[FFmpeg-devel] [PATCH v4 4/4] lavc/vp9dsp: R-V V mc tap hv

[uk7b at foxmail.com]

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

                                                     C908   X60
vp9_avg_8tap_smooth_4hv_8bpp_c                     :   32.0   28.0
vp9_avg_8tap_smooth_4hv_8bpp_rvv_i32               :   15.0   13.2
vp9_avg_8tap_smooth_8hv_8bpp_c                     :   98.0   86.2
vp9_avg_8tap_smooth_8hv_8bpp_rvv_i32               :   23.7   21.2
vp9_avg_8tap_smooth_16hv_8bpp_c                    :  355.7  297.0
vp9_avg_8tap_smooth_16hv_8bpp_rvv_i32              :   47.0   41.5
vp9_avg_8tap_smooth_32hv_8bpp_c                    : 1272.7 1099.7
vp9_avg_8tap_smooth_32hv_8bpp_rvv_i32              :  134.7  119.7
vp9_avg_8tap_smooth_64hv_8bpp_c                    : 4937.0 4224.2
vp9_avg_8tap_smooth_64hv_8bpp_rvv_i32              :  528.5  228.5
vp9_put_8tap_smooth_4hv_8bpp_c                     :   30.2   26.7
vp9_put_8tap_smooth_4hv_8bpp_rvv_i32               :   30.5   12.5
vp9_put_8tap_smooth_8hv_8bpp_c                     :   91.5   81.2
vp9_put_8tap_smooth_8hv_8bpp_rvv_i32               :   22.7   20.2
vp9_put_8tap_smooth_16hv_8bpp_c                    :  313.2  277.5
vp9_put_8tap_smooth_16hv_8bpp_rvv_i32              :   45.2   40.2
vp9_put_8tap_smooth_32hv_8bpp_c                    : 1166.7 1022.2
vp9_put_8tap_smooth_32hv_8bpp_rvv_i32              :  131.7  117.2
vp9_put_8tap_smooth_64hv_8bpp_c                    : 4560.5 3961.7
vp9_put_8tap_smooth_64hv_8bpp_rvv_i32              :  517.0  223.2
---
 libavcodec/riscv/vp9_mc_rvv.S  | 75 ++++++++++++++++++++++++++++++++++
 libavcodec/riscv/vp9dsp_init.c |  8 ++++
 2 files changed, 83 insertions(+)

diff --git a/libavcodec/riscv/vp9_mc_rvv.S b/libavcodec/riscv/vp9_mc_rvv.S
index 5e81301aa5..474c9035ae 100644
--- a/libavcodec/riscv/vp9_mc_rvv.S
+++ b/libavcodec/riscv/vp9_mc_rvv.S
@@ -373,6 +373,77 @@ func ff_\op\()_vp9_8tap_\name\()_\len\()\type\()_rvv\vlen\(), zve32x
 endfunc
 .endm
 
+#if __riscv_xlen == 64
+.macro epel_hv_once len, name, op
+        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, \op, \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, \op, \name, v, 0, s
+        vse8.v          v30, (a0)
+.endm
+
+.macro epel_hv op, name, len, vlen
+func ff_\op\()_vp9_8tap_\name\()_\len\()hv_rvv\vlen\(), zve32x
+        addi            sp, sp, -64
+        .irp n,0,1,2,3,4,5,6,7
+        sd              s\n, \n\()<<3(sp)
+        .endr
+.if \len == 64 && \vlen < 256
+        addi            sp, sp, -48
+        .irp n,0,1,2,3,4,5
+        sd              a\n, \n\()<<3(sp)
+        .endr
+.endif
+.ifc \op,avg
+        csrwi           vxrm, 0
+.endif
+        epel_filter     \name, h, t
+        epel_filter     \name, v, s
+.if \vlen < 256
+        vsetvlstatic8   \len, a6, 32, m2
+.else
+        vsetvlstatic8   \len, a6, 64, m2
+.endif
+        epel_hv_once    \len, \name, \op
+.if \len == 64 && \vlen < 256
+        .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, \op
+.endif
+        .irp n,0,1,2,3,4,5,6,7
+        ld              s\n, \n\()<<3(sp)
+        .endr
+        addi            sp, sp, 64
+
+        ret
+endfunc
+.endm
+#endif
+
 .irp len, 64, 32, 16, 8, 4
         copy_avg \len
         .irp op, put, avg
@@ -381,6 +452,10 @@ endfunc
                                 epel \len, \op, \name, \type, 128
                                 epel \len, \op, \name, \type, 256
                         .endr
+                        #if __riscv_xlen == 64
+                        epel_hv \op, \name, \len, 128
+                        epel_hv \op, \name, \len, 256
+                        #endif
                 .endr
         .endr
 .endr
diff --git a/libavcodec/riscv/vp9dsp_init.c b/libavcodec/riscv/vp9dsp_init.c
index 3669070fca..7b090c9889 100644
--- a/libavcodec/riscv/vp9dsp_init.c
+++ b/libavcodec/riscv/vp9dsp_init.c
@@ -119,6 +119,10 @@ static av_cold void vp9dsp_mc_init_riscv(VP9DSPContext *dsp, int bpp)
     if (flags & AV_CPU_FLAG_RVB_ADDR) {
         init_subpel2(0, 0, 1, v, put, 128);
         init_subpel2(1, 0, 1, v, avg, 128);
+# if __riscv_xlen == 64
+        init_subpel2(0, 1, 1, hv, put, 128);
+        init_subpel2(1, 1, 1, hv, avg, 128);
+# endif
     }
 
     }
@@ -129,6 +133,10 @@ static av_cold void vp9dsp_mc_init_riscv(VP9DSPContext *dsp, int bpp)
         if (flags & AV_CPU_FLAG_RVB_ADDR) {
             init_subpel2(0, 0, 1, v, put, 256);
             init_subpel2(1, 0, 1, v, avg, 256);
+# if __riscv_xlen == 64
+            init_subpel2(0, 1, 1, hv, put, 256);
+            init_subpel2(1, 1, 1, hv, avg, 256);
+# endif
         }
     }
     }
-- 
2.45.2