1 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --include-generated-funcs --replace-function-regex "__omp_offloading_[0-9a-z]+_[0-9a-z]+" 2 // Test target codegen - host bc file has to be created first. 3 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc 4 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix CHECK1 5 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - -fopenmp-cuda-parallel-target-regions | FileCheck %s --check-prefix CHECK2 6 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm-bc %s -o %t-x86-host.bc 7 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK3 8 // RUN: %clang_cc1 -verify -fopenmp -fexceptions -fcxx-exceptions -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -fopenmp-cuda-teams-reduction-recs-num=2048 -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK4 9 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - -fopenmp-cuda-parallel-target-regions | FileCheck %s --check-prefix CHECK5 10 // RUN: %clang_cc1 -verify -fopenmp -fexceptions -fcxx-exceptions -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -fopenmp-cuda-teams-reduction-recs-num=2048 -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - -fopenmp-cuda-parallel-target-regions | FileCheck %s --check-prefix CHECK6 11 // expected-no-diagnostics 12 #ifndef HEADER 13 #define HEADER 14 15 template<typename tx> 16 tx ftemplate(int n) { 17 int a; 18 short b; 19 tx c; 20 float d; 21 double e; 22 23 #pragma omp target 24 #pragma omp teams reduction(+: e) 25 { 26 e += 5; 27 } 28 29 #pragma omp target 30 #pragma omp teams reduction(^: c) reduction(*: d) 31 { 32 c ^= 2; 33 d *= 33; 34 } 35 36 #pragma omp target 37 #pragma omp teams reduction(|: a) reduction(max: b) 38 #pragma omp parallel reduction(|: a) reduction(max: b) 39 { 40 a |= 1; 41 b = 99 > b ? 99 : b; 42 } 43 44 return a+b+c+d+e; 45 } 46 47 int bar(int n){ 48 int a = 0; 49 50 a += ftemplate<char>(n); 51 52 return a; 53 } 54 55 #endif 56 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l23_worker 57 // CHECK1-SAME: () #[[ATTR0:[0-9]+]] { 58 // CHECK1-NEXT: entry: 59 // CHECK1-NEXT: [[WORK_FN:%.*]] = alloca i8*, align 8 60 // CHECK1-NEXT: [[EXEC_STATUS:%.*]] = alloca i8, align 1 61 // CHECK1-NEXT: store i8* null, i8** [[WORK_FN]], align 8 62 // CHECK1-NEXT: store i8 0, i8* [[EXEC_STATUS]], align 1 63 // CHECK1-NEXT: br label [[DOTAWAIT_WORK:%.*]] 64 // CHECK1: .await.work: 65 // CHECK1-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 66 // CHECK1-NEXT: [[TMP0:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORK_FN]]) 67 // CHECK1-NEXT: [[TMP1:%.*]] = zext i1 [[TMP0]] to i8 68 // CHECK1-NEXT: store i8 [[TMP1]], i8* [[EXEC_STATUS]], align 1 69 // CHECK1-NEXT: [[TMP2:%.*]] = load i8*, i8** [[WORK_FN]], align 8 70 // CHECK1-NEXT: [[SHOULD_TERMINATE:%.*]] = icmp eq i8* [[TMP2]], null 71 // CHECK1-NEXT: br i1 [[SHOULD_TERMINATE]], label [[DOTEXIT:%.*]], label [[DOTSELECT_WORKERS:%.*]] 72 // CHECK1: .select.workers: 73 // CHECK1-NEXT: [[TMP3:%.*]] = load i8, i8* [[EXEC_STATUS]], align 1 74 // CHECK1-NEXT: [[IS_ACTIVE:%.*]] = icmp ne i8 [[TMP3]], 0 75 // CHECK1-NEXT: br i1 [[IS_ACTIVE]], label [[DOTEXECUTE_PARALLEL:%.*]], label [[DOTBARRIER_PARALLEL:%.*]] 76 // CHECK1: .execute.parallel: 77 // CHECK1-NEXT: [[TMP4:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]]) 78 // CHECK1-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP2]] to void (i16, i32)* 79 // CHECK1-NEXT: call void [[TMP5]](i16 0, i32 [[TMP4]]) 80 // CHECK1-NEXT: br label [[DOTTERMINATE_PARALLEL:%.*]] 81 // CHECK1: .terminate.parallel: 82 // CHECK1-NEXT: call void @__kmpc_kernel_end_parallel() 83 // CHECK1-NEXT: br label [[DOTBARRIER_PARALLEL]] 84 // CHECK1: .barrier.parallel: 85 // CHECK1-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 86 // CHECK1-NEXT: br label [[DOTAWAIT_WORK]] 87 // CHECK1: .exit: 88 // CHECK1-NEXT: ret void 89 // 90 // 91 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l23 92 // CHECK1-SAME: (i64 [[E:%.*]]) #[[ATTR1:[0-9]+]] { 93 // CHECK1-NEXT: entry: 94 // CHECK1-NEXT: [[E_ADDR:%.*]] = alloca i64, align 8 95 // CHECK1-NEXT: [[DOTZERO_ADDR:%.*]] = alloca i32, align 4 96 // CHECK1-NEXT: [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4 97 // CHECK1-NEXT: store i32 0, i32* [[DOTZERO_ADDR]], align 4 98 // CHECK1-NEXT: store i64 [[E]], i64* [[E_ADDR]], align 8 99 // CHECK1-NEXT: [[CONV:%.*]] = bitcast i64* [[E_ADDR]] to double* 100 // CHECK1-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 101 // CHECK1-NEXT: [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 102 // CHECK1-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 103 // CHECK1-NEXT: [[THREAD_LIMIT:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS]], [[NVPTX_WARP_SIZE]] 104 // CHECK1-NEXT: [[TMP0:%.*]] = icmp ult i32 [[NVPTX_TID]], [[THREAD_LIMIT]] 105 // CHECK1-NEXT: br i1 [[TMP0]], label [[DOTWORKER:%.*]], label [[DOTMASTERCHECK:%.*]] 106 // CHECK1: .worker: 107 // CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l23_worker() #[[ATTR3:[0-9]+]] 108 // CHECK1-NEXT: br label [[DOTEXIT:%.*]] 109 // CHECK1: .mastercheck: 110 // CHECK1-NEXT: [[NVPTX_TID1:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 111 // CHECK1-NEXT: [[NVPTX_NUM_THREADS2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 112 // CHECK1-NEXT: [[NVPTX_WARP_SIZE3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 113 // CHECK1-NEXT: [[TMP1:%.*]] = sub nuw i32 [[NVPTX_WARP_SIZE3]], 1 114 // CHECK1-NEXT: [[TMP2:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS2]], 1 115 // CHECK1-NEXT: [[TMP3:%.*]] = xor i32 [[TMP1]], -1 116 // CHECK1-NEXT: [[MASTER_TID:%.*]] = and i32 [[TMP2]], [[TMP3]] 117 // CHECK1-NEXT: [[TMP4:%.*]] = icmp eq i32 [[NVPTX_TID1]], [[MASTER_TID]] 118 // CHECK1-NEXT: br i1 [[TMP4]], label [[DOTMASTER:%.*]], label [[DOTEXIT]] 119 // CHECK1: .master: 120 // CHECK1-NEXT: [[NVPTX_NUM_THREADS4:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 121 // CHECK1-NEXT: [[NVPTX_WARP_SIZE5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 122 // CHECK1-NEXT: [[THREAD_LIMIT6:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS4]], [[NVPTX_WARP_SIZE5]] 123 // CHECK1-NEXT: call void @__kmpc_kernel_init(i32 [[THREAD_LIMIT6]], i16 1) 124 // CHECK1-NEXT: call void @__kmpc_data_sharing_init_stack() 125 // CHECK1-NEXT: [[TMP5:%.*]] = load i16, i16* @"_openmp_static_kernel$is_shared", align 2 126 // CHECK1-NEXT: [[TMP6:%.*]] = load i64, i64* @"_openmp_static_kernel$size", align 8 127 // CHECK1-NEXT: call void @__kmpc_get_team_static_memory(i16 0, i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds (%"union._shared_openmp_static_memory_type_$_", %"union._shared_openmp_static_memory_type_$_" addrspace(3)* @"_openmp_shared_static_glob_rd_$_", i32 0, i32 0, i32 0) to i8*), i64 [[TMP6]], i16 [[TMP5]], i8** addrspacecast (i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr" to i8**)) 128 // CHECK1-NEXT: [[TMP7:%.*]] = load i8*, i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr", align 8 129 // CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds i8, i8* [[TMP7]], i64 0 130 // CHECK1-NEXT: [[TMP9:%.*]] = bitcast i8* [[TMP8]] to %struct._globalized_locals_ty* 131 // CHECK1-NEXT: [[TMP10:%.*]] = load double, double* [[CONV]], align 8 132 // CHECK1-NEXT: [[E7:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY:%.*]], %struct._globalized_locals_ty* [[TMP9]], i32 0, i32 0 133 // CHECK1-NEXT: store double [[TMP10]], double* [[E7]], align 8 134 // CHECK1-NEXT: [[TMP11:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 135 // CHECK1-NEXT: store i32 [[TMP11]], i32* [[DOTTHREADID_TEMP_]], align 4 136 // CHECK1-NEXT: call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], double* [[E7]]) #[[ATTR3]] 137 // CHECK1-NEXT: [[TMP12:%.*]] = load i16, i16* @"_openmp_static_kernel$is_shared", align 2 138 // CHECK1-NEXT: call void @__kmpc_restore_team_static_memory(i16 0, i16 [[TMP12]]) 139 // CHECK1-NEXT: br label [[DOTTERMINATION_NOTIFIER:%.*]] 140 // CHECK1: .termination.notifier: 141 // CHECK1-NEXT: call void @__kmpc_kernel_deinit(i16 1) 142 // CHECK1-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 143 // CHECK1-NEXT: br label [[DOTEXIT]] 144 // CHECK1: .exit: 145 // CHECK1-NEXT: ret void 146 // 147 // 148 // CHECK1-LABEL: define {{[^@]+}}@__omp_outlined__ 149 // CHECK1-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], double* nonnull align 8 dereferenceable(8) [[E:%.*]]) #[[ATTR1]] { 150 // CHECK1-NEXT: entry: 151 // CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8 152 // CHECK1-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8 153 // CHECK1-NEXT: [[E_ADDR:%.*]] = alloca double*, align 8 154 // CHECK1-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x i8*], align 8 155 // CHECK1-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8 156 // CHECK1-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8 157 // CHECK1-NEXT: store double* [[E]], double** [[E_ADDR]], align 8 158 // CHECK1-NEXT: [[TMP0:%.*]] = load double*, double** [[E_ADDR]], align 8 159 // CHECK1-NEXT: [[TMP1:%.*]] = load i8*, i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr", align 8 160 // CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds i8, i8* [[TMP1]], i64 8 161 // CHECK1-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to %struct._globalized_locals_ty.0* 162 // CHECK1-NEXT: [[E1:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_0:%.*]], %struct._globalized_locals_ty.0* [[TMP3]], i32 0, i32 0 163 // CHECK1-NEXT: store double 0.000000e+00, double* [[E1]], align 8 164 // CHECK1-NEXT: [[TMP4:%.*]] = load double, double* [[E1]], align 8 165 // CHECK1-NEXT: [[ADD:%.*]] = fadd double [[TMP4]], 5.000000e+00 166 // CHECK1-NEXT: store double [[ADD]], double* [[E1]], align 8 167 // CHECK1-NEXT: [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8 168 // CHECK1-NEXT: [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4 169 // CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0 170 // CHECK1-NEXT: [[TMP8:%.*]] = bitcast double* [[E1]] to i8* 171 // CHECK1-NEXT: store i8* [[TMP8]], i8** [[TMP7]], align 8 172 // CHECK1-NEXT: [[TMP9:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 173 // CHECK1-NEXT: [[TMP10:%.*]] = load i8*, i8** @"_openmp_teams_reductions_buffer_$_$ptr", align 8 174 // CHECK1-NEXT: [[TMP11:%.*]] = call i32 @__kmpc_nvptx_teams_reduce_nowait_v2(%struct.ident_t* @[[GLOB1]], i32 [[TMP6]], i8* [[TMP10]], i32 1024, i8* [[TMP9]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_copy_func, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_reduce_func, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_copy_func, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_reduce_func) 175 // CHECK1-NEXT: [[TMP12:%.*]] = icmp eq i32 [[TMP11]], 1 176 // CHECK1-NEXT: br i1 [[TMP12]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]] 177 // CHECK1: .omp.reduction.then: 178 // CHECK1-NEXT: [[TMP13:%.*]] = load double, double* [[TMP0]], align 8 179 // CHECK1-NEXT: [[TMP14:%.*]] = load double, double* [[E1]], align 8 180 // CHECK1-NEXT: [[ADD2:%.*]] = fadd double [[TMP13]], [[TMP14]] 181 // CHECK1-NEXT: store double [[ADD2]], double* [[TMP0]], align 8 182 // CHECK1-NEXT: call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP6]]) 183 // CHECK1-NEXT: br label [[DOTOMP_REDUCTION_DONE]] 184 // CHECK1: .omp.reduction.done: 185 // CHECK1-NEXT: ret void 186 // 187 // 188 // CHECK1-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func 189 // CHECK1-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR0]] { 190 // CHECK1-NEXT: entry: 191 // CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 192 // CHECK1-NEXT: [[DOTADDR1:%.*]] = alloca i16, align 2 193 // CHECK1-NEXT: [[DOTADDR2:%.*]] = alloca i16, align 2 194 // CHECK1-NEXT: [[DOTADDR3:%.*]] = alloca i16, align 2 195 // CHECK1-NEXT: [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [1 x i8*], align 8 196 // CHECK1-NEXT: [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca double, align 8 197 // CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 198 // CHECK1-NEXT: store i16 [[TMP1]], i16* [[DOTADDR1]], align 2 199 // CHECK1-NEXT: store i16 [[TMP2]], i16* [[DOTADDR2]], align 2 200 // CHECK1-NEXT: store i16 [[TMP3]], i16* [[DOTADDR3]], align 2 201 // CHECK1-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 8 202 // CHECK1-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [1 x i8*]* 203 // CHECK1-NEXT: [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2 204 // CHECK1-NEXT: [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2 205 // CHECK1-NEXT: [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2 206 // CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i64 0, i64 0 207 // CHECK1-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8 208 // CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0 209 // CHECK1-NEXT: [[TMP12:%.*]] = bitcast i8* [[TMP10]] to double* 210 // CHECK1-NEXT: [[TMP13:%.*]] = getelementptr double, double* [[TMP12]], i64 1 211 // CHECK1-NEXT: [[TMP14:%.*]] = bitcast double* [[TMP13]] to i8* 212 // CHECK1-NEXT: [[TMP15:%.*]] = bitcast double* [[TMP12]] to i64* 213 // CHECK1-NEXT: [[TMP16:%.*]] = bitcast double* [[DOTOMP_REDUCTION_ELEMENT]] to i64* 214 // CHECK1-NEXT: [[TMP17:%.*]] = load i64, i64* [[TMP15]], align 8 215 // CHECK1-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 216 // CHECK1-NEXT: [[TMP18:%.*]] = trunc i32 [[NVPTX_WARP_SIZE]] to i16 217 // CHECK1-NEXT: [[TMP19:%.*]] = call i64 @__kmpc_shuffle_int64(i64 [[TMP17]], i16 [[TMP7]], i16 [[TMP18]]) 218 // CHECK1-NEXT: store i64 [[TMP19]], i64* [[TMP16]], align 8 219 // CHECK1-NEXT: [[TMP20:%.*]] = getelementptr i64, i64* [[TMP15]], i64 1 220 // CHECK1-NEXT: [[TMP21:%.*]] = getelementptr i64, i64* [[TMP16]], i64 1 221 // CHECK1-NEXT: [[TMP22:%.*]] = bitcast double* [[DOTOMP_REDUCTION_ELEMENT]] to i8* 222 // CHECK1-NEXT: store i8* [[TMP22]], i8** [[TMP11]], align 8 223 // CHECK1-NEXT: [[TMP23:%.*]] = icmp eq i16 [[TMP8]], 0 224 // CHECK1-NEXT: [[TMP24:%.*]] = icmp eq i16 [[TMP8]], 1 225 // CHECK1-NEXT: [[TMP25:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]] 226 // CHECK1-NEXT: [[TMP26:%.*]] = and i1 [[TMP24]], [[TMP25]] 227 // CHECK1-NEXT: [[TMP27:%.*]] = icmp eq i16 [[TMP8]], 2 228 // CHECK1-NEXT: [[TMP28:%.*]] = and i16 [[TMP6]], 1 229 // CHECK1-NEXT: [[TMP29:%.*]] = icmp eq i16 [[TMP28]], 0 230 // CHECK1-NEXT: [[TMP30:%.*]] = and i1 [[TMP27]], [[TMP29]] 231 // CHECK1-NEXT: [[TMP31:%.*]] = icmp sgt i16 [[TMP7]], 0 232 // CHECK1-NEXT: [[TMP32:%.*]] = and i1 [[TMP30]], [[TMP31]] 233 // CHECK1-NEXT: [[TMP33:%.*]] = or i1 [[TMP23]], [[TMP26]] 234 // CHECK1-NEXT: [[TMP34:%.*]] = or i1 [[TMP33]], [[TMP32]] 235 // CHECK1-NEXT: br i1 [[TMP34]], label [[THEN:%.*]], label [[ELSE:%.*]] 236 // CHECK1: then: 237 // CHECK1-NEXT: [[TMP35:%.*]] = bitcast [1 x i8*]* [[TMP5]] to i8* 238 // CHECK1-NEXT: [[TMP36:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8* 239 // CHECK1-NEXT: call void @"_omp$reduction$reduction_func"(i8* [[TMP35]], i8* [[TMP36]]) #[[ATTR3]] 240 // CHECK1-NEXT: br label [[IFCONT:%.*]] 241 // CHECK1: else: 242 // CHECK1-NEXT: br label [[IFCONT]] 243 // CHECK1: ifcont: 244 // CHECK1-NEXT: [[TMP37:%.*]] = icmp eq i16 [[TMP8]], 1 245 // CHECK1-NEXT: [[TMP38:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]] 246 // CHECK1-NEXT: [[TMP39:%.*]] = and i1 [[TMP37]], [[TMP38]] 247 // CHECK1-NEXT: br i1 [[TMP39]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 248 // CHECK1: then4: 249 // CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0 250 // CHECK1-NEXT: [[TMP41:%.*]] = load i8*, i8** [[TMP40]], align 8 251 // CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i64 0, i64 0 252 // CHECK1-NEXT: [[TMP43:%.*]] = load i8*, i8** [[TMP42]], align 8 253 // CHECK1-NEXT: [[TMP44:%.*]] = bitcast i8* [[TMP41]] to double* 254 // CHECK1-NEXT: [[TMP45:%.*]] = bitcast i8* [[TMP43]] to double* 255 // CHECK1-NEXT: [[TMP46:%.*]] = load double, double* [[TMP44]], align 8 256 // CHECK1-NEXT: store double [[TMP46]], double* [[TMP45]], align 8 257 // CHECK1-NEXT: br label [[IFCONT6:%.*]] 258 // CHECK1: else5: 259 // CHECK1-NEXT: br label [[IFCONT6]] 260 // CHECK1: ifcont6: 261 // CHECK1-NEXT: ret void 262 // 263 // 264 // CHECK1-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func 265 // CHECK1-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] { 266 // CHECK1-NEXT: entry: 267 // CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 268 // CHECK1-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 269 // CHECK1-NEXT: [[DOTCNT_ADDR:%.*]] = alloca i32, align 4 270 // CHECK1-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 271 // CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 272 // CHECK1-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 273 // CHECK1-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 274 // CHECK1-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 275 // CHECK1-NEXT: [[NVPTX_LANE_ID:%.*]] = and i32 [[NVPTX_TID2]], 31 276 // CHECK1-NEXT: [[NVPTX_TID3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 277 // CHECK1-NEXT: [[NVPTX_WARP_ID:%.*]] = ashr i32 [[NVPTX_TID3]], 5 278 // CHECK1-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8 279 // CHECK1-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [1 x i8*]* 280 // CHECK1-NEXT: store i32 0, i32* [[DOTCNT_ADDR]], align 4 281 // CHECK1-NEXT: br label [[PRECOND:%.*]] 282 // CHECK1: precond: 283 // CHECK1-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTCNT_ADDR]], align 4 284 // CHECK1-NEXT: [[TMP6:%.*]] = icmp ult i32 [[TMP5]], 2 285 // CHECK1-NEXT: br i1 [[TMP6]], label [[BODY:%.*]], label [[EXIT:%.*]] 286 // CHECK1: body: 287 // CHECK1-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP2]]) 288 // CHECK1-NEXT: [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 289 // CHECK1-NEXT: br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]] 290 // CHECK1: then: 291 // CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP4]], i64 0, i64 0 292 // CHECK1-NEXT: [[TMP8:%.*]] = load i8*, i8** [[TMP7]], align 8 293 // CHECK1-NEXT: [[TMP9:%.*]] = bitcast i8* [[TMP8]] to i32* 294 // CHECK1-NEXT: [[TMP10:%.*]] = getelementptr i32, i32* [[TMP9]], i32 [[TMP5]] 295 // CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 296 // CHECK1-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP10]], align 4 297 // CHECK1-NEXT: store volatile i32 [[TMP12]], i32 addrspace(3)* [[TMP11]], align 4 298 // CHECK1-NEXT: br label [[IFCONT:%.*]] 299 // CHECK1: else: 300 // CHECK1-NEXT: br label [[IFCONT]] 301 // CHECK1: ifcont: 302 // CHECK1-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 303 // CHECK1-NEXT: [[TMP13:%.*]] = load i32, i32* [[DOTADDR1]], align 4 304 // CHECK1-NEXT: [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP13]] 305 // CHECK1-NEXT: br i1 [[IS_ACTIVE_THREAD]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 306 // CHECK1: then4: 307 // CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 308 // CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP4]], i64 0, i64 0 309 // CHECK1-NEXT: [[TMP16:%.*]] = load i8*, i8** [[TMP15]], align 8 310 // CHECK1-NEXT: [[TMP17:%.*]] = bitcast i8* [[TMP16]] to i32* 311 // CHECK1-NEXT: [[TMP18:%.*]] = getelementptr i32, i32* [[TMP17]], i32 [[TMP5]] 312 // CHECK1-NEXT: [[TMP19:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP14]], align 4 313 // CHECK1-NEXT: store i32 [[TMP19]], i32* [[TMP18]], align 4 314 // CHECK1-NEXT: br label [[IFCONT6:%.*]] 315 // CHECK1: else5: 316 // CHECK1-NEXT: br label [[IFCONT6]] 317 // CHECK1: ifcont6: 318 // CHECK1-NEXT: [[TMP20:%.*]] = add nsw i32 [[TMP5]], 1 319 // CHECK1-NEXT: store i32 [[TMP20]], i32* [[DOTCNT_ADDR]], align 4 320 // CHECK1-NEXT: br label [[PRECOND]] 321 // CHECK1: exit: 322 // CHECK1-NEXT: ret void 323 // 324 // 325 // CHECK1-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_copy_func 326 // CHECK1-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 327 // CHECK1-NEXT: entry: 328 // CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 329 // CHECK1-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 330 // CHECK1-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 8 331 // CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 332 // CHECK1-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 333 // CHECK1-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 8 334 // CHECK1-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 8 335 // CHECK1-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [1 x i8*]* 336 // CHECK1-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 8 337 // CHECK1-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.1* 338 // CHECK1-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 339 // CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP4]], i64 0, i64 0 340 // CHECK1-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 8 341 // CHECK1-NEXT: [[TMP10:%.*]] = bitcast i8* [[TMP9]] to double* 342 // CHECK1-NEXT: [[E:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_1:%.*]], %struct._globalized_locals_ty.1* [[TMP6]], i32 0, i32 0 343 // CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [1024 x double], [1024 x double]* [[E]], i32 0, i32 [[TMP7]] 344 // CHECK1-NEXT: [[TMP12:%.*]] = load double, double* [[TMP10]], align 8 345 // CHECK1-NEXT: store double [[TMP12]], double* [[TMP11]], align 128 346 // CHECK1-NEXT: ret void 347 // 348 // 349 // CHECK1-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_reduce_func 350 // CHECK1-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 351 // CHECK1-NEXT: entry: 352 // CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 353 // CHECK1-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 354 // CHECK1-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 8 355 // CHECK1-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x i8*], align 8 356 // CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 357 // CHECK1-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 358 // CHECK1-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 8 359 // CHECK1-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8 360 // CHECK1-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.1* 361 // CHECK1-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 362 // CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0 363 // CHECK1-NEXT: [[E:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_1:%.*]], %struct._globalized_locals_ty.1* [[TMP4]], i32 0, i32 0 364 // CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1024 x double], [1024 x double]* [[E]], i32 0, i32 [[TMP5]] 365 // CHECK1-NEXT: [[TMP8:%.*]] = bitcast double* [[TMP7]] to i8* 366 // CHECK1-NEXT: store i8* [[TMP8]], i8** [[TMP6]], align 8 367 // CHECK1-NEXT: [[TMP9:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 368 // CHECK1-NEXT: [[TMP10:%.*]] = load i8*, i8** [[DOTADDR2]], align 8 369 // CHECK1-NEXT: call void @"_omp$reduction$reduction_func"(i8* [[TMP9]], i8* [[TMP10]]) #[[ATTR3]] 370 // CHECK1-NEXT: ret void 371 // 372 // 373 // CHECK1-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_copy_func 374 // CHECK1-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 375 // CHECK1-NEXT: entry: 376 // CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 377 // CHECK1-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 378 // CHECK1-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 8 379 // CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 380 // CHECK1-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 381 // CHECK1-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 8 382 // CHECK1-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 8 383 // CHECK1-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [1 x i8*]* 384 // CHECK1-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 8 385 // CHECK1-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.1* 386 // CHECK1-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 387 // CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP4]], i64 0, i64 0 388 // CHECK1-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 8 389 // CHECK1-NEXT: [[TMP10:%.*]] = bitcast i8* [[TMP9]] to double* 390 // CHECK1-NEXT: [[E:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_1:%.*]], %struct._globalized_locals_ty.1* [[TMP6]], i32 0, i32 0 391 // CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [1024 x double], [1024 x double]* [[E]], i32 0, i32 [[TMP7]] 392 // CHECK1-NEXT: [[TMP12:%.*]] = load double, double* [[TMP11]], align 128 393 // CHECK1-NEXT: store double [[TMP12]], double* [[TMP10]], align 8 394 // CHECK1-NEXT: ret void 395 // 396 // 397 // CHECK1-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_reduce_func 398 // CHECK1-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 399 // CHECK1-NEXT: entry: 400 // CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 401 // CHECK1-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 402 // CHECK1-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 8 403 // CHECK1-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x i8*], align 8 404 // CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 405 // CHECK1-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 406 // CHECK1-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 8 407 // CHECK1-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8 408 // CHECK1-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.1* 409 // CHECK1-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 410 // CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0 411 // CHECK1-NEXT: [[E:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_1:%.*]], %struct._globalized_locals_ty.1* [[TMP4]], i32 0, i32 0 412 // CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1024 x double], [1024 x double]* [[E]], i32 0, i32 [[TMP5]] 413 // CHECK1-NEXT: [[TMP8:%.*]] = bitcast double* [[TMP7]] to i8* 414 // CHECK1-NEXT: store i8* [[TMP8]], i8** [[TMP6]], align 8 415 // CHECK1-NEXT: [[TMP9:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 416 // CHECK1-NEXT: [[TMP10:%.*]] = load i8*, i8** [[DOTADDR2]], align 8 417 // CHECK1-NEXT: call void @"_omp$reduction$reduction_func"(i8* [[TMP10]], i8* [[TMP9]]) #[[ATTR3]] 418 // CHECK1-NEXT: ret void 419 // 420 // 421 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29_worker 422 // CHECK1-SAME: () #[[ATTR0]] { 423 // CHECK1-NEXT: entry: 424 // CHECK1-NEXT: [[WORK_FN:%.*]] = alloca i8*, align 8 425 // CHECK1-NEXT: [[EXEC_STATUS:%.*]] = alloca i8, align 1 426 // CHECK1-NEXT: store i8* null, i8** [[WORK_FN]], align 8 427 // CHECK1-NEXT: store i8 0, i8* [[EXEC_STATUS]], align 1 428 // CHECK1-NEXT: br label [[DOTAWAIT_WORK:%.*]] 429 // CHECK1: .await.work: 430 // CHECK1-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 431 // CHECK1-NEXT: [[TMP0:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORK_FN]]) 432 // CHECK1-NEXT: [[TMP1:%.*]] = zext i1 [[TMP0]] to i8 433 // CHECK1-NEXT: store i8 [[TMP1]], i8* [[EXEC_STATUS]], align 1 434 // CHECK1-NEXT: [[TMP2:%.*]] = load i8*, i8** [[WORK_FN]], align 8 435 // CHECK1-NEXT: [[SHOULD_TERMINATE:%.*]] = icmp eq i8* [[TMP2]], null 436 // CHECK1-NEXT: br i1 [[SHOULD_TERMINATE]], label [[DOTEXIT:%.*]], label [[DOTSELECT_WORKERS:%.*]] 437 // CHECK1: .select.workers: 438 // CHECK1-NEXT: [[TMP3:%.*]] = load i8, i8* [[EXEC_STATUS]], align 1 439 // CHECK1-NEXT: [[IS_ACTIVE:%.*]] = icmp ne i8 [[TMP3]], 0 440 // CHECK1-NEXT: br i1 [[IS_ACTIVE]], label [[DOTEXECUTE_PARALLEL:%.*]], label [[DOTBARRIER_PARALLEL:%.*]] 441 // CHECK1: .execute.parallel: 442 // CHECK1-NEXT: [[TMP4:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 443 // CHECK1-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP2]] to void (i16, i32)* 444 // CHECK1-NEXT: call void [[TMP5]](i16 0, i32 [[TMP4]]) 445 // CHECK1-NEXT: br label [[DOTTERMINATE_PARALLEL:%.*]] 446 // CHECK1: .terminate.parallel: 447 // CHECK1-NEXT: call void @__kmpc_kernel_end_parallel() 448 // CHECK1-NEXT: br label [[DOTBARRIER_PARALLEL]] 449 // CHECK1: .barrier.parallel: 450 // CHECK1-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 451 // CHECK1-NEXT: br label [[DOTAWAIT_WORK]] 452 // CHECK1: .exit: 453 // CHECK1-NEXT: ret void 454 // 455 // 456 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29 457 // CHECK1-SAME: (i64 [[C:%.*]], i64 [[D:%.*]]) #[[ATTR1]] { 458 // CHECK1-NEXT: entry: 459 // CHECK1-NEXT: [[C_ADDR:%.*]] = alloca i64, align 8 460 // CHECK1-NEXT: [[D_ADDR:%.*]] = alloca i64, align 8 461 // CHECK1-NEXT: [[DOTZERO_ADDR:%.*]] = alloca i32, align 4 462 // CHECK1-NEXT: [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4 463 // CHECK1-NEXT: store i32 0, i32* [[DOTZERO_ADDR]], align 4 464 // CHECK1-NEXT: store i64 [[C]], i64* [[C_ADDR]], align 8 465 // CHECK1-NEXT: store i64 [[D]], i64* [[D_ADDR]], align 8 466 // CHECK1-NEXT: [[CONV:%.*]] = bitcast i64* [[C_ADDR]] to i8* 467 // CHECK1-NEXT: [[CONV1:%.*]] = bitcast i64* [[D_ADDR]] to float* 468 // CHECK1-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 469 // CHECK1-NEXT: [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 470 // CHECK1-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 471 // CHECK1-NEXT: [[THREAD_LIMIT:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS]], [[NVPTX_WARP_SIZE]] 472 // CHECK1-NEXT: [[TMP0:%.*]] = icmp ult i32 [[NVPTX_TID]], [[THREAD_LIMIT]] 473 // CHECK1-NEXT: br i1 [[TMP0]], label [[DOTWORKER:%.*]], label [[DOTMASTERCHECK:%.*]] 474 // CHECK1: .worker: 475 // CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29_worker() #[[ATTR3]] 476 // CHECK1-NEXT: br label [[DOTEXIT:%.*]] 477 // CHECK1: .mastercheck: 478 // CHECK1-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 479 // CHECK1-NEXT: [[NVPTX_NUM_THREADS3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 480 // CHECK1-NEXT: [[NVPTX_WARP_SIZE4:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 481 // CHECK1-NEXT: [[TMP1:%.*]] = sub nuw i32 [[NVPTX_WARP_SIZE4]], 1 482 // CHECK1-NEXT: [[TMP2:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS3]], 1 483 // CHECK1-NEXT: [[TMP3:%.*]] = xor i32 [[TMP1]], -1 484 // CHECK1-NEXT: [[MASTER_TID:%.*]] = and i32 [[TMP2]], [[TMP3]] 485 // CHECK1-NEXT: [[TMP4:%.*]] = icmp eq i32 [[NVPTX_TID2]], [[MASTER_TID]] 486 // CHECK1-NEXT: br i1 [[TMP4]], label [[DOTMASTER:%.*]], label [[DOTEXIT]] 487 // CHECK1: .master: 488 // CHECK1-NEXT: [[NVPTX_NUM_THREADS5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 489 // CHECK1-NEXT: [[NVPTX_WARP_SIZE6:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 490 // CHECK1-NEXT: [[THREAD_LIMIT7:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS5]], [[NVPTX_WARP_SIZE6]] 491 // CHECK1-NEXT: call void @__kmpc_kernel_init(i32 [[THREAD_LIMIT7]], i16 1) 492 // CHECK1-NEXT: call void @__kmpc_data_sharing_init_stack() 493 // CHECK1-NEXT: [[TMP5:%.*]] = load i16, i16* @"_openmp_static_kernel$is_shared1", align 2 494 // CHECK1-NEXT: [[TMP6:%.*]] = load i64, i64* @"_openmp_static_kernel$size2", align 8 495 // CHECK1-NEXT: call void @__kmpc_get_team_static_memory(i16 0, i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds (%"union._shared_openmp_static_memory_type_$_", %"union._shared_openmp_static_memory_type_$_" addrspace(3)* @"_openmp_shared_static_glob_rd_$_", i32 0, i32 0, i32 0) to i8*), i64 [[TMP6]], i16 [[TMP5]], i8** addrspacecast (i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr" to i8**)) 496 // CHECK1-NEXT: [[TMP7:%.*]] = load i8*, i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr", align 8 497 // CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds i8, i8* [[TMP7]], i64 0 498 // CHECK1-NEXT: [[TMP9:%.*]] = bitcast i8* [[TMP8]] to %struct._globalized_locals_ty.2* 499 // CHECK1-NEXT: [[TMP10:%.*]] = load i8, i8* [[CONV]], align 8 500 // CHECK1-NEXT: [[C8:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_2:%.*]], %struct._globalized_locals_ty.2* [[TMP9]], i32 0, i32 1 501 // CHECK1-NEXT: store i8 [[TMP10]], i8* [[C8]], align 4 502 // CHECK1-NEXT: [[TMP11:%.*]] = load float, float* [[CONV1]], align 8 503 // CHECK1-NEXT: [[D9:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_2]], %struct._globalized_locals_ty.2* [[TMP9]], i32 0, i32 0 504 // CHECK1-NEXT: store float [[TMP11]], float* [[D9]], align 4 505 // CHECK1-NEXT: [[TMP12:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 506 // CHECK1-NEXT: store i32 [[TMP12]], i32* [[DOTTHREADID_TEMP_]], align 4 507 // CHECK1-NEXT: call void @__omp_outlined__3(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i8* [[C8]], float* [[D9]]) #[[ATTR3]] 508 // CHECK1-NEXT: [[TMP13:%.*]] = load i16, i16* @"_openmp_static_kernel$is_shared1", align 2 509 // CHECK1-NEXT: call void @__kmpc_restore_team_static_memory(i16 0, i16 [[TMP13]]) 510 // CHECK1-NEXT: br label [[DOTTERMINATION_NOTIFIER:%.*]] 511 // CHECK1: .termination.notifier: 512 // CHECK1-NEXT: call void @__kmpc_kernel_deinit(i16 1) 513 // CHECK1-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 514 // CHECK1-NEXT: br label [[DOTEXIT]] 515 // CHECK1: .exit: 516 // CHECK1-NEXT: ret void 517 // 518 // 519 // CHECK1-LABEL: define {{[^@]+}}@__omp_outlined__3 520 // CHECK1-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i8* nonnull align 1 dereferenceable(1) [[C:%.*]], float* nonnull align 4 dereferenceable(4) [[D:%.*]]) #[[ATTR1]] { 521 // CHECK1-NEXT: entry: 522 // CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8 523 // CHECK1-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8 524 // CHECK1-NEXT: [[C_ADDR:%.*]] = alloca i8*, align 8 525 // CHECK1-NEXT: [[D_ADDR:%.*]] = alloca float*, align 8 526 // CHECK1-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 8 527 // CHECK1-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8 528 // CHECK1-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8 529 // CHECK1-NEXT: store i8* [[C]], i8** [[C_ADDR]], align 8 530 // CHECK1-NEXT: store float* [[D]], float** [[D_ADDR]], align 8 531 // CHECK1-NEXT: [[TMP0:%.*]] = load i8*, i8** [[C_ADDR]], align 8 532 // CHECK1-NEXT: [[TMP1:%.*]] = load float*, float** [[D_ADDR]], align 8 533 // CHECK1-NEXT: [[TMP2:%.*]] = load i8*, i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr", align 8 534 // CHECK1-NEXT: [[TMP3:%.*]] = getelementptr inbounds i8, i8* [[TMP2]], i64 8 535 // CHECK1-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.3* 536 // CHECK1-NEXT: [[C1:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3:%.*]], %struct._globalized_locals_ty.3* [[TMP4]], i32 0, i32 1 537 // CHECK1-NEXT: [[D2:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3]], %struct._globalized_locals_ty.3* [[TMP4]], i32 0, i32 0 538 // CHECK1-NEXT: store i8 0, i8* [[C1]], align 4 539 // CHECK1-NEXT: store float 1.000000e+00, float* [[D2]], align 4 540 // CHECK1-NEXT: [[TMP5:%.*]] = load i8, i8* [[C1]], align 4 541 // CHECK1-NEXT: [[CONV:%.*]] = sext i8 [[TMP5]] to i32 542 // CHECK1-NEXT: [[XOR:%.*]] = xor i32 [[CONV]], 2 543 // CHECK1-NEXT: [[CONV3:%.*]] = trunc i32 [[XOR]] to i8 544 // CHECK1-NEXT: store i8 [[CONV3]], i8* [[C1]], align 4 545 // CHECK1-NEXT: [[TMP6:%.*]] = load float, float* [[D2]], align 4 546 // CHECK1-NEXT: [[MUL:%.*]] = fmul float [[TMP6]], 3.300000e+01 547 // CHECK1-NEXT: store float [[MUL]], float* [[D2]], align 4 548 // CHECK1-NEXT: [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8 549 // CHECK1-NEXT: [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4 550 // CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0 551 // CHECK1-NEXT: store i8* [[C1]], i8** [[TMP9]], align 8 552 // CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 1 553 // CHECK1-NEXT: [[TMP11:%.*]] = bitcast float* [[D2]] to i8* 554 // CHECK1-NEXT: store i8* [[TMP11]], i8** [[TMP10]], align 8 555 // CHECK1-NEXT: [[TMP12:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 556 // CHECK1-NEXT: [[TMP13:%.*]] = load i8*, i8** @"_openmp_teams_reductions_buffer_$_$ptr", align 8 557 // CHECK1-NEXT: [[TMP14:%.*]] = call i32 @__kmpc_nvptx_teams_reduce_nowait_v2(%struct.ident_t* @[[GLOB1]], i32 [[TMP8]], i8* [[TMP13]], i32 1024, i8* [[TMP12]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func5, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func6, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_copy_func7, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_reduce_func8, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_copy_func9, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_reduce_func10) 558 // CHECK1-NEXT: [[TMP15:%.*]] = icmp eq i32 [[TMP14]], 1 559 // CHECK1-NEXT: br i1 [[TMP15]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]] 560 // CHECK1: .omp.reduction.then: 561 // CHECK1-NEXT: [[TMP16:%.*]] = load i8, i8* [[TMP0]], align 1 562 // CHECK1-NEXT: [[CONV4:%.*]] = sext i8 [[TMP16]] to i32 563 // CHECK1-NEXT: [[TMP17:%.*]] = load i8, i8* [[C1]], align 4 564 // CHECK1-NEXT: [[CONV5:%.*]] = sext i8 [[TMP17]] to i32 565 // CHECK1-NEXT: [[XOR6:%.*]] = xor i32 [[CONV4]], [[CONV5]] 566 // CHECK1-NEXT: [[CONV7:%.*]] = trunc i32 [[XOR6]] to i8 567 // CHECK1-NEXT: store i8 [[CONV7]], i8* [[TMP0]], align 1 568 // CHECK1-NEXT: [[TMP18:%.*]] = load float, float* [[TMP1]], align 4 569 // CHECK1-NEXT: [[TMP19:%.*]] = load float, float* [[D2]], align 4 570 // CHECK1-NEXT: [[MUL8:%.*]] = fmul float [[TMP18]], [[TMP19]] 571 // CHECK1-NEXT: store float [[MUL8]], float* [[TMP1]], align 4 572 // CHECK1-NEXT: call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP8]]) 573 // CHECK1-NEXT: br label [[DOTOMP_REDUCTION_DONE]] 574 // CHECK1: .omp.reduction.done: 575 // CHECK1-NEXT: ret void 576 // 577 // 578 // CHECK1-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func5 579 // CHECK1-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR0]] { 580 // CHECK1-NEXT: entry: 581 // CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 582 // CHECK1-NEXT: [[DOTADDR1:%.*]] = alloca i16, align 2 583 // CHECK1-NEXT: [[DOTADDR2:%.*]] = alloca i16, align 2 584 // CHECK1-NEXT: [[DOTADDR3:%.*]] = alloca i16, align 2 585 // CHECK1-NEXT: [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [2 x i8*], align 8 586 // CHECK1-NEXT: [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca i8, align 1 587 // CHECK1-NEXT: [[DOTOMP_REDUCTION_ELEMENT4:%.*]] = alloca float, align 4 588 // CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 589 // CHECK1-NEXT: store i16 [[TMP1]], i16* [[DOTADDR1]], align 2 590 // CHECK1-NEXT: store i16 [[TMP2]], i16* [[DOTADDR2]], align 2 591 // CHECK1-NEXT: store i16 [[TMP3]], i16* [[DOTADDR3]], align 2 592 // CHECK1-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 8 593 // CHECK1-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [2 x i8*]* 594 // CHECK1-NEXT: [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2 595 // CHECK1-NEXT: [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2 596 // CHECK1-NEXT: [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2 597 // CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i64 0, i64 0 598 // CHECK1-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8 599 // CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0 600 // CHECK1-NEXT: [[TMP12:%.*]] = getelementptr i8, i8* [[TMP10]], i64 1 601 // CHECK1-NEXT: [[TMP13:%.*]] = load i8, i8* [[TMP10]], align 1 602 // CHECK1-NEXT: [[TMP14:%.*]] = sext i8 [[TMP13]] to i32 603 // CHECK1-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 604 // CHECK1-NEXT: [[TMP15:%.*]] = trunc i32 [[NVPTX_WARP_SIZE]] to i16 605 // CHECK1-NEXT: [[TMP16:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP14]], i16 [[TMP7]], i16 [[TMP15]]) 606 // CHECK1-NEXT: [[TMP17:%.*]] = trunc i32 [[TMP16]] to i8 607 // CHECK1-NEXT: store i8 [[TMP17]], i8* [[DOTOMP_REDUCTION_ELEMENT]], align 1 608 // CHECK1-NEXT: [[TMP18:%.*]] = getelementptr i8, i8* [[TMP10]], i64 1 609 // CHECK1-NEXT: [[TMP19:%.*]] = getelementptr i8, i8* [[DOTOMP_REDUCTION_ELEMENT]], i64 1 610 // CHECK1-NEXT: store i8* [[DOTOMP_REDUCTION_ELEMENT]], i8** [[TMP11]], align 8 611 // CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i64 0, i64 1 612 // CHECK1-NEXT: [[TMP21:%.*]] = load i8*, i8** [[TMP20]], align 8 613 // CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 1 614 // CHECK1-NEXT: [[TMP23:%.*]] = bitcast i8* [[TMP21]] to float* 615 // CHECK1-NEXT: [[TMP24:%.*]] = getelementptr float, float* [[TMP23]], i64 1 616 // CHECK1-NEXT: [[TMP25:%.*]] = bitcast float* [[TMP24]] to i8* 617 // CHECK1-NEXT: [[TMP26:%.*]] = bitcast float* [[TMP23]] to i32* 618 // CHECK1-NEXT: [[TMP27:%.*]] = bitcast float* [[DOTOMP_REDUCTION_ELEMENT4]] to i32* 619 // CHECK1-NEXT: [[TMP28:%.*]] = load i32, i32* [[TMP26]], align 4 620 // CHECK1-NEXT: [[NVPTX_WARP_SIZE5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 621 // CHECK1-NEXT: [[TMP29:%.*]] = trunc i32 [[NVPTX_WARP_SIZE5]] to i16 622 // CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP28]], i16 [[TMP7]], i16 [[TMP29]]) 623 // CHECK1-NEXT: store i32 [[TMP30]], i32* [[TMP27]], align 4 624 // CHECK1-NEXT: [[TMP31:%.*]] = getelementptr i32, i32* [[TMP26]], i64 1 625 // CHECK1-NEXT: [[TMP32:%.*]] = getelementptr i32, i32* [[TMP27]], i64 1 626 // CHECK1-NEXT: [[TMP33:%.*]] = bitcast float* [[DOTOMP_REDUCTION_ELEMENT4]] to i8* 627 // CHECK1-NEXT: store i8* [[TMP33]], i8** [[TMP22]], align 8 628 // CHECK1-NEXT: [[TMP34:%.*]] = icmp eq i16 [[TMP8]], 0 629 // CHECK1-NEXT: [[TMP35:%.*]] = icmp eq i16 [[TMP8]], 1 630 // CHECK1-NEXT: [[TMP36:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]] 631 // CHECK1-NEXT: [[TMP37:%.*]] = and i1 [[TMP35]], [[TMP36]] 632 // CHECK1-NEXT: [[TMP38:%.*]] = icmp eq i16 [[TMP8]], 2 633 // CHECK1-NEXT: [[TMP39:%.*]] = and i16 [[TMP6]], 1 634 // CHECK1-NEXT: [[TMP40:%.*]] = icmp eq i16 [[TMP39]], 0 635 // CHECK1-NEXT: [[TMP41:%.*]] = and i1 [[TMP38]], [[TMP40]] 636 // CHECK1-NEXT: [[TMP42:%.*]] = icmp sgt i16 [[TMP7]], 0 637 // CHECK1-NEXT: [[TMP43:%.*]] = and i1 [[TMP41]], [[TMP42]] 638 // CHECK1-NEXT: [[TMP44:%.*]] = or i1 [[TMP34]], [[TMP37]] 639 // CHECK1-NEXT: [[TMP45:%.*]] = or i1 [[TMP44]], [[TMP43]] 640 // CHECK1-NEXT: br i1 [[TMP45]], label [[THEN:%.*]], label [[ELSE:%.*]] 641 // CHECK1: then: 642 // CHECK1-NEXT: [[TMP46:%.*]] = bitcast [2 x i8*]* [[TMP5]] to i8* 643 // CHECK1-NEXT: [[TMP47:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8* 644 // CHECK1-NEXT: call void @"_omp$reduction$reduction_func4"(i8* [[TMP46]], i8* [[TMP47]]) #[[ATTR3]] 645 // CHECK1-NEXT: br label [[IFCONT:%.*]] 646 // CHECK1: else: 647 // CHECK1-NEXT: br label [[IFCONT]] 648 // CHECK1: ifcont: 649 // CHECK1-NEXT: [[TMP48:%.*]] = icmp eq i16 [[TMP8]], 1 650 // CHECK1-NEXT: [[TMP49:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]] 651 // CHECK1-NEXT: [[TMP50:%.*]] = and i1 [[TMP48]], [[TMP49]] 652 // CHECK1-NEXT: br i1 [[TMP50]], label [[THEN6:%.*]], label [[ELSE7:%.*]] 653 // CHECK1: then6: 654 // CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0 655 // CHECK1-NEXT: [[TMP52:%.*]] = load i8*, i8** [[TMP51]], align 8 656 // CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i64 0, i64 0 657 // CHECK1-NEXT: [[TMP54:%.*]] = load i8*, i8** [[TMP53]], align 8 658 // CHECK1-NEXT: [[TMP55:%.*]] = load i8, i8* [[TMP52]], align 1 659 // CHECK1-NEXT: store i8 [[TMP55]], i8* [[TMP54]], align 1 660 // CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 1 661 // CHECK1-NEXT: [[TMP57:%.*]] = load i8*, i8** [[TMP56]], align 8 662 // CHECK1-NEXT: [[TMP58:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i64 0, i64 1 663 // CHECK1-NEXT: [[TMP59:%.*]] = load i8*, i8** [[TMP58]], align 8 664 // CHECK1-NEXT: [[TMP60:%.*]] = bitcast i8* [[TMP57]] to float* 665 // CHECK1-NEXT: [[TMP61:%.*]] = bitcast i8* [[TMP59]] to float* 666 // CHECK1-NEXT: [[TMP62:%.*]] = load float, float* [[TMP60]], align 4 667 // CHECK1-NEXT: store float [[TMP62]], float* [[TMP61]], align 4 668 // CHECK1-NEXT: br label [[IFCONT8:%.*]] 669 // CHECK1: else7: 670 // CHECK1-NEXT: br label [[IFCONT8]] 671 // CHECK1: ifcont8: 672 // CHECK1-NEXT: ret void 673 // 674 // 675 // CHECK1-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func6 676 // CHECK1-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] { 677 // CHECK1-NEXT: entry: 678 // CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 679 // CHECK1-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 680 // CHECK1-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 681 // CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 682 // CHECK1-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 683 // CHECK1-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 684 // CHECK1-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 685 // CHECK1-NEXT: [[NVPTX_LANE_ID:%.*]] = and i32 [[NVPTX_TID2]], 31 686 // CHECK1-NEXT: [[NVPTX_TID3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 687 // CHECK1-NEXT: [[NVPTX_WARP_ID:%.*]] = ashr i32 [[NVPTX_TID3]], 5 688 // CHECK1-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8 689 // CHECK1-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 690 // CHECK1-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 691 // CHECK1-NEXT: [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 692 // CHECK1-NEXT: br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]] 693 // CHECK1: then: 694 // CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 0 695 // CHECK1-NEXT: [[TMP6:%.*]] = load i8*, i8** [[TMP5]], align 8 696 // CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 697 // CHECK1-NEXT: [[TMP8:%.*]] = bitcast i32 addrspace(3)* [[TMP7]] to i8 addrspace(3)* 698 // CHECK1-NEXT: [[TMP9:%.*]] = load i8, i8* [[TMP6]], align 1 699 // CHECK1-NEXT: store volatile i8 [[TMP9]], i8 addrspace(3)* [[TMP8]], align 1 700 // CHECK1-NEXT: br label [[IFCONT:%.*]] 701 // CHECK1: else: 702 // CHECK1-NEXT: br label [[IFCONT]] 703 // CHECK1: ifcont: 704 // CHECK1-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 705 // CHECK1-NEXT: [[TMP10:%.*]] = load i32, i32* [[DOTADDR1]], align 4 706 // CHECK1-NEXT: [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP10]] 707 // CHECK1-NEXT: br i1 [[IS_ACTIVE_THREAD]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 708 // CHECK1: then4: 709 // CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 710 // CHECK1-NEXT: [[TMP12:%.*]] = bitcast i32 addrspace(3)* [[TMP11]] to i8 addrspace(3)* 711 // CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 0 712 // CHECK1-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 8 713 // CHECK1-NEXT: [[TMP15:%.*]] = load volatile i8, i8 addrspace(3)* [[TMP12]], align 1 714 // CHECK1-NEXT: store i8 [[TMP15]], i8* [[TMP14]], align 1 715 // CHECK1-NEXT: br label [[IFCONT6:%.*]] 716 // CHECK1: else5: 717 // CHECK1-NEXT: br label [[IFCONT6]] 718 // CHECK1: ifcont6: 719 // CHECK1-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 720 // CHECK1-NEXT: [[WARP_MASTER7:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 721 // CHECK1-NEXT: br i1 [[WARP_MASTER7]], label [[THEN8:%.*]], label [[ELSE9:%.*]] 722 // CHECK1: then8: 723 // CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 1 724 // CHECK1-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 8 725 // CHECK1-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i32* 726 // CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 727 // CHECK1-NEXT: [[TMP20:%.*]] = load i32, i32* [[TMP18]], align 4 728 // CHECK1-NEXT: store volatile i32 [[TMP20]], i32 addrspace(3)* [[TMP19]], align 4 729 // CHECK1-NEXT: br label [[IFCONT10:%.*]] 730 // CHECK1: else9: 731 // CHECK1-NEXT: br label [[IFCONT10]] 732 // CHECK1: ifcont10: 733 // CHECK1-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 734 // CHECK1-NEXT: [[TMP21:%.*]] = load i32, i32* [[DOTADDR1]], align 4 735 // CHECK1-NEXT: [[IS_ACTIVE_THREAD11:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP21]] 736 // CHECK1-NEXT: br i1 [[IS_ACTIVE_THREAD11]], label [[THEN12:%.*]], label [[ELSE13:%.*]] 737 // CHECK1: then12: 738 // CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 739 // CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 1 740 // CHECK1-NEXT: [[TMP24:%.*]] = load i8*, i8** [[TMP23]], align 8 741 // CHECK1-NEXT: [[TMP25:%.*]] = bitcast i8* [[TMP24]] to i32* 742 // CHECK1-NEXT: [[TMP26:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP22]], align 4 743 // CHECK1-NEXT: store i32 [[TMP26]], i32* [[TMP25]], align 4 744 // CHECK1-NEXT: br label [[IFCONT14:%.*]] 745 // CHECK1: else13: 746 // CHECK1-NEXT: br label [[IFCONT14]] 747 // CHECK1: ifcont14: 748 // CHECK1-NEXT: ret void 749 // 750 // 751 // CHECK1-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_copy_func7 752 // CHECK1-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 753 // CHECK1-NEXT: entry: 754 // CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 755 // CHECK1-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 756 // CHECK1-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 8 757 // CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 758 // CHECK1-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 759 // CHECK1-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 8 760 // CHECK1-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 8 761 // CHECK1-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 762 // CHECK1-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 8 763 // CHECK1-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.4* 764 // CHECK1-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 765 // CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 0 766 // CHECK1-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 8 767 // CHECK1-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4:%.*]], %struct._globalized_locals_ty.4* [[TMP6]], i32 0, i32 0 768 // CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[C]], i32 0, i32 [[TMP7]] 769 // CHECK1-NEXT: [[TMP11:%.*]] = load i8, i8* [[TMP9]], align 1 770 // CHECK1-NEXT: store i8 [[TMP11]], i8* [[TMP10]], align 128 771 // CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 1 772 // CHECK1-NEXT: [[TMP13:%.*]] = load i8*, i8** [[TMP12]], align 8 773 // CHECK1-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to float* 774 // CHECK1-NEXT: [[D:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4]], %struct._globalized_locals_ty.4* [[TMP6]], i32 0, i32 1 775 // CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1024 x float], [1024 x float]* [[D]], i32 0, i32 [[TMP7]] 776 // CHECK1-NEXT: [[TMP16:%.*]] = load float, float* [[TMP14]], align 4 777 // CHECK1-NEXT: store float [[TMP16]], float* [[TMP15]], align 128 778 // CHECK1-NEXT: ret void 779 // 780 // 781 // CHECK1-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_reduce_func8 782 // CHECK1-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 783 // CHECK1-NEXT: entry: 784 // CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 785 // CHECK1-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 786 // CHECK1-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 8 787 // CHECK1-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 8 788 // CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 789 // CHECK1-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 790 // CHECK1-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 8 791 // CHECK1-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8 792 // CHECK1-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.4* 793 // CHECK1-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 794 // CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0 795 // CHECK1-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4:%.*]], %struct._globalized_locals_ty.4* [[TMP4]], i32 0, i32 0 796 // CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[C]], i32 0, i32 [[TMP5]] 797 // CHECK1-NEXT: store i8* [[TMP7]], i8** [[TMP6]], align 8 798 // CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 1 799 // CHECK1-NEXT: [[D:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4]], %struct._globalized_locals_ty.4* [[TMP4]], i32 0, i32 1 800 // CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1024 x float], [1024 x float]* [[D]], i32 0, i32 [[TMP5]] 801 // CHECK1-NEXT: [[TMP10:%.*]] = bitcast float* [[TMP9]] to i8* 802 // CHECK1-NEXT: store i8* [[TMP10]], i8** [[TMP8]], align 8 803 // CHECK1-NEXT: [[TMP11:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 804 // CHECK1-NEXT: [[TMP12:%.*]] = load i8*, i8** [[DOTADDR2]], align 8 805 // CHECK1-NEXT: call void @"_omp$reduction$reduction_func4"(i8* [[TMP11]], i8* [[TMP12]]) #[[ATTR3]] 806 // CHECK1-NEXT: ret void 807 // 808 // 809 // CHECK1-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_copy_func9 810 // CHECK1-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 811 // CHECK1-NEXT: entry: 812 // CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 813 // CHECK1-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 814 // CHECK1-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 8 815 // CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 816 // CHECK1-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 817 // CHECK1-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 8 818 // CHECK1-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 8 819 // CHECK1-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 820 // CHECK1-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 8 821 // CHECK1-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.4* 822 // CHECK1-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 823 // CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 0 824 // CHECK1-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 8 825 // CHECK1-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4:%.*]], %struct._globalized_locals_ty.4* [[TMP6]], i32 0, i32 0 826 // CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[C]], i32 0, i32 [[TMP7]] 827 // CHECK1-NEXT: [[TMP11:%.*]] = load i8, i8* [[TMP10]], align 128 828 // CHECK1-NEXT: store i8 [[TMP11]], i8* [[TMP9]], align 1 829 // CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 1 830 // CHECK1-NEXT: [[TMP13:%.*]] = load i8*, i8** [[TMP12]], align 8 831 // CHECK1-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to float* 832 // CHECK1-NEXT: [[D:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4]], %struct._globalized_locals_ty.4* [[TMP6]], i32 0, i32 1 833 // CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1024 x float], [1024 x float]* [[D]], i32 0, i32 [[TMP7]] 834 // CHECK1-NEXT: [[TMP16:%.*]] = load float, float* [[TMP15]], align 128 835 // CHECK1-NEXT: store float [[TMP16]], float* [[TMP14]], align 4 836 // CHECK1-NEXT: ret void 837 // 838 // 839 // CHECK1-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_reduce_func10 840 // CHECK1-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 841 // CHECK1-NEXT: entry: 842 // CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 843 // CHECK1-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 844 // CHECK1-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 8 845 // CHECK1-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 8 846 // CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 847 // CHECK1-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 848 // CHECK1-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 8 849 // CHECK1-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8 850 // CHECK1-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.4* 851 // CHECK1-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 852 // CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0 853 // CHECK1-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4:%.*]], %struct._globalized_locals_ty.4* [[TMP4]], i32 0, i32 0 854 // CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[C]], i32 0, i32 [[TMP5]] 855 // CHECK1-NEXT: store i8* [[TMP7]], i8** [[TMP6]], align 8 856 // CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 1 857 // CHECK1-NEXT: [[D:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4]], %struct._globalized_locals_ty.4* [[TMP4]], i32 0, i32 1 858 // CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1024 x float], [1024 x float]* [[D]], i32 0, i32 [[TMP5]] 859 // CHECK1-NEXT: [[TMP10:%.*]] = bitcast float* [[TMP9]] to i8* 860 // CHECK1-NEXT: store i8* [[TMP10]], i8** [[TMP8]], align 8 861 // CHECK1-NEXT: [[TMP11:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 862 // CHECK1-NEXT: [[TMP12:%.*]] = load i8*, i8** [[DOTADDR2]], align 8 863 // CHECK1-NEXT: call void @"_omp$reduction$reduction_func4"(i8* [[TMP12]], i8* [[TMP11]]) #[[ATTR3]] 864 // CHECK1-NEXT: ret void 865 // 866 // 867 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l36 868 // CHECK1-SAME: (i64 [[A:%.*]], i64 [[B:%.*]]) #[[ATTR1]] { 869 // CHECK1-NEXT: entry: 870 // CHECK1-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8 871 // CHECK1-NEXT: [[B_ADDR:%.*]] = alloca i64, align 8 872 // CHECK1-NEXT: [[DOTZERO_ADDR:%.*]] = alloca i32, align 4 873 // CHECK1-NEXT: [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4 874 // CHECK1-NEXT: store i32 0, i32* [[DOTZERO_ADDR]], align 4 875 // CHECK1-NEXT: store i64 [[A]], i64* [[A_ADDR]], align 8 876 // CHECK1-NEXT: store i64 [[B]], i64* [[B_ADDR]], align 8 877 // CHECK1-NEXT: [[CONV:%.*]] = bitcast i64* [[A_ADDR]] to i32* 878 // CHECK1-NEXT: [[CONV1:%.*]] = bitcast i64* [[B_ADDR]] to i16* 879 // CHECK1-NEXT: [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 880 // CHECK1-NEXT: call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 1) 881 // CHECK1-NEXT: call void @__kmpc_data_sharing_init_stack_spmd() 882 // CHECK1-NEXT: br label [[DOTEXECUTE:%.*]] 883 // CHECK1: .execute: 884 // CHECK1-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3:[0-9]+]]) 885 // CHECK1-NEXT: store i32 [[TMP0]], i32* [[DOTTHREADID_TEMP_]], align 4 886 // CHECK1-NEXT: call void @__omp_outlined__11(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32* [[CONV]], i16* [[CONV1]]) #[[ATTR3]] 887 // CHECK1-NEXT: br label [[DOTOMP_DEINIT:%.*]] 888 // CHECK1: .omp.deinit: 889 // CHECK1-NEXT: call void @__kmpc_spmd_kernel_deinit_v2(i16 1) 890 // CHECK1-NEXT: br label [[DOTEXIT:%.*]] 891 // CHECK1: .exit: 892 // CHECK1-NEXT: ret void 893 // 894 // 895 // CHECK1-LABEL: define {{[^@]+}}@__omp_outlined__11 896 // CHECK1-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32* nonnull align 4 dereferenceable(4) [[A:%.*]], i16* nonnull align 2 dereferenceable(2) [[B:%.*]]) #[[ATTR1]] { 897 // CHECK1-NEXT: entry: 898 // CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8 899 // CHECK1-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8 900 // CHECK1-NEXT: [[A_ADDR:%.*]] = alloca i32*, align 8 901 // CHECK1-NEXT: [[B_ADDR:%.*]] = alloca i16*, align 8 902 // CHECK1-NEXT: [[A1:%.*]] = alloca i32, align 4 903 // CHECK1-NEXT: [[B2:%.*]] = alloca i16, align 2 904 // CHECK1-NEXT: [[CAPTURED_VARS_ADDRS:%.*]] = alloca [2 x i8*], align 8 905 // CHECK1-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 8 906 // CHECK1-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8 907 // CHECK1-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8 908 // CHECK1-NEXT: store i32* [[A]], i32** [[A_ADDR]], align 8 909 // CHECK1-NEXT: store i16* [[B]], i16** [[B_ADDR]], align 8 910 // CHECK1-NEXT: [[TMP0:%.*]] = load i32*, i32** [[A_ADDR]], align 8 911 // CHECK1-NEXT: [[TMP1:%.*]] = load i16*, i16** [[B_ADDR]], align 8 912 // CHECK1-NEXT: store i32 0, i32* [[A1]], align 4 913 // CHECK1-NEXT: store i16 -32768, i16* [[B2]], align 2 914 // CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0 915 // CHECK1-NEXT: [[TMP3:%.*]] = bitcast i32* [[A1]] to i8* 916 // CHECK1-NEXT: store i8* [[TMP3]], i8** [[TMP2]], align 8 917 // CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1 918 // CHECK1-NEXT: [[TMP5:%.*]] = bitcast i16* [[B2]] to i8* 919 // CHECK1-NEXT: store i8* [[TMP5]], i8** [[TMP4]], align 8 920 // CHECK1-NEXT: [[TMP6:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8 921 // CHECK1-NEXT: [[TMP7:%.*]] = load i32, i32* [[TMP6]], align 4 922 // CHECK1-NEXT: [[TMP8:%.*]] = bitcast [2 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8** 923 // CHECK1-NEXT: call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP7]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32*, i16*)* @__omp_outlined__12 to i8*), i8* null, i8** [[TMP8]], i64 2) 924 // CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0 925 // CHECK1-NEXT: [[TMP10:%.*]] = bitcast i32* [[A1]] to i8* 926 // CHECK1-NEXT: store i8* [[TMP10]], i8** [[TMP9]], align 8 927 // CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 1 928 // CHECK1-NEXT: [[TMP12:%.*]] = bitcast i16* [[B2]] to i8* 929 // CHECK1-NEXT: store i8* [[TMP12]], i8** [[TMP11]], align 8 930 // CHECK1-NEXT: [[TMP13:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 931 // CHECK1-NEXT: [[TMP14:%.*]] = load i8*, i8** @"_openmp_teams_reductions_buffer_$_$ptr", align 8 932 // CHECK1-NEXT: [[TMP15:%.*]] = call i32 @__kmpc_nvptx_teams_reduce_nowait_v2(%struct.ident_t* @[[GLOB3]], i32 [[TMP7]], i8* [[TMP14]], i32 1024, i8* [[TMP13]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func17, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func18, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_copy_func19, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_reduce_func20, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_copy_func21, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_reduce_func22) 933 // CHECK1-NEXT: [[TMP16:%.*]] = icmp eq i32 [[TMP15]], 1 934 // CHECK1-NEXT: br i1 [[TMP16]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]] 935 // CHECK1: .omp.reduction.then: 936 // CHECK1-NEXT: [[TMP17:%.*]] = load i32, i32* [[TMP0]], align 4 937 // CHECK1-NEXT: [[TMP18:%.*]] = load i32, i32* [[A1]], align 4 938 // CHECK1-NEXT: [[OR:%.*]] = or i32 [[TMP17]], [[TMP18]] 939 // CHECK1-NEXT: store i32 [[OR]], i32* [[TMP0]], align 4 940 // CHECK1-NEXT: [[TMP19:%.*]] = load i16, i16* [[TMP1]], align 2 941 // CHECK1-NEXT: [[CONV:%.*]] = sext i16 [[TMP19]] to i32 942 // CHECK1-NEXT: [[TMP20:%.*]] = load i16, i16* [[B2]], align 2 943 // CHECK1-NEXT: [[CONV3:%.*]] = sext i16 [[TMP20]] to i32 944 // CHECK1-NEXT: [[CMP:%.*]] = icmp sgt i32 [[CONV]], [[CONV3]] 945 // CHECK1-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] 946 // CHECK1: cond.true: 947 // CHECK1-NEXT: [[TMP21:%.*]] = load i16, i16* [[TMP1]], align 2 948 // CHECK1-NEXT: br label [[COND_END:%.*]] 949 // CHECK1: cond.false: 950 // CHECK1-NEXT: [[TMP22:%.*]] = load i16, i16* [[B2]], align 2 951 // CHECK1-NEXT: br label [[COND_END]] 952 // CHECK1: cond.end: 953 // CHECK1-NEXT: [[COND:%.*]] = phi i16 [ [[TMP21]], [[COND_TRUE]] ], [ [[TMP22]], [[COND_FALSE]] ] 954 // CHECK1-NEXT: store i16 [[COND]], i16* [[TMP1]], align 2 955 // CHECK1-NEXT: call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP7]]) 956 // CHECK1-NEXT: br label [[DOTOMP_REDUCTION_DONE]] 957 // CHECK1: .omp.reduction.done: 958 // CHECK1-NEXT: ret void 959 // 960 // 961 // CHECK1-LABEL: define {{[^@]+}}@__omp_outlined__12 962 // CHECK1-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32* nonnull align 4 dereferenceable(4) [[A:%.*]], i16* nonnull align 2 dereferenceable(2) [[B:%.*]]) #[[ATTR1]] { 963 // CHECK1-NEXT: entry: 964 // CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8 965 // CHECK1-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8 966 // CHECK1-NEXT: [[A_ADDR:%.*]] = alloca i32*, align 8 967 // CHECK1-NEXT: [[B_ADDR:%.*]] = alloca i16*, align 8 968 // CHECK1-NEXT: [[A1:%.*]] = alloca i32, align 4 969 // CHECK1-NEXT: [[B2:%.*]] = alloca i16, align 2 970 // CHECK1-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 8 971 // CHECK1-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8 972 // CHECK1-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8 973 // CHECK1-NEXT: store i32* [[A]], i32** [[A_ADDR]], align 8 974 // CHECK1-NEXT: store i16* [[B]], i16** [[B_ADDR]], align 8 975 // CHECK1-NEXT: [[TMP0:%.*]] = load i32*, i32** [[A_ADDR]], align 8 976 // CHECK1-NEXT: [[TMP1:%.*]] = load i16*, i16** [[B_ADDR]], align 8 977 // CHECK1-NEXT: store i32 0, i32* [[A1]], align 4 978 // CHECK1-NEXT: store i16 -32768, i16* [[B2]], align 2 979 // CHECK1-NEXT: [[TMP2:%.*]] = load i32, i32* [[A1]], align 4 980 // CHECK1-NEXT: [[OR:%.*]] = or i32 [[TMP2]], 1 981 // CHECK1-NEXT: store i32 [[OR]], i32* [[A1]], align 4 982 // CHECK1-NEXT: [[TMP3:%.*]] = load i16, i16* [[B2]], align 2 983 // CHECK1-NEXT: [[CONV:%.*]] = sext i16 [[TMP3]] to i32 984 // CHECK1-NEXT: [[CMP:%.*]] = icmp sgt i32 99, [[CONV]] 985 // CHECK1-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] 986 // CHECK1: cond.true: 987 // CHECK1-NEXT: br label [[COND_END:%.*]] 988 // CHECK1: cond.false: 989 // CHECK1-NEXT: [[TMP4:%.*]] = load i16, i16* [[B2]], align 2 990 // CHECK1-NEXT: [[CONV3:%.*]] = sext i16 [[TMP4]] to i32 991 // CHECK1-NEXT: br label [[COND_END]] 992 // CHECK1: cond.end: 993 // CHECK1-NEXT: [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[CONV3]], [[COND_FALSE]] ] 994 // CHECK1-NEXT: [[CONV4:%.*]] = trunc i32 [[COND]] to i16 995 // CHECK1-NEXT: store i16 [[CONV4]], i16* [[B2]], align 2 996 // CHECK1-NEXT: [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8 997 // CHECK1-NEXT: [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4 998 // CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0 999 // CHECK1-NEXT: [[TMP8:%.*]] = bitcast i32* [[A1]] to i8* 1000 // CHECK1-NEXT: store i8* [[TMP8]], i8** [[TMP7]], align 8 1001 // CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 1 1002 // CHECK1-NEXT: [[TMP10:%.*]] = bitcast i16* [[B2]] to i8* 1003 // CHECK1-NEXT: store i8* [[TMP10]], i8** [[TMP9]], align 8 1004 // CHECK1-NEXT: [[TMP11:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 1005 // CHECK1-NEXT: [[TMP12:%.*]] = call i32 @__kmpc_nvptx_parallel_reduce_nowait_v2(%struct.ident_t* @[[GLOB3]], i32 [[TMP6]], i32 2, i64 16, i8* [[TMP11]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func14, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func15) 1006 // CHECK1-NEXT: [[TMP13:%.*]] = icmp eq i32 [[TMP12]], 1 1007 // CHECK1-NEXT: br i1 [[TMP13]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]] 1008 // CHECK1: .omp.reduction.then: 1009 // CHECK1-NEXT: [[TMP14:%.*]] = load i32, i32* [[TMP0]], align 4 1010 // CHECK1-NEXT: [[TMP15:%.*]] = load i32, i32* [[A1]], align 4 1011 // CHECK1-NEXT: [[OR5:%.*]] = or i32 [[TMP14]], [[TMP15]] 1012 // CHECK1-NEXT: store i32 [[OR5]], i32* [[TMP0]], align 4 1013 // CHECK1-NEXT: [[TMP16:%.*]] = load i16, i16* [[TMP1]], align 2 1014 // CHECK1-NEXT: [[CONV6:%.*]] = sext i16 [[TMP16]] to i32 1015 // CHECK1-NEXT: [[TMP17:%.*]] = load i16, i16* [[B2]], align 2 1016 // CHECK1-NEXT: [[CONV7:%.*]] = sext i16 [[TMP17]] to i32 1017 // CHECK1-NEXT: [[CMP8:%.*]] = icmp sgt i32 [[CONV6]], [[CONV7]] 1018 // CHECK1-NEXT: br i1 [[CMP8]], label [[COND_TRUE9:%.*]], label [[COND_FALSE10:%.*]] 1019 // CHECK1: cond.true9: 1020 // CHECK1-NEXT: [[TMP18:%.*]] = load i16, i16* [[TMP1]], align 2 1021 // CHECK1-NEXT: br label [[COND_END11:%.*]] 1022 // CHECK1: cond.false10: 1023 // CHECK1-NEXT: [[TMP19:%.*]] = load i16, i16* [[B2]], align 2 1024 // CHECK1-NEXT: br label [[COND_END11]] 1025 // CHECK1: cond.end11: 1026 // CHECK1-NEXT: [[COND12:%.*]] = phi i16 [ [[TMP18]], [[COND_TRUE9]] ], [ [[TMP19]], [[COND_FALSE10]] ] 1027 // CHECK1-NEXT: store i16 [[COND12]], i16* [[TMP1]], align 2 1028 // CHECK1-NEXT: call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP6]]) 1029 // CHECK1-NEXT: br label [[DOTOMP_REDUCTION_DONE]] 1030 // CHECK1: .omp.reduction.done: 1031 // CHECK1-NEXT: ret void 1032 // 1033 // 1034 // CHECK1-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func14 1035 // CHECK1-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR0]] { 1036 // CHECK1-NEXT: entry: 1037 // CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 1038 // CHECK1-NEXT: [[DOTADDR1:%.*]] = alloca i16, align 2 1039 // CHECK1-NEXT: [[DOTADDR2:%.*]] = alloca i16, align 2 1040 // CHECK1-NEXT: [[DOTADDR3:%.*]] = alloca i16, align 2 1041 // CHECK1-NEXT: [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [2 x i8*], align 8 1042 // CHECK1-NEXT: [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca i32, align 4 1043 // CHECK1-NEXT: [[DOTOMP_REDUCTION_ELEMENT4:%.*]] = alloca i16, align 2 1044 // CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 1045 // CHECK1-NEXT: store i16 [[TMP1]], i16* [[DOTADDR1]], align 2 1046 // CHECK1-NEXT: store i16 [[TMP2]], i16* [[DOTADDR2]], align 2 1047 // CHECK1-NEXT: store i16 [[TMP3]], i16* [[DOTADDR3]], align 2 1048 // CHECK1-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 8 1049 // CHECK1-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [2 x i8*]* 1050 // CHECK1-NEXT: [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2 1051 // CHECK1-NEXT: [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2 1052 // CHECK1-NEXT: [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2 1053 // CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i64 0, i64 0 1054 // CHECK1-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8 1055 // CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0 1056 // CHECK1-NEXT: [[TMP12:%.*]] = bitcast i8* [[TMP10]] to i32* 1057 // CHECK1-NEXT: [[TMP13:%.*]] = getelementptr i32, i32* [[TMP12]], i64 1 1058 // CHECK1-NEXT: [[TMP14:%.*]] = bitcast i32* [[TMP13]] to i8* 1059 // CHECK1-NEXT: [[TMP15:%.*]] = load i32, i32* [[TMP12]], align 4 1060 // CHECK1-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 1061 // CHECK1-NEXT: [[TMP16:%.*]] = trunc i32 [[NVPTX_WARP_SIZE]] to i16 1062 // CHECK1-NEXT: [[TMP17:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP15]], i16 [[TMP7]], i16 [[TMP16]]) 1063 // CHECK1-NEXT: store i32 [[TMP17]], i32* [[DOTOMP_REDUCTION_ELEMENT]], align 4 1064 // CHECK1-NEXT: [[TMP18:%.*]] = getelementptr i32, i32* [[TMP12]], i64 1 1065 // CHECK1-NEXT: [[TMP19:%.*]] = getelementptr i32, i32* [[DOTOMP_REDUCTION_ELEMENT]], i64 1 1066 // CHECK1-NEXT: [[TMP20:%.*]] = bitcast i32* [[DOTOMP_REDUCTION_ELEMENT]] to i8* 1067 // CHECK1-NEXT: store i8* [[TMP20]], i8** [[TMP11]], align 8 1068 // CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i64 0, i64 1 1069 // CHECK1-NEXT: [[TMP22:%.*]] = load i8*, i8** [[TMP21]], align 8 1070 // CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 1 1071 // CHECK1-NEXT: [[TMP24:%.*]] = bitcast i8* [[TMP22]] to i16* 1072 // CHECK1-NEXT: [[TMP25:%.*]] = getelementptr i16, i16* [[TMP24]], i64 1 1073 // CHECK1-NEXT: [[TMP26:%.*]] = bitcast i16* [[TMP25]] to i8* 1074 // CHECK1-NEXT: [[TMP27:%.*]] = load i16, i16* [[TMP24]], align 2 1075 // CHECK1-NEXT: [[TMP28:%.*]] = sext i16 [[TMP27]] to i32 1076 // CHECK1-NEXT: [[NVPTX_WARP_SIZE5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 1077 // CHECK1-NEXT: [[TMP29:%.*]] = trunc i32 [[NVPTX_WARP_SIZE5]] to i16 1078 // CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP28]], i16 [[TMP7]], i16 [[TMP29]]) 1079 // CHECK1-NEXT: [[TMP31:%.*]] = trunc i32 [[TMP30]] to i16 1080 // CHECK1-NEXT: store i16 [[TMP31]], i16* [[DOTOMP_REDUCTION_ELEMENT4]], align 2 1081 // CHECK1-NEXT: [[TMP32:%.*]] = getelementptr i16, i16* [[TMP24]], i64 1 1082 // CHECK1-NEXT: [[TMP33:%.*]] = getelementptr i16, i16* [[DOTOMP_REDUCTION_ELEMENT4]], i64 1 1083 // CHECK1-NEXT: [[TMP34:%.*]] = bitcast i16* [[DOTOMP_REDUCTION_ELEMENT4]] to i8* 1084 // CHECK1-NEXT: store i8* [[TMP34]], i8** [[TMP23]], align 8 1085 // CHECK1-NEXT: [[TMP35:%.*]] = icmp eq i16 [[TMP8]], 0 1086 // CHECK1-NEXT: [[TMP36:%.*]] = icmp eq i16 [[TMP8]], 1 1087 // CHECK1-NEXT: [[TMP37:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]] 1088 // CHECK1-NEXT: [[TMP38:%.*]] = and i1 [[TMP36]], [[TMP37]] 1089 // CHECK1-NEXT: [[TMP39:%.*]] = icmp eq i16 [[TMP8]], 2 1090 // CHECK1-NEXT: [[TMP40:%.*]] = and i16 [[TMP6]], 1 1091 // CHECK1-NEXT: [[TMP41:%.*]] = icmp eq i16 [[TMP40]], 0 1092 // CHECK1-NEXT: [[TMP42:%.*]] = and i1 [[TMP39]], [[TMP41]] 1093 // CHECK1-NEXT: [[TMP43:%.*]] = icmp sgt i16 [[TMP7]], 0 1094 // CHECK1-NEXT: [[TMP44:%.*]] = and i1 [[TMP42]], [[TMP43]] 1095 // CHECK1-NEXT: [[TMP45:%.*]] = or i1 [[TMP35]], [[TMP38]] 1096 // CHECK1-NEXT: [[TMP46:%.*]] = or i1 [[TMP45]], [[TMP44]] 1097 // CHECK1-NEXT: br i1 [[TMP46]], label [[THEN:%.*]], label [[ELSE:%.*]] 1098 // CHECK1: then: 1099 // CHECK1-NEXT: [[TMP47:%.*]] = bitcast [2 x i8*]* [[TMP5]] to i8* 1100 // CHECK1-NEXT: [[TMP48:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8* 1101 // CHECK1-NEXT: call void @"_omp$reduction$reduction_func13"(i8* [[TMP47]], i8* [[TMP48]]) #[[ATTR3]] 1102 // CHECK1-NEXT: br label [[IFCONT:%.*]] 1103 // CHECK1: else: 1104 // CHECK1-NEXT: br label [[IFCONT]] 1105 // CHECK1: ifcont: 1106 // CHECK1-NEXT: [[TMP49:%.*]] = icmp eq i16 [[TMP8]], 1 1107 // CHECK1-NEXT: [[TMP50:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]] 1108 // CHECK1-NEXT: [[TMP51:%.*]] = and i1 [[TMP49]], [[TMP50]] 1109 // CHECK1-NEXT: br i1 [[TMP51]], label [[THEN6:%.*]], label [[ELSE7:%.*]] 1110 // CHECK1: then6: 1111 // CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0 1112 // CHECK1-NEXT: [[TMP53:%.*]] = load i8*, i8** [[TMP52]], align 8 1113 // CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i64 0, i64 0 1114 // CHECK1-NEXT: [[TMP55:%.*]] = load i8*, i8** [[TMP54]], align 8 1115 // CHECK1-NEXT: [[TMP56:%.*]] = bitcast i8* [[TMP53]] to i32* 1116 // CHECK1-NEXT: [[TMP57:%.*]] = bitcast i8* [[TMP55]] to i32* 1117 // CHECK1-NEXT: [[TMP58:%.