virtualenv/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2024-12-18 03:19:31 +02:00
Code Issues Releases Activity
bdf7093bd0
FFmpeg/libavcodec/mips/aaccoder_mips.c

2499 lines
103 KiB
C
Raw Normal View History

mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
/*
* 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)
* Szabolcs Pal (sabolc@mips.com)
*
* AAC coefficients encoder 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/aaccoder.c
*/
#include "libavutil/libm.h"
#include <float.h>
#include "libavutil/mathematics.h"
#include "libavcodec/avcodec.h"
#include "libavcodec/put_bits.h"
#include "libavcodec/aac.h"
#include "libavcodec/aacenc.h"
#include "libavcodec/aactab.h"
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
#include "libavcodec/aacenctab.h"
avcodec/mips/aaccoder_mips: Sync with the generic code This patch fixes build of AAC encoder optimized for mips that was broken due to some changes in generic code that were not propagated to the optimized code. Also, some functions in the optimized code are basically duplicate of functions from generic code. Since they do not bring enough improvement to the optimized code to justify their existence, they are removed (which improves maintainability of the optimized code). Optimizations disabled in 97437bd are enabled again. Signed-off-by: Nedeljko Babic <nedeljko.babic@imgtec.com> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2015-10-13 16:14:51 +02:00
#include "libavcodec/aacenc_utils.h"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
#if HAVE_INLINE_ASM
typedef struct BandCodingPath {
int prev_idx;
float cost;
int run;
} BandCodingPath;
static const uint8_t uquad_sign_bits[81] = {
0, 1, 1, 1, 2, 2, 1, 2, 2,
1, 2, 2, 2, 3, 3, 2, 3, 3,
1, 2, 2, 2, 3, 3, 2, 3, 3,
1, 2, 2, 2, 3, 3, 2, 3, 3,
2, 3, 3, 3, 4, 4, 3, 4, 4,
2, 3, 3, 3, 4, 4, 3, 4, 4,
1, 2, 2, 2, 3, 3, 2, 3, 3,
2, 3, 3, 3, 4, 4, 3, 4, 4,
2, 3, 3, 3, 4, 4, 3, 4, 4
};
static const uint8_t upair7_sign_bits[64] = {
0, 1, 1, 1, 1, 1, 1, 1,
1, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2,
};
static const uint8_t upair12_sign_bits[169] = {
0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2
};
static const uint8_t esc_sign_bits[289] = {
0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2
};
/**
* Functions developed from template function and optimized for quantizing and encoding band
*/
static void quantize_and_encode_band_cost_SQUAD_mips(struct AACEncContext *s,
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
PutBitContext *pb, const float *in, float *out,
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
int *bits, float *energy, const float ROUNDING)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
{
const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
const float IQ = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
int i;
int qc1, qc2, qc3, qc4;
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
float qenergy = 0.0f;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1];
uint16_t *p_codes = (uint16_t *)ff_aac_spectral_codes[cb-1];
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
float *p_vec = (float *)ff_aac_codebook_vectors[cb-1];
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
abs_pow34_v(s->scoefs, in, size);
scaled = s->scoefs;
for (i = 0; i < size; i += 4) {
int curidx;
int *in_int = (int *)&in[i];
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
int t0, t1, t2, t3, t4, t5, t6, t7;
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
const float *vec;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
qc1 = scaled[i ] * Q34 + ROUND_STANDARD;
qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;
qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;
qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
"slt %[qc1], $zero, %[qc1] \n\t"
"slt %[qc2], $zero, %[qc2] \n\t"
"slt %[qc3], $zero, %[qc3] \n\t"
"slt %[qc4], $zero, %[qc4] \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"lw %[t0], 0(%[in_int]) \n\t"
"lw %[t1], 4(%[in_int]) \n\t"
"lw %[t2], 8(%[in_int]) \n\t"
"lw %[t3], 12(%[in_int]) \n\t"
"srl %[t0], %[t0], 31 \n\t"
"srl %[t1], %[t1], 31 \n\t"
"srl %[t2], %[t2], 31 \n\t"
"srl %[t3], %[t3], 31 \n\t"
"subu %[t4], $zero, %[qc1] \n\t"
"subu %[t5], $zero, %[qc2] \n\t"
"subu %[t6], $zero, %[qc3] \n\t"
"subu %[t7], $zero, %[qc4] \n\t"
"movn %[qc1], %[t4], %[t0] \n\t"
"movn %[qc2], %[t5], %[t1] \n\t"
"movn %[qc3], %[t6], %[t2] \n\t"
"movn %[qc4], %[t7], %[t3] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
".set pop \n\t"
: [qc1]"+r"(qc1), [qc2]"+r"(qc2),
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
[qc3]"+r"(qc3), [qc4]"+r"(qc4),
[t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),
[t4]"=&r"(t4), [t5]"=&r"(t5), [t6]"=&r"(t6), [t7]"=&r"(t7)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
: [in_int]"r"(in_int)
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
: "memory"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
);
curidx = qc1;
curidx *= 3;
curidx += qc2;
curidx *= 3;
curidx += qc3;
curidx *= 3;
curidx += qc4;
curidx += 40;
put_bits(pb, p_bits[curidx], p_codes[curidx]);
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
if (out || energy) {
float e1,e2,e3,e4;
vec = &p_vec[curidx*4];
e1 = vec[0] * IQ;
e2 = vec[1] * IQ;
e3 = vec[2] * IQ;
e4 = vec[3] * IQ;
if (out) {
out[i+0] = e1;
out[i+1] = e2;
out[i+2] = e3;
out[i+3] = e4;
}
if (energy)
qenergy += (e1*e1 + e2*e2) + (e3*e3 + e4*e4);
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
}
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
}
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
if (energy)
*energy = qenergy;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
}
static void quantize_and_encode_band_cost_UQUAD_mips(struct AACEncContext *s,
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
PutBitContext *pb, const float *in, float *out,
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
int *bits, float *energy, const float ROUNDING)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
{
const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
const float IQ = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
int i;
int qc1, qc2, qc3, qc4;
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
float qenergy = 0.0f;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1];
uint16_t *p_codes = (uint16_t *)ff_aac_spectral_codes[cb-1];
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
float *p_vec = (float *)ff_aac_codebook_vectors[cb-1];
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
abs_pow34_v(s->scoefs, in, size);
scaled = s->scoefs;
for (i = 0; i < size; i += 4) {
int curidx, sign, count;
int *in_int = (int *)&in[i];
uint8_t v_bits;
unsigned int v_codes;
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
int t0, t1, t2, t3, t4;
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
const float *vec;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
qc1 = scaled[i ] * Q34 + ROUND_STANDARD;
qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;
qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;
qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"ori %[t4], $zero, 2 \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
"ori %[sign], $zero, 0 \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"slt %[t0], %[t4], %[qc1] \n\t"
"slt %[t1], %[t4], %[qc2] \n\t"
"slt %[t2], %[t4], %[qc3] \n\t"
"slt %[t3], %[t4], %[qc4] \n\t"
"movn %[qc1], %[t4], %[t0] \n\t"
"movn %[qc2], %[t4], %[t1] \n\t"
"movn %[qc3], %[t4], %[t2] \n\t"
"movn %[qc4], %[t4], %[t3] \n\t"
"lw %[t0], 0(%[in_int]) \n\t"
"lw %[t1], 4(%[in_int]) \n\t"
"lw %[t2], 8(%[in_int]) \n\t"
"lw %[t3], 12(%[in_int]) \n\t"
"slt %[t0], %[t0], $zero \n\t"
"movn %[sign], %[t0], %[qc1] \n\t"
"slt %[t1], %[t1], $zero \n\t"
"slt %[t2], %[t2], $zero \n\t"
"slt %[t3], %[t3], $zero \n\t"
"sll %[t0], %[sign], 1 \n\t"
"or %[t0], %[t0], %[t1] \n\t"
"movn %[sign], %[t0], %[qc2] \n\t"
"slt %[t4], $zero, %[qc1] \n\t"
"slt %[t1], $zero, %[qc2] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
"slt %[count], $zero, %[qc3] \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"sll %[t0], %[sign], 1 \n\t"
"or %[t0], %[t0], %[t2] \n\t"
"movn %[sign], %[t0], %[qc3] \n\t"
"slt %[t2], $zero, %[qc4] \n\t"
"addu %[count], %[count], %[t4] \n\t"
"addu %[count], %[count], %[t1] \n\t"
"sll %[t0], %[sign], 1 \n\t"
"or %[t0], %[t0], %[t3] \n\t"
"movn %[sign], %[t0], %[qc4] \n\t"
"addu %[count], %[count], %[t2] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
".set pop \n\t"
: [qc1]"+r"(qc1), [qc2]"+r"(qc2),
[qc3]"+r"(qc3), [qc4]"+r"(qc4),
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
[sign]"=&r"(sign), [count]"=&r"(count),
[t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),
[t4]"=&r"(t4)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
: [in_int]"r"(in_int)
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
: "memory"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
);
curidx = qc1;
curidx *= 3;
curidx += qc2;
curidx *= 3;
curidx += qc3;
curidx *= 3;
curidx += qc4;
v_codes = (p_codes[curidx] << count) | (sign & ((1 << count) - 1));
v_bits = p_bits[curidx] + count;
put_bits(pb, v_bits, v_codes);
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
if (out || energy) {
float e1,e2,e3,e4;
vec = &p_vec[curidx*4];
e1 = copysignf(vec[0] * IQ, in[i+0]);
e2 = copysignf(vec[1] * IQ, in[i+1]);
e3 = copysignf(vec[2] * IQ, in[i+2]);
e4 = copysignf(vec[3] * IQ, in[i+3]);
if (out) {
out[i+0] = e1;
out[i+1] = e2;
out[i+2] = e3;
out[i+3] = e4;
}
if (energy)
qenergy += (e1*e1 + e2*e2) + (e3*e3 + e4*e4);
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
}
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
}
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
if (energy)
*energy = qenergy;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
}
static void quantize_and_encode_band_cost_SPAIR_mips(struct AACEncContext *s,
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
PutBitContext *pb, const float *in, float *out,
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
int *bits, float *energy, const float ROUNDING)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
{
const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
const float IQ = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
int i;
int qc1, qc2, qc3, qc4;
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
float qenergy = 0.0f;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1];
uint16_t *p_codes = (uint16_t *)ff_aac_spectral_codes[cb-1];
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
float *p_vec = (float *)ff_aac_codebook_vectors[cb-1];
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
abs_pow34_v(s->scoefs, in, size);
scaled = s->scoefs;
for (i = 0; i < size; i += 4) {
int curidx, curidx2;
int *in_int = (int *)&in[i];
uint8_t v_bits;
unsigned int v_codes;
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
int t0, t1, t2, t3, t4, t5, t6, t7;
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
const float *vec1, *vec2;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
qc1 = scaled[i ] * Q34 + ROUND_STANDARD;
qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;
qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;
qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"ori %[t4], $zero, 4 \n\t"
"slt %[t0], %[t4], %[qc1] \n\t"
"slt %[t1], %[t4], %[qc2] \n\t"
"slt %[t2], %[t4], %[qc3] \n\t"
"slt %[t3], %[t4], %[qc4] \n\t"
"movn %[qc1], %[t4], %[t0] \n\t"
"movn %[qc2], %[t4], %[t1] \n\t"
"movn %[qc3], %[t4], %[t2] \n\t"
"movn %[qc4], %[t4], %[t3] \n\t"
"lw %[t0], 0(%[in_int]) \n\t"
"lw %[t1], 4(%[in_int]) \n\t"
"lw %[t2], 8(%[in_int]) \n\t"
"lw %[t3], 12(%[in_int]) \n\t"
"srl %[t0], %[t0], 31 \n\t"
"srl %[t1], %[t1], 31 \n\t"
"srl %[t2], %[t2], 31 \n\t"
"srl %[t3], %[t3], 31 \n\t"
"subu %[t4], $zero, %[qc1] \n\t"
"subu %[t5], $zero, %[qc2] \n\t"
"subu %[t6], $zero, %[qc3] \n\t"
"subu %[t7], $zero, %[qc4] \n\t"
"movn %[qc1], %[t4], %[t0] \n\t"
"movn %[qc2], %[t5], %[t1] \n\t"
"movn %[qc3], %[t6], %[t2] \n\t"
"movn %[qc4], %[t7], %[t3] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
".set pop \n\t"
: [qc1]"+r"(qc1), [qc2]"+r"(qc2),
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
[qc3]"+r"(qc3), [qc4]"+r"(qc4),
[t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),
[t4]"=&r"(t4), [t5]"=&r"(t5), [t6]"=&r"(t6), [t7]"=&r"(t7)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
: [in_int]"r"(in_int)
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
: "memory"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
);
curidx = 9 * qc1;
curidx += qc2 + 40;
curidx2 = 9 * qc3;
curidx2 += qc4 + 40;
v_codes = (p_codes[curidx] << p_bits[curidx2]) | (p_codes[curidx2]);
v_bits = p_bits[curidx] + p_bits[curidx2];
put_bits(pb, v_bits, v_codes);
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
if (out || energy) {
float e1,e2,e3,e4;
vec1 = &p_vec[curidx*2 ];
vec2 = &p_vec[curidx2*2];
e1 = vec1[0] * IQ;
e2 = vec1[1] * IQ;
e3 = vec2[0] * IQ;
e4 = vec2[1] * IQ;
if (out) {
out[i+0] = e1;
out[i+1] = e2;
out[i+2] = e3;
out[i+3] = e4;
}
if (energy)
qenergy += (e1*e1 + e2*e2) + (e3*e3 + e4*e4);
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
}
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
}
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
if (energy)
*energy = qenergy;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
}
static void quantize_and_encode_band_cost_UPAIR7_mips(struct AACEncContext *s,
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
PutBitContext *pb, const float *in, float *out,
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
int *bits, float *energy, const float ROUNDING)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
{
const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
const float IQ = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
int i;
int qc1, qc2, qc3, qc4;
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
float qenergy = 0.0f;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
uint8_t *p_bits = (uint8_t*) ff_aac_spectral_bits[cb-1];
uint16_t *p_codes = (uint16_t*)ff_aac_spectral_codes[cb-1];
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
float *p_vec = (float *)ff_aac_codebook_vectors[cb-1];
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
abs_pow34_v(s->scoefs, in, size);
scaled = s->scoefs;
for (i = 0; i < size; i += 4) {
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
int curidx1, curidx2, sign1, count1, sign2, count2;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
int *in_int = (int *)&in[i];
uint8_t v_bits;
unsigned int v_codes;
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
int t0, t1, t2, t3, t4;
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
const float *vec1, *vec2;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
qc1 = scaled[i ] * Q34 + ROUND_STANDARD;
qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;
qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;
qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"ori %[t4], $zero, 7 \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
"ori %[sign1], $zero, 0 \n\t"
"ori %[sign2], $zero, 0 \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"slt %[t0], %[t4], %[qc1] \n\t"
"slt %[t1], %[t4], %[qc2] \n\t"
"slt %[t2], %[t4], %[qc3] \n\t"
"slt %[t3], %[t4], %[qc4] \n\t"
"movn %[qc1], %[t4], %[t0] \n\t"
"movn %[qc2], %[t4], %[t1] \n\t"
"movn %[qc3], %[t4], %[t2] \n\t"
"movn %[qc4], %[t4], %[t3] \n\t"
"lw %[t0], 0(%[in_int]) \n\t"
"lw %[t1], 4(%[in_int]) \n\t"
"lw %[t2], 8(%[in_int]) \n\t"
"lw %[t3], 12(%[in_int]) \n\t"
"slt %[t0], %[t0], $zero \n\t"
