| blob | 71fecd0fa5fd0a78a75c6b3e86e391b86d4988c5 |
1 // RUN: mlir-translate -mlir-to-llvmir %s | FileCheck %s
3 module attributes {omp.is_target_device = false, omp.target_triples = ["amdgcn-amd-amdhsa"]} {
4 llvm.func @_QQmain() attributes {fir.bindc_name = "main"} {
5 %0 = llvm.mlir.constant(39 : index) : i64
6 %1 = llvm.mlir.constant(0 : index) : i64
7 %2 = llvm.mlir.constant(1 : index) : i64
8 %3 = llvm.mlir.constant(40 : index) : i64
9 %4 = llvm.mlir.addressof @_QFEa : !llvm.ptr
10 %5 = llvm.mlir.addressof @_QFEb : !llvm.ptr
11 %6 = llvm.mlir.constant(1 : i64) : i64
12 %7 = llvm.alloca %6 x i32 {bindc_name = "i"} : (i64) -> !llvm.ptr
13 %8 = llvm.mlir.addressof @_QFEn : !llvm.ptr
14 omp.task {
15 %14 = llvm.mlir.constant(1 : i64) : i64
16 %15 = llvm.alloca %14 x i32 {bindc_name = "i", pinned} : (i64) -> !llvm.ptr
17 %16 = llvm.load %8 : !llvm.ptr -> i32
18 %17 = llvm.sext %16 : i32 to i64
19 %18 = llvm.trunc %2 : i64 to i32
20 llvm.br ^bb1(%18, %17 : i32, i64)
21 ^bb1(%19: i32, %20: i64): // 2 preds: ^bb0, ^bb2
22 %21 = llvm.icmp "sgt" %20, %1 : i64
23 llvm.cond_br %21, ^bb2, ^bb3
24 ^bb2: // pred: ^bb1
25 llvm.store %19, %15 : i32, !llvm.ptr
26 %22 = llvm.load %15 : !llvm.ptr -> i32
27 %23 = llvm.sext %22 : i32 to i64
28 %24 = llvm.mlir.constant(1 : i64) : i64
29 %25 = llvm.mlir.constant(0 : i64) : i64
30 %26 = llvm.sub %23, %24 overflow<nsw> : i64
31 %27 = llvm.mul %26, %24 overflow<nsw> : i64
32 %28 = llvm.mul %27, %24 overflow<nsw> : i64
33 %29 = llvm.add %28, %25 overflow<nsw> : i64
34 %30 = llvm.mul %24, %3 overflow<nsw> : i64
35 %31 = llvm.getelementptr %4[%29] : (!llvm.ptr, i64) -> !llvm.ptr, i32
36 llvm.store %22, %31 : i32, !llvm.ptr
37 %32 = llvm.load %15 : !llvm.ptr -> i32
38 %33 = llvm.add %32, %18 : i32
39 %34 = llvm.sub %20, %2 : i64
40 llvm.br ^bb1(%33, %34 : i32, i64)
41 ^bb3: // pred: ^bb1
42 llvm.store %19, %15 : i32, !llvm.ptr
43 omp.terminator
44 }
45 %9 = omp.map.bounds lower_bound(%1 : i64) upper_bound(%0 : i64) extent(%3 : i64) stride(%2 : i64) start_idx(%2 : i64)
46 %10 = omp.map.info var_ptr(%4 : !llvm.ptr, !llvm.array<40 x i32>) map_clauses(to) capture(ByRef) bounds(%9) -> !llvm.ptr {name = "a"}
47 %11 = omp.map.info var_ptr(%5 : !llvm.ptr, !llvm.array<40 x i32>) map_clauses(from) capture(ByRef) bounds(%9) -> !llvm.ptr {name = "b"}
48 %12 = omp.map.info var_ptr(%7 : !llvm.ptr, i32) map_clauses(implicit, exit_release_or_enter_alloc) capture(ByCopy) -> !llvm.ptr {name = "i"}