*]] = load i32, i32* [[TMP56]], align 4 1118 // CHECK1-NEXT: store i32 [[TMP58]], i32* [[TMP57]], align 4 1119 // CHECK1-NEXT: [[TMP59:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 1 1120 // CHECK1-NEXT: [[TMP60:%.*]] = load i8*, i8** [[TMP59]], align 8 1121 // CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i64 0, i64 1 1122 // CHECK1-NEXT: [[TMP62:%.*]] = load i8*, i8** [[TMP61]], align 8 1123 // CHECK1-NEXT: [[TMP63:%.*]] = bitcast i8* [[TMP60]] to i16* 1124 // CHECK1-NEXT: [[TMP64:%.*]] = bitcast i8* [[TMP62]] to i16* 1125 // CHECK1-NEXT: [[TMP65:%.*]] = load i16, i16* [[TMP63]], align 2 1126 // CHECK1-NEXT: store i16 [[TMP65]], i16* [[TMP64]], align 2 1127 // CHECK1-NEXT: br label [[IFCONT8:%.*]] 1128 // CHECK1: else7: 1129 // CHECK1-NEXT: br label [[IFCONT8]] 1130 // CHECK1: ifcont8: 1131 // CHECK1-NEXT: ret void 1132 // 1133 // 1134 // CHECK1-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func15 1135 // CHECK1-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] { 1136 // CHECK1-NEXT: entry: 1137 // CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 1138 // CHECK1-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 1139 // CHECK1-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]]) 1140 // CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 1141 // CHECK1-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 1142 // CHECK1-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 1143 // CHECK1-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 1144 // CHECK1-NEXT: [[NVPTX_LANE_ID:%.*]] = and i32 [[NVPTX_TID2]], 31 1145 // CHECK1-NEXT: [[NVPTX_TID3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 1146 // CHECK1-NEXT: [[NVPTX_WARP_ID:%.*]] = ashr i32 [[NVPTX_TID3]], 5 1147 // CHECK1-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8 1148 // CHECK1-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 1149 // CHECK1-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4:[0-9]+]], i32 [[TMP2]]) 1150 // CHECK1-NEXT: [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 1151 // CHECK1-NEXT: br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]] 1152 // CHECK1: then: 1153 // CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 0 1154 // CHECK1-NEXT: [[TMP6:%.*]] = load i8*, i8** [[TMP5]], align 8 1155 // CHECK1-NEXT: [[TMP7:%.*]] = bitcast i8* [[TMP6]] to i32* 1156 // CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 1157 // CHECK1-NEXT: [[TMP9:%.*]] = load i32, i32* [[TMP7]], align 4 1158 // CHECK1-NEXT: store volatile i32 [[TMP9]], i32 addrspace(3)* [[TMP8]], align 4 1159 // CHECK1-NEXT: br label [[IFCONT:%.*]] 1160 // CHECK1: else: 1161 // CHECK1-NEXT: br label [[IFCONT]] 1162 // CHECK1: ifcont: 1163 // CHECK1-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 1164 // CHECK1-NEXT: [[TMP10:%.*]] = load i32, i32* [[DOTADDR1]], align 4 1165 // CHECK1-NEXT: [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP10]] 1166 // CHECK1-NEXT: br i1 [[IS_ACTIVE_THREAD]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 1167 // CHECK1: then4: 1168 // CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 1169 // CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 0 1170 // CHECK1-NEXT: [[TMP13:%.*]] = load i8*, i8** [[TMP12]], align 8 1171 // CHECK1-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to i32* 1172 // CHECK1-NEXT: [[TMP15:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP11]], align 4 1173 // CHECK1-NEXT: store i32 [[TMP15]], i32* [[TMP14]], align 4 1174 // CHECK1-NEXT: br label [[IFCONT6:%.*]] 1175 // CHECK1: else5: 1176 // CHECK1-NEXT: br label [[IFCONT6]] 1177 // CHECK1: ifcont6: 1178 // CHECK1-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 1179 // CHECK1-NEXT: [[WARP_MASTER7:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 1180 // CHECK1-NEXT: br i1 [[WARP_MASTER7]], label [[THEN8:%.*]], label [[ELSE9:%.*]] 1181 // CHECK1: then8: 1182 // CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 1 1183 // CHECK1-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 8 1184 // CHECK1-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i16* 1185 // CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 1186 // CHECK1-NEXT: [[TMP20:%.*]] = bitcast i32 addrspace(3)* [[TMP19]] to i16 addrspace(3)* 1187 // CHECK1-NEXT: [[TMP21:%.*]] = load i16, i16* [[TMP18]], align 2 1188 // CHECK1-NEXT: store volatile i16 [[TMP21]], i16 addrspace(3)* [[TMP20]], align 2 1189 // CHECK1-NEXT: br label [[IFCONT10:%.*]] 1190 // CHECK1: else9: 1191 // CHECK1-NEXT: br label [[IFCONT10]] 1192 // CHECK1: ifcont10: 1193 // CHECK1-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 1194 // CHECK1-NEXT: [[TMP22:%.*]] = load i32, i32* [[DOTADDR1]], align 4 1195 // CHECK1-NEXT: [[IS_ACTIVE_THREAD11:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP22]] 1196 // CHECK1-NEXT: br i1 [[IS_ACTIVE_THREAD11]], label [[THEN12:%.*]], label [[ELSE13:%.*]] 1197 // CHECK1: then12: 1198 // CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 1199 // CHECK1-NEXT: [[TMP24:%.*]] = bitcast i32 addrspace(3)* [[TMP23]] to i16 addrspace(3)* 1200 // CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 1 1201 // CHECK1-NEXT: [[TMP26:%.*]] = load i8*, i8** [[TMP25]], align 8 1202 // CHECK1-NEXT: [[TMP27:%.*]] = bitcast i8* [[TMP26]] to i16* 1203 // CHECK1-NEXT: [[TMP28:%.*]] = load volatile i16, i16 addrspace(3)* [[TMP24]], align 2 1204 // CHECK1-NEXT: store i16 [[TMP28]], i16* [[TMP27]], align 2 1205 // CHECK1-NEXT: br label [[IFCONT14:%.*]] 1206 // CHECK1: else13: 1207 // CHECK1-NEXT: br label [[IFCONT14]] 1208 // CHECK1: ifcont14: 1209 // CHECK1-NEXT: ret void 1210 // 1211 // 1212 // CHECK1-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func17 1213 // CHECK1-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR0]] { 1214 // CHECK1-NEXT: entry: 1215 // CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 1216 // CHECK1-NEXT: [[DOTADDR1:%.*]] = alloca i16, align 2 1217 // CHECK1-NEXT: [[DOTADDR2:%.*]] = alloca i16, align 2 1218 // CHECK1-NEXT: [[DOTADDR3:%.*]] = alloca i16, align 2 1219 // CHECK1-NEXT: [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [2 x i8*], align 8 1220 // CHECK1-NEXT: [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca i32, align 4 1221 // CHECK1-NEXT: [[DOTOMP_REDUCTION_ELEMENT4:%.*]] = alloca i16, align 2 1222 // CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 1223 // CHECK1-NEXT: store i16 [[TMP1]], i16* [[DOTADDR1]], align 2 1224 // CHECK1-NEXT: store i16 [[TMP2]], i16* [[DOTADDR2]], align 2 1225 // CHECK1-NEXT: store i16 [[TMP3]], i16* [[DOTADDR3]], align 2 1226 // CHECK1-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 8 1227 // CHECK1-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [2 x i8*]* 1228 // CHECK1-NEXT: [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2 1229 // CHECK1-NEXT: [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2 1230 // CHECK1-NEXT: [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2 1231 // CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i64 0, i64 0 1232 // CHECK1-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8 1233 // CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0 1234 // CHECK1-NEXT: [[TMP12:%.*]] = bitcast i8* [[TMP10]] to i32* 1235 // CHECK1-NEXT: [[TMP13:%.*]] = getelementptr i32, i32* [[TMP12]], i64 1 1236 // CHECK1-NEXT: [[TMP14:%.*]] = bitcast i32* [[TMP13]] to i8* 1237 // CHECK1-NEXT: [[TMP15:%.*]] = load i32, i32* [[TMP12]], align 4 1238 // CHECK1-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 1239 // CHECK1-NEXT: [[TMP16:%.*]] = trunc i32 [[NVPTX_WARP_SIZE]] to i16 1240 // CHECK1-NEXT: [[TMP17:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP15]], i16 [[TMP7]], i16 [[TMP16]]) 1241 // CHECK1-NEXT: store i32 [[TMP17]], i32* [[DOTOMP_REDUCTION_ELEMENT]], align 4 1242 // CHECK1-NEXT: [[TMP18:%.*]] = getelementptr i32, i32* [[TMP12]], i64 1 1243 // CHECK1-NEXT: [[TMP19:%.*]] = getelementptr i32, i32* [[DOTOMP_REDUCTION_ELEMENT]], i64 1 1244 // CHECK1-NEXT: [[TMP20:%.*]] = bitcast i32* [[DOTOMP_REDUCTION_ELEMENT]] to i8* 1245 // CHECK1-NEXT: store i8* [[TMP20]], i8** [[TMP11]], align 8 1246 // CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i64 0, i64 1 1247 // CHECK1-NEXT: [[TMP22:%.*]] = load i8*, i8** [[TMP21]], align 8 1248 // CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 1 1249 // CHECK1-NEXT: [[TMP24:%.*]] = bitcast i8* [[TMP22]] to i16* 1250 // CHECK1-NEXT: [[TMP25:%.*]] = getelementptr i16, i16* [[TMP24]], i64 1 1251 // CHECK1-NEXT: [[TMP26:%.*]] = bitcast i16* [[TMP25]] to i8* 1252 // CHECK1-NEXT: [[TMP27:%.*]] = load i16, i16* [[TMP24]], align 2 1253 // CHECK1-NEXT: [[TMP28:%.*]] = sext i16 [[TMP27]] to i32 1254 // CHECK1-NEXT: [[NVPTX_WARP_SIZE5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 1255 // CHECK1-NEXT: [[TMP29:%.*]] = trunc i32 [[NVPTX_WARP_SIZE5]] to i16 1256 // CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP28]], i16 [[TMP7]], i16 [[TMP29]]) 1257 // CHECK1-NEXT: [[TMP31:%.*]] = trunc i32 [[TMP30]] to i16 1258 // CHECK1-NEXT: store i16 [[TMP31]], i16* [[DOTOMP_REDUCTION_ELEMENT4]], align 2 1259 // CHECK1-NEXT: [[TMP32:%.*]] = getelementptr i16, i16* [[TMP24]], i64 1 1260 // CHECK1-NEXT: [[TMP33:%.*]] = getelementptr i16, i16* [[DOTOMP_REDUCTION_ELEMENT4]], i64 1 1261 // CHECK1-NEXT: [[TMP34:%.*]] = bitcast i16* [[DOTOMP_REDUCTION_ELEMENT4]] to i8* 1262 // CHECK1-NEXT: store i8* [[TMP34]], i8** [[TMP23]], align 8 1263 // CHECK1-NEXT: [[TMP35:%.*]] = icmp eq i16 [[TMP8]], 0 1264 // CHECK1-NEXT: [[TMP36:%.*]] = icmp eq i16 [[TMP8]], 1 1265 // CHECK1-NEXT: [[TMP37:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]] 1266 // CHECK1-NEXT: [[TMP38:%.*]] = and i1 [[TMP36]], [[TMP37]] 1267 // CHECK1-NEXT: [[TMP39:%.*]] = icmp eq i16 [[TMP8]], 2 1268 // CHECK1-NEXT: [[TMP40:%.*]] = and i16 [[TMP6]], 1 1269 // CHECK1-NEXT: [[TMP41:%.*]] = icmp eq i16 [[TMP40]], 0 1270 // CHECK1-NEXT: [[TMP42:%.*]] = and i1 [[TMP39]], [[TMP41]] 1271 // CHECK1-NEXT: [[TMP43:%.*]] = icmp sgt i16 [[TMP7]], 0 1272 // CHECK1-NEXT: [[TMP44:%.*]] = and i1 [[TMP42]], [[TMP43]] 1273 // CHECK1-NEXT: [[TMP45:%.*]] = or i1 [[TMP35]], [[TMP38]] 1274 // CHECK1-NEXT: [[TMP46:%.*]] = or i1 [[TMP45]], [[TMP44]] 1275 // CHECK1-NEXT: br i1 [[TMP46]], label [[THEN:%.*]], label [[ELSE:%.*]] 1276 // CHECK1: then: 1277 // CHECK1-NEXT: [[TMP47:%.*]] = bitcast [2 x i8*]* [[TMP5]] to i8* 1278 // CHECK1-NEXT: [[TMP48:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8* 1279 // CHECK1-NEXT: call void @"_omp$reduction$reduction_func16"(i8* [[TMP47]], i8* [[TMP48]]) #[[ATTR3]] 1280 // CHECK1-NEXT: br label [[IFCONT:%.*]] 1281 // CHECK1: else: 1282 // CHECK1-NEXT: br label [[IFCONT]] 1283 // CHECK1: ifcont: 1284 // CHECK1-NEXT: [[TMP49:%.*]] = icmp eq i16 [[TMP8]], 1 1285 // CHECK1-NEXT: [[TMP50:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]] 1286 // CHECK1-NEXT: [[TMP51:%.*]] = and i1 [[TMP49]], [[TMP50]] 1287 // CHECK1-NEXT: br i1 [[TMP51]], label [[THEN6:%.*]], label [[ELSE7:%.*]] 1288 // CHECK1: then6: 1289 // CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0 1290 // CHECK1-NEXT: [[TMP53:%.*]] = load i8*, i8** [[TMP52]], align 8 1291 // CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i64 0, i64 0 1292 // CHECK1-NEXT: [[TMP55:%.*]] = load i8*, i8** [[TMP54]], align 8 1293 // CHECK1-NEXT: [[TMP56:%.*]] = bitcast i8* [[TMP53]] to i32* 1294 // CHECK1-NEXT: [[TMP57:%.*]] = bitcast i8* [[TMP55]] to i32* 1295 // CHECK1-NEXT: [[TMP58:%.*]] = load i32, i32* [[TMP56]], align 4 1296 // CHECK1-NEXT: store i32 [[TMP58]], i32* [[TMP57]], align 4 1297 // CHECK1-NEXT: [[TMP59:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 1 1298 // CHECK1-NEXT: [[TMP60:%.*]] = load i8*, i8** [[TMP59]], align 8 1299 // CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i64 0, i64 1 1300 // CHECK1-NEXT: [[TMP62:%.*]] = load i8*, i8** [[TMP61]], align 8 1301 // CHECK1-NEXT: [[TMP63:%.*]] = bitcast i8* [[TMP60]] to i16* 1302 // CHECK1-NEXT: [[TMP64:%.*]] = bitcast i8* [[TMP62]] to i16* 1303 // CHECK1-NEXT: [[TMP65:%.*]] = load i16, i16* [[TMP63]], align 2 1304 // CHECK1-NEXT: store i16 [[TMP65]], i16* [[TMP64]], align 2 1305 // CHECK1-NEXT: br label [[IFCONT8:%.*]] 1306 // CHECK1: else7: 1307 // CHECK1-NEXT: br label [[IFCONT8]] 1308 // CHECK1: ifcont8: 1309 // CHECK1-NEXT: ret void 1310 // 1311 // 1312 // CHECK1-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func18 1313 // CHECK1-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] { 1314 // CHECK1-NEXT: entry: 1315 // CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 1316 // CHECK1-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 1317 // CHECK1-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]]) 1318 // CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 1319 // CHECK1-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 1320 // CHECK1-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 1321 // CHECK1-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 1322 // CHECK1-NEXT: [[NVPTX_LANE_ID:%.*]] = and i32 [[NVPTX_TID2]], 31 1323 // CHECK1-NEXT: [[NVPTX_TID3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 1324 // CHECK1-NEXT: [[NVPTX_WARP_ID:%.*]] = ashr i32 [[NVPTX_TID3]], 5 1325 // CHECK1-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8 1326 // CHECK1-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 1327 // CHECK1-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 1328 // CHECK1-NEXT: [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 1329 // CHECK1-NEXT: br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]] 1330 // CHECK1: then: 1331 // CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 0 1332 // CHECK1-NEXT: [[TMP6:%.*]] = load i8*, i8** [[TMP5]], align 8 1333 // CHECK1-NEXT: [[TMP7:%.*]] = bitcast i8* [[TMP6]] to i32* 1334 // CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 1335 // CHECK1-NEXT: [[TMP9:%.*]] = load i32, i32* [[TMP7]], align 4 1336 // CHECK1-NEXT: store volatile i32 [[TMP9]], i32 addrspace(3)* [[TMP8]], align 4 1337 // CHECK1-NEXT: br label [[IFCONT:%.*]] 1338 // CHECK1: else: 1339 // CHECK1-NEXT: br label [[IFCONT]] 1340 // CHECK1: ifcont: 1341 // CHECK1-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 1342 // CHECK1-NEXT: [[TMP10:%.*]] = load i32, i32* [[DOTADDR1]], align 4 1343 // CHECK1-NEXT: [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP10]] 1344 // CHECK1-NEXT: br i1 [[IS_ACTIVE_THREAD]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 1345 // CHECK1: then4: 1346 // CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 1347 // CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 0 1348 // CHECK1-NEXT: [[TMP13:%.*]] = load i8*, i8** [[TMP12]], align 8 1349 // CHECK1-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to i32* 1350 // CHECK1-NEXT: [[TMP15:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP11]], align 4 1351 // CHECK1-NEXT: store i32 [[TMP15]], i32* [[TMP14]], align 4 1352 // CHECK1-NEXT: br label [[IFCONT6:%.*]] 1353 // CHECK1: else5: 1354 // CHECK1-NEXT: br label [[IFCONT6]] 1355 // CHECK1: ifcont6: 1356 // CHECK1-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 1357 // CHECK1-NEXT: [[WARP_MASTER7:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 1358 // CHECK1-NEXT: br i1 [[WARP_MASTER7]], label [[THEN8:%.*]], label [[ELSE9:%.*]] 1359 // CHECK1: then8: 1360 // CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 1 1361 // CHECK1-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 8 1362 // CHECK1-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i16* 1363 // CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 1364 // CHECK1-NEXT: [[TMP20:%.*]] = bitcast i32 addrspace(3)* [[TMP19]] to i16 addrspace(3)* 1365 // CHECK1-NEXT: [[TMP21:%.*]] = load i16, i16* [[TMP18]], align 2 1366 // CHECK1-NEXT: store volatile i16 [[TMP21]], i16 addrspace(3)* [[TMP20]], align 2 1367 // CHECK1-NEXT: br label [[IFCONT10:%.*]] 1368 // CHECK1: else9: 1369 // CHECK1-NEXT: br label [[IFCONT10]] 1370 // CHECK1: ifcont10: 1371 // CHECK1-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 1372 // CHECK1-NEXT: [[TMP22:%.*]] = load i32, i32* [[DOTADDR1]], align 4 1373 // CHECK1-NEXT: [[IS_ACTIVE_THREAD11:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP22]] 1374 // CHECK1-NEXT: br i1 [[IS_ACTIVE_THREAD11]], label [[THEN12:%.*]], label [[ELSE13:%.*]] 1375 // CHECK1: then12: 1376 // CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 1377 // CHECK1-NEXT: [[TMP24:%.*]] = bitcast i32 addrspace(3)* [[TMP23]] to i16 addrspace(3)* 1378 // CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 1 1379 // CHECK1-NEXT: [[TMP26:%.*]] = load i8*, i8** [[TMP25]], align 8 1380 // CHECK1-NEXT: [[TMP27:%.*]] = bitcast i8* [[TMP26]] to i16* 1381 // CHECK1-NEXT: [[TMP28:%.*]] = load volatile i16, i16 addrspace(3)* [[TMP24]], align 2 1382 // CHECK1-NEXT: store i16 [[TMP28]], i16* [[TMP27]], align 2 1383 // CHECK1-NEXT: br label [[IFCONT14:%.*]] 1384 // CHECK1: else13: 1385 // CHECK1-NEXT: br label [[IFCONT14]] 1386 // CHECK1: ifcont14: 1387 // CHECK1-NEXT: ret void 1388 // 1389 // 1390 // CHECK1-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_copy_func19 1391 // CHECK1-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 1392 // CHECK1-NEXT: entry: 1393 // CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 1394 // CHECK1-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 1395 // CHECK1-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 8 1396 // CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 1397 // CHECK1-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 1398 // CHECK1-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 8 1399 // CHECK1-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 8 1400 // CHECK1-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 1401 // CHECK1-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 8 1402 // CHECK1-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.5* 1403 // CHECK1-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 1404 // CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 0 1405 // CHECK1-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 8 1406 // CHECK1-NEXT: [[TMP10:%.*]] = bitcast i8* [[TMP9]] to i32* 1407 // CHECK1-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_5:%.*]], %struct._globalized_locals_ty.5* [[TMP6]], i32 0, i32 0 1408 // CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [1024 x i32], [1024 x i32]* [[A]], i32 0, i32 [[TMP7]] 1409 // CHECK1-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP10]], align 4 1410 // CHECK1-NEXT: store i32 [[TMP12]], i32* [[TMP11]], align 128 1411 // CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 1 1412 // CHECK1-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 8 1413 // CHECK1-NEXT: [[TMP15:%.*]] = bitcast i8* [[TMP14]] to i16* 1414 // CHECK1-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_5]], %struct._globalized_locals_ty.5* [[TMP6]], i32 0, i32 1 1415 // CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1024 x i16], [1024 x i16]* [[B]], i32 0, i32 [[TMP7]] 1416 // CHECK1-NEXT: [[TMP17:%.*]] = load i16, i16* [[TMP15]], align 2 1417 // CHECK1-NEXT: store i16 [[TMP17]], i16* [[TMP16]], align 128 1418 // CHECK1-NEXT: ret void 1419 // 1420 // 1421 // CHECK1-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_reduce_func20 1422 // CHECK1-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 1423 // CHECK1-NEXT: entry: 1424 // CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 1425 // CHECK1-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 1426 // CHECK1-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 8 1427 // CHECK1-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 8 1428 // CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 1429 // CHECK1-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 1430 // CHECK1-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 8 1431 // CHECK1-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8 1432 // CHECK1-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.5* 1433 // CHECK1-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 1434 // CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0 1435 // CHECK1-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_5:%.*]], %struct._globalized_locals_ty.5* [[TMP4]], i32 0, i32 0 1436 // CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1024 x i32], [1024 x i32]* [[A]], i32 0, i32 [[TMP5]] 1437 // CHECK1-NEXT: [[TMP8:%.*]] = bitcast i32* [[TMP7]] to i8* 1438 // CHECK1-NEXT: store i8* [[TMP8]], i8** [[TMP6]], align 8 1439 // CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 1 1440 // CHECK1-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_5]], %struct._globalized_locals_ty.5* [[TMP4]], i32 0, i32 1 1441 // CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [1024 x i16], [1024 x i16]* [[B]], i32 0, i32 [[TMP5]] 1442 // CHECK1-NEXT: [[TMP11:%.*]] = bitcast i16* [[TMP10]] to i8* 1443 // CHECK1-NEXT: store i8* [[TMP11]], i8** [[TMP9]], align 8 1444 // CHECK1-NEXT: [[TMP12:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 1445 // CHECK1-NEXT: [[TMP13:%.*]] = load i8*, i8** [[DOTADDR2]], align 8 1446 // CHECK1-NEXT: call void @"_omp$reduction$reduction_func16"(i8* [[TMP12]], i8* [[TMP13]]) #[[ATTR3]] 1447 // CHECK1-NEXT: ret void 1448 // 1449 // 1450 // CHECK1-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_copy_func21 1451 // CHECK1-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 1452 // CHECK1-NEXT: entry: 1453 // CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 1454 // CHECK1-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 1455 // CHECK1-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 8 1456 // CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 1457 // CHECK1-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 1458 // CHECK1-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 8 1459 // CHECK1-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 8 1460 // CHECK1-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 1461 // CHECK1-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 8 1462 // CHECK1-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.5* 1463 // CHECK1-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 1464 // CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 0 1465 // CHECK1-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 8 1466 // CHECK1-NEXT: [[TMP10:%.*]] = bitcast i8* [[TMP9]] to i32* 1467 // CHECK1-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_5:%.*]], %struct._globalized_locals_ty.5* [[TMP6]], i32 0, i32 0 1468 // CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [1024 x i32], [1024 x i32]* [[A]], i32 0, i32 [[TMP7]] 1469 // CHECK1-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 128 1470 // CHECK1-NEXT: store i32 [[TMP12]], i32* [[TMP10]], align 4 1471 // CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 1 1472 // CHECK1-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 8 1473 // CHECK1-NEXT: [[TMP15:%.*]] = bitcast i8* [[TMP14]] to i16* 1474 // CHECK1-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_5]], %struct._globalized_locals_ty.5* [[TMP6]], i32 0, i32 1 1475 // CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1024 x i16], [1024 x i16]* [[B]], i32 0, i32 [[TMP7]] 1476 // CHECK1-NEXT: [[TMP17:%.*]] = load i16, i16* [[TMP16]], align 128 1477 // CHECK1-NEXT: store i16 [[TMP17]], i16* [[TMP15]], align 2 1478 // CHECK1-NEXT: ret void 1479 // 1480 // 1481 // CHECK1-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_reduce_func22 1482 // CHECK1-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 1483 // CHECK1-NEXT: entry: 1484 // CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 1485 // CHECK1-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 1486 // CHECK1-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 8 1487 // CHECK1-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 8 1488 // CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 1489 // CHECK1-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 1490 // CHECK1-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 8 1491 // CHECK1-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8 1492 // CHECK1-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.5* 1493 // CHECK1-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 1494 // CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0 1495 // CHECK1-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_5:%.*]], %struct._globalized_locals_ty.5* [[TMP4]], i32 0, i32 0 1496 // CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1024 x i32], [1024 x i32]* [[A]], i32 0, i32 [[TMP5]] 1497 // CHECK1-NEXT: [[TMP8:%.*]] = bitcast i32* [[TMP7]] to i8* 1498 // CHECK1-NEXT: store i8* [[TMP8]], i8** [[TMP6]], align 8 1499 // CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 1 1500 // CHECK1-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_5]], %struct._globalized_locals_ty.5* [[TMP4]], i32 0, i32 1 1501 // CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [1024 x i16], [1024 x i16]* [[B]], i32 0, i32 [[TMP5]] 1502 // CHECK1-NEXT: [[TMP11:%.*]] = bitcast i16* [[TMP10]] to i8* 1503 // CHECK1-NEXT: store i8* [[TMP11]], i8** [[TMP9]], align 8 1504 // CHECK1-NEXT: [[TMP12:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 1505 // CHECK1-NEXT: [[TMP13:%.*]] = load i8*, i8** [[DOTADDR2]], align 8 1506 // CHECK1-NEXT: call void @"_omp$reduction$reduction_func16"(i8* [[TMP13]], i8* [[TMP12]]) #[[ATTR3]] 1507 // CHECK1-NEXT: ret void 1508 // 1509 // 1510 // CHECK2-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l23_worker 1511 // CHECK2-SAME: () #[[ATTR0:[0-9]+]] { 1512 // CHECK2-NEXT: entry: 1513 // CHECK2-NEXT: [[WORK_FN:%.*]] = alloca i8*, align 8 1514 // CHECK2-NEXT: [[EXEC_STATUS:%.*]] = alloca i8, align 1 1515 // CHECK2-NEXT: store i8* null, i8** [[WORK_FN]], align 8 1516 // CHECK2-NEXT: store i8 0, i8* [[EXEC_STATUS]], align 1 1517 // CHECK2-NEXT: br label [[DOTAWAIT_WORK:%.*]] 1518 // CHECK2: .await.work: 1519 // CHECK2-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 1520 // CHECK2-NEXT: [[TMP0:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORK_FN]]) 1521 // CHECK2-NEXT: [[TMP1:%.*]] = zext i1 [[TMP0]] to i8 1522 // CHECK2-NEXT: store i8 [[TMP1]], i8* [[EXEC_STATUS]], align 1 1523 // CHECK2-NEXT: [[TMP2:%.*]] = load i8*, i8** [[WORK_FN]], align 8 1524 // CHECK2-NEXT: [[SHOULD_TERMINATE:%.*]] = icmp eq i8* [[TMP2]], null 1525 // CHECK2-NEXT: br i1 [[SHOULD_TERMINATE]], label [[DOTEXIT:%.*]], label [[DOTSELECT_WORKERS:%.*]] 1526 // CHECK2: .select.workers: 1527 // CHECK2-NEXT: [[TMP3:%.*]] = load i8, i8* [[EXEC_STATUS]], align 1 1528 // CHECK2-NEXT: [[IS_ACTIVE:%.*]] = icmp ne i8 [[TMP3]], 0 1529 // CHECK2-NEXT: br i1 [[IS_ACTIVE]], label [[DOTEXECUTE_PARALLEL:%.*]], label [[DOTBARRIER_PARALLEL:%.*]] 1530 // CHECK2: .execute.parallel: 1531 // CHECK2-NEXT: [[TMP4:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]]) 1532 // CHECK2-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP2]] to void (i16, i32)* 1533 // CHECK2-NEXT: call void [[TMP5]](i16 0, i32 [[TMP4]]) 1534 // CHECK2-NEXT: br label [[DOTTERMINATE_PARALLEL:%.*]] 1535 // CHECK2: .terminate.parallel: 1536 // CHECK2-NEXT: call void @__kmpc_kernel_end_parallel() 1537 // CHECK2-NEXT: br label [[DOTBARRIER_PARALLEL]] 1538 // CHECK2: .barrier.parallel: 1539 // CHECK2-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 1540 // CHECK2-NEXT: br label [[DOTAWAIT_WORK]] 1541 // CHECK2: .exit: 1542 // CHECK2-NEXT: ret void 1543 // 1544 // 1545 // CHECK2-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l23 1546 // CHECK2-SAME: (i64 [[E:%.*]]) #[[ATTR1:[0-9]+]] { 1547 // CHECK2-NEXT: entry: 1548 // CHECK2-NEXT: [[E_ADDR:%.*]] = alloca i64, align 8 1549 // CHECK2-NEXT: [[DOTZERO_ADDR:%.*]] = alloca i32, align 4 1550 // CHECK2-NEXT: [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4 1551 // CHECK2-NEXT: store i32 0, i32* [[DOTZERO_ADDR]], align 4 1552 // CHECK2-NEXT: store i64 [[E]], i64* [[E_ADDR]], align 8 1553 // CHECK2-NEXT: [[CONV:%.*]] = bitcast i64* [[E_ADDR]] to double* 1554 // CHECK2-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 1555 // CHECK2-NEXT: [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 1556 // CHECK2-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 1557 // CHECK2-NEXT: [[THREAD_LIMIT:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS]], [[NVPTX_WARP_SIZE]] 1558 // CHECK2-NEXT: [[TMP0:%.*]] = icmp ult i32 [[NVPTX_TID]], [[THREAD_LIMIT]] 1559 // CHECK2-NEXT: br i1 [[TMP0]], label [[DOTWORKER:%.*]], label [[DOTMASTERCHECK:%.*]] 1560 // CHECK2: .worker: 1561 // CHECK2-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l23_worker() #[[ATTR3:[0-9]+]] 1562 // CHECK2-NEXT: br label [[DOTEXIT:%.*]] 1563 // CHECK2: .mastercheck: 1564 // CHECK2-NEXT: [[NVPTX_TID1:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 1565 // CHECK2-NEXT: [[NVPTX_NUM_THREADS2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 1566 // CHECK2-NEXT: [[NVPTX_WARP_SIZE3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 1567 // CHECK2-NEXT: [[TMP1:%.*]] = sub nuw i32 [[NVPTX_WARP_SIZE3]], 1 1568 // CHECK2-NEXT: [[TMP2:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS2]], 1 1569 // CHECK2-NEXT: [[TMP3:%.*]] = xor i32 [[TMP1]], -1 1570 // CHECK2-NEXT: [[MASTER_TID:%.*]] = and i32 [[TMP2]], [[TMP3]] 1571 // CHECK2-NEXT: [[TMP4:%.*]] = icmp eq i32 [[NVPTX_TID1]], [[MASTER_TID]] 1572 // CHECK2-NEXT: br i1 [[TMP4]], label [[DOTMASTER:%.*]], label [[DOTEXIT]] 1573 // CHECK2: .master: 1574 // CHECK2-NEXT: [[NVPTX_NUM_THREADS4:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 1575 // CHECK2-NEXT: [[NVPTX_WARP_SIZE5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 1576 // CHECK2-NEXT: [[THREAD_LIMIT6:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS4]], [[NVPTX_WARP_SIZE5]] 1577 // CHECK2-NEXT: call void @__kmpc_kernel_init(i32 [[THREAD_LIMIT6]], i16 1) 1578 // CHECK2-NEXT: call void @__kmpc_data_sharing_init_stack() 1579 // CHECK2-NEXT: [[TMP5:%.*]] = call i8* @__kmpc_data_sharing_push_stack(i64 8, i16 1) 1580 // CHECK2-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty* 1581 // CHECK2-NEXT: [[TMP7:%.*]] = load double, double* [[CONV]], align 8 1582 // CHECK2-NEXT: [[E7:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY:%.*]], %struct._globalized_locals_ty* [[TMP6]], i32 0, i32 0 1583 // CHECK2-NEXT: store double [[TMP7]], double* [[E7]], align 8 1584 // CHECK2-NEXT: [[TMP8:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 1585 // CHECK2-NEXT: store i32 [[TMP8]], i32* [[DOTTHREADID_TEMP_]], align 4 1586 // CHECK2-NEXT: call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], double* [[E7]]) #[[ATTR3]] 1587 // CHECK2-NEXT: call void @__kmpc_data_sharing_pop_stack(i8* [[TMP5]]) 1588 // CHECK2-NEXT: br label [[DOTTERMINATION_NOTIFIER:%.*]] 1589 // CHECK2: .termination.notifier: 1590 // CHECK2-NEXT: call void @__kmpc_kernel_deinit(i16 1) 1591 // CHECK2-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 1592 // CHECK2-NEXT: br label [[DOTEXIT]] 1593 // CHECK2: .exit: 1594 // CHECK2-NEXT: ret void 1595 // 1596 // 1597 // CHECK2-LABEL: define {{[^@]+}}@__omp_outlined__ 1598 // CHECK2-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], double* nonnull align 8 dereferenceable(8) [[E:%.*]]) #[[ATTR1]] { 1599 // CHECK2-NEXT: entry: 1600 // CHECK2-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8 1601 // CHECK2-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8 1602 // CHECK2-NEXT: [[E_ADDR:%.*]] = alloca double*, align 8 1603 // CHECK2-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x i8*], align 8 1604 // CHECK2-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8 1605 // CHECK2-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8 1606 // CHECK2-NEXT: store double* [[E]], double** [[E_ADDR]], align 8 1607 // CHECK2-NEXT: [[TMP0:%.*]] = load double*, double** [[E_ADDR]], align 8 1608 // CHECK2-NEXT: [[TMP1:%.*]] = call i8* @__kmpc_data_sharing_push_stack(i64 8, i16 1) 1609 // CHECK2-NEXT: [[TMP2:%.*]] = bitcast i8* [[TMP1]] to %struct._globalized_locals_ty.0* 1610 // CHECK2-NEXT: [[E1:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_0:%.*]], %struct._globalized_locals_ty.0* [[TMP2]], i32 0, i32 0 1611 // CHECK2-NEXT: store double 0.000000e+00, double* [[E1]], align 8 1612 // CHECK2-NEXT: [[TMP3:%.*]] = load double, double* [[E1]], align 8 1613 // CHECK2-NEXT: [[ADD:%.*]] = fadd double [[TMP3]], 5.000000e+00 1614 // CHECK2-NEXT: store double [[ADD]], double* [[E1]], align 8 1615 // CHECK2-NEXT: [[TMP4:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8 1616 // CHECK2-NEXT: [[TMP5:%.*]] = load i32, i32* [[TMP4]], align 4 1617 // CHECK2-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0 1618 // CHECK2-NEXT: [[TMP7:%.*]] = bitcast double* [[E1]] to i8* 1619 // CHECK2-NEXT: store i8* [[TMP7]], i8** [[TMP6]], align 8 1620 // CHECK2-NEXT: [[TMP8:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 1621 // CHECK2-NEXT: [[TMP9:%.*]] = load i8*, i8** @"_openmp_teams_reductions_buffer_$_$ptr", align 8 1622 // CHECK2-NEXT: [[TMP10:%.*]] = call i32 @__kmpc_nvptx_teams_reduce_nowait_v2(%struct.ident_t* @[[GLOB1]], i32 [[TMP5]], i8* [[TMP9]], i32 1024, i8* [[TMP8]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_copy_func, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_reduce_func, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_copy_func, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_reduce_func) 1623 // CHECK2-NEXT: [[TMP11:%.*]] = icmp eq i32 [[TMP10]], 1 1624 // CHECK2-NEXT: br i1 [[TMP11]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]] 1625 // CHECK2: .omp.reduction.then: 1626 // CHECK2-NEXT: [[TMP12:%.*]] = load double, double* [[TMP0]], align 8 1627 // CHECK2-NEXT: [[TMP13:%.*]] = load double, double* [[E1]], align 8 1628 // CHECK2-NEXT: [[ADD2:%.*]] = fadd double [[TMP12]], [[TMP13]] 1629 // CHECK2-NEXT: store double [[ADD2]], double* [[TMP0]], align 8 1630 // CHECK2-NEXT: call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP5]]) 1631 // CHECK2-NEXT: br label [[DOTOMP_REDUCTION_DONE]] 1632 // CHECK2: .omp.reduction.done: 1633 // CHECK2-NEXT: call void @__kmpc_data_sharing_pop_stack(i8* [[TMP1]]) 1634 // CHECK2-NEXT: ret void 1635 // 1636 // 1637 // CHECK2-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func 1638 // CHECK2-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR0]] { 1639 // CHECK2-NEXT: entry: 1640 // CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 1641 // CHECK2-NEXT: [[DOTADDR1:%.*]] = alloca i16, align 2 1642 // CHECK2-NEXT: [[DOTADDR2:%.*]] = alloca i16, align 2 1643 // CHECK2-NEXT: [[DOTADDR3:%.*]] = alloca i16, align 2 1644 // CHECK2-NEXT: [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [1 x i8*], align 8 1645 // CHECK2-NEXT: [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca double, align 8 1646 // CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 1647 // CHECK2-NEXT: store i16 [[TMP1]], i16* [[DOTADDR1]], align 2 1648 // CHECK2-NEXT: store i16 [[TMP2]], i16* [[DOTADDR2]], align 2 1649 // CHECK2-NEXT: store i16 [[TMP3]], i16* [[DOTADDR3]], align 2 1650 // CHECK2-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 8 1651 // CHECK2-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [1 x i8*]* 1652 // CHECK2-NEXT: [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2 1653 // CHECK2-NEXT: [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2 1654 // CHECK2-NEXT: [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2 1655 // CHECK2-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i64 0, i64 0 1656 // CHECK2-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8 1657 // CHECK2-NEXT: [[TMP11:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0 1658 // CHECK2-NEXT: [[TMP12:%.*]] = bitcast i8* [[TMP10]] to double* 1659 // CHECK2-NEXT: [[TMP13:%.*]] = getelementptr double, double* [[TMP12]], i64 1 1660 // CHECK2-NEXT: [[TMP14:%.*]] = bitcast double* [[TMP13]] to i8* 1661 // CHECK2-NEXT: [[TMP15:%.*]] = bitcast double* [[TMP12]] to i64* 1662 // CHECK2-NEXT: [[TMP16:%.*]] = bitcast double* [[DOTOMP_REDUCTION_ELEMENT]] to i64* 1663 // CHECK2-NEXT: [[TMP17:%.*]] = load i64, i64* [[TMP15]], align 8 1664 // CHECK2-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 1665 // CHECK2-NEXT: [[TMP18:%.*]] = trunc i32 [[NVPTX_WARP_SIZE]] to i16 1666 // CHECK2-NEXT: [[TMP19:%.*]] = call i64 @__kmpc_shuffle_int64(i64 [[TMP17]], i16 [[TMP7]], i16 [[TMP18]]) 1667 // CHECK2-NEXT: store i64 [[TMP19]], i64* [[TMP16]], align 8 1668 // CHECK2-NEXT: [[TMP20:%.*]] = getelementptr i64, i64* [[TMP15]], i64 1 1669 // CHECK2-NEXT: [[TMP21:%.*]] = getelementptr i64, i64* [[TMP16]], i64 1 1670 // CHECK2-NEXT: [[TMP22:%.*]] = bitcast double* [[DOTOMP_REDUCTION_ELEMENT]] to i8* 1671 // CHECK2-NEXT: store i8* [[TMP22]], i8** [[TMP11]], align 8 1672 // CHECK2-NEXT: [[TMP23:%.*]] = icmp eq i16 [[TMP8]], 0 1673 // CHECK2-NEXT: [[TMP24:%.*]] = icmp eq i16 [[TMP8]], 1 1674 // CHECK2-NEXT: [[TMP25:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]] 1675 // CHECK2-NEXT: [[TMP26:%.*]] = and i1 [[TMP24]], [[TMP25]] 1676 // CHECK2-NEXT: [[TMP27:%.*]] = icmp eq i16 [[TMP8]], 2 1677 // CHECK2-NEXT: [[TMP28:%.*]] = and i16 [[TMP6]], 1 1678 // CHECK2-NEXT: [[TMP29:%.*]] = icmp eq i16 [[TMP28]], 0 1679 // CHECK2-NEXT: [[TMP30:%.*]] = and i1 [[TMP27]], [[TMP29]] 1680 // CHECK2-NEXT: [[TMP31:%.*]] = icmp sgt i16 [[TMP7]], 0 1681 // CHECK2-NEXT: [[TMP32:%.*]] = and i1 [[TMP30]], [[TMP31]] 1682 // CHECK2-NEXT: [[TMP33:%.*]] = or i1 [[TMP23]], [[TMP26]] 1683 // CHECK2-NEXT: [[TMP34:%.*]] = or i1 [[TMP33]], [[TMP32]] 1684 // CHECK2-NEXT: br i1 [[TMP34]], label [[THEN:%.*]], label [[ELSE:%.*]] 1685 // CHECK2: then: 1686 // CHECK2-NEXT: [[TMP35:%.*]] = bitcast [1 x i8*]* [[TMP5]] to i8* 1687 // CHECK2-NEXT: [[TMP36:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8* 1688 // CHECK2-NEXT: call void @"_omp$reduction$reduction_func"(i8* [[TMP35]], i8* [[TMP36]]) #[[ATTR3]] 1689 // CHECK2-NEXT: br label [[IFCONT:%.*]] 1690 // CHECK2: else: 1691 // CHECK2-NEXT: br label [[IFCONT]] 1692 // CHECK2: ifcont: 1693 // CHECK2-NEXT: [[TMP37:%.*]] = icmp eq i16 [[TMP8]], 1 1694 // CHECK2-NEXT: [[TMP38:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]] 1695 // CHECK2-NEXT: [[TMP39:%.*]] = and i1 [[TMP37]], [[TMP38]] 1696 // CHECK2-NEXT: br i1 [[TMP39]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 1697 // CHECK2: then4: 1698 // CHECK2-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0 1699 // CHECK2-NEXT: [[TMP41:%.*]] = load i8*, i8** [[TMP40]], align 8 1700 // CHECK2-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i64 0, i64 0 1701 // CHECK2-NEXT: [[TMP43:%.*]] = load i8*, i8** [[TMP42]], align 8 1702 // CHECK2-NEXT: [[TMP44:%.*]] = bitcast i8* [[TMP41]] to double* 1703 // CHECK2-NEXT: [[TMP45:%.*]] = bitcast i8* [[TMP43]] to double* 1704 // CHECK2-NEXT: [[TMP46:%.*]] = load double, double* [[TMP44]], align 8 1705 // CHECK2-NEXT: store double [[TMP46]], double* [[TMP45]], align 8 1706 // CHECK2-NEXT: br label [[IFCONT6:%.*]] 1707 // CHECK2: else5: 1708 // CHECK2-NEXT: br label [[IFCONT6]] 1709 // CHECK2: ifcont6: 1710 // CHECK2-NEXT: ret void 1711 // 1712 // 1713 // CHECK2-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func 1714 // CHECK2-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] { 1715 // CHECK2-NEXT: entry: 1716 // CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 1717 // CHECK2-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 1718 // CHECK2-NEXT: [[DOTCNT_ADDR:%.*]] = alloca i32, align 4 1719 // CHECK2-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 1720 // CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 1721 // CHECK2-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 1722 // CHECK2-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 1723 // CHECK2-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 1724 // CHECK2-NEXT: [[NVPTX_LANE_ID:%.*]] = and i32 [[NVPTX_TID2]], 31 1725 // CHECK2-NEXT: [[NVPTX_TID3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 1726 // CHECK2-NEXT: [[NVPTX_WARP_ID:%.*]] = ashr i32 [[NVPTX_TID3]], 5 1727 // CHECK2-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8 1728 // CHECK2-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [1 x i8*]* 1729 // CHECK2-NEXT: store i32 0, i32* [[DOTCNT_ADDR]], align 4 1730 // CHECK2-NEXT: br label [[PRECOND:%.*]] 1731 // CHECK2: precond: 1732 // CHECK2-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTCNT_ADDR]], align 4 1733 // CHECK2-NEXT: [[TMP6:%.*]] = icmp ult i32 [[TMP5]], 2 1734 // CHECK2-NEXT: br i1 [[TMP6]], label [[BODY:%.*]], label [[EXIT:%.*]] 1735 // CHECK2: body: 1736 // CHECK2-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP2]]) 1737 // CHECK2-NEXT: [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 1738 // CHECK2-NEXT: br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]] 1739 // CHECK2: then: 1740 // CHECK2-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP4]], i64 0, i64 0 1741 // CHECK2-NEXT: [[TMP8:%.*]] = load i8*, i8** [[TMP7]], align 8 1742 // CHECK2-NEXT: [[TMP9:%.*]] = bitcast i8* [[TMP8]] to i32* 1743 // CHECK2-NEXT: [[TMP10:%.*]] = getelementptr i32, i32* [[TMP9]], i32 [[TMP5]] 1744 // CHECK2-NEXT: [[TMP11:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 1745 // CHECK2-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP10]], align 4 1746 // CHECK2-NEXT: store volatile i32 [[TMP12]], i32 addrspace(3)* [[TMP11]], align 4 1747 // CHECK2-NEXT: br label [[IFCONT:%.*]] 1748 // CHECK2: else: 1749 // CHECK2-NEXT: br label [[IFCONT]] 1750 // CHECK2: ifcont: 1751 // CHECK2-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 1752 // CHECK2-NEXT: [[TMP13:%.*]] = load i32, i32* [[DOTADDR1]], align 4 1753 // CHECK2-NEXT: [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP13]] 1754 // CHECK2-NEXT: br i1 [[IS_ACTIVE_THREAD]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 1755 // CHECK2: then4: 1756 // CHECK2-NEXT: [[TMP14:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 1757 // CHECK2-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP4]], i64 0, i64 0 1758 // CHECK2-NEXT: [[TMP16:%.*]] = load i8*, i8** [[TMP15]], align 8 1759 // CHECK2-NEXT: [[TMP17:%.*]] = bitcast i8* [[TMP16]] to i32* 1760 // CHECK2-NEXT: [[TMP18:%.*]] = getelementptr i32, i32* [[TMP17]], i32 [[TMP5]] 1761 // CHECK2-NEXT: [[TMP19:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP14]], align 4 1762 // CHECK2-NEXT: store i32 [[TMP19]], i32* [[TMP18]], align 4 1763 // CHECK2-NEXT: br label [[IFCONT6:%.*]] 1764 // CHECK2: else5: 1765 // CHECK2-NEXT: br label [[IFCONT6]] 1766 // CHECK2: ifcont6: 1767 // CHECK2-NEXT: [[TMP20:%.*]] = add nsw i32 [[TMP5]], 1 1768 // CHECK2-NEXT: store i32 [[TMP20]], i32* [[DOTCNT_ADDR]], align 4 1769 // CHECK2-NEXT: br label [[PRECOND]] 1770 // CHECK2: exit: 1771 // CHECK2-NEXT: ret void 1772 // 1773 // 1774 // CHECK2-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_copy_func 1775 // CHECK2-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 1776 // CHECK2-NEXT: entry: 1777 // CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 1778 // CHECK2-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 1779 // CHECK2-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 8 1780 // CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 1781 // CHECK2-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 1782 // CHECK2-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 8 1783 // CHECK2-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 8 1784 // CHECK2-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [1 x i8*]* 1785 // CHECK2-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 8 1786 // CHECK2-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.1* 1787 // CHECK2-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 1788 // CHECK2-NEXT: [[TMP8:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP4]], i64 0, i64 0 1789 // CHECK2-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 8 1790 // CHECK2-NEXT: [[TMP10:%.*]] = bitcast i8* [[TMP9]] to double* 1791 // CHECK2-NEXT: [[E:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_1:%.*]], %struct._globalized_locals_ty.1* [[TMP6]], i32 0, i32 0 1792 // CHECK2-NEXT: [[TMP11:%.*]] = getelementptr inbounds [1024 x double], [1024 x double]* [[E]], i32 0, i32 [[TMP7]] 1793 // CHECK2-NEXT: [[TMP12:%.*]] = load double, double* [[TMP10]], align 8 1794 // CHECK2-NEXT: store double [[TMP12]], double* [[TMP11]], align 128 1795 // CHECK2-NEXT: ret void 1796 // 1797 // 1798 // CHECK2-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_reduce_func 1799 // CHECK2-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 1800 // CHECK2-NEXT: entry: 1801 // CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 1802 // CHECK2-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 1803 // CHECK2-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 8 1804 // CHECK2-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x i8*], align 8 1805 // CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 1806 // CHECK2-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 1807 // CHECK2-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 8 1808 // CHECK2-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8 1809 // CHECK2-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.1* 1810 // CHECK2-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 1811 // CHECK2-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0 1812 // CHECK2-NEXT: [[E:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_1:%.*]], %struct._globalized_locals_ty.1* [[TMP4]], i32 0, i32 0 1813 // CHECK2-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1024 x double], [1024 x double]* [[E]], i32 0, i32 [[TMP5]] 1814 // CHECK2-NEXT: [[TMP8:%.*]] = bitcast double* [[TMP7]] to i8* 1815 // CHECK2-NEXT: store i8* [[TMP8]], i8** [[TMP6]], align 8 1816 // CHECK2-NEXT: [[TMP9:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 1817 // CHECK2-NEXT: [[TMP10:%.*]] = load i8*, i8** [[DOTADDR2]], align 8 1818 // CHECK2-NEXT: call void @"_omp$reduction$reduction_func"(i8* [[TMP9]], i8* [[TMP10]]) #[[ATTR3]] 1819 // CHECK2-NEXT: ret void 1820 // 1821 // 1822 // CHECK2-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_copy_func 1823 // CHECK2-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 1824 // CHECK2-NEXT: entry: 1825 // CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 1826 // CHECK2-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 1827 // CHECK2-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 8 1828 // CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 1829 // CHECK2-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 1830 // CHECK2-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 8 1831 // CHECK2-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 8 1832 // CHECK2-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [1 x i8*]* 1833 // CHECK2-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 8 1834 // CHECK2-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.1* 1835 // CHECK2-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 1836 // CHECK2-NEXT: [[TMP8:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP4]], i64 0, i64 0 1837 // CHECK2-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 8 1838 // CHECK2-NEXT: [[TMP10:%.*]] = bitcast i8* [[TMP9]] to double* 1839 // CHECK2-NEXT: [[E:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_1:%.*]], %struct._globalized_locals_ty.1* [[TMP6]], i32 0, i32 0 1840 // CHECK2-NEXT: [[TMP11:%.*]] = getelementptr inbounds [1024 x double], [1024 x double]* [[E]], i32 0, i32 [[TMP7]] 1841 // CHECK2-NEXT: [[TMP12:%.*]] = load double, double* [[TMP11]], align 128 1842 // CHECK2-NEXT: store double [[TMP12]], double* [[TMP10]], align 8 1843 // CHECK2-NEXT: ret void 1844 // 1845 // 1846 // CHECK2-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_reduce_func 1847 // CHECK2-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 1848 // CHECK2-NEXT: entry: 1849 // CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 1850 // CHECK2-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 1851 // CHECK2-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 8 1852 // CHECK2-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x i8*], align 8 1853 // CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 1854 // CHECK2-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 1855 // CHECK2-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 8 1856 // CHECK2-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8 1857 // CHECK2-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.1* 1858 // CHECK2-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 1859 // CHECK2-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0 1860 // CHECK2-NEXT: [[E:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_1:%.*]], %struct._globalized_locals_ty.1* [[TMP4]], i32 0, i32 0 1861 // CHECK2-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1024 x double], [1024 x double]* [[E]], i32 0, i32 [[TMP5]] 1862 // CHECK2-NEXT: [[TMP8:%.*]] = bitcast double* [[TMP7]] to i8* 1863 // CHECK2-NEXT: store i8* [[TMP8]], i8** [[TMP6]], align 8 1864 // CHECK2-NEXT: [[TMP9:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 1865 // CHECK2-NEXT: [[TMP10:%.*]] = load i8*, i8** [[DOTADDR2]], align 8 1866 // CHECK2-NEXT: call void @"_omp$reduction$reduction_func"(i8* [[TMP10]], i8* [[TMP9]]) #[[ATTR3]] 1867 // CHECK2-NEXT: ret void 1868 // 1869 // 1870 // CHECK2-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29_worker 1871 // CHECK2-SAME: () #[[ATTR0]] { 1872 // CHECK2-NEXT: entry: 1873 // CHECK2-NEXT: [[WORK_FN:%.*]] = alloca i8*, align 8 1874 // CHECK2-NEXT: [[EXEC_STATUS:%.*]] = alloca i8, align 1 1875 // CHECK2-NEXT: store i8* null, i8** [[WORK_FN]], align 8 1876 // CHECK2-NEXT: store i8 0, i8* [[EXEC_STATUS]], align 1 1877 // CHECK2-NEXT: br label [[DOTAWAIT_WORK:%.*]] 1878 // CHECK2: .await.work: 1879 // CHECK2-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 1880 // CHECK2-NEXT: [[TMP0:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORK_FN]]) 1881 // CHECK2-NEXT: [[TMP1:%.*]] = zext i1 [[TMP0]] to i8 1882 // CHECK2-NEXT: store i8 [[TMP1]], i8* [[EXEC_STATUS]], align 1 1883 // CHECK2-NEXT: [[TMP2:%.*]] = load i8*, i8** [[WORK_FN]], align 8 1884 // CHECK2-NEXT: [[SHOULD_TERMINATE:%.*]] = icmp eq i8* [[TMP2]], null 1885 // CHECK2-NEXT: br i1 [[SHOULD_TERMINATE]], label [[DOTEXIT:%.*]], label [[DOTSELECT_WORKERS:%.*]] 1886 // CHECK2: .select.workers: 1887 // CHECK2-NEXT: [[TMP3:%.*]] = load i8, i8* [[EXEC_STATUS]], align 1 1888 // CHECK2-NEXT: [[IS_ACTIVE:%.*]] = icmp ne i8 [[TMP3]], 0 1889 // CHECK2-NEXT: br i1 [[IS_ACTIVE]], label [[DOTEXECUTE_PARALLEL:%.*]], label [[DOTBARRIER_PARALLEL:%.*]] 1890 // CHECK2: .execute.parallel: 1891 // CHECK2-NEXT: [[TMP4:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 1892 // CHECK2-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP2]] to void (i16, i32)* 1893 // CHECK2-NEXT: call void [[TMP5]](i16 0, i32 [[TMP4]]) 1894 // CHECK2-NEXT: br label [[DOTTERMINATE_PARALLEL:%.*]] 1895 // CHECK2: .terminate.parallel: 1896 // CHECK2-NEXT: call void @__kmpc_kernel_end_parallel() 1897 // CHECK2-NEXT: br label [[DOTBARRIER_PARALLEL]] 1898 // CHECK2: .barrier.parallel: 1899 // CHECK2-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 1900 // CHECK2-NEXT: br label [[DOTAWAIT_WORK]] 1901 // CHECK2: .exit: 1902 // CHECK2-NEXT: ret void 1903 // 1904 // 1905 // CHECK2-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29 1906 // CHECK2-SAME: (i64 [[C:%.*]], i64 [[D:%.*]]) #[[ATTR1]] { 1907 // CHECK2-NEXT: entry: 1908 // CHECK2-NEXT: [[C_ADDR:%.*]] = alloca i64, align 8 1909 // CHECK2-NEXT: [[D_ADDR:%.*]] = alloca i64, align 8 1910 // CHECK2-NEXT: [[DOTZERO_ADDR:%.*]] = alloca i32, align 4 1911 // CHECK2-NEXT: [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4 1912 // CHECK2-NEXT: store i32 0, i32* [[DOTZERO_ADDR]], align 4 1913 // CHECK2-NEXT: store i64 [[C]], i64* [[C_ADDR]], align 8 1914 // CHECK2-NEXT: store i64 [[D]], i64* [[D_ADDR]], align 8 1915 // CHECK2-NEXT: [[CONV:%.*]] = bitcast i64* [[C_ADDR]] to i8* 1916 // CHECK2-NEXT: [[CONV1:%.*]] = bitcast i64* [[D_ADDR]] to float* 1917 // CHECK2-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 1918 // CHECK2-NEXT: [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 1919 // CHECK2-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 1920 // CHECK2-NEXT: [[THREAD_LIMIT:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS]], [[NVPTX_WARP_SIZE]] 1921 // CHECK2-NEXT: [[TMP0:%.*]] = icmp ult i32 [[NVPTX_TID]], [[THREAD_LIMIT]] 1922 // CHECK2-NEXT: br i1 [[TMP0]], label [[DOTWORKER:%.*]], label [[DOTMASTERCHECK:%.*]] 1923 // CHECK2: .worker: 1924 // CHECK2-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29_worker() #[[ATTR3]] 1925 // CHECK2-NEXT: br label [[DOTEXIT:%.*]] 1926 // CHECK2: .mastercheck: 1927 // CHECK2-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 1928 // CHECK2-NEXT: [[NVPTX_NUM_THREADS3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 1929 // CHECK2-NEXT: [[NVPTX_WARP_SIZE4:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 1930 // CHECK2-NEXT: [[TMP1:%.*]] = sub nuw i32 [[NVPTX_WARP_SIZE4]], 1 1931 // CHECK2-NEXT: [[TMP2:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS3]], 1 1932 // CHECK2-NEXT: [[TMP3:%.*]] = xor i32 [[TMP1]], -1 1933 // CHECK2-NEXT: [[MASTER_TID:%.*]] = and i32 [[TMP2]], [[TMP3]] 1934 // CHECK2-NEXT: [[TMP4:%.*]] = icmp eq i32 [[NVPTX_TID2]], [[MASTER_TID]] 1935 // CHECK2-NEXT: br i1 [[TMP4]], label [[DOTMASTER:%.*]], label [[DOTEXIT]] 1936 // CHECK2: .master: 1937 // CHECK2-NEXT: [[NVPTX_NUM_THREADS5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 1938 // CHECK2-NEXT: [[NVPTX_WARP_SIZE6:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 1939 // CHECK2-NEXT: [[THREAD_LIMIT7:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS5]], [[NVPTX_WARP_SIZE6]] 1940 // CHECK2-NEXT: call void @__kmpc_kernel_init(i32 [[THREAD_LIMIT7]], i16 1) 1941 // CHECK2-NEXT: call void @__kmpc_data_sharing_init_stack() 1942 // CHECK2-NEXT: [[TMP5:%.*]] = call i8* @__kmpc_data_sharing_push_stack(i64 8, i16 1) 1943 // CHECK2-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.2* 1944 // CHECK2-NEXT: [[TMP7:%.*]] = load i8, i8* [[CONV]], align 8 1945 // CHECK2-NEXT: [[C8:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_2:%.*]], %struct._globalized_locals_ty.2* [[TMP6]], i32 0, i32 1 1946 // CHECK2-NEXT: store i8 [[TMP7]], i8* [[C8]], align 4 1947 // CHECK2-NEXT: [[TMP8:%.*]] = load float, float* [[CONV1]], align 8 1948 // CHECK2-NEXT: [[D9:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_2]], %struct._globalized_locals_ty.2* [[TMP6]], i32 0, i32 0 1949 // CHECK2-NEXT: store float [[TMP8]], float* [[D9]], align 4 1950 // CHECK2-NEXT: [[TMP9:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 1951 // CHECK2-NEXT: store i32 [[TMP9]], i32* [[DOTTHREADID_TEMP_]], align 4 1952 // CHECK2-NEXT: call void @__omp_outlined__1(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i8* [[C8]], float* [[D9]]) #[[ATTR3]] 1953 // CHECK2-NEXT: call void @__kmpc_data_sharing_pop_stack(i8* [[TMP5]]) 1954 // CHECK2-NEXT: br label [[DOTTERMINATION_NOTIFIER:%.*]] 1955 // CHECK2: .termination.notifier: 1956 // CHECK2-NEXT: call void @__kmpc_kernel_deinit(i16 1) 1957 // CHECK2-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 1958 // CHECK2-NEXT: br label [[DOTEXIT]] 1959 // CHECK2: .exit: 1960 // CHECK2-NEXT: ret void 1961 // 1962 // 1963 // CHECK2-LABEL: define {{[^@]+}}@__omp_outlined__1 1964 // CHECK2-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i8* nonnull align 1 dereferenceable(1) [[C:%.*]], float* nonnull align 4 dereferenceable(4) [[D:%.*]]) #[[ATTR1]] { 1965 // CHECK2-NEXT: entry: 1966 // CHECK2-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8 1967 // CHECK2-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8 1968 // CHECK2-NEXT: [[C_ADDR:%.*]] = alloca i8*, align 8 1969 // CHECK2-NEXT: [[D_ADDR:%.*]] = alloca float*, align 8 1970 // CHECK2-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 8 1971 // CHECK2-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8 1972 // CHECK2-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8 1973 // CHECK2-NEXT: store i8* [[C]], i8** [[C_ADDR]], align 8 1974 // CHECK2-NEXT: store float* [[D]], float** [[D_ADDR]], align 8 1975 // CHECK2-NEXT: [[TMP0:%.*]] = load i8*, i8** [[C_ADDR]], align 8 1976 // CHECK2-NEXT: [[TMP1:%.*]] = load float*, float** [[D_ADDR]], align 8 1977 // CHECK2-NEXT: [[TMP2:%.*]] = call i8* @__kmpc_data_sharing_push_stack(i64 8, i16 1) 1978 // CHECK2-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to %struct._globalized_locals_ty.3* 1979 // CHECK2-NEXT: [[C1:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3:%.*]], %struct._globalized_locals_ty.3* [[TMP3]], i32 0, i32 1 1980 // CHECK2-NEXT: [[D2:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3]], %struct._globalized_locals_ty.3* [[TMP3]], i32 0, i32 0 1981 // CHECK2-NEXT: store i8 0, i8* [[C1]], align 4 1982 // CHECK2-NEXT: store float 1.000000e+00, float* [[D2]], align 4 1983 // CHECK2-NEXT: [[TMP4:%.*]] = load i8, i8* [[C1]], align 4 1984 // CHECK2-NEXT: [[CONV:%.*]] = sext i8 [[TMP4]] to i32 1985 // CHECK2-NEXT: [[XOR:%.*]] = xor i32 [[CONV]], 2 1986 // CHECK2-NEXT: [[CONV3:%.*]] = trunc i32 [[XOR]] to i8 1987 // CHECK2-NEXT: store i8 [[CONV3]], i8* [[C1]], align 4 1988 // CHECK2-NEXT: [[TMP5:%.*]] = load float, float* [[D2]], align 4 1989 // CHECK2-NEXT: [[MUL:%.*]] = fmul float [[TMP5]], 3.300000e+01 1990 // CHECK2-NEXT: store float [[MUL]], float* [[D2]], align 4 1991 // CHECK2-NEXT: [[TMP6:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8 1992 // CHECK2-NEXT: [[TMP7:%.*]] = load i32, i32* [[TMP6]], align 4 1993 // CHECK2-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0 1994 // CHECK2-NEXT: store i8* [[C1]], i8** [[TMP8]], align 8 1995 // CHECK2-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 1 1996 // CHECK2-NEXT: [[TMP10:%.*]] = bitcast float* [[D2]] to i8* 1997 // CHECK2-NEXT: store i8* [[TMP10]], i8** [[TMP9]], align 8 1998 // CHECK2-NEXT: [[TMP11:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 1999 // CHECK2-NEXT: [[TMP12:%.*]] = load i8*, i8** @"_openmp_teams_reductions_buffer_$_$ptr", align 8 2000 // CHECK2-NEXT: [[TMP13:%.*]] = call i32 @__kmpc_nvptx_teams_reduce_nowait_v2(%struct.ident_t* @[[GLOB1]], i32 [[TMP7]], i8* [[TMP12]], i32 1024, i8* [[TMP11]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func3, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func4, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_copy_func5, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_reduce_func6, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_copy_func7, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_reduce_func8) 2001 // CHECK2-NEXT: [[TMP14:%.*]] = icmp eq i32 [[TMP13]], 1 2002 // CHECK2-NEXT: br i1 [[TMP14]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]] 2003 // CHECK2: .omp.reduction.then: 2004 // CHECK2-NEXT: [[TMP15:%.*]] = load i8, i8* [[TMP0]], align 1 2005 // CHECK2-NEXT: [[CONV4:%.*]] = sext i8 [[TMP15]] to i32 2006 // CHECK2-NEXT: [[TMP16:%.*]] = load i8, i8* [[C1]], align 4 2007 // CHECK2-NEXT: [[CONV5:%.*]] = sext i8 [[TMP16]] to i32 2008 // CHECK2-NEXT: [[XOR6:%.*]] = xor i32 [[CONV4]], [[CONV5]] 2009 // CHECK2-NEXT: [[CONV7:%.*]] = trunc i32 [[XOR6]] to i8 2010 // CHECK2-NEXT: store i8 [[CONV7]], i8* [[TMP0]], align 1 2011 // CHECK2-NEXT: [[TMP17:%.*]] = load float, float* [[TMP1]], align 4 2012 // CHECK2-NEXT: [[TMP18:%.*]] = load float, float* [[D2]], align 4 2013 // CHECK2-NEXT: [[MUL8:%.*]] = fmul float [[TMP17]], [[TMP18]] 2014 // CHECK2-NEXT: store float [[MUL8]], float* [[TMP1]], align 4 2015 // CHECK2-NEXT: call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP7]]) 2016 // CHECK2-NEXT: br label [[DOTOMP_REDUCTION_DONE]] 2017 // CHECK2: .omp.reduction.done: 2018 // CHECK2-NEXT: call void @__kmpc_data_sharing_pop_stack(i8* [[TMP2]]) 2019 // CHECK2-NEXT: ret void 2020 // 2021 // 2022 // CHECK2-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func3 2023 // CHECK2-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR0]] { 2024 // CHECK2-NEXT: entry: 2025 // CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 2026 // CHECK2-NEXT: [[DOTADDR1:%.*]] = alloca i16, align 2 2027 // CHECK2-NEXT: [[DOTADDR2:%.*]] = alloca i16, align 2 2028 // CHECK2-NEXT: [[DOTADDR3:%.*]] = alloca i16, align 2 2029 // CHECK2-NEXT: [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [2 x i8*], align 8 2030 // CHECK2-NEXT: [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca i8, align 1 2031 // CHECK2-NEXT: [[DOTOMP_REDUCTION_ELEMENT4:%.*]] = alloca float, align 4 2032 // CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 2033 // CHECK2-NEXT: store i16 [[TMP1]], i16* [[DOTADDR1]], align 2 2034 // CHECK2-NEXT: store i16 [[TMP2]], i16* [[DOTADDR2]], align 2 2035 // CHECK2-NEXT: store i16 [[TMP3]], i16* [[DOTADDR3]], align 2 2036 // CHECK2-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 8 2037 // CHECK2-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [2 x i8*]* 2038 // CHECK2-NEXT: [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2 2039 // CHECK2-NEXT: [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2 2040 // CHECK2-NEXT: [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2 2041 // CHECK2-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i64 0, i64 0 2042 // CHECK2-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8 2043 // CHECK2-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0 2044 // CHECK2-NEXT: [[TMP12:%.*]] = getelementptr i8, i8* [[TMP10]], i64 1 2045 // CHECK2-NEXT: [[TMP13:%.*]] = load i8, i8* [[TMP10]], align 1 2046 // CHECK2-NEXT: [[TMP14:%.*]] = sext i8 [[TMP13]] to i32 2047 // CHECK2-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 2048 // CHECK2-NEXT: [[TMP15:%.*]] = trunc i32 [[NVPTX_WARP_SIZE]] to i16 2049 // CHECK2-NEXT: [[TMP16:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP14]], i16 [[TMP7]], i16 [[TMP15]]) 2050 // CHECK2-NEXT: [[TMP17:%.*]] = trunc i32 [[TMP16]] to i8 2051 // CHECK2-NEXT: store i8 [[TMP17]], i8* [[DOTOMP_REDUCTION_ELEMENT]], align 1 2052 // CHECK2-NEXT: [[TMP18:%.*]] = getelementptr i8, i8* [[TMP10]], i64 1 2053 // CHECK2-NEXT: [[TMP19:%.*]] = getelementptr i8, i8* [[DOTOMP_REDUCTION_ELEMENT]], i64 1 2054 // CHECK2-NEXT: store i8* [[DOTOMP_REDUCTION_ELEMENT]], i8** [[TMP11]], align 8 2055 // CHECK2-NEXT: [[TMP20:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i64 0, i64 1 2056 // CHECK2-NEXT: [[TMP21:%.*]] = load i8*, i8** [[TMP20]], align 8 2057 // CHECK2-NEXT: [[TMP22:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 1 2058 // CHECK2-NEXT: [[TMP23:%.*]] = bitcast i8* [[TMP21]] to float* 2059 // CHECK2-NEXT: [[TMP24:%.*]] = getelementptr float, float* [[TMP23]], i64 1 2060 // CHECK2-NEXT: [[TMP25:%.*]] = bitcast float* [[TMP24]] to i8* 2061 // CHECK2-NEXT: [[TMP26:%.*]] = bitcast float* [[TMP23]] to i32* 2062 // CHECK2-NEXT: [[TMP27:%.*]] = bitcast float* [[DOTOMP_REDUCTION_ELEMENT4]] to i32* 2063 // CHECK2-NEXT: [[TMP28:%.*]] = load i32, i32* [[TMP26]], align 4 2064 // CHECK2-NEXT: [[NVPTX_WARP_SIZE5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 2065 // CHECK2-NEXT: [[TMP29:%.*]] = trunc i32 [[NVPTX_WARP_SIZE5]] to i16 2066 // CHECK2-NEXT: [[TMP30:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP28]], i16 [[TMP7]], i16 [[TMP29]]) 2067 // CHECK2-NEXT: store i32 [[TMP30]], i32* [[TMP27]], align 4 2068 // CHECK2-NEXT: [[TMP31:%.*]] = getelementptr i32, i32* [[TMP26]], i64 1 2069 // CHECK2-NEXT: [[TMP32:%.*]] = getelementptr i32, i32* [[TMP27]], i64 1 2070 // CHECK2-NEXT: [[TMP33:%.*]] = bitcast float* [[DOTOMP_REDUCTION_ELEMENT4]] to i8* 2071 // CHECK2-NEXT: store i8* [[TMP33]], i8** [[TMP22]], align 8 2072 // CHECK2-NEXT: [[TMP34:%.*]] = icmp eq i16 [[TMP8]], 0 2073 // CHECK2-NEXT: [[TMP35:%.*]] = icmp eq i16 [[TMP8]], 1 2074 // CHECK2-NEXT: [[TMP36:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]] 2075 // CHECK2-NEXT: [[TMP37:%.*]] = and i1 [[TMP35]], [[TMP36]] 2076 // CHECK2-NEXT: [[TMP38:%.*]] = icmp eq i16 [[TMP8]], 2 2077 // CHECK2-NEXT: [[TMP39:%.*]] = and i16 [[TMP6]], 1 2078 // CHECK2-NEXT: [[TMP40:%.*]] = icmp eq i16 [[TMP39]], 0 2079 // CHECK2-NEXT: [[TMP41:%.*]] = and i1 [[TMP38]], [[TMP40]] 2080 // CHECK2-NEXT: [[TMP42:%.*]] = icmp sgt i16 [[TMP7]], 0 2081 // CHECK2-NEXT: [[TMP43:%.*]] = and i1 [[TMP41]], [[TMP42]] 2082 // CHECK2-NEXT: [[TMP44:%.*]] = or i1 [[TMP34]], [[TMP37]] 2083 // CHECK2-NEXT: [[TMP45:%.*]] = or i1 [[TMP44]], [[TMP43]] 2084 // CHECK2-NEXT: br i1 [[TMP45]], label [[THEN:%.*]], label [[ELSE:%.*]] 2085 // CHECK2: then: 2086 // CHECK2-NEXT: [[TMP46:%.*]] = bitcast [2 x i8*]* [[TMP5]] to i8* 2087 // CHECK2-NEXT: [[TMP47:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8* 2088 // CHECK2-NEXT: call void @"_omp$reduction$reduction_func2"(i8* [[TMP46]], i8* [[TMP47]]) #[[ATTR3]] 2089 // CHECK2-NEXT: br label [[IFCONT:%.*]] 2090 // CHECK2: else: 2091 // CHECK2-NEXT: br label [[IFCONT]] 2092 // CHECK2: ifcont: 2093 // CHECK2-NEXT: [[TMP48:%.*]] = icmp eq i16 [[TMP8]], 1 2094 // CHECK2-NEXT: [[TMP49:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]] 2095 // CHECK2-NEXT: [[TMP50:%.*]] = and i1 [[TMP48]], [[TMP49]] 2096 // CHECK2-NEXT: br i1 [[TMP50]], label [[THEN6:%.*]], label [[ELSE7:%.*]] 2097 // CHECK2: then6: 2098 // CHECK2-NEXT: [[TMP51:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0 2099 // CHECK2-NEXT: [[TMP52:%.*]] = load i8*, i8** [[TMP51]], align 8 2100 // CHECK2-NEXT: [[TMP53:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i64 0, i64 0 2101 // CHECK2-NEXT: [[TMP54:%.*]] = load i8*, i8** [[TMP53]], align 8 2102 // CHECK2-NEXT: [[TMP55:%.*]] = load i8, i8* [[TMP52]], align 1 2103 // CHECK2-NEXT: store i8 [[TMP55]], i8* [[TMP54]], align 1 2104 // CHECK2-NEXT: [[TMP56:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 1 2105 // CHECK2-NEXT: [[TMP57:%.*]] = load i8*, i8** [[TMP56]], align 8 2106 // CHECK2-NEXT: [[TMP58:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i64 0, i64 1 2107 // CHECK2-NEXT: [[TMP59:%.*]] = load i8*, i8** [[TMP58]], align 8 2108 // CHECK2-NEXT: [[TMP60:%.*]] = bitcast i8* [[TMP57]] to float* 2109 // CHECK2-NEXT: [[TMP61:%.*]] = bitcast i8* [[TMP59]] to float* 2110 // CHECK2-NEXT: [[TMP62:%.*]] = load float, float* [[TMP60]], align 4 2111 // CHECK2-NEXT: store float [[TMP62]], float* [[TMP61]], align 4 2112 // CHECK2-NEXT: br label [[IFCONT8:%.*]] 2113 // CHECK2: else7: 2114 // CHECK2-NEXT: br label [[IFCONT8]] 2115 // CHECK2: ifcont8: 2116 // CHECK2-NEXT: ret void 2117 // 2118 // 2119 // CHECK2-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func4 2120 // CHECK2-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] { 2121 // CHECK2-NEXT: entry: 2122 // CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 2123 // CHECK2-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 2124 // CHECK2-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 2125 // CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 2126 // CHECK2-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 2127 // CHECK2-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 2128 // CHECK2-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 2129 // CHECK2-NEXT: [[NVPTX_LANE_ID:%.*]] = and i32 [[NVPTX_TID2]], 31 2130 // CHECK2-NEXT: [[NVPTX_TID3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 2131 // CHECK2-NEXT: [[NVPTX_WARP_ID:%.*]] = ashr i32 [[NVPTX_TID3]], 5 2132 // CHECK2-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8 2133 // CHECK2-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 2134 // CHECK2-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 2135 // CHECK2-NEXT: [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 2136 // CHECK2-NEXT: br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]] 2137 // CHECK2: then: 2138 // CHECK2-NEXT: [[TMP5:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 0 2139 // CHECK2-NEXT: [[TMP6:%.*]] = load i8*, i8** [[TMP5]], align 8 2140 // CHECK2-NEXT: [[TMP7:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 2141 // CHECK2-NEXT: [[TMP8:%.*]] = bitcast i32 addrspace(3)* [[TMP7]] to i8 addrspace(3)* 2142 // CHECK2-NEXT: [[TMP9:%.*]] = load i8, i8* [[TMP6]], align 1 2143 // CHECK2-NEXT: store volatile i8 [[TMP9]], i8 addrspace(3)* [[TMP8]], align 1 2144 // CHECK2-NEXT: br label [[IFCONT:%.*]] 2145 // CHECK2: else: 2146 // CHECK2-NEXT: br label [[IFCONT]] 2147 // CHECK2: ifcont: 2148 // CHECK2-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 2149 // CHECK2-NEXT: [[TMP10:%.*]] = load i32, i32* [[DOTADDR1]], align 4 2150 // CHECK2-NEXT: [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP10]] 2151 // CHECK2-NEXT: br i1 [[IS_ACTIVE_THREAD]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 2152 // CHECK2: then4: 2153 // CHECK2-NEXT: [[TMP11:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 2154 // CHECK2-NEXT: [[TMP12:%.*]] = bitcast i32 addrspace(3)* [[TMP11]] to i8 addrspace(3)* 2155 // CHECK2-NEXT: [[TMP13:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 0 2156 // CHECK2-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 8 2157 // CHECK2-NEXT: [[TMP15:%.*]] = load volatile i8, i8 addrspace(3)* [[TMP12]], align 1 2158 // CHECK2-NEXT: store i8 [[TMP15]], i8* [[TMP14]], align 1 2159 // CHECK2-NEXT: br label [[IFCONT6:%.*]] 2160 // CHECK2: else5: 2161 // CHECK2-NEXT: br label [[IFCONT6]] 2162 // CHECK2: ifcont6: 2163 // CHECK2-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 2164 // CHECK2-NEXT: [[WARP_MASTER7:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 2165 // CHECK2-NEXT: br i1 [[WARP_MASTER7]], label [[THEN8:%.*]], label [[ELSE9:%.*]] 2166 // CHECK2: then8: 2167 // CHECK2-NEXT: [[TMP16:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 1 2168 // CHECK2-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 8 2169 // CHECK2-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i32* 2170 // CHECK2-NEXT: [[TMP19:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 2171 // CHECK2-NEXT: [[TMP20:%.*]] = load i32, i32* [[TMP18]], align 4 2172 // CHECK2-NEXT: store volatile i32 [[TMP20]], i32 addrspace(3)* [[TMP19]], align 4 2173 // CHECK2-NEXT: br label [[IFCONT10:%.*]] 2174 // CHECK2: else9: 2175 // CHECK2-NEXT: br label [[IFCONT10]] 2176 // CHECK2: ifcont10: 2177 // CHECK2-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 2178 // CHECK2-NEXT: [[TMP21:%.*]] = load i32, i32* [[DOTADDR1]], align 4 2179 // CHECK2-NEXT: [[IS_ACTIVE_THREAD11:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP21]] 2180 // CHECK2-NEXT: br i1 [[IS_ACTIVE_THREAD11]], label [[THEN12:%.*]], label [[ELSE13:%.*]] 2181 // CHECK2: then12: 2182 // CHECK2-NEXT: [[TMP22:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 2183 // CHECK2-NEXT: [[TMP23:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 1 2184 // CHECK2-NEXT: [[TMP24:%.*]] = load i8*, i8** [[TMP23]], align 8 2185 // CHECK2-NEXT: [[TMP25:%.*]] = bitcast i8* [[TMP24]] to i32* 2186 // CHECK2-NEXT: [[TMP26:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP22]], align 4 2187 // CHECK2-NEXT: store i32 [[TMP26]], i32* [[TMP25]], align 4 2188 // CHECK2-NEXT: br label [[IFCONT14:%.*]] 2189 // CHECK2: else13: 2190 // CHECK2-NEXT: br label [[IFCONT14]] 2191 // CHECK2: ifcont14: 2192 // CHECK2-NEXT: ret void 2193 // 2194 // 2195 // CHECK2-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_copy_func5 2196 // CHECK2-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 2197 // CHECK2-NEXT: entry: 2198 // CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 2199 // CHECK2-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 2200 // CHECK2-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 8 2201 // CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 2202 // CHECK2-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 2203 // CHECK2-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 8 2204 // CHECK2-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 8 2205 // CHECK2-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 2206 // CHECK2-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 8 2207 // CHECK2-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.4* 2208 // CHECK2-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 2209 // CHECK2-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 0 2210 // CHECK2-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 8 2211 // CHECK2-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4:%.*]], %struct._globalized_locals_ty.4* [[TMP6]], i32 0, i32 0 2212 // CHECK2-NEXT: [[TMP10:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[C]], i32 0, i32 [[TMP7]] 2213 // CHECK2-NEXT: [[TMP11:%.*]] = load i8, i8* [[TMP9]], align 1 2214 // CHECK2-NEXT: store i8 [[TMP11]], i8* [[TMP10]], align 128 2215 // CHECK2-NEXT: [[TMP12:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 1 2216 // CHECK2-NEXT: [[TMP13:%.*]] = load i8*, i8** [[TMP12]], align 8 2217 // CHECK2-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to float* 2218 // CHECK2-NEXT: [[D:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4]], %struct._globalized_locals_ty.4* [[TMP6]], i32 0, i32 1 2219 // CHECK2-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1024 x float], [1024 x float]* [[D]], i32 0, i32 [[TMP7]] 2220 // CHECK2-NEXT: [[TMP16:%.