[FFmpeg-devel] [PATCH 11/12] lavu/riscv: float vector dot product with RVV

[remi at remlab.net]

From: Rémi Denis-Courmont <remi at remlab.net>

---
 libavutil/riscv/float_dsp_init.c |  2 ++
 libavutil/riscv/float_dsp_rvv.S  | 21 +++++++++++++++++++++
 2 files changed, 23 insertions(+)

diff --git a/libavutil/riscv/float_dsp_init.c b/libavutil/riscv/float_dsp_init.c
index 887706d899..7c2fc10e99 100644
--- a/libavutil/riscv/float_dsp_init.c
+++ b/libavutil/riscv/float_dsp_init.c
@@ -35,6 +35,7 @@ void ff_vector_fmul_add_rvv(float *dst, const float *src0, const float *src1,
 void ff_vector_fmul_reverse_rvv(float *dst, const float *src0,
                                  const float *src1, int len);
 void ff_butterflies_float_rvv(float *v1, float *v2, int len);
+float ff_scalarproduct_float_rvv(const float *v1, const float *v2, int len);
 
 void ff_vector_dmul_rvv(double *dst, const double *src0, const double *src1,
                          int len);
@@ -55,6 +56,7 @@ av_cold void ff_float_dsp_init_riscv(AVFloatDSPContext *fdsp)
         fdsp->vector_fmul_add = ff_vector_fmul_add_rvv;
         fdsp->vector_fmul_reverse = ff_vector_fmul_reverse_rvv;
         fdsp->butterflies_float = ff_butterflies_float_rvv;
+        fdsp->scalarproduct_float = ff_scalarproduct_float_rvv;
 
         if (flags & AV_CPU_FLAG_ZVE64D) {
             fdsp->vector_dmul = ff_vector_dmul_rvv;
diff --git a/libavutil/riscv/float_dsp_rvv.S b/libavutil/riscv/float_dsp_rvv.S
index 65daaa2d27..81bd0e510a 100644
--- a/libavutil/riscv/float_dsp_rvv.S
+++ b/libavutil/riscv/float_dsp_rvv.S
@@ -167,6 +167,27 @@ func ff_butterflies_float_rvv, zve32f
         ret
 endfunc
 
+// a0 = (a0).(a1) [0..a2-1]
+func ff_scalarproduct_float_rvv, zve32f
+        vsetvli      zero, zero, e32, m8, ta, ma
+        vmv.s.x      v8, zero
+
+1:      vsetvli      t0, a2, e32, m8, ta, ma
+        slli         t1, t0, 2
+        vle32.v      v16, (a0)
+        add          a0, a0, t1
+        vle32.v      v24, (a1)
+        add          a1, a1, t1
+        vfmul.vv     v16, v16, v24
+        sub          a2, a2, t0
+        vfredusum.vs v8, v16, v8
+        bnez         a2, 1b
+
+        vfmv.f.s fa0, v8
+NOHWF   fmv.x.w  a0, fa0
+        ret
+endfunc
+
 // (a0) = (a1) * (a2) [0..a3-1]
 func ff_vector_dmul_rvv, zve64d
 1:      vsetvli  t0, a3, e64, m8, ta, ma
-- 
2.37.2