[llvm-dev] lifetime_start/end

[Michael Kruse via llvm-dev]

If removing them, they still must be consistent. That is, for instance, for
any execution path that reaches an @llvm.lifetime.end, all must go through
a @llvm.lifetime.start (or none of them). I remember problem with loop
versioning in Polly, which violated this. StackColoring was not expected to
handle inconsistencies.

Michael


Am Mi., 30. Sept. 2020 um 08:16 Uhr schrieb Johannes Doerfert via llvm-dev <
llvm-dev at lists.llvm.org>:

> lifetime markers should always be OK to remove, as you noted.
>
> They restrict life ranges and removing them "just" makes the life ranges
> larger,
> regardless if the variables are used or not.
>
> That said, original code or the transformation you apply is probably
> the reason for the non-determinism, it's impossible to say from a snipped
> like this.
>
> ~ Johannes
>
>
> On 9/30/20 4:28 AM, Abid Malik via llvm-dev wrote:
> > Hello,
> >
> > What intrinsics "@llvm.lifetime.start/@llvm.lifetime.end" really do? As
> per
> > my knowledge, they define the live ranges of variables. In the following
> > code section, they seem redundant. However, when I remove them, the
> > behavior of the code becomes non-deterministic. The live ranges of the
> > variables defined by them are never used in the code.
> >
> > Thanks,
> > ---------------
> > %37 = bitcast i32* %7 to i8*
> >    call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %37) #5
> >    store i32 0, i32* %7, align 4, !tbaa !4
> >    %38 = bitcast i32* %8 to i8*
> >    call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %38) #5
> >    store i32 99, i32* %8, align 4, !tbaa !4
> >    %39 = bitcast i32* %9 to i8*
> >    call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %39) #5
> >    store i32 1, i32* %9, align 4, !tbaa !4
> >    %40 = bitcast i32* %10 to i8*
> >    call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %40) #5
> >    store i32 0, i32* %10, align 4, !tbaa !4
> >    call void @__kmpc_for_static_init_4(%struct.ident_t* nonnull @0, i32
> %23,
> > i32 34, i32* nonnull %10, i32* nonnull %7, i32* nonnull %8, i32* nonnull
> > %9, i32 1, i32 1) #5
> >    %41 = load i32, i32* %8, align 4, !tbaa !4
> >    %42 = icmp slt i32 %41, 99
> >    %43 = select i1 %42, i32 %41, i32 99
> >    store i32 %43, i32* %8, align 4, !tbaa !4
> >    %44 = load i32, i32* %7, align 4, !tbaa !4
> >    br label %45
> >
> > 45:                                               ; preds = %49, %36
> >    %46 = phi i32 [ %44, %36 ], [ %52, %49 ]
> >    %47 = load i32, i32* %8, align 4, !tbaa !4
> >    %48 = icmp sgt i32 %46, %47
> >    br i1 %48, label %53, label %49
> >
> > 49:                                               ; preds = %45
> >    %50 = shl nsw i32 %46, 1
> >    %51 = call i32 (i8*, ...) @printf(i8* nonnull dereferenceable(1)
> > getelementptr inbounds ([12 x i8], [12 x i8]* @.str.2, i64 0, i64 0), i32
> > %50)
> >    %52 = add nsw i32 %46, 1
> >    br label %45
> >
> > 53:                                               ; preds = %45
> >    call void @__kmpc_for_static_fini(%struct.ident_t* nonnull @0, i32
> %23) #5
> >    call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %40) #5
> >    call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %39) #5
> >    call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %38) #5
> >    call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %37) #5
> >    call void @__kmpc_barrier(%struct.ident_t* nonnull @1, i32 %23) #5
> >
> > _______________________________________________
> > LLVM Developers mailing list
> > llvm-dev at lists.llvm.org
> > https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20200930/e69e5118/attachment.html>