* [FFmpeg-devel] [PATCH 7/7] lavc/me_cmp: R-V V nsse
@ 2024-02-06 15:56 flow gg
2024-02-22 15:20 ` Rémi Denis-Courmont
0 siblings, 1 reply; 3+ messages in thread
From: flow gg @ 2024-02-06 15:56 UTC (permalink / raw)
To: FFmpeg development discussions and patches
[-- Attachment #1: Type: text/plain, Size: 1 bytes --]
[-- Attachment #2: 0007-lavc-me_cmp-R-V-V-nsse.patch --]
[-- Type: text/x-patch, Size: 6439 bytes --]
From 31635394e89318c554a9653bd22791336309951e Mon Sep 17 00:00:00 2001
From: sunyuechi <sunyuechi@iscas.ac.cn>
Date: Tue, 6 Feb 2024 22:51:47 +0800
Subject: [PATCH 7/7] lavc/me_cmp: R-V V nsse
C908:
nsse_0_c: 1990.0
nsse_0_rvv_i32: 572.0
nsse_1_c: 910.0
nsse_1_rvv_i32: 456.0
---
libavcodec/riscv/me_cmp_init.c | 30 +++++++++
libavcodec/riscv/me_cmp_rvv.S | 118 +++++++++++++++++++++++++++++++++
2 files changed, 148 insertions(+)
diff --git a/libavcodec/riscv/me_cmp_init.c b/libavcodec/riscv/me_cmp_init.c
index 24e78e3eeb..48c0d3d827 100644
--- a/libavcodec/riscv/me_cmp_init.c
+++ b/libavcodec/riscv/me_cmp_init.c
@@ -55,6 +55,15 @@ int ff_vsad8_rvv(MpegEncContext *c, const uint8_t *s1, const uint8_t *s2, ptrdif
int ff_vsad_intra16_rvv(MpegEncContext *c, const uint8_t *s, const uint8_t *dummy, ptrdiff_t stride, int h);
int ff_vsad_intra8_rvv(MpegEncContext *c, const uint8_t *s, const uint8_t *dummy, ptrdiff_t stride, int h);
+int ff_nsse16_rvv(int multiplier, const uint8_t *s1, const uint8_t *s2,
+ ptrdiff_t stride, int h);
+int nsse16_rvv_wrapper(MpegEncContext *c, const uint8_t *s1, const uint8_t *s2,
+ ptrdiff_t stride, int h);
+int ff_nsse8_rvv(int multiplier, const uint8_t *s1, const uint8_t *s2,
+ ptrdiff_t stride, int h);
+int nsse8_rvv_wrapper(MpegEncContext *c, const uint8_t *s1, const uint8_t *s2,
+ ptrdiff_t stride, int h);
+
av_cold void ff_me_cmp_init_riscv(MECmpContext *c, AVCodecContext *avctx)
{
#if HAVE_RVV
@@ -82,6 +91,27 @@ av_cold void ff_me_cmp_init_riscv(MECmpContext *c, AVCodecContext *avctx)
c->vsad[1] = ff_vsad8_rvv;
c->vsad[4] = ff_vsad_intra16_rvv;
c->vsad[5] = ff_vsad_intra8_rvv;
+
+ c->nsse[0] = nsse16_rvv_wrapper;
+ c->nsse[1] = nsse8_rvv_wrapper;
}
#endif
}
+
+int nsse16_rvv_wrapper(MpegEncContext *c, const uint8_t *s1, const uint8_t *s2,
+ ptrdiff_t stride, int h)
+{
+ if (c)
+ return ff_nsse16_rvv(c->avctx->nsse_weight, s1, s2, stride, h);
+ else
+ return ff_nsse16_rvv(8, s1, s2, stride, h);
+}
+
+int nsse8_rvv_wrapper(MpegEncContext *c, const uint8_t *s1, const uint8_t *s2,
