123 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals all --version 62; RUN: opt -S -mtriple=amdgcn-- -passes=amdgpu-lower-exec-sync,amdgpu-lower-module-lds < %s 2>&1 | FileCheck %s3; RUN: llc < %s -enable-new-pm -stop-after=amdgpu-lower-module-lds -mtriple=amdgcn-amd-amdhsa | FileCheck %s4 5; Test to ensure that LDS variables like named barriers are lowered correctly,6; where amdgpu-lower-module-lds pass runs in pipeline after amdgpu-lower-exec-sync pass.7 8%class.ExpAmdWorkgroupWaveBarrier = type { target("amdgcn.named.barrier", 0) }9@bar2 = internal addrspace(3) global [2 x target("amdgcn.named.barrier", 0)] poison10@bar3 = internal addrspace(3) global target("amdgcn.named.barrier", 0) poison11@bar1 = internal addrspace(3) global [4 x %class.ExpAmdWorkgroupWaveBarrier] poison12@lds1 = internal addrspace(3) global [1 x i8] poison, align 413 14;.15; CHECK: @bar2 = internal addrspace(3) global [2 x target("amdgcn.named.barrier", 0)] poison, !absolute_symbol [[META0:![0-9]+]]16; CHECK: @bar3 = internal addrspace(3) global target("amdgcn.named.barrier", 0) poison, !absolute_symbol [[META1:![0-9]+]]17; CHECK: @bar1 = internal addrspace(3) global [4 x %class.ExpAmdWorkgroupWaveBarrier] poison, !absolute_symbol [[META2:![0-9]+]]18; CHECK: @bar1.kernel1 = internal addrspace(3) global [4 x %class.ExpAmdWorkgroupWaveBarrier] poison, !absolute_symbol [[META2]]19; CHECK: @llvm.amdgcn.module.lds = internal addrspace(3) global %llvm.amdgcn.module.lds.t poison, align 4, !absolute_symbol [[META3:![0-9]+]]20; CHECK: @llvm.compiler.used = appending addrspace(1) global [1 x ptr] [ptr addrspacecast (ptr addrspace(3) @llvm.amdgcn.module.lds to ptr)], section "llvm.metadata"21;.22define void @func1() #0 {23; CHECK-LABEL: define void @func1(24; CHECK-SAME: ) #[[ATTR0:[0-9]+]] {25; CHECK-NEXT: call void @llvm.amdgcn.s.barrier.signal.var(ptr addrspace(3) @bar3, i32 7)26; CHECK-NEXT: call void @llvm.amdgcn.s.barrier.join(ptr addrspace(3) @bar3)27; CHECK-NEXT: call void @llvm.amdgcn.s.barrier.wait(i16 1)28; CHECK-NEXT: ret void29;30 call void @llvm.amdgcn.s.barrier.signal.var(ptr addrspace(3) @bar3, i32 7)31 call void @llvm.amdgcn.s.barrier.join(ptr addrspace(3) @bar3)32 call void @llvm.amdgcn.s.barrier.wait(i16 1)33 ret void34}35 36define void @func2() #0 {37; CHECK-LABEL: define void @func2(38; CHECK-SAME: ) #[[ATTR0]] {39; CHECK-NEXT: call void @llvm.amdgcn.s.barrier.signal.var(ptr addrspace(3) @bar2, i32 7)40; CHECK-NEXT: call void @llvm.amdgcn.s.barrier.join(ptr addrspace(3) @bar2)41; CHECK-NEXT: call void @llvm.amdgcn.s.barrier.wait(i16 1)42; CHECK-NEXT: store i8 7, ptr addrspace(3) @llvm.amdgcn.module.lds, align 443; CHECK-NEXT: ret void44;45 call void @llvm.amdgcn.s.barrier.signal.var(ptr addrspace(3) @bar2, i32 7)46 call void @llvm.amdgcn.s.barrier.join(ptr addrspace(3) @bar2)47 call void @llvm.amdgcn.s.barrier.wait(i16 1)48 store i8 7, ptr addrspace(3) @lds1, align 449 ret void50}51 52define amdgpu_kernel void @kernel1() #0 {53; CHECK-LABEL: define amdgpu_kernel void @kernel1(54; CHECK-SAME: ) #[[ATTR1:[0-9]+]] {55; CHECK-NEXT: call void @llvm.donothing() [ "ExplicitUse"(ptr addrspace(3) @llvm.amdgcn.module.lds) ]56; CHECK-NEXT: call void @llvm.amdgcn.s.barrier.signal.var(ptr addrspace(3) @bar1.kernel1, i32 11)57; CHECK-NEXT: call void @llvm.amdgcn.s.barrier.join(ptr addrspace(3) @bar1.kernel1)58; CHECK-NEXT: call void @llvm.amdgcn.s.barrier.wait(i16 1)59; CHECK-NEXT: [[STATE:%.