clang/test/OpenMP/nvptx_target_parallel_reduction_codegen.cpp

   1 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --include-generated-funcs --replace-value-regex "__omp_offloading_[0-9a-z]+_[0-9a-z]+" "reduction_size[.].+[.]" "pl_cond[.].+[.|,]" --prefix-filecheck-ir-name _
   2 // Test target codegen - host bc file has to be created first.
   3 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-cuda-mode -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 -fopenmp-cuda-mode -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix=CHECK-64
   5 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-cuda-mode -x c++ -triple i386-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm-bc %s -o %t-x86-host.bc
   6 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-cuda-mode -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix=CHECK-32
   7 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-cuda-mode -fexceptions -fcxx-exceptions -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix=CHECK-32-EX
   8 // expected-no-diagnostics
   9 #ifndef HEADER
  10 #define HEADER
  11
  12 // Check for the data transfer medium in shared memory to transfer the reduction list to the first warp.
  13
  14 // Check that the execution mode of all 3 target regions is set to Spmd Mode.
  15
  16 template<typename tx>
  17 tx ftemplate(int n) {
  18   int a;
  19   short b;
  20   tx c;
  21   float d;
  22   double e;
  23
  24   #pragma omp target parallel reduction(+: e)
  25   {
  26     e += 5;
  27   }
  28
  29   #pragma omp target parallel reduction(^: c) reduction(*: d)
  30   {
  31     c ^= 2;
  32     d *= 33;
  33   }
  34
  35   #pragma omp target parallel reduction(|: a) reduction(max: b)
  36   {
  37     a |= 1;
  38     b = 99 > b ? 99 : b;
  39   }
  40
  41   return a+b+c+d+e;
  42 }
  43
  44 int bar(int n){
  45   int a = 0;
  46
  47   a += ftemplate<char>(n);
  48
  49   return a;
  50 }
  51
  52 // define internal void [[PFN]](
  53
  54
  55 // Reduction function
  56
  57 // Shuffle and reduce function
  58 // Condition to reduce
  59 // Now check if we should just copy over the remote reduction list
  60
  61 // Inter warp copy function
  62 // [[DO_COPY]]
  63 // Barrier after copy to shared memory storage medium.
  64 // Read into warp 0.
  65
  66 // define internal void [[PFN1]](
  67
  68 // Reduction function
  69
  70 // Shuffle and reduce function
  71 // Condition to reduce
  72 // Now check if we should just copy over the remote reduction list
  73
  74 // Inter warp copy function
  75 // [[DO_COPY]]
  76 // Barrier after copy to shared memory storage medium.
  77 // Read into warp 0.
  78 // [[DO_COPY]]
  79 // Barrier after copy to shared memory storage medium.
  80 // Read into warp 0.
  81
  82 // define internal void [[PFN2]](
  83
  84
  85 // Reduction function
  86
  87 // Shuffle and reduce function
  88 // Condition to reduce
  89 // Now check if we should just copy over the remote reduction list
  90
  91 // Inter warp copy function
  92 // [[DO_COPY]]
  93 // Barrier after copy to shared memory storage medium.
  94 // Read into warp 0.
  95 // [[DO_COPY]]
  96 // Barrier after copy to shared memory storage medium.
  97 // Read into warp 0.
  98
  99 #endif
 100 // CHECK-64-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l24
 101 // CHECK-64-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[E:%.*]]) #[[ATTR0:[0-9]+]] {
 102 // CHECK-64-NEXT:  entry:
 103 // CHECK-64-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8
 104 // CHECK-64-NEXT:    [[E_ADDR:%.*]] = alloca ptr, align 8
 105 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x ptr], align 8
 106 // CHECK-64-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 8
 107 // CHECK-64-NEXT:    store ptr [[E]], ptr [[E_ADDR]], align 8
 108 // CHECK-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[E_ADDR]], align 8
 109 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l24_kernel_environment, ptr [[DYN_PTR]])
 110 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 111 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 112 // CHECK-64:       user_code.entry:
 113 // CHECK-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 114 // CHECK-64-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
 115 // CHECK-64-NEXT:    store ptr [[TMP0]], ptr [[TMP3]], align 8
 116 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l24_omp_outlined, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 1)
 117 // CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 118 // CHECK-64-NEXT:    ret void
 119 // CHECK-64:       worker.exit:
 120 // CHECK-64-NEXT:    ret void
 121 //
 122 //
 123 // CHECK-64-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l24_omp_outlined
 124 // CHECK-64-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[E:%.*]]) #[[ATTR1:[0-9]+]] {
 125 // CHECK-64-NEXT:  entry:
 126 // CHECK-64-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 127 // CHECK-64-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 128 // CHECK-64-NEXT:    [[E_ADDR:%.*]] = alloca ptr, align 8
 129 // CHECK-64-NEXT:    [[E1:%.*]] = alloca double, align 8
 130 // CHECK-64-NEXT:    [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x ptr], align 8
 131 // CHECK-64-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 132 // CHECK-64-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 133 // CHECK-64-NEXT:    store ptr [[E]], ptr [[E_ADDR]], align 8