49 %13 = omp.map.info var_ptr(%8 : !llvm.ptr, i32) map_clauses(implicit, exit_release_or_enter_alloc) capture(ByCopy) -> !llvm.ptr {name = "n"}
50 omp.target depend(taskdependin -> %4 : !llvm.ptr) map_entries(%10 -> %arg0, %11 -> %arg1, %12 -> %arg2, %13 -> %arg3 : !llvm.ptr, !llvm.ptr, !llvm.ptr, !llvm.ptr) {
51 %14 = llvm.mlir.constant(0 : index) : i64
52 %15 = llvm.mlir.constant(10 : i32) : i32
53 %16 = llvm.mlir.constant(1 : index) : i64
54 %17 = llvm.mlir.constant(40 : index) : i64
55 %18 = llvm.load %arg3 : !llvm.ptr -> i32
56 %19 = llvm.sext %18 : i32 to i64
57 %20 = llvm.trunc %16 : i64 to i32
58 llvm.br ^bb1(%20, %19 : i32, i64)
59 ^bb1(%21: i32, %22: i64): // 2 preds: ^bb0, ^bb2
60 %23 = llvm.icmp "sgt" %22, %14 : i64
61 llvm.cond_br %23, ^bb2, ^bb3
62 ^bb2: // pred: ^bb1
63 llvm.store %21, %arg2 : i32, !llvm.ptr
64 %24 = llvm.load %arg2 : !llvm.ptr -> i32
65 %25 = llvm.sext %24 : i32 to i64
66 %26 = llvm.mlir.constant(1 : i64) : i64
67 %27 = llvm.mlir.constant(0 : i64) : i64
68 %28 = llvm.sub %25, %26 overflow<nsw> : i64
69 %29 = llvm.mul %28, %26 overflow<nsw> : i64
70 %30 = llvm.mul %29, %26 overflow<nsw> : i64
71 %31 = llvm.add %30, %27 overflow<nsw> : i64
72 %32 = llvm.mul %26, %17 overflow<nsw> : i64
73 %33 = llvm.getelementptr %arg0[%31] : (!llvm.ptr, i64) -> !llvm.ptr, i32
74 %34 = llvm.load %33 : !llvm.ptr -> i32
75 %35 = llvm.add %34, %15 : i32
76 %36 = llvm.mlir.constant(1 : i64) : i64
77 %37 = llvm.mlir.constant(0 : i64) : i64
78 %38 = llvm.sub %25, %36 overflow<nsw> : i64
79 %39 = llvm.mul %38, %36 overflow<nsw> : i64
80 %40 = llvm.mul %39, %36 overflow<nsw> : i64
81 %41 = llvm.add %40, %37 overflow<nsw> : i64
82 %42 = llvm.mul %36, %17 overflow<nsw> : i64
83 %43 = llvm.getelementptr %arg1[%41] : (!llvm.ptr, i64) -> !llvm.ptr, i32
84 llvm.store %35, %43 : i32, !llvm.ptr
85 %44 = llvm.load %arg2 : !llvm.ptr -> i32
86 %45 = llvm.add %44, %20 : i32
87 %46 = llvm.sub %22, %16 : i64
88 llvm.br ^bb1(%45, %46 : i32, i64)
89 ^bb3: // pred: ^bb1
90 llvm.store %21, %arg2 : i32, !llvm.ptr
91 omp.terminator
92 }
93 llvm.return
94 }
95 llvm.mlir.global internal @_QFEa() {addr_space = 0 : i32} : !llvm.array<40 x i32> {
96 %0 = llvm.mlir.zero : !llvm.array<40 x i32>
97 llvm.return %0 : !llvm.array<40 x i32>
98 }
99 llvm.mlir.global internal @_QFEb() {addr_space = 0 : i32} : !llvm.array<40 x i32> {
100 %0 = llvm.mlir.zero : !llvm.array<40 x i32>
101 llvm.return %0 : !llvm.array<40 x i32>
102 }
103 llvm.mlir.global internal @_QFEc() {addr_space = 0 : i32} : !llvm.array<40 x i32> {
