| 1 | // RUN: %clang_cc1 -std=c++11 -triple x86_64-apple-darwin %s -emit-llvm -o - | FileCheck -check-prefixes=X64,CHECK %s |
| 2 | // RUN: %clang_cc1 -std=c++11 -triple amdgcn %s -emit-llvm -o - | FileCheck -check-prefixes=AMDGCN,CHECK %s |
| 3 | |
| 4 | template<typename T> |
| 5 | struct S { |
| 6 | static int n; |
| 7 | }; |
| 8 | template<typename T> int S<T>::n = 5; |
| 9 | |
| 10 | int f() { |
| 11 | // Make sure that the reference here is enough to trigger the instantiation of |
| 12 | // the static data member. |
| 13 | // CHECK: @_ZN1SIiE1nE = linkonce_odr{{.*}} global i32 5 |
| 14 | int a[S<int>::n]; |
| 15 | return sizeof a; |
| 16 | } |
| 17 | |
| 18 | // rdar://problem/9506377 |
| 19 | void test0(void *array, int n) { |
| 20 | // CHECK-LABEL: define void @_Z5test0Pvi( |
| 21 | // X64: [[ARRAY:%.*]] = alloca i8*, align 8 |
| 22 | // AMDGCN: [[ARRAY0:%.*]] = alloca i8*, align 8, addrspace(5) |
| 23 | // AMDGCN-NEXT: [[ARRAY:%.*]] = addrspacecast i8* addrspace(5)* [[ARRAY0]] to i8** |
| 24 | // X64-NEXT: [[N:%.*]] = alloca i32, align 4 |
| 25 | // AMDGCN: [[N0:%.*]] = alloca i32, align 4, addrspace(5) |
| 26 | // AMDGCN-NEXT: [[N:%.*]] = addrspacecast i32 addrspace(5)* [[N0]] to i32* |
| 27 | // X64-NEXT: [[REF:%.*]] = alloca i16*, align 8 |
| 28 | // AMDGCN: [[REF0:%.*]] = alloca i16*, align 8, addrspace(5) |
| 29 | // AMDGCN-NEXT: [[REF:%.*]] = addrspacecast i16* addrspace(5)* [[REF0]] to i16** |
| 30 | // X64-NEXT: [[S:%.*]] = alloca i16, align 2 |
| 31 | // AMDGCN: [[S0:%.*]] = alloca i16, align 2, addrspace(5) |
| 32 | // AMDGCN-NEXT: [[S:%.*]] = addrspacecast i16 addrspace(5)* [[S0]] to i16* |
| 33 | // CHECK-NEXT: store i8* |
| 34 | // CHECK-NEXT: store i32 |
| 35 | |
| 36 | // Capture the bounds. |
| 37 | // CHECK-NEXT: [[T0:%.*]] = load i32, i32* [[N]], align 4 |
| 38 | // CHECK-NEXT: [[DIM0:%.*]] = zext i32 [[T0]] to i64 |
| 39 | // CHECK-NEXT: [[T0:%.*]] = load i32, i32* [[N]], align 4 |
| 40 | // CHECK-NEXT: [[T1:%.*]] = add nsw i32 [[T0]], 1 |
| 41 | // CHECK-NEXT: [[DIM1:%.*]] = zext i32 [[T1]] to i64 |
| 42 | typedef short array_t[n][n+1]; |
| 43 | |
| 44 | // CHECK-NEXT: [[T0:%.*]] = load i8*, i8** [[ARRAY]], align 8 |
| 45 | // CHECK-NEXT: [[T1:%.*]] = bitcast i8* [[T0]] to i16* |
| 46 | // CHECK-NEXT: store i16* [[T1]], i16** [[REF]], align 8 |
| 47 | array_t &ref = *(array_t*) array; |
| 48 | |
| 49 | // CHECK-NEXT: [[T0:%.*]] = load i16*, i16** [[REF]] |
| 50 | // CHECK-NEXT: [[T1:%.*]] = mul nsw i64 1, [[DIM1]] |
| 51 | // CHECK-NEXT: [[T2:%.*]] = getelementptr inbounds i16, i16* [[T0]], i64 [[T1]] |
| 52 | // CHECK-NEXT: [[T3:%.*]] = getelementptr inbounds i16, i16* [[T2]], i64 2 |
| 53 | // CHECK-NEXT: store i16 3, i16* [[T3]] |
| 54 | ref[1][2] = 3; |
| 55 | |
| 56 | // CHECK-NEXT: [[T0:%.*]] = load i16*, i16** [[REF]] |
| 57 | // CHECK-NEXT: [[T1:%.*]] = mul nsw i64 4, [[DIM1]] |
| 58 | // CHECK-NEXT: [[T2:%.*]] = getelementptr inbounds i16, i16* [[T0]], i64 [[T1]] |
| 59 | // CHECK-NEXT: [[T3:%.*]] = getelementptr inbounds i16, i16* [[T2]], i64 5 |
| 60 | // CHECK-NEXT: [[T4:%.*]] = load i16, i16* [[T3]] |
| 61 | // CHECK-NEXT: store i16 [[T4]], i16* [[S]], align 2 |
| 62 | short s = ref[4][5]; |
| 63 | |
| 64 | // CHECK-NEXT: ret void |
| 65 | } |