*]] = load float, float* [[TMP14]], align 4 2221 // CHECK2-NEXT: store float [[TMP16]], float* [[TMP15]], align 128 2222 // CHECK2-NEXT: ret void 2223 // 2224 // 2225 // CHECK2-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_reduce_func6 2226 // CHECK2-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 2227 // CHECK2-NEXT: entry: 2228 // CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 2229 // CHECK2-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 2230 // CHECK2-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 8 2231 // CHECK2-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 8 2232 // CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 2233 // CHECK2-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 2234 // CHECK2-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 8 2235 // CHECK2-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8 2236 // CHECK2-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.4* 2237 // CHECK2-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 2238 // CHECK2-NEXT: [[TMP6:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0 2239 // CHECK2-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4:%.*]], %struct._globalized_locals_ty.4* [[TMP4]], i32 0, i32 0 2240 // CHECK2-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[C]], i32 0, i32 [[TMP5]] 2241 // CHECK2-NEXT: store i8* [[TMP7]], i8** [[TMP6]], align 8 2242 // CHECK2-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 1 2243 // CHECK2-NEXT: [[D:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4]], %struct._globalized_locals_ty.4* [[TMP4]], i32 0, i32 1 2244 // CHECK2-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1024 x float], [1024 x float]* [[D]], i32 0, i32 [[TMP5]] 2245 // CHECK2-NEXT: [[TMP10:%.*]] = bitcast float* [[TMP9]] to i8* 2246 // CHECK2-NEXT: store i8* [[TMP10]], i8** [[TMP8]], align 8 2247 // CHECK2-NEXT: [[TMP11:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 2248 // CHECK2-NEXT: [[TMP12:%.*]] = load i8*, i8** [[DOTADDR2]], align 8 2249 // CHECK2-NEXT: call void @"_omp$reduction$reduction_func2"(i8* [[TMP11]], i8* [[TMP12]]) #[[ATTR3]] 2250 // CHECK2-NEXT: ret void 2251 // 2252 // 2253 // CHECK2-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_copy_func7 2254 // CHECK2-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 2255 // CHECK2-NEXT: entry: 2256 // CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 2257 // CHECK2-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 2258 // CHECK2-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 8 2259 // CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 2260 // CHECK2-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 2261 // CHECK2-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 8 2262 // CHECK2-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 8 2263 // CHECK2-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 2264 // CHECK2-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 8 2265 // CHECK2-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.4* 2266 // CHECK2-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 2267 // CHECK2-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 0 2268 // CHECK2-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 8 2269 // CHECK2-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4:%.*]], %struct._globalized_locals_ty.4* [[TMP6]], i32 0, i32 0 2270 // CHECK2-NEXT: [[TMP10:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[C]], i32 0, i32 [[TMP7]] 2271 // CHECK2-NEXT: [[TMP11:%.*]] = load i8, i8* [[TMP10]], align 128 2272 // CHECK2-NEXT: store i8 [[TMP11]], i8* [[TMP9]], align 1 2273 // CHECK2-NEXT: [[TMP12:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 1 2274 // CHECK2-NEXT: [[TMP13:%.*]] = load i8*, i8** [[TMP12]], align 8 2275 // CHECK2-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to float* 2276 // CHECK2-NEXT: [[D:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4]], %struct._globalized_locals_ty.4* [[TMP6]], i32 0, i32 1 2277 // CHECK2-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1024 x float], [1024 x float]* [[D]], i32 0, i32 [[TMP7]] 2278 // CHECK2-NEXT: [[TMP16:%.*]] = load float, float* [[TMP15]], align 128 2279 // CHECK2-NEXT: store float [[TMP16]], float* [[TMP14]], align 4 2280 // CHECK2-NEXT: ret void 2281 // 2282 // 2283 // CHECK2-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_reduce_func8 2284 // CHECK2-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 2285 // CHECK2-NEXT: entry: 2286 // CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 2287 // CHECK2-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 2288 // CHECK2-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 8 2289 // CHECK2-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 8 2290 // CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 2291 // CHECK2-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 2292 // CHECK2-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 8 2293 // CHECK2-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8 2294 // CHECK2-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.4* 2295 // CHECK2-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 2296 // CHECK2-NEXT: [[TMP6:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0 2297 // CHECK2-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4:%.*]], %struct._globalized_locals_ty.4* [[TMP4]], i32 0, i32 0 2298 // CHECK2-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[C]], i32 0, i32 [[TMP5]] 2299 // CHECK2-NEXT: store i8* [[TMP7]], i8** [[TMP6]], align 8 2300 // CHECK2-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 1 2301 // CHECK2-NEXT: [[D:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4]], %struct._globalized_locals_ty.4* [[TMP4]], i32 0, i32 1 2302 // CHECK2-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1024 x float], [1024 x float]* [[D]], i32 0, i32 [[TMP5]] 2303 // CHECK2-NEXT: [[TMP10:%.*]] = bitcast float* [[TMP9]] to i8* 2304 // CHECK2-NEXT: store i8* [[TMP10]], i8** [[TMP8]], align 8 2305 // CHECK2-NEXT: [[TMP11:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 2306 // CHECK2-NEXT: [[TMP12:%.*]] = load i8*, i8** [[DOTADDR2]], align 8 2307 // CHECK2-NEXT: call void @"_omp$reduction$reduction_func2"(i8* [[TMP12]], i8* [[TMP11]]) #[[ATTR3]] 2308 // CHECK2-NEXT: ret void 2309 // 2310 // 2311 // CHECK2-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l36 2312 // CHECK2-SAME: (i64 [[A:%.*]], i64 [[B:%.*]]) #[[ATTR1]] { 2313 // CHECK2-NEXT: entry: 2314 // CHECK2-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8 2315 // CHECK2-NEXT: [[B_ADDR:%.*]] = alloca i64, align 8 2316 // CHECK2-NEXT: [[DOTZERO_ADDR:%.*]] = alloca i32, align 4 2317 // CHECK2-NEXT: [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4 2318 // CHECK2-NEXT: store i32 0, i32* [[DOTZERO_ADDR]], align 4 2319 // CHECK2-NEXT: store i64 [[A]], i64* [[A_ADDR]], align 8 2320 // CHECK2-NEXT: store i64 [[B]], i64* [[B_ADDR]], align 8 2321 // CHECK2-NEXT: [[CONV:%.*]] = bitcast i64* [[A_ADDR]] to i32* 2322 // CHECK2-NEXT: [[CONV1:%.*]] = bitcast i64* [[B_ADDR]] to i16* 2323 // CHECK2-NEXT: [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 2324 // CHECK2-NEXT: call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 1) 2325 // CHECK2-NEXT: call void @__kmpc_data_sharing_init_stack_spmd() 2326 // CHECK2-NEXT: br label [[DOTEXECUTE:%.*]] 2327 // CHECK2: .execute: 2328 // CHECK2-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3:[0-9]+]]) 2329 // CHECK2-NEXT: store i32 [[TMP0]], i32* [[DOTTHREADID_TEMP_]], align 4 2330 // CHECK2-NEXT: call void @__omp_outlined__9(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32* [[CONV]], i16* [[CONV1]]) #[[ATTR3]] 2331 // CHECK2-NEXT: br label [[DOTOMP_DEINIT:%.*]] 2332 // CHECK2: .omp.deinit: 2333 // CHECK2-NEXT: call void @__kmpc_spmd_kernel_deinit_v2(i16 1) 2334 // CHECK2-NEXT: br label [[DOTEXIT:%.*]] 2335 // CHECK2: .exit: 2336 // CHECK2-NEXT: ret void 2337 // 2338 // 2339 // CHECK2-LABEL: define {{[^@]+}}@__omp_outlined__9 2340 // CHECK2-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32* nonnull align 4 dereferenceable(4) [[A:%.*]], i16* nonnull align 2 dereferenceable(2) [[B:%.*]]) #[[ATTR1]] { 2341 // CHECK2-NEXT: entry: 2342 // CHECK2-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8 2343 // CHECK2-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8 2344 // CHECK2-NEXT: [[A_ADDR:%.*]] = alloca i32*, align 8 2345 // CHECK2-NEXT: [[B_ADDR:%.*]] = alloca i16*, align 8 2346 // CHECK2-NEXT: [[A1:%.*]] = alloca i32, align 4 2347 // CHECK2-NEXT: [[B2:%.*]] = alloca i16, align 2 2348 // CHECK2-NEXT: [[CAPTURED_VARS_ADDRS:%.*]] = alloca [2 x i8*], align 8 2349 // CHECK2-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 8 2350 // CHECK2-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8 2351 // CHECK2-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8 2352 // CHECK2-NEXT: store i32* [[A]], i32** [[A_ADDR]], align 8 2353 // CHECK2-NEXT: store i16* [[B]], i16** [[B_ADDR]], align 8 2354 // CHECK2-NEXT: [[TMP0:%.*]] = load i32*, i32** [[A_ADDR]], align 8 2355 // CHECK2-NEXT: [[TMP1:%.*]] = load i16*, i16** [[B_ADDR]], align 8 2356 // CHECK2-NEXT: store i32 0, i32* [[A1]], align 4 2357 // CHECK2-NEXT: store i16 -32768, i16* [[B2]], align 2 2358 // CHECK2-NEXT: [[TMP2:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0 2359 // CHECK2-NEXT: [[TMP3:%.*]] = bitcast i32* [[A1]] to i8* 2360 // CHECK2-NEXT: store i8* [[TMP3]], i8** [[TMP2]], align 8 2361 // CHECK2-NEXT: [[TMP4:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1 2362 // CHECK2-NEXT: [[TMP5:%.*]] = bitcast i16* [[B2]] to i8* 2363 // CHECK2-NEXT: store i8* [[TMP5]], i8** [[TMP4]], align 8 2364 // CHECK2-NEXT: [[TMP6:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8 2365 // CHECK2-NEXT: [[TMP7:%.*]] = load i32, i32* [[TMP6]], align 4 2366 // CHECK2-NEXT: [[TMP8:%.*]] = bitcast [2 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8** 2367 // CHECK2-NEXT: call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP7]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32*, i16*)* @__omp_outlined__10 to i8*), i8* null, i8** [[TMP8]], i64 2) 2368 // CHECK2-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0 2369 // CHECK2-NEXT: [[TMP10:%.*]] = bitcast i32* [[A1]] to i8* 2370 // CHECK2-NEXT: store i8* [[TMP10]], i8** [[TMP9]], align 8 2371 // CHECK2-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 1 2372 // CHECK2-NEXT: [[TMP12:%.*]] = bitcast i16* [[B2]] to i8* 2373 // CHECK2-NEXT: store i8* [[TMP12]], i8** [[TMP11]], align 8 2374 // CHECK2-NEXT: [[TMP13:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 2375 // CHECK2-NEXT: [[TMP14:%.*]] = load i8*, i8** @"_openmp_teams_reductions_buffer_$_$ptr", align 8 2376 // CHECK2-NEXT: [[TMP15:%.*]] = call i32 @__kmpc_nvptx_teams_reduce_nowait_v2(%struct.ident_t* @[[GLOB3]], i32 [[TMP7]], i8* [[TMP14]], i32 1024, i8* [[TMP13]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func15, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func16, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_copy_func17, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_reduce_func18, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_copy_func19, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_reduce_func20) 2377 // CHECK2-NEXT: [[TMP16:%.*]] = icmp eq i32 [[TMP15]], 1 2378 // CHECK2-NEXT: br i1 [[TMP16]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]] 2379 // CHECK2: .omp.reduction.then: 2380 // CHECK2-NEXT: [[TMP17:%.*]] = load i32, i32* [[TMP0]], align 4 2381 // CHECK2-NEXT: [[TMP18:%.*]] = load i32, i32* [[A1]], align 4 2382 // CHECK2-NEXT: [[OR:%.*]] = or i32 [[TMP17]], [[TMP18]] 2383 // CHECK2-NEXT: store i32 [[OR]], i32* [[TMP0]], align 4 2384 // CHECK2-NEXT: [[TMP19:%.*]] = load i16, i16* [[TMP1]], align 2 2385 // CHECK2-NEXT: [[CONV:%.*]] = sext i16 [[TMP19]] to i32 2386 // CHECK2-NEXT: [[TMP20:%.*]] = load i16, i16* [[B2]], align 2 2387 // CHECK2-NEXT: [[CONV3:%.*]] = sext i16 [[TMP20]] to i32 2388 // CHECK2-NEXT: [[CMP:%.*]] = icmp sgt i32 [[CONV]], [[CONV3]] 2389 // CHECK2-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] 2390 // CHECK2: cond.true: 2391 // CHECK2-NEXT: [[TMP21:%.*]] = load i16, i16* [[TMP1]], align 2 2392 // CHECK2-NEXT: br label [[COND_END:%.*]] 2393 // CHECK2: cond.false: 2394 // CHECK2-NEXT: [[TMP22:%.*]] = load i16, i16* [[B2]], align 2 2395 // CHECK2-NEXT: br label [[COND_END]] 2396 // CHECK2: cond.end: 2397 // CHECK2-NEXT: [[COND:%.*]] = phi i16 [ [[TMP21]], [[COND_TRUE]] ], [ [[TMP22]], [[COND_FALSE]] ] 2398 // CHECK2-NEXT: store i16 [[COND]], i16* [[TMP1]], align 2 2399 // CHECK2-NEXT: call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP7]]) 2400 // CHECK2-NEXT: br label [[DOTOMP_REDUCTION_DONE]] 2401 // CHECK2: .omp.reduction.done: 2402 // CHECK2-NEXT: ret void 2403 // 2404 // 2405 // CHECK2-LABEL: define {{[^@]+}}@__omp_outlined__10 2406 // CHECK2-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32* nonnull align 4 dereferenceable(4) [[A:%.*]], i16* nonnull align 2 dereferenceable(2) [[B:%.*]]) #[[ATTR1]] { 2407 // CHECK2-NEXT: entry: 2408 // CHECK2-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8 2409 // CHECK2-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8 2410 // CHECK2-NEXT: [[A_ADDR:%.*]] = alloca i32*, align 8 2411 // CHECK2-NEXT: [[B_ADDR:%.*]] = alloca i16*, align 8 2412 // CHECK2-NEXT: [[A1:%.*]] = alloca i32, align 4 2413 // CHECK2-NEXT: [[B2:%.*]] = alloca i16, align 2 2414 // CHECK2-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 8 2415 // CHECK2-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8 2416 // CHECK2-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8 2417 // CHECK2-NEXT: store i32* [[A]], i32** [[A_ADDR]], align 8 2418 // CHECK2-NEXT: store i16* [[B]], i16** [[B_ADDR]], align 8 2419 // CHECK2-NEXT: [[TMP0:%.*]] = load i32*, i32** [[A_ADDR]], align 8 2420 // CHECK2-NEXT: [[TMP1:%.*]] = load i16*, i16** [[B_ADDR]], align 8 2421 // CHECK2-NEXT: store i32 0, i32* [[A1]], align 4 2422 // CHECK2-NEXT: store i16 -32768, i16* [[B2]], align 2 2423 // CHECK2-NEXT: [[TMP2:%.*]] = load i32, i32* [[A1]], align 4 2424 // CHECK2-NEXT: [[OR:%.*]] = or i32 [[TMP2]], 1 2425 // CHECK2-NEXT: store i32 [[OR]], i32* [[A1]], align 4 2426 // CHECK2-NEXT: [[TMP3:%.*]] = load i16, i16* [[B2]], align 2 2427 // CHECK2-NEXT: [[CONV:%.*]] = sext i16 [[TMP3]] to i32 2428 // CHECK2-NEXT: [[CMP:%.*]] = icmp sgt i32 99, [[CONV]] 2429 // CHECK2-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] 2430 // CHECK2: cond.true: 2431 // CHECK2-NEXT: br label [[COND_END:%.*]] 2432 // CHECK2: cond.false: 2433 // CHECK2-NEXT: [[TMP4:%.*]] = load i16, i16* [[B2]], align 2 2434 // CHECK2-NEXT: [[CONV3:%.*]] = sext i16 [[TMP4]] to i32 2435 // CHECK2-NEXT: br label [[COND_END]] 2436 // CHECK2: cond.end: 2437 // CHECK2-NEXT: [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[CONV3]], [[COND_FALSE]] ] 2438 // CHECK2-NEXT: [[CONV4:%.*]] = trunc i32 [[COND]] to i16 2439 // CHECK2-NEXT: store i16 [[CONV4]], i16* [[B2]], align 2 2440 // CHECK2-NEXT: [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8 2441 // CHECK2-NEXT: [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4 2442 // CHECK2-NEXT: [[TMP7:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0 2443 // CHECK2-NEXT: [[TMP8:%.*]] = bitcast i32* [[A1]] to i8* 2444 // CHECK2-NEXT: store i8* [[TMP8]], i8** [[TMP7]], align 8 2445 // CHECK2-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 1 2446 // CHECK2-NEXT: [[TMP10:%.*]] = bitcast i16* [[B2]] to i8* 2447 // CHECK2-NEXT: store i8* [[TMP10]], i8** [[TMP9]], align 8 2448 // CHECK2-NEXT: [[TMP11:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 2449 // CHECK2-NEXT: [[TMP12:%.*]] = call i32 @__kmpc_nvptx_parallel_reduce_nowait_v2(%struct.ident_t* @[[GLOB3]], i32 [[TMP6]], i32 2, i64 16, i8* [[TMP11]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func12, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func13) 2450 // CHECK2-NEXT: [[TMP13:%.*]] = icmp eq i32 [[TMP12]], 1 2451 // CHECK2-NEXT: br i1 [[TMP13]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]] 2452 // CHECK2: .omp.reduction.then: 2453 // CHECK2-NEXT: [[TMP14:%.*]] = load i32, i32* [[TMP0]], align 4 2454 // CHECK2-NEXT: [[TMP15:%.*]] = load i32, i32* [[A1]], align 4 2455 // CHECK2-NEXT: [[OR5:%.*]] = or i32 [[TMP14]], [[TMP15]] 2456 // CHECK2-NEXT: store i32 [[OR5]], i32* [[TMP0]], align 4 2457 // CHECK2-NEXT: [[TMP16:%.*]] = load i16, i16* [[TMP1]], align 2 2458 // CHECK2-NEXT: [[CONV6:%.*]] = sext i16 [[TMP16]] to i32 2459 // CHECK2-NEXT: [[TMP17:%.*]] = load i16, i16* [[B2]], align 2 2460 // CHECK2-NEXT: [[CONV7:%.*]] = sext i16 [[TMP17]] to i32 2461 // CHECK2-NEXT: [[CMP8:%.*]] = icmp sgt i32 [[CONV6]], [[CONV7]] 2462 // CHECK2-NEXT: br i1 [[CMP8]], label [[COND_TRUE9:%.*]], label [[COND_FALSE10:%.*]] 2463 // CHECK2: cond.true9: 2464 // CHECK2-NEXT: [[TMP18:%.*]] = load i16, i16* [[TMP1]], align 2 2465 // CHECK2-NEXT: br label [[COND_END11:%.*]] 2466 // CHECK2: cond.false10: 2467 // CHECK2-NEXT: [[TMP19:%.*]] = load i16, i16* [[B2]], align 2 2468 // CHECK2-NEXT: br label [[COND_END11]] 2469 // CHECK2: cond.end11: 2470 // CHECK2-NEXT: [[COND12:%.*]] = phi i16 [ [[TMP18]], [[COND_TRUE9]] ], [ [[TMP19]], [[COND_FALSE10]] ] 2471 // CHECK2-NEXT: store i16 [[COND12]], i16* [[TMP1]], align 2 2472 // CHECK2-NEXT: call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP6]]) 2473 // CHECK2-NEXT: br label [[DOTOMP_REDUCTION_DONE]] 2474 // CHECK2: .omp.reduction.done: 2475 // CHECK2-NEXT: ret void 2476 // 2477 // 2478 // CHECK2-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func12 2479 // CHECK2-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR0]] { 2480 // CHECK2-NEXT: entry: 2481 // CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 2482 // CHECK2-NEXT: [[DOTADDR1:%.*]] = alloca i16, align 2 2483 // CHECK2-NEXT: [[DOTADDR2:%.*]] = alloca i16, align 2 2484 // CHECK2-NEXT: [[DOTADDR3:%.*]] = alloca i16, align 2 2485 // CHECK2-NEXT: [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [2 x i8*], align 8 2486 // CHECK2-NEXT: [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca i32, align 4 2487 // CHECK2-NEXT: [[DOTOMP_REDUCTION_ELEMENT4:%.*]] = alloca i16, align 2 2488 // CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 2489 // CHECK2-NEXT: store i16 [[TMP1]], i16* [[DOTADDR1]], align 2 2490 // CHECK2-NEXT: store i16 [[TMP2]], i16* [[DOTADDR2]], align 2 2491 // CHECK2-NEXT: store i16 [[TMP3]], i16* [[DOTADDR3]], align 2 2492 // CHECK2-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 8 2493 // CHECK2-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [2 x i8*]* 2494 // CHECK2-NEXT: [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2 2495 // CHECK2-NEXT: [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2 2496 // CHECK2-NEXT: [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2 2497 // CHECK2-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i64 0, i64 0 2498 // CHECK2-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8 2499 // CHECK2-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0 2500 // CHECK2-NEXT: [[TMP12:%.*]] = bitcast i8* [[TMP10]] to i32* 2501 // CHECK2-NEXT: [[TMP13:%.*]] = getelementptr i32, i32* [[TMP12]], i64 1 2502 // CHECK2-NEXT: [[TMP14:%.*]] = bitcast i32* [[TMP13]] to i8* 2503 // CHECK2-NEXT: [[TMP15:%.*]] = load i32, i32* [[TMP12]], align 4 2504 // CHECK2-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 2505 // CHECK2-NEXT: [[TMP16:%.*]] = trunc i32 [[NVPTX_WARP_SIZE]] to i16 2506 // CHECK2-NEXT: [[TMP17:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP15]], i16 [[TMP7]], i16 [[TMP16]]) 2507 // CHECK2-NEXT: store i32 [[TMP17]], i32* [[DOTOMP_REDUCTION_ELEMENT]], align 4 2508 // CHECK2-NEXT: [[TMP18:%.*]] = getelementptr i32, i32* [[TMP12]], i64 1 2509 // CHECK2-NEXT: [[TMP19:%.*]] = getelementptr i32, i32* [[DOTOMP_REDUCTION_ELEMENT]], i64 1 2510 // CHECK2-NEXT: [[TMP20:%.*]] = bitcast i32* [[DOTOMP_REDUCTION_ELEMENT]] to i8* 2511 // CHECK2-NEXT: store i8* [[TMP20]], i8** [[TMP11]], align 8 2512 // CHECK2-NEXT: [[TMP21:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i64 0, i64 1 2513 // CHECK2-NEXT: [[TMP22:%.*]] = load i8*, i8** [[TMP21]], align 8 2514 // CHECK2-NEXT: [[TMP23:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 1 2515 // CHECK2-NEXT: [[TMP24:%.*]] = bitcast i8* [[TMP22]] to i16* 2516 // CHECK2-NEXT: [[TMP25:%.*]] = getelementptr i16, i16* [[TMP24]], i64 1 2517 // CHECK2-NEXT: [[TMP26:%.*]] = bitcast i16* [[TMP25]] to i8* 2518 // CHECK2-NEXT: [[TMP27:%.*]] = load i16, i16* [[TMP24]], align 2 2519 // CHECK2-NEXT: [[TMP28:%.*]] = sext i16 [[TMP27]] to i32 2520 // CHECK2-NEXT: [[NVPTX_WARP_SIZE5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 2521 // CHECK2-NEXT: [[TMP29:%.*]] = trunc i32 [[NVPTX_WARP_SIZE5]] to i16 2522 // CHECK2-NEXT: [[TMP30:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP28]], i16 [[TMP7]], i16 [[TMP29]]) 2523 // CHECK2-NEXT: [[TMP31:%.*]] = trunc i32 [[TMP30]] to i16 2524 // CHECK2-NEXT: store i16 [[TMP31]], i16* [[DOTOMP_REDUCTION_ELEMENT4]], align 2 2525 // CHECK2-NEXT: [[TMP32:%.*]] = getelementptr i16, i16* [[TMP24]], i64 1 2526 // CHECK2-NEXT: [[TMP33:%.*]] = getelementptr i16, i16* [[DOTOMP_REDUCTION_ELEMENT4]], i64 1 2527 // CHECK2-NEXT: [[TMP34:%.*]] = bitcast i16* [[DOTOMP_REDUCTION_ELEMENT4]] to i8* 2528 // CHECK2-NEXT: store i8* [[TMP34]], i8** [[TMP23]], align 8 2529 // CHECK2-NEXT: [[TMP35:%.*]] = icmp eq i16 [[TMP8]], 0 2530 // CHECK2-NEXT: [[TMP36:%.*]] = icmp eq i16 [[TMP8]], 1 2531 // CHECK2-NEXT: [[TMP37:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]] 2532 // CHECK2-NEXT: [[TMP38:%.*]] = and i1 [[TMP36]], [[TMP37]] 2533 // CHECK2-NEXT: [[TMP39:%.*]] = icmp eq i16 [[TMP8]], 2 2534 // CHECK2-NEXT: [[TMP40:%.*]] = and i16 [[TMP6]], 1 2535 // CHECK2-NEXT: [[TMP41:%.*]] = icmp eq i16 [[TMP40]], 0 2536 // CHECK2-NEXT: [[TMP42:%.*]] = and i1 [[TMP39]], [[TMP41]] 2537 // CHECK2-NEXT: [[TMP43:%.*]] = icmp sgt i16 [[TMP7]], 0 2538 // CHECK2-NEXT: [[TMP44:%.*]] = and i1 [[TMP42]], [[TMP43]] 2539 // CHECK2-NEXT: [[TMP45:%.*]] = or i1 [[TMP35]], [[TMP38]] 2540 // CHECK2-NEXT: [[TMP46:%.*]] = or i1 [[TMP45]], [[TMP44]] 2541 // CHECK2-NEXT: br i1 [[TMP46]], label [[THEN:%.*]], label [[ELSE:%.*]] 2542 // CHECK2: then: 2543 // CHECK2-NEXT: [[TMP47:%.*]] = bitcast [2 x i8*]* [[TMP5]] to i8* 2544 // CHECK2-NEXT: [[TMP48:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8* 2545 // CHECK2-NEXT: call void @"_omp$reduction$reduction_func11"(i8* [[TMP47]], i8* [[TMP48]]) #[[ATTR3]] 2546 // CHECK2-NEXT: br label [[IFCONT:%.*]] 2547 // CHECK2: else: 2548 // CHECK2-NEXT: br label [[IFCONT]] 2549 // CHECK2: ifcont: 2550 // CHECK2-NEXT: [[TMP49:%.*]] = icmp eq i16 [[TMP8]], 1 2551 // CHECK2-NEXT: [[TMP50:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]] 2552 // CHECK2-NEXT: [[TMP51:%.*]] = and i1 [[TMP49]], [[TMP50]] 2553 // CHECK2-NEXT: br i1 [[TMP51]], label [[THEN6:%.*]], label [[ELSE7:%.*]] 2554 // CHECK2: then6: 2555 // CHECK2-NEXT: [[TMP52:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0 2556 // CHECK2-NEXT: [[TMP53:%.*]] = load i8*, i8** [[TMP52]], align 8 2557 // CHECK2-NEXT: [[TMP54:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i64 0, i64 0 2558 // CHECK2-NEXT: [[TMP55:%.*]] = load i8*, i8** [[TMP54]], align 8 2559 // CHECK2-NEXT: [[TMP56:%.*]] = bitcast i8* [[TMP53]] to i32* 2560 // CHECK2-NEXT: [[TMP57:%.*]] = bitcast i8* [[TMP55]] to i32* 2561 // CHECK2-NEXT: [[TMP58:%.*]] = load i32, i32* [[TMP56]], align 4 2562 // CHECK2-NEXT: store i32 [[TMP58]], i32* [[TMP57]], align 4 2563 // CHECK2-NEXT: [[TMP59:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 1 2564 // CHECK2-NEXT: [[TMP60:%.*]] = load i8*, i8** [[TMP59]], align 8 2565 // CHECK2-NEXT: [[TMP61:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i64 0, i64 1 2566 // CHECK2-NEXT: [[TMP62:%.*]] = load i8*, i8** [[TMP61]], align 8 2567 // CHECK2-NEXT: [[TMP63:%.*]] = bitcast i8* [[TMP60]] to i16* 2568 // CHECK2-NEXT: [[TMP64:%.*]] = bitcast i8* [[TMP62]] to i16* 2569 // CHECK2-NEXT: [[TMP65:%.*]] = load i16, i16* [[TMP63]], align 2 2570 // CHECK2-NEXT: store i16 [[TMP65]], i16* [[TMP64]], align 2 2571 // CHECK2-NEXT: br label [[IFCONT8:%.*]] 2572 // CHECK2: else7: 2573 // CHECK2-NEXT: br label [[IFCONT8]] 2574 // CHECK2: ifcont8: 2575 // CHECK2-NEXT: ret void 2576 // 2577 // 2578 // CHECK2-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func13 2579 // CHECK2-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] { 2580 // CHECK2-NEXT: entry: 2581 // CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 2582 // CHECK2-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 2583 // CHECK2-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]]) 2584 // CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 2585 // CHECK2-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 2586 // CHECK2-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 2587 // CHECK2-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 2588 // CHECK2-NEXT: [[NVPTX_LANE_ID:%.*]] = and i32 [[NVPTX_TID2]], 31 2589 // CHECK2-NEXT: [[NVPTX_TID3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 2590 // CHECK2-NEXT: [[NVPTX_WARP_ID:%.*]] = ashr i32 [[NVPTX_TID3]], 5 2591 // CHECK2-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8 2592 // CHECK2-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 2593 // CHECK2-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4:[0-9]+]], i32 [[TMP2]]) 2594 // CHECK2-NEXT: [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 2595 // CHECK2-NEXT: br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]] 2596 // CHECK2: then: 2597 // CHECK2-NEXT: [[TMP5:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 0 2598 // CHECK2-NEXT: [[TMP6:%.*]] = load i8*, i8** [[TMP5]], align 8 2599 // CHECK2-NEXT: [[TMP7:%.*]] = bitcast i8* [[TMP6]] to i32* 2600 // CHECK2-NEXT: [[TMP8:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 2601 // CHECK2-NEXT: [[TMP9:%.*]] = load i32, i32* [[TMP7]], align 4 2602 // CHECK2-NEXT: store volatile i32 [[TMP9]], i32 addrspace(3)* [[TMP8]], align 4 2603 // CHECK2-NEXT: br label [[IFCONT:%.*]] 2604 // CHECK2: else: 2605 // CHECK2-NEXT: br label [[IFCONT]] 2606 // CHECK2: ifcont: 2607 // CHECK2-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 2608 // CHECK2-NEXT: [[TMP10:%.*]] = load i32, i32* [[DOTADDR1]], align 4 2609 // CHECK2-NEXT: [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP10]] 2610 // CHECK2-NEXT: br i1 [[IS_ACTIVE_THREAD]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 2611 // CHECK2: then4: 2612 // CHECK2-NEXT: [[TMP11:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 2613 // CHECK2-NEXT: [[TMP12:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 0 2614 // CHECK2-NEXT: [[TMP13:%.*]] = load i8*, i8** [[TMP12]], align 8 2615 // CHECK2-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to i32* 2616 // CHECK2-NEXT: [[TMP15:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP11]], align 4 2617 // CHECK2-NEXT: store i32 [[TMP15]], i32* [[TMP14]], align 4 2618 // CHECK2-NEXT: br label [[IFCONT6:%.*]] 2619 // CHECK2: else5: 2620 // CHECK2-NEXT: br label [[IFCONT6]] 2621 // CHECK2: ifcont6: 2622 // CHECK2-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 2623 // CHECK2-NEXT: [[WARP_MASTER7:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 2624 // CHECK2-NEXT: br i1 [[WARP_MASTER7]], label [[THEN8:%.*]], label [[ELSE9:%.*]] 2625 // CHECK2: then8: 2626 // CHECK2-NEXT: [[TMP16:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 1 2627 // CHECK2-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 8 2628 // CHECK2-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i16* 2629 // CHECK2-NEXT: [[TMP19:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 2630 // CHECK2-NEXT: [[TMP20:%.*]] = bitcast i32 addrspace(3)* [[TMP19]] to i16 addrspace(3)* 2631 // CHECK2-NEXT: [[TMP21:%.*]] = load i16, i16* [[TMP18]], align 2 2632 // CHECK2-NEXT: store volatile i16 [[TMP21]], i16 addrspace(3)* [[TMP20]], align 2 2633 // CHECK2-NEXT: br label [[IFCONT10:%.*]] 2634 // CHECK2: else9: 2635 // CHECK2-NEXT: br label [[IFCONT10]] 2636 // CHECK2: ifcont10: 2637 // CHECK2-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 2638 // CHECK2-NEXT: [[TMP22:%.*]] = load i32, i32* [[DOTADDR1]], align 4 2639 // CHECK2-NEXT: [[IS_ACTIVE_THREAD11:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP22]] 2640 // CHECK2-NEXT: br i1 [[IS_ACTIVE_THREAD11]], label [[THEN12:%.*]], label [[ELSE13:%.*]] 2641 // CHECK2: then12: 2642 // CHECK2-NEXT: [[TMP23:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 2643 // CHECK2-NEXT: [[TMP24:%.*]] = bitcast i32 addrspace(3)* [[TMP23]] to i16 addrspace(3)* 2644 // CHECK2-NEXT: [[TMP25:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 1 2645 // CHECK2-NEXT: [[TMP26:%.*]] = load i8*, i8** [[TMP25]], align 8 2646 // CHECK2-NEXT: [[TMP27:%.*]] = bitcast i8* [[TMP26]] to i16* 2647 // CHECK2-NEXT: [[TMP28:%.*]] = load volatile i16, i16 addrspace(3)* [[TMP24]], align 2 2648 // CHECK2-NEXT: store i16 [[TMP28]], i16* [[TMP27]], align 2 2649 // CHECK2-NEXT: br label [[IFCONT14:%.*]] 2650 // CHECK2: else13: 2651 // CHECK2-NEXT: br label [[IFCONT14]] 2652 // CHECK2: ifcont14: 2653 // CHECK2-NEXT: ret void 2654 // 2655 // 2656 // CHECK2-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func15 2657 // CHECK2-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR0]] { 2658 // CHECK2-NEXT: entry: 2659 // CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 2660 // CHECK2-NEXT: [[DOTADDR1:%.*]] = alloca i16, align 2 2661 // CHECK2-NEXT: [[DOTADDR2:%.*]] = alloca i16, align 2 2662 // CHECK2-NEXT: [[DOTADDR3:%.*]] = alloca i16, align 2 2663 // CHECK2-NEXT: [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [2 x i8*], align 8 2664 // CHECK2-NEXT: [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca i32, align 4 2665 // CHECK2-NEXT: [[DOTOMP_REDUCTION_ELEMENT4:%.*]] = alloca i16, align 2 2666 // CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 2667 // CHECK2-NEXT: store i16 [[TMP1]], i16* [[DOTADDR1]], align 2 2668 // CHECK2-NEXT: store i16 [[TMP2]], i16* [[DOTADDR2]], align 2 2669 // CHECK2-NEXT: store i16 [[TMP3]], i16* [[DOTADDR3]], align 2 2670 // CHECK2-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 8 2671 // CHECK2-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [2 x i8*]* 2672 // CHECK2-NEXT: [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2 2673 // CHECK2-NEXT: [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2 2674 // CHECK2-NEXT: [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2 2675 // CHECK2-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i64 0, i64 0 2676 // CHECK2-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8 2677 // CHECK2-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0 2678 // CHECK2-NEXT: [[TMP12:%.*]] = bitcast i8* [[TMP10]] to i32* 2679 // CHECK2-NEXT: [[TMP13:%.*]] = getelementptr i32, i32* [[TMP12]], i64 1 2680 // CHECK2-NEXT: [[TMP14:%.*]] = bitcast i32* [[TMP13]] to i8* 2681 // CHECK2-NEXT: [[TMP15:%.*]] = load i32, i32* [[TMP12]], align 4 2682 // CHECK2-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 2683 // CHECK2-NEXT: [[TMP16:%.*]] = trunc i32 [[NVPTX_WARP_SIZE]] to i16 2684 // CHECK2-NEXT: [[TMP17:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP15]], i16 [[TMP7]], i16 [[TMP16]]) 2685 // CHECK2-NEXT: store i32 [[TMP17]], i32* [[DOTOMP_REDUCTION_ELEMENT]], align 4 2686 // CHECK2-NEXT: [[TMP18:%.*]] = getelementptr i32, i32* [[TMP12]], i64 1 2687 // CHECK2-NEXT: [[TMP19:%.*]] = getelementptr i32, i32* [[DOTOMP_REDUCTION_ELEMENT]], i64 1 2688 // CHECK2-NEXT: [[TMP20:%.*]] = bitcast i32* [[DOTOMP_REDUCTION_ELEMENT]] to i8* 2689 // CHECK2-NEXT: store i8* [[TMP20]], i8** [[TMP11]], align 8 2690 // CHECK2-NEXT: [[TMP21:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i64 0, i64 1 2691 // CHECK2-NEXT: [[TMP22:%.*]] = load i8*, i8** [[TMP21]], align 8 2692 // CHECK2-NEXT: [[TMP23:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 1 2693 // CHECK2-NEXT: [[TMP24:%.*]] = bitcast i8* [[TMP22]] to i16* 2694 // CHECK2-NEXT: [[TMP25:%.*]] = getelementptr i16, i16* [[TMP24]], i64 1 2695 // CHECK2-NEXT: [[TMP26:%.*]] = bitcast i16* [[TMP25]] to i8* 2696 // CHECK2-NEXT: [[TMP27:%.*]] = load i16, i16* [[TMP24]], align 2 2697 // CHECK2-NEXT: [[TMP28:%.*]] = sext i16 [[TMP27]] to i32 2698 // CHECK2-NEXT: [[NVPTX_WARP_SIZE5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 2699 // CHECK2-NEXT: [[TMP29:%.*]] = trunc i32 [[NVPTX_WARP_SIZE5]] to i16 2700 // CHECK2-NEXT: [[TMP30:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP28]], i16 [[TMP7]], i16 [[TMP29]]) 2701 // CHECK2-NEXT: [[TMP31:%.*]] = trunc i32 [[TMP30]] to i16 2702 // CHECK2-NEXT: store i16 [[TMP31]], i16* [[DOTOMP_REDUCTION_ELEMENT4]], align 2 2703 // CHECK2-NEXT: [[TMP32:%.*]] = getelementptr i16, i16* [[TMP24]], i64 1 2704 // CHECK2-NEXT: [[TMP33:%.*]] = getelementptr i16, i16* [[DOTOMP_REDUCTION_ELEMENT4]], i64 1 2705 // CHECK2-NEXT: [[TMP34:%.*]] = bitcast i16* [[DOTOMP_REDUCTION_ELEMENT4]] to i8* 2706 // CHECK2-NEXT: store i8* [[TMP34]], i8** [[TMP23]], align 8 2707 // CHECK2-NEXT: [[TMP35:%.*]] = icmp eq i16 [[TMP8]], 0 2708 // CHECK2-NEXT: [[TMP36:%.*]] = icmp eq i16 [[TMP8]], 1 2709 // CHECK2-NEXT: [[TMP37:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]] 2710 // CHECK2-NEXT: [[TMP38:%.*]] = and i1 [[TMP36]], [[TMP37]] 2711 // CHECK2-NEXT: [[TMP39:%.*]] = icmp eq i16 [[TMP8]], 2 2712 // CHECK2-NEXT: [[TMP40:%.*]] = and i16 [[TMP6]], 1 2713 // CHECK2-NEXT: [[TMP41:%.*]] = icmp eq i16 [[TMP40]], 0 2714 // CHECK2-NEXT: [[TMP42:%.*]] = and i1 [[TMP39]], [[TMP41]] 2715 // CHECK2-NEXT: [[TMP43:%.*]] = icmp sgt i16 [[TMP7]], 0 2716 // CHECK2-NEXT: [[TMP44:%.*]] = and i1 [[TMP42]], [[TMP43]] 2717 // CHECK2-NEXT: [[TMP45:%.*]] = or i1 [[TMP35]], [[TMP38]] 2718 // CHECK2-NEXT: [[TMP46:%.*]] = or i1 [[TMP45]], [[TMP44]] 2719 // CHECK2-NEXT: br i1 [[TMP46]], label [[THEN:%.*]], label [[ELSE:%.*]] 2720 // CHECK2: then: 2721 // CHECK2-NEXT: [[TMP47:%.*]] = bitcast [2 x i8*]* [[TMP5]] to i8* 2722 // CHECK2-NEXT: [[TMP48:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8* 2723 // CHECK2-NEXT: call void @"_omp$reduction$reduction_func14"(i8* [[TMP47]], i8* [[TMP48]]) #[[ATTR3]] 2724 // CHECK2-NEXT: br label [[IFCONT:%.*]] 2725 // CHECK2: else: 2726 // CHECK2-NEXT: br label [[IFCONT]] 2727 // CHECK2: ifcont: 2728 // CHECK2-NEXT: [[TMP49:%.*]] = icmp eq i16 [[TMP8]], 1 2729 // CHECK2-NEXT: [[TMP50:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]] 2730 // CHECK2-NEXT: [[TMP51:%.*]] = and i1 [[TMP49]], [[TMP50]] 2731 // CHECK2-NEXT: br i1 [[TMP51]], label [[THEN6:%.*]], label [[ELSE7:%.*]] 2732 // CHECK2: then6: 2733 // CHECK2-NEXT: [[TMP52:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0 2734 // CHECK2-NEXT: [[TMP53:%.*]] = load i8*, i8** [[TMP52]], align 8 2735 // CHECK2-NEXT: [[TMP54:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i64 0, i64 0 2736 // CHECK2-NEXT: [[TMP55:%.*]] = load i8*, i8** [[TMP54]], align 8 2737 // CHECK2-NEXT: [[TMP56:%.*]] = bitcast i8* [[TMP53]] to i32* 2738 // CHECK2-NEXT: [[TMP57:%.*]] = bitcast i8* [[TMP55]] to i32* 2739 // CHECK2-NEXT: [[TMP58:%.*]] = load i32, i32* [[TMP56]], align 4 2740 // CHECK2-NEXT: store i32 [[TMP58]], i32* [[TMP57]], align 4 2741 // CHECK2-NEXT: [[TMP59:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 1 2742 // CHECK2-NEXT: [[TMP60:%.*]] = load i8*, i8** [[TMP59]], align 8 2743 // CHECK2-NEXT: [[TMP61:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i64 0, i64 1 2744 // CHECK2-NEXT: [[TMP62:%.*]] = load i8*, i8** [[TMP61]], align 8 2745 // CHECK2-NEXT: [[TMP63:%.*]] = bitcast i8* [[TMP60]] to i16* 2746 // CHECK2-NEXT: [[TMP64:%.*]] = bitcast i8* [[TMP62]] to i16* 2747 // CHECK2-NEXT: [[TMP65:%.*]] = load i16, i16* [[TMP63]], align 2 2748 // CHECK2-NEXT: store i16 [[TMP65]], i16* [[TMP64]], align 2 2749 // CHECK2-NEXT: br label [[IFCONT8:%.*]] 2750 // CHECK2: else7: 2751 // CHECK2-NEXT: br label [[IFCONT8]] 2752 // CHECK2: ifcont8: 2753 // CHECK2-NEXT: ret void 2754 // 2755 // 2756 // CHECK2-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func16 2757 // CHECK2-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] { 2758 // CHECK2-NEXT: entry: 2759 // CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 2760 // CHECK2-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 2761 // CHECK2-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]]) 2762 // CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 2763 // CHECK2-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 2764 // CHECK2-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 2765 // CHECK2-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 2766 // CHECK2-NEXT: [[NVPTX_LANE_ID:%.*]] = and i32 [[NVPTX_TID2]], 31 2767 // CHECK2-NEXT: [[NVPTX_TID3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 2768 // CHECK2-NEXT: [[NVPTX_WARP_ID:%.*]] = ashr i32 [[NVPTX_TID3]], 5 2769 // CHECK2-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8 2770 // CHECK2-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 2771 // CHECK2-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 2772 // CHECK2-NEXT: [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 2773 // CHECK2-NEXT: br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]] 2774 // CHECK2: then: 2775 // CHECK2-NEXT: [[TMP5:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 0 2776 // CHECK2-NEXT: [[TMP6:%.*]] = load i8*, i8** [[TMP5]], align 8 2777 // CHECK2-NEXT: [[TMP7:%.*]] = bitcast i8* [[TMP6]] to i32* 2778 // CHECK2-NEXT: [[TMP8:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 2779 // CHECK2-NEXT: [[TMP9:%.*]] = load i32, i32* [[TMP7]], align 4 2780 // CHECK2-NEXT: store volatile i32 [[TMP9]], i32 addrspace(3)* [[TMP8]], align 4 2781 // CHECK2-NEXT: br label [[IFCONT:%.*]] 2782 // CHECK2: else: 2783 // CHECK2-NEXT: br label [[IFCONT]] 2784 // CHECK2: ifcont: 2785 // CHECK2-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 2786 // CHECK2-NEXT: [[TMP10:%.*]] = load i32, i32* [[DOTADDR1]], align 4 2787 // CHECK2-NEXT: [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP10]] 2788 // CHECK2-NEXT: br i1 [[IS_ACTIVE_THREAD]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 2789 // CHECK2: then4: 2790 // CHECK2-NEXT: [[TMP11:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 2791 // CHECK2-NEXT: [[TMP12:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 0 2792 // CHECK2-NEXT: [[TMP13:%.*]] = load i8*, i8** [[TMP12]], align 8 2793 // CHECK2-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to i32* 2794 // CHECK2-NEXT: [[TMP15:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP11]], align 4 2795 // CHECK2-NEXT: store i32 [[TMP15]], i32* [[TMP14]], align 4 2796 // CHECK2-NEXT: br label [[IFCONT6:%.*]] 2797 // CHECK2: else5: 2798 // CHECK2-NEXT: br label [[IFCONT6]] 2799 // CHECK2: ifcont6: 2800 // CHECK2-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 2801 // CHECK2-NEXT: [[WARP_MASTER7:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 2802 // CHECK2-NEXT: br i1 [[WARP_MASTER7]], label [[THEN8:%.*]], label [[ELSE9:%.*]] 2803 // CHECK2: then8: 2804 // CHECK2-NEXT: [[TMP16:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 1 2805 // CHECK2-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 8 2806 // CHECK2-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i16* 2807 // CHECK2-NEXT: [[TMP19:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 2808 // CHECK2-NEXT: [[TMP20:%.*]] = bitcast i32 addrspace(3)* [[TMP19]] to i16 addrspace(3)* 2809 // CHECK2-NEXT: [[TMP21:%.*]] = load i16, i16* [[TMP18]], align 2 2810 // CHECK2-NEXT: store volatile i16 [[TMP21]], i16 addrspace(3)* [[TMP20]], align 2 2811 // CHECK2-NEXT: br label [[IFCONT10:%.*]] 2812 // CHECK2: else9: 2813 // CHECK2-NEXT: br label [[IFCONT10]] 2814 // CHECK2: ifcont10: 2815 // CHECK2-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 2816 // CHECK2-NEXT: [[TMP22:%.*]] = load i32, i32* [[DOTADDR1]], align 4 2817 // CHECK2-NEXT: [[IS_ACTIVE_THREAD11:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP22]] 2818 // CHECK2-NEXT: br i1 [[IS_ACTIVE_THREAD11]], label [[THEN12:%.*]], label [[ELSE13:%.*]] 2819 // CHECK2: then12: 2820 // CHECK2-NEXT: [[TMP23:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 2821 // CHECK2-NEXT: [[TMP24:%.*]] = bitcast i32 addrspace(3)* [[TMP23]] to i16 addrspace(3)* 2822 // CHECK2-NEXT: [[TMP25:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 1 2823 // CHECK2-NEXT: [[TMP26:%.*]] = load i8*, i8** [[TMP25]], align 8 2824 // CHECK2-NEXT: [[TMP27:%.*]] = bitcast i8* [[TMP26]] to i16* 2825 // CHECK2-NEXT: [[TMP28:%.*]] = load volatile i16, i16 addrspace(3)* [[TMP24]], align 2 2826 // CHECK2-NEXT: store i16 [[TMP28]], i16* [[TMP27]], align 2 2827 // CHECK2-NEXT: br label [[IFCONT14:%.*]] 2828 // CHECK2: else13: 2829 // CHECK2-NEXT: br label [[IFCONT14]] 2830 // CHECK2: ifcont14: 2831 // CHECK2-NEXT: ret void 2832 // 2833 // 2834 // CHECK2-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_copy_func17 2835 // CHECK2-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 2836 // CHECK2-NEXT: entry: 2837 // CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 2838 // CHECK2-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 2839 // CHECK2-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 8 2840 // CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 2841 // CHECK2-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 2842 // CHECK2-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 8 2843 // CHECK2-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 8 2844 // CHECK2-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 2845 // CHECK2-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 8 2846 // CHECK2-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.5* 2847 // CHECK2-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 2848 // CHECK2-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 0 2849 // CHECK2-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 8 2850 // CHECK2-NEXT: [[TMP10:%.*]] = bitcast i8* [[TMP9]] to i32* 2851 // CHECK2-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_5:%.*]], %struct._globalized_locals_ty.5* [[TMP6]], i32 0, i32 0 2852 // CHECK2-NEXT: [[TMP11:%.*]] = getelementptr inbounds [1024 x i32], [1024 x i32]* [[A]], i32 0, i32 [[TMP7]] 2853 // CHECK2-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP10]], align 4 2854 // CHECK2-NEXT: store i32 [[TMP12]], i32* [[TMP11]], align 128 2855 // CHECK2-NEXT: [[TMP13:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 1 2856 // CHECK2-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 8 2857 // CHECK2-NEXT: [[TMP15:%.*]] = bitcast i8* [[TMP14]] to i16* 2858 // CHECK2-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_5]], %struct._globalized_locals_ty.5* [[TMP6]], i32 0, i32 1 2859 // CHECK2-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1024 x i16], [1024 x i16]* [[B]], i32 0, i32 [[TMP7]] 2860 // CHECK2-NEXT: [[TMP17:%.*]] = load i16, i16* [[TMP15]], align 2 2861 // CHECK2-NEXT: store i16 [[TMP17]], i16* [[TMP16]], align 128 2862 // CHECK2-NEXT: ret void 2863 // 2864 // 2865 // CHECK2-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_reduce_func18 2866 // CHECK2-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 2867 // CHECK2-NEXT: entry: 2868 // CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 2869 // CHECK2-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 2870 // CHECK2-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 8 2871 // CHECK2-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 8 2872 // CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 2873 // CHECK2-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 2874 // CHECK2-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 8 2875 // CHECK2-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8 2876 // CHECK2-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.5* 2877 // CHECK2-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 2878 // CHECK2-NEXT: [[TMP6:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0 2879 // CHECK2-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_5:%.*]], %struct._globalized_locals_ty.5* [[TMP4]], i32 0, i32 0 2880 // CHECK2-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1024 x i32], [1024 x i32]* [[A]], i32 0, i32 [[TMP5]] 2881 // CHECK2-NEXT: [[TMP8:%.*]] = bitcast i32* [[TMP7]] to i8* 2882 // CHECK2-NEXT: store i8* [[TMP8]], i8** [[TMP6]], align 8 2883 // CHECK2-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 1 2884 // CHECK2-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_5]], %struct._globalized_locals_ty.5* [[TMP4]], i32 0, i32 1 2885 // CHECK2-NEXT: [[TMP10:%.*]] = getelementptr inbounds [1024 x i16], [1024 x i16]* [[B]], i32 0, i32 [[TMP5]] 2886 // CHECK2-NEXT: [[TMP11:%.*]] = bitcast i16* [[TMP10]] to i8* 2887 // CHECK2-NEXT: store i8* [[TMP11]], i8** [[TMP9]], align 8 2888 // CHECK2-NEXT: [[TMP12:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 2889 // CHECK2-NEXT: [[TMP13:%.*]] = load i8*, i8** [[DOTADDR2]], align 8 2890 // CHECK2-NEXT: call void @"_omp$reduction$reduction_func14"(i8* [[TMP12]], i8* [[TMP13]]) #[[ATTR3]] 2891 // CHECK2-NEXT: ret void 2892 // 2893 // 2894 // CHECK2-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_copy_func19 2895 // CHECK2-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 2896 // CHECK2-NEXT: entry: 2897 // CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 2898 // CHECK2-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 2899 // CHECK2-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 8 2900 // CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 2901 // CHECK2-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 2902 // CHECK2-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 8 2903 // CHECK2-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 8 2904 // CHECK2-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 2905 // CHECK2-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 8 2906 // CHECK2-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.5* 2907 // CHECK2-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 2908 // CHECK2-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 0 2909 // CHECK2-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 8 2910 // CHECK2-NEXT: [[TMP10:%.*]] = bitcast i8* [[TMP9]] to i32* 2911 // CHECK2-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_5:%.*]], %struct._globalized_locals_ty.5* [[TMP6]], i32 0, i32 0 2912 // CHECK2-NEXT: [[TMP11:%.*]] = getelementptr inbounds [1024 x i32], [1024 x i32]* [[A]], i32 0, i32 [[TMP7]] 2913 // CHECK2-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 128 2914 // CHECK2-NEXT: store i32 [[TMP12]], i32* [[TMP10]], align 4 2915 // CHECK2-NEXT: [[TMP13:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i64 0, i64 1 2916 // CHECK2-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 8 2917 // CHECK2-NEXT: [[TMP15:%.*]] = bitcast i8* [[TMP14]] to i16* 2918 // CHECK2-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_5]], %struct._globalized_locals_ty.5* [[TMP6]], i32 0, i32 1 2919 // CHECK2-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1024 x i16], [1024 x i16]* [[B]], i32 0, i32 [[TMP7]] 2920 // CHECK2-NEXT: [[TMP17:%.*]] = load i16, i16* [[TMP16]], align 128 2921 // CHECK2-NEXT: store i16 [[TMP17]], i16* [[TMP15]], align 2 2922 // CHECK2-NEXT: ret void 2923 // 2924 // 2925 // CHECK2-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_reduce_func20 2926 // CHECK2-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 2927 // CHECK2-NEXT: entry: 2928 // CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8 2929 // CHECK2-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 2930 // CHECK2-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 8 2931 // CHECK2-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 8 2932 // CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8 2933 // CHECK2-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 2934 // CHECK2-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 8 2935 // CHECK2-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8 2936 // CHECK2-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.5* 2937 // CHECK2-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 2938 // CHECK2-NEXT: [[TMP6:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0 2939 // CHECK2-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_5:%.*]], %struct._globalized_locals_ty.5* [[TMP4]], i32 0, i32 0 2940 // CHECK2-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1024 x i32], [1024 x i32]* [[A]], i32 0, i32 [[TMP5]] 2941 // CHECK2-NEXT: [[TMP8:%.*]] = bitcast i32* [[TMP7]] to i8* 2942 // CHECK2-NEXT: store i8* [[TMP8]], i8** [[TMP6]], align 8 2943 // CHECK2-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 1 2944 // CHECK2-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_5]], %struct._globalized_locals_ty.5* [[TMP4]], i32 0, i32 1 2945 // CHECK2-NEXT: [[TMP10:%.*]] = getelementptr inbounds [1024 x i16], [1024 x i16]* [[B]], i32 0, i32 [[TMP5]] 2946 // CHECK2-NEXT: [[TMP11:%.*]] = bitcast i16* [[TMP10]] to i8* 2947 // CHECK2-NEXT: store i8* [[TMP11]], i8** [[TMP9]], align 8 2948 // CHECK2-NEXT: [[TMP12:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 2949 // CHECK2-NEXT: [[TMP13:%.*]] = load i8*, i8** [[DOTADDR2]], align 8 2950 // CHECK2-NEXT: call void @"_omp$reduction$reduction_func14"(i8* [[TMP13]], i8* [[TMP12]]) #[[ATTR3]] 2951 // CHECK2-NEXT: ret void 2952 // 2953 // 2954 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l23_worker 2955 // CHECK3-SAME: () #[[ATTR0:[0-9]+]] { 2956 // CHECK3-NEXT: entry: 2957 // CHECK3-NEXT: [[WORK_FN:%.*]] = alloca i8*, align 4 2958 // CHECK3-NEXT: [[EXEC_STATUS:%.*]] = alloca i8, align 1 2959 // CHECK3-NEXT: store i8* null, i8** [[WORK_FN]], align 4 2960 // CHECK3-NEXT: store i8 0, i8* [[EXEC_STATUS]], align 1 2961 // CHECK3-NEXT: br label [[DOTAWAIT_WORK:%.*]] 2962 // CHECK3: .await.work: 2963 // CHECK3-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 2964 // CHECK3-NEXT: [[TMP0:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORK_FN]]) 2965 // CHECK3-NEXT: [[TMP1:%.*]] = zext i1 [[TMP0]] to i8 2966 // CHECK3-NEXT: store i8 [[TMP1]], i8* [[EXEC_STATUS]], align 1 2967 // CHECK3-NEXT: [[TMP2:%.*]] = load i8*, i8** [[WORK_FN]], align 4 2968 // CHECK3-NEXT: [[SHOULD_TERMINATE:%.*]] = icmp eq i8* [[TMP2]], null 2969 // CHECK3-NEXT: br i1 [[SHOULD_TERMINATE]], label [[DOTEXIT:%.*]], label [[DOTSELECT_WORKERS:%.*]] 2970 // CHECK3: .select.workers: 2971 // CHECK3-NEXT: [[TMP3:%.*]] = load i8, i8* [[EXEC_STATUS]], align 1 2972 // CHECK3-NEXT: [[IS_ACTIVE:%.*]] = icmp ne i8 [[TMP3]], 0 2973 // CHECK3-NEXT: br i1 [[IS_ACTIVE]], label [[DOTEXECUTE_PARALLEL:%.*]], label [[DOTBARRIER_PARALLEL:%.*]] 2974 // CHECK3: .execute.parallel: 2975 // CHECK3-NEXT: [[TMP4:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]]) 2976 // CHECK3-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP2]] to void (i16, i32)* 2977 // CHECK3-NEXT: call void [[TMP5]](i16 0, i32 [[TMP4]]) 2978 // CHECK3-NEXT: br label [[DOTTERMINATE_PARALLEL:%.*]] 2979 // CHECK3: .terminate.parallel: 2980 // CHECK3-NEXT: call void @__kmpc_kernel_end_parallel() 2981 // CHECK3-NEXT: br label [[DOTBARRIER_PARALLEL]] 2982 // CHECK3: .barrier.parallel: 2983 // CHECK3-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 2984 // CHECK3-NEXT: br label [[DOTAWAIT_WORK]] 2985 // CHECK3: .exit: 2986 // CHECK3-NEXT: ret void 2987 // 2988 // 2989 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l23 2990 // CHECK3-SAME: (double* nonnull align 8 dereferenceable(8) [[E:%.*]]) #[[ATTR1:[0-9]+]] { 2991 // CHECK3-NEXT: entry: 2992 // CHECK3-NEXT: [[E_ADDR:%.*]] = alloca double*, align 4 2993 // CHECK3-NEXT: [[E7:%.*]] = alloca double, align 8 2994 // CHECK3-NEXT: [[DOTZERO_ADDR:%.*]] = alloca i32, align 4 2995 // CHECK3-NEXT: [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4 2996 // CHECK3-NEXT: store i32 0, i32* [[DOTZERO_ADDR]], align 4 2997 // CHECK3-NEXT: store double* [[E]], double** [[E_ADDR]], align 4 2998 // CHECK3-NEXT: [[TMP0:%.*]] = load double*, double** [[E_ADDR]], align 4 2999 // CHECK3-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 3000 // CHECK3-NEXT: [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 3001 // CHECK3-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 3002 // CHECK3-NEXT: [[THREAD_LIMIT:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS]], [[NVPTX_WARP_SIZE]] 3003 // CHECK3-NEXT: [[TMP1:%.*]] = icmp ult i32 [[NVPTX_TID]], [[THREAD_LIMIT]] 3004 // CHECK3-NEXT: br i1 [[TMP1]], label [[DOTWORKER:%.*]], label [[DOTMASTERCHECK:%.*]] 3005 // CHECK3: .worker: 3006 // CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l23_worker() #[[ATTR3:[0-9]+]] 3007 // CHECK3-NEXT: br label [[DOTEXIT:%.*]] 3008 // CHECK3: .mastercheck: 3009 // CHECK3-NEXT: [[NVPTX_TID1:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 3010 // CHECK3-NEXT: [[NVPTX_NUM_THREADS2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 3011 // CHECK3-NEXT: [[NVPTX_WARP_SIZE3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 3012 // CHECK3-NEXT: [[TMP2:%.*]] = sub nuw i32 [[NVPTX_WARP_SIZE3]], 1 3013 // CHECK3-NEXT: [[TMP3:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS2]], 1 3014 // CHECK3-NEXT: [[TMP4:%.*]] = xor i32 [[TMP2]], -1 3015 // CHECK3-NEXT: [[MASTER_TID:%.*]] = and i32 [[TMP3]], [[TMP4]] 3016 // CHECK3-NEXT: [[TMP5:%.*]] = icmp eq i32 [[NVPTX_TID1]], [[MASTER_TID]] 3017 // CHECK3-NEXT: br i1 [[TMP5]], label [[DOTMASTER:%.*]], label [[DOTEXIT]] 3018 // CHECK3: .master: 3019 // CHECK3-NEXT: [[NVPTX_NUM_THREADS4:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 3020 // CHECK3-NEXT: [[NVPTX_WARP_SIZE5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 3021 // CHECK3-NEXT: [[THREAD_LIMIT6:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS4]], [[NVPTX_WARP_SIZE5]] 3022 // CHECK3-NEXT: call void @__kmpc_kernel_init(i32 [[THREAD_LIMIT6]], i16 1) 3023 // CHECK3-NEXT: call void @__kmpc_data_sharing_init_stack() 3024 // CHECK3-NEXT: [[TMP6:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 3025 // CHECK3-NEXT: [[TMP7:%.*]] = load double, double* [[TMP0]], align 8 3026 // CHECK3-NEXT: store double [[TMP7]], double* [[E7]], align 8 3027 // CHECK3-NEXT: store i32 [[TMP6]], i32* [[DOTTHREADID_TEMP_]], align 4 3028 // CHECK3-NEXT: call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], double* [[E7]]) #[[ATTR3]] 3029 // CHECK3-NEXT: br label [[DOTTERMINATION_NOTIFIER:%.*]] 3030 // CHECK3: .termination.notifier: 3031 // CHECK3-NEXT: call void @__kmpc_kernel_deinit(i16 1) 3032 // CHECK3-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 3033 // CHECK3-NEXT: br label [[DOTEXIT]] 3034 // CHECK3: .exit: 3035 // CHECK3-NEXT: ret void 3036 // 3037 // 3038 // CHECK3-LABEL: define {{[^@]+}}@__omp_outlined__ 3039 // CHECK3-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], double* nonnull align 8 dereferenceable(8) [[E:%.*]]) #[[ATTR1]] { 3040 // CHECK3-NEXT: entry: 3041 // CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4 3042 // CHECK3-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4 3043 // CHECK3-NEXT: [[E_ADDR:%.*]] = alloca double*, align 4 3044 // CHECK3-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x i8*], align 4 3045 // CHECK3-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4 3046 // CHECK3-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4 3047 // CHECK3-NEXT: store double* [[E]], double** [[E_ADDR]], align 4 3048 // CHECK3-NEXT: [[TMP0:%.*]] = load double*, double** [[E_ADDR]], align 4 3049 // CHECK3-NEXT: [[TMP1:%.*]] = load i16, i16* @"_openmp_static_kernel$is_shared", align 2 3050 // CHECK3-NEXT: [[TMP2:%.*]] = load i32, i32* @"_openmp_static_kernel$size", align 4 3051 // CHECK3-NEXT: call void @__kmpc_get_team_static_memory(i16 0, i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds (%"union._shared_openmp_static_memory_type_$_", %"union._shared_openmp_static_memory_type_$_" addrspace(3)* @"_openmp_shared_static_glob_rd_$_", i32 0, i32 0, i32 0) to i8*), i32 [[TMP2]], i16 [[TMP1]], i8** addrspacecast (i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr" to i8**)) 3052 // CHECK3-NEXT: [[TMP3:%.*]] = load i8*, i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr", align 4 3053 // CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds i8, i8* [[TMP3]], i32 0 3054 // CHECK3-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to %struct._globalized_locals_ty* 3055 // CHECK3-NEXT: [[E1:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY:%.*]], %struct._globalized_locals_ty* [[TMP5]], i32 0, i32 0 3056 // CHECK3-NEXT: store double 0.000000e+00, double* [[E1]], align 8 3057 // CHECK3-NEXT: [[TMP6:%.*]] = load double, double* [[E1]], align 8 3058 // CHECK3-NEXT: [[ADD:%.*]] = fadd double [[TMP6]], 5.000000e+00 3059 // CHECK3-NEXT: store double [[ADD]], double* [[E1]], align 8 3060 // CHECK3-NEXT: [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4 3061 // CHECK3-NEXT: [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4 3062 // CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 3063 // CHECK3-NEXT: [[TMP10:%.*]] = bitcast double* [[E1]] to i8* 3064 // CHECK3-NEXT: store i8* [[TMP10]], i8** [[TMP9]], align 4 3065 // CHECK3-NEXT: [[TMP11:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 3066 // CHECK3-NEXT: [[TMP12:%.*]] = load i8*, i8** @"_openmp_teams_reductions_buffer_$_$ptr", align 4 3067 // CHECK3-NEXT: [[TMP13:%.*]] = call i32 @__kmpc_nvptx_teams_reduce_nowait_v2(%struct.ident_t* @[[GLOB1]], i32 [[TMP8]], i8* [[TMP12]], i32 1024, i8* [[TMP11]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_copy_func, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_reduce_func, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_copy_func, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_reduce_func) 3068 // CHECK3-NEXT: [[TMP14:%.*]] = icmp eq i32 [[TMP13]], 1 3069 // CHECK3-NEXT: br i1 [[TMP14]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]] 3070 // CHECK3: .omp.reduction.then: 3071 // CHECK3-NEXT: [[TMP15:%.*]] = load double, double* [[TMP0]], align 8 3072 // CHECK3-NEXT: [[TMP16:%.*]] = load double, double* [[E1]], align 8 3073 // CHECK3-NEXT: [[ADD2:%.*]] = fadd double [[TMP15]], [[TMP16]] 3074 // CHECK3-NEXT: store double [[ADD2]], double* [[TMP0]], align 8 3075 // CHECK3-NEXT: call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP8]]) 3076 // CHECK3-NEXT: br label [[DOTOMP_REDUCTION_DONE]] 3077 // CHECK3: .omp.reduction.done: 3078 // CHECK3-NEXT: [[TMP17:%.*]] = load i16, i16* @"_openmp_static_kernel$is_shared", align 2 3079 // CHECK3-NEXT: call void @__kmpc_restore_team_static_memory(i16 0, i16 [[TMP17]]) 3080 // CHECK3-NEXT: ret void 3081 // 3082 // 3083 // CHECK3-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func 3084 // CHECK3-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR0]] { 3085 // CHECK3-NEXT: entry: 3086 // CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 3087 // CHECK3-NEXT: [[DOTADDR1:%.*]] = alloca i16, align 2 3088 // CHECK3-NEXT: [[DOTADDR2:%.*]] = alloca i16, align 2 3089 // CHECK3-NEXT: [[DOTADDR3:%.*]] = alloca i16, align 2 3090 // CHECK3-NEXT: [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [1 x i8*], align 4 3091 // CHECK3-NEXT: [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca double, align 8 3092 // CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 3093 // CHECK3-NEXT: store i16 [[TMP1]], i16* [[DOTADDR1]], align 2 3094 // CHECK3-NEXT: store i16 [[TMP2]], i16* [[DOTADDR2]], align 2 3095 // CHECK3-NEXT: store i16 [[TMP3]], i16* [[DOTADDR3]], align 2 3096 // CHECK3-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 4 3097 // CHECK3-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [1 x i8*]* 3098 // CHECK3-NEXT: [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2 3099 // CHECK3-NEXT: [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2 3100 // CHECK3-NEXT: [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2 3101 // CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i32 0, i32 0 3102 // CHECK3-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 4 3103 // CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 3104 // CHECK3-NEXT: [[TMP12:%.*]] = bitcast i8* [[TMP10]] to double* 3105 // CHECK3-NEXT: [[TMP13:%.*]] = getelementptr double, double* [[TMP12]], i32 1 3106 // CHECK3-NEXT: [[TMP14:%.*]] = bitcast double* [[TMP13]] to i8* 3107 // CHECK3-NEXT: [[TMP15:%.*]] = bitcast double* [[TMP12]] to i64* 3108 // CHECK3-NEXT: [[TMP16:%.*]] = bitcast double* [[DOTOMP_REDUCTION_ELEMENT]] to i64* 3109 // CHECK3-NEXT: [[TMP17:%.*]] = load i64, i64* [[TMP15]], align 8 3110 // CHECK3-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 3111 // CHECK3-NEXT: [[TMP18:%.*]] = trunc i32 [[NVPTX_WARP_SIZE]] to i16 3112 // CHECK3-NEXT: [[TMP19:%.*]] = call i64 @__kmpc_shuffle_int64(i64 [[TMP17]], i16 [[TMP7]], i16 [[TMP18]]) 3113 // CHECK3-NEXT: store i64 [[TMP19]], i64* [[TMP16]], align 8 3114 // CHECK3-NEXT: [[TMP20:%.*]] = getelementptr i64, i64* [[TMP15]], i32 1 3115 // CHECK3-NEXT: [[TMP21:%.*]] = getelementptr i64, i64* [[TMP16]], i32 1 3116 // CHECK3-NEXT: [[TMP22:%.*]] = bitcast double* [[DOTOMP_REDUCTION_ELEMENT]] to i8* 3117 // CHECK3-NEXT: store i8* [[TMP22]], i8** [[TMP11]], align 4 3118 // CHECK3-NEXT: [[TMP23:%.*]] = icmp eq i16 [[TMP8]], 0 3119 // CHECK3-NEXT: [[TMP24:%.*]] = icmp eq i16 [[TMP8]], 1 3120 // CHECK3-NEXT: [[TMP25:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]] 3121 // CHECK3-NEXT: [[TMP26:%.*]] = and i1 [[TMP24]], [[TMP25]] 3122 // CHECK3-NEXT: [[TMP27:%.*]] = icmp eq i16 [[TMP8]], 2 3123 // CHECK3-NEXT: [[TMP28:%.*]] = and i16 [[TMP6]], 1 3124 // CHECK3-NEXT: [[TMP29:%.*]] = icmp eq i16 [[TMP28]], 0 3125 // CHECK3-NEXT: [[TMP30:%.*]] = and i1 [[TMP27]], [[TMP29]] 3126 // CHECK3-NEXT: [[TMP31:%.*]] = icmp sgt i16 [[TMP7]], 0 3127 // CHECK3-NEXT: [[TMP32:%.*]] = and i1 [[TMP30]], [[TMP31]] 3128 // CHECK3-NEXT: [[TMP33:%.*]] = or i1 [[TMP23]], [[TMP26]] 3129 // CHECK3-NEXT: [[TMP34:%.*]] = or i1 [[TMP33]], [[TMP32]] 3130 // CHECK3-NEXT: br i1 [[TMP34]], label [[THEN:%.*]], label [[ELSE:%.*]] 3131 // CHECK3: then: 3132 // CHECK3-NEXT: [[TMP35:%.*]] = bitcast [1 x i8*]* [[TMP5]] to i8* 3133 // CHECK3-NEXT: [[TMP36:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8* 3134 // CHECK3-NEXT: call void @"_omp$reduction$reduction_func"(i8* [[TMP35]], i8* [[TMP36]]) #[[ATTR3]] 3135 // CHECK3-NEXT: br label [[IFCONT:%.*]] 3136 // CHECK3: else: 3137 // CHECK3-NEXT: br label [[IFCONT]] 3138 // CHECK3: ifcont: 3139 // CHECK3-NEXT: [[TMP37:%.*]] = icmp eq i16 [[TMP8]], 1 3140 // CHECK3-NEXT: [[TMP38:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]] 3141 // CHECK3-NEXT: [[TMP39:%.*]] = and i1 [[TMP37]], [[TMP38]] 3142 // CHECK3-NEXT: br i1 [[TMP39]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 3143 // CHECK3: then4: 3144 // CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 3145 // CHECK3-NEXT: [[TMP41:%.*]] = load i8*, i8** [[TMP40]], align 4 3146 // CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i32 0, i32 0 3147 // CHECK3-NEXT: [[TMP43:%.*]] = load i8*, i8** [[TMP42]], align 4 3148 // CHECK3-NEXT: [[TMP44:%.*]] = bitcast i8* [[TMP41]] to double* 3149 // CHECK3-NEXT: [[TMP45:%.*]] = bitcast i8* [[TMP43]] to double* 3150 // CHECK3-NEXT: [[TMP46:%.*]] = load double, double* [[TMP44]], align 8 3151 // CHECK3-NEXT: store double [[TMP46]], double* [[TMP45]], align 8 3152 // CHECK3-NEXT: br label [[IFCONT6:%.*]] 3153 // CHECK3: else5: 3154 // CHECK3-NEXT: br label [[IFCONT6]] 3155 // CHECK3: ifcont6: 3156 // CHECK3-NEXT: ret void 3157 // 3158 // 3159 // CHECK3-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func 3160 // CHECK3-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] { 3161 // CHECK3-NEXT: entry: 3162 // CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 3163 // CHECK3-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 3164 // CHECK3-NEXT: [[DOTCNT_ADDR:%.*]] = alloca i32, align 4 3165 // CHECK3-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 3166 // CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 3167 // CHECK3-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 3168 // CHECK3-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 3169 // CHECK3-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 3170 // CHECK3-NEXT: [[NVPTX_LANE_ID:%.*]] = and i32 [[NVPTX_TID2]], 31 3171 // CHECK3-NEXT: [[NVPTX_TID3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 3172 // CHECK3-NEXT: [[NVPTX_WARP_ID:%.*]] = ashr i32 [[NVPTX_TID3]], 5 3173 // CHECK3-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 3174 // CHECK3-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [1 x i8*]* 3175 // CHECK3-NEXT: store i32 0, i32* [[DOTCNT_ADDR]], align 4 3176 // CHECK3-NEXT: br label [[PRECOND:%.*]] 3177 // CHECK3: precond: 3178 // CHECK3-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTCNT_ADDR]], align 4 3179 // CHECK3-NEXT: [[TMP6:%.*]] = icmp ult i32 [[TMP5]], 2 3180 // CHECK3-NEXT: br i1 [[TMP6]], label [[BODY:%.*]], label [[EXIT:%.*]] 3181 // CHECK3: body: 3182 // CHECK3-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP2]]) 3183 // CHECK3-NEXT: [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 3184 // CHECK3-NEXT: br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]] 3185 // CHECK3: then: 3186 // CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP4]], i32 0, i32 0 3187 // CHECK3-NEXT: [[TMP8:%.*]] = load i8*, i8** [[TMP7]], align 4 3188 // CHECK3-NEXT: [[TMP9:%.*]] = bitcast i8* [[TMP8]] to i32* 3189 // CHECK3-NEXT: [[TMP10:%.*]] = getelementptr i32, i32* [[TMP9]], i32 [[TMP5]] 3190 // CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 3191 // CHECK3-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP10]], align 4 3192 // CHECK3-NEXT: store volatile i32 [[TMP12]], i32 addrspace(3)* [[TMP11]], align 4 3193 // CHECK3-NEXT: br label [[IFCONT:%.*]] 3194 // CHECK3: else: 3195 // CHECK3-NEXT: br label [[IFCONT]] 3196 // CHECK3: ifcont: 3197 // CHECK3-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 3198 // CHECK3-NEXT: [[TMP13:%.*]] = load i32, i32* [[DOTADDR1]], align 4 3199 // CHECK3-NEXT: [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP13]] 3200 // CHECK3-NEXT: br i1 [[IS_ACTIVE_THREAD]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 3201 // CHECK3: then4: 3202 // CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 3203 // CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP4]], i32 0, i32 0 3204 // CHECK3-NEXT: [[TMP16:%.*]] = load i8*, i8** [[TMP15]], align 4 3205 // CHECK3-NEXT: [[TMP17:%.*]] = bitcast i8* [[TMP16]] to i32* 3206 // CHECK3-NEXT: [[TMP18:%.*]] = getelementptr i32, i32* [[TMP17]], i32 [[TMP5]] 3207 // CHECK3-NEXT: [[TMP19:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP14]], align 4 3208 // CHECK3-NEXT: store i32 [[TMP19]], i32* [[TMP18]], align 4 3209 // CHECK3-NEXT: br label [[IFCONT6:%.*]] 3210 // CHECK3: else5: 3211 // CHECK3-NEXT: br label [[IFCONT6]] 3212 // CHECK3: ifcont6: 3213 // CHECK3-NEXT: [[TMP20:%.*]] = add nsw i32 [[TMP5]], 1 3214 // CHECK3-NEXT: store i32 [[TMP20]], i32* [[DOTCNT_ADDR]], align 4 3215 // CHECK3-NEXT: br label [[PRECOND]] 3216 // CHECK3: exit: 3217 // CHECK3-NEXT: ret void 3218 // 3219 // 3220 // CHECK3-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_copy_func 3221 // CHECK3-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 3222 // CHECK3-NEXT: entry: 3223 // CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 3224 // CHECK3-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 3225 // CHECK3-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 3226 // CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 3227 // CHECK3-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 3228 // CHECK3-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 3229 // CHECK3-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 3230 // CHECK3-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [1 x i8*]* 3231 // CHECK3-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 4 3232 // CHECK3-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.0* 3233 // CHECK3-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 3234 // CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP4]], i32 0, i32 0 3235 // CHECK3-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 4 3236 // CHECK3-NEXT: [[TMP10:%.*]] = bitcast i8* [[TMP9]] to double* 3237 // CHECK3-NEXT: [[E:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_0:%.*]], %struct._globalized_locals_ty.0* [[TMP6]], i32 0, i32 0 3238 // CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [1024 x double], [1024 x double]* [[E]], i32 0, i32 [[TMP7]] 3239 // CHECK3-NEXT: [[TMP12:%.*]] = load double, double* [[TMP10]], align 8 3240 // CHECK3-NEXT: store double [[TMP12]], double* [[TMP11]], align 128 3241 // CHECK3-NEXT: ret void 3242 // 3243 // 3244 // CHECK3-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_reduce_func 3245 // CHECK3-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 3246 // CHECK3-NEXT: entry: 3247 // CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 3248 // CHECK3-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 3249 // CHECK3-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 3250 // CHECK3-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x i8*], align 4 3251 // CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 3252 // CHECK3-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 3253 // CHECK3-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 3254 // CHECK3-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 3255 // CHECK3-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.0* 3256 // CHECK3-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 3257 // CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 3258 // CHECK3-NEXT: [[E:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_0:%.*]], %struct._globalized_locals_ty.0* [[TMP4]], i32 0, i32 0 3259 // CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1024 x double], [1024 x double]* [[E]], i32 0, i32 [[TMP5]] 3260 // CHECK3-NEXT: [[TMP8:%.*]] = bitcast double* [[TMP7]] to i8* 3261 // CHECK3-NEXT: store i8* [[TMP8]], i8** [[TMP6]], align 4 3262 // CHECK3-NEXT: [[TMP9:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 3263 // CHECK3-NEXT: [[TMP10:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 3264 // CHECK3-NEXT: call void @"_omp$reduction$reduction_func"(i8* [[TMP9]], i8* [[TMP10]]) #[[ATTR3]] 3265 // CHECK3-NEXT: ret void 3266 // 3267 // 3268 // CHECK3-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_copy_func 3269 // CHECK3-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 3270 // CHECK3-NEXT: entry: 3271 // CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 3272 // CHECK3-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 3273 // CHECK3-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 3274 // CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 3275 // CHECK3-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 3276 // CHECK3-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 3277 // CHECK3-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 3278 // CHECK3-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [1 x i8*]* 3279 // CHECK3-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 4 3280 // CHECK3-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.0* 3281 // CHECK3-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 3282 // CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP4]], i32 0, i32 0 3283 // CHECK3-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 4 3284 // CHECK3-NEXT: [[TMP10:%.*]] = bitcast i8* [[TMP9]] to double* 3285 // CHECK3-NEXT: [[E:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_0:%.*]], %struct._globalized_locals_ty.0* [[TMP6]], i32 0, i32 0 3286 // CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [1024 x double], [1024 x double]* [[E]], i32 0, i32 [[TMP7]] 3287 // CHECK3-NEXT: [[TMP12:%.*]] = load double, double* [[TMP11]], align 128 3288 // CHECK3-NEXT: store double [[TMP12]], double* [[TMP10]], align 8 3289 // CHECK3-NEXT: ret void 3290 // 3291 // 3292 // CHECK3-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_reduce_func 3293 // CHECK3-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 3294 // CHECK3-NEXT: entry: 3295 // CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 3296 // CHECK3-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 3297 // CHECK3-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 3298 // CHECK3-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x i8*], align 4 3299 // CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 3300 // CHECK3-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 3301 // CHECK3-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 3302 // CHECK3-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 3303 // CHECK3-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.0* 3304 // CHECK3-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 3305 // CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 3306 // CHECK3-NEXT: [[E:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_0:%.*]], %struct._globalized_locals_ty.0* [[TMP4]], i32 0, i32 0 3307 // CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1024 x double], [1024 x double]* [[E]], i32 0, i32 [[TMP5]] 3308 // CHECK3-NEXT: [[TMP8:%.*]] = bitcast double* [[TMP7]] to i8* 3309 // CHECK3-NEXT: store i8* [[TMP8]], i8** [[TMP6]], align 4 3310 // CHECK3-NEXT: [[TMP9:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 3311 // CHECK3-NEXT: [[TMP10:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 3312 // CHECK3-NEXT: call void @"_omp$reduction$reduction_func"(i8* [[TMP10]], i8* [[TMP9]]) #[[ATTR3]] 3313 // CHECK3-NEXT: ret void 3314 // 3315 // 3316 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29_worker 3317 // CHECK3-SAME: () #[[ATTR0]] { 3318 // CHECK3-NEXT: entry: 3319 // CHECK3-NEXT: [[WORK_FN:%.*]] = alloca i8*, align 4 3320 // CHECK3-NEXT: [[EXEC_STATUS:%.*]] = alloca i8, align 1 3321 // CHECK3-NEXT: store i8* null, i8** [[WORK_FN]], align 4 3322 // CHECK3-NEXT: store i8 0, i8* [[EXEC_STATUS]], align 1 3323 // CHECK3-NEXT: br label [[DOTAWAIT_WORK:%.*]] 3324 // CHECK3: .await.work: 3325 // CHECK3-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 3326 // CHECK3-NEXT: [[TMP0:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORK_FN]]) 3327 // CHECK3-NEXT: [[TMP1:%.*]] = zext i1 [[TMP0]] to i8 3328 // CHECK3-NEXT: store i8 [[TMP1]], i8* [[EXEC_STATUS]], align 1 3329 // CHECK3-NEXT: [[TMP2:%.*]] = load i8*, i8** [[WORK_FN]], align 4 3330 // CHECK3-NEXT: [[SHOULD_TERMINATE:%.*]] = icmp eq i8* [[TMP2]], null 3331 // CHECK3-NEXT: br i1 [[SHOULD_TERMINATE]], label [[DOTEXIT:%.*]], label [[DOTSELECT_WORKERS:%.*]] 3332 // CHECK3: .select.workers: 3333 // CHECK3-NEXT: [[TMP3:%.*]] = load i8, i8* [[EXEC_STATUS]], align 1 3334 // CHECK3-NEXT: [[IS_ACTIVE:%.*]] = icmp ne i8 [[TMP3]], 0 3335 // CHECK3-NEXT: br i1 [[IS_ACTIVE]], label [[DOTEXECUTE_PARALLEL:%.*]], label [[DOTBARRIER_PARALLEL:%.*]] 3336 // CHECK3: .execute.parallel: 3337 // CHECK3-NEXT: [[TMP4:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 3338 // CHECK3-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP2]] to void (i16, i32)* 3339 // CHECK3-NEXT: call void [[TMP5]](i16 0, i32 [[TMP4]]) 3340 // CHECK3-NEXT: br label [[DOTTERMINATE_PARALLEL:%.*]] 3341 // CHECK3: .terminate.parallel: 3342 // CHECK3-NEXT: call void @__kmpc_kernel_end_parallel() 3343 // CHECK3-NEXT: br label [[DOTBARRIER_PARALLEL]] 3344 // CHECK3: .barrier.parallel: 3345 // CHECK3-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 3346 // CHECK3-NEXT: br label [[DOTAWAIT_WORK]] 3347 // CHECK3: .exit: 3348 // CHECK3-NEXT: ret void 3349 // 3350 // 3351 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29 3352 // CHECK3-SAME: (i32 [[C:%.*]], i32 [[D:%.*]]) #[[ATTR1]] { 3353 // CHECK3-NEXT: entry: 3354 // CHECK3-NEXT: [[C_ADDR:%.*]] = alloca i32, align 4 3355 // CHECK3-NEXT: [[D_ADDR:%.*]] = alloca i32, align 4 3356 // CHECK3-NEXT: [[DOTZERO_ADDR:%.*]] = alloca i32, align 4 3357 // CHECK3-NEXT: [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4 3358 // CHECK3-NEXT: store i32 0, i32* [[DOTZERO_ADDR]], align 4 3359 // CHECK3-NEXT: store i32 [[C]], i32* [[C_ADDR]], align 4 3360 // CHECK3-NEXT: store i32 [[D]], i32* [[D_ADDR]], align 4 3361 // CHECK3-NEXT: [[CONV:%.*]] = bitcast i32* [[C_ADDR]] to i8* 3362 // CHECK3-NEXT: [[CONV1:%.*]] = bitcast i32* [[D_ADDR]] to float* 3363 // CHECK3-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 3364 // CHECK3-NEXT: [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 3365 // CHECK3-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 3366 // CHECK3-NEXT: [[THREAD_LIMIT:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS]], [[NVPTX_WARP_SIZE]] 3367 // CHECK3-NEXT: [[TMP0:%.*]] = icmp ult i32 [[NVPTX_TID]], [[THREAD_LIMIT]] 3368 // CHECK3-NEXT: br i1 [[TMP0]], label [[DOTWORKER:%.*]], label [[DOTMASTERCHECK:%.*]] 3369 // CHECK3: .worker: 3370 // CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29_worker() #[[ATTR3]] 3371 // CHECK3-NEXT: br label [[DOTEXIT:%.*]] 3372 // CHECK3: .mastercheck: 3373 // CHECK3-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 3374 // CHECK3-NEXT: [[NVPTX_NUM_THREADS3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 3375 // CHECK3-NEXT: [[NVPTX_WARP_SIZE4:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 3376 // CHECK3-NEXT: [[TMP1:%.*]] = sub nuw i32 [[NVPTX_WARP_SIZE4]], 1 3377 // CHECK3-NEXT: [[TMP2:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS3]], 1 3378 // CHECK3-NEXT: [[TMP3:%.*]] = xor i32 [[TMP1]], -1 3379 // CHECK3-NEXT: [[MASTER_TID:%.*]] = and i32 [[TMP2]], [[TMP3]] 3380 // CHECK3-NEXT: [[TMP4:%.*]] = icmp eq i32 [[NVPTX_TID2]], [[MASTER_TID]] 3381 // CHECK3-NEXT: br i1 [[TMP4]], label [[DOTMASTER:%.*]], label [[DOTEXIT]] 3382 // CHECK3: .master: 3383 // CHECK3-NEXT: [[NVPTX_NUM_THREADS5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 3384 // CHECK3-NEXT: [[NVPTX_WARP_SIZE6:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 3385 // CHECK3-NEXT: [[THREAD_LIMIT7:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS5]], [[NVPTX_WARP_SIZE6]] 3386 // CHECK3-NEXT: call void @__kmpc_kernel_init(i32 [[THREAD_LIMIT7]], i16 1) 3387 // CHECK3-NEXT: call void @__kmpc_data_sharing_init_stack() 3388 // CHECK3-NEXT: [[TMP5:%.*]] = load i16, i16* @"_openmp_static_kernel$is_shared1", align 2 3389 // CHECK3-NEXT: [[TMP6:%.*]] = load i32, i32* @"_openmp_static_kernel$size2", align 4 3390 // CHECK3-NEXT: call void @__kmpc_get_team_static_memory(i16 0, i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds (%"union._shared_openmp_static_memory_type_$_", %"union._shared_openmp_static_memory_type_$_" addrspace(3)* @"_openmp_shared_static_glob_rd_$_", i32 0, i32 0, i32 0) to i8*), i32 [[TMP6]], i16 [[TMP5]], i8** addrspacecast (i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr" to i8**)) 3391 // CHECK3-NEXT: [[TMP7:%.*]] = load i8*, i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr", align 4 3392 // CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds i8, i8* [[TMP7]], i32 0 3393 // CHECK3-NEXT: [[TMP9:%.*]] = bitcast i8* [[TMP8]] to %struct._globalized_locals_ty.1* 3394 // CHECK3-NEXT: [[TMP10:%.*]] = load i8, i8* [[CONV]], align 4 3395 // CHECK3-NEXT: [[C8:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_1:%.*]], %struct._globalized_locals_ty.1* [[TMP9]], i32 0, i32 1 3396 // CHECK3-NEXT: store i8 [[TMP10]], i8* [[C8]], align 4 3397 // CHECK3-NEXT: [[TMP11:%.*]] = load float, float* [[CONV1]], align 4 3398 // CHECK3-NEXT: [[D9:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_1]], %struct._globalized_locals_ty.1* [[TMP9]], i32 0, i32 0 3399 // CHECK3-NEXT: store float [[TMP11]], float* [[D9]], align 4 3400 // CHECK3-NEXT: [[TMP12:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 3401 // CHECK3-NEXT: store i32 [[TMP12]], i32* [[DOTTHREADID_TEMP_]], align 4 3402 // CHECK3-NEXT: call void @__omp_outlined__3(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i8* [[C8]], float* [[D9]]) #[[ATTR3]] 3403 // CHECK3-NEXT: [[TMP13:%.*]] = load i16, i16* @"_openmp_static_kernel$is_shared1", align 2 3404 // CHECK3-NEXT: call void @__kmpc_restore_team_static_memory(i16 0, i16 [[TMP13]]) 3405 // CHECK3-NEXT: br label [[DOTTERMINATION_NOTIFIER:%.*]] 3406 // CHECK3: .termination.notifier: 3407 // CHECK3-NEXT: call void @__kmpc_kernel_deinit(i16 1) 3408 // CHECK3-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 3409 // CHECK3-NEXT: br label [[DOTEXIT]] 3410 // CHECK3: .exit: 3411 // CHECK3-NEXT: ret void 3412 // 3413 // 3414 // CHECK3-LABEL: define {{[^@]+}}@__omp_outlined__3 3415 // CHECK3-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i8* nonnull align 1 dereferenceable(1) [[C:%.*]], float* nonnull align 4 dereferenceable(4) [[D:%.*]]) #[[ATTR1]] { 3416 // CHECK3-NEXT: entry: 3417 // CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4 3418 // CHECK3-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4 3419 // CHECK3-NEXT: [[C_ADDR:%.*]] = alloca i8*, align 4 3420 // CHECK3-NEXT: [[D_ADDR:%.*]] = alloca float*, align 4 3421 // CHECK3-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 3422 // CHECK3-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4 3423 // CHECK3-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4 3424 // CHECK3-NEXT: store i8* [[C]], i8** [[C_ADDR]], align 4 3425 // CHECK3-NEXT: store float* [[D]], float** [[D_ADDR]], align 4 3426 // CHECK3-NEXT: [[TMP0:%.*]] = load i8*, i8** [[C_ADDR]], align 4 3427 // CHECK3-NEXT: [[TMP1:%.*]] = load float*, float** [[D_ADDR]], align 4 3428 // CHECK3-NEXT: [[TMP2:%.*]] = load i8*, i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr", align 4 3429 // CHECK3-NEXT: [[TMP3:%.*]] = getelementptr inbounds i8, i8* [[TMP2]], i32 8 3430 // CHECK3-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.2* 3431 // CHECK3-NEXT: [[C1:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_2:%.*]], %struct._globalized_locals_ty.2* [[TMP4]], i32 0, i32 1 3432 // CHECK3-NEXT: [[D2:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_2]], %struct._globalized_locals_ty.2* [[TMP4]], i32 0, i32 0 3433 // CHECK3-NEXT: store i8 0, i8* [[C1]], align 4 3434 // CHECK3-NEXT: store float 1.000000e+00, float* [[D2]], align 4 3435 // CHECK3-NEXT: [[TMP5:%.*]] = load i8, i8* [[C1]], align 4 3436 // CHECK3-NEXT: [[CONV:%.*]] = sext i8 [[TMP5]] to i32 3437 // CHECK3-NEXT: [[XOR:%.*]] = xor i32 [[CONV]], 2 3438 // CHECK3-NEXT: [[CONV3:%.*]] = trunc i32 [[XOR]] to i8 3439 // CHECK3-NEXT: store i8 [[CONV3]], i8* [[C1]], align 4 3440 // CHECK3-NEXT: [[TMP6:%.*]] = load float, float* [[D2]], align 4 3441 // CHECK3-NEXT: [[MUL:%.*]] = fmul float [[TMP6]], 3.300000e+01 3442 // CHECK3-NEXT: store float [[MUL]], float* [[D2]], align 4 3443 // CHECK3-NEXT: [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4 3444 // CHECK3-NEXT: [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4 3445 // CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 3446 // CHECK3-NEXT: store i8* [[C1]], i8** [[TMP9]], align 4 3447 // CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 3448 // CHECK3-NEXT: [[TMP11:%.*]] = bitcast float* [[D2]] to i8* 3449 // CHECK3-NEXT: store i8* [[TMP11]], i8** [[TMP10]], align 4 3450 // CHECK3-NEXT: [[TMP12:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 3451 // CHECK3-NEXT: [[TMP13:%.*]] = load i8*, i8** @"_openmp_teams_reductions_buffer_$_$ptr", align 4 3452 // CHECK3-NEXT: [[TMP14:%.*]] = call i32 @__kmpc_nvptx_teams_reduce_nowait_v2(%struct.ident_t* @[[GLOB1]], i32 [[TMP8]], i8* [[TMP13]], i32 1024, i8* [[TMP12]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func5, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func6, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_copy_func7, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_reduce_func8, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_copy_func9, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_reduce_func10) 3453 // CHECK3-NEXT: [[TMP15:%.*]] = icmp eq i32 [[TMP14]], 1 3454 // CHECK3-NEXT: br i1 [[TMP15]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]] 3455 // CHECK3: .omp.reduction.then: 3456 // CHECK3-NEXT: [[TMP16:%.*]] = load i8, i8* [[TMP0]], align 1 3457 // CHECK3-NEXT: [[CONV4:%.*]] = sext i8 [[TMP16]] to i32 3458 // CHECK3-NEXT: [[TMP17:%.*]] = load i8, i8* [[C1]], align 4 3459 // CHECK3-NEXT: [[CONV5:%.*]] = sext i8 [[TMP17]] to i32 3460 // CHECK3-NEXT: [[XOR6:%.*]] = xor i32 [[CONV4]], [[CONV5]] 3461 // CHECK3-NEXT: [[CONV7:%.*]] = trunc i32 [[XOR6]] to i8 3462 // CHECK3-NEXT: store i8 [[CONV7]], i8* [[TMP0]], align 1 3463 // CHECK3-NEXT: [[TMP18:%.*]] = load float, float* [[TMP1]], align 4 3464 // CHECK3-NEXT: [[TMP19:%.*]] = load float, float* [[D2]], align 4 3465 // CHECK3-NEXT: [[MUL8:%.*]] = fmul float [[TMP18]], [[TMP19]] 3466 // CHECK3-NEXT: store float [[MUL8]], float* [[TMP1]], align 4 3467 // CHECK3-NEXT: call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP8]]) 3468 // CHECK3-NEXT: br label [[DOTOMP_REDUCTION_DONE]] 3469 // CHECK3: .omp.reduction.done: 3470 // CHECK3-NEXT: ret void 3471 // 3472 // 3473 // CHECK3-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func5 3474 // CHECK3-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR0]] { 3475 // CHECK3-NEXT: entry: 3476 // CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 3477 // CHECK3-NEXT: [[DOTADDR1:%.*]] = alloca i16, align 2 3478 // CHECK3-NEXT: [[DOTADDR2:%.*]] = alloca i16, align 2 3479 // CHECK3-NEXT: [[DOTADDR3:%.*]] = alloca i16, align 2 3480 // CHECK3-NEXT: [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [2 x i8*], align 4 3481 // CHECK3-NEXT: [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca i8, align 1 3482 // CHECK3-NEXT: [[DOTOMP_REDUCTION_ELEMENT4:%.*]] = alloca float, align 4 3483 // CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 3484 // CHECK3-NEXT: store i16 [[TMP1]], i16* [[DOTADDR1]], align 2 3485 // CHECK3-NEXT: store i16 [[TMP2]], i16* [[DOTADDR2]], align 2 3486 // CHECK3-NEXT: store i16 [[TMP3]], i16* [[DOTADDR3]], align 2 3487 // CHECK3-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 4 3488 // CHECK3-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [2 x i8*]* 3489 // CHECK3-NEXT: [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2 3490 // CHECK3-NEXT: [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2 3491 // CHECK3-NEXT: [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2 3492 // CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 0 3493 // CHECK3-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 4 3494 // CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 3495 // CHECK3-NEXT: [[TMP12:%.*]] = getelementptr i8, i8* [[TMP10]], i32 1 3496 // CHECK3-NEXT: [[TMP13:%.*]] = load i8, i8* [[TMP10]], align 1 3497 // CHECK3-NEXT: [[TMP14:%.*]] = sext i8 [[TMP13]] to i32 3498 // CHECK3-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 3499 // CHECK3-NEXT: [[TMP15:%.*]] = trunc i32 [[NVPTX_WARP_SIZE]] to i16 3500 // CHECK3-NEXT: [[TMP16:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP14]], i16 [[TMP7]], i16 [[TMP15]]) 3501 // CHECK3-NEXT: [[TMP17:%.*]] = trunc i32 [[TMP16]] to i8 3502 // CHECK3-NEXT: store i8 [[TMP17]], i8* [[DOTOMP_REDUCTION_ELEMENT]], align 1 3503 // CHECK3-NEXT: [[TMP18:%.*]] = getelementptr i8, i8* [[TMP10]], i32 1 3504 // CHECK3-NEXT: [[TMP19:%.*]] = getelementptr i8, i8* [[DOTOMP_REDUCTION_ELEMENT]], i32 1 3505 // CHECK3-NEXT: store i8* [[DOTOMP_REDUCTION_ELEMENT]], i8** [[TMP11]], align 4 3506 // CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 1 3507 // CHECK3-NEXT: [[TMP21:%.*]] = load i8*, i8** [[TMP20]], align 4 3508 // CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1 3509 // CHECK3-NEXT: [[TMP23:%.*]] = bitcast i8* [[TMP21]] to float* 3510 // CHECK3-NEXT: [[TMP24:%.*]] = getelementptr float, float* [[TMP23]], i32 1 3511 // CHECK3-NEXT: [[TMP25:%.*]] = bitcast float* [[TMP24]] to i8* 3512 // CHECK3-NEXT: [[TMP26:%.*]] = bitcast float* [[TMP23]] to i32* 3513 // CHECK3-NEXT: [[TMP27:%.*]] = bitcast float* [[DOTOMP_REDUCTION_ELEMENT4]] to i32* 3514 // CHECK3-NEXT: [[TMP28:%.*]] = load i32, i32* [[TMP26]], align 4 3515 // CHECK3-NEXT: [[NVPTX_WARP_SIZE5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 3516 // CHECK3-NEXT: [[TMP29:%.*]] = trunc i32 [[NVPTX_WARP_SIZE5]] to i16 3517 // CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP28]], i16 [[TMP7]], i16 [[TMP29]]) 3518 // CHECK3-NEXT: store i32 [[TMP30]], i32* [[TMP27]], align 4 3519 // CHECK3-NEXT: [[TMP31:%.*]] = getelementptr i32, i32* [[TMP26]], i32 1 3520 // CHECK3-NEXT: [[TMP32:%.*]] = getelementptr i32, i32* [[TMP27]], i32 1 3521 // CHECK3-NEXT: [[TMP33:%.*]] = bitcast float* [[DOTOMP_REDUCTION_ELEMENT4]] to i8* 3522 // CHECK3-NEXT: store i8* [[TMP33]], i8** [[TMP22]], align 4 3523 // CHECK3-NEXT: [[TMP34:%.*]] = icmp eq i16 [[TMP8]], 0 3524 // CHECK3-NEXT: [[TMP35:%.*]] = icmp eq i16 [[TMP8]], 1 3525 // CHECK3-NEXT: [[TMP36:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]] 3526 // CHECK3-NEXT: [[TMP37:%.*]] = and i1 [[TMP35]], [[TMP36]] 3527 // CHECK3-NEXT: [[TMP38:%.*]] = icmp eq i16 [[TMP8]], 2 3528 // CHECK3-NEXT: [[TMP39:%.*]] = and i16 [[TMP6]], 1 3529 // CHECK3-NEXT: [[TMP40:%.*]] = icmp eq i16 [[TMP39]], 0 3530 // CHECK3-NEXT: [[TMP41:%.*]] = and i1 [[TMP38]], [[TMP40]] 3531 // CHECK3-NEXT: [[TMP42:%.*]] = icmp sgt i16 [[TMP7]], 0 3532 // CHECK3-NEXT: [[TMP43:%.*]] = and i1 [[TMP41]], [[TMP42]] 3533 // CHECK3-NEXT: [[TMP44:%.*]] = or i1 [[TMP34]], [[TMP37]] 3534 // CHECK3-NEXT: [[TMP45:%.*]] = or i1 [[TMP44]], [[TMP43]] 3535 // CHECK3-NEXT: br i1 [[TMP45]], label [[THEN:%.*]], label [[ELSE:%.*]] 3536 // CHECK3: then: 3537 // CHECK3-NEXT: [[TMP46:%.*]] = bitcast [2 x i8*]* [[TMP5]] to i8* 3538 // CHECK3-NEXT: [[TMP47:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8* 3539 // CHECK3-NEXT: call void @"_omp$reduction$reduction_func4"(i8* [[TMP46]], i8* [[TMP47]]) #[[ATTR3]] 3540 // CHECK3-NEXT: br label [[IFCONT:%.*]] 3541 // CHECK3: else: 3542 // CHECK3-NEXT: br label [[IFCONT]] 3543 // CHECK3: ifcont: 3544 // CHECK3-NEXT: [[TMP48:%.*]] = icmp eq i16 [[TMP8]], 1 3545 // CHECK3-NEXT: [[TMP49:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]] 3546 // CHECK3-NEXT: [[TMP50:%.*]] = and i1 [[TMP48]], [[TMP49]] 3547 // CHECK3-NEXT: br i1 [[TMP50]], label [[THEN6:%.*]], label [[ELSE7:%.*]] 3548 // CHECK3: then6: 3549 // CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 3550 // CHECK3-NEXT: [[TMP52:%.*]] = load i8*, i8** [[TMP51]], align 4 3551 // CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 0 3552 // CHECK3-NEXT: [[TMP54:%.*]] = load i8*, i8** [[TMP53]], align 4 3553 // CHECK3-NEXT: [[TMP55:%.*]] = load i8, i8* [[TMP52]], align 1 3554 // CHECK3-NEXT: store i8 [[TMP55]], i8* [[TMP54]], align 1 3555 // CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1 3556 // CHECK3-NEXT: [[TMP57:%.*]] = load i8*, i8** [[TMP56]], align 4 3557 // CHECK3-NEXT: [[TMP58:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 1 3558 // CHECK3-NEXT: [[TMP59:%.*]] = load i8*, i8** [[TMP58]], align 4 3559 // CHECK3-NEXT: [[TMP60:%.*]] = bitcast i8* [[TMP57]] to float* 3560 // CHECK3-NEXT: [[TMP61:%.*]] = bitcast i8* [[TMP59]] to float* 3561 // CHECK3-NEXT: [[TMP62:%.*]] = load float, float* [[TMP60]], align 4 3562 // CHECK3-NEXT: store float [[TMP62]], float* [[TMP61]], align 4 3563 // CHECK3-NEXT: br label [[IFCONT8:%.*]] 3564 // CHECK3: else7: 3565 // CHECK3-NEXT: br label [[IFCONT8]] 3566 // CHECK3: ifcont8: 3567 // CHECK3-NEXT: ret void 3568 // 3569 // 3570 // CHECK3-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func6 3571 // CHECK3-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] { 3572 // CHECK3-NEXT: entry: 3573 // CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 3574 // CHECK3-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 3575 // CHECK3-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 3576 // CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 3577 // CHECK3-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 3578 // CHECK3-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 3579 // CHECK3-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 3580 // CHECK3-NEXT: [[NVPTX_LANE_ID:%.*]] = and i32 [[NVPTX_TID2]], 31 3581 // CHECK3-NEXT: [[NVPTX_TID3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 3582 // CHECK3-NEXT: [[NVPTX_WARP_ID:%.*]] = ashr i32 [[NVPTX_TID3]], 5 3583 // CHECK3-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 3584 // CHECK3-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 3585 // CHECK3-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 3586 // CHECK3-NEXT: [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 3587 // CHECK3-NEXT: br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]] 3588 // CHECK3: then: 3589 // CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 3590 // CHECK3-NEXT: [[TMP6:%.*]] = load i8*, i8** [[TMP5]], align 4 3591 // CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 3592 // CHECK3-NEXT: [[TMP8:%.*]] = bitcast i32 addrspace(3)* [[TMP7]] to i8 addrspace(3)* 3593 // CHECK3-NEXT: [[TMP9:%.*]] = load i8, i8* [[TMP6]], align 1 3594 // CHECK3-NEXT: store volatile i8 [[TMP9]], i8 addrspace(3)* [[TMP8]], align 1 3595 // CHECK3-NEXT: br label [[IFCONT:%.*]] 3596 // CHECK3: else: 3597 // CHECK3-NEXT: br label [[IFCONT]] 3598 // CHECK3: ifcont: 3599 // CHECK3-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 3600 // CHECK3-NEXT: [[TMP10:%.*]] = load i32, i32* [[DOTADDR1]], align 4 3601 // CHECK3-NEXT: [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP10]] 3602 // CHECK3-NEXT: br i1 [[IS_ACTIVE_THREAD]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 3603 // CHECK3: then4: 3604 // CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 3605 // CHECK3-NEXT: [[TMP12:%.*]] = bitcast i32 addrspace(3)* [[TMP11]] to i8 addrspace(3)* 3606 // CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 3607 // CHECK3-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 4 3608 // CHECK3-NEXT: [[TMP15:%.*]] = load volatile i8, i8 addrspace(3)* [[TMP12]], align 1 3609 // CHECK3-NEXT: store i8 [[TMP15]], i8* [[TMP14]], align 1 3610 // CHECK3-NEXT: br label [[IFCONT6:%.*]] 3611 // CHECK3: else5: 3612 // CHECK3-NEXT: br label [[IFCONT6]] 3613 // CHECK3: ifcont6: 3614 // CHECK3-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 3615 // CHECK3-NEXT: [[WARP_MASTER7:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 3616 // CHECK3-NEXT: br i1 [[WARP_MASTER7]], label [[THEN8:%.*]], label [[ELSE9:%.*]] 3617 // CHECK3: then8: 3618 // CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 3619 // CHECK3-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 4 3620 // CHECK3-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i32* 3621 // CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 3622 // CHECK3-NEXT: [[TMP20:%.*]] = load i32, i32* [[TMP18]], align 4 3623 // CHECK3-NEXT: store volatile i32 [[TMP20]], i32 addrspace(3)* [[TMP19]], align 4 3624 // CHECK3-NEXT: br label [[IFCONT10:%.*]] 3625 // CHECK3: else9: 3626 // CHECK3-NEXT: br label [[IFCONT10]] 3627 // CHECK3: ifcont10: 3628 // CHECK3-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 3629 // CHECK3-NEXT: [[TMP21:%.*]] = load i32, i32* [[DOTADDR1]], align 4 3630 // CHECK3-NEXT: [[IS_ACTIVE_THREAD11:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP21]] 3631 // CHECK3-NEXT: br i1 [[IS_ACTIVE_THREAD11]], label [[THEN12:%.*]], label [[ELSE13:%.*]] 3632 // CHECK3: then12: 3633 // CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 3634 // CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 3635 // CHECK3-NEXT: [[TMP24:%.*]] = load i8*, i8** [[TMP23]], align 4 3636 // CHECK3-NEXT: [[TMP25:%.*]] = bitcast i8* [[TMP24]] to i32* 3637 // CHECK3-NEXT: [[TMP26:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP22]], align 4 3638 // CHECK3-NEXT: store i32 [[TMP26]], i32* [[TMP25]], align 4 3639 // CHECK3-NEXT: br label [[IFCONT14:%.*]] 3640 // CHECK3: else13: 3641 // CHECK3-NEXT: br label [[IFCONT14]] 3642 // CHECK3: ifcont14: 3643 // CHECK3-NEXT: ret void 3644 // 3645 // 3646 // CHECK3-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_copy_func7 3647 // CHECK3-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 3648 // CHECK3-NEXT: entry: 3649 // CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 3650 // CHECK3-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 3651 // CHECK3-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 3652 // CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 3653 // CHECK3-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 3654 // CHECK3-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 3655 // CHECK3-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 3656 // CHECK3-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 3657 // CHECK3-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 4 3658 // CHECK3-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.3* 3659 // CHECK3-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 3660 // CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 3661 // CHECK3-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 4 3662 // CHECK3-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3:%.*]], %struct._globalized_locals_ty.3* [[TMP6]], i32 0, i32 0 3663 // CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[C]], i32 0, i32 [[TMP7]] 3664 // CHECK3-NEXT: [[TMP11:%.*]] = load i8, i8* [[TMP9]], align 1 3665 // CHECK3-NEXT: store i8 [[TMP11]], i8* [[TMP10]], align 128 3666 // CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 3667 // CHECK3-NEXT: [[TMP13:%.*]] = load i8*, i8** [[TMP12]], align 4 3668 // CHECK3-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to float* 3669 // CHECK3-NEXT: [[D:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3]], %struct._globalized_locals_ty.3* [[TMP6]], i32 0, i32 1 3670 // CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1024 x float], [1024 x float]* [[D]], i32 0, i32 [[TMP7]] 3671 // CHECK3-NEXT: [[TMP16:%.*]] = load float, float* [[TMP14]], align 4 3672 // CHECK3-NEXT: store float [[TMP16]], float* [[TMP15]], align 128 3673 // CHECK3-NEXT: ret void 3674 // 3675 // 3676 // CHECK3-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_reduce_func8 3677 // CHECK3-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 3678 // CHECK3-NEXT: entry: 3679 // CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 3680 // CHECK3-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 3681 // CHECK3-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 3682 // CHECK3-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 3683 // CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 3684 // CHECK3-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 3685 // CHECK3-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 3686 // CHECK3-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 3687 // CHECK3-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.3* 3688 // CHECK3-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 3689 // CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 3690 // CHECK3-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3:%.*]], %struct._globalized_locals_ty.3* [[TMP4]], i32 0, i32 0 3691 // CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[C]], i32 0, i32 [[TMP5]] 3692 // CHECK3-NEXT: store i8* [[TMP7]], i8** [[TMP6]], align 4 3693 // CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 3694 // CHECK3-NEXT: [[D:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3]], %struct._globalized_locals_ty.3* [[TMP4]], i32 0, i32 1 3695 // CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1024 x float], [1024 x float]* [[D]], i32 0, i32 [[TMP5]] 3696 // CHECK3-NEXT: [[TMP10:%.*]] = bitcast float* [[TMP9]] to i8* 3697 // CHECK3-NEXT: store i8* [[TMP10]], i8** [[TMP8]], align 4 3698 // CHECK3-NEXT: [[TMP11:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 3699 // CHECK3-NEXT: [[TMP12:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 3700 // CHECK3-NEXT: call void @"_omp$reduction$reduction_func4"(i8* [[TMP11]], i8* [[TMP12]]) #[[ATTR3]] 3701 // CHECK3-NEXT: ret void 3702 // 3703 // 3704 // CHECK3-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_copy_func9 3705 // CHECK3-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 3706 // CHECK3-NEXT: entry: 3707 // CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 3708 // CHECK3-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 3709 // CHECK3-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 3710 // CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 3711 // CHECK3-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 3712 // CHECK3-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 3713 // CHECK3-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 3714 // CHECK3-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 3715 // CHECK3-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 4 3716 // CHECK3-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.3* 3717 // CHECK3-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 3718 // CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 3719 // CHECK3-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 4 3720 // CHECK3-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3:%.*]], %struct._globalized_locals_ty.3* [[TMP6]], i32 0, i32 0 3721 // CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[C]], i32 0, i32 [[TMP7]] 3722 // CHECK3-NEXT: [[TMP11:%.*]] = load i8, i8* [[TMP10]], align 128 3723 // CHECK3-NEXT: store i8 [[TMP11]], i8* [[TMP9]], align 1 3724 // CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 3725 // CHECK3-NEXT: [[TMP13:%.*]] = load i8*, i8** [[TMP12]], align 4 3726 // CHECK3-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to float* 3727 // CHECK3-NEXT: [[D:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3]], %struct._globalized_locals_ty.3* [[TMP6]], i32 0, i32 1 3728 // CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1024 x float], [1024 x float]* [[D]], i32 0, i32 [[TMP7]] 3729 // CHECK3-NEXT: [[TMP16:%.*]] = load float, float* [[TMP15]], align 128 3730 // CHECK3-NEXT: store float [[TMP16]], float* [[TMP14]], align 4 3731 // CHECK3-NEXT: ret void 3732 // 3733 // 3734 // CHECK3-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_reduce_func10 3735 // CHECK3-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 3736 // CHECK3-NEXT: entry: 3737 // CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 3738 // CHECK3-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 3739 // CHECK3-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 3740 // CHECK3-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 3741 // CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 3742 // CHECK3-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 3743 // CHECK3-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 3744 // CHECK3-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 3745 // CHECK3-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.3* 3746 // CHECK3-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 3747 // CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 3748 // CHECK3-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3:%.*]], %struct._globalized_locals_ty.3* [[TMP4]], i32 0, i32 0 3749 // CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[C]], i32 0, i32 [[TMP5]] 3750 // CHECK3-NEXT: store i8* [[TMP7]], i8** [[TMP6]], align 4 3751 // CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 3752 // CHECK3-NEXT: [[D:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3]], %struct._globalized_locals_ty.3* [[TMP4]], i32 0, i32 1 3753 // CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1024 x float], [1024 x float]* [[D]], i32 0, i32 [[TMP5]] 3754 // CHECK3-NEXT: [[TMP10:%.*]] = bitcast float* [[TMP9]] to i8* 3755 // CHECK3-NEXT: store i8* [[TMP10]], i8** [[TMP8]], align 4 3756 // CHECK3-NEXT: [[TMP11:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 3757 // CHECK3-NEXT: [[TMP12:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 3758 // CHECK3-NEXT: call void @"_omp$reduction$reduction_func4"(i8* [[TMP12]], i8* [[TMP11]]) #[[ATTR3]] 3759 // CHECK3-NEXT: ret void 3760 // 3761 // 3762 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l36 3763 // CHECK3-SAME: (i32 [[A:%.*]], i32 [[B:%.*]]) #[[ATTR1]] { 3764 // CHECK3-NEXT: entry: 3765 // CHECK3-NEXT: [[A_ADDR:%.*]] = alloca i32, align 4 3766 // CHECK3-NEXT: [[B_ADDR:%.*]] = alloca i32, align 4 3767 // CHECK3-NEXT: [[DOTZERO_ADDR:%.*]] = alloca i32, align 4 3768 // CHECK3-NEXT: [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4 3769 // CHECK3-NEXT: store i32 0, i32* [[DOTZERO_ADDR]], align 4 3770 // CHECK3-NEXT: store i32 [[A]], i32* [[A_ADDR]], align 4 3771 // CHECK3-NEXT: store i32 [[B]], i32* [[B_ADDR]], align 4 3772 // CHECK3-NEXT: [[CONV:%.*]] = bitcast i32* [[B_ADDR]] to i16* 3773 // CHECK3-NEXT: [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 3774 // CHECK3-NEXT: call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 1) 3775 // CHECK3-NEXT: call void @__kmpc_data_sharing_init_stack_spmd() 3776 // CHECK3-NEXT: br label [[DOTEXECUTE:%.*]] 3777 // CHECK3: .execute: 3778 // CHECK3-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3:[0-9]+]]) 3779 // CHECK3-NEXT: store i32 [[TMP0]], i32* [[DOTTHREADID_TEMP_]], align 4 3780 // CHECK3-NEXT: call void @__omp_outlined__11(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32* [[A_ADDR]], i16* [[CONV]]) #[[ATTR3]] 3781 // CHECK3-NEXT: br label [[DOTOMP_DEINIT:%.*]] 3782 // CHECK3: .omp.deinit: 3783 // CHECK3-NEXT: call void @__kmpc_spmd_kernel_deinit_v2(i16 1) 3784 // CHECK3-NEXT: br label [[DOTEXIT:%.*]] 3785 // CHECK3: .exit: 3786 // CHECK3-NEXT: ret void 3787 // 3788 // 3789 // CHECK3-LABEL: define {{[^@]+}}@__omp_outlined__11 3790 // CHECK3-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32* nonnull align 4 dereferenceable(4) [[A:%.*]], i16* nonnull align 2 dereferenceable(2) [[B:%.*]]) #[[ATTR1]] { 3791 // CHECK3-NEXT: entry: 3792 // CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4 3793 // CHECK3-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4 3794 // CHECK3-NEXT: [[A_ADDR:%.*]] = alloca i32*, align 4 3795 // CHECK3-NEXT: [[B_ADDR:%.*]] = alloca i16*, align 4 3796 // CHECK3-NEXT: [[A1:%.*]] = alloca i32, align 4 3797 // CHECK3-NEXT: [[B2:%.*]] = alloca i16, align 2 3798 // CHECK3-NEXT: [[CAPTURED_VARS_ADDRS:%.*]] = alloca [2 x i8*], align 4 3799 // CHECK3-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 3800 // CHECK3-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4 3801 // CHECK3-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4 3802 // CHECK3-NEXT: store i32* [[A]], i32** [[A_ADDR]], align 4 3803 // CHECK3-NEXT: store i16* [[B]], i16** [[B_ADDR]], align 4 3804 // CHECK3-NEXT: [[TMP0:%.*]] = load i32*, i32** [[A_ADDR]], align 4 3805 // CHECK3-NEXT: [[TMP1:%.*]] = load i16*, i16** [[B_ADDR]], align 4 3806 // CHECK3-NEXT: store i32 0, i32* [[A1]], align 4 3807 // CHECK3-NEXT: store i16 -32768, i16* [[B2]], align 2 3808 // CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0 3809 // CHECK3-NEXT: [[TMP3:%.*]] = bitcast i32* [[A1]] to i8* 3810 // CHECK3-NEXT: store i8* [[TMP3]], i8** [[TMP2]], align 4 3811 // CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1 3812 // CHECK3-NEXT: [[TMP5:%.*]] = bitcast i16* [[B2]] to i8* 3813 // CHECK3-NEXT: store i8* [[TMP5]], i8** [[TMP4]], align 4 3814 // CHECK3-NEXT: [[TMP6:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4 3815 // CHECK3-NEXT: [[TMP7:%.*]] = load i32, i32* [[TMP6]], align 4 3816 // CHECK3-NEXT: [[TMP8:%.*]] = bitcast [2 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8** 3817 // CHECK3-NEXT: call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP7]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32*, i16*)* @__omp_outlined__12 to i8*), i8* null, i8** [[TMP8]], i32 2) 3818 // CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 3819 // CHECK3-NEXT: [[TMP10:%.*]] = bitcast i32* [[A1]] to i8* 3820 // CHECK3-NEXT: store i8* [[TMP10]], i8** [[TMP9]], align 4 3821 // CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 3822 // CHECK3-NEXT: [[TMP12:%.*]] = bitcast i16* [[B2]] to i8* 3823 // CHECK3-NEXT: store i8* [[TMP12]], i8** [[TMP11]], align 4 3824 // CHECK3-NEXT: [[TMP13:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 3825 // CHECK3-NEXT: [[TMP14:%.*]] = load i8*, i8** @"_openmp_teams_reductions_buffer_$_$ptr", align 4 3826 // CHECK3-NEXT: [[TMP15:%.*]] = call i32 @__kmpc_nvptx_teams_reduce_nowait_v2(%struct.ident_t* @[[GLOB3]], i32 [[TMP7]], i8* [[TMP14]], i32 1024, i8* [[TMP13]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func17, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func18, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_copy_func19, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_reduce_func20, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_copy_func21, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_reduce_func22) 3827 // CHECK3-NEXT: [[TMP16:%.*]] = icmp eq i32 [[TMP15]], 1 3828 // CHECK3-NEXT: br i1 [[TMP16]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]] 3829 // CHECK3: .omp.reduction.then: 3830 // CHECK3-NEXT: [[TMP17:%.*]] = load i32, i32* [[TMP0]], align 4 3831 // CHECK3-NEXT: [[TMP18:%.*]] = load i32, i32* [[A1]], align 4 3832 // CHECK3-NEXT: [[OR:%.*]] = or i32 [[TMP17]], [[TMP18]] 3833 // CHECK3-NEXT: store i32 [[OR]], i32* [[TMP0]], align 4 3834 // CHECK3-NEXT: [[TMP19:%.*]] = load i16, i16* [[TMP1]], align 2 3835 // CHECK3-NEXT: [[CONV:%.*]] = sext i16 [[TMP19]] to i32 3836 // CHECK3-NEXT: [[TMP20:%.*]] = load i16, i16* [[B2]], align 2 3837 // CHECK3-NEXT: [[CONV3:%.*]] = sext i16 [[TMP20]] to i32 3838 // CHECK3-NEXT: [[CMP:%.*]] = icmp sgt i32 [[CONV]], [[CONV3]] 3839 // CHECK3-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] 3840 // CHECK3: cond.true: 3841 // CHECK3-NEXT: [[TMP21:%.*]] = load i16, i16* [[TMP1]], align 2 3842 // CHECK3-NEXT: br label [[COND_END:%.*]] 3843 // CHECK3: cond.false: 3844 // CHECK3-NEXT: [[TMP22:%.*]] = load i16, i16* [[B2]], align 2 3845 // CHECK3-NEXT: br label [[COND_END]] 3846 // CHECK3: cond.end: 3847 // CHECK3-NEXT: [[COND:%.*]] = phi i16 [ [[TMP21]], [[COND_TRUE]] ], [ [[TMP22]], [[COND_FALSE]] ] 3848 // CHECK3-NEXT: store i16 [[COND]], i16* [[TMP1]], align 2 3849 // CHECK3-NEXT: call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP7]]) 3850 // CHECK3-NEXT: br label [[DOTOMP_REDUCTION_DONE]] 3851 // CHECK3: .omp.reduction.done: 3852 // CHECK3-NEXT: ret void 3853 // 3854 // 3855 // CHECK3-LABEL: define {{[^@]+}}@__omp_outlined__12 3856 // CHECK3-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32* nonnull align 4 dereferenceable(4) [[A:%.*]], i16* nonnull align 2 dereferenceable(2) [[B:%.*]]) #[[ATTR1]] { 3857 // CHECK3-NEXT: entry: 3858 // CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4 3859 // CHECK3-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4 3860 // CHECK3-NEXT: [[A_ADDR:%.*]] = alloca i32*, align 4 3861 // CHECK3-NEXT: [[B_ADDR:%.*]] = alloca i16*, align 4 3862 // CHECK3-NEXT: [[A1:%.*]] = alloca i32, align 4 3863 // CHECK3-NEXT: [[B2:%.*]] = alloca i16, align 2 3864 // CHECK3-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 3865 // CHECK3-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4 3866 // CHECK3-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4 3867 // CHECK3-NEXT: store i32* [[A]], i32** [[A_ADDR]], align 4 3868 // CHECK3-NEXT: store i16* [[B]], i16** [[B_ADDR]], align 4 3869 // CHECK3-NEXT: [[TMP0:%.*]] = load i32*, i32** [[A_ADDR]], align 4 3870 // CHECK3-NEXT: [[TMP1:%.*]] = load i16*, i16** [[B_ADDR]], align 4 3871 // CHECK3-NEXT: store i32 0, i32* [[A1]], align 4 3872 // CHECK3-NEXT: store i16 -32768, i16* [[B2]], align 2 3873 // CHECK3-NEXT: [[TMP2:%.*]] = load i32, i32* [[A1]], align 4 3874 // CHECK3-NEXT: [[OR:%.*]] = or i32 [[TMP2]], 1 3875 // CHECK3-NEXT: store i32 [[OR]], i32* [[A1]], align 4 3876 // CHECK3-NEXT: [[TMP3:%.*]] = load i16, i16* [[B2]], align 2 3877 // CHECK3-NEXT: [[CONV:%.*]] = sext i16 [[TMP3]] to i32 3878 // CHECK3-NEXT: [[CMP:%.*]] = icmp sgt i32 99, [[CONV]] 3879 // CHECK3-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] 3880 // CHECK3: cond.true: 3881 // CHECK3-NEXT: br label [[COND_END:%.*]] 3882 // CHECK3: cond.false: 3883 // CHECK3-NEXT: [[TMP4:%.*]] = load i16, i16* [[B2]], align 2 3884 // CHECK3-NEXT: [[CONV3:%.*]] = sext i16 [[TMP4]] to i32 3885 // CHECK3-NEXT: br label [[COND_END]] 3886 // CHECK3: cond.end: 3887 // CHECK3-NEXT: [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[CONV3]], [[COND_FALSE]] ] 3888 // CHECK3-NEXT: [[CONV4:%.*]] = trunc i32 [[COND]] to i16 3889 // CHECK3-NEXT: store i16 [[CONV4]], i16* [[B2]], align 2 3890 // CHECK3-NEXT: [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4 3891 // CHECK3-NEXT: [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4 3892 // CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 3893 // CHECK3-NEXT: [[TMP8:%.*]] = bitcast i32* [[A1]] to i8* 3894 // CHECK3-NEXT: store i8* [[TMP8]], i8** [[TMP7]], align 4 3895 // CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 3896 // CHECK3-NEXT: [[TMP10:%.*]] = bitcast i16* [[B2]] to i8* 3897 // CHECK3-NEXT: store i8* [[TMP10]], i8** [[TMP9]], align 4 3898 // CHECK3-NEXT: [[TMP11:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 3899 // CHECK3-NEXT: [[TMP12:%.*]] = call i32 @__kmpc_nvptx_parallel_reduce_nowait_v2(%struct.ident_t* @[[GLOB3]], i32 [[TMP6]], i32 2, i32 8, i8* [[TMP11]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func14, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func15) 3900 // CHECK3-NEXT: [[TMP13:%.*]] = icmp eq i32 [[TMP12]], 1 3901 // CHECK3-NEXT: br i1 [[TMP13]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]] 3902 // CHECK3: .omp.reduction.then: 3903 // CHECK3-NEXT: [[TMP14:%.*]] = load i32, i32* [[TMP0]], align 4 3904 // CHECK3-NEXT: [[TMP15:%.*]] = load i32, i32* [[A1]], align 4 3905 // CHECK3-NEXT: [[OR5:%.*]] = or i32 [[TMP14]], [[TMP15]] 3906 // CHECK3-NEXT: store i32 [[OR5]], i32* [[TMP0]], align 4 3907 // CHECK3-NEXT: [[TMP16:%.*]] = load i16, i16* [[TMP1]], align 2 3908 // CHECK3-NEXT: [[CONV6:%.*]] = sext i16 [[TMP16]] to i32 3909 // CHECK3-NEXT: [[TMP17:%.*]] = load i16, i16* [[B2]], align 2 3910 // CHECK3-NEXT: [[CONV7:%.*]] = sext i16 [[TMP17]] to i32 3911 // CHECK3-NEXT: [[CMP8:%.*]] = icmp sgt i32 [[CONV6]], [[CONV7]] 3912 // CHECK3-NEXT: br i1 [[CMP8]], label [[COND_TRUE9:%.*]], label [[COND_FALSE10:%.*]] 3913 // CHECK3: cond.true9: 3914 // CHECK3-NEXT: [[TMP18:%.*]] = load i16, i16* [[TMP1]], align 2 3915 // CHECK3-NEXT: br label [[COND_END11:%.*]] 3916 // CHECK3: cond.false10: 3917 // CHECK3-NEXT: [[TMP19:%.*]] = load i16, i16* [[B2]], align 2 3918 // CHECK3-NEXT: br label [[COND_END11]] 3919 // CHECK3: cond.end11: 3920 // CHECK3-NEXT: [[COND12:%.*]] = phi i16 [ [[TMP18]], [[COND_TRUE9]] ], [ [[TMP19]], [[COND_FALSE10]] ] 3921 // CHECK3-NEXT: store i16 [[COND12]], i16* [[TMP1]], align 2 3922 // CHECK3-NEXT: call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP6]]) 3923 // CHECK3-NEXT: br label [[DOTOMP_REDUCTION_DONE]] 3924 // CHECK3: .omp.reduction.done: 3925 // CHECK3-NEXT: ret void 3926 // 3927 // 3928 // CHECK3-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func14 3929 // CHECK3-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR0]] { 3930 // CHECK3-NEXT: entry: 3931 // CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 3932 // CHECK3-NEXT: [[DOTADDR1:%.*]] = alloca i16, align 2 3933 // CHECK3-NEXT: [[DOTADDR2:%.*]] = alloca i16, align 2 3934 // CHECK3-NEXT: [[DOTADDR3:%.*]] = alloca i16, align 2 3935 // CHECK3-NEXT: [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [2 x i8*], align 4 3936 // CHECK3-NEXT: [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca i32, align 4 3937 // CHECK3-NEXT: [[DOTOMP_REDUCTION_ELEMENT4:%.*]] = alloca i16, align 2 3938 // CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 3939 // CHECK3-NEXT: store i16 [[TMP1]], i16* [[DOTADDR1]], align 2 3940 // CHECK3-NEXT: store i16 [[TMP2]], i16* [[DOTADDR2]], align 2 3941 // CHECK3-NEXT: store i16 [[TMP3]], i16* [[DOTADDR3]], align 2 3942 // CHECK3-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 4 3943 // CHECK3-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [2 x i8*]* 3944 // CHECK3-NEXT: [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2 3945 // CHECK3-NEXT: [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2 3946 // CHECK3-NEXT: [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2 3947 // CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 0 3948 // CHECK3-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 4 3949 // CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 3950 // CHECK3-NEXT: [[TMP12:%.*]] = bitcast i8* [[TMP10]] to i32* 3951 // CHECK3-NEXT: [[TMP13:%.*]] = getelementptr i32, i32* [[TMP12]], i32 1 3952 // CHECK3-NEXT: [[TMP14:%.*]] = bitcast i32* [[TMP13]] to i8* 3953 // CHECK3-NEXT: [[TMP15:%.*]] = load i32, i32* [[TMP12]], align 4 3954 // CHECK3-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 3955 // CHECK3-NEXT: [[TMP16:%.*]] = trunc i32 [[NVPTX_WARP_SIZE]] to i16 3956 // CHECK3-NEXT: [[TMP17:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP15]], i16 [[TMP7]], i16 [[TMP16]]) 3957 // CHECK3-NEXT: store i32 [[TMP17]], i32* [[DOTOMP_REDUCTION_ELEMENT]], align 4 3958 // CHECK3-NEXT: [[TMP18:%.*]] = getelementptr i32, i32* [[TMP12]], i32 1 3959 // CHECK3-NEXT: [[TMP19:%.*]] = getelementptr i32, i32* [[DOTOMP_REDUCTION_ELEMENT]], i32 1 3960 // CHECK3-NEXT: [[TMP20:%.*]] = bitcast i32* [[DOTOMP_REDUCTION_ELEMENT]] to i8* 3961 // CHECK3-NEXT: store i8* [[TMP20]], i8** [[TMP11]], align 4 3962 // CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 1 3963 // CHECK3-NEXT: [[TMP22:%.*]] = load i8*, i8** [[TMP21]], align 4 3964 // CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1 3965 // CHECK3-NEXT: [[TMP24:%.*]] = bitcast i8* [[TMP22]] to i16* 3966 // CHECK3-NEXT: [[TMP25:%.*]] = getelementptr i16, i16* [[TMP24]], i32 1 3967 // CHECK3-NEXT: [[TMP26:%.*]] = bitcast i16* [[TMP25]] to i8* 3968 // CHECK3-NEXT: [[TMP27:%.*]] = load i16, i16* [[TMP24]], align 2 3969 // CHECK3-NEXT: [[TMP28:%.*]] = sext i16 [[TMP27]] to i32 3970 // CHECK3-NEXT: [[NVPTX_WARP_SIZE5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 3971 // CHECK3-NEXT: [[TMP29:%.*]] = trunc i32 [[NVPTX_WARP_SIZE5]] to i16 3972 // CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP28]], i16 [[TMP7]], i16 [[TMP29]]) 3973 // CHECK3-NEXT: [[TMP31:%.*]] = trunc i32 [[TMP30]] to i16 3974 // CHECK3-NEXT: store i16 [[TMP31]], i16* [[DOTOMP_REDUCTION_ELEMENT4]], align 2 3975 // CHECK3-NEXT: [[TMP32:%.*]] = getelementptr i16, i16* [[TMP24]], i32 1 3976 // CHECK3-NEXT: [[TMP33:%.*]] = getelementptr i16, i16* [[DOTOMP_REDUCTION_ELEMENT4]], i32 1 3977 // CHECK3-NEXT: [[TMP34:%.*]] = bitcast i16* [[DOTOMP_REDUCTION_ELEMENT4]] to i8* 3978 // CHECK3-NEXT: store i8* [[TMP34]], i8** [[TMP23]], align 4 3979 // CHECK3-NEXT: [[TMP35:%.*]] = icmp eq i16 [[TMP8]], 0 3980 // CHECK3-NEXT: [[TMP36:%.*]] = icmp eq i16 [[TMP8]], 1 3981 // CHECK3-NEXT: [[TMP37:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]] 3982 // CHECK3-NEXT: [[TMP38:%.*]] = and i1 [[TMP36]], [[TMP37]] 3983 // CHECK3-NEXT: [[TMP39:%.*]] = icmp eq i16 [[TMP8]], 2 3984 // CHECK3-NEXT: [[TMP40:%.*]] = and i16 [[TMP6]], 1 3985 // CHECK3-NEXT: [[TMP41:%.*]] = icmp eq i16 [[TMP40]], 0 3986 // CHECK3-NEXT: [[TMP42:%.*]] = and i1 [[TMP39]], [[TMP41]] 3987 // CHECK3-NEXT: [[TMP43:%.*]] = icmp sgt i16 [[TMP7]], 0 3988 // CHECK3-NEXT: [[TMP44:%.*]] = and i1 [[TMP42]], [[TMP43]] 3989 // CHECK3-NEXT: [[TMP45:%.*]] = or i1 [[TMP35]], [[TMP38]] 3990 // CHECK3-NEXT: [[TMP46:%.*]] = or i1 [[TMP45]], [[TMP44]] 3991 // CHECK3-NEXT: br i1 [[TMP46]], label [[THEN:%.*]], label [[ELSE:%.*]] 3992 // CHECK3: then: 3993 // CHECK3-NEXT: [[TMP47:%.*]] = bitcast [2 x i8*]* [[TMP5]] to i8* 3994 // CHECK3-NEXT: [[TMP48:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8* 3995 // CHECK3-NEXT: call void @"_omp$reduction$reduction_func13"(i8* [[TMP47]], i8* [[TMP48]]) #[[ATTR3]] 3996 // CHECK3-NEXT: br label [[IFCONT:%.*]] 3997 // CHECK3: else: 3998 // CHECK3-NEXT: br label [[IFCONT]] 3999 // CHECK3: ifcont: 4000 // CHECK3-NEXT: [[TMP49:%.*]] = icmp eq i16 [[TMP8]], 1 4001 // CHECK3-NEXT: [[TMP50:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]] 4002 // CHECK3-NEXT: [[TMP51:%.*]] = and i1 [[TMP49]], [[TMP50]] 4003 // CHECK3-NEXT: br i1 [[TMP51]], label [[THEN6:%.*]], label [[ELSE7:%.*]] 4004 // CHECK3: then6: 4005 // CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 4006 // CHECK3-NEXT: [[TMP53:%.*]] = load i8*, i8** [[TMP52]], align 4 4007 // CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 0 4008 // CHECK3-NEXT: [[TMP55:%.*]] = load i8*, i8** [[TMP54]], align 4 4009 // CHECK3-NEXT: [[TMP56:%.*]] = bitcast i8* [[TMP53]] to i32* 4010 // CHECK3-NEXT: [[TMP57:%.*]] = bitcast i8* [[TMP55]] to i32* 4011 // CHECK3-NEXT: [[TMP58:%.*]] = load i32, i32* [[TMP56]], align 4 4012 // CHECK3-NEXT: store i32 [[TMP58]], i32* [[TMP57]], align 4 4013 // CHECK3-NEXT: [[TMP59:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1 4014 // CHECK3-NEXT: [[TMP60:%.*]] = load i8*, i8** [[TMP59]], align 4 4015 // CHECK3-NEXT: [[TMP61:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 1 4016 // CHECK3-NEXT: [[TMP62:%.*]] = load i8*, i8** [[TMP61]], align 4 4017 // CHECK3-NEXT: [[TMP63:%.*]] = bitcast i8* [[TMP60]] to i16* 4018 // CHECK3-NEXT: [[TMP64:%.*]] = bitcast i8* [[TMP62]] to i16* 4019 // CHECK3-NEXT: [[TMP65:%.*]] = load i16, i16* [[TMP63]], align 2 4020 // CHECK3-NEXT: store i16 [[TMP65]], i16* [[TMP64]], align 2 4021 // CHECK3-NEXT: br label [[IFCONT8:%.*]] 4022 // CHECK3: else7: 4023 // CHECK3-NEXT: br label [[IFCONT8]] 4024 // CHECK3: ifcont8: 4025 // CHECK3-NEXT: ret void 4026 // 4027 // 4028 // CHECK3-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func15 4029 // CHECK3-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] { 4030 // CHECK3-NEXT: entry: 4031 // CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 4032 // CHECK3-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 4033 // CHECK3-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]]) 4034 // CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 4035 // CHECK3-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 4036 // CHECK3-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 4037 // CHECK3-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 4038 // CHECK3-NEXT: [[NVPTX_LANE_ID:%.*]] = and i32 [[NVPTX_TID2]], 31 4039 // CHECK3-NEXT: [[NVPTX_TID3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 4040 // CHECK3-NEXT: [[NVPTX_WARP_ID:%.*]] = ashr i32 [[NVPTX_TID3]], 5 4041 // CHECK3-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 4042 // CHECK3-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 4043 // CHECK3-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4:[0-9]+]], i32 [[TMP2]]) 4044 // CHECK3-NEXT: [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 4045 // CHECK3-NEXT: br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]] 4046 // CHECK3: then: 4047 // CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 4048 // CHECK3-NEXT: [[TMP6:%.*]] = load i8*, i8** [[TMP5]], align 4 4049 // CHECK3-NEXT: [[TMP7:%.*]] = bitcast i8* [[TMP6]] to i32* 4050 // CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 4051 // CHECK3-NEXT: [[TMP9:%.*]] = load i32, i32* [[TMP7]], align 4 4052 // CHECK3-NEXT: store volatile i32 [[TMP9]], i32 addrspace(3)* [[TMP8]], align 4 4053 // CHECK3-NEXT: br label [[IFCONT:%.*]] 4054 // CHECK3: else: 4055 // CHECK3-NEXT: br label [[IFCONT]] 4056 // CHECK3: ifcont: 4057 // CHECK3-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 4058 // CHECK3-NEXT: [[TMP10:%.*]] = load i32, i32* [[DOTADDR1]], align 4 4059 // CHECK3-NEXT: [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP10]] 4060 // CHECK3-NEXT: br i1 [[IS_ACTIVE_THREAD]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 4061 // CHECK3: then4: 4062 // CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 4063 // CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 4064 // CHECK3-NEXT: [[TMP13:%.*]] = load i8*, i8** [[TMP12]], align 4 4065 // CHECK3-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to i32* 4066 // CHECK3-NEXT: [[TMP15:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP11]], align 4 4067 // CHECK3-NEXT: store i32 [[TMP15]], i32* [[TMP14]], align 4 4068 // CHECK3-NEXT: br label [[IFCONT6:%.*]] 4069 // CHECK3: else5: 4070 // CHECK3-NEXT: br label [[IFCONT6]] 4071 // CHECK3: ifcont6: 4072 // CHECK3-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 4073 // CHECK3-NEXT: [[WARP_MASTER7:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 4074 // CHECK3-NEXT: br i1 [[WARP_MASTER7]], label [[THEN8:%.*]], label [[ELSE9:%.*]] 4075 // CHECK3: then8: 4076 // CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 4077 // CHECK3-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 4 4078 // CHECK3-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i16* 4079 // CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 4080 // CHECK3-NEXT: [[TMP20:%.*]] = bitcast i32 addrspace(3)* [[TMP19]] to i16 addrspace(3)* 4081 // CHECK3-NEXT: [[TMP21:%.*]] = load i16, i16* [[TMP18]], align 2 4082 // CHECK3-NEXT: store volatile i16 [[TMP21]], i16 addrspace(3)* [[TMP20]], align 2 4083 // CHECK3-NEXT: br label [[IFCONT10:%.*]] 4084 // CHECK3: else9: 4085 // CHECK3-NEXT: br label [[IFCONT10]] 4086 // CHECK3: ifcont10: 4087 // CHECK3-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 4088 // CHECK3-NEXT: [[TMP22:%.*]] = load i32, i32* [[DOTADDR1]], align 4 4089 // CHECK3-NEXT: [[IS_ACTIVE_THREAD11:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP22]] 4090 // CHECK3-NEXT: br i1 [[IS_ACTIVE_THREAD11]], label [[THEN12:%.*]], label [[ELSE13:%.*]] 4091 // CHECK3: then12: 4092 // CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 4093 // CHECK3-NEXT: [[TMP24:%.*]] = bitcast i32 addrspace(3)* [[TMP23]] to i16 addrspace(3)* 4094 // CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 4095 // CHECK3-NEXT: [[TMP26:%.*]] = load i8*, i8** [[TMP25]], align 4 4096 // CHECK3-NEXT: [[TMP27:%.*]] = bitcast i8* [[TMP26]] to i16* 4097 // CHECK3-NEXT: [[TMP28:%.*]] = load volatile i16, i16 addrspace(3)* [[TMP24]], align 2 4098 // CHECK3-NEXT: store i16 [[TMP28]], i16* [[TMP27]], align 2 4099 // CHECK3-NEXT: br label [[IFCONT14:%.*]] 4100 // CHECK3: else13: 4101 // CHECK3-NEXT: br label [[IFCONT14]] 4102 // CHECK3: ifcont14: 4103 // CHECK3-NEXT: ret void 4104 // 4105 // 4106 // CHECK3-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func17 4107 // CHECK3-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR0]] { 4108 // CHECK3-NEXT: entry: 4109 // CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 4110 // CHECK3-NEXT: [[DOTADDR1:%.*]] = alloca i16, align 2 4111 // CHECK3-NEXT: [[DOTADDR2:%.*]] = alloca i16, align 2 4112 // CHECK3-NEXT: [[DOTADDR3:%.*]] = alloca i16, align 2 4113 // CHECK3-NEXT: [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [2 x i8*], align 4 4114 // CHECK3-NEXT: [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca i32, align 4 4115 // CHECK3-NEXT: [[DOTOMP_REDUCTION_ELEMENT4:%.*]] = alloca i16, align 2 4116 // CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 4117 // CHECK3-NEXT: store i16 [[TMP1]], i16* [[DOTADDR1]], align 2 4118 // CHECK3-NEXT: store i16 [[TMP2]], i16* [[DOTADDR2]], align 2 4119 // CHECK3-NEXT: store i16 [[TMP3]], i16* [[DOTADDR3]], align 2 4120 // CHECK3-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 4 4121 // CHECK3-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [2 x i8*]* 4122 // CHECK3-NEXT: [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2 4123 // CHECK3-NEXT: [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2 4124 // CHECK3-NEXT: [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2 4125 // CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 0 4126 // CHECK3-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 4 4127 // CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 4128 // CHECK3-NEXT: [[TMP12:%.*]] = bitcast i8* [[TMP10]] to i32* 4129 // CHECK3-NEXT: [[TMP13:%.*]] = getelementptr i32, i32* [[TMP12]], i32 1 4130 // CHECK3-NEXT: [[TMP14:%.*]] = bitcast i32* [[TMP13]] to i8* 4131 // CHECK3-NEXT: [[TMP15:%.*]] = load i32, i32* [[TMP12]], align 4 4132 // CHECK3-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 4133 // CHECK3-NEXT: [[TMP16:%.*]] = trunc i32 [[NVPTX_WARP_SIZE]] to i16 4134 // CHECK3-NEXT: [[TMP17:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP15]], i16 [[TMP7]], i16 [[TMP16]]) 4135 // CHECK3-NEXT: store i32 [[TMP17]], i32* [[DOTOMP_REDUCTION_ELEMENT]], align 4 4136 // CHECK3-NEXT: [[TMP18:%.*]] = getelementptr i32, i32* [[TMP12]], i32 1 4137 // CHECK3-NEXT: [[TMP19:%.*]] = getelementptr i32, i32* [[DOTOMP_REDUCTION_ELEMENT]], i32 1 4138 // CHECK3-NEXT: [[TMP20:%.*]] = bitcast i32* [[DOTOMP_REDUCTION_ELEMENT]] to i8* 4139 // CHECK3-NEXT: store i8* [[TMP20]], i8** [[TMP11]], align 4 4140 // CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 1 4141 // CHECK3-NEXT: [[TMP22:%.*]] = load i8*, i8** [[TMP21]], align 4 4142 // CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1 4143 // CHECK3-NEXT: [[TMP24:%.*]] = bitcast i8* [[TMP22]] to i16* 4144 // CHECK3-NEXT: [[TMP25:%.*]] = getelementptr i16, i16* [[TMP24]], i32 1 4145 // CHECK3-NEXT: [[TMP26:%.*]] = bitcast i16* [[TMP25]] to i8* 4146 // CHECK3-NEXT: [[TMP27:%.*]] = load i16, i16* [[TMP24]], align 2 4147 // CHECK3-NEXT: [[TMP28:%.*]] = sext i16 [[TMP27]] to i32 4148 // CHECK3-NEXT: [[NVPTX_WARP_SIZE5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 4149 // CHECK3-NEXT: [[TMP29:%.*]] = trunc i32 [[NVPTX_WARP_SIZE5]] to i16 4150 // CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP28]], i16 [[TMP7]], i16 [[TMP29]]) 4151 // CHECK3-NEXT: [[TMP31:%.*]] = trunc i32 [[TMP30]] to i16 4152 // CHECK3-NEXT: store i16 [[TMP31]], i16* [[DOTOMP_REDUCTION_ELEMENT4]], align 2 4153 // CHECK3-NEXT: [[TMP32:%.*]] = getelementptr i16, i16* [[TMP24]], i32 1 4154 // CHECK3-NEXT: [[TMP33:%.*]] = getelementptr i16, i16* [[DOTOMP_REDUCTION_ELEMENT4]], i32 1 4155 // CHECK3-NEXT: [[TMP34:%.*]] = bitcast i16* [[DOTOMP_REDUCTION_ELEMENT4]] to i8* 4156 // CHECK3-NEXT: store i8* [[TMP34]], i8** [[TMP23]], align 4 4157 // CHECK3-NEXT: [[TMP35:%.*]] = icmp eq i16 [[TMP8]], 0 4158 // CHECK3-NEXT: [[TMP36:%.*]] = icmp eq i16 [[TMP8]], 1 4159 // CHECK3-NEXT: [[TMP37:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]] 4160 // CHECK3-NEXT: [[TMP38:%.*]] = and i1 [[TMP36]], [[TMP37]] 4161 // CHECK3-NEXT: [[TMP39:%.*]] = icmp eq i16 [[TMP8]], 2 4162 // CHECK3-NEXT: [[TMP40:%.*]] = and i16 [[TMP6]], 1 4163 // CHECK3-NEXT: [[TMP41:%.*]] = icmp eq i16 [[TMP40]], 0 4164 // CHECK3-NEXT: [[TMP42:%.*]] = and i1 [[TMP39]], [[TMP41]] 4165 // CHECK3-NEXT: [[TMP43:%.*]] = icmp sgt i16 [[TMP7]], 0 4166 // CHECK3-NEXT: [[TMP44:%.*]] = and i1 [[TMP42]], [[TMP43]] 4167 // CHECK3-NEXT: [[TMP45:%.*]] = or i1 [[TMP35]], [[TMP38]] 4168 // CHECK3-NEXT: [[TMP46:%.*]] = or i1 [[TMP45]], [[TMP44]] 4169 // CHECK3-NEXT: br i1 [[TMP46]], label [[THEN:%.*]], label [[ELSE:%.*]] 4170 // CHECK3: then: 4171 // CHECK3-NEXT: [[TMP47:%.*]] = bitcast [2 x i8*]* [[TMP5]] to i8* 4172 // CHECK3-NEXT: [[TMP48:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8* 4173 // CHECK3-NEXT: call void @"_omp$reduction$reduction_func16"(i8* [[TMP47]], i8* [[TMP48]]) #[[ATTR3]] 4174 // CHECK3-NEXT: br label [[IFCONT:%.*]] 4175 // CHECK3: else: 4176 // CHECK3-NEXT: br label [[IFCONT]] 4177 // CHECK3: ifcont: 4178 // CHECK3-NEXT: [[TMP49:%.*]] = icmp eq i16 [[TMP8]], 1 4179 // CHECK3-NEXT: [[TMP50:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]] 4180 // CHECK3-NEXT: [[TMP51:%.*]] = and i1 [[TMP49]], [[TMP50]] 4181 // CHECK3-NEXT: br i1 [[TMP51]], label [[THEN6:%.*]], label [[ELSE7:%.*]] 4182 // CHECK3: then6: 4183 // CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 4184 // CHECK3-NEXT: [[TMP53:%.*]] = load i8*, i8** [[TMP52]], align 4 4185 // CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 0 4186 // CHECK3-NEXT: [[TMP55:%.*]] = load i8*, i8** [[TMP54]], align 4 4187 // CHECK3-NEXT: [[TMP56:%.*]] = bitcast i8* [[TMP53]] to i32* 4188 // CHECK3-NEXT: [[TMP57:%.*]] = bitcast i8* [[TMP55]] to i32* 4189 // CHECK3-NEXT: [[TMP58:%.*]] = load i32, i32* [[TMP56]], align 4 4190 // CHECK3-NEXT: store i32 [[TMP58]], i32* [[TMP57]], align 4 4191 // CHECK3-NEXT: [[TMP59:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1 4192 // CHECK3-NEXT: [[TMP60:%.*]] = load i8*, i8** [[TMP59]], align 4 4193 // CHECK3-NEXT: [[TMP61:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 1 4194 // CHECK3-NEXT: [[TMP62:%.*]] = load i8*, i8** [[TMP61]], align 4 4195 // CHECK3-NEXT: [[TMP63:%.*]] = bitcast i8* [[TMP60]] to i16* 4196 // CHECK3-NEXT: [[TMP64:%.*]] = bitcast i8* [[TMP62]] to i16* 4197 // CHECK3-NEXT: [[TMP65:%.*]] = load i16, i16* [[TMP63]], align 2 4198 // CHECK3-NEXT: store i16 [[TMP65]], i16* [[TMP64]], align 2 4199 // CHECK3-NEXT: br label [[IFCONT8:%.*]] 4200 // CHECK3: else7: 4201 // CHECK3-NEXT: br label [[IFCONT8]] 4202 // CHECK3: ifcont8: 4203 // CHECK3-NEXT: ret void 4204 // 4205 // 4206 // CHECK3-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func18 4207 // CHECK3-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] { 4208 // CHECK3-NEXT: entry: 4209 // CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 4210 // CHECK3-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 4211 // CHECK3-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]]) 4212 // CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 4213 // CHECK3-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 4214 // CHECK3-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 4215 // CHECK3-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 4216 // CHECK3-NEXT: [[NVPTX_LANE_ID:%.*]] = and i32 [[NVPTX_TID2]], 31 4217 // CHECK3-NEXT: [[NVPTX_TID3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 4218 // CHECK3-NEXT: [[NVPTX_WARP_ID:%.*]] = ashr i32 [[NVPTX_TID3]], 5 4219 // CHECK3-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 4220 // CHECK3-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 4221 // CHECK3-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 4222 // CHECK3-NEXT: [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 4223 // CHECK3-NEXT: br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]] 4224 // CHECK3: then: 4225 // CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 4226 // CHECK3-NEXT: [[TMP6:%.*]] = load i8*, i8** [[TMP5]], align 4 4227 // CHECK3-NEXT: [[TMP7:%.*]] = bitcast i8* [[TMP6]] to i32* 4228 // CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 4229 // CHECK3-NEXT: [[TMP9:%.*]] = load i32, i32* [[TMP7]], align 4 4230 // CHECK3-NEXT: store volatile i32 [[TMP9]], i32 addrspace(3)* [[TMP8]], align 4 4231 // CHECK3-NEXT: br label [[IFCONT:%.*]] 4232 // CHECK3: else: 4233 // CHECK3-NEXT: br label [[IFCONT]] 4234 // CHECK3: ifcont: 4235 // CHECK3-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 4236 // CHECK3-NEXT: [[TMP10:%.*]] = load i32, i32* [[DOTADDR1]], align 4 4237 // CHECK3-NEXT: [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP10]] 4238 // CHECK3-NEXT: br i1 [[IS_ACTIVE_THREAD]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 4239 // CHECK3: then4: 4240 // CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 4241 // CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 4242 // CHECK3-NEXT: [[TMP13:%.*]] = load i8*, i8** [[TMP12]], align 4 4243 // CHECK3-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to i32* 4244 // CHECK3-NEXT: [[TMP15:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP11]], align 4 4245 // CHECK3-NEXT: store i32 [[TMP15]], i32* [[TMP14]], align 4 4246 // CHECK3-NEXT: br label [[IFCONT6:%.*]] 4247 // CHECK3: else5: 4248 // CHECK3-NEXT: br label [[IFCONT6]] 4249 // CHECK3: ifcont6: 4250 // CHECK3-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 4251 // CHECK3-NEXT: [[WARP_MASTER7:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 4252 // CHECK3-NEXT: br i1 [[WARP_MASTER7]], label [[THEN8:%.*]], label [[ELSE9:%.*]] 4253 // CHECK3: then8: 4254 // CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 4255 // CHECK3-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 4 4256 // CHECK3-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i16* 4257 // CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 4258 // CHECK3-NEXT: [[TMP20:%.*]] = bitcast i32 addrspace(3)* [[TMP19]] to i16 addrspace(3)* 4259 // CHECK3-NEXT: [[TMP21:%.*]] = load i16, i16* [[TMP18]], align 2 4260 // CHECK3-NEXT: store volatile i16 [[TMP21]], i16 addrspace(3)* [[TMP20]], align 2 4261 // CHECK3-NEXT: br label [[IFCONT10:%.*]] 4262 // CHECK3: else9: 4263 // CHECK3-NEXT: br label [[IFCONT10]] 4264 // CHECK3: ifcont10: 4265 // CHECK3-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 4266 // CHECK3-NEXT: [[TMP22:%.*]] = load i32, i32* [[DOTADDR1]], align 4 4267 // CHECK3-NEXT: [[IS_ACTIVE_THREAD11:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP22]] 4268 // CHECK3-NEXT: br i1 [[IS_ACTIVE_THREAD11]], label [[THEN12:%.*]], label [[ELSE13:%.*]] 4269 // CHECK3: then12: 4270 // CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 4271 // CHECK3-NEXT: [[TMP24:%.*]] = bitcast i32 addrspace(3)* [[TMP23]] to i16 addrspace(3)* 4272 // CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 4273 // CHECK3-NEXT: [[TMP26:%.*]] = load i8*, i8** [[TMP25]], align 4 4274 // CHECK3-NEXT: [[TMP27:%.*]] = bitcast i8* [[TMP26]] to i16* 4275 // CHECK3-NEXT: [[TMP28:%.*]] = load volatile i16, i16 addrspace(3)* [[TMP24]], align 2 4276 // CHECK3-NEXT: store i16 [[TMP28]], i16* [[TMP27]], align 2 4277 // CHECK3-NEXT: br label [[IFCONT14:%.*]] 4278 // CHECK3: else13: 4279 // CHECK3-NEXT: br label [[IFCONT14]] 4280 // CHECK3: ifcont14: 4281 // CHECK3-NEXT: ret void 4282 // 4283 // 4284 // CHECK3-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_copy_func19 4285 // CHECK3-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 4286 // CHECK3-NEXT: entry: 4287 // CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 4288 // CHECK3-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 4289 // CHECK3-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 4290 // CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 4291 // CHECK3-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 4292 // CHECK3-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 4293 // CHECK3-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 4294 // CHECK3-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 4295 // CHECK3-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 4 4296 // CHECK3-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.4* 4297 // CHECK3-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 4298 // CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 4299 // CHECK3-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 4 4300 // CHECK3-NEXT: [[TMP10:%.*]] = bitcast i8* [[TMP9]] to i32* 4301 // CHECK3-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4:%.*]], %struct._globalized_locals_ty.4* [[TMP6]], i32 0, i32 0 4302 // CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [1024 x i32], [1024 x i32]* [[A]], i32 0, i32 [[TMP7]] 4303 // CHECK3-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP10]], align 4 4304 // CHECK3-NEXT: store i32 [[TMP12]], i32* [[TMP11]], align 128 4305 // CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 4306 // CHECK3-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 4 4307 // CHECK3-NEXT: [[TMP15:%.*]] = bitcast i8* [[TMP14]] to i16* 4308 // CHECK3-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4]], %struct._globalized_locals_ty.4* [[TMP6]], i32 0, i32 1 4309 // CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1024 x i16], [1024 x i16]* [[B]], i32 0, i32 [[TMP7]] 4310 // CHECK3-NEXT: [[TMP17:%.*]] = load i16, i16* [[TMP15]], align 2 4311 // CHECK3-NEXT: store i16 [[TMP17]], i16* [[TMP16]], align 128 4312 // CHECK3-NEXT: ret void 4313 // 4314 // 4315 // CHECK3-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_reduce_func20 4316 // CHECK3-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 4317 // CHECK3-NEXT: entry: 4318 // CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 4319 // CHECK3-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 4320 // CHECK3-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 4321 // CHECK3-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 4322 // CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 4323 // CHECK3-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 4324 // CHECK3-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 4325 // CHECK3-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 4326 // CHECK3-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.4* 4327 // CHECK3-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 4328 // CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 4329 // CHECK3-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4:%.*]], %struct._globalized_locals_ty.4* [[TMP4]], i32 0, i32 0 4330 // CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1024 x i32], [1024 x i32]* [[A]], i32 0, i32 [[TMP5]] 4331 // CHECK3-NEXT: [[TMP8:%.*]] = bitcast i32* [[TMP7]] to i8* 4332 // CHECK3-NEXT: store i8* [[TMP8]], i8** [[TMP6]], align 4 4333 // CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 4334 // CHECK3-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4]], %struct._globalized_locals_ty.4* [[TMP4]], i32 0, i32 1 4335 // CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [1024 x i16], [1024 x i16]* [[B]], i32 0, i32 [[TMP5]] 4336 // CHECK3-NEXT: [[TMP11:%.*]] = bitcast i16* [[TMP10]] to i8* 4337 // CHECK3-NEXT: store i8* [[TMP11]], i8** [[TMP9]], align 4 4338 // CHECK3-NEXT: [[TMP12:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 4339 // CHECK3-NEXT: [[TMP13:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 4340 // CHECK3-NEXT: call void @"_omp$reduction$reduction_func16"(i8* [[TMP12]], i8* [[TMP13]]) #[[ATTR3]] 4341 // CHECK3-NEXT: ret void 4342 // 4343 // 4344 // CHECK3-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_copy_func21 4345 // CHECK3-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 4346 // CHECK3-NEXT: entry: 4347 // CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 4348 // CHECK3-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 4349 // CHECK3-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 4350 // CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 4351 // CHECK3-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 4352 // CHECK3-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 4353 // CHECK3-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 4354 // CHECK3-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 4355 // CHECK3-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 4 4356 // CHECK3-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.4* 4357 // CHECK3-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 4358 // CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 4359 // CHECK3-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 4 4360 // CHECK3-NEXT: [[TMP10:%.*]] = bitcast i8* [[TMP9]] to i32* 4361 // CHECK3-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4:%.*]], %struct._globalized_locals_ty.4* [[TMP6]], i32 0, i32 0 4362 // CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [1024 x i32], [1024 x i32]* [[A]], i32 0, i32 [[TMP7]] 4363 // CHECK3-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 128 4364 // CHECK3-NEXT: store i32 [[TMP12]], i32* [[TMP10]], align 4 4365 // CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 4366 // CHECK3-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 4 4367 // CHECK3-NEXT: [[TMP15:%.*]] = bitcast i8* [[TMP14]] to i16* 4368 // CHECK3-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4]], %struct._globalized_locals_ty.4* [[TMP6]], i32 0, i32 1 4369 // CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1024 x i16], [1024 x i16]* [[B]], i32 0, i32 [[TMP7]] 4370 // CHECK3-NEXT: [[TMP17:%.*]] = load i16, i16* [[TMP16]], align 128 4371 // CHECK3-NEXT: store i16 [[TMP17]], i16* [[TMP15]], align 2 4372 // CHECK3-NEXT: ret void 4373 // 4374 // 4375 // CHECK3-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_reduce_func22 4376 // CHECK3-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 4377 // CHECK3-NEXT: entry: 4378 // CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 4379 // CHECK3-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 4380 // CHECK3-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 4381 // CHECK3-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 4382 // CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 4383 // CHECK3-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 4384 // CHECK3-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 4385 // CHECK3-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 4386 // CHECK3-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.4* 4387 // CHECK3-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 4388 // CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 4389 // CHECK3-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4:%.*]], %struct._globalized_locals_ty.4* [[TMP4]], i32 0, i32 0 4390 // CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1024 x i32], [1024 x i32]* [[A]], i32 0, i32 [[TMP5]] 4391 // CHECK3-NEXT: [[TMP8:%.*]] = bitcast i32* [[TMP7]] to i8* 4392 // CHECK3-NEXT: store i8* [[TMP8]], i8** [[TMP6]], align 4 4393 // CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 4394 // CHECK3-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4]], %struct._globalized_locals_ty.4* [[TMP4]], i32 0, i32 1 4395 // CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [1024 x i16], [1024 x i16]* [[B]], i32 0, i32 [[TMP5]] 4396 // CHECK3-NEXT: [[TMP11:%.