80 lines · plain
1// RUN: mlir-translate -mlir-to-llvmir %s | FileCheck %s2 3// Only check the overall shape of the code and the presence of relevant4// runtime calls. Actual IR checking is done at the OpenMPIRBuilder level.5 6module attributes {dlti.dl_spec = #dlti.dl_spec<#dlti.dl_entry<"dlti.alloca_memory_space", 5 : ui32>>, llvm.data_layout = "e-p:64:64-p1:64:64-p2:32:32-p3:32:32-p4:64:64-p5:32:32-p6:32:32-p7:160:256:256:32-p8:128:128:128:48-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64-S32-A5-G1-ni:7:8", llvm.target_triple = "amdgcn-amd-amdhsa", omp.is_gpu = true, omp.is_target_device = true } {7 omp.declare_reduction @add_reduction_i32 : i32 init {8 ^bb0(%arg0: i32):9 %0 = llvm.mlir.constant(0 : i32) : i3210 omp.yield(%0 : i32)11 } combiner {12 ^bb0(%arg0: i32, %arg1: i32):13 %0 = llvm.add %arg0, %arg1 : i3214 omp.yield(%0 : i32)15 }16 llvm.func @simple_target_teams_only_reduction_() attributes {fir.internal_name = "_QPsimple_target_teams_only_reduction", frame_pointer = #llvm.framePointerKind<all>, omp.declare_target = #omp.declaretarget<device_type = (host), capture_clause = (to)>, target_cpu = "gfx1030", target_features = #llvm.target_features<["+16-bit-insts", "+ci-insts", "+dl-insts", "+dot1-insts", "+dot10-insts", "+dot2-insts", "+dot5-insts", "+dot6-insts", "+dot7-insts", "+dpp", "+gfx10-3-insts", "+gfx10-insts", "+gfx8-insts", "+gfx9-insts", "+gws", "+image-insts", "+s-memrealtime", "+s-memtime-inst", "+wavefrontsize32"]>} {17 %0 = llvm.mlir.constant(1 : i64) : i6418 %1 = llvm.alloca %0 x i32 {bindc_name = "sum"} : (i64) -> !llvm.ptr<5>19 %2 = llvm.addrspacecast %1 : !llvm.ptr<5> to !llvm.ptr20 %3 = llvm.mlir.constant(1 : i64) : i6421 %4 = llvm.alloca %3 x i32 {bindc_name = "index_"} : (i64) -> !llvm.ptr<5>22 %5 = llvm.addrspacecast %4 : !llvm.ptr<5> to !llvm.ptr23 %6 = llvm.mlir.constant(0 : i32) : i3224 %7 = llvm.mlir.constant(1 : i64) : i6425 %8 = llvm.mlir.constant(1 : i64) : i6426 llvm.store %6, %2 : i32, !llvm.ptr27 %9 = omp.map.info var_ptr(%2 : !llvm.ptr, i32) map_clauses(tofrom) capture(ByRef) -> !llvm.ptr {name = "sum"}28 %10 = omp.map.info var_ptr(%5 : !llvm.ptr, i32) map_clauses(implicit, exit_release_or_enter_alloc) capture(ByCopy) -> !llvm.ptr {name = "index_"}29 omp.target map_entries(%9 -> %arg0, %10 -> %arg1 : !llvm.ptr, !llvm.ptr) {30 %11 = llvm.mlir.constant(0 : index) : i6431 %12 = llvm.mlir.constant(10000 : index) : i6432 %13 = llvm.mlir.constant(1 : index) : i6433 omp.teams reduction(@add_reduction_i32 %arg0 -> %arg2 : !llvm.ptr) {34 %14 = llvm.trunc %13 : i64 to i3235 llvm.br ^bb1(%14, %12 : i32, i64)36 ^bb1(%15: i32, %16: i64): // 2 preds: ^bb0, ^bb237 %17 = llvm.icmp "sgt" %16, %11 : i6438 llvm.cond_br %17, ^bb2, ^bb339 ^bb2: // pred: ^bb140 llvm.store %15, %arg1 : i32, !llvm.ptr41 %18 = llvm.load %arg2 : !llvm.ptr -> i3242 %19 = llvm.load %arg1 : !llvm.ptr -> i3243 %20 = llvm.add %18, %19 : i3244 llvm.store %20, %arg2 : i32, !llvm.ptr45 %21 = llvm.load %arg1 : !llvm.ptr -> i3246 %22 = llvm.add %21, %14 overflow<nsw> : i3247 %23 = llvm.sub %16, %13 : i6448 llvm.br ^bb1(%22, %23 : i32, i64)49 ^bb3: // pred: ^bb150 llvm.store %15, %arg1 : i32, !llvm.ptr51 omp.terminator52 }53 omp.terminator54 }55 llvm.return56 }57}58 59// CHECK: call i32 @__kmpc_target_init60// CHECK: call void @[[OUTLINED:__omp_offloading_[A-Za-z0-9_.]*]]61// CHECK: %[[MASTER:.+]] = call i32 @__kmpc_nvptx_teams_reduce_nowait_v262// CHECK: icmp eq i32 %[[MASTER]], 163// CHECK: i1 %{{.+}}, label %[[THEN:[A-Za-z0-9_.]*]], label %[[DONE:[A-Za-z0-9_.]*]]64// CHECK: [[THEN]]:65// CHECK-NEXT: %[[FINAL_LHS:[A-Za-z0-9_.]*]] = load i3266// CHECK-NEXT: %[[FINAL_RHS:[A-Za-z0-9_.]*]] = load i3267// CHECK-NEXT: %[[FINAL_RESULT:[A-Za-z0-9_.]*]] = add i32 %[[FINAL_LHS]], %[[FINAL_RHS]]68// CHECK-NEXT: store i32 %[[FINAL_RESULT]]69 70// CHECK: call void @__kmpc_barrier71// CHECK: call void @__kmpc_target_deinit72 73// CHECK: define internal void @[[OUTLINED]]74// Skip to the loop75// CHECK: br i176// CHECK: %[[TEAM_RHS:[A-Za-z0-9_.]*]] = load i3277// CHECK-NEXT: %[[TEAM_LHS:[A-Za-z0-9_.]*]] = load i3278// CHECK-NEXT: %[[TEAM_RESULT:[A-Za-z0-9_.]*]] = add i32 %[[TEAM_RHS]], %[[TEAM_LHS]]79// CHECK-NEXT: store i32 %[[TEAM_RESULT]]80