xref: /llvm-project/mlir/test/Target/LLVMIR/openmp-reduction-sections.mlir (revision 6b3ba6677d64e394b9c929ea0d1f7c54e3146fda)
1// RUN: mlir-translate -mlir-to-llvmir %s | FileCheck %s
2
3omp.declare_reduction @add_reduction_f32 : f32 init {
4^bb0(%arg0: f32):
5  %0 = llvm.mlir.constant(0.000000e+00 : f32) : f32
6  omp.yield(%0 : f32)
7} combiner {
8^bb0(%arg0: f32, %arg1: f32):
9  %0 = llvm.fadd %arg0, %arg1  {fastmathFlags = #llvm.fastmath<contract>} : f32
10  omp.yield(%0 : f32)
11}
12llvm.func @sections_(%arg0: !llvm.ptr {fir.bindc_name = "x"}) attributes {fir.internal_name = "_QPsections"} {
13  %0 = llvm.mlir.constant(2.000000e+00 : f32) : f32
14  %1 = llvm.mlir.constant(1.000000e+00 : f32) : f32
15  omp.parallel {
16    omp.sections reduction(@add_reduction_f32 %arg0 -> %arg1 : !llvm.ptr) {
17      omp.section {
18      ^bb0(%arg2: !llvm.ptr):
19        %2 = llvm.load %arg2 : !llvm.ptr -> f32
20        %3 = llvm.fadd %2, %1  {fastmathFlags = #llvm.fastmath<contract>} : f32
21        llvm.store %3, %arg2 : f32, !llvm.ptr
22        omp.terminator
23      }
24      omp.section {
25      ^bb0(%arg2: !llvm.ptr):
26        %2 = llvm.load %arg2 : !llvm.ptr -> f32
27        %3 = llvm.fadd %2, %0  {fastmathFlags = #llvm.fastmath<contract>} : f32
28        llvm.store %3, %arg2 : f32, !llvm.ptr
29        omp.terminator
30      }
31      omp.terminator
32    }
33    omp.terminator
34  }
35  llvm.return
36}
37
38// CHECK-LABEL: define internal void @sections_..omp_par
39// CHECK:       [[PAR_ENTRY:omp.par.entry]]:
40// CHECK:         %[[VAL_9:.*]] = getelementptr { ptr }, ptr %[[VAL_10:.*]], i32 0, i32 0
41// CHECK:         %[[VAL_11:.*]] = load ptr, ptr %[[VAL_9]], align 8
42// CHECK:         %[[VAL_12:.*]] = alloca i32, align 4
43// CHECK:         %[[VAL_13:.*]] = alloca i32, align 4
44// CHECK:         %[[VAL_14:.*]] = alloca i32, align 4
45// CHECK:         %[[VAL_15:.*]] = alloca i32, align 4
46// CHECK:         %[[VAL_16:.*]] = alloca i32, align 4
47// CHECK:         %[[VAL_17:.*]] = load i32, ptr %[[VAL_18:.*]], align 4
48// CHECK:         store i32 %[[VAL_17]], ptr %[[VAL_16]], align 4
49// CHECK:         %[[VAL_19:.*]] = load i32, ptr %[[VAL_16]], align 4
50// CHECK:         %[[VAL_20:.*]] = alloca float, align 4
51// CHECK:         %[[VAL_21:.*]] = alloca [1 x ptr], align 8
52// CHECK:         br label %[[VAL_22:.*]]
53
54// CHECK:       [[VAL_22]]:
55// CHECK:         br label %[[PAR_REG:omp.par.region]]
56
57// CHECK:       [[PAR_REG]]:                                   ; preds = %[[VAL_22]]
58// CHECK:         br label %[[VAL_25:.*]]
59// CHECK:       omp.par.region1:                                  ; preds = %[[PAR_REG]]
60// CHECK:         br label %[[VAL_26:.*]]
61
62// CHECK:       [[RED_INIT:omp.reduction.init]]:
63// CHECK:         store float 0.000000e+00, ptr %[[VAL_20]], align 4
64// CHECK:         br label %[[VAL_24:.*]]
65
66// CHECK:       omp_section_loop.preheader:                       ; preds = %[[RED_INIT]]
67// CHECK:         store i32 0, ptr %[[VAL_13]], align 4
68// CHECK:         store i32 1, ptr %[[VAL_14]], align 4
69// CHECK:         store i32 1, ptr %[[VAL_15]], align 4