104 %0 = llvm.mlir.zero : !llvm.array<40 x i32>
105 llvm.return %0 : !llvm.array<40 x i32>
106 }
107 llvm.mlir.global internal @_QFEn() {addr_space = 0 : i32} : i32 {
108 %0 = llvm.mlir.constant(40 : i32) : i32
109 llvm.return %0 : i32
110 }
111 llvm.func @_FortranAProgramStart(i32, !llvm.ptr, !llvm.ptr, !llvm.ptr) attributes {sym_visibility = "private"}
112 llvm.func @_FortranAProgramEndStatement() attributes {sym_visibility = "private"}
113 llvm.func @main(%arg0: i32, %arg1: !llvm.ptr, %arg2: !llvm.ptr) -> i32 {
114 %0 = llvm.mlir.constant(0 : i32) : i32
115 %1 = llvm.mlir.zero : !llvm.ptr
116 llvm.call @_FortranAProgramStart(%arg0, %arg1, %arg2, %1) {fastmathFlags = #llvm.fastmath<contract>} : (i32, !llvm.ptr, !llvm.ptr, !llvm.ptr) -> ()
117 llvm.call @_QQmain() {fastmathFlags = #llvm.fastmath<contract>} : () -> ()
118 llvm.call @_FortranAProgramEndStatement() {fastmathFlags = #llvm.fastmath<contract>} : () -> ()
119 llvm.return %0 : i32
120 }
121 }
123 // %strucArg holds pointers to shared data.
124 // CHECK: define void @_QQmain() {
125 // CHECK-DAG: %[[STRUCTARG:.+]] = alloca { ptr, ptr, ptr }, align 8
126 // CHECK-DAG: %[[DEP_ARRAY:.+]] = alloca [1 x %struct.kmp_dep_info], align 8
127 // CHECK: %[[DEP_INFO:.+]] = getelementptr inbounds [1 x %struct.kmp_dep_info], ptr %[[DEP_ARRAY]], i64 0, i64 0
128 // CHECK: %[[PTR0:.+]] = getelementptr inbounds nuw %struct.kmp_dep_info, ptr %[[DEP_INFO]], i32 0, i32 0
129 // CHECK: store i64 ptrtoint (ptr @_QFEa to i64), ptr %[[PTR0]], align 4
130 // CHECK: %[[PTR1:.+]] = getelementptr inbounds nuw %struct.kmp_dep_info, ptr %[[DEP_INFO]], i32 0, i32 1
131 // CHECK: store i64 8, ptr %[[PTR1]], align 4
132 // CHECK: %[[PTR2:.+]] = getelementptr inbounds nuw %struct.kmp_dep_info, ptr %[[DEP_INFO]], i32 0, i32 2
133 // CHECK: store i8 1, ptr %[[PTR2]], align 1
135 // CHECK: %[[TASKDATA:.+]] = call ptr @__kmpc_omp_task_alloc({{.+}}, ptr @.omp_target_task_proxy_func)
136 // CHECK: %[[SHARED_DATA:.+]] = load ptr, ptr %[[TASKDATA]], align 8
137 // CHECK: call void @llvm.memcpy.p0.p0.i64(ptr align 1 %[[SHARED_DATA]], ptr align 1 %[[STRUCTARG]], i64 24, i1 false)
138 // CHECK: call void @__kmpc_omp_wait_deps({{.+}}, i32 1, ptr %[[DEP_ARRAY]], i32 0, ptr null)
139 // CHECK: call void @__kmpc_omp_task_begin_if0({{.+}}, ptr %[[TASKDATA]])
140 // CHECK: call void @.omp_target_task_proxy_func({{.+}}, ptr %[[TASKDATA]])
141 // CHECK: call void @__kmpc_omp_task_complete_if0({{.+}}, ptr %[[TASKDATA]])