"movn %[sign1], %[t0], %[qc1] \n\t"
"slt %[t2], %[t2], $zero \n\t"
"movn %[sign2], %[t2], %[qc3] \n\t"
"slt %[t1], %[t1], $zero \n\t"
"sll %[t0], %[sign1], 1 \n\t"
"or %[t0], %[t0], %[t1] \n\t"
"movn %[sign1], %[t0], %[qc2] \n\t"
"slt %[t3], %[t3], $zero \n\t"
"sll %[t0], %[sign2], 1 \n\t"
"or %[t0], %[t0], %[t3] \n\t"
"movn %[sign2], %[t0], %[qc4] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
"slt %[count1], $zero, %[qc1] \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"slt %[t1], $zero, %[qc2] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
"slt %[count2], $zero, %[qc3] \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"slt %[t2], $zero, %[qc4] \n\t"
"addu %[count1], %[count1], %[t1] \n\t"
"addu %[count2], %[count2], %[t2] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
".set pop \n\t"
: [qc1]"+r"(qc1), [qc2]"+r"(qc2),
[qc3]"+r"(qc3), [qc4]"+r"(qc4),
[sign1]"=&r"(sign1), [count1]"=&r"(count1),
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
[sign2]"=&r"(sign2), [count2]"=&r"(count2),
[t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),
[t4]"=&r"(t4)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
: [in_int]"r"(in_int)
: "t0", "t1", "t2", "t3", "t4",
"memory"
);
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
curidx1 = 8 * qc1;
curidx1 += qc2;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
v_codes = (p_codes[curidx1] << count1) | sign1;
v_bits = p_bits[curidx1] + count1;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
put_bits(pb, v_bits, v_codes);
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
curidx2 = 8 * qc3;
curidx2 += qc4;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
v_codes = (p_codes[curidx2] << count2) | sign2;
v_bits = p_bits[curidx2] + count2;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
put_bits(pb, v_bits, v_codes);
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
if (out || energy) {
float e1,e2,e3,e4;
vec1 = &p_vec[curidx1*2];
vec2 = &p_vec[curidx2*2];
e1 = copysignf(vec1[0] * IQ, in[i+0]);
e2 = copysignf(vec1[1] * IQ, in[i+1]);
e3 = copysignf(vec2[0] * IQ, in[i+2]);
e4 = copysignf(vec2[1] * IQ, in[i+3]);
if (out) {
out[i+0] = e1;
out[i+1] = e2;
out[i+2] = e3;
out[i+3] = e4;
}
if (energy)
qenergy += (e1*e1 + e2*e2) + (e3*e3 + e4*e4);
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
}
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
}
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
if (energy)
*energy = qenergy;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
}
static void quantize_and_encode_band_cost_UPAIR12_mips(struct AACEncContext *s,
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
PutBitContext *pb, const float *in, float *out,
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
int *bits, float *energy, const float ROUNDING)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
{
const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
const float IQ = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
int i;
int qc1, qc2, qc3, qc4;
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
float qenergy = 0.0f;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
uint8_t *p_bits = (uint8_t*) ff_aac_spectral_bits[cb-1];
uint16_t *p_codes = (uint16_t*)ff_aac_spectral_codes[cb-1];
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
float *p_vec = (float *)ff_aac_codebook_vectors[cb-1];
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
abs_pow34_v(s->scoefs, in, size);
scaled = s->scoefs;
for (i = 0; i < size; i += 4) {
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
int curidx1, curidx2, sign1, count1, sign2, count2;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
int *in_int = (int *)&in[i];
uint8_t v_bits;
unsigned int v_codes;
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
int t0, t1, t2, t3, t4;
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
const float *vec1, *vec2;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
qc1 = scaled[i ] * Q34 + ROUND_STANDARD;
qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;
qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;
qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"ori %[t4], $zero, 12 \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
"ori %[sign1], $zero, 0 \n\t"
"ori %[sign2], $zero, 0 \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"slt %[t0], %[t4], %[qc1] \n\t"
"slt %[t1], %[t4], %[qc2] \n\t"
"slt %[t2], %[t4], %[qc3] \n\t"
"slt %[t3], %[t4], %[qc4] \n\t"
"movn %[qc1], %[t4], %[t0] \n\t"
"movn %[qc2], %[t4], %[t1] \n\t"
"movn %[qc3], %[t4], %[t2] \n\t"
"movn %[qc4], %[t4], %[t3] \n\t"
"lw %[t0], 0(%[in_int]) \n\t"
"lw %[t1], 4(%[in_int]) \n\t"
"lw %[t2], 8(%[in_int]) \n\t"
"lw %[t3], 12(%[in_int]) \n\t"
"slt %[t0], %[t0], $zero \n\t"
"movn %[sign1], %[t0], %[qc1] \n\t"
"slt %[t2], %[t2], $zero \n\t"
"movn %[sign2], %[t2], %[qc3] \n\t"
"slt %[t1], %[t1], $zero \n\t"
"sll %[t0], %[sign1], 1 \n\t"
"or %[t0], %[t0], %[t1] \n\t"
"movn %[sign1], %[t0], %[qc2] \n\t"
"slt %[t3], %[t3], $zero \n\t"
"sll %[t0], %[sign2], 1 \n\t"
"or %[t0], %[t0], %[t3] \n\t"
"movn %[sign2], %[t0], %[qc4] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
"slt %[count1], $zero, %[qc1] \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"slt %[t1], $zero, %[qc2] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
"slt %[count2], $zero, %[qc3] \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"slt %[t2], $zero, %[qc4] \n\t"
"addu %[count1], %[count1], %[t1] \n\t"
"addu %[count2], %[count2], %[t2] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
".set pop \n\t"
: [qc1]"+r"(qc1), [qc2]"+r"(qc2),
[qc3]"+r"(qc3), [qc4]"+r"(qc4),
[sign1]"=&r"(sign1), [count1]"=&r"(count1),
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
[sign2]"=&r"(sign2), [count2]"=&r"(count2),
[t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),
[t4]"=&r"(t4)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
: [in_int]"r"(in_int)
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
: "memory"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
);
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
curidx1 = 13 * qc1;
curidx1 += qc2;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
v_codes = (p_codes[curidx1] << count1) | sign1;
v_bits = p_bits[curidx1] + count1;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
put_bits(pb, v_bits, v_codes);
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
curidx2 = 13 * qc3;
curidx2 += qc4;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
v_codes = (p_codes[curidx2] << count2) | sign2;
v_bits = p_bits[curidx2] + count2;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
put_bits(pb, v_bits, v_codes);
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
if (out || energy) {
float e1,e2,e3,e4;
vec1 = &p_vec[curidx1*2];
vec2 = &p_vec[curidx2*2];
e1 = copysignf(vec1[0] * IQ, in[i+0]);
e2 = copysignf(vec1[1] * IQ, in[i+1]);
e3 = copysignf(vec2[0] * IQ, in[i+2]);
e4 = copysignf(vec2[1] * IQ, in[i+3]);
if (out) {
out[i+0] = e1;
out[i+1] = e2;
out[i+2] = e3;
out[i+3] = e4;
}
if (energy)
qenergy += (e1*e1 + e2*e2) + (e3*e3 + e4*e4);
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
}
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
}
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
if (energy)
*energy = qenergy;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
}
static void quantize_and_encode_band_cost_ESC_mips(struct AACEncContext *s,
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
PutBitContext *pb, const float *in, float *out,
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
int *bits, float *energy, const float ROUNDING)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
{
const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
const float IQ = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
int i;
int qc1, qc2, qc3, qc4;
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
float qenergy = 0.0f;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
uint8_t *p_bits = (uint8_t* )ff_aac_spectral_bits[cb-1];
uint16_t *p_codes = (uint16_t*)ff_aac_spectral_codes[cb-1];
float *p_vectors = (float* )ff_aac_codebook_vectors[cb-1];
abs_pow34_v(s->scoefs, in, size);
scaled = s->scoefs;
if (cb < 11) {
for (i = 0; i < size; i += 4) {
int curidx, curidx2, sign1, count1, sign2, count2;
int *in_int = (int *)&in[i];
uint8_t v_bits;
unsigned int v_codes;
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
int t0, t1, t2, t3, t4;
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
const float *vec1, *vec2;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
qc1 = scaled[i ] * Q34 + ROUNDING;
qc2 = scaled[i+1] * Q34 + ROUNDING;
qc3 = scaled[i+2] * Q34 + ROUNDING;
qc4 = scaled[i+3] * Q34 + ROUNDING;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"ori %[t4], $zero, 16 \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
"ori %[sign1], $zero, 0 \n\t"
"ori %[sign2], $zero, 0 \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"slt %[t0], %[t4], %[qc1] \n\t"
"slt %[t1], %[t4], %[qc2] \n\t"
"slt %[t2], %[t4], %[qc3] \n\t"
"slt %[t3], %[t4], %[qc4] \n\t"
"movn %[qc1], %[t4], %[t0] \n\t"
"movn %[qc2], %[t4], %[t1] \n\t"
"movn %[qc3], %[t4], %[t2] \n\t"
"movn %[qc4], %[t4], %[t3] \n\t"
"lw %[t0], 0(%[in_int]) \n\t"
"lw %[t1], 4(%[in_int]) \n\t"
"lw %[t2], 8(%[in_int]) \n\t"
"lw %[t3], 12(%[in_int]) \n\t"
"slt %[t0], %[t0], $zero \n\t"
"movn %[sign1], %[t0], %[qc1] \n\t"
"slt %[t2], %[t2], $zero \n\t"
"movn %[sign2], %[t2], %[qc3] \n\t"
"slt %[t1], %[t1], $zero \n\t"
"sll %[t0], %[sign1], 1 \n\t"
"or %[t0], %[t0], %[t1] \n\t"
"movn %[sign1], %[t0], %[qc2] \n\t"
"slt %[t3], %[t3], $zero \n\t"
"sll %[t0], %[sign2], 1 \n\t"
"or %[t0], %[t0], %[t3] \n\t"
"movn %[sign2], %[t0], %[qc4] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
"slt %[count1], $zero, %[qc1] \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"slt %[t1], $zero, %[qc2] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
"slt %[count2], $zero, %[qc3] \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"slt %[t2], $zero, %[qc4] \n\t"
"addu %[count1], %[count1], %[t1] \n\t"
"addu %[count2], %[count2], %[t2] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
".set pop \n\t"
: [qc1]"+r"(qc1), [qc2]"+r"(qc2),
[qc3]"+r"(qc3), [qc4]"+r"(qc4),
[sign1]"=&r"(sign1), [count1]"=&r"(count1),
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
[sign2]"=&r"(sign2), [count2]"=&r"(count2),
[t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),
[t4]"=&r"(t4)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
: [in_int]"r"(in_int)
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
: "memory"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
);
curidx = 17 * qc1;
curidx += qc2;
curidx2 = 17 * qc3;
curidx2 += qc4;
v_codes = (p_codes[curidx] << count1) | sign1;
v_bits = p_bits[curidx] + count1;
put_bits(pb, v_bits, v_codes);
v_codes = (p_codes[curidx2] << count2) | sign2;
v_bits = p_bits[curidx2] + count2;
put_bits(pb, v_bits, v_codes);
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
if (out || energy) {
float e1,e2,e3,e4;
vec1 = &p_vectors[curidx*2 ];
vec2 = &p_vectors[curidx2*2];
e1 = copysignf(vec1[0] * IQ, in[i+0]);
e2 = copysignf(vec1[1] * IQ, in[i+1]);
e3 = copysignf(vec2[0] * IQ, in[i+2]);
e4 = copysignf(vec2[1] * IQ, in[i+3]);
if (out) {
out[i+0] = e1;
out[i+1] = e2;
out[i+2] = e3;
out[i+3] = e4;
}
if (energy)
qenergy += (e1*e1 + e2*e2) + (e3*e3 + e4*e4);
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
}
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
}
} else {
for (i = 0; i < size; i += 4) {
int curidx, curidx2, sign1, count1, sign2, count2;
int *in_int = (int *)&in[i];
uint8_t v_bits;
unsigned int v_codes;
int c1, c2, c3, c4;
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
int t0, t1, t2, t3, t4;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
qc1 = scaled[i ] * Q34 + ROUNDING;
qc2 = scaled[i+1] * Q34 + ROUNDING;
qc3 = scaled[i+2] * Q34 + ROUNDING;
qc4 = scaled[i+3] * Q34 + ROUNDING;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"ori %[t4], $zero, 16 \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
"ori %[sign1], $zero, 0 \n\t"
"ori %[sign2], $zero, 0 \n\t"
"shll_s.w %[c1], %[qc1], 18 \n\t"
"shll_s.w %[c2], %[qc2], 18 \n\t"
"shll_s.w %[c3], %[qc3], 18 \n\t"
"shll_s.w %[c4], %[qc4], 18 \n\t"
"srl %[c1], %[c1], 18 \n\t"
"srl %[c2], %[c2], 18 \n\t"
"srl %[c3], %[c3], 18 \n\t"
"srl %[c4], %[c4], 18 \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"slt %[t0], %[t4], %[qc1] \n\t"
"slt %[t1], %[t4], %[qc2] \n\t"
"slt %[t2], %[t4], %[qc3] \n\t"
"slt %[t3], %[t4], %[qc4] \n\t"
"movn %[qc1], %[t4], %[t0] \n\t"
"movn %[qc2], %[t4], %[t1] \n\t"
"movn %[qc3], %[t4], %[t2] \n\t"
"movn %[qc4], %[t4], %[t3] \n\t"
"lw %[t0], 0(%[in_int]) \n\t"
"lw %[t1], 4(%[in_int]) \n\t"
"lw %[t2], 8(%[in_int]) \n\t"
"lw %[t3], 12(%[in_int]) \n\t"
"slt %[t0], %[t0], $zero \n\t"
"movn %[sign1], %[t0], %[qc1] \n\t"
"slt %[t2], %[t2], $zero \n\t"
"movn %[sign2], %[t2], %[qc3] \n\t"
"slt %[t1], %[t1], $zero \n\t"
"sll %[t0], %[sign1], 1 \n\t"
"or %[t0], %[t0], %[t1] \n\t"
"movn %[sign1], %[t0], %[qc2] \n\t"
"slt %[t3], %[t3], $zero \n\t"
"sll %[t0], %[sign2], 1 \n\t"
"or %[t0], %[t0], %[t3] \n\t"
"movn %[sign2], %[t0], %[qc4] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
"slt %[count1], $zero, %[qc1] \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"slt %[t1], $zero, %[qc2] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
"slt %[count2], $zero, %[qc3] \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"slt %[t2], $zero, %[qc4] \n\t"
"addu %[count1], %[count1], %[t1] \n\t"
"addu %[count2], %[count2], %[t2] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
".set pop \n\t"
: [qc1]"+r"(qc1), [qc2]"+r"(qc2),
[qc3]"+r"(qc3), [qc4]"+r"(qc4),
[sign1]"=&r"(sign1), [count1]"=&r"(count1),
[sign2]"=&r"(sign2), [count2]"=&r"(count2),
[c1]"=&r"(c1), [c2]"=&r"(c2),
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
[c3]"=&r"(c3), [c4]"=&r"(c4),
[t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),
[t4]"=&r"(t4)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
: [in_int]"r"(in_int)
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
: "memory"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
);
curidx = 17 * qc1;
curidx += qc2;
curidx2 = 17 * qc3;
curidx2 += qc4;
v_codes = (p_codes[curidx] << count1) | sign1;
v_bits = p_bits[curidx] + count1;
put_bits(pb, v_bits, v_codes);
if (p_vectors[curidx*2 ] == 64.0f) {
int len = av_log2(c1);
v_codes = (((1 << (len - 3)) - 2) << len) | (c1 & ((1 << len) - 1));
put_bits(pb, len * 2 - 3, v_codes);
}
if (p_vectors[curidx*2+1] == 64.0f) {
int len = av_log2(c2);
v_codes = (((1 << (len - 3)) - 2) << len) | (c2 & ((1 << len) - 1));
put_bits(pb, len*2-3, v_codes);
}
v_codes = (p_codes[curidx2] << count2) | sign2;
v_bits = p_bits[curidx2] + count2;
put_bits(pb, v_bits, v_codes);
if (p_vectors[curidx2*2 ] == 64.0f) {
int len = av_log2(c3);
v_codes = (((1 << (len - 3)) - 2) << len) | (c3 & ((1 << len) - 1));
put_bits(pb, len* 2 - 3, v_codes);
}
if (p_vectors[curidx2*2+1] == 64.0f) {
int len = av_log2(c4);
v_codes = (((1 << (len - 3)) - 2) << len) | (c4 & ((1 << len) - 1));
put_bits(pb, len * 2 - 3, v_codes);
}
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
if (out || energy) {
float e1, e2, e3, e4;
e1 = copysignf(c1 * cbrtf(c1) * IQ, in[i+0]);
e2 = copysignf(c2 * cbrtf(c2) * IQ, in[i+1]);
e3 = copysignf(c3 * cbrtf(c3) * IQ, in[i+2]);
e4 = copysignf(c4 * cbrtf(c4) * IQ, in[i+3]);
if (out) {
out[i+0] = e1;
out[i+1] = e2;
out[i+2] = e3;
out[i+3] = e4;
}
if (energy)
