-
Notifications
You must be signed in to change notification settings - Fork 136
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.
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 a new tester that utilizes tlfloat #574
Comments
Hello, Sorry for the delay in replying. Here are a few general comments:
Some answers/reactions to your issue:
I still wonder what sort of environment does not have access to MPFR? This did not occur to me that this could be a difficult dependency to satisfy.
Restriction to C++20 and clang 16+, might be acceptable although not ideal. Maybe some of these bugs can be addressed?
If the user can choose between old and new tester then I suppose the C++ compiler dependency is not a huge issue. |
If so, could you briefly explain the plan in a new issue?
Yeah. We should add documentation across the board.
MPFR is quite troublesome to use on Windows. |
Sure, we will post about it in an issue shortly.
We have the ability to test on WoA internally. We work under Msys2 (which a unix-like environment) and it is relatively easy to install gmp, mpc and mpfr using the package manager |
Are you testing building SLEEF with MSVC? |
We are building with clang, it's proved simpler on many occasions. |
In any case, using MPFR with MSVC is a bit tricky, and I think it would be ideal to test SLEEF properly with MSVC. The new tester should be useful for this purpose. |
Sure! Looking forward to the PR. |
For now, we didn't create a issue as #9 is still open.
Should upload a PR introducing this new tool soon - will keep you updated. |
Why a new feature? Is your feature request related to a problem? Please describe it.
The new tester will work without a pipe connecting two processes. Introducing this tester will resolve the posix dependency problem.
The new tester can be used in environments where MPFR is not available. It will be much easier to do testing on Windows.
I expect the new tester to work faster than the current tester/iut combo.
What feature would you like?
How would you like this feature to be implemented?
As described in Incorporating tlfloat as a submodule #568, I will incorporate TLFloat as a submodule of SLEEF.
The structure will be like a combination of the existing IUT and tester, and will be built with different settings for each vector extension.
How portable is the feature across architectures and platforms?
The purpose of introducing this tester is to address the current portability issues associated with MPFR and pipe.
TLFloat has fewer portability issues than MPFR and can be used in a wider range of environments.
However, it will require a C++20 compiler and cannot be compiled with clang 16 or earlier due to compiler bugs.
Describe alternative approaches you have considered
The current tester/iut will not be deleted.
At the time of build, the user can select either the current tester/iut or the new tester to be introduced.
I will allow the user to choose whether to build the current tester/iut and the new tester by cmake options.
Additional context
The addition of the new tester will not constitute a major code change.
However, the addition of C++ code and the introduction of submodules may affect the user's build environment and future development.
The default setting is to test with the existing tester/iut, as is the case at present. Only when new cmake options are specified, the new tester will be built and used for testing instead of the current tester/iut.
The text was updated successfully, but these errors were encountered: