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

move invert simd out

Browse source
This commit is contained in:
Ryan Oldenburg 2022-06-29 00:13:38 -05:00
parent 9f2c018c9e
commit e2a966288d
2 changed files with 34 additions and 17 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

26
src/pixie/images.nim
View file

@ -421,28 +421,20 @@ proc applyOpacity*(target: Image | Mask, opacity: float32) {.raises: [].} =
proc invert*(image: Image) {.raises: [].} = proc invert*(image: Image) {.raises: [].} =
## Inverts all of the colors and alpha. ## Inverts all of the colors and alpha.
var i: int if allowSimd and compiles(invertSimd):
when defined(amd64) and allowSimd: invertSimd(
let vec255 = mm_set1_epi8(cast[int8](255)) cast[ptr UncheckedArray[ColorRGBX]](image.data[0].addr),
for _ in 0 ..< image.data.len div 16: image.data.len
let )
a = mm_loadu_si128(image.data[i + 0].addr) return
b = mm_loadu_si128(image.data[i + 4].addr)
c = mm_loadu_si128(image.data[i + 8].addr)
d = mm_loadu_si128(image.data[i + 12].addr)
mm_storeu_si128(image.data[i + 0].addr, mm_sub_epi8(vec255, a))
mm_storeu_si128(image.data[i + 4].addr, mm_sub_epi8(vec255, b))
mm_storeu_si128(image.data[i + 8].addr, mm_sub_epi8(vec255, c))
mm_storeu_si128(image.data[i + 12].addr, mm_sub_epi8(vec255, d))
i += 16
for j in i ..< image.data.len: for i in 0 ..< image.data.len:
var rgbx = image.data[j] var rgbx = image.data[i]
rgbx.r = 255 - rgbx.r rgbx.r = 255 - rgbx.r
rgbx.g = 255 - rgbx.g rgbx.g = 255 - rgbx.g
rgbx.b = 255 - rgbx.b rgbx.b = 255 - rgbx.b
rgbx.a = 255 - rgbx.a rgbx.a = 255 - rgbx.a
image.data[j] = rgbx image.data[i] = rgbx
# Inverting rgbx(50, 100, 150, 200) becomes rgbx(205, 155, 105, 55). This # Inverting rgbx(50, 100, 150, 200) becomes rgbx(205, 155, 105, 55). This
# is not a valid premultiplied alpha color. # is not a valid premultiplied alpha color.

25
src/pixie/simd.nim
View file

@ -170,3 +170,28 @@ when defined(amd64):
c.g = ((c.g.uint32 * c.a) div 255).uint8 c.g = ((c.g.uint32 * c.a) div 255).uint8
c.b = ((c.b.uint32 * c.a) div 255).uint8 c.b = ((c.b.uint32 * c.a) div 255).uint8
copyMem(data[i].addr, c.addr, 4) copyMem(data[i].addr, c.addr, 4)
proc invertSimd*(data: ptr UncheckedArray[ColorRGBX], len: int) =
var i: int
let vec255 = mm_set1_epi8(cast[int8](255))
for _ in 0 ..< len div 16:
let
a = mm_loadu_si128(data[i + 0].addr)
b = mm_loadu_si128(data[i + 4].addr)
c = mm_loadu_si128(data[i + 8].addr)
d = mm_loadu_si128(data[i + 12].addr)
mm_storeu_si128(data[i + 0].addr, mm_sub_epi8(vec255, a))
mm_storeu_si128(data[i + 4].addr, mm_sub_epi8(vec255, b))
mm_storeu_si128(data[i + 8].addr, mm_sub_epi8(vec255, c))
mm_storeu_si128(data[i + 12].addr, mm_sub_epi8(vec255, d))
i += 16
for i in i ..< len:
var rgbx = data[i]
rgbx.r = 255 - rgbx.r
rgbx.g = 255 - rgbx.g
rgbx.b = 255 - rgbx.b
rgbx.a = 255 - rgbx.a
data[i] = rgbx
toPremultipliedAlphaSimd(cast[ptr UncheckedArray[uint32]](data), len)