Harmen Stoppels wrote:
> > The test failures with floating-point numbers apparently come from the fact
> > that in this compiler's runtime environment, denormalized numbers don't
> > exist: they are mapped to zero.
> 
> This is documented here: 
> https://www.intel.com/content/www/us/en/docs/dpcpp-cpp-compiler/developer-guide-reference/2023-1/fp-model-fp.html,
>  it happens from -O1.
> 
> Some projects explicitly add `icx -fp-model=strict` to fix this: 
> https://github.com/python/cpython/blob/e02ac1d907d7cf32db5592593bd0c19a88cc6131/configure.ac#L2832-L2833.

Thanks. With this option, all floating-points tests pass, indeed.

I've documented it in this wiki now:
https://gitlab.com/ghwiki/gnow-how/-/wikis/Platforms/Configuration

Bruno





Reply via email to