Run DCE after a LoopFlatten test to reduce spurious output [nfc]
[llvm-project.git] / clang / test / CodeGenCXX / vla.cpp
blob4cf2b3b445b406122eacb55f74d1217c385b49e2
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
4 template<typename T>
5 struct S {
6 static int n;
7 };
8 template<typename T> int S<T>::n = 5;
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;
18 void test0(void *array, int n) {
19 // CHECK-LABEL: define{{.*}} void @_Z5test0Pvi(
20 // AMDGCN: [[ARRAY0:%.*]] = alloca ptr, align 8, addrspace(5)
21 // AMDGCN-NEXT: [[N0:%.*]] = alloca i32, align 4, addrspace(5)
22 // AMDGCN-NEXT: [[REF0:%.*]] = alloca ptr, align 8, addrspace(5)
23 // AMDGCN-NEXT: [[S0:%.*]] = alloca i16, align 2, addrspace(5)
24 // AMDGCN-NEXT: [[ARRAY:%.*]] = addrspacecast ptr addrspace(5) [[ARRAY0]] to ptr
25 // AMDGCN-NEXT: [[N:%.*]] = addrspacecast ptr addrspace(5) [[N0]] to ptr
26 // AMDGCN-NEXT: [[REF:%.*]] = addrspacecast ptr addrspace(5) [[REF0]] to ptr
27 // AMDGCN-NEXT: [[S:%.*]] = addrspacecast ptr addrspace(5) [[S0]] to ptr
28 // X64: [[ARRAY:%.*]] = alloca ptr, align 8
29 // X64-NEXT: [[N:%.*]] = alloca i32, align 4
30 // X64-NEXT: [[REF:%.*]] = alloca ptr, align 8
31 // X64-NEXT: [[S:%.*]] = alloca i16, align 2
32 // CHECK-NEXT: store ptr
33 // CHECK-NEXT: store i32
35 // Capture the bounds.
36 // CHECK-NEXT: [[T0:%.*]] = load i32, ptr [[N]], align 4
37 // CHECK-NEXT: [[DIM0:%.*]] = zext i32 [[T0]] to i64
38 // CHECK-NEXT: [[T0:%.*]] = load i32, ptr [[N]], align 4
39 // CHECK-NEXT: [[T1:%.*]] = add nsw i32 [[T0]], 1
40 // CHECK-NEXT: [[DIM1:%.*]] = zext i32 [[T1]] to i64
41 typedef short array_t[n][n+1];
43 // CHECK-NEXT: [[T0:%.*]] = load ptr, ptr [[ARRAY]], align 8
44 // CHECK-NEXT: store ptr [[T0]], ptr [[REF]], align 8
45 array_t &ref = *(array_t*) array;
47 // CHECK-NEXT: [[T0:%.*]] = load ptr, ptr [[REF]]
48 // CHECK-NEXT: [[T1:%.*]] = mul nsw i64 1, [[DIM1]]
49 // CHECK-NEXT: [[T2:%.*]] = getelementptr inbounds i16, ptr [[T0]], i64 [[T1]]
50 // CHECK-NEXT: [[T3:%.*]] = getelementptr inbounds i16, ptr [[T2]], i64 2
51 // CHECK-NEXT: store i16 3, ptr [[T3]]
52 ref[1][2] = 3;
54 // CHECK-NEXT: [[T0:%.*]] = load ptr, ptr [[REF]]
55 // CHECK-NEXT: [[T1:%.*]] = mul nsw i64 4, [[DIM1]]
56 // CHECK-NEXT: [[T2:%.*]] = getelementptr inbounds i16, ptr [[T0]], i64 [[T1]]
57 // CHECK-NEXT: [[T3:%.*]] = getelementptr inbounds i16, ptr [[T2]], i64 5
58 // CHECK-NEXT: [[T4:%.*]] = load i16, ptr [[T3]]
59 // CHECK-NEXT: store i16 [[T4]], ptr [[S]], align 2
60 short s = ref[4][5];
62 // CHECK-NEXT: ret void
66 void test2(int b) {
67 // CHECK-LABEL: define{{.*}} void {{.*}}test2{{.*}}(i32 noundef %b)
68 int varr[b];
