1; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=hawaii -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,CI %s 2; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX9 %s 3 4; GCN-LABEL: {{^}}is_local_vgpr: 5; GCN-DAG: {{flat|global}}_load_dwordx2 v{{\[[0-9]+}}:[[PTR_HI:[0-9]+]]{{\]}} 6; CI-DAG: s_load_dword [[APERTURE:s[0-9]+]], s[4:5], 0x10 7; GFX9-DAG: s_getreg_b32 [[APERTURE:s[0-9]+]], hwreg(HW_REG_SH_MEM_BASES, 16, 16) 8; GFX9: s_lshl_b32 [[APERTURE]], [[APERTURE]], 16 9 10; GCN: v_cmp_eq_u32_e32 vcc, [[APERTURE]], v[[PTR_HI]] 11; GCN: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 1, vcc 12define amdgpu_kernel void @is_local_vgpr(i8* addrspace(1)* %ptr.ptr) { 13 %id = call i32 @llvm.amdgcn.workitem.id.x() 14 %gep = getelementptr inbounds i8*, i8* addrspace(1)* %ptr.ptr, i32 %id 15 %ptr = load volatile i8*, i8* addrspace(1)* %gep 16 %val = call i1 @llvm.amdgcn.is.shared(i8* %ptr) 17 %ext = zext i1 %val to i32 18 store i32 %ext, i32 addrspace(1)* undef 19 ret void 20} 21 22; FIXME: setcc (zero_extend (setcc)), 1) not folded out, resulting in 23; select and vcc branch. 24 25; GCN-LABEL: {{^}}is_local_sgpr: 26; CI-DAG: s_load_dword [[APERTURE:s[0-9]+]], s[4:5], 0x10{{$}} 27; GFX9-DAG: s_getreg_b32 [[APERTURE:s[0-9]+]], hwreg(HW_REG_SH_MEM_BASES, 16, 16) 28; GFX9-DAG: s_lshl_b32 [[APERTURE]], [[APERTURE]], 16 29 30; CI-DAG: s_load_dword [[PTR_HI:s[0-9]+]], s[6:7], 0x1{{$}} 31; GFX9-DAG: s_load_dword [[PTR_HI:s[0-9]+]], s[6:7], 0x4{{$}} 32 33; GCN: v_mov_b32_e32 [[V_APERTURE:v[0-9]+]], [[APERTURE]] 34; GCN: v_cmp_eq_u32_e32 vcc, [[PTR_HI]], [[V_APERTURE]] 35; GCN: s_cbranch_vccnz 36define amdgpu_kernel void @is_local_sgpr(i8* %ptr) { 37 %val = call i1 @llvm.amdgcn.is.shared(i8* %ptr) 38 br i1 %val, label %bb0, label %bb1 39 40bb0: 41 store volatile i32 0, i32 addrspace(1)* undef 42 br label %bb1 43 44bb1: 45 ret void 46} 47 48declare i32 @llvm.amdgcn.workitem.id.x() #0 49declare i1 @llvm.amdgcn.is.shared(i8* nocapture) #0 50 51attributes #0 = { nounwind readnone speculatable } 52