*]] = call i32 @llvm.amdgcn.s.get.named.barrier.state(ptr addrspace(3) @bar1.kernel1)60; CHECK-NEXT: call void @llvm.amdgcn.s.barrier()61; CHECK-NEXT: call void @func1()62; CHECK-NEXT: call void @func2()63; CHECK-NEXT: store i8 9, ptr addrspace(3) @llvm.amdgcn.module.lds, align 464; CHECK-NEXT: ret void65;66 call void @llvm.amdgcn.s.barrier.signal.var(ptr addrspace(3) @bar1, i32 11)67 call void @llvm.amdgcn.s.barrier.join(ptr addrspace(3) @bar1)68 call void @llvm.amdgcn.s.barrier.wait(i16 1)69 %state = call i32 @llvm.amdgcn.s.get.named.barrier.state(ptr addrspace(3) @bar1)70 call void @llvm.amdgcn.s.barrier()71 call void @func1()72 call void @func2()73 store i8 9, ptr addrspace(3) @lds1, align 474 ret void75}76 77define amdgpu_kernel void @kernel2() #0 {78; CHECK-LABEL: define amdgpu_kernel void @kernel2(79; CHECK-SAME: ) #[[ATTR1]] {80; CHECK-NEXT: call void @llvm.donothing() [ "ExplicitUse"(ptr addrspace(3) @llvm.amdgcn.module.lds) ]81; CHECK-NEXT: call void @llvm.amdgcn.s.barrier.signal.var(ptr addrspace(3) @bar1, i32 9)82; CHECK-NEXT: call void @llvm.amdgcn.s.barrier.join(ptr addrspace(3) @bar1)83; CHECK-NEXT: call void @llvm.amdgcn.s.barrier.wait(i16 1)84; CHECK-NEXT: call void @func2()85; CHECK-NEXT: store i8 10, ptr addrspace(3) @llvm.amdgcn.module.lds, align 486; CHECK-NEXT: ret void87;88 call void @llvm.amdgcn.s.barrier.signal.var(ptr addrspace(3) @bar1, i32 9)89 call void @llvm.amdgcn.s.barrier.join(ptr addrspace(3) @bar1)90 call void @llvm.amdgcn.s.barrier.wait(i16 1)91 call void @func2()92 store i8 10, ptr addrspace(3) @lds1, align 493 ret void94}95 96declare void @llvm.amdgcn.s.barrier() #197declare void @llvm.amdgcn.s.barrier.wait(i16) #198declare void @llvm.amdgcn.s.barrier.signal(i32) #199declare void @llvm.amdgcn.s.barrier.signal.var(ptr addrspace(3), i32) #1100declare i1 @llvm.amdgcn.s.barrier.signal.isfirst(i32) #1101declare void @llvm.amdgcn.s.barrier.init(ptr addrspace(3), i32) #1102declare void @llvm.amdgcn.s.barrier.join(ptr addrspace(3)) #1103declare void @llvm.amdgcn.s.barrier.leave(i16) #1104declare void @llvm.amdgcn.s.wakeup.barrier(ptr addrspace(3)) #1105declare i32 @llvm.amdgcn.s.get.named.barrier.state(ptr addrspace(3)) #1106 107attributes #0 = { nounwind }108attributes #1 = { convergent nounwind }109attributes #2 = { nounwind readnone }110 111;.112; CHECK: attributes #[[ATTR0]] = { nounwind }113; CHECK: attributes #[[ATTR1]] = { nounwind "amdgpu-lds-size"="1" }114; CHECK: attributes #[[ATTR2:[0-9]+]] = { convergent nocallback nofree nounwind willreturn }115; CHECK: attributes #[[ATTR3:[0-9]+]] = { convergent nounwind }116; CHECK: attributes #[[ATTR4:[0-9]+]] = { nocallback nofree nosync nounwind willreturn memory(none) }117;.118; CHECK: [[META0]] = !{i32 8396816, i32 8396817}119; CHECK: [[META1]] = !{i32 8396912, i32 8396913}120; CHECK: [[META2]] = !{i32 8396848, i32 8396849}121; CHECK: [[META3]] = !{i32 0, i32 1}122;.123