+ ptrdiff_t stride, int h)
+{
+ if (c)
+ return ff_nsse8_rvv(c->avctx->nsse_weight, s1, s2, stride, h);
+ else
+ return ff_nsse8_rvv(8, s1, s2, stride, h);
+}
diff --git a/libavcodec/riscv/me_cmp_rvv.S b/libavcodec/riscv/me_cmp_rvv.S
index f32ae6b259..c9ae5bb6fc 100644
--- a/libavcodec/riscv/me_cmp_rvv.S
+++ b/libavcodec/riscv/me_cmp_rvv.S
@@ -407,3 +407,121 @@ endfunc
func ff_vsad_intra8_rvv, zve32x
vsad_vsse_intra8 abs
endfunc
+
+func ff_nsse16_rvv, zve32x
+ .macro squarediff16
+ vsetivli zero, 16, e8, m1, tu, ma
+ vle8.v v4, (a1)
+ vle8.v v12, (a2)
+ vwsubu.vv v16, v4, v12
+ vsetvli zero, zero, e16, m2, tu, ma
+ vwmacc.vv v24, v16, v16
+ .endm
+
+ .macro gradiff16 srcx srcv
+ vsetivli zero, 16, e8, m1, tu, ma
+ vle8.v v8, (\srcx)
+ vslide1down.vx v0, \srcv, t5
+ vslide1down.vx v16, v8, t5
+ vwsubu.vv v20, \srcv, v0
+ vwsubu.wv v0, v20, v8
+ vwaddu.wv v20, v0, v16
+ vsetivli zero, 15, e16, m2, tu, ma
+ vneg.v v0, v20
+ vmax.vv v0, v20, v0
+ .endm
+
+ csrwi vxrm, 0
+ vsetivli t0, 16, e32, m4, ta, ma
+ addi a4, a4, -1
+ li t5, 1
+ vmv.v.x v24, zero
+ vmv.v.x v28, zero
+1:
+ add t1, a1, a3
+ add t2, a2, a3
+ addi a4, a4, -1
+ squarediff16
+ gradiff16 t1, v4
+ vwaddu.wv v28, v28, v0
+ gradiff16 t2, v12
+ vwsubu.wv v28, v28, v0
+ add a1, a1, a3
+ add a2, a2, a3
+ bnez a4, 1b
+
+ squarediff16
+ vsetivli zero, 16, e32, m4, tu, ma
+ vmv.s.x v0, zero
+ vmv.s.x v4, zero
+ vredsum.vs v0, v24, v0
+ vredsum.vs v4, v28, v4
+ vmv.x.s t1, v0
+ vmv.x.s t2, v4
+ srai t3, t2, 31
+ xor t2, t3, t2
+ sub t2, t2, t3
+ mul t2, t2, a0
+ add a0, t2, t1
+
+ ret
+endfunc
+
+func ff_nsse8_rvv, zve32x
+ .macro squarediff8
+ vsetivli zero, 8, e8, mf2, tu, ma
+ vle8.v v4, (a1)
+ vle8.v v12, (a2)
+ vwsubu.vv v16, v4, v12
+ vsetvli zero, zero, e16, m1, tu, ma
+ vwmacc.vv v24, v16, v16
+ .endm
+
+ .macro gradiff8 srcx srcv
+ vsetivli zero, 8, e8, mf2, tu, ma
+ vle8.v v8, (\srcx)
+ vslide1down.vx v0, \srcv, t5
+ vslide1down.vx v16, v8, t5
+ vwsubu.vv v20, \srcv, v0
+ vwsubu.wv v0, v20, v8
+ vwaddu.wv v20, v0, v16
+ vsetivli zero, 7, e16, m1, tu, ma
+ vneg.v v0, v20
+ vmax.vv v0, v20, v0
+ .endm
+
+ csrwi vxrm, 0
+ vsetivli t0, 8, e32, m2, ta, ma
+ addi a4, a4, -1
+ li t5, 1
+ vmv.v.x v24, zero
+ vmv.v.x v28, zero
+1:
+ add t1, a1, a3
+ add t2, a2, a3
+ addi a4, a4, -1
+ squarediff8
+ gradiff8 t1, v4
+ vwaddu.wv v28, v28, v0
+ gradiff8 t2, v12
+ vwsubu.wv v28, v28, v0
+ add a1, a1, a3
+ add a2, a2, a3
+ bnez a4, 1b
+
+ squarediff8
