate two samples at the same time to avoid one memory + access per two sample */ + + for(j = 1; j < 16; j++) { + __asm__ volatile ( + "mthi $0, $ac1 \n\t" + "mtlo $0, $ac1 \n\t" + "mthi $0 \n\t" + "mtlo %[temp1] \n\t" + "addi %[p_temp1], %[p_temp1], 4 \n\t" + "lw %[w_asm], 0(%[w]) \n\t" + "lw %[p_asm], 0(%[p_temp1]) \n\t" + "lw %[w2_asm], 0(%[w2]) \n\t" + "lw %[w_asm1], 64*4(%[w]) \n\t" + "lw %[p_asm1], 64*4(%[p_temp1]) \n\t" + "lw %[w2_asm1], 64*4(%[w2]) \n\t" + "madd %[w_asm], %[p_asm] \n\t" + "msub $ac1, %[w2_asm], %[p_asm] \n\t" + "madd %[w_asm1], %[p_asm1] \n\t" + "msub $ac1, %[w2_asm1], %[p_asm1] \n\t" + "lw %[w_asm], 128*4(%[w]) \n\t" + "lw %[p_asm], 128*4(%[p_temp1]) \n\t" + "lw %[w2_asm], 128*4(%[w2]) \n\t" + "lw %[w_asm1], 192*4(%[w]) \n\t" + "lw %[p_asm1], 192*4(%[p_temp1]) \n\t" + "lw %[w2_asm1], 192*4(%[w2]) \n\t" + "madd %[w_asm], %[p_asm] \n\t" + "msub $ac1, %[w2_asm], %[p_asm] \n\t" + "madd %[w_asm1], %[p_asm1] \n\t" + "msub $ac1, %[w2_asm1], %[p_asm1] \n\t" + "lw %[w_asm], 256*4(%[w]) \n\t" + "lw %[p_asm], 256*4(%[p_temp1]) \n\t" + "lw %[w2_asm], 256*4(%[w2]) \n\t" + "lw %[w_asm1], 320*4(%[w]) \n\t" + "lw %[p_asm1], 320*4(%[p_temp1]) \n\t" + "lw %[w2_asm1], 320*4(%[w2]) \n\t" + "madd %[w_asm], %[p_asm] \n\t" + "msub $ac1, %[w2_asm], %[p_asm] \n\t" + "madd %[w_asm1], %[p_asm1] \n\t" + "msub $ac1, %[w2_asm1], %[p_asm1] \n\t" + "lw %[w_asm], 384*4(%[w]) \n\t" + "lw %[p_asm], 384*4(%[p_temp1]) \n\t" + "lw %[w2_asm], 384*4(%[w2]) \n\t" + "lw %[w_asm1], 448*4(%[w]) \n\t" + "lw %[p_asm1], 448*4(%[p_temp1]) \n\t" + "lw %[w2_asm1], 448*4(%[w2]) \n\t" + "madd %[w_asm], %[p_asm] \n\t" + "msub $ac1, %[w2_asm], %[p_asm] \n\t" + "madd %[w_asm1], %[p_asm1] \n\t" + "msub $ac1, %[w2_asm1], %[p_asm1] \n\t" + "addi %[p_temp2], %[p_temp2], -4 \n\t" + "lw %[w_asm], 32*4(%[w]) \n\t" + "lw %[p_asm], 0(%[p_temp2]) \n\t" + "lw %[w2_asm], 32*4(%[w2]) \n\t" + "lw %[w_asm1], 96*4(%[w]) \n\t" + "lw %[p_asm1], 64*4(%[p_temp2]) \n\t" + "lw %[w2_asm1], 96*4(%[w2]) \n\t" + "msub %[w_asm], %[p_asm] \n\t" + "msub $ac1, %[w2_asm], %[p_asm] \n\t" + "msub %[w_asm1], %[p_asm1] \n\t" + "msub $ac1, %[w2_asm1], %[p_asm1] \n\t" + "lw %[w_asm], 160*4(%[w]) \n\t" + "lw %[p_asm], 128*4(%[p_temp2]) \n\t" + "lw %[w2_asm], 160*4(%[w2]) \n\t" + "lw %[w_asm1], 224*4(%[w]) \n\t" + "lw %[p_asm1], 192*4(%[p_temp2]) \n\t" + "lw %[w2_asm1], 224*4(%[w2]) \n\t" + "msub %[w_asm], %[p_asm] \n\t" + "msub $ac1, %[w2_asm], %[p_asm] \n\t" + "msub %[w_asm1], %[p_asm1] \n\t" + "msub $ac1, %[w2_asm1], %[p_asm1] \n\t" + "lw %[w_asm], 288*4(%[w]) \n\t" + "lw %[p_asm], 256*4(%[p_temp2]) \n\t" + "lw %[w2_asm], 288*4(%[w2]) \n\t" + "lw %[w_asm1], 352*4(%[w]) \n\t" + "lw %[p_asm1], 320*4(%[p_temp2]) \n\t" + "lw %[w2_asm1], 352*4(%[w2]) \n\t" + "msub %[w_asm], %[p_asm] \n\t" + "msub $ac1, %[w2_asm], %[p_asm] \n\t" + "msub %[w_asm1], %[p_asm1] \n\t" + "msub $ac1, %[w2_asm1], %[p_asm1] \n\t" + "lw %[w_asm], 416*4(%[w]) \n\t" + "lw %[p_asm], 384*4(%[p_temp2]) \n\t" + "lw %[w2_asm], 416*4(%[w2]) \n\t" + "lw %[w_asm1], 480*4(%[w]) \n\t" + "lw %[p_asm1], 448*4(%[p_temp2]) \n\t" + "lw %[w2_asm1], 480*4(%[w2]) \n\t" + "msub %[w_asm], %[p_asm] \n\t" + "msub %[w_asm1], %[p_asm1] \n\t" + "msub $ac1, %[w2_asm], %[p_asm] \n\t" + "msub $ac1, %[w2_asm1], %[p_asm1] \n\t" + "addi %[w], %[w], 4 \n\t" + "addi %[w2], %[w2], -4 \n\t" + "mflo %[temp2] \n\t" + "extr.w %[sum1], $ac0, 24 \n\t" + "li %[temp3], 1 \n\t" + "and %[temp1], %[temp2], 0x00ffffff \n\t" + "madd $ac1, %[temp1], %[temp3] \n\t" + "slt %[temp2], %[sum1], %[min_asm] \n\t" + "movn %[sum1], %[min_asm], %[temp2] \n\t" + "slt %[temp2], %[max_asm], %[sum1] \n\t" + "movn %[sum1], %[max_asm], %[temp2] \n\t" + "sh %[sum1], 0(%[samples]) \n\t" + "mflo %[temp3], $ac1 \n\t" + "extr.w %[sum1], $ac1, 24 \n\t" + "and %[temp1], %[temp3], 0x00ffffff \n\t" + "slt %[temp2], %[sum1], %[min_asm] \n\t" + "movn %[sum1], %[min_asm], %[temp2] \n\t" + "slt %[temp2], %[max_asm], %[sum1] \n\t" + "movn %[sum1], %[max_asm], %[temp2] \n\t" + "sh %[sum1], 0(%[samples2]) \n\t" + + : [w_asm] "=&r" (w_asm), [p_asm] "=&r" (p_asm), [w_asm1] "=&r" (w_asm1), + [p_asm1] "=&r" (p_asm1), [w2_asm1] "=&r" (w2_asm1), + [w2_asm] "=&r" (w2_asm), [temp1] "+r" (temp1), [temp2] "+r" (temp2), + [p_temp1] "+r" (p_temp1), [p_temp2] "+r" (p_temp2), [sum1] "+r" (sum1), + [w] "+r" (w), [w2] "+r" (w2), [samples] "+r" (samples), + [samples2] "+r" (samples2), [temp3] "+r" (temp3) + : [min_asm] "r" (min_asm), [max_asm] "r" (max_asm) + : "hi", "lo" + ); + + samples += incr; + samples2 -= incr; + } + + p = synth_buf + 32; + + __asm__ volatile ( + "mthi $0 \n\t" + "mtlo %[temp1] \n\t" + "lw %[w_asm], 32*4(%[w]) \n\t" + "lw %[p_asm], 0(%[p]) \n\t" + "lw %[w_asm1], 96*4(%[w]) \n\t" + "lw %[p_asm1], 64*4(%[p]) \n\t" + "lw %[w_asm2], 160*4(%[w]) \n\t" + "lw %[p_asm2], 128*4(%[p]) \n\t" + "msub %[w_asm], %[p_asm] \n\t" + "msub %[w_asm1], %[p_asm1] \n\t" + "msub %[w_asm2], %[p_asm2] \n\t" + "lw %[w_asm], 224*4(%[w]) \n\t" + "lw %[p_asm], 192*4(%[p]) \n\t" + "lw %[w_asm1], 288*4(%[w]) \n\t" + "lw %[p_asm1], 256*4(%[p]) \n\t" + "lw %[w_asm2], 352*4(%[w]) \n\t" + "lw %[p_asm2], 320*4(%[p]) \n\t" + "msub %[w_asm], %[p_asm] \n\t" + "msub %[w_asm1], %[p_asm1] \n\t" + "msub %[w_asm2], %[p_asm2] \n\t" + "lw %[w_asm], 416*4(%[w]) \n\t" + "lw %[p_asm], 384*4(%[p]) \n\t" + "lw %[w_asm1], 480*4(%[w]) \n\t" + "lw %[p_asm1], 448*4(%[p]) \n\t" + "msub %[w_asm], %[p_asm] \n\t" + "msub %[w_asm1], %[p_asm1] \n\t" + "extr.w %[sum1], $ac0, 24 \n\t" + "mflo %[temp2] \n\t" + "and %[temp1], %[temp2], 0x00ffffff \n\t" + "slt %[temp2], %[sum1], %[min_asm] \n\t" + "movn %[sum1], %[min_asm], %[temp2] \n\t" + "slt %[temp2], %[max_asm], %[sum1] \n\t" + "movn %[sum1], %[max_asm], %[temp2] \n\t" + "sh %[sum1], 0(%[samples]) \n\t" + + : [w_asm] "=&r" (w_asm), [p_asm] "=&r" (p_asm), [w_asm1] "=&r" (w_asm1), + [p_asm1] "=&r" (p_asm1), [temp1] "+r" (temp1), [temp2] "+r" (temp2), + [w_asm2] "=&r" (w_asm2), [p_asm2] "=&r" (p_asm2), [sum1] "+r" (sum1) + : [w] "r" (w), [p] "r" (p), [samples] "r" (samples), [min_asm] "r" (min_asm), + [max_asm] "r" (max_asm) + : "hi", "lo" + ); + + *dither_state= temp1; +} + +static void imdct36_mips_fixed(int *out, int *buf, int *in, int *win) +{ + int j; + int t0, t1, t2, t3, s0, s1, s2, s3; + int tmp[18], *tmp1, *in1; + /* temporary variables */ + int temp_reg1, temp_reg2, temp_reg3, temp_reg4, temp_reg5, temp_reg6; + int t4, t5, t6, t8, t7; + + /* values defined in macros and tables are + * eliminated - they are directly loaded in appropriate variables + */ + int const C_1 = 4229717092; /* cos(pi*1/18)*2 */ + int const C_2 = 4035949074; /* cos(pi*2/18)*2 */ + int const C_3 = 575416510; /* -cos(pi*3/18)*2 */ + int const C_3A = 3719550786; /* cos(pi*3/18)*2 */ + int const C_4 = 1004831466; /* -cos(pi*4/18)*2 */ + int const C_5 = 1534215534; /* -cos(pi*5/18)*2 */ + int const C_7 = -1468965330; /* -cos(pi*7/18)*2 */ + int const C_8 = -745813244; /* -cos(pi*8/18)*2 */ + + /* + * instructions of the first two loops are reorganized and loops are unrolled, + * in order to eliminate unnecessary readings and writings in array + */ + + __asm__ volatile ( + "lw %[t1], 17*4(%[in]) \n\t" + "lw %[t2], 16*4(%[in]) \n\t" + "lw %[t3], 15*4(%[in]) \n\t" + "lw %[t4], 14*4(%[in]) \n\t" + "addu %[t1], %[t1], %[t2] \n\t" + "addu %[t2], %[t2], %[t3] \n\t" + "addu %[t3], %[t3], %[t4] \n\t" + "lw %[t5], 13*4(%[in]) \n\t" + "addu %[t1], %[t1], %[t3] \n\t" + "sw %[t2], 16*4(%[in]) \n\t" + "lw %[t6], 12*4(%[in]) \n\t" + "sw %[t1], 17*4(%[in]) \n\t" + "addu %[t4], %[t4], %[t5] \n\t" + "addu %[t5], %[t5], %[t6] \n\t" + "lw %[t7], 11*4(%[in]) \n\t" + "addu %[t3], %[t3], %[t5] \n\t" + "sw %[t4], 14*4(%[in]) \n\t" + "lw %[t8], 10*4(%[in]) \n\t" + "sw %[t3], 15*4(%[in]) \n\t" + "addu %[t6], %[t6], %[t7] \n\t" + "addu %[t7], %[t7], %[t8] \n\t" + "sw %[t6], 12*4(%[in]) \n\t" + "addu %[t5], %[t5], %[t7] \n\t" + "lw %[t1], 9*4(%[in]) \n\t" + "lw %[t2], 8*4(%[in]) \n\t" + "sw %[t5], 13*4(%[in]) \n\t" + "addu %[t8], %[t8], %[t1] \n\t" + "addu %[t1], %[t1], %[t2] \n\t" + "sw %[t8], 10*4(%[in]) \n\t" + "addu %[t7], %[t7], %[t1] \n\t" + "lw %[t3], 7*4(%[in]) \n\t" + "lw %[t4], 6*4(%[in]) \n\t" + "sw %[t7], 11*4(%[in]) \n\t" + "addu %[t2], %[t2], %[t3] \n\t" + "addu %[t3], %[t3], %[t4] \n\t" + "sw %[t2], 8*4(%[in]) \n\t" + "addu %[t1], %[t1], %[t3] \n\t" + "lw %[t5], 5*4(%[in]) \n\t" + "lw %[t6], 4*4(%[in]) \n\t" + "sw %[t1], 9*4(%[in]) \n\t" + "addu %[t4], %[t4], %[t5] \n\t" + "addu %[t5], %[t5], %[t6] \n\t" + "sw %[t4], 6*4(%[in]) \n\t" + "addu %[t3], %[t3], %[t5] \n\t" + "lw %[t7], 3*4(%[in]) \n\t" + "lw %[t8], 2*4(%[in]) \n\t" + "sw %[t3], 7*4(%[in]) \n\t" + "addu %[t6], %[t6], %[t7] \n\t" + "addu %[t7], %[t7], %[t8] \n\t" + "sw %[t6], 4*4(%[in]) \n\t" + "addu %[t5], %[t5], %[t7] \n\t" + "lw %[t1], 1*4(%[in]) \n\t" + "lw %[t2], 0*4(%[in]) \n\t" + "sw %[t5], 5*4(%[in]) \n\t" + "addu %[t8], %[t8], %[t1] \n\t" + "addu %[t1], %[t1], %[t2] \n\t" + "sw %[t8], 2*4(%[in]) \n\t" + "addu %[t7], %[t7], %[t1] \n\t" + "sw %[t7], 