[FFmpeg-cvslog] lavc/h264dsp: stick R-V V weight to 16-bit precision

[Rémi Denis-Courmont]

ffmpeg | branch: master | Rémi Denis-Courmont <remi at remlab.net> | Tue Jul 30 20:29:02 2024 +0300| [677f28b310ac8cbf96e3a02865c2aff49c6950ff] | committer: Rémi Denis-Courmont

lavc/h264dsp: stick R-V V weight to 16-bit precision

T-Head C908 (ns):
h264_weight2_8_c:        1607.8
h264_weight2_8_rvv_i32:   515.0 (before)
h264_weight2_8_rvv_i32:   348.5 (after)
h264_weight4_8_c:        2255.8
h264_weight4_8_rvv_i32:  1015.0 (before)
h264_weight4_8_rvv_i32:   691.0 (after)
h264_weight8_8_c:        3857.5
h264_weight8_8_rvv_i32:  2218.8 (before)
h264_weight8_8_rvv_i32:  1561.3 (after)
h264_weight16_8_c:       7431.5
h264_weight16_8_rvv_i32: 2737.3 (before)
h264_weight16_8_rvv_i32: 1848.3 (after)

SpacemiT X60 (ns):
h264_weight2_8_c:        1624.1
h264_weight2_8_rvv_i32:   352.6 (before)
h264_weight2_8_rvv_i32:   259.3 (after)
h264_weight4_8_c:        2259.3
h264_weight4_8_rvv_i32:   685.8 (before)
h264_weight4_8_rvv_i32:   530.3 (after)
h264_weight8_8_c:        4103.3
h264_weight8_8_rvv_i32:  1581.8 (before)
h264_weight8_8_rvv_i32:  1238.6 (after)
h264_weight16_8_c:       7624.3
h264_weight16_8_rvv_i32: 2738.1 (before)
h264_weight16_8_rvv_i32: 1853.3 (after)

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=677f28b310ac8cbf96e3a02865c2aff49c6950ff
---

 libavcodec/riscv/h264dsp_rvv.S | 37 ++++++++++++++++---------------------
 1 file changed, 16 insertions(+), 21 deletions(-)

diff --git a/libavcodec/riscv/h264dsp_rvv.S b/libavcodec/riscv/h264dsp_rvv.S
index 0e08de43e4..0d641008cb 100644
--- a/libavcodec/riscv/h264dsp_rvv.S
+++ b/libavcodec/riscv/h264dsp_rvv.S
@@ -32,17 +32,15 @@ func ff_h264_weight_pixels_simple_8_rvv, zve32x
         csrwi   vxrm, 0
         sll     a5, a5, a3
 1:
-        vsetvli zero, a6, e32, m4, ta, ma
+        vsetvli     zero, a6, e16, m2, ta, ma
         vle8.v  v8, (a0)
         addi    a2, a2, -1
-        vmv.v.x v16, a5
-        vsetvli zero, zero, e16, m2, ta, ma
         vzext.vf2   v24, v8
-        vwmaccsu.vx v16, a4, v24
-        vnclip.wx   v16, v16, a3
+        vmul.vx     v16, v24, a4
+        vsadd.vx    v16, v16, a5
         vmax.vx v16, v16, zero
         vsetvli zero, zero, e8, m1, ta, ma
-        vnclipu.wi  v8, v16, 0
+        vnclipu.wx  v8, v16, a3
         vse8.v  v8, (a0)
         add     a0, a0, a1
         bnez    a2, 1b
@@ -85,23 +83,20 @@ func ff_h264_weight_pixels_8_rvv, zve32x
         mv      t0, a0
         mv      t6, a6
 2:
-        vsetvli t2, a2, e32, m8, ta, ma
+        vsetvli     t2, a2, e16, m8, ta, ma
         vlsseg2e8.v v0, (t0), a1
         addi    t6, t6, -2
-        vmv.v.x v16, a5
-        vmv.v.x v24, a5
-        vsetvli zero, zero, e16, m4, ta, ma
-        vzext.vf2   v8, v0
-        vzext.vf2   v12, v2
-        vwmaccsu.vx v16, a4, v8
-        vwmaccsu.vx v24, a4, v12
-        vnclip.wx   v8, v16, a3
-        vnclip.wx   v12, v24, a3
-        vmax.vx v8, v8, zero
-        vmax.vx v12, v12, zero
-        vsetvli zero, zero, e8, m2, ta, ma
-        vnclipu.wi  v0, v8, 0
-        vnclipu.wi  v2, v12, 0
+        vzext.vf2   v16, v0
+        vzext.vf2   v24, v4
+        vmul.vx     v16, v16, a4
+        vmul.vx     v24, v24, a4
+        vsadd.vx    v16, v16, a5
+        vsadd.vx    v24, v24, a5
+        vmax.vx     v16, v16, zero
+        vmax.vx     v24, v24, zero
+        vsetvli     zero, zero, e8, m4, ta, ma
+        vnclipu.wx  v0, v16, a3
+        vnclipu.wx  v4, v24, a3
         vssseg2e8.v v0, (t0), a1
         addi    t0, t0, 2
         bnez    t6, 2b