qenergy += (e1*e1 + e2*e2) + (e3*e3 + e4*e4);
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
}
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
}
}
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
if (energy)
*energy = qenergy;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
}
AAC: MIPS: Add missing codebooks in quantize funcs Add entries on the quantize function tables for the missing codebooks (which all behave like ZERO)
2015-09-03 15:49:17 +02:00
static void quantize_and_encode_band_cost_NONE_mips(struct AACEncContext *s,
PutBitContext *pb, const float *in, float *out,
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
int *bits, float *energy, const float ROUNDING) {
AAC: MIPS: Add missing codebooks in quantize funcs Add entries on the quantize function tables for the missing codebooks (which all behave like ZERO)
2015-09-03 15:49:17 +02:00
av_assert0(0);
}
static void quantize_and_encode_band_cost_ZERO_mips(struct AACEncContext *s,
PutBitContext *pb, const float *in, float *out,
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
int *bits, float *energy, const float ROUNDING) {
AAC: MIPS: Add missing codebooks in quantize funcs Add entries on the quantize function tables for the missing codebooks (which all behave like ZERO)
2015-09-03 15:49:17 +02:00
int i;
if (bits)
*bits = 0;
if (out) {
for (i = 0; i < size; i += 4) {
out[i ] = 0.0f;
out[i+1] = 0.0f;
out[i+2] = 0.0f;
out[i+3] = 0.0f;
}
}
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
if (energy)
*energy = 0.0f;
AAC: MIPS: Add missing codebooks in quantize funcs Add entries on the quantize function tables for the missing codebooks (which all behave like ZERO)
2015-09-03 15:49:17 +02:00
}
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
static void (*const quantize_and_encode_band_cost_arr[])(struct AACEncContext *s,
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
PutBitContext *pb, const float *in, float *out,
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
int *bits, float *energy, const float ROUNDING) = {
AAC: MIPS: Add missing codebooks in quantize funcs Add entries on the quantize function tables for the missing codebooks (which all behave like ZERO)
2015-09-03 15:49:17 +02:00
quantize_and_encode_band_cost_ZERO_mips,
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
quantize_and_encode_band_cost_SQUAD_mips,
quantize_and_encode_band_cost_SQUAD_mips,
quantize_and_encode_band_cost_UQUAD_mips,
quantize_and_encode_band_cost_UQUAD_mips,
quantize_and_encode_band_cost_SPAIR_mips,
quantize_and_encode_band_cost_SPAIR_mips,
quantize_and_encode_band_cost_UPAIR7_mips,
quantize_and_encode_band_cost_UPAIR7_mips,
quantize_and_encode_band_cost_UPAIR12_mips,
quantize_and_encode_band_cost_UPAIR12_mips,
quantize_and_encode_band_cost_ESC_mips,
AAC: MIPS: Add missing codebooks in quantize funcs Add entries on the quantize function tables for the missing codebooks (which all behave like ZERO)
2015-09-03 15:49:17 +02:00
quantize_and_encode_band_cost_NONE_mips, /* cb 12 doesn't exist */
quantize_and_encode_band_cost_ZERO_mips,
quantize_and_encode_band_cost_ZERO_mips,
quantize_and_encode_band_cost_ZERO_mips,
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
};
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
#define quantize_and_encode_band_cost( \
s, pb, in, out, scaled, size, scale_idx, cb, \
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
lambda, uplim, bits, energy, ROUNDING) \
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
quantize_and_encode_band_cost_arr[cb]( \
s, pb, in, out, scaled, size, scale_idx, cb, \
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
lambda, uplim, bits, energy, ROUNDING)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
static void quantize_and_encode_band_mips(struct AACEncContext *s, PutBitContext *pb,
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
const float *in, float *out, int size, int scale_idx,
int cb, const float lambda, int rtz)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
{
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
quantize_and_encode_band_cost(s, pb, in, out, NULL, size, scale_idx, cb, lambda,
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
INFINITY, NULL, NULL, (rtz) ? ROUND_TO_ZERO : ROUND_STANDARD);
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
}
/**
* Functions developed from template function and optimized for getting the number of bits
*/
static float get_band_numbits_ZERO_mips(struct AACEncContext *s,
PutBitContext *pb, const float *in,
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
int *bits)
{
return 0;
}
AAC: MIPS: Add missing codebooks in quantize funcs Add entries on the quantize function tables for the missing codebooks (which all behave like ZERO)
2015-09-03 15:49:17 +02:00
static float get_band_numbits_NONE_mips(struct AACEncContext *s,
PutBitContext *pb, const float *in,
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
int *bits)
{
av_assert0(0);
return 0;
}
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
static float get_band_numbits_SQUAD_mips(struct AACEncContext *s,
PutBitContext *pb, const float *in,
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
int *bits)
{
const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];
int i;
int qc1, qc2, qc3, qc4;
int curbits = 0;
uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1];
for (i = 0; i < size; i += 4) {
int curidx;
int *in_int = (int *)&in[i];
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
int t0, t1, t2, t3, t4, t5, t6, t7;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
qc1 = scaled[i ] * Q34 + ROUND_STANDARD;
qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;
qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;
qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
"slt %[qc1], $zero, %[qc1] \n\t"
"slt %[qc2], $zero, %[qc2] \n\t"
"slt %[qc3], $zero, %[qc3] \n\t"
"slt %[qc4], $zero, %[qc4] \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"lw %[t0], 0(%[in_int]) \n\t"
"lw %[t1], 4(%[in_int]) \n\t"
"lw %[t2], 8(%[in_int]) \n\t"
"lw %[t3], 12(%[in_int]) \n\t"
"srl %[t0], %[t0], 31 \n\t"
"srl %[t1], %[t1], 31 \n\t"
"srl %[t2], %[t2], 31 \n\t"
"srl %[t3], %[t3], 31 \n\t"
"subu %[t4], $zero, %[qc1] \n\t"
"subu %[t5], $zero, %[qc2] \n\t"
"subu %[t6], $zero, %[qc3] \n\t"
"subu %[t7], $zero, %[qc4] \n\t"
"movn %[qc1], %[t4], %[t0] \n\t"
"movn %[qc2], %[t5], %[t1] \n\t"
"movn %[qc3], %[t6], %[t2] \n\t"
"movn %[qc4], %[t7], %[t3] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
".set pop \n\t"
: [qc1]"+r"(qc1), [qc2]"+r"(qc2),
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
[qc3]"+r"(qc3), [qc4]"+r"(qc4),
[t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),
[t4]"=&r"(t4), [t5]"=&r"(t5), [t6]"=&r"(t6), [t7]"=&r"(t7)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
: [in_int]"r"(in_int)
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
: "memory"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
);
curidx = qc1;
curidx *= 3;
curidx += qc2;
curidx *= 3;
curidx += qc3;
curidx *= 3;
curidx += qc4;
curidx += 40;
curbits += p_bits[curidx];
}
return curbits;
}
static float get_band_numbits_UQUAD_mips(struct AACEncContext *s,
PutBitContext *pb, const float *in,
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
int *bits)
{
const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];
int i;
int curbits = 0;
int qc1, qc2, qc3, qc4;
uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1];
for (i = 0; i < size; i += 4) {
int curidx;
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
int t0, t1, t2, t3, t4;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
qc1 = scaled[i ] * Q34 + ROUND_STANDARD;
qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;
qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;
qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"ori %[t4], $zero, 2 \n\t"
"slt %[t0], %[t4], %[qc1] \n\t"
"slt %[t1], %[t4], %[qc2] \n\t"
"slt %[t2], %[t4], %[qc3] \n\t"
"slt %[t3], %[t4], %[qc4] \n\t"
"movn %[qc1], %[t4], %[t0] \n\t"
"movn %[qc2], %[t4], %[t1] \n\t"
"movn %[qc3], %[t4], %[t2] \n\t"
"movn %[qc4], %[t4], %[t3] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
".set pop \n\t"
: [qc1]"+r"(qc1), [qc2]"+r"(qc2),
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
[qc3]"+r"(qc3), [qc4]"+r"(qc4),
[t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),
[t4]"=&r"(t4)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
);
curidx = qc1;
curidx *= 3;
curidx += qc2;
curidx *= 3;
curidx += qc3;
curidx *= 3;
curidx += qc4;
curbits += p_bits[curidx];
curbits += uquad_sign_bits[curidx];
}
return curbits;
}
static float get_band_numbits_SPAIR_mips(struct AACEncContext *s,
PutBitContext *pb, const float *in,
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
int *bits)
{
const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];
int i;
int qc1, qc2, qc3, qc4;
int curbits = 0;
uint8_t *p_bits = (uint8_t*)ff_aac_spectral_bits[cb-1];
for (i = 0; i < size; i += 4) {
int curidx, curidx2;
int *in_int = (int *)&in[i];
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
int t0, t1, t2, t3, t4, t5, t6, t7;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
qc1 = scaled[i ] * Q34 + ROUND_STANDARD;
qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;
qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;
qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"ori %[t4], $zero, 4 \n\t"
"slt %[t0], %[t4], %[qc1] \n\t"
"slt %[t1], %[t4], %[qc2] \n\t"
"slt %[t2], %[t4], %[qc3] \n\t"
"slt %[t3], %[t4], %[qc4] \n\t"
"movn %[qc1], %[t4], %[t0] \n\t"
"movn %[qc2], %[t4], %[t1] \n\t"
"movn %[qc3], %[t4], %[t2] \n\t"
"movn %[qc4], %[t4], %[t3] \n\t"
"lw %[t0], 0(%[in_int]) \n\t"
"lw %[t1], 4(%[in_int]) \n\t"
"lw %[t2], 8(%[in_int]) \n\t"
"lw %[t3], 12(%[in_int]) \n\t"
"srl %[t0], %[t0], 31 \n\t"
"srl %[t1], %[t1], 31 \n\t"
"srl %[t2], %[t2], 31 \n\t"
"srl %[t3], %[t3], 31 \n\t"
"subu %[t4], $zero, %[qc1] \n\t"
"subu %[t5], $zero, %[qc2] \n\t"
"subu %[t6], $zero, %[qc3] \n\t"
"subu %[t7], $zero, %[qc4] \n\t"
"movn %[qc1], %[t4], %[t0] \n\t"
"movn %[qc2], %[t5], %[t1] \n\t"
"movn %[qc3], %[t6], %[t2] \n\t"
"movn %[qc4], %[t7], %[t3] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
".set pop \n\t"
: [qc1]"+r"(qc1), [qc2]"+r"(qc2),
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
[qc3]"+r"(qc3), [qc4]"+r"(qc4),
[t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),
[t4]"=&r"(t4), [t5]"=&r"(t5), [t6]"=&r"(t6), [t7]"=&r"(t7)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
: [in_int]"r"(in_int)
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
: "memory"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
);
curidx = 9 * qc1;
curidx += qc2 + 40;
curidx2 = 9 * qc3;
curidx2 += qc4 + 40;
curbits += p_bits[curidx] + p_bits[curidx2];
}
return curbits;
}
static float get_band_numbits_UPAIR7_mips(struct AACEncContext *s,
PutBitContext *pb, const float *in,
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
int *bits)
{
const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];
int i;
int qc1, qc2, qc3, qc4;
int curbits = 0;
uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1];
for (i = 0; i < size; i += 4) {
int curidx, curidx2;
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
int t0, t1, t2, t3, t4;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
qc1 = scaled[i ] * Q34 + ROUND_STANDARD;
qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;
qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;
qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"ori %[t4], $zero, 7 \n\t"
"slt %[t0], %[t4], %[qc1] \n\t"
"slt %[t1], %[t4], %[qc2] \n\t"
"slt %[t2], %[t4], %[qc3] \n\t"
"slt %[t3], %[t4], %[qc4] \n\t"
"movn %[qc1], %[t4], %[t0] \n\t"
"movn %[qc2], %[t4], %[t1] \n\t"
"movn %[qc3], %[t4], %[t2] \n\t"
"movn %[qc4], %[t4], %[t3] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
".set pop \n\t"
: [qc1]"+r"(qc1), [qc2]"+r"(qc2),
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
[qc3]"+r"(qc3), [qc4]"+r"(qc4),
[t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),
[t4]"=&r"(t4)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
);
curidx = 8 * qc1;
curidx += qc2;
curidx2 = 8 * qc3;
curidx2 += qc4;
curbits += p_bits[curidx] +
upair7_sign_bits[curidx] +
p_bits[curidx2] +
upair7_sign_bits[curidx2];
}
return curbits;
}
static float get_band_numbits_UPAIR12_mips(struct AACEncContext *s,
PutBitContext *pb, const float *in,
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
int *bits)
{
const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];
int i;
int qc1, qc2, qc3, qc4;
int curbits = 0;
uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1];
for (i = 0; i < size; i += 4) {
int curidx, curidx2;
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
int t0, t1, t2, t3, t4;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
qc1 = scaled[i ] * Q34 + ROUND_STANDARD;
qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;
qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;
qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"ori %[t4], $zero, 12 \n\t"
"slt %[t0], %[t4], %[qc1] \n\t"
"slt %[t1], %[t4], %[qc2] \n\t"
"slt %[t2], %[t4], %[qc3] \n\t"
"slt %[t3], %[t4], %[qc4] \n\t"
"movn %[qc1], %[t4], %[t0] \n\t"
"movn %[qc2], %[t4], %[t1] \n\t"
"movn %[qc3], %[t4], %[t2] \n\t"
"movn %[qc4], %[t4], %[t3] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
".set pop \n\t"
: [qc1]"+r"(qc1), [qc2]"+r"(qc2),
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
[qc3]"+r"(qc3), [qc4]"+r"(qc4),
[t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),
[t4]"=&r"(t4)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
);
curidx = 13 * qc1;
curidx += qc2;
curidx2 = 13 * qc3;
curidx2 += qc4;
curbits += p_bits[curidx] +
p_bits[curidx2] +
upair12_sign_bits[curidx] +
upair12_sign_bits[curidx2];
}
return curbits;
}
static float get_band_numbits_ESC_mips(struct AACEncContext *s,
PutBitContext *pb, const float *in,
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
int *bits)
{
const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];
int i;
int qc1, qc2, qc3, qc4;
int curbits = 0;
uint8_t *p_bits = (uint8_t*)ff_aac_spectral_bits[cb-1];
for (i = 0; i < size; i += 4) {
int curidx, curidx2;
int cond0, cond1, cond2, cond3;
int c1, c2, c3, c4;
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
int t4, t5;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
qc1 = scaled[i ] * Q34 + ROUND_STANDARD;
qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;
qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;
qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"ori %[t4], $zero, 15 \n\t"
"ori %[t5], $zero, 16 \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
"shll_s.w %[c1], %[qc1], 18 \n\t"
"shll_s.w %[c2], %[qc2], 18 \n\t"
"shll_s.w %[c3], %[qc3], 18 \n\t"
"shll_s.w %[c4], %[qc4], 18 \n\t"
"srl %[c1], %[c1], 18 \n\t"
"srl %[c2], %[c2], 18 \n\t"
"srl %[c3], %[c3], 18 \n\t"
"srl %[c4], %[c4], 18 \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"slt %[cond0], %[t4], %[qc1] \n\t"
"slt %[cond1], %[t4], %[qc2] \n\t"
"slt %[cond2], %[t4], %[qc3] \n\t"
"slt %[cond3], %[t4], %[qc4] \n\t"
"movn %[qc1], %[t5], %[cond0] \n\t"
"movn %[qc2], %[t5], %[cond1] \n\t"
"movn %[qc3], %[t5], %[cond2] \n\t"
"movn %[qc4], %[t5], %[cond3] \n\t"
"ori %[t5], $zero, 31 \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
"clz %[c1], %[c1] \n\t"
"clz %[c2], %[c2] \n\t"
"clz %[c3], %[c3] \n\t"
"clz %[c4], %[c4] \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"subu %[c1], %[t5], %[c1] \n\t"
"subu %[c2], %[t5], %[c2] \n\t"
"subu %[c3], %[t5], %[c3] \n\t"
