FFmpeg/libavcodec/loongarch/hevcdsp_lsx.h
jinbo 9239081db3
avcodec/hevc: Add asm opt for the following functions
tests/checkasm/checkasm:           C       LSX     LASX
put_hevc_qpel_uni_h4_8_c:          5.7     1.2
put_hevc_qpel_uni_h6_8_c:          12.2    2.7
put_hevc_qpel_uni_h8_8_c:          21.5    3.2
put_hevc_qpel_uni_h12_8_c:         47.2    9.2     7.2
put_hevc_qpel_uni_h16_8_c:         87.0    11.7    9.0
put_hevc_qpel_uni_h24_8_c:         188.2   27.5    21.0
put_hevc_qpel_uni_h32_8_c:         335.2   46.7    28.5
put_hevc_qpel_uni_h48_8_c:         772.5   104.5   65.2
put_hevc_qpel_uni_h64_8_c:         1383.2  142.2   109.0

put_hevc_epel_uni_w_v4_8_c:        5.0     1.5
put_hevc_epel_uni_w_v6_8_c:        10.7    3.5     2.5
put_hevc_epel_uni_w_v8_8_c:        18.2    3.7     3.0
put_hevc_epel_uni_w_v12_8_c:       40.2    10.7    7.5
put_hevc_epel_uni_w_v16_8_c:       70.2    13.0    9.2
put_hevc_epel_uni_w_v24_8_c:       158.2   30.2    22.5
put_hevc_epel_uni_w_v32_8_c:       281.0   52.0    36.5
put_hevc_epel_uni_w_v48_8_c:       631.7   116.7   82.7
put_hevc_epel_uni_w_v64_8_c:       1108.2  207.5   142.2

put_hevc_epel_uni_w_h4_8_c:        4.7     1.2
put_hevc_epel_uni_w_h6_8_c:        9.7     3.5     2.7
put_hevc_epel_uni_w_h8_8_c:        17.2    4.2     3.5
put_hevc_epel_uni_w_h12_8_c:       38.0    11.5    7.2
put_hevc_epel_uni_w_h16_8_c:       69.2    14.5    9.2
put_hevc_epel_uni_w_h24_8_c:       152.0   34.7    22.5
put_hevc_epel_uni_w_h32_8_c:       271.0   58.0    40.0
put_hevc_epel_uni_w_h48_8_c:       597.5   136.7   95.0
put_hevc_epel_uni_w_h64_8_c:       1074.0  252.2   168.0

put_hevc_epel_bi_h4_8_c:           4.5     0.7
put_hevc_epel_bi_h6_8_c:           9.0     1.5
put_hevc_epel_bi_h8_8_c:           15.2    1.7
put_hevc_epel_bi_h12_8_c:          33.5    4.2     3.7
put_hevc_epel_bi_h16_8_c:          59.7    5.2     4.7
put_hevc_epel_bi_h24_8_c:          132.2   11.0
put_hevc_epel_bi_h32_8_c:          232.7   20.2    13.2
put_hevc_epel_bi_h48_8_c:          521.7   45.2    31.2
put_hevc_epel_bi_h64_8_c:          949.0   71.5    51.0

After this patch, the peformance of decoding H265 4K 30FPS
30Mbps on 3A6000 with 8 threads improves 1fps(55fps-->56fsp).

Change-Id: I8cc1e41daa63ca478039bc55d1ee8934a7423f51
Reviewed-by: yinshiyou-hf@loongson.cn
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2024-01-12 23:35:40 +01:00

