26 lines · plain
1; RUN: opt -aa-pipeline=globals-aa -passes='require<globals-aa>,gvn' -S < %s | FileCheck %s2; RUN: opt -aa-pipeline=basic-aa,globals-aa -passes='require<globals-aa>,gvn' -S < %s | FileCheck %s3;4; Functions w/o `nosync` attribute may communicate via memory and must be5; treated conservatively. Taken from https://reviews.llvm.org/D115302.6 7target datalayout = "e-i64:64-i128:128-v16:16-v32:32-n16:32:64"8target triple = "nvptx64-nvidia-cuda"9 10@s = internal local_unnamed_addr addrspace(3) global i32 undef, align 411 12; CHECK-LABEL: @bar_sync13; CHECK: store14; CHECK: tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0)15; CHECK: load16define dso_local i32 @bar_sync(i32 %0) local_unnamed_addr {17 store i32 %0, ptr addrspacecast (ptr addrspace(3) @s to ptr), align 418 tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0)19 %2 = load i32, ptr addrspacecast (ptr addrspace(3) @s to ptr), align 420 ret i32 %221}22 23declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #024 25attributes #0 = { convergent nounwind }26