[FFmpeg-devel] [PATCH 10/10] lavc/vp9dsp: R-V V mc tap hv
uk7b at foxmail.com
uk7b at foxmail.com
Sat May 4 18:03:13 EEST 2024
From: sunyuechi <sunyuechi at iscas.ac.cn>
C908:
vp9_avg_8tap_smooth_4hv_8bpp_c: 32.2
vp9_avg_8tap_smooth_4hv_8bpp_rvv_i64: 15.2
vp9_avg_8tap_smooth_8hv_8bpp_c: 98.5
vp9_avg_8tap_smooth_8hv_8bpp_rvv_i64: 23.5
vp9_avg_8tap_smooth_16hv_8bpp_c: 355.5
vp9_avg_8tap_smooth_16hv_8bpp_rvv_i64: 46.2
vp9_avg_8tap_smooth_32hv_8bpp_c: 1270.7
vp9_avg_8tap_smooth_32hv_8bpp_rvv_i64: 133.2
vp9_avg_8tap_smooth_64hv_8bpp_c: 4936.5
vp9_avg_8tap_smooth_64hv_8bpp_rvv_i64: 521.7
vp9_put_8tap_smooth_4hv_8bpp_c: 30.2
vp9_put_8tap_smooth_4hv_8bpp_rvv_i64: 14.2
vp9_put_8tap_smooth_8hv_8bpp_c: 91.5
vp9_put_8tap_smooth_8hv_8bpp_rvv_i64: 22.7
vp9_put_8tap_smooth_16hv_8bpp_c: 330.0
vp9_put_8tap_smooth_16hv_8bpp_rvv_i64: 45.0
vp9_put_8tap_smooth_32hv_8bpp_c: 1296.5
vp9_put_8tap_smooth_32hv_8bpp_rvv_i64: 131.0
vp9_put_8tap_smooth_64hv_8bpp_c: 4497.7
vp9_put_8tap_smooth_64hv_8bpp_rvv_i64: 513.2
---
libavcodec/riscv/vp9_mc_rvv.S | 79 ++++++++++++++++++++++++++++++++++
libavcodec/riscv/vp9dsp_init.c | 3 +-
2 files changed, 81 insertions(+), 1 deletion(-)
diff --git a/libavcodec/riscv/vp9_mc_rvv.S b/libavcodec/riscv/vp9_mc_rvv.S
index c8a42c7159..6ad7ea2433 100644
--- a/libavcodec/riscv/vp9_mc_rvv.S
+++ b/libavcodec/riscv/vp9_mc_rvv.S
@@ -446,12 +446,90 @@ endconst
ret
.endm
+.macro epel_hv_once len name do
+ sub a2, a2, a3
+ sub a2, a2, a3
+ sub a2, a2, a3
+ .irp n 0 2 4 6 8 10 12 14
+ epel_load_inc v\n \len put \name h 1 t
+ .endr
+ addi a4, a4, -1
+1:
+ addi a4, a4, -1
+ epel_load v30 \len \do \name v 0 s
+ vse8.v v30, (a0)
+ vmv.v.v v0, v2
+ vmv.v.v v2, v4
+ vmv.v.v v4, v6
+ vmv.v.v v6, v8
+ vmv.v.v v8, v10
+ vmv.v.v v10, v12
+ vmv.v.v v12, v14
+ epel_load v14 \len put \name h 1 t
+ add a2, a2, a3
+ add a0, a0, a1
+ bnez a4, 1b
+ epel_load v30 \len \do \name v 0 s
+ vse8.v v30, (a0)
+.endm
+
+.macro epel_hv do name len
+ addi sp, sp, -64
+ .irp n 0,1,2,3,4,5,6,7
+ sd s\n, \n\()<<3(sp)
+ .endr
+.ifc \len,64
+ addi sp, sp, -48
+ .irp n 0,1,2,3,4,5
+ sd a\n, \n\()<<3(sp)
+ .endr
+.endif
+.ifc \do,avg
+ csrwi vxrm, 0
+.endif
+ epel_filter \name h t
+ epel_filter \name v s
+.ifc \len,4
+ vsetivli zero, 4, e8, mf4, ta, ma
+.elseif \len == 8
+ vsetivli zero, 8, e8, mf2, ta, ma
+.elseif \len == 16
+ vsetivli zero, 16, e8, m1, ta, ma
+.else
+ li a6, 32
+ vsetvli zero, a6, e8, m2, ta, ma
+.endif
+ epel_hv_once \len \name \do
+.ifc \len,64
+ .irp n 0,1,2,3,4,5
+ ld a\n, \n\()<<3(sp)
+ .endr
+ addi sp, sp, 48
+ addi a0, a0, 32
+ addi a2, a2, 32
+ epel_filter \name h t
+ epel_hv_once \len \name \do
+.endif
+ .irp n 0,1,2,3,4,5,6,7
+ ld s\n, \n\()<<3(sp)
+ .endr
+ addi sp, sp, 64
+
+ ret
+.endm
+
.macro gen_epel len do name type
func ff_\do\()_8tap_\name\()_\len\()\type\()_rvv, zve32x
epel \len \do \name \type
endfunc
.endm
+.macro gen_epelhv len name do
+func ff_\do\()_8tap_\name\()_\len\()hv_rvv, zve32x
+ epel_hv \do \name \len
+endfunc
+.endm
+
.irp len 64, 32, 16, 8, 4
func ff_avg\len\()_rvv, zve32x
copy_avg \len avg
@@ -481,6 +559,7 @@ endfunc
.irp type h v
gen_epel \len \do \name \type
.endr
+ gen_epelhv \len \name \do
.endr
.endr
.endr
diff --git a/libavcodec/riscv/vp9dsp_init.c b/libavcodec/riscv/vp9dsp_init.c
index ff7d445f6a..0c75ef38dc 100644
--- a/libavcodec/riscv/vp9dsp_init.c
+++ b/libavcodec/riscv/vp9dsp_init.c
@@ -126,7 +126,8 @@ static av_cold void vp9dsp_mc_init_rvv(VP9DSPContext *dsp, int bpp)
#define init_subpel3(idx, type) \
init_subpel2(idx, 1, 0, h, type); \
- init_subpel2(idx, 0, 1, v, type)
+ init_subpel2(idx, 0, 1, v, type); \
+ init_subpel2(idx, 1, 1, hv, type)
init_subpel3(0, put);
init_subpel3(1, avg);
--
2.45.0
More information about the ffmpeg-devel
mailing list