*]] = bitcast i16* [[TMP10]] to i8* 4397 // CHECK3-NEXT: store i8* [[TMP11]], i8** [[TMP9]], align 4 4398 // CHECK3-NEXT: [[TMP12:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 4399 // CHECK3-NEXT: [[TMP13:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 4400 // CHECK3-NEXT: call void @"_omp$reduction$reduction_func16"(i8* [[TMP13]], i8* [[TMP12]]) #[[ATTR3]] 4401 // CHECK3-NEXT: ret void 4402 // 4403 // 4404 // CHECK4-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l23_worker 4405 // CHECK4-SAME: () #[[ATTR0:[0-9]+]] { 4406 // CHECK4-NEXT: entry: 4407 // CHECK4-NEXT: [[WORK_FN:%.*]] = alloca i8*, align 4 4408 // CHECK4-NEXT: [[EXEC_STATUS:%.*]] = alloca i8, align 1 4409 // CHECK4-NEXT: store i8* null, i8** [[WORK_FN]], align 4 4410 // CHECK4-NEXT: store i8 0, i8* [[EXEC_STATUS]], align 1 4411 // CHECK4-NEXT: br label [[DOTAWAIT_WORK:%.*]] 4412 // CHECK4: .await.work: 4413 // CHECK4-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 4414 // CHECK4-NEXT: [[TMP0:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORK_FN]]) 4415 // CHECK4-NEXT: [[TMP1:%.*]] = zext i1 [[TMP0]] to i8 4416 // CHECK4-NEXT: store i8 [[TMP1]], i8* [[EXEC_STATUS]], align 1 4417 // CHECK4-NEXT: [[TMP2:%.*]] = load i8*, i8** [[WORK_FN]], align 4 4418 // CHECK4-NEXT: [[SHOULD_TERMINATE:%.*]] = icmp eq i8* [[TMP2]], null 4419 // CHECK4-NEXT: br i1 [[SHOULD_TERMINATE]], label [[DOTEXIT:%.*]], label [[DOTSELECT_WORKERS:%.*]] 4420 // CHECK4: .select.workers: 4421 // CHECK4-NEXT: [[TMP3:%.*]] = load i8, i8* [[EXEC_STATUS]], align 1 4422 // CHECK4-NEXT: [[IS_ACTIVE:%.*]] = icmp ne i8 [[TMP3]], 0 4423 // CHECK4-NEXT: br i1 [[IS_ACTIVE]], label [[DOTEXECUTE_PARALLEL:%.*]], label [[DOTBARRIER_PARALLEL:%.*]] 4424 // CHECK4: .execute.parallel: 4425 // CHECK4-NEXT: [[TMP4:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]]) 4426 // CHECK4-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP2]] to void (i16, i32)* 4427 // CHECK4-NEXT: call void [[TMP5]](i16 0, i32 [[TMP4]]) 4428 // CHECK4-NEXT: br label [[DOTTERMINATE_PARALLEL:%.*]] 4429 // CHECK4: .terminate.parallel: 4430 // CHECK4-NEXT: call void @__kmpc_kernel_end_parallel() 4431 // CHECK4-NEXT: br label [[DOTBARRIER_PARALLEL]] 4432 // CHECK4: .barrier.parallel: 4433 // CHECK4-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 4434 // CHECK4-NEXT: br label [[DOTAWAIT_WORK]] 4435 // CHECK4: .exit: 4436 // CHECK4-NEXT: ret void 4437 // 4438 // 4439 // CHECK4-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l23 4440 // CHECK4-SAME: (double* nonnull align 8 dereferenceable(8) [[E:%.*]]) #[[ATTR1:[0-9]+]] { 4441 // CHECK4-NEXT: entry: 4442 // CHECK4-NEXT: [[E_ADDR:%.*]] = alloca double*, align 4 4443 // CHECK4-NEXT: [[E7:%.*]] = alloca double, align 8 4444 // CHECK4-NEXT: [[DOTZERO_ADDR:%.*]] = alloca i32, align 4 4445 // CHECK4-NEXT: [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4 4446 // CHECK4-NEXT: store i32 0, i32* [[DOTZERO_ADDR]], align 4 4447 // CHECK4-NEXT: store double* [[E]], double** [[E_ADDR]], align 4 4448 // CHECK4-NEXT: [[TMP0:%.*]] = load double*, double** [[E_ADDR]], align 4 4449 // CHECK4-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 4450 // CHECK4-NEXT: [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 4451 // CHECK4-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 4452 // CHECK4-NEXT: [[THREAD_LIMIT:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS]], [[NVPTX_WARP_SIZE]] 4453 // CHECK4-NEXT: [[TMP1:%.*]] = icmp ult i32 [[NVPTX_TID]], [[THREAD_LIMIT]] 4454 // CHECK4-NEXT: br i1 [[TMP1]], label [[DOTWORKER:%.*]], label [[DOTMASTERCHECK:%.*]] 4455 // CHECK4: .worker: 4456 // CHECK4-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l23_worker() #[[ATTR3:[0-9]+]] 4457 // CHECK4-NEXT: br label [[DOTEXIT:%.*]] 4458 // CHECK4: .mastercheck: 4459 // CHECK4-NEXT: [[NVPTX_TID1:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 4460 // CHECK4-NEXT: [[NVPTX_NUM_THREADS2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 4461 // CHECK4-NEXT: [[NVPTX_WARP_SIZE3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 4462 // CHECK4-NEXT: [[TMP2:%.*]] = sub nuw i32 [[NVPTX_WARP_SIZE3]], 1 4463 // CHECK4-NEXT: [[TMP3:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS2]], 1 4464 // CHECK4-NEXT: [[TMP4:%.*]] = xor i32 [[TMP2]], -1 4465 // CHECK4-NEXT: [[MASTER_TID:%.*]] = and i32 [[TMP3]], [[TMP4]] 4466 // CHECK4-NEXT: [[TMP5:%.*]] = icmp eq i32 [[NVPTX_TID1]], [[MASTER_TID]] 4467 // CHECK4-NEXT: br i1 [[TMP5]], label [[DOTMASTER:%.*]], label [[DOTEXIT]] 4468 // CHECK4: .master: 4469 // CHECK4-NEXT: [[NVPTX_NUM_THREADS4:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 4470 // CHECK4-NEXT: [[NVPTX_WARP_SIZE5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 4471 // CHECK4-NEXT: [[THREAD_LIMIT6:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS4]], [[NVPTX_WARP_SIZE5]] 4472 // CHECK4-NEXT: call void @__kmpc_kernel_init(i32 [[THREAD_LIMIT6]], i16 1) 4473 // CHECK4-NEXT: call void @__kmpc_data_sharing_init_stack() 4474 // CHECK4-NEXT: [[TMP6:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 4475 // CHECK4-NEXT: [[TMP7:%.*]] = load double, double* [[TMP0]], align 8 4476 // CHECK4-NEXT: store double [[TMP7]], double* [[E7]], align 8 4477 // CHECK4-NEXT: store i32 [[TMP6]], i32* [[DOTTHREADID_TEMP_]], align 4 4478 // CHECK4-NEXT: call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], double* [[E7]]) #[[ATTR3]] 4479 // CHECK4-NEXT: br label [[DOTTERMINATION_NOTIFIER:%.*]] 4480 // CHECK4: .termination.notifier: 4481 // CHECK4-NEXT: call void @__kmpc_kernel_deinit(i16 1) 4482 // CHECK4-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 4483 // CHECK4-NEXT: br label [[DOTEXIT]] 4484 // CHECK4: .exit: 4485 // CHECK4-NEXT: ret void 4486 // 4487 // 4488 // CHECK4-LABEL: define {{[^@]+}}@__omp_outlined__ 4489 // CHECK4-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], double* nonnull align 8 dereferenceable(8) [[E:%.*]]) #[[ATTR1]] { 4490 // CHECK4-NEXT: entry: 4491 // CHECK4-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4 4492 // CHECK4-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4 4493 // CHECK4-NEXT: [[E_ADDR:%.*]] = alloca double*, align 4 4494 // CHECK4-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x i8*], align 4 4495 // CHECK4-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4 4496 // CHECK4-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4 4497 // CHECK4-NEXT: store double* [[E]], double** [[E_ADDR]], align 4 4498 // CHECK4-NEXT: [[TMP0:%.*]] = load double*, double** [[E_ADDR]], align 4 4499 // CHECK4-NEXT: [[TMP1:%.*]] = load i16, i16* @"_openmp_static_kernel$is_shared", align 2 4500 // CHECK4-NEXT: [[TMP2:%.*]] = load i32, i32* @"_openmp_static_kernel$size", align 4 4501 // CHECK4-NEXT: call void @__kmpc_get_team_static_memory(i16 0, i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds (%"union._shared_openmp_static_memory_type_$_", %"union._shared_openmp_static_memory_type_$_" addrspace(3)* @"_openmp_shared_static_glob_rd_$_", i32 0, i32 0, i32 0) to i8*), i32 [[TMP2]], i16 [[TMP1]], i8** addrspacecast (i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr" to i8**)) 4502 // CHECK4-NEXT: [[TMP3:%.*]] = load i8*, i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr", align 4 4503 // CHECK4-NEXT: [[TMP4:%.*]] = getelementptr inbounds i8, i8* [[TMP3]], i32 0 4504 // CHECK4-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to %struct._globalized_locals_ty* 4505 // CHECK4-NEXT: [[E1:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY:%.*]], %struct._globalized_locals_ty* [[TMP5]], i32 0, i32 0 4506 // CHECK4-NEXT: store double 0.000000e+00, double* [[E1]], align 8 4507 // CHECK4-NEXT: [[TMP6:%.*]] = load double, double* [[E1]], align 8 4508 // CHECK4-NEXT: [[ADD:%.*]] = fadd double [[TMP6]], 5.000000e+00 4509 // CHECK4-NEXT: store double [[ADD]], double* [[E1]], align 8 4510 // CHECK4-NEXT: [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4 4511 // CHECK4-NEXT: [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4 4512 // CHECK4-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 4513 // CHECK4-NEXT: [[TMP10:%.*]] = bitcast double* [[E1]] to i8* 4514 // CHECK4-NEXT: store i8* [[TMP10]], i8** [[TMP9]], align 4 4515 // CHECK4-NEXT: [[TMP11:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 4516 // CHECK4-NEXT: [[TMP12:%.*]] = load i8*, i8** @"_openmp_teams_reductions_buffer_$_$ptr", align 4 4517 // CHECK4-NEXT: [[TMP13:%.*]] = call i32 @__kmpc_nvptx_teams_reduce_nowait_v2(%struct.ident_t* @[[GLOB1]], i32 [[TMP8]], i8* [[TMP12]], i32 2048, i8* [[TMP11]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_copy_func, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_reduce_func, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_copy_func, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_reduce_func) 4518 // CHECK4-NEXT: [[TMP14:%.*]] = icmp eq i32 [[TMP13]], 1 4519 // CHECK4-NEXT: br i1 [[TMP14]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]] 4520 // CHECK4: .omp.reduction.then: 4521 // CHECK4-NEXT: [[TMP15:%.*]] = load double, double* [[TMP0]], align 8 4522 // CHECK4-NEXT: [[TMP16:%.*]] = load double, double* [[E1]], align 8 4523 // CHECK4-NEXT: [[ADD2:%.*]] = fadd double [[TMP15]], [[TMP16]] 4524 // CHECK4-NEXT: store double [[ADD2]], double* [[TMP0]], align 8 4525 // CHECK4-NEXT: call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP8]]) 4526 // CHECK4-NEXT: br label [[DOTOMP_REDUCTION_DONE]] 4527 // CHECK4: .omp.reduction.done: 4528 // CHECK4-NEXT: [[TMP17:%.*]] = load i16, i16* @"_openmp_static_kernel$is_shared", align 2 4529 // CHECK4-NEXT: call void @__kmpc_restore_team_static_memory(i16 0, i16 [[TMP17]]) 4530 // CHECK4-NEXT: ret void 4531 // 4532 // 4533 // CHECK4-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func 4534 // CHECK4-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR0]] { 4535 // CHECK4-NEXT: entry: 4536 // CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 4537 // CHECK4-NEXT: [[DOTADDR1:%.*]] = alloca i16, align 2 4538 // CHECK4-NEXT: [[DOTADDR2:%.*]] = alloca i16, align 2 4539 // CHECK4-NEXT: [[DOTADDR3:%.*]] = alloca i16, align 2 4540 // CHECK4-NEXT: [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [1 x i8*], align 4 4541 // CHECK4-NEXT: [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca double, align 8 4542 // CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 4543 // CHECK4-NEXT: store i16 [[TMP1]], i16* [[DOTADDR1]], align 2 4544 // CHECK4-NEXT: store i16 [[TMP2]], i16* [[DOTADDR2]], align 2 4545 // CHECK4-NEXT: store i16 [[TMP3]], i16* [[DOTADDR3]], align 2 4546 // CHECK4-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 4 4547 // CHECK4-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [1 x i8*]* 4548 // CHECK4-NEXT: [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2 4549 // CHECK4-NEXT: [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2 4550 // CHECK4-NEXT: [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2 4551 // CHECK4-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i32 0, i32 0 4552 // CHECK4-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 4 4553 // CHECK4-NEXT: [[TMP11:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 4554 // CHECK4-NEXT: [[TMP12:%.*]] = bitcast i8* [[TMP10]] to double* 4555 // CHECK4-NEXT: [[TMP13:%.*]] = getelementptr double, double* [[TMP12]], i32 1 4556 // CHECK4-NEXT: [[TMP14:%.*]] = bitcast double* [[TMP13]] to i8* 4557 // CHECK4-NEXT: [[TMP15:%.*]] = bitcast double* [[TMP12]] to i64* 4558 // CHECK4-NEXT: [[TMP16:%.*]] = bitcast double* [[DOTOMP_REDUCTION_ELEMENT]] to i64* 4559 // CHECK4-NEXT: [[TMP17:%.*]] = load i64, i64* [[TMP15]], align 8 4560 // CHECK4-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 4561 // CHECK4-NEXT: [[TMP18:%.*]] = trunc i32 [[NVPTX_WARP_SIZE]] to i16 4562 // CHECK4-NEXT: [[TMP19:%.*]] = call i64 @__kmpc_shuffle_int64(i64 [[TMP17]], i16 [[TMP7]], i16 [[TMP18]]) 4563 // CHECK4-NEXT: store i64 [[TMP19]], i64* [[TMP16]], align 8 4564 // CHECK4-NEXT: [[TMP20:%.*]] = getelementptr i64, i64* [[TMP15]], i32 1 4565 // CHECK4-NEXT: [[TMP21:%.*]] = getelementptr i64, i64* [[TMP16]], i32 1 4566 // CHECK4-NEXT: [[TMP22:%.*]] = bitcast double* [[DOTOMP_REDUCTION_ELEMENT]] to i8* 4567 // CHECK4-NEXT: store i8* [[TMP22]], i8** [[TMP11]], align 4 4568 // CHECK4-NEXT: [[TMP23:%.*]] = icmp eq i16 [[TMP8]], 0 4569 // CHECK4-NEXT: [[TMP24:%.*]] = icmp eq i16 [[TMP8]], 1 4570 // CHECK4-NEXT: [[TMP25:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]] 4571 // CHECK4-NEXT: [[TMP26:%.*]] = and i1 [[TMP24]], [[TMP25]] 4572 // CHECK4-NEXT: [[TMP27:%.*]] = icmp eq i16 [[TMP8]], 2 4573 // CHECK4-NEXT: [[TMP28:%.*]] = and i16 [[TMP6]], 1 4574 // CHECK4-NEXT: [[TMP29:%.*]] = icmp eq i16 [[TMP28]], 0 4575 // CHECK4-NEXT: [[TMP30:%.*]] = and i1 [[TMP27]], [[TMP29]] 4576 // CHECK4-NEXT: [[TMP31:%.*]] = icmp sgt i16 [[TMP7]], 0 4577 // CHECK4-NEXT: [[TMP32:%.*]] = and i1 [[TMP30]], [[TMP31]] 4578 // CHECK4-NEXT: [[TMP33:%.*]] = or i1 [[TMP23]], [[TMP26]] 4579 // CHECK4-NEXT: [[TMP34:%.*]] = or i1 [[TMP33]], [[TMP32]] 4580 // CHECK4-NEXT: br i1 [[TMP34]], label [[THEN:%.*]], label [[ELSE:%.*]] 4581 // CHECK4: then: 4582 // CHECK4-NEXT: [[TMP35:%.*]] = bitcast [1 x i8*]* [[TMP5]] to i8* 4583 // CHECK4-NEXT: [[TMP36:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8* 4584 // CHECK4-NEXT: call void @"_omp$reduction$reduction_func"(i8* [[TMP35]], i8* [[TMP36]]) #[[ATTR3]] 4585 // CHECK4-NEXT: br label [[IFCONT:%.*]] 4586 // CHECK4: else: 4587 // CHECK4-NEXT: br label [[IFCONT]] 4588 // CHECK4: ifcont: 4589 // CHECK4-NEXT: [[TMP37:%.*]] = icmp eq i16 [[TMP8]], 1 4590 // CHECK4-NEXT: [[TMP38:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]] 4591 // CHECK4-NEXT: [[TMP39:%.*]] = and i1 [[TMP37]], [[TMP38]] 4592 // CHECK4-NEXT: br i1 [[TMP39]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 4593 // CHECK4: then4: 4594 // CHECK4-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 4595 // CHECK4-NEXT: [[TMP41:%.*]] = load i8*, i8** [[TMP40]], align 4 4596 // CHECK4-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i32 0, i32 0 4597 // CHECK4-NEXT: [[TMP43:%.*]] = load i8*, i8** [[TMP42]], align 4 4598 // CHECK4-NEXT: [[TMP44:%.*]] = bitcast i8* [[TMP41]] to double* 4599 // CHECK4-NEXT: [[TMP45:%.*]] = bitcast i8* [[TMP43]] to double* 4600 // CHECK4-NEXT: [[TMP46:%.*]] = load double, double* [[TMP44]], align 8 4601 // CHECK4-NEXT: store double [[TMP46]], double* [[TMP45]], align 8 4602 // CHECK4-NEXT: br label [[IFCONT6:%.*]] 4603 // CHECK4: else5: 4604 // CHECK4-NEXT: br label [[IFCONT6]] 4605 // CHECK4: ifcont6: 4606 // CHECK4-NEXT: ret void 4607 // 4608 // 4609 // CHECK4-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func 4610 // CHECK4-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] { 4611 // CHECK4-NEXT: entry: 4612 // CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 4613 // CHECK4-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 4614 // CHECK4-NEXT: [[DOTCNT_ADDR:%.*]] = alloca i32, align 4 4615 // CHECK4-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 4616 // CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 4617 // CHECK4-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 4618 // CHECK4-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 4619 // CHECK4-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 4620 // CHECK4-NEXT: [[NVPTX_LANE_ID:%.*]] = and i32 [[NVPTX_TID2]], 31 4621 // CHECK4-NEXT: [[NVPTX_TID3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 4622 // CHECK4-NEXT: [[NVPTX_WARP_ID:%.*]] = ashr i32 [[NVPTX_TID3]], 5 4623 // CHECK4-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 4624 // CHECK4-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [1 x i8*]* 4625 // CHECK4-NEXT: store i32 0, i32* [[DOTCNT_ADDR]], align 4 4626 // CHECK4-NEXT: br label [[PRECOND:%.*]] 4627 // CHECK4: precond: 4628 // CHECK4-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTCNT_ADDR]], align 4 4629 // CHECK4-NEXT: [[TMP6:%.*]] = icmp ult i32 [[TMP5]], 2 4630 // CHECK4-NEXT: br i1 [[TMP6]], label [[BODY:%.*]], label [[EXIT:%.*]] 4631 // CHECK4: body: 4632 // CHECK4-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP2]]) 4633 // CHECK4-NEXT: [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 4634 // CHECK4-NEXT: br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]] 4635 // CHECK4: then: 4636 // CHECK4-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP4]], i32 0, i32 0 4637 // CHECK4-NEXT: [[TMP8:%.*]] = load i8*, i8** [[TMP7]], align 4 4638 // CHECK4-NEXT: [[TMP9:%.*]] = bitcast i8* [[TMP8]] to i32* 4639 // CHECK4-NEXT: [[TMP10:%.*]] = getelementptr i32, i32* [[TMP9]], i32 [[TMP5]] 4640 // CHECK4-NEXT: [[TMP11:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 4641 // CHECK4-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP10]], align 4 4642 // CHECK4-NEXT: store volatile i32 [[TMP12]], i32 addrspace(3)* [[TMP11]], align 4 4643 // CHECK4-NEXT: br label [[IFCONT:%.*]] 4644 // CHECK4: else: 4645 // CHECK4-NEXT: br label [[IFCONT]] 4646 // CHECK4: ifcont: 4647 // CHECK4-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 4648 // CHECK4-NEXT: [[TMP13:%.*]] = load i32, i32* [[DOTADDR1]], align 4 4649 // CHECK4-NEXT: [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP13]] 4650 // CHECK4-NEXT: br i1 [[IS_ACTIVE_THREAD]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 4651 // CHECK4: then4: 4652 // CHECK4-NEXT: [[TMP14:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 4653 // CHECK4-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP4]], i32 0, i32 0 4654 // CHECK4-NEXT: [[TMP16:%.*]] = load i8*, i8** [[TMP15]], align 4 4655 // CHECK4-NEXT: [[TMP17:%.*]] = bitcast i8* [[TMP16]] to i32* 4656 // CHECK4-NEXT: [[TMP18:%.*]] = getelementptr i32, i32* [[TMP17]], i32 [[TMP5]] 4657 // CHECK4-NEXT: [[TMP19:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP14]], align 4 4658 // CHECK4-NEXT: store i32 [[TMP19]], i32* [[TMP18]], align 4 4659 // CHECK4-NEXT: br label [[IFCONT6:%.*]] 4660 // CHECK4: else5: 4661 // CHECK4-NEXT: br label [[IFCONT6]] 4662 // CHECK4: ifcont6: 4663 // CHECK4-NEXT: [[TMP20:%.*]] = add nsw i32 [[TMP5]], 1 4664 // CHECK4-NEXT: store i32 [[TMP20]], i32* [[DOTCNT_ADDR]], align 4 4665 // CHECK4-NEXT: br label [[PRECOND]] 4666 // CHECK4: exit: 4667 // CHECK4-NEXT: ret void 4668 // 4669 // 4670 // CHECK4-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_copy_func 4671 // CHECK4-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 4672 // CHECK4-NEXT: entry: 4673 // CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 4674 // CHECK4-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 4675 // CHECK4-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 4676 // CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 4677 // CHECK4-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 4678 // CHECK4-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 4679 // CHECK4-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 4680 // CHECK4-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [1 x i8*]* 4681 // CHECK4-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 4 4682 // CHECK4-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.0* 4683 // CHECK4-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 4684 // CHECK4-NEXT: [[TMP8:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP4]], i32 0, i32 0 4685 // CHECK4-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 4 4686 // CHECK4-NEXT: [[TMP10:%.*]] = bitcast i8* [[TMP9]] to double* 4687 // CHECK4-NEXT: [[E:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_0:%.*]], %struct._globalized_locals_ty.0* [[TMP6]], i32 0, i32 0 4688 // CHECK4-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2048 x double], [2048 x double]* [[E]], i32 0, i32 [[TMP7]] 4689 // CHECK4-NEXT: [[TMP12:%.*]] = load double, double* [[TMP10]], align 8 4690 // CHECK4-NEXT: store double [[TMP12]], double* [[TMP11]], align 128 4691 // CHECK4-NEXT: ret void 4692 // 4693 // 4694 // CHECK4-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_reduce_func 4695 // CHECK4-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 4696 // CHECK4-NEXT: entry: 4697 // CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 4698 // CHECK4-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 4699 // CHECK4-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 4700 // CHECK4-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x i8*], align 4 4701 // CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 4702 // CHECK4-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 4703 // CHECK4-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 4704 // CHECK4-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 4705 // CHECK4-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.0* 4706 // CHECK4-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 4707 // CHECK4-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 4708 // CHECK4-NEXT: [[E:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_0:%.*]], %struct._globalized_locals_ty.0* [[TMP4]], i32 0, i32 0 4709 // CHECK4-NEXT: [[TMP7:%.*]] = getelementptr inbounds [2048 x double], [2048 x double]* [[E]], i32 0, i32 [[TMP5]] 4710 // CHECK4-NEXT: [[TMP8:%.*]] = bitcast double* [[TMP7]] to i8* 4711 // CHECK4-NEXT: store i8* [[TMP8]], i8** [[TMP6]], align 4 4712 // CHECK4-NEXT: [[TMP9:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 4713 // CHECK4-NEXT: [[TMP10:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 4714 // CHECK4-NEXT: call void @"_omp$reduction$reduction_func"(i8* [[TMP9]], i8* [[TMP10]]) #[[ATTR3]] 4715 // CHECK4-NEXT: ret void 4716 // 4717 // 4718 // CHECK4-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_copy_func 4719 // CHECK4-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 4720 // CHECK4-NEXT: entry: 4721 // CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 4722 // CHECK4-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 4723 // CHECK4-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 4724 // CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 4725 // CHECK4-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 4726 // CHECK4-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 4727 // CHECK4-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 4728 // CHECK4-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [1 x i8*]* 4729 // CHECK4-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 4 4730 // CHECK4-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.0* 4731 // CHECK4-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 4732 // CHECK4-NEXT: [[TMP8:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP4]], i32 0, i32 0 4733 // CHECK4-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 4 4734 // CHECK4-NEXT: [[TMP10:%.*]] = bitcast i8* [[TMP9]] to double* 4735 // CHECK4-NEXT: [[E:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_0:%.*]], %struct._globalized_locals_ty.0* [[TMP6]], i32 0, i32 0 4736 // CHECK4-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2048 x double], [2048 x double]* [[E]], i32 0, i32 [[TMP7]] 4737 // CHECK4-NEXT: [[TMP12:%.*]] = load double, double* [[TMP11]], align 128 4738 // CHECK4-NEXT: store double [[TMP12]], double* [[TMP10]], align 8 4739 // CHECK4-NEXT: ret void 4740 // 4741 // 4742 // CHECK4-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_reduce_func 4743 // CHECK4-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 4744 // CHECK4-NEXT: entry: 4745 // CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 4746 // CHECK4-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 4747 // CHECK4-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 4748 // CHECK4-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x i8*], align 4 4749 // CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 4750 // CHECK4-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 4751 // CHECK4-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 4752 // CHECK4-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 4753 // CHECK4-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.0* 4754 // CHECK4-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 4755 // CHECK4-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 4756 // CHECK4-NEXT: [[E:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_0:%.*]], %struct._globalized_locals_ty.0* [[TMP4]], i32 0, i32 0 4757 // CHECK4-NEXT: [[TMP7:%.*]] = getelementptr inbounds [2048 x double], [2048 x double]* [[E]], i32 0, i32 [[TMP5]] 4758 // CHECK4-NEXT: [[TMP8:%.*]] = bitcast double* [[TMP7]] to i8* 4759 // CHECK4-NEXT: store i8* [[TMP8]], i8** [[TMP6]], align 4 4760 // CHECK4-NEXT: [[TMP9:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 4761 // CHECK4-NEXT: [[TMP10:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 4762 // CHECK4-NEXT: call void @"_omp$reduction$reduction_func"(i8* [[TMP10]], i8* [[TMP9]]) #[[ATTR3]] 4763 // CHECK4-NEXT: ret void 4764 // 4765 // 4766 // CHECK4-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29_worker 4767 // CHECK4-SAME: () #[[ATTR0]] { 4768 // CHECK4-NEXT: entry: 4769 // CHECK4-NEXT: [[WORK_FN:%.*]] = alloca i8*, align 4 4770 // CHECK4-NEXT: [[EXEC_STATUS:%.*]] = alloca i8, align 1 4771 // CHECK4-NEXT: store i8* null, i8** [[WORK_FN]], align 4 4772 // CHECK4-NEXT: store i8 0, i8* [[EXEC_STATUS]], align 1 4773 // CHECK4-NEXT: br label [[DOTAWAIT_WORK:%.*]] 4774 // CHECK4: .await.work: 4775 // CHECK4-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 4776 // CHECK4-NEXT: [[TMP0:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORK_FN]]) 4777 // CHECK4-NEXT: [[TMP1:%.*]] = zext i1 [[TMP0]] to i8 4778 // CHECK4-NEXT: store i8 [[TMP1]], i8* [[EXEC_STATUS]], align 1 4779 // CHECK4-NEXT: [[TMP2:%.*]] = load i8*, i8** [[WORK_FN]], align 4 4780 // CHECK4-NEXT: [[SHOULD_TERMINATE:%.*]] = icmp eq i8* [[TMP2]], null 4781 // CHECK4-NEXT: br i1 [[SHOULD_TERMINATE]], label [[DOTEXIT:%.*]], label [[DOTSELECT_WORKERS:%.*]] 4782 // CHECK4: .select.workers: 4783 // CHECK4-NEXT: [[TMP3:%.*]] = load i8, i8* [[EXEC_STATUS]], align 1 4784 // CHECK4-NEXT: [[IS_ACTIVE:%.*]] = icmp ne i8 [[TMP3]], 0 4785 // CHECK4-NEXT: br i1 [[IS_ACTIVE]], label [[DOTEXECUTE_PARALLEL:%.*]], label [[DOTBARRIER_PARALLEL:%.*]] 4786 // CHECK4: .execute.parallel: 4787 // CHECK4-NEXT: [[TMP4:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 4788 // CHECK4-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP2]] to void (i16, i32)* 4789 // CHECK4-NEXT: call void [[TMP5]](i16 0, i32 [[TMP4]]) 4790 // CHECK4-NEXT: br label [[DOTTERMINATE_PARALLEL:%.*]] 4791 // CHECK4: .terminate.parallel: 4792 // CHECK4-NEXT: call void @__kmpc_kernel_end_parallel() 4793 // CHECK4-NEXT: br label [[DOTBARRIER_PARALLEL]] 4794 // CHECK4: .barrier.parallel: 4795 // CHECK4-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 4796 // CHECK4-NEXT: br label [[DOTAWAIT_WORK]] 4797 // CHECK4: .exit: 4798 // CHECK4-NEXT: ret void 4799 // 4800 // 4801 // CHECK4-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29 4802 // CHECK4-SAME: (i32 [[C:%.*]], i32 [[D:%.*]]) #[[ATTR1]] { 4803 // CHECK4-NEXT: entry: 4804 // CHECK4-NEXT: [[C_ADDR:%.*]] = alloca i32, align 4 4805 // CHECK4-NEXT: [[D_ADDR:%.*]] = alloca i32, align 4 4806 // CHECK4-NEXT: [[DOTZERO_ADDR:%.*]] = alloca i32, align 4 4807 // CHECK4-NEXT: [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4 4808 // CHECK4-NEXT: store i32 0, i32* [[DOTZERO_ADDR]], align 4 4809 // CHECK4-NEXT: store i32 [[C]], i32* [[C_ADDR]], align 4 4810 // CHECK4-NEXT: store i32 [[D]], i32* [[D_ADDR]], align 4 4811 // CHECK4-NEXT: [[CONV:%.*]] = bitcast i32* [[C_ADDR]] to i8* 4812 // CHECK4-NEXT: [[CONV1:%.*]] = bitcast i32* [[D_ADDR]] to float* 4813 // CHECK4-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 4814 // CHECK4-NEXT: [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 4815 // CHECK4-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 4816 // CHECK4-NEXT: [[THREAD_LIMIT:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS]], [[NVPTX_WARP_SIZE]] 4817 // CHECK4-NEXT: [[TMP0:%.*]] = icmp ult i32 [[NVPTX_TID]], [[THREAD_LIMIT]] 4818 // CHECK4-NEXT: br i1 [[TMP0]], label [[DOTWORKER:%.*]], label [[DOTMASTERCHECK:%.*]] 4819 // CHECK4: .worker: 4820 // CHECK4-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29_worker() #[[ATTR3]] 4821 // CHECK4-NEXT: br label [[DOTEXIT:%.*]] 4822 // CHECK4: .mastercheck: 4823 // CHECK4-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 4824 // CHECK4-NEXT: [[NVPTX_NUM_THREADS3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 4825 // CHECK4-NEXT: [[NVPTX_WARP_SIZE4:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 4826 // CHECK4-NEXT: [[TMP1:%.*]] = sub nuw i32 [[NVPTX_WARP_SIZE4]], 1 4827 // CHECK4-NEXT: [[TMP2:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS3]], 1 4828 // CHECK4-NEXT: [[TMP3:%.*]] = xor i32 [[TMP1]], -1 4829 // CHECK4-NEXT: [[MASTER_TID:%.*]] = and i32 [[TMP2]], [[TMP3]] 4830 // CHECK4-NEXT: [[TMP4:%.*]] = icmp eq i32 [[NVPTX_TID2]], [[MASTER_TID]] 4831 // CHECK4-NEXT: br i1 [[TMP4]], label [[DOTMASTER:%.*]], label [[DOTEXIT]] 4832 // CHECK4: .master: 4833 // CHECK4-NEXT: [[NVPTX_NUM_THREADS5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 4834 // CHECK4-NEXT: [[NVPTX_WARP_SIZE6:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 4835 // CHECK4-NEXT: [[THREAD_LIMIT7:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS5]], [[NVPTX_WARP_SIZE6]] 4836 // CHECK4-NEXT: call void @__kmpc_kernel_init(i32 [[THREAD_LIMIT7]], i16 1) 4837 // CHECK4-NEXT: call void @__kmpc_data_sharing_init_stack() 4838 // CHECK4-NEXT: [[TMP5:%.*]] = load i16, i16* @"_openmp_static_kernel$is_shared1", align 2 4839 // CHECK4-NEXT: [[TMP6:%.*]] = load i32, i32* @"_openmp_static_kernel$size2", align 4 4840 // CHECK4-NEXT: call void @__kmpc_get_team_static_memory(i16 0, i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds (%"union._shared_openmp_static_memory_type_$_", %"union._shared_openmp_static_memory_type_$_" addrspace(3)* @"_openmp_shared_static_glob_rd_$_", i32 0, i32 0, i32 0) to i8*), i32 [[TMP6]], i16 [[TMP5]], i8** addrspacecast (i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr" to i8**)) 4841 // CHECK4-NEXT: [[TMP7:%.*]] = load i8*, i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr", align 4 4842 // CHECK4-NEXT: [[TMP8:%.*]] = getelementptr inbounds i8, i8* [[TMP7]], i32 0 4843 // CHECK4-NEXT: [[TMP9:%.*]] = bitcast i8* [[TMP8]] to %struct._globalized_locals_ty.1* 4844 // CHECK4-NEXT: [[TMP10:%.*]] = load i8, i8* [[CONV]], align 4 4845 // CHECK4-NEXT: [[C8:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_1:%.*]], %struct._globalized_locals_ty.1* [[TMP9]], i32 0, i32 1 4846 // CHECK4-NEXT: store i8 [[TMP10]], i8* [[C8]], align 4 4847 // CHECK4-NEXT: [[TMP11:%.*]] = load float, float* [[CONV1]], align 4 4848 // CHECK4-NEXT: [[D9:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_1]], %struct._globalized_locals_ty.1* [[TMP9]], i32 0, i32 0 4849 // CHECK4-NEXT: store float [[TMP11]], float* [[D9]], align 4 4850 // CHECK4-NEXT: [[TMP12:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 4851 // CHECK4-NEXT: store i32 [[TMP12]], i32* [[DOTTHREADID_TEMP_]], align 4 4852 // CHECK4-NEXT: call void @__omp_outlined__3(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i8* [[C8]], float* [[D9]]) #[[ATTR3]] 4853 // CHECK4-NEXT: [[TMP13:%.*]] = load i16, i16* @"_openmp_static_kernel$is_shared1", align 2 4854 // CHECK4-NEXT: call void @__kmpc_restore_team_static_memory(i16 0, i16 [[TMP13]]) 4855 // CHECK4-NEXT: br label [[DOTTERMINATION_NOTIFIER:%.*]] 4856 // CHECK4: .termination.notifier: 4857 // CHECK4-NEXT: call void @__kmpc_kernel_deinit(i16 1) 4858 // CHECK4-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 4859 // CHECK4-NEXT: br label [[DOTEXIT]] 4860 // CHECK4: .exit: 4861 // CHECK4-NEXT: ret void 4862 // 4863 // 4864 // CHECK4-LABEL: define {{[^@]+}}@__omp_outlined__3 4865 // CHECK4-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i8* nonnull align 1 dereferenceable(1) [[C:%.*]], float* nonnull align 4 dereferenceable(4) [[D:%.*]]) #[[ATTR1]] { 4866 // CHECK4-NEXT: entry: 4867 // CHECK4-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4 4868 // CHECK4-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4 4869 // CHECK4-NEXT: [[C_ADDR:%.*]] = alloca i8*, align 4 4870 // CHECK4-NEXT: [[D_ADDR:%.*]] = alloca float*, align 4 4871 // CHECK4-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 4872 // CHECK4-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4 4873 // CHECK4-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4 4874 // CHECK4-NEXT: store i8* [[C]], i8** [[C_ADDR]], align 4 4875 // CHECK4-NEXT: store float* [[D]], float** [[D_ADDR]], align 4 4876 // CHECK4-NEXT: [[TMP0:%.*]] = load i8*, i8** [[C_ADDR]], align 4 4877 // CHECK4-NEXT: [[TMP1:%.*]] = load float*, float** [[D_ADDR]], align 4 4878 // CHECK4-NEXT: [[TMP2:%.*]] = load i8*, i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr", align 4 4879 // CHECK4-NEXT: [[TMP3:%.*]] = getelementptr inbounds i8, i8* [[TMP2]], i32 8 4880 // CHECK4-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.2* 4881 // CHECK4-NEXT: [[C1:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_2:%.*]], %struct._globalized_locals_ty.2* [[TMP4]], i32 0, i32 1 4882 // CHECK4-NEXT: [[D2:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_2]], %struct._globalized_locals_ty.2* [[TMP4]], i32 0, i32 0 4883 // CHECK4-NEXT: store i8 0, i8* [[C1]], align 4 4884 // CHECK4-NEXT: store float 1.000000e+00, float* [[D2]], align 4 4885 // CHECK4-NEXT: [[TMP5:%.*]] = load i8, i8* [[C1]], align 4 4886 // CHECK4-NEXT: [[CONV:%.*]] = sext i8 [[TMP5]] to i32 4887 // CHECK4-NEXT: [[XOR:%.*]] = xor i32 [[CONV]], 2 4888 // CHECK4-NEXT: [[CONV3:%.*]] = trunc i32 [[XOR]] to i8 4889 // CHECK4-NEXT: store i8 [[CONV3]], i8* [[C1]], align 4 4890 // CHECK4-NEXT: [[TMP6:%.*]] = load float, float* [[D2]], align 4 4891 // CHECK4-NEXT: [[MUL:%.*]] = fmul float [[TMP6]], 3.300000e+01 4892 // CHECK4-NEXT: store float [[MUL]], float* [[D2]], align 4 4893 // CHECK4-NEXT: [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4 4894 // CHECK4-NEXT: [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4 4895 // CHECK4-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 4896 // CHECK4-NEXT: store i8* [[C1]], i8** [[TMP9]], align 4 4897 // CHECK4-NEXT: [[TMP10:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 4898 // CHECK4-NEXT: [[TMP11:%.*]] = bitcast float* [[D2]] to i8* 4899 // CHECK4-NEXT: store i8* [[TMP11]], i8** [[TMP10]], align 4 4900 // CHECK4-NEXT: [[TMP12:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 4901 // CHECK4-NEXT: [[TMP13:%.*]] = load i8*, i8** @"_openmp_teams_reductions_buffer_$_$ptr", align 4 4902 // CHECK4-NEXT: [[TMP14:%.*]] = call i32 @__kmpc_nvptx_teams_reduce_nowait_v2(%struct.ident_t* @[[GLOB1]], i32 [[TMP8]], i8* [[TMP13]], i32 2048, i8* [[TMP12]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func5, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func6, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_copy_func7, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_reduce_func8, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_copy_func9, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_reduce_func10) 4903 // CHECK4-NEXT: [[TMP15:%.*]] = icmp eq i32 [[TMP14]], 1 4904 // CHECK4-NEXT: br i1 [[TMP15]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]] 4905 // CHECK4: .omp.reduction.then: 4906 // CHECK4-NEXT: [[TMP16:%.*]] = load i8, i8* [[TMP0]], align 1 4907 // CHECK4-NEXT: [[CONV4:%.*]] = sext i8 [[TMP16]] to i32 4908 // CHECK4-NEXT: [[TMP17:%.*]] = load i8, i8* [[C1]], align 4 4909 // CHECK4-NEXT: [[CONV5:%.*]] = sext i8 [[TMP17]] to i32 4910 // CHECK4-NEXT: [[XOR6:%.*]] = xor i32 [[CONV4]], [[CONV5]] 4911 // CHECK4-NEXT: [[CONV7:%.*]] = trunc i32 [[XOR6]] to i8 4912 // CHECK4-NEXT: store i8 [[CONV7]], i8* [[TMP0]], align 1 4913 // CHECK4-NEXT: [[TMP18:%.*]] = load float, float* [[TMP1]], align 4 4914 // CHECK4-NEXT: [[TMP19:%.*]] = load float, float* [[D2]], align 4 4915 // CHECK4-NEXT: [[MUL8:%.*]] = fmul float [[TMP18]], [[TMP19]] 4916 // CHECK4-NEXT: store float [[MUL8]], float* [[TMP1]], align 4 4917 // CHECK4-NEXT: call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP8]]) 4918 // CHECK4-NEXT: br label [[DOTOMP_REDUCTION_DONE]] 4919 // CHECK4: .omp.reduction.done: 4920 // CHECK4-NEXT: ret void 4921 // 4922 // 4923 // CHECK4-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func5 4924 // CHECK4-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR0]] { 4925 // CHECK4-NEXT: entry: 4926 // CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 4927 // CHECK4-NEXT: [[DOTADDR1:%.*]] = alloca i16, align 2 4928 // CHECK4-NEXT: [[DOTADDR2:%.*]] = alloca i16, align 2 4929 // CHECK4-NEXT: [[DOTADDR3:%.*]] = alloca i16, align 2 4930 // CHECK4-NEXT: [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [2 x i8*], align 4 4931 // CHECK4-NEXT: [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca i8, align 1 4932 // CHECK4-NEXT: [[DOTOMP_REDUCTION_ELEMENT4:%.*]] = alloca float, align 4 4933 // CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 4934 // CHECK4-NEXT: store i16 [[TMP1]], i16* [[DOTADDR1]], align 2 4935 // CHECK4-NEXT: store i16 [[TMP2]], i16* [[DOTADDR2]], align 2 4936 // CHECK4-NEXT: store i16 [[TMP3]], i16* [[DOTADDR3]], align 2 4937 // CHECK4-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 4 4938 // CHECK4-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [2 x i8*]* 4939 // CHECK4-NEXT: [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2 4940 // CHECK4-NEXT: [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2 4941 // CHECK4-NEXT: [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2 4942 // CHECK4-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 0 4943 // CHECK4-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 4 4944 // CHECK4-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 4945 // CHECK4-NEXT: [[TMP12:%.*]] = getelementptr i8, i8* [[TMP10]], i32 1 4946 // CHECK4-NEXT: [[TMP13:%.*]] = load i8, i8* [[TMP10]], align 1 4947 // CHECK4-NEXT: [[TMP14:%.*]] = sext i8 [[TMP13]] to i32 4948 // CHECK4-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 4949 // CHECK4-NEXT: [[TMP15:%.*]] = trunc i32 [[NVPTX_WARP_SIZE]] to i16 4950 // CHECK4-NEXT: [[TMP16:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP14]], i16 [[TMP7]], i16 [[TMP15]]) 4951 // CHECK4-NEXT: [[TMP17:%.*]] = trunc i32 [[TMP16]] to i8 4952 // CHECK4-NEXT: store i8 [[TMP17]], i8* [[DOTOMP_REDUCTION_ELEMENT]], align 1 4953 // CHECK4-NEXT: [[TMP18:%.*]] = getelementptr i8, i8* [[TMP10]], i32 1 4954 // CHECK4-NEXT: [[TMP19:%.*]] = getelementptr i8, i8* [[DOTOMP_REDUCTION_ELEMENT]], i32 1 4955 // CHECK4-NEXT: store i8* [[DOTOMP_REDUCTION_ELEMENT]], i8** [[TMP11]], align 4 4956 // CHECK4-NEXT: [[TMP20:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 1 4957 // CHECK4-NEXT: [[TMP21:%.*]] = load i8*, i8** [[TMP20]], align 4 4958 // CHECK4-NEXT: [[TMP22:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1 4959 // CHECK4-NEXT: [[TMP23:%.*]] = bitcast i8* [[TMP21]] to float* 4960 // CHECK4-NEXT: [[TMP24:%.*]] = getelementptr float, float* [[TMP23]], i32 1 4961 // CHECK4-NEXT: [[TMP25:%.*]] = bitcast float* [[TMP24]] to i8* 4962 // CHECK4-NEXT: [[TMP26:%.*]] = bitcast float* [[TMP23]] to i32* 4963 // CHECK4-NEXT: [[TMP27:%.*]] = bitcast float* [[DOTOMP_REDUCTION_ELEMENT4]] to i32* 4964 // CHECK4-NEXT: [[TMP28:%.*]] = load i32, i32* [[TMP26]], align 4 4965 // CHECK4-NEXT: [[NVPTX_WARP_SIZE5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 4966 // CHECK4-NEXT: [[TMP29:%.*]] = trunc i32 [[NVPTX_WARP_SIZE5]] to i16 4967 // CHECK4-NEXT: [[TMP30:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP28]], i16 [[TMP7]], i16 [[TMP29]]) 4968 // CHECK4-NEXT: store i32 [[TMP30]], i32* [[TMP27]], align 4 4969 // CHECK4-NEXT: [[TMP31:%.*]] = getelementptr i32, i32* [[TMP26]], i32 1 4970 // CHECK4-NEXT: [[TMP32:%.*]] = getelementptr i32, i32* [[TMP27]], i32 1 4971 // CHECK4-NEXT: [[TMP33:%.*]] = bitcast float* [[DOTOMP_REDUCTION_ELEMENT4]] to i8* 4972 // CHECK4-NEXT: store i8* [[TMP33]], i8** [[TMP22]], align 4 4973 // CHECK4-NEXT: [[TMP34:%.*]] = icmp eq i16 [[TMP8]], 0 4974 // CHECK4-NEXT: [[TMP35:%.*]] = icmp eq i16 [[TMP8]], 1 4975 // CHECK4-NEXT: [[TMP36:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]] 4976 // CHECK4-NEXT: [[TMP37:%.*]] = and i1 [[TMP35]], [[TMP36]] 4977 // CHECK4-NEXT: [[TMP38:%.*]] = icmp eq i16 [[TMP8]], 2 4978 // CHECK4-NEXT: [[TMP39:%.*]] = and i16 [[TMP6]], 1 4979 // CHECK4-NEXT: [[TMP40:%.*]] = icmp eq i16 [[TMP39]], 0 4980 // CHECK4-NEXT: [[TMP41:%.*]] = and i1 [[TMP38]], [[TMP40]] 4981 // CHECK4-NEXT: [[TMP42:%.*]] = icmp sgt i16 [[TMP7]], 0 4982 // CHECK4-NEXT: [[TMP43:%.*]] = and i1 [[TMP41]], [[TMP42]] 4983 // CHECK4-NEXT: [[TMP44:%.*]] = or i1 [[TMP34]], [[TMP37]] 4984 // CHECK4-NEXT: [[TMP45:%.*]] = or i1 [[TMP44]], [[TMP43]] 4985 // CHECK4-NEXT: br i1 [[TMP45]], label [[THEN:%.*]], label [[ELSE:%.*]] 4986 // CHECK4: then: 4987 // CHECK4-NEXT: [[TMP46:%.*]] = bitcast [2 x i8*]* [[TMP5]] to i8* 4988 // CHECK4-NEXT: [[TMP47:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8* 4989 // CHECK4-NEXT: call void @"_omp$reduction$reduction_func4"(i8* [[TMP46]], i8* [[TMP47]]) #[[ATTR3]] 4990 // CHECK4-NEXT: br label [[IFCONT:%.*]] 4991 // CHECK4: else: 4992 // CHECK4-NEXT: br label [[IFCONT]] 4993 // CHECK4: ifcont: 4994 // CHECK4-NEXT: [[TMP48:%.*]] = icmp eq i16 [[TMP8]], 1 4995 // CHECK4-NEXT: [[TMP49:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]] 4996 // CHECK4-NEXT: [[TMP50:%.*]] = and i1 [[TMP48]], [[TMP49]] 4997 // CHECK4-NEXT: br i1 [[TMP50]], label [[THEN6:%.*]], label [[ELSE7:%.*]] 4998 // CHECK4: then6: 4999 // CHECK4-NEXT: [[TMP51:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 5000 // CHECK4-NEXT: [[TMP52:%.*]] = load i8*, i8** [[TMP51]], align 4 5001 // CHECK4-NEXT: [[TMP53:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 0 5002 // CHECK4-NEXT: [[TMP54:%.*]] = load i8*, i8** [[TMP53]], align 4 5003 // CHECK4-NEXT: [[TMP55:%.*]] = load i8, i8* [[TMP52]], align 1 5004 // CHECK4-NEXT: store i8 [[TMP55]], i8* [[TMP54]], align 1 5005 // CHECK4-NEXT: [[TMP56:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1 5006 // CHECK4-NEXT: [[TMP57:%.*]] = load i8*, i8** [[TMP56]], align 4 5007 // CHECK4-NEXT: [[TMP58:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 1 5008 // CHECK4-NEXT: [[TMP59:%.*]] = load i8*, i8** [[TMP58]], align 4 5009 // CHECK4-NEXT: [[TMP60:%.*]] = bitcast i8* [[TMP57]] to float* 5010 // CHECK4-NEXT: [[TMP61:%.*]] = bitcast i8* [[TMP59]] to float* 5011 // CHECK4-NEXT: [[TMP62:%.*]] = load float, float* [[TMP60]], align 4 5012 // CHECK4-NEXT: store float [[TMP62]], float* [[TMP61]], align 4 5013 // CHECK4-NEXT: br label [[IFCONT8:%.*]] 5014 // CHECK4: else7: 5015 // CHECK4-NEXT: br label [[IFCONT8]] 5016 // CHECK4: ifcont8: 5017 // CHECK4-NEXT: ret void 5018 // 5019 // 5020 // CHECK4-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func6 5021 // CHECK4-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] { 5022 // CHECK4-NEXT: entry: 5023 // CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 5024 // CHECK4-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 5025 // CHECK4-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 5026 // CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 5027 // CHECK4-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 5028 // CHECK4-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 5029 // CHECK4-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 5030 // CHECK4-NEXT: [[NVPTX_LANE_ID:%.*]] = and i32 [[NVPTX_TID2]], 31 5031 // CHECK4-NEXT: [[NVPTX_TID3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 5032 // CHECK4-NEXT: [[NVPTX_WARP_ID:%.*]] = ashr i32 [[NVPTX_TID3]], 5 5033 // CHECK4-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 5034 // CHECK4-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 5035 // CHECK4-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 5036 // CHECK4-NEXT: [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 5037 // CHECK4-NEXT: br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]] 5038 // CHECK4: then: 5039 // CHECK4-NEXT: [[TMP5:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 5040 // CHECK4-NEXT: [[TMP6:%.*]] = load i8*, i8** [[TMP5]], align 4 5041 // CHECK4-NEXT: [[TMP7:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 5042 // CHECK4-NEXT: [[TMP8:%.*]] = bitcast i32 addrspace(3)* [[TMP7]] to i8 addrspace(3)* 5043 // CHECK4-NEXT: [[TMP9:%.*]] = load i8, i8* [[TMP6]], align 1 5044 // CHECK4-NEXT: store volatile i8 [[TMP9]], i8 addrspace(3)* [[TMP8]], align 1 5045 // CHECK4-NEXT: br label [[IFCONT:%.*]] 5046 // CHECK4: else: 5047 // CHECK4-NEXT: br label [[IFCONT]] 5048 // CHECK4: ifcont: 5049 // CHECK4-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 5050 // CHECK4-NEXT: [[TMP10:%.*]] = load i32, i32* [[DOTADDR1]], align 4 5051 // CHECK4-NEXT: [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP10]] 5052 // CHECK4-NEXT: br i1 [[IS_ACTIVE_THREAD]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 5053 // CHECK4: then4: 5054 // CHECK4-NEXT: [[TMP11:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 5055 // CHECK4-NEXT: [[TMP12:%.*]] = bitcast i32 addrspace(3)* [[TMP11]] to i8 addrspace(3)* 5056 // CHECK4-NEXT: [[TMP13:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 5057 // CHECK4-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 4 5058 // CHECK4-NEXT: [[TMP15:%.*]] = load volatile i8, i8 addrspace(3)* [[TMP12]], align 1 5059 // CHECK4-NEXT: store i8 [[TMP15]], i8* [[TMP14]], align 1 5060 // CHECK4-NEXT: br label [[IFCONT6:%.*]] 5061 // CHECK4: else5: 5062 // CHECK4-NEXT: br label [[IFCONT6]] 5063 // CHECK4: ifcont6: 5064 // CHECK4-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 5065 // CHECK4-NEXT: [[WARP_MASTER7:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 5066 // CHECK4-NEXT: br i1 [[WARP_MASTER7]], label [[THEN8:%.*]], label [[ELSE9:%.*]] 5067 // CHECK4: then8: 5068 // CHECK4-NEXT: [[TMP16:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 5069 // CHECK4-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 4 5070 // CHECK4-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i32* 5071 // CHECK4-NEXT: [[TMP19:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 5072 // CHECK4-NEXT: [[TMP20:%.*]] = load i32, i32* [[TMP18]], align 4 5073 // CHECK4-NEXT: store volatile i32 [[TMP20]], i32 addrspace(3)* [[TMP19]], align 4 5074 // CHECK4-NEXT: br label [[IFCONT10:%.*]] 5075 // CHECK4: else9: 5076 // CHECK4-NEXT: br label [[IFCONT10]] 5077 // CHECK4: ifcont10: 5078 // CHECK4-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 5079 // CHECK4-NEXT: [[TMP21:%.*]] = load i32, i32* [[DOTADDR1]], align 4 5080 // CHECK4-NEXT: [[IS_ACTIVE_THREAD11:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP21]] 5081 // CHECK4-NEXT: br i1 [[IS_ACTIVE_THREAD11]], label [[THEN12:%.*]], label [[ELSE13:%.*]] 5082 // CHECK4: then12: 5083 // CHECK4-NEXT: [[TMP22:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 5084 // CHECK4-NEXT: [[TMP23:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 5085 // CHECK4-NEXT: [[TMP24:%.*]] = load i8*, i8** [[TMP23]], align 4 5086 // CHECK4-NEXT: [[TMP25:%.*]] = bitcast i8* [[TMP24]] to i32* 5087 // CHECK4-NEXT: [[TMP26:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP22]], align 4 5088 // CHECK4-NEXT: store i32 [[TMP26]], i32* [[TMP25]], align 4 5089 // CHECK4-NEXT: br label [[IFCONT14:%.*]] 5090 // CHECK4: else13: 5091 // CHECK4-NEXT: br label [[IFCONT14]] 5092 // CHECK4: ifcont14: 5093 // CHECK4-NEXT: ret void 5094 // 5095 // 5096 // CHECK4-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_copy_func7 5097 // CHECK4-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 5098 // CHECK4-NEXT: entry: 5099 // CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 5100 // CHECK4-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 5101 // CHECK4-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 5102 // CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 5103 // CHECK4-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 5104 // CHECK4-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 5105 // CHECK4-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 5106 // CHECK4-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 5107 // CHECK4-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 4 5108 // CHECK4-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.3* 5109 // CHECK4-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 5110 // CHECK4-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 5111 // CHECK4-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 4 5112 // CHECK4-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3:%.*]], %struct._globalized_locals_ty.3* [[TMP6]], i32 0, i32 0 5113 // CHECK4-NEXT: [[TMP10:%.*]] = getelementptr inbounds [2048 x i8], [2048 x i8]* [[C]], i32 0, i32 [[TMP7]] 5114 // CHECK4-NEXT: [[TMP11:%.*]] = load i8, i8* [[TMP9]], align 1 5115 // CHECK4-NEXT: store i8 [[TMP11]], i8* [[TMP10]], align 128 5116 // CHECK4-NEXT: [[TMP12:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 5117 // CHECK4-NEXT: [[TMP13:%.*]] = load i8*, i8** [[TMP12]], align 4 5118 // CHECK4-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to float* 5119 // CHECK4-NEXT: [[D:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3]], %struct._globalized_locals_ty.3* [[TMP6]], i32 0, i32 1 5120 // CHECK4-NEXT: [[TMP15:%.*]] = getelementptr inbounds [2048 x float], [2048 x float]* [[D]], i32 0, i32 [[TMP7]] 5121 // CHECK4-NEXT: [[TMP16:%.*]] = load float, float* [[TMP14]], align 4 5122 // CHECK4-NEXT: store float [[TMP16]], float* [[TMP15]], align 128 5123 // CHECK4-NEXT: ret void 5124 // 5125 // 5126 // CHECK4-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_reduce_func8 5127 // CHECK4-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 5128 // CHECK4-NEXT: entry: 5129 // CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 5130 // CHECK4-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 5131 // CHECK4-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 5132 // CHECK4-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 5133 // CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 5134 // CHECK4-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 5135 // CHECK4-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 5136 // CHECK4-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 5137 // CHECK4-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.3* 5138 // CHECK4-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 5139 // CHECK4-NEXT: [[TMP6:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 5140 // CHECK4-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3:%.*]], %struct._globalized_locals_ty.3* [[TMP4]], i32 0, i32 0 5141 // CHECK4-NEXT: [[TMP7:%.*]] = getelementptr inbounds [2048 x i8], [2048 x i8]* [[C]], i32 0, i32 [[TMP5]] 5142 // CHECK4-NEXT: store i8* [[TMP7]], i8** [[TMP6]], align 4 5143 // CHECK4-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 5144 // CHECK4-NEXT: [[D:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3]], %struct._globalized_locals_ty.3* [[TMP4]], i32 0, i32 1 5145 // CHECK4-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2048 x float], [2048 x float]* [[D]], i32 0, i32 [[TMP5]] 5146 // CHECK4-NEXT: [[TMP10:%.*]] = bitcast float* [[TMP9]] to i8* 5147 // CHECK4-NEXT: store i8* [[TMP10]], i8** [[TMP8]], align 4 5148 // CHECK4-NEXT: [[TMP11:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 5149 // CHECK4-NEXT: [[TMP12:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 5150 // CHECK4-NEXT: call void @"_omp$reduction$reduction_func4"(i8* [[TMP11]], i8* [[TMP12]]) #[[ATTR3]] 5151 // CHECK4-NEXT: ret void 5152 // 5153 // 5154 // CHECK4-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_copy_func9 5155 // CHECK4-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 5156 // CHECK4-NEXT: entry: 5157 // CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 5158 // CHECK4-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 5159 // CHECK4-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 5160 // CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 5161 // CHECK4-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 5162 // CHECK4-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 5163 // CHECK4-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 5164 // CHECK4-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 5165 // CHECK4-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 4 5166 // CHECK4-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.3* 5167 // CHECK4-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 5168 // CHECK4-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 5169 // CHECK4-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 4 5170 // CHECK4-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3:%.*]], %struct._globalized_locals_ty.3* [[TMP6]], i32 0, i32 0 5171 // CHECK4-NEXT: [[TMP10:%.*]] = getelementptr inbounds [2048 x i8], [2048 x i8]* [[C]], i32 0, i32 [[TMP7]] 5172 // CHECK4-NEXT: [[TMP11:%.*]] = load i8, i8* [[TMP10]], align 128 5173 // CHECK4-NEXT: store i8 [[TMP11]], i8* [[TMP9]], align 1 5174 // CHECK4-NEXT: [[TMP12:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 5175 // CHECK4-NEXT: [[TMP13:%.*]] = load i8*, i8** [[TMP12]], align 4 5176 // CHECK4-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to float* 5177 // CHECK4-NEXT: [[D:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3]], %struct._globalized_locals_ty.3* [[TMP6]], i32 0, i32 1 5178 // CHECK4-NEXT: [[TMP15:%.*]] = getelementptr inbounds [2048 x float], [2048 x float]* [[D]], i32 0, i32 [[TMP7]] 5179 // CHECK4-NEXT: [[TMP16:%.*]] = load float, float* [[TMP15]], align 128 5180 // CHECK4-NEXT: store float [[TMP16]], float* [[TMP14]], align 4 5181 // CHECK4-NEXT: ret void 5182 // 5183 // 5184 // CHECK4-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_reduce_func10 5185 // CHECK4-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 5186 // CHECK4-NEXT: entry: 5187 // CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 5188 // CHECK4-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 5189 // CHECK4-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 5190 // CHECK4-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 5191 // CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 5192 // CHECK4-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 5193 // CHECK4-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 5194 // CHECK4-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 5195 // CHECK4-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.3* 5196 // CHECK4-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 5197 // CHECK4-NEXT: [[TMP6:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 5198 // CHECK4-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3:%.*]], %struct._globalized_locals_ty.3* [[TMP4]], i32 0, i32 0 5199 // CHECK4-NEXT: [[TMP7:%.*]] = getelementptr inbounds [2048 x i8], [2048 x i8]* [[C]], i32 0, i32 [[TMP5]] 5200 // CHECK4-NEXT: store i8* [[TMP7]], i8** [[TMP6]], align 4 5201 // CHECK4-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 5202 // CHECK4-NEXT: [[D:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3]], %struct._globalized_locals_ty.3* [[TMP4]], i32 0, i32 1 5203 // CHECK4-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2048 x float], [2048 x float]* [[D]], i32 0, i32 [[TMP5]] 5204 // CHECK4-NEXT: [[TMP10:%.*]] = bitcast float* [[TMP9]] to i8* 5205 // CHECK4-NEXT: store i8* [[TMP10]], i8** [[TMP8]], align 4 5206 // CHECK4-NEXT: [[TMP11:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 5207 // CHECK4-NEXT: [[TMP12:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 5208 // CHECK4-NEXT: call void @"_omp$reduction$reduction_func4"(i8* [[TMP12]], i8* [[TMP11]]) #[[ATTR3]] 5209 // CHECK4-NEXT: ret void 5210 // 5211 // 5212 // CHECK4-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l36 5213 // CHECK4-SAME: (i32 [[A:%.*]], i32 [[B:%.*]]) #[[ATTR1]] { 5214 // CHECK4-NEXT: entry: 5215 // CHECK4-NEXT: [[A_ADDR:%.*]] = alloca i32, align 4 5216 // CHECK4-NEXT: [[B_ADDR:%.*]] = alloca i32, align 4 5217 // CHECK4-NEXT: [[DOTZERO_ADDR:%.*]] = alloca i32, align 4 5218 // CHECK4-NEXT: [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4 5219 // CHECK4-NEXT: store i32 0, i32* [[DOTZERO_ADDR]], align 4 5220 // CHECK4-NEXT: store i32 [[A]], i32* [[A_ADDR]], align 4 5221 // CHECK4-NEXT: store i32 [[B]], i32* [[B_ADDR]], align 4 5222 // CHECK4-NEXT: [[CONV:%.*]] = bitcast i32* [[B_ADDR]] to i16* 5223 // CHECK4-NEXT: [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 5224 // CHECK4-NEXT: call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 1) 5225 // CHECK4-NEXT: call void @__kmpc_data_sharing_init_stack_spmd() 5226 // CHECK4-NEXT: br label [[DOTEXECUTE:%.*]] 5227 // CHECK4: .execute: 5228 // CHECK4-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3:[0-9]+]]) 5229 // CHECK4-NEXT: store i32 [[TMP0]], i32* [[DOTTHREADID_TEMP_]], align 4 5230 // CHECK4-NEXT: call void @__omp_outlined__11(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32* [[A_ADDR]], i16* [[CONV]]) #[[ATTR3]] 5231 // CHECK4-NEXT: br label [[DOTOMP_DEINIT:%.*]] 5232 // CHECK4: .omp.deinit: 5233 // CHECK4-NEXT: call void @__kmpc_spmd_kernel_deinit_v2(i16 1) 5234 // CHECK4-NEXT: br label [[DOTEXIT:%.*]] 5235 // CHECK4: .exit: 5236 // CHECK4-NEXT: ret void 5237 // 5238 // 5239 // CHECK4-LABEL: define {{[^@]+}}@__omp_outlined__11 5240 // CHECK4-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32* nonnull align 4 dereferenceable(4) [[A:%.*]], i16* nonnull align 2 dereferenceable(2) [[B:%.*]]) #[[ATTR1]] { 5241 // CHECK4-NEXT: entry: 5242 // CHECK4-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4 5243 // CHECK4-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4 5244 // CHECK4-NEXT: [[A_ADDR:%.*]] = alloca i32*, align 4 5245 // CHECK4-NEXT: [[B_ADDR:%.*]] = alloca i16*, align 4 5246 // CHECK4-NEXT: [[A1:%.*]] = alloca i32, align 4 5247 // CHECK4-NEXT: [[B2:%.*]] = alloca i16, align 2 5248 // CHECK4-NEXT: [[CAPTURED_VARS_ADDRS:%.*]] = alloca [2 x i8*], align 4 5249 // CHECK4-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 5250 // CHECK4-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4 5251 // CHECK4-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4 5252 // CHECK4-NEXT: store i32* [[A]], i32** [[A_ADDR]], align 4 5253 // CHECK4-NEXT: store i16* [[B]], i16** [[B_ADDR]], align 4 5254 // CHECK4-NEXT: [[TMP0:%.*]] = load i32*, i32** [[A_ADDR]], align 4 5255 // CHECK4-NEXT: [[TMP1:%.*]] = load i16*, i16** [[B_ADDR]], align 4 5256 // CHECK4-NEXT: store i32 0, i32* [[A1]], align 4 5257 // CHECK4-NEXT: store i16 -32768, i16* [[B2]], align 2 5258 // CHECK4-NEXT: [[TMP2:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0 5259 // CHECK4-NEXT: [[TMP3:%.*]] = bitcast i32* [[A1]] to i8* 5260 // CHECK4-NEXT: store i8* [[TMP3]], i8** [[TMP2]], align 4 5261 // CHECK4-NEXT: [[TMP4:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1 5262 // CHECK4-NEXT: [[TMP5:%.*]] = bitcast i16* [[B2]] to i8* 5263 // CHECK4-NEXT: store i8* [[TMP5]], i8** [[TMP4]], align 4 5264 // CHECK4-NEXT: [[TMP6:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4 5265 // CHECK4-NEXT: [[TMP7:%.*]] = load i32, i32* [[TMP6]], align 4 5266 // CHECK4-NEXT: [[TMP8:%.*]] = bitcast [2 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8** 5267 // CHECK4-NEXT: call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP7]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32*, i16*)* @__omp_outlined__12 to i8*), i8* null, i8** [[TMP8]], i32 2) 5268 // CHECK4-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 5269 // CHECK4-NEXT: [[TMP10:%.*]] = bitcast i32* [[A1]] to i8* 5270 // CHECK4-NEXT: store i8* [[TMP10]], i8** [[TMP9]], align 4 5271 // CHECK4-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 5272 // CHECK4-NEXT: [[TMP12:%.*]] = bitcast i16* [[B2]] to i8* 5273 // CHECK4-NEXT: store i8* [[TMP12]], i8** [[TMP11]], align 4 5274 // CHECK4-NEXT: [[TMP13:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 5275 // CHECK4-NEXT: [[TMP14:%.*]] = load i8*, i8** @"_openmp_teams_reductions_buffer_$_$ptr", align 4 5276 // CHECK4-NEXT: [[TMP15:%.*]] = call i32 @__kmpc_nvptx_teams_reduce_nowait_v2(%struct.ident_t* @[[GLOB3]], i32 [[TMP7]], i8* [[TMP14]], i32 2048, i8* [[TMP13]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func17, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func18, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_copy_func19, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_reduce_func20, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_copy_func21, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_reduce_func22) 5277 // CHECK4-NEXT: [[TMP16:%.*]] = icmp eq i32 [[TMP15]], 1 5278 // CHECK4-NEXT: br i1 [[TMP16]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]] 5279 // CHECK4: .omp.reduction.then: 5280 // CHECK4-NEXT: [[TMP17:%.*]] = load i32, i32* [[TMP0]], align 4 5281 // CHECK4-NEXT: [[TMP18:%.*]] = load i32, i32* [[A1]], align 4 5282 // CHECK4-NEXT: [[OR:%.*]] = or i32 [[TMP17]], [[TMP18]] 5283 // CHECK4-NEXT: store i32 [[OR]], i32* [[TMP0]], align 4 5284 // CHECK4-NEXT: [[TMP19:%.*]] = load i16, i16* [[TMP1]], align 2 5285 // CHECK4-NEXT: [[CONV:%.*]] = sext i16 [[TMP19]] to i32 5286 // CHECK4-NEXT: [[TMP20:%.*]] = load i16, i16* [[B2]], align 2 5287 // CHECK4-NEXT: [[CONV3:%.*]] = sext i16 [[TMP20]] to i32 5288 // CHECK4-NEXT: [[CMP:%.*]] = icmp sgt i32 [[CONV]], [[CONV3]] 5289 // CHECK4-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] 5290 // CHECK4: cond.true: 5291 // CHECK4-NEXT: [[TMP21:%.*]] = load i16, i16* [[TMP1]], align 2 5292 // CHECK4-NEXT: br label [[COND_END:%.*]] 5293 // CHECK4: cond.false: 5294 // CHECK4-NEXT: [[TMP22:%.*]] = load i16, i16* [[B2]], align 2 5295 // CHECK4-NEXT: br label [[COND_END]] 5296 // CHECK4: cond.end: 5297 // CHECK4-NEXT: [[COND:%.*]] = phi i16 [ [[TMP21]], [[COND_TRUE]] ], [ [[TMP22]], [[COND_FALSE]] ] 5298 // CHECK4-NEXT: store i16 [[COND]], i16* [[TMP1]], align 2 5299 // CHECK4-NEXT: call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP7]]) 5300 // CHECK4-NEXT: br label [[DOTOMP_REDUCTION_DONE]] 5301 // CHECK4: .omp.reduction.done: 5302 // CHECK4-NEXT: ret void 5303 // 5304 // 5305 // CHECK4-LABEL: define {{[^@]+}}@__omp_outlined__12 5306 // CHECK4-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32* nonnull align 4 dereferenceable(4) [[A:%.*]], i16* nonnull align 2 dereferenceable(2) [[B:%.*]]) #[[ATTR1]] { 5307 // CHECK4-NEXT: entry: 5308 // CHECK4-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4 5309 // CHECK4-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4 5310 // CHECK4-NEXT: [[A_ADDR:%.*]] = alloca i32*, align 4 5311 // CHECK4-NEXT: [[B_ADDR:%.*]] = alloca i16*, align 4 5312 // CHECK4-NEXT: [[A1:%.*]] = alloca i32, align 4 5313 // CHECK4-NEXT: [[B2:%.*]] = alloca i16, align 2 5314 // CHECK4-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 5315 // CHECK4-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4 5316 // CHECK4-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4 5317 // CHECK4-NEXT: store i32* [[A]], i32** [[A_ADDR]], align 4 5318 // CHECK4-NEXT: store i16* [[B]], i16** [[B_ADDR]], align 4 5319 // CHECK4-NEXT: [[TMP0:%.*]] = load i32*, i32** [[A_ADDR]], align 4 5320 // CHECK4-NEXT: [[TMP1:%.*]] = load i16*, i16** [[B_ADDR]], align 4 5321 // CHECK4-NEXT: store i32 0, i32* [[A1]], align 4 5322 // CHECK4-NEXT: store i16 -32768, i16* [[B2]], align 2 5323 // CHECK4-NEXT: [[TMP2:%.*]] = load i32, i32* [[A1]], align 4 5324 // CHECK4-NEXT: [[OR:%.*]] = or i32 [[TMP2]], 1 5325 // CHECK4-NEXT: store i32 [[OR]], i32* [[A1]], align 4 5326 // CHECK4-NEXT: [[TMP3:%.*]] = load i16, i16* [[B2]], align 2 5327 // CHECK4-NEXT: [[CONV:%.*]] = sext i16 [[TMP3]] to i32 5328 // CHECK4-NEXT: [[CMP:%.*]] = icmp sgt i32 99, [[CONV]] 5329 // CHECK4-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] 5330 // CHECK4: cond.true: 5331 // CHECK4-NEXT: br label [[COND_END:%.*]] 5332 // CHECK4: cond.false: 5333 // CHECK4-NEXT: [[TMP4:%.*]] = load i16, i16* [[B2]], align 2 5334 // CHECK4-NEXT: [[CONV3:%.*]] = sext i16 [[TMP4]] to i32 5335 // CHECK4-NEXT: br label [[COND_END]] 5336 // CHECK4: cond.end: 5337 // CHECK4-NEXT: [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[CONV3]], [[COND_FALSE]] ] 5338 // CHECK4-NEXT: [[CONV4:%.*]] = trunc i32 [[COND]] to i16 5339 // CHECK4-NEXT: store i16 [[CONV4]], i16* [[B2]], align 2 5340 // CHECK4-NEXT: [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4 5341 // CHECK4-NEXT: [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4 5342 // CHECK4-NEXT: [[TMP7:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 5343 // CHECK4-NEXT: [[TMP8:%.*]] = bitcast i32* [[A1]] to i8* 5344 // CHECK4-NEXT: store i8* [[TMP8]], i8** [[TMP7]], align 4 5345 // CHECK4-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 5346 // CHECK4-NEXT: [[TMP10:%.*]] = bitcast i16* [[B2]] to i8* 5347 // CHECK4-NEXT: store i8* [[TMP10]], i8** [[TMP9]], align 4 5348 // CHECK4-NEXT: [[TMP11:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 5349 // CHECK4-NEXT: [[TMP12:%.*]] = call i32 @__kmpc_nvptx_parallel_reduce_nowait_v2(%struct.ident_t* @[[GLOB3]], i32 [[TMP6]], i32 2, i32 8, i8* [[TMP11]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func14, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func15) 5350 // CHECK4-NEXT: [[TMP13:%.*]] = icmp eq i32 [[TMP12]], 1 5351 // CHECK4-NEXT: br i1 [[TMP13]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]] 5352 // CHECK4: .omp.reduction.then: 5353 // CHECK4-NEXT: [[TMP14:%.*]] = load i32, i32* [[TMP0]], align 4 5354 // CHECK4-NEXT: [[TMP15:%.*]] = load i32, i32* [[A1]], align 4 5355 // CHECK4-NEXT: [[OR5:%.*]] = or i32 [[TMP14]], [[TMP15]] 5356 // CHECK4-NEXT: store i32 [[OR5]], i32* [[TMP0]], align 4 5357 // CHECK4-NEXT: [[TMP16:%.*]] = load i16, i16* [[TMP1]], align 2 5358 // CHECK4-NEXT: [[CONV6:%.*]] = sext i16 [[TMP16]] to i32 5359 // CHECK4-NEXT: [[TMP17:%.*]] = load i16, i16* [[B2]], align 2 5360 // CHECK4-NEXT: [[CONV7:%.*]] = sext i16 [[TMP17]] to i32 5361 // CHECK4-NEXT: [[CMP8:%.*]] = icmp sgt i32 [[CONV6]], [[CONV7]] 5362 // CHECK4-NEXT: br i1 [[CMP8]], label [[COND_TRUE9:%.*]], label [[COND_FALSE10:%.*]] 5363 // CHECK4: cond.true9: 5364 // CHECK4-NEXT: [[TMP18:%.*]] = load i16, i16* [[TMP1]], align 2 5365 // CHECK4-NEXT: br label [[COND_END11:%.*]] 5366 // CHECK4: cond.false10: 5367 // CHECK4-NEXT: [[TMP19:%.*]] = load i16, i16* [[B2]], align 2 5368 // CHECK4-NEXT: br label [[COND_END11]] 5369 // CHECK4: cond.end11: 5370 // CHECK4-NEXT: [[COND12:%.*]] = phi i16 [ [[TMP18]], [[COND_TRUE9]] ], [ [[TMP19]], [[COND_FALSE10]] ] 5371 // CHECK4-NEXT: store i16 [[COND12]], i16* [[TMP1]], align 2 5372 // CHECK4-NEXT: call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP6]]) 5373 // CHECK4-NEXT: br label [[DOTOMP_REDUCTION_DONE]] 5374 // CHECK4: .omp.reduction.done: 5375 // CHECK4-NEXT: ret void 5376 // 5377 // 5378 // CHECK4-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func14 5379 // CHECK4-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR0]] { 5380 // CHECK4-NEXT: entry: 5381 // CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 5382 // CHECK4-NEXT: [[DOTADDR1:%.*]] = alloca i16, align 2 5383 // CHECK4-NEXT: [[DOTADDR2:%.*]] = alloca i16, align 2 5384 // CHECK4-NEXT: [[DOTADDR3:%.*]] = alloca i16, align 2 5385 // CHECK4-NEXT: [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [2 x i8*], align 4 5386 // CHECK4-NEXT: [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca i32, align 4 5387 // CHECK4-NEXT: [[DOTOMP_REDUCTION_ELEMENT4:%.*]] = alloca i16, align 2 5388 // CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 5389 // CHECK4-NEXT: store i16 [[TMP1]], i16* [[DOTADDR1]], align 2 5390 // CHECK4-NEXT: store i16 [[TMP2]], i16* [[DOTADDR2]], align 2 5391 // CHECK4-NEXT: store i16 [[TMP3]], i16* [[DOTADDR3]], align 2 5392 // CHECK4-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 4 5393 // CHECK4-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [2 x i8*]* 5394 // CHECK4-NEXT: [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2 5395 // CHECK4-NEXT: [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2 5396 // CHECK4-NEXT: [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2 5397 // CHECK4-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 0 5398 // CHECK4-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 4 5399 // CHECK4-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 5400 // CHECK4-NEXT: [[TMP12:%.*]] = bitcast i8* [[TMP10]] to i32* 5401 // CHECK4-NEXT: [[TMP13:%.*]] = getelementptr i32, i32* [[TMP12]], i32 1 5402 // CHECK4-NEXT: [[TMP14:%.*]] = bitcast i32* [[TMP13]] to i8* 5403 // CHECK4-NEXT: [[TMP15:%.*]] = load i32, i32* [[TMP12]], align 4 5404 // CHECK4-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 5405 // CHECK4-NEXT: [[TMP16:%.*]] = trunc i32 [[NVPTX_WARP_SIZE]] to i16 5406 // CHECK4-NEXT: [[TMP17:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP15]], i16 [[TMP7]], i16 [[TMP16]]) 5407 // CHECK4-NEXT: store i32 [[TMP17]], i32* [[DOTOMP_REDUCTION_ELEMENT]], align 4 5408 // CHECK4-NEXT: [[TMP18:%.*]] = getelementptr i32, i32* [[TMP12]], i32 1 5409 // CHECK4-NEXT: [[TMP19:%.*]] = getelementptr i32, i32* [[DOTOMP_REDUCTION_ELEMENT]], i32 1 5410 // CHECK4-NEXT: [[TMP20:%.*]] = bitcast i32* [[DOTOMP_REDUCTION_ELEMENT]] to i8* 5411 // CHECK4-NEXT: store i8* [[TMP20]], i8** [[TMP11]], align 4 5412 // CHECK4-NEXT: [[TMP21:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 1 5413 // CHECK4-NEXT: [[TMP22:%.*]] = load i8*, i8** [[TMP21]], align 4 5414 // CHECK4-NEXT: [[TMP23:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1 5415 // CHECK4-NEXT: [[TMP24:%.*]] = bitcast i8* [[TMP22]] to i16* 5416 // CHECK4-NEXT: [[TMP25:%.*]] = getelementptr i16, i16* [[TMP24]], i32 1 5417 // CHECK4-NEXT: [[TMP26:%.*]] = bitcast i16* [[TMP25]] to i8* 5418 // CHECK4-NEXT: [[TMP27:%.*]] = load i16, i16* [[TMP24]], align 2 5419 // CHECK4-NEXT: [[TMP28:%.*]] = sext i16 [[TMP27]] to i32 5420 // CHECK4-NEXT: [[NVPTX_WARP_SIZE5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 5421 // CHECK4-NEXT: [[TMP29:%.*]] = trunc i32 [[NVPTX_WARP_SIZE5]] to i16 5422 // CHECK4-NEXT: [[TMP30:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP28]], i16 [[TMP7]], i16 [[TMP29]]) 5423 // CHECK4-NEXT: [[TMP31:%.*]] = trunc i32 [[TMP30]] to i16 5424 // CHECK4-NEXT: store i16 [[TMP31]], i16* [[DOTOMP_REDUCTION_ELEMENT4]], align 2 5425 // CHECK4-NEXT: [[TMP32:%.*]] = getelementptr i16, i16* [[TMP24]], i32 1 5426 // CHECK4-NEXT: [[TMP33:%.*]] = getelementptr i16, i16* [[DOTOMP_REDUCTION_ELEMENT4]], i32 1 5427 // CHECK4-NEXT: [[TMP34:%.*]] = bitcast i16* [[DOTOMP_REDUCTION_ELEMENT4]] to i8* 5428 // CHECK4-NEXT: store i8* [[TMP34]], i8** [[TMP23]], align 4 5429 // CHECK4-NEXT: [[TMP35:%.*]] = icmp eq i16 [[TMP8]], 0 5430 // CHECK4-NEXT: [[TMP36:%.*]] = icmp eq i16 [[TMP8]], 1 5431 // CHECK4-NEXT: [[TMP37:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]] 5432 // CHECK4-NEXT: [[TMP38:%.*]] = and i1 [[TMP36]], [[TMP37]] 5433 // CHECK4-NEXT: [[TMP39:%.*]] = icmp eq i16 [[TMP8]], 2 5434 // CHECK4-NEXT: [[TMP40:%.*]] = and i16 [[TMP6]], 1 5435 // CHECK4-NEXT: [[TMP41:%.*]] = icmp eq i16 [[TMP40]], 0 5436 // CHECK4-NEXT: [[TMP42:%.*]] = and i1 [[TMP39]], [[TMP41]] 5437 // CHECK4-NEXT: [[TMP43:%.*]] = icmp sgt i16 [[TMP7]], 0 5438 // CHECK4-NEXT: [[TMP44:%.*]] = and i1 [[TMP42]], [[TMP43]] 5439 // CHECK4-NEXT: [[TMP45:%.*]] = or i1 [[TMP35]], [[TMP38]] 5440 // CHECK4-NEXT: [[TMP46:%.*]] = or i1 [[TMP45]], [[TMP44]] 5441 // CHECK4-NEXT: br i1 [[TMP46]], label [[THEN:%.*]], label [[ELSE:%.*]] 5442 // CHECK4: then: 5443 // CHECK4-NEXT: [[TMP47:%.*]] = bitcast [2 x i8*]* [[TMP5]] to i8* 5444 // CHECK4-NEXT: [[TMP48:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8* 5445 // CHECK4-NEXT: call void @"_omp$reduction$reduction_func13"(i8* [[TMP47]], i8* [[TMP48]]) #[[ATTR3]] 5446 // CHECK4-NEXT: br label [[IFCONT:%.*]] 5447 // CHECK4: else: 5448 // CHECK4-NEXT: br label [[IFCONT]] 5449 // CHECK4: ifcont: 5450 // CHECK4-NEXT: [[TMP49:%.*]] = icmp eq i16 [[TMP8]], 1 5451 // CHECK4-NEXT: [[TMP50:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]] 5452 // CHECK4-NEXT: [[TMP51:%.*]] = and i1 [[TMP49]], [[TMP50]] 5453 // CHECK4-NEXT: br i1 [[TMP51]], label [[THEN6:%.*]], label [[ELSE7:%.*]] 5454 // CHECK4: then6: 5455 // CHECK4-NEXT: [[TMP52:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 5456 // CHECK4-NEXT: [[TMP53:%.*]] = load i8*, i8** [[TMP52]], align 4 5457 // CHECK4-NEXT: [[TMP54:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 0 5458 // CHECK4-NEXT: [[TMP55:%.*]] = load i8*, i8** [[TMP54]], align 4 5459 // CHECK4-NEXT: [[TMP56:%.*]] = bitcast i8* [[TMP53]] to i32* 5460 // CHECK4-NEXT: [[TMP57:%.*]] = bitcast i8* [[TMP55]] to i32* 5461 // CHECK4-NEXT: [[TMP58:%.