66 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-amdhsa -passes=amdgpu-attributor %s | FileCheck -check-prefix=ATTRIBUTOR_GCN %s3 4; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefix=GFX9 %s5 6define internal void @indirect() {7; ATTRIBUTOR_GCN-LABEL: define {{[^@]+}}@indirect8; ATTRIBUTOR_GCN-SAME: () #[[ATTR0:[0-9]+]] {9; ATTRIBUTOR_GCN-NEXT: ret void10;11; GFX9-LABEL: indirect:12; GFX9: ; %bb.0:13; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)14; GFX9-NEXT: s_setpc_b64 s[30:31]15 ret void16}17 18define amdgpu_kernel void @test_simple_indirect_call() {19; ATTRIBUTOR_GCN-LABEL: define {{[^@]+}}@test_simple_indirect_call20; ATTRIBUTOR_GCN-SAME: () #[[ATTR1:[0-9]+]] {21; ATTRIBUTOR_GCN-NEXT: [[FPTR:%.*]] = alloca ptr, align 8, addrspace(5)22; ATTRIBUTOR_GCN-NEXT: store ptr @indirect, ptr addrspace(5) [[FPTR]], align 8, !noalias.addrspace [[META0:![0-9]+]]23; ATTRIBUTOR_GCN-NEXT: [[FP:%.*]] = load ptr, ptr addrspace(5) [[FPTR]], align 8, !noalias.addrspace [[META0]]24; ATTRIBUTOR_GCN-NEXT: call void [[FP]]()25; ATTRIBUTOR_GCN-NEXT: ret void26;27; GFX9-LABEL: test_simple_indirect_call:28; GFX9: ; %bb.0:29; GFX9-NEXT: s_add_u32 flat_scratch_lo, s12, s1730; GFX9-NEXT: s_addc_u32 flat_scratch_hi, s13, 031; GFX9-NEXT: s_add_u32 s0, s0, s1732; GFX9-NEXT: s_addc_u32 s1, s1, 033; GFX9-NEXT: s_getpc_b64 s[18:19]34; GFX9-NEXT: s_add_u32 s18, s18, indirect@rel32@lo+435; GFX9-NEXT: s_addc_u32 s19, s19, indirect@rel32@hi+1236; GFX9-NEXT: s_mov_b32 s13, s1537; GFX9-NEXT: s_mov_b32 s12, s1438; GFX9-NEXT: s_mov_b64 s[14:15], src_private_base39; GFX9-NEXT: v_mov_b32_e32 v5, s1840; GFX9-NEXT: v_lshlrev_b32_e32 v2, 20, v241; GFX9-NEXT: v_lshlrev_b32_e32 v1, 10, v142; GFX9-NEXT: v_mov_b32_e32 v3, 043; GFX9-NEXT: v_mov_b32_e32 v4, s1544; GFX9-NEXT: v_mov_b32_e32 v6, s1945; GFX9-NEXT: v_or3_b32 v31, v0, v1, v246; GFX9-NEXT: s_mov_b32 s14, s1647; GFX9-NEXT: s_movk_i32 s32, 0x40048; GFX9-NEXT: flat_store_dwordx2 v[3:4], v[5:6]49; GFX9-NEXT: s_swappc_b64 s[30:31], s[18:19]50; GFX9-NEXT: s_endpgm51 %fptr = alloca ptr, addrspace(5)52 %fptr.cast = addrspacecast ptr addrspace(5) %fptr to ptr53 store ptr @indirect, ptr %fptr.cast54 %fp = load ptr, ptr %fptr.cast55 call void %fp()56 ret void57}58 59 60;.61; ATTRIBUTOR_GCN: attributes #[[ATTR0]] = { "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" }62; ATTRIBUTOR_GCN: attributes #[[ATTR1]] = { "uniform-work-group-size"="false" }63;.64; ATTRIBUTOR_GCN: [[META0]] = !{i32 1, i32 5, i32 6, i32 10}65;.66