+ vsetivli zero, 8, e32, m2, tu, ma
+ vmv.s.x v0, zero
+ vmv.s.x v4, zero
+ vredsum.vs v0, v24, v0
+ vredsum.vs v4, v28, v4
+ vmv.x.s t1, v0
+ vmv.x.s t2, v4
+ srai t3, t2, 31
+ xor t2, t3, t2
+ sub t2, t2, t3
+ mul t2, t2, a0
+ add a0, t2, t1
+
+ ret
+endfunc
--
2.43.0
[-- Attachment #3: Type: text/plain, Size: 251 bytes --]
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
^ permalink raw reply [flat|nested] 3+ messages in thread* Re: [FFmpeg-devel] [PATCH 7/7] lavc/me_cmp: R-V V nsse
2024-02-06 15:56 [FFmpeg-devel] [PATCH 7/7] lavc/me_cmp: R-V V nsse flow gg
@ 2024-02-22 15:20 ` Rémi Denis-Courmont
2024-02-22 15:41 ` flow gg
0 siblings, 1 reply; 3+ messages in thread
From: Rémi Denis-Courmont @ 2024-02-22 15:20 UTC (permalink / raw)
To: FFmpeg development discussions and patches
Le tiistaina 6. helmikuuta 2024, 17.56.59 EET flow gg a écrit :
>
Use 'static' functions where possible.
--
レミ・デニ-クールモン
http://www.remlab.net/
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [FFmpeg-devel] [PATCH 7/7] lavc/me_cmp: R-V V nsse
2024-02-22 15:20 ` Rémi Denis-Courmont
@ 2024-02-22 15:41 ` flow gg
0 siblings, 0 replies; 3+ messages in thread
From: flow gg @ 2024-02-22 15:41 UTC (permalink / raw)
To: FFmpeg development discussions and patches
[-- Attachment #1: Type: text/plain, Size: 578 bytes --]
Okay, updated it in the reply
Rémi Denis-Courmont <remi@remlab.net> 于2024年2月22日周四 23:20写道：
> Le tiistaina 6. helmikuuta 2024, 17.56.59 EET flow gg a écrit :
> >
>
> Use 'static' functions where possible.
>
> --
> レミ・デニ-クールモン
> http://www.remlab.net/
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel@ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
>
[-- Attachment #2: 0007-lavc-me_cmp-R-V-V-nsse.patch --]
[-- Type: text/x-patch, Size: 6273 bytes --]
From b5cd8bf8557b410dcdfa65acee9e841fd88fd0a9 Mon Sep 17 00:00:00 2001
From: sunyuechi <sunyuechi@iscas.ac.cn>
Date: Tue, 6 Feb 2024 22:51:47 +0800
Subject: [PATCH 7/7] lavc/me_cmp: R-V V nsse
C908:
nsse_0_c: 1990.0
nsse_0_rvv_i32: 572.0
nsse_1_c: 910.0
nsse_1_rvv_i32: 456.0
---
libavcodec/riscv/me_cmp_init.c | 25 +++++++
libavcodec/riscv/me_cmp_rvv.S | 118 +++++++++++++++++++++++++++++++++
2 files changed, 143 insertions(+)
diff --git a/libavcodec/riscv/me_cmp_init.c b/libavcodec/riscv/me_cmp_init.c
index 24e78e3eeb..858e2ccdb8 100644
--- a/libavcodec/riscv/me_cmp_init.c
+++ b/libavcodec/riscv/me_cmp_init.c
@@ -54,6 +54,28 @@ int ff_vsad16_rvv(MpegEncContext *c, const uint8_t *s1, const uint8_t *s2, ptrdi
