[FFmpeg-devel] [PATCH 31/31] lavc/aacpsdsp: RISC-V V stereo_interpolate[0]

[remi at remlab.net]

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

---
 libavcodec/riscv/aacpsdsp_init.c |  4 ++
 libavcodec/riscv/aacpsdsp_rvv.S  | 65 ++++++++++++++++++++++++++++++++
 2 files changed, 69 insertions(+)

diff --git a/libavcodec/riscv/aacpsdsp_init.c b/libavcodec/riscv/aacpsdsp_init.c
index 20b1a12741..58a4c61121 100644
--- a/libavcodec/riscv/aacpsdsp_init.c
+++ b/libavcodec/riscv/aacpsdsp_init.c
@@ -34,6 +34,9 @@ void ff_ps_hybrid_analysis_ileave_rvv(float (*out)[32][2], float L[2][38][64],
 void ff_ps_hybrid_synthesis_deint_rvv(float out[2][38][64], float (*in)[32][2],
                                       int i, int len);
 
+void ff_ps_stereo_interpolate_rvv(float (*l)[2], float (*r)[2],
+                                  float h[2][4], float h_step[2][4], int len);
+
 av_cold void ff_psdsp_init_riscv(PSDSPContext *c)
 {
 #if HAVE_RVV
@@ -47,6 +50,7 @@ av_cold void ff_psdsp_init_riscv(PSDSPContext *c)
             c->add_squares = ff_ps_add_squares_rvv;
             c->mul_pair_single = ff_ps_mul_pair_single_rvv;
             c->hybrid_analysis = ff_ps_hybrid_analysis_rvv;
+            c->stereo_interpolate[0] = ff_ps_stereo_interpolate_rvv;
         }
     }
 #endif
diff --git a/libavcodec/riscv/aacpsdsp_rvv.S b/libavcodec/riscv/aacpsdsp_rvv.S
index 0cbe4c1d3c..a236dfe43c 100644
--- a/libavcodec/riscv/aacpsdsp_rvv.S
+++ b/libavcodec/riscv/aacpsdsp_rvv.S
@@ -219,3 +219,68 @@ func ff_ps_hybrid_synthesis_deint_rvv, zve32x
 3:
         ret
 endfunc
+
+func ff_ps_stereo_interpolate_rvv, zve32f
+        vsetvli      t0, zero, e32, m1, ta, ma
+        vid.v        v24
+        flw          ft0,   (a2)
+        vadd.vi      v24, v24, 1   // v24[i] = i + 1
+        flw          ft1,  4(a2)
+        vfcvt.f.xu.v v24, v24
+        flw          ft2,  8(a2)
+        vfmv.v.f     v16, ft0
+        flw          ft3, 12(a2)
+        vfmv.v.f     v17, ft1
+        flw          ft0,   (a3)
+        vfmv.v.f     v18, ft2
+        flw          ft1,  4(a3)
+        vfmv.v.f     v19, ft3
+        flw          ft2,  8(a3)
+        vfmv.v.f     v20, ft0
+        flw          ft3, 12(a3)
+        vfmv.v.f     v21, ft1
+        fcvt.s.wu    ft4, t0       // (float)(vlenb / sizeof (float))
+        vfmv.v.f     v22, ft2
+        li           t1, 8
+        vfmv.v.f     v23, ft3
+        addi         a6, a0, 4     // l[*][1]
+        vfmacc.vv    v16, v24, v20 // h0 += (i + 1) * h0_step
+        addi         a7, a1, 4     // r[*][1]
+        vfmacc.vv    v17, v24, v21
+        fmul.s       ft0, ft0, ft4
+        vfmacc.vv    v18, v24, v22
+        fmul.s       ft1, ft1, ft4
+        vfmacc.vv    v19, v24, v23
+        fmul.s       ft2, ft2, ft4
+        fmul.s       ft3, ft3, ft4
+1:
+        vsetvli   t0, a4, e32, m1, ta, ma
+        vlse32.v  v8, (a0), t1   // l_re
+        sub       a4, a4, t0
+        vlse32.v  v9, (a6), t1   // l_im
+        vlse32.v  v10, (a1), t1  // r_re
+        vlse32.v  v11, (a7), t1  // r_im
+        vfmul.vv  v12, v8, v16
+        vfmul.vv  v13, v9, v16
+        vfmul.vv  v14, v8, v17
+        vfmul.vv  v15, v9, v17
+        vfmacc.vv v12, v10, v18
+        vfmacc.vv v13, v11, v18
+        vfmacc.vv v14, v10, v19
+        vfmacc.vv v15, v11, v19
+        vsse32.v  v12, (a0), t1
+        sh3add    a0, t0, a0
+        vsse32.v  v13, (a6), t1
+        sh3add    a6, t0, a6
+        vsse32.v  v14, (a1), t1
+        sh3add    a1, t0, a1
+        vsse32.v  v15, (a7), t1
+        sh3add    a7, t0, a7
+        vfadd.vf  v16, v16, ft0 // h0 += (vlenb / sizeof (float)) * h0_step
+        vfadd.vf  v17, v17, ft1
+        vfadd.vf  v18, v18, ft2
+        vfadd.vf  v19, v19, ft3
+        bnez      a4, 1b
+
+        ret
+endfunc
-- 
2.37.2