 134 // CHECK-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[E_ADDR]], align 8
 135 // CHECK-64-NEXT:    store double 0.000000e+00, ptr [[E1]], align 8
 136 // CHECK-64-NEXT:    [[TMP1:%.*]] = load double, ptr [[E1]], align 8
 137 // CHECK-64-NEXT:    [[ADD:%.*]] = fadd double [[TMP1]], 5.000000e+00
 138 // CHECK-64-NEXT:    store double [[ADD]], ptr [[E1]], align 8
 139 // CHECK-64-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0
 140 // CHECK-64-NEXT:    store ptr [[E1]], ptr [[TMP2]], align 8
 141 // CHECK-64-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_nvptx_parallel_reduce_nowait_v2(ptr @[[GLOB1]], i64 8, ptr [[DOTOMP_REDUCTION_RED_LIST]], ptr @_omp_reduction_shuffle_and_reduce_func, ptr @_omp_reduction_inter_warp_copy_func)
 142 // CHECK-64-NEXT:    [[TMP4:%.*]] = icmp eq i32 [[TMP3]], 1
 143 // CHECK-64-NEXT:    br i1 [[TMP4]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]]
 144 // CHECK-64:       .omp.reduction.then:
 145 // CHECK-64-NEXT:    [[TMP5:%.*]] = load double, ptr [[TMP0]], align 8
 146 // CHECK-64-NEXT:    [[TMP6:%.*]] = load double, ptr [[E1]], align 8
 147 // CHECK-64-NEXT:    [[ADD2:%.*]] = fadd double [[TMP5]], [[TMP6]]
 148 // CHECK-64-NEXT:    store double [[ADD2]], ptr [[TMP0]], align 8
 149 // CHECK-64-NEXT:    br label [[DOTOMP_REDUCTION_DONE]]
 150 // CHECK-64:       .omp.reduction.done:
 151 // CHECK-64-NEXT:    ret void
 152 //
 153 //
 154 // CHECK-64-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func
 155 // CHECK-64-SAME: (ptr noundef [[TMP0:%.*]], i16 noundef signext [[TMP1:%.*]], i16 noundef signext [[TMP2:%.*]], i16 noundef signext [[TMP3:%.*]]) #[[ATTR2:[0-9]+]] {
 156 // CHECK-64-NEXT:  entry:
 157 // CHECK-64-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 158 // CHECK-64-NEXT:    [[DOTADDR1:%.*]] = alloca i16, align 2
 159 // CHECK-64-NEXT:    [[DOTADDR2:%.*]] = alloca i16, align 2
 160 // CHECK-64-NEXT:    [[DOTADDR3:%.*]] = alloca i16, align 2
 161 // CHECK-64-NEXT:    [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [1 x ptr], align 8
 162 // CHECK-64-NEXT:    [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca double, align 8
 163 // CHECK-64-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 164 // CHECK-64-NEXT:    store i16 [[TMP1]], ptr [[DOTADDR1]], align 2
 165 // CHECK-64-NEXT:    store i16 [[TMP2]], ptr [[DOTADDR2]], align 2
 166 // CHECK-64-NEXT:    store i16 [[TMP3]], ptr [[DOTADDR3]], align 2
 167 // CHECK-64-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 168 // CHECK-64-NEXT:    [[TMP5:%.*]] = load i16, ptr [[DOTADDR1]], align 2
 169 // CHECK-64-NEXT:    [[TMP6:%.*]] = load i16, ptr [[DOTADDR2]], align 2
 170 // CHECK-64-NEXT:    [[TMP7:%.*]] = load i16, ptr [[DOTADDR3]], align 2
 171 // CHECK-64-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP4]], i64 0, i64 0
 172 // CHECK-64-NEXT:    [[TMP9:%.*]] = load ptr, ptr [[TMP8]], align 8
 173 // CHECK-64-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0
 174 // CHECK-64-NEXT:    [[TMP11:%.*]] = getelementptr double, ptr [[TMP9]], i64 1
 175 // CHECK-64-NEXT:    [[TMP12:%.*]] = load i64, ptr [[TMP9]], align 8
 176 // CHECK-64-NEXT:    [[TMP13:%.*]] = call i32 @__kmpc_get_warp_size()
 177 // CHECK-64-NEXT:    [[TMP14:%.*]] = trunc i32 [[TMP13]] to i16
 178 // CHECK-64-NEXT:    [[TMP15:%.*]] = call i64 @__kmpc_shuffle_int64(i64 [[TMP12]], i16 [[TMP6]], i16 [[TMP14]])
 179 // CHECK-64-NEXT:    store i64 [[TMP15]], ptr [[DOTOMP_REDUCTION_ELEMENT]], align 8
 180 // CHECK-64-NEXT:    [[TMP16:%.*]] = getelementptr i64, ptr [[TMP9]], i64 1
 181 // CHECK-64-NEXT:    [[TMP17:%.*]] = getelementptr i64, ptr [[DOTOMP_REDUCTION_ELEMENT]], i64 1
 182 // CHECK-64-NEXT:    store ptr [[DOTOMP_REDUCTION_ELEMENT]], ptr [[TMP10]], align 8
 183 // CHECK-64-NEXT:    [[TMP18:%.*]] = icmp eq i16 [[TMP7]], 0
 184 // CHECK-64-NEXT:    [[TMP19:%.*]] = icmp eq i16 [[TMP7]], 1
 185 // CHECK-64-NEXT:    [[TMP20:%.*]] = icmp ult i16 [[TMP5]], [[TMP6]]
 186 // CHECK-64-NEXT:    [[TMP21:%.*]] = and i1 [[TMP19]], [[TMP20]]
 187 // CHECK-64-NEXT:    [[TMP22:%.*]] = icmp eq i16 [[TMP7]], 2
 188 // CHECK-64-NEXT:    [[TMP23:%.*]] = and i16 [[TMP5]], 1
 189 // CHECK-64-NEXT:    [[TMP24:%.*]] = icmp eq i16 [[TMP23]], 0
 190 // CHECK-64-NEXT:    [[TMP25:%.*]] = and i1 [[TMP22]], [[TMP24]]
 191 // CHECK-64-NEXT:    [[TMP26:%.*]] = icmp sgt i16 [[TMP6]], 0
 192 // CHECK-64-NEXT:    [[TMP27:%.*]] = and i1 [[TMP25]], [[TMP26]]
 193 // CHECK-64-NEXT:    [[TMP28:%.*]] = or i1 [[TMP18]], [[TMP21]]
 194 // CHECK-64-NEXT:    [[TMP29:%.*]] = or i1 [[TMP28]], [[TMP27]]
 195 // CHECK-64-NEXT:    br i1 [[TMP29]], label [[THEN:%.*]], label [[ELSE:%.*]]
 196 // CHECK-64:       then:
 197 // CHECK-64-NEXT:    call void @"{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l24_omp_outlined_omp$reduction$reduction_func"(ptr [[TMP4]], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]]) #[[ATTR3:[0-9]+]]
 198 // CHECK-64-NEXT:    br label [[IFCONT:%.*]]
 199 // CHECK-64:       else:
 200 // CHECK-64-NEXT:    br label [[IFCONT]]
 201 // CHECK-64:       ifcont:
 202 // CHECK-64-NEXT:    [[TMP30:%.*]] = icmp eq i16 [[TMP7]], 1
 203 // CHECK-64-NEXT:    [[TMP31:%.*]] = icmp uge i16 [[TMP5]], [[TMP6]]
 204 // CHECK-64-NEXT:    [[TMP32:%.*]] = and i1 [[TMP30]], [[TMP31]]
 205 // CHECK-64-NEXT:    br i1 [[TMP32]], label [[THEN4:%.*]], label [[ELSE5:%.*]]
 206 // CHECK-64:       then4:
 207 // CHECK-64-NEXT:    [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0
 208 // CHECK-64-NEXT:    [[TMP34:%.*]] = load ptr, ptr [[TMP33]], align 8
