ci: Compile with checking enabled

[powerboat9]

No description provided.

[powerboat9]

This doesn't help resolve https://builder.sourceware.org/buildbot/#/builders/27 like I was hoping, but it should still be nice to have.

[tschwinge]

Build of multilibs is enabled by default, and also checking (yes,extra) is enabled by default, gcc/configure.ac:

# Determine the default checks.                                                                                                                                                                                                               
if test x$is_release = x ; then                                                                                                                                                                                                               
  ac_checking_flags=yes,extra                                                                                                                                                                                                                 
else                                                                                                                                                                                                                                          
  ac_checking_flags=release                                                                                                                                                                                                                   
fi])                                                                                                                                                                                                                                          

[tschwinge]

The fails are for a 32-bit x86 host system. We have .github/workflows/ccpp32alpine.yml, so what is different here, hmm...

[tschwinge]

Haha! "GCC Rust build and test (Alpine 32-bit) / build-alpine-32bit-and-check-alpine-32bit" produced:

UNSUPPORTED: rust/execute/torture/sip-hasher.rs   -O0 
UNSUPPORTED: rust/execute/torture/sip-hasher.rs   -O1 
UNSUPPORTED: rust/execute/torture/sip-hasher.rs   -O2 
UNSUPPORTED: rust/execute/torture/sip-hasher.rs   -O3 -g 
UNSUPPORTED: rust/execute/torture/sip-hasher.rs   -Os 
UNSUPPORTED: rust/execute/torture/sip-hasher.rs   -O2 -flto -fno-use-linker-plugin -flto-partition=none 
UNSUPPORTED: rust/execute/torture/sip-hasher.rs   -O2 -flto -fuse-linker-plugin -fno-fat-lto-objects 

..., because of // { dg-skip-if "" { *-*-* } { "-m32" } { "" } }, and this testing here is running with explicit -m32 (nothing wrong with that). That's why we don't see the internal compiler error: in expand_shift_1, at expmed.cc:2713 here.

That test case was added in #3915, #1247 commit 9526906 "gccrs: fix bad monomophization of generic paths". @philberty: What's the rationale behind the -m32 skip? If anything, this should skip for effective-target ilp32 (?) or only allow for lp64 (?), but that all sounds dubious, papering over an actual problem -- like this ICE (I suppose?), so yet better, instead of dg-skip-if, this should (conditionally) mark up the ICE with dg-ice.

[philberty]

@tschwinge this is my excitement where this goal test case now works but its uncovered a bug with how we do the intrinsic u32 rotate which is causing this ICE on m32

I think the issue as far as i can tell is we need to change the shift count to be masked by the size of the type but it was an related change to that patch. I'll open an issue