[FFmpeg-cvslog] lavc/llauddsp: R-V V scalarproduct_and_madd_int32

[Rémi Denis-Courmont]

ffmpeg | branch: master | Rémi Denis-Courmont <remi at remlab.net> | Sun Nov 12 22:12:53 2023 +0200| [d076517056ca5914ecbb1bcbd50799615c568b4c] | committer: Rémi Denis-Courmont

lavc/llauddsp: R-V V scalarproduct_and_madd_int32

scalarproduct_and_madd_int32_c:      10899.7
scalarproduct_and_madd_int32_rvv_i32: 1749.0

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=d076517056ca5914ecbb1bcbd50799615c568b4c
---

 libavcodec/riscv/llauddsp_init.c |  4 ++++
 libavcodec/riscv/llauddsp_rvv.S  | 26 ++++++++++++++++++++++++++
 2 files changed, 30 insertions(+)

diff --git a/libavcodec/riscv/llauddsp_init.c b/libavcodec/riscv/llauddsp_init.c
index ea023f73e6..1924b36821 100644
--- a/libavcodec/riscv/llauddsp_init.c
+++ b/libavcodec/riscv/llauddsp_init.c
@@ -27,6 +27,9 @@
 int32_t ff_scalarproduct_and_madd_int16_rvv(int16_t *v1, const int16_t *v2,
                                             const int16_t *v3, int len,
                                             int mul);
+int32_t ff_scalarproduct_and_madd_int32_rvv(int16_t *v1, const int32_t *v2,
+                                            const int16_t *v3, int len,
+                                            int mul);
 
 av_cold void ff_llauddsp_init_riscv(LLAudDSPContext *c)
 {
@@ -35,6 +38,7 @@ av_cold void ff_llauddsp_init_riscv(LLAudDSPContext *c)
 
     if ((flags & AV_CPU_FLAG_RVV_I32)  && (flags & AV_CPU_FLAG_RVB_ADDR)) {
         c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_rvv;
+        c->scalarproduct_and_madd_int32 = ff_scalarproduct_and_madd_int32_rvv;
     }
 #endif
 }
diff --git a/libavcodec/riscv/llauddsp_rvv.S b/libavcodec/riscv/llauddsp_rvv.S
index 74adb338a9..5569864832 100644
--- a/libavcodec/riscv/llauddsp_rvv.S
+++ b/libavcodec/riscv/llauddsp_rvv.S
@@ -43,3 +43,29 @@ func ff_scalarproduct_and_madd_int16_rvv, zve32x
         vmv.x.s a0, v0
         ret
 endfunc
+
+func ff_scalarproduct_and_madd_int32_rvv, zve32x
+        vsetvli t0, zero, e32, m8, ta, ma
+        vmv.v.x v0, zero
+1:
+        vsetvli t0, a3, e32, m8, tu, ma
+        vle16.v v8, (a0)
+        sub     a3, a3, t0
+        vsext.vf2 v24, v8
+        vle32.v v16, (a1)
+        sh2add  a1, t0, a1
+        vmacc.vv v0, v16, v24
+        vsetvli zero, zero, e16, m4, ta, ma
+        vle16.v v24, (a2)
+        sh1add  a2, t0, a2
+        vmacc.vx v8, a4, v24
+        vse16.v v8, (a0)
+        sh1add  a0, t0, a0
+        bnez    a3, 1b
+
+        vsetvli t0, zero, e32, m8, ta, ma
+        vmv.s.x v8, zero
+        vredsum.vs v0, v0, v8
+        vmv.x.s a0, v0
+        ret
+endfunc