 209 // CHECK-64-NEXT:    [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP4]], i64 0, i64 0
 210 // CHECK-64-NEXT:    [[TMP36:%.*]] = load ptr, ptr [[TMP35]], align 8
 211 // CHECK-64-NEXT:    [[TMP37:%.*]] = load double, ptr [[TMP34]], align 8
 212 // CHECK-64-NEXT:    store double [[TMP37]], ptr [[TMP36]], align 8
 213 // CHECK-64-NEXT:    br label [[IFCONT6:%.*]]
 214 // CHECK-64:       else5:
 215 // CHECK-64-NEXT:    br label [[IFCONT6]]
 216 // CHECK-64:       ifcont6:
 217 // CHECK-64-NEXT:    ret void
 218 //
 219 //
 220 // CHECK-64-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func
 221 // CHECK-64-SAME: (ptr noundef [[TMP0:%.*]], i32 noundef [[TMP1:%.*]]) #[[ATTR2]] {
 222 // CHECK-64-NEXT:  entry:
 223 // CHECK-64-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 224 // CHECK-64-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
 225 // CHECK-64-NEXT:    [[DOTCNT_ADDR:%.*]] = alloca i32, align 4
 226 // CHECK-64-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 227 // CHECK-64-NEXT:    store i32 [[TMP1]], ptr [[DOTADDR1]], align 4
 228 // CHECK-64-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
 229 // CHECK-64-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
 230 // CHECK-64-NEXT:    [[NVPTX_LANE_ID:%.*]] = and i32 [[TMP4]], 31
 231 // CHECK-64-NEXT:    [[TMP5:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
 232 // CHECK-64-NEXT:    [[NVPTX_WARP_ID:%.*]] = ashr i32 [[TMP5]], 5
 233 // CHECK-64-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 234 // CHECK-64-NEXT:    store i32 0, ptr [[DOTCNT_ADDR]], align 4
 235 // CHECK-64-NEXT:    br label [[PRECOND:%.*]]
 236 // CHECK-64:       precond:
 237 // CHECK-64-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCNT_ADDR]], align 4
 238 // CHECK-64-NEXT:    [[TMP8:%.*]] = icmp ult i32 [[TMP7]], 2
 239 // CHECK-64-NEXT:    br i1 [[TMP8]], label [[BODY:%.*]], label [[EXIT:%.*]]
 240 // CHECK-64:       body:
 241 // CHECK-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
 242 // CHECK-64-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2:[0-9]+]], i32 [[TMP2]])
 243 // CHECK-64-NEXT:    [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0
 244 // CHECK-64-NEXT:    br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]]
 245 // CHECK-64:       then:
 246 // CHECK-64-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP6]], i64 0, i64 0
 247 // CHECK-64-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[TMP9]], align 8
 248 // CHECK-64-NEXT:    [[TMP11:%.*]] = getelementptr i32, ptr [[TMP10]], i32 [[TMP7]]
 249 // CHECK-64-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]]
 250 // CHECK-64-NEXT:    [[TMP13:%.*]] = load i32, ptr [[TMP11]], align 4
 251 // CHECK-64-NEXT:    store volatile i32 [[TMP13]], ptr addrspace(3) [[TMP12]], align 4
 252 // CHECK-64-NEXT:    br label [[IFCONT:%.*]]
 253 // CHECK-64:       else:
 254 // CHECK-64-NEXT:    br label [[IFCONT]]
 255 // CHECK-64:       ifcont:
 256 // CHECK-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
 257 // CHECK-64-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP2]])
 258 // CHECK-64-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTADDR1]], align 4
 259 // CHECK-64-NEXT:    [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[TMP3]], [[TMP14]]
 260 // CHECK-64-NEXT:    br i1 [[IS_ACTIVE_THREAD]], label [[THEN2:%.*]], label [[ELSE3:%.*]]
 261 // CHECK-64:       then3:
 262 // CHECK-64-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[TMP3]]
 263 // CHECK-64-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP6]], i64 0, i64 0
 264 // CHECK-64-NEXT:    [[TMP17:%.*]] = load ptr, ptr [[TMP16]], align 8
 265 // CHECK-64-NEXT:    [[TMP18:%.*]] = getelementptr i32, ptr [[TMP17]], i32 [[TMP7]]
 266 // CHECK-64-NEXT:    [[TMP19:%.*]] = load volatile i32, ptr addrspace(3) [[TMP15]], align 4
 267 // CHECK-64-NEXT:    store i32 [[TMP19]], ptr [[TMP18]], align 4
 268 // CHECK-64-NEXT:    br label [[IFCONT4:%.*]]
 269 // CHECK-64:       else4:
 270 // CHECK-64-NEXT:    br label [[IFCONT4]]
 271 // CHECK-64:       ifcont5:
 272 // CHECK-64-NEXT:    [[TMP20:%.*]] = add nsw i32 [[TMP7]], 1
 273 // CHECK-64-NEXT:    store i32 [[TMP20]], ptr [[DOTCNT_ADDR]], align 4
 274 // CHECK-64-NEXT:    br label [[PRECOND]]
 275 // CHECK-64:       exit:
 276 // CHECK-64-NEXT:    ret void
 277 //
 278 //
 279 // CHECK-64-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29
 280 // CHECK-64-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], ptr noundef nonnull align 1 dereferenceable(1) [[C:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[D:%.*]]) #[[ATTR0]] {
 281 // CHECK-64-NEXT:  entry:
 282 // CHECK-64-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8
 283 // CHECK-64-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 284 // CHECK-64-NEXT:    [[D_ADDR:%.*]] = alloca ptr, align 8
 285 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [2 x ptr], align 8
 286 // CHECK-64-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 8
 287 // CHECK-64-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 288 // CHECK-64-NEXT:    store ptr [[D]], ptr [[D_ADDR]], align 8
 289 // CHECK-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 290 // CHECK-64-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[D_ADDR]], align 8
 291 // CHECK-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29_kernel_environment, ptr [[DYN_PTR]])
 292 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP2]], -1