"subu %[c4], %[t5], %[c4] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
"sll %[c1], %[c1], 1 \n\t"
"sll %[c2], %[c2], 1 \n\t"
"sll %[c3], %[c3], 1 \n\t"
"sll %[c4], %[c4], 1 \n\t"
"addiu %[c1], %[c1], -3 \n\t"
"addiu %[c2], %[c2], -3 \n\t"
"addiu %[c3], %[c3], -3 \n\t"
"addiu %[c4], %[c4], -3 \n\t"
"subu %[cond0], $zero, %[cond0] \n\t"
"subu %[cond1], $zero, %[cond1] \n\t"
"subu %[cond2], $zero, %[cond2] \n\t"
"subu %[cond3], $zero, %[cond3] \n\t"
"and %[c1], %[c1], %[cond0] \n\t"
"and %[c2], %[c2], %[cond1] \n\t"
"and %[c3], %[c3], %[cond2] \n\t"
"and %[c4], %[c4], %[cond3] \n\t"
".set pop \n\t"
: [qc1]"+r"(qc1), [qc2]"+r"(qc2),
[qc3]"+r"(qc3), [qc4]"+r"(qc4),
[cond0]"=&r"(cond0), [cond1]"=&r"(cond1),
[cond2]"=&r"(cond2), [cond3]"=&r"(cond3),
[c1]"=&r"(c1), [c2]"=&r"(c2),
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
[c3]"=&r"(c3), [c4]"=&r"(c4),
[t4]"=&r"(t4), [t5]"=&r"(t5)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
);
curidx = 17 * qc1;
curidx += qc2;
curidx2 = 17 * qc3;
curidx2 += qc4;
curbits += p_bits[curidx];
curbits += esc_sign_bits[curidx];
curbits += p_bits[curidx2];
curbits += esc_sign_bits[curidx2];
curbits += c1;
curbits += c2;
curbits += c3;
curbits += c4;
}
return curbits;
}
static float (*const get_band_numbits_arr[])(struct AACEncContext *s,
PutBitContext *pb, const float *in,
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
int *bits) = {
get_band_numbits_ZERO_mips,
get_band_numbits_SQUAD_mips,
get_band_numbits_SQUAD_mips,
get_band_numbits_UQUAD_mips,
get_band_numbits_UQUAD_mips,
get_band_numbits_SPAIR_mips,
get_band_numbits_SPAIR_mips,
get_band_numbits_UPAIR7_mips,
get_band_numbits_UPAIR7_mips,
get_band_numbits_UPAIR12_mips,
get_band_numbits_UPAIR12_mips,
get_band_numbits_ESC_mips,
AAC: MIPS: Add missing codebooks in quantize funcs Add entries on the quantize function tables for the missing codebooks (which all behave like ZERO)
2015-09-03 15:49:17 +02:00
get_band_numbits_NONE_mips, /* cb 12 doesn't exist */
get_band_numbits_ZERO_mips,
get_band_numbits_ZERO_mips,
get_band_numbits_ZERO_mips,
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
};
#define get_band_numbits( \
s, pb, in, scaled, size, scale_idx, cb, \
lambda, uplim, bits) \
get_band_numbits_arr[cb]( \
s, pb, in, scaled, size, scale_idx, cb, \
lambda, uplim, bits)
static float quantize_band_cost_bits(struct AACEncContext *s, const float *in,
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
int *bits, float *energy, int rtz)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
{
return get_band_numbits(s, NULL, in, scaled, size, scale_idx, cb, lambda, uplim, bits);
}
/**
* Functions developed from template function and optimized for getting the band cost
*/
#if HAVE_MIPSFPU
static float get_band_cost_ZERO_mips(struct AACEncContext *s,
PutBitContext *pb, const float *in,
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
int *bits, float *energy)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
{
int i;
float cost = 0;
for (i = 0; i < size; i += 4) {
cost += in[i ] * in[i ];
cost += in[i+1] * in[i+1];
cost += in[i+2] * in[i+2];
cost += in[i+3] * in[i+3];
}
if (bits)
*bits = 0;
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
if (energy)
*energy = 0.0f;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
return cost * lambda;
}
AAC: MIPS: Add missing codebooks in quantize funcs Add entries on the quantize function tables for the missing codebooks (which all behave like ZERO)
2015-09-03 15:49:17 +02:00
static float get_band_cost_NONE_mips(struct AACEncContext *s,
PutBitContext *pb, const float *in,
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
int *bits, float *energy)
AAC: MIPS: Add missing codebooks in quantize funcs Add entries on the quantize function tables for the missing codebooks (which all behave like ZERO)
2015-09-03 15:49:17 +02:00
{
av_assert0(0);
return 0;
}
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
static float get_band_cost_SQUAD_mips(struct AACEncContext *s,
PutBitContext *pb, const float *in,
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
int *bits, float *energy)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
{
const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];
const float IQ = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];
int i;
float cost = 0;
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
float qenergy = 0.0f;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
int qc1, qc2, qc3, qc4;
int curbits = 0;
uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1];
float *p_codes = (float *)ff_aac_codebook_vectors[cb-1];
for (i = 0; i < size; i += 4) {
const float *vec;
int curidx;
int *in_int = (int *)&in[i];
float *in_pos = (float *)&in[i];
float di0, di1, di2, di3;
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
int t0, t1, t2, t3, t4, t5, t6, t7;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
qc1 = scaled[i ] * Q34 + ROUND_STANDARD;
qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;
qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;
qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
"slt %[qc1], $zero, %[qc1] \n\t"
"slt %[qc2], $zero, %[qc2] \n\t"
"slt %[qc3], $zero, %[qc3] \n\t"
"slt %[qc4], $zero, %[qc4] \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"lw %[t0], 0(%[in_int]) \n\t"
"lw %[t1], 4(%[in_int]) \n\t"
"lw %[t2], 8(%[in_int]) \n\t"
"lw %[t3], 12(%[in_int]) \n\t"
"srl %[t0], %[t0], 31 \n\t"
"srl %[t1], %[t1], 31 \n\t"
"srl %[t2], %[t2], 31 \n\t"
"srl %[t3], %[t3], 31 \n\t"
"subu %[t4], $zero, %[qc1] \n\t"
"subu %[t5], $zero, %[qc2] \n\t"
"subu %[t6], $zero, %[qc3] \n\t"
"subu %[t7], $zero, %[qc4] \n\t"
"movn %[qc1], %[t4], %[t0] \n\t"
"movn %[qc2], %[t5], %[t1] \n\t"
"movn %[qc3], %[t6], %[t2] \n\t"
"movn %[qc4], %[t7], %[t3] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
".set pop \n\t"
: [qc1]"+r"(qc1), [qc2]"+r"(qc2),
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
[qc3]"+r"(qc3), [qc4]"+r"(qc4),
[t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),
[t4]"=&r"(t4), [t5]"=&r"(t5), [t6]"=&r"(t6), [t7]"=&r"(t7)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
: [in_int]"r"(in_int)
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
: "memory"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
);
curidx = qc1;
curidx *= 3;
curidx += qc2;
curidx *= 3;
curidx += qc3;
curidx *= 3;
curidx += qc4;
curidx += 40;
curbits += p_bits[curidx];
vec = &p_codes[curidx*4];
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
qenergy += vec[0]*vec[0] + vec[1]*vec[1]
+ vec[2]*vec[2] + vec[3]*vec[3];
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
"lwc1 $f0, 0(%[in_pos]) \n\t"
"lwc1 $f1, 0(%[vec]) \n\t"
"lwc1 $f2, 4(%[in_pos]) \n\t"
"lwc1 $f3, 4(%[vec]) \n\t"
"lwc1 $f4, 8(%[in_pos]) \n\t"
"lwc1 $f5, 8(%[vec]) \n\t"
"lwc1 $f6, 12(%[in_pos]) \n\t"
"lwc1 $f7, 12(%[vec]) \n\t"
"nmsub.s %[di0], $f0, $f1, %[IQ] \n\t"
"nmsub.s %[di1], $f2, $f3, %[IQ] \n\t"
"nmsub.s %[di2], $f4, $f5, %[IQ] \n\t"
"nmsub.s %[di3], $f6, $f7, %[IQ] \n\t"
".set pop \n\t"
: [di0]"=&f"(di0), [di1]"=&f"(di1),
[di2]"=&f"(di2), [di3]"=&f"(di3)
: [in_pos]"r"(in_pos), [vec]"r"(vec),
[IQ]"f"(IQ)
: "$f0", "$f1", "$f2", "$f3",
"$f4", "$f5", "$f6", "$f7",
"memory"
);
cost += di0 * di0 + di1 * di1
+ di2 * di2 + di3 * di3;
}
if (bits)
*bits = curbits;
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
if (energy)
*energy = qenergy * (IQ*IQ);
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
return cost * lambda + curbits;
}
static float get_band_cost_UQUAD_mips(struct AACEncContext *s,
PutBitContext *pb, const float *in,
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
int *bits, float *energy)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
{
const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];
const float IQ = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];
int i;
float cost = 0;
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
float qenergy = 0.0f;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
int curbits = 0;
int qc1, qc2, qc3, qc4;
uint8_t *p_bits = (uint8_t*)ff_aac_spectral_bits[cb-1];
float *p_codes = (float *)ff_aac_codebook_vectors[cb-1];
for (i = 0; i < size; i += 4) {
const float *vec;
int curidx;
float *in_pos = (float *)&in[i];
float di0, di1, di2, di3;
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
int t0, t1, t2, t3, t4;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
qc1 = scaled[i ] * Q34 + ROUND_STANDARD;
qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;
qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;
qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"ori %[t4], $zero, 2 \n\t"
"slt %[t0], %[t4], %[qc1] \n\t"
"slt %[t1], %[t4], %[qc2] \n\t"
"slt %[t2], %[t4], %[qc3] \n\t"
"slt %[t3], %[t4], %[qc4] \n\t"
"movn %[qc1], %[t4], %[t0] \n\t"
"movn %[qc2], %[t4], %[t1] \n\t"
"movn %[qc3], %[t4], %[t2] \n\t"
"movn %[qc4], %[t4], %[t3] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
".set pop \n\t"
: [qc1]"+r"(qc1), [qc2]"+r"(qc2),
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
[qc3]"+r"(qc3), [qc4]"+r"(qc4),
[t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),
[t4]"=&r"(t4)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
);
curidx = qc1;
curidx *= 3;
curidx += qc2;
curidx *= 3;
curidx += qc3;
curidx *= 3;
curidx += qc4;
curbits += p_bits[curidx];
curbits += uquad_sign_bits[curidx];
vec = &p_codes[curidx*4];
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
qenergy += vec[0]*vec[0] + vec[1]*vec[1]
+ vec[2]*vec[2] + vec[3]*vec[3];
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
"lwc1 %[di0], 0(%[in_pos]) \n\t"
"lwc1 %[di1], 4(%[in_pos]) \n\t"
"lwc1 %[di2], 8(%[in_pos]) \n\t"
"lwc1 %[di3], 12(%[in_pos]) \n\t"
"abs.s %[di0], %[di0] \n\t"
"abs.s %[di1], %[di1] \n\t"
"abs.s %[di2], %[di2] \n\t"
"abs.s %[di3], %[di3] \n\t"
"lwc1 $f0, 0(%[vec]) \n\t"
"lwc1 $f1, 4(%[vec]) \n\t"
"lwc1 $f2, 8(%[vec]) \n\t"
"lwc1 $f3, 12(%[vec]) \n\t"
"nmsub.s %[di0], %[di0], $f0, %[IQ] \n\t"
"nmsub.s %[di1], %[di1], $f1, %[IQ] \n\t"
"nmsub.s %[di2], %[di2], $f2, %[IQ] \n\t"
"nmsub.s %[di3], %[di3], $f3, %[IQ] \n\t"
".set pop \n\t"
: [di0]"=&f"(di0), [di1]"=&f"(di1),
[di2]"=&f"(di2), [di3]"=&f"(di3)
: [in_pos]"r"(in_pos), [vec]"r"(vec),
[IQ]"f"(IQ)
: "$f0", "$f1", "$f2", "$f3",
"memory"
);
cost += di0 * di0 + di1 * di1
+ di2 * di2 + di3 * di3;
}
if (bits)
*bits = curbits;
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
if (energy)
*energy = qenergy * (IQ*IQ);
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
return cost * lambda + curbits;
}
static float get_band_cost_SPAIR_mips(struct AACEncContext *s,
PutBitContext *pb, const float *in,
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
int *bits, float *energy)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
{
const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];
const float IQ = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];
int i;
float cost = 0;
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
float qenergy = 0.0f;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
int qc1, qc2, qc3, qc4;
int curbits = 0;
uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1];
float *p_codes = (float *)ff_aac_codebook_vectors[cb-1];
for (i = 0; i < size; i += 4) {
const float *vec, *vec2;
int curidx, curidx2;
int *in_int = (int *)&in[i];
float *in_pos = (float *)&in[i];
float di0, di1, di2, di3;
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
int t0, t1, t2, t3, t4, t5, t6, t7;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
qc1 = scaled[i ] * Q34 + ROUND_STANDARD;
qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;
qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;
qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"ori %[t4], $zero, 4 \n\t"
"slt %[t0], %[t4], %[qc1] \n\t"
"slt %[t1], %[t4], %[qc2] \n\t"
"slt %[t2], %[t4], %[qc3] \n\t"
"slt %[t3], %[t4], %[qc4] \n\t"
"movn %[qc1], %[t4], %[t0] \n\t"
"movn %[qc2], %[t4], %[t1] \n\t"
"movn %[qc3], %[t4], %[t2] \n\t"
"movn %[qc4], %[t4], %[t3] \n\t"
"lw %[t0], 0(%[in_int]) \n\t"
"lw %[t1], 4(%[in_int]) \n\t"
"lw %[t2], 8(%[in_int]) \n\t"
"lw %[t3], 12(%[in_int]) \n\t"
"srl %[t0], %[t0], 31 \n\t"
"srl %[t1], %[t1], 31 \n\t"
"srl %[t2], %[t2], 31 \n\t"
"srl %[t3], %[t3], 31 \n\t"
"subu %[t4], $zero, %[qc1] \n\t"
"subu %[t5], $zero, %[qc2] \n\t"
"subu %[t6], $zero, %[qc3] \n\t"
"subu %[t7], $zero, %[qc4] \n\t"
"movn %[qc1], %[t4], %[t0] \n\t"
"movn %[qc2], %[t5], %[t1] \n\t"
"movn %[qc3], %[t6], %[t2] \n\t"
"movn %[qc4], %[t7], %[t3] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
".set pop \n\t"
: [qc1]"+r"(qc1), [qc2]"+r"(qc2),
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
[qc3]"+r"(qc3), [qc4]"+r"(qc4),
[t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),
[t4]"=&r"(t4), [t5]"=&r"(t5), [t6]"=&r"(t6), [t7]"=&r"(t7)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
: [in_int]"r"(in_int)
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
: "memory"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
);
curidx = 9 * qc1;
curidx += qc2 + 40;
curidx2 = 9 * qc3;
curidx2 += qc4 + 40;
curbits += p_bits[curidx];
curbits += p_bits[curidx2];
vec = &p_codes[curidx*2];
vec2 = &p_codes[curidx2*2];
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
qenergy += vec[0]*vec[0] + vec[1]*vec[1]
+ vec2[0]*vec2[0] + vec2[1]*vec2[1];
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
"lwc1 $f0, 0(%[in_pos]) \n\t"
"lwc1 $f1, 0(%[vec]) \n\t"
"lwc1 $f2, 4(%[in_pos]) \n\t"
"lwc1 $f3, 4(%[vec]) \n\t"
"lwc1 $f4, 8(%[in_pos]) \n\t"
"lwc1 $f5, 0(%[vec2]) \n\t"
"lwc1 $f6, 12(%[in_pos]) \n\t"
"lwc1 $f7, 4(%[vec2]) \n\t"
"nmsub.s %[di0], $f0, $f1, %[IQ] \n\t"
"nmsub.s %[di1], $f2, $f3, %[IQ] \n\t"
"nmsub.s %[di2], $f4, $f5, %[IQ] \n\t"
"nmsub.s %[di3], $f6, $f7, %[IQ] \n\t"
".set pop \n\t"
: [di0]"=&f"(di0), [di1]"=&f"(di1),
[di2]"=&f"(di2), [di3]"=&f"(di3)
: [in_pos]"r"(in_pos), [vec]"r"(vec),
[vec2]"r"(vec2), [IQ]"f"(IQ)
: "$f0", "$f1", "$f2", "$f3",
"$f4", "$f5", "$f6", "$f7",
"memory"
);
cost += di0 * di0 + di1 * di1
+ di2 * di2 + di3 * di3;
}
if (bits)
*bits = curbits;
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
if (energy)
*energy = qenergy * (IQ*IQ);
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
return cost * lambda + curbits;
}
static float get_band_cost_UPAIR7_mips(struct AACEncContext *s,
PutBitContext *pb, const float *in,
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
int *bits, float *energy)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
{
const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];
const float IQ = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];
int i;
float cost = 0;
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
float qenergy = 0.0f;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
int qc1, qc2, qc3, qc4;
int curbits = 0;
uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1];
float *p_codes = (float *)ff_aac_codebook_vectors[cb-1];
for (i = 0; i < size; i += 4) {
const float *vec, *vec2;
int curidx, curidx2, sign1, count1, sign2, count2;
int *in_int = (int *)&in[i];
float *in_pos = (float *)&in[i];
float di0, di1, di2, di3;
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
int t0, t1, t2, t3, t4;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
qc1 = scaled[i ] * Q34 + ROUND_STANDARD;
qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;
qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;
qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"ori %[t4], $zero, 7 \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
"ori %[sign1], $zero, 0 \n\t"
