Skip to content
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

Add bang to set_sqrt_type and set_log_type #2031

Merged
merged 2 commits into from
Aug 7, 2024
Merged

Add bang to set_sqrt_type and set_log_type #2031

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
6beea29
add bang to set_sqrt_type and set_log_type
JoshuaLampert Aug 5, 2024
0084407
Apply suggestions from code review
JoshuaLampert Aug 7, 2024
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
52 changes: 29 additions & 23 deletions src/auxiliary/math.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,44 +23,47 @@ function set_polyester!(toggle::Bool; force = true)
end

"""
Trixi.set_sqrt_type(type; force = true)
Trixi.set_sqrt_type!(type; force = true)

Set the `type` of the square root function to be used in Trixi.jl.
The default is `"sqrt_Trixi_NaN"` which returns `NaN` for negative arguments
instead of throwing an error.
Alternatively, you can set `type` to `"sqrt_Base"` to use the Julia built-in `sqrt` function
Alternatively, you can set `type` to `"sqrt_Base"` to use the Julia built-in `sqrt` function
which provides a stack-trace of the error which might come in handy when debugging code.
"""
function set_sqrt_type(type; force = true)
function set_sqrt_type!(type; force = true)
@assert type == "sqrt_Trixi_NaN"||type == "sqrt_Base" "Only allowed `sqrt` function types are `\"sqrt_Trixi_NaN\"` and `\"sqrt_Base\"`"
set_preferences!(TRIXI_UUID, "sqrt" => type, force = force)
@info "Please restart Julia and reload Trixi.jl for the `sqrt` computation change to take effect"
end

# TODO: deprecation introduced in v0.8
@deprecate set_sqrt_type(type; force = true) set_sqrt_type!(type; force = true) false
JoshuaLampert marked this conversation as resolved.
Show resolved Hide resolved

@static if _PREFERENCE_SQRT == "sqrt_Trixi_NaN"
"""
Trixi.sqrt(x::Real)

Custom square root function which returns `NaN` for negative arguments instead of throwing an error.
This is required to ensure [correct results for multithreaded computations](https://github.com/trixi-framework/Trixi.jl/issues/1766)
when using the [`Polyester` package](https://github.com/JuliaSIMD/Polyester.jl),
This is required to ensure [correct results for multithreaded computations](https://github.com/trixi-framework/Trixi.jl/issues/1766)
when using the [`Polyester` package](https://github.com/JuliaSIMD/Polyester.jl),
i.e., using the `@batch` macro instead of the Julia built-in `@threads` macro, see [`@threaded`](@ref).

We dispatch this function for `Float64, Float32, Float16` to the LLVM intrinsics
`llvm.sqrt.f64`, `llvm.sqrt.f32`, `llvm.sqrt.f16` as for these the LLVM functions can be used out-of the box,
We dispatch this function for `Float64, Float32, Float16` to the LLVM intrinsics
`llvm.sqrt.f64`, `llvm.sqrt.f32`, `llvm.sqrt.f16` as for these the LLVM functions can be used out-of the box,
i.e., they return `NaN` for negative arguments.
In principle, one could also use the `sqrt_llvm` call, but for transparency and consistency with [`log`](@ref) we
spell out the datatype-dependent functions here.
In principle, one could also use the `sqrt_llvm` call, but for transparency and consistency with [`log`](@ref) we
spell out the datatype-dependent functions here.
For other types, such as integers or dual numbers required for algorithmic differentiation, we
fall back to the Julia built-in `sqrt` function after a check for negative arguments.
Since these cases are not performance critical, the check for negativity does not hurt here
Since these cases are not performance critical, the check for negativity does not hurt here
and can (as of now) even be optimized away by the compiler due to the implementation of `sqrt` in Julia.

When debugging code, it might be useful to change the implementation of this function to redirect to
the Julia built-in `sqrt` function, as this reports the exact place in code where the domain is violated
When debugging code, it might be useful to change the implementation of this function to redirect to
the Julia built-in `sqrt` function, as this reports the exact place in code where the domain is violated
in the stacktrace.

See also [`Trixi.set_sqrt_type`](@ref).
See also [`Trixi.set_sqrt_type!`](@ref).
"""
@inline sqrt(x::Real) = x < zero(x) ? oftype(x, NaN) : Base.sqrt(x)

Expand All @@ -74,41 +77,44 @@ end
end

"""
Trixi.set_log_type(type; force = true)
Trixi.set_log_type!(type; force = true)

Set the `type` of the (natural) `log` function to be used in Trixi.jl.
The default is `"sqrt_Trixi_NaN"` which returns `NaN` for negative arguments
instead of throwing an error.
Alternatively, you can set `type` to `"sqrt_Base"` to use the Julia built-in `sqrt` function
Alternatively, you can set `type` to `"sqrt_Base"` to use the Julia built-in `sqrt` function
which provides a stack-trace of the error which might come in handy when debugging code.
"""
function set_log_type(type; force = true)
function set_log_type!(type; force = true)
@assert type == "log_Trixi_NaN"||type == "log_Base" "Only allowed log function types are `\"log_Trixi_NaN\"` and `\"log_Base\"`."
set_preferences!(TRIXI_UUID, "log" => type, force = force)
@info "Please restart Julia and reload Trixi.jl for the `log` computation change to take effect"
end

# TODO: deprecation introduced in v0.8
@deprecate set_log_type(type; force = true) set_log_type!(type; force = true) false
JoshuaLampert marked this conversation as resolved.
Show resolved Hide resolved

@static if _PREFERENCE_LOG == "log_Trixi_NaN"
"""
Trixi.log(x::Real)

Custom natural logarithm function which returns `NaN` for negative arguments instead of throwing an error.
This is required to ensure [correct results for multithreaded computations](https://github.com/trixi-framework/Trixi.jl/issues/1766)
when using the [`Polyester` package](https://github.com/JuliaSIMD/Polyester.jl),
This is required to ensure [correct results for multithreaded computations](https://github.com/trixi-framework/Trixi.jl/issues/1766)
when using the [`Polyester` package](https://github.com/JuliaSIMD/Polyester.jl),
i.e., using the `@batch` macro instead of the Julia built-in `@threads` macro, see [`@threaded`](@ref).

We dispatch this function for `Float64, Float32, Float16` to the respective LLVM intrinsics
`llvm.log.f64`, `llvm.log.f32`, `llvm.log.f16` as for this the LLVM functions can be used out-of the box, i.e.,
We dispatch this function for `Float64, Float32, Float16` to the respective LLVM intrinsics
`llvm.log.f64`, `llvm.log.f32`, `llvm.log.f16` as for this the LLVM functions can be used out-of the box, i.e.,
they return `NaN` for negative arguments.
For other types, such as integers or dual numbers required for algorithmic differentiation, we
fall back to the Julia built-in `log` function after a check for negative arguments.
Since these cases are not performance critical, the check for negativity does not hurt here.

When debugging code, it might be useful to change the implementation of this function to redirect to
the Julia built-in `log` function, as this reports the exact place in code where the domain is violated
When debugging code, it might be useful to change the implementation of this function to redirect to
the Julia built-in `log` function, as this reports the exact place in code where the domain is violated
in the stacktrace.

See also [`Trixi.set_log_type`](@ref).
See also [`Trixi.set_log_type!`](@ref).
"""
@inline log(x::Real) = x < zero(x) ? oftype(x, NaN) : Base.log(x)

Expand Down
Loading