 293 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 294 // CHECK-64:       user_code.entry:
 295 // CHECK-64-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
 296 // CHECK-64-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [2 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
 297 // CHECK-64-NEXT:    store ptr [[TMP0]], ptr [[TMP4]], align 8
 298 // CHECK-64-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [2 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
 299 // CHECK-64-NEXT:    store ptr [[TMP1]], ptr [[TMP5]], align 8
 300 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP3]], i32 1, i32 -1, i32 -1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29_omp_outlined, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 2)
 301 // CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 302 // CHECK-64-NEXT:    ret void
 303 // CHECK-64:       worker.exit:
 304 // CHECK-64-NEXT:    ret void
 305 //
 306 //
 307 // CHECK-64-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29_omp_outlined
 308 // CHECK-64-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 1 dereferenceable(1) [[C:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[D:%.*]]) #[[ATTR1]] {
 309 // CHECK-64-NEXT:  entry:
 310 // CHECK-64-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 311 // CHECK-64-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 312 // CHECK-64-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 313 // CHECK-64-NEXT:    [[D_ADDR:%.*]] = alloca ptr, align 8
 314 // CHECK-64-NEXT:    [[C1:%.*]] = alloca i8, align 1
 315 // CHECK-64-NEXT:    [[D2:%.*]] = alloca float, align 4
 316 // CHECK-64-NEXT:    [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x ptr], align 8
 317 // CHECK-64-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 318 // CHECK-64-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 319 // CHECK-64-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 320 // CHECK-64-NEXT:    store ptr [[D]], ptr [[D_ADDR]], align 8
 321 // CHECK-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 322 // CHECK-64-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[D_ADDR]], align 8
 323 // CHECK-64-NEXT:    store i8 0, ptr [[C1]], align 1
 324 // CHECK-64-NEXT:    store float 1.000000e+00, ptr [[D2]], align 4
 325 // CHECK-64-NEXT:    [[TMP2:%.*]] = load i8, ptr [[C1]], align 1
 326 // CHECK-64-NEXT:    [[CONV:%.*]] = sext i8 [[TMP2]] to i32
 327 // CHECK-64-NEXT:    [[XOR:%.*]] = xor i32 [[CONV]], 2
 328 // CHECK-64-NEXT:    [[CONV3:%.*]] = trunc i32 [[XOR]] to i8
 329 // CHECK-64-NEXT:    store i8 [[CONV3]], ptr [[C1]], align 1
 330 // CHECK-64-NEXT:    [[TMP3:%.*]] = load float, ptr [[D2]], align 4
 331 // CHECK-64-NEXT:    [[MUL:%.*]] = fmul float [[TMP3]], 3.300000e+01
 332 // CHECK-64-NEXT:    store float [[MUL]], ptr [[D2]], align 4
 333 // CHECK-64-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0
 334 // CHECK-64-NEXT:    store ptr [[C1]], ptr [[TMP4]], align 8
 335 // CHECK-64-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 1
 336 // CHECK-64-NEXT:    store ptr [[D2]], ptr [[TMP5]], align 8
 337 // CHECK-64-NEXT:    [[TMP6:%.*]] = call i32 @__kmpc_nvptx_parallel_reduce_nowait_v2(ptr @[[GLOB1]], i64 8, ptr [[DOTOMP_REDUCTION_RED_LIST]], ptr @_omp_reduction_shuffle_and_reduce_func1, ptr @_omp_reduction_inter_warp_copy_func2)
 338 // CHECK-64-NEXT:    [[TMP7:%.*]] = icmp eq i32 [[TMP6]], 1
 339 // CHECK-64-NEXT:    br i1 [[TMP7]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]]
 340 // CHECK-64:       .omp.reduction.then:
 341 // CHECK-64-NEXT:    [[TMP8:%.*]] = load i8, ptr [[TMP0]], align 1
 342 // CHECK-64-NEXT:    [[CONV4:%.*]] = sext i8 [[TMP8]] to i32
 343 // CHECK-64-NEXT:    [[TMP9:%.*]] = load i8, ptr [[C1]], align 1
 344 // CHECK-64-NEXT:    [[CONV5:%.*]] = sext i8 [[TMP9]] to i32
 345 // CHECK-64-NEXT:    [[XOR6:%.*]] = xor i32 [[CONV4]], [[CONV5]]
 346 // CHECK-64-NEXT:    [[CONV7:%.*]] = trunc i32 [[XOR6]] to i8
 347 // CHECK-64-NEXT:    store i8 [[CONV7]], ptr [[TMP0]], align 1
 348 // CHECK-64-NEXT:    [[TMP10:%.*]] = load float, ptr [[TMP1]], align 4
 349 // CHECK-64-NEXT:    [[TMP11:%.*]] = load float, ptr [[D2]], align 4
 350 // CHECK-64-NEXT:    [[MUL8:%.*]] = fmul float [[TMP10]], [[TMP11]]
 351 // CHECK-64-NEXT:    store float [[MUL8]], ptr [[TMP1]], align 4
 352 // CHECK-64-NEXT:    br label [[DOTOMP_REDUCTION_DONE]]
 353 // CHECK-64:       .omp.reduction.done:
 354 // CHECK-64-NEXT:    ret void
 355 //
 356 //
 357 // CHECK-64-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func1
 358 // CHECK-64-SAME: (ptr noundef [[TMP0:%.*]], i16 noundef signext [[TMP1:%.*]], i16 noundef signext [[TMP2:%.*]], i16 noundef signext [[TMP3:%.*]]) #[[ATTR2]] {
 359 // CHECK-64-NEXT:  entry:
 360 // CHECK-64-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 361 // CHECK-64-NEXT:    [[DOTADDR1:%.*]] = alloca i16, align 2
 362 // CHECK-64-NEXT:    [[DOTADDR2:%.*]] = alloca i16, align 2
 363 // CHECK-64-NEXT:    [[DOTADDR3:%.*]] = alloca i16, align 2
 364 // CHECK-64-NEXT:    [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [2 x ptr], align 8
 365 // CHECK-64-NEXT:    [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca i8, align 1
 366 // CHECK-64-NEXT:    [[DOTOMP_REDUCTION_ELEMENT4:%.*]] = alloca float, align 4
 367 // CHECK-64-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 368 // CHECK-64-NEXT:    store i16 [[TMP1]], ptr [[DOTADDR1]], align 2
 369 // CHECK-64-NEXT:    store i16 [[TMP2]], ptr [[DOTADDR2]], align 2