3*4(%[in]) \n\t" + "sw %[t1], 1*4(%[in]) \n\t" + + : [in] "+r" (in), [t1] "=&r" (t1), [t2] "=&r" (t2), [t3] "=&r" (t3), + [t4] "=&r" (t4), [t5] "=&r" (t5), [t6] "=&r" (t6), + [t7] "=&r" (t7), [t8] "=&r" (t8) + ); + + for(j = 0; j < 2; j++) { + + tmp1 = tmp + j; + in1 = in + j; + + /** + * Original constants are multiplied by two in advanced + * for assembly optimization (e.g. C_2 = 2 * C2). + * That can lead to overflow in operations where they are used. + * + * Example of the solution: + * + * in original code: + * t0 = ((int64_t)(in1[2*2] + in1[2*4]) * (int64_t)(2*C2))>>32 + * + * in assembly: + * C_2 = 2 * C2; + * . + * . + * "lw %[t7], 4*4(%[in1]) \n\t" + * "lw %[t8], 8*4(%[in1]) \n\t" + * "addu %[temp_reg2],%[t7], %[t8] \n\t" + * "multu %[C_2], %[temp_reg2] \n\t" + * "mfhi %[temp_reg1] \n\t" + * "sra %[temp_reg2],%[temp_reg2],31 \n\t" + * "move %[t0], $0 \n\t" + * "movn %[t0], %[C_2], %[temp_reg2] \n\t" + * "sub %[t0], %[temp_reg1],%[t0] \n\t" + */ + + __asm__ volatile ( + "lw %[t7], 4*4(%[in1]) \n\t" + "lw %[t8], 8*4(%[in1]) \n\t" + "lw %[t6], 16*4(%[in1]) \n\t" + "lw %[t4], 0*4(%[in1]) \n\t" + "addu %[temp_reg2], %[t7], %[t8] \n\t" + "addu %[t2], %[t6], %[t8] \n\t" + "multu %[C_2], %[temp_reg2] \n\t" + "lw %[t5], 12*4(%[in1]) \n\t" + "sub %[t2], %[t2], %[t7] \n\t" + "sub %[t1], %[t4], %[t5] \n\t" + "sra %[t3], %[t5], 1 \n\t" + "sra %[temp_reg1], %[t2], 1 \n\t" + "addu %[t3], %[t3], %[t4] \n\t" + "sub %[temp_reg1], %[t1], %[temp_reg1] \n\t" + "sra %[temp_reg2], %[temp_reg2], 31 \n\t" + "sw %[temp_reg1], 6*4(%[tmp1]) \n\t" + "move %[t0], $0 \n\t" + "movn %[t0], %[C_2], %[temp_reg2] \n\t" + "mfhi %[temp_reg1] \n\t" + "addu %[t1], %[t1], %[t2] \n\t" + "sw %[t1], 16*4(%[tmp1]) \n\t" + "sub %[temp_reg4], %[t8], %[t6] \n\t" + "add %[temp_reg2], %[t7], %[t6] \n\t" + "mult $ac1, %[C_8], %[temp_reg4] \n\t" + "multu $ac2, %[C_4], %[temp_reg2] \n\t" + "sub %[t0], %[temp_reg1], %[t0] \n\t" + "sra %[temp_reg1], %[temp_reg2], 31 \n\t" + "move %[t2], $0 \n\t" + "movn %[t2], %[C_4], %[temp_reg1] \n\t" + "mfhi %[t1], $ac1 \n\t" + "mfhi %[temp_reg1], $ac2 \n\t" + "lw %[t6], 10*4(%[in1]) \n\t" + "lw %[t8], 14*4(%[in1]) \n\t" + "lw %[t7], 2*4(%[in1]) \n\t" + "lw %[t4], 6*4(%[in1]) \n\t" + "sub %[temp_reg3], %[t3], %[t0] \n\t" + "add %[temp_reg4], %[t3], %[t0] \n\t" + "sub %[temp_reg1], %[temp_reg1], %[temp_reg2] \n\t" + "add %[temp_reg4], %[temp_reg4], %[t1] \n\t" + "sub %[t2], %[temp_reg1], %[t2] \n\t" + "sw %[temp_reg4], 2*4(%[tmp1]) \n\t" + "sub %[temp_reg3], %[temp_reg3], %[t2] \n\t" + "add %[temp_reg1], %[t3], %[t2] \n\t" + "sw %[temp_reg3], 10*4(%[tmp1]) \n\t" + "sub %[temp_reg1], %[temp_reg1], %[t1] \n\t" + "addu %[temp_reg2], %[t6], %[t8] \n\t" + "sw %[temp_reg1], 14*4(%[tmp1]) \n\t" + "sub %[temp_reg2], %[temp_reg2], %[t7] \n\t" + "addu %[temp_reg3], %[t7], %[t6] \n\t" + "multu $ac3, %[C_3], %[temp_reg2] \n\t" + "multu %[C_1], %[temp_reg3] \n\t" + "sra %[temp_reg1], %[temp_reg2], 31 \n\t" + "move %[t1], $0 \n\t" + "sra %[temp_reg3], %[temp_reg3], 31 \n\t" + "movn %[t1], %[C_3], %[temp_reg1] \n\t" + "mfhi %[temp_reg1], $ac3 \n\t" + "mfhi %[temp_reg4] \n\t" + "move %[t2], $0 \n\t" + "movn %[t2], %[C_1], %[temp_reg3] \n\t" + "sub %[temp_reg3], %[t6], %[t8] \n\t" + "sub %[t2], %[temp_reg4], %[t2] \n\t" + "multu $ac1, %[C_7], %[temp_reg3] \n\t" + "sub %[temp_reg1], %[temp_reg1], %[temp_reg2] \n\t" + "sra %[temp_reg4], %[temp_reg3], 31 \n\t" + "sub %[t1], %[temp_reg1], %[t1] \n\t" + "move %[t3], $0 \n\t" + "sw %[t1], 4*4(%[tmp1]) \n\t" + "movn %[t3], %[C_7], %[temp_reg4] \n\t" + "multu $ac2, %[C_3A], %[t4] \n\t" + "add %[temp_reg2], %[t7], %[t8] \n\t" + "move %[t1], $0 \n\t" + "mfhi %[temp_reg4], $ac1 \n\t" + "multu $ac3,%[C_5], %[temp_reg2] \n\t" + "move %[t0], $0 \n\t" + "sra %[temp_reg1], %[temp_reg2], 31 \n\t" + "movn %[t1],%[C_5], %[temp_reg1] \n\t" + "sub %[temp_reg4], %[temp_reg4], %[temp_reg3] \n\t" + "mfhi %[temp_reg1], $ac3 \n\t" + "sra %[temp_reg3], %[t4], 31 \n\t" + "movn %[t0], %[C_3A], %[temp_reg3] \n\t" + "mfhi %[temp_reg3], $ac2 \n\t" + "sub %[t3], %[temp_reg4], %[t3] \n\t" + "add %[temp_reg4], %[t3], %[t2] \n\t" + "sub %[temp_reg1], %[temp_reg1], %[temp_reg2] \n\t" + "sub %[t1], %[temp_reg1], %[t1] \n\t" + "sub %[t0], %[temp_reg3], %[t0] \n\t" + "add %[temp_reg1], %[t2], %[t1] \n\t" + "add %[temp_reg4], %[temp_reg4], %[t0] \n\t" + "sub %[temp_reg2], %[t3], %[t1] \n\t" + "sw %[temp_reg4], 0*4(%[tmp1]) \n\t" + "sub %[temp_reg1], %[temp_reg1], %[t0] \n\t" + "sub %[temp_reg2], %[temp_reg2], %[t0] \n\t" + "sw %[temp_reg1], 12*4(%[tmp1]) \n\t" + "sw %[temp_reg2], 8*4(%[tmp1]) \n\t" + + : [t7] "=&r" (t7), [temp_reg1] "=&r" (temp_reg1), + [temp_reg2] "=&r" (temp_reg2), [temp_reg4] "=&r" (temp_reg4), + [temp_reg3] "=&r" (temp_reg3), [t8] "=&r" (t8), [t0] "=&r" (t0), + [t4] "=&r" (t4), [t5] "=&r" (t5), [t6] "=&r"(t6), [t2] "=&r" (t2), + [t3] "=&r" (t3), [t1] "=&r" (t1) + : [C_2] "r" (C_2), [in1] "r" (in1), [tmp1] "r" (tmp1), [C_8] "r" (C_8), + [C_4] "r" (C_4), [C_3] "r" (C_3), [C_1] "r" (C_1), [C_7] "r" (C_7), + [C_3A] "r" (C_3A), [C_5] "r" (C_5) + : "hi", "lo" + ); + } + + /** + * loop is unrolled four times + * + * values defined in tables(icos36[] and icos36h[]) are not loaded from + * these tables - they are directly loaded in appropriate registers + * + */ + + __asm__ volatile ( + "lw %[t2], 1*4(%[tmp]) \n\t" + "lw %[t3], 3*4(%[tmp]) \n\t" + "lw %[t0], 0*4(%[tmp]) \n\t" + "lw %[t1], 2*4(%[tmp]) \n\t" + "addu %[temp_reg1], %[t3], %[t2] \n\t" + "li %[temp_reg2], 0x807D2B1E \n\t" + "move %[s1], $0 \n\t" + "multu %[temp_reg2], %[temp_reg1] \n\t" + "sra %[temp_reg1], %[temp_reg1], 31 \n\t" + "movn %[s1], %[temp_reg2], %[temp_reg1] \n\t" + "sub %[temp_reg3], %[t3], %[t2] \n\t" + "li %[temp_reg4], 0x2de5151 \n\t" + "mfhi %[temp_reg2] \n\t" + "addu %[s0], %[t1], %[t0] \n\t" + "lw %[temp_reg5], 9*4(%[win]) \n\t" + "mult $ac1, %[temp_reg4], %[temp_reg3] \n\t" + "lw %[temp_reg6], 4*9*4(%[buf]) \n\t" + "sub %[s2], %[t1], %[t0] \n\t" + "lw %[temp_reg3], 29*4(%[win]) \n\t" + "subu %[s1], %[temp_reg2], %[s1] \n\t" + "lw %[temp_reg4], 28*4(%[win]) \n\t" + "add %[t0], %[s0], %[s1] \n\t" + "extr.w %[s3], $ac1,23 \n\t" + "mult $ac2, %[t0], %[temp_reg3] \n\t" + "sub %[t1], %[s0], %[s1] \n\t" + "lw %[temp_reg1], 4*8*4(%[buf]) \n\t" + "mult %[t1], %[temp_reg5] \n\t" + "lw %[temp_reg2], 8*4(%[win]) \n\t" + "mfhi %[temp_reg3], $ac2 \n\t" + "mult $ac3, %[t0], %[temp_reg4] \n\t" + "add %[t0], %[s2], %[s3] \n\t" + "mfhi %[temp_reg5] \n\t" + "mult $ac1, %[t1], %[temp_reg2] \n\t" + "sub %[t1], %[s2], %[s3] \n\t" + "sw %[temp_reg3], 4*9*4(%[buf]) \n\t" + "mfhi %[temp_reg4], $ac3 \n\t" + "lw %[temp_reg3], 37*4(%[win]) \n\t" + "mfhi %[temp_reg2], $ac1 \n\t" + "add %[temp_reg5], %[temp_reg5], %[temp_reg6] \n\t" + "lw %[temp_reg6], 17*4(%[win]) \n\t" + "sw %[temp_reg5], 32*9*4(%[out]) \n\t" + "sw %[temp_reg4], 4*8*4(%[buf]) \n\t" + "mult %[t1], %[temp_reg6] \n\t" + "add %[temp_reg1], %[temp_reg1], %[temp_reg2] \n\t" + "lw %[temp_reg2], 0*4(%[win]) \n\t" + "lw %[temp_reg5], 4*17*4(%[buf]) \n\t" + "sw %[temp_reg1], 8*32*4(%[out]) \n\t" + "mfhi %[temp_reg6] \n\t" + "mult $ac1, %[t1], %[temp_reg2] \n\t" + "lw %[temp_reg4], 20*4(%[win]) \n\t" + "lw %[temp_reg1], 0(%[buf]) \n\t" + "mult $ac2, %[t0], %[temp_reg3] \n\t" + "mult %[t0], %[temp_reg4] \n\t" + "mfhi %[temp_reg2], $ac1 \n\t" + "lw %[t0], 4*4(%[tmp]) \n\t" + "add %[temp_reg5], %[temp_reg5], %[temp_reg6] \n\t" + "mfhi %[temp_reg3], $ac2 \n\t" + "mfhi %[temp_reg4] \n\t" + "sw %[temp_reg5], 17*32*4(%[out]) \n\t" + "lw %[t1], 6*4(%[tmp]) \n\t" + "add %[temp_reg1], %[temp_reg1], %[temp_reg2] \n\t" + "lw %[t2], 5*4(%[tmp]) \n\t" + "sw %[temp_reg1], 0*32*4(%[out]) \n\t" + "addu %[s0], %[t1], %[t0] \n\t" + "sw %[temp_reg3], 4*17*4(%[buf]) \n\t" + "lw %[t3], 7*4(%[tmp]) \n\t" + "sub %[s2], %[t1], %[t0] \n\t" + "sw %[temp_reg4], 0(%[buf]) \n\t" + "addu %[temp_reg5], %[t3], %[t2] \n\t" + "li %[temp_reg6], 0x8483EE0C \n\t" + "move %[s1], $0 \n\t" + "multu %[temp_reg6], %[temp_reg5] \n\t" + "sub %[temp_reg1], %[t3], %[t2] \n\t" + "li %[temp_reg2], 0xf746ea \n\t" + "sra %[temp_reg5], %[temp_reg5], 31 \n\t" + "mult $ac1, %[temp_reg2], %[temp_reg1] \n\t" + "movn %[s1], %[temp_reg6], %[temp_reg5] \n\t" + "mfhi %[temp_reg5] \n\t" + "lw %[temp_reg3], 10*4(%[win]) \n\t" + "lw %[temp_reg4], 4*10*4(%[buf]) \n\t" + "extr.w %[s3], $ac1, 23 \n\t" + "lw %[temp_reg1], 4*7*4(%[buf]) \n\t" + "lw %[temp_reg2], 7*4(%[win]) \n\t" + "lw %[temp_reg6], 30*4(%[win]) \n\t" + "subu %[s1], %[temp_reg5], %[s1] \n\t" + "sub %[t1], %[s0], %[s1] \n\t" + "add %[t0], %[s0], %[s1] \n\t" + "mult $ac2, %[t1], %[temp_reg3] \n\t" + "mult $ac3, %[t1], %[temp_reg2] \n\t" + "mult %[t0], %[temp_reg6] \n\t" + "lw %[temp_reg5], 27*4(%[win]) \n\t" + "mult $ac1, %[t0], %[temp_reg5] \n\t" + "mfhi %[temp_reg3], $ac2 \n\t" + "mfhi %[temp_reg2], $ac3 \n\t" + "mfhi %[temp_reg6] \n\t" + "add %[t0], %[s2], %[s3] \n\t" + "sub %[t1], %[s2], %[s3] \n\t" + "add %[temp_reg3], %[temp_reg3], %[temp_reg4] \n\t" + "lw %[temp_reg4], 