[FFmpeg-devel] [PATCH 2/4] lavc/vp8dsp: R-V V loop_filter_simple

[uk7b at foxmail.com]

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

                                                     C908   X60
vp8_loop_filter_simple_h_c                         :    7.0    6.0
vp8_loop_filter_simple_h_rvv_i32                   :    3.2    2.7
vp8_loop_filter_simple_v_c                         :    7.2    6.5
vp8_loop_filter_simple_v_rvv_i32                   :    1.7    1.2
---
 libavcodec/riscv/vp8dsp_init.c | 18 ++++++-
 libavcodec/riscv/vp8dsp_rvv.S  | 87 ++++++++++++++++++++++++++++++++++
 2 files changed, 104 insertions(+), 1 deletion(-)

diff --git a/libavcodec/riscv/vp8dsp_init.c b/libavcodec/riscv/vp8dsp_init.c
index dcb6307d5b..8c5b2c8b04 100644
--- a/libavcodec/riscv/vp8dsp_init.c
+++ b/libavcodec/riscv/vp8dsp_init.c
@@ -49,6 +49,9 @@ VP8_BILIN(16, rvv256);
 VP8_BILIN(8,  rvv256);
 VP8_BILIN(4,  rvv256);
 
+VP8_LF(rvv128);
+VP8_LF(rvv256);
+
 av_cold void ff_vp78dsp_init_riscv(VP8DSPContext *c)
 {
 #if HAVE_RV
@@ -147,9 +150,15 @@ av_cold void ff_vp78dsp_init_riscv(VP8DSPContext *c)
 av_cold void ff_vp8dsp_init_riscv(VP8DSPContext *c)
 {
 #if HAVE_RVV
+    int vlenb = ff_get_rv_vlenb();
+
+#define init_loop_filter(vlen)                                           \
+    c->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter16_simple_rvv##vlen; \
+    c->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter16_simple_rvv##vlen;
+
     int flags = av_get_cpu_flags();
 
-    if (flags & AV_CPU_FLAG_RVV_I32 && ff_rv_vlen_least(128)) {
+    if (flags & AV_CPU_FLAG_RVV_I32 && vlenb >= 16) {
 #if __riscv_xlen >= 64
         if (flags & AV_CPU_FLAG_RVV_I64)
             c->vp8_luma_dc_wht = ff_vp8_luma_dc_wht_rvv;
@@ -159,6 +168,13 @@ av_cold void ff_vp8dsp_init_riscv(VP8DSPContext *c)
         c->vp8_idct_dc_add4y = ff_vp8_idct_dc_add4y_rvv;
         if (flags & AV_CPU_FLAG_RVV_I64)
             c->vp8_idct_dc_add4uv = ff_vp8_idct_dc_add4uv_rvv;
+
+        if (vlenb >= 32) {
+            init_loop_filter(256);
+        } else {
+            init_loop_filter(128);
+        }
     }
+#undef init_loop_filter
 #endif
 }
diff --git a/libavcodec/riscv/vp8dsp_rvv.S b/libavcodec/riscv/vp8dsp_rvv.S
index 0cbf1672f7..b5f8bb31b4 100644
--- a/libavcodec/riscv/vp8dsp_rvv.S
+++ b/libavcodec/riscv/vp8dsp_rvv.S
@@ -275,6 +275,93 @@ func ff_vp78_idct_dc_add4uv_rvv, zve64x
         ret
 endfunc
 
+.macro filter_fmin len, vlen, a, f1, p0f2, q0f1
+        vsetvlstatic16  \len, \vlen
+        vsext.vf2       \q0f1, \a
+        vmin.vx         \p0f2, \q0f1, a7
+        vmin.vx         \q0f1, \q0f1, t3
+        vadd.vi         \p0f2, \p0f2, 3
+        vadd.vi         \q0f1, \q0f1, 4
+        vsra.vi         \p0f2, \p0f2, 3
+        vsra.vi         \f1,   \q0f1, 3
+        vadd.vv         \p0f2, \p0f2, v8
+        vsub.vv         \q0f1, v16, \f1
+        vmax.vx         \p0f2, \p0f2, zero
+        vmax.vx         \q0f1, \q0f1, zero
+.endm
+
+.macro filter len, vlen, type, normal, inner, dst, stride, fE, fI, thresh
+.ifc \type,v
+        slli            a6, \stride, 1
+        sub             t2, \dst, a6
+        add             t4, \dst, \stride
+        sub             t1, \dst, \stride
+        vle8.v          v1, (t2)
+        vle8.v          v11, (t4)
+        vle8.v          v17, (t1)
+        vle8.v          v22, (\dst)
+.else
+        addi            t1, \dst, -1
+        addi            a6, \dst, -2
+        addi            t4, \dst, 1
+        vlse8.v         v1, (a6), \stride
+        vlse8.v         v11, (t4), \stride
+        vlse8.v         v17, (t1), \stride
+        vlse8.v         v22, (\dst), \stride
+.endif
+        vwsubu.vv       v12, v1, v11             // p1-q1
+        vwsubu.vv       v24, v22, v17            // q0-p0
+        vnclip.wi       v23, v12, 0
+        vsetvlstatic16  \len, \vlen
+        // vp8_simple_limit(dst + i, stride, flim)
+        li              a7, 2
+        vneg.v          v18, v12
+        vmax.vv         v18, v18, v12
+        vneg.v          v8, v24
+        vmax.vv         v8, v8, v24
+        vsrl.vi         v18, v18, 1
+        vmacc.vx        v18, a7, v8
+        vmsleu.vx       v0, v18, \fE
+
+        li              t5, 3
+        li              a7, 124
+        li              t3, 123
+        vmul.vx         v30, v24, t5
+        vsext.vf2       v4, v23
+        vzext.vf2       v8, v17                  // p0
+        vzext.vf2       v16, v22                 // q0
+        vadd.vv         v12, v30, v4
+        vsetvlstatic8   \len, \vlen
+        vnclip.wi       v11, v12, 0
+        filter_fmin     \len, \vlen, v11, v24, v4, v6
+        vsetvlstatic8   \len, \vlen
+        vnclipu.wi      v4, v4, 0
+        vnclipu.wi      v6, v6, 0
+
+.ifc \type,v
+        vse8.v          v4, (t1), v0.t
+        vse8.v          v6, (\dst), v0.t
+.else
+        vsse8.v         v4, (t1), \stride, v0.t
+        vsse8.v         v6, (\dst), \stride, v0.t
+.endif
+
+.endm
+
+.irp vlen,256,128
+func ff_vp8_v_loop_filter16_simple_rvv\vlen, zve32x
+        vsetvlstatic8   16, \vlen
+        filter 16, \vlen, v, 0, 0, a0, a1, a2, a3, a4
+        ret
+endfunc
+
+func ff_vp8_h_loop_filter16_simple_rvv\vlen, zve32x
+        vsetvlstatic8   16, \vlen
+        filter 16, \vlen, h, 0, 0, a0, a1, a2, a3, a4
+        ret
+endfunc
+.endr
+
 .macro bilin_load_h dst mn
         addi            t5, a2, 1
         vle8.v          \dst, (a2)
-- 
2.45.2