Skip to content

Optimize pow5_12(::Float32) #493

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
kimikage merged 1 commit into from
Jun 23, 2021
Merged

Optimize pow5_12(::Float32) #493

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 2 additions & 1 deletion src/conversions.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -81,8 +81,9 @@ correct_gamut(c::CV) where {CV<:TransparentRGB} =
@inline function srgb_compand(v)
F = typeof(0.5f0v) === Float32 ? Float32 : promote_type(Float64, typeof(v))
vf = F(v)
vc = @fastmath max(vf, F(0.0031308))
# `pow5_12` is an optimized function to get `v^(1/2.4)`
vf > F(0.0031308) ? muladd(F(1.055), F(pow5_12(vf)), F(-0.055)) : F(12.92) * vf
vf > F(0.0031308) ? muladd(F(1.055), F(pow5_12(vc)), F(-0.055)) : F(12.92) * vf
end

function _hsx_to_rgb(im::UInt8, v, n, m)
Expand Down
18 changes: 16 additions & 2 deletions src/utilities.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -92,8 +92,8 @@ pow3_4(x) = (y = @fastmath(sqrt(x)); y*@fastmath(sqrt(y))) # x^(3/4)

# `pow5_12` is called from `srgb_compand`.
pow5_12(x) = pow3_4(x) / cbrt(x) # 5/12 == 1/2 + 1/4 - 1/3 == 3/4 - 1/3
pow5_12(x::Float32) = Float32(pow5_12(Float64(x)))
@inline function pow5_12(x::Float64)
@noinline _cbrt(x) = cbrt01(x)
p3_4 = pow3_4(x)
# x^(-1/6)
if x < 0.02
Expand All @@ -106,7 +106,7 @@ pow5_12(x::Float32) = Float32(pow5_12(Float64(x)))
t0 = @evalpoly(x, 1.7047813285940905, -3.1261253501167308,
7.498744828350077, -10.100319516746419, 6.820601476522508, -1.7978894213531524)
else
return p3_4 / cbrt(x)
return p3_4 / _cbrt(x)
end
# x^(-1/3)
t1 = t0 * t0
Expand All @@ -117,6 +117,20 @@ pow5_12(x::Float32) = Float32(pow5_12(Float64(x)))
# x^(3/4) * x^(-1/3)
muladd(p3_4, t2, p3_4 * t2h)
end
@inline function pow5_12(x::Float32)
# x^(-1/3)
rc = rcbrt(x)
rcx = -rc * x
rch = muladd(muladd(rc, x, rcx), -rc^2, muladd(rc^2, rcx, 1.0f0)) # 1 - x * rc^3
rce = muladd(2/9f0, rch, 1/3f0) * rch * rc
# x^(3/4)
p3_4_f64 = pow3_4(Float64(x))
p3_4r = reinterpret(Float64, reinterpret(UInt64, p3_4_f64) & 0xffffffff_e0000000)
p3_4 = Float32(p3_4r)
p3_4e = Float32(p3_4_f64 - p3_4r)
# x^(3/4) * x^(-1/3)
muladd(p3_4, rc, muladd(p3_4, rce, p3_4e * rc))
end

# `pow12_5` is called from `invert_srgb_compand`.
pow12_5(x) = pow12_5(Float64(x))
Expand Down