wrapper/FFmpeg
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

arm/aarch64: vp9lpf: Keep the comparison to E within 8 bit

Browse source 
The theoretical maximum value of E is 193, so we can just
saturate the addition to 255.

Before:                     Cortex A7      A8      A9     A53  A53/AArch64
vp9_loop_filter_v_4_8_neon:     143.0   127.7   114.8    88.0         87.7
vp9_loop_filter_v_8_8_neon:     241.0   197.2   173.7   140.0        136.7
vp9_loop_filter_v_16_8_neon:    497.0   419.5   379.7   293.0        275.7
vp9_loop_filter_v_16_16_neon:   965.2   818.7   731.4   579.0        452.0
After:
vp9_loop_filter_v_4_8_neon:     136.0   125.7   112.6    84.0         83.0
vp9_loop_filter_v_8_8_neon:     234.0   195.5   171.5   136.0        133.7
vp9_loop_filter_v_16_8_neon:    490.0   417.5   377.7   289.0        271.0
vp9_loop_filter_v_16_16_neon:   951.2   814.7   732.3   571.0        446.7

This is cherrypicked from libav commit
c582cb8537.

Signed-off-by: Martin Storsjö <martin@martin.st>
This commit is contained in:
Martin Storsjö 2017-01-14 20:49:19 +02:00
parent dda45c087b
commit 3fbbad2984
2 changed files with 14 additions and 37 deletions
Download patch file Download diff file Expand all files Collapse all files

11
libavcodec/arm/vp9lpf_neon.S
View file

@ -51,7 +51,7 @@
@ and d28-d31 as temp registers, or d8-d15.
@ tmp1,tmp2 = tmpq1, tmp3,tmp4 = tmpq2, tmp5,tmp6 = tmpq3, tmp7,tmp8 = tmpq4
.macro loop_filter wd, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8, tmpq1, tmpq2, tmpq3, tmpq4
vdup.u16 q0, r2 @ E
vdup.u8 d0, r2 @ E
vdup.u8 d2, r3 @ I
ldr r3, [sp]
@ -64,16 +64,15 @@
vmax.u8 d4, d4, d5
vmax.u8 d5, d6, d7
vmax.u8 \tmp1, \tmp1, \tmp2
vabdl.u8 q3, d23, d24 @ abs(p0 - q0)
vabd.u8 d6, d23, d24 @ abs(p0 - q0)
vmax.u8 d4, d4, d5
vadd.u16 q3, q3, q3 @ abs(p0 - q0) * 2
vqadd.u8 d6, d6, d6 @ abs(p0 - q0) * 2
vabd.u8 d5, d22, d25 @ abs(p1 - q1)
vmax.u8 d4, d4, \tmp1 @ max(abs(p3 - p2), ..., abs(q2 - q3))
vshr.u8 d5, d5, #1
vcle.u8 d4, d4, d2 @ max(abs()) <= I
vaddw.u8 q3, q3, d5 @ abs(p0 - q0) * 2 + abs(p1 - q1) >> 1
vcle.u16 q3, q3, q0
vmovn.u16 d5, q3
vqadd.u8 d6, d6, d5 @ abs(p0 - q0) * 2 + abs(p1 - q1) >> 1
vcle.u8 d5, d6, d0
vand d4, d4, d5 @ fm
vdup.u8 d3, r3 @ H