"ori %[sign2], $zero, 0 \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"slt %[t0], %[t4], %[qc1] \n\t"
"slt %[t1], %[t4], %[qc2] \n\t"
"slt %[t2], %[t4], %[qc3] \n\t"
"slt %[t3], %[t4], %[qc4] \n\t"
"movn %[qc1], %[t4], %[t0] \n\t"
"movn %[qc2], %[t4], %[t1] \n\t"
"movn %[qc3], %[t4], %[t2] \n\t"
"movn %[qc4], %[t4], %[t3] \n\t"
"lw %[t0], 0(%[in_int]) \n\t"
"lw %[t1], 4(%[in_int]) \n\t"
"lw %[t2], 8(%[in_int]) \n\t"
"lw %[t3], 12(%[in_int]) \n\t"
"slt %[t0], %[t0], $zero \n\t"
"movn %[sign1], %[t0], %[qc1] \n\t"
"slt %[t2], %[t2], $zero \n\t"
"movn %[sign2], %[t2], %[qc3] \n\t"
"slt %[t1], %[t1], $zero \n\t"
"sll %[t0], %[sign1], 1 \n\t"
"or %[t0], %[t0], %[t1] \n\t"
"movn %[sign1], %[t0], %[qc2] \n\t"
"slt %[t3], %[t3], $zero \n\t"
"sll %[t0], %[sign2], 1 \n\t"
"or %[t0], %[t0], %[t3] \n\t"
"movn %[sign2], %[t0], %[qc4] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
"slt %[count1], $zero, %[qc1] \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"slt %[t1], $zero, %[qc2] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
"slt %[count2], $zero, %[qc3] \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"slt %[t2], $zero, %[qc4] \n\t"
"addu %[count1], %[count1], %[t1] \n\t"
"addu %[count2], %[count2], %[t2] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
".set pop \n\t"
: [qc1]"+r"(qc1), [qc2]"+r"(qc2),
[qc3]"+r"(qc3), [qc4]"+r"(qc4),
[sign1]"=&r"(sign1), [count1]"=&r"(count1),
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
[sign2]"=&r"(sign2), [count2]"=&r"(count2),
[t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),
[t4]"=&r"(t4)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
: [in_int]"r"(in_int)
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
: "memory"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
);
curidx = 8 * qc1;
curidx += qc2;
curidx2 = 8 * qc3;
curidx2 += qc4;
curbits += p_bits[curidx];
curbits += upair7_sign_bits[curidx];
vec = &p_codes[curidx*2];
curbits += p_bits[curidx2];
curbits += upair7_sign_bits[curidx2];
vec2 = &p_codes[curidx2*2];
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
qenergy += vec[0]*vec[0] + vec[1]*vec[1]
+ vec2[0]*vec2[0] + vec2[1]*vec2[1];
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
"lwc1 %[di0], 0(%[in_pos]) \n\t"
"lwc1 %[di1], 4(%[in_pos]) \n\t"
"lwc1 %[di2], 8(%[in_pos]) \n\t"
"lwc1 %[di3], 12(%[in_pos]) \n\t"
"abs.s %[di0], %[di0] \n\t"
"abs.s %[di1], %[di1] \n\t"
"abs.s %[di2], %[di2] \n\t"
"abs.s %[di3], %[di3] \n\t"
"lwc1 $f0, 0(%[vec]) \n\t"
"lwc1 $f1, 4(%[vec]) \n\t"
"lwc1 $f2, 0(%[vec2]) \n\t"
"lwc1 $f3, 4(%[vec2]) \n\t"
"nmsub.s %[di0], %[di0], $f0, %[IQ] \n\t"
"nmsub.s %[di1], %[di1], $f1, %[IQ] \n\t"
"nmsub.s %[di2], %[di2], $f2, %[IQ] \n\t"
"nmsub.s %[di3], %[di3], $f3, %[IQ] \n\t"
".set pop \n\t"
: [di0]"=&f"(di0), [di1]"=&f"(di1),
[di2]"=&f"(di2), [di3]"=&f"(di3)
: [in_pos]"r"(in_pos), [vec]"r"(vec),
[vec2]"r"(vec2), [IQ]"f"(IQ)
: "$f0", "$f1", "$f2", "$f3",
"memory"
);
cost += di0 * di0 + di1 * di1
+ di2 * di2 + di3 * di3;
}
if (bits)
*bits = curbits;
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
if (energy)
*energy = qenergy * (IQ*IQ);
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
return cost * lambda + curbits;
}
static float get_band_cost_UPAIR12_mips(struct AACEncContext *s,
PutBitContext *pb, const float *in,
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
int *bits, float *energy)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
{
const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];
const float IQ = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];
int i;
float cost = 0;
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
float qenergy = 0.0f;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
int qc1, qc2, qc3, qc4;
int curbits = 0;
uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1];
float *p_codes = (float *)ff_aac_codebook_vectors[cb-1];
for (i = 0; i < size; i += 4) {
const float *vec, *vec2;
int curidx, curidx2;
int sign1, count1, sign2, count2;
int *in_int = (int *)&in[i];
float *in_pos = (float *)&in[i];
float di0, di1, di2, di3;
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
int t0, t1, t2, t3, t4;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
qc1 = scaled[i ] * Q34 + ROUND_STANDARD;
qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;
qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;
qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"ori %[t4], $zero, 12 \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
"ori %[sign1], $zero, 0 \n\t"
"ori %[sign2], $zero, 0 \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"slt %[t0], %[t4], %[qc1] \n\t"
"slt %[t1], %[t4], %[qc2] \n\t"
"slt %[t2], %[t4], %[qc3] \n\t"
"slt %[t3], %[t4], %[qc4] \n\t"
"movn %[qc1], %[t4], %[t0] \n\t"
"movn %[qc2], %[t4], %[t1] \n\t"
"movn %[qc3], %[t4], %[t2] \n\t"
"movn %[qc4], %[t4], %[t3] \n\t"
"lw %[t0], 0(%[in_int]) \n\t"
"lw %[t1], 4(%[in_int]) \n\t"
"lw %[t2], 8(%[in_int]) \n\t"
"lw %[t3], 12(%[in_int]) \n\t"
"slt %[t0], %[t0], $zero \n\t"
"movn %[sign1], %[t0], %[qc1] \n\t"
"slt %[t2], %[t2], $zero \n\t"
"movn %[sign2], %[t2], %[qc3] \n\t"
"slt %[t1], %[t1], $zero \n\t"
"sll %[t0], %[sign1], 1 \n\t"
"or %[t0], %[t0], %[t1] \n\t"
"movn %[sign1], %[t0], %[qc2] \n\t"
"slt %[t3], %[t3], $zero \n\t"
"sll %[t0], %[sign2], 1 \n\t"
"or %[t0], %[t0], %[t3] \n\t"
"movn %[sign2], %[t0], %[qc4] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
"slt %[count1], $zero, %[qc1] \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"slt %[t1], $zero, %[qc2] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
"slt %[count2], $zero, %[qc3] \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"slt %[t2], $zero, %[qc4] \n\t"
"addu %[count1], %[count1], %[t1] \n\t"
"addu %[count2], %[count2], %[t2] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
".set pop \n\t"
: [qc1]"+r"(qc1), [qc2]"+r"(qc2),
[qc3]"+r"(qc3), [qc4]"+r"(qc4),
[sign1]"=&r"(sign1), [count1]"=&r"(count1),
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
[sign2]"=&r"(sign2), [count2]"=&r"(count2),
[t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),
[t4]"=&r"(t4)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
: [in_int]"r"(in_int)
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
: "memory"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
);
curidx = 13 * qc1;
curidx += qc2;
curidx2 = 13 * qc3;
curidx2 += qc4;
curbits += p_bits[curidx];
curbits += p_bits[curidx2];
curbits += upair12_sign_bits[curidx];
curbits += upair12_sign_bits[curidx2];
vec = &p_codes[curidx*2];
vec2 = &p_codes[curidx2*2];
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
qenergy += vec[0]*vec[0] + vec[1]*vec[1]
+ vec2[0]*vec2[0] + vec2[1]*vec2[1];
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
"lwc1 %[di0], 0(%[in_pos]) \n\t"
"lwc1 %[di1], 4(%[in_pos]) \n\t"
"lwc1 %[di2], 8(%[in_pos]) \n\t"
"lwc1 %[di3], 12(%[in_pos]) \n\t"
"abs.s %[di0], %[di0] \n\t"
"abs.s %[di1], %[di1] \n\t"
"abs.s %[di2], %[di2] \n\t"
"abs.s %[di3], %[di3] \n\t"
"lwc1 $f0, 0(%[vec]) \n\t"
"lwc1 $f1, 4(%[vec]) \n\t"
"lwc1 $f2, 0(%[vec2]) \n\t"
"lwc1 $f3, 4(%[vec2]) \n\t"
"nmsub.s %[di0], %[di0], $f0, %[IQ] \n\t"
"nmsub.s %[di1], %[di1], $f1, %[IQ] \n\t"
"nmsub.s %[di2], %[di2], $f2, %[IQ] \n\t"
"nmsub.s %[di3], %[di3], $f3, %[IQ] \n\t"
".set pop \n\t"
: [di0]"=&f"(di0), [di1]"=&f"(di1),
[di2]"=&f"(di2), [di3]"=&f"(di3)
: [in_pos]"r"(in_pos), [vec]"r"(vec),
[vec2]"r"(vec2), [IQ]"f"(IQ)
: "$f0", "$f1", "$f2", "$f3",
"memory"
);
cost += di0 * di0 + di1 * di1
+ di2 * di2 + di3 * di3;
}
if (bits)
*bits = curbits;
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
if (energy)
*energy = qenergy * (IQ*IQ);
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
return cost * lambda + curbits;
}
static float get_band_cost_ESC_mips(struct AACEncContext *s,
PutBitContext *pb, const float *in,
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
int *bits, float *energy)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
{
const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];
const float IQ = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];
const float CLIPPED_ESCAPE = 165140.0f * IQ;
int i;
float cost = 0;
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
float qenergy = 0.0f;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
int qc1, qc2, qc3, qc4;
int curbits = 0;
uint8_t *p_bits = (uint8_t*)ff_aac_spectral_bits[cb-1];
float *p_codes = (float* )ff_aac_codebook_vectors[cb-1];
for (i = 0; i < size; i += 4) {
const float *vec, *vec2;
int curidx, curidx2;
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
float t1, t2, t3, t4, V;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
float di1, di2, di3, di4;
int cond0, cond1, cond2, cond3;
int c1, c2, c3, c4;
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
int t6, t7;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
qc1 = scaled[i ] * Q34 + ROUND_STANDARD;
qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;
qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;
qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"ori %[t6], $zero, 15 \n\t"
"ori %[t7], $zero, 16 \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
"shll_s.w %[c1], %[qc1], 18 \n\t"
"shll_s.w %[c2], %[qc2], 18 \n\t"
"shll_s.w %[c3], %[qc3], 18 \n\t"
"shll_s.w %[c4], %[qc4], 18 \n\t"
"srl %[c1], %[c1], 18 \n\t"
"srl %[c2], %[c2], 18 \n\t"
"srl %[c3], %[c3], 18 \n\t"
"srl %[c4], %[c4], 18 \n\t"
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
"slt %[cond0], %[t6], %[qc1] \n\t"
"slt %[cond1], %[t6], %[qc2] \n\t"
"slt %[cond2], %[t6], %[qc3] \n\t"
"slt %[cond3], %[t6], %[qc4] \n\t"
"movn %[qc1], %[t7], %[cond0] \n\t"
"movn %[qc2], %[t7], %[cond1] \n\t"
"movn %[qc3], %[t7], %[cond2] \n\t"
"movn %[qc4], %[t7], %[cond3] \n\t"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
".set pop \n\t"
: [qc1]"+r"(qc1), [qc2]"+r"(qc2),
[qc3]"+r"(qc3), [qc4]"+r"(qc4),
[cond0]"=&r"(cond0), [cond1]"=&r"(cond1),
[cond2]"=&r"(cond2), [cond3]"=&r"(cond3),
[c1]"=&r"(c1), [c2]"=&r"(c2),
mips/aaccoder: use variables instead of using register names directly On mips64, the registers t[4-7] do not exist. Instead of using a lot of #ifdef or defines to handle differing register names, use variables and let GCC allocate the registers automatically (like in the other mips assembly files). In get_band_cost_ESC_mips, t4 and t5 were renamed to t6 and t7 to avoid a variable name conflict. Signed-off-by: James Cowgill <james410@cowgill.org.uk> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-02-26 15:42:53 +02:00
[c3]"=&r"(c3), [c4]"=&r"(c4),
[t6]"=&r"(t6), [t7]"=&r"(t7)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
);
curidx = 17 * qc1;
curidx += qc2;
curidx2 = 17 * qc3;
curidx2 += qc4;
curbits += p_bits[curidx];
curbits += esc_sign_bits[curidx];
vec = &p_codes[curidx*2];
curbits += p_bits[curidx2];
curbits += esc_sign_bits[curidx2];
vec2 = &p_codes[curidx2*2];
curbits += (av_log2(c1) * 2 - 3) & (-cond0);
curbits += (av_log2(c2) * 2 - 3) & (-cond1);
curbits += (av_log2(c3) * 2 - 3) & (-cond2);
curbits += (av_log2(c4) * 2 - 3) & (-cond3);
t1 = fabsf(in[i ]);
t2 = fabsf(in[i+1]);
t3 = fabsf(in[i+2]);
t4 = fabsf(in[i+3]);
if (cond0) {
if (t1 >= CLIPPED_ESCAPE) {
di1 = t1 - CLIPPED_ESCAPE;
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
qenergy += CLIPPED_ESCAPE*CLIPPED_ESCAPE;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
} else {
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
di1 = t1 - (V = c1 * cbrtf(c1) * IQ);
qenergy += V*V;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
}
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
} else {
di1 = t1 - (V = vec[0] * IQ);
qenergy += V*V;
}
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
if (cond1) {
if (t2 >= CLIPPED_ESCAPE) {
di2 = t2 - CLIPPED_ESCAPE;
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
qenergy += CLIPPED_ESCAPE*CLIPPED_ESCAPE;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
} else {
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
di2 = t2 - (V = c2 * cbrtf(c2) * IQ);
qenergy += V*V;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
}
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
} else {
di2 = t2 - (V = vec[1] * IQ);
qenergy += V*V;
}
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
if (cond2) {
if (t3 >= CLIPPED_ESCAPE) {
di3 = t3 - CLIPPED_ESCAPE;
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
qenergy += CLIPPED_ESCAPE*CLIPPED_ESCAPE;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
} else {
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
di3 = t3 - (V = c3 * cbrtf(c3) * IQ);
qenergy += V*V;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
}
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
} else {
di3 = t3 - (V = vec2[0] * IQ);
qenergy += V*V;
}
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
if (cond3) {
if (t4 >= CLIPPED_ESCAPE) {
di4 = t4 - CLIPPED_ESCAPE;
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
qenergy += CLIPPED_ESCAPE*CLIPPED_ESCAPE;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
} else {
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
di4 = t4 - (V = c4 * cbrtf(c4) * IQ);
qenergy += V*V;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
}
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
} else {
di4 = t4 - (V = vec2[1]*IQ);
qenergy += V*V;
}
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
cost += di1 * di1 + di2 * di2
+ di3 * di3 + di4 * di4;
}
if (bits)
*bits = curbits;
return cost * lambda + curbits;
}
static float (*const get_band_cost_arr[])(struct AACEncContext *s,
PutBitContext *pb, const float *in,
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
int *bits, float *energy) = {
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
get_band_cost_ZERO_mips,
get_band_cost_SQUAD_mips,
get_band_cost_SQUAD_mips,
get_band_cost_UQUAD_mips,
get_band_cost_UQUAD_mips,
get_band_cost_SPAIR_mips,
get_band_cost_SPAIR_mips,
get_band_cost_UPAIR7_mips,
get_band_cost_UPAIR7_mips,
get_band_cost_UPAIR12_mips,
get_band_cost_UPAIR12_mips,
get_band_cost_ESC_mips,
AAC: MIPS: Add missing codebooks in quantize funcs Add entries on the quantize function tables for the missing codebooks (which all behave like ZERO)
2015-09-03 15:49:17 +02:00
get_band_cost_NONE_mips, /* cb 12 doesn't exist */
get_band_cost_ZERO_mips,
get_band_cost_ZERO_mips,
get_band_cost_ZERO_mips,
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
};
#define get_band_cost( \
s, pb, in, scaled, size, scale_idx, cb, \
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
lambda, uplim, bits, energy) \
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
get_band_cost_arr[cb]( \
s, pb, in, scaled, size, scale_idx, cb, \
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
lambda, uplim, bits, energy)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
static float quantize_band_cost(struct AACEncContext *s, const float *in,
const float *scaled, int size, int scale_idx,
int cb, const float lambda, const float uplim,
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
int *bits, float *energy, int rtz)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
{
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
return get_band_cost(s, NULL, in, scaled, size, scale_idx, cb, lambda, uplim, bits, energy);
}
AAC encoder: memoize quantize_band_cost The bulk of calls to quantize_band_cost are replaced by a call to a version that memoizes, greatly improving performance, since during coefficient search there is a great deal of repeat work. Memoization cannot always be applied, so do this in a different function, and leave the original as-is.
