[PATCH] D83397: [flang] Replace uses of _Complex with std::complex

[Eric Christopher via llvm-commits]

IIRC the two types are guaranteed to be layout compatible between C and C++
so there should be no abstraction penalty here?

-eric

On Fri, Jul 24, 2020 at 9:03 AM Jean Perier via Phabricator <
reviews at reviews.llvm.org> wrote:

> jeanPerier added a comment.
>
> In this very spot, the goal of the template is to digest the runtime
> headers in order to lower their function signature to an `mlir::FuncOp`
> declaration that is compatible with the runtime and will be call in user
> programs.
> So if we use `std::complex` here and introduce a wrapper indirection in C
> around the runtime, then the cost of this indirection will be paid by the
> user program. I am not comfortable with having a Fortran programmer pay the
> price of a C/C++ compatibility issues.
>
> However, you are absolutely correct that in this very code spot we are
> discussing, using `_Complex` is not essential. All we need to know "this
> runtime interface is using C99 complex", so we could define a type like
> `Fortran::common::C99complex<T>` instead, and  use it as a tag here.
>
> In folding the the C99 `_Complex` functions are called from C++ the code
> around the comment I linked above is dealing with this interface (or was,
> since part of it was reverted). Your wrappers would be a solution here
> given folding performance is not as critical, but given folding with pgmath
> is the non-default option (under ifdef, I think it is easier to only enable
> it if `_Complex` of `_FComplex` is available and resort to the default
> usage of cmath in the other cases.
>
> Thanks for pointing out we do not really use the interface in this very
> spot ! I think I can now put up a solution that should fulfill all wishes:
>
> - no warnings, compiles on all C++ compliant compiler.
> - no silent undefined behavior for X86_32 and the whatever other ABI out
> there deals with _Complex in a special way
> - no user program performance hit
> - still possible to fold with pgmath if you are using a mainstream C++
> compiler to compile flang (g++, clang++, msvc when possible)
> - no hard error if using pgmath complex functions is not possible
> - not too much additional code complexity in flang or the runtime.
>
>
> Repository:
>   rG LLVM Github Monorepo
>
> CHANGES SINCE LAST ACTION
>   https://reviews.llvm.org/D83397/new/
>
> https://reviews.llvm.org/D83397
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20200726/e2f2ee56/attachment.html>