69 // AMDGCN: %__end1 = alloca ptr, align 8, addrspace(5)
70 // AMDGCN: [[END:%.*]] = addrspacecast ptr addrspace(5) %__end1 to ptr
71 // get the address of %b by checking the first store that stores it
72 //CHECK: store i32 %b, ptr [[PTR_B:%.*]]
74 // get the size of the VLA by getting the first load of the PTR_B
75 //CHECK: [[VLA_NUM_ELEMENTS_PREZEXT:%.*]] = load i32, ptr [[PTR_B]]
76 //CHECK-NEXT: [[VLA_NUM_ELEMENTS_PRE:%.*]] = zext i32 [[VLA_NUM_ELEMENTS_PREZEXT]]
78 b = 15;
79 //CHECK: store i32 15, ptr [[PTR_B]]
81 // Now get the sizeof, and then divide by the element size
84 //CHECK: [[VLA_SIZEOF:%.*]] = mul nuw i64 4, [[VLA_NUM_ELEMENTS_PRE]]
85 //CHECK-NEXT: [[VLA_NUM_ELEMENTS_POST:%.*]] = udiv i64 [[VLA_SIZEOF]], 4
86 //CHECK-NEXT: [[VLA_END_PTR:%.*]] = getelementptr inbounds i32, ptr {{%.*}}, i64 [[VLA_NUM_ELEMENTS_POST]]
87 //X64-NEXT: store ptr [[VLA_END_PTR]], ptr %__end1
88 //AMDGCN-NEXT: store ptr [[VLA_END_PTR]], ptr [[END]]
89 for (int d : varr) 0;
92 void test3(int b, int c) {
93 // CHECK-LABEL: define{{.*}} void {{.*}}test3{{.*}}(i32 noundef %b, i32 noundef %c)
94 int varr[b][c];
95 // AMDGCN: %__end1 = alloca ptr, align 8, addrspace(5)
96 // AMDGCN: [[END:%.*]] = addrspacecast ptr addrspace(5) %__end1 to ptr
97 // get the address of %b by checking the first store that stores it
98 //CHECK: store i32 %b, ptr [[PTR_B:%.*]]
99 //CHECK-NEXT: store i32 %c, ptr [[PTR_C:%.*]]
101 // get the size of the VLA by getting the first load of the PTR_B
102 //CHECK: [[VLA_DIM1_PREZEXT:%.*]] = load i32, ptr [[PTR_B]]
103 //CHECK-NEXT: [[VLA_DIM1_PRE:%.*]] = zext i32 [[VLA_DIM1_PREZEXT]]
104 //CHECK: [[VLA_DIM2_PREZEXT:%.*]] = load i32, ptr [[PTR_C]]
105 //CHECK-NEXT: [[VLA_DIM2_PRE:%.*]] = zext i32 [[VLA_DIM2_PREZEXT]]
107 b = 15;
108 c = 15;
109 //CHECK: store i32 15, ptr [[PTR_B]]
110 //CHECK: store i32 15, ptr [[PTR_C]]
111 // Now get the sizeof, and then divide by the element size
113 // multiply the two dimensions, then by the element type and then divide by the sizeof dim2
114 //CHECK: [[VLA_DIM1_X_DIM2:%.*]] = mul nuw i64 [[VLA_DIM1_PRE]], [[VLA_DIM2_PRE]]
115 //CHECK-NEXT: [[VLA_SIZEOF:%.*]] = mul nuw i64 4, [[VLA_DIM1_X_DIM2]]
116 //CHECK-NEXT: [[VLA_SIZEOF_DIM2:%.*]] = mul nuw i64 4, [[VLA_DIM2_PRE]]
117 //CHECK-NEXT: [[VLA_NUM_ELEMENTS:%.*]] = udiv i64 [[VLA_SIZEOF]], [[VLA_SIZEOF_DIM2]]
118 //CHECK-NEXT: [[VLA_END_INDEX:%.*]] = mul nsw i64 [[VLA_NUM_ELEMENTS]], [[VLA_DIM2_PRE]]
119 //CHECK-NEXT: [[VLA_END_PTR:%.*]] = getelementptr inbounds i32, ptr {{%.*}}, i64 [[VLA_END_INDEX]]
120 //X64-NEXT: store ptr [[VLA_END_PTR]], ptr %__end
121 //AMDGCN-NEXT: store ptr [[VLA_END_PTR]], ptr [[END]]
123 for (auto &d : varr) 0;