2015-10-12 08:56:22 +02:00
#include "libavcodec/aacenc_quantization_misc.h"
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
#include "libavcodec/aaccoder_twoloop.h"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
aaccoder_mips: update function definitions This commit updates the function definitions in the aaccoder_mips.c file. This was broken around a month or so ago with the addition of the rounding argument. The previous commit in this series also introduced a separate array to put the quantization error in, this also needed to be updated, albeit non-functional, in the MIPS optimized aaccoder file. Credits for the rounding goes to Claudio Freire. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-21 20:00:15 +02:00
static void search_for_ms_mips(AACEncContext *s, ChannelElement *cpe)
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
{
AAC encoder: improve SF range utilization This patch does 4 things, all of which interact and thus it woudln't be possible to commit them separately without causing either quality regressions or assertion failures. Fate comparison targets don't all reflect improvements in quality, yet listening tests show substantially improved quality and stability. 1. Increase SF range utilization. The spec requires SF delta values to be constrained within the range -60..60. The previous code was applying that range to the whole SF array and not only the deltas of consecutive values, because doing so requires smarter code: zeroing or otherwise skipping a band may invalidate lots of SF choices. This patch implements that logic to allow the coders to utilize the full dynamic range of scalefactors, increasing quality quite considerably, and fixing delta-SF-related assertion failures, since now the limitation is enforced rather than asserted. 2. PNS tweaks The previous modification makes big improvements in twoloop's efficiency, and every time that happens PNS logic needs to be tweaked accordingly to avoid it from stepping all over twoloop's decisions. This patch includes modifications of the sort. 3. Account for lowpass cutoff during PSY analysis The closer PSY's allocation is to final allocation the better the quality is, and given these modifications, twoloop is now very efficient at avoiding holes. Thus, to compute accurate thresholds, PSY needs to account for the lowpass applied implicitly during twoloop (by zeroing high bands). This patch makes twoloop set the cutoff in psymodel's context the first time it runs, and makes PSY account for it during threshold computation, making PE and threshold computations closer to the final allocation and thus achieving better subjective quality. 4. Tweaks to RC lambda tracking loop in relation to PNS Without this tweak some corner cases cause quality regressions. Basically, lambda needs to react faster to overall bitrate efficiency changes since now PNS can be quite successful in enforcing maximum bitrates, when PSY allocates too many bits to the lower bands, suppressing the signals RC logic uses to lower lambda in those cases and causing aggressive PNS. This tweak makes PNS much less aggressive, though it can still use some further tweaks. Also update MIPS specializations and adjust fuzz Also in lavc/mips/aacpsy_mips.h: remove trailing whitespace
2015-12-01 08:28:36 +02:00
int start = 0, i, w, w2, g, sid_sf_boost, prev_mid, prev_side;
uint8_t nextband0[128], nextband1[128];
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
float M[128], S[128];
float *L34 = s->scoefs, *R34 = s->scoefs + 128, *M34 = s->scoefs + 128*2, *S34 = s->scoefs + 128*3;
avcodec/mips/aaccoder_mips: Sync with generic aaccoder file. Code in aaccoder_mips.c was not synced with changes in aaccoder.c for some time. That was cause for some fate-aac tests failing. This patch fixes the problems. Optimizations disabled in 933309a are enabled again. Signed-off-by: Nedeljko Babic <nedeljko.babic@imgtec.com> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2015-09-11 15:16:16 +02:00
const float lambda = s->lambda;
AAC encoder: improve SF range utilization This patch does 4 things, all of which interact and thus it woudln't be possible to commit them separately without causing either quality regressions or assertion failures. Fate comparison targets don't all reflect improvements in quality, yet listening tests show substantially improved quality and stability. 1. Increase SF range utilization. The spec requires SF delta values to be constrained within the range -60..60. The previous code was applying that range to the whole SF array and not only the deltas of consecutive values, because doing so requires smarter code: zeroing or otherwise skipping a band may invalidate lots of SF choices. This patch implements that logic to allow the coders to utilize the full dynamic range of scalefactors, increasing quality quite considerably, and fixing delta-SF-related assertion failures, since now the limitation is enforced rather than asserted. 2. PNS tweaks The previous modification makes big improvements in twoloop's efficiency, and every time that happens PNS logic needs to be tweaked accordingly to avoid it from stepping all over twoloop's decisions. This patch includes modifications of the sort. 3. Account for lowpass cutoff during PSY analysis The closer PSY's allocation is to final allocation the better the quality is, and given these modifications, twoloop is now very efficient at avoiding holes. Thus, to compute accurate thresholds, PSY needs to account for the lowpass applied implicitly during twoloop (by zeroing high bands). This patch makes twoloop set the cutoff in psymodel's context the first time it runs, and makes PSY account for it during threshold computation, making PE and threshold computations closer to the final allocation and thus achieving better subjective quality. 4. Tweaks to RC lambda tracking loop in relation to PNS Without this tweak some corner cases cause quality regressions. Basically, lambda needs to react faster to overall bitrate efficiency changes since now PNS can be quite successful in enforcing maximum bitrates, when PSY allocates too many bits to the lower bands, suppressing the signals RC logic uses to lower lambda in those cases and causing aggressive PNS. This tweak makes PNS much less aggressive, though it can still use some further tweaks. Also update MIPS specializations and adjust fuzz Also in lavc/mips/aacpsy_mips.h: remove trailing whitespace
2015-12-01 08:28:36 +02:00
const float mslambda = FFMIN(1.0f, lambda / 120.f);
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
SingleChannelElement *sce0 = &cpe->ch[0];
SingleChannelElement *sce1 = &cpe->ch[1];
if (!cpe->common_window)
return;
AAC encoder: improve SF range utilization This patch does 4 things, all of which interact and thus it woudln't be possible to commit them separately without causing either quality regressions or assertion failures. Fate comparison targets don't all reflect improvements in quality, yet listening tests show substantially improved quality and stability. 1. Increase SF range utilization. The spec requires SF delta values to be constrained within the range -60..60. The previous code was applying that range to the whole SF array and not only the deltas of consecutive values, because doing so requires smarter code: zeroing or otherwise skipping a band may invalidate lots of SF choices. This patch implements that logic to allow the coders to utilize the full dynamic range of scalefactors, increasing quality quite considerably, and fixing delta-SF-related assertion failures, since now the limitation is enforced rather than asserted. 2. PNS tweaks The previous modification makes big improvements in twoloop's efficiency, and every time that happens PNS logic needs to be tweaked accordingly to avoid it from stepping all over twoloop's decisions. This patch includes modifications of the sort. 3. Account for lowpass cutoff during PSY analysis The closer PSY's allocation is to final allocation the better the quality is, and given these modifications, twoloop is now very efficient at avoiding holes. Thus, to compute accurate thresholds, PSY needs to account for the lowpass applied implicitly during twoloop (by zeroing high bands). This patch makes twoloop set the cutoff in psymodel's context the first time it runs, and makes PSY account for it during threshold computation, making PE and threshold computations closer to the final allocation and thus achieving better subjective quality. 4. Tweaks to RC lambda tracking loop in relation to PNS Without this tweak some corner cases cause quality regressions. Basically, lambda needs to react faster to overall bitrate efficiency changes since now PNS can be quite successful in enforcing maximum bitrates, when PSY allocates too many bits to the lower bands, suppressing the signals RC logic uses to lower lambda in those cases and causing aggressive PNS. This tweak makes PNS much less aggressive, though it can still use some further tweaks. Also update MIPS specializations and adjust fuzz Also in lavc/mips/aacpsy_mips.h: remove trailing whitespace
2015-12-01 08:28:36 +02:00
/** Scout out next nonzero bands */
ff_init_nextband_map(sce0, nextband0);
ff_init_nextband_map(sce1, nextband1);
prev_mid = sce0->sf_idx[0];
prev_side = sce1->sf_idx[0];
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
for (w = 0; w < sce0->ics.num_windows; w += sce0->ics.group_len[w]) {
avcodec/mips/aaccoder_mips: Sync with generic aaccoder file. Code in aaccoder_mips.c was not synced with changes in aaccoder.c for some time. That was cause for some fate-aac tests failing. This patch fixes the problems. Optimizations disabled in 933309a are enabled again. Signed-off-by: Nedeljko Babic <nedeljko.babic@imgtec.com> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2015-09-11 15:16:16 +02:00
start = 0;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
for (g = 0; g < sce0->ics.num_swb; g++) {
AAC encoder: improve SF range utilization This patch does 4 things, all of which interact and thus it woudln't be possible to commit them separately without causing either quality regressions or assertion failures. Fate comparison targets don't all reflect improvements in quality, yet listening tests show substantially improved quality and stability. 1. Increase SF range utilization. The spec requires SF delta values to be constrained within the range -60..60. The previous code was applying that range to the whole SF array and not only the deltas of consecutive values, because doing so requires smarter code: zeroing or otherwise skipping a band may invalidate lots of SF choices. This patch implements that logic to allow the coders to utilize the full dynamic range of scalefactors, increasing quality quite considerably, and fixing delta-SF-related assertion failures, since now the limitation is enforced rather than asserted. 2. PNS tweaks The previous modification makes big improvements in twoloop's efficiency, and every time that happens PNS logic needs to be tweaked accordingly to avoid it from stepping all over twoloop's decisions. This patch includes modifications of the sort. 3. Account for lowpass cutoff during PSY analysis The closer PSY's allocation is to final allocation the better the quality is, and given these modifications, twoloop is now very efficient at avoiding holes. Thus, to compute accurate thresholds, PSY needs to account for the lowpass applied implicitly during twoloop (by zeroing high bands). This patch makes twoloop set the cutoff in psymodel's context the first time it runs, and makes PSY account for it during threshold computation, making PE and threshold computations closer to the final allocation and thus achieving better subjective quality. 4. Tweaks to RC lambda tracking loop in relation to PNS Without this tweak some corner cases cause quality regressions. Basically, lambda needs to react faster to overall bitrate efficiency changes since now PNS can be quite successful in enforcing maximum bitrates, when PSY allocates too many bits to the lower bands, suppressing the signals RC logic uses to lower lambda in those cases and causing aggressive PNS. This tweak makes PNS much less aggressive, though it can still use some further tweaks. Also update MIPS specializations and adjust fuzz Also in lavc/mips/aacpsy_mips.h: remove trailing whitespace
2015-12-01 08:28:36 +02:00
float bmax = bval2bmax(g * 17.0f / sce0->ics.num_swb) / 0.0045f;
AAC encoder: various fixes in M/S coding 1. Fix sf_idx and band_type addressing to address only the first subwindow in the group (others could hold garbage values) 2. Don't step on ms_mask when is_mask is set. I/S selection already sets the ms_mask properly and shouldn't be overridden. 3. Use mid/sid cb/sf when computing coding error, as should be since those are the cb/sfs that will eventually be set. 4. Fix distortion computation on multi-subwindow groups (was subtracting the bits terms multiple times) 5. Clear ms_mask when one side uses PNS and the other doesn't. When using PNS, ms_mask signals correlated noise, which can be detected just like regular M/S detection, so we don't skip noise bands, but when only one side uses PNS setting the flag can confuse some encoders, so avoid that.
2016-01-08 10:31:32 +02:00
if (!cpe->is_mask[w*16+g])
cpe->ms_mask[w*16+g] = 0;
if (!sce0->zeroes[w*16+g] && !sce1->zeroes[w*16+g] && !cpe->is_mask[w*16+g]) {
AAC encoder: improve SF range utilization This patch does 4 things, all of which interact and thus it woudln't be possible to commit them separately without causing either quality regressions or assertion failures. Fate comparison targets don't all reflect improvements in quality, yet listening tests show substantially improved quality and stability. 1. Increase SF range utilization. The spec requires SF delta values to be constrained within the range -60..60. The previous code was applying that range to the whole SF array and not only the deltas of consecutive values, because doing so requires smarter code: zeroing or otherwise skipping a band may invalidate lots of SF choices. This patch implements that logic to allow the coders to utilize the full dynamic range of scalefactors, increasing quality quite considerably, and fixing delta-SF-related assertion failures, since now the limitation is enforced rather than asserted. 2. PNS tweaks The previous modification makes big improvements in twoloop's efficiency, and every time that happens PNS logic needs to be tweaked accordingly to avoid it from stepping all over twoloop's decisions. This patch includes modifications of the sort. 3. Account for lowpass cutoff during PSY analysis The closer PSY's allocation is to final allocation the better the quality is, and given these modifications, twoloop is now very efficient at avoiding holes. Thus, to compute accurate thresholds, PSY needs to account for the lowpass applied implicitly during twoloop (by zeroing high bands). This patch makes twoloop set the cutoff in psymodel's context the first time it runs, and makes PSY account for it during threshold computation, making PE and threshold computations closer to the final allocation and thus achieving better subjective quality. 4. Tweaks to RC lambda tracking loop in relation to PNS Without this tweak some corner cases cause quality regressions. Basically, lambda needs to react faster to overall bitrate efficiency changes since now PNS can be quite successful in enforcing maximum bitrates, when PSY allocates too many bits to the lower bands, suppressing the signals RC logic uses to lower lambda in those cases and causing aggressive PNS. This tweak makes PNS much less aggressive, though it can still use some further tweaks. Also update MIPS specializations and adjust fuzz Also in lavc/mips/aacpsy_mips.h: remove trailing whitespace
2015-12-01 08:28:36 +02:00
float Mmax = 0.0f, Smax = 0.0f;
/* Must compute mid/side SF and book for the whole window group */
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
for (w2 = 0; w2 < sce0->ics.group_len[w]; w2++) {
AAC encoder: improve SF range utilization This patch does 4 things, all of which interact and thus it woudln't be possible to commit them separately without causing either quality regressions or assertion failures. Fate comparison targets don't all reflect improvements in quality, yet listening tests show substantially improved quality and stability. 1. Increase SF range utilization. The spec requires SF delta values to be constrained within the range -60..60. The previous code was applying that range to the whole SF array and not only the deltas of consecutive values, because doing so requires smarter code: zeroing or otherwise skipping a band may invalidate lots of SF choices. This patch implements that logic to allow the coders to utilize the full dynamic range of scalefactors, increasing quality quite considerably, and fixing delta-SF-related assertion failures, since now the limitation is enforced rather than asserted. 2. PNS tweaks The previous modification makes big improvements in twoloop's efficiency, and every time that happens PNS logic needs to be tweaked accordingly to avoid it from stepping all over twoloop's decisions. This patch includes modifications of the sort. 3. Account for lowpass cutoff during PSY analysis The closer PSY's allocation is to final allocation the better the quality is, and given these modifications, twoloop is now very efficient at avoiding holes. Thus, to compute accurate thresholds, PSY needs to account for the lowpass applied implicitly during twoloop (by zeroing high bands). This patch makes twoloop set the cutoff in psymodel's context the first time it runs, and makes PSY account for it during threshold computation, making PE and threshold computations closer to the final allocation and thus achieving better subjective quality. 4. Tweaks to RC lambda tracking loop in relation to PNS Without this tweak some corner cases cause quality regressions. Basically, lambda needs to react faster to overall bitrate efficiency changes since now PNS can be quite successful in enforcing maximum bitrates, when PSY allocates too many bits to the lower bands, suppressing the signals RC logic uses to lower lambda in those cases and causing aggressive PNS. This tweak makes PNS much less aggressive, though it can still use some further tweaks. Also update MIPS specializations and adjust fuzz Also in lavc/mips/aacpsy_mips.h: remove trailing whitespace
2015-12-01 08:28:36 +02:00
for (i = 0; i < sce0->ics.swb_sizes[g]; i++) {
M[i] = (sce0->coeffs[start+(w+w2)*128+i]
+ sce1->coeffs[start+(w+w2)*128+i]) * 0.5;
S[i] = M[i]
- sce1->coeffs[start+(w+w2)*128+i];
}
abs_pow34_v(M34, M, sce0->ics.swb_sizes[g]);
abs_pow34_v(S34, S, sce0->ics.swb_sizes[g]);
for (i = 0; i < sce0->ics.swb_sizes[g]; i++ ) {
Mmax = FFMAX(Mmax, M34[i]);
Smax = FFMAX(Smax, S34[i]);
}
}
for (sid_sf_boost = 0; sid_sf_boost < 4; sid_sf_boost++) {
float dist1 = 0.0f, dist2 = 0.0f;
int B0 = 0, B1 = 0;
int minidx;
int mididx, sididx;
int midcb, sidcb;
minidx = FFMIN(sce0->sf_idx[w*16+g], sce1->sf_idx[w*16+g]);
mididx = av_clip(minidx, 0, SCALE_MAX_POS - SCALE_DIV_512);
sididx = av_clip(minidx - sid_sf_boost * 3, 0, SCALE_MAX_POS - SCALE_DIV_512);
AAC encoder: various fixes in M/S coding 1. Fix sf_idx and band_type addressing to address only the first subwindow in the group (others could hold garbage values) 2. Don't step on ms_mask when is_mask is set. I/S selection already sets the ms_mask properly and shouldn't be overridden. 3. Use mid/sid cb/sf when computing coding error, as should be since those are the cb/sfs that will eventually be set. 4. Fix distortion computation on multi-subwindow groups (was subtracting the bits terms multiple times) 5. Clear ms_mask when one side uses PNS and the other doesn't. When using PNS, ms_mask signals correlated noise, which can be detected just like regular M/S detection, so we don't skip noise bands, but when only one side uses PNS setting the flag can confuse some encoders, so avoid that.
