virtualenv/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2024-11-21 10:55:51 +02:00
Code Issues Releases Activity

libavcodec/hevc: add hevc idct4x4 neon of aarch64

Browse Source 
Signed-off-by: Martin Storsjö <martin@martin.st>
This commit is contained in:
xufuji456 2023-02-24 15:43:13 +08:00 committed by Martin Storsjö
parent ec7fa13eb0
commit 4b4de07721
2 changed files with 44 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

40
libavcodec/aarch64/hevcdsp_idct_neon.S
View File

@ -246,6 +246,43 @@ function hevc_add_residual_32x32_16_neon, export=0
ret
endfunc
.macro tr_4x4 in0, in1, in2, in3, out0, out1, out2, out3, shift
sshll v20.4s, \in0, #6
sshll v21.4s, \in0, #6
smull v22.4s, \in1, v4.h[1]
smull v23.4s, \in1, v4.h[3]
smlal v20.4s, \in2, v4.h[0] //e0
smlsl v21.4s, \in2, v4.h[0] //e1
smlal v22.4s, \in3, v4.h[3] //o0
smlsl v23.4s, \in3, v4.h[1] //o1
add v24.4s, v20.4s, v22.4s
sub v20.4s, v20.4s, v22.4s
add v22.4s, v21.4s, v23.4s
sub v21.4s, v21.4s, v23.4s
sqrshrn \out0, v24.4s, #\shift
sqrshrn \out3, v20.4s, #\shift
sqrshrn \out1, v22.4s, #\shift
sqrshrn \out2, v21.4s, #\shift
.endm
.macro idct_4x4 bitdepth
function ff_hevc_idct_4x4_\bitdepth\()_neon, export=1
ld1 {v0.4h-v3.4h}, [x0]
movrel x1, trans
ld1 {v4.4h}, [x1]
tr_4x4 v0.4h, v1.4h, v2.4h, v3.4h, v16.4h, v17.4h, v18.4h, v19.4h, 7
transpose_4x8H v16, v17, v18, v19, v26, v27, v28, v29
tr_4x4 v16.4h, v17.4h, v18.4h, v19.4h, v0.4h, v1.4h, v2.4h, v3.4h, 20 - \bitdepth
transpose_4x8H v0, v1, v2, v3, v26, v27, v28, v29
st1 {v0.4h-v3.4h}, [x0]
ret
endfunc
.endm
.macro sum_sub out, in, c, op, p
.ifc \op, +
smlal\p \out, \in, \c
@ -568,6 +605,9 @@ function ff_hevc_idct_16x16_\bitdepth\()_neon, export=1
endfunc
.endm
idct_4x4 8
idct_4x4 10
idct_8x8 8
idct_8x8 10

4
libavcodec/aarch64/hevcdsp_init_aarch64.c
View File

@ -49,6 +49,8 @@ void ff_hevc_add_residual_32x32_10_neon(uint8_t *_dst, const int16_t *coeffs,
ptrdiff_t stride);
void ff_hevc_add_residual_32x32_12_neon(uint8_t *_dst, const int16_t *coeffs,
ptrdiff_t stride);
void ff_hevc_idct_4x4_8_neon(int16_t *coeffs, int col_limit);
void ff_hevc_idct_4x4_10_neon(int16_t *coeffs, int col_limit);
void ff_hevc_idct_8x8_8_neon(int16_t *coeffs, int col_limit);
void ff_hevc_idct_8x8_10_neon(int16_t *coeffs, int col_limit);
void ff_hevc_idct_16x16_8_neon(int16_t *coeffs, int col_limit);
@ -119,6 +121,7 @@ av_cold void ff_hevc_dsp_init_aarch64(HEVCDSPContext *c, const int bit_depth)
c->add_residual[1] = ff_hevc_add_residual_8x8_8_neon;
c->add_residual[2] = ff_hevc_add_residual_16x16_8_neon;
c->add_residual[3] = ff_hevc_add_residual_32x32_8_neon;
c->idct[0] = ff_hevc_idct_4x4_8_neon;
c->idct[1] = ff_hevc_idct_8x8_8_neon;
c->idct[2] = ff_hevc_idct_16x16_8_neon;
c->idct_dc[0] = ff_hevc_idct_4x4_dc_8_neon;
@ -168,6 +171,7 @@ av_cold void ff_hevc_dsp_init_aarch64(HEVCDSPContext *c, const int bit_depth)
c->add_residual[1] = ff_hevc_add_residual_8x8_10_neon;
c->add_residual[2] = ff_hevc_add_residual_16x16_10_neon;
c->add_residual[3] = ff_hevc_add_residual_32x32_10_neon;
c->idct[0] = ff_hevc_idct_4x4_10_neon;
c->idct[1] = ff_hevc_idct_8x8_10_neon;
c->idct[2] = ff_hevc_idct_16x16_10_neon;
c->idct_dc[0] = ff_hevc_idct_4x4_dc_10_neon;