*]] = load i32, i32* [[TMP56]], align 4 5462 // CHECK4-NEXT: store i32 [[TMP58]], i32* [[TMP57]], align 4 5463 // CHECK4-NEXT: [[TMP59:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1 5464 // CHECK4-NEXT: [[TMP60:%.*]] = load i8*, i8** [[TMP59]], align 4 5465 // CHECK4-NEXT: [[TMP61:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 1 5466 // CHECK4-NEXT: [[TMP62:%.*]] = load i8*, i8** [[TMP61]], align 4 5467 // CHECK4-NEXT: [[TMP63:%.*]] = bitcast i8* [[TMP60]] to i16* 5468 // CHECK4-NEXT: [[TMP64:%.*]] = bitcast i8* [[TMP62]] to i16* 5469 // CHECK4-NEXT: [[TMP65:%.*]] = load i16, i16* [[TMP63]], align 2 5470 // CHECK4-NEXT: store i16 [[TMP65]], i16* [[TMP64]], align 2 5471 // CHECK4-NEXT: br label [[IFCONT8:%.*]] 5472 // CHECK4: else7: 5473 // CHECK4-NEXT: br label [[IFCONT8]] 5474 // CHECK4: ifcont8: 5475 // CHECK4-NEXT: ret void 5476 // 5477 // 5478 // CHECK4-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func15 5479 // CHECK4-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] { 5480 // CHECK4-NEXT: entry: 5481 // CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 5482 // CHECK4-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 5483 // CHECK4-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]]) 5484 // CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 5485 // CHECK4-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 5486 // CHECK4-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 5487 // CHECK4-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 5488 // CHECK4-NEXT: [[NVPTX_LANE_ID:%.*]] = and i32 [[NVPTX_TID2]], 31 5489 // CHECK4-NEXT: [[NVPTX_TID3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 5490 // CHECK4-NEXT: [[NVPTX_WARP_ID:%.*]] = ashr i32 [[NVPTX_TID3]], 5 5491 // CHECK4-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 5492 // CHECK4-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 5493 // CHECK4-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4:[0-9]+]], i32 [[TMP2]]) 5494 // CHECK4-NEXT: [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 5495 // CHECK4-NEXT: br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]] 5496 // CHECK4: then: 5497 // CHECK4-NEXT: [[TMP5:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 5498 // CHECK4-NEXT: [[TMP6:%.*]] = load i8*, i8** [[TMP5]], align 4 5499 // CHECK4-NEXT: [[TMP7:%.*]] = bitcast i8* [[TMP6]] to i32* 5500 // CHECK4-NEXT: [[TMP8:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 5501 // CHECK4-NEXT: [[TMP9:%.*]] = load i32, i32* [[TMP7]], align 4 5502 // CHECK4-NEXT: store volatile i32 [[TMP9]], i32 addrspace(3)* [[TMP8]], align 4 5503 // CHECK4-NEXT: br label [[IFCONT:%.*]] 5504 // CHECK4: else: 5505 // CHECK4-NEXT: br label [[IFCONT]] 5506 // CHECK4: ifcont: 5507 // CHECK4-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 5508 // CHECK4-NEXT: [[TMP10:%.*]] = load i32, i32* [[DOTADDR1]], align 4 5509 // CHECK4-NEXT: [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP10]] 5510 // CHECK4-NEXT: br i1 [[IS_ACTIVE_THREAD]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 5511 // CHECK4: then4: 5512 // CHECK4-NEXT: [[TMP11:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 5513 // CHECK4-NEXT: [[TMP12:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 5514 // CHECK4-NEXT: [[TMP13:%.*]] = load i8*, i8** [[TMP12]], align 4 5515 // CHECK4-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to i32* 5516 // CHECK4-NEXT: [[TMP15:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP11]], align 4 5517 // CHECK4-NEXT: store i32 [[TMP15]], i32* [[TMP14]], align 4 5518 // CHECK4-NEXT: br label [[IFCONT6:%.*]] 5519 // CHECK4: else5: 5520 // CHECK4-NEXT: br label [[IFCONT6]] 5521 // CHECK4: ifcont6: 5522 // CHECK4-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 5523 // CHECK4-NEXT: [[WARP_MASTER7:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 5524 // CHECK4-NEXT: br i1 [[WARP_MASTER7]], label [[THEN8:%.*]], label [[ELSE9:%.*]] 5525 // CHECK4: then8: 5526 // CHECK4-NEXT: [[TMP16:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 5527 // CHECK4-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 4 5528 // CHECK4-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i16* 5529 // CHECK4-NEXT: [[TMP19:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 5530 // CHECK4-NEXT: [[TMP20:%.*]] = bitcast i32 addrspace(3)* [[TMP19]] to i16 addrspace(3)* 5531 // CHECK4-NEXT: [[TMP21:%.*]] = load i16, i16* [[TMP18]], align 2 5532 // CHECK4-NEXT: store volatile i16 [[TMP21]], i16 addrspace(3)* [[TMP20]], align 2 5533 // CHECK4-NEXT: br label [[IFCONT10:%.*]] 5534 // CHECK4: else9: 5535 // CHECK4-NEXT: br label [[IFCONT10]] 5536 // CHECK4: ifcont10: 5537 // CHECK4-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 5538 // CHECK4-NEXT: [[TMP22:%.*]] = load i32, i32* [[DOTADDR1]], align 4 5539 // CHECK4-NEXT: [[IS_ACTIVE_THREAD11:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP22]] 5540 // CHECK4-NEXT: br i1 [[IS_ACTIVE_THREAD11]], label [[THEN12:%.*]], label [[ELSE13:%.*]] 5541 // CHECK4: then12: 5542 // CHECK4-NEXT: [[TMP23:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 5543 // CHECK4-NEXT: [[TMP24:%.*]] = bitcast i32 addrspace(3)* [[TMP23]] to i16 addrspace(3)* 5544 // CHECK4-NEXT: [[TMP25:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 5545 // CHECK4-NEXT: [[TMP26:%.*]] = load i8*, i8** [[TMP25]], align 4 5546 // CHECK4-NEXT: [[TMP27:%.*]] = bitcast i8* [[TMP26]] to i16* 5547 // CHECK4-NEXT: [[TMP28:%.*]] = load volatile i16, i16 addrspace(3)* [[TMP24]], align 2 5548 // CHECK4-NEXT: store i16 [[TMP28]], i16* [[TMP27]], align 2 5549 // CHECK4-NEXT: br label [[IFCONT14:%.*]] 5550 // CHECK4: else13: 5551 // CHECK4-NEXT: br label [[IFCONT14]] 5552 // CHECK4: ifcont14: 5553 // CHECK4-NEXT: ret void 5554 // 5555 // 5556 // CHECK4-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func17 5557 // CHECK4-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR0]] { 5558 // CHECK4-NEXT: entry: 5559 // CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 5560 // CHECK4-NEXT: [[DOTADDR1:%.*]] = alloca i16, align 2 5561 // CHECK4-NEXT: [[DOTADDR2:%.*]] = alloca i16, align 2 5562 // CHECK4-NEXT: [[DOTADDR3:%.*]] = alloca i16, align 2 5563 // CHECK4-NEXT: [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [2 x i8*], align 4 5564 // CHECK4-NEXT: [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca i32, align 4 5565 // CHECK4-NEXT: [[DOTOMP_REDUCTION_ELEMENT4:%.*]] = alloca i16, align 2 5566 // CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 5567 // CHECK4-NEXT: store i16 [[TMP1]], i16* [[DOTADDR1]], align 2 5568 // CHECK4-NEXT: store i16 [[TMP2]], i16* [[DOTADDR2]], align 2 5569 // CHECK4-NEXT: store i16 [[TMP3]], i16* [[DOTADDR3]], align 2 5570 // CHECK4-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 4 5571 // CHECK4-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [2 x i8*]* 5572 // CHECK4-NEXT: [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2 5573 // CHECK4-NEXT: [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2 5574 // CHECK4-NEXT: [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2 5575 // CHECK4-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 0 5576 // CHECK4-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 4 5577 // CHECK4-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 5578 // CHECK4-NEXT: [[TMP12:%.*]] = bitcast i8* [[TMP10]] to i32* 5579 // CHECK4-NEXT: [[TMP13:%.*]] = getelementptr i32, i32* [[TMP12]], i32 1 5580 // CHECK4-NEXT: [[TMP14:%.*]] = bitcast i32* [[TMP13]] to i8* 5581 // CHECK4-NEXT: [[TMP15:%.*]] = load i32, i32* [[TMP12]], align 4 5582 // CHECK4-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 5583 // CHECK4-NEXT: [[TMP16:%.*]] = trunc i32 [[NVPTX_WARP_SIZE]] to i16 5584 // CHECK4-NEXT: [[TMP17:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP15]], i16 [[TMP7]], i16 [[TMP16]]) 5585 // CHECK4-NEXT: store i32 [[TMP17]], i32* [[DOTOMP_REDUCTION_ELEMENT]], align 4 5586 // CHECK4-NEXT: [[TMP18:%.*]] = getelementptr i32, i32* [[TMP12]], i32 1 5587 // CHECK4-NEXT: [[TMP19:%.*]] = getelementptr i32, i32* [[DOTOMP_REDUCTION_ELEMENT]], i32 1 5588 // CHECK4-NEXT: [[TMP20:%.*]] = bitcast i32* [[DOTOMP_REDUCTION_ELEMENT]] to i8* 5589 // CHECK4-NEXT: store i8* [[TMP20]], i8** [[TMP11]], align 4 5590 // CHECK4-NEXT: [[TMP21:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 1 5591 // CHECK4-NEXT: [[TMP22:%.*]] = load i8*, i8** [[TMP21]], align 4 5592 // CHECK4-NEXT: [[TMP23:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1 5593 // CHECK4-NEXT: [[TMP24:%.*]] = bitcast i8* [[TMP22]] to i16* 5594 // CHECK4-NEXT: [[TMP25:%.*]] = getelementptr i16, i16* [[TMP24]], i32 1 5595 // CHECK4-NEXT: [[TMP26:%.*]] = bitcast i16* [[TMP25]] to i8* 5596 // CHECK4-NEXT: [[TMP27:%.*]] = load i16, i16* [[TMP24]], align 2 5597 // CHECK4-NEXT: [[TMP28:%.*]] = sext i16 [[TMP27]] to i32 5598 // CHECK4-NEXT: [[NVPTX_WARP_SIZE5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 5599 // CHECK4-NEXT: [[TMP29:%.*]] = trunc i32 [[NVPTX_WARP_SIZE5]] to i16 5600 // CHECK4-NEXT: [[TMP30:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP28]], i16 [[TMP7]], i16 [[TMP29]]) 5601 // CHECK4-NEXT: [[TMP31:%.*]] = trunc i32 [[TMP30]] to i16 5602 // CHECK4-NEXT: store i16 [[TMP31]], i16* [[DOTOMP_REDUCTION_ELEMENT4]], align 2 5603 // CHECK4-NEXT: [[TMP32:%.*]] = getelementptr i16, i16* [[TMP24]], i32 1 5604 // CHECK4-NEXT: [[TMP33:%.*]] = getelementptr i16, i16* [[DOTOMP_REDUCTION_ELEMENT4]], i32 1 5605 // CHECK4-NEXT: [[TMP34:%.*]] = bitcast i16* [[DOTOMP_REDUCTION_ELEMENT4]] to i8* 5606 // CHECK4-NEXT: store i8* [[TMP34]], i8** [[TMP23]], align 4 5607 // CHECK4-NEXT: [[TMP35:%.*]] = icmp eq i16 [[TMP8]], 0 5608 // CHECK4-NEXT: [[TMP36:%.*]] = icmp eq i16 [[TMP8]], 1 5609 // CHECK4-NEXT: [[TMP37:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]] 5610 // CHECK4-NEXT: [[TMP38:%.*]] = and i1 [[TMP36]], [[TMP37]] 5611 // CHECK4-NEXT: [[TMP39:%.*]] = icmp eq i16 [[TMP8]], 2 5612 // CHECK4-NEXT: [[TMP40:%.*]] = and i16 [[TMP6]], 1 5613 // CHECK4-NEXT: [[TMP41:%.*]] = icmp eq i16 [[TMP40]], 0 5614 // CHECK4-NEXT: [[TMP42:%.*]] = and i1 [[TMP39]], [[TMP41]] 5615 // CHECK4-NEXT: [[TMP43:%.*]] = icmp sgt i16 [[TMP7]], 0 5616 // CHECK4-NEXT: [[TMP44:%.*]] = and i1 [[TMP42]], [[TMP43]] 5617 // CHECK4-NEXT: [[TMP45:%.*]] = or i1 [[TMP35]], [[TMP38]] 5618 // CHECK4-NEXT: [[TMP46:%.*]] = or i1 [[TMP45]], [[TMP44]] 5619 // CHECK4-NEXT: br i1 [[TMP46]], label [[THEN:%.*]], label [[ELSE:%.*]] 5620 // CHECK4: then: 5621 // CHECK4-NEXT: [[TMP47:%.*]] = bitcast [2 x i8*]* [[TMP5]] to i8* 5622 // CHECK4-NEXT: [[TMP48:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8* 5623 // CHECK4-NEXT: call void @"_omp$reduction$reduction_func16"(i8* [[TMP47]], i8* [[TMP48]]) #[[ATTR3]] 5624 // CHECK4-NEXT: br label [[IFCONT:%.*]] 5625 // CHECK4: else: 5626 // CHECK4-NEXT: br label [[IFCONT]] 5627 // CHECK4: ifcont: 5628 // CHECK4-NEXT: [[TMP49:%.*]] = icmp eq i16 [[TMP8]], 1 5629 // CHECK4-NEXT: [[TMP50:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]] 5630 // CHECK4-NEXT: [[TMP51:%.*]] = and i1 [[TMP49]], [[TMP50]] 5631 // CHECK4-NEXT: br i1 [[TMP51]], label [[THEN6:%.*]], label [[ELSE7:%.*]] 5632 // CHECK4: then6: 5633 // CHECK4-NEXT: [[TMP52:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 5634 // CHECK4-NEXT: [[TMP53:%.*]] = load i8*, i8** [[TMP52]], align 4 5635 // CHECK4-NEXT: [[TMP54:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 0 5636 // CHECK4-NEXT: [[TMP55:%.*]] = load i8*, i8** [[TMP54]], align 4 5637 // CHECK4-NEXT: [[TMP56:%.*]] = bitcast i8* [[TMP53]] to i32* 5638 // CHECK4-NEXT: [[TMP57:%.*]] = bitcast i8* [[TMP55]] to i32* 5639 // CHECK4-NEXT: [[TMP58:%.*]] = load i32, i32* [[TMP56]], align 4 5640 // CHECK4-NEXT: store i32 [[TMP58]], i32* [[TMP57]], align 4 5641 // CHECK4-NEXT: [[TMP59:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1 5642 // CHECK4-NEXT: [[TMP60:%.*]] = load i8*, i8** [[TMP59]], align 4 5643 // CHECK4-NEXT: [[TMP61:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 1 5644 // CHECK4-NEXT: [[TMP62:%.*]] = load i8*, i8** [[TMP61]], align 4 5645 // CHECK4-NEXT: [[TMP63:%.*]] = bitcast i8* [[TMP60]] to i16* 5646 // CHECK4-NEXT: [[TMP64:%.*]] = bitcast i8* [[TMP62]] to i16* 5647 // CHECK4-NEXT: [[TMP65:%.*]] = load i16, i16* [[TMP63]], align 2 5648 // CHECK4-NEXT: store i16 [[TMP65]], i16* [[TMP64]], align 2 5649 // CHECK4-NEXT: br label [[IFCONT8:%.*]] 5650 // CHECK4: else7: 5651 // CHECK4-NEXT: br label [[IFCONT8]] 5652 // CHECK4: ifcont8: 5653 // CHECK4-NEXT: ret void 5654 // 5655 // 5656 // CHECK4-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func18 5657 // CHECK4-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] { 5658 // CHECK4-NEXT: entry: 5659 // CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 5660 // CHECK4-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 5661 // CHECK4-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]]) 5662 // CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 5663 // CHECK4-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 5664 // CHECK4-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 5665 // CHECK4-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 5666 // CHECK4-NEXT: [[NVPTX_LANE_ID:%.*]] = and i32 [[NVPTX_TID2]], 31 5667 // CHECK4-NEXT: [[NVPTX_TID3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 5668 // CHECK4-NEXT: [[NVPTX_WARP_ID:%.*]] = ashr i32 [[NVPTX_TID3]], 5 5669 // CHECK4-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 5670 // CHECK4-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 5671 // CHECK4-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 5672 // CHECK4-NEXT: [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 5673 // CHECK4-NEXT: br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]] 5674 // CHECK4: then: 5675 // CHECK4-NEXT: [[TMP5:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 5676 // CHECK4-NEXT: [[TMP6:%.*]] = load i8*, i8** [[TMP5]], align 4 5677 // CHECK4-NEXT: [[TMP7:%.*]] = bitcast i8* [[TMP6]] to i32* 5678 // CHECK4-NEXT: [[TMP8:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 5679 // CHECK4-NEXT: [[TMP9:%.*]] = load i32, i32* [[TMP7]], align 4 5680 // CHECK4-NEXT: store volatile i32 [[TMP9]], i32 addrspace(3)* [[TMP8]], align 4 5681 // CHECK4-NEXT: br label [[IFCONT:%.*]] 5682 // CHECK4: else: 5683 // CHECK4-NEXT: br label [[IFCONT]] 5684 // CHECK4: ifcont: 5685 // CHECK4-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 5686 // CHECK4-NEXT: [[TMP10:%.*]] = load i32, i32* [[DOTADDR1]], align 4 5687 // CHECK4-NEXT: [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP10]] 5688 // CHECK4-NEXT: br i1 [[IS_ACTIVE_THREAD]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 5689 // CHECK4: then4: 5690 // CHECK4-NEXT: [[TMP11:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 5691 // CHECK4-NEXT: [[TMP12:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 5692 // CHECK4-NEXT: [[TMP13:%.*]] = load i8*, i8** [[TMP12]], align 4 5693 // CHECK4-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to i32* 5694 // CHECK4-NEXT: [[TMP15:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP11]], align 4 5695 // CHECK4-NEXT: store i32 [[TMP15]], i32* [[TMP14]], align 4 5696 // CHECK4-NEXT: br label [[IFCONT6:%.*]] 5697 // CHECK4: else5: 5698 // CHECK4-NEXT: br label [[IFCONT6]] 5699 // CHECK4: ifcont6: 5700 // CHECK4-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 5701 // CHECK4-NEXT: [[WARP_MASTER7:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 5702 // CHECK4-NEXT: br i1 [[WARP_MASTER7]], label [[THEN8:%.*]], label [[ELSE9:%.*]] 5703 // CHECK4: then8: 5704 // CHECK4-NEXT: [[TMP16:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 5705 // CHECK4-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 4 5706 // CHECK4-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i16* 5707 // CHECK4-NEXT: [[TMP19:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 5708 // CHECK4-NEXT: [[TMP20:%.*]] = bitcast i32 addrspace(3)* [[TMP19]] to i16 addrspace(3)* 5709 // CHECK4-NEXT: [[TMP21:%.*]] = load i16, i16* [[TMP18]], align 2 5710 // CHECK4-NEXT: store volatile i16 [[TMP21]], i16 addrspace(3)* [[TMP20]], align 2 5711 // CHECK4-NEXT: br label [[IFCONT10:%.*]] 5712 // CHECK4: else9: 5713 // CHECK4-NEXT: br label [[IFCONT10]] 5714 // CHECK4: ifcont10: 5715 // CHECK4-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 5716 // CHECK4-NEXT: [[TMP22:%.*]] = load i32, i32* [[DOTADDR1]], align 4 5717 // CHECK4-NEXT: [[IS_ACTIVE_THREAD11:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP22]] 5718 // CHECK4-NEXT: br i1 [[IS_ACTIVE_THREAD11]], label [[THEN12:%.*]], label [[ELSE13:%.*]] 5719 // CHECK4: then12: 5720 // CHECK4-NEXT: [[TMP23:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 5721 // CHECK4-NEXT: [[TMP24:%.*]] = bitcast i32 addrspace(3)* [[TMP23]] to i16 addrspace(3)* 5722 // CHECK4-NEXT: [[TMP25:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 5723 // CHECK4-NEXT: [[TMP26:%.*]] = load i8*, i8** [[TMP25]], align 4 5724 // CHECK4-NEXT: [[TMP27:%.*]] = bitcast i8* [[TMP26]] to i16* 5725 // CHECK4-NEXT: [[TMP28:%.*]] = load volatile i16, i16 addrspace(3)* [[TMP24]], align 2 5726 // CHECK4-NEXT: store i16 [[TMP28]], i16* [[TMP27]], align 2 5727 // CHECK4-NEXT: br label [[IFCONT14:%.*]] 5728 // CHECK4: else13: 5729 // CHECK4-NEXT: br label [[IFCONT14]] 5730 // CHECK4: ifcont14: 5731 // CHECK4-NEXT: ret void 5732 // 5733 // 5734 // CHECK4-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_copy_func19 5735 // CHECK4-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 5736 // CHECK4-NEXT: entry: 5737 // CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 5738 // CHECK4-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 5739 // CHECK4-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 5740 // CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 5741 // CHECK4-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 5742 // CHECK4-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 5743 // CHECK4-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 5744 // CHECK4-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 5745 // CHECK4-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 4 5746 // CHECK4-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.4* 5747 // CHECK4-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 5748 // CHECK4-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 5749 // CHECK4-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 4 5750 // CHECK4-NEXT: [[TMP10:%.*]] = bitcast i8* [[TMP9]] to i32* 5751 // CHECK4-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4:%.*]], %struct._globalized_locals_ty.4* [[TMP6]], i32 0, i32 0 5752 // CHECK4-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2048 x i32], [2048 x i32]* [[A]], i32 0, i32 [[TMP7]] 5753 // CHECK4-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP10]], align 4 5754 // CHECK4-NEXT: store i32 [[TMP12]], i32* [[TMP11]], align 128 5755 // CHECK4-NEXT: [[TMP13:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 5756 // CHECK4-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 4 5757 // CHECK4-NEXT: [[TMP15:%.*]] = bitcast i8* [[TMP14]] to i16* 5758 // CHECK4-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4]], %struct._globalized_locals_ty.4* [[TMP6]], i32 0, i32 1 5759 // CHECK4-NEXT: [[TMP16:%.*]] = getelementptr inbounds [2048 x i16], [2048 x i16]* [[B]], i32 0, i32 [[TMP7]] 5760 // CHECK4-NEXT: [[TMP17:%.*]] = load i16, i16* [[TMP15]], align 2 5761 // CHECK4-NEXT: store i16 [[TMP17]], i16* [[TMP16]], align 128 5762 // CHECK4-NEXT: ret void 5763 // 5764 // 5765 // CHECK4-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_reduce_func20 5766 // CHECK4-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 5767 // CHECK4-NEXT: entry: 5768 // CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 5769 // CHECK4-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 5770 // CHECK4-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 5771 // CHECK4-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 5772 // CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 5773 // CHECK4-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 5774 // CHECK4-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 5775 // CHECK4-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 5776 // CHECK4-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.4* 5777 // CHECK4-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 5778 // CHECK4-NEXT: [[TMP6:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 5779 // CHECK4-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4:%.*]], %struct._globalized_locals_ty.4* [[TMP4]], i32 0, i32 0 5780 // CHECK4-NEXT: [[TMP7:%.*]] = getelementptr inbounds [2048 x i32], [2048 x i32]* [[A]], i32 0, i32 [[TMP5]] 5781 // CHECK4-NEXT: [[TMP8:%.*]] = bitcast i32* [[TMP7]] to i8* 5782 // CHECK4-NEXT: store i8* [[TMP8]], i8** [[TMP6]], align 4 5783 // CHECK4-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 5784 // CHECK4-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4]], %struct._globalized_locals_ty.4* [[TMP4]], i32 0, i32 1 5785 // CHECK4-NEXT: [[TMP10:%.*]] = getelementptr inbounds [2048 x i16], [2048 x i16]* [[B]], i32 0, i32 [[TMP5]] 5786 // CHECK4-NEXT: [[TMP11:%.*]] = bitcast i16* [[TMP10]] to i8* 5787 // CHECK4-NEXT: store i8* [[TMP11]], i8** [[TMP9]], align 4 5788 // CHECK4-NEXT: [[TMP12:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 5789 // CHECK4-NEXT: [[TMP13:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 5790 // CHECK4-NEXT: call void @"_omp$reduction$reduction_func16"(i8* [[TMP12]], i8* [[TMP13]]) #[[ATTR3]] 5791 // CHECK4-NEXT: ret void 5792 // 5793 // 5794 // CHECK4-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_copy_func21 5795 // CHECK4-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 5796 // CHECK4-NEXT: entry: 5797 // CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 5798 // CHECK4-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 5799 // CHECK4-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 5800 // CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 5801 // CHECK4-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 5802 // CHECK4-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 5803 // CHECK4-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 5804 // CHECK4-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 5805 // CHECK4-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 4 5806 // CHECK4-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.4* 5807 // CHECK4-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 5808 // CHECK4-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 5809 // CHECK4-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 4 5810 // CHECK4-NEXT: [[TMP10:%.*]] = bitcast i8* [[TMP9]] to i32* 5811 // CHECK4-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4:%.*]], %struct._globalized_locals_ty.4* [[TMP6]], i32 0, i32 0 5812 // CHECK4-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2048 x i32], [2048 x i32]* [[A]], i32 0, i32 [[TMP7]] 5813 // CHECK4-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 128 5814 // CHECK4-NEXT: store i32 [[TMP12]], i32* [[TMP10]], align 4 5815 // CHECK4-NEXT: [[TMP13:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 5816 // CHECK4-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 4 5817 // CHECK4-NEXT: [[TMP15:%.*]] = bitcast i8* [[TMP14]] to i16* 5818 // CHECK4-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4]], %struct._globalized_locals_ty.4* [[TMP6]], i32 0, i32 1 5819 // CHECK4-NEXT: [[TMP16:%.*]] = getelementptr inbounds [2048 x i16], [2048 x i16]* [[B]], i32 0, i32 [[TMP7]] 5820 // CHECK4-NEXT: [[TMP17:%.*]] = load i16, i16* [[TMP16]], align 128 5821 // CHECK4-NEXT: store i16 [[TMP17]], i16* [[TMP15]], align 2 5822 // CHECK4-NEXT: ret void 5823 // 5824 // 5825 // CHECK4-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_reduce_func22 5826 // CHECK4-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 5827 // CHECK4-NEXT: entry: 5828 // CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 5829 // CHECK4-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 5830 // CHECK4-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 5831 // CHECK4-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 5832 // CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 5833 // CHECK4-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 5834 // CHECK4-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 5835 // CHECK4-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 5836 // CHECK4-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.4* 5837 // CHECK4-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 5838 // CHECK4-NEXT: [[TMP6:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 5839 // CHECK4-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4:%.*]], %struct._globalized_locals_ty.4* [[TMP4]], i32 0, i32 0 5840 // CHECK4-NEXT: [[TMP7:%.*]] = getelementptr inbounds [2048 x i32], [2048 x i32]* [[A]], i32 0, i32 [[TMP5]] 5841 // CHECK4-NEXT: [[TMP8:%.*]] = bitcast i32* [[TMP7]] to i8* 5842 // CHECK4-NEXT: store i8* [[TMP8]], i8** [[TMP6]], align 4 5843 // CHECK4-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 5844 // CHECK4-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4]], %struct._globalized_locals_ty.4* [[TMP4]], i32 0, i32 1 5845 // CHECK4-NEXT: [[TMP10:%.*]] = getelementptr inbounds [2048 x i16], [2048 x i16]* [[B]], i32 0, i32 [[TMP5]] 5846 // CHECK4-NEXT: [[TMP11:%.*]] = bitcast i16* [[TMP10]] to i8* 5847 // CHECK4-NEXT: store i8* [[TMP11]], i8** [[TMP9]], align 4 5848 // CHECK4-NEXT: [[TMP12:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 5849 // CHECK4-NEXT: [[TMP13:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 5850 // CHECK4-NEXT: call void @"_omp$reduction$reduction_func16"(i8* [[TMP13]], i8* [[TMP12]]) #[[ATTR3]] 5851 // CHECK4-NEXT: ret void 5852 // 5853 // 5854 // CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l23_worker 5855 // CHECK5-SAME: () #[[ATTR0:[0-9]+]] { 5856 // CHECK5-NEXT: entry: 5857 // CHECK5-NEXT: [[WORK_FN:%.*]] = alloca i8*, align 4 5858 // CHECK5-NEXT: [[EXEC_STATUS:%.*]] = alloca i8, align 1 5859 // CHECK5-NEXT: store i8* null, i8** [[WORK_FN]], align 4 5860 // CHECK5-NEXT: store i8 0, i8* [[EXEC_STATUS]], align 1 5861 // CHECK5-NEXT: br label [[DOTAWAIT_WORK:%.*]] 5862 // CHECK5: .await.work: 5863 // CHECK5-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 5864 // CHECK5-NEXT: [[TMP0:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORK_FN]]) 5865 // CHECK5-NEXT: [[TMP1:%.*]] = zext i1 [[TMP0]] to i8 5866 // CHECK5-NEXT: store i8 [[TMP1]], i8* [[EXEC_STATUS]], align 1 5867 // CHECK5-NEXT: [[TMP2:%.*]] = load i8*, i8** [[WORK_FN]], align 4 5868 // CHECK5-NEXT: [[SHOULD_TERMINATE:%.*]] = icmp eq i8* [[TMP2]], null 5869 // CHECK5-NEXT: br i1 [[SHOULD_TERMINATE]], label [[DOTEXIT:%.*]], label [[DOTSELECT_WORKERS:%.*]] 5870 // CHECK5: .select.workers: 5871 // CHECK5-NEXT: [[TMP3:%.*]] = load i8, i8* [[EXEC_STATUS]], align 1 5872 // CHECK5-NEXT: [[IS_ACTIVE:%.*]] = icmp ne i8 [[TMP3]], 0 5873 // CHECK5-NEXT: br i1 [[IS_ACTIVE]], label [[DOTEXECUTE_PARALLEL:%.*]], label [[DOTBARRIER_PARALLEL:%.*]] 5874 // CHECK5: .execute.parallel: 5875 // CHECK5-NEXT: [[TMP4:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]]) 5876 // CHECK5-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP2]] to void (i16, i32)* 5877 // CHECK5-NEXT: call void [[TMP5]](i16 0, i32 [[TMP4]]) 5878 // CHECK5-NEXT: br label [[DOTTERMINATE_PARALLEL:%.*]] 5879 // CHECK5: .terminate.parallel: 5880 // CHECK5-NEXT: call void @__kmpc_kernel_end_parallel() 5881 // CHECK5-NEXT: br label [[DOTBARRIER_PARALLEL]] 5882 // CHECK5: .barrier.parallel: 5883 // CHECK5-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 5884 // CHECK5-NEXT: br label [[DOTAWAIT_WORK]] 5885 // CHECK5: .exit: 5886 // CHECK5-NEXT: ret void 5887 // 5888 // 5889 // CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l23 5890 // CHECK5-SAME: (double* nonnull align 8 dereferenceable(8) [[E:%.*]]) #[[ATTR1:[0-9]+]] { 5891 // CHECK5-NEXT: entry: 5892 // CHECK5-NEXT: [[E_ADDR:%.*]] = alloca double*, align 4 5893 // CHECK5-NEXT: [[E7:%.*]] = alloca double, align 8 5894 // CHECK5-NEXT: [[DOTZERO_ADDR:%.*]] = alloca i32, align 4 5895 // CHECK5-NEXT: [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4 5896 // CHECK5-NEXT: store i32 0, i32* [[DOTZERO_ADDR]], align 4 5897 // CHECK5-NEXT: store double* [[E]], double** [[E_ADDR]], align 4 5898 // CHECK5-NEXT: [[TMP0:%.*]] = load double*, double** [[E_ADDR]], align 4 5899 // CHECK5-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 5900 // CHECK5-NEXT: [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 5901 // CHECK5-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 5902 // CHECK5-NEXT: [[THREAD_LIMIT:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS]], [[NVPTX_WARP_SIZE]] 5903 // CHECK5-NEXT: [[TMP1:%.*]] = icmp ult i32 [[NVPTX_TID]], [[THREAD_LIMIT]] 5904 // CHECK5-NEXT: br i1 [[TMP1]], label [[DOTWORKER:%.*]], label [[DOTMASTERCHECK:%.*]] 5905 // CHECK5: .worker: 5906 // CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l23_worker() #[[ATTR3:[0-9]+]] 5907 // CHECK5-NEXT: br label [[DOTEXIT:%.*]] 5908 // CHECK5: .mastercheck: 5909 // CHECK5-NEXT: [[NVPTX_TID1:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 5910 // CHECK5-NEXT: [[NVPTX_NUM_THREADS2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 5911 // CHECK5-NEXT: [[NVPTX_WARP_SIZE3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 5912 // CHECK5-NEXT: [[TMP2:%.*]] = sub nuw i32 [[NVPTX_WARP_SIZE3]], 1 5913 // CHECK5-NEXT: [[TMP3:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS2]], 1 5914 // CHECK5-NEXT: [[TMP4:%.*]] = xor i32 [[TMP2]], -1 5915 // CHECK5-NEXT: [[MASTER_TID:%.*]] = and i32 [[TMP3]], [[TMP4]] 5916 // CHECK5-NEXT: [[TMP5:%.*]] = icmp eq i32 [[NVPTX_TID1]], [[MASTER_TID]] 5917 // CHECK5-NEXT: br i1 [[TMP5]], label [[DOTMASTER:%.*]], label [[DOTEXIT]] 5918 // CHECK5: .master: 5919 // CHECK5-NEXT: [[NVPTX_NUM_THREADS4:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 5920 // CHECK5-NEXT: [[NVPTX_WARP_SIZE5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 5921 // CHECK5-NEXT: [[THREAD_LIMIT6:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS4]], [[NVPTX_WARP_SIZE5]] 5922 // CHECK5-NEXT: call void @__kmpc_kernel_init(i32 [[THREAD_LIMIT6]], i16 1) 5923 // CHECK5-NEXT: call void @__kmpc_data_sharing_init_stack() 5924 // CHECK5-NEXT: [[TMP6:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 5925 // CHECK5-NEXT: [[TMP7:%.*]] = load double, double* [[TMP0]], align 8 5926 // CHECK5-NEXT: store double [[TMP7]], double* [[E7]], align 8 5927 // CHECK5-NEXT: store i32 [[TMP6]], i32* [[DOTTHREADID_TEMP_]], align 4 5928 // CHECK5-NEXT: call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], double* [[E7]]) #[[ATTR3]] 5929 // CHECK5-NEXT: br label [[DOTTERMINATION_NOTIFIER:%.*]] 5930 // CHECK5: .termination.notifier: 5931 // CHECK5-NEXT: call void @__kmpc_kernel_deinit(i16 1) 5932 // CHECK5-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 5933 // CHECK5-NEXT: br label [[DOTEXIT]] 5934 // CHECK5: .exit: 5935 // CHECK5-NEXT: ret void 5936 // 5937 // 5938 // CHECK5-LABEL: define {{[^@]+}}@__omp_outlined__ 5939 // CHECK5-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], double* nonnull align 8 dereferenceable(8) [[E:%.*]]) #[[ATTR1]] { 5940 // CHECK5-NEXT: entry: 5941 // CHECK5-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4 5942 // CHECK5-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4 5943 // CHECK5-NEXT: [[E_ADDR:%.*]] = alloca double*, align 4 5944 // CHECK5-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x i8*], align 4 5945 // CHECK5-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4 5946 // CHECK5-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4 5947 // CHECK5-NEXT: store double* [[E]], double** [[E_ADDR]], align 4 5948 // CHECK5-NEXT: [[TMP0:%.*]] = load double*, double** [[E_ADDR]], align 4 5949 // CHECK5-NEXT: [[TMP1:%.*]] = call i8* @__kmpc_data_sharing_push_stack(i32 8, i16 1) 5950 // CHECK5-NEXT: [[TMP2:%.*]] = bitcast i8* [[TMP1]] to %struct._globalized_locals_ty* 5951 // CHECK5-NEXT: [[E1:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY:%.*]], %struct._globalized_locals_ty* [[TMP2]], i32 0, i32 0 5952 // CHECK5-NEXT: store double 0.000000e+00, double* [[E1]], align 8 5953 // CHECK5-NEXT: [[TMP3:%.*]] = load double, double* [[E1]], align 8 5954 // CHECK5-NEXT: [[ADD:%.*]] = fadd double [[TMP3]], 5.000000e+00 5955 // CHECK5-NEXT: store double [[ADD]], double* [[E1]], align 8 5956 // CHECK5-NEXT: [[TMP4:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4 5957 // CHECK5-NEXT: [[TMP5:%.*]] = load i32, i32* [[TMP4]], align 4 5958 // CHECK5-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 5959 // CHECK5-NEXT: [[TMP7:%.*]] = bitcast double* [[E1]] to i8* 5960 // CHECK5-NEXT: store i8* [[TMP7]], i8** [[TMP6]], align 4 5961 // CHECK5-NEXT: [[TMP8:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 5962 // CHECK5-NEXT: [[TMP9:%.*]] = load i8*, i8** @"_openmp_teams_reductions_buffer_$_$ptr", align 4 5963 // CHECK5-NEXT: [[TMP10:%.*]] = call i32 @__kmpc_nvptx_teams_reduce_nowait_v2(%struct.ident_t* @[[GLOB1]], i32 [[TMP5]], i8* [[TMP9]], i32 1024, i8* [[TMP8]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_copy_func, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_reduce_func, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_copy_func, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_reduce_func) 5964 // CHECK5-NEXT: [[TMP11:%.*]] = icmp eq i32 [[TMP10]], 1 5965 // CHECK5-NEXT: br i1 [[TMP11]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]] 5966 // CHECK5: .omp.reduction.then: 5967 // CHECK5-NEXT: [[TMP12:%.*]] = load double, double* [[TMP0]], align 8 5968 // CHECK5-NEXT: [[TMP13:%.*]] = load double, double* [[E1]], align 8 5969 // CHECK5-NEXT: [[ADD2:%.*]] = fadd double [[TMP12]], [[TMP13]] 5970 // CHECK5-NEXT: store double [[ADD2]], double* [[TMP0]], align 8 5971 // CHECK5-NEXT: call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP5]]) 5972 // CHECK5-NEXT: br label [[DOTOMP_REDUCTION_DONE]] 5973 // CHECK5: .omp.reduction.done: 5974 // CHECK5-NEXT: call void @__kmpc_data_sharing_pop_stack(i8* [[TMP1]]) 5975 // CHECK5-NEXT: ret void 5976 // 5977 // 5978 // CHECK5-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func 5979 // CHECK5-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR0]] { 5980 // CHECK5-NEXT: entry: 5981 // CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 5982 // CHECK5-NEXT: [[DOTADDR1:%.*]] = alloca i16, align 2 5983 // CHECK5-NEXT: [[DOTADDR2:%.*]] = alloca i16, align 2 5984 // CHECK5-NEXT: [[DOTADDR3:%.*]] = alloca i16, align 2 5985 // CHECK5-NEXT: [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [1 x i8*], align 4 5986 // CHECK5-NEXT: [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca double, align 8 5987 // CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 5988 // CHECK5-NEXT: store i16 [[TMP1]], i16* [[DOTADDR1]], align 2 5989 // CHECK5-NEXT: store i16 [[TMP2]], i16* [[DOTADDR2]], align 2 5990 // CHECK5-NEXT: store i16 [[TMP3]], i16* [[DOTADDR3]], align 2 5991 // CHECK5-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 4 5992 // CHECK5-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [1 x i8*]* 5993 // CHECK5-NEXT: [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2 5994 // CHECK5-NEXT: [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2 5995 // CHECK5-NEXT: [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2 5996 // CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i32 0, i32 0 5997 // CHECK5-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 4 5998 // CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 5999 // CHECK5-NEXT: [[TMP12:%.*]] = bitcast i8* [[TMP10]] to double* 6000 // CHECK5-NEXT: [[TMP13:%.*]] = getelementptr double, double* [[TMP12]], i32 1 6001 // CHECK5-NEXT: [[TMP14:%.*]] = bitcast double* [[TMP13]] to i8* 6002 // CHECK5-NEXT: [[TMP15:%.*]] = bitcast double* [[TMP12]] to i64* 6003 // CHECK5-NEXT: [[TMP16:%.*]] = bitcast double* [[DOTOMP_REDUCTION_ELEMENT]] to i64* 6004 // CHECK5-NEXT: [[TMP17:%.*]] = load i64, i64* [[TMP15]], align 8 6005 // CHECK5-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 6006 // CHECK5-NEXT: [[TMP18:%.*]] = trunc i32 [[NVPTX_WARP_SIZE]] to i16 6007 // CHECK5-NEXT: [[TMP19:%.*]] = call i64 @__kmpc_shuffle_int64(i64 [[TMP17]], i16 [[TMP7]], i16 [[TMP18]]) 6008 // CHECK5-NEXT: store i64 [[TMP19]], i64* [[TMP16]], align 8 6009 // CHECK5-NEXT: [[TMP20:%.*]] = getelementptr i64, i64* [[TMP15]], i32 1 6010 // CHECK5-NEXT: [[TMP21:%.*]] = getelementptr i64, i64* [[TMP16]], i32 1 6011 // CHECK5-NEXT: [[TMP22:%.*]] = bitcast double* [[DOTOMP_REDUCTION_ELEMENT]] to i8* 6012 // CHECK5-NEXT: store i8* [[TMP22]], i8** [[TMP11]], align 4 6013 // CHECK5-NEXT: [[TMP23:%.*]] = icmp eq i16 [[TMP8]], 0 6014 // CHECK5-NEXT: [[TMP24:%.*]] = icmp eq i16 [[TMP8]], 1 6015 // CHECK5-NEXT: [[TMP25:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]] 6016 // CHECK5-NEXT: [[TMP26:%.*]] = and i1 [[TMP24]], [[TMP25]] 6017 // CHECK5-NEXT: [[TMP27:%.*]] = icmp eq i16 [[TMP8]], 2 6018 // CHECK5-NEXT: [[TMP28:%.*]] = and i16 [[TMP6]], 1 6019 // CHECK5-NEXT: [[TMP29:%.*]] = icmp eq i16 [[TMP28]], 0 6020 // CHECK5-NEXT: [[TMP30:%.*]] = and i1 [[TMP27]], [[TMP29]] 6021 // CHECK5-NEXT: [[TMP31:%.*]] = icmp sgt i16 [[TMP7]], 0 6022 // CHECK5-NEXT: [[TMP32:%.*]] = and i1 [[TMP30]], [[TMP31]] 6023 // CHECK5-NEXT: [[TMP33:%.*]] = or i1 [[TMP23]], [[TMP26]] 6024 // CHECK5-NEXT: [[TMP34:%.*]] = or i1 [[TMP33]], [[TMP32]] 6025 // CHECK5-NEXT: br i1 [[TMP34]], label [[THEN:%.*]], label [[ELSE:%.*]] 6026 // CHECK5: then: 6027 // CHECK5-NEXT: [[TMP35:%.*]] = bitcast [1 x i8*]* [[TMP5]] to i8* 6028 // CHECK5-NEXT: [[TMP36:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8* 6029 // CHECK5-NEXT: call void @"_omp$reduction$reduction_func"(i8* [[TMP35]], i8* [[TMP36]]) #[[ATTR3]] 6030 // CHECK5-NEXT: br label [[IFCONT:%.*]] 6031 // CHECK5: else: 6032 // CHECK5-NEXT: br label [[IFCONT]] 6033 // CHECK5: ifcont: 6034 // CHECK5-NEXT: [[TMP37:%.*]] = icmp eq i16 [[TMP8]], 1 6035 // CHECK5-NEXT: [[TMP38:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]] 6036 // CHECK5-NEXT: [[TMP39:%.*]] = and i1 [[TMP37]], [[TMP38]] 6037 // CHECK5-NEXT: br i1 [[TMP39]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 6038 // CHECK5: then4: 6039 // CHECK5-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 6040 // CHECK5-NEXT: [[TMP41:%.*]] = load i8*, i8** [[TMP40]], align 4 6041 // CHECK5-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i32 0, i32 0 6042 // CHECK5-NEXT: [[TMP43:%.*]] = load i8*, i8** [[TMP42]], align 4 6043 // CHECK5-NEXT: [[TMP44:%.*]] = bitcast i8* [[TMP41]] to double* 6044 // CHECK5-NEXT: [[TMP45:%.*]] = bitcast i8* [[TMP43]] to double* 6045 // CHECK5-NEXT: [[TMP46:%.*]] = load double, double* [[TMP44]], align 8 6046 // CHECK5-NEXT: store double [[TMP46]], double* [[TMP45]], align 8 6047 // CHECK5-NEXT: br label [[IFCONT6:%.*]] 6048 // CHECK5: else5: 6049 // CHECK5-NEXT: br label [[IFCONT6]] 6050 // CHECK5: ifcont6: 6051 // CHECK5-NEXT: ret void 6052 // 6053 // 6054 // CHECK5-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func 6055 // CHECK5-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] { 6056 // CHECK5-NEXT: entry: 6057 // CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 6058 // CHECK5-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 6059 // CHECK5-NEXT: [[DOTCNT_ADDR:%.*]] = alloca i32, align 4 6060 // CHECK5-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 6061 // CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 6062 // CHECK5-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 6063 // CHECK5-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 6064 // CHECK5-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 6065 // CHECK5-NEXT: [[NVPTX_LANE_ID:%.*]] = and i32 [[NVPTX_TID2]], 31 6066 // CHECK5-NEXT: [[NVPTX_TID3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 6067 // CHECK5-NEXT: [[NVPTX_WARP_ID:%.*]] = ashr i32 [[NVPTX_TID3]], 5 6068 // CHECK5-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 6069 // CHECK5-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [1 x i8*]* 6070 // CHECK5-NEXT: store i32 0, i32* [[DOTCNT_ADDR]], align 4 6071 // CHECK5-NEXT: br label [[PRECOND:%.*]] 6072 // CHECK5: precond: 6073 // CHECK5-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTCNT_ADDR]], align 4 6074 // CHECK5-NEXT: [[TMP6:%.*]] = icmp ult i32 [[TMP5]], 2 6075 // CHECK5-NEXT: br i1 [[TMP6]], label [[BODY:%.*]], label [[EXIT:%.*]] 6076 // CHECK5: body: 6077 // CHECK5-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP2]]) 6078 // CHECK5-NEXT: [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 6079 // CHECK5-NEXT: br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]] 6080 // CHECK5: then: 6081 // CHECK5-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP4]], i32 0, i32 0 6082 // CHECK5-NEXT: [[TMP8:%.*]] = load i8*, i8** [[TMP7]], align 4 6083 // CHECK5-NEXT: [[TMP9:%.*]] = bitcast i8* [[TMP8]] to i32* 6084 // CHECK5-NEXT: [[TMP10:%.*]] = getelementptr i32, i32* [[TMP9]], i32 [[TMP5]] 6085 // CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 6086 // CHECK5-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP10]], align 4 6087 // CHECK5-NEXT: store volatile i32 [[TMP12]], i32 addrspace(3)* [[TMP11]], align 4 6088 // CHECK5-NEXT: br label [[IFCONT:%.*]] 6089 // CHECK5: else: 6090 // CHECK5-NEXT: br label [[IFCONT]] 6091 // CHECK5: ifcont: 6092 // CHECK5-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 6093 // CHECK5-NEXT: [[TMP13:%.*]] = load i32, i32* [[DOTADDR1]], align 4 6094 // CHECK5-NEXT: [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP13]] 6095 // CHECK5-NEXT: br i1 [[IS_ACTIVE_THREAD]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 6096 // CHECK5: then4: 6097 // CHECK5-NEXT: [[TMP14:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 6098 // CHECK5-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP4]], i32 0, i32 0 6099 // CHECK5-NEXT: [[TMP16:%.*]] = load i8*, i8** [[TMP15]], align 4 6100 // CHECK5-NEXT: [[TMP17:%.*]] = bitcast i8* [[TMP16]] to i32* 6101 // CHECK5-NEXT: [[TMP18:%.*]] = getelementptr i32, i32* [[TMP17]], i32 [[TMP5]] 6102 // CHECK5-NEXT: [[TMP19:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP14]], align 4 6103 // CHECK5-NEXT: store i32 [[TMP19]], i32* [[TMP18]], align 4 6104 // CHECK5-NEXT: br label [[IFCONT6:%.*]] 6105 // CHECK5: else5: 6106 // CHECK5-NEXT: br label [[IFCONT6]] 6107 // CHECK5: ifcont6: 6108 // CHECK5-NEXT: [[TMP20:%.*]] = add nsw i32 [[TMP5]], 1 6109 // CHECK5-NEXT: store i32 [[TMP20]], i32* [[DOTCNT_ADDR]], align 4 6110 // CHECK5-NEXT: br label [[PRECOND]] 6111 // CHECK5: exit: 6112 // CHECK5-NEXT: ret void 6113 // 6114 // 6115 // CHECK5-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_copy_func 6116 // CHECK5-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 6117 // CHECK5-NEXT: entry: 6118 // CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 6119 // CHECK5-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 6120 // CHECK5-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 6121 // CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 6122 // CHECK5-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 6123 // CHECK5-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 6124 // CHECK5-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 6125 // CHECK5-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [1 x i8*]* 6126 // CHECK5-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 4 6127 // CHECK5-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.0* 6128 // CHECK5-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 6129 // CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP4]], i32 0, i32 0 6130 // CHECK5-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 4 6131 // CHECK5-NEXT: [[TMP10:%.*]] = bitcast i8* [[TMP9]] to double* 6132 // CHECK5-NEXT: [[E:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_0:%.*]], %struct._globalized_locals_ty.0* [[TMP6]], i32 0, i32 0 6133 // CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [1024 x double], [1024 x double]* [[E]], i32 0, i32 [[TMP7]] 6134 // CHECK5-NEXT: [[TMP12:%.*]] = load double, double* [[TMP10]], align 8 6135 // CHECK5-NEXT: store double [[TMP12]], double* [[TMP11]], align 128 6136 // CHECK5-NEXT: ret void 6137 // 6138 // 6139 // CHECK5-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_reduce_func 6140 // CHECK5-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 6141 // CHECK5-NEXT: entry: 6142 // CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 6143 // CHECK5-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 6144 // CHECK5-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 6145 // CHECK5-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x i8*], align 4 6146 // CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 6147 // CHECK5-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 6148 // CHECK5-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 6149 // CHECK5-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 6150 // CHECK5-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.0* 6151 // CHECK5-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 6152 // CHECK5-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 6153 // CHECK5-NEXT: [[E:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_0:%.*]], %struct._globalized_locals_ty.0* [[TMP4]], i32 0, i32 0 6154 // CHECK5-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1024 x double], [1024 x double]* [[E]], i32 0, i32 [[TMP5]] 6155 // CHECK5-NEXT: [[TMP8:%.*]] = bitcast double* [[TMP7]] to i8* 6156 // CHECK5-NEXT: store i8* [[TMP8]], i8** [[TMP6]], align 4 6157 // CHECK5-NEXT: [[TMP9:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 6158 // CHECK5-NEXT: [[TMP10:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 6159 // CHECK5-NEXT: call void @"_omp$reduction$reduction_func"(i8* [[TMP9]], i8* [[TMP10]]) #[[ATTR3]] 6160 // CHECK5-NEXT: ret void 6161 // 6162 // 6163 // CHECK5-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_copy_func 6164 // CHECK5-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 6165 // CHECK5-NEXT: entry: 6166 // CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 6167 // CHECK5-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 6168 // CHECK5-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 6169 // CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 6170 // CHECK5-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 6171 // CHECK5-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 6172 // CHECK5-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 6173 // CHECK5-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [1 x i8*]* 6174 // CHECK5-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 4 6175 // CHECK5-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.0* 6176 // CHECK5-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 6177 // CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP4]], i32 0, i32 0 6178 // CHECK5-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 4 6179 // CHECK5-NEXT: [[TMP10:%.*]] = bitcast i8* [[TMP9]] to double* 6180 // CHECK5-NEXT: [[E:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_0:%.*]], %struct._globalized_locals_ty.0* [[TMP6]], i32 0, i32 0 6181 // CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [1024 x double], [1024 x double]* [[E]], i32 0, i32 [[TMP7]] 6182 // CHECK5-NEXT: [[TMP12:%.*]] = load double, double* [[TMP11]], align 128 6183 // CHECK5-NEXT: store double [[TMP12]], double* [[TMP10]], align 8 6184 // CHECK5-NEXT: ret void 6185 // 6186 // 6187 // CHECK5-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_reduce_func 6188 // CHECK5-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 6189 // CHECK5-NEXT: entry: 6190 // CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 6191 // CHECK5-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 6192 // CHECK5-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 6193 // CHECK5-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x i8*], align 4 6194 // CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 6195 // CHECK5-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 6196 // CHECK5-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 6197 // CHECK5-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 6198 // CHECK5-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.0* 6199 // CHECK5-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 6200 // CHECK5-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 6201 // CHECK5-NEXT: [[E:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_0:%.*]], %struct._globalized_locals_ty.0* [[TMP4]], i32 0, i32 0 6202 // CHECK5-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1024 x double], [1024 x double]* [[E]], i32 0, i32 [[TMP5]] 6203 // CHECK5-NEXT: [[TMP8:%.*]] = bitcast double* [[TMP7]] to i8* 6204 // CHECK5-NEXT: store i8* [[TMP8]], i8** [[TMP6]], align 4 6205 // CHECK5-NEXT: [[TMP9:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 6206 // CHECK5-NEXT: [[TMP10:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 6207 // CHECK5-NEXT: call void @"_omp$reduction$reduction_func"(i8* [[TMP10]], i8* [[TMP9]]) #[[ATTR3]] 6208 // CHECK5-NEXT: ret void 6209 // 6210 // 6211 // CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29_worker 6212 // CHECK5-SAME: () #[[ATTR0]] { 6213 // CHECK5-NEXT: entry: 6214 // CHECK5-NEXT: [[WORK_FN:%.*]] = alloca i8*, align 4 6215 // CHECK5-NEXT: [[EXEC_STATUS:%.*]] = alloca i8, align 1 6216 // CHECK5-NEXT: store i8* null, i8** [[WORK_FN]], align 4 6217 // CHECK5-NEXT: store i8 0, i8* [[EXEC_STATUS]], align 1 6218 // CHECK5-NEXT: br label [[DOTAWAIT_WORK:%.*]] 6219 // CHECK5: .await.work: 6220 // CHECK5-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 6221 // CHECK5-NEXT: [[TMP0:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORK_FN]]) 6222 // CHECK5-NEXT: [[TMP1:%.*]] = zext i1 [[TMP0]] to i8 6223 // CHECK5-NEXT: store i8 [[TMP1]], i8* [[EXEC_STATUS]], align 1 6224 // CHECK5-NEXT: [[TMP2:%.*]] = load i8*, i8** [[WORK_FN]], align 4 6225 // CHECK5-NEXT: [[SHOULD_TERMINATE:%.*]] = icmp eq i8* [[TMP2]], null 6226 // CHECK5-NEXT: br i1 [[SHOULD_TERMINATE]], label [[DOTEXIT:%.*]], label [[DOTSELECT_WORKERS:%.*]] 6227 // CHECK5: .select.workers: 6228 // CHECK5-NEXT: [[TMP3:%.*]] = load i8, i8* [[EXEC_STATUS]], align 1 6229 // CHECK5-NEXT: [[IS_ACTIVE:%.*]] = icmp ne i8 [[TMP3]], 0 6230 // CHECK5-NEXT: br i1 [[IS_ACTIVE]], label [[DOTEXECUTE_PARALLEL:%.*]], label [[DOTBARRIER_PARALLEL:%.*]] 6231 // CHECK5: .execute.parallel: 6232 // CHECK5-NEXT: [[TMP4:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 6233 // CHECK5-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP2]] to void (i16, i32)* 6234 // CHECK5-NEXT: call void [[TMP5]](i16 0, i32 [[TMP4]]) 6235 // CHECK5-NEXT: br label [[DOTTERMINATE_PARALLEL:%.*]] 6236 // CHECK5: .terminate.parallel: 6237 // CHECK5-NEXT: call void @__kmpc_kernel_end_parallel() 6238 // CHECK5-NEXT: br label [[DOTBARRIER_PARALLEL]] 6239 // CHECK5: .barrier.parallel: 6240 // CHECK5-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 6241 // CHECK5-NEXT: br label [[DOTAWAIT_WORK]] 6242 // CHECK5: .exit: 6243 // CHECK5-NEXT: ret void 6244 // 6245 // 6246 // CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29 6247 // CHECK5-SAME: (i32 [[C:%.*]], i32 [[D:%.*]]) #[[ATTR1]] { 6248 // CHECK5-NEXT: entry: 6249 // CHECK5-NEXT: [[C_ADDR:%.*]] = alloca i32, align 4 6250 // CHECK5-NEXT: [[D_ADDR:%.*]] = alloca i32, align 4 6251 // CHECK5-NEXT: [[DOTZERO_ADDR:%.*]] = alloca i32, align 4 6252 // CHECK5-NEXT: [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4 6253 // CHECK5-NEXT: store i32 0, i32* [[DOTZERO_ADDR]], align 4 6254 // CHECK5-NEXT: store i32 [[C]], i32* [[C_ADDR]], align 4 6255 // CHECK5-NEXT: store i32 [[D]], i32* [[D_ADDR]], align 4 6256 // CHECK5-NEXT: [[CONV:%.*]] = bitcast i32* [[C_ADDR]] to i8* 6257 // CHECK5-NEXT: [[CONV1:%.*]] = bitcast i32* [[D_ADDR]] to float* 6258 // CHECK5-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 6259 // CHECK5-NEXT: [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 6260 // CHECK5-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 6261 // CHECK5-NEXT: [[THREAD_LIMIT:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS]], [[NVPTX_WARP_SIZE]] 6262 // CHECK5-NEXT: [[TMP0:%.*]] = icmp ult i32 [[NVPTX_TID]], [[THREAD_LIMIT]] 6263 // CHECK5-NEXT: br i1 [[TMP0]], label [[DOTWORKER:%.*]], label [[DOTMASTERCHECK:%.*]] 6264 // CHECK5: .worker: 6265 // CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29_worker() #[[ATTR3]] 6266 // CHECK5-NEXT: br label [[DOTEXIT:%.*]] 6267 // CHECK5: .mastercheck: 6268 // CHECK5-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 6269 // CHECK5-NEXT: [[NVPTX_NUM_THREADS3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 6270 // CHECK5-NEXT: [[NVPTX_WARP_SIZE4:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 6271 // CHECK5-NEXT: [[TMP1:%.*]] = sub nuw i32 [[NVPTX_WARP_SIZE4]], 1 6272 // CHECK5-NEXT: [[TMP2:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS3]], 1 6273 // CHECK5-NEXT: [[TMP3:%.*]] = xor i32 [[TMP1]], -1 6274 // CHECK5-NEXT: [[MASTER_TID:%.*]] = and i32 [[TMP2]], [[TMP3]] 6275 // CHECK5-NEXT: [[TMP4:%.*]] = icmp eq i32 [[NVPTX_TID2]], [[MASTER_TID]] 6276 // CHECK5-NEXT: br i1 [[TMP4]], label [[DOTMASTER:%.*]], label [[DOTEXIT]] 6277 // CHECK5: .master: 6278 // CHECK5-NEXT: [[NVPTX_NUM_THREADS5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 6279 // CHECK5-NEXT: [[NVPTX_WARP_SIZE6:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 6280 // CHECK5-NEXT: [[THREAD_LIMIT7:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS5]], [[NVPTX_WARP_SIZE6]] 6281 // CHECK5-NEXT: call void @__kmpc_kernel_init(i32 [[THREAD_LIMIT7]], i16 1) 6282 // CHECK5-NEXT: call void @__kmpc_data_sharing_init_stack() 6283 // CHECK5-NEXT: [[TMP5:%.*]] = call i8* @__kmpc_data_sharing_push_stack(i32 8, i16 1) 6284 // CHECK5-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.1* 6285 // CHECK5-NEXT: [[TMP7:%.*]] = load i8, i8* [[CONV]], align 4 6286 // CHECK5-NEXT: [[C8:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_1:%.*]], %struct._globalized_locals_ty.1* [[TMP6]], i32 0, i32 1 6287 // CHECK5-NEXT: store i8 [[TMP7]], i8* [[C8]], align 4 6288 // CHECK5-NEXT: [[TMP8:%.*]] = load float, float* [[CONV1]], align 4 6289 // CHECK5-NEXT: [[D9:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_1]], %struct._globalized_locals_ty.1* [[TMP6]], i32 0, i32 0 6290 // CHECK5-NEXT: store float [[TMP8]], float* [[D9]], align 4 6291 // CHECK5-NEXT: [[TMP9:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 6292 // CHECK5-NEXT: store i32 [[TMP9]], i32* [[DOTTHREADID_TEMP_]], align 4 6293 // CHECK5-NEXT: call void @__omp_outlined__1(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i8* [[C8]], float* [[D9]]) #[[ATTR3]] 6294 // CHECK5-NEXT: call void @__kmpc_data_sharing_pop_stack(i8* [[TMP5]]) 6295 // CHECK5-NEXT: br label [[DOTTERMINATION_NOTIFIER:%.*]] 6296 // CHECK5: .termination.notifier: 6297 // CHECK5-NEXT: call void @__kmpc_kernel_deinit(i16 1) 6298 // CHECK5-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 6299 // CHECK5-NEXT: br label [[DOTEXIT]] 6300 // CHECK5: .exit: 6301 // CHECK5-NEXT: ret void 6302 // 6303 // 6304 // CHECK5-LABEL: define {{[^@]+}}@__omp_outlined__1 6305 // CHECK5-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i8* nonnull align 1 dereferenceable(1) [[C:%.*]], float* nonnull align 4 dereferenceable(4) [[D:%.*]]) #[[ATTR1]] { 6306 // CHECK5-NEXT: entry: 6307 // CHECK5-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4 6308 // CHECK5-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4 6309 // CHECK5-NEXT: [[C_ADDR:%.*]] = alloca i8*, align 4 6310 // CHECK5-NEXT: [[D_ADDR:%.*]] = alloca float*, align 4 6311 // CHECK5-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 6312 // CHECK5-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4 6313 // CHECK5-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4 6314 // CHECK5-NEXT: store i8* [[C]], i8** [[C_ADDR]], align 4 6315 // CHECK5-NEXT: store float* [[D]], float** [[D_ADDR]], align 4 6316 // CHECK5-NEXT: [[TMP0:%.*]] = load i8*, i8** [[C_ADDR]], align 4 6317 // CHECK5-NEXT: [[TMP1:%.*]] = load float*, float** [[D_ADDR]], align 4 6318 // CHECK5-NEXT: [[TMP2:%.*]] = call i8* @__kmpc_data_sharing_push_stack(i32 8, i16 1) 6319 // CHECK5-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to %struct._globalized_locals_ty.2* 6320 // CHECK5-NEXT: [[C1:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_2:%.*]], %struct._globalized_locals_ty.2* [[TMP3]], i32 0, i32 1 6321 // CHECK5-NEXT: [[D2:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_2]], %struct._globalized_locals_ty.2* [[TMP3]], i32 0, i32 0 6322 // CHECK5-NEXT: store i8 0, i8* [[C1]], align 4 6323 // CHECK5-NEXT: store float 1.000000e+00, float* [[D2]], align 4 6324 // CHECK5-NEXT: [[TMP4:%.*]] = load i8, i8* [[C1]], align 4 6325 // CHECK5-NEXT: [[CONV:%.*]] = sext i8 [[TMP4]] to i32 6326 // CHECK5-NEXT: [[XOR:%.*]] = xor i32 [[CONV]], 2 6327 // CHECK5-NEXT: [[CONV3:%.*]] = trunc i32 [[XOR]] to i8 6328 // CHECK5-NEXT: store i8 [[CONV3]], i8* [[C1]], align 4 6329 // CHECK5-NEXT: [[TMP5:%.*]] = load float, float* [[D2]], align 4 6330 // CHECK5-NEXT: [[MUL:%.*]] = fmul float [[TMP5]], 3.300000e+01 6331 // CHECK5-NEXT: store float [[MUL]], float* [[D2]], align 4 6332 // CHECK5-NEXT: [[TMP6:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4 6333 // CHECK5-NEXT: [[TMP7:%.*]] = load i32, i32* [[TMP6]], align 4 6334 // CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 6335 // CHECK5-NEXT: store i8* [[C1]], i8** [[TMP8]], align 4 6336 // CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 6337 // CHECK5-NEXT: [[TMP10:%.*]] = bitcast float* [[D2]] to i8* 6338 // CHECK5-NEXT: store i8* [[TMP10]], i8** [[TMP9]], align 4 6339 // CHECK5-NEXT: [[TMP11:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 6340 // CHECK5-NEXT: [[TMP12:%.*]] = load i8*, i8** @"_openmp_teams_reductions_buffer_$_$ptr", align 4 6341 // CHECK5-NEXT: [[TMP13:%.*]] = call i32 @__kmpc_nvptx_teams_reduce_nowait_v2(%struct.ident_t* @[[GLOB1]], i32 [[TMP7]], i8* [[TMP12]], i32 1024, i8* [[TMP11]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func3, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func4, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_copy_func5, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_reduce_func6, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_copy_func7, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_reduce_func8) 6342 // CHECK5-NEXT: [[TMP14:%.*]] = icmp eq i32 [[TMP13]], 1 6343 // CHECK5-NEXT: br i1 [[TMP14]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]] 6344 // CHECK5: .omp.reduction.then: 6345 // CHECK5-NEXT: [[TMP15:%.*]] = load i8, i8* [[TMP0]], align 1 6346 // CHECK5-NEXT: [[CONV4:%.*]] = sext i8 [[TMP15]] to i32 6347 // CHECK5-NEXT: [[TMP16:%.*]] = load i8, i8* [[C1]], align 4 6348 // CHECK5-NEXT: [[CONV5:%.*]] = sext i8 [[TMP16]] to i32 6349 // CHECK5-NEXT: [[XOR6:%.*]] = xor i32 [[CONV4]], [[CONV5]] 6350 // CHECK5-NEXT: [[CONV7:%.*]] = trunc i32 [[XOR6]] to i8 6351 // CHECK5-NEXT: store i8 [[CONV7]], i8* [[TMP0]], align 1 6352 // CHECK5-NEXT: [[TMP17:%.*]] = load float, float* [[TMP1]], align 4 6353 // CHECK5-NEXT: [[TMP18:%.*]] = load float, float* [[D2]], align 4 6354 // CHECK5-NEXT: [[MUL8:%.*]] = fmul float [[TMP17]], [[TMP18]] 6355 // CHECK5-NEXT: store float [[MUL8]], float* [[TMP1]], align 4 6356 // CHECK5-NEXT: call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP7]]) 6357 // CHECK5-NEXT: br label [[DOTOMP_REDUCTION_DONE]] 6358 // CHECK5: .omp.reduction.done: 6359 // CHECK5-NEXT: call void @__kmpc_data_sharing_pop_stack(i8* [[TMP2]]) 6360 // CHECK5-NEXT: ret void 6361 // 6362 // 6363 // CHECK5-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func3 6364 // CHECK5-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR0]] { 6365 // CHECK5-NEXT: entry: 6366 // CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 6367 // CHECK5-NEXT: [[DOTADDR1:%.*]] = alloca i16, align 2 6368 // CHECK5-NEXT: [[DOTADDR2:%.*]] = alloca i16, align 2 6369 // CHECK5-NEXT: [[DOTADDR3:%.*]] = alloca i16, align 2 6370 // CHECK5-NEXT: [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [2 x i8*], align 4 6371 // CHECK5-NEXT: [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca i8, align 1 6372 // CHECK5-NEXT: [[DOTOMP_REDUCTION_ELEMENT4:%.*]] = alloca float, align 4 6373 // CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 6374 // CHECK5-NEXT: store i16 [[TMP1]], i16* [[DOTADDR1]], align 2 6375 // CHECK5-NEXT: store i16 [[TMP2]], i16* [[DOTADDR2]], align 2 6376 // CHECK5-NEXT: store i16 [[TMP3]], i16* [[DOTADDR3]], align 2 6377 // CHECK5-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 4 6378 // CHECK5-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [2 x i8*]* 6379 // CHECK5-NEXT: [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2 6380 // CHECK5-NEXT: [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2 6381 // CHECK5-NEXT: [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2 6382 // CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 0 6383 // CHECK5-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 4 6384 // CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 6385 // CHECK5-NEXT: [[TMP12:%.*]] = getelementptr i8, i8* [[TMP10]], i32 1 6386 // CHECK5-NEXT: [[TMP13:%.*]] = load i8, i8* [[TMP10]], align 1 6387 // CHECK5-NEXT: [[TMP14:%.*]] = sext i8 [[TMP13]] to i32 6388 // CHECK5-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 6389 // CHECK5-NEXT: [[TMP15:%.*]] = trunc i32 [[NVPTX_WARP_SIZE]] to i16 6390 // CHECK5-NEXT: [[TMP16:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP14]], i16 [[TMP7]], i16 [[TMP15]]) 6391 // CHECK5-NEXT: [[TMP17:%.*]] = trunc i32 [[TMP16]] to i8 6392 // CHECK5-NEXT: store i8 [[TMP17]], i8* [[DOTOMP_REDUCTION_ELEMENT]], align 1 6393 // CHECK5-NEXT: [[TMP18:%.*]] = getelementptr i8, i8* [[TMP10]], i32 1 6394 // CHECK5-NEXT: [[TMP19:%.*]] = getelementptr i8, i8* [[DOTOMP_REDUCTION_ELEMENT]], i32 1 6395 // CHECK5-NEXT: store i8* [[DOTOMP_REDUCTION_ELEMENT]], i8** [[TMP11]], align 4 6396 // CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 1 6397 // CHECK5-NEXT: [[TMP21:%.*]] = load i8*, i8** [[TMP20]], align 4 6398 // CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1 6399 // CHECK5-NEXT: [[TMP23:%.*]] = bitcast i8* [[TMP21]] to float* 6400 // CHECK5-NEXT: [[TMP24:%.*]] = getelementptr float, float* [[TMP23]], i32 1 6401 // CHECK5-NEXT: [[TMP25:%.*]] = bitcast float* [[TMP24]] to i8* 6402 // CHECK5-NEXT: [[TMP26:%.*]] = bitcast float* [[TMP23]] to i32* 6403 // CHECK5-NEXT: [[TMP27:%.*]] = bitcast float* [[DOTOMP_REDUCTION_ELEMENT4]] to i32* 6404 // CHECK5-NEXT: [[TMP28:%.*]] = load i32, i32* [[TMP26]], align 4 6405 // CHECK5-NEXT: [[NVPTX_WARP_SIZE5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 6406 // CHECK5-NEXT: [[TMP29:%.*]] = trunc i32 [[NVPTX_WARP_SIZE5]] to i16 6407 // CHECK5-NEXT: [[TMP30:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP28]], i16 [[TMP7]], i16 [[TMP29]]) 6408 // CHECK5-NEXT: store i32 [[TMP30]], i32* [[TMP27]], align 4 6409 // CHECK5-NEXT: [[TMP31:%.*]] = getelementptr i32, i32* [[TMP26]], i32 1 6410 // CHECK5-NEXT: [[TMP32:%.*]] = getelementptr i32, i32* [[TMP27]], i32 1 6411 // CHECK5-NEXT: [[TMP33:%.*]] = bitcast float* [[DOTOMP_REDUCTION_ELEMENT4]] to i8* 6412 // CHECK5-NEXT: store i8* [[TMP33]], i8** [[TMP22]], align 4 6413 // CHECK5-NEXT: [[TMP34:%.*]] = icmp eq i16 [[TMP8]], 0 6414 // CHECK5-NEXT: [[TMP35:%.*]] = icmp eq i16 [[TMP8]], 1 6415 // CHECK5-NEXT: [[TMP36:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]] 6416 // CHECK5-NEXT: [[TMP37:%.*]] = and i1 [[TMP35]], [[TMP36]] 6417 // CHECK5-NEXT: [[TMP38:%.*]] = icmp eq i16 [[TMP8]], 2 6418 // CHECK5-NEXT: [[TMP39:%.*]] = and i16 [[TMP6]], 1 6419 // CHECK5-NEXT: [[TMP40:%.*]] = icmp eq i16 [[TMP39]], 0 6420 // CHECK5-NEXT: [[TMP41:%.*]] = and i1 [[TMP38]], [[TMP40]] 6421 // CHECK5-NEXT: [[TMP42:%.*]] = icmp sgt i16 [[TMP7]], 0 6422 // CHECK5-NEXT: [[TMP43:%.*]] = and i1 [[TMP41]], [[TMP42]] 6423 // CHECK5-NEXT: [[TMP44:%.*]] = or i1 [[TMP34]], [[TMP37]] 6424 // CHECK5-NEXT: [[TMP45:%.*]] = or i1 [[TMP44]], [[TMP43]] 6425 // CHECK5-NEXT: br i1 [[TMP45]], label [[THEN:%.*]], label [[ELSE:%.*]] 6426 // CHECK5: then: 6427 // CHECK5-NEXT: [[TMP46:%.*]] = bitcast [2 x i8*]* [[TMP5]] to i8* 6428 // CHECK5-NEXT: [[TMP47:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8* 6429 // CHECK5-NEXT: call void @"_omp$reduction$reduction_func2"(i8* [[TMP46]], i8* [[TMP47]]) #[[ATTR3]] 6430 // CHECK5-NEXT: br label [[IFCONT:%.*]] 6431 // CHECK5: else: 6432 // CHECK5-NEXT: br label [[IFCONT]] 6433 // CHECK5: ifcont: 6434 // CHECK5-NEXT: [[TMP48:%.*]] = icmp eq i16 [[TMP8]], 1 6435 // CHECK5-NEXT: [[TMP49:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]] 6436 // CHECK5-NEXT: [[TMP50:%.*]] = and i1 [[TMP48]], [[TMP49]] 6437 // CHECK5-NEXT: br i1 [[TMP50]], label [[THEN6:%.*]], label [[ELSE7:%.*]] 6438 // CHECK5: then6: 6439 // CHECK5-NEXT: [[TMP51:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 6440 // CHECK5-NEXT: [[TMP52:%.*]] = load i8*, i8** [[TMP51]], align 4 6441 // CHECK5-NEXT: [[TMP53:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 0 6442 // CHECK5-NEXT: [[TMP54:%.*]] = load i8*, i8** [[TMP53]], align 4 6443 // CHECK5-NEXT: [[TMP55:%.*]] = load i8, i8* [[TMP52]], align 1 6444 // CHECK5-NEXT: store i8 [[TMP55]], i8* [[TMP54]], align 1 6445 // CHECK5-NEXT: [[TMP56:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1 6446 // CHECK5-NEXT: [[TMP57:%.*]] = load i8*, i8** [[TMP56]], align 4 6447 // CHECK5-NEXT: [[TMP58:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 1 6448 // CHECK5-NEXT: [[TMP59:%.*]] = load i8*, i8** [[TMP58]], align 4 6449 // CHECK5-NEXT: [[TMP60:%.*]] = bitcast i8* [[TMP57]] to float* 6450 // CHECK5-NEXT: [[TMP61:%.*]] = bitcast i8* [[TMP59]] to float* 6451 // CHECK5-NEXT: [[TMP62:%.*]] = load float, float* [[TMP60]], align 4 6452 // CHECK5-NEXT: store float [[TMP62]], float* [[TMP61]], align 4 6453 // CHECK5-NEXT: br label [[IFCONT8:%.*]] 6454 // CHECK5: else7: 6455 // CHECK5-NEXT: br label [[IFCONT8]] 6456 // CHECK5: ifcont8: 6457 // CHECK5-NEXT: ret void 6458 // 6459 // 6460 // CHECK5-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func4 6461 // CHECK5-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] { 6462 // CHECK5-NEXT: entry: 6463 // CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 6464 // CHECK5-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 6465 // CHECK5-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 6466 // CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 6467 // CHECK5-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 6468 // CHECK5-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 6469 // CHECK5-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 6470 // CHECK5-NEXT: [[NVPTX_LANE_ID:%.*]] = and i32 [[NVPTX_TID2]], 31 6471 // CHECK5-NEXT: [[NVPTX_TID3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 6472 // CHECK5-NEXT: [[NVPTX_WARP_ID:%.*]] = ashr i32 [[NVPTX_TID3]], 5 6473 // CHECK5-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 6474 // CHECK5-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 6475 // CHECK5-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 6476 // CHECK5-NEXT: [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 6477 // CHECK5-NEXT: br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]] 6478 // CHECK5: then: 6479 // CHECK5-NEXT: [[TMP5:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 6480 // CHECK5-NEXT: [[TMP6:%.*]] = load i8*, i8** [[TMP5]], align 4 6481 // CHECK5-NEXT: [[TMP7:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 6482 // CHECK5-NEXT: [[TMP8:%.*]] = bitcast i32 addrspace(3)* [[TMP7]] to i8 addrspace(3)* 6483 // CHECK5-NEXT: [[TMP9:%.*]] = load i8, i8* [[TMP6]], align 1 6484 // CHECK5-NEXT: store volatile i8 [[TMP9]], i8 addrspace(3)* [[TMP8]], align 1 6485 // CHECK5-NEXT: br label [[IFCONT:%.*]] 6486 // CHECK5: else: 6487 // CHECK5-NEXT: br label [[IFCONT]] 6488 // CHECK5: ifcont: 6489 // CHECK5-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 6490 // CHECK5-NEXT: [[TMP10:%.*]] = load i32, i32* [[DOTADDR1]], align 4 6491 // CHECK5-NEXT: [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP10]] 6492 // CHECK5-NEXT: br i1 [[IS_ACTIVE_THREAD]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 6493 // CHECK5: then4: 6494 // CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 6495 // CHECK5-NEXT: [[TMP12:%.*]] = bitcast i32 addrspace(3)* [[TMP11]] to i8 addrspace(3)* 6496 // CHECK5-NEXT: [[TMP13:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 6497 // CHECK5-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 4 6498 // CHECK5-NEXT: [[TMP15:%.*]] = load volatile i8, i8 addrspace(3)* [[TMP12]], align 1 6499 // CHECK5-NEXT: store i8 [[TMP15]], i8* [[TMP14]], align 1 6500 // CHECK5-NEXT: br label [[IFCONT6:%.*]] 6501 // CHECK5: else5: 6502 // CHECK5-NEXT: br label [[IFCONT6]] 6503 // CHECK5: ifcont6: 6504 // CHECK5-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 6505 // CHECK5-NEXT: [[WARP_MASTER7:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 6506 // CHECK5-NEXT: br i1 [[WARP_MASTER7]], label [[THEN8:%.*]], label [[ELSE9:%.*]] 6507 // CHECK5: then8: 6508 // CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 6509 // CHECK5-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 4 6510 // CHECK5-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i32* 6511 // CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 6512 // CHECK5-NEXT: [[TMP20:%.*]] = load i32, i32* [[TMP18]], align 4 6513 // CHECK5-NEXT: store volatile i32 [[TMP20]], i32 addrspace(3)* [[TMP19]], align 4 6514 // CHECK5-NEXT: br label [[IFCONT10:%.*]] 6515 // CHECK5: else9: 6516 // CHECK5-NEXT: br label [[IFCONT10]] 6517 // CHECK5: ifcont10: 6518 // CHECK5-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 6519 // CHECK5-NEXT: [[TMP21:%.*]] = load i32, i32* [[DOTADDR1]], align 4 6520 // CHECK5-NEXT: [[IS_ACTIVE_THREAD11:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP21]] 6521 // CHECK5-NEXT: br i1 [[IS_ACTIVE_THREAD11]], label [[THEN12:%.*]], label [[ELSE13:%.*]] 6522 // CHECK5: then12: 6523 // CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 6524 // CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 6525 // CHECK5-NEXT: [[TMP24:%.*]] = load i8*, i8** [[TMP23]], align 4 6526 // CHECK5-NEXT: [[TMP25:%.*]] = bitcast i8* [[TMP24]] to i32* 6527 // CHECK5-NEXT: [[TMP26:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP22]], align 4 6528 // CHECK5-NEXT: store i32 [[TMP26]], i32* [[TMP25]], align 4 6529 // CHECK5-NEXT: br label [[IFCONT14:%.*]] 6530 // CHECK5: else13: 6531 // CHECK5-NEXT: br label [[IFCONT14]] 6532 // CHECK5: ifcont14: 6533 // CHECK5-NEXT: ret void 6534 // 6535 // 6536 // CHECK5-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_copy_func5 6537 // CHECK5-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 6538 // CHECK5-NEXT: entry: 6539 // CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 6540 // CHECK5-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 6541 // CHECK5-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 6542 // CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 6543 // CHECK5-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 6544 // CHECK5-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 6545 // CHECK5-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 6546 // CHECK5-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 6547 // CHECK5-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 4 6548 // CHECK5-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.3* 6549 // CHECK5-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 6550 // CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 6551 // CHECK5-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 4 6552 // CHECK5-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3:%.*]], %struct._globalized_locals_ty.3* [[TMP6]], i32 0, i32 0 6553 // CHECK5-NEXT: [[TMP10:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[C]], i32 0, i32 [[TMP7]] 6554 // CHECK5-NEXT: [[TMP11:%.*]] = load i8, i8* [[TMP9]], align 1 6555 // CHECK5-NEXT: store i8 [[TMP11]], i8* [[TMP10]], align 128 6556 // CHECK5-NEXT: [[TMP12:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 6557 // CHECK5-NEXT: [[TMP13:%.*]] = load i8*, i8** [[TMP12]], align 4 6558 // CHECK5-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to float* 6559 // CHECK5-NEXT: [[D:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3]], %struct._globalized_locals_ty.3* [[TMP6]], i32 0, i32 1 6560 // CHECK5-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1024 x float], [1024 x float]* [[D]], i32 0, i32 [[TMP7]] 6561 // CHECK5-NEXT: [[TMP16:%.