[FFmpeg-devel] [PATCH 13/18] lavu/riscv: float vector multiplication-addition with RVV

[remi at remlab.net]

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

---
 libavutil/riscv/float_dsp_init.c |  3 +++
 libavutil/riscv/float_dsp_rvv.S  | 19 +++++++++++++++++++
 2 files changed, 22 insertions(+)

diff --git a/libavutil/riscv/float_dsp_init.c b/libavutil/riscv/float_dsp_init.c
index b63da72acd..9b31ed2ed1 100644
--- a/libavutil/riscv/float_dsp_init.c
+++ b/libavutil/riscv/float_dsp_init.c
@@ -28,6 +28,8 @@ void ff_vector_fmac_scalar_rvv(float *dst, const float *src, float mul,
                                 int len);
 void ff_vector_fmul_scalar_rvv(float *dst, const float *src, float mul,
                                 int len);
+void ff_vector_fmul_add_rvv(float *dst, const float *src0, const float *src1,
+                             const float *src2, int len);
 
 void ff_vector_dmul_rvv(double *dst, const double *src0, const double *src1,
                          int len);
@@ -45,6 +47,7 @@ av_cold void ff_float_dsp_init_riscv(AVFloatDSPContext *fdsp)
         fdsp->vector_fmul = ff_vector_fmul_rvv;
         fdsp->vector_fmac_scalar = ff_vector_fmac_scalar_rvv;
         fdsp->vector_fmul_scalar = ff_vector_fmul_scalar_rvv;
+        fdsp->vector_fmul_add = ff_vector_fmul_add_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 5a7d92abd6..efbf12179f 100644
--- a/libavutil/riscv/float_dsp_rvv.S
+++ b/libavutil/riscv/float_dsp_rvv.S
@@ -73,6 +73,25 @@ NOHWF   mv       a2, a3
         ret
 endfunc
 
+// (a0) = (a1) * (a2) + (a3) [0..a4-1]
+func ff_vector_fmul_add_rvv, zve32f
+1:      vsetvli   t0, a4, e32, m8, ta, ma
+        slli      t1, t0, 2
+        vle32.v   v8, (a1)
+        add       a1, a1, t1
+        vle32.v   v16, (a2)
+        add       a2, a2, t1
+        vle32.v   v24, (a3)
+        add       a3, a3, t1
+        vfmadd.vv v8, v16, v24
+        sub       a4, a4, t0
+        vse32.v   v8, (a0)
+        add       a0, a0, t1
+        bnez      a4, 1b
+
+        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