Home | History | Annotate | Download | only in AMDGPU
      1 ; RUN: llc < %s -march=amdgcn -verify-machineinstrs | FileCheck %s
      2 
      3 ; This tests that the llvm.SI.end.cf intrinsic is not inserted into the
      4 ; loop block.  This intrinsic will be lowered to s_or_b64 by the code
      5 ; generator.
      6 
      7 ; CHECK-LABEL: {{^}}test:
      8 
      9 ; This is was lowered from the llvm.SI.end.cf intrinsic:
     10 ; CHECK: s_or_b64 exec, exec
     11 
     12 ; CHECK: [[LOOP_LABEL:[0-9A-Za-z_]+]]: ; %loop{{$}}
     13 ; CHECK-NOT: s_or_b64 exec, exec
     14 ; CHECK: s_cbranch_execnz [[LOOP_LABEL]]
     15 define amdgpu_kernel void @test(i32 addrspace(1)* %out) {
     16 entry:
     17   %cond = call i32 @llvm.amdgcn.workitem.id.x() #0
     18   %tmp0 = icmp eq i32 %cond, 0
     19   br i1 %tmp0, label %if, label %loop
     20 
     21 if:
     22   store i32 0, i32 addrspace(1)* %out
     23   br label %loop
     24 
     25 loop:
     26   %tmp1 = phi i32 [0, %entry], [0, %if], [%inc, %loop]
     27   %inc = add i32 %tmp1, %cond
     28   %tmp2 = icmp ugt i32 %inc, 10
     29   br i1 %tmp2, label %done, label %loop
     30 
     31 done:
     32   %tmp3 = getelementptr i32, i32 addrspace(1)* %out, i64 1
     33   store i32 %inc, i32 addrspace(1)* %tmp3
     34   ret void
     35 }
     36 
     37 declare i32 @llvm.amdgcn.workitem.id.x() #0
     38 
     39 attributes #0 = { nounwind readnone }
     40