*]] = load float, float* [[TMP14]], align 4 6562 // CHECK5-NEXT: store float [[TMP16]], float* [[TMP15]], align 128 6563 // CHECK5-NEXT: ret void 6564 // 6565 // 6566 // CHECK5-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_reduce_func6 6567 // CHECK5-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 6568 // CHECK5-NEXT: entry: 6569 // CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 6570 // CHECK5-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 6571 // CHECK5-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 6572 // CHECK5-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 6573 // CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 6574 // CHECK5-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 6575 // CHECK5-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 6576 // CHECK5-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 6577 // CHECK5-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.3* 6578 // CHECK5-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 6579 // CHECK5-NEXT: [[TMP6:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 6580 // CHECK5-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3:%.*]], %struct._globalized_locals_ty.3* [[TMP4]], i32 0, i32 0 6581 // CHECK5-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[C]], i32 0, i32 [[TMP5]] 6582 // CHECK5-NEXT: store i8* [[TMP7]], i8** [[TMP6]], align 4 6583 // CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 6584 // CHECK5-NEXT: [[D:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3]], %struct._globalized_locals_ty.3* [[TMP4]], i32 0, i32 1 6585 // CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1024 x float], [1024 x float]* [[D]], i32 0, i32 [[TMP5]] 6586 // CHECK5-NEXT: [[TMP10:%.*]] = bitcast float* [[TMP9]] to i8* 6587 // CHECK5-NEXT: store i8* [[TMP10]], i8** [[TMP8]], align 4 6588 // CHECK5-NEXT: [[TMP11:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 6589 // CHECK5-NEXT: [[TMP12:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 6590 // CHECK5-NEXT: call void @"_omp$reduction$reduction_func2"(i8* [[TMP11]], i8* [[TMP12]]) #[[ATTR3]] 6591 // CHECK5-NEXT: ret void 6592 // 6593 // 6594 // CHECK5-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_copy_func7 6595 // CHECK5-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 6596 // CHECK5-NEXT: entry: 6597 // CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 6598 // CHECK5-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 6599 // CHECK5-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 6600 // CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 6601 // CHECK5-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 6602 // CHECK5-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 6603 // CHECK5-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 6604 // CHECK5-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 6605 // CHECK5-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 4 6606 // CHECK5-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.3* 6607 // CHECK5-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 6608 // CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 6609 // CHECK5-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 4 6610 // CHECK5-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3:%.*]], %struct._globalized_locals_ty.3* [[TMP6]], i32 0, i32 0 6611 // CHECK5-NEXT: [[TMP10:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[C]], i32 0, i32 [[TMP7]] 6612 // CHECK5-NEXT: [[TMP11:%.*]] = load i8, i8* [[TMP10]], align 128 6613 // CHECK5-NEXT: store i8 [[TMP11]], i8* [[TMP9]], align 1 6614 // CHECK5-NEXT: [[TMP12:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 6615 // CHECK5-NEXT: [[TMP13:%.*]] = load i8*, i8** [[TMP12]], align 4 6616 // CHECK5-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to float* 6617 // CHECK5-NEXT: [[D:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3]], %struct._globalized_locals_ty.3* [[TMP6]], i32 0, i32 1 6618 // CHECK5-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1024 x float], [1024 x float]* [[D]], i32 0, i32 [[TMP7]] 6619 // CHECK5-NEXT: [[TMP16:%.*]] = load float, float* [[TMP15]], align 128 6620 // CHECK5-NEXT: store float [[TMP16]], float* [[TMP14]], align 4 6621 // CHECK5-NEXT: ret void 6622 // 6623 // 6624 // CHECK5-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_reduce_func8 6625 // CHECK5-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 6626 // CHECK5-NEXT: entry: 6627 // CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 6628 // CHECK5-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 6629 // CHECK5-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 6630 // CHECK5-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 6631 // CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 6632 // CHECK5-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 6633 // CHECK5-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 6634 // CHECK5-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 6635 // CHECK5-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.3* 6636 // CHECK5-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 6637 // CHECK5-NEXT: [[TMP6:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 6638 // CHECK5-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3:%.*]], %struct._globalized_locals_ty.3* [[TMP4]], i32 0, i32 0 6639 // CHECK5-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[C]], i32 0, i32 [[TMP5]] 6640 // CHECK5-NEXT: store i8* [[TMP7]], i8** [[TMP6]], align 4 6641 // CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 6642 // CHECK5-NEXT: [[D:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3]], %struct._globalized_locals_ty.3* [[TMP4]], i32 0, i32 1 6643 // CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1024 x float], [1024 x float]* [[D]], i32 0, i32 [[TMP5]] 6644 // CHECK5-NEXT: [[TMP10:%.*]] = bitcast float* [[TMP9]] to i8* 6645 // CHECK5-NEXT: store i8* [[TMP10]], i8** [[TMP8]], align 4 6646 // CHECK5-NEXT: [[TMP11:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 6647 // CHECK5-NEXT: [[TMP12:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 6648 // CHECK5-NEXT: call void @"_omp$reduction$reduction_func2"(i8* [[TMP12]], i8* [[TMP11]]) #[[ATTR3]] 6649 // CHECK5-NEXT: ret void 6650 // 6651 // 6652 // CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l36 6653 // CHECK5-SAME: (i32 [[A:%.*]], i32 [[B:%.*]]) #[[ATTR1]] { 6654 // CHECK5-NEXT: entry: 6655 // CHECK5-NEXT: [[A_ADDR:%.*]] = alloca i32, align 4 6656 // CHECK5-NEXT: [[B_ADDR:%.*]] = alloca i32, align 4 6657 // CHECK5-NEXT: [[DOTZERO_ADDR:%.*]] = alloca i32, align 4 6658 // CHECK5-NEXT: [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4 6659 // CHECK5-NEXT: store i32 0, i32* [[DOTZERO_ADDR]], align 4 6660 // CHECK5-NEXT: store i32 [[A]], i32* [[A_ADDR]], align 4 6661 // CHECK5-NEXT: store i32 [[B]], i32* [[B_ADDR]], align 4 6662 // CHECK5-NEXT: [[CONV:%.*]] = bitcast i32* [[B_ADDR]] to i16* 6663 // CHECK5-NEXT: [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 6664 // CHECK5-NEXT: call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 1) 6665 // CHECK5-NEXT: call void @__kmpc_data_sharing_init_stack_spmd() 6666 // CHECK5-NEXT: br label [[DOTEXECUTE:%.*]] 6667 // CHECK5: .execute: 6668 // CHECK5-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3:[0-9]+]]) 6669 // CHECK5-NEXT: store i32 [[TMP0]], i32* [[DOTTHREADID_TEMP_]], align 4 6670 // CHECK5-NEXT: call void @__omp_outlined__9(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32* [[A_ADDR]], i16* [[CONV]]) #[[ATTR3]] 6671 // CHECK5-NEXT: br label [[DOTOMP_DEINIT:%.*]] 6672 // CHECK5: .omp.deinit: 6673 // CHECK5-NEXT: call void @__kmpc_spmd_kernel_deinit_v2(i16 1) 6674 // CHECK5-NEXT: br label [[DOTEXIT:%.*]] 6675 // CHECK5: .exit: 6676 // CHECK5-NEXT: ret void 6677 // 6678 // 6679 // CHECK5-LABEL: define {{[^@]+}}@__omp_outlined__9 6680 // CHECK5-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32* nonnull align 4 dereferenceable(4) [[A:%.*]], i16* nonnull align 2 dereferenceable(2) [[B:%.*]]) #[[ATTR1]] { 6681 // CHECK5-NEXT: entry: 6682 // CHECK5-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4 6683 // CHECK5-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4 6684 // CHECK5-NEXT: [[A_ADDR:%.*]] = alloca i32*, align 4 6685 // CHECK5-NEXT: [[B_ADDR:%.*]] = alloca i16*, align 4 6686 // CHECK5-NEXT: [[A1:%.*]] = alloca i32, align 4 6687 // CHECK5-NEXT: [[B2:%.*]] = alloca i16, align 2 6688 // CHECK5-NEXT: [[CAPTURED_VARS_ADDRS:%.*]] = alloca [2 x i8*], align 4 6689 // CHECK5-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 6690 // CHECK5-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4 6691 // CHECK5-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4 6692 // CHECK5-NEXT: store i32* [[A]], i32** [[A_ADDR]], align 4 6693 // CHECK5-NEXT: store i16* [[B]], i16** [[B_ADDR]], align 4 6694 // CHECK5-NEXT: [[TMP0:%.*]] = load i32*, i32** [[A_ADDR]], align 4 6695 // CHECK5-NEXT: [[TMP1:%.*]] = load i16*, i16** [[B_ADDR]], align 4 6696 // CHECK5-NEXT: store i32 0, i32* [[A1]], align 4 6697 // CHECK5-NEXT: store i16 -32768, i16* [[B2]], align 2 6698 // CHECK5-NEXT: [[TMP2:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0 6699 // CHECK5-NEXT: [[TMP3:%.*]] = bitcast i32* [[A1]] to i8* 6700 // CHECK5-NEXT: store i8* [[TMP3]], i8** [[TMP2]], align 4 6701 // CHECK5-NEXT: [[TMP4:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1 6702 // CHECK5-NEXT: [[TMP5:%.*]] = bitcast i16* [[B2]] to i8* 6703 // CHECK5-NEXT: store i8* [[TMP5]], i8** [[TMP4]], align 4 6704 // CHECK5-NEXT: [[TMP6:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4 6705 // CHECK5-NEXT: [[TMP7:%.*]] = load i32, i32* [[TMP6]], align 4 6706 // CHECK5-NEXT: [[TMP8:%.*]] = bitcast [2 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8** 6707 // CHECK5-NEXT: call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP7]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32*, i16*)* @__omp_outlined__10 to i8*), i8* null, i8** [[TMP8]], i32 2) 6708 // CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 6709 // CHECK5-NEXT: [[TMP10:%.*]] = bitcast i32* [[A1]] to i8* 6710 // CHECK5-NEXT: store i8* [[TMP10]], i8** [[TMP9]], align 4 6711 // CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 6712 // CHECK5-NEXT: [[TMP12:%.*]] = bitcast i16* [[B2]] to i8* 6713 // CHECK5-NEXT: store i8* [[TMP12]], i8** [[TMP11]], align 4 6714 // CHECK5-NEXT: [[TMP13:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 6715 // CHECK5-NEXT: [[TMP14:%.*]] = load i8*, i8** @"_openmp_teams_reductions_buffer_$_$ptr", align 4 6716 // CHECK5-NEXT: [[TMP15:%.*]] = call i32 @__kmpc_nvptx_teams_reduce_nowait_v2(%struct.ident_t* @[[GLOB3]], i32 [[TMP7]], i8* [[TMP14]], i32 1024, i8* [[TMP13]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func15, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func16, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_copy_func17, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_reduce_func18, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_copy_func19, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_reduce_func20) 6717 // CHECK5-NEXT: [[TMP16:%.*]] = icmp eq i32 [[TMP15]], 1 6718 // CHECK5-NEXT: br i1 [[TMP16]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]] 6719 // CHECK5: .omp.reduction.then: 6720 // CHECK5-NEXT: [[TMP17:%.*]] = load i32, i32* [[TMP0]], align 4 6721 // CHECK5-NEXT: [[TMP18:%.*]] = load i32, i32* [[A1]], align 4 6722 // CHECK5-NEXT: [[OR:%.*]] = or i32 [[TMP17]], [[TMP18]] 6723 // CHECK5-NEXT: store i32 [[OR]], i32* [[TMP0]], align 4 6724 // CHECK5-NEXT: [[TMP19:%.*]] = load i16, i16* [[TMP1]], align 2 6725 // CHECK5-NEXT: [[CONV:%.*]] = sext i16 [[TMP19]] to i32 6726 // CHECK5-NEXT: [[TMP20:%.*]] = load i16, i16* [[B2]], align 2 6727 // CHECK5-NEXT: [[CONV3:%.*]] = sext i16 [[TMP20]] to i32 6728 // CHECK5-NEXT: [[CMP:%.*]] = icmp sgt i32 [[CONV]], [[CONV3]] 6729 // CHECK5-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] 6730 // CHECK5: cond.true: 6731 // CHECK5-NEXT: [[TMP21:%.*]] = load i16, i16* [[TMP1]], align 2 6732 // CHECK5-NEXT: br label [[COND_END:%.*]] 6733 // CHECK5: cond.false: 6734 // CHECK5-NEXT: [[TMP22:%.*]] = load i16, i16* [[B2]], align 2 6735 // CHECK5-NEXT: br label [[COND_END]] 6736 // CHECK5: cond.end: 6737 // CHECK5-NEXT: [[COND:%.*]] = phi i16 [ [[TMP21]], [[COND_TRUE]] ], [ [[TMP22]], [[COND_FALSE]] ] 6738 // CHECK5-NEXT: store i16 [[COND]], i16* [[TMP1]], align 2 6739 // CHECK5-NEXT: call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP7]]) 6740 // CHECK5-NEXT: br label [[DOTOMP_REDUCTION_DONE]] 6741 // CHECK5: .omp.reduction.done: 6742 // CHECK5-NEXT: ret void 6743 // 6744 // 6745 // CHECK5-LABEL: define {{[^@]+}}@__omp_outlined__10 6746 // CHECK5-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32* nonnull align 4 dereferenceable(4) [[A:%.*]], i16* nonnull align 2 dereferenceable(2) [[B:%.*]]) #[[ATTR1]] { 6747 // CHECK5-NEXT: entry: 6748 // CHECK5-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4 6749 // CHECK5-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4 6750 // CHECK5-NEXT: [[A_ADDR:%.*]] = alloca i32*, align 4 6751 // CHECK5-NEXT: [[B_ADDR:%.*]] = alloca i16*, align 4 6752 // CHECK5-NEXT: [[A1:%.*]] = alloca i32, align 4 6753 // CHECK5-NEXT: [[B2:%.*]] = alloca i16, align 2 6754 // CHECK5-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 6755 // CHECK5-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4 6756 // CHECK5-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4 6757 // CHECK5-NEXT: store i32* [[A]], i32** [[A_ADDR]], align 4 6758 // CHECK5-NEXT: store i16* [[B]], i16** [[B_ADDR]], align 4 6759 // CHECK5-NEXT: [[TMP0:%.*]] = load i32*, i32** [[A_ADDR]], align 4 6760 // CHECK5-NEXT: [[TMP1:%.*]] = load i16*, i16** [[B_ADDR]], align 4 6761 // CHECK5-NEXT: store i32 0, i32* [[A1]], align 4 6762 // CHECK5-NEXT: store i16 -32768, i16* [[B2]], align 2 6763 // CHECK5-NEXT: [[TMP2:%.*]] = load i32, i32* [[A1]], align 4 6764 // CHECK5-NEXT: [[OR:%.*]] = or i32 [[TMP2]], 1 6765 // CHECK5-NEXT: store i32 [[OR]], i32* [[A1]], align 4 6766 // CHECK5-NEXT: [[TMP3:%.*]] = load i16, i16* [[B2]], align 2 6767 // CHECK5-NEXT: [[CONV:%.*]] = sext i16 [[TMP3]] to i32 6768 // CHECK5-NEXT: [[CMP:%.*]] = icmp sgt i32 99, [[CONV]] 6769 // CHECK5-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] 6770 // CHECK5: cond.true: 6771 // CHECK5-NEXT: br label [[COND_END:%.*]] 6772 // CHECK5: cond.false: 6773 // CHECK5-NEXT: [[TMP4:%.*]] = load i16, i16* [[B2]], align 2 6774 // CHECK5-NEXT: [[CONV3:%.*]] = sext i16 [[TMP4]] to i32 6775 // CHECK5-NEXT: br label [[COND_END]] 6776 // CHECK5: cond.end: 6777 // CHECK5-NEXT: [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[CONV3]], [[COND_FALSE]] ] 6778 // CHECK5-NEXT: [[CONV4:%.*]] = trunc i32 [[COND]] to i16 6779 // CHECK5-NEXT: store i16 [[CONV4]], i16* [[B2]], align 2 6780 // CHECK5-NEXT: [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4 6781 // CHECK5-NEXT: [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4 6782 // CHECK5-NEXT: [[TMP7:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 6783 // CHECK5-NEXT: [[TMP8:%.*]] = bitcast i32* [[A1]] to i8* 6784 // CHECK5-NEXT: store i8* [[TMP8]], i8** [[TMP7]], align 4 6785 // CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 6786 // CHECK5-NEXT: [[TMP10:%.*]] = bitcast i16* [[B2]] to i8* 6787 // CHECK5-NEXT: store i8* [[TMP10]], i8** [[TMP9]], align 4 6788 // CHECK5-NEXT: [[TMP11:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 6789 // CHECK5-NEXT: [[TMP12:%.*]] = call i32 @__kmpc_nvptx_parallel_reduce_nowait_v2(%struct.ident_t* @[[GLOB3]], i32 [[TMP6]], i32 2, i32 8, i8* [[TMP11]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func12, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func13) 6790 // CHECK5-NEXT: [[TMP13:%.*]] = icmp eq i32 [[TMP12]], 1 6791 // CHECK5-NEXT: br i1 [[TMP13]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]] 6792 // CHECK5: .omp.reduction.then: 6793 // CHECK5-NEXT: [[TMP14:%.*]] = load i32, i32* [[TMP0]], align 4 6794 // CHECK5-NEXT: [[TMP15:%.*]] = load i32, i32* [[A1]], align 4 6795 // CHECK5-NEXT: [[OR5:%.*]] = or i32 [[TMP14]], [[TMP15]] 6796 // CHECK5-NEXT: store i32 [[OR5]], i32* [[TMP0]], align 4 6797 // CHECK5-NEXT: [[TMP16:%.*]] = load i16, i16* [[TMP1]], align 2 6798 // CHECK5-NEXT: [[CONV6:%.*]] = sext i16 [[TMP16]] to i32 6799 // CHECK5-NEXT: [[TMP17:%.*]] = load i16, i16* [[B2]], align 2 6800 // CHECK5-NEXT: [[CONV7:%.*]] = sext i16 [[TMP17]] to i32 6801 // CHECK5-NEXT: [[CMP8:%.*]] = icmp sgt i32 [[CONV6]], [[CONV7]] 6802 // CHECK5-NEXT: br i1 [[CMP8]], label [[COND_TRUE9:%.*]], label [[COND_FALSE10:%.*]] 6803 // CHECK5: cond.true9: 6804 // CHECK5-NEXT: [[TMP18:%.*]] = load i16, i16* [[TMP1]], align 2 6805 // CHECK5-NEXT: br label [[COND_END11:%.*]] 6806 // CHECK5: cond.false10: 6807 // CHECK5-NEXT: [[TMP19:%.*]] = load i16, i16* [[B2]], align 2 6808 // CHECK5-NEXT: br label [[COND_END11]] 6809 // CHECK5: cond.end11: 6810 // CHECK5-NEXT: [[COND12:%.*]] = phi i16 [ [[TMP18]], [[COND_TRUE9]] ], [ [[TMP19]], [[COND_FALSE10]] ] 6811 // CHECK5-NEXT: store i16 [[COND12]], i16* [[TMP1]], align 2 6812 // CHECK5-NEXT: call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP6]]) 6813 // CHECK5-NEXT: br label [[DOTOMP_REDUCTION_DONE]] 6814 // CHECK5: .omp.reduction.done: 6815 // CHECK5-NEXT: ret void 6816 // 6817 // 6818 // CHECK5-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func12 6819 // CHECK5-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR0]] { 6820 // CHECK5-NEXT: entry: 6821 // CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 6822 // CHECK5-NEXT: [[DOTADDR1:%.*]] = alloca i16, align 2 6823 // CHECK5-NEXT: [[DOTADDR2:%.*]] = alloca i16, align 2 6824 // CHECK5-NEXT: [[DOTADDR3:%.*]] = alloca i16, align 2 6825 // CHECK5-NEXT: [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [2 x i8*], align 4 6826 // CHECK5-NEXT: [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca i32, align 4 6827 // CHECK5-NEXT: [[DOTOMP_REDUCTION_ELEMENT4:%.*]] = alloca i16, align 2 6828 // CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 6829 // CHECK5-NEXT: store i16 [[TMP1]], i16* [[DOTADDR1]], align 2 6830 // CHECK5-NEXT: store i16 [[TMP2]], i16* [[DOTADDR2]], align 2 6831 // CHECK5-NEXT: store i16 [[TMP3]], i16* [[DOTADDR3]], align 2 6832 // CHECK5-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 4 6833 // CHECK5-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [2 x i8*]* 6834 // CHECK5-NEXT: [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2 6835 // CHECK5-NEXT: [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2 6836 // CHECK5-NEXT: [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2 6837 // CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 0 6838 // CHECK5-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 4 6839 // CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 6840 // CHECK5-NEXT: [[TMP12:%.*]] = bitcast i8* [[TMP10]] to i32* 6841 // CHECK5-NEXT: [[TMP13:%.*]] = getelementptr i32, i32* [[TMP12]], i32 1 6842 // CHECK5-NEXT: [[TMP14:%.*]] = bitcast i32* [[TMP13]] to i8* 6843 // CHECK5-NEXT: [[TMP15:%.*]] = load i32, i32* [[TMP12]], align 4 6844 // CHECK5-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 6845 // CHECK5-NEXT: [[TMP16:%.*]] = trunc i32 [[NVPTX_WARP_SIZE]] to i16 6846 // CHECK5-NEXT: [[TMP17:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP15]], i16 [[TMP7]], i16 [[TMP16]]) 6847 // CHECK5-NEXT: store i32 [[TMP17]], i32* [[DOTOMP_REDUCTION_ELEMENT]], align 4 6848 // CHECK5-NEXT: [[TMP18:%.*]] = getelementptr i32, i32* [[TMP12]], i32 1 6849 // CHECK5-NEXT: [[TMP19:%.*]] = getelementptr i32, i32* [[DOTOMP_REDUCTION_ELEMENT]], i32 1 6850 // CHECK5-NEXT: [[TMP20:%.*]] = bitcast i32* [[DOTOMP_REDUCTION_ELEMENT]] to i8* 6851 // CHECK5-NEXT: store i8* [[TMP20]], i8** [[TMP11]], align 4 6852 // CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 1 6853 // CHECK5-NEXT: [[TMP22:%.*]] = load i8*, i8** [[TMP21]], align 4 6854 // CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1 6855 // CHECK5-NEXT: [[TMP24:%.*]] = bitcast i8* [[TMP22]] to i16* 6856 // CHECK5-NEXT: [[TMP25:%.*]] = getelementptr i16, i16* [[TMP24]], i32 1 6857 // CHECK5-NEXT: [[TMP26:%.*]] = bitcast i16* [[TMP25]] to i8* 6858 // CHECK5-NEXT: [[TMP27:%.*]] = load i16, i16* [[TMP24]], align 2 6859 // CHECK5-NEXT: [[TMP28:%.*]] = sext i16 [[TMP27]] to i32 6860 // CHECK5-NEXT: [[NVPTX_WARP_SIZE5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 6861 // CHECK5-NEXT: [[TMP29:%.*]] = trunc i32 [[NVPTX_WARP_SIZE5]] to i16 6862 // CHECK5-NEXT: [[TMP30:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP28]], i16 [[TMP7]], i16 [[TMP29]]) 6863 // CHECK5-NEXT: [[TMP31:%.*]] = trunc i32 [[TMP30]] to i16 6864 // CHECK5-NEXT: store i16 [[TMP31]], i16* [[DOTOMP_REDUCTION_ELEMENT4]], align 2 6865 // CHECK5-NEXT: [[TMP32:%.*]] = getelementptr i16, i16* [[TMP24]], i32 1 6866 // CHECK5-NEXT: [[TMP33:%.*]] = getelementptr i16, i16* [[DOTOMP_REDUCTION_ELEMENT4]], i32 1 6867 // CHECK5-NEXT: [[TMP34:%.*]] = bitcast i16* [[DOTOMP_REDUCTION_ELEMENT4]] to i8* 6868 // CHECK5-NEXT: store i8* [[TMP34]], i8** [[TMP23]], align 4 6869 // CHECK5-NEXT: [[TMP35:%.*]] = icmp eq i16 [[TMP8]], 0 6870 // CHECK5-NEXT: [[TMP36:%.*]] = icmp eq i16 [[TMP8]], 1 6871 // CHECK5-NEXT: [[TMP37:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]] 6872 // CHECK5-NEXT: [[TMP38:%.*]] = and i1 [[TMP36]], [[TMP37]] 6873 // CHECK5-NEXT: [[TMP39:%.*]] = icmp eq i16 [[TMP8]], 2 6874 // CHECK5-NEXT: [[TMP40:%.*]] = and i16 [[TMP6]], 1 6875 // CHECK5-NEXT: [[TMP41:%.*]] = icmp eq i16 [[TMP40]], 0 6876 // CHECK5-NEXT: [[TMP42:%.*]] = and i1 [[TMP39]], [[TMP41]] 6877 // CHECK5-NEXT: [[TMP43:%.*]] = icmp sgt i16 [[TMP7]], 0 6878 // CHECK5-NEXT: [[TMP44:%.*]] = and i1 [[TMP42]], [[TMP43]] 6879 // CHECK5-NEXT: [[TMP45:%.*]] = or i1 [[TMP35]], [[TMP38]] 6880 // CHECK5-NEXT: [[TMP46:%.*]] = or i1 [[TMP45]], [[TMP44]] 6881 // CHECK5-NEXT: br i1 [[TMP46]], label [[THEN:%.*]], label [[ELSE:%.*]] 6882 // CHECK5: then: 6883 // CHECK5-NEXT: [[TMP47:%.*]] = bitcast [2 x i8*]* [[TMP5]] to i8* 6884 // CHECK5-NEXT: [[TMP48:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8* 6885 // CHECK5-NEXT: call void @"_omp$reduction$reduction_func11"(i8* [[TMP47]], i8* [[TMP48]]) #[[ATTR3]] 6886 // CHECK5-NEXT: br label [[IFCONT:%.*]] 6887 // CHECK5: else: 6888 // CHECK5-NEXT: br label [[IFCONT]] 6889 // CHECK5: ifcont: 6890 // CHECK5-NEXT: [[TMP49:%.*]] = icmp eq i16 [[TMP8]], 1 6891 // CHECK5-NEXT: [[TMP50:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]] 6892 // CHECK5-NEXT: [[TMP51:%.*]] = and i1 [[TMP49]], [[TMP50]] 6893 // CHECK5-NEXT: br i1 [[TMP51]], label [[THEN6:%.*]], label [[ELSE7:%.*]] 6894 // CHECK5: then6: 6895 // CHECK5-NEXT: [[TMP52:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 6896 // CHECK5-NEXT: [[TMP53:%.*]] = load i8*, i8** [[TMP52]], align 4 6897 // CHECK5-NEXT: [[TMP54:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 0 6898 // CHECK5-NEXT: [[TMP55:%.*]] = load i8*, i8** [[TMP54]], align 4 6899 // CHECK5-NEXT: [[TMP56:%.*]] = bitcast i8* [[TMP53]] to i32* 6900 // CHECK5-NEXT: [[TMP57:%.*]] = bitcast i8* [[TMP55]] to i32* 6901 // CHECK5-NEXT: [[TMP58:%.*]] = load i32, i32* [[TMP56]], align 4 6902 // CHECK5-NEXT: store i32 [[TMP58]], i32* [[TMP57]], align 4 6903 // CHECK5-NEXT: [[TMP59:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1 6904 // CHECK5-NEXT: [[TMP60:%.*]] = load i8*, i8** [[TMP59]], align 4 6905 // CHECK5-NEXT: [[TMP61:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 1 6906 // CHECK5-NEXT: [[TMP62:%.*]] = load i8*, i8** [[TMP61]], align 4 6907 // CHECK5-NEXT: [[TMP63:%.*]] = bitcast i8* [[TMP60]] to i16* 6908 // CHECK5-NEXT: [[TMP64:%.*]] = bitcast i8* [[TMP62]] to i16* 6909 // CHECK5-NEXT: [[TMP65:%.*]] = load i16, i16* [[TMP63]], align 2 6910 // CHECK5-NEXT: store i16 [[TMP65]], i16* [[TMP64]], align 2 6911 // CHECK5-NEXT: br label [[IFCONT8:%.*]] 6912 // CHECK5: else7: 6913 // CHECK5-NEXT: br label [[IFCONT8]] 6914 // CHECK5: ifcont8: 6915 // CHECK5-NEXT: ret void 6916 // 6917 // 6918 // CHECK5-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func13 6919 // CHECK5-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] { 6920 // CHECK5-NEXT: entry: 6921 // CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 6922 // CHECK5-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 6923 // CHECK5-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]]) 6924 // CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 6925 // CHECK5-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 6926 // CHECK5-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 6927 // CHECK5-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 6928 // CHECK5-NEXT: [[NVPTX_LANE_ID:%.*]] = and i32 [[NVPTX_TID2]], 31 6929 // CHECK5-NEXT: [[NVPTX_TID3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 6930 // CHECK5-NEXT: [[NVPTX_WARP_ID:%.*]] = ashr i32 [[NVPTX_TID3]], 5 6931 // CHECK5-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 6932 // CHECK5-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 6933 // CHECK5-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4:[0-9]+]], i32 [[TMP2]]) 6934 // CHECK5-NEXT: [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 6935 // CHECK5-NEXT: br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]] 6936 // CHECK5: then: 6937 // CHECK5-NEXT: [[TMP5:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 6938 // CHECK5-NEXT: [[TMP6:%.*]] = load i8*, i8** [[TMP5]], align 4 6939 // CHECK5-NEXT: [[TMP7:%.*]] = bitcast i8* [[TMP6]] to i32* 6940 // CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 6941 // CHECK5-NEXT: [[TMP9:%.*]] = load i32, i32* [[TMP7]], align 4 6942 // CHECK5-NEXT: store volatile i32 [[TMP9]], i32 addrspace(3)* [[TMP8]], align 4 6943 // CHECK5-NEXT: br label [[IFCONT:%.*]] 6944 // CHECK5: else: 6945 // CHECK5-NEXT: br label [[IFCONT]] 6946 // CHECK5: ifcont: 6947 // CHECK5-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 6948 // CHECK5-NEXT: [[TMP10:%.*]] = load i32, i32* [[DOTADDR1]], align 4 6949 // CHECK5-NEXT: [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP10]] 6950 // CHECK5-NEXT: br i1 [[IS_ACTIVE_THREAD]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 6951 // CHECK5: then4: 6952 // CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 6953 // CHECK5-NEXT: [[TMP12:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 6954 // CHECK5-NEXT: [[TMP13:%.*]] = load i8*, i8** [[TMP12]], align 4 6955 // CHECK5-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to i32* 6956 // CHECK5-NEXT: [[TMP15:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP11]], align 4 6957 // CHECK5-NEXT: store i32 [[TMP15]], i32* [[TMP14]], align 4 6958 // CHECK5-NEXT: br label [[IFCONT6:%.*]] 6959 // CHECK5: else5: 6960 // CHECK5-NEXT: br label [[IFCONT6]] 6961 // CHECK5: ifcont6: 6962 // CHECK5-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 6963 // CHECK5-NEXT: [[WARP_MASTER7:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 6964 // CHECK5-NEXT: br i1 [[WARP_MASTER7]], label [[THEN8:%.*]], label [[ELSE9:%.*]] 6965 // CHECK5: then8: 6966 // CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 6967 // CHECK5-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 4 6968 // CHECK5-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i16* 6969 // CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 6970 // CHECK5-NEXT: [[TMP20:%.*]] = bitcast i32 addrspace(3)* [[TMP19]] to i16 addrspace(3)* 6971 // CHECK5-NEXT: [[TMP21:%.*]] = load i16, i16* [[TMP18]], align 2 6972 // CHECK5-NEXT: store volatile i16 [[TMP21]], i16 addrspace(3)* [[TMP20]], align 2 6973 // CHECK5-NEXT: br label [[IFCONT10:%.*]] 6974 // CHECK5: else9: 6975 // CHECK5-NEXT: br label [[IFCONT10]] 6976 // CHECK5: ifcont10: 6977 // CHECK5-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 6978 // CHECK5-NEXT: [[TMP22:%.*]] = load i32, i32* [[DOTADDR1]], align 4 6979 // CHECK5-NEXT: [[IS_ACTIVE_THREAD11:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP22]] 6980 // CHECK5-NEXT: br i1 [[IS_ACTIVE_THREAD11]], label [[THEN12:%.*]], label [[ELSE13:%.*]] 6981 // CHECK5: then12: 6982 // CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 6983 // CHECK5-NEXT: [[TMP24:%.*]] = bitcast i32 addrspace(3)* [[TMP23]] to i16 addrspace(3)* 6984 // CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 6985 // CHECK5-NEXT: [[TMP26:%.*]] = load i8*, i8** [[TMP25]], align 4 6986 // CHECK5-NEXT: [[TMP27:%.*]] = bitcast i8* [[TMP26]] to i16* 6987 // CHECK5-NEXT: [[TMP28:%.*]] = load volatile i16, i16 addrspace(3)* [[TMP24]], align 2 6988 // CHECK5-NEXT: store i16 [[TMP28]], i16* [[TMP27]], align 2 6989 // CHECK5-NEXT: br label [[IFCONT14:%.*]] 6990 // CHECK5: else13: 6991 // CHECK5-NEXT: br label [[IFCONT14]] 6992 // CHECK5: ifcont14: 6993 // CHECK5-NEXT: ret void 6994 // 6995 // 6996 // CHECK5-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func15 6997 // CHECK5-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR0]] { 6998 // CHECK5-NEXT: entry: 6999 // CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 7000 // CHECK5-NEXT: [[DOTADDR1:%.*]] = alloca i16, align 2 7001 // CHECK5-NEXT: [[DOTADDR2:%.*]] = alloca i16, align 2 7002 // CHECK5-NEXT: [[DOTADDR3:%.*]] = alloca i16, align 2 7003 // CHECK5-NEXT: [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [2 x i8*], align 4 7004 // CHECK5-NEXT: [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca i32, align 4 7005 // CHECK5-NEXT: [[DOTOMP_REDUCTION_ELEMENT4:%.*]] = alloca i16, align 2 7006 // CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 7007 // CHECK5-NEXT: store i16 [[TMP1]], i16* [[DOTADDR1]], align 2 7008 // CHECK5-NEXT: store i16 [[TMP2]], i16* [[DOTADDR2]], align 2 7009 // CHECK5-NEXT: store i16 [[TMP3]], i16* [[DOTADDR3]], align 2 7010 // CHECK5-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 4 7011 // CHECK5-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [2 x i8*]* 7012 // CHECK5-NEXT: [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2 7013 // CHECK5-NEXT: [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2 7014 // CHECK5-NEXT: [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2 7015 // CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 0 7016 // CHECK5-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 4 7017 // CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 7018 // CHECK5-NEXT: [[TMP12:%.*]] = bitcast i8* [[TMP10]] to i32* 7019 // CHECK5-NEXT: [[TMP13:%.*]] = getelementptr i32, i32* [[TMP12]], i32 1 7020 // CHECK5-NEXT: [[TMP14:%.*]] = bitcast i32* [[TMP13]] to i8* 7021 // CHECK5-NEXT: [[TMP15:%.*]] = load i32, i32* [[TMP12]], align 4 7022 // CHECK5-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 7023 // CHECK5-NEXT: [[TMP16:%.*]] = trunc i32 [[NVPTX_WARP_SIZE]] to i16 7024 // CHECK5-NEXT: [[TMP17:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP15]], i16 [[TMP7]], i16 [[TMP16]]) 7025 // CHECK5-NEXT: store i32 [[TMP17]], i32* [[DOTOMP_REDUCTION_ELEMENT]], align 4 7026 // CHECK5-NEXT: [[TMP18:%.*]] = getelementptr i32, i32* [[TMP12]], i32 1 7027 // CHECK5-NEXT: [[TMP19:%.*]] = getelementptr i32, i32* [[DOTOMP_REDUCTION_ELEMENT]], i32 1 7028 // CHECK5-NEXT: [[TMP20:%.*]] = bitcast i32* [[DOTOMP_REDUCTION_ELEMENT]] to i8* 7029 // CHECK5-NEXT: store i8* [[TMP20]], i8** [[TMP11]], align 4 7030 // CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 1 7031 // CHECK5-NEXT: [[TMP22:%.*]] = load i8*, i8** [[TMP21]], align 4 7032 // CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1 7033 // CHECK5-NEXT: [[TMP24:%.*]] = bitcast i8* [[TMP22]] to i16* 7034 // CHECK5-NEXT: [[TMP25:%.*]] = getelementptr i16, i16* [[TMP24]], i32 1 7035 // CHECK5-NEXT: [[TMP26:%.*]] = bitcast i16* [[TMP25]] to i8* 7036 // CHECK5-NEXT: [[TMP27:%.*]] = load i16, i16* [[TMP24]], align 2 7037 // CHECK5-NEXT: [[TMP28:%.*]] = sext i16 [[TMP27]] to i32 7038 // CHECK5-NEXT: [[NVPTX_WARP_SIZE5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 7039 // CHECK5-NEXT: [[TMP29:%.*]] = trunc i32 [[NVPTX_WARP_SIZE5]] to i16 7040 // CHECK5-NEXT: [[TMP30:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP28]], i16 [[TMP7]], i16 [[TMP29]]) 7041 // CHECK5-NEXT: [[TMP31:%.*]] = trunc i32 [[TMP30]] to i16 7042 // CHECK5-NEXT: store i16 [[TMP31]], i16* [[DOTOMP_REDUCTION_ELEMENT4]], align 2 7043 // CHECK5-NEXT: [[TMP32:%.*]] = getelementptr i16, i16* [[TMP24]], i32 1 7044 // CHECK5-NEXT: [[TMP33:%.*]] = getelementptr i16, i16* [[DOTOMP_REDUCTION_ELEMENT4]], i32 1 7045 // CHECK5-NEXT: [[TMP34:%.*]] = bitcast i16* [[DOTOMP_REDUCTION_ELEMENT4]] to i8* 7046 // CHECK5-NEXT: store i8* [[TMP34]], i8** [[TMP23]], align 4 7047 // CHECK5-NEXT: [[TMP35:%.*]] = icmp eq i16 [[TMP8]], 0 7048 // CHECK5-NEXT: [[TMP36:%.*]] = icmp eq i16 [[TMP8]], 1 7049 // CHECK5-NEXT: [[TMP37:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]] 7050 // CHECK5-NEXT: [[TMP38:%.*]] = and i1 [[TMP36]], [[TMP37]] 7051 // CHECK5-NEXT: [[TMP39:%.*]] = icmp eq i16 [[TMP8]], 2 7052 // CHECK5-NEXT: [[TMP40:%.*]] = and i16 [[TMP6]], 1 7053 // CHECK5-NEXT: [[TMP41:%.*]] = icmp eq i16 [[TMP40]], 0 7054 // CHECK5-NEXT: [[TMP42:%.*]] = and i1 [[TMP39]], [[TMP41]] 7055 // CHECK5-NEXT: [[TMP43:%.*]] = icmp sgt i16 [[TMP7]], 0 7056 // CHECK5-NEXT: [[TMP44:%.*]] = and i1 [[TMP42]], [[TMP43]] 7057 // CHECK5-NEXT: [[TMP45:%.*]] = or i1 [[TMP35]], [[TMP38]] 7058 // CHECK5-NEXT: [[TMP46:%.*]] = or i1 [[TMP45]], [[TMP44]] 7059 // CHECK5-NEXT: br i1 [[TMP46]], label [[THEN:%.*]], label [[ELSE:%.*]] 7060 // CHECK5: then: 7061 // CHECK5-NEXT: [[TMP47:%.*]] = bitcast [2 x i8*]* [[TMP5]] to i8* 7062 // CHECK5-NEXT: [[TMP48:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8* 7063 // CHECK5-NEXT: call void @"_omp$reduction$reduction_func14"(i8* [[TMP47]], i8* [[TMP48]]) #[[ATTR3]] 7064 // CHECK5-NEXT: br label [[IFCONT:%.*]] 7065 // CHECK5: else: 7066 // CHECK5-NEXT: br label [[IFCONT]] 7067 // CHECK5: ifcont: 7068 // CHECK5-NEXT: [[TMP49:%.*]] = icmp eq i16 [[TMP8]], 1 7069 // CHECK5-NEXT: [[TMP50:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]] 7070 // CHECK5-NEXT: [[TMP51:%.*]] = and i1 [[TMP49]], [[TMP50]] 7071 // CHECK5-NEXT: br i1 [[TMP51]], label [[THEN6:%.*]], label [[ELSE7:%.*]] 7072 // CHECK5: then6: 7073 // CHECK5-NEXT: [[TMP52:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 7074 // CHECK5-NEXT: [[TMP53:%.*]] = load i8*, i8** [[TMP52]], align 4 7075 // CHECK5-NEXT: [[TMP54:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 0 7076 // CHECK5-NEXT: [[TMP55:%.*]] = load i8*, i8** [[TMP54]], align 4 7077 // CHECK5-NEXT: [[TMP56:%.*]] = bitcast i8* [[TMP53]] to i32* 7078 // CHECK5-NEXT: [[TMP57:%.*]] = bitcast i8* [[TMP55]] to i32* 7079 // CHECK5-NEXT: [[TMP58:%.*]] = load i32, i32* [[TMP56]], align 4 7080 // CHECK5-NEXT: store i32 [[TMP58]], i32* [[TMP57]], align 4 7081 // CHECK5-NEXT: [[TMP59:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1 7082 // CHECK5-NEXT: [[TMP60:%.*]] = load i8*, i8** [[TMP59]], align 4 7083 // CHECK5-NEXT: [[TMP61:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 1 7084 // CHECK5-NEXT: [[TMP62:%.*]] = load i8*, i8** [[TMP61]], align 4 7085 // CHECK5-NEXT: [[TMP63:%.*]] = bitcast i8* [[TMP60]] to i16* 7086 // CHECK5-NEXT: [[TMP64:%.*]] = bitcast i8* [[TMP62]] to i16* 7087 // CHECK5-NEXT: [[TMP65:%.*]] = load i16, i16* [[TMP63]], align 2 7088 // CHECK5-NEXT: store i16 [[TMP65]], i16* [[TMP64]], align 2 7089 // CHECK5-NEXT: br label [[IFCONT8:%.*]] 7090 // CHECK5: else7: 7091 // CHECK5-NEXT: br label [[IFCONT8]] 7092 // CHECK5: ifcont8: 7093 // CHECK5-NEXT: ret void 7094 // 7095 // 7096 // CHECK5-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func16 7097 // CHECK5-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] { 7098 // CHECK5-NEXT: entry: 7099 // CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 7100 // CHECK5-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 7101 // CHECK5-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]]) 7102 // CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 7103 // CHECK5-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 7104 // CHECK5-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 7105 // CHECK5-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 7106 // CHECK5-NEXT: [[NVPTX_LANE_ID:%.*]] = and i32 [[NVPTX_TID2]], 31 7107 // CHECK5-NEXT: [[NVPTX_TID3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 7108 // CHECK5-NEXT: [[NVPTX_WARP_ID:%.*]] = ashr i32 [[NVPTX_TID3]], 5 7109 // CHECK5-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 7110 // CHECK5-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 7111 // CHECK5-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 7112 // CHECK5-NEXT: [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 7113 // CHECK5-NEXT: br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]] 7114 // CHECK5: then: 7115 // CHECK5-NEXT: [[TMP5:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 7116 // CHECK5-NEXT: [[TMP6:%.*]] = load i8*, i8** [[TMP5]], align 4 7117 // CHECK5-NEXT: [[TMP7:%.*]] = bitcast i8* [[TMP6]] to i32* 7118 // CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 7119 // CHECK5-NEXT: [[TMP9:%.*]] = load i32, i32* [[TMP7]], align 4 7120 // CHECK5-NEXT: store volatile i32 [[TMP9]], i32 addrspace(3)* [[TMP8]], align 4 7121 // CHECK5-NEXT: br label [[IFCONT:%.*]] 7122 // CHECK5: else: 7123 // CHECK5-NEXT: br label [[IFCONT]] 7124 // CHECK5: ifcont: 7125 // CHECK5-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 7126 // CHECK5-NEXT: [[TMP10:%.*]] = load i32, i32* [[DOTADDR1]], align 4 7127 // CHECK5-NEXT: [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP10]] 7128 // CHECK5-NEXT: br i1 [[IS_ACTIVE_THREAD]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 7129 // CHECK5: then4: 7130 // CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 7131 // CHECK5-NEXT: [[TMP12:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 7132 // CHECK5-NEXT: [[TMP13:%.*]] = load i8*, i8** [[TMP12]], align 4 7133 // CHECK5-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to i32* 7134 // CHECK5-NEXT: [[TMP15:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP11]], align 4 7135 // CHECK5-NEXT: store i32 [[TMP15]], i32* [[TMP14]], align 4 7136 // CHECK5-NEXT: br label [[IFCONT6:%.*]] 7137 // CHECK5: else5: 7138 // CHECK5-NEXT: br label [[IFCONT6]] 7139 // CHECK5: ifcont6: 7140 // CHECK5-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 7141 // CHECK5-NEXT: [[WARP_MASTER7:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 7142 // CHECK5-NEXT: br i1 [[WARP_MASTER7]], label [[THEN8:%.*]], label [[ELSE9:%.*]] 7143 // CHECK5: then8: 7144 // CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 7145 // CHECK5-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 4 7146 // CHECK5-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i16* 7147 // CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 7148 // CHECK5-NEXT: [[TMP20:%.*]] = bitcast i32 addrspace(3)* [[TMP19]] to i16 addrspace(3)* 7149 // CHECK5-NEXT: [[TMP21:%.*]] = load i16, i16* [[TMP18]], align 2 7150 // CHECK5-NEXT: store volatile i16 [[TMP21]], i16 addrspace(3)* [[TMP20]], align 2 7151 // CHECK5-NEXT: br label [[IFCONT10:%.*]] 7152 // CHECK5: else9: 7153 // CHECK5-NEXT: br label [[IFCONT10]] 7154 // CHECK5: ifcont10: 7155 // CHECK5-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 7156 // CHECK5-NEXT: [[TMP22:%.*]] = load i32, i32* [[DOTADDR1]], align 4 7157 // CHECK5-NEXT: [[IS_ACTIVE_THREAD11:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP22]] 7158 // CHECK5-NEXT: br i1 [[IS_ACTIVE_THREAD11]], label [[THEN12:%.*]], label [[ELSE13:%.*]] 7159 // CHECK5: then12: 7160 // CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 7161 // CHECK5-NEXT: [[TMP24:%.*]] = bitcast i32 addrspace(3)* [[TMP23]] to i16 addrspace(3)* 7162 // CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 7163 // CHECK5-NEXT: [[TMP26:%.*]] = load i8*, i8** [[TMP25]], align 4 7164 // CHECK5-NEXT: [[TMP27:%.*]] = bitcast i8* [[TMP26]] to i16* 7165 // CHECK5-NEXT: [[TMP28:%.*]] = load volatile i16, i16 addrspace(3)* [[TMP24]], align 2 7166 // CHECK5-NEXT: store i16 [[TMP28]], i16* [[TMP27]], align 2 7167 // CHECK5-NEXT: br label [[IFCONT14:%.*]] 7168 // CHECK5: else13: 7169 // CHECK5-NEXT: br label [[IFCONT14]] 7170 // CHECK5: ifcont14: 7171 // CHECK5-NEXT: ret void 7172 // 7173 // 7174 // CHECK5-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_copy_func17 7175 // CHECK5-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 7176 // CHECK5-NEXT: entry: 7177 // CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 7178 // CHECK5-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 7179 // CHECK5-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 7180 // CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 7181 // CHECK5-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 7182 // CHECK5-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 7183 // CHECK5-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 7184 // CHECK5-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 7185 // CHECK5-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 4 7186 // CHECK5-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.4* 7187 // CHECK5-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 7188 // CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 7189 // CHECK5-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 4 7190 // CHECK5-NEXT: [[TMP10:%.*]] = bitcast i8* [[TMP9]] to i32* 7191 // CHECK5-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4:%.*]], %struct._globalized_locals_ty.4* [[TMP6]], i32 0, i32 0 7192 // CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [1024 x i32], [1024 x i32]* [[A]], i32 0, i32 [[TMP7]] 7193 // CHECK5-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP10]], align 4 7194 // CHECK5-NEXT: store i32 [[TMP12]], i32* [[TMP11]], align 128 7195 // CHECK5-NEXT: [[TMP13:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 7196 // CHECK5-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 4 7197 // CHECK5-NEXT: [[TMP15:%.*]] = bitcast i8* [[TMP14]] to i16* 7198 // CHECK5-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4]], %struct._globalized_locals_ty.4* [[TMP6]], i32 0, i32 1 7199 // CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1024 x i16], [1024 x i16]* [[B]], i32 0, i32 [[TMP7]] 7200 // CHECK5-NEXT: [[TMP17:%.*]] = load i16, i16* [[TMP15]], align 2 7201 // CHECK5-NEXT: store i16 [[TMP17]], i16* [[TMP16]], align 128 7202 // CHECK5-NEXT: ret void 7203 // 7204 // 7205 // CHECK5-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_reduce_func18 7206 // CHECK5-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 7207 // CHECK5-NEXT: entry: 7208 // CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 7209 // CHECK5-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 7210 // CHECK5-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 7211 // CHECK5-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 7212 // CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 7213 // CHECK5-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 7214 // CHECK5-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 7215 // CHECK5-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 7216 // CHECK5-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.4* 7217 // CHECK5-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 7218 // CHECK5-NEXT: [[TMP6:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 7219 // CHECK5-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4:%.*]], %struct._globalized_locals_ty.4* [[TMP4]], i32 0, i32 0 7220 // CHECK5-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1024 x i32], [1024 x i32]* [[A]], i32 0, i32 [[TMP5]] 7221 // CHECK5-NEXT: [[TMP8:%.*]] = bitcast i32* [[TMP7]] to i8* 7222 // CHECK5-NEXT: store i8* [[TMP8]], i8** [[TMP6]], align 4 7223 // CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 7224 // CHECK5-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4]], %struct._globalized_locals_ty.4* [[TMP4]], i32 0, i32 1 7225 // CHECK5-NEXT: [[TMP10:%.*]] = getelementptr inbounds [1024 x i16], [1024 x i16]* [[B]], i32 0, i32 [[TMP5]] 7226 // CHECK5-NEXT: [[TMP11:%.*]] = bitcast i16* [[TMP10]] to i8* 7227 // CHECK5-NEXT: store i8* [[TMP11]], i8** [[TMP9]], align 4 7228 // CHECK5-NEXT: [[TMP12:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 7229 // CHECK5-NEXT: [[TMP13:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 7230 // CHECK5-NEXT: call void @"_omp$reduction$reduction_func14"(i8* [[TMP12]], i8* [[TMP13]]) #[[ATTR3]] 7231 // CHECK5-NEXT: ret void 7232 // 7233 // 7234 // CHECK5-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_copy_func19 7235 // CHECK5-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 7236 // CHECK5-NEXT: entry: 7237 // CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 7238 // CHECK5-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 7239 // CHECK5-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 7240 // CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 7241 // CHECK5-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 7242 // CHECK5-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 7243 // CHECK5-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 7244 // CHECK5-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 7245 // CHECK5-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 4 7246 // CHECK5-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.4* 7247 // CHECK5-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 7248 // CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 7249 // CHECK5-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 4 7250 // CHECK5-NEXT: [[TMP10:%.*]] = bitcast i8* [[TMP9]] to i32* 7251 // CHECK5-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4:%.*]], %struct._globalized_locals_ty.4* [[TMP6]], i32 0, i32 0 7252 // CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [1024 x i32], [1024 x i32]* [[A]], i32 0, i32 [[TMP7]] 7253 // CHECK5-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 128 7254 // CHECK5-NEXT: store i32 [[TMP12]], i32* [[TMP10]], align 4 7255 // CHECK5-NEXT: [[TMP13:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 7256 // CHECK5-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 4 7257 // CHECK5-NEXT: [[TMP15:%.*]] = bitcast i8* [[TMP14]] to i16* 7258 // CHECK5-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4]], %struct._globalized_locals_ty.4* [[TMP6]], i32 0, i32 1 7259 // CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1024 x i16], [1024 x i16]* [[B]], i32 0, i32 [[TMP7]] 7260 // CHECK5-NEXT: [[TMP17:%.*]] = load i16, i16* [[TMP16]], align 128 7261 // CHECK5-NEXT: store i16 [[TMP17]], i16* [[TMP15]], align 2 7262 // CHECK5-NEXT: ret void 7263 // 7264 // 7265 // CHECK5-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_reduce_func20 7266 // CHECK5-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 7267 // CHECK5-NEXT: entry: 7268 // CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 7269 // CHECK5-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 7270 // CHECK5-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 7271 // CHECK5-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 7272 // CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 7273 // CHECK5-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 7274 // CHECK5-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 7275 // CHECK5-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 7276 // CHECK5-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.4* 7277 // CHECK5-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 7278 // CHECK5-NEXT: [[TMP6:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 7279 // CHECK5-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4:%.*]], %struct._globalized_locals_ty.4* [[TMP4]], i32 0, i32 0 7280 // CHECK5-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1024 x i32], [1024 x i32]* [[A]], i32 0, i32 [[TMP5]] 7281 // CHECK5-NEXT: [[TMP8:%.*]] = bitcast i32* [[TMP7]] to i8* 7282 // CHECK5-NEXT: store i8* [[TMP8]], i8** [[TMP6]], align 4 7283 // CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 7284 // CHECK5-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4]], %struct._globalized_locals_ty.4* [[TMP4]], i32 0, i32 1 7285 // CHECK5-NEXT: [[TMP10:%.*]] = getelementptr inbounds [1024 x i16], [1024 x i16]* [[B]], i32 0, i32 [[TMP5]] 7286 // CHECK5-NEXT: [[TMP11:%.*]] = bitcast i16* [[TMP10]] to i8* 7287 // CHECK5-NEXT: store i8* [[TMP11]], i8** [[TMP9]], align 4 7288 // CHECK5-NEXT: [[TMP12:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 7289 // CHECK5-NEXT: [[TMP13:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 7290 // CHECK5-NEXT: call void @"_omp$reduction$reduction_func14"(i8* [[TMP13]], i8* [[TMP12]]) #[[ATTR3]] 7291 // CHECK5-NEXT: ret void 7292 // 7293 // 7294 // CHECK6-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l23_worker 7295 // CHECK6-SAME: () #[[ATTR0:[0-9]+]] { 7296 // CHECK6-NEXT: entry: 7297 // CHECK6-NEXT: [[WORK_FN:%.*]] = alloca i8*, align 4 7298 // CHECK6-NEXT: [[EXEC_STATUS:%.*]] = alloca i8, align 1 7299 // CHECK6-NEXT: store i8* null, i8** [[WORK_FN]], align 4 7300 // CHECK6-NEXT: store i8 0, i8* [[EXEC_STATUS]], align 1 7301 // CHECK6-NEXT: br label [[DOTAWAIT_WORK:%.*]] 7302 // CHECK6: .await.work: 7303 // CHECK6-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 7304 // CHECK6-NEXT: [[TMP0:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORK_FN]]) 7305 // CHECK6-NEXT: [[TMP1:%.*]] = zext i1 [[TMP0]] to i8 7306 // CHECK6-NEXT: store i8 [[TMP1]], i8* [[EXEC_STATUS]], align 1 7307 // CHECK6-NEXT: [[TMP2:%.*]] = load i8*, i8** [[WORK_FN]], align 4 7308 // CHECK6-NEXT: [[SHOULD_TERMINATE:%.*]] = icmp eq i8* [[TMP2]], null 7309 // CHECK6-NEXT: br i1 [[SHOULD_TERMINATE]], label [[DOTEXIT:%.*]], label [[DOTSELECT_WORKERS:%.*]] 7310 // CHECK6: .select.workers: 7311 // CHECK6-NEXT: [[TMP3:%.*]] = load i8, i8* [[EXEC_STATUS]], align 1 7312 // CHECK6-NEXT: [[IS_ACTIVE:%.*]] = icmp ne i8 [[TMP3]], 0 7313 // CHECK6-NEXT: br i1 [[IS_ACTIVE]], label [[DOTEXECUTE_PARALLEL:%.*]], label [[DOTBARRIER_PARALLEL:%.*]] 7314 // CHECK6: .execute.parallel: 7315 // CHECK6-NEXT: [[TMP4:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]]) 7316 // CHECK6-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP2]] to void (i16, i32)* 7317 // CHECK6-NEXT: call void [[TMP5]](i16 0, i32 [[TMP4]]) 7318 // CHECK6-NEXT: br label [[DOTTERMINATE_PARALLEL:%.*]] 7319 // CHECK6: .terminate.parallel: 7320 // CHECK6-NEXT: call void @__kmpc_kernel_end_parallel() 7321 // CHECK6-NEXT: br label [[DOTBARRIER_PARALLEL]] 7322 // CHECK6: .barrier.parallel: 7323 // CHECK6-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 7324 // CHECK6-NEXT: br label [[DOTAWAIT_WORK]] 7325 // CHECK6: .exit: 7326 // CHECK6-NEXT: ret void 7327 // 7328 // 7329 // CHECK6-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l23 7330 // CHECK6-SAME: (double* nonnull align 8 dereferenceable(8) [[E:%.*]]) #[[ATTR1:[0-9]+]] { 7331 // CHECK6-NEXT: entry: 7332 // CHECK6-NEXT: [[E_ADDR:%.*]] = alloca double*, align 4 7333 // CHECK6-NEXT: [[E7:%.*]] = alloca double, align 8 7334 // CHECK6-NEXT: [[DOTZERO_ADDR:%.*]] = alloca i32, align 4 7335 // CHECK6-NEXT: [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4 7336 // CHECK6-NEXT: store i32 0, i32* [[DOTZERO_ADDR]], align 4 7337 // CHECK6-NEXT: store double* [[E]], double** [[E_ADDR]], align 4 7338 // CHECK6-NEXT: [[TMP0:%.*]] = load double*, double** [[E_ADDR]], align 4 7339 // CHECK6-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 7340 // CHECK6-NEXT: [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 7341 // CHECK6-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 7342 // CHECK6-NEXT: [[THREAD_LIMIT:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS]], [[NVPTX_WARP_SIZE]] 7343 // CHECK6-NEXT: [[TMP1:%.*]] = icmp ult i32 [[NVPTX_TID]], [[THREAD_LIMIT]] 7344 // CHECK6-NEXT: br i1 [[TMP1]], label [[DOTWORKER:%.*]], label [[DOTMASTERCHECK:%.*]] 7345 // CHECK6: .worker: 7346 // CHECK6-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l23_worker() #[[ATTR3:[0-9]+]] 7347 // CHECK6-NEXT: br label [[DOTEXIT:%.*]] 7348 // CHECK6: .mastercheck: 7349 // CHECK6-NEXT: [[NVPTX_TID1:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 7350 // CHECK6-NEXT: [[NVPTX_NUM_THREADS2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 7351 // CHECK6-NEXT: [[NVPTX_WARP_SIZE3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 7352 // CHECK6-NEXT: [[TMP2:%.*]] = sub nuw i32 [[NVPTX_WARP_SIZE3]], 1 7353 // CHECK6-NEXT: [[TMP3:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS2]], 1 7354 // CHECK6-NEXT: [[TMP4:%.*]] = xor i32 [[TMP2]], -1 7355 // CHECK6-NEXT: [[MASTER_TID:%.*]] = and i32 [[TMP3]], [[TMP4]] 7356 // CHECK6-NEXT: [[TMP5:%.*]] = icmp eq i32 [[NVPTX_TID1]], [[MASTER_TID]] 7357 // CHECK6-NEXT: br i1 [[TMP5]], label [[DOTMASTER:%.*]], label [[DOTEXIT]] 7358 // CHECK6: .master: 7359 // CHECK6-NEXT: [[NVPTX_NUM_THREADS4:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 7360 // CHECK6-NEXT: [[NVPTX_WARP_SIZE5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 7361 // CHECK6-NEXT: [[THREAD_LIMIT6:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS4]], [[NVPTX_WARP_SIZE5]] 7362 // CHECK6-NEXT: call void @__kmpc_kernel_init(i32 [[THREAD_LIMIT6]], i16 1) 7363 // CHECK6-NEXT: call void @__kmpc_data_sharing_init_stack() 7364 // CHECK6-NEXT: [[TMP6:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 7365 // CHECK6-NEXT: [[TMP7:%.*]] = load double, double* [[TMP0]], align 8 7366 // CHECK6-NEXT: store double [[TMP7]], double* [[E7]], align 8 7367 // CHECK6-NEXT: store i32 [[TMP6]], i32* [[DOTTHREADID_TEMP_]], align 4 7368 // CHECK6-NEXT: call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], double* [[E7]]) #[[ATTR3]] 7369 // CHECK6-NEXT: br label [[DOTTERMINATION_NOTIFIER:%.*]] 7370 // CHECK6: .termination.notifier: 7371 // CHECK6-NEXT: call void @__kmpc_kernel_deinit(i16 1) 7372 // CHECK6-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 7373 // CHECK6-NEXT: br label [[DOTEXIT]] 7374 // CHECK6: .exit: 7375 // CHECK6-NEXT: ret void 7376 // 7377 // 7378 // CHECK6-LABEL: define {{[^@]+}}@__omp_outlined__ 7379 // CHECK6-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], double* nonnull align 8 dereferenceable(8) [[E:%.*]]) #[[ATTR1]] { 7380 // CHECK6-NEXT: entry: 7381 // CHECK6-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4 7382 // CHECK6-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4 7383 // CHECK6-NEXT: [[E_ADDR:%.*]] = alloca double*, align 4 7384 // CHECK6-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x i8*], align 4 7385 // CHECK6-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4 7386 // CHECK6-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4 7387 // CHECK6-NEXT: store double* [[E]], double** [[E_ADDR]], align 4 7388 // CHECK6-NEXT: [[TMP0:%.*]] = load double*, double** [[E_ADDR]], align 4 7389 // CHECK6-NEXT: [[TMP1:%.*]] = call i8* @__kmpc_data_sharing_push_stack(i32 8, i16 1) 7390 // CHECK6-NEXT: [[TMP2:%.*]] = bitcast i8* [[TMP1]] to %struct._globalized_locals_ty* 7391 // CHECK6-NEXT: [[E1:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY:%.*]], %struct._globalized_locals_ty* [[TMP2]], i32 0, i32 0 7392 // CHECK6-NEXT: store double 0.000000e+00, double* [[E1]], align 8 7393 // CHECK6-NEXT: [[TMP3:%.*]] = load double, double* [[E1]], align 8 7394 // CHECK6-NEXT: [[ADD:%.*]] = fadd double [[TMP3]], 5.000000e+00 7395 // CHECK6-NEXT: store double [[ADD]], double* [[E1]], align 8 7396 // CHECK6-NEXT: [[TMP4:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4 7397 // CHECK6-NEXT: [[TMP5:%.*]] = load i32, i32* [[TMP4]], align 4 7398 // CHECK6-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 7399 // CHECK6-NEXT: [[TMP7:%.*]] = bitcast double* [[E1]] to i8* 7400 // CHECK6-NEXT: store i8* [[TMP7]], i8** [[TMP6]], align 4 7401 // CHECK6-NEXT: [[TMP8:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 7402 // CHECK6-NEXT: [[TMP9:%.*]] = load i8*, i8** @"_openmp_teams_reductions_buffer_$_$ptr", align 4 7403 // CHECK6-NEXT: [[TMP10:%.*]] = call i32 @__kmpc_nvptx_teams_reduce_nowait_v2(%struct.ident_t* @[[GLOB1]], i32 [[TMP5]], i8* [[TMP9]], i32 2048, i8* [[TMP8]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_copy_func, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_reduce_func, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_copy_func, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_reduce_func) 7404 // CHECK6-NEXT: [[TMP11:%.*]] = icmp eq i32 [[TMP10]], 1 7405 // CHECK6-NEXT: br i1 [[TMP11]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]] 7406 // CHECK6: .omp.reduction.then: 7407 // CHECK6-NEXT: [[TMP12:%.*]] = load double, double* [[TMP0]], align 8 7408 // CHECK6-NEXT: [[TMP13:%.*]] = load double, double* [[E1]], align 8 7409 // CHECK6-NEXT: [[ADD2:%.*]] = fadd double [[TMP12]], [[TMP13]] 7410 // CHECK6-NEXT: store double [[ADD2]], double* [[TMP0]], align 8 7411 // CHECK6-NEXT: call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP5]]) 7412 // CHECK6-NEXT: br label [[DOTOMP_REDUCTION_DONE]] 7413 // CHECK6: .omp.reduction.done: 7414 // CHECK6-NEXT: call void @__kmpc_data_sharing_pop_stack(i8* [[TMP1]]) 7415 // CHECK6-NEXT: ret void 7416 // 7417 // 7418 // CHECK6-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func 7419 // CHECK6-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR0]] { 7420 // CHECK6-NEXT: entry: 7421 // CHECK6-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 7422 // CHECK6-NEXT: [[DOTADDR1:%.*]] = alloca i16, align 2 7423 // CHECK6-NEXT: [[DOTADDR2:%.*]] = alloca i16, align 2 7424 // CHECK6-NEXT: [[DOTADDR3:%.*]] = alloca i16, align 2 7425 // CHECK6-NEXT: [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [1 x i8*], align 4 7426 // CHECK6-NEXT: [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca double, align 8 7427 // CHECK6-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 7428 // CHECK6-NEXT: store i16 [[TMP1]], i16* [[DOTADDR1]], align 2 7429 // CHECK6-NEXT: store i16 [[TMP2]], i16* [[DOTADDR2]], align 2 7430 // CHECK6-NEXT: store i16 [[TMP3]], i16* [[DOTADDR3]], align 2 7431 // CHECK6-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 4 7432 // CHECK6-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [1 x i8*]* 7433 // CHECK6-NEXT: [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2 7434 // CHECK6-NEXT: [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2 7435 // CHECK6-NEXT: [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2 7436 // CHECK6-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i32 0, i32 0 7437 // CHECK6-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 4 7438 // CHECK6-NEXT: [[TMP11:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 7439 // CHECK6-NEXT: [[TMP12:%.*]] = bitcast i8* [[TMP10]] to double* 7440 // CHECK6-NEXT: [[TMP13:%.*]] = getelementptr double, double* [[TMP12]], i32 1 7441 // CHECK6-NEXT: [[TMP14:%.*]] = bitcast double* [[TMP13]] to i8* 7442 // CHECK6-NEXT: [[TMP15:%.*]] = bitcast double* [[TMP12]] to i64* 7443 // CHECK6-NEXT: [[TMP16:%.*]] = bitcast double* [[DOTOMP_REDUCTION_ELEMENT]] to i64* 7444 // CHECK6-NEXT: [[TMP17:%.*]] = load i64, i64* [[TMP15]], align 8 7445 // CHECK6-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 7446 // CHECK6-NEXT: [[TMP18:%.*]] = trunc i32 [[NVPTX_WARP_SIZE]] to i16 7447 // CHECK6-NEXT: [[TMP19:%.*]] = call i64 @__kmpc_shuffle_int64(i64 [[TMP17]], i16 [[TMP7]], i16 [[TMP18]]) 7448 // CHECK6-NEXT: store i64 [[TMP19]], i64* [[TMP16]], align 8 7449 // CHECK6-NEXT: [[TMP20:%.*]] = getelementptr i64, i64* [[TMP15]], i32 1 7450 // CHECK6-NEXT: [[TMP21:%.*]] = getelementptr i64, i64* [[TMP16]], i32 1 7451 // CHECK6-NEXT: [[TMP22:%.*]] = bitcast double* [[DOTOMP_REDUCTION_ELEMENT]] to i8* 7452 // CHECK6-NEXT: store i8* [[TMP22]], i8** [[TMP11]], align 4 7453 // CHECK6-NEXT: [[TMP23:%.*]] = icmp eq i16 [[TMP8]], 0 7454 // CHECK6-NEXT: [[TMP24:%.*]] = icmp eq i16 [[TMP8]], 1 7455 // CHECK6-NEXT: [[TMP25:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]] 7456 // CHECK6-NEXT: [[TMP26:%.*]] = and i1 [[TMP24]], [[TMP25]] 7457 // CHECK6-NEXT: [[TMP27:%.*]] = icmp eq i16 [[TMP8]], 2 7458 // CHECK6-NEXT: [[TMP28:%.*]] = and i16 [[TMP6]], 1 7459 // CHECK6-NEXT: [[TMP29:%.*]] = icmp eq i16 [[TMP28]], 0 7460 // CHECK6-NEXT: [[TMP30:%.*]] = and i1 [[TMP27]], [[TMP29]] 7461 // CHECK6-NEXT: [[TMP31:%.*]] = icmp sgt i16 [[TMP7]], 0 7462 // CHECK6-NEXT: [[TMP32:%.*]] = and i1 [[TMP30]], [[TMP31]] 7463 // CHECK6-NEXT: [[TMP33:%.*]] = or i1 [[TMP23]], [[TMP26]] 7464 // CHECK6-NEXT: [[TMP34:%.*]] = or i1 [[TMP33]], [[TMP32]] 7465 // CHECK6-NEXT: br i1 [[TMP34]], label [[THEN:%.*]], label [[ELSE:%.*]] 7466 // CHECK6: then: 7467 // CHECK6-NEXT: [[TMP35:%.*]] = bitcast [1 x i8*]* [[TMP5]] to i8* 7468 // CHECK6-NEXT: [[TMP36:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8* 7469 // CHECK6-NEXT: call void @"_omp$reduction$reduction_func"(i8* [[TMP35]], i8* [[TMP36]]) #[[ATTR3]] 7470 // CHECK6-NEXT: br label [[IFCONT:%.*]] 7471 // CHECK6: else: 7472 // CHECK6-NEXT: br label [[IFCONT]] 7473 // CHECK6: ifcont: 7474 // CHECK6-NEXT: [[TMP37:%.*]] = icmp eq i16 [[TMP8]], 1 7475 // CHECK6-NEXT: [[TMP38:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]] 7476 // CHECK6-NEXT: [[TMP39:%.*]] = and i1 [[TMP37]], [[TMP38]] 7477 // CHECK6-NEXT: br i1 [[TMP39]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 7478 // CHECK6: then4: 7479 // CHECK6-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 7480 // CHECK6-NEXT: [[TMP41:%.*]] = load i8*, i8** [[TMP40]], align 4 7481 // CHECK6-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i32 0, i32 0 7482 // CHECK6-NEXT: [[TMP43:%.*]] = load i8*, i8** [[TMP42]], align 4 7483 // CHECK6-NEXT: [[TMP44:%.*]] = bitcast i8* [[TMP41]] to double* 7484 // CHECK6-NEXT: [[TMP45:%.*]] = bitcast i8* [[TMP43]] to double* 7485 // CHECK6-NEXT: [[TMP46:%.*]] = load double, double* [[TMP44]], align 8 7486 // CHECK6-NEXT: store double [[TMP46]], double* [[TMP45]], align 8 7487 // CHECK6-NEXT: br label [[IFCONT6:%.*]] 7488 // CHECK6: else5: 7489 // CHECK6-NEXT: br label [[IFCONT6]] 7490 // CHECK6: ifcont6: 7491 // CHECK6-NEXT: ret void 7492 // 7493 // 7494 // CHECK6-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func 7495 // CHECK6-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] { 7496 // CHECK6-NEXT: entry: 7497 // CHECK6-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 7498 // CHECK6-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 7499 // CHECK6-NEXT: [[DOTCNT_ADDR:%.*]] = alloca i32, align 4 7500 // CHECK6-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 7501 // CHECK6-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 7502 // CHECK6-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 7503 // CHECK6-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 7504 // CHECK6-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 7505 // CHECK6-NEXT: [[NVPTX_LANE_ID:%.*]] = and i32 [[NVPTX_TID2]], 31 7506 // CHECK6-NEXT: [[NVPTX_TID3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 7507 // CHECK6-NEXT: [[NVPTX_WARP_ID:%.*]] = ashr i32 [[NVPTX_TID3]], 5 7508 // CHECK6-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 7509 // CHECK6-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [1 x i8*]* 7510 // CHECK6-NEXT: store i32 0, i32* [[DOTCNT_ADDR]], align 4 7511 // CHECK6-NEXT: br label [[PRECOND:%.*]] 7512 // CHECK6: precond: 7513 // CHECK6-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTCNT_ADDR]], align 4 7514 // CHECK6-NEXT: [[TMP6:%.*]] = icmp ult i32 [[TMP5]], 2 7515 // CHECK6-NEXT: br i1 [[TMP6]], label [[BODY:%.*]], label [[EXIT:%.*]] 7516 // CHECK6: body: 7517 // CHECK6-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP2]]) 7518 // CHECK6-NEXT: [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 7519 // CHECK6-NEXT: br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]] 7520 // CHECK6: then: 7521 // CHECK6-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP4]], i32 0, i32 0 7522 // CHECK6-NEXT: [[TMP8:%.*]] = load i8*, i8** [[TMP7]], align 4 7523 // CHECK6-NEXT: [[TMP9:%.*]] = bitcast i8* [[TMP8]] to i32* 7524 // CHECK6-NEXT: [[TMP10:%.*]] = getelementptr i32, i32* [[TMP9]], i32 [[TMP5]] 7525 // CHECK6-NEXT: [[TMP11:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 7526 // CHECK6-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP10]], align 4 7527 // CHECK6-NEXT: store volatile i32 [[TMP12]], i32 addrspace(3)* [[TMP11]], align 4 7528 // CHECK6-NEXT: br label [[IFCONT:%.*]] 7529 // CHECK6: else: 7530 // CHECK6-NEXT: br label [[IFCONT]] 7531 // CHECK6: ifcont: 7532 // CHECK6-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 7533 // CHECK6-NEXT: [[TMP13:%.*]] = load i32, i32* [[DOTADDR1]], align 4 7534 // CHECK6-NEXT: [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP13]] 7535 // CHECK6-NEXT: br i1 [[IS_ACTIVE_THREAD]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 7536 // CHECK6: then4: 7537 // CHECK6-NEXT: [[TMP14:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 7538 // CHECK6-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP4]], i32 0, i32 0 7539 // CHECK6-NEXT: [[TMP16:%.*]] = load i8*, i8** [[TMP15]], align 4 7540 // CHECK6-NEXT: [[TMP17:%.*]] = bitcast i8* [[TMP16]] to i32* 7541 // CHECK6-NEXT: [[TMP18:%.*]] = getelementptr i32, i32* [[TMP17]], i32 [[TMP5]] 7542 // CHECK6-NEXT: [[TMP19:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP14]], align 4 7543 // CHECK6-NEXT: store i32 [[TMP19]], i32* [[TMP18]], align 4 7544 // CHECK6-NEXT: br label [[IFCONT6:%.*]] 7545 // CHECK6: else5: 7546 // CHECK6-NEXT: br label [[IFCONT6]] 7547 // CHECK6: ifcont6: 7548 // CHECK6-NEXT: [[TMP20:%.*]] = add nsw i32 [[TMP5]], 1 7549 // CHECK6-NEXT: store i32 [[TMP20]], i32* [[DOTCNT_ADDR]], align 4 7550 // CHECK6-NEXT: br label [[PRECOND]] 7551 // CHECK6: exit: 7552 // CHECK6-NEXT: ret void 7553 // 7554 // 7555 // CHECK6-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_copy_func 7556 // CHECK6-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 7557 // CHECK6-NEXT: entry: 7558 // CHECK6-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 7559 // CHECK6-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 7560 // CHECK6-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 7561 // CHECK6-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 7562 // CHECK6-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 7563 // CHECK6-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 7564 // CHECK6-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 7565 // CHECK6-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [1 x i8*]* 7566 // CHECK6-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 4 7567 // CHECK6-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.0* 7568 // CHECK6-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 7569 // CHECK6-NEXT: [[TMP8:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP4]], i32 0, i32 0 7570 // CHECK6-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 4 7571 // CHECK6-NEXT: [[TMP10:%.*]] = bitcast i8* [[TMP9]] to double* 7572 // CHECK6-NEXT: [[E:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_0:%.*]], %struct._globalized_locals_ty.0* [[TMP6]], i32 0, i32 0 7573 // CHECK6-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2048 x double], [2048 x double]* [[E]], i32 0, i32 [[TMP7]] 7574 // CHECK6-NEXT: [[TMP12:%.*]] = load double, double* [[TMP10]], align 8 7575 // CHECK6-NEXT: store double [[TMP12]], double* [[TMP11]], align 128 7576 // CHECK6-NEXT: ret void 7577 // 7578 // 7579 // CHECK6-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_reduce_func 7580 // CHECK6-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 7581 // CHECK6-NEXT: entry: 7582 // CHECK6-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 7583 // CHECK6-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 7584 // CHECK6-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 7585 // CHECK6-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x i8*], align 4 7586 // CHECK6-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 7587 // CHECK6-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 7588 // CHECK6-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 7589 // CHECK6-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 7590 // CHECK6-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.0* 7591 // CHECK6-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 7592 // CHECK6-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 7593 // CHECK6-NEXT: [[E:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_0:%.*]], %struct._globalized_locals_ty.0* [[TMP4]], i32 0, i32 0 7594 // CHECK6-NEXT: [[TMP7:%.*]] = getelementptr inbounds [2048 x double], [2048 x double]* [[E]], i32 0, i32 [[TMP5]] 7595 // CHECK6-NEXT: [[TMP8:%.*]] = bitcast double* [[TMP7]] to i8* 7596 // CHECK6-NEXT: store i8* [[TMP8]], i8** [[TMP6]], align 4 7597 // CHECK6-NEXT: [[TMP9:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 7598 // CHECK6-NEXT: [[TMP10:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 7599 // CHECK6-NEXT: call void @"_omp$reduction$reduction_func"(i8* [[TMP9]], i8* [[TMP10]]) #[[ATTR3]] 7600 // CHECK6-NEXT: ret void 7601 // 7602 // 7603 // CHECK6-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_copy_func 7604 // CHECK6-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 7605 // CHECK6-NEXT: entry: 7606 // CHECK6-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 7607 // CHECK6-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 7608 // CHECK6-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 7609 // CHECK6-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 7610 // CHECK6-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 7611 // CHECK6-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 7612 // CHECK6-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 7613 // CHECK6-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [1 x i8*]* 7614 // CHECK6-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 4 7615 // CHECK6-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.0* 7616 // CHECK6-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 7617 // CHECK6-NEXT: [[TMP8:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP4]], i32 0, i32 0 7618 // CHECK6-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 4 7619 // CHECK6-NEXT: [[TMP10:%.*]] = bitcast i8* [[TMP9]] to double* 7620 // CHECK6-NEXT: [[E:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_0:%.*]], %struct._globalized_locals_ty.0* [[TMP6]], i32 0, i32 0 7621 // CHECK6-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2048 x double], [2048 x double]* [[E]], i32 0, i32 [[TMP7]] 7622 // CHECK6-NEXT: [[TMP12:%.*]] = load double, double* [[TMP11]], align 128 7623 // CHECK6-NEXT: store double [[TMP12]], double* [[TMP10]], align 8 7624 // CHECK6-NEXT: ret void 7625 // 7626 // 7627 // CHECK6-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_reduce_func 7628 // CHECK6-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 7629 // CHECK6-NEXT: entry: 7630 // CHECK6-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 7631 // CHECK6-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 7632 // CHECK6-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 7633 // CHECK6-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x i8*], align 4 7634 // CHECK6-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 7635 // CHECK6-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 7636 // CHECK6-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 7637 // CHECK6-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 7638 // CHECK6-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.0* 7639 // CHECK6-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 7640 // CHECK6-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 7641 // CHECK6-NEXT: [[E:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_0:%.*]], %struct._globalized_locals_ty.0* [[TMP4]], i32 0, i32 0 7642 // CHECK6-NEXT: [[TMP7:%.*]] = getelementptr inbounds [2048 x double], [2048 x double]* [[E]], i32 0, i32 [[TMP5]] 7643 // CHECK6-NEXT: [[TMP8:%.*]] = bitcast double* [[TMP7]] to i8* 7644 // CHECK6-NEXT: store i8* [[TMP8]], i8** [[TMP6]], align 4 7645 // CHECK6-NEXT: [[TMP9:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 7646 // CHECK6-NEXT: [[TMP10:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 7647 // CHECK6-NEXT: call void @"_omp$reduction$reduction_func"(i8* [[TMP10]], i8* [[TMP9]]) #[[ATTR3]] 7648 // CHECK6-NEXT: ret void 7649 // 7650 // 7651 // CHECK6-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29_worker 7652 // CHECK6-SAME: () #[[ATTR0]] { 7653 // CHECK6-NEXT: entry: 7654 // CHECK6-NEXT: [[WORK_FN:%.*]] = alloca i8*, align 4 7655 // CHECK6-NEXT: [[EXEC_STATUS:%.*]] = alloca i8, align 1 7656 // CHECK6-NEXT: store i8* null, i8** [[WORK_FN]], align 4 7657 // CHECK6-NEXT: store i8 0, i8* [[EXEC_STATUS]], align 1 7658 // CHECK6-NEXT: br label [[DOTAWAIT_WORK:%.