[FFmpeg-cvslog] lavc/aacpsdsp: rework R-V V add_squares

[Rémi Denis-Courmont]

ffmpeg | branch: master | Rémi Denis-Courmont <remi at remlab.net> | Sat Jul 15 23:23:50 2023 +0300| [eb508702a899f4a41d3653d90b077b487083d2ab] | committer: Rémi Denis-Courmont

lavc/aacpsdsp: rework R-V V add_squares

Segmented loads may be slower than not. So this advantageously uses a
unit-strided load and narrowing shifts instead.

Before:
ps_add_squares_c: 60757.7
ps_add_squares_rvv_f32: 22242.5

After:
ps_add_squares_c: 60516.0
ps_add_squares_rvv_i64: 17067.7

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=eb508702a899f4a41d3653d90b077b487083d2ab
---

 libavcodec/riscv/aacpsdsp_init.c | 3 ++-
 libavcodec/riscv/aacpsdsp_rvv.S  | 9 ++++++---
 2 files changed, 8 insertions(+), 4 deletions(-)

diff --git a/libavcodec/riscv/aacpsdsp_init.c b/libavcodec/riscv/aacpsdsp_init.c
index c5ec796232..f72d1bc330 100644
--- a/libavcodec/riscv/aacpsdsp_init.c
+++ b/libavcodec/riscv/aacpsdsp_init.c
@@ -46,7 +46,8 @@ av_cold void ff_psdsp_init_riscv(PSDSPContext *c)
         c->hybrid_analysis = ff_ps_hybrid_analysis_rvv;
 
         if (flags & AV_CPU_FLAG_RVB_ADDR) {
-            c->add_squares = ff_ps_add_squares_rvv;
+            if (flags & AV_CPU_FLAG_RVV_I64)
+                c->add_squares = ff_ps_add_squares_rvv;
             c->mul_pair_single = ff_ps_mul_pair_single_rvv;
             c->stereo_interpolate[0] = ff_ps_stereo_interpolate_rvv;
         }
diff --git a/libavcodec/riscv/aacpsdsp_rvv.S b/libavcodec/riscv/aacpsdsp_rvv.S
index fe250cd83b..cf872599c8 100644
--- a/libavcodec/riscv/aacpsdsp_rvv.S
+++ b/libavcodec/riscv/aacpsdsp_rvv.S
@@ -1,5 +1,5 @@
 /*
- * Copyright © 2022 Rémi Denis-Courmont.
+ * Copyright © 2022-2023 Rémi Denis-Courmont.
  *
  * This file is part of FFmpeg.
  *
@@ -20,13 +20,16 @@
 
 #include "libavutil/riscv/asm.S"
 
-func ff_ps_add_squares_rvv, zve32f
+func ff_ps_add_squares_rvv, zve64f
+        li          t1, 32
 1:
         vsetvli     t0, a2, e32, m4, ta, ma
-        vlseg2e32.v v24, (a1)
+        vle64.v     v8, (a1)
         sub         a2, a2, t0
+        vnsrl.wx    v24, v8, zero
         vle32.v     v16, (a0)
         sh3add      a1, t0, a1
+        vnsrl.wx    v28, v8, t1
         vfmacc.vv   v16, v24, v24
         vfmacc.vv   v16, v28, v28
         vse32.v     v16, (a0)