 370 // CHECK-64-NEXT:    store i16 [[TMP3]], ptr [[DOTADDR3]], align 2
 371 // CHECK-64-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 372 // CHECK-64-NEXT:    [[TMP5:%.*]] = load i16, ptr [[DOTADDR1]], align 2
 373 // CHECK-64-NEXT:    [[TMP6:%.*]] = load i16, ptr [[DOTADDR2]], align 2
 374 // CHECK-64-NEXT:    [[TMP7:%.*]] = load i16, ptr [[DOTADDR3]], align 2
 375 // CHECK-64-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP4]], i64 0, i64 0
 376 // CHECK-64-NEXT:    [[TMP9:%.*]] = load ptr, ptr [[TMP8]], align 8
 377 // CHECK-64-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0
 378 // CHECK-64-NEXT:    [[TMP11:%.*]] = getelementptr i8, ptr [[TMP9]], i64 1
 379 // CHECK-64-NEXT:    [[TMP12:%.*]] = load i8, ptr [[TMP9]], align 1
 380 // CHECK-64-NEXT:    [[TMP13:%.*]] = sext i8 [[TMP12]] to i32
 381 // CHECK-64-NEXT:    [[TMP14:%.*]] = call i32 @__kmpc_get_warp_size()
 382 // CHECK-64-NEXT:    [[TMP15:%.*]] = trunc i32 [[TMP14]] to i16
 383 // CHECK-64-NEXT:    [[TMP16:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP13]], i16 [[TMP6]], i16 [[TMP15]])
 384 // CHECK-64-NEXT:    [[TMP17:%.*]] = trunc i32 [[TMP16]] to i8
 385 // CHECK-64-NEXT:    store i8 [[TMP17]], ptr [[DOTOMP_REDUCTION_ELEMENT]], align 1
 386 // CHECK-64-NEXT:    [[TMP18:%.*]] = getelementptr i8, ptr [[TMP9]], i64 1
 387 // CHECK-64-NEXT:    [[TMP19:%.*]] = getelementptr i8, ptr [[DOTOMP_REDUCTION_ELEMENT]], i64 1
 388 // CHECK-64-NEXT:    store ptr [[DOTOMP_REDUCTION_ELEMENT]], ptr [[TMP10]], align 8
 389 // CHECK-64-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP4]], i64 0, i64 1
 390 // CHECK-64-NEXT:    [[TMP21:%.*]] = load ptr, ptr [[TMP20]], align 8
 391 // CHECK-64-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 1
 392 // CHECK-64-NEXT:    [[TMP23:%.*]] = getelementptr float, ptr [[TMP21]], i64 1
 393 // CHECK-64-NEXT:    [[TMP24:%.*]] = load i32, ptr [[TMP21]], align 4
 394 // CHECK-64-NEXT:    [[TMP25:%.*]] = call i32 @__kmpc_get_warp_size()
 395 // CHECK-64-NEXT:    [[TMP26:%.*]] = trunc i32 [[TMP25]] to i16
 396 // CHECK-64-NEXT:    [[TMP27:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP24]], i16 [[TMP6]], i16 [[TMP26]])
 397 // CHECK-64-NEXT:    store i32 [[TMP27]], ptr [[DOTOMP_REDUCTION_ELEMENT4]], align 4
 398 // CHECK-64-NEXT:    [[TMP28:%.*]] = getelementptr i32, ptr [[TMP21]], i64 1
 399 // CHECK-64-NEXT:    [[TMP29:%.*]] = getelementptr i32, ptr [[DOTOMP_REDUCTION_ELEMENT4]], i64 1
 400 // CHECK-64-NEXT:    store ptr [[DOTOMP_REDUCTION_ELEMENT4]], ptr [[TMP22]], align 8
 401 // CHECK-64-NEXT:    [[TMP30:%.*]] = icmp eq i16 [[TMP7]], 0
 402 // CHECK-64-NEXT:    [[TMP31:%.*]] = icmp eq i16 [[TMP7]], 1
 403 // CHECK-64-NEXT:    [[TMP32:%.*]] = icmp ult i16 [[TMP5]], [[TMP6]]
 404 // CHECK-64-NEXT:    [[TMP33:%.*]] = and i1 [[TMP31]], [[TMP32]]
 405 // CHECK-64-NEXT:    [[TMP34:%.*]] = icmp eq i16 [[TMP7]], 2
 406 // CHECK-64-NEXT:    [[TMP35:%.*]] = and i16 [[TMP5]], 1
 407 // CHECK-64-NEXT:    [[TMP36:%.*]] = icmp eq i16 [[TMP35]], 0
 408 // CHECK-64-NEXT:    [[TMP37:%.*]] = and i1 [[TMP34]], [[TMP36]]
 409 // CHECK-64-NEXT:    [[TMP38:%.*]] = icmp sgt i16 [[TMP6]], 0
 410 // CHECK-64-NEXT:    [[TMP39:%.*]] = and i1 [[TMP37]], [[TMP38]]
 411 // CHECK-64-NEXT:    [[TMP40:%.*]] = or i1 [[TMP30]], [[TMP33]]
 412 // CHECK-64-NEXT:    [[TMP41:%.*]] = or i1 [[TMP40]], [[TMP39]]
 413 // CHECK-64-NEXT:    br i1 [[TMP41]], label [[THEN:%.*]], label [[ELSE:%.*]]
 414 // CHECK-64:       then:
 415 // CHECK-64-NEXT:    call void @"{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29_omp_outlined_omp$reduction$reduction_func"(ptr [[TMP4]], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]]) #[[ATTR3]]
 416 // CHECK-64-NEXT:    br label [[IFCONT:%.*]]
 417 // CHECK-64:       else:
 418 // CHECK-64-NEXT:    br label [[IFCONT]]
 419 // CHECK-64:       ifcont:
 420 // CHECK-64-NEXT:    [[TMP42:%.*]] = icmp eq i16 [[TMP7]], 1
 421 // CHECK-64-NEXT:    [[TMP43:%.*]] = icmp uge i16 [[TMP5]], [[TMP6]]
 422 // CHECK-64-NEXT:    [[TMP44:%.*]] = and i1 [[TMP42]], [[TMP43]]
 423 // CHECK-64-NEXT:    br i1 [[TMP44]], label [[THEN5:%.*]], label [[ELSE6:%.*]]
 424 // CHECK-64:       then5:
 425 // CHECK-64-NEXT:    [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0
 426 // CHECK-64-NEXT:    [[TMP46:%.*]] = load ptr, ptr [[TMP45]], align 8
 427 // CHECK-64-NEXT:    [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP4]], i64 0, i64 0
 428 // CHECK-64-NEXT:    [[TMP48:%.*]] = load ptr, ptr [[TMP47]], align 8
 429 // CHECK-64-NEXT:    [[TMP49:%.*]] = load i8, ptr [[TMP46]], align 1
 430 // CHECK-64-NEXT:    store i8 [[TMP49]], ptr [[TMP48]], align 1
 431 // CHECK-64-NEXT:    [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 1
 432 // CHECK-64-NEXT:    [[TMP51:%.*]] = load ptr, ptr [[TMP50]], align 8
 433 // CHECK-64-NEXT:    [[TMP52:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP4]], i64 0, i64 1
 434 // CHECK-64-NEXT:    [[TMP53:%.*]] = load ptr, ptr [[TMP52]], align 8
 435 // CHECK-64-NEXT:    [[TMP54:%.*]] = load float, ptr [[TMP51]], align 4
 436 // CHECK-64-NEXT:    store float [[TMP54]], ptr [[TMP53]], align 4
 437 // CHECK-64-NEXT:    br label [[IFCONT7:%.*]]
 438 // CHECK-64:       else6:
 439 // CHECK-64-NEXT:    br label [[IFCONT7]]
 440 // CHECK-64:       ifcont7:
 441 // CHECK-64-NEXT:    ret void
 442 //
 443 //
 444 // CHECK-64-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func2
 445 // CHECK-64-SAME: (ptr noundef [[TMP0:%.*]], i32 noundef [[TMP1:%.*]]) #[[ATTR2]] {
