[FFmpeg-devel] [PATCH 15/18] lavu/riscv: float reversed vector multiplication 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  | 22 ++++++++++++++++++++++
 2 files changed, 25 insertions(+)

diff --git a/libavutil/riscv/float_dsp_init.c b/libavutil/riscv/float_dsp_init.c
index 4980214821..e6a5efbf68 100644
--- a/libavutil/riscv/float_dsp_init.c
+++ b/libavutil/riscv/float_dsp_init.c
@@ -30,6 +30,8 @@ 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_fmul_reverse_rvv(float *dst, const float *src0,
+                                 const float *src1, int len);
 void ff_butterflies_float_rvv(float *v1, float *v2, int len);
 
 void ff_vector_dmul_rvv(double *dst, const double *src0, const double *src1,
@@ -49,6 +51,7 @@ av_cold void ff_float_dsp_init_riscv(AVFloatDSPContext *fdsp)
         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;
+        fdsp->vector_fmul_reverse = ff_vector_fmul_reverse_rvv;
         fdsp->butterflies_float = ff_butterflies_float_rvv;
 
         if (flags & AV_CPU_FLAG_ZVE64D) {
diff --git a/libavutil/riscv/float_dsp_rvv.S b/libavutil/riscv/float_dsp_rvv.S
index 1c3b08b94f..b376392294 100644
--- a/libavutil/riscv/float_dsp_rvv.S
+++ b/libavutil/riscv/float_dsp_rvv.S
@@ -92,6 +92,28 @@ func ff_vector_fmul_add_rvv, zve32f
         ret
 endfunc
 
+// (a0) = (a1) * reverse(a2) [0..a3-1]
+func ff_vector_fmul_reverse_rvv, zve32f
+        add      t3, a3, -1
+        li       t2, -4 // byte stride
+        slli     t3, t3, 2
+        add      a2, a2, t3
+
+1:      vsetvli  t0, a3, e32, m8, ta, ma
+        slli     t1, t0, 2
+        vle32.v  v16, (a1)
+        add      a1, a1, t1
+        vlse32.v v24, (a2), t2
+        sub      a2, a2, t1
+        vfmul.vv v16, v16, v24
+        sub      a3, a3, t0
+        vse32.v  v16, (a0)
+        add      a0, a0, t1
+        bnez     a3, 1b
+
+        ret
+endfunc
+
 // (a0) = (a0) + (a1), (a1) = (a0) - (a1) [0..a2-1]
 func ff_butterflies_float_rvv, zve32f
 1:      vsetvli  t0, a2, e32, m8, ta, ma
-- 
2.37.2