2x faster mask minifyBy2

This commit is contained in:
Ryan Oldenburg 2021-06-18 16:03:25 -05:00
parent 280089b1b3
commit 51d38c7e37
3 changed files with 76 additions and 5 deletions

View file

@ -73,17 +73,80 @@ proc minifyBy2*(mask: Mask, power = 1): Mask =
if power == 0:
return mask.copy()
var src = mask
for i in 1 .. power:
result = newMask(mask.width div 2, mask.height div 2)
for y in 0 ..< result.height:
for x in 0 ..< result.width:
var x: int
when defined(amd64) and not defined(pixieNoSimd):
let
oddMask = mm_set1_epi16(cast[int16](0xff00))
first8 = cast[M128i]([uint8.high, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0])
for _ in countup(0, result.width - 16, 8):
let
top = mm_loadu_si128(src.data[src.dataIndex(x * 2, y * 2 + 0)].addr)
btm = mm_loadu_si128(src.data[src.dataIndex(x * 2, y * 2 + 1)].addr)
topShifted = mm_srli_si128(top, 1)
btmShifted = mm_srli_si128(btm, 1)
topEven = mm_andnot_si128(oddMask, top)
topOdd = mm_srli_epi16(mm_and_si128(top, oddMask), 8)
btmEven = mm_andnot_si128(oddMask, btm)
btmOdd = mm_srli_epi16(mm_and_si128(btm, oddMask), 8)
topShiftedEven = mm_andnot_si128(oddMask, topShifted)
topShiftedOdd = mm_srli_epi16(mm_and_si128(topShifted, oddMask), 8)
btmShiftedEven = mm_andnot_si128(oddMask, btmShifted)
btmShiftedOdd = mm_srli_epi16(mm_and_si128(btmShifted, oddMask), 8)
topAddedEven = mm_add_epi16(topEven, topShiftedEven)
btmAddedEven = mm_add_epi16(btmEven, btmShiftedEven)
topAddedOdd = mm_add_epi16(topOdd, topShiftedOdd)
bottomAddedOdd = mm_add_epi16(btmOdd, btmShiftedOdd)
addedEven = mm_add_epi16(topAddedEven, btmAddedEven)
addedOdd = mm_add_epi16(topAddedOdd, bottomAddedOdd)
addedEvenDiv4 = mm_srli_epi16(addedEven, 2)
addedOddDiv4 = mm_srli_epi16(addedOdd, 2)
merged = mm_or_si128(addedEvenDiv4, mm_slli_epi16(addedOddDiv4, 8))
# merged has the correct values in the even indices
a = mm_and_si128(merged, first8)
b = mm_and_si128(mm_srli_si128(merged, 2), first8)
c = mm_and_si128(mm_srli_si128(merged, 4), first8)
d = mm_and_si128(mm_srli_si128(merged, 6), first8)
e = mm_and_si128(mm_srli_si128(merged, 8), first8)
f = mm_and_si128(mm_srli_si128(merged, 10), first8)
g = mm_and_si128(mm_srli_si128(merged, 12), first8)
h = mm_and_si128(mm_srli_si128(merged, 14), first8)
ab = mm_or_si128(a, mm_slli_si128(b, 1))
cd = mm_or_si128(c, mm_slli_si128(d, 1))
ef = mm_or_si128(e, mm_slli_si128(f, 1))
gh = mm_or_si128(g, mm_slli_si128(h, 1))
abcd = mm_or_si128(ab, mm_slli_si128(cd, 2))
efgh = mm_or_si128(ef, mm_slli_si128(gh, 2))
abcdefgh = mm_or_si128(abcd, mm_slli_si128(efgh, 4))
mm_storeu_si128(result.data[result.dataIndex(x, y)].addr, abcdefgh)
x += 8
for x in x ..< result.width:
let value =
mask.getValueUnsafe(x * 2 + 0, y * 2 + 0).uint32 +
mask.getValueUnsafe(x * 2 + 1, y * 2 + 0) +
mask.getValueUnsafe(x * 2 + 1, y * 2 + 1) +
mask.getValueUnsafe(x * 2 + 0, y * 2 + 1)
src.getValueUnsafe(x * 2 + 0, y * 2 + 0).uint32 +
src.getValueUnsafe(x * 2 + 1, y * 2 + 0) +
src.getValueUnsafe(x * 2 + 1, y * 2 + 1) +
src.getValueUnsafe(x * 2 + 0, y * 2 + 1)
result.setValueUnsafe(x, y, (value div 4).uint8)
# Set src as this result for if we do another power
src = result
proc fillUnsafe*(data: var seq[uint8], value: uint8, start, len: int) =
## Fills the mask data with the parameter value starting at index start and
## continuing for len indices.

Binary file not shown.

After

Width:  |  Height:  |  Size: 1.3 KiB

View file

@ -156,3 +156,11 @@ block:
mask.fillRect(rect(25, 25, 50, 50))
mask.blur(20)
writeFile("tests/images/maskblur20.png", mask.encodePng())
block:
let mask = newMask(200, 200)
mask.fillRect(rect(25, 25, 150, 150))
mask.blur(25)
let minified = mask.minifyBy2()
writeFile("tests/images/masks/minifiedBlur.png", minified.encodePng())