 446 // CHECK-64-NEXT:  entry:
 447 // CHECK-64-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 448 // CHECK-64-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
 449 // CHECK-64-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 450 // CHECK-64-NEXT:    store i32 [[TMP1]], ptr [[DOTADDR1]], align 4
 451 // CHECK-64-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
 452 // CHECK-64-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
 453 // CHECK-64-NEXT:    [[NVPTX_LANE_ID:%.*]] = and i32 [[TMP4]], 31
 454 // CHECK-64-NEXT:    [[TMP5:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
 455 // CHECK-64-NEXT:    [[NVPTX_WARP_ID:%.*]] = ashr i32 [[TMP5]], 5
 456 // CHECK-64-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 457 // CHECK-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
 458 // CHECK-64-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP2]])
 459 // CHECK-64-NEXT:    [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0
 460 // CHECK-64-NEXT:    br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]]
 461 // CHECK-64:       then:
 462 // CHECK-64-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP6]], i64 0, i64 0
 463 // CHECK-64-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[TMP7]], align 8
 464 // CHECK-64-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]]
 465 // CHECK-64-NEXT:    [[TMP10:%.*]] = load i8, ptr [[TMP8]], align 1
 466 // CHECK-64-NEXT:    store volatile i8 [[TMP10]], ptr addrspace(3) [[TMP9]], align 1
 467 // CHECK-64-NEXT:    br label [[IFCONT:%.*]]
 468 // CHECK-64:       else:
 469 // CHECK-64-NEXT:    br label [[IFCONT]]
 470 // CHECK-64:       ifcont:
 471 // CHECK-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
 472 // CHECK-64-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP2]])
 473 // CHECK-64-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTADDR1]], align 4
 474 // CHECK-64-NEXT:    [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[TMP3]], [[TMP11]]
 475 // CHECK-64-NEXT:    br i1 [[IS_ACTIVE_THREAD]], label [[THEN2:%.*]], label [[ELSE3:%.*]]
 476 // CHECK-64:       then3:
 477 // CHECK-64-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[TMP3]]
 478 // CHECK-64-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP6]], i64 0, i64 0
 479 // CHECK-64-NEXT:    [[TMP14:%.*]] = load ptr, ptr [[TMP13]], align 8
 480 // CHECK-64-NEXT:    [[TMP15:%.*]] = load volatile i8, ptr addrspace(3) [[TMP12]], align 1
 481 // CHECK-64-NEXT:    store i8 [[TMP15]], ptr [[TMP14]], align 1
 482 // CHECK-64-NEXT:    br label [[IFCONT4:%.*]]
 483 // CHECK-64:       else4:
 484 // CHECK-64-NEXT:    br label [[IFCONT4]]
 485 // CHECK-64:       ifcont5:
 486 // CHECK-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
 487 // CHECK-64-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP2]])
 488 // CHECK-64-NEXT:    [[WARP_MASTER5:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0
 489 // CHECK-64-NEXT:    br i1 [[WARP_MASTER5]], label [[THEN6:%.*]], label [[ELSE7:%.*]]
 490 // CHECK-64:       then8:
 491 // CHECK-64-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP6]], i64 0, i64 1
 492 // CHECK-64-NEXT:    [[TMP17:%.*]] = load ptr, ptr [[TMP16]], align 8
 493 // CHECK-64-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]]
 494 // CHECK-64-NEXT:    [[TMP19:%.*]] = load i32, ptr [[TMP17]], align 4
 495 // CHECK-64-NEXT:    store volatile i32 [[TMP19]], ptr addrspace(3) [[TMP18]], align 4
 496 // CHECK-64-NEXT:    br label [[IFCONT8:%.*]]
 497 // CHECK-64:       else9:
 498 // CHECK-64-NEXT:    br label [[IFCONT8]]
 499 // CHECK-64:       ifcont10:
 500 // CHECK-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
 501 // CHECK-64-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP2]])
 502 // CHECK-64-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTADDR1]], align 4
 503 // CHECK-64-NEXT:    [[IS_ACTIVE_THREAD9:%.*]] = icmp ult i32 [[TMP3]], [[TMP20]]
 504 // CHECK-64-NEXT:    br i1 [[IS_ACTIVE_THREAD9]], label [[THEN10:%.*]], label [[ELSE11:%.*]]
 505 // CHECK-64:       then13:
 506 // CHECK-64-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[TMP3]]
 507 // CHECK-64-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP6]], i64 0, i64 1
 508 // CHECK-64-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP22]], align 8
 509 // CHECK-64-NEXT:    [[TMP24:%.*]] = load volatile i32, ptr addrspace(3) [[TMP21]], align 4
 510 // CHECK-64-NEXT:    store i32 [[TMP24]], ptr [[TMP23]], align 4
 511 // CHECK-64-NEXT:    br label [[IFCONT12:%.*]]
 512 // CHECK-64:       else14:
 513 // CHECK-64-NEXT:    br label [[IFCONT12]]
 514 // CHECK-64:       ifcont15:
 515 // CHECK-64-NEXT:    ret void
 516 //
 517 //
 518 // CHECK-64-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l35
 519 // CHECK-64-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 2 dereferenceable(2) [[B:%.*]]) #[[ATTR0]] {
 520 // CHECK-64-NEXT:  entry:
 521 // CHECK-64-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8
 522 // CHECK-64-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 523 // CHECK-64-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 524 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [2 x ptr], align 8
 525 // CHECK-64-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 8
 526 // CHECK-64-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 527 // CHECK-64-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 528 // CHECK-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 529 // CHECK-64-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 530 // CHECK-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l35_kernel_environment, ptr [[DYN_PTR]])