70// CHECK:         %[[VAL_27:.*]] = call i32 @__kmpc_global_thread_num(ptr @1)
71// CHECK:         call void @__kmpc_for_static_init_4u(ptr @1, i32 %[[VAL_27]], i32 34, ptr %[[VAL_12]], ptr %[[VAL_13]], ptr %[[VAL_14]], ptr %[[VAL_15]], i32 1, i32 0)
72// CHECK:         %[[VAL_28:.*]] = load i32, ptr %[[VAL_13]], align 4
73// CHECK:         %[[VAL_29:.*]] = load i32, ptr %[[VAL_14]], align 4
74// CHECK:         %[[VAL_30:.*]] = sub i32 %[[VAL_29]], %[[VAL_28]]
75// CHECK:         %[[VAL_31:.*]] = add i32 %[[VAL_30]], 1
76// CHECK:         br label %[[VAL_32:.*]]
77// CHECK:       omp_section_loop.header:                          ; preds = %[[VAL_33:.*]], %[[VAL_24]]
78// CHECK:         %[[VAL_34:.*]] = phi i32 [ 0, %[[VAL_24]] ], [ %[[VAL_35:.*]], %[[VAL_33]] ]
79// CHECK:         br label %[[VAL_36:.*]]
80// CHECK:       omp_section_loop.cond:                            ; preds = %[[VAL_32]]
81// CHECK:         %[[VAL_37:.*]] = icmp ult i32 %[[VAL_34]], %[[VAL_31]]
82// CHECK:         br i1 %[[VAL_37]], label %[[VAL_38:.*]], label %[[VAL_39:.*]]
83// CHECK:       omp_section_loop.exit:                            ; preds = %[[VAL_36]]
84// CHECK:         call void @__kmpc_for_static_fini(ptr @1, i32 %[[VAL_27]])
85// CHECK:         %[[VAL_40:.*]] = call i32 @__kmpc_global_thread_num(ptr @1)
86// CHECK:         call void @__kmpc_barrier(ptr @2, i32 %[[VAL_40]])
87// CHECK:         br label %[[VAL_41:.*]]
88// CHECK:       omp_section_loop.after:                           ; preds = %[[VAL_39]]
89// CHECK:         br label %[[VAL_42:.*]]
90// CHECK:       omp_section_loop.aftersections.fini:              ; preds = %[[VAL_41]]
91// CHECK:         %[[VAL_43:.*]] = getelementptr inbounds [1 x ptr], ptr %[[VAL_21]], i64 0, i64 0
92// CHECK:         store ptr %[[VAL_20]], ptr %[[VAL_43]], align 8
93// CHECK:         %[[VAL_44:.*]] = call i32 @__kmpc_global_thread_num(ptr @1)
94// CHECK:         %[[VAL_45:.*]] = call i32 @__kmpc_reduce(ptr @1, i32 %[[VAL_44]], i32 1, i64 8, ptr %[[VAL_21]], ptr @.omp.reduction.func, ptr @.gomp_critical_user_.reduction.var)
95// CHECK:         switch i32 %[[VAL_45]], label %[[VAL_46:.*]] [
96// CHECK:           i32 1, label %[[VAL_47:.*]]
97// CHECK:           i32 2, label %[[VAL_48:.*]]
98// CHECK:         ]
99// CHECK:       reduce.switch.atomic:                             ; preds = %[[VAL_42]]
100// CHECK:         unreachable
101// CHECK:       reduce.switch.nonatomic:                          ; preds = %[[VAL_42]]
102// CHECK:         %[[VAL_49:.*]] = load float, ptr %[[VAL_11]], align 4
103// CHECK:         %[[VAL_50:.*]] = load float, ptr %[[VAL_20]], align 4
104// CHECK:         %[[VAL_51:.*]] = fadd contract float %[[VAL_49]], %[[VAL_50]]
105// CHECK:         store float %[[VAL_51]], ptr %[[VAL_11]], align 4
106// CHECK:         call void @__kmpc_end_reduce(ptr @1, i32 %[[VAL_44]], ptr @.gomp_critical_user_.reduction.var)
107// CHECK:         br label %[[VAL_46]]
108// CHECK:       reduce.finalize:                                  ; preds = %[[VAL_47]], %[[VAL_42]]
109// CHECK:         %[[VAL_52:.*]] = call i32 @__kmpc_global_thread_num(ptr @1)
110// CHECK:         call void @__kmpc_barrier(ptr @2, i32 %[[VAL_52]])
111// CHECK:         br label %[[VAL_53:.*]]
