[CIR] Add builtin_setjmp codegen support

[badumbatish]

No description provided.

[badumbatish]

LLVM generated from default clang

define dso_local void @_Z9BI_setjmpPPv(ptr noundef initializes((0, 8), (16, 24)) %env) local_unnamed_addr #0 !dbg !10 {
entry:
    #dbg_value(ptr %env, !16, !DIExpression(), !17)
  %0 = tail call ptr @llvm.frameaddress.p0(i32 0), !dbg !18
  store ptr %0, ptr %env, align 8, !dbg !18
  %1 = tail call ptr @llvm.stacksave.p0(), !dbg !18
  %2 = getelementptr inbounds nuw i8, ptr %env, i64 16, !dbg !18
  store ptr %1, ptr %2, align 8, !dbg !18
  %3 = tail call i32 @llvm.eh.sjlj.setjmp(ptr nonnull %env), !dbg !18
  ret void, !dbg !19
}

LLVM generated from clangir clang

define dso_local void @_Z9BI_setjmpPPv(ptr initializes((0, 16)) %0) local_unnamed_addr #0 !dbg !4 {
  %2 = tail call ptr @llvm.frameaddress.p0(i32 0), !dbg !8
  store ptr %2, ptr %0, align 8, !dbg !8
  %3 = tail call ptr @llvm.stacksave.p0(), !dbg !8
  %4 = getelementptr i8, ptr %0, i64 8, !dbg !8
  store ptr %3, ptr %4, align 8, !dbg !8
  %5 = tail call i32 @llvm.eh.sjlj.setjmp(ptr nonnull %0), !dbg !8
  ret void, !dbg !9
}

I'm not sure how the ptr stride op can generate the inbounds nuw flag just like in the default clang

[badumbatish]

I can try for the setjmp in "setjmp.h" once i got my hands on a x86 linux, i'm not sure quite how to cross compile for x86_64 through llvm-lit and with x86_64's sysroot from a MacOS?!

[tommymcm]

Closes #1807

[bcardosolopes]

Thanks for your first contribution and welcome to CIR! Some comments inline.

[bcardosolopes]

I'm not sure how the ptr stride op can generate the inbounds nuw flag just like in the default clang

You should add assert(!cir::MissingFeatures::emitCheckedInBoundsGEP()); to the appropriate CIRGen place so we know how to come fix this later.

[badumbatish]

the OG behavior defers BI_setjmp codegen to the predefined lib function codegen, the equivalent in CIR/ is

  // If this is a predefined lib function (e.g. malloc), emit the call
  // using exactly the normal call path.
  if (BI.isPredefinedLibFunction(BuiltinID))
    return emitLibraryCall(*this, FD, E, CGM.getRawFunctionPointer(FD));

the latest commit replicates this behavior. I replicate the structure but didn't put up the actual implementation for MSVCRT.

[bcardosolopes]

Thanks for the latest round of updates!