Optimize png.decoder.filter_1_distance_4_sse128

wuffs_png_decode_filter_1_sub/clang9  1.26GB/s ± 0%  4.59GB/s ± 0%  +263.24%  (p=0.008 n=5+5)
wuffs_png_decode_filter_1_sub/gcc10   1.85GB/s ± 0%  1.85GB/s ± 0%      ~     (p=0.333 n=5+5)
This commit is contained in:
Nigel Tao committed 2021-01-15 23:28:58 +11:00
1 parent f58db5a4d8
commit 0b6865e92c
4 files changed
+39 -29

No files matched your search

+13 -14
View File
@@ -20,25 +20,24 @@ pri func decoder.filter_1_distance_4_sse128!(curr: slice base.u8),
choose cpu_arch >= sse128,
{
var c : slice base.u8
var fa0 : base.u8
var fa1 : base.u8
var fa2 : base.u8
var fa3 : base.u8
var x32 : base.u32
var x128 : base.sse128_i
var a128 : base.sse128_i
iterate (c = args.curr)(length: 4, advance: 4, unroll: 1) {
x32 = ((c[0] as base.u32) << 0x00) |
((c[1] as base.u32) << 0x08) |
((c[2] as base.u32) << 0x10) |
((c[3] as base.u32) << 0x18)
x128.load_u32!(a: x32)
x128 = x128._mm_add_epi8!(b: a128)
a128 = x128
x32 = x128.truncate_u32()
if x32 == 0 {
}
fa0 ~mod+= c[0]
c[0] = fa0
fa1 ~mod+= c[1]
c[1] = fa1
fa2 ~mod+= c[2]
c[2] = fa2
fa3 ~mod+= c[3]
c[3] = fa3
c[0] = (0xFF & (x32 >> 0x00)) as base.u8
c[1] = (0xFF & (x32 >> 0x08)) as base.u8
c[2] = (0xFF & (x32 >> 0x10)) as base.u8
c[3] = (0xFF & (x32 >> 0x18)) as base.u8
}
}