MyouProject/pixie
3
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Merge pull request #444 from guzba/pngupsimd

Browse source 
encode filter 2 pngs, simd unfilter filter 2 pngs
This commit is contained in:
Andre von Houck 2022-06-16 18:07:54 -07:00 committed by GitHub
commit 7718063d00
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 20 additions and 8 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

28
src/pixie/fileformats/png.nim
View file

@ -1,6 +1,9 @@
import chroma, flatty/binny, math, pixie/common, pixie/images, pixie/internal, import chroma, flatty/binny, math, pixie/common, pixie/images, pixie/internal,
pixie/masks, zippy, zippy/crc pixie/masks, zippy, zippy/crc
when defined(amd64) and allowSimd:
import nimsimd/sse2
# See http://www.libpng.org/pub/png/spec/1.2/PNG-Contents.html # See http://www.libpng.org/pub/png/spec/1.2/PNG-Contents.html
const const
@ -128,7 +131,19 @@ proc unfilter(
let let
uncompressedStartIdx = uncompressedIdx(1, y) uncompressedStartIdx = uncompressedIdx(1, y)
unfilteredStartIx = unfiteredIdx(0, y) unfilteredStartIx = unfiteredIdx(0, y)
for x in 0 ..< rowBytes: var x: int
when allowSimd and defined(amd64):
if y - 1 >= 0:
for _ in 0 ..< rowBytes div 16:
let
bytes = mm_loadu_si128(uncompressed[uncompressedStartIdx + x].addr)
up = mm_loadu_si128(result[unfilteredStartIx + x - rowBytes].addr)
mm_storeu_si128(
result[unfilteredStartIx + x].addr,
mm_add_epi8(bytes, up)
)
x += 16
for x in x ..< rowBytes:
var value = uncompressed[uncompressedStartIdx + x] var value = uncompressed[uncompressedStartIdx + x]
if y - 1 >= 0: if y - 1 >= 0:
value += result[unfilteredStartIx + x - rowBytes] value += result[unfilteredStartIx + x - rowBytes]
@ -556,19 +571,16 @@ proc encodePng*(
# Add room for 1 byte before each row for the filter type. # Add room for 1 byte before each row for the filter type.
var filtered = newString(width * height * channels + height) var filtered = newString(width * height * channels + height)
for y in 0 ..< height: for y in 0 ..< height:
filtered[y * width * channels + y] = 3.char # Average filtered[y * width * channels + y] = 2.char # Up filter type
for x in 0 ..< width * channels: for x in 0 ..< width * channels:
# Move through the image data byte-by-byte # Move through the image data byte-by-byte
let let
dataPos = y * width * channels + x dataPos = y * width * channels + x
filteredPos = y * width * channels + y + 1 + x filteredPos = y * width * channels + y + 1 + x
var left, up: int var up: uint8
if x - channels >= 0:
left = data[dataPos - channels].int
if y - 1 >= 0: if y - 1 >= 0:
up = data[(y - 1) * width * channels + x].int up = data[(y - 1) * width * channels + x]
let avg = ((left + up) div 2).uint8 filtered[filteredPos] = (data[dataPos] - up).char
filtered[filteredPos] = (data[dataPos] - avg).char
let compressed = let compressed =
try: try: