[FFmpeg-devel] [PATCH 06/12] lavu/riscv: float vector multiply-accumulate with RVV

[remi at remlab.net]

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

---
 libavutil/riscv/float_dsp_init.c |  6 +++++
 libavutil/riscv/float_dsp_rvv.S  | 38 ++++++++++++++++++++++++++++++++
 2 files changed, 44 insertions(+)

diff --git a/libavutil/riscv/float_dsp_init.c b/libavutil/riscv/float_dsp_init.c
index 4135284c76..a1bb112ec7 100644
--- a/libavutil/riscv/float_dsp_init.c
+++ b/libavutil/riscv/float_dsp_init.c
@@ -24,11 +24,15 @@
 
 void ff_vector_fmul_rvv(float *dst, const float *src0, const float *src1,
                          int len);
+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_dmul_rvv(double *dst, const double *src0, const double *src1,
                          int len);
+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);
 
@@ -38,10 +42,12 @@ av_cold void ff_float_dsp_init_riscv(AVFloatDSPContext *fdsp)
 
     if (flags & AV_CPU_FLAG_ZVE32F) {
         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;
 
         if (flags & AV_CPU_FLAG_ZVE64D) {
             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;
         }
     }
diff --git a/libavutil/riscv/float_dsp_rvv.S b/libavutil/riscv/float_dsp_rvv.S
index 65c3a77b01..5a7d92abd6 100644
--- a/libavutil/riscv/float_dsp_rvv.S
+++ b/libavutil/riscv/float_dsp_rvv.S
@@ -36,6 +36,25 @@ func ff_vector_fmul_rvv, zve32f
         ret
 endfunc
 
+// (a0) += (a1) * fa0 [0..a2-1]
+func ff_vector_fmac_scalar_rvv, zve32f
+NOHWF   fmv.w.x   fa0, a2
+NOHWF   mv        a2, a3
+
+1:      vsetvli   t0, a2, e32, m8, ta, ma
+        slli      t1, t0, 2
+        vle32.v   v24, (a1)
+        add       a1, a1, t1
+        vle32.v   v16, (a0)
+        vfmacc.vf v16, fa0, v24
+        sub       a2, a2, t0
+        vse32.v   v16, (a0)
+        add       a0, a0, t1
+        bnez      a2, 1b
+
+        ret
+endfunc
+
 // (a0) = (a1) * fa0 [0..a2-1]
 func ff_vector_fmul_scalar_rvv, zve32f
 NOHWF   fmv.w.x  fa0, a2
@@ -71,6 +90,25 @@ func ff_vector_dmul_rvv, zve64d
         ret
 endfunc
 
+// (a0) += (a1) * fa0 [0..a2-1]
+func ff_vector_dmac_scalar_rvv, zve64d
+NOHWD   fmv.d.x   fa0, a2
+NOHWD   mv        a2, a3
+
+1:      vsetvli   t0, a2, e64, m8, ta, ma
+        slli      t1, t0, 3
+        vle64.v   v24, (a1)
+        add       a1, a1, t1
+        vle64.v   v16, (a0)
+        vfmacc.vf v16, fa0, v24
+        sub       a2, a2, t0
+        vse64.v   v16, (a0)
+        add       a0, a0, t1
+        bnez      a2, 1b
+
+        ret
+endfunc
+
 // (a0) = (a1) * fa0 [0..a2-1]
 func ff_vector_dmul_scalar_rvv, zve64d
 NOHWD   fmv.d.x  fa0, a2
-- 
2.37.2