Skip to content

Commit

Permalink
Merge pull request #61 from milankl/conversion
Browse files Browse the repository at this point in the history 
make new conversion default
  • Loading branch information
@milankl
milankl committed Oct 20, 2021
2 parents 7c3af06 + 1839850 commit 817f977
Show file tree
Hide file tree
Showing 4 changed files with 73 additions and 63 deletions.
2 changes: 1 addition & 1 deletion src/SoftPosit.jl
View file
Original file line number Diff line number Diff line change
Expand Up @@ -5,7 +5,7 @@ module SoftPosit
Posit8_2, Posit16_2, Posit24_2,
notareal, minusone,
AbstractQuire, Quire8, Quire16, Quire32, fms,
Posit16_new, Float32_new
Posit16_old, Float32_old

import Base: Float64, Float32, Float16, Int32, Int64,
UInt8, UInt16, UInt32,
Expand Down
60 changes: 33 additions & 27 deletions src/conversionFloatToPosit.jl
View file
Original file line number Diff line number Diff line change
@@ -1,39 +1,16 @@
# from Float64 to Posit8,16,32
Posit8(x::Float64) = ccall((:convertDoubleToP8, SoftPositPath), Posit8, (Float64,),x)
Posit16(x::Float64) = ccall((:convertDoubleToP16, SoftPositPath), Posit16, (Float64,),x)
Posit16_old(x::Float64) = ccall((:convertDoubleToP16, SoftPositPath), Posit16, (Float64,),x)
Posit32(x::Float64) = ccall((:convertDoubleToP32, SoftPositPath), Posit32, (Float64,),x)

# from Float32/16 to Posit8,16,32
Posit8(x::T where {T <: Float16or32}) = Posit8(Float64(x))
Posit16(x::T where {T <: Float16or32}) = Posit16(Float64(x))
Posit32(x::T where {T <: Float16or32}) = Posit32(Float64(x))

# from Float64 to Posit_1 (convertDoubleToPX1 currently not available)
# Posit8_1(x::Float64) = ccall((:convertDoubleToPX1, SoftPositPath), Posit8_1, (Float64,Int64),x,8)
# Posit16_1(x::Float64) = ccall((:convertDoubleToPX1, SoftPositPath), Posit16_1, (Float64,Int64),x,16)
# Posit24_1(x::Float64) = ccall((:convertDoubleToPX1, SoftPositPath), Posit24_1, (Float64,Int64),x,24)
Posit16(x::Float16) = Posit16(Float32(x))
Posit16(x::Float64) = Posit16(Float32(x))

# use detour
Posit8_1(x::Float64) = Posit8_1(Posit32(x))
Posit16_1(x::Float64) = Posit16_1(Posit32(x))
Posit24_1(x::Float64) = Posit24_1(Posit32(x))

# from Float32/16 to Posit_1
Posit8_1(x::T where {T <: Float16or32}) = Posit8_1(Float64(x))
Posit16_1(x::T where {T <: Float16or32}) = Posit16_1(Float64(x))
Posit24_1(x::T where {T <: Float16or32}) = Posit24_1(Float64(x))

# from Float64 to Posit_2
Posit8_2(x::Float64) = ccall((:convertDoubleToPX2, SoftPositPath), Posit8_2, (Float64,Int64),x,8)
Posit16_2(x::Float64) = ccall((:convertDoubleToPX2, SoftPositPath), Posit16_2, (Float64,Int64),x,16)
Posit24_2(x::Float64) = ccall((:convertDoubleToPX2, SoftPositPath), Posit24_2, (Float64,Int64),x,24)

# from Float32/16 to Posit_2
Posit8_2(x::T where {T <: Float16or32}) = Posit8_2(Float64(x))
Posit16_2(x::T where {T <: Float16or32}) = Posit16_2(Float64(x))
Posit24_2(x::T where {T <: Float16or32}) = Posit24_2(Float64(x))

function Posit16_new(x::Float32)
function Posit16(x::Float32)
ui = reinterpret(UInt32,x)

# REGIME AND EXPONENT BITS
Expand Down Expand Up @@ -64,3 +41,32 @@ function Posit16_new(x::Float32)
p16 = signbit(x) ? -p16 : p16
return reinterpret(Posit16,p16)
end