int ff_vsad8_rvv(MpegEncContext *c, const uint8_t *s1, const uint8_t *s2, ptrdiff_t stride, int h);
int ff_vsad_intra16_rvv(MpegEncContext *c, const uint8_t *s, const uint8_t *dummy, ptrdiff_t stride, int h);
int ff_vsad_intra8_rvv(MpegEncContext *c, const uint8_t *s, const uint8_t *dummy, ptrdiff_t stride, int h);
+int ff_nsse16_rvv(int multiplier, const uint8_t *s1, const uint8_t *s2,
+ ptrdiff_t stride, int h);
+int ff_nsse8_rvv(int multiplier, const uint8_t *s1, const uint8_t *s2,
+ ptrdiff_t stride, int h);
+
+static int nsse16_rvv_wrapper(MpegEncContext *c, const uint8_t *s1, const uint8_t *s2,
+ ptrdiff_t stride, int h)
+{
+ if (c)
+ return ff_nsse16_rvv(c->avctx->nsse_weight, s1, s2, stride, h);
+ else
+ return ff_nsse16_rvv(8, s1, s2, stride, h);
+}
+
+static int nsse8_rvv_wrapper(MpegEncContext *c, const uint8_t *s1, const uint8_t *s2,
+ ptrdiff_t stride, int h)
+{
+ if (c)
+ return ff_nsse8_rvv(c->avctx->nsse_weight, s1, s2, stride, h);
+ else
+ return ff_nsse8_rvv(8, s1, s2, stride, h);
+}
av_cold void ff_me_cmp_init_riscv(MECmpContext *c, AVCodecContext *avctx)
{
@@ -82,6 +104,9 @@ av_cold void ff_me_cmp_init_riscv(MECmpContext *c, AVCodecContext *avctx)
c->vsad[1] = ff_vsad8_rvv;
c->vsad[4] = ff_vsad_intra16_rvv;
c->vsad[5] = ff_vsad_intra8_rvv;
+
+ c->nsse[0] = nsse16_rvv_wrapper;
+ c->nsse[1] = nsse8_rvv_wrapper;
}
#endif
}
diff --git a/libavcodec/riscv/me_cmp_rvv.S b/libavcodec/riscv/me_cmp_rvv.S
index f32ae6b259..c9ae5bb6fc 100644
--- a/libavcodec/riscv/me_cmp_rvv.S
+++ b/libavcodec/riscv/me_cmp_rvv.S
@@ -407,3 +407,121 @@ endfunc
func ff_vsad_intra8_rvv, zve32x
vsad_vsse_intra8 abs
endfunc
+
+func ff_nsse16_rvv, zve32x
+ .macro squarediff16
+ vsetivli zero, 16, e8, m1, tu, ma
+ vle8.v v4, (a1)
+ vle8.v v12, (a2)
+ vwsubu.vv v16, v4, v12
+ vsetvli zero, zero, e16, m2, tu, ma
+ vwmacc.vv v24, v16, v16
+ .endm
+
+ .macro gradiff16 srcx srcv
+ vsetivli zero, 16, e8, m1, tu, ma
+ vle8.v v8, (\srcx)
+ vslide1down.vx v0, \srcv, t5
+ vslide1down.vx v16, v8, t5
+ vwsubu.vv v20, \srcv, v0
+ vwsubu.wv v0, v20, v8
+ vwaddu.wv v20, v0, v16
+ vsetivli zero, 15, e16, m2, tu, ma
+ vneg.v v0, v20
+ vmax.vv v0, v20, v0
+ .endm
+
+ csrwi vxrm, 0
+ vsetivli t0, 16, e32, m4, ta, ma
+ addi a4, a4, -1
+ li t5, 1
+ vmv.v.x v24, zero
+ vmv.v.x v28, zero
+1:
+ add t1, a1, a3
+ add t2, a2, a3
+ addi a4, a4, -1
+ squarediff16
+ gradiff16 t1, v4
+ vwaddu.wv v28, v28, v0
