75 lines · plain
1; RUN: opt -S -mtriple=amdgcn-- -passes=amdgpu-lower-exec-sync,amdgpu-lower-module-lds < %s 2>&1 | FileCheck %s2; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 -verify-machineinstrs -o - %s | FileCheck -check-prefixes=SOUT %s3 4%class.ExpAmdWorkgroupWaveBarrier = type { target("amdgcn.named.barrier", 0) }5 6@bar2 = internal addrspace(3) global [2 x target("amdgcn.named.barrier", 0)] poison7@bar3 = internal addrspace(3) global target("amdgcn.named.barrier", 0) poison8@bar1 = internal addrspace(3) global [4 x %class.ExpAmdWorkgroupWaveBarrier] poison9 10; CHECK: @bar2 = internal addrspace(3) global [2 x target("amdgcn.named.barrier", 0)] poison, !absolute_symbol !011; CHECK-NEXT: @bar3 = internal addrspace(3) global target("amdgcn.named.barrier", 0) poison, !absolute_symbol !112; CHECK-NEXT: @bar1 = internal addrspace(3) global [4 x %class.ExpAmdWorkgroupWaveBarrier] poison, !absolute_symbol !213; CHECK-NEXT: @bar1.kernel1 = internal addrspace(3) global [4 x %class.ExpAmdWorkgroupWaveBarrier] poison, !absolute_symbol !214 15; SOUT: .set func1.num_named_barrier, 716define void @func1() {17 call void @llvm.amdgcn.s.barrier.signal.var(ptr addrspace(3) @bar3, i32 7)18 call void @llvm.amdgcn.s.barrier.join(ptr addrspace(3) @bar3)19 call void @llvm.amdgcn.s.barrier.wait(i16 1)20 ret void21}22 23; SOUT: .set func2.num_named_barrier, 224define void @func2() {25 call void @llvm.amdgcn.s.barrier.signal.var(ptr addrspace(3) @bar2, i32 7)26 call void @llvm.amdgcn.s.barrier.join(ptr addrspace(3) @bar2)27 call void @llvm.amdgcn.s.barrier.wait(i16 1)28 ret void29}30 31; SOUT: .amdhsa_named_barrier_count 232; SOUT: .set kernel1.num_named_barrier, max(6, func1.num_named_barrier, func2.num_named_barrier)33define amdgpu_kernel void @kernel1() #0 {34; CHECK-DAG: call void @llvm.amdgcn.s.barrier.signal.var(ptr addrspace(3) @bar1.kernel1, i32 11)35 call void @llvm.amdgcn.s.barrier.signal.var(ptr addrspace(3) @bar1, i32 11)36 call void @llvm.amdgcn.s.barrier.join(ptr addrspace(3) @bar1)37 call void @llvm.amdgcn.s.barrier.wait(i16 1)38 %state = call i32 @llvm.amdgcn.s.get.named.barrier.state(ptr addrspace(3) @bar1)39 call void @llvm.amdgcn.s.barrier()40 call void @func1()41 call void @func2()42 ret void43}44 45; SOUT: .amdhsa_named_barrier_count 246; SOUT: .set kernel2.num_named_barrier, max(6, func2.num_named_barrier)47define amdgpu_kernel void @kernel2() #0 {48; CHECK-DAG: call void @llvm.amdgcn.s.barrier.signal.var(ptr addrspace(3) @bar1, i32 9)49 call void @llvm.amdgcn.s.barrier.signal.var(ptr addrspace(3) @bar1, i32 9)50 call void @llvm.amdgcn.s.barrier.join(ptr addrspace(3) @bar1)51 call void @llvm.amdgcn.s.barrier.wait(i16 1)52 53 call void @func2()54 ret void55}56 57declare void @llvm.amdgcn.s.barrier() #158declare void @llvm.amdgcn.s.barrier.wait(i16) #159declare void @llvm.amdgcn.s.barrier.signal(i32) #160declare void @llvm.amdgcn.s.barrier.signal.var(ptr addrspace(3), i32) #161declare i1 @llvm.amdgcn.s.barrier.signal.isfirst(i32) #162declare void @llvm.amdgcn.s.barrier.init(ptr addrspace(3), i32) #163declare void @llvm.amdgcn.s.barrier.join(ptr addrspace(3)) #164declare void @llvm.amdgcn.s.barrier.leave(i16) #165declare void @llvm.amdgcn.s.wakeup.barrier(ptr addrspace(3)) #166declare i32 @llvm.amdgcn.s.get.named.barrier.state(ptr addrspace(3)) #167 68attributes #0 = { nounwind }69attributes #1 = { convergent nounwind }70attributes #2 = { nounwind readnone }71 72; CHECK: !0 = !{i32 8396816, i32 8396817}73; CHECK-NEXT: !1 = !{i32 8396912, i32 8396913}74; CHECK-NEXT: !2 = !{i32 8396848, i32 8396849}75