| 66 | |
| 67 | |
| 68 | void test2(int b) { |
| 69 | // CHECK-LABEL: define void {{.*}}test2{{.*}}(i32 %b) |
| 70 | int varr[b]; |
| 71 | // AMDGCN: %__end1 = alloca i32*, align 8, addrspace(5) |
| 72 | // AMDGCN: [[END:%.*]] = addrspacecast i32* addrspace(5)* %__end1 to i32** |
| 73 | // get the address of %b by checking the first store that stores it |
| 74 | //CHECK: store i32 %b, i32* [[PTR_B:%.*]] |
| 75 | |
| 76 | // get the size of the VLA by getting the first load of the PTR_B |
| 77 | //CHECK: [[VLA_NUM_ELEMENTS_PREZEXT:%.*]] = load i32, i32* [[PTR_B]] |
| 78 | //CHECK-NEXT: [[VLA_NUM_ELEMENTS_PRE:%.*]] = zext i32 [[VLA_NUM_ELEMENTS_PREZEXT]] |
| 79 | |
| 80 | b = 15; |
| 81 | //CHECK: store i32 15, i32* [[PTR_B]] |
| 82 | |
| 83 | // Now get the sizeof, and then divide by the element size |
| 84 | |
| 85 | |
| 86 | //CHECK: [[VLA_SIZEOF:%.*]] = mul nuw i64 4, [[VLA_NUM_ELEMENTS_PRE]] |
| 87 | //CHECK-NEXT: [[VLA_NUM_ELEMENTS_POST:%.*]] = udiv i64 [[VLA_SIZEOF]], 4 |
| 88 | //CHECK-NEXT: [[VLA_END_PTR:%.*]] = getelementptr inbounds i32, i32* {{%.*}}, i64 [[VLA_NUM_ELEMENTS_POST]] |
| 89 | //X64-NEXT: store i32* [[VLA_END_PTR]], i32** %__end1 |
| 90 | //AMDGCN-NEXT: store i32* [[VLA_END_PTR]], i32** [[END]] |
| 91 | for (int d : varr) 0; |
| 92 | } |
| 93 | |
| 94 | void test3(int b, int c) { |
| 95 | // CHECK-LABEL: define void {{.*}}test3{{.*}}(i32 %b, i32 %c) |
| 96 | int varr[b][c]; |
| 97 | // AMDGCN: %__end1 = alloca i32*, align 8, addrspace(5) |
| 98 | // AMDGCN: [[END:%.*]] = addrspacecast i32* addrspace(5)* %__end1 to i32** |
| 99 | // get the address of %b by checking the first store that stores it |
| 100 | //CHECK: store i32 %b, i32* [[PTR_B:%.*]] |
| 101 | //CHECK-NEXT: store i32 %c, i32* [[PTR_C:%.*]] |
| 102 | |
| 103 | // get the size of the VLA by getting the first load of the PTR_B |
| 104 | //CHECK: [[VLA_DIM1_PREZEXT:%.*]] = load i32, i32* [[PTR_B]] |
| 105 | //CHECK-NEXT: [[VLA_DIM1_PRE:%.*]] = zext i32 [[VLA_DIM1_PREZEXT]] |
| 106 | //CHECK: [[VLA_DIM2_PREZEXT:%.*]] = load i32, i32* [[PTR_C]] |
| 107 | //CHECK-NEXT: [[VLA_DIM2_PRE:%.*]] = zext i32 [[VLA_DIM2_PREZEXT]] |
| 108 | |
| 109 | b = 15; |
| 110 | c = 15; |
| 111 | //CHECK: store i32 15, i32* [[PTR_B]] |
| 112 | //CHECK: store i32 15, i32* [[PTR_C]] |
| 113 | // Now get the sizeof, and then divide by the element size |
| 114 | |
| 115 | // multiply the two dimensions, then by the element type and then divide by the sizeof dim2 |
| 116 | //CHECK: [[VLA_DIM1_X_DIM2:%.*]] = mul nuw i64 [[VLA_DIM1_PRE]], [[VLA_DIM2_PRE]] |
| 117 | //CHECK-NEXT: [[VLA_SIZEOF:%.*]] = mul nuw i64 4, [[VLA_DIM1_X_DIM2]] |
| 118 | //CHECK-NEXT: [[VLA_SIZEOF_DIM2:%.*]] = mul nuw i64 4, [[VLA_DIM2_PRE]] |
| 119 | //CHECK-NEXT: [[VLA_NUM_ELEMENTS:%.*]] = udiv i64 [[VLA_SIZEOF]], [[VLA_SIZEOF_DIM2]] |
| 120 | //CHECK-NEXT: [[VLA_END_INDEX:%.*]] = mul nsw i64 [[VLA_NUM_ELEMENTS]], [[VLA_DIM2_PRE]] |
| 121 | //CHECK-NEXT: [[VLA_END_PTR:%.*]] = getelementptr inbounds i32, i32* {{%.*}}, i64 [[VLA_END_INDEX]] |
| 122 | //X64-NEXT: store i32* [[VLA_END_PTR]], i32** %__end |
| 123 | //AMDGCN-NEXT: store i32* [[VLA_END_PTR]], i32** [[END]] |
| 124 | |
| 125 | for (auto &d : varr) 0; |
| 126 | } |
| 127 | |
| 128 | |
| 129 | |