virtualenv/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2024-12-23 12:43:46 +02:00
Code Issues Releases Activity

avcodec: [loongarch] Optimize Hevc_mc_uni/w with LSX.

Browse Source 
ffmpeg -i 5_h265_1080p_60fps_3Mbps.mkv -f rawvideo -y /dev/null -an
before: 182fps
after : 191fps

Signed-off-by: Hao Chen <chenhao@loongson.cn>
Reviewed-by: 殷时友 <yinshiyou-hf@loongson.cn>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
This commit is contained in:
Lu Wang 2022-02-17 19:11:51 +08:00 committed by Michael Niedermayer
parent a70a5b7c62
commit 72604b10f4
5 changed files with 1813 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
libavcodec/loongarch/Makefile
View File

@ -28,4 +28,6 @@ LSX-OBJS-$(CONFIG_VP9_DECODER) += loongarch/vp9_mc_lsx.o \
LSX-OBJS-$(CONFIG_HEVC_DECODER) += loongarch/hevcdsp_lsx.o \
loongarch/hevc_idct_lsx.o \
loongarch/hevc_lpf_sao_lsx.o \
loongarch/hevc_mc_bi_lsx.o
loongarch/hevc_mc_bi_lsx.o \
loongarch/hevc_mc_uni_lsx.o \
loongarch/hevc_mc_uniw_lsx.o

