Merge pull request #440 from guzba/master
much faster toStraightAlpha + other stuff
This commit is contained in:
commit
0af140b047
2 changed files with 28 additions and 13 deletions
|
@ -171,7 +171,7 @@ proc decodeGif*(data: string): Gif {.raises: [PixieError].} =
|
|||
colorIndexes: seq[int]
|
||||
codeSize = minCodeSize + 1
|
||||
table = newSeq[(int, int)](endCode + 1)
|
||||
prev: (int, int)
|
||||
prev: tuple[offset, len: int]
|
||||
|
||||
while true:
|
||||
let code = b.readBits(codeSize).int
|
||||
|
@ -194,14 +194,14 @@ proc decodeGif*(data: string): Gif {.raises: [PixieError].} =
|
|||
if code < table.len: # If we have seen the code before
|
||||
if code < clearCode:
|
||||
colorIndexes.add(code)
|
||||
if prev[1] > 0:
|
||||
table.add((prev[0], prev[1] + 1))
|
||||
if prev.len > 0:
|
||||
table.add((prev.offset, prev.len + 1))
|
||||
prev = (start, 1)
|
||||
else:
|
||||
let (offset, len) = table[code]
|
||||
for i in 0 ..< len:
|
||||
colorIndexes.add(colorIndexes[offset + i])
|
||||
table.add((prev[0], prev[1] + 1))
|
||||
table.add((prev.offset, prev.len + 1))
|
||||
prev = (start, len)
|
||||
else:
|
||||
if prev[1] == 0:
|
||||
|
@ -209,8 +209,8 @@ proc decodeGif*(data: string): Gif {.raises: [PixieError].} =
|
|||
for i in 0 ..< prev[1]:
|
||||
colorIndexes.add(colorIndexes[prev[0] + i])
|
||||
colorIndexes.add(colorIndexes[prev[0]])
|
||||
table.add((start, prev[1] + 1))
|
||||
prev = (start, prev[1] + 1)
|
||||
table.add((start, prev.len + 1))
|
||||
prev = (start, prev.len + 1)
|
||||
|
||||
if colorIndexes.len != imageWidth * imageHeight:
|
||||
failInvalid()
|
||||
|
@ -243,6 +243,8 @@ proc decodeGif*(data: string): Gif {.raises: [PixieError].} =
|
|||
image.data[i] = globalColorTable[colorIndex]
|
||||
|
||||
if interlaced:
|
||||
# Just copyMem the rows into the right place. I've only ever seen
|
||||
# interlaced for the first frame so this is unlikely to be a hot path.
|
||||
let deinterlaced = newImage(image.width, image.height)
|
||||
var
|
||||
y: int
|
||||
|
|
|
@ -10,6 +10,9 @@ template currentExceptionAsPixieError*(): untyped =
|
|||
let e = getCurrentException()
|
||||
newException(PixieError, e.getStackTrace & e.msg, e)
|
||||
|
||||
when defined(release):
|
||||
{.push checks: off.}
|
||||
|
||||
proc gaussianKernel*(radius: int): seq[uint16] {.raises: [].} =
|
||||
## Compute lookup table for 1d Gaussian kernel.
|
||||
## Values are [0, 255] * 256.
|
||||
|
@ -81,16 +84,23 @@ proc fillUnsafe*(
|
|||
for j in i ..< start + len:
|
||||
data[j] = rgbx
|
||||
|
||||
const straightAlphaTable = block:
|
||||
var table: array[256, array[256, uint8]]
|
||||
for a in 0 ..< 256:
|
||||
let multiplier = if a > 0: (255 / a.float32) else: 0
|
||||
for c in 0 ..< 256:
|
||||
table[a][c] = min(round((c.float32 * multiplier)), 255).uint8
|
||||
table
|
||||
|
||||
proc toStraightAlpha*(data: var seq[ColorRGBA | ColorRGBX]) {.raises: [].} =
|
||||
## Converts an image from premultiplied alpha to straight alpha.
|
||||
## This is expensive for large images.
|
||||
for c in data.mitems:
|
||||
if c.a == 0 or c.a == 255:
|
||||
continue
|
||||
let multiplier = ((255 / c.a.float32) * 255).uint32
|
||||
c.r = ((c.r.uint32 * multiplier) div 255).uint8
|
||||
c.g = ((c.g.uint32 * multiplier) div 255).uint8
|
||||
c.b = ((c.b.uint32 * multiplier) div 255).uint8
|
||||
for i in 0 ..< data.len:
|
||||
var c = data[i]
|
||||
c.r = straightAlphaTable[c.a][c.r]
|
||||
c.g = straightAlphaTable[c.a][c.g]
|
||||
c.b = straightAlphaTable[c.a][c.b]
|
||||
data[i] = c
|
||||
|
||||
proc toPremultipliedAlpha*(data: var seq[ColorRGBA | ColorRGBX]) {.raises: [].} =
|
||||
## Converts an image to premultiplied alpha from straight alpha.
|
||||
|
@ -187,3 +197,6 @@ when defined(amd64) and allowSimd:
|
|||
a = mm_unpacklo_epi8(v, mm_setzero_si128())
|
||||
b = mm_unpacklo_epi8(a, mm_setzero_si128())
|
||||
result = mm_slli_epi32(b, 24) # Shift the values to uint32 `a`
|
||||
|
||||
when defined(release):
|
||||
{.pop.}
|
||||
|
|
Loading…
Reference in a new issue