[FFmpeg-devel] [PATCH 16/18] lavu/riscv: float vector windowed overlap/add 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  | 35 ++++++++++++++++++++++++++++++++
 2 files changed, 38 insertions(+)

diff --git a/libavutil/riscv/float_dsp_init.c b/libavutil/riscv/float_dsp_init.c
index ae089d2fdb..cf8c995d7c 100644
--- a/libavutil/riscv/float_dsp_init.c
+++ b/libavutil/riscv/float_dsp_init.c
@@ -29,6 +29,8 @@ 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_window_rvv(float *dst, const float *src0,
+                                const float *src1, const float *win, 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,
@@ -51,6 +53,7 @@ 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_window = ff_vector_fmul_window_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;
diff --git a/libavutil/riscv/float_dsp_rvv.S b/libavutil/riscv/float_dsp_rvv.S
index b376392294..65daaa2d27 100644
--- a/libavutil/riscv/float_dsp_rvv.S
+++ b/libavutil/riscv/float_dsp_rvv.S
@@ -73,6 +73,41 @@ NOHWF   mv       a2, a3
         ret
 endfunc
 
+func ff_vector_fmul_window_rvv, zve32f
+        // a0: dst, a1: src0, a2: src1, a3: window, a4: length
+        addi       t0, a4, -1
+        add        t1, t0, a4
+        slli       t0, t0, 2
+        slli       t1, t1, 2
+        add        a2, a2, t0
+        add        t0, a0, t1
+        add        t3, a3, t1
+        li         t1, -4 // byte stride
+
+1:      vsetvli    t2, a4, e32, m4, ta, ma
+        slli       t4, t2, 2
+        vle32.v    v16, (a1)
+        add        a1, a1, t4
+        vlse32.v   v20, (a2), t1
+        sub        a2, a2, t4
+        vle32.v    v24, (a3)
+        add        a3, a3, t4
+        vlse32.v   v28, (t3), t1
+        sub        t3, t3, t4
+        vfmul.vv   v0, v16, v28
+        sub        a4, a4, t2
+        vfmul.vv   v8, v16, v24
+        vfnmsac.vv v0, v20, v24
+        vfmacc.vv  v8, v20, v28
+        vse32.v    v0, (a0)
+        add        a0, a0, t4
+        vsse32.v   v8, (t0), t1
+        sub        t0, t0, t4
+        bnez       a4, 1b
+
+        ret
+endfunc
+
 // (a0) = (a1) * (a2) + (a3) [0..a4-1]
 func ff_vector_fmul_add_rvv, zve32f
 1:      vsetvli   t0, a4, e32, m8, ta, ma
-- 
2.37.2