[FFmpeg-devel] [PATCH v2 4/4] lavc/vp8dsp: R-V V loop_filter

[uk7b at foxmail.com]

From: sunyuechi <sunyuechi at iscas.ac.cn>

                                                     C908   X60
vp8_loop_filter8uv_h_c                             :   12.2   10.0
vp8_loop_filter8uv_h_rvv_i32                       :   11.5    9.7
vp8_loop_filter8uv_v_c                             :   13.2   11.2
vp8_loop_filter8uv_v_rvv_i32                       :    8.0    6.5
vp8_loop_filter16y_h_c                             :   11.7   10.5
vp8_loop_filter16y_h_rvv_i32                       :    9.2    7.2
vp8_loop_filter16y_v_c                             :   11.5   10.5
vp8_loop_filter16y_v_rvv_i32                       :    5.5    3.7
---
 libavcodec/riscv/vp8dsp_init.c |  6 +++-
 libavcodec/riscv/vp8dsp_rvv.S  | 59 ++++++++++++++++++++++++++++++++++
 2 files changed, 64 insertions(+), 1 deletion(-)

diff --git a/libavcodec/riscv/vp8dsp_init.c b/libavcodec/riscv/vp8dsp_init.c
index 8cb21b8ceb..c53223a0e8 100644
--- a/libavcodec/riscv/vp8dsp_init.c
+++ b/libavcodec/riscv/vp8dsp_init.c
@@ -158,7 +158,11 @@ av_cold void ff_vp8dsp_init_riscv(VP8DSPContext *c)
     c->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16_inner_rvv##vlen; \
     c->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16_inner_rvv##vlen; \
     c->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_rvv##vlen; \
-    c->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_rvv##vlen;
+    c->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_rvv##vlen; \
+    c->vp8_v_loop_filter16y = ff_vp8_v_loop_filter16_rvv##vlen; \
+    c->vp8_h_loop_filter16y = ff_vp8_h_loop_filter16_rvv##vlen; \
+    c->vp8_v_loop_filter8uv = ff_vp8_v_loop_filter8uv_rvv##vlen; \
+    c->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_rvv##vlen;
 
     int flags = av_get_cpu_flags();
 
diff --git a/libavcodec/riscv/vp8dsp_rvv.S b/libavcodec/riscv/vp8dsp_rvv.S
index 036872a29e..298910ea90 100644
--- a/libavcodec/riscv/vp8dsp_rvv.S
+++ b/libavcodec/riscv/vp8dsp_rvv.S
@@ -390,12 +390,43 @@ endfunc
 .if \normal
         vmand.mm        v0, v1, v3              // vp8_normal_limit & !hv
 
+        .if \inner
         vnclip.wi       v22, v22, 0             // clip_int8(a);
         filter_fmin     \len, \vlen, v22, v12, v26, v10, v24, v20
         vadd.vi         v12, v12, 1
         vsra.vi         v12, v12, 1             // (f1 + 1) >> 1;
         vadd.vv         v8, v8, v12             // p1 + a
         vsub.vv         v14, v14, v12           // q1 - a
+        .else
+        li              t6, 27
+        li              a7, 18
+        li              a6, 9
+        vwmul.vx        v22, v28, t6
+        vwmul.vx        v4, v28, a7
+        vwmul.vx        v26, v28, a6
+        vsetvlstatic16  \len, \vlen
+        li              a7, 63
+        vzext.vf2       v18, v2                  // p2
+        vzext.vf2       v28, v7                  // q2
+        vadd.vx         v22, v22, a7
+        vadd.vx         v4, v4, a7
+        vadd.vx         v26, v26, a7
+        vsra.vi         v22, v22, 7              // a0
+        vsra.vi         v4, v4, 7                // a1
+        vsra.vi         v26, v26, 7              // a2
+        vadd.vv         v18, v18, v26
+        vadd.vv         v8, v8, v4
+        vadd.vv         v30, v24, v22
+        vsub.vv         v28, v28, v26
+        vsub.vv         v10, v20, v22
+        vsub.vv         v14, v14, v4
+        vmax.vx         v18, v18, zero
+        vmax.vx         v8, v8, zero
+        vmax.vx         v26, v30, zero
+        vmax.vx         v10, v10, zero
+        vmax.vx         v14, v14, zero
+        vmax.vx         v28, v28, zero
+        .endif
 
         vmax.vx         v8, v8, zero
         vmax.vx         v14, v14, zero
@@ -404,14 +435,27 @@ endfunc
         vnclipu.wi      v4, v26, 0              // -1
         vnclipu.wi      v5, v10, 0              // 0
         vnclipu.wi      v6, v14, 0              // 1
+        .if !\inner
+        vnclipu.wi      v2, v18, 0
+        vnclipu.wi      v7, v28, 0
+        addi            t0, \dst, -3
+        .endif
 
         .ifc \type,v
         vse8.v          v3, (t2), v0.t          // -2
         vse8.v          v4, (t3), v0.t          // -1
         vse8.v          v5, (\dst), v0.t        // 0
         vse8.v          v6, (t4), v0.t          // 1
+        .if !\inner
+        vse8.v          v2, (t1), v0.t
+        vse8.v          v7, (t5), v0.t
+        .endif
         .else
+        .if \inner
         vssseg4e8.v     v3, (t2), \stride, v0.t
+        .else
+        vssseg6e8.v     v2, (t0), \stride, v0.t
+        .endif
         .endif
 .endif
 .endm
@@ -439,6 +483,21 @@ func ff_vp8_\type\()_loop_filter8uv_inner_rvv\vlen, zve32x
         filter 8, \vlen, \type, 1, 1, a1, a2, a3, a4, a5
         ret
 endfunc
+
+func ff_vp8_\type\()_loop_filter16_rvv\vlen, zve32x
+        csrwi           vxrm, 0
+        vsetvlstatic8   16, \vlen
+        filter 16, \vlen, \type, 1, 0, a0, a1, a2, a3, a4
+        ret
+endfunc
+
+func ff_vp8_\type\()_loop_filter8uv_rvv\vlen, zve32x
+        csrwi           vxrm, 0
+        vsetvlstatic8   8, \vlen
+        filter 8, \vlen, \type, 1, 0, a0, a2, a3, a4, a5
+        filter 8, \vlen, \type, 1, 0, a1, a2, a3, a4, a5
+        ret
+endfunc
 .endr
 .endr
 
-- 
2.45.2