| 1 |
; RUN: opt -mtriple=amdgcn-unknown-amdhsa -mcpu=hawaii -passes=loop-unroll -S < %s | FileCheck %s |
| 2 |
|
| 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: br |
| 9 |
define amdgpu_kernel void @test_unroll_convergent_barrier(ptr addrspace(1) noalias nocapture %out, ptr addrspace(1) noalias nocapture %in) #0 { |
| 10 |
entry: |
| 11 |
br label %for.body |
| 12 |
|
| 13 |
for.body: ; preds = %for.body, %entry |
| 14 |
%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.iv |
| 17 |
%arrayidx.out = getelementptr inbounds i32, ptr addrspace(1) %out, i32 %indvars.iv |
| 18 |
%load = load i32, ptr addrspace(1) %arrayidx.in |
| 19 |
call void @llvm.amdgcn.s.barrier() #1 |
| 20 |
%add = add i32 %load, %sum.02 |
| 21 |
store i32 %add, ptr addrspace(1) %arrayidx.out |
| 22 |
%indvars.iv.next = add i32 %indvars.iv, 1 |
| 23 |
%exitcond = icmp eq i32 %indvars.iv.next, 4 |
| 24 |
br i1 %exitcond, label %for.end, label %for.body |
| 25 |
|
| 26 |
for.end: ; preds = %for.body, %entry |
| 27 |
ret void |
| 28 |
} |
| 29 |
|
| 30 |
declare void @llvm.amdgcn.s.barrier() #1 |
| 31 |
|
| 32 |
attributes #0 = { nounwind } |
| 33 |
attributes #1 = { nounwind convergent } |