65 lines · plain
1; RUN: opt -mtriple amdgcn-unknown-amdhsa -passes='print<uniformity>' -disable-output %s 2>&1 | FileCheck %s2 3define amdgpu_kernel void @hidden_diverge(i32 %n, i32 %a, i32 %b) #0 {4; CHECK-LABEL: for function 'hidden_diverge'5entry:6 %tid = call i32 @llvm.amdgcn.workitem.id.x()7 %cond.var = icmp slt i32 %tid, 08 br i1 %cond.var, label %B, label %C ; divergent9; CHECK: DIVERGENT: %cond.var =10; CHECK: DIVERGENT: br i1 %cond.var,11B:12 %cond.uni = icmp slt i32 %n, 013 br i1 %cond.uni, label %C, label %merge ; uniform14; CHECK-NOT: DIVERGENT: br i1 %cond.uni,15C:16 %phi.var.hidden = phi i32 [ 1, %entry ], [ 2, %B ]17; CHECK: DIVERGENT: %phi.var.hidden = phi i3218 br label %merge19merge:20 %phi.ipd = phi i32 [ %a, %B ], [ %b, %C ]21; CHECK: DIVERGENT: %phi.ipd = phi i3222 ret void23}24 25define amdgpu_kernel void @hidden_loop_ipd(i32 %n, i32 %a, i32 %b) #0 {26; CHECK-LABEL: for function 'hidden_loop_ipd'27entry:28 %tid = call i32 @llvm.amdgcn.workitem.id.x()29 %cond.var = icmp slt i32 %tid, 030; CHECK: DIVERGENT: %cond.var = icmp31 %cond.uni = icmp slt i32 %n, 032; CHECK-NOT: DIVERGENT: %cond.uni = icmp33 br label %for.header34for.header:35 br i1 %cond.var, label %A, label %B36A:37 br label %C38B:39 br label %C40C:41 br i1 %cond.uni, label %E, label %D42D:43 br i1 %cond.var, label %for.header, label %F44 45E:46 %e.lcssa.uni = phi i32 [ 0, %C ]47; CHECK-NOT: DIVERGENT: %e.lcssa.uni = phi i3248 br label %G49 50F:51 %f.lcssa.uni = phi i32 [ 1, %D ]52; CHECK-NOT: DIVERGENT: %f.lcssa.uni = phi i3253 br label %G54 55G:56 %g.join.var = phi i32 [ %e.lcssa.uni, %E ], [ %f.lcssa.uni, %F ]57; CHECK: DIVERGENT: %g.join.var = phi i3258 ret void59}60 61 62declare i32 @llvm.amdgcn.workitem.id.x() #063 64attributes #0 = { nounwind readnone }65