1423
libavcodec/loongarch/hevc_mc_uni_lsx.c Normal file
View File

File diff suppressed because it is too large Load Diff

298
libavcodec/loongarch/hevc_mc_uniw_lsx.c Normal file
View File

@ -0,0 +1,298 @@
/*
* 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
*/
#include "libavutil/loongarch/loongson_intrinsics.h"
#include "hevcdsp_lsx.h"
static const uint8_t ff_hevc_mask_arr[16 * 2] __attribute__((aligned(0x40))) = {
/* 8 width cases */
0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8,
/* 4 width cases */
0, 1, 1, 2, 2, 3, 3, 4, 16, 17, 17, 18, 18, 19, 19, 20
};
static av_always_inline
void hevc_hv_8t_8x2_lsx(uint8_t *src, int32_t src_stride, uint8_t *dst,
int32_t dst_stride, const int8_t *filter_x,
const int8_t *filter_y, int32_t height, int32_t weight,
int32_t offset, int32_t rnd_val, int32_t width)
{
uint32_t loop_cnt, cnt;
uint8_t *src_tmp;
uint8_t *dst_tmp;
const int32_t src_stride_2x = (src_stride << 1);
const int32_t dst_stride_2x = (dst_stride << 1);
const int32_t src_stride_4x = (src_stride << 2);
const int32_t src_stride_3x = src_stride_2x + src_stride;
__m128i src0, src1, src2, src3, src4, src5, src6, src7, src8;
__m128i filt0, filt1, filt2, filt3;
__m128i filt_h0, filt_h1, filt_h2, filt_h3;
__m128i mask1, mask2, mask3;
__m128i filter_vec;
__m128i vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
__m128i vec8, vec9, vec10, vec11, vec12, vec13, vec14, vec15;
__m128i dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, dst8;
__m128i dst0_r, dst0_l, dst1_r, dst1_l;
__m128i dst10_r, dst32_r, dst54_r, dst76_r;
__m128i dst10_l, dst32_l, dst54_l, dst76_l;
__m128i dst21_r, dst43_r, dst65_r, dst87_r;
__m128i dst21_l, dst43_l, dst65_l, dst87_l;
__m128i weight_vec, offset_vec, rnd_vec;
__m128i mask0 = __lsx_vld(ff_hevc_mask_arr, 0);
src -= (src_stride_3x + 3);
weight_vec = __lsx_vreplgr2vr_w(weight);
offset_vec = __lsx_vreplgr2vr_w(offset);
rnd_vec = __lsx_vreplgr2vr_w(rnd_val);
DUP4_ARG2(__lsx_vldrepl_h, filter_x, 0, filter_x, 2, filter_x, 4,
filter_x, 6, filt0, filt1, filt2, filt3);
filter_vec = __lsx_vld(filter_y, 0);
filter_vec = __lsx_vsllwil_h_b(filter_vec, 0);
DUP4_ARG2(__lsx_vreplvei_w, filter_vec, 0, filter_vec, 1, filter_vec, 2,
filter_vec, 3, filt_h0, filt_h1, filt_h2, filt_h3);
DUP2_ARG2(__lsx_vaddi_bu, mask0, 2, mask0, 4, mask1, mask2);
mask3 = __lsx_vaddi_bu(mask0, 6);
for (cnt = width >> 3; cnt--;) {
src_tmp = src;
dst_tmp = dst;
src0 = __lsx_vld(src_tmp, 0);
DUP2_ARG2(__lsx_vldx, src_tmp, src_stride, src_tmp, src_stride_2x,
src1, src2);
src3 = __lsx_vldx(src_tmp, src_stride_3x);
src_tmp += src_stride_4x;
src4 = __lsx_vld(src_tmp, 0);
DUP2_ARG2(__lsx_vldx, src_tmp, src_stride, src_tmp, src_stride_2x,
src5, src6);
src_tmp += src_stride_3x;
DUP4_ARG3(__lsx_vshuf_b, src0, src0, mask0, src0, src0, mask1, src0,
src0, mask2, src0, src0, mask3, vec0, vec1, vec2, vec3);
DUP4_ARG3(__lsx_vshuf_b, src1, src1, mask0, src1, src1, mask1, src1,
src1, mask2, src1, src1, mask3, vec4, vec5, vec6, vec7);
DUP4_ARG3(__lsx_vshuf_b, src2, src2, mask0, src2, src2, mask1, src2,
src2, mask2, src2, src2, mask3, vec8, vec9, vec10, vec11);
DUP4_ARG3(__lsx_vshuf_b, src3, src3, mask0, src3, src3, mask1, src3,
src3, mask2, src3, src3, mask3, vec12, vec13, vec14, vec15);
DUP4_ARG2(__lsx_vdp2_h_bu_b, vec0, filt0, vec4, filt0, vec8, filt0,
vec12, filt0, dst0, dst1, dst2, dst3);
DUP4_ARG3(__lsx_vdp2add_h_bu_b, dst0, vec1, filt1, dst1, vec5, filt1,
dst2, vec9, filt1, dst3, vec13, filt1, dst0, dst1, dst2, dst3);
