26 lines · plain
1; RUN: llc < %s | FileCheck %s2; RUN: %if ptxas %{ llc < %s | %ptxas-verify %}3 4target triple = "nvptx64-nvidia-cuda"5 6declare void @llvm.nvvm.barrier0()7 8; Load a value, then syncthreads. Branch, and use the loaded value only on one9; side of the branch. The load shouldn't be sunk beneath the call, because10; syncthreads is modeled as maystore.11define i32 @f(i32 %x, ptr %ptr, i1 %cond) {12Start:13 ; CHECK: ld.b3214 %ptr_val = load i32, ptr %ptr15 ; CHECK: bar.sync16 call void @llvm.nvvm.barrier0()17 br i1 %cond, label %L1, label %L218L1:19 %ptr_val2 = add i32 %ptr_val, 10020 br label %L221L2:22 %v4 = phi i32 [ %x, %Start ], [ %ptr_val2, %L1 ]23 %v5 = add i32 %v4, 100024 ret i32 %v525}26