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

faster

Browse source
This commit is contained in:
Ryan Oldenburg 2022-07-30 21:31:52 -05:00
parent 0df68b6b65
commit 0f0a60f5cc
1 changed files with 15 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

18
src/pixie/simd/avx2.nim
View file

@ -107,6 +107,10 @@ proc toPremultipliedAlphaAvx2*(data: var seq[ColorRGBA | ColorRGBX]) {.simd.} =
let let
alphaMask = mm256_set1_epi32(cast[int32](0xff000000)) alphaMask = mm256_set1_epi32(cast[int32](0xff000000))
shuffleControl = mm256_set_epi8(
15, 0, 15, 0, 11, 0, 11, 0, 7, 0, 7, 0, 3, 0, 3, 0,
15, 0, 15, 0, 11, 0, 11, 0, 7, 0, 7, 0, 3, 0, 3, 0
)
oddMask = mm256_set1_epi16(0xff00) oddMask = mm256_set1_epi16(0xff00)
vec128 = mm256_set1_epi16(128) vec128 = mm256_set1_epi16(128)
hiMask = mm256_set1_epi16(255 shl 8) hiMask = mm256_set1_epi16(255 shl 8)
@ -118,7 +122,7 @@ proc toPremultipliedAlphaAvx2*(data: var seq[ColorRGBA | ColorRGBX]) {.simd.} =
eq = mm256_cmpeq_epi8(values, alphaMask) eq = mm256_cmpeq_epi8(values, alphaMask)
if (mm256_movemask_epi8(eq) and 0x88888888) != 0x88888888: if (mm256_movemask_epi8(eq) and 0x88888888) != 0x88888888:
let let
evenMultiplier = mm256_or_si256(alpha, mm256_srli_epi32(alpha, 16)) evenMultiplier = mm256_shuffle_epi8(alpha, shuffleControl)
oddMultiplier = mm256_or_si256(evenMultiplier, alphaMask) oddMultiplier = mm256_or_si256(evenMultiplier, alphaMask)
var var
colorsEven = mm256_slli_epi16(values, 8) colorsEven = mm256_slli_epi16(values, 8)
@ -385,6 +389,10 @@ proc blitLineNormalAvx2*(
div255 = mm256_set1_epi16(cast[int16](0x8081)) div255 = mm256_set1_epi16(cast[int16](0x8081))
vec255 = mm256_set1_epi8(255) vec255 = mm256_set1_epi8(255)
vecAlpha255 = mm256_set1_epi32(cast[int32]([0.uint8, 255, 0, 255])) vecAlpha255 = mm256_set1_epi32(cast[int32]([0.uint8, 255, 0, 255]))
shuffleControl = mm256_set_epi8(
15, 0, 15, 0, 11, 0, 11, 0, 7, 0, 7, 0, 3, 0, 3, 0,
15, 0, 15, 0, 11, 0, 11, 0, 7, 0, 7, 0, 3, 0, 3, 0
)
var i: int var i: int
while i < len - 8: while i < len - 8:
@ -401,7 +409,7 @@ proc blitLineNormalAvx2*(
backdropEven = mm256_slli_epi16(backdrop, 8) backdropEven = mm256_slli_epi16(backdrop, 8)
backdropOdd = mm256_and_si256(backdrop, oddMask) backdropOdd = mm256_and_si256(backdrop, oddMask)
sourceAlpha = mm256_or_si256(sourceAlpha, mm256_srli_epi32(sourceAlpha, 16)) sourceAlpha = mm256_shuffle_epi8(sourceAlpha, shuffleControl)
let multiplier = mm256_sub_epi32(vecAlpha255, sourceAlpha) let multiplier = mm256_sub_epi32(vecAlpha255, sourceAlpha)
@ -430,6 +438,10 @@ proc blitLineMaskAvx2*(
oddMask = mm256_set1_epi16(cast[int16](0xff00)) oddMask = mm256_set1_epi16(cast[int16](0xff00))
div255 = mm256_set1_epi16(cast[int16](0x8081)) div255 = mm256_set1_epi16(cast[int16](0x8081))
vec255 = mm256_set1_epi8(255) vec255 = mm256_set1_epi8(255)
shuffleControl = mm256_set_epi8(
15, 0, 15, 0, 11, 0, 11, 0, 7, 0, 7, 0, 3, 0, 3, 0,
15, 0, 15, 0, 11, 0, 11, 0, 7, 0, 7, 0, 3, 0, 3, 0
)
var i: int var i: int
while i < len - 8: while i < len - 8:
@ -446,7 +458,7 @@ proc blitLineMaskAvx2*(
backdropEven = mm256_slli_epi16(backdrop, 8) backdropEven = mm256_slli_epi16(backdrop, 8)
backdropOdd = mm256_and_si256(backdrop, oddMask) backdropOdd = mm256_and_si256(backdrop, oddMask)
sourceAlpha = mm256_or_si256(sourceAlpha, mm256_srli_epi32(sourceAlpha, 16)) sourceAlpha = mm256_shuffle_epi8(sourceAlpha, shuffleControl)
backdropEven = mm256_mulhi_epu16(backdropEven, sourceAlpha) backdropEven = mm256_mulhi_epu16(backdropEven, sourceAlpha)
backdropOdd = mm256_mulhi_epu16(backdropOdd, sourceAlpha) backdropOdd = mm256_mulhi_epu16(backdropOdd, sourceAlpha)