[flang-commits] [flang] [flang][test] Restrict Semantics/kinds04_q10.f90 to x86_64 (PR #103724)

[Rainer Orth via flang-commits]

rorth wrote:

> I wonder why this test seems to be failing only on SPARC, since there weren't complaints of it failing on other hosts. AArch64 hosts also don't support `REAL(10)`, but, at least on the machine that I used, the test on #102890 worked correctly when targeting `x86_64`. In that test case, however, AArch64 and other targets were already being skipped.
> 
> I'm OK with restricting this test to `x86_64` only, as long as an issue is opened to track this problem, of programs using `REAL(10)` failing to compile on some hosts, even when they are targeting `x86_64`. But it would be good to wait for other reviewers' comments on this.

Given @DanielCChen's findings and similar ones by myself when compiling a minimal version of `kinds04_q10.f90` which contains only the `REAL(10)` cases, where there are differences like
```
- at _QFsEdoublevar3 = internal global double 4.000000e+06
+ at _QFsEdoublevar3 = internal global double 0x7FFC000000000000
```
between the `x86_64` and `sparcv9` outputs, I wonder how we should proceed with this patch?  Restrict it to `x86_64` for now as suggested?

Maybe @DanielCChen is in a better position to file an Issue than I am, given that he's well familiar with the code?

https://github.com/llvm/llvm-project/pull/103724