2016-01-08 10:31:32 +02:00
if (sce0->band_type[w*16+g] != NOISE_BT && sce1->band_type[w*16+g] != NOISE_BT
AAC encoder: improve SF range utilization This patch does 4 things, all of which interact and thus it woudln't be possible to commit them separately without causing either quality regressions or assertion failures. Fate comparison targets don't all reflect improvements in quality, yet listening tests show substantially improved quality and stability. 1. Increase SF range utilization. The spec requires SF delta values to be constrained within the range -60..60. The previous code was applying that range to the whole SF array and not only the deltas of consecutive values, because doing so requires smarter code: zeroing or otherwise skipping a band may invalidate lots of SF choices. This patch implements that logic to allow the coders to utilize the full dynamic range of scalefactors, increasing quality quite considerably, and fixing delta-SF-related assertion failures, since now the limitation is enforced rather than asserted. 2. PNS tweaks The previous modification makes big improvements in twoloop's efficiency, and every time that happens PNS logic needs to be tweaked accordingly to avoid it from stepping all over twoloop's decisions. This patch includes modifications of the sort. 3. Account for lowpass cutoff during PSY analysis The closer PSY's allocation is to final allocation the better the quality is, and given these modifications, twoloop is now very efficient at avoiding holes. Thus, to compute accurate thresholds, PSY needs to account for the lowpass applied implicitly during twoloop (by zeroing high bands). This patch makes twoloop set the cutoff in psymodel's context the first time it runs, and makes PSY account for it during threshold computation, making PE and threshold computations closer to the final allocation and thus achieving better subjective quality. 4. Tweaks to RC lambda tracking loop in relation to PNS Without this tweak some corner cases cause quality regressions. Basically, lambda needs to react faster to overall bitrate efficiency changes since now PNS can be quite successful in enforcing maximum bitrates, when PSY allocates too many bits to the lower bands, suppressing the signals RC logic uses to lower lambda in those cases and causing aggressive PNS. This tweak makes PNS much less aggressive, though it can still use some further tweaks. Also update MIPS specializations and adjust fuzz Also in lavc/mips/aacpsy_mips.h: remove trailing whitespace
2015-12-01 08:28:36 +02:00
&& ( !ff_sfdelta_can_replace(sce0, nextband0, prev_mid, mididx, w*16+g)
|| !ff_sfdelta_can_replace(sce1, nextband1, prev_side, sididx, w*16+g))) {
/* scalefactor range violation, bad stuff, will decrease quality unacceptably */
continue;
}
midcb = find_min_book(Mmax, mididx);
sidcb = find_min_book(Smax, sididx);
/* No CB can be zero */
midcb = FFMAX(1,midcb);
sidcb = FFMAX(1,sidcb);
for (w2 = 0; w2 < sce0->ics.group_len[w]; w2++) {
FFPsyBand *band0 = &s->psy.ch[s->cur_channel+0].psy_bands[(w+w2)*16+g];
FFPsyBand *band1 = &s->psy.ch[s->cur_channel+1].psy_bands[(w+w2)*16+g];
float minthr = FFMIN(band0->threshold, band1->threshold);
int b1,b2,b3,b4;
for (i = 0; i < sce0->ics.swb_sizes[g]; i++) {
M[i] = (sce0->coeffs[start+(w+w2)*128+i]
+ sce1->coeffs[start+(w+w2)*128+i]) * 0.5;
S[i] = M[i]
- sce1->coeffs[start+(w+w2)*128+i];
}
abs_pow34_v(L34, sce0->coeffs+start+(w+w2)*128, sce0->ics.swb_sizes[g]);
abs_pow34_v(R34, sce1->coeffs+start+(w+w2)*128, sce0->ics.swb_sizes[g]);
abs_pow34_v(M34, M, sce0->ics.swb_sizes[g]);
abs_pow34_v(S34, S, sce0->ics.swb_sizes[g]);
dist1 += quantize_band_cost(s, &sce0->coeffs[start + (w+w2)*128],
L34,
sce0->ics.swb_sizes[g],
AAC encoder: various fixes in M/S coding 1. Fix sf_idx and band_type addressing to address only the first subwindow in the group (others could hold garbage values) 2. Don't step on ms_mask when is_mask is set. I/S selection already sets the ms_mask properly and shouldn't be overridden. 3. Use mid/sid cb/sf when computing coding error, as should be since those are the cb/sfs that will eventually be set. 4. Fix distortion computation on multi-subwindow groups (was subtracting the bits terms multiple times) 5. Clear ms_mask when one side uses PNS and the other doesn't. When using PNS, ms_mask signals correlated noise, which can be detected just like regular M/S detection, so we don't skip noise bands, but when only one side uses PNS setting the flag can confuse some encoders, so avoid that.
2016-01-08 10:31:32 +02:00
sce0->sf_idx[w*16+g],
sce0->band_type[w*16+g],
AAC encoder: improve SF range utilization This patch does 4 things, all of which interact and thus it woudln't be possible to commit them separately without causing either quality regressions or assertion failures. Fate comparison targets don't all reflect improvements in quality, yet listening tests show substantially improved quality and stability. 1. Increase SF range utilization. The spec requires SF delta values to be constrained within the range -60..60. The previous code was applying that range to the whole SF array and not only the deltas of consecutive values, because doing so requires smarter code: zeroing or otherwise skipping a band may invalidate lots of SF choices. This patch implements that logic to allow the coders to utilize the full dynamic range of scalefactors, increasing quality quite considerably, and fixing delta-SF-related assertion failures, since now the limitation is enforced rather than asserted. 2. PNS tweaks The previous modification makes big improvements in twoloop's efficiency, and every time that happens PNS logic needs to be tweaked accordingly to avoid it from stepping all over twoloop's decisions. This patch includes modifications of the sort. 3. Account for lowpass cutoff during PSY analysis The closer PSY's allocation is to final allocation the better the quality is, and given these modifications, twoloop is now very efficient at avoiding holes. Thus, to compute accurate thresholds, PSY needs to account for the lowpass applied implicitly during twoloop (by zeroing high bands). This patch makes twoloop set the cutoff in psymodel's context the first time it runs, and makes PSY account for it during threshold computation, making PE and threshold computations closer to the final allocation and thus achieving better subjective quality. 4. Tweaks to RC lambda tracking loop in relation to PNS Without this tweak some corner cases cause quality regressions. Basically, lambda needs to react faster to overall bitrate efficiency changes since now PNS can be quite successful in enforcing maximum bitrates, when PSY allocates too many bits to the lower bands, suppressing the signals RC logic uses to lower lambda in those cases and causing aggressive PNS. This tweak makes PNS much less aggressive, though it can still use some further tweaks. Also update MIPS specializations and adjust fuzz Also in lavc/mips/aacpsy_mips.h: remove trailing whitespace
2015-12-01 08:28:36 +02:00
lambda / band0->threshold, INFINITY, &b1, NULL, 0);
dist1 += quantize_band_cost(s, &sce1->coeffs[start + (w+w2)*128],
R34,
sce1->ics.swb_sizes[g],
AAC encoder: various fixes in M/S coding 1. Fix sf_idx and band_type addressing to address only the first subwindow in the group (others could hold garbage values) 2. Don't step on ms_mask when is_mask is set. I/S selection already sets the ms_mask properly and shouldn't be overridden. 3. Use mid/sid cb/sf when computing coding error, as should be since those are the cb/sfs that will eventually be set. 4. Fix distortion computation on multi-subwindow groups (was subtracting the bits terms multiple times) 5. Clear ms_mask when one side uses PNS and the other doesn't. When using PNS, ms_mask signals correlated noise, which can be detected just like regular M/S detection, so we don't skip noise bands, but when only one side uses PNS setting the flag can confuse some encoders, so avoid that.
2016-01-08 10:31:32 +02:00
sce1->sf_idx[w*16+g],
sce1->band_type[w*16+g],
AAC encoder: improve SF range utilization This patch does 4 things, all of which interact and thus it woudln't be possible to commit them separately without causing either quality regressions or assertion failures. Fate comparison targets don't all reflect improvements in quality, yet listening tests show substantially improved quality and stability. 1. Increase SF range utilization. The spec requires SF delta values to be constrained within the range -60..60. The previous code was applying that range to the whole SF array and not only the deltas of consecutive values, because doing so requires smarter code: zeroing or otherwise skipping a band may invalidate lots of SF choices. This patch implements that logic to allow the coders to utilize the full dynamic range of scalefactors, increasing quality quite considerably, and fixing delta-SF-related assertion failures, since now the limitation is enforced rather than asserted. 2. PNS tweaks The previous modification makes big improvements in twoloop's efficiency, and every time that happens PNS logic needs to be tweaked accordingly to avoid it from stepping all over twoloop's decisions. This patch includes modifications of the sort. 3. Account for lowpass cutoff during PSY analysis The closer PSY's allocation is to final allocation the better the quality is, and given these modifications, twoloop is now very efficient at avoiding holes. Thus, to compute accurate thresholds, PSY needs to account for the lowpass applied implicitly during twoloop (by zeroing high bands). This patch makes twoloop set the cutoff in psymodel's context the first time it runs, and makes PSY account for it during threshold computation, making PE and threshold computations closer to the final allocation and thus achieving better subjective quality. 4. Tweaks to RC lambda tracking loop in relation to PNS Without this tweak some corner cases cause quality regressions. Basically, lambda needs to react faster to overall bitrate efficiency changes since now PNS can be quite successful in enforcing maximum bitrates, when PSY allocates too many bits to the lower bands, suppressing the signals RC logic uses to lower lambda in those cases and causing aggressive PNS. This tweak makes PNS much less aggressive, though it can still use some further tweaks. Also update MIPS specializations and adjust fuzz Also in lavc/mips/aacpsy_mips.h: remove trailing whitespace
2015-12-01 08:28:36 +02:00
lambda / band1->threshold, INFINITY, &b2, NULL, 0);
dist2 += quantize_band_cost(s, M,
M34,
sce0->ics.swb_sizes[g],
AAC encoder: various fixes in M/S coding 1. Fix sf_idx and band_type addressing to address only the first subwindow in the group (others could hold garbage values) 2. Don't step on ms_mask when is_mask is set. I/S selection already sets the ms_mask properly and shouldn't be overridden. 3. Use mid/sid cb/sf when computing coding error, as should be since those are the cb/sfs that will eventually be set. 4. Fix distortion computation on multi-subwindow groups (was subtracting the bits terms multiple times) 5. Clear ms_mask when one side uses PNS and the other doesn't. When using PNS, ms_mask signals correlated noise, which can be detected just like regular M/S detection, so we don't skip noise bands, but when only one side uses PNS setting the flag can confuse some encoders, so avoid that.
2016-01-08 10:31:32 +02:00
mididx,
midcb,
AAC encoder: improve SF range utilization This patch does 4 things, all of which interact and thus it woudln't be possible to commit them separately without causing either quality regressions or assertion failures. Fate comparison targets don't all reflect improvements in quality, yet listening tests show substantially improved quality and stability. 1. Increase SF range utilization. The spec requires SF delta values to be constrained within the range -60..60. The previous code was applying that range to the whole SF array and not only the deltas of consecutive values, because doing so requires smarter code: zeroing or otherwise skipping a band may invalidate lots of SF choices. This patch implements that logic to allow the coders to utilize the full dynamic range of scalefactors, increasing quality quite considerably, and fixing delta-SF-related assertion failures, since now the limitation is enforced rather than asserted. 2. PNS tweaks The previous modification makes big improvements in twoloop's efficiency, and every time that happens PNS logic needs to be tweaked accordingly to avoid it from stepping all over twoloop's decisions. This patch includes modifications of the sort. 3. Account for lowpass cutoff during PSY analysis The closer PSY's allocation is to final allocation the better the quality is, and given these modifications, twoloop is now very efficient at avoiding holes. Thus, to compute accurate thresholds, PSY needs to account for the lowpass applied implicitly during twoloop (by zeroing high bands). This patch makes twoloop set the cutoff in psymodel's context the first time it runs, and makes PSY account for it during threshold computation, making PE and threshold computations closer to the final allocation and thus achieving better subjective quality. 4. Tweaks to RC lambda tracking loop in relation to PNS Without this tweak some corner cases cause quality regressions. Basically, lambda needs to react faster to overall bitrate efficiency changes since now PNS can be quite successful in enforcing maximum bitrates, when PSY allocates too many bits to the lower bands, suppressing the signals RC logic uses to lower lambda in those cases and causing aggressive PNS. This tweak makes PNS much less aggressive, though it can still use some further tweaks. Also update MIPS specializations and adjust fuzz Also in lavc/mips/aacpsy_mips.h: remove trailing whitespace
2015-12-01 08:28:36 +02:00
lambda / minthr, INFINITY, &b3, NULL, 0);
dist2 += quantize_band_cost(s, S,
S34,
sce1->ics.swb_sizes[g],
AAC encoder: various fixes in M/S coding 1. Fix sf_idx and band_type addressing to address only the first subwindow in the group (others could hold garbage values) 2. Don't step on ms_mask when is_mask is set. I/S selection already sets the ms_mask properly and shouldn't be overridden. 3. Use mid/sid cb/sf when computing coding error, as should be since those are the cb/sfs that will eventually be set. 4. Fix distortion computation on multi-subwindow groups (was subtracting the bits terms multiple times) 5. Clear ms_mask when one side uses PNS and the other doesn't. When using PNS, ms_mask signals correlated noise, which can be detected just like regular M/S detection, so we don't skip noise bands, but when only one side uses PNS setting the flag can confuse some encoders, so avoid that.