DUP4_ARG3(__lsx_vdp2add_h_bu_b, dst0, vec2, filt2, dst1, vec6, filt2,
dst2, vec10, filt2, dst3, vec14, filt2, dst0, dst1, dst2, dst3);
DUP4_ARG3(__lsx_vdp2add_h_bu_b, dst0, vec3, filt3, dst1, vec7, filt3,
dst2, vec11, filt3, dst3, vec15, filt3, dst0, dst1, dst2, dst3);
DUP4_ARG3(__lsx_vshuf_b, src4, src4, mask0, src4, src4, mask1, src4,
src4, mask2, src4, src4, mask3, vec0, vec1, vec2, vec3);
DUP4_ARG3(__lsx_vshuf_b, src5, src5, mask0, src5, src5, mask1, src5,
src5, mask2, src5, src5, mask3, vec4, vec5, vec6, vec7);
DUP4_ARG3(__lsx_vshuf_b, src6, src6, mask0, src6, src6, mask1, src6,
src6, mask2, src6, src6, mask3, vec8, vec9, vec10, vec11);
DUP2_ARG2(__lsx_vdp2_h_bu_b, vec0, filt0, vec4, filt0, dst4, dst5);
dst6 = __lsx_vdp2_h_bu_b(vec8, filt0);
DUP4_ARG3(__lsx_vdp2add_h_bu_b, dst4, vec1, filt1, dst5, vec5, filt1,
dst6, vec9, filt1, dst4, vec2, filt2, dst4, dst5, dst6, dst4);
DUP4_ARG3(__lsx_vdp2add_h_bu_b, dst5, vec6, filt2, dst6, vec10, filt2,
dst4, vec3, filt3, dst5, vec7, filt3, dst5, dst6, dst4, dst5);
dst6 = __lsx_vdp2add_h_bu_b(dst6, vec11, filt3);
DUP4_ARG2(__lsx_vilvl_h, dst1, dst0, dst3, dst2, dst5, dst4, dst2,
dst1, dst10_r, dst32_r, dst54_r, dst21_r);
DUP2_ARG2(__lsx_vilvl_h, dst4, dst3, dst6, dst5, dst43_r, dst65_r);
DUP4_ARG2(__lsx_vilvh_h, dst1, dst0, dst3, dst2, dst5, dst4, dst2,
dst1, dst10_l, dst32_l, dst54_l, dst21_l);
DUP2_ARG2(__lsx_vilvh_h, dst4, dst3, dst6, dst5, dst43_l, dst65_l);
for (loop_cnt = height >> 1; loop_cnt--;) {
src7 = __lsx_vld(src_tmp, 0);
src8 = __lsx_vldx(src_tmp, src_stride);
src_tmp += src_stride_2x;
DUP4_ARG3(__lsx_vshuf_b, src7, src7, mask0, src7, src7, mask1, src7,
src7, mask2, src7, src7, mask3, vec0, vec1, vec2, vec3);
dst7 = __lsx_vdp2_h_bu_b(vec0, filt0);
DUP2_ARG3(__lsx_vdp2add_h_bu_b, dst7, vec1, filt1, dst7, vec2,
filt2, dst7, dst7);
dst7 = __lsx_vdp2add_h_bu_b(dst7, vec3, filt3);
dst76_r = __lsx_vilvl_h(dst7, dst6);
dst76_l = __lsx_vilvh_h(dst7, dst6);
DUP2_ARG2(__lsx_vdp2_w_h, dst10_r, filt_h0, dst10_l, filt_h0,
dst0_r, dst0_l);
DUP4_ARG3(__lsx_vdp2add_w_h, dst0_r, dst32_r, filt_h1, dst0_l,
dst32_l, filt_h1, dst0_r, dst54_r, filt_h2, dst0_l,
dst54_l, filt_h2, dst0_r, dst0_l, dst0_r, dst0_l);
DUP2_ARG3(__lsx_vdp2add_w_h, dst0_r, dst76_r, filt_h3, dst0_l,
dst76_l, filt_h3, dst0_r, dst0_l);
DUP2_ARG2(__lsx_vsrai_w, dst0_r, 6, dst0_l, 6, dst0_r, dst0_l);
/* row 8 */
DUP4_ARG3(__lsx_vshuf_b, src8, src8, mask0, src8, src8, mask1, src8,
src8, mask2, src8, src8, mask3, vec0, vec1, vec2, vec3);
dst8 = __lsx_vdp2_h_bu_b(vec0, filt0);
DUP2_ARG3(__lsx_vdp2add_h_bu_b, dst8, vec1, filt1, dst8, vec2,
filt2, dst8, dst8);
dst8 = __lsx_vdp2add_h_bu_b(dst8, vec3, filt3);
dst87_r = __lsx_vilvl_h(dst8, dst7);
dst87_l = __lsx_vilvh_h(dst8, dst7);
DUP2_ARG2(__lsx_vdp2_w_h, dst21_r, filt_h0, dst21_l, filt_h0,
dst1_r, dst1_l);
DUP4_ARG3(__lsx_vdp2add_w_h, dst1_r, dst43_r, filt_h1, dst1_l,
dst43_l, filt_h1, dst1_r, dst65_r, filt_h2, dst1_l,
dst65_l, filt_h2, dst1_r, dst1_l, dst1_r, dst1_l);
DUP2_ARG3(__lsx_vdp2add_w_h, dst1_r, dst87_r, filt_h3, dst1_l,
dst87_l, filt_h3, dst1_r, dst1_l);
DUP2_ARG2(__lsx_vsrai_w, dst1_r, 6, dst1_l, 6, dst1_r, dst1_l);
DUP2_ARG2(__lsx_vmul_w, dst0_r, weight_vec, dst0_l, weight_vec,
dst0_r, dst0_l);
DUP2_ARG2(__lsx_vmul_w, dst1_r, weight_vec, dst1_l, weight_vec,
dst1_r, dst1_l);
DUP4_ARG2(__lsx_vsrar_w, dst0_r, rnd_vec, dst1_r, rnd_vec, dst0_l,
rnd_vec, dst1_l, rnd_vec, dst0_r, dst1_r, dst0_l, dst1_l);
