MyouProject/pixie
2
0
Fork 0
Code Issues Pull requests Actions Packages Projects Releases Wiki Activity

isOpaqueSimd

Browse source
This commit is contained in:
Ryan Oldenburg 2022-06-30 10:21:53 -05:00
parent d76550052e
commit 1c6fa86ac0
3 changed files with 14 additions and 15 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

5
src/pixie/internal.nim
View file

@ -126,10 +126,7 @@ proc toPremultipliedAlpha*(data: var seq[ColorRGBA | ColorRGBX]) {.raises: [].}
proc isOpaque*(data: var seq[ColorRGBX], start, len: int): bool =
when allowSimd and compiles(isOpaqueSimd):
return isOpaqueSimd(
cast[ptr UncheckedArray[ColorRGBX]](data[start].addr),
len
)
return isOpaqueSimd(data, start, len)
result = true

11
src/pixie/runtimechecked/avx2.nim
View file

@ -61,18 +61,19 @@ proc isTransparentAvx2*(data: ptr UncheckedArray[ColorRGBX], len: int): bool =
if data[i].a != 0:
return false
proc isOpaqueAvx2*(data: ptr UncheckedArray[ColorRGBX], len: int): bool =
proc isOpaqueAvx2*(data: var seq[ColorRGBX], start, len: int): bool =
result = true
var i: int
while i < len and (cast[uint](data[i].addr) and 31) != 0: # Align to 32 bytes
var i = start
# Align to 32 bytes
while i < (start + len) and (cast[uint](data[i].addr) and 31) != 0:
if data[i].a != 255:
return false
inc i
let
vec255 = mm256_set1_epi8(255)
iterations = (len - i) div 16
iterations = (start + len - i) div 16
for _ in 0 ..< iterations:
let
values0 = mm256_load_si256(data[i].addr)
@ -83,7 +84,7 @@ proc isOpaqueAvx2*(data: ptr UncheckedArray[ColorRGBX], len: int): bool =
return false
i += 16
for i in i ..< len:
for i in i ..< start + len:
if data[i].a != 255:
return false

13
src/pixie/simd.nim
View file

@ -130,21 +130,22 @@ when defined(amd64):
if data[i].a != 0:
return false
proc isOpaqueSimd*(data: ptr UncheckedArray[ColorRGBX], len: int): bool =
proc isOpaqueSimd*(data: var seq[ColorRGBX], start, len: int): bool =
if cpuHasAvx2:
return isOpaqueAvx2(data, len)
return isOpaqueAvx2(data, start, len)
result = true
var i: int
while i < len and (cast[uint](data[i].addr) and 15) != 0: # Align to 16 bytes
var i = start
# Align to 16 bytes
while i < (start + len) and (cast[uint](data[i].addr) and 15) != 0:
if data[i].a != 255:
return false
inc i
let
vec255 = mm_set1_epi8(255)
iterations = (len - i) div 16
iterations = (start + len - i) div 16
for _ in 0 ..< iterations:
let
values0 = mm_load_si128(data[i].addr)
@ -159,7 +160,7 @@ when defined(amd64):
return false
i += 16
for i in i ..< len:
for i in i ..< start + len:
if data[i].a != 255:
return false