16*4(%[win]) \n\t" + "mfhi %[temp_reg5], $ac1 \n\t" + "sw %[temp_reg3], 32*10*4(%[out]) \n\t" + "add %[temp_reg1], %[temp_reg1], %[temp_reg2] \n\t" + "lw %[temp_reg3], 4*16*4(%[buf]) \n\t" + "sw %[temp_reg6], 4*10*4(%[buf]) \n\t" + "sw %[temp_reg1], 7*32*4(%[out]) \n\t" + "mult $ac2, %[t1], %[temp_reg4] \n\t" + "sw %[temp_reg5], 4*7*4(%[buf]) \n\t" + "lw %[temp_reg6], 1*4(%[win]) \n\t" + "lw %[temp_reg5], 4*1*4(%[buf]) \n\t" + "lw %[temp_reg1], 36*4(%[win]) \n\t" + "mult $ac3, %[t1], %[temp_reg6] \n\t" + "lw %[temp_reg2], 21*4(%[win]) \n\t" + "mfhi %[temp_reg4], $ac2 \n\t" + "mult %[t0], %[temp_reg1] \n\t" + "mult $ac1, %[t0],%[temp_reg2] \n\t" + "lw %[t0], 8*4(%[tmp]) \n\t" + "mfhi %[temp_reg6], $ac3 \n\t" + "lw %[t1], 10*4(%[tmp]) \n\t" + "lw %[t3], 11*4(%[tmp]) \n\t" + "mfhi %[temp_reg1] \n\t" + "add %[temp_reg3], %[temp_reg3], %[temp_reg4] \n\t" + "lw %[t2], 9*4(%[tmp]) \n\t" + "mfhi %[temp_reg2], $ac1 \n\t" + "add %[temp_reg5], %[temp_reg5], %[temp_reg6] \n\t" + "sw %[temp_reg3], 16*32*4(%[out]) \n\t" + "sw %[temp_reg5], 1*32*4(%[out]) \n\t" + "sw %[temp_reg1], 4*16*4(%[buf]) \n\t" + "addu %[temp_reg3], %[t3], %[t2] \n\t" + "li %[temp_reg4], 0x8D3B7CD6 \n\t" + "sw %[temp_reg2], 4*1*4(%[buf]) \n\t" + "multu %[temp_reg4],%[temp_reg3] \n\t" + "sra %[temp_reg3], %[temp_reg3], 31 \n\t" + "move %[s1], $0 \n\t" + "movn %[s1], %[temp_reg4], %[temp_reg3] \n\t" + "addu %[s0], %[t1], %[t0] \n\t" + "mfhi %[temp_reg3] \n\t" + "sub %[s2], %[t1], %[t0] \n\t" + "sub %[temp_reg5], %[t3], %[t2] \n\t" + "li %[temp_reg6], 0x976fd9 \n\t" + "lw %[temp_reg2], 11*4(%[win]) \n\t" + "lw %[temp_reg1], 4*11*4(%[buf]) \n\t" + "mult $ac1, %[temp_reg6], %[temp_reg5] \n\t" + "subu %[s1], %[temp_reg3], %[s1] \n\t" + "lw %[temp_reg5], 31*4(%[win]) \n\t" + "sub %[t1], %[s0], %[s1] \n\t" + "add %[t0], %[s0], %[s1] \n\t" + "mult $ac2, %[t1], %[temp_reg2] \n\t" + "mult %[t0], %[temp_reg5] \n\t" + "lw %[temp_reg4], 6*4(%[win]) \n\t" + "extr.w %[s3], $ac1, 23 \n\t" + "lw %[temp_reg3], 4*6*4(%[buf]) \n\t" + "mfhi %[temp_reg2], $ac2 \n\t" + "lw %[temp_reg6], 26*4(%[win]) \n\t" + "mfhi %[temp_reg5] \n\t" + "mult $ac3, %[t1], %[temp_reg4] \n\t" + "mult $ac1, %[t0], %[temp_reg6] \n\t" + "add %[t0], %[s2], %[s3] \n\t" + "sub %[t1], %[s2], %[s3] \n\t" + "add %[temp_reg2], %[temp_reg2], %[temp_reg1] \n\t" + "mfhi %[temp_reg4], $ac3 \n\t" + "mfhi %[temp_reg6], $ac1 \n\t" + "sw %[temp_reg5], 4*11*4(%[buf]) \n\t" + "sw %[temp_reg2], 32*11*4(%[out]) \n\t" + "lw %[temp_reg1], 4*15*4(%[buf]) \n\t" + "add %[temp_reg3], %[temp_reg3], %[temp_reg4] \n\t" + "lw %[temp_reg2], 15*4(%[win]) \n\t" + "sw %[temp_reg3], 6*32*4(%[out]) \n\t" + "sw %[temp_reg6], 4*6*4(%[buf]) \n\t" + "mult %[t1], %[temp_reg2] \n\t" + "lw %[temp_reg3], 2*4(%[win]) \n\t" + "lw %[temp_reg4], 4*2*4(%[buf]) \n\t" + "lw %[temp_reg5], 35*4(%[win]) \n\t" + "mult $ac1, %[t1], %[temp_reg3] \n\t" + "mfhi %[temp_reg2] \n\t" + "lw %[temp_reg6], 22*4(%[win]) \n\t" + "mult $ac2, %[t0], %[temp_reg5] \n\t" + "lw %[t1], 14*4(%[tmp]) \n\t" + "mult $ac3, %[t0], %[temp_reg6] \n\t" + "lw %[t0], 12*4(%[tmp]) \n\t" + "mfhi %[temp_reg3], $ac1 \n\t" + "add %[temp_reg1], %[temp_reg1], %[temp_reg2] \n\t" + "mfhi %[temp_reg5], $ac2 \n\t" + "sw %[temp_reg1], 15*32*4(%[out]) \n\t" + "mfhi %[temp_reg6], $ac3 \n\t" + "lw %[t2], 13*4(%[tmp]) \n\t" + "lw %[t3], 15*4(%[tmp]) \n\t" + "add %[temp_reg4], %[temp_reg4], %[temp_reg3] \n\t" + "sw %[temp_reg5], 4*15*4(%[buf]) \n\t" + "addu %[temp_reg1], %[t3], %[t2] \n\t" + "li %[temp_reg2], 0x9C42577C \n\t" + "move %[s1], $0 \n\t" + "multu %[temp_reg2], %[temp_reg1] \n\t" + "sw %[temp_reg4], 2*32*4(%[out]) \n\t" + "sra %[temp_reg1], %[temp_reg1], 31 \n\t" + "movn %[s1], %[temp_reg2], %[temp_reg1] \n\t" + "sub %[temp_reg3], %[t3], %[t2] \n\t" + "li %[temp_reg4], 0x6f94a2 \n\t" + "mfhi %[temp_reg1] \n\t" + "addu %[s0], %[t1], %[t0] \n\t" + "sw %[temp_reg6], 4*2*4(%[buf]) \n\t" + "mult $ac1, %[temp_reg4], %[temp_reg3] \n\t" + "sub %[s2], %[t1], %[t0] \n\t" + "lw %[temp_reg5], 12*4(%[win]) \n\t" + "lw %[temp_reg6], 4*12*4(%[buf]) \n\t" + "subu %[s1], %[temp_reg1], %[s1] \n\t" + "sub %[t1], %[s0], %[s1] \n\t" + "lw %[temp_reg3], 32*4(%[win]) \n\t" + "mult $ac2, %[t1], %[temp_reg5] \n\t" + "add %[t0], %[s0], %[s1] \n\t" + "extr.w %[s3], $ac1, 23 \n\t" + "lw %[temp_reg2], 5*4(%[win]) \n\t" + "mult %[t0], %[temp_reg3] \n\t" + "mfhi %[temp_reg5], $ac2 \n\t" + "lw %[temp_reg4], 25*4(%[win]) \n\t" + "lw %[temp_reg1], 4*5*4(%[buf]) \n\t" + "mult $ac3, %[t1], %[temp_reg2] \n\t" + "mult $ac1, %[t0], %[temp_reg4] \n\t" + "mfhi %[temp_reg3] \n\t" + "add %[t0], %[s2], %[s3] \n\t" + "add %[temp_reg5], %[temp_reg5], %[temp_reg6] \n\t" + "mfhi %[temp_reg2], $ac3 \n\t" + "mfhi %[temp_reg4], $ac1 \n\t" + "sub %[t1], %[s2], %[s3] \n\t" + "sw %[temp_reg5], 32*12*4(%[out]) \n\t" + "sw %[temp_reg3], 4*12*4(%[buf]) \n\t" + "lw %[temp_reg6], 14*4(%[win]) \n\t" + "lw %[temp_reg5], 4*14*4(%[buf]) \n\t" + "add %[temp_reg1], %[temp_reg1], %[temp_reg2] \n\t" + "sw %[temp_reg4], 4*5*4(%[buf]) \n\t" + "sw %[temp_reg1], 5*32*4(%[out]) \n\t" + "mult %[t1], %[temp_reg6] \n\t" + "lw %[temp_reg4], 34*4(%[win]) \n\t" + "lw %[temp_reg2], 3*4(%[win]) \n\t" + "lw %[temp_reg1], 4*3*4(%[buf]) \n\t" + "mult $ac2, %[t0], %[temp_reg4] \n\t" + "mfhi %[temp_reg6] \n\t" + "mult $ac1, %[t1], %[temp_reg2] \n\t" + "lw %[temp_reg3], 23*4(%[win]) \n\t" + "lw %[s0], 16*4(%[tmp]) \n\t" + "mfhi %[temp_reg4], $ac2 \n\t" + "lw %[t1], 17*4(%[tmp]) \n\t" + "mult $ac3, %[t0], %[temp_reg3] \n\t" + "move %[s1], $0 \n\t" + "add %[temp_reg5], %[temp_reg5], %[temp_reg6] \n\t" + "mfhi %[temp_reg2], $ac1 \n\t" + "sw %[temp_reg5], 14*32*4(%[out]) \n\t" + "sw %[temp_reg4], 4*14*4(%[buf]) \n\t" + "mfhi %[temp_reg3], $ac3 \n\t" + "li %[temp_reg5], 0xB504F334 \n\t" + "add %[temp_reg1], %[temp_reg1], %[temp_reg2] \n\t" + "multu %[temp_reg5], %[t1] \n\t" + "lw %[temp_reg2], 4*13*4(%[buf]) \n\t" + "sw %[temp_reg1], 3*32*4(%[out]) \n\t" + "sra %[t1], %[t1], 31 \n\t" + "mfhi %[temp_reg6] \n\t" + "movn %[s1], %[temp_reg5], %[t1] \n\t" + "sw %[temp_reg3], 4*3*4(%[buf]) \n\t" + "lw %[temp_reg1], 13*4(%[win]) \n\t" + "lw %[temp_reg4], 4*4*4(%[buf]) \n\t" + "lw %[temp_reg3], 4*4(%[win]) \n\t" + "lw %[temp_reg5], 33*4(%[win]) \n\t" + "subu %[s1], %[temp_reg6], %[s1] \n\t" + "lw %[temp_reg6], 24*4(%[win]) \n\t" + "sub %[t1], %[s0], %[s1] \n\t" + "add %[t0], %[s0], %[s1] \n\t" + "mult $ac1, %[t1], %[temp_reg1] \n\t" + "mult $ac2, %[t1], %[temp_reg3] \n\t" + "mult $ac3, %[t0], %[temp_reg5] \n\t" + "mult %[t0], %[temp_reg6] \n\t" + "mfhi %[temp_reg1], $ac1 \n\t" + "mfhi %[temp_reg3], $ac2 \n\t" + "mfhi %[temp_reg5], $ac3 \n\t" + "mfhi %[temp_reg6] \n\t" + "add %[temp_reg2], %[temp_reg2], %[temp_reg1] \n\t" + "add %[temp_reg4], %[temp_reg4], %[temp_reg3] \n\t" + "sw %[temp_reg2], 13*32*4(%[out]) \n\t" + "sw %[temp_reg4], 4*32*4(%[out]) \n\t" + "sw %[temp_reg5], 4*13*4(%[buf]) \n\t" + "sw %[temp_reg6], 4*4*4(%[buf]) \n\t" + + : [t0] "=&r" (t0), [t1] "=&r" (t1), [t2] "=&r" (t2), [t3] "=&r" (t3), + [s0] "=&r" (s0), [s2] "=&r" (s2), [temp_reg1] "=&r" (temp_reg1), + [temp_reg2] "=&r" (temp_reg2), [s1] "=&r" (s1), [s3] "=&r" (s3), + [temp_reg3] "=&r" (temp_reg3), [temp_reg4] "=&r" (temp_reg4), + [temp_reg5] "=&r" (temp_reg5), [temp_reg6] "=&r" (temp_reg6), + [out] "+r" (out) + : [tmp] "r" (tmp), [win] "r" (win), [buf] "r" (buf) + : "hi", "lo" + ); +} + +static void ff_imdct36_blocks_mips_fixed(int *out, int *buf, int *in, + int count, int switch_point, int block_type) +{ + int j; + for (j=0 ; j < count; j++) { + /* apply window & overlap with previous buffer */ + + /* select window */ + int win_idx = (switch_point && j < 2) ? 0 : block_type; + int *win = ff_mdct_win_fixed[win_idx + (4 & -(j & 1))]; + + imdct36_mips_fixed(out, buf, in, win); + + in += 18; + buf += ((j&3) != 3 ? 1 : (72-3)); + out++; + } +} + +void ff_mpadsp_init_mipsdspr1(MPADSPContext *s) +{ + s->apply_window_fixed = ff_mpadsp_apply_window_mips_fixed; + s->imdct36_blocks_fixed = ff_imdct36_blocks_mips_fixed; +} diff --git a/libavcodec/mips/mpegaudiodsp_mips_float.c b/libavcodec/mips/mpegaudiodsp_mips_float.c new file mode 100644 index 0000000000..00a7f20c1e --- /dev/null +++ b/libavcodec/mips/mpegaudiodsp_mips_float.c @@ -0,0 +1,1242 @@ +/* + * Copyright (c) 2012 + * MIPS Technologies, Inc., California. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its + * contributors may be used to endorse or promote products derived from + * this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * Author: Bojan Zivkovic (bojan@mips.com) + * + * MPEG Audio decoder optimized for MIPS floating-point architecture + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +/** + * @file + * Reference: libavcodec/mpegaudiodsp_template.c + * libavcodec/dct32.c + */ + +#include <string.h> + +#include "libavcodec/mpegaudiodsp.h" + +static void ff_mpadsp_apply_window_mips_float(float *synth_buf, float *window, + int *dither_state, float *samples, int incr) +{ + register const float *w, *w2, *p; + int j; + float *samples2; + float sum, sum2; + /* temporary variables */ + int incr1 = incr << 2; + int t_sample; + float in1, in2, in3, in4, in5, in6, in7, in8; + float *p2; + + /* copy to avoid wrap */ + memcpy(synth_buf + 512, synth_buf, 32 * sizeof(*synth_buf)); + + /** + * instructions are scheduled to minimize pipeline stall. + * use of round_sample function from the original code is + * changed with appropriate assembly instructions. + */ + + __asm__ volatile ( + "lwc1 %[sum], 0(%[dither_state]) \t\n" + "sll %[t_sample], %[incr1], 5 \t\n" + "sub %[t_sample], %[t_sample], %[incr1] \n\t" + "li %[j], 4 \t\n" + "lwc1 %[in1], 0(%[window]) \t\n" + "lwc1 %[in2], 16*4(%[synth_buf]) \t\n" + "sw $zero, 0(%[dither_state]) \t\n" + "lwc1 %[in3], 64*4(%[window]) \t\n" + "lwc1 %[in4], 80*4(%[synth_buf]) \t\n" + "addu %[samples2], %[samples], %[t_sample] \t\n" + "madd.s %[sum], %[sum], %[in1], %[in2] \t\n" + "lwc1 %[in5], 128*4(%[window]) \t\n" + "lwc1 %[in6], 144*4(%[synth_buf]) \t\n" + "lwc1 %[in7], 192*4(%[window]) \t\n" + "madd.s %[sum], %[sum], %[in3], %[in4] \t\n" + "lwc1 %[in8], 208*4(%[synth_buf]) \t\n" + "lwc1 %[in1], 256*4(%[window]) \t\n" + "lwc1 %[in2], 272*4(%[synth_buf]) \t\n" + "madd.s %[sum], %[sum], %[in5], %[in6] \t\n" + "lwc1 %[in3], 320*4(%[window]) \t\n" + "lwc1 %[in4], 336*4(%[synth_buf]) \t\n" + "lwc1 %[in5], 384*4(%[window]) \t\n" + "madd.s %[sum], %[sum], %[in7], %[in8] \t\n" + "lwc1 %[in6], 400*4(%[synth_buf]) \t\n" + "lwc1 %[in7], 448*4(%[window]) \t\n" + "lwc1 %[in8], 464*4(%[synth_buf]) \t\n" + "madd.s %[sum], %[sum], %[in1], %[in2] \t\n" + "lwc1 %[in1], 32*4(%[window]) \t\n" + "lwc1 %[in2], 48*4(%[synth_buf]) \t\n" + "madd.s %[sum], %[sum], %[in3], %[in4] \t\n" + "lwc1 %[in3], 96*4(%[window]) \t\n" + "lwc1 %[in4], 112*4(%[synth_buf]) \t\n" + "madd.s %[sum], %[sum], %[in5], %[in6] \t\n" + "lwc1 %[in5], 160*4(%[window]) \t\n" + "lwc1 %[in6], 176*4(%[synth_buf]) \t\n" + "madd.s %[sum], %[sum], %[in7], %[in8] \t\n" + "lwc1 %[in7], 224*4(%[window]) \t\n" + "lwc1 %[in8], 240*4(%[synth_buf]) \t\n" + "nmsub.s %[sum], %[sum], %[in1], %[in2] \t\n" + "lwc1 %[in1], 288*4(%[window]) \t\n" + "lwc1 %[in2], 304*4(%[synth_buf]) \t\n" + "nmsub.s %[sum], %[sum], %[in3], %[in4] \t\n" + "lwc1 %[in3], 352*4(%[window]) \t\n" + "lwc1 %[in4], 368*4(%[synth_buf]) \t\n" + "nmsub.s %[sum], %[sum], %[in5], %[in6] \t\n" + "lwc1 %[in5], 416*4(%[window]) \t\n" + "lwc1 %[in6], 432*4(%[synth_buf]) \t\n" + "nmsub.s %[sum], %[sum], %[in7], %[in8] \t\n" + "lwc1 %[in7], 480*4(%[window]) \t\n" + "lwc1 %[in8], 496*4(%[synth_buf]) \t\n" + "nmsub.s %[sum], %[sum], %[in1], %[in2] \t\n" + "addu %[w], %[window], 4 \t\n" + "nmsub.s %[sum], %[sum], %[in3], %[in4] \t\n" + "addu %[w2], %[window], 124 \t\n" + "addiu %[p], %[synth_buf], 68 \t\n" + "addiu %[p2], %[synth_buf], 188 \t\n" + "nmsub.s %[sum], %[sum], %[in5], %[in6] \t\n" + "nmsub.s %[sum], %[sum], %[in7], %[in8] \t\n" + "swc1 %[sum], 0(%[samples]) \t\n" + "addu %[samples], %[samples], %[incr1] \t\n" + + /* calculate two samples at the same time to avoid one memory + access per two sample */ + + "ff_mpadsp_apply_window_loop%=: \t\n" + "lwc1 %[in1], 0(%[w]) \t\n" + "lwc1 %[in2], 0(%[p]) \t\n" + "lwc1 %[in3], 0(%[w2]) \t\n" + "lwc1 %[in4], 64*4(%[w]) \t\n" + "lwc1 %[in5], 64*4(%[p]) \t\n" + "lwc1 %[in6], 64*4(%[w2]) \t\n" + "mul.s %[sum], %[in1], %[in2] \t\n" + "mul.s %[sum2], %[in2], %[in3] \t\n" + "lwc1 %[in1], 128*4(%[w]) \t\n" + "lwc1 %[in2], 128*4(%[p]) \t\n" + "madd.s %[sum], %[sum], %[in4], %[in5] \t\n" + "nmadd.s %[sum2], %[sum2], %[in5], %[in6] \t\n" + "lwc1 %[in3], 128*4(%[w2]) \t\n" + "lwc1 %[in4], 192*4(%[w]) \t\n" + "madd.s %[sum], %[sum], %[in1], %[in2] \t\n" + "lwc1 %[in5], 192*4(%[p]) \t\n" + "lwc1 %[in6], 192*4(%[w2]) \t\n" + "nmsub.s %[sum2], %[sum2], %[in2], %[in3] \t\n" + "lwc1 %[in1], 256*4(%[w]) \t\n" + "lwc1 %[in2], 256*4(%[p]) \t\n" + "madd.s %[sum], %[sum], %[in4], %[in5] \t\n" + "nmsub.s %[sum2], %[sum2], %[in5], %[in6] \t\n" + "lwc1 %[in3], 256*4(%[w2]) \t\n" + "lwc1 %[in4], 320*4(%[w]) \t\n" + "madd.s %[sum], %[sum], %[in1], %[in2] \t\n" + "lwc1 %[in5], 320*4(%[p]) \t\n" + "lwc1 %[in6], 320*4(%[w2]) \t\n" + "nmsub.s %[sum2], %[sum2], %[in2], %[in3] \t\n" + "lwc1 %[in1], 384*4(%[w]) \t\n" + "lwc1 %[in2], 384*4(%[p]) \t\n" + "madd.s %[sum], %[sum], %[in4], %[in5] \t\n" + "nmsub.s %[sum2], %[sum2], %[in5], %[in6] \t\n" + "lwc1 %[in3], 384*4(%[w2]) \t\n" + "lwc1 %[in4], 448*4(%[w]) \t\n" + "madd.s %[sum], %[sum], %[in1], %[in2] \t\n" + "lwc1 %[in5], 448*4(%[p]) \t\n" + "lwc1 %[in6], 448*4(%[w2]) \t\n" + "nmsub.s %[sum2], %[sum2], %[in2], %[in3] \t\n" + "madd.s %[sum], %[sum], %[in4], %[in5] \t\n" + "lwc1 %[in1], 32*4(%[w]) \t\n" + "lwc1 %[in2], 0(%[p2]) \t\n" + "nmsub.s %[sum2], %[sum2], %[in5], %[in6] \t\n" + "lwc1 %[in3], 32*4(%[w2]) \t\n" + "lwc1 %[in4], 96*4(%[w]) \t\n" + "lwc1 %[in5], 64*4(%[p2]) \t\n" + "nmsub.s %[sum], %[sum], %[in1], %[in2] \t\n" + "lwc1 %[in6], 96*4(%[w2]) \t\n" + "nmsub.s %[sum2], %[sum2], %[in2], %[in3] \t\n" + "lwc1 %[in1], 160*4(%[w]) \t\n" + "nmsub.s %[sum], %[sum], %[in4], %[in5] \t\n" + "lwc1 %[in2], 128*4(%[p2]) \t\n" + "nmsub.s %[sum2], %[sum2], %[in5], %[in6] \t\n" + "lwc1 %[in3], 160*4(%[w2]) \t\n" + "lwc1 %[in4], 224*4(%[w]) \t\n" + "lwc1 %[in5], 192*4(%[p2]) \t\n" + "nmsub.s %[sum], %[sum], %[in1], %[in2] \t\n" + "lwc1 %[in6], 224*4(%[w2]) \t\n" + "nmsub.s %[sum2], %[sum2], %[in2], %[in3] \t\n" + "lwc1 %[in1], 288*4(%[w]) \t\n" + "nmsub.s %[sum], %[sum], %[in4], %[in5] \t\n" + "lwc1 %[in2], 256*4(%[p2]) \t\n" + "nmsub.s %[sum2], %[sum2], %[in5], %[in6] \t\n" + "lwc1 %[in3], 288*4(%[w2]) \t\n" + "lwc1 %[in4], 352*4(%[w]) \t\n" + "lwc1 %[in5], 320*4(%[p2]) \t\n" + "nmsub.s %[sum], %[sum], %[in1], %[in2] \t\n" + "lwc1 %[in6], 352*4(%[w2]) \t\n" + "nmsub.s %[sum2], %[sum2], %[in2], %[in3] \t\n" + "lwc1 %[in1], 416*4(%[w]) \t\n" + "nmsub.s %[sum], %[sum], %[in4], %[in5] \t\n" + "lwc1 %[in2], 384*4(%[p2]) \t\n" + "nmsub.s %[sum2], %[sum2], %[in5], %[in6] \t\n" + "lwc1 %[in3], 416*4(%[w2]) \t\n" + "lwc1 %[in4], 480*4(%[w]) \t\n" + "lwc1 %[in5], 448*4(%[p2]) \t\n" + "nmsub.s %[sum], %[sum], %[in1], %[in2] \t\n" + "lwc1 %[in6], 480*4(%[w2]) \t\n" + "nmsub.s %[sum2], %[sum2], %[in2], %[in3] \t\n" + "addiu %[w], %[w], 4 \t\n" + "nmsub.s %[sum], %[sum], %[in4], %[in5] \t\n" + "addiu %[w2], %[w2], -4 \t\n" + "nmsub.s %[sum2], %[sum2], %[in5], %[in6] \t\n" + "addu %[j], %[j], 4 \t\n" + "addiu %[p], 4 \t\n" + "swc1 %[sum], 0(%[samples]) \t\n" + "addiu %[p2], -4 \t\n" + "swc1 %[sum2], 0(%[samples2]) \t\n" + "addu %[samples], %[samples], %[incr1] \t\n" + "subu %[samples2], %[samples2], %[incr1] \t\n" + "bne %[j], 64, ff_mpadsp_apply_window_loop%= \t\n" + + "lwc1 %[in1], 48*4(%[window]) \t\n" + "lwc1 %[in2], 32*4(%[synth_buf]) \t\n" + "lwc1 %[in3], 112*4(%[window]) \t\n" + "lwc1 %[in4], 96*4(%[synth_buf]) \t\n" + "lwc1 %[in5], 176*4(%[window]) \t\n" + "lwc1 %[in6], 160*4(%[synth_buf]) \t\n" + "mul.s %[sum], %[in1], %[in2] \t\n" + "lwc1 %[in7], 240*4(%[window]) \t\n" + "lwc1 %[in8], 224*4(%[synth_buf]) \t\n" + "lwc1 %[in1], 304*4(%[window]) \t\n" + "nmadd.s %[sum], %[sum], %[in3], %[in4] \t\n" + "lwc1 %[in2], 288*4(%[synth_buf]) \t\n" + "lwc1 %[in3], 368*4(%[window]) \t\n" + "lwc1 %[in4], 352*4(%[synth_buf]) \t\n" + "nmsub.s %[sum], %[sum], %[in5], %[in6] \t\n" + "nmsub.s %[sum], %[sum], %[in7], %[in8] \t\n" + "lwc1 %[in5], 432*4(%[window]) \t\n" + "lwc1 %[in6], 416*4(%[synth_buf]) \t\n" + "nmsub.s %[sum], %[sum], %[in1], %[in2] \t\n" + "lwc1 %[in7], 496*4(%[window]) \t\n" + "lwc1 %[in8], 480*4(%[synth_buf]) \t\n" + "nmsub.s %[sum], %[sum], %[in3], %[in4] \t\n" + "nmsub.s %[sum], %[sum], %[in5], %[in6] \t\n" + "nmsub.s %[sum], %[sum], %[in7], %[in8] \t\n" + "swc1 %[sum], 0(%[samples]) \t\n" + + : [sum] "=&f" (sum), [sum2] "=&f" (sum2), + [w2] "=&r" (w2), [w] "=&r" (w), + [p] "=&r" (p), [p2] "=&r" (p2), [j] "=&r" (j), + [samples] "+r" (samples), [samples2] "=&r" (samples2), + [in1] "=&f" (in1), [in2] "=&f" (in2), + [in3] "=&f" (in3), [in4] "=&f" (in4), + [in5] "=&f" (in5), [in6] "=&f" (in6), + [in7] "=&f" (in7), [in8] "=&f" (in8), + [t_sample] "=&r" (t_sample) + : [synth_buf] "r" (synth_buf), [window] "r" (window), + [dither_state] "r" (dither_state), [incr1] "r" (incr1) + ); +} + +static void ff_dct32_mips_float(float *out, const float *tab) +{ + float val0 , val1 , val2 , val3 , val4 , val5 , val6 , val7, + val8 , val9 , val10, val11, val12, val13, val14, val15, + val16, val17, val18, val19, val20, val21, val22, val23, + val24, val25, val26, val27, val28, val29, val30, val31; + float fTmp1, fTmp2, fTmp3, fTmp4, fTmp5, fTmp6, fTmp7, fTmp8, + fTmp9, fTmp10, fTmp11; + + /** + * instructions are scheduled to minimize pipeline stall. + */ + __asm__ volatile ( + "lwc1 %[fTmp1], 0*4(%[tab]) \n\t" + "lwc1 %[fTmp2], 31*4(%[tab]) \n\t" + "lwc1 %[fTmp3], 15*4(%[tab]) \n\t" + "lwc1 %[fTmp4], 16*4(%[tab]) \n\t" + "li.s %[fTmp7], 0.50241928618815570551 \n\t" + "add.s %[fTmp5], %[fTmp1], %[fTmp2] \n\t" + "sub.s %[fTmp8], %[fTmp1], %[fTmp2] \n\t" + "add.s %[fTmp6], %[fTmp3], %[fTmp4] \n\t" + "sub.s %[fTmp9], %[fTmp3], %[fTmp4] \n\t" + "li.s %[fTmp10], 0.50060299823519630134 \n\t" + "li.s %[fTmp11], 10.19000812354805681150 \n\t" + "mul.s %[fTmp8], %[fTmp8], %[fTmp10] \n\t" + "add.s %[val0], %[fTmp5], %[fTmp6] \n\t" + "sub.s %[val15], %[fTmp5], %[fTmp6] \n\t" + "lwc1 %[fTmp1], 7*4(%[tab]) \n\t" + "lwc1 %[fTmp2], 24*4(%[tab]) \n\t" + "madd.s %[val16], %[fTmp8], %[fTmp9], %[fTmp11] \n\t" + "nmsub.s %[val31], %[fTmp8], %[fTmp9], %[fTmp11] \n\t" + "mul.s %[val15], %[val15], %[fTmp7] \n\t" + "lwc1 %[fTmp3], 8*4(%[tab]) \n\t" + "lwc1 %[fTmp4], 23*4(%[tab]) \n\t" + "add.s %[fTmp5], %[fTmp1], %[fTmp2] \n\t" + "mul.s %[val31], %[val31], %[fTmp7] \n\t" + "sub.s %[fTmp8], %[fTmp1], %[fTmp2] \n\t" + "add.s %[fTmp6], %[fTmp3], %[fTmp4] \n\t" + "sub.s %[fTmp9], %[fTmp3], %[fTmp4] \n\t" + "li.s %[fTmp7], 5.10114861868916385802 \n\t" + "li.s %[fTmp10], 0.67480834145500574602 \n\t" + "li.s %[fTmp11], 0.74453627100229844977 \n\t" + "add.s %[val7], %[fTmp5], %[fTmp6] \n\t" + "sub.s %[val8], %[fTmp5], %[fTmp6] \n\t" + "mul.s %[fTmp8], %[fTmp8], %[fTmp10] \n\t" + "li.s %[fTmp1], 0.50979557910415916894 \n\t" + "sub.s %[fTmp2], %[val0], %[val7] \n\t" + "mul.s %[val8], %[val8], %[fTmp7] \n\t" + "madd.s %[val23], %[fTmp8], %[fTmp9], %[fTmp11] \n\t" + "nmsub.s %[val24], %[fTmp8], %[fTmp9], %[fTmp11] \n\t" + "add.s %[val0], %[val0], %[val7] \n\t" + "mul.s %[val7], %[fTmp1], %[fTmp2] \n\t" + "sub.s %[fTmp2], %[val15], %[val8] \n\t" + "add.s %[val8], %[val15], %[val8] \n\t" + "mul.s %[val24], %[val24], %[fTmp7] \n\t" + "sub.s %[fTmp3], %[val16], %[val23] \n\t" + "add.s %[val16], %[val16], %[val23] \n\t" + "mul.s %[val15], %[fTmp1], %[fTmp2] \n\t" + "sub.s %[fTmp4], %[val31], %[val24] \n\t" + "mul.s %[val23], %[fTmp1], %[fTmp3] \n\t" + "add.s %[val24], %[val31], %[val24] \n\t" + "mul.s %[val31], %[fTmp1], %[fTmp4] \n\t" + + : [fTmp1] "=&f" (fTmp1), [fTmp2] "=&f" (fTmp2), [fTmp3] "=&f" (fTmp3), + [fTmp4] "=&f" (fTmp4), [fTmp5] "=&f" (fTmp5), [fTmp6] "=&f" (fTmp6), + [fTmp7] "=&f" (fTmp7), [fTmp8] "=&f" (fTmp8), [fTmp9] "=&f" (fTmp9), + [fTmp10] "=&f" (fTmp10), [fTmp11] "=&f" (fTmp11), + [val0] "=f" (val0), [val7] "=f" (val7), + [val8] "=f" (val8), [val15] "=f" (val15), + [val16] "=f" (val16), [val23] "=f" (val23), + [val24] "=f" (val24), [val31] "=f" (val31) + : [tab] "r" (tab) + ); + + __asm__ volatile ( + "lwc1 %[fTmp1], 3*4(%[tab]) \n\t" + "lwc1 %[fTmp2], 28*4(%[tab]) \n\t" + "lwc1 %[fTmp3], 12*4(%[tab]) \n\t" + "lwc1 %[fTmp4], 19*4(%[tab]) \n\t" + "li.s %[fTmp7], 0.64682178335999012954 \n\t" + "add.s %[fTmp5], %[fTmp1], %[fTmp2] \n\t" + "sub.s %[fTmp8], %[fTmp1], %[fTmp2] \n\t" + "add.s %[fTmp6], %[fTmp3], %[fTmp4] \n\t" + "sub.s %[fTmp9], %[fTmp3], %[fTmp4] \n\t" + "li.s %[fTmp10], 0.53104259108978417447 \n\t" + "li.s %[fTmp11], 1.48416461631416627724 \n\t" + "mul.s %[fTmp8], %[fTmp8], %[fTmp10] \n\t" + "add.s %[val3], %[fTmp5], %[fTmp6] \n\t" + "sub.s %[val12], %[fTmp5], %[fTmp6] \n\t" + "lwc1 %[fTmp1], 4*4(%[tab]) \n\t" + "lwc1 %[fTmp2], 27*4(%[tab]) \n\t" + "madd.s %[val19], %[fTmp8], %[fTmp9], %[fTmp11] \n\t" + "nmsub.s %[val28], %[fTmp8], %[fTmp9], %[fTmp11] \n\t" + "mul.s %[val12], %[val12], %[fTmp7] \n\t" + "lwc1 %[fTmp3], 11*4(%[tab]) \n\t" + "lwc1 %[fTmp4], 20*4(%[tab]) \n\t" + "add.s %[fTmp5], %[fTmp1], %[fTmp2] \n\t" + "mul.s %[val28], %[val28], %[fTmp7] \n\t" + "sub.s %[fTmp8], %[fTmp1], %[fTmp2] \n\t" + "li.s %[fTmp7], 0.78815462345125022473 \n\t" + "add.s %[fTmp6], %[fTmp3], %[fTmp4] \n\t" + "sub.s %[fTmp9], %[fTmp3], %[fTmp4] \n\t" + "li.s %[fTmp10], 0.55310389603444452782 \n\t" + "li.s %[fTmp11], 1.16943993343288495515 \n\t" + "mul.s %[fTmp8], %[fTmp8], %[fTmp10] \n\t" + "add.s %[val4], %[fTmp5], %[fTmp6] \n\t" + "sub.s %[val11], %[fTmp5], %[fTmp6] \n\t" + "li.s %[fTmp1], 2.56291544774150617881 \n\t" + "madd.s %[val20], %[fTmp8], %[fTmp9], %[fTmp11] \n\t" + "nmsub.s %[val27], %[fTmp8], %[fTmp9], %[fTmp11] \n\t" + "mul.s %[val11], %[val11], %[fTmp7] \n\t" + "sub.s %[fTmp2], %[val3], %[val4] \n\t" + "add.s %[val3], %[val3], %[val4] \n\t" + "sub.s %[fTmp4], %[val19], %[val20] \n\t" + "mul.s %[val27], %[val27], %[fTmp7] \n\t" + "sub.s %[fTmp3], %[val12], %[val11] \n\t" + "mul.s %[val4], %[fTmp1], %[fTmp2] \n\t" + "add.s %[val11], %[val12], %[val11] \n\t" + "add.s %[val19], %[val19], %[val20] \n\t" + "mul.s %[val20], %[fTmp1], %[fTmp4] \n\t" + "mul.s %[val12], %[fTmp1], %[fTmp3] \n\t" + "sub.s %[fTmp2], %[val28], %[val27] \n\t" + "add.s %[val27], %[val28], %[val27] \n\t" + "mul.s %[val28], %[fTmp1], %[fTmp2] \n\t" + + : [fTmp1] "=&f" (fTmp1), [fTmp2] "=&f" (fTmp2), [fTmp3] "=&f" (fTmp3), + [fTmp4] "=&f" (fTmp4), [fTmp5] "=&f" (fTmp5), [fTmp6] "=&f" (fTmp6), + [fTmp7] "=&f" (fTmp7), [fTmp8] "=&f" (fTmp8), [fTmp9] "=&f" (fTmp9), + [fTmp10] "=&f" (fTmp10), [fTmp11] "=&f" (fTmp11), + [val3] "=f" (val3), [val4] "=f" (val4), + [val11] "=f" (val11), [val12] "=f" (val12), + [val19] "=f" (val19), [val20] "=f" (val20), + [val27] "=f" (val27), [val28] "=f" (val28) + : [tab] "r" (tab) + ); + + __asm__ volatile ( + "li.s %[fTmp1], 0.54119610014619698439 \n\t" + "sub.s %[fTmp2], %[val0], %[val3] \n\t" + "add.s %[val0], %[val0], %[val3] \n\t" + "sub.s %[fTmp3], %[val7], %[val4] \n\t" + "add.s %[val4], %[val7], %[val4] \n\t" + "sub.s %[fTmp4], %[val8], %[val11] \n\t" + "mul.s %[val3], %[fTmp1], %[fTmp2] \n\t" + "add.s %[val8], %[val8], %[val11] \n\t" + "mul.s %[val7], %[fTmp1], %[fTmp3] \n\t" + "sub.s %[fTmp2], %[val15], %[val12] \n\t" + "mul.s %[val11], %[fTmp1], %[fTmp4] \n\t" + "add.s %[val12], %[val15], %[val12] \n\t" + "mul.s %[val15], %[fTmp1], %[fTmp2] \n\t" + + : [val0] "+f" (val0), [val3] "+f" (val3), + [val4] "+f" (val4), [val7] "+f" (val7), + [val8] "+f" (val8), [val11] "+f" (val11), + [val12] "+f" (val12), [val15] "+f" (val15), + [fTmp1] "=f" (fTmp1), [fTmp2] "=&f" (fTmp2), + [fTmp3] "=&f" (fTmp3), [fTmp4] "=&f" (fTmp4) + : + ); + + __asm__ volatile ( + "sub.s %[fTmp2], %[val16], %[val19] \n\t" + "add.s %[val16], %[val16], %[val19] \n\t" + "sub.s %[fTmp3], %[val23], %[val20] \n\t" + "add.s %[val20], %[val23], %[val20] \n\t" + "sub.s %[fTmp4], %[val24], %[val27] \n\t" + "mul.s %[val19], %[fTmp1], %[fTmp2] \n\t" + "add.s %[val24], %[val24], %[val27] \n\t" + "mul.s %[val23], %[fTmp1], %[fTmp3] \n\t" + "sub.s %[fTmp2], %[val31], %[val28] \n\t" + "mul.s %[val27], %[fTmp1], %[fTmp4] \n\t" + "add.s %[val28], %[val31], %[val28] \n\t" + "mul.s %[val31], %[fTmp1], %[fTmp2] \n\t" + + : [fTmp2] "=&f" (fTmp2), [fTmp3] "=&f" (fTmp3), [fTmp4] "=&f" (fTmp4), + [val16] "+f" (val16), [val19] "+f" (val19), [val20] "+f" (val20), + [val23] "+f" (val23), [val24] "+f" (val24), [val27] "+f" (val27), + [val28] "+f" (val28), [val31] "+f" (val31) + : [fTmp1] "f" (fTmp1) + ); + + __asm__ volatile ( + "lwc1 %[fTmp1], 1*4(%[tab]) \n\t" + "lwc1 %[fTmp2], 30*4(%[tab]) \n\t" + "lwc1 %[fTmp3], 14*4(%[tab]) \n\t" + "lwc1 %[fTmp4], 17*4(%[tab]) \n\t" + "li.s %[fTmp7], 0.52249861493968888062 \n\t" + "add.s %[fTmp5], %[fTmp1], %[fTmp2] \n\t" + "sub.s %[fTmp8], %[fTmp1], %[fTmp2] \n\t" + "add.s %[fTmp6], %[fTmp3], %[fTmp4] \n\t" + "sub.s %[fTmp9], %[fTmp3], %[fTmp4] \n\t" + "li.s %[fTmp10], 0.50547095989754365998 \n\t" + "li.s %[fTmp11], 3.40760841846871878570 \n\t" + "mul.s %[fTmp8], %[fTmp8], %[fTmp10] \n\t" + "add.s %[val1], %[fTmp5], %[fTmp6] \n\t" + "sub.s %[val14], %[fTmp5], %[fTmp6] \n\t" + "lwc1 %[fTmp1], 6*4(%[tab]) \n\t" + "lwc1 %[fTmp2], 25*4(%[tab]) \n\t" + "madd.s %[val17], %[fTmp8], %[fTmp9], %[fTmp11] \n\t" + "nmsub.s %[val30], %[fTmp8], %[fTmp9], %[fTmp11] \n\t" + "mul.s %[val14], %[val14], %[fTmp7] \n\t" + "lwc1 %[fTmp3], 9*4(%[tab]) \n\t" + "lwc1 %[fTmp4], 22*4(%[tab]) \n\t" + "add.s %[fTmp5], %[fTmp1], %[fTmp2] \n\t" + "mul.s %[val30], %[val30], %[fTmp7] \n\t" + "sub.s %[fTmp8], %[fTmp1], %[fTmp2] \n\t" + "add.s %[fTmp6], %[fTmp3], %[fTmp4] \n\t" + "sub.s %[fTmp9], %[fTmp3], %[fTmp4] \n\t" + "li.s %[fTmp7], 1.72244709823833392782 \n\t" + "li.s %[fTmp10], 0.62250412303566481615 \n\t" + "li.s %[fTmp11], 0.83934964541552703873 \n\t" + "add.s %[val6], %[fTmp5], %[fTmp6] \n\t" + "sub.s %[val9], %[fTmp5], %[fTmp6] \n\t" + "mul.s %[fTmp8], %[fTmp8], %[fTmp10] \n\t" + "li.s %[fTmp1], 0.60134488693504528054 \n\t" + "sub.s %[fTmp2], %[val1], %[val6] \n\t" + "add.s %[val1], %[val1], %[val6] \n\t" + "mul.s %[val9], %[val9], %[fTmp7] \n\t" + "madd.s %[val22], %[fTmp8], %[fTmp9], %[fTmp11] \n\t" + "nmsub.s %[val25], %[fTmp8], %[fTmp9], %[fTmp11] \n\t" + "mul.s %[val6], %[fTmp1], %[fTmp2] \n\t" + "sub.s %[fTmp2], %[val14], %[val9] \n\t" + "add.s %[val9], %[val14], %[val9] \n\t" + "mul.s %[val25], %[val25], %[fTmp7] \n\t" + "sub.s %[fTmp3], %[val17], %[val22] \n\t" + "add.s %[val17], %[val17], %[val22] \n\t" + "mul.s %[val14], %[fTmp1], %[fTmp2] \n\t" + "sub.s %[fTmp2], %[val30], %[val25] \n\t" + "mul.s %[val22], %[fTmp1], %[fTmp3] \n\t" + "add.s %[val25], %[val30], %[val25] \n\t" + "mul.s %[val30], %[fTmp1], %[fTmp2] \n\t" + + : [fTmp1] "=&f" (fTmp1), [fTmp2] "=&f" (fTmp2), [fTmp3] "=&f" (fTmp3), + [fTmp4] "=&f" (fTmp4), [fTmp5] "=&f" (fTmp5), [fTmp6] "=&f" (fTmp6), + [fTmp7] "=&f" (fTmp7), [fTmp8] "=&f" (fTmp8), [fTmp9] "=&f" (fTmp9), + [fTmp10] "=&f" (fTmp10), [fTmp11] "=&f" (fTmp11), + [val1] "=f" (val1), [val6] "=f" (val6), + [val9] "=f" (val9), [val14] "=f" (val14), + [val17] "=f" (val17), [val22] "=f" (val22), + [val25] "=f" (val25), [val30] "=f" (val30) + : [tab] "r" (tab) + ); + + __asm__ volatile ( + "lwc1 %[fTmp1], 2*4(%[tab]) \n\t" + "lwc1 %[fTmp2], 29*4(%[tab]) \n\t" + "lwc1 %[fTmp3], 13*4(%[tab]) \n\t" + "lwc1 %[fTmp4], 18*4(%[tab]) \n\t" + "li.s %[fTmp7], 0.56694403481635770368 \n\t" + "add.s %[fTmp5], %[fTmp1], %[fTmp2] \n\t" + "sub.s %[fTmp8], %[fTmp1], %[fTmp2] \n\t" + "add.s %[fTmp6], %[fTmp3], %[fTmp4] \n\t" + "sub.s %[fTmp9], %[fTmp3], %[fTmp4] \n\t" + "li.s %[fTmp10], 0.51544730992262454697 \n\t" + "li.s %[fTmp11], 2.05778100995341155085 \n\t" + "mul.s %[fTmp8], %[fTmp8], %[fTmp10] \n\t" + "add.s %[val2], %[fTmp5], %[fTmp6] \n\t" + "sub.s %[val13], %[fTmp5], %[fTmp6] \n\t" + "lwc1 %[fTmp1], 5*4(%[tab]) \n\t" + "lwc1 %[fTmp2], 26*4(%[tab]) \n\t" + "madd.s %[val18], %[fTmp8], %[fTmp9], %[fTmp11] \n\t" + "nmsub.s %[val29], %[fTmp8], %[fTmp9], %[fTmp11] \n\t" + "mul.s %[val13], %[val13], %[fTmp7] \n\t" + "lwc1 %[fTmp3], 10*4(%[tab]) \n\t" + "lwc1 %[fTmp4], 21*4(%[tab]) \n\t" + "mul.s %[val29], %[val29], %[fTmp7] \n\t" + "add.s %[fTmp5], %[fTmp1], %[fTmp2] \n\t" + "sub.s %[fTmp8], %[fTmp1], %[fTmp2] \n\t" + "add.s %[fTmp6], %[fTmp3], %[fTmp4] \n\t" + "sub.s %[fTmp9], %[fTmp3], %[fTmp4] \n\t" + "li.s %[fTmp7], 1.06067768599034747134 \n\t" + "li.s %[fTmp10], 0.58293496820613387367 \n\t" + "li.s %[fTmp11], 0.97256823786196069369 \n\t" + "add.s %[val5], %[fTmp5], %[fTmp6] \n\t" + "sub.s %[val10], %[fTmp5], %[fTmp6] \n\t" + "mul.s %[fTmp8], %[fTmp8], %[fTmp10] \n\t" + "li.s %[fTmp1], 0.89997622313641570463 \n\t" + "sub.s %[fTmp2], %[val2], %[val5] \n\t" + "mul.s %[val10], %[val10], %[fTmp7] \n\t" + "madd.s %[val21], %[fTmp8], %[fTmp9], %[fTmp11] \n\t" + "nmsub.s %[val26], %[fTmp8], %[fTmp9], %[fTmp11] \n\t" + "add.s %[val2], %[val2], %[val5] \n\t" + "mul.s %[val5], %[fTmp1], %[fTmp2] \n\t" + "sub.s %[fTmp3], %[val13], %[val10] \n\t" + "add.s %[val10], %[val13], %[val10] \n\t" + "mul.s %[val26], %[val26], %[fTmp7] \n\t" + "sub.s %[fTmp4], %[val18], %[val21] \n\t" + "add.s %[val18], %[val18], %[val21] \n\t" + "mul.s %[val13], %[fTmp1], %[fTmp3] \n\t" + "sub.s %[fTmp2], %[val29], %[val26] \n\t" + "add.s %[val26], %[val29], %[val26] \n\t" + "mul.s %[val21], %[fTmp1], %[fTmp4] \n\t" + "mul.s %[val29], %[fTmp1], %[fTmp2] \n\t" + + : [fTmp1] "=&f" (fTmp1), [fTmp2] "=&f" (fTmp2), [fTmp3] "=&f" (fTmp3), + [fTmp4] "=&f" (fTmp4), [fTmp5] "=&f" (fTmp5), [fTmp6] "=&f" (fTmp6), + [fTmp7] "=&f" (fTmp7), [fTmp8] "=&f" (fTmp8), [fTmp9] "=&f" (fTmp9), + [fTmp10] "=&f" (fTmp10), [fTmp11] "=&f" (fTmp11), + [val2] "=f" (val2), [val5] "=f" (val5), + [val10] "=f" (val10), [val13] "=f" (val13), + [val18] "=f" (val18), [val21] "=f" (val21), + [val26] "=f" (val26), [val29] "=f" (val29) + : [tab] "r" (tab) + ); + + __asm__ volatile ( + "li.s %[fTmp1], 1.30656296487637652785 \n\t" + "sub.s %[fTmp2], %[val1], %[val2] \n\t" + "add.s %[val1], %[val1], %[val2] \n\t" + "sub.s %[fTmp3], %[val6], %[val5] \n\t" + "add.s %[val5], %[val6], %[val5] \n\t" + "sub.s %[fTmp4], %[val9], %[val10] \n\t" + "mul.s %[val2], %[fTmp1], %[fTmp2] \n\t" + "add.s %[val9], %[val9], %[val10] \n\t" + "mul.s %[val6], %[fTmp1], %[fTmp3] \n\t" + "sub.s %[fTmp2], %[val14], %[val13] \n\t" + "mul.s %[val10], %[fTmp1], %[fTmp4] \n\t" + "add.s %[val13], %[val14], %[val13] \n\t" + "mul.s %[val14], %[fTmp1], %[fTmp2] \n\t" + + : [fTmp1] "=f" (fTmp1), [fTmp2] "=&f" (fTmp2), + [fTmp3] "=&f" (fTmp3), [fTmp4] "=&f" (fTmp4), + [val1] "+f" (val1), [val2] "+f" (val2), + [val5] "+f" (val5), [val6] "+f" (val6), + [val9] "+f" (val9), [val10] "+f" (val10), + [val13] "+f" (val13), [val14] "+f" (val14) + : + ); + + __asm__ volatile ( + "sub.s %[fTmp2], %[val17], %[val18] \n\t" + "add.s %[val17], %[val17], %[val18] \n\t" + "sub.s %[fTmp3], %[val22], %[val21] \n\t" + "add.s %[val21], %[val22], %[val21] \n\t" + "sub.s %[fTmp4], %[val25], %[val26] \n\t" + "mul.s %[val18], %[fTmp1], %[fTmp2] \n\t" + "add.s %[val25], %[val25], %[val26] \n\t" + "mul.s %[val22], %[fTmp1], %[fTmp3] \n\t" + "sub.s %[fTmp2], %[val30], %[val29] \n\t" + "mul.s %[val26], %[fTmp1], %[fTmp4] \n\t" + "add.s %[val29], %[val30], %[val29] \n\t" + "mul.s %[val30], %[fTmp1], %[fTmp2] \n\t" + + : [fTmp2] "=&f" (fTmp2), [fTmp3] "=&f" (fTmp3), [fTmp4] "=&f" (fTmp4), + [val17] "+f" (val17), [val18] "+f" (val18), [val21] "+f" (val21), + [val22] "+f" (val22), [val25] "+f" (val25), [val26] "+f" (val26), + [val29] "+f" (val29), [val30] "+f" (val30) + : [fTmp1] "f" (fTmp1) + ); + + __asm__ volatile ( + "li.s %[fTmp1], 0.70710678118654752439 \n\t" + "sub.s %[fTmp2], %[val0], %[val1] \n\t" + "add.s %[val0], %[val0], %[val1] \n\t" + "sub.s %[fTmp3], %[val3], %[val2] \n\t" + "add.s %[val2], %[val3], %[val2] \n\t" + "sub.s %[fTmp4], %[val4], %[val5] \n\t" + "mul.s %[val1], %[fTmp1], %[fTmp2] \n\t" + "swc1 %[val0], 0(%[out]) \n\t" + "mul.s %[val3], %[fTmp3], %[fTmp1] \n\t" + "add.s %[val4], %[val4], %[val5] \n\t" + "mul.s %[val5], %[fTmp1], %[fTmp4] \n\t" + "swc1 %[val1], 16*4(%[out]) \n\t" + "sub.s %[fTmp2], %[val7], %[val6] \n\t" + "add.s %[val2], %[val2], %[val3] \n\t" + "swc1 %[val3], 24*4(%[out]) \n\t" + "add.s %[val6], %[val7], %[val6] \n\t" + "mul.s %[val7], %[fTmp1], %[fTmp2] \n\t" + "swc1 %[val2], 8*4(%[out]) \n\t" + "add.s %[val6], %[val6], %[val7] \n\t" + "swc1 %[val7], 28*4(%[out]) \n\t" + "add.s %[val4], %[val4], %[val6] \n\t" + "add.s %[val6], %[val6], %[val5] \n\t" + "add.s %[val5], %[val5], %[val7] \n\t" + "swc1 %[val4], 4*4(%[out]) \n\t" + "swc1 %[val5], 20*4(%[out]) \n\t" + "swc1 %[val6], 12*4(%[out]) \n\t" + + : [fTmp1] "=f" (fTmp1), [fTmp2] "=&f" (fTmp2), + [fTmp3] "=&f" (fTmp3), [fTmp4] "=&f" (fTmp4), + [val0] "+f" (val0), [val1] "+f" (val1), + [val2] "+f" (val2), [val3] "+f" (val3), + [val4] "+f" (val4), [val5] "+f" (val5), + [val6] "+f" (val6), [val7] "+f" (val7) + : [out] "r" (out) + ); + + __asm__ volatile ( + "sub.s %[fTmp2], %[val8], %[val9] \n\t" + "add.s %[val8], %[val8], %[val9] \n\t" + "sub.s %[fTmp3], %[val11], %[val10] \n\t" + "add.s %[val10], %[val11], %[val10] \n\t" + "sub.s %[fTmp4], %[val12], %[val13] \n\t" + "mul.s %[val9], %[fTmp1], %[fTmp2] \n\t" + "add.s %[val12], %[val12], %[val13] \n\t" + "mul.s %[val11], %[fTmp1], %[fTmp3] \n\t" + "sub.s %[fTmp2], %[val15], %[val14] \n\t" + "mul.s %[val13], %[fTmp1], %[fTmp4] \n\t" + "add.s %[val14], %[val15], %[val14] \n\t" + "add.s %[val10], %[val10], %[val11] \n\t" + "mul.s %[val15], %[fTmp1], %[fTmp2] \n\t" + "add.s %[val14], %[val14], %[val15] \n\t" + "add.s %[val12], %[val12], %[val14] \n\t" + "add.s %[val14], %[val14], %[val13] \n\t" + "add.s %[val13], %[val13], %[val15] \n\t" + "add.s %[val8], %[val8], %[val12] \n\t" + "add.s %[val12], %[val12], %[val10] \n\t" + "add.s %[val10], %[val10], %[val14] \n\t" + "add.s %[val14], %[val14], %[val9] \n\t" + "add.s %[val9], %[val9], %[val13] \n\t" + "add.s %[val13], %[val13], %[val11] \n\t" + "add.s %[val11], %[val11], %[val15] \n\t" + "swc1 %[val8], 2*4(%[out]) \n\t" + "swc1 %[val9], 18*4(%[out]) \n\t" + "swc1 %[val10], 10*4(%[out]) \n\t" + "swc1 %[val11], 26*4(%[out]) \n\t" + "swc1 %[val12], 6*4(%[out]) \n\t" + "swc1 %[val13], 22*4(%[out]) \n\t" + "swc1 %[val14], 14*4(%[out]) \n\t" + "swc1 %[val15], 30*4(%[out]) \n\t" + + : [fTmp2] "=&f" (fTmp2), [fTmp3] "=&f" (fTmp3), [fTmp4] "=&f" (fTmp4), + [val8] "+f" (val8), [val9] "+f" (val9), [val10] "+f" (val10), + [val11] "+f" (val11), [val12] "+f" (val12), [val13] "+f" (val13), + [val14] "+f" (val14), [val15] "+f" (val15) + : [fTmp1] "f" (fTmp1), [out] "r" (out) + ); + + __asm__ volatile ( + "sub.s %[fTmp2], %[val16], %[val17] \n\t" + "add.s %[val16], %[val16], %[val17] \n\t" + "sub.s %[fTmp3], %[val19], %[val18] \n\t" + "add.s %[val18], %[val19], %[val18] \n\t" + "sub.s %[fTmp4], %[val20], %[val21] \n\t" + "mul.s %[val17], %[fTmp1], %[fTmp2] \n\t" + "add.s %[val20], %[val20], %[val21] \n\t" + "mul.s %[val19], %[fTmp1], %[fTmp3] \n\t" + "sub.s %[fTmp2], %[val23], %[val22] \n\t" + "mul.s %[val21], %[fTmp1], %[fTmp4] \n\t" + "add.s %[val22], %[val23], %[val22] \n\t" + "add.s %[val18], %[val18], %[val19] \n\t" + "mul.s %[val23], %[fTmp1], %[fTmp2] \n\t" + "add.s %[val22], %[val22], %[val23] \n\t" + "add.s %[val20], %[val20], %[val22] \n\t" + "add.s %[val22], %[val22], %[val21] \n\t" + "add.s %[val21], %[val21], %[val23] \n\t" + + : [fTmp2] "=&f" (fTmp2), [fTmp3] "=&f" (fTmp3), [fTmp4] "=&f" (fTmp4), + [val16] "+f" (val16), [val17] "+f" (val17), [val18] "+f" (val18), + [val19] "+f" (val19), [val20] "+f" (val20), [val21] "+f" (val21), + [val22] "+f" (val22), [val23] "+f" (val23) + : [fTmp1] "f" (fTmp1) + ); + + __asm__ volatile ( + "sub.s %[fTmp2], %[val24], %[val25] \n\t" + "add.s %[val24], %[val24], %[val25] \n\t" + "sub.s %[fTmp3], %[val27], %[val26] \n\t" + "add.s %[val26], %[val27], %[val26] \n\t" + "sub.s %[fTmp4], %[val28], %[val29] \n\t" + "mul.s %[val25], %[fTmp1], %[fTmp2] \n\t" + "add.s %[val28], %[val28], %[val29] \n\t" + "mul.s %[val27], %[fTmp1], %[fTmp3] \n\t" + "sub.s %[fTmp2], %[val31], %[val30] \n\t" + "mul.s %[val29], %[fTmp1], %[fTmp4] \n\t" + "add.s %[val30], %[val31], %[val30] \n\t" + "add.s %[val26], %[val26], %[val27] \n\t" + "mul.s %[val31], %[fTmp1], %[fTmp2] \n\t" + "add.s %[val30], %[val30], %[val31] \n\t" + "add.s %[val28], %[val28], %[val30] \n\t" + "add.s %[val30], %[val30], %[val29] \n\t" + "add.s %[val29], %[val29], %[val31] \n\t" + "add.s %[val24], %[val24], %[val28] \n\t" + "add.s %[val28], %[val28], %[val26] \n\t" + "add.s %[val26], %[val26], %[val30] \n\t" + "add.s %[val30], %[val30], %[val25] \n\t" + "add.s %[val25], %[val25], %[val29] \n\t" + "add.s %[val29], %[val29], %[val27] \n\t" + "add.s %[val27], %[val27], %[val31] \n\t" + + : [fTmp2] "=&f" (fTmp2), [fTmp3] "=&f" (fTmp3), [fTmp4] "=&f" (fTmp4), + [val24] "+f" (val24), [val25] "+f" (val25), [val26] "+f" (val26), + [val27] "+f" (val27), [val28] "+f" (val28), [val29] "+f" (val29), + [val30] "+f" (val30), [val31] "+f" (val31) + : [fTmp1] "f" (fTmp1) + ); + + out[ 1] = val16 + val24; + out[17] = val17 + val25; + out[ 9] = val18 + val26; + out[25] = val19 + val27; + out[ 5] = val20 + val28; + out[21] = val21 + val29; + out[13] = val22 + val30; + out[29] = val23 + val31; + out[ 3] = val24 + val20; + out[19] = val25 + val21; + out[11] = val26 + val22; + out[27] = val27 + val23; + out[ 7] = val28 + val18; + out[23] = val29 + val19; + out[15] = val30 + val17; + out[31] = val31; +} + +static void imdct36_mips_float(float *out, float *buf, float *in, float *win) +{ + float t0, t1, t2, t3, s0, s1, s2, s3; + float tmp[18]; + /* temporary variables */ + float in1, in2, in3, in4, in5, in6; + float out1, out2, out3, out4, out5; + float c1, c2, c3, c4, c5, c6, c7, c8, c9; + + /** + * all loops are unrolled totally, and instructions are scheduled to + * minimize pipeline stall. instructions of the first two loops are + * reorganized, in order to eliminate unnecessary readings and + * writings into array. values defined in macros and tables are + * eliminated - they are directly loaded in appropriate variables + */ + + /* loop 1 and 2 */ + __asm__ volatile ( + "lwc1 %[in1], 17*4(%[in]) \t\n" + "lwc1 %[in2], 16*4(%[in]) \t\n" + "lwc1 %[in3], 15*4(%[in]) \t\n" + "lwc1 %[in4], 14*4(%[in]) \t\n" + "lwc1 %[in5], 13*4(%[in]) \t\n" + "lwc1 %[in6], 12*4(%[in]) \t\n" + "add.s %[out1], %[in1], %[in2] \t\n" + "add.s %[out2], %[in2], %[in3] \t\n" + "add.s %[out3], %[in3], %[in4] \t\n" + "add.s %[out4], %[in4], %[in5] \t\n" + "add.s %[out5], %[in5], %[in6] \t\n" + "lwc1 %[in1], 11*4(%[in]) \t\n" + "swc1 %[out2], 16*4(%[in]) \t\n" + "add.s %[out1], %[out1], %[out3] \t\n" + "swc1 %[out4], 14*4(%[in]) \t\n" + "add.s %[out3], %[out3], %[out5] \t\n" + "lwc1 %[in2], 10*4(%[in]) \t\n" + "lwc1 %[in3], 9*4(%[in]) \t\n" + "swc1 %[out1], 17*4(%[in]) \t\n" + "lwc1 %[in4], 8*4(%[in]) \t\n" + "swc1 %[out3], 15*4(%[in]) \t\n" + "add.s %[out1], %[in6], %[in1] \t\n" + "add.s %[out2], %[in1], %[in2] \t\n" + "add.s %[out3], %[in2], %[in3] \t\n" + "add.s %[out4], %[in3], %[in4] \t\n" + "lwc1 %[in5], 7*4(%[in]) \t\n" + "swc1 %[out1], 12*4(%[in]) \t\n" + "add.s %[out5], %[out5], %[out2] \t\n" + "swc1 %[out3], 10*4(%[in]) \t\n" + "add.s %[out2], %[out2], %[out4] \t\n" + "lwc1 %[in6], 6*4(%[in]) \t\n" + "lwc1 %[in1], 5*4(%[in]) \t\n" + "swc1 %[out5], 13*4(%[in]) \t\n" + "lwc1 %[in2], 4*4(%[in]) \t\n" + "swc1 %[out2], 11*4(%[in]) \t\n" + "add.s %[out5], %[in4], %[in5] \t\n" + "add.s %[out1], %[in5], %[in6] \t\n" + "add.s %[out2], %[in6], %[in1] \t\n" + "add.s %[out3], %[in1], %[in2] \t\n" + "lwc1 %[in3], 3*4(%[in]) \t\n" + "swc1 %[out5], 8*4(%[in]) \t\n" + "add.s %[out4], %[out4], %[out1] \t\n" + "swc1 %[out2], 6*4(%[in]) \t\n" + "add.s %[out1], %[out1], %[out3] \t\n" + "lwc1 %[in4], 2*4(%[in]) \t\n" + "lwc1 %[in5], 1*4(%[in]) \t\n" + "swc1 %[out4], 9*4(%[in]) \t\n" + "lwc1 %[in6], 0(%[in]) \t\n" + "swc1 %[out1], 7*4(%[in]) \t\n" + "add.s %[out4], %[in2], %[in3] \t\n" + "add.s %[out5], %[in3], %[in4] \t\n" + "add.s %[out1], %[in4], %[in5] \t\n" + "add.s %[out2], %[in5], %[in6] \t\n" + "swc1 %[out4], 4*4(%[in]) \t\n" + "add.s %[out3], %[out3], %[out5] \t\n" + "swc1 %[out1], 2*4(%[in]) \t\n" + "add.s %[out5], %[out5], %[out2] \t\n" + "swc1 %[out2], 1*4(%[in]) \t\n" + "swc1 %[out3], 5*4(%[in]) \t\n" + "swc1 %[out5], 3*4(%[in]) \t\n" + + : [in1] "=&f" (in1), [in2] "=&f" (in2), + [in3] "=&f" (in3), [in4] "=&f" (in4), + [in5] "=&f" (in5), [in6] "=&f" (in6), + [out1] "=&f" (out1), [out2] "=&f" (out2), + [out3] "=&f" (out3), [out4] "=&f" (out4), + [out5] "=&f" (out5) + : [in] "r" (in) + ); + + /* loop 3 */ + __asm__ volatile ( + "li.s %[c1], 0.5 \t\n" + "lwc1 %[in1], 8*4(%[in]) \t\n" + "lwc1 %[in2], 16*4(%[in]) \t\n" + "lwc1 %[in3], 4*4(%[in]) \t\n" + "lwc1 %[in4], 0(%[in]) \t\n" + "lwc1 %[in5], 12*4(%[in]) \t\n" + "li.s %[c2], 0.93969262078590838405 \t\n" + "add.s %[t2], %[in1], %[in2] \t\n" + "add.s %[t0], %[in1], %[in3] \t\n" + "li.s %[c3], -0.76604444311897803520 \t\n" + "madd.s %[t3], %[in4], %[in5], %[c1] \t\n" + "sub.s %[t1], %[in4], %[in5] \t\n" + "sub.s %[t2], %[t2], %[in3] \t\n" + "mul.s %[t0], %[t0], %[c2] \t\n" + "li.s %[c4], -0.17364817766693034885 \t\n" + "li.s %[c5], -0.86602540378443864676 \t\n" + "li.s %[c6], 0.98480775301220805936 \t\n" + "nmsub.s %[out1], %[t1], %[t2], %[c1] \t\n" + "add.s %[out2], %[t1], %[t2] \t\n" + "add.s %[t2], %[in2], %[in3] \t\n" + "sub.s %[t1], %[in1], %[in2] \t\n" + "sub.s %[out3], %[t3], %[t0] \t\n" + "swc1 %[out1], 6*4(%[tmp]) \t\n" + "swc1 %[out2], 16*4(%[tmp]) \t\n" + "mul.s %[t2], %[t2], %[c3] \t\n" + "mul.s %[t1], %[t1], %[c4] \t\n" + "add.s %[out1], %[t3], %[t0] \t\n" + "lwc1 %[in1], 10*4(%[in]) \t\n" + "lwc1 %[in2], 14*4(%[in]) \t\n" + "sub.s %[out3], %[out3], %[t2] \t\n" + "add.s %[out2], %[t3], %[t2] \t\n" + "add.s %[out1], %[out1], %[t1] \t\n" + "lwc1 %[in3], 2*4(%[in]) \t\n" + "lwc1 %[in4], 6*4(%[in]) \t\n" + "swc1 %[out3], 10*4(%[tmp]) \t\n" + "sub.s %[out2], %[out2], %[t1] \t\n" + "swc1 %[out1], 2*4(%[tmp]) \t\n" + "add.s %[out1], %[in1], %[in2] \t\n" + "add.s %[t2], %[in1], %[in3] \t\n" + "sub.s %[t3], %[in1], %[in2] \t\n" + "swc1 %[out2], 14*4(%[tmp]) \t\n" + "li.s %[c7], -0.34202014332566873304 \t\n" + "sub.s %[out1], %[out1], %[in3] \t\n" + "mul.s %[t2], %[t2], %[c6] \t\n" + "mul.s %[t3], %[t3], %[c7] \t\n" + "li.s %[c8], 0.86602540378443864676 \t\n" + "mul.s %[t0], %[in4], %[c8] \t\n" + "mul.s %[out1], %[out1], %[c5] \t\n" + "add.s %[t1], %[in2], %[in3] \t\n" + "li.s %[c9], -0.64278760968653932632 \t\n" + "add.s %[out2], %[t2], %[t3] \t\n" + "lwc1 %[in1], 9*4(%[in]) \t\n" + "swc1 %[out1], 4*4(%[tmp]) \t\n" + "mul.s %[t1], %[t1], %[c9] \t\n" + "lwc1 %[in2], 17*4(%[in]) \t\n" + "add.s %[out2], %[out2], %[t0] \t\n" + "lwc1 %[in3], 5*4(%[in]) \t\n" + "lwc1 %[in4], 1*4(%[in]) \t\n" + "add.s %[out3], %[t2], %[t1] \t\n" + "sub.s %[out1], %[t3], %[t1] \t\n" + "swc1 %[out2], 0(%[tmp]) \t\n" + "lwc1 %[in5], 13*4(%[in]) \t\n" + "add.s %[t2], %[in1], %[in2] \t\n" + "sub.s %[out3], %[out3], %[t0] \t\n" + "sub.s %[out1], %[out1], %[t0] \t\n" + "add.s %[t0], %[in1], %[in3] \t\n" + "madd.s %[t3], %[in4], %[in5], %[c1] \t\n" + "sub.s %[t2], %[t2], %[in3] \t\n" + "swc1 %[out3], 12*4(%[tmp]) \t\n" + "swc1 %[out1], 8*4(%[tmp]) \t\n" + "sub.s %[t1], %[in4], %[in5] \t\n" + "mul.s %[t0], %[t0], %[c2] \t\n" + "nmsub.s %[out1], %[t1], %[t2], %[c1] \t\n" + "add.s %[out2], %[t1], %[t2] \t\n" + "add.s %[t2], %[in2], %[in3] \t\n" + "sub.s %[t1], %[in1], %[in2] \t\n" + "sub.s %[out3], %[t3], %[t0] \t\n" + "swc1 %[out1], 7*4(%[tmp]) \t\n" + "swc1 %[out2], 17*4(%[tmp]) \t\n" + "mul.s %[t2], %[t2], %[c3] \t\n" + "mul.s %[t1], %[t1], %[c4] \t\n" + "add.s %[out1], %[t3], %[t0] \t\n" + "lwc1 %[in1], 11*4(%[in]) \t\n" + "lwc1 %[in2], 15*4(%[in]) \t\n" + "sub.s %[out3], %[out3], %[t2] \t\n" + "add.s %[out2], %[t3], %[t2] \t\n" + "add.s %[out1], %[out1], %[t1] \t\n" + "lwc1 %[in3], 3*4(%[in]) \t\n" + "lwc1 %[in4], 7*4(%[in]) \t\n" + "swc1 %[out3], 11*4(%[tmp]) \t\n" + "sub.s %[out2], %[out2], %[t1] \t\n" + "swc1 %[out1], 3*4(%[tmp]) \t\n" + "add.s %[out3], %[in1], %[in2] \t\n" + "add.s %[t2], %[in1], %[in3] \t\n" + "sub.s %[t3], %[in1], %[in2] \t\n" + "swc1 %[out2], 15*4(%[tmp]) \t\n" + "mul.s %[t0], %[in4], %[c8] \t\n" + "sub.s %[out3], %[out3], %[in3] \t\n" + "mul.s %[t2], %[t2], %[c6] \t\n" + "mul.s %[t3], %[t3], %[c7] \t\n" + "add.s %[t1], %[in2], %[in3] \t\n" + "mul.s %[out3], %[out3], %[c5] \t\n" + "add.s %[out1], %[t2], %[t3] \t\n" + "mul.s %[t1], %[t1], %[c9] \t\n" + "swc1 %[out3], 5*4(%[tmp]) \t\n" + "add.s %[out1], %[out1], %[t0] \t\n" + "add.s %[out2], %[t2], %[t1] \t\n" + "sub.s %[out3], %[t3], %[t1] \t\n" + "swc1 %[out1], 1*4(%[tmp]) \t\n" + "sub.s %[out2], %[out2], %[t0] \t\n" + "sub.s %[out3], %[out3], %[t0] \t\n" + "swc1 %[out2], 13*4(%[tmp]) \t\n" + "swc1 %[out3], 9*4(%[tmp]) \t\n" + + : [t0] "=&f" (t0), [t1] "=&f" (t1), + [t2] "=&f" (t2), [t3] "=&f" (t3), + [in1] "=&f" (in1), [in2] "=&f" (in2), + [in3] "=&f" (in3), [in4] "=&f" (in4), + [in5] "=&f" (in5), + [out1] "=&f" (out1), [out2] "=&f" (out2), + [out3] "=&f" (out3), + [c1] "=&f" (c1), [c2] "=&f" (c2), + [c3] "=&f" (c3), [c4] "=&f" (c4), + [c5] "=&f" (c5), [c6] "=&f" (c6), + [c7] "=&f" (c7), [c8] "=&f" (c8), + [c9] "=&f" (c9) + : [in] "r" (in), [tmp] "r" (tmp) + ); + + /* loop 4 */ + __asm__ volatile ( + "lwc1 %[in1], 2*4(%[tmp]) \t\n" + "lwc1 %[in2], 0(%[tmp]) \t\n" + "lwc1 %[in3], 3*4(%[tmp]) \t\n" + "lwc1 %[in4], 1*4(%[tmp]) \t\n" + "li.s %[c1], 0.50190991877167369479 \t\n" + "li.s %[c2], 5.73685662283492756461 \t\n" + "add.s %[s0], %[in1], %[in2] \t\n" + "sub.s %[s2], %[in1], %[in2] \t\n" + "add.s %[s1], %[in3], %[in4] \t\n" + "sub.s %[s3], %[in3], %[in4] \t\n" + "lwc1 %[in1], 9*4(%[win]) \t\n" + "lwc1 %[in2], 4*9*4(%[buf]) \t\n" + "lwc1 %[in3], 8*4(%[win]) \t\n" + "mul.s %[s1], %[s1], %[c1] \t\n" + "mul.s %[s3], %[s3], %[c2] \t\n" + "lwc1 %[in4], 4*8*4(%[buf]) \t\n" + "lwc1 %[in5], 29*4(%[win]) \t\n" + "lwc1 %[in6], 28*4(%[win]) \t\n" + "add.s %[t0], %[s0], %[s1] \t\n" + "sub.s %[t1], %[s0], %[s1] \t\n" + "li.s %[c1], 0.51763809020504152469 \t\n" + "li.s %[c2], 1.93185165257813657349 \t\n" + "mul.s %[out3], %[in5], %[t0] \t\n" + "madd.s %[out1], %[in2], %[in1], %[t1] \t\n" + "madd.s %[out2], %[in4], %[in3], %[t1] \t\n" + "mul.s %[out4], %[in6], %[t0] \t\n" + "add.s %[t0], %[s2], %[s3] \t\n" + "swc1 %[out3], 4*9*4(%[buf]) \t\n" + "swc1 %[out1], 288*4(%[out]) \t\n" + "swc1 %[out2], 256*4(%[out]) \t\n" + "swc1 %[out4], 4*8*4(%[buf]) \t\n" + "sub.s %[t1], %[s2], %[s3] \t\n" + "lwc1 %[in1], 17*4(%[win]) \t\n" + "lwc1 %[in2], 4*17*4(%[buf]) \t\n" + "lwc1 %[in3], 0(%[win]) \t\n" + "lwc1 %[in4], 0(%[buf]) \t\n" + "lwc1 %[in5], 37*4(%[win]) \t\n" + "lwc1 %[in6], 20*4(%[win]) \t\n" + "madd.s %[out1], %[in2], %[in1], %[t1] \t\n" + "lwc1 %[in1], 6*4(%[tmp]) \t\n" + "madd.s %[out2], %[in4], %[in3], %[t1] \t\n" + "mul.s %[out3], %[t0], %[in5] \t\n" + "mul.s %[out4], %[t0], %[in6] \t\n" + "swc1 %[out1], 544*4(%[out]) \t\n" + "lwc1 %[in2], 4*4(%[tmp]) \t\n" + "swc1 %[out2], 0(%[out]) \t\n" + "swc1 %[out3], 4*17*4(%[buf]) \t\n" + "swc1 %[out4], 0(%[buf]) \t\n" + "lwc1 %[in3], 7*4(%[tmp]) \t\n" + "add.s %[s0], %[in1], %[in2] \t\n" + "sub.s %[s2], %[in1], %[in2] \t\n" + "lwc1 %[in4], 5*4(%[tmp]) \t\n" + "add.s %[s1], %[in3], %[in4] \t\n" + "sub.s %[s3], %[in3], %[in4] \t\n" + "lwc1 %[in1], 10*4(%[win]) \t\n" + "lwc1 %[in2], 4*10*4(%[buf]) \t\n" + "lwc1 %[in3], 7*4(%[win]) \t\n" + "mul.s %[s1], %[s1], %[c1] \t\n" + "mul.s %[s3], %[s3], %[c2] \t\n" + "add.s %[t0], %[s0], %[s1] \t\n" + "sub.s %[t1], %[s0], %[s1] \t\n" + "lwc1 %[in4], 4*7*4(%[buf]) \t\n" + "lwc1 %[in5], 30*4(%[win]) \t\n" + "lwc1 %[in6], 27*4(%[win]) \t\n" + "li.s %[c1], 0.55168895948124587824 \t\n" + "madd.s %[out1], %[in2], %[in1], %[t1] \t\n" + "madd.s %[out2], %[in4], %[in3], %[t1] \t\n" + "mul.s %[out3], %[t0], %[in5] \t\n" + "mul.s %[out4], %[t0], %[in6] \t\n" + "add.s %[t0], %[s2], %[s3] \t\n" + "swc1 %[out1], 320*4(%[out]) \t\n" + "swc1 %[out2], 224*4(%[out]) \t\n" + "swc1 %[out3], 4*10*4(%[buf]) \t\n" + "swc1 %[out4], 4*7*4(%[buf]) \t\n" + "sub.s %[t1], %[s2], %[s3] \t\n" + "lwc1 %[in1], 16*4(%[win]) \t\n" + "lwc1 %[in2], 4*16*4(%[buf]) \t\n" + "lwc1 %[in3], 1*4(%[win]) \t\n" + "lwc1 %[in4], 4*1*4(%[buf]) \t\n" + "lwc1 %[in5], 36*4(%[win]) \t\n" + "lwc1 %[in6], 21*4(%[win]) \t\n" + "madd.s %[out1], %[in2], %[in1], %[t1] \t\n" + "lwc1 %[in1], 10*4(%[tmp]) \t\n" + "madd.s %[out2], %[in4], %[in3], %[t1] \t\n" + "mul.s %[out3], %[in5], %[t0] \t\n" + "mul.s %[out4], %[in6], %[t0] \t\n" + "swc1 %[out1], 512*4(%[out]) \t\n" + "lwc1 %[in2], 8*4(%[tmp]) \t\n" + "swc1 %[out2], 32*4(%[out]) \t\n" + "swc1 %[out3], 4*16*4(%[buf]) \t\n" + "swc1 %[out4], 4*1*4(%[buf]) \t\n" + "li.s %[c2], 1.18310079157624925896 \t\n" + "add.s %[s0], %[in1], %[in2] \t\n" + "sub.s %[s2], %[in1], %[in2] \t\n" + "lwc1 %[in3], 11*4(%[tmp]) \t\n" + "lwc1 %[in4], 9*4(%[tmp]) \t\n" + "add.s %[s1], %[in3], %[in4] \t\n" + "sub.s %[s3], %[in3], %[in4] \t\n" + "lwc1 %[in1], 11*4(%[win]) \t\n" + "lwc1 %[in2], 4*11*4(%[buf]) \t\n" + "lwc1 %[in3], 6*4(%[win]) \t\n" + "mul.s %[s1], %[s1], %[c1] \t\n" + "mul.s %[s3], %[s3], %[c2] \t\n" + "lwc1 %[in4], 4*6*4(%[buf]) \t\n" + "lwc1 %[in5], 31*4(%[win]) \t\n" + "lwc1 %[in6], 26*4(%[win]) \t\n" + "add.s %[t0], %[s0], %[s1] \t\n" + "sub.s %[t1], %[s0], %[s1] \t\n" + "mul.s %[out3], %[t0], %[in5] \t\n" + "mul.s %[out4], %[t0], %[in6] \t\n" + "add.s %[t0], %[s2], %[s3] \t\n" + "madd.s %[out1], %[in2], %[in1], %[t1] \t\n" + "madd.s %[out2], %[in4], %[in3], %[t1] \t\n" + "swc1 %[out3], 4*11*4(%[buf]) \t\n" + "swc1 %[out4], 4*6*4(%[buf]) \t\n" + "sub.s %[t1], %[s2], %[s3] \t\n" + "swc1 %[out1], 352*4(%[out]) \t\n" + "swc1 %[out2], 192*4(%[out]) \t\n" + "lwc1 %[in1], 15*4(%[win]) \t\n" + "lwc1 %[in2], 4*15*4(%[buf]) \t\n" + "lwc1 %[in3], 2*4(%[win]) \t\n" + "lwc1 %[in4], 4*2*4(%[buf]) \t\n" + "lwc1 %[in5], 35*4(%[win]) \t\n" + "lwc1 %[in6], 22*4(%[win]) \t\n" + "madd.s %[out1], %[in2], %[in1], %[t1] \t\n" + "lwc1 %[in1], 14*4(%[tmp]) \t\n" + "madd.s %[out2], %[in4], %[in3], %[t1] \t\n" + "mul.s %[out3], %[t0], %[in5] \t\n" + "mul.s %[out4], %[t0], %[in6] \t\n" + "swc1 %[out1], 480*4(%[out]) \t\n" + "lwc1 %[in2], 12*4(%[tmp]) \t\n" + "swc1 %[out2], 64*4(%[out]) \t\n" + "swc1 %[out3], 4*15*4(%[buf]) \t\n" + "swc1 %[out4], 4*2*4(%[buf]) \t\n" + "lwc1 %[in3], 15*4(%[tmp]) \t\n" + "add.s %[s0], %[in1], %[in2] \t\n" + "sub.s %[s2], %[in1], %[in2] \t\n" + "lwc1 %[in4], 13*4(%[tmp]) \t\n" + "li.s %[c1], 0.61038729438072803416 \t\n" + "li.s %[c2], 0.87172339781054900991 \t\n" + "add.s %[s1], %[in3], %[in4] \t\n" + "sub.s %[s3], %[in3], %[in4] \t\n" + "lwc1 %[in1], 12*4(%[win]) \t\n" + "lwc1 %[in2], 4*12*4(%[buf]) \t\n" + "lwc1 %[in3], 5*4(%[win]) \t\n" + "mul.s %[s1], %[s1], %[c1] \t\n" + "mul.s %[s3], %[s3], %[c2] \t\n" + "lwc1 %[in4], 4*5*4(%[buf]) \t\n" + "lwc1 %[in5], 32*4(%[win]) \t\n" + "lwc1 %[in6], 25*4(%[win]) \t\n" + "add.s %[t0], %[s0], %[s1] \t\n" + "sub.s %[t1], %[s0], %[s1] \t\n" + "lwc1 %[s0], 16*4(%[tmp]) \t\n" + "lwc1 %[s1], 17*4(%[tmp]) \t\n" + "li.s %[c1], 0.70710678118654752439 \t\n" + "mul.s %[out3], %[t0], %[in5] \t\n" + "madd.s %[out1], %[in2], %[in1], %[t1] \t\n" + "madd.s %[out2], %[in4], %[in3], %[t1] \t\n" + "mul.s %[out4], %[t0], %[in6] \t\n" + "add.s %[t0], %[s2], %[s3] \t\n" + "swc1 %[out3], 4*12*4(%[buf]) \t\n" + "swc1 %[out1], 384*4(%[out]) \t\n" + "swc1 %[out2], 160*4(%[out]) \t\n" + "swc1 %[out4], 4*5*4(%[buf]) \t\n" + "sub.s %[t1], %[s2], %[s3] \t\n" + "lwc1 %[in1], 14*4(%[win]) \t\n" + "lwc1 %[in2], 4*14*4(%[buf]) \t\n" + "lwc1 %[in3], 3*4(%[win]) \t\n" + "lwc1 %[in4], 4*3*4(%[buf]) \t\n" + "lwc1 %[in5], 34*4(%[win]) \t\n" + "lwc1 %[in6], 23*4(%[win]) \t\n" + "madd.s %[out1], %[in2], %[in1], %[t1] \t\n" + "mul.s %[s1], %[s1], %[c1] \t\n" + "madd.s %[out2], %[in4], %[in3], %[t1] \t\n" + "mul.s %[out3], %[in5], %[t0] \t\n" + "mul.s %[out4], %[in6], %[t0] \t\n" + "swc1 %[out1], 448*4(%[out]) \t\n" + "add.s %[t0], %[s0], %[s1] \t\n" + "swc1 %[out2], 96*4(%[out]) \t\n" + "swc1 %[out3], 4*14*4(%[buf]) \t\n" + "swc1 %[out4], 4*3*4(%[buf]) \t\n" + "sub.s %[t1], %[s0], %[s1] \t\n" + "lwc1 %[in1], 13*4(%[win]) \t\n" + "lwc1 %[in2], 4*13*4(%[buf]) \t\n" + "lwc1 %[in3], 4*4(%[win]) \t\n" + "lwc1 %[in4], 4*4*4(%[buf]) \t\n" + "lwc1 %[in5], 33*4(%[win]) \t\n" + "lwc1 %[in6], 24*4(%[win]) \t\n" + "madd.s %[out1], %[in2], %[in1], %[t1] \t\n" + "madd.s %[out2], %[in4], %[in3], %[t1] \t\n" + "mul.s %[out3], %[t0], %[in5] \t\n" + "mul.s %[out4], %[t0], %[in6] \t\n" + "swc1 %[out1], 416*4(%[out]) \t\n" + "swc1 %[out2], 128*4(%[out]) \t\n" + "swc1 %[out3], 4*13*4(%[buf]) \t\n" + "swc1 %[out4], 4*4*4(%[buf]) \t\n" + + : [c1] "=&f" (c1), [c2] "=&f" (c2), + [in1] "=&f" (in1), [in2] "=&f" (in2), + [in3] "=&f" (in3), [in4] "=&f" (in4), + [in5] "=&f" (in5), [in6] "=&f" (in6), + [out1] "=&f" (out1), [out2] "=&f" (out2), + [out3] "=&f" (out3), [out4] "=&f" (out4), + [t0] "=&f" (t0), [t1] "=&f" (t1), + [t2] "=&f" (t2), [t3] "=&f" (t3), + [s0] "=&f" (s0), [s1] "=&f" (s1), + [s2] "=&f" (s2), [s3] "=&f" (s3) + : [tmp] "r" (tmp), [win] "r" (win), + [buf] "r" (buf), [out] "r" (out) + ); +} + +static void ff_imdct36_blocks_mips_float(float *out, float *buf, float *in, + int count, int switch_point, int block_type) +{ + int j; + for (j=0 ; j < count; j++) { + /* apply window & overlap with previous buffer */ + + /* select window */ + int win_idx = (switch_point && j < 2) ? 0 : block_type; + float *win = ff_mdct_win_float[win_idx + (4 & -(j & 1))]; + + imdct36_mips_float(out, buf, in, win); + + in += 18; + buf += ((j&3) != 3 ? 1 : (72-3)); + out++; + } +} + +void ff_mpadsp_init_mipsfpu(MPADSPContext *s) +{ + s->apply_window_float = ff_mpadsp_apply_window_mips_float; + s->imdct36_blocks_float = ff_imdct36_blocks_mips_float; + s->dct32_float = ff_dct32_mips_float; +} |