2016-01-08 10:31:32 +02:00
sididx,
sidcb,
AAC encoder: improve SF range utilization This patch does 4 things, all of which interact and thus it woudln't be possible to commit them separately without causing either quality regressions or assertion failures. Fate comparison targets don't all reflect improvements in quality, yet listening tests show substantially improved quality and stability. 1. Increase SF range utilization. The spec requires SF delta values to be constrained within the range -60..60. The previous code was applying that range to the whole SF array and not only the deltas of consecutive values, because doing so requires smarter code: zeroing or otherwise skipping a band may invalidate lots of SF choices. This patch implements that logic to allow the coders to utilize the full dynamic range of scalefactors, increasing quality quite considerably, and fixing delta-SF-related assertion failures, since now the limitation is enforced rather than asserted. 2. PNS tweaks The previous modification makes big improvements in twoloop's efficiency, and every time that happens PNS logic needs to be tweaked accordingly to avoid it from stepping all over twoloop's decisions. This patch includes modifications of the sort. 3. Account for lowpass cutoff during PSY analysis The closer PSY's allocation is to final allocation the better the quality is, and given these modifications, twoloop is now very efficient at avoiding holes. Thus, to compute accurate thresholds, PSY needs to account for the lowpass applied implicitly during twoloop (by zeroing high bands). This patch makes twoloop set the cutoff in psymodel's context the first time it runs, and makes PSY account for it during threshold computation, making PE and threshold computations closer to the final allocation and thus achieving better subjective quality. 4. Tweaks to RC lambda tracking loop in relation to PNS Without this tweak some corner cases cause quality regressions. Basically, lambda needs to react faster to overall bitrate efficiency changes since now PNS can be quite successful in enforcing maximum bitrates, when PSY allocates too many bits to the lower bands, suppressing the signals RC logic uses to lower lambda in those cases and causing aggressive PNS. This tweak makes PNS much less aggressive, though it can still use some further tweaks. Also update MIPS specializations and adjust fuzz Also in lavc/mips/aacpsy_mips.h: remove trailing whitespace
2015-12-01 08:28:36 +02:00
mslambda / (minthr * bmax), INFINITY, &b4, NULL, 0);
B0 += b1+b2;
B1 += b3+b4;
AAC encoder: various fixes in M/S coding 1. Fix sf_idx and band_type addressing to address only the first subwindow in the group (others could hold garbage values) 2. Don't step on ms_mask when is_mask is set. I/S selection already sets the ms_mask properly and shouldn't be overridden. 3. Use mid/sid cb/sf when computing coding error, as should be since those are the cb/sfs that will eventually be set. 4. Fix distortion computation on multi-subwindow groups (was subtracting the bits terms multiple times) 5. Clear ms_mask when one side uses PNS and the other doesn't. When using PNS, ms_mask signals correlated noise, which can be detected just like regular M/S detection, so we don't skip noise bands, but when only one side uses PNS setting the flag can confuse some encoders, so avoid that.
2016-01-08 10:31:32 +02:00
dist1 -= b1+b2;
dist2 -= b3+b4;
AAC encoder: improve SF range utilization This patch does 4 things, all of which interact and thus it woudln't be possible to commit them separately without causing either quality regressions or assertion failures. Fate comparison targets don't all reflect improvements in quality, yet listening tests show substantially improved quality and stability. 1. Increase SF range utilization. The spec requires SF delta values to be constrained within the range -60..60. The previous code was applying that range to the whole SF array and not only the deltas of consecutive values, because doing so requires smarter code: zeroing or otherwise skipping a band may invalidate lots of SF choices. This patch implements that logic to allow the coders to utilize the full dynamic range of scalefactors, increasing quality quite considerably, and fixing delta-SF-related assertion failures, since now the limitation is enforced rather than asserted. 2. PNS tweaks The previous modification makes big improvements in twoloop's efficiency, and every time that happens PNS logic needs to be tweaked accordingly to avoid it from stepping all over twoloop's decisions. This patch includes modifications of the sort. 3. Account for lowpass cutoff during PSY analysis The closer PSY's allocation is to final allocation the better the quality is, and given these modifications, twoloop is now very efficient at avoiding holes. Thus, to compute accurate thresholds, PSY needs to account for the lowpass applied implicitly during twoloop (by zeroing high bands). This patch makes twoloop set the cutoff in psymodel's context the first time it runs, and makes PSY account for it during threshold computation, making PE and threshold computations closer to the final allocation and thus achieving better subjective quality. 4. Tweaks to RC lambda tracking loop in relation to PNS Without this tweak some corner cases cause quality regressions. Basically, lambda needs to react faster to overall bitrate efficiency changes since now PNS can be quite successful in enforcing maximum bitrates, when PSY allocates too many bits to the lower bands, suppressing the signals RC logic uses to lower lambda in those cases and causing aggressive PNS. This tweak makes PNS much less aggressive, though it can still use some further tweaks. Also update MIPS specializations and adjust fuzz Also in lavc/mips/aacpsy_mips.h: remove trailing whitespace
2015-12-01 08:28:36 +02:00
}
cpe->ms_mask[w*16+g] = dist2 <= dist1 && B1 < B0;
if (cpe->ms_mask[w*16+g]) {
AAC encoder: various fixes in M/S coding 1. Fix sf_idx and band_type addressing to address only the first subwindow in the group (others could hold garbage values) 2. Don't step on ms_mask when is_mask is set. I/S selection already sets the ms_mask properly and shouldn't be overridden. 3. Use mid/sid cb/sf when computing coding error, as should be since those are the cb/sfs that will eventually be set. 4. Fix distortion computation on multi-subwindow groups (was subtracting the bits terms multiple times) 5. Clear ms_mask when one side uses PNS and the other doesn't. When using PNS, ms_mask signals correlated noise, which can be detected just like regular M/S detection, so we don't skip noise bands, but when only one side uses PNS setting the flag can confuse some encoders, so avoid that.
2016-01-08 10:31:32 +02:00
if (sce0->band_type[w*16+g] != NOISE_BT && sce1->band_type[w*16+g] != NOISE_BT) {
AAC encoder: improve SF range utilization This patch does 4 things, all of which interact and thus it woudln't be possible to commit them separately without causing either quality regressions or assertion failures. Fate comparison targets don't all reflect improvements in quality, yet listening tests show substantially improved quality and stability. 1. Increase SF range utilization. The spec requires SF delta values to be constrained within the range -60..60. The previous code was applying that range to the whole SF array and not only the deltas of consecutive values, because doing so requires smarter code: zeroing or otherwise skipping a band may invalidate lots of SF choices. This patch implements that logic to allow the coders to utilize the full dynamic range of scalefactors, increasing quality quite considerably, and fixing delta-SF-related assertion failures, since now the limitation is enforced rather than asserted. 2. PNS tweaks The previous modification makes big improvements in twoloop's efficiency, and every time that happens PNS logic needs to be tweaked accordingly to avoid it from stepping all over twoloop's decisions. This patch includes modifications of the sort. 3. Account for lowpass cutoff during PSY analysis The closer PSY's allocation is to final allocation the better the quality is, and given these modifications, twoloop is now very efficient at avoiding holes. Thus, to compute accurate thresholds, PSY needs to account for the lowpass applied implicitly during twoloop (by zeroing high bands). This patch makes twoloop set the cutoff in psymodel's context the first time it runs, and makes PSY account for it during threshold computation, making PE and threshold computations closer to the final allocation and thus achieving better subjective quality. 4. Tweaks to RC lambda tracking loop in relation to PNS Without this tweak some corner cases cause quality regressions. Basically, lambda needs to react faster to overall bitrate efficiency changes since now PNS can be quite successful in enforcing maximum bitrates, when PSY allocates too many bits to the lower bands, suppressing the signals RC logic uses to lower lambda in those cases and causing aggressive PNS. This tweak makes PNS much less aggressive, though it can still use some further tweaks. Also update MIPS specializations and adjust fuzz Also in lavc/mips/aacpsy_mips.h: remove trailing whitespace
2015-12-01 08:28:36 +02:00
sce0->sf_idx[w*16+g] = mididx;
sce1->sf_idx[w*16+g] = sididx;
sce0->band_type[w*16+g] = midcb;
sce1->band_type[w*16+g] = sidcb;
AAC encoder: various fixes in M/S coding 1. Fix sf_idx and band_type addressing to address only the first subwindow in the group (others could hold garbage values) 2. Don't step on ms_mask when is_mask is set. I/S selection already sets the ms_mask properly and shouldn't be overridden. 3. Use mid/sid cb/sf when computing coding error, as should be since those are the cb/sfs that will eventually be set. 4. Fix distortion computation on multi-subwindow groups (was subtracting the bits terms multiple times) 5. Clear ms_mask when one side uses PNS and the other doesn't. When using PNS, ms_mask signals correlated noise, which can be detected just like regular M/S detection, so we don't skip noise bands, but when only one side uses PNS setting the flag can confuse some encoders, so avoid that.
2016-01-08 10:31:32 +02:00
} else if ((sce0->band_type[w*16+g] != NOISE_BT) ^ (sce1->band_type[w*16+g] != NOISE_BT)) {
/* ms_mask unneeded, and it confuses some decoders */
cpe->ms_mask[w*16+g] = 0;
AAC encoder: improve SF range utilization This patch does 4 things, all of which interact and thus it woudln't be possible to commit them separately without causing either quality regressions or assertion failures. Fate comparison targets don't all reflect improvements in quality, yet listening tests show substantially improved quality and stability. 1. Increase SF range utilization. The spec requires SF delta values to be constrained within the range -60..60. The previous code was applying that range to the whole SF array and not only the deltas of consecutive values, because doing so requires smarter code: zeroing or otherwise skipping a band may invalidate lots of SF choices. This patch implements that logic to allow the coders to utilize the full dynamic range of scalefactors, increasing quality quite considerably, and fixing delta-SF-related assertion failures, since now the limitation is enforced rather than asserted. 2. PNS tweaks The previous modification makes big improvements in twoloop's efficiency, and every time that happens PNS logic needs to be tweaked accordingly to avoid it from stepping all over twoloop's decisions. This patch includes modifications of the sort. 3. Account for lowpass cutoff during PSY analysis The closer PSY's allocation is to final allocation the better the quality is, and given these modifications, twoloop is now very efficient at avoiding holes. Thus, to compute accurate thresholds, PSY needs to account for the lowpass applied implicitly during twoloop (by zeroing high bands). This patch makes twoloop set the cutoff in psymodel's context the first time it runs, and makes PSY account for it during threshold computation, making PE and threshold computations closer to the final allocation and thus achieving better subjective quality. 4. Tweaks to RC lambda tracking loop in relation to PNS Without this tweak some corner cases cause quality regressions. Basically, lambda needs to react faster to overall bitrate efficiency changes since now PNS can be quite successful in enforcing maximum bitrates, when PSY allocates too many bits to the lower bands, suppressing the signals RC logic uses to lower lambda in those cases and causing aggressive PNS. This tweak makes PNS much less aggressive, though it can still use some further tweaks. Also update MIPS specializations and adjust fuzz Also in lavc/mips/aacpsy_mips.h: remove trailing whitespace
2015-12-01 08:28:36 +02:00
}
break;
} else if (B1 > B0) {
/* More boost won't fix this */
break;
}
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
}
}
AAC encoder: improve SF range utilization This patch does 4 things, all of which interact and thus it woudln't be possible to commit them separately without causing either quality regressions or assertion failures. Fate comparison targets don't all reflect improvements in quality, yet listening tests show substantially improved quality and stability. 1. Increase SF range utilization. The spec requires SF delta values to be constrained within the range -60..60. The previous code was applying that range to the whole SF array and not only the deltas of consecutive values, because doing so requires smarter code: zeroing or otherwise skipping a band may invalidate lots of SF choices. This patch implements that logic to allow the coders to utilize the full dynamic range of scalefactors, increasing quality quite considerably, and fixing delta-SF-related assertion failures, since now the limitation is enforced rather than asserted. 2. PNS tweaks The previous modification makes big improvements in twoloop's efficiency, and every time that happens PNS logic needs to be tweaked accordingly to avoid it from stepping all over twoloop's decisions. This patch includes modifications of the sort. 3. Account for lowpass cutoff during PSY analysis The closer PSY's allocation is to final allocation the better the quality is, and given these modifications, twoloop is now very efficient at avoiding holes. Thus, to compute accurate thresholds, PSY needs to account for the lowpass applied implicitly during twoloop (by zeroing high bands). This patch makes twoloop set the cutoff in psymodel's context the first time it runs, and makes PSY account for it during threshold computation, making PE and threshold computations closer to the final allocation and thus achieving better subjective quality. 4. Tweaks to RC lambda tracking loop in relation to PNS Without this tweak some corner cases cause quality regressions. Basically, lambda needs to react faster to overall bitrate efficiency changes since now PNS can be quite successful in enforcing maximum bitrates, when PSY allocates too many bits to the lower bands, suppressing the signals RC logic uses to lower lambda in those cases and causing aggressive PNS. This tweak makes PNS much less aggressive, though it can still use some further tweaks. Also update MIPS specializations and adjust fuzz Also in lavc/mips/aacpsy_mips.h: remove trailing whitespace
2015-12-01 08:28:36 +02:00
if (!sce0->zeroes[w*16+g] && sce0->band_type[w*16+g] < RESERVED_BT)
prev_mid = sce0->sf_idx[w*16+g];
if (!sce1->zeroes[w*16+g] && !cpe->is_mask[w*16+g] && sce1->band_type[w*16+g] < RESERVED_BT)
prev_side = sce1->sf_idx[w*16+g];
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
start += sce0->ics.swb_sizes[g];
}
}
}
#endif /*HAVE_MIPSFPU */
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
#include "libavcodec/aaccoder_trellis.h"
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
#endif /* HAVE_INLINE_ASM */
void ff_aac_coder_init_mips(AACEncContext *c) {
avcodec/mips/aaccoder_mips: Sync with the generic code This patch fixes build of AAC encoder optimized for mips that was broken due to some changes in generic code that were not propagated to the optimized code. Also, some functions in the optimized code are basically duplicate of functions from generic code. Since they do not bring enough improvement to the optimized code to justify their existence, they are removed (which improves maintainability of the optimized code). Optimizations disabled in 97437bd are enabled again. Signed-off-by: Nedeljko Babic <nedeljko.babic@imgtec.com> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2015-10-13 16:14:51 +02:00
#if HAVE_INLINE_ASM
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
AACCoefficientsEncoder *e = c->coder;
avcodec/mips/aaccoder_mips: Sync with the generic code This patch fixes build of AAC encoder optimized for mips that was broken due to some changes in generic code that were not propagated to the optimized code. Also, some functions in the optimized code are basically duplicate of functions from generic code. Since they do not bring enough improvement to the optimized code to justify their existence, they are removed (which improves maintainability of the optimized code). Optimizations disabled in 97437bd are enabled again. Signed-off-by: Nedeljko Babic <nedeljko.babic@imgtec.com> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2015-10-13 16:14:51 +02:00
int option = c->options.coder;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
if (option == 2) {
aaccoder_mips: Fix indentation
2015-09-14 17:54:08 +02:00
e->quantize_and_encode_band = quantize_and_encode_band_mips;
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
e->encode_window_bands_info = codebook_trellis_rate;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
#if HAVE_MIPSFPU
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
e->search_for_quantizers = search_for_quantizers_twoloop;
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
#endif /* HAVE_MIPSFPU */
}
AAC encoder: refactor to resynchronize MIPS port This patch refactors the AAC coders to reuse code between the MIPS port and the regular, portable C code. There were two main functions that had to use hand-optimized versions of quantization code: - search_for_quantizers_twoloop - codebook_trellis_rate Those two were split into their own template header files so they can be inlined inside both the MIPS port and the generic code. In each context, they'll link to their specialized implementations, and thus be optimized by the compiler. This approach I believe is better than maintaining several copies of each function. As past experience has proven, having to keep those in sync was error prone. In this way, they will remain in sync by default. Also, an implementation of the dequantized output argument for the optimized quantize_and_encode functions is included in the patch. While the current implementation of search_for_pred still isn't using it, future iterations of main prediction probably will. It should not imply any measurable performance hit while not being used.
2015-09-15 08:59:45 +02:00
#if HAVE_MIPSFPU
e->search_for_ms = search_for_ms_mips;
#endif /* HAVE_MIPSFPU */
mips: Optimization of AAC coefficients encoder functions Signed-off-by: Bojan Zivkovic <bojan@mips.com> Reviewed-by: Nedeljko Babic <Nedeljko.Babic@imgtec.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2013-03-06 16:55:05 +03:00
#endif /* HAVE_INLINE_ASM */
}
Reference in New Issue Copy Permalink