 531 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP2]], -1
 532 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 533 // CHECK-64:       user_code.entry:
 534 // CHECK-64-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
 535 // CHECK-64-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [2 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
 536 // CHECK-64-NEXT:    store ptr [[TMP0]], ptr [[TMP4]], align 8
 537 // CHECK-64-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [2 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
 538 // CHECK-64-NEXT:    store ptr [[TMP1]], ptr [[TMP5]], align 8
 539 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP3]], i32 1, i32 -1, i32 -1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l35_omp_outlined, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 2)
 540 // CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 541 // CHECK-64-NEXT:    ret void
 542 // CHECK-64:       worker.exit:
 543 // CHECK-64-NEXT:    ret void
 544 //
 545 //
 546 // CHECK-64-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l35_omp_outlined
 547 // CHECK-64-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 2 dereferenceable(2) [[B:%.*]]) #[[ATTR1]] {
 548 // CHECK-64-NEXT:  entry:
 549 // CHECK-64-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 550 // CHECK-64-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 551 // CHECK-64-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 552 // CHECK-64-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 553 // CHECK-64-NEXT:    [[A1:%.*]] = alloca i32, align 4
 554 // CHECK-64-NEXT:    [[B2:%.*]] = alloca i16, align 2
 555 // CHECK-64-NEXT:    [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x ptr], align 8
 556 // CHECK-64-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 557 // CHECK-64-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 558 // CHECK-64-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 559 // CHECK-64-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 560 // CHECK-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 561 // CHECK-64-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 562 // CHECK-64-NEXT:    store i32 0, ptr [[A1]], align 4
 563 // CHECK-64-NEXT:    store i16 -32768, ptr [[B2]], align 2
 564 // CHECK-64-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A1]], align 4
 565 // CHECK-64-NEXT:    [[OR:%.*]] = or i32 [[TMP2]], 1
 566 // CHECK-64-NEXT:    store i32 [[OR]], ptr [[A1]], align 4
 567 // CHECK-64-NEXT:    [[TMP3:%.*]] = load i16, ptr [[B2]], align 2
 568 // CHECK-64-NEXT:    [[CONV:%.*]] = sext i16 [[TMP3]] to i32
 569 // CHECK-64-NEXT:    [[CMP:%.*]] = icmp sgt i32 99, [[CONV]]
 570 // CHECK-64-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 571 // CHECK-64:       cond.true:
 572 // CHECK-64-NEXT:    br label [[COND_END:%.*]]
 573 // CHECK-64:       cond.false:
 574 // CHECK-64-NEXT:    [[TMP4:%.*]] = load i16, ptr [[B2]], align 2
 575 // CHECK-64-NEXT:    [[CONV3:%.*]] = sext i16 [[TMP4]] to i32
 576 // CHECK-64-NEXT:    br label [[COND_END]]
 577 // CHECK-64:       cond.end:
 578 // CHECK-64-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[CONV3]], [[COND_FALSE]] ]
 579 // CHECK-64-NEXT:    [[CONV4:%.*]] = trunc i32 [[COND]] to i16
 580 // CHECK-64-NEXT:    store i16 [[CONV4]], ptr [[B2]], align 2
 581 // CHECK-64-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0
 582 // CHECK-64-NEXT:    store ptr [[A1]], ptr [[TMP5]], align 8
 583 // CHECK-64-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 1
 584 // CHECK-64-NEXT:    store ptr [[B2]], ptr [[TMP6]], align 8
 585 // CHECK-64-NEXT:    [[TMP7:%.*]] = call i32 @__kmpc_nvptx_parallel_reduce_nowait_v2(ptr @[[GLOB1]], i64 8, ptr [[DOTOMP_REDUCTION_RED_LIST]], ptr @_omp_reduction_shuffle_and_reduce_func3, ptr @_omp_reduction_inter_warp_copy_func4)
 586 // CHECK-64-NEXT:    [[TMP8:%.*]] = icmp eq i32 [[TMP7]], 1
 587 // CHECK-64-NEXT:    br i1 [[TMP8]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]]
 588 // CHECK-64:       .omp.reduction.then:
 589 // CHECK-64-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP0]], align 4
 590 // CHECK-64-NEXT:    [[TMP10:%.*]] = load i32, ptr [[A1]], align 4
 591 // CHECK-64-NEXT:    [[OR5:%.*]] = or i32 [[TMP9]], [[TMP10]]
 592 // CHECK-64-NEXT:    store i32 [[OR5]], ptr [[TMP0]], align 4
 593 // CHECK-64-NEXT:    [[TMP11:%.*]] = load i16, ptr [[TMP1]], align 2
 594 // CHECK-64-NEXT:    [[CONV6:%.*]] = sext i16 [[TMP11]] to i32
 595 // CHECK-64-NEXT:    [[TMP12:%.*]] = load i16, ptr [[B2]], align 2
 596 // CHECK-64-NEXT:    [[CONV7:%.*]] = sext i16 [[TMP12]] to i32
 597 // CHECK-64-NEXT:    [[CMP8:%.*]] = icmp sgt i32 [[CONV6]], [[CONV7]]
 598 // CHECK-64-NEXT:    br i1 [[CMP8]], label [[COND_TRUE9:%.*]], label [[COND_FALSE10:%.*]]
 599 // CHECK-64:       cond.true9:
 600 // CHECK-64-NEXT:    [[TMP13:%.*]] = load i16, ptr [[TMP1]], align 2
 601 // CHECK-64-NEXT:    br label [[COND_END11:%.*]]
 602 // CHECK-64:       cond.false10:
 603 // CHECK-64-NEXT:    [[TMP14:%.*]] = load i16, ptr [[B2]], align 2
 604 // CHECK-64-NEXT:    br label [[COND_END11]]
 605 // CHECK-64:       cond.end11:
 606 // CHECK-64-NEXT:    [[COND12:%.*]] = phi i16 [ [[TMP13]], [[COND_TRUE9]] ], [ [[TMP14]], [[COND_FALSE10]] ]
 607 // CHECK-64-NEXT:    store i16 [[COND12]], ptr [[TMP1]], align 2
