68 lines · plain
1; RUN: llc < %s -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -start-before=amdgpu-isel -stop-after=amdgpu-isel | FileCheck %s --check-prefixes=CHECK2; RUN: llc < %s -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -start-before=amdgpu-isel -stop-after=amdgpu-isel -enable-new-pm | FileCheck %s --check-prefixes=CHECK3 4; This caused failure in infinite cycle in Selection DAG (combine) due to missing insert_subvector.5;6; CHECK-LABEL: name: test17; CHECK: GLOBAL_LOAD_DWORDX48; CHECK: GLOBAL_LOAD_DWORDX49; CHECK: GLOBAL_STORE_DWORDX410define protected amdgpu_kernel void @test1(ptr addrspace(4) %ptr) local_unnamed_addr !kernel_arg_addr_space !0 !kernel_arg_access_qual !1 !kernel_arg_type !2 !kernel_arg_base_type !2 !kernel_arg_type_qual !3 !kernel_arg_name !4 {11entry:12 %tmp = load <3 x i64>, ptr addrspace(4) %ptr, align 16, !invariant.load !513 %srcA.load2 = extractelement <3 x i64> %tmp, i32 014 %tmp1 = inttoptr i64 %srcA.load2 to ptr addrspace(1)15 %tmp2 = getelementptr inbounds double, ptr addrspace(1) %tmp1, i64 016 %tmp4 = load <3 x double>, ptr addrspace(1) %tmp2, align 8, !tbaa !617 %tmp5 = extractelement <3 x double> %tmp4, i32 118 %tmp6 = insertelement <3 x double> poison, double %tmp5, i32 119 %tmp7 = insertelement <3 x double> %tmp6, double poison, i32 220 %tmp8 = load <3 x double>, ptr addrspace(1) poison, align 8, !tbaa !621 %tmp9 = extractelement <3 x double> %tmp8, i32 222 %tmp10 = insertelement <3 x double> poison, double %tmp9, i32 223 %tmp11 = fcmp olt <3 x double> %tmp10, %tmp724 %tmp12 = select <3 x i1> %tmp11, <3 x double> zeroinitializer, <3 x double> <double 1.000000e+00, double 1.000000e+00, double 1.000000e+00>25 %tmp13 = extractelement <3 x double> %tmp12, i64 126 %tmp14 = insertelement <2 x double> poison, double %tmp13, i32 127 store <2 x double> %tmp14, ptr addrspace(1) poison, align 8, !tbaa !628 ret void29}30 31; This caused failure in Selection DAG due to lack of insert_subvector implementation.32;33; CHECK-LABEL: name: test234; CHECK: GLOBAL_LOAD_DWORDX235; CHECK: GLOBAL_LOAD_DWORDX236; CHECK: GLOBAL_STORE_DWORDX237define protected amdgpu_kernel void @test2(ptr addrspace(4) %ptr) local_unnamed_addr !kernel_arg_addr_space !0 !kernel_arg_access_qual !1 !kernel_arg_type !2 !kernel_arg_base_type !2 !kernel_arg_type_qual !3 !kernel_arg_name !4 {38entry:39 %tmp = load <3 x i64>, ptr addrspace(4) %ptr, align 16, !invariant.load !540 %srcA.load2 = extractelement <3 x i64> %tmp, i32 041 %tmp1 = inttoptr i64 %srcA.load2 to ptr addrspace(1)42 %tmp2 = getelementptr inbounds double, ptr addrspace(1) %tmp1, i64 043 %tmp4 = load <3 x double>, ptr addrspace(1) %tmp2, align 8, !tbaa !644 %tmp5 = extractelement <3 x double> %tmp4, i32 145 %tmp6 = insertelement <3 x double> poison, double %tmp5, i32 146 %tmp7 = insertelement <3 x double> %tmp6, double poison, i32 247 %tmp8 = load <3 x double>, ptr addrspace(1) poison, align 8, !tbaa !648 %tmp9 = extractelement <3 x double> %tmp8, i32 149 %tmp10 = insertelement <3 x double> poison, double %tmp9, i32 150 %tmp11 = insertelement <3 x double> %tmp10, double poison, i32 251 %tmp12 = fcmp olt <3 x double> %tmp11, %tmp752 %tmp13 = select <3 x i1> %tmp12, <3 x double> zeroinitializer, <3 x double> <double 1.000000e+00, double 1.000000e+00, double 1.000000e+00>53 %tmp14 = extractelement <3 x double> %tmp13, i64 254 store double %tmp14, ptr addrspace(1) poison, align 8, !tbaa !655 ret void56}57 58!0 = !{i32 1, i32 1, i32 1}59!1 = !{!"none", !"none", !"none"}60!2 = !{!"ptr", !"ptr", !"ptr"}61!3 = !{!"", !"", !""}62!4 = !{!"srcA", !"srcB", !"dst"}63!5 = !{}64!6 = !{!7, !7, i64 0}65!7 = !{!"double", !8, i64 0}66!8 = !{!"omnipotent char", !9, i64 0}67!9 = !{!"Simple C/C++ TBAA"}68