63 lines · plain
1; RUN: opt -S -passes=jump-threading,verify < %s | FileCheck %s2 3target datalayout = "e-i64:64-i128:128-v16:16-v32:32-n16:32:64"4target triple = "nvptx64-nvidia-cuda"5 6$wrapped_tid = comdat any7 8$foo = comdat any9 10define i32 @wrapped_tid() #0 comdat align 32 {11 %1 = call i32 @tid()12 ret i32 %113}14 15declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #116 17; We had a bug where we duplicated basic blocks containing convergent18; functions like @llvm.nvvm.barrier0 below. Verify that we don't do19; that.20define void @foo() local_unnamed_addr #2 comdat align 32 {21; CHECK-LABEL: @foo22 %1 = call i32 @tid()23 %2 = urem i32 %1, 724 br label %325 263:27 %4 = icmp eq i32 %1, 028 br i1 %4, label %5, label %629 305:31 call void @bar()32 br label %633 346:35; CHECK: call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0)36; CHECK-NOT: call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0)37 call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0)38 %7 = icmp eq i32 %2, 039 br i1 %7, label %11, label %840 418:42 %9 = icmp ult i32 %1, 4943 br i1 %9, label %10, label %1144 4510:46 call void @llvm.trap()47 unreachable48 4911:50 br label %351}52 53declare i32 @tid() #254 55declare void @bar()56 57declare void @llvm.trap() #358 59attributes #1 = { convergent }60attributes #2 = { readnone }61attributes #3 = { noreturn }62attributes #4 = { convergent }63