DUP2_ARG2(__lsx_vadd_w, dst0_r, offset_vec, dst0_l, offset_vec,
dst0_r, dst0_l);
DUP2_ARG2(__lsx_vadd_w, dst1_r, offset_vec, dst1_l, offset_vec,
dst1_r, dst1_l);
DUP4_ARG1(__lsx_vclip255_w, dst0_r, dst1_r, dst0_l, dst1_l, dst0_r,
dst1_r, dst0_l, dst1_l);
DUP2_ARG2(__lsx_vpickev_h, dst0_l, dst0_r, dst1_l, dst1_r,
dst0_r, dst1_r);
dst0_r = __lsx_vpickev_b(dst1_r, dst0_r);
__lsx_vstelm_d(dst0_r, dst_tmp, 0, 0);
__lsx_vstelm_d(dst0_r, dst_tmp + dst_stride, 0, 1);
dst_tmp += dst_stride_2x;
dst10_r = dst32_r;
dst32_r = dst54_r;
dst54_r = dst76_r;
dst10_l = dst32_l;
dst32_l = dst54_l;
dst54_l = dst76_l;
dst21_r = dst43_r;
dst43_r = dst65_r;
dst65_r = dst87_r;
dst21_l = dst43_l;
dst43_l = dst65_l;
dst65_l = dst87_l;
dst6 = dst8;
}
src += 8;
dst += 8;
}
}
static
void hevc_hv_8t_8w_lsx(uint8_t *src, int32_t src_stride, uint8_t *dst,
int32_t dst_stride, const int8_t *filter_x,
const int8_t *filter_y, int32_t height, int32_t weight,
int32_t offset, int32_t rnd_val)
{
hevc_hv_8t_8x2_lsx(src, src_stride, dst, dst_stride, filter_x,
filter_y, height, weight, offset, rnd_val, 8);
}
static
void hevc_hv_8t_16w_lsx(uint8_t *src, int32_t src_stride, uint8_t *dst,
int32_t dst_stride, const int8_t *filter_x,
const int8_t *filter_y, int32_t height, int32_t weight,
int32_t offset, int32_t rnd_val)
{
hevc_hv_8t_8x2_lsx(src, src_stride, dst, dst_stride, filter_x,
filter_y, height, weight, offset, rnd_val, 16);
}
static
void hevc_hv_8t_24w_lsx(uint8_t *src, int32_t src_stride, uint8_t *dst,
int32_t dst_stride, const int8_t *filter_x,
const int8_t *filter_y, int32_t height, int32_t weight,
int32_t offset, int32_t rnd_val)
{
hevc_hv_8t_8x2_lsx(src, src_stride, dst, dst_stride, filter_x,
filter_y, height, weight, offset, rnd_val, 24);
}
static
void hevc_hv_8t_32w_lsx(uint8_t *src, int32_t src_stride, uint8_t *dst,
int32_t dst_stride, const int8_t *filter_x,
const int8_t *filter_y, int32_t height, int32_t weight,
int32_t offset, int32_t rnd_val)
{
hevc_hv_8t_8x2_lsx(src, src_stride, dst, dst_stride, filter_x,
filter_y, height, weight, offset, rnd_val, 32);
}
static
void hevc_hv_8t_48w_lsx(uint8_t *src, int32_t src_stride, uint8_t *dst,
int32_t dst_stride, const int8_t *filter_x,
const int8_t *filter_y, int32_t height, int32_t weight,
int32_t offset, int32_t rnd_val)
{
hevc_hv_8t_8x2_lsx(src, src_stride, dst, dst_stride, filter_x,
filter_y, height, weight, offset, rnd_val, 48);
}
static
void hevc_hv_8t_64w_lsx(uint8_t *src, int32_t src_stride, uint8_t *dst,
int32_t dst_stride, const int8_t *filter_x,
const int8_t *filter_y, int32_t height, int32_t weight,
int32_t offset, int32_t rnd_val)
{
hevc_hv_8t_8x2_lsx(src, src_stride, dst, dst_stride, filter_x,
filter_y, height, weight, offset, rnd_val, 64);
}
#define UNI_W_MC_HV(PEL, WIDTH, TAP) \
void ff_hevc_put_hevc_uni_w_##PEL##_hv##WIDTH##_8_lsx(uint8_t *dst, \
ptrdiff_t dst_stride, \
uint8_t *src, \
ptrdiff_t src_stride, \
int height, \
int denom, \
int weight, \
int offset, \
intptr_t mx, \
intptr_t my, \
int width) \
{ \
const int8_t *filter_x = ff_hevc_##PEL##_filters[mx - 1]; \
const int8_t *filter_y = ff_hevc_##PEL##_filters[my - 1]; \
int shift = denom + 14 - 8; \
\
hevc_hv_##TAP##t_##WIDTH##w_lsx(src, src_stride, dst, dst_stride, filter_x,\
filter_y, height, weight, offset, shift); \
}
UNI_W_MC_HV(qpel, 8, 8);
UNI_W_MC_HV(qpel, 16, 8);
UNI_W_MC_HV(qpel, 24, 8);
UNI_W_MC_HV(qpel, 32, 8);
UNI_W_MC_HV(qpel, 48, 8);
UNI_W_MC_HV(qpel, 64, 8);
#undef UNI_W_MC_HV