327 lines
10 KiB
C

/*
* Copyright (c) 2022 Loongson Technology Corporation Limited
* Contributed by Lu Wang <wanglu@loongson.cn>
* Hao Chen <chenhao@loongson.cn>
*
* 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
*/
#ifndef AVCODEC_LOONGARCH_HEVCDSP_LSX_H
#define AVCODEC_LOONGARCH_HEVCDSP_LSX_H
#include "libavcodec/hevcdsp.h"
#define MC(PEL, DIR, WIDTH) \
void ff_hevc_put_hevc_##PEL##_##DIR##WIDTH##_8_lsx(int16_t *dst, \
const uint8_t *src, \
ptrdiff_t src_stride, \
int height, \
intptr_t mx, \
intptr_t my, \
int width)
MC(pel, pixels, 4);
MC(pel, pixels, 6);
MC(pel, pixels, 8);
MC(pel, pixels, 12);
MC(pel, pixels, 16);
MC(pel, pixels, 24);
MC(pel, pixels, 32);
MC(pel, pixels, 48);
MC(pel, pixels, 64);
MC(qpel, h, 4);
MC(qpel, h, 8);
MC(qpel, h, 12);
MC(qpel, h, 16);
MC(qpel, h, 24);
MC(qpel, h, 32);
MC(qpel, h, 48);
MC(qpel, h, 64);
MC(qpel, v, 4);
MC(qpel, v, 8);
MC(qpel, v, 12);
MC(qpel, v, 16);
MC(qpel, v, 24);
MC(qpel, v, 32);
MC(qpel, v, 48);
MC(qpel, v, 64);
MC(qpel, hv, 4);
MC(qpel, hv, 8);
MC(qpel, hv, 12);
MC(qpel, hv, 16);
MC(qpel, hv, 24);
MC(qpel, hv, 32);
MC(qpel, hv, 48);
MC(qpel, hv, 64);
MC(epel, h, 32);
MC(epel, v, 16);
MC(epel, v, 24);
MC(epel, v, 32);
MC(epel, hv, 8);
MC(epel, hv, 12);
MC(epel, hv, 16);
MC(epel, hv, 24);
MC(epel, hv, 32);
#undef MC
#define BI_MC(PEL, DIR, WIDTH) \
void ff_hevc_put_hevc_bi_##PEL##_##DIR##WIDTH##_8_lsx(uint8_t *dst, \
ptrdiff_t dst_stride, \
const uint8_t *src, \
ptrdiff_t src_stride, \
const int16_t *src_16bit, \
int height, \
intptr_t mx, \
intptr_t my, \
int width)
BI_MC(pel, pixels, 4);
BI_MC(pel, pixels, 6);
BI_MC(pel, pixels, 8);
BI_MC(pel, pixels, 12);
BI_MC(pel, pixels, 16);
BI_MC(pel, pixels, 24);
BI_MC(pel, pixels, 32);
BI_MC(pel, pixels, 48);
BI_MC(pel, pixels, 64);
BI_MC(qpel, h, 16);
BI_MC(qpel, h, 24);
BI_MC(qpel, h, 32);
BI_MC(qpel, h, 48);
BI_MC(qpel, h, 64);
BI_MC(qpel, v, 8);
BI_MC(qpel, v, 16);
BI_MC(qpel, v, 24);
BI_MC(qpel, v, 32);
BI_MC(qpel, v, 48);
BI_MC(qpel, v, 64);
BI_MC(qpel, hv, 8);
BI_MC(qpel, hv, 16);
BI_MC(qpel, hv, 24);
BI_MC(qpel, hv, 32);
BI_MC(qpel, hv, 48);
BI_MC(qpel, hv, 64);
BI_MC(epel, h, 4);
BI_MC(epel, h, 6);
BI_MC(epel, h, 8);
BI_MC(epel, h, 12);
BI_MC(epel, h, 16);
BI_MC(epel, h, 24);
BI_MC(epel, h, 32);
BI_MC(epel, h, 48);
BI_MC(epel, h, 64);
BI_MC(epel, v, 12);
BI_MC(epel, v, 16);
BI_MC(epel, v, 24);
BI_MC(epel, v, 32);
BI_MC(epel, hv, 6);
BI_MC(epel, hv, 8);
BI_MC(epel, hv, 16);
BI_MC(epel, hv, 24);
BI_MC(epel, hv, 32);
#undef BI_MC
#define UNI_MC(PEL, DIR, WIDTH) \
void ff_hevc_put_hevc_uni_##PEL##_##DIR##WIDTH##_8_lsx(uint8_t *dst, \
ptrdiff_t dst_stride, \
const uint8_t *src, \
ptrdiff_t src_stride, \
int height, \
intptr_t mx, \
intptr_t my, \
int width)
UNI_MC(qpel, h, 4);
UNI_MC(qpel, h, 6);
UNI_MC(qpel, h, 8);
UNI_MC(qpel, h, 12);
UNI_MC(qpel, h, 16);
UNI_MC(qpel, h, 24);
UNI_MC(qpel, h, 32);
UNI_MC(qpel, h, 48);
UNI_MC(qpel, h, 64);
UNI_MC(qpel, v, 24);
UNI_MC(qpel, v, 32);
UNI_MC(qpel, v, 48);
UNI_MC(qpel, v, 64);
UNI_MC(qpel, hv, 8);
UNI_MC(qpel, hv, 16);
UNI_MC(qpel, hv, 24);
UNI_MC(qpel, hv, 32);
UNI_MC(qpel, hv, 48);
UNI_MC(qpel, hv, 64);
UNI_MC(epel, v, 24);
UNI_MC(epel, v, 32);
UNI_MC(epel, hv, 8);
UNI_MC(epel, hv, 12);
UNI_MC(epel, hv, 16);
UNI_MC(epel, hv, 24);
UNI_MC(epel, hv, 32);
#undef UNI_MC
#define UNI_W_MC(PEL, DIR, WIDTH) \
void ff_hevc_put_hevc_uni_w_##PEL##_##DIR##WIDTH##_8_lsx(uint8_t *dst, \