112// CHECK:       omp.region.cont:                                  ; preds = %[[VAL_46]]
113// CHECK:         br label %[[VAL_54:.*]]
114// CHECK:       omp.par.pre_finalize:                             ; preds = %[[VAL_53]]
115// CHECK:         br label %[[VAL_55:.*]]
116// CHECK:       omp_section_loop.body:                            ; preds = %[[VAL_36]]
117// CHECK:         %[[VAL_56:.*]] = add i32 %[[VAL_34]], %[[VAL_28]]
118// CHECK:         %[[VAL_57:.*]] = mul i32 %[[VAL_56]], 1
119// CHECK:         %[[VAL_58:.*]] = add i32 %[[VAL_57]], 0
120// CHECK:         switch i32 %[[VAL_58]], label %[[VAL_59:.*]] [
121// CHECK:           i32 0, label %[[VAL_60:.*]]
122// CHECK:           i32 1, label %[[VAL_61:.*]]
123// CHECK:         ]
124// CHECK:       omp_section_loop.body.case3:                      ; preds = %[[VAL_38]]
125// CHECK:         br label %[[VAL_62:.*]]
126// CHECK:       omp.section.region5:                              ; preds = %[[VAL_61]]
127// CHECK:         %[[VAL_63:.*]] = load float, ptr %[[VAL_20]], align 4
128// CHECK:         %[[VAL_64:.*]] = fadd contract float %[[VAL_63]], 2.000000e+00
129// CHECK:         store float %[[VAL_64]], ptr %[[VAL_20]], align 4
130// CHECK:         br label %[[VAL_65:.*]]
131// CHECK:       omp.region.cont4:                                 ; preds = %[[VAL_62]]
132// CHECK:         br label %[[VAL_59]]
133// CHECK:       omp_section_loop.body.case:                       ; preds = %[[VAL_38]]
134// CHECK:         br label %[[VAL_66:.*]]
135// CHECK:       omp.section.region:                               ; preds = %[[VAL_60]]
136// CHECK:         %[[VAL_67:.*]] = load float, ptr %[[VAL_20]], align 4
137// CHECK:         %[[VAL_68:.*]] = fadd contract float %[[VAL_67]], 1.000000e+00
138// CHECK:         store float %[[VAL_68]], ptr %[[VAL_20]], align 4
139// CHECK:         br label %[[VAL_69:.*]]
140// CHECK:       omp.region.cont2:                                 ; preds = %[[VAL_66]]
141// CHECK:         br label %[[VAL_59]]
142// CHECK:       omp_section_loop.body.sections.after:             ; preds = %[[VAL_65]], %[[VAL_69]], %[[VAL_38]]
143// CHECK:         br label %[[VAL_33]]
144// CHECK:       omp_section_loop.inc:                             ; preds = %[[VAL_59]]
145// CHECK:         %[[VAL_35]] = add nuw i32 %[[VAL_34]], 1
146// CHECK:         br label %[[VAL_32]]
147// CHECK:       omp.par.outlined.exit.exitStub:                   ; preds = %[[VAL_54]]
148// CHECK:         ret void
149// CHECK:         %[[VAL_70:.*]] = getelementptr inbounds [1 x ptr], ptr %[[VAL_71:.*]], i64 0, i64 0
150// CHECK:         %[[VAL_72:.*]] = load ptr, ptr %[[VAL_70]], align 8
151// CHECK:         %[[VAL_73:.*]] = load float, ptr %[[VAL_72]], align 4
152// CHECK:         %[[VAL_74:.*]] = getelementptr inbounds [1 x ptr], ptr %[[VAL_75:.*]], i64 0, i64 0
153// CHECK:         %[[VAL_76:.*]] = load ptr, ptr %[[VAL_74]], align 8
154// CHECK:         %[[VAL_77:.*]] = load float, ptr %[[VAL_76]], align 4
155// CHECK:         %[[VAL_78:.*]] = fadd contract float %[[VAL_73]], %[[VAL_77]]
156// CHECK:         store float %[[VAL_78]], ptr %[[VAL_72]], align 4
157// CHECK:         ret void
158