1 ; RUN: llc < %s | FileCheck %s 2 target triple = "nvptx64-nvidia-cuda" 3 4 declare void @llvm.nvvm.barrier0() 5 6 ; Load a value, then syncthreads. Branch, and use the loaded value only on one 7 ; side of the branch. The load shouldn't be sunk beneath the call, because 8 ; syncthreads is modeled as maystore. 9 define i32 @f(i32 %x, i32* %ptr, i1 %cond) { 10 Start: 11 ; CHECK: ld.u32 12 %ptr_val = load i32, i32* %ptr 13 ; CHECK: bar.sync 14 call void @llvm.nvvm.barrier0() 15 br i1 %cond, label %L1, label %L2 16 L1: 17 %ptr_val2 = add i32 %ptr_val, 100 18 br label %L2 19 L2: 20 %v4 = phi i32 [ %x, %Start ], [ %ptr_val2, %L1 ] 21 %v5 = add i32 %v4, 1000 22 ret i32 %v5 23 } 24