30
libavcodec/loongarch/hevcdsp_init_loongarch.c
View File

@ -137,6 +137,36 @@ void ff_hevc_dsp_init_loongarch(HEVCDSPContext *c, const int bit_depth)
c->put_hevc_epel_bi[6][1][1] = ff_hevc_put_hevc_bi_epel_hv24_8_lsx;
c->put_hevc_epel_bi[7][1][1] = ff_hevc_put_hevc_bi_epel_hv32_8_lsx;
c->put_hevc_qpel_uni[9][0][1] = ff_hevc_put_hevc_uni_qpel_h64_8_lsx;
c->put_hevc_qpel_uni[6][1][0] = ff_hevc_put_hevc_uni_qpel_v24_8_lsx;
c->put_hevc_qpel_uni[7][1][0] = ff_hevc_put_hevc_uni_qpel_v32_8_lsx;
c->put_hevc_qpel_uni[8][1][0] = ff_hevc_put_hevc_uni_qpel_v48_8_lsx;
c->put_hevc_qpel_uni[9][1][0] = ff_hevc_put_hevc_uni_qpel_v64_8_lsx;
c->put_hevc_qpel_uni[3][1][1] = ff_hevc_put_hevc_uni_qpel_hv8_8_lsx;
c->put_hevc_qpel_uni[5][1][1] = ff_hevc_put_hevc_uni_qpel_hv16_8_lsx;
c->put_hevc_qpel_uni[6][1][1] = ff_hevc_put_hevc_uni_qpel_hv24_8_lsx;
c->put_hevc_qpel_uni[7][1][1] = ff_hevc_put_hevc_uni_qpel_hv32_8_lsx;
c->put_hevc_qpel_uni[8][1][1] = ff_hevc_put_hevc_uni_qpel_hv48_8_lsx;
c->put_hevc_qpel_uni[9][1][1] = ff_hevc_put_hevc_uni_qpel_hv64_8_lsx;
c->put_hevc_epel_uni[6][1][0] = ff_hevc_put_hevc_uni_epel_v24_8_lsx;
c->put_hevc_epel_uni[7][1][0] = ff_hevc_put_hevc_uni_epel_v32_8_lsx;
c->put_hevc_epel_uni[3][1][1] = ff_hevc_put_hevc_uni_epel_hv8_8_lsx;
c->put_hevc_epel_uni[4][1][1] = ff_hevc_put_hevc_uni_epel_hv12_8_lsx;
c->put_hevc_epel_uni[5][1][1] = ff_hevc_put_hevc_uni_epel_hv16_8_lsx;
c->put_hevc_epel_uni[6][1][1] = ff_hevc_put_hevc_uni_epel_hv24_8_lsx;
c->put_hevc_epel_uni[7][1][1] = ff_hevc_put_hevc_uni_epel_hv32_8_lsx;
c->put_hevc_qpel_uni_w[3][1][1] = ff_hevc_put_hevc_uni_w_qpel_hv8_8_lsx;
c->put_hevc_qpel_uni_w[5][1][1] = ff_hevc_put_hevc_uni_w_qpel_hv16_8_lsx;
c->put_hevc_qpel_uni_w[6][1][1] = ff_hevc_put_hevc_uni_w_qpel_hv24_8_lsx;
c->put_hevc_qpel_uni_w[7][1][1] = ff_hevc_put_hevc_uni_w_qpel_hv32_8_lsx;
c->put_hevc_qpel_uni_w[8][1][1] = ff_hevc_put_hevc_uni_w_qpel_hv48_8_lsx;
c->put_hevc_qpel_uni_w[9][1][1] = ff_hevc_put_hevc_uni_w_qpel_hv64_8_lsx;
c->sao_edge_filter[4] = ff_hevc_sao_edge_filter_8_lsx;
c->hevc_h_loop_filter_luma = ff_hevc_loop_filter_luma_h_8_lsx;

59
libavcodec/loongarch/hevcdsp_lsx.h
View File

@ -142,6 +142,65 @@ 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, \
uint8_t *src, \
ptrdiff_t src_stride, \
int height, \
intptr_t mx, \
intptr_t my, \
int width)
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, \
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, int32_t *tc,
uint8_t *p_is_pcm, uint8_t *q_is_pcm);