34 lines · plain
1; RUN: opt -mtriple=amdgcn-unknown-amdhsa -mcpu=hawaii -passes=loop-unroll -S < %s | FileCheck %s2 3; CHECK-LABEL: @test_unroll_convergent_barrier(4; CHECK: call void @llvm.amdgcn.s.barrier()5; CHECK: call void @llvm.amdgcn.s.barrier()6; CHECK: call void @llvm.amdgcn.s.barrier()7; CHECK: call void @llvm.amdgcn.s.barrier()8; CHECK-NOT: br9define amdgpu_kernel void @test_unroll_convergent_barrier(ptr addrspace(1) noalias nocapture %out, ptr addrspace(1) noalias nocapture %in) #0 {10entry:11 br label %for.body12 13for.body: ; preds = %for.body, %entry14 %indvars.iv = phi i32 [ %indvars.iv.next, %for.body ], [ 0, %entry ]15 %sum.02 = phi i32 [ %add, %for.body ], [ 0, %entry ]16 %arrayidx.in = getelementptr inbounds i32, ptr addrspace(1) %in, i32 %indvars.iv17 %arrayidx.out = getelementptr inbounds i32, ptr addrspace(1) %out, i32 %indvars.iv18 %load = load i32, ptr addrspace(1) %arrayidx.in19 call void @llvm.amdgcn.s.barrier() #120 %add = add i32 %load, %sum.0221 store i32 %add, ptr addrspace(1) %arrayidx.out22 %indvars.iv.next = add i32 %indvars.iv, 123 %exitcond = icmp eq i32 %indvars.iv.next, 424 br i1 %exitcond, label %for.end, label %for.body25 26for.end: ; preds = %for.body, %entry27 ret void28}29 30declare void @llvm.amdgcn.s.barrier() #131 32attributes #0 = { nounwind }33attributes #1 = { nounwind convergent }34