RepoMirrors/ffmpeg
1
0
Fork 0
mirror of https://git.ffmpeg.org/ffmpeg.git synced 2024-12-28 10:22:10 +00:00
Code Issues Projects Releases Wiki Activity

avcodec: [loongarch] Optimize Hevc_mc_bi with LSX.

Browse Source 
ffmpeg -i 5_h265_1080p_60fps_3Mbps.mkv -f rawvideo -y /dev/null -an
before: 124fps
after : 182fps

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:
Hao Chen 2022-02-17 19:11:50 +08:00 committed by Michael Niedermayer
parent b6ceeee16b
commit a70a5b7c62
4 changed files with 2399 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
libavcodec/loongarch/Makefile
View File

@ -27,4 +27,5 @@ LSX-OBJS-$(CONFIG_VP9_DECODER) += loongarch/vp9_mc_lsx.o \
loongarch/vp9_idct_lsx.o
LSX-OBJS-$(CONFIG_HEVC_DECODER) += loongarch/hevcdsp_lsx.o \
loongarch/hevc_idct_lsx.o \
loongarch/hevc_lpf_sao_lsx.o
loongarch/hevc_lpf_sao_lsx.o \
loongarch/hevc_mc_bi_lsx.o

2289
libavcodec/loongarch/hevc_mc_bi_lsx.c Normal file
View File

File diff suppressed because it is too large Load Diff

51
libavcodec/loongarch/hevcdsp_init_loongarch.c
View File

@ -86,6 +86,57 @@ void ff_hevc_dsp_init_loongarch(HEVCDSPContext *c, const int bit_depth)
c->put_hevc_epel[6][1][1] = ff_hevc_put_hevc_epel_hv24_8_lsx;
c->put_hevc_epel[7][1][1] = ff_hevc_put_hevc_epel_hv32_8_lsx;
c->put_hevc_qpel_bi[1][0][0] = ff_hevc_put_hevc_bi_pel_pixels4_8_lsx;
c->put_hevc_qpel_bi[3][0][0] = ff_hevc_put_hevc_bi_pel_pixels8_8_lsx;
c->put_hevc_qpel_bi[4][0][0] = ff_hevc_put_hevc_bi_pel_pixels12_8_lsx;
c->put_hevc_qpel_bi[5][0][0] = ff_hevc_put_hevc_bi_pel_pixels16_8_lsx;
c->put_hevc_qpel_bi[6][0][0] = ff_hevc_put_hevc_bi_pel_pixels24_8_lsx;
c->put_hevc_qpel_bi[7][0][0] = ff_hevc_put_hevc_bi_pel_pixels32_8_lsx;
c->put_hevc_qpel_bi[8][0][0] = ff_hevc_put_hevc_bi_pel_pixels48_8_lsx;
c->put_hevc_qpel_bi[9][0][0] = ff_hevc_put_hevc_bi_pel_pixels64_8_lsx;
c->put_hevc_epel_bi[1][0][0] = ff_hevc_put_hevc_bi_pel_pixels4_8_lsx;
c->put_hevc_epel_bi[2][0][0] = ff_hevc_put_hevc_bi_pel_pixels6_8_lsx;
c->put_hevc_epel_bi[3][0][0] = ff_hevc_put_hevc_bi_pel_pixels8_8_lsx;
c->put_hevc_epel_bi[4][0][0] = ff_hevc_put_hevc_bi_pel_pixels12_8_lsx;
c->put_hevc_epel_bi[5][0][0] = ff_hevc_put_hevc_bi_pel_pixels16_8_lsx;
c->put_hevc_epel_bi[6][0][0] = ff_hevc_put_hevc_bi_pel_pixels24_8_lsx;
c->put_hevc_epel_bi[7][0][0] = ff_hevc_put_hevc_bi_pel_pixels32_8_lsx;
c->put_hevc_qpel_bi[3][1][0] = ff_hevc_put_hevc_bi_qpel_v8_8_lsx;
c->put_hevc_qpel_bi[5][1][0] = ff_hevc_put_hevc_bi_qpel_v16_8_lsx;
c->put_hevc_qpel_bi[6][1][0] = ff_hevc_put_hevc_bi_qpel_v24_8_lsx;
c->put_hevc_qpel_bi[7][1][0] = ff_hevc_put_hevc_bi_qpel_v32_8_lsx;
c->put_hevc_qpel_bi[8][1][0] = ff_hevc_put_hevc_bi_qpel_v48_8_lsx;
c->put_hevc_qpel_bi[9][1][0] = ff_hevc_put_hevc_bi_qpel_v64_8_lsx;
c->put_hevc_qpel_bi[3][1][1] = ff_hevc_put_hevc_bi_qpel_hv8_8_lsx;
c->put_hevc_qpel_bi[5][1][1] = ff_hevc_put_hevc_bi_qpel_hv16_8_lsx;
c->put_hevc_qpel_bi[6][1][1] = ff_hevc_put_hevc_bi_qpel_hv24_8_lsx;
c->put_hevc_qpel_bi[7][1][1] = ff_hevc_put_hevc_bi_qpel_hv32_8_lsx;
c->put_hevc_qpel_bi[8][1][1] = ff_hevc_put_hevc_bi_qpel_hv48_8_lsx;
c->put_hevc_qpel_bi[9][1][1] = ff_hevc_put_hevc_bi_qpel_hv64_8_lsx;
c->put_hevc_qpel_bi[5][0][1] = ff_hevc_put_hevc_bi_qpel_h16_8_lsx;
c->put_hevc_qpel_bi[6][0][1] = ff_hevc_put_hevc_bi_qpel_h24_8_lsx;
c->put_hevc_qpel_bi[7][0][1] = ff_hevc_put_hevc_bi_qpel_h32_8_lsx;
c->put_hevc_qpel_bi[8][0][1] = ff_hevc_put_hevc_bi_qpel_h48_8_lsx;
c->put_hevc_qpel_bi[9][0][1] = ff_hevc_put_hevc_bi_qpel_h64_8_lsx;
c->put_hevc_epel_bi[6][0][1] = ff_hevc_put_hevc_bi_epel_h24_8_lsx;
c->put_hevc_epel_bi[7][0][1] = ff_hevc_put_hevc_bi_epel_h32_8_lsx;
c->put_hevc_epel_bi[4][1][0] = ff_hevc_put_hevc_bi_epel_v12_8_lsx;
c->put_hevc_epel_bi[5][1][0] = ff_hevc_put_hevc_bi_epel_v16_8_lsx;
c->put_hevc_epel_bi[6][1][0] = ff_hevc_put_hevc_bi_epel_v24_8_lsx;
c->put_hevc_epel_bi[7][1][0] = ff_hevc_put_hevc_bi_epel_v32_8_lsx;
c->put_hevc_epel_bi[2][1][1] = ff_hevc_put_hevc_bi_epel_hv6_8_lsx;
c->put_hevc_epel_bi[3][1][1] = ff_hevc_put_hevc_bi_epel_hv8_8_lsx;
c->put_hevc_epel_bi[5][1][1] = ff_hevc_put_hevc_bi_epel_hv16_8_lsx;
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->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;

57
libavcodec/loongarch/hevcdsp_lsx.h
View File

@ -85,6 +85,63 @@ 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, \
uint8_t *src, \
ptrdiff_t src_stride, \
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, 24);
BI_MC(epel, h, 32);
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
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);