Re: [committed] amdgcn: disable TImode

[Andrew Stubbs <ams@codesourcery.com>]

Indeed, libgomp fails to build:

configure: error: unsupported system, cannot find Fortran int kind=16, 
needed for omp_depend_kind

I've reverted the patch for now. We'll just have to put up with a lot of 
new test failures in the stand-alone toolchain so that the offload 
toolchain will at least build.

I suspect we'll see some real failures here soon though.

Andrew

On 07/05/2021 23:45, Andrew Stubbs wrote:
> On 07/05/2021 18:11, Tobias Burnus wrote:
>> On 07.05.21 18:35, Andrew Stubbs wrote:
>>
>>> TImode has always been a problem on amdgcn, and now it is causing many
>>> new test failures, so I'm disabling it.
>>
>> Does still still work with libgomp?
>>
>> The patch sounds as if it might cause problems, but on the other hand,
>> I assume you did test it? To recall:
>  >
>> The problem is that OpenMP's depobj as implemented in GCC has
>> sizeof() = 2*sizeof(void*) and is implemented as a two-element struct 
>> in C/C++.
>> But the OpenMP spec mandates that it is an integer type in Fortran, i.e.
>> integer(kind=omp_depend_kind).
> 
> I was focussing on getting the raw amdgcn toolchain toolchain to work 
> again. I had forgotten about that little detail with Fortran. :-(
> 
> Is there another way we can fix this without implementing all of TImode 
> support or tracking down every place in the middle-end that wants to use 
> TImode without checking the optab?
> 
>> Combining the impl choice and the type requirements that means that
>> on 64bit systems, this requires __int128 support, cf. commit
>> https://gcc.gnu.org/g:8d0b2b33748014ee57973c1d7bc9fd7706bb3da9
>> and https://gcc.gnu.org/PR96306
>>
>> (Side note: The definition in OpenMP is bad - it should have been
>> some opaque derived type but that's a mistake done in OpenMP 5.0.)
> 
> :-(
> 
> Andrew