Skip to content

Stochasically round on conversions from larger types #82

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

Open
wants to merge 3 commits into
base: main
Choose a base branch
from
Open

Stochasically round on conversions from larger types #82

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
ffcbc88
Stochasically round on conversions from larger types
pjaap Apr 29, 2025
59929da
Add test for deterministic -> stochastic
pjaap May 7, 2025
31c5355
Improve formatting a bit
pjaap Sep 8, 2025
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
2 changes: 1 addition & 1 deletion Project.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,7 +1,7 @@
name = "StochasticRounding"
uuid = "3843c9a1-1f18-49ff-9d99-1b4c8a8e97ed"
authors = ["Milan Kloewer and StochasticRounding.jl contributors"]
version = "0.8.3"
version = "0.9.0"

[deps]
BFloat16s = "ab4f0b2a-ad5b-11e8-123f-65d77653426b"
Expand Down
10 changes: 5 additions & 5 deletions src/general.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,10 +12,10 @@ for t in (Int8, Int16, Int32, Int64, Int128, UInt8, UInt16, UInt32, UInt64, UInt
end

# other floats, irrational and rationals
(::Type{T})(x::Real) where {T<:AbstractStochasticFloat} = stochastic_float(float(T)(x))
(::Type{T})(x::Rational) where {T<:AbstractStochasticFloat} = stochastic_float(float(T)(x))
(::Type{T})(x::Real) where {T<:AbstractStochasticFloat} = stochastic_round(T,x)
(::Type{T})(x::Rational) where {T<:AbstractStochasticFloat} = stochastic_round(stochastic_float(float(T)),x)
(::Type{T})(x::AbstractStochasticFloat) where {T<:AbstractFloat} = convert(T,float(x))
(::Type{T})(x::AbstractStochasticFloat) where {T<:AbstractStochasticFloat} = stochastic_float(convert(float(T),float(x)))
(::Type{T})(x::AbstractStochasticFloat) where {T<:AbstractStochasticFloat} = stochastic_round(T,float(x))
DoubleFloats.Double64(x::T) where T<:AbstractStochasticFloat = Double64(float(x))

# masks same as for deterministic floats
Expand Down Expand Up @@ -104,8 +104,8 @@ Base.abs(x::AbstractStochasticFloat) = stochastic_float(abs(float(x)))

# stochastic rounding
export stochastic_round
stochastic_round(T::Type{<:AbstractFloat},x::Real) = stochastic_round(T,widen(stochastic_float(T))(x))
stochastic_round(T::Type{<:AbstractStochasticFloat},x::AbstractFloat) = stochastic_float(stochastic_round(float(T),x))
stochastic_round(T::Type{<:AbstractFloat}, x::Real) = stochastic_round(T,widen(stochastic_float(T))(x))
stochastic_round(T::Type{<:AbstractStochasticFloat}, x::Real) = stochastic_float(stochastic_round(float(T),x))

# Comparison
for op in (:(==), :<, :<=, :isless)
Expand Down
48 changes: 39 additions & 9 deletions test/conversions.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,9 +1,9 @@
@testset "Converting Stochastic FP to BFloat16sr" begin
@testset "Deterministic conversion of Stochastic FP to BFloat16sr" begin
for k in 1:100000
trueVal = randn(Float64)
# Convert to each deterministic type to ensure the result is
# Convert to each deterministic type to ensure the result is
# representable in all precisions
trueVal=Float16(Float32(BFloat16(trueVal)))
trueVal = Float16(Float32(BFloat16(trueVal)))
bfloatVal = BFloat16sr(trueVal)
float16Val = Float16sr(trueVal)
float32Val = Float32sr(trueVal)
Expand All @@ -14,12 +14,12 @@
end
end

@testset "Converting Stochastic FP to Float16sr" begin
@testset "Deterministic conversion of Stochastic FP to Float16sr" begin
for k in 1:100000
trueVal = randn(Float64)
# Convert to each deterministic type to ensure the result is
# Convert to each deterministic type to ensure the result is
# representable in all precisions
trueVal=Float16(Float32(BFloat16(trueVal)))
trueVal = Float16(Float32(BFloat16(trueVal)))
bfloatVal = BFloat16sr(trueVal)
float16Val = Float16sr(trueVal)
float32Val = Float32sr(trueVal)
Expand All @@ -30,12 +30,12 @@ end
end
end

@testset "Converting Stochastic FP to Float32sr" begin
@testset "Deterministic conversion of Stochastic FP to Float32sr" begin
for k in 1:100000
trueVal = randn(Float64)
# Convert to each deterministic type to ensure the result is
# Convert to each deterministic type to ensure the result is
# representable in all precisions
trueVal=Float16(Float32(BFloat16(trueVal)))
trueVal = Float16(Float32(BFloat16(trueVal)))
bfloatVal = BFloat16sr(trueVal)
float16Val = Float16sr(trueVal)
float32Val = Float32sr(trueVal)
Expand All @@ -45,3 +45,33 @@ end
@test float32Val == Float32sr(float16Val)
end
end


@testset "Stochastic conversion of Float64 to $(SR)" for SR in [Float16sr, BFloat16sr, Float32sr]

# corresponding FP type
FP = float(SR)

for k in 1:100
trueVal = randn(Float64)

# make sure that this is not representable by FP
if trueVal == FP(trueVal)
trueVal = nextfloat(trueVal)
end

roundedVal = SR(trueVal)

# check that rounding is not deterministic
is_stochastic = false
for l in 1:10000
if roundedVal != SR(trueVal)
is_stochastic = true
break
end
end

# the redundant "== true" produces a better error message in the test log
@test is_stochastic == true
end
end