108 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --function-signature --check-globals2; RUN: opt -S -mtriple=amdgcn-amd- -passes=amdgpu-attributor %s | FileCheck %s3 4; Test to ensure recursive functions exhibit proper behaviour5; Test to generate fibonacci numbers6 7define i32 @fib(i32 %n) #0 {8; CHECK-LABEL: define {{[^@]+}}@fib9; CHECK-SAME: (i32 [[N:%.*]]) #[[ATTR0:[0-9]+]] {10; CHECK-NEXT: [[CMP1:%.*]] = icmp eq i32 [[N]], 011; CHECK-NEXT: br i1 [[CMP1]], label [[EXIT:%.*]], label [[CONT1:%.*]]12; CHECK: cont1:13; CHECK-NEXT: [[CMP2:%.*]] = icmp eq i32 [[N]], 114; CHECK-NEXT: br i1 [[CMP2]], label [[EXIT]], label [[CONT2:%.*]]15; CHECK: cont2:16; CHECK-NEXT: [[NM1:%.*]] = sub i32 [[N]], 117; CHECK-NEXT: [[FIBM1:%.*]] = call i32 @fib(i32 [[NM1]])18; CHECK-NEXT: [[NM2:%.*]] = sub i32 [[N]], 219; CHECK-NEXT: [[FIBM2:%.*]] = call i32 @fib(i32 [[NM2]])20; CHECK-NEXT: [[RETVAL:%.*]] = add i32 [[FIBM1]], [[FIBM2]]21; CHECK-NEXT: ret i32 [[RETVAL]]22; CHECK: exit:23; CHECK-NEXT: ret i32 124;25 %cmp1 = icmp eq i32 %n, 026 br i1 %cmp1, label %exit, label %cont127 28cont1:29 %cmp2 = icmp eq i32 %n, 130 br i1 %cmp2, label %exit, label %cont231 32cont2:33 %nm1 = sub i32 %n, 134 %fibm1 = call i32 @fib(i32 %nm1)35 %nm2 = sub i32 %n, 236 %fibm2 = call i32 @fib(i32 %nm2)37 %retval = add i32 %fibm1, %fibm238 39 ret i32 %retval40 41exit:42 ret i32 143}44 45define internal i32 @fib_internal(i32 %n) #0 {46; CHECK-LABEL: define {{[^@]+}}@fib_internal47; CHECK-SAME: (i32 [[N:%.*]]) #[[ATTR1:[0-9]+]] {48; CHECK-NEXT: [[CMP1:%.*]] = icmp eq i32 [[N]], 049; CHECK-NEXT: br i1 [[CMP1]], label [[EXIT:%.*]], label [[CONT1:%.*]]50; CHECK: cont1:51; CHECK-NEXT: [[CMP2:%.*]] = icmp eq i32 [[N]], 152; CHECK-NEXT: br i1 [[CMP2]], label [[EXIT]], label [[CONT2:%.*]]53; CHECK: cont2:54; CHECK-NEXT: [[NM1:%.*]] = sub i32 [[N]], 155; CHECK-NEXT: [[FIBM1:%.*]] = call i32 @fib_internal(i32 [[NM1]])56; CHECK-NEXT: [[NM2:%.*]] = sub i32 [[N]], 257; CHECK-NEXT: [[FIBM2:%.*]] = call i32 @fib_internal(i32 [[NM2]])58; CHECK-NEXT: [[RETVAL:%.*]] = add i32 [[FIBM1]], [[FIBM2]]59; CHECK-NEXT: ret i32 [[RETVAL]]60; CHECK: exit:61; CHECK-NEXT: ret i32 162;63 %cmp1 = icmp eq i32 %n, 064 br i1 %cmp1, label %exit, label %cont165 66cont1:67 %cmp2 = icmp eq i32 %n, 168 br i1 %cmp2, label %exit, label %cont269 70cont2:71 %nm1 = sub i32 %n, 172 %fibm1 = call i32 @fib_internal(i32 %nm1)73 %nm2 = sub i32 %n, 274 %fibm2 = call i32 @fib_internal(i32 %nm2)75 %retval = add i32 %fibm1, %fibm276 77 ret i32 %retval78 79exit:80 ret i32 181}82 83define amdgpu_kernel void @kernel(ptr addrspace(1) %m) #1 {84; CHECK-LABEL: define {{[^@]+}}@kernel85; CHECK-SAME: (ptr addrspace(1) [[M:%.*]]) #[[ATTR2:[0-9]+]] {86; CHECK-NEXT: [[R:%.*]] = call i32 @fib(i32 5)87; CHECK-NEXT: [[R2:%.*]] = call i32 @fib_internal(i32 5)88; CHECK-NEXT: store i32 [[R]], ptr addrspace(1) [[M]], align 489; CHECK-NEXT: store i32 [[R2]], ptr addrspace(1) [[M]], align 490; CHECK-NEXT: ret void91;92 %r = call i32 @fib(i32 5)93 %r2 = call i32 @fib_internal(i32 5)94 95 store i32 %r, ptr addrspace(1) %m96 store i32 %r2, ptr addrspace(1) %m97 ret void98}99 100; nounwind and readnone are added to match attributor results.101attributes #0 = { nounwind readnone }102attributes #1 = { "uniform-work-group-size"="true" }103;.104; CHECK: attributes #[[ATTR0]] = { nounwind memory(none) "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" }105; CHECK: attributes #[[ATTR1]] = { nounwind memory(none) "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="true" }106; CHECK: attributes #[[ATTR2]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="true" }107;.108