+ gradiff16 t2, v12
+ vwsubu.wv v28, v28, v0
+ add a1, a1, a3
+ add a2, a2, a3
+ bnez a4, 1b
+
+ squarediff16
+ vsetivli zero, 16, e32, m4, tu, ma
+ vmv.s.x v0, zero
+ vmv.s.x v4, zero
+ vredsum.vs v0, v24, v0
+ vredsum.vs v4, v28, v4
+ vmv.x.s t1, v0
+ vmv.x.s t2, v4
+ srai t3, t2, 31
+ xor t2, t3, t2
+ sub t2, t2, t3
+ mul t2, t2, a0
+ add a0, t2, t1
+
+ ret
+endfunc
+
+func ff_nsse8_rvv, zve32x
+ .macro squarediff8
+ vsetivli zero, 8, e8, mf2, tu, ma
+ vle8.v v4, (a1)
+ vle8.v v12, (a2)
+ vwsubu.vv v16, v4, v12
+ vsetvli zero, zero, e16, m1, tu, ma
+ vwmacc.vv v24, v16, v16
+ .endm
+
+ .macro gradiff8 srcx srcv
+ vsetivli zero, 8, e8, mf2, tu, ma
+ vle8.v v8, (\srcx)
+ vslide1down.vx v0, \srcv, t5
+ vslide1down.vx v16, v8, t5
+ vwsubu.vv v20, \srcv, v0
+ vwsubu.wv v0, v20, v8
+ vwaddu.wv v20, v0, v16
+ vsetivli zero, 7, e16, m1, tu, ma
+ vneg.v v0, v20
+ vmax.vv v0, v20, v0
+ .endm
+
+ csrwi vxrm, 0
+ vsetivli t0, 8, e32, m2, ta, ma
+ addi a4, a4, -1
+ li t5, 1
+ vmv.v.x v24, zero
+ vmv.v.x v28, zero
+1:
+ add t1, a1, a3
+ add t2, a2, a3
+ addi a4, a4, -1
+ squarediff8
+ gradiff8 t1, v4
+ vwaddu.wv v28, v28, v0
+ gradiff8 t2, v12
+ vwsubu.wv v28, v28, v0
+ add a1, a1, a3
+ add a2, a2, a3
+ bnez a4, 1b
+
+ squarediff8
+ vsetivli zero, 8, e32, m2, tu, ma
+ vmv.s.x v0, zero
+ vmv.s.x v4, zero
+ vredsum.vs v0, v24, v0
+ vredsum.vs v4, v28, v4
+ vmv.x.s t1, v0
+ vmv.x.s t2, v4
+ srai t3, t2, 31
+ xor t2, t3, t2
+ sub t2, t2, t3
+ mul t2, t2, a0
+ add a0, t2, t1
+
+ ret
+endfunc
--
2.43.2
[-- Attachment #3: Type: text/plain, Size: 251 bytes --]
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2024-02-22 15:42 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2024-02-06 15:56 [FFmpeg-devel] [PATCH 7/7] lavc/me_cmp: R-V V nsse flow gg
2024-02-22 15:20 ` Rémi Denis-Courmont
2024-02-22 15:41 ` flow gg
Git Inbox Mirror of the ffmpeg-devel mailing list - see https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
This inbox may be cloned and mirrored by anyone:
git clone --mirror https://master.gitmailbox.com/ffmpegdev/0 ffmpegdev/git/0.git
# If you have public-inbox 1.1+ installed, you may
# initialize and index your mirror using the following commands:
public-inbox-init -V2 ffmpegdev ffmpegdev/ https://master.gitmailbox.com/ffmpegdev \
ffmpegdev@gitmailbox.com
public-inbox-index ffmpegdev
Example config snippet for mirrors.
AGPL code for this site: git clone https://public-inbox.org/public-inbox.git