[FFmpeg-devel] [PATCH 5/5] lavu/float_dsp: R-V V scalarproduct_double

[Rémi Denis-Courmont]

C908:
scalarproduct_double_c:       39.2
scalarproduct_double_rvv_f64: 10.5

X60:
scalarproduct_double_c:       35.0
scalarproduct_double_rvv_f64:  5.2
---
 libavutil/riscv/float_dsp_init.c |  3 +++
 libavutil/riscv/float_dsp_rvv.S  | 21 +++++++++++++++++++++
 2 files changed, 24 insertions(+)

diff --git a/libavutil/riscv/float_dsp_init.c b/libavutil/riscv/float_dsp_init.c
index 585f237225..155496fa6b 100644
--- a/libavutil/riscv/float_dsp_init.c
+++ b/libavutil/riscv/float_dsp_init.c
@@ -46,6 +46,8 @@ void ff_vector_dmac_scalar_rvv(double *dst, const double *src, double mul,
                                 int len);
 void ff_vector_dmul_scalar_rvv(double *dst, const double *src, double mul,
                                 int len);
+double ff_scalarproduct_double_rvv(const double *v1, const double *v2,
+                                   size_t len);
 
 av_cold void ff_float_dsp_init_riscv(AVFloatDSPContext *fdsp)
 {
@@ -68,6 +70,7 @@ av_cold void ff_float_dsp_init_riscv(AVFloatDSPContext *fdsp)
             fdsp->vector_dmul = ff_vector_dmul_rvv;
             fdsp->vector_dmac_scalar = ff_vector_dmac_scalar_rvv;
             fdsp->vector_dmul_scalar = ff_vector_dmul_scalar_rvv;
+            fdsp->scalarproduct_double = ff_scalarproduct_double_rvv;
         }
     }
 #endif
diff --git a/libavutil/riscv/float_dsp_rvv.S b/libavutil/riscv/float_dsp_rvv.S
index e6ec182a7a..2f0ade6db6 100644
--- a/libavutil/riscv/float_dsp_rvv.S
+++ b/libavutil/riscv/float_dsp_rvv.S
@@ -249,3 +249,24 @@ NOHWD   mv       a2, a3
 
         ret
 endfunc
+
+func ff_scalarproduct_double_rvv, zve64f
+        vsetvli      t0, zero, e64, m8, ta, ma
+        vmv.v.x      v8, zero
+        vmv.s.x      v0, zero
+1:
+        vsetvli      t0, a2, e64, m8, tu, ma
+        vle64.v      v16, (a0)
+        sub          a2, a2, t0
+        vle64.v      v24, (a1)
+        sh3add       a0, t0, a0
+        vfmacc.vv    v8, v16, v24
+        sh3add       a1, t0, a1
+        bnez         a2, 1b
+
+        vsetvli      t0, zero, e64, m8, ta, ma
+        vfredusum.vs v0, v8, v0
+        vfmv.f.s     fa0, v0
+NOHWD   fmv.x.w      a0, fa0
+        ret
+endfunc
-- 
2.45.1