ptrdiff_t \
dst_stride, \
const uint8_t *src, \
ptrdiff_t \
src_stride, \
int height, \
int denom, \
int weight, \
int offset, \
intptr_t mx, \
intptr_t my, \
int width)
UNI_W_MC(qpel, hv, 8);
UNI_W_MC(qpel, hv, 16);
UNI_W_MC(qpel, hv, 24);
UNI_W_MC(qpel, hv, 32);
UNI_W_MC(qpel, hv, 48);
UNI_W_MC(qpel, hv, 64);
#undef UNI_W_MC
void ff_hevc_loop_filter_luma_h_8_lsx(uint8_t *src, ptrdiff_t stride,
int32_t beta, const int32_t *tc,
const uint8_t *p_is_pcm, const uint8_t *q_is_pcm);
void ff_hevc_loop_filter_luma_v_8_lsx(uint8_t *src, ptrdiff_t stride,
int32_t beta, const int32_t *tc,
const uint8_t *p_is_pcm, const uint8_t *q_is_pcm);
void ff_hevc_loop_filter_chroma_h_8_lsx(uint8_t *src, ptrdiff_t stride,
const int32_t *tc, const uint8_t *p_is_pcm,
const uint8_t *q_is_pcm);
void ff_hevc_loop_filter_chroma_v_8_lsx(uint8_t *src, ptrdiff_t stride,
const int32_t *tc, const uint8_t *p_is_pcm,
const uint8_t *q_is_pcm);
void ff_hevc_sao_edge_filter_8_lsx(uint8_t *dst, const uint8_t *src,
ptrdiff_t stride_dst,
const int16_t *sao_offset_val,
int eo, int width, int height);
void ff_hevc_idct_4x4_lsx(int16_t *coeffs, int col_limit);
void ff_hevc_idct_8x8_lsx(int16_t *coeffs, int col_limit);
void ff_hevc_idct_16x16_lsx(int16_t *coeffs, int col_limit);
void ff_hevc_idct_32x32_lsx(int16_t *coeffs, int col_limit);
void ff_hevc_add_residual4x4_8_lsx(uint8_t *dst, const int16_t *res, ptrdiff_t stride);
void ff_hevc_add_residual8x8_8_lsx(uint8_t *dst, const int16_t *res, ptrdiff_t stride);
void ff_hevc_add_residual16x16_8_lsx(uint8_t *dst, const int16_t *res, ptrdiff_t stride);
void ff_hevc_add_residual32x32_8_lsx(uint8_t *dst, const int16_t *res, ptrdiff_t stride);
#define PEL_UNI_W(PEL, DIR, WIDTH) \
void ff_hevc_put_hevc_##PEL##_uni_w_##DIR##WIDTH##_8_lsx(uint8_t *dst, \
ptrdiff_t \
dst_stride, \
const uint8_t *src, \
ptrdiff_t \
src_stride, \
int height, \
int denom, \
int wx, \
int ox, \
intptr_t mx, \
intptr_t my, \
int width)
PEL_UNI_W(pel, pixels, 4);
PEL_UNI_W(pel, pixels, 6);
PEL_UNI_W(pel, pixels, 8);
PEL_UNI_W(pel, pixels, 12);
PEL_UNI_W(pel, pixels, 16);
PEL_UNI_W(pel, pixels, 24);
PEL_UNI_W(pel, pixels, 32);
PEL_UNI_W(pel, pixels, 48);
PEL_UNI_W(pel, pixels, 64);
PEL_UNI_W(qpel, v, 4);
PEL_UNI_W(qpel, v, 6);
PEL_UNI_W(qpel, v, 8);
PEL_UNI_W(qpel, v, 12);
PEL_UNI_W(qpel, v, 16);
PEL_UNI_W(qpel, v, 24);
PEL_UNI_W(qpel, v, 32);
PEL_UNI_W(qpel, v, 48);
PEL_UNI_W(qpel, v, 64);
PEL_UNI_W(qpel, h, 4);
PEL_UNI_W(qpel, h, 6);
PEL_UNI_W(qpel, h, 8);
PEL_UNI_W(qpel, h, 12);
PEL_UNI_W(qpel, h, 16);
PEL_UNI_W(qpel, h, 24);
PEL_UNI_W(qpel, h, 32);
PEL_UNI_W(qpel, h, 48);
PEL_UNI_W(qpel, h, 64);
PEL_UNI_W(epel, hv, 4);
PEL_UNI_W(epel, hv, 6);
PEL_UNI_W(epel, hv, 8);
PEL_UNI_W(epel, hv, 12);
PEL_UNI_W(epel, hv, 16);
PEL_UNI_W(epel, hv, 24);
PEL_UNI_W(epel, hv, 32);
PEL_UNI_W(epel, hv, 48);
PEL_UNI_W(epel, hv, 64);
PEL_UNI_W(epel, h, 4);
PEL_UNI_W(epel, h, 6);
PEL_UNI_W(epel, h, 8);
PEL_UNI_W(epel, h, 12);
PEL_UNI_W(epel, h, 16);
PEL_UNI_W(epel, h, 24);
PEL_UNI_W(epel, h, 32);
PEL_UNI_W(epel, h, 48);
PEL_UNI_W(epel, h, 64);
PEL_UNI_W(epel, v, 4);
PEL_UNI_W(epel, v, 6);
PEL_UNI_W(epel, v, 8);
PEL_UNI_W(epel, v, 12);
PEL_UNI_W(epel, v, 16);
PEL_UNI_W(epel, v, 24);
PEL_UNI_W(epel, v, 32);
PEL_UNI_W(epel, v, 48);
PEL_UNI_W(epel, v, 64);
#undef PEL_UNI_W
#endif // #ifndef AVCODEC_LOONGARCH_HEVCDSP_LSX_H