# legacy
Posit16_old(x::Float32) = Posit16_old(Float64(x))
Posit16_old(x::Float16) = Posit16_old(Float64(x))

# from Float64 to Posit_1 (convertDoubleToPX1 currently not available)
# Posit8_1(x::Float64) = ccall((:convertDoubleToPX1, SoftPositPath), Posit8_1, (Float64,Int64),x,8)
# Posit16_1(x::Float64) = ccall((:convertDoubleToPX1, SoftPositPath), Posit16_1, (Float64,Int64),x,16)
# Posit24_1(x::Float64) = ccall((:convertDoubleToPX1, SoftPositPath), Posit24_1, (Float64,Int64),x,24)

# use detour
Posit8_1(x::Float64) = Posit8_1(Posit32(x))
Posit16_1(x::Float64) = Posit16_1(Posit32(x))
Posit24_1(x::Float64) = Posit24_1(Posit32(x))

# from Float32/16 to Posit_1
Posit8_1(x::T where {T <: Float16or32}) = Posit8_1(Float64(x))
Posit16_1(x::T where {T <: Float16or32}) = Posit16_1(Float64(x))
Posit24_1(x::T where {T <: Float16or32}) = Posit24_1(Float64(x))

# from Float64 to Posit_2
Posit8_2(x::Float64) = ccall((:convertDoubleToPX2, SoftPositPath), Posit8_2, (Float64,Int64),x,8)
Posit16_2(x::Float64) = ccall((:convertDoubleToPX2, SoftPositPath), Posit16_2, (Float64,Int64),x,16)
Posit24_2(x::Float64) = ccall((:convertDoubleToPX2, SoftPositPath), Posit24_2, (Float64,Int64),x,24)

# from Float32/16 to Posit_2
Posit8_2(x::T where {T <: Float16or32}) = Posit8_2(Float64(x))
Posit16_2(x::T where {T <: Float16or32}) = Posit16_2(Float64(x))
Posit24_2(x::T where {T <: Float16or32}) = Posit24_2(Float64(x))
10 changes: 7 additions & 3 deletions src/conversionPositToFloat.jl
View file
Original file line number Diff line number Diff line change
@@ -1,6 +1,5 @@
# from Posit to Float64,32,16
Float64(x::Posit8) = ccall((:convertP8ToDouble, SoftPositPath), Float64, (Posit8,),x)
Float64(x::Posit16) = ccall((:convertP16ToDouble, SoftPositPath), Float64, (Posit16,),x)
Float64(x::Posit32) = ccall((:convertP32ToDouble, SoftPositPath), Float64, (Posit32,),x)

# conversion from PX2 to Float64,32,16
Expand All @@ -14,10 +13,12 @@ Float64(x::Posit16_1) = ccall((:convertPX1ToDouble, SoftPositPath), Float64, (Po
Float64(x::Posit24_1) = ccall((:convertPX1ToDouble, SoftPositPath), Float64, (Posit24_1,),x)

# conversion to Float32,16
Float32(x::AbstractPosit) = Float32(Float64(x))
Float16(x::AbstractPosit) = Float16(Float64(x))
Float32(x::AbstractPosit) = Float32(Float64(x))
Float16(x::Posit16) = Float16(Float32(x))
Float64(x::Posit16) = Float64(Float32(x))

function Float32_new(x::Posit16)
function Float32(x::Posit16)
ui = reinterpret(UInt16,x)

signbitx = signbit(x) # sign of number
Expand All @@ -44,3 +45,6 @@ function Float32_new(x::Posit16)
return reinterpret(Float32,f32)
end

# legacy
Float32_old(x::AbstractPosit) = Float32(Float64_old(x))
Float64_old(x::Posit16) = ccall((:convertP16ToDouble, SoftPositPath), Float64, (Posit16,),x)
64 changes: 32 additions & 32 deletions test/test_conversions.jl
View file
Original file line number Diff line number Diff line change
Expand Up @@ -38,16 +38,16 @@ end

@testset "0,±1,±2,±4" begin
for f in Float32[-4,-2,-1,0,1,2,4]
@test Posit16(f) == Posit16_new(f)
@test f == Float32_new(Posit16_new(f))
@test Posit16(f) == Posit16_old(f)
@test f == Float32(Posit16(f))
end
end


@testset "0,±1/2,±1/4,±1/8" begin
for f in Float32[-1/8,-1/4,-1/2,0,1/2,1/4,1/8]
@test Posit16(f) == Posit16_new(f)
@test f == Float32_new(Posit16_new(f))
@test Posit16(f) == Posit16_old(f)
@test f == Float32(Posit16(f))
end
end

Expand All @@ -63,73 +63,73 @@ end
0.125,
0.25,
0.5]
@test Posit16(f) == Posit16_new(f)
@test Posit16(-f) == Posit16_new(-f)
@test Posit16(1/f) == Posit16_new(1/f)
@test Posit16(-1/f) == Posit16_new(-1/f)

@test f == Float32_new(Posit16_new(f))
@test -f == Float32_new(Posit16_new(-f))
@test 1/f == Float32_new(Posit16_new(1/f))
@test -1/f == Float32_new(Posit16_new(-1/f))
@test Posit16(f) == Posit16_old(f)
@test Posit16(-f) == Posit16_old(-f)
@test Posit16(1/f) == Posit16_old(1/f)
@test Posit16(-1/f) == Posit16_old(-1/f)

@test f == Float32(Posit16(f))
@test -f == Float32(Posit16(-f))
@test 1/f == Float32(Posit16(1/f))
@test -1/f == Float32(Posit16(-1/f))
end
end

@testset "NaN, Inf" begin
for f in [NaN32,Inf32,-Inf32]
@test Posit16(f) == Posit16_new(f)
@test isnan(Float32_new(Posit16_new(f)))
@test Posit16(f) == Posit16_old(f)
@test isnan(Float32(Posit16(f)))
end

for _ in 1:10
# create various NaNs
f = reinterpret(Float32,reinterpret(UInt32,NaN32)+(rand(UInt32)>>10))
@test Posit16(f) == Posit16_new(f)
@test isnan(Posit16_new(f))
@test Posit16(f) == Posit16_old(f)
@test isnan(Posit16(f))
end
end

@testset "U(0,1)" begin
for f in rand(Float32,100)
@test Posit16(f) == Posit16_new(f)
@test Float32(Posit16(f)) == Float32_new(Posit16_new(f))
@test Posit16(f) == Posit16_old(f)
@test Float32(Posit16(f)) == Float32_old(Posit16_old(f))

# idempotence
f32 = Float32(Posit16(f))
@test f32 == Float32_new(Posit16_new(f32))
@test f32 == Float32(Posit16(f32))
end
end

@testset "U(1,21)" begin
for f in 1 .+ 20*rand(Float32,100)
@test Posit16(-f) == Posit16_new(-f)
@test Posit16(f) == Posit16_new(f)
@test Float32(Posit16(f)) == Float32_new(Posit16_new(f))
@test Posit16(-f) == Posit16_old(-f)
@test Posit16(f) == Posit16_old(f)
@test Float32(Posit16(f)) == Float32_old(Posit16_old(f))

# idempotence
f32 = Float32(Posit16(f))
@test f32 == Float32_new(Posit16_new(f32))
@test f32 == Float32_old(Posit16_old(f32))
end
end

@testset "N(0,1)" begin
for f in randn(Float32,100)
@test Posit16(f) == Posit16_new(f)
@test Float32(Posit16(f)) == Float32_new(Posit16_new(f))
@test Posit16(f) == Posit16_old(f)
@test Float32(Posit16(f)) == Float32_old(Posit16_old(f))

# idempotence
f32 = Float32(Posit16(f))
@test f32 == Float32_new(Posit16_new(f32))
@test f32 == Float32(Posit16(f32))
end
end

@testset "Overflow and underflow" begin
for f in Float32[2f8,3f8,4f8,5f8,1f9,1f10,1f15,1f20,1f25,1f30,1f35,floatmax(Float32)/4]
@test floatmax(Posit16) == Posit16_new(f)
@test -floatmax(Posit16) == Posit16_new(-f)
@test floatmax(Posit16) == Posit16(f)
@test -floatmax(Posit16) == Posit16(-f)
end

@test isnan(Posit16_new(NaN32))
@test isnan(Posit16_new(Inf32))
@test isnan(Posit16_new(-Inf32))
@test isnan(Posit16(NaN32))
@test isnan(Posit16(Inf32))
@test isnan(Posit16(-Inf32))
end

2 comments on commit 817f977

@milankl
Copy link
Owner Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@JuliaRegistrator register

@JuliaRegistrator
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Registration pull request created: JuliaRegistries/General/47081

After the above pull request is merged, it is recommended that a tag is created on this repository for the registered package version.

This will be done automatically if the Julia TagBot GitHub Action is installed, or can be done manually through the github interface, or via:

git tag -a v0.4.0 -m "<description of version>" 817f97743a6a2a7380a70d0721c4d9c9b69909ae
git push origin v0.4.0

Please sign in to comment.