[FFmpeg-devel] [PATCH] lavc/audiodsp: rework RISC-V V scalar product

[Rémi Denis-Courmont]

Take vector reduction out of the loop and unroll.

Before:
audiodsp.scalarproduct_int16_c: 12321.0
audiodsp.scalarproduct_int16_rvv_i32: 4175.7

After:
audiodsp.scalarproduct_int16_c: 12320.5
audiodsp.scalarproduct_int16_rvv_i32: 1230.2
---
 libavcodec/riscv/audiodsp_rvv.S | 15 ++++++++-------
 1 file changed, 8 insertions(+), 7 deletions(-)

diff --git a/libavcodec/riscv/audiodsp_rvv.S b/libavcodec/riscv/audiodsp_rvv.S
index af1e07bef9..f7eba2114f 100644
--- a/libavcodec/riscv/audiodsp_rvv.S
+++ b/libavcodec/riscv/audiodsp_rvv.S
@@ -21,21 +21,22 @@
 #include "libavutil/riscv/asm.S"
 
 func ff_scalarproduct_int16_rvv, zve32x
-        vsetivli    zero, 1, e32, m1, ta, ma
-        vmv.s.x     v8, zero
+        vsetvli     t0, zero, e32, m8, ta, ma
+        vmv.v.x     v8, zero
+        vmv.s.x     v0, zero
 1:
-        vsetvli     t0, a2, e16, m1, ta, ma
+        vsetvli     t0, a2, e16, m4, tu, ma
         vle16.v     v16, (a0)
         sub         a2, a2, t0
         vle16.v     v24, (a1)
         sh1add      a0, t0, a0
-        vwmul.vv    v0, v16, v24
+        vwmacc.vv   v8, v16, v24
         sh1add      a1, t0, a1
-        vsetvli     zero, t0, e32, m2, ta, ma
-        vredsum.vs  v8, v0, v8
         bnez        a2, 1b
 
-        vmv.x.s     a0, v8
+        vsetvli     t0, zero, e32, m8, ta, ma
+        vredsum.vs  v0, v8, v0
+        vmv.x.s     a0, v0
         ret
 endfunc
 
-- 
2.40.1