[FFmpeg-cvslog] lavc/vp8dsp: rework R-V V idct_dc_add4y

[Rémi Denis-Courmont]

ffmpeg | branch: master | Rémi Denis-Courmont <remi at remlab.net> | Sun Jun  2 12:13:25 2024 +0300| [225de53c9d446ddf1cc3ece6e99c06c8cce9b78f] | committer: Rémi Denis-Courmont

lavc/vp8dsp: rework R-V V idct_dc_add4y

DCT-related FFmpeg functions often add an unsigned 8-bit sample to a
signed 16-bit coefficient, then clip the result back to an unsigned
8-bit value. RISC-V has no signed 16-bit to unsigned 8-bit clip, so
instead our most common sequence is:
    VWADDU.WV
    set SEW to 16 bits
    VMAX.VV zero     # clip negative values to 0
    set SEW to 8 bits
    VNCLIPU.WI       # clip values over 255 to 255 and narrow

Here we use a different sequence which does not require toggling the
vector type. This assumes that the wide addend vector is biased by
-128:
    VWADDU.WV
    VNCLIP.WI    # clip values to signed 8-bit and narrow
    VXOR.VX 0x80 # flip sign bit (convert signed to unsigned)

Also the VMAX is effectively replaced by a VXOR of half-width. In this
function, this comes for free as we anyway add a constant to the wide
vector in the prologue.

On C908, this has no observable effects. On X60, this improves
microbenchmarks by about 20%.

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=225de53c9d446ddf1cc3ece6e99c06c8cce9b78f
---

 libavcodec/riscv/vp7dsp_rvv.S |  2 +-
 libavcodec/riscv/vp8dsp_rvv.S | 14 +++++++-------
 2 files changed, 8 insertions(+), 8 deletions(-)

diff --git a/libavcodec/riscv/vp7dsp_rvv.S b/libavcodec/riscv/vp7dsp_rvv.S
index 39b23c2e79..09dcbf3857 100644
--- a/libavcodec/riscv/vp7dsp_rvv.S
+++ b/libavcodec/riscv/vp7dsp_rvv.S
@@ -134,7 +134,7 @@ func ff_vp7_idct_dc_add4y_rvv, zve32x
         li       t1, 23170
         vlse16.v v8, (a1), t0 # block[0..3][0]
         vwmul.vx v0, v8, t1
-        li       t2, 0x20000
+        li       t2, 0x20000 - (128 << 18)
         vsetvli  zero, zero, e32, m1, ta, ma
         vsra.vi  v0, v0, 14
         vmul.vx  v0, v0, t1
diff --git a/libavcodec/riscv/vp8dsp_rvv.S b/libavcodec/riscv/vp8dsp_rvv.S
index 8ea0a0c9bd..458eebb306 100644
--- a/libavcodec/riscv/vp8dsp_rvv.S
+++ b/libavcodec/riscv/vp8dsp_rvv.S
@@ -125,31 +125,31 @@ endfunc
 func ff_vp8_idct_dc_add4y_rvv, zve32x
         li       t0, 32
         vsetivli zero, 4, e16, mf2, ta, ma
+        li       t1, 4 - (128 << 3)
         vlse16.v v8, (a1), t0
-        vadd.vi  v8, v8, 4
+        vadd.vx  v8, v8, t1
         vsra.vi  v8, v8, 3
         # fall through
 endfunc
 
         .variant_cc ff_vp78_idct_dc_add4y_rvv
-# v8 = [dc0, dc1, dc2, dc3]
+# v8 = [dc0 - 128, dc1 - 128, dc2 - 128, dc3 - 128]
 func ff_vp78_idct_dc_add4y_rvv, zve32x
         vsetivli    zero, 16, e16, m2, ta, ma
         vid.v       v4
+        li          a4, 4
         vsrl.vi     v4, v4, 2
+        li          t1, 128
         vrgather.vv v0, v8, v4 # replicate each DC four times
         vsetvli     zero, zero, e8, m1, ta, ma
-        li          a4, 4
 1:
         vle8.v      v8, (a0)
         addi        a4, a4, -1
         vwaddu.wv   v16, v0, v8
         sh          zero, (a1)
-        vsetvli     zero, zero, e16, m2, ta, ma
-        vmax.vx     v16, v16, zero
+        vnclip.wi   v8, v16, 0
         addi        a1, a1, 32
-        vsetvli     zero, zero, e8, m1, ta, ma
-        vnclipu.wi  v8, v16, 0
+        vxor.vx     v8, v8, t1
         vse8.v      v8, (a0)
         add         a0, a0, a2
         bnez        a4, 1b