*]] 7659 // CHECK6: .await.work: 7660 // CHECK6-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 7661 // CHECK6-NEXT: [[TMP0:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORK_FN]]) 7662 // CHECK6-NEXT: [[TMP1:%.*]] = zext i1 [[TMP0]] to i8 7663 // CHECK6-NEXT: store i8 [[TMP1]], i8* [[EXEC_STATUS]], align 1 7664 // CHECK6-NEXT: [[TMP2:%.*]] = load i8*, i8** [[WORK_FN]], align 4 7665 // CHECK6-NEXT: [[SHOULD_TERMINATE:%.*]] = icmp eq i8* [[TMP2]], null 7666 // CHECK6-NEXT: br i1 [[SHOULD_TERMINATE]], label [[DOTEXIT:%.*]], label [[DOTSELECT_WORKERS:%.*]] 7667 // CHECK6: .select.workers: 7668 // CHECK6-NEXT: [[TMP3:%.*]] = load i8, i8* [[EXEC_STATUS]], align 1 7669 // CHECK6-NEXT: [[IS_ACTIVE:%.*]] = icmp ne i8 [[TMP3]], 0 7670 // CHECK6-NEXT: br i1 [[IS_ACTIVE]], label [[DOTEXECUTE_PARALLEL:%.*]], label [[DOTBARRIER_PARALLEL:%.*]] 7671 // CHECK6: .execute.parallel: 7672 // CHECK6-NEXT: [[TMP4:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 7673 // CHECK6-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP2]] to void (i16, i32)* 7674 // CHECK6-NEXT: call void [[TMP5]](i16 0, i32 [[TMP4]]) 7675 // CHECK6-NEXT: br label [[DOTTERMINATE_PARALLEL:%.*]] 7676 // CHECK6: .terminate.parallel: 7677 // CHECK6-NEXT: call void @__kmpc_kernel_end_parallel() 7678 // CHECK6-NEXT: br label [[DOTBARRIER_PARALLEL]] 7679 // CHECK6: .barrier.parallel: 7680 // CHECK6-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 7681 // CHECK6-NEXT: br label [[DOTAWAIT_WORK]] 7682 // CHECK6: .exit: 7683 // CHECK6-NEXT: ret void 7684 // 7685 // 7686 // CHECK6-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29 7687 // CHECK6-SAME: (i32 [[C:%.*]], i32 [[D:%.*]]) #[[ATTR1]] { 7688 // CHECK6-NEXT: entry: 7689 // CHECK6-NEXT: [[C_ADDR:%.*]] = alloca i32, align 4 7690 // CHECK6-NEXT: [[D_ADDR:%.*]] = alloca i32, align 4 7691 // CHECK6-NEXT: [[DOTZERO_ADDR:%.*]] = alloca i32, align 4 7692 // CHECK6-NEXT: [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4 7693 // CHECK6-NEXT: store i32 0, i32* [[DOTZERO_ADDR]], align 4 7694 // CHECK6-NEXT: store i32 [[C]], i32* [[C_ADDR]], align 4 7695 // CHECK6-NEXT: store i32 [[D]], i32* [[D_ADDR]], align 4 7696 // CHECK6-NEXT: [[CONV:%.*]] = bitcast i32* [[C_ADDR]] to i8* 7697 // CHECK6-NEXT: [[CONV1:%.*]] = bitcast i32* [[D_ADDR]] to float* 7698 // CHECK6-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 7699 // CHECK6-NEXT: [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 7700 // CHECK6-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 7701 // CHECK6-NEXT: [[THREAD_LIMIT:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS]], [[NVPTX_WARP_SIZE]] 7702 // CHECK6-NEXT: [[TMP0:%.*]] = icmp ult i32 [[NVPTX_TID]], [[THREAD_LIMIT]] 7703 // CHECK6-NEXT: br i1 [[TMP0]], label [[DOTWORKER:%.*]], label [[DOTMASTERCHECK:%.*]] 7704 // CHECK6: .worker: 7705 // CHECK6-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29_worker() #[[ATTR3]] 7706 // CHECK6-NEXT: br label [[DOTEXIT:%.*]] 7707 // CHECK6: .mastercheck: 7708 // CHECK6-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 7709 // CHECK6-NEXT: [[NVPTX_NUM_THREADS3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 7710 // CHECK6-NEXT: [[NVPTX_WARP_SIZE4:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 7711 // CHECK6-NEXT: [[TMP1:%.*]] = sub nuw i32 [[NVPTX_WARP_SIZE4]], 1 7712 // CHECK6-NEXT: [[TMP2:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS3]], 1 7713 // CHECK6-NEXT: [[TMP3:%.*]] = xor i32 [[TMP1]], -1 7714 // CHECK6-NEXT: [[MASTER_TID:%.*]] = and i32 [[TMP2]], [[TMP3]] 7715 // CHECK6-NEXT: [[TMP4:%.*]] = icmp eq i32 [[NVPTX_TID2]], [[MASTER_TID]] 7716 // CHECK6-NEXT: br i1 [[TMP4]], label [[DOTMASTER:%.*]], label [[DOTEXIT]] 7717 // CHECK6: .master: 7718 // CHECK6-NEXT: [[NVPTX_NUM_THREADS5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 7719 // CHECK6-NEXT: [[NVPTX_WARP_SIZE6:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 7720 // CHECK6-NEXT: [[THREAD_LIMIT7:%.*]] = sub nuw i32 [[NVPTX_NUM_THREADS5]], [[NVPTX_WARP_SIZE6]] 7721 // CHECK6-NEXT: call void @__kmpc_kernel_init(i32 [[THREAD_LIMIT7]], i16 1) 7722 // CHECK6-NEXT: call void @__kmpc_data_sharing_init_stack() 7723 // CHECK6-NEXT: [[TMP5:%.*]] = call i8* @__kmpc_data_sharing_push_stack(i32 8, i16 1) 7724 // CHECK6-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.1* 7725 // CHECK6-NEXT: [[TMP7:%.*]] = load i8, i8* [[CONV]], align 4 7726 // CHECK6-NEXT: [[C8:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_1:%.*]], %struct._globalized_locals_ty.1* [[TMP6]], i32 0, i32 1 7727 // CHECK6-NEXT: store i8 [[TMP7]], i8* [[C8]], align 4 7728 // CHECK6-NEXT: [[TMP8:%.*]] = load float, float* [[CONV1]], align 4 7729 // CHECK6-NEXT: [[D9:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_1]], %struct._globalized_locals_ty.1* [[TMP6]], i32 0, i32 0 7730 // CHECK6-NEXT: store float [[TMP8]], float* [[D9]], align 4 7731 // CHECK6-NEXT: [[TMP9:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 7732 // CHECK6-NEXT: store i32 [[TMP9]], i32* [[DOTTHREADID_TEMP_]], align 4 7733 // CHECK6-NEXT: call void @__omp_outlined__1(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i8* [[C8]], float* [[D9]]) #[[ATTR3]] 7734 // CHECK6-NEXT: call void @__kmpc_data_sharing_pop_stack(i8* [[TMP5]]) 7735 // CHECK6-NEXT: br label [[DOTTERMINATION_NOTIFIER:%.*]] 7736 // CHECK6: .termination.notifier: 7737 // CHECK6-NEXT: call void @__kmpc_kernel_deinit(i16 1) 7738 // CHECK6-NEXT: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 7739 // CHECK6-NEXT: br label [[DOTEXIT]] 7740 // CHECK6: .exit: 7741 // CHECK6-NEXT: ret void 7742 // 7743 // 7744 // CHECK6-LABEL: define {{[^@]+}}@__omp_outlined__1 7745 // CHECK6-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i8* nonnull align 1 dereferenceable(1) [[C:%.*]], float* nonnull align 4 dereferenceable(4) [[D:%.*]]) #[[ATTR1]] { 7746 // CHECK6-NEXT: entry: 7747 // CHECK6-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4 7748 // CHECK6-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4 7749 // CHECK6-NEXT: [[C_ADDR:%.*]] = alloca i8*, align 4 7750 // CHECK6-NEXT: [[D_ADDR:%.*]] = alloca float*, align 4 7751 // CHECK6-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 7752 // CHECK6-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4 7753 // CHECK6-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4 7754 // CHECK6-NEXT: store i8* [[C]], i8** [[C_ADDR]], align 4 7755 // CHECK6-NEXT: store float* [[D]], float** [[D_ADDR]], align 4 7756 // CHECK6-NEXT: [[TMP0:%.*]] = load i8*, i8** [[C_ADDR]], align 4 7757 // CHECK6-NEXT: [[TMP1:%.*]] = load float*, float** [[D_ADDR]], align 4 7758 // CHECK6-NEXT: [[TMP2:%.*]] = call i8* @__kmpc_data_sharing_push_stack(i32 8, i16 1) 7759 // CHECK6-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to %struct._globalized_locals_ty.2* 7760 // CHECK6-NEXT: [[C1:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_2:%.*]], %struct._globalized_locals_ty.2* [[TMP3]], i32 0, i32 1 7761 // CHECK6-NEXT: [[D2:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_2]], %struct._globalized_locals_ty.2* [[TMP3]], i32 0, i32 0 7762 // CHECK6-NEXT: store i8 0, i8* [[C1]], align 4 7763 // CHECK6-NEXT: store float 1.000000e+00, float* [[D2]], align 4 7764 // CHECK6-NEXT: [[TMP4:%.*]] = load i8, i8* [[C1]], align 4 7765 // CHECK6-NEXT: [[CONV:%.*]] = sext i8 [[TMP4]] to i32 7766 // CHECK6-NEXT: [[XOR:%.*]] = xor i32 [[CONV]], 2 7767 // CHECK6-NEXT: [[CONV3:%.*]] = trunc i32 [[XOR]] to i8 7768 // CHECK6-NEXT: store i8 [[CONV3]], i8* [[C1]], align 4 7769 // CHECK6-NEXT: [[TMP5:%.*]] = load float, float* [[D2]], align 4 7770 // CHECK6-NEXT: [[MUL:%.*]] = fmul float [[TMP5]], 3.300000e+01 7771 // CHECK6-NEXT: store float [[MUL]], float* [[D2]], align 4 7772 // CHECK6-NEXT: [[TMP6:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4 7773 // CHECK6-NEXT: [[TMP7:%.*]] = load i32, i32* [[TMP6]], align 4 7774 // CHECK6-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 7775 // CHECK6-NEXT: store i8* [[C1]], i8** [[TMP8]], align 4 7776 // CHECK6-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 7777 // CHECK6-NEXT: [[TMP10:%.*]] = bitcast float* [[D2]] to i8* 7778 // CHECK6-NEXT: store i8* [[TMP10]], i8** [[TMP9]], align 4 7779 // CHECK6-NEXT: [[TMP11:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 7780 // CHECK6-NEXT: [[TMP12:%.*]] = load i8*, i8** @"_openmp_teams_reductions_buffer_$_$ptr", align 4 7781 // CHECK6-NEXT: [[TMP13:%.*]] = call i32 @__kmpc_nvptx_teams_reduce_nowait_v2(%struct.ident_t* @[[GLOB1]], i32 [[TMP7]], i8* [[TMP12]], i32 2048, i8* [[TMP11]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func3, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func4, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_copy_func5, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_reduce_func6, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_copy_func7, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_reduce_func8) 7782 // CHECK6-NEXT: [[TMP14:%.*]] = icmp eq i32 [[TMP13]], 1 7783 // CHECK6-NEXT: br i1 [[TMP14]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]] 7784 // CHECK6: .omp.reduction.then: 7785 // CHECK6-NEXT: [[TMP15:%.*]] = load i8, i8* [[TMP0]], align 1 7786 // CHECK6-NEXT: [[CONV4:%.*]] = sext i8 [[TMP15]] to i32 7787 // CHECK6-NEXT: [[TMP16:%.*]] = load i8, i8* [[C1]], align 4 7788 // CHECK6-NEXT: [[CONV5:%.*]] = sext i8 [[TMP16]] to i32 7789 // CHECK6-NEXT: [[XOR6:%.*]] = xor i32 [[CONV4]], [[CONV5]] 7790 // CHECK6-NEXT: [[CONV7:%.*]] = trunc i32 [[XOR6]] to i8 7791 // CHECK6-NEXT: store i8 [[CONV7]], i8* [[TMP0]], align 1 7792 // CHECK6-NEXT: [[TMP17:%.*]] = load float, float* [[TMP1]], align 4 7793 // CHECK6-NEXT: [[TMP18:%.*]] = load float, float* [[D2]], align 4 7794 // CHECK6-NEXT: [[MUL8:%.*]] = fmul float [[TMP17]], [[TMP18]] 7795 // CHECK6-NEXT: store float [[MUL8]], float* [[TMP1]], align 4 7796 // CHECK6-NEXT: call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP7]]) 7797 // CHECK6-NEXT: br label [[DOTOMP_REDUCTION_DONE]] 7798 // CHECK6: .omp.reduction.done: 7799 // CHECK6-NEXT: call void @__kmpc_data_sharing_pop_stack(i8* [[TMP2]]) 7800 // CHECK6-NEXT: ret void 7801 // 7802 // 7803 // CHECK6-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func3 7804 // CHECK6-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR0]] { 7805 // CHECK6-NEXT: entry: 7806 // CHECK6-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 7807 // CHECK6-NEXT: [[DOTADDR1:%.*]] = alloca i16, align 2 7808 // CHECK6-NEXT: [[DOTADDR2:%.*]] = alloca i16, align 2 7809 // CHECK6-NEXT: [[DOTADDR3:%.*]] = alloca i16, align 2 7810 // CHECK6-NEXT: [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [2 x i8*], align 4 7811 // CHECK6-NEXT: [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca i8, align 1 7812 // CHECK6-NEXT: [[DOTOMP_REDUCTION_ELEMENT4:%.*]] = alloca float, align 4 7813 // CHECK6-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 7814 // CHECK6-NEXT: store i16 [[TMP1]], i16* [[DOTADDR1]], align 2 7815 // CHECK6-NEXT: store i16 [[TMP2]], i16* [[DOTADDR2]], align 2 7816 // CHECK6-NEXT: store i16 [[TMP3]], i16* [[DOTADDR3]], align 2 7817 // CHECK6-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 4 7818 // CHECK6-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [2 x i8*]* 7819 // CHECK6-NEXT: [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2 7820 // CHECK6-NEXT: [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2 7821 // CHECK6-NEXT: [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2 7822 // CHECK6-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 0 7823 // CHECK6-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 4 7824 // CHECK6-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 7825 // CHECK6-NEXT: [[TMP12:%.*]] = getelementptr i8, i8* [[TMP10]], i32 1 7826 // CHECK6-NEXT: [[TMP13:%.*]] = load i8, i8* [[TMP10]], align 1 7827 // CHECK6-NEXT: [[TMP14:%.*]] = sext i8 [[TMP13]] to i32 7828 // CHECK6-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 7829 // CHECK6-NEXT: [[TMP15:%.*]] = trunc i32 [[NVPTX_WARP_SIZE]] to i16 7830 // CHECK6-NEXT: [[TMP16:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP14]], i16 [[TMP7]], i16 [[TMP15]]) 7831 // CHECK6-NEXT: [[TMP17:%.*]] = trunc i32 [[TMP16]] to i8 7832 // CHECK6-NEXT: store i8 [[TMP17]], i8* [[DOTOMP_REDUCTION_ELEMENT]], align 1 7833 // CHECK6-NEXT: [[TMP18:%.*]] = getelementptr i8, i8* [[TMP10]], i32 1 7834 // CHECK6-NEXT: [[TMP19:%.*]] = getelementptr i8, i8* [[DOTOMP_REDUCTION_ELEMENT]], i32 1 7835 // CHECK6-NEXT: store i8* [[DOTOMP_REDUCTION_ELEMENT]], i8** [[TMP11]], align 4 7836 // CHECK6-NEXT: [[TMP20:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 1 7837 // CHECK6-NEXT: [[TMP21:%.*]] = load i8*, i8** [[TMP20]], align 4 7838 // CHECK6-NEXT: [[TMP22:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1 7839 // CHECK6-NEXT: [[TMP23:%.*]] = bitcast i8* [[TMP21]] to float* 7840 // CHECK6-NEXT: [[TMP24:%.*]] = getelementptr float, float* [[TMP23]], i32 1 7841 // CHECK6-NEXT: [[TMP25:%.*]] = bitcast float* [[TMP24]] to i8* 7842 // CHECK6-NEXT: [[TMP26:%.*]] = bitcast float* [[TMP23]] to i32* 7843 // CHECK6-NEXT: [[TMP27:%.*]] = bitcast float* [[DOTOMP_REDUCTION_ELEMENT4]] to i32* 7844 // CHECK6-NEXT: [[TMP28:%.*]] = load i32, i32* [[TMP26]], align 4 7845 // CHECK6-NEXT: [[NVPTX_WARP_SIZE5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 7846 // CHECK6-NEXT: [[TMP29:%.*]] = trunc i32 [[NVPTX_WARP_SIZE5]] to i16 7847 // CHECK6-NEXT: [[TMP30:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP28]], i16 [[TMP7]], i16 [[TMP29]]) 7848 // CHECK6-NEXT: store i32 [[TMP30]], i32* [[TMP27]], align 4 7849 // CHECK6-NEXT: [[TMP31:%.*]] = getelementptr i32, i32* [[TMP26]], i32 1 7850 // CHECK6-NEXT: [[TMP32:%.*]] = getelementptr i32, i32* [[TMP27]], i32 1 7851 // CHECK6-NEXT: [[TMP33:%.*]] = bitcast float* [[DOTOMP_REDUCTION_ELEMENT4]] to i8* 7852 // CHECK6-NEXT: store i8* [[TMP33]], i8** [[TMP22]], align 4 7853 // CHECK6-NEXT: [[TMP34:%.*]] = icmp eq i16 [[TMP8]], 0 7854 // CHECK6-NEXT: [[TMP35:%.*]] = icmp eq i16 [[TMP8]], 1 7855 // CHECK6-NEXT: [[TMP36:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]] 7856 // CHECK6-NEXT: [[TMP37:%.*]] = and i1 [[TMP35]], [[TMP36]] 7857 // CHECK6-NEXT: [[TMP38:%.*]] = icmp eq i16 [[TMP8]], 2 7858 // CHECK6-NEXT: [[TMP39:%.*]] = and i16 [[TMP6]], 1 7859 // CHECK6-NEXT: [[TMP40:%.*]] = icmp eq i16 [[TMP39]], 0 7860 // CHECK6-NEXT: [[TMP41:%.*]] = and i1 [[TMP38]], [[TMP40]] 7861 // CHECK6-NEXT: [[TMP42:%.*]] = icmp sgt i16 [[TMP7]], 0 7862 // CHECK6-NEXT: [[TMP43:%.*]] = and i1 [[TMP41]], [[TMP42]] 7863 // CHECK6-NEXT: [[TMP44:%.*]] = or i1 [[TMP34]], [[TMP37]] 7864 // CHECK6-NEXT: [[TMP45:%.*]] = or i1 [[TMP44]], [[TMP43]] 7865 // CHECK6-NEXT: br i1 [[TMP45]], label [[THEN:%.*]], label [[ELSE:%.*]] 7866 // CHECK6: then: 7867 // CHECK6-NEXT: [[TMP46:%.*]] = bitcast [2 x i8*]* [[TMP5]] to i8* 7868 // CHECK6-NEXT: [[TMP47:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8* 7869 // CHECK6-NEXT: call void @"_omp$reduction$reduction_func2"(i8* [[TMP46]], i8* [[TMP47]]) #[[ATTR3]] 7870 // CHECK6-NEXT: br label [[IFCONT:%.*]] 7871 // CHECK6: else: 7872 // CHECK6-NEXT: br label [[IFCONT]] 7873 // CHECK6: ifcont: 7874 // CHECK6-NEXT: [[TMP48:%.*]] = icmp eq i16 [[TMP8]], 1 7875 // CHECK6-NEXT: [[TMP49:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]] 7876 // CHECK6-NEXT: [[TMP50:%.*]] = and i1 [[TMP48]], [[TMP49]] 7877 // CHECK6-NEXT: br i1 [[TMP50]], label [[THEN6:%.*]], label [[ELSE7:%.*]] 7878 // CHECK6: then6: 7879 // CHECK6-NEXT: [[TMP51:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 7880 // CHECK6-NEXT: [[TMP52:%.*]] = load i8*, i8** [[TMP51]], align 4 7881 // CHECK6-NEXT: [[TMP53:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 0 7882 // CHECK6-NEXT: [[TMP54:%.*]] = load i8*, i8** [[TMP53]], align 4 7883 // CHECK6-NEXT: [[TMP55:%.*]] = load i8, i8* [[TMP52]], align 1 7884 // CHECK6-NEXT: store i8 [[TMP55]], i8* [[TMP54]], align 1 7885 // CHECK6-NEXT: [[TMP56:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1 7886 // CHECK6-NEXT: [[TMP57:%.*]] = load i8*, i8** [[TMP56]], align 4 7887 // CHECK6-NEXT: [[TMP58:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 1 7888 // CHECK6-NEXT: [[TMP59:%.*]] = load i8*, i8** [[TMP58]], align 4 7889 // CHECK6-NEXT: [[TMP60:%.*]] = bitcast i8* [[TMP57]] to float* 7890 // CHECK6-NEXT: [[TMP61:%.*]] = bitcast i8* [[TMP59]] to float* 7891 // CHECK6-NEXT: [[TMP62:%.*]] = load float, float* [[TMP60]], align 4 7892 // CHECK6-NEXT: store float [[TMP62]], float* [[TMP61]], align 4 7893 // CHECK6-NEXT: br label [[IFCONT8:%.*]] 7894 // CHECK6: else7: 7895 // CHECK6-NEXT: br label [[IFCONT8]] 7896 // CHECK6: ifcont8: 7897 // CHECK6-NEXT: ret void 7898 // 7899 // 7900 // CHECK6-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func4 7901 // CHECK6-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] { 7902 // CHECK6-NEXT: entry: 7903 // CHECK6-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 7904 // CHECK6-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 7905 // CHECK6-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) 7906 // CHECK6-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 7907 // CHECK6-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 7908 // CHECK6-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 7909 // CHECK6-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 7910 // CHECK6-NEXT: [[NVPTX_LANE_ID:%.*]] = and i32 [[NVPTX_TID2]], 31 7911 // CHECK6-NEXT: [[NVPTX_TID3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 7912 // CHECK6-NEXT: [[NVPTX_WARP_ID:%.*]] = ashr i32 [[NVPTX_TID3]], 5 7913 // CHECK6-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 7914 // CHECK6-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 7915 // CHECK6-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 7916 // CHECK6-NEXT: [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 7917 // CHECK6-NEXT: br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]] 7918 // CHECK6: then: 7919 // CHECK6-NEXT: [[TMP5:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 7920 // CHECK6-NEXT: [[TMP6:%.*]] = load i8*, i8** [[TMP5]], align 4 7921 // CHECK6-NEXT: [[TMP7:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 7922 // CHECK6-NEXT: [[TMP8:%.*]] = bitcast i32 addrspace(3)* [[TMP7]] to i8 addrspace(3)* 7923 // CHECK6-NEXT: [[TMP9:%.*]] = load i8, i8* [[TMP6]], align 1 7924 // CHECK6-NEXT: store volatile i8 [[TMP9]], i8 addrspace(3)* [[TMP8]], align 1 7925 // CHECK6-NEXT: br label [[IFCONT:%.*]] 7926 // CHECK6: else: 7927 // CHECK6-NEXT: br label [[IFCONT]] 7928 // CHECK6: ifcont: 7929 // CHECK6-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 7930 // CHECK6-NEXT: [[TMP10:%.*]] = load i32, i32* [[DOTADDR1]], align 4 7931 // CHECK6-NEXT: [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP10]] 7932 // CHECK6-NEXT: br i1 [[IS_ACTIVE_THREAD]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 7933 // CHECK6: then4: 7934 // CHECK6-NEXT: [[TMP11:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 7935 // CHECK6-NEXT: [[TMP12:%.*]] = bitcast i32 addrspace(3)* [[TMP11]] to i8 addrspace(3)* 7936 // CHECK6-NEXT: [[TMP13:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 7937 // CHECK6-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 4 7938 // CHECK6-NEXT: [[TMP15:%.*]] = load volatile i8, i8 addrspace(3)* [[TMP12]], align 1 7939 // CHECK6-NEXT: store i8 [[TMP15]], i8* [[TMP14]], align 1 7940 // CHECK6-NEXT: br label [[IFCONT6:%.*]] 7941 // CHECK6: else5: 7942 // CHECK6-NEXT: br label [[IFCONT6]] 7943 // CHECK6: ifcont6: 7944 // CHECK6-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 7945 // CHECK6-NEXT: [[WARP_MASTER7:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 7946 // CHECK6-NEXT: br i1 [[WARP_MASTER7]], label [[THEN8:%.*]], label [[ELSE9:%.*]] 7947 // CHECK6: then8: 7948 // CHECK6-NEXT: [[TMP16:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 7949 // CHECK6-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 4 7950 // CHECK6-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i32* 7951 // CHECK6-NEXT: [[TMP19:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 7952 // CHECK6-NEXT: [[TMP20:%.*]] = load i32, i32* [[TMP18]], align 4 7953 // CHECK6-NEXT: store volatile i32 [[TMP20]], i32 addrspace(3)* [[TMP19]], align 4 7954 // CHECK6-NEXT: br label [[IFCONT10:%.*]] 7955 // CHECK6: else9: 7956 // CHECK6-NEXT: br label [[IFCONT10]] 7957 // CHECK6: ifcont10: 7958 // CHECK6-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]]) 7959 // CHECK6-NEXT: [[TMP21:%.*]] = load i32, i32* [[DOTADDR1]], align 4 7960 // CHECK6-NEXT: [[IS_ACTIVE_THREAD11:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP21]] 7961 // CHECK6-NEXT: br i1 [[IS_ACTIVE_THREAD11]], label [[THEN12:%.*]], label [[ELSE13:%.*]] 7962 // CHECK6: then12: 7963 // CHECK6-NEXT: [[TMP22:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 7964 // CHECK6-NEXT: [[TMP23:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 7965 // CHECK6-NEXT: [[TMP24:%.*]] = load i8*, i8** [[TMP23]], align 4 7966 // CHECK6-NEXT: [[TMP25:%.*]] = bitcast i8* [[TMP24]] to i32* 7967 // CHECK6-NEXT: [[TMP26:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP22]], align 4 7968 // CHECK6-NEXT: store i32 [[TMP26]], i32* [[TMP25]], align 4 7969 // CHECK6-NEXT: br label [[IFCONT14:%.*]] 7970 // CHECK6: else13: 7971 // CHECK6-NEXT: br label [[IFCONT14]] 7972 // CHECK6: ifcont14: 7973 // CHECK6-NEXT: ret void 7974 // 7975 // 7976 // CHECK6-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_copy_func5 7977 // CHECK6-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 7978 // CHECK6-NEXT: entry: 7979 // CHECK6-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 7980 // CHECK6-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 7981 // CHECK6-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 7982 // CHECK6-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 7983 // CHECK6-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 7984 // CHECK6-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 7985 // CHECK6-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 7986 // CHECK6-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 7987 // CHECK6-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 4 7988 // CHECK6-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.3* 7989 // CHECK6-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 7990 // CHECK6-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 7991 // CHECK6-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 4 7992 // CHECK6-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3:%.*]], %struct._globalized_locals_ty.3* [[TMP6]], i32 0, i32 0 7993 // CHECK6-NEXT: [[TMP10:%.*]] = getelementptr inbounds [2048 x i8], [2048 x i8]* [[C]], i32 0, i32 [[TMP7]] 7994 // CHECK6-NEXT: [[TMP11:%.*]] = load i8, i8* [[TMP9]], align 1 7995 // CHECK6-NEXT: store i8 [[TMP11]], i8* [[TMP10]], align 128 7996 // CHECK6-NEXT: [[TMP12:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 7997 // CHECK6-NEXT: [[TMP13:%.*]] = load i8*, i8** [[TMP12]], align 4 7998 // CHECK6-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to float* 7999 // CHECK6-NEXT: [[D:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3]], %struct._globalized_locals_ty.3* [[TMP6]], i32 0, i32 1 8000 // CHECK6-NEXT: [[TMP15:%.*]] = getelementptr inbounds [2048 x float], [2048 x float]* [[D]], i32 0, i32 [[TMP7]] 8001 // CHECK6-NEXT: [[TMP16:%.*]] = load float, float* [[TMP14]], align 4 8002 // CHECK6-NEXT: store float [[TMP16]], float* [[TMP15]], align 128 8003 // CHECK6-NEXT: ret void 8004 // 8005 // 8006 // CHECK6-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_reduce_func6 8007 // CHECK6-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 8008 // CHECK6-NEXT: entry: 8009 // CHECK6-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 8010 // CHECK6-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 8011 // CHECK6-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 8012 // CHECK6-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 8013 // CHECK6-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 8014 // CHECK6-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 8015 // CHECK6-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 8016 // CHECK6-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 8017 // CHECK6-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.3* 8018 // CHECK6-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 8019 // CHECK6-NEXT: [[TMP6:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 8020 // CHECK6-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3:%.*]], %struct._globalized_locals_ty.3* [[TMP4]], i32 0, i32 0 8021 // CHECK6-NEXT: [[TMP7:%.*]] = getelementptr inbounds [2048 x i8], [2048 x i8]* [[C]], i32 0, i32 [[TMP5]] 8022 // CHECK6-NEXT: store i8* [[TMP7]], i8** [[TMP6]], align 4 8023 // CHECK6-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 8024 // CHECK6-NEXT: [[D:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3]], %struct._globalized_locals_ty.3* [[TMP4]], i32 0, i32 1 8025 // CHECK6-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2048 x float], [2048 x float]* [[D]], i32 0, i32 [[TMP5]] 8026 // CHECK6-NEXT: [[TMP10:%.*]] = bitcast float* [[TMP9]] to i8* 8027 // CHECK6-NEXT: store i8* [[TMP10]], i8** [[TMP8]], align 4 8028 // CHECK6-NEXT: [[TMP11:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 8029 // CHECK6-NEXT: [[TMP12:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 8030 // CHECK6-NEXT: call void @"_omp$reduction$reduction_func2"(i8* [[TMP11]], i8* [[TMP12]]) #[[ATTR3]] 8031 // CHECK6-NEXT: ret void 8032 // 8033 // 8034 // CHECK6-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_copy_func7 8035 // CHECK6-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 8036 // CHECK6-NEXT: entry: 8037 // CHECK6-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 8038 // CHECK6-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 8039 // CHECK6-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 8040 // CHECK6-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 8041 // CHECK6-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 8042 // CHECK6-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 8043 // CHECK6-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 8044 // CHECK6-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 8045 // CHECK6-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 4 8046 // CHECK6-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.3* 8047 // CHECK6-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 8048 // CHECK6-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 8049 // CHECK6-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 4 8050 // CHECK6-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3:%.*]], %struct._globalized_locals_ty.3* [[TMP6]], i32 0, i32 0 8051 // CHECK6-NEXT: [[TMP10:%.*]] = getelementptr inbounds [2048 x i8], [2048 x i8]* [[C]], i32 0, i32 [[TMP7]] 8052 // CHECK6-NEXT: [[TMP11:%.*]] = load i8, i8* [[TMP10]], align 128 8053 // CHECK6-NEXT: store i8 [[TMP11]], i8* [[TMP9]], align 1 8054 // CHECK6-NEXT: [[TMP12:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 8055 // CHECK6-NEXT: [[TMP13:%.*]] = load i8*, i8** [[TMP12]], align 4 8056 // CHECK6-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to float* 8057 // CHECK6-NEXT: [[D:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3]], %struct._globalized_locals_ty.3* [[TMP6]], i32 0, i32 1 8058 // CHECK6-NEXT: [[TMP15:%.*]] = getelementptr inbounds [2048 x float], [2048 x float]* [[D]], i32 0, i32 [[TMP7]] 8059 // CHECK6-NEXT: [[TMP16:%.*]] = load float, float* [[TMP15]], align 128 8060 // CHECK6-NEXT: store float [[TMP16]], float* [[TMP14]], align 4 8061 // CHECK6-NEXT: ret void 8062 // 8063 // 8064 // CHECK6-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_reduce_func8 8065 // CHECK6-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 8066 // CHECK6-NEXT: entry: 8067 // CHECK6-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 8068 // CHECK6-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 8069 // CHECK6-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 8070 // CHECK6-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 8071 // CHECK6-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 8072 // CHECK6-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 8073 // CHECK6-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 8074 // CHECK6-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 8075 // CHECK6-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.3* 8076 // CHECK6-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 8077 // CHECK6-NEXT: [[TMP6:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 8078 // CHECK6-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3:%.*]], %struct._globalized_locals_ty.3* [[TMP4]], i32 0, i32 0 8079 // CHECK6-NEXT: [[TMP7:%.*]] = getelementptr inbounds [2048 x i8], [2048 x i8]* [[C]], i32 0, i32 [[TMP5]] 8080 // CHECK6-NEXT: store i8* [[TMP7]], i8** [[TMP6]], align 4 8081 // CHECK6-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 8082 // CHECK6-NEXT: [[D:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_3]], %struct._globalized_locals_ty.3* [[TMP4]], i32 0, i32 1 8083 // CHECK6-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2048 x float], [2048 x float]* [[D]], i32 0, i32 [[TMP5]] 8084 // CHECK6-NEXT: [[TMP10:%.*]] = bitcast float* [[TMP9]] to i8* 8085 // CHECK6-NEXT: store i8* [[TMP10]], i8** [[TMP8]], align 4 8086 // CHECK6-NEXT: [[TMP11:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 8087 // CHECK6-NEXT: [[TMP12:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 8088 // CHECK6-NEXT: call void @"_omp$reduction$reduction_func2"(i8* [[TMP12]], i8* [[TMP11]]) #[[ATTR3]] 8089 // CHECK6-NEXT: ret void 8090 // 8091 // 8092 // CHECK6-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l36 8093 // CHECK6-SAME: (i32 [[A:%.*]], i32 [[B:%.*]]) #[[ATTR1]] { 8094 // CHECK6-NEXT: entry: 8095 // CHECK6-NEXT: [[A_ADDR:%.*]] = alloca i32, align 4 8096 // CHECK6-NEXT: [[B_ADDR:%.*]] = alloca i32, align 4 8097 // CHECK6-NEXT: [[DOTZERO_ADDR:%.*]] = alloca i32, align 4 8098 // CHECK6-NEXT: [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4 8099 // CHECK6-NEXT: store i32 0, i32* [[DOTZERO_ADDR]], align 4 8100 // CHECK6-NEXT: store i32 [[A]], i32* [[A_ADDR]], align 4 8101 // CHECK6-NEXT: store i32 [[B]], i32* [[B_ADDR]], align 4 8102 // CHECK6-NEXT: [[CONV:%.*]] = bitcast i32* [[B_ADDR]] to i16* 8103 // CHECK6-NEXT: [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() 8104 // CHECK6-NEXT: call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 1) 8105 // CHECK6-NEXT: call void @__kmpc_data_sharing_init_stack_spmd() 8106 // CHECK6-NEXT: br label [[DOTEXECUTE:%.*]] 8107 // CHECK6: .execute: 8108 // CHECK6-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3:[0-9]+]]) 8109 // CHECK6-NEXT: store i32 [[TMP0]], i32* [[DOTTHREADID_TEMP_]], align 4 8110 // CHECK6-NEXT: call void @__omp_outlined__9(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32* [[A_ADDR]], i16* [[CONV]]) #[[ATTR3]] 8111 // CHECK6-NEXT: br label [[DOTOMP_DEINIT:%.*]] 8112 // CHECK6: .omp.deinit: 8113 // CHECK6-NEXT: call void @__kmpc_spmd_kernel_deinit_v2(i16 1) 8114 // CHECK6-NEXT: br label [[DOTEXIT:%.*]] 8115 // CHECK6: .exit: 8116 // CHECK6-NEXT: ret void 8117 // 8118 // 8119 // CHECK6-LABEL: define {{[^@]+}}@__omp_outlined__9 8120 // CHECK6-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32* nonnull align 4 dereferenceable(4) [[A:%.*]], i16* nonnull align 2 dereferenceable(2) [[B:%.*]]) #[[ATTR1]] { 8121 // CHECK6-NEXT: entry: 8122 // CHECK6-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4 8123 // CHECK6-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4 8124 // CHECK6-NEXT: [[A_ADDR:%.*]] = alloca i32*, align 4 8125 // CHECK6-NEXT: [[B_ADDR:%.*]] = alloca i16*, align 4 8126 // CHECK6-NEXT: [[A1:%.*]] = alloca i32, align 4 8127 // CHECK6-NEXT: [[B2:%.*]] = alloca i16, align 2 8128 // CHECK6-NEXT: [[CAPTURED_VARS_ADDRS:%.*]] = alloca [2 x i8*], align 4 8129 // CHECK6-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 8130 // CHECK6-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4 8131 // CHECK6-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4 8132 // CHECK6-NEXT: store i32* [[A]], i32** [[A_ADDR]], align 4 8133 // CHECK6-NEXT: store i16* [[B]], i16** [[B_ADDR]], align 4 8134 // CHECK6-NEXT: [[TMP0:%.*]] = load i32*, i32** [[A_ADDR]], align 4 8135 // CHECK6-NEXT: [[TMP1:%.*]] = load i16*, i16** [[B_ADDR]], align 4 8136 // CHECK6-NEXT: store i32 0, i32* [[A1]], align 4 8137 // CHECK6-NEXT: store i16 -32768, i16* [[B2]], align 2 8138 // CHECK6-NEXT: [[TMP2:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0 8139 // CHECK6-NEXT: [[TMP3:%.*]] = bitcast i32* [[A1]] to i8* 8140 // CHECK6-NEXT: store i8* [[TMP3]], i8** [[TMP2]], align 4 8141 // CHECK6-NEXT: [[TMP4:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1 8142 // CHECK6-NEXT: [[TMP5:%.*]] = bitcast i16* [[B2]] to i8* 8143 // CHECK6-NEXT: store i8* [[TMP5]], i8** [[TMP4]], align 4 8144 // CHECK6-NEXT: [[TMP6:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4 8145 // CHECK6-NEXT: [[TMP7:%.*]] = load i32, i32* [[TMP6]], align 4 8146 // CHECK6-NEXT: [[TMP8:%.*]] = bitcast [2 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8** 8147 // CHECK6-NEXT: call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP7]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32*, i16*)* @__omp_outlined__10 to i8*), i8* null, i8** [[TMP8]], i32 2) 8148 // CHECK6-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 8149 // CHECK6-NEXT: [[TMP10:%.*]] = bitcast i32* [[A1]] to i8* 8150 // CHECK6-NEXT: store i8* [[TMP10]], i8** [[TMP9]], align 4 8151 // CHECK6-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 8152 // CHECK6-NEXT: [[TMP12:%.*]] = bitcast i16* [[B2]] to i8* 8153 // CHECK6-NEXT: store i8* [[TMP12]], i8** [[TMP11]], align 4 8154 // CHECK6-NEXT: [[TMP13:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 8155 // CHECK6-NEXT: [[TMP14:%.*]] = load i8*, i8** @"_openmp_teams_reductions_buffer_$_$ptr", align 4 8156 // CHECK6-NEXT: [[TMP15:%.*]] = call i32 @__kmpc_nvptx_teams_reduce_nowait_v2(%struct.ident_t* @[[GLOB3]], i32 [[TMP7]], i8* [[TMP14]], i32 2048, i8* [[TMP13]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func15, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func16, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_copy_func17, void (i8*, i32, i8*)* @_omp_reduction_list_to_global_reduce_func18, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_copy_func19, void (i8*, i32, i8*)* @_omp_reduction_global_to_list_reduce_func20) 8157 // CHECK6-NEXT: [[TMP16:%.*]] = icmp eq i32 [[TMP15]], 1 8158 // CHECK6-NEXT: br i1 [[TMP16]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]] 8159 // CHECK6: .omp.reduction.then: 8160 // CHECK6-NEXT: [[TMP17:%.*]] = load i32, i32* [[TMP0]], align 4 8161 // CHECK6-NEXT: [[TMP18:%.*]] = load i32, i32* [[A1]], align 4 8162 // CHECK6-NEXT: [[OR:%.*]] = or i32 [[TMP17]], [[TMP18]] 8163 // CHECK6-NEXT: store i32 [[OR]], i32* [[TMP0]], align 4 8164 // CHECK6-NEXT: [[TMP19:%.*]] = load i16, i16* [[TMP1]], align 2 8165 // CHECK6-NEXT: [[CONV:%.*]] = sext i16 [[TMP19]] to i32 8166 // CHECK6-NEXT: [[TMP20:%.*]] = load i16, i16* [[B2]], align 2 8167 // CHECK6-NEXT: [[CONV3:%.*]] = sext i16 [[TMP20]] to i32 8168 // CHECK6-NEXT: [[CMP:%.*]] = icmp sgt i32 [[CONV]], [[CONV3]] 8169 // CHECK6-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] 8170 // CHECK6: cond.true: 8171 // CHECK6-NEXT: [[TMP21:%.*]] = load i16, i16* [[TMP1]], align 2 8172 // CHECK6-NEXT: br label [[COND_END:%.*]] 8173 // CHECK6: cond.false: 8174 // CHECK6-NEXT: [[TMP22:%.*]] = load i16, i16* [[B2]], align 2 8175 // CHECK6-NEXT: br label [[COND_END]] 8176 // CHECK6: cond.end: 8177 // CHECK6-NEXT: [[COND:%.*]] = phi i16 [ [[TMP21]], [[COND_TRUE]] ], [ [[TMP22]], [[COND_FALSE]] ] 8178 // CHECK6-NEXT: store i16 [[COND]], i16* [[TMP1]], align 2 8179 // CHECK6-NEXT: call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP7]]) 8180 // CHECK6-NEXT: br label [[DOTOMP_REDUCTION_DONE]] 8181 // CHECK6: .omp.reduction.done: 8182 // CHECK6-NEXT: ret void 8183 // 8184 // 8185 // CHECK6-LABEL: define {{[^@]+}}@__omp_outlined__10 8186 // CHECK6-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32* nonnull align 4 dereferenceable(4) [[A:%.*]], i16* nonnull align 2 dereferenceable(2) [[B:%.*]]) #[[ATTR1]] { 8187 // CHECK6-NEXT: entry: 8188 // CHECK6-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4 8189 // CHECK6-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4 8190 // CHECK6-NEXT: [[A_ADDR:%.*]] = alloca i32*, align 4 8191 // CHECK6-NEXT: [[B_ADDR:%.*]] = alloca i16*, align 4 8192 // CHECK6-NEXT: [[A1:%.*]] = alloca i32, align 4 8193 // CHECK6-NEXT: [[B2:%.*]] = alloca i16, align 2 8194 // CHECK6-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 8195 // CHECK6-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4 8196 // CHECK6-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4 8197 // CHECK6-NEXT: store i32* [[A]], i32** [[A_ADDR]], align 4 8198 // CHECK6-NEXT: store i16* [[B]], i16** [[B_ADDR]], align 4 8199 // CHECK6-NEXT: [[TMP0:%.*]] = load i32*, i32** [[A_ADDR]], align 4 8200 // CHECK6-NEXT: [[TMP1:%.*]] = load i16*, i16** [[B_ADDR]], align 4 8201 // CHECK6-NEXT: store i32 0, i32* [[A1]], align 4 8202 // CHECK6-NEXT: store i16 -32768, i16* [[B2]], align 2 8203 // CHECK6-NEXT: [[TMP2:%.*]] = load i32, i32* [[A1]], align 4 8204 // CHECK6-NEXT: [[OR:%.*]] = or i32 [[TMP2]], 1 8205 // CHECK6-NEXT: store i32 [[OR]], i32* [[A1]], align 4 8206 // CHECK6-NEXT: [[TMP3:%.*]] = load i16, i16* [[B2]], align 2 8207 // CHECK6-NEXT: [[CONV:%.*]] = sext i16 [[TMP3]] to i32 8208 // CHECK6-NEXT: [[CMP:%.*]] = icmp sgt i32 99, [[CONV]] 8209 // CHECK6-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] 8210 // CHECK6: cond.true: 8211 // CHECK6-NEXT: br label [[COND_END:%.*]] 8212 // CHECK6: cond.false: 8213 // CHECK6-NEXT: [[TMP4:%.*]] = load i16, i16* [[B2]], align 2 8214 // CHECK6-NEXT: [[CONV3:%.*]] = sext i16 [[TMP4]] to i32 8215 // CHECK6-NEXT: br label [[COND_END]] 8216 // CHECK6: cond.end: 8217 // CHECK6-NEXT: [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[CONV3]], [[COND_FALSE]] ] 8218 // CHECK6-NEXT: [[CONV4:%.*]] = trunc i32 [[COND]] to i16 8219 // CHECK6-NEXT: store i16 [[CONV4]], i16* [[B2]], align 2 8220 // CHECK6-NEXT: [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4 8221 // CHECK6-NEXT: [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4 8222 // CHECK6-NEXT: [[TMP7:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 8223 // CHECK6-NEXT: [[TMP8:%.*]] = bitcast i32* [[A1]] to i8* 8224 // CHECK6-NEXT: store i8* [[TMP8]], i8** [[TMP7]], align 4 8225 // CHECK6-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 8226 // CHECK6-NEXT: [[TMP10:%.*]] = bitcast i16* [[B2]] to i8* 8227 // CHECK6-NEXT: store i8* [[TMP10]], i8** [[TMP9]], align 4 8228 // CHECK6-NEXT: [[TMP11:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 8229 // CHECK6-NEXT: [[TMP12:%.*]] = call i32 @__kmpc_nvptx_parallel_reduce_nowait_v2(%struct.ident_t* @[[GLOB3]], i32 [[TMP6]], i32 2, i32 8, i8* [[TMP11]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func12, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func13) 8230 // CHECK6-NEXT: [[TMP13:%.*]] = icmp eq i32 [[TMP12]], 1 8231 // CHECK6-NEXT: br i1 [[TMP13]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]] 8232 // CHECK6: .omp.reduction.then: 8233 // CHECK6-NEXT: [[TMP14:%.*]] = load i32, i32* [[TMP0]], align 4 8234 // CHECK6-NEXT: [[TMP15:%.*]] = load i32, i32* [[A1]], align 4 8235 // CHECK6-NEXT: [[OR5:%.*]] = or i32 [[TMP14]], [[TMP15]] 8236 // CHECK6-NEXT: store i32 [[OR5]], i32* [[TMP0]], align 4 8237 // CHECK6-NEXT: [[TMP16:%.*]] = load i16, i16* [[TMP1]], align 2 8238 // CHECK6-NEXT: [[CONV6:%.*]] = sext i16 [[TMP16]] to i32 8239 // CHECK6-NEXT: [[TMP17:%.*]] = load i16, i16* [[B2]], align 2 8240 // CHECK6-NEXT: [[CONV7:%.*]] = sext i16 [[TMP17]] to i32 8241 // CHECK6-NEXT: [[CMP8:%.*]] = icmp sgt i32 [[CONV6]], [[CONV7]] 8242 // CHECK6-NEXT: br i1 [[CMP8]], label [[COND_TRUE9:%.*]], label [[COND_FALSE10:%.*]] 8243 // CHECK6: cond.true9: 8244 // CHECK6-NEXT: [[TMP18:%.*]] = load i16, i16* [[TMP1]], align 2 8245 // CHECK6-NEXT: br label [[COND_END11:%.*]] 8246 // CHECK6: cond.false10: 8247 // CHECK6-NEXT: [[TMP19:%.*]] = load i16, i16* [[B2]], align 2 8248 // CHECK6-NEXT: br label [[COND_END11]] 8249 // CHECK6: cond.end11: 8250 // CHECK6-NEXT: [[COND12:%.*]] = phi i16 [ [[TMP18]], [[COND_TRUE9]] ], [ [[TMP19]], [[COND_FALSE10]] ] 8251 // CHECK6-NEXT: store i16 [[COND12]], i16* [[TMP1]], align 2 8252 // CHECK6-NEXT: call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP6]]) 8253 // CHECK6-NEXT: br label [[DOTOMP_REDUCTION_DONE]] 8254 // CHECK6: .omp.reduction.done: 8255 // CHECK6-NEXT: ret void 8256 // 8257 // 8258 // CHECK6-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func12 8259 // CHECK6-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR0]] { 8260 // CHECK6-NEXT: entry: 8261 // CHECK6-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 8262 // CHECK6-NEXT: [[DOTADDR1:%.*]] = alloca i16, align 2 8263 // CHECK6-NEXT: [[DOTADDR2:%.*]] = alloca i16, align 2 8264 // CHECK6-NEXT: [[DOTADDR3:%.*]] = alloca i16, align 2 8265 // CHECK6-NEXT: [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [2 x i8*], align 4 8266 // CHECK6-NEXT: [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca i32, align 4 8267 // CHECK6-NEXT: [[DOTOMP_REDUCTION_ELEMENT4:%.*]] = alloca i16, align 2 8268 // CHECK6-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 8269 // CHECK6-NEXT: store i16 [[TMP1]], i16* [[DOTADDR1]], align 2 8270 // CHECK6-NEXT: store i16 [[TMP2]], i16* [[DOTADDR2]], align 2 8271 // CHECK6-NEXT: store i16 [[TMP3]], i16* [[DOTADDR3]], align 2 8272 // CHECK6-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 4 8273 // CHECK6-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [2 x i8*]* 8274 // CHECK6-NEXT: [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2 8275 // CHECK6-NEXT: [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2 8276 // CHECK6-NEXT: [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2 8277 // CHECK6-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 0 8278 // CHECK6-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 4 8279 // CHECK6-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 8280 // CHECK6-NEXT: [[TMP12:%.*]] = bitcast i8* [[TMP10]] to i32* 8281 // CHECK6-NEXT: [[TMP13:%.*]] = getelementptr i32, i32* [[TMP12]], i32 1 8282 // CHECK6-NEXT: [[TMP14:%.*]] = bitcast i32* [[TMP13]] to i8* 8283 // CHECK6-NEXT: [[TMP15:%.*]] = load i32, i32* [[TMP12]], align 4 8284 // CHECK6-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 8285 // CHECK6-NEXT: [[TMP16:%.*]] = trunc i32 [[NVPTX_WARP_SIZE]] to i16 8286 // CHECK6-NEXT: [[TMP17:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP15]], i16 [[TMP7]], i16 [[TMP16]]) 8287 // CHECK6-NEXT: store i32 [[TMP17]], i32* [[DOTOMP_REDUCTION_ELEMENT]], align 4 8288 // CHECK6-NEXT: [[TMP18:%.*]] = getelementptr i32, i32* [[TMP12]], i32 1 8289 // CHECK6-NEXT: [[TMP19:%.*]] = getelementptr i32, i32* [[DOTOMP_REDUCTION_ELEMENT]], i32 1 8290 // CHECK6-NEXT: [[TMP20:%.*]] = bitcast i32* [[DOTOMP_REDUCTION_ELEMENT]] to i8* 8291 // CHECK6-NEXT: store i8* [[TMP20]], i8** [[TMP11]], align 4 8292 // CHECK6-NEXT: [[TMP21:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 1 8293 // CHECK6-NEXT: [[TMP22:%.*]] = load i8*, i8** [[TMP21]], align 4 8294 // CHECK6-NEXT: [[TMP23:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1 8295 // CHECK6-NEXT: [[TMP24:%.*]] = bitcast i8* [[TMP22]] to i16* 8296 // CHECK6-NEXT: [[TMP25:%.*]] = getelementptr i16, i16* [[TMP24]], i32 1 8297 // CHECK6-NEXT: [[TMP26:%.*]] = bitcast i16* [[TMP25]] to i8* 8298 // CHECK6-NEXT: [[TMP27:%.*]] = load i16, i16* [[TMP24]], align 2 8299 // CHECK6-NEXT: [[TMP28:%.*]] = sext i16 [[TMP27]] to i32 8300 // CHECK6-NEXT: [[NVPTX_WARP_SIZE5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 8301 // CHECK6-NEXT: [[TMP29:%.*]] = trunc i32 [[NVPTX_WARP_SIZE5]] to i16 8302 // CHECK6-NEXT: [[TMP30:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP28]], i16 [[TMP7]], i16 [[TMP29]]) 8303 // CHECK6-NEXT: [[TMP31:%.*]] = trunc i32 [[TMP30]] to i16 8304 // CHECK6-NEXT: store i16 [[TMP31]], i16* [[DOTOMP_REDUCTION_ELEMENT4]], align 2 8305 // CHECK6-NEXT: [[TMP32:%.*]] = getelementptr i16, i16* [[TMP24]], i32 1 8306 // CHECK6-NEXT: [[TMP33:%.*]] = getelementptr i16, i16* [[DOTOMP_REDUCTION_ELEMENT4]], i32 1 8307 // CHECK6-NEXT: [[TMP34:%.*]] = bitcast i16* [[DOTOMP_REDUCTION_ELEMENT4]] to i8* 8308 // CHECK6-NEXT: store i8* [[TMP34]], i8** [[TMP23]], align 4 8309 // CHECK6-NEXT: [[TMP35:%.*]] = icmp eq i16 [[TMP8]], 0 8310 // CHECK6-NEXT: [[TMP36:%.*]] = icmp eq i16 [[TMP8]], 1 8311 // CHECK6-NEXT: [[TMP37:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]] 8312 // CHECK6-NEXT: [[TMP38:%.*]] = and i1 [[TMP36]], [[TMP37]] 8313 // CHECK6-NEXT: [[TMP39:%.*]] = icmp eq i16 [[TMP8]], 2 8314 // CHECK6-NEXT: [[TMP40:%.*]] = and i16 [[TMP6]], 1 8315 // CHECK6-NEXT: [[TMP41:%.*]] = icmp eq i16 [[TMP40]], 0 8316 // CHECK6-NEXT: [[TMP42:%.*]] = and i1 [[TMP39]], [[TMP41]] 8317 // CHECK6-NEXT: [[TMP43:%.*]] = icmp sgt i16 [[TMP7]], 0 8318 // CHECK6-NEXT: [[TMP44:%.*]] = and i1 [[TMP42]], [[TMP43]] 8319 // CHECK6-NEXT: [[TMP45:%.*]] = or i1 [[TMP35]], [[TMP38]] 8320 // CHECK6-NEXT: [[TMP46:%.*]] = or i1 [[TMP45]], [[TMP44]] 8321 // CHECK6-NEXT: br i1 [[TMP46]], label [[THEN:%.*]], label [[ELSE:%.*]] 8322 // CHECK6: then: 8323 // CHECK6-NEXT: [[TMP47:%.*]] = bitcast [2 x i8*]* [[TMP5]] to i8* 8324 // CHECK6-NEXT: [[TMP48:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8* 8325 // CHECK6-NEXT: call void @"_omp$reduction$reduction_func11"(i8* [[TMP47]], i8* [[TMP48]]) #[[ATTR3]] 8326 // CHECK6-NEXT: br label [[IFCONT:%.*]] 8327 // CHECK6: else: 8328 // CHECK6-NEXT: br label [[IFCONT]] 8329 // CHECK6: ifcont: 8330 // CHECK6-NEXT: [[TMP49:%.*]] = icmp eq i16 [[TMP8]], 1 8331 // CHECK6-NEXT: [[TMP50:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]] 8332 // CHECK6-NEXT: [[TMP51:%.*]] = and i1 [[TMP49]], [[TMP50]] 8333 // CHECK6-NEXT: br i1 [[TMP51]], label [[THEN6:%.*]], label [[ELSE7:%.*]] 8334 // CHECK6: then6: 8335 // CHECK6-NEXT: [[TMP52:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 8336 // CHECK6-NEXT: [[TMP53:%.*]] = load i8*, i8** [[TMP52]], align 4 8337 // CHECK6-NEXT: [[TMP54:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 0 8338 // CHECK6-NEXT: [[TMP55:%.*]] = load i8*, i8** [[TMP54]], align 4 8339 // CHECK6-NEXT: [[TMP56:%.*]] = bitcast i8* [[TMP53]] to i32* 8340 // CHECK6-NEXT: [[TMP57:%.*]] = bitcast i8* [[TMP55]] to i32* 8341 // CHECK6-NEXT: [[TMP58:%.*]] = load i32, i32* [[TMP56]], align 4 8342 // CHECK6-NEXT: store i32 [[TMP58]], i32* [[TMP57]], align 4 8343 // CHECK6-NEXT: [[TMP59:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1 8344 // CHECK6-NEXT: [[TMP60:%.*]] = load i8*, i8** [[TMP59]], align 4 8345 // CHECK6-NEXT: [[TMP61:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 1 8346 // CHECK6-NEXT: [[TMP62:%.*]] = load i8*, i8** [[TMP61]], align 4 8347 // CHECK6-NEXT: [[TMP63:%.*]] = bitcast i8* [[TMP60]] to i16* 8348 // CHECK6-NEXT: [[TMP64:%.*]] = bitcast i8* [[TMP62]] to i16* 8349 // CHECK6-NEXT: [[TMP65:%.*]] = load i16, i16* [[TMP63]], align 2 8350 // CHECK6-NEXT: store i16 [[TMP65]], i16* [[TMP64]], align 2 8351 // CHECK6-NEXT: br label [[IFCONT8:%.*]] 8352 // CHECK6: else7: 8353 // CHECK6-NEXT: br label [[IFCONT8]] 8354 // CHECK6: ifcont8: 8355 // CHECK6-NEXT: ret void 8356 // 8357 // 8358 // CHECK6-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func13 8359 // CHECK6-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] { 8360 // CHECK6-NEXT: entry: 8361 // CHECK6-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 8362 // CHECK6-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 8363 // CHECK6-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]]) 8364 // CHECK6-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 8365 // CHECK6-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 8366 // CHECK6-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 8367 // CHECK6-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 8368 // CHECK6-NEXT: [[NVPTX_LANE_ID:%.*]] = and i32 [[NVPTX_TID2]], 31 8369 // CHECK6-NEXT: [[NVPTX_TID3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 8370 // CHECK6-NEXT: [[NVPTX_WARP_ID:%.*]] = ashr i32 [[NVPTX_TID3]], 5 8371 // CHECK6-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 8372 // CHECK6-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 8373 // CHECK6-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4:[0-9]+]], i32 [[TMP2]]) 8374 // CHECK6-NEXT: [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 8375 // CHECK6-NEXT: br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]] 8376 // CHECK6: then: 8377 // CHECK6-NEXT: [[TMP5:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 8378 // CHECK6-NEXT: [[TMP6:%.*]] = load i8*, i8** [[TMP5]], align 4 8379 // CHECK6-NEXT: [[TMP7:%.*]] = bitcast i8* [[TMP6]] to i32* 8380 // CHECK6-NEXT: [[TMP8:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 8381 // CHECK6-NEXT: [[TMP9:%.*]] = load i32, i32* [[TMP7]], align 4 8382 // CHECK6-NEXT: store volatile i32 [[TMP9]], i32 addrspace(3)* [[TMP8]], align 4 8383 // CHECK6-NEXT: br label [[IFCONT:%.*]] 8384 // CHECK6: else: 8385 // CHECK6-NEXT: br label [[IFCONT]] 8386 // CHECK6: ifcont: 8387 // CHECK6-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 8388 // CHECK6-NEXT: [[TMP10:%.*]] = load i32, i32* [[DOTADDR1]], align 4 8389 // CHECK6-NEXT: [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP10]] 8390 // CHECK6-NEXT: br i1 [[IS_ACTIVE_THREAD]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 8391 // CHECK6: then4: 8392 // CHECK6-NEXT: [[TMP11:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 8393 // CHECK6-NEXT: [[TMP12:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 8394 // CHECK6-NEXT: [[TMP13:%.*]] = load i8*, i8** [[TMP12]], align 4 8395 // CHECK6-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to i32* 8396 // CHECK6-NEXT: [[TMP15:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP11]], align 4 8397 // CHECK6-NEXT: store i32 [[TMP15]], i32* [[TMP14]], align 4 8398 // CHECK6-NEXT: br label [[IFCONT6:%.*]] 8399 // CHECK6: else5: 8400 // CHECK6-NEXT: br label [[IFCONT6]] 8401 // CHECK6: ifcont6: 8402 // CHECK6-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 8403 // CHECK6-NEXT: [[WARP_MASTER7:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 8404 // CHECK6-NEXT: br i1 [[WARP_MASTER7]], label [[THEN8:%.*]], label [[ELSE9:%.*]] 8405 // CHECK6: then8: 8406 // CHECK6-NEXT: [[TMP16:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 8407 // CHECK6-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 4 8408 // CHECK6-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i16* 8409 // CHECK6-NEXT: [[TMP19:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 8410 // CHECK6-NEXT: [[TMP20:%.*]] = bitcast i32 addrspace(3)* [[TMP19]] to i16 addrspace(3)* 8411 // CHECK6-NEXT: [[TMP21:%.*]] = load i16, i16* [[TMP18]], align 2 8412 // CHECK6-NEXT: store volatile i16 [[TMP21]], i16 addrspace(3)* [[TMP20]], align 2 8413 // CHECK6-NEXT: br label [[IFCONT10:%.*]] 8414 // CHECK6: else9: 8415 // CHECK6-NEXT: br label [[IFCONT10]] 8416 // CHECK6: ifcont10: 8417 // CHECK6-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 8418 // CHECK6-NEXT: [[TMP22:%.*]] = load i32, i32* [[DOTADDR1]], align 4 8419 // CHECK6-NEXT: [[IS_ACTIVE_THREAD11:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP22]] 8420 // CHECK6-NEXT: br i1 [[IS_ACTIVE_THREAD11]], label [[THEN12:%.*]], label [[ELSE13:%.*]] 8421 // CHECK6: then12: 8422 // CHECK6-NEXT: [[TMP23:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 8423 // CHECK6-NEXT: [[TMP24:%.*]] = bitcast i32 addrspace(3)* [[TMP23]] to i16 addrspace(3)* 8424 // CHECK6-NEXT: [[TMP25:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 8425 // CHECK6-NEXT: [[TMP26:%.*]] = load i8*, i8** [[TMP25]], align 4 8426 // CHECK6-NEXT: [[TMP27:%.*]] = bitcast i8* [[TMP26]] to i16* 8427 // CHECK6-NEXT: [[TMP28:%.*]] = load volatile i16, i16 addrspace(3)* [[TMP24]], align 2 8428 // CHECK6-NEXT: store i16 [[TMP28]], i16* [[TMP27]], align 2 8429 // CHECK6-NEXT: br label [[IFCONT14:%.*]] 8430 // CHECK6: else13: 8431 // CHECK6-NEXT: br label [[IFCONT14]] 8432 // CHECK6: ifcont14: 8433 // CHECK6-NEXT: ret void 8434 // 8435 // 8436 // CHECK6-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func15 8437 // CHECK6-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR0]] { 8438 // CHECK6-NEXT: entry: 8439 // CHECK6-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 8440 // CHECK6-NEXT: [[DOTADDR1:%.*]] = alloca i16, align 2 8441 // CHECK6-NEXT: [[DOTADDR2:%.*]] = alloca i16, align 2 8442 // CHECK6-NEXT: [[DOTADDR3:%.*]] = alloca i16, align 2 8443 // CHECK6-NEXT: [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [2 x i8*], align 4 8444 // CHECK6-NEXT: [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca i32, align 4 8445 // CHECK6-NEXT: [[DOTOMP_REDUCTION_ELEMENT4:%.*]] = alloca i16, align 2 8446 // CHECK6-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 8447 // CHECK6-NEXT: store i16 [[TMP1]], i16* [[DOTADDR1]], align 2 8448 // CHECK6-NEXT: store i16 [[TMP2]], i16* [[DOTADDR2]], align 2 8449 // CHECK6-NEXT: store i16 [[TMP3]], i16* [[DOTADDR3]], align 2 8450 // CHECK6-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 4 8451 // CHECK6-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [2 x i8*]* 8452 // CHECK6-NEXT: [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2 8453 // CHECK6-NEXT: [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2 8454 // CHECK6-NEXT: [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2 8455 // CHECK6-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 0 8456 // CHECK6-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 4 8457 // CHECK6-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 8458 // CHECK6-NEXT: [[TMP12:%.*]] = bitcast i8* [[TMP10]] to i32* 8459 // CHECK6-NEXT: [[TMP13:%.*]] = getelementptr i32, i32* [[TMP12]], i32 1 8460 // CHECK6-NEXT: [[TMP14:%.*]] = bitcast i32* [[TMP13]] to i8* 8461 // CHECK6-NEXT: [[TMP15:%.*]] = load i32, i32* [[TMP12]], align 4 8462 // CHECK6-NEXT: [[NVPTX_WARP_SIZE:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 8463 // CHECK6-NEXT: [[TMP16:%.*]] = trunc i32 [[NVPTX_WARP_SIZE]] to i16 8464 // CHECK6-NEXT: [[TMP17:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP15]], i16 [[TMP7]], i16 [[TMP16]]) 8465 // CHECK6-NEXT: store i32 [[TMP17]], i32* [[DOTOMP_REDUCTION_ELEMENT]], align 4 8466 // CHECK6-NEXT: [[TMP18:%.*]] = getelementptr i32, i32* [[TMP12]], i32 1 8467 // CHECK6-NEXT: [[TMP19:%.*]] = getelementptr i32, i32* [[DOTOMP_REDUCTION_ELEMENT]], i32 1 8468 // CHECK6-NEXT: [[TMP20:%.*]] = bitcast i32* [[DOTOMP_REDUCTION_ELEMENT]] to i8* 8469 // CHECK6-NEXT: store i8* [[TMP20]], i8** [[TMP11]], align 4 8470 // CHECK6-NEXT: [[TMP21:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 1 8471 // CHECK6-NEXT: [[TMP22:%.*]] = load i8*, i8** [[TMP21]], align 4 8472 // CHECK6-NEXT: [[TMP23:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1 8473 // CHECK6-NEXT: [[TMP24:%.*]] = bitcast i8* [[TMP22]] to i16* 8474 // CHECK6-NEXT: [[TMP25:%.*]] = getelementptr i16, i16* [[TMP24]], i32 1 8475 // CHECK6-NEXT: [[TMP26:%.*]] = bitcast i16* [[TMP25]] to i8* 8476 // CHECK6-NEXT: [[TMP27:%.*]] = load i16, i16* [[TMP24]], align 2 8477 // CHECK6-NEXT: [[TMP28:%.*]] = sext i16 [[TMP27]] to i32 8478 // CHECK6-NEXT: [[NVPTX_WARP_SIZE5:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() 8479 // CHECK6-NEXT: [[TMP29:%.*]] = trunc i32 [[NVPTX_WARP_SIZE5]] to i16 8480 // CHECK6-NEXT: [[TMP30:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP28]], i16 [[TMP7]], i16 [[TMP29]]) 8481 // CHECK6-NEXT: [[TMP31:%.*]] = trunc i32 [[TMP30]] to i16 8482 // CHECK6-NEXT: store i16 [[TMP31]], i16* [[DOTOMP_REDUCTION_ELEMENT4]], align 2 8483 // CHECK6-NEXT: [[TMP32:%.*]] = getelementptr i16, i16* [[TMP24]], i32 1 8484 // CHECK6-NEXT: [[TMP33:%.*]] = getelementptr i16, i16* [[DOTOMP_REDUCTION_ELEMENT4]], i32 1 8485 // CHECK6-NEXT: [[TMP34:%.*]] = bitcast i16* [[DOTOMP_REDUCTION_ELEMENT4]] to i8* 8486 // CHECK6-NEXT: store i8* [[TMP34]], i8** [[TMP23]], align 4 8487 // CHECK6-NEXT: [[TMP35:%.*]] = icmp eq i16 [[TMP8]], 0 8488 // CHECK6-NEXT: [[TMP36:%.*]] = icmp eq i16 [[TMP8]], 1 8489 // CHECK6-NEXT: [[TMP37:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]] 8490 // CHECK6-NEXT: [[TMP38:%.*]] = and i1 [[TMP36]], [[TMP37]] 8491 // CHECK6-NEXT: [[TMP39:%.*]] = icmp eq i16 [[TMP8]], 2 8492 // CHECK6-NEXT: [[TMP40:%.*]] = and i16 [[TMP6]], 1 8493 // CHECK6-NEXT: [[TMP41:%.*]] = icmp eq i16 [[TMP40]], 0 8494 // CHECK6-NEXT: [[TMP42:%.*]] = and i1 [[TMP39]], [[TMP41]] 8495 // CHECK6-NEXT: [[TMP43:%.*]] = icmp sgt i16 [[TMP7]], 0 8496 // CHECK6-NEXT: [[TMP44:%.*]] = and i1 [[TMP42]], [[TMP43]] 8497 // CHECK6-NEXT: [[TMP45:%.*]] = or i1 [[TMP35]], [[TMP38]] 8498 // CHECK6-NEXT: [[TMP46:%.*]] = or i1 [[TMP45]], [[TMP44]] 8499 // CHECK6-NEXT: br i1 [[TMP46]], label [[THEN:%.*]], label [[ELSE:%.*]] 8500 // CHECK6: then: 8501 // CHECK6-NEXT: [[TMP47:%.*]] = bitcast [2 x i8*]* [[TMP5]] to i8* 8502 // CHECK6-NEXT: [[TMP48:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8* 8503 // CHECK6-NEXT: call void @"_omp$reduction$reduction_func14"(i8* [[TMP47]], i8* [[TMP48]]) #[[ATTR3]] 8504 // CHECK6-NEXT: br label [[IFCONT:%.*]] 8505 // CHECK6: else: 8506 // CHECK6-NEXT: br label [[IFCONT]] 8507 // CHECK6: ifcont: 8508 // CHECK6-NEXT: [[TMP49:%.*]] = icmp eq i16 [[TMP8]], 1 8509 // CHECK6-NEXT: [[TMP50:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]] 8510 // CHECK6-NEXT: [[TMP51:%.*]] = and i1 [[TMP49]], [[TMP50]] 8511 // CHECK6-NEXT: br i1 [[TMP51]], label [[THEN6:%.*]], label [[ELSE7:%.*]] 8512 // CHECK6: then6: 8513 // CHECK6-NEXT: [[TMP52:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0 8514 // CHECK6-NEXT: [[TMP53:%.*]] = load i8*, i8** [[TMP52]], align 4 8515 // CHECK6-NEXT: [[TMP54:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 0 8516 // CHECK6-NEXT: [[TMP55:%.*]] = load i8*, i8** [[TMP54]], align 4 8517 // CHECK6-NEXT: [[TMP56:%.*]] = bitcast i8* [[TMP53]] to i32* 8518 // CHECK6-NEXT: [[TMP57:%.*]] = bitcast i8* [[TMP55]] to i32* 8519 // CHECK6-NEXT: [[TMP58:%.*]] = load i32, i32* [[TMP56]], align 4 8520 // CHECK6-NEXT: store i32 [[TMP58]], i32* [[TMP57]], align 4 8521 // CHECK6-NEXT: [[TMP59:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1 8522 // CHECK6-NEXT: [[TMP60:%.*]] = load i8*, i8** [[TMP59]], align 4 8523 // CHECK6-NEXT: [[TMP61:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i32 0, i32 1 8524 // CHECK6-NEXT: [[TMP62:%.*]] = load i8*, i8** [[TMP61]], align 4 8525 // CHECK6-NEXT: [[TMP63:%.*]] = bitcast i8* [[TMP60]] to i16* 8526 // CHECK6-NEXT: [[TMP64:%.*]] = bitcast i8* [[TMP62]] to i16* 8527 // CHECK6-NEXT: [[TMP65:%.*]] = load i16, i16* [[TMP63]], align 2 8528 // CHECK6-NEXT: store i16 [[TMP65]], i16* [[TMP64]], align 2 8529 // CHECK6-NEXT: br label [[IFCONT8:%.*]] 8530 // CHECK6: else7: 8531 // CHECK6-NEXT: br label [[IFCONT8]] 8532 // CHECK6: ifcont8: 8533 // CHECK6-NEXT: ret void 8534 // 8535 // 8536 // CHECK6-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func16 8537 // CHECK6-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] { 8538 // CHECK6-NEXT: entry: 8539 // CHECK6-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 8540 // CHECK6-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 8541 // CHECK6-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]]) 8542 // CHECK6-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 8543 // CHECK6-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 8544 // CHECK6-NEXT: [[NVPTX_TID:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 8545 // CHECK6-NEXT: [[NVPTX_TID2:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 8546 // CHECK6-NEXT: [[NVPTX_LANE_ID:%.*]] = and i32 [[NVPTX_TID2]], 31 8547 // CHECK6-NEXT: [[NVPTX_TID3:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() 8548 // CHECK6-NEXT: [[NVPTX_WARP_ID:%.*]] = ashr i32 [[NVPTX_TID3]], 5 8549 // CHECK6-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 8550 // CHECK6-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 8551 // CHECK6-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 8552 // CHECK6-NEXT: [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 8553 // CHECK6-NEXT: br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]] 8554 // CHECK6: then: 8555 // CHECK6-NEXT: [[TMP5:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 8556 // CHECK6-NEXT: [[TMP6:%.*]] = load i8*, i8** [[TMP5]], align 4 8557 // CHECK6-NEXT: [[TMP7:%.*]] = bitcast i8* [[TMP6]] to i32* 8558 // CHECK6-NEXT: [[TMP8:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 8559 // CHECK6-NEXT: [[TMP9:%.*]] = load i32, i32* [[TMP7]], align 4 8560 // CHECK6-NEXT: store volatile i32 [[TMP9]], i32 addrspace(3)* [[TMP8]], align 4 8561 // CHECK6-NEXT: br label [[IFCONT:%.*]] 8562 // CHECK6: else: 8563 // CHECK6-NEXT: br label [[IFCONT]] 8564 // CHECK6: ifcont: 8565 // CHECK6-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 8566 // CHECK6-NEXT: [[TMP10:%.*]] = load i32, i32* [[DOTADDR1]], align 4 8567 // CHECK6-NEXT: [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP10]] 8568 // CHECK6-NEXT: br i1 [[IS_ACTIVE_THREAD]], label [[THEN4:%.*]], label [[ELSE5:%.*]] 8569 // CHECK6: then4: 8570 // CHECK6-NEXT: [[TMP11:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 8571 // CHECK6-NEXT: [[TMP12:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 8572 // CHECK6-NEXT: [[TMP13:%.*]] = load i8*, i8** [[TMP12]], align 4 8573 // CHECK6-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to i32* 8574 // CHECK6-NEXT: [[TMP15:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP11]], align 4 8575 // CHECK6-NEXT: store i32 [[TMP15]], i32* [[TMP14]], align 4 8576 // CHECK6-NEXT: br label [[IFCONT6:%.*]] 8577 // CHECK6: else5: 8578 // CHECK6-NEXT: br label [[IFCONT6]] 8579 // CHECK6: ifcont6: 8580 // CHECK6-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 8581 // CHECK6-NEXT: [[WARP_MASTER7:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0 8582 // CHECK6-NEXT: br i1 [[WARP_MASTER7]], label [[THEN8:%.*]], label [[ELSE9:%.*]] 8583 // CHECK6: then8: 8584 // CHECK6-NEXT: [[TMP16:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 8585 // CHECK6-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 4 8586 // CHECK6-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i16* 8587 // CHECK6-NEXT: [[TMP19:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]] 8588 // CHECK6-NEXT: [[TMP20:%.*]] = bitcast i32 addrspace(3)* [[TMP19]] to i16 addrspace(3)* 8589 // CHECK6-NEXT: [[TMP21:%.*]] = load i16, i16* [[TMP18]], align 2 8590 // CHECK6-NEXT: store volatile i16 [[TMP21]], i16 addrspace(3)* [[TMP20]], align 2 8591 // CHECK6-NEXT: br label [[IFCONT10:%.*]] 8592 // CHECK6: else9: 8593 // CHECK6-NEXT: br label [[IFCONT10]] 8594 // CHECK6: ifcont10: 8595 // CHECK6-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]]) 8596 // CHECK6-NEXT: [[TMP22:%.*]] = load i32, i32* [[DOTADDR1]], align 4 8597 // CHECK6-NEXT: [[IS_ACTIVE_THREAD11:%.*]] = icmp ult i32 [[NVPTX_TID]], [[TMP22]] 8598 // CHECK6-NEXT: br i1 [[IS_ACTIVE_THREAD11]], label [[THEN12:%.*]], label [[ELSE13:%.*]] 8599 // CHECK6: then12: 8600 // CHECK6-NEXT: [[TMP23:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_TID]] 8601 // CHECK6-NEXT: [[TMP24:%.*]] = bitcast i32 addrspace(3)* [[TMP23]] to i16 addrspace(3)* 8602 // CHECK6-NEXT: [[TMP25:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 8603 // CHECK6-NEXT: [[TMP26:%.*]] = load i8*, i8** [[TMP25]], align 4 8604 // CHECK6-NEXT: [[TMP27:%.*]] = bitcast i8* [[TMP26]] to i16* 8605 // CHECK6-NEXT: [[TMP28:%.*]] = load volatile i16, i16 addrspace(3)* [[TMP24]], align 2 8606 // CHECK6-NEXT: store i16 [[TMP28]], i16* [[TMP27]], align 2 8607 // CHECK6-NEXT: br label [[IFCONT14:%.*]] 8608 // CHECK6: else13: 8609 // CHECK6-NEXT: br label [[IFCONT14]] 8610 // CHECK6: ifcont14: 8611 // CHECK6-NEXT: ret void 8612 // 8613 // 8614 // CHECK6-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_copy_func17 8615 // CHECK6-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 8616 // CHECK6-NEXT: entry: 8617 // CHECK6-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 8618 // CHECK6-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 8619 // CHECK6-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 8620 // CHECK6-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 8621 // CHECK6-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 8622 // CHECK6-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 8623 // CHECK6-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 8624 // CHECK6-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 8625 // CHECK6-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 4 8626 // CHECK6-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.4* 8627 // CHECK6-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 8628 // CHECK6-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 8629 // CHECK6-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 4 8630 // CHECK6-NEXT: [[TMP10:%.*]] = bitcast i8* [[TMP9]] to i32* 8631 // CHECK6-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4:%.*]], %struct._globalized_locals_ty.4* [[TMP6]], i32 0, i32 0 8632 // CHECK6-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2048 x i32], [2048 x i32]* [[A]], i32 0, i32 [[TMP7]] 8633 // CHECK6-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP10]], align 4 8634 // CHECK6-NEXT: store i32 [[TMP12]], i32* [[TMP11]], align 128 8635 // CHECK6-NEXT: [[TMP13:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 8636 // CHECK6-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 4 8637 // CHECK6-NEXT: [[TMP15:%.*]] = bitcast i8* [[TMP14]] to i16* 8638 // CHECK6-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4]], %struct._globalized_locals_ty.4* [[TMP6]], i32 0, i32 1 8639 // CHECK6-NEXT: [[TMP16:%.*]] = getelementptr inbounds [2048 x i16], [2048 x i16]* [[B]], i32 0, i32 [[TMP7]] 8640 // CHECK6-NEXT: [[TMP17:%.*]] = load i16, i16* [[TMP15]], align 2 8641 // CHECK6-NEXT: store i16 [[TMP17]], i16* [[TMP16]], align 128 8642 // CHECK6-NEXT: ret void 8643 // 8644 // 8645 // CHECK6-LABEL: define {{[^@]+}}@_omp_reduction_list_to_global_reduce_func18 8646 // CHECK6-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 8647 // CHECK6-NEXT: entry: 8648 // CHECK6-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 8649 // CHECK6-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 8650 // CHECK6-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 8651 // CHECK6-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 8652 // CHECK6-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 8653 // CHECK6-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 8654 // CHECK6-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 8655 // CHECK6-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 8656 // CHECK6-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.4* 8657 // CHECK6-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 8658 // CHECK6-NEXT: [[TMP6:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 8659 // CHECK6-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4:%.*]], %struct._globalized_locals_ty.4* [[TMP4]], i32 0, i32 0 8660 // CHECK6-NEXT: [[TMP7:%.*]] = getelementptr inbounds [2048 x i32], [2048 x i32]* [[A]], i32 0, i32 [[TMP5]] 8661 // CHECK6-NEXT: [[TMP8:%.*]] = bitcast i32* [[TMP7]] to i8* 8662 // CHECK6-NEXT: store i8* [[TMP8]], i8** [[TMP6]], align 4 8663 // CHECK6-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 8664 // CHECK6-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4]], %struct._globalized_locals_ty.4* [[TMP4]], i32 0, i32 1 8665 // CHECK6-NEXT: [[TMP10:%.*]] = getelementptr inbounds [2048 x i16], [2048 x i16]* [[B]], i32 0, i32 [[TMP5]] 8666 // CHECK6-NEXT: [[TMP11:%.*]] = bitcast i16* [[TMP10]] to i8* 8667 // CHECK6-NEXT: store i8* [[TMP11]], i8** [[TMP9]], align 4 8668 // CHECK6-NEXT: [[TMP12:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 8669 // CHECK6-NEXT: [[TMP13:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 8670 // CHECK6-NEXT: call void @"_omp$reduction$reduction_func14"(i8* [[TMP12]], i8* [[TMP13]]) #[[ATTR3]] 8671 // CHECK6-NEXT: ret void 8672 // 8673 // 8674 // CHECK6-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_copy_func19 8675 // CHECK6-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 8676 // CHECK6-NEXT: entry: 8677 // CHECK6-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 8678 // CHECK6-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 8679 // CHECK6-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 8680 // CHECK6-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 8681 // CHECK6-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 8682 // CHECK6-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 8683 // CHECK6-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 8684 // CHECK6-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to [2 x i8*]* 8685 // CHECK6-NEXT: [[TMP5:%.*]] = load i8*, i8** [[DOTADDR]], align 4 8686 // CHECK6-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP5]] to %struct._globalized_locals_ty.4* 8687 // CHECK6-NEXT: [[TMP7:%.*]] = load i32, i32* [[DOTADDR1]], align 4 8688 // CHECK6-NEXT: [[TMP8:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 0 8689 // CHECK6-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 4 8690 // CHECK6-NEXT: [[TMP10:%.*]] = bitcast i8* [[TMP9]] to i32* 8691 // CHECK6-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4:%.*]], %struct._globalized_locals_ty.4* [[TMP6]], i32 0, i32 0 8692 // CHECK6-NEXT: [[TMP11:%.*]] = getelementptr inbounds [2048 x i32], [2048 x i32]* [[A]], i32 0, i32 [[TMP7]] 8693 // CHECK6-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 128 8694 // CHECK6-NEXT: store i32 [[TMP12]], i32* [[TMP10]], align 4 8695 // CHECK6-NEXT: [[TMP13:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP4]], i32 0, i32 1 8696 // CHECK6-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 4 8697 // CHECK6-NEXT: [[TMP15:%.*]] = bitcast i8* [[TMP14]] to i16* 8698 // CHECK6-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4]], %struct._globalized_locals_ty.4* [[TMP6]], i32 0, i32 1 8699 // CHECK6-NEXT: [[TMP16:%.*]] = getelementptr inbounds [2048 x i16], [2048 x i16]* [[B]], i32 0, i32 [[TMP7]] 8700 // CHECK6-NEXT: [[TMP17:%.*]] = load i16, i16* [[TMP16]], align 128 8701 // CHECK6-NEXT: store i16 [[TMP17]], i16* [[TMP15]], align 2 8702 // CHECK6-NEXT: ret void 8703 // 8704 // 8705 // CHECK6-LABEL: define {{[^@]+}}@_omp_reduction_global_to_list_reduce_func20 8706 // CHECK6-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]], i8* [[TMP2:%.*]]) #[[ATTR0]] { 8707 // CHECK6-NEXT: entry: 8708 // CHECK6-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 4 8709 // CHECK6-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4 8710 // CHECK6-NEXT: [[DOTADDR2:%.*]] = alloca i8*, align 4 8711 // CHECK6-NEXT: [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x i8*], align 4 8712 // CHECK6-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 4 8713 // CHECK6-NEXT: store i32 [[TMP1]], i32* [[DOTADDR1]], align 4 8714 // CHECK6-NEXT: store i8* [[TMP2]], i8** [[DOTADDR2]], align 4 8715 // CHECK6-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 4 8716 // CHECK6-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to %struct._globalized_locals_ty.4* 8717 // CHECK6-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTADDR1]], align 4 8718 // CHECK6-NEXT: [[TMP6:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0 8719 // CHECK6-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4:%.*]], %struct._globalized_locals_ty.4* [[TMP4]], i32 0, i32 0 8720 // CHECK6-NEXT: [[TMP7:%.*]] = getelementptr inbounds [2048 x i32], [2048 x i32]* [[A]], i32 0, i32 [[TMP5]] 8721 // CHECK6-NEXT: [[TMP8:%.*]] = bitcast i32* [[TMP7]] to i8* 8722 // CHECK6-NEXT: store i8* [[TMP8]], i8** [[TMP6]], align 4 8723 // CHECK6-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1 8724 // CHECK6-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_4]], %struct._globalized_locals_ty.4* [[TMP4]], i32 0, i32 1 8725 // CHECK6-NEXT: [[TMP10:%.*]] = getelementptr inbounds [2048 x i16], [2048 x i16]* [[B]], i32 0, i32 [[TMP5]] 8726 // CHECK6-NEXT: [[TMP11:%.*]] = bitcast i16* [[TMP10]] to i8* 8727 // CHECK6-NEXT: store i8* [[TMP11]], i8** [[TMP9]], align 4 8728 // CHECK6-NEXT: [[TMP12:%.*]] = bitcast [2 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8* 8729 // CHECK6-NEXT: [[TMP13:%.*]] = load i8*, i8** [[DOTADDR2]], align 4 8730 // CHECK6-NEXT: call void @"_omp$reduction$reduction_func14"(i8* [[TMP13]], i8* [[TMP12]]) #[[ATTR3]] 8731 // CHECK6-NEXT: ret void 8732 // 8733