virtualenv/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2025-07-16 22:42:38 +02:00
Code Issues Releases Activity

lavc/mpegvideoencdsp: R-V V pix_norm1

Browse Source 
T-Head C908:
pix_norm1_c:       480.2
pix_norm1_rvv_i64: 146.9

SpacemiT X60:
pix_norm1_c:       478.2
pix_norm1_rvv_i64:  92.7
This commit is contained in:
Rémi Denis-Courmont
2024-08-09 22:03:49 +03:00
parent 63d016aea5
commit 0fd37c00d7
2 changed files with 33 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
libavcodec/riscv/mpegvideoencdsp_init.c
View File

@ -24,6 +24,7 @@
#include "libavcodec/mpegvideoencdsp.h" #include "libavcodec/mpegvideoencdsp.h"
int ff_pix_sum_rvv(const uint8_t *pix, int line_size); int ff_pix_sum_rvv(const uint8_t *pix, int line_size);
int ff_pix_norm1_rvv(const uint8_t *pix, int line_size);
av_cold void ff_mpegvideoencdsp_init_riscv(MpegvideoEncDSPContext *c, av_cold void ff_mpegvideoencdsp_init_riscv(MpegvideoEncDSPContext *c,
AVCodecContext *avctx) AVCodecContext *avctx)
@ -31,8 +32,10 @@ av_cold void ff_mpegvideoencdsp_init_riscv(MpegvideoEncDSPContext *c,
#if HAVE_RVV #if HAVE_RVV
int flags = av_get_cpu_flags(); int flags = av_get_cpu_flags();
if ((flags & AV_CPU_FLAG_RVV_I64) && (flags & AV_CPU_FLAG_RVB) && if (flags & AV_CPU_FLAG_RVV_I64) {
ff_rv_vlen_least(128)) if ((flags & AV_CPU_FLAG_RVB) && ff_rv_vlen_least(128))
c->pix_sum = ff_pix_sum_rvv; c->pix_sum = ff_pix_sum_rvv;
c->pix_norm1 = ff_pix_norm1_rvv;
}
#endif #endif
} }

27
libavcodec/riscv/mpegvideoencdsp_rvv.S
View File

@ -44,3 +44,30 @@ func ff_pix_sum_rvv, zve64x, b
zext.h a0, a0 zext.h a0, a0
ret ret
endfunc endfunc
func ff_pix_norm1_rvv, zve64x
lpad 0
li t1, 16
vsetvli t0, zero, e32, m8, ta, ma
vmv.v.x v16, zero
vmv.s.x v1, zero
1:
vsetvli t0, t1, e64, m2, ta, ma
vlsseg2e64.v v8, (a0), a1
slli t2, t0, 3
vsetvli zero, t2, e8, m2, ta, ma
vwmulu.vv v24, v8, v8
sub t1, t1, t0
vwmulu.vv v28, v10, v10
mul t3, a1, t0
vsetvli zero, zero, e16, m4, tu, ma
vwaddu.wv v16, v16, v24
add a0, t3, a0
vwaddu.wv v16, v16, v28
bnez t1, 1b
vsetvli zero, zero, e32, m8, ta, ma
vredsum.vs v1, v16, v1
vmv.x.s a0, v1
ret
endfunc