 608 // CHECK-64-NEXT:    br label [[DOTOMP_REDUCTION_DONE]]
 609 // CHECK-64:       .omp.reduction.done:
 610 // CHECK-64-NEXT:    ret void
 611 //
 612 //
 613 // CHECK-64-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func3
 614 // CHECK-64-SAME: (ptr noundef [[TMP0:%.*]], i16 noundef signext [[TMP1:%.*]], i16 noundef signext [[TMP2:%.*]], i16 noundef signext [[TMP3:%.*]]) #[[ATTR2]] {
 615 // CHECK-64-NEXT:  entry:
 616 // CHECK-64-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 617 // CHECK-64-NEXT:    [[DOTADDR1:%.*]] = alloca i16, align 2
 618 // CHECK-64-NEXT:    [[DOTADDR2:%.*]] = alloca i16, align 2
 619 // CHECK-64-NEXT:    [[DOTADDR3:%.*]] = alloca i16, align 2
 620 // CHECK-64-NEXT:    [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [2 x ptr], align 8
 621 // CHECK-64-NEXT:    [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca i32, align 4
 622 // CHECK-64-NEXT:    [[DOTOMP_REDUCTION_ELEMENT4:%.*]] = alloca i16, align 2
 623 // CHECK-64-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 624 // CHECK-64-NEXT:    store i16 [[TMP1]], ptr [[DOTADDR1]], align 2
 625 // CHECK-64-NEXT:    store i16 [[TMP2]], ptr [[DOTADDR2]], align 2
 626 // CHECK-64-NEXT:    store i16 [[TMP3]], ptr [[DOTADDR3]], align 2
 627 // CHECK-64-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 628 // CHECK-64-NEXT:    [[TMP5:%.*]] = load i16, ptr [[DOTADDR1]], align 2
 629 // CHECK-64-NEXT:    [[TMP6:%.*]] = load i16, ptr [[DOTADDR2]], align 2
 630 // CHECK-64-NEXT:    [[TMP7:%.*]] = load i16, ptr [[DOTADDR3]], align 2
 631 // CHECK-64-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP4]], i64 0, i64 0
 632 // CHECK-64-NEXT:    [[TMP9:%.*]] = load ptr, ptr [[TMP8]], align 8
 633 // CHECK-64-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0
 634 // CHECK-64-NEXT:    [[TMP11:%.*]] = getelementptr i32, ptr [[TMP9]], i64 1
 635 // CHECK-64-NEXT:    [[TMP12:%.*]] = load i32, ptr [[TMP9]], align 4
 636 // CHECK-64-NEXT:    [[TMP13:%.*]] = call i32 @__kmpc_get_warp_size()
 637 // CHECK-64-NEXT:    [[TMP14:%.*]] = trunc i32 [[TMP13]] to i16
 638 // CHECK-64-NEXT:    [[TMP15:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP12]], i16 [[TMP6]], i16 [[TMP14]])
 639 // CHECK-64-NEXT:    store i32 [[TMP15]], ptr [[DOTOMP_REDUCTION_ELEMENT]], align 4
 640 // CHECK-64-NEXT:    [[TMP16:%.*]] = getelementptr i32, ptr [[TMP9]], i64 1
 641 // CHECK-64-NEXT:    [[TMP17:%.*]] = getelementptr i32, ptr [[DOTOMP_REDUCTION_ELEMENT]], i64 1
 642 // CHECK-64-NEXT:    store ptr [[DOTOMP_REDUCTION_ELEMENT]], ptr [[TMP10]], align 8
 643 // CHECK-64-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP4]], i64 0, i64 1
 644 // CHECK-64-NEXT:    [[TMP19:%.*]] = load ptr, ptr [[TMP18]], align 8
 645 // CHECK-64-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 1
 646 // CHECK-64-NEXT:    [[TMP21:%.*]] = getelementptr i16, ptr [[TMP19]], i64 1
 647 // CHECK-64-NEXT:    [[TMP22:%.*]] = load i16, ptr [[TMP19]], align 2
 648 // CHECK-64-NEXT:    [[TMP23:%.*]] = sext i16 [[TMP22]] to i32
 649 // CHECK-64-NEXT:    [[TMP24:%.*]] = call i32 @__kmpc_get_warp_size()
 650 // CHECK-64-NEXT:    [[TMP25:%.*]] = trunc i32 [[TMP24]] to i16
 651 // CHECK-64-NEXT:    [[TMP26:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP23]], i16 [[TMP6]], i16 [[TMP25]])
 652 // CHECK-64-NEXT:    [[TMP27:%.*]] = trunc i32 [[TMP26]] to i16
 653 // CHECK-64-NEXT:    store i16 [[TMP27]], ptr [[DOTOMP_REDUCTION_ELEMENT4]], align 2
 654 // CHECK-64-NEXT:    [[TMP28:%.*]] = getelementptr i16, ptr [[TMP19]], i64 1
 655 // CHECK-64-NEXT:    [[TMP29:%.*]] = getelementptr i16, ptr [[DOTOMP_REDUCTION_ELEMENT4]], i64 1
 656 // CHECK-64-NEXT:    store ptr [[DOTOMP_REDUCTION_ELEMENT4]], ptr [[TMP20]], align 8
 657 // CHECK-64-NEXT:    [[TMP30:%.*]] = icmp eq i16 [[TMP7]], 0
 658 // CHECK-64-NEXT:    [[TMP31:%.*]] = icmp eq i16 [[TMP7]], 1
 659 // CHECK-64-NEXT:    [[TMP32:%.*]] = icmp ult i16 [[TMP5]], [[TMP6]]
 660 // CHECK-64-NEXT:    [[TMP33:%.*]] = and i1 [[TMP31]], [[TMP32]]
 661 // CHECK-64-NEXT:    [[TMP34:%.*]] = icmp eq i16 [[TMP7]], 2
 662 // CHECK-64-NEXT:    [[TMP35:%.*]] = and i16 [[TMP5]], 1
 663 // CHECK-64-NEXT:    [[TMP36:%.*]] = icmp eq i16 [[TMP35]], 0
 664 // CHECK-64-NEXT:    [[TMP37:%.*]] = and i1 [[TMP34]], [[TMP36]]
 665 // CHECK-64-NEXT:    [[TMP38:%.*]] = icmp sgt i16 [[TMP6]], 0
 666 // CHECK-64-NEXT:    [[TMP39:%.*]] = and i1 [[TMP37]], [[TMP38]]
 667 // CHECK-64-NEXT:    [[TMP40:%.*]] = or i1 [[TMP30]], [[TMP33]]
 668 // CHECK-64-NEXT:    [[TMP41:%.*]] = or i1 [[TMP40]], [[TMP39]]
 669 // CHECK-64-NEXT:    br i1 [[TMP41]], label [[THEN:%.*]], label [[ELSE:%.*]]
 670 // CHECK-64:       then:
 671 // CHECK-64-NEXT:    call void @"{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l35_omp_outlined_omp$reduction$reduction_func"(ptr [[TMP4]], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]]) #[[ATTR3]]
 672 // CHECK-64-NEXT:    br label [[IFCONT:%.*]]
 673 // CHECK-64:       else:
 674 // CHECK-64-NEXT:    br label [[IFCONT]]
 675 // CHECK-64:       ifcont:
 676 // CHECK-64-NEXT:    [[TMP42:%.*]] = icmp eq i16 [[TMP7]], 1
 677 // CHECK-64-NEXT:    [[TMP43:%.*]] = icmp uge i16 [[TMP5]], [[TMP6]]
 678 // CHECK-64-NEXT:    [[TMP44:%.*]] = and i1 [[TMP42]], [[TMP43]]
 679 // CHECK-64-NEXT:    br i1 [[TMP44]], label [[THEN5:%.*]], label [[ELSE6:%.*]]
 680 // CHECK-64:       then5:
 681 // CHECK-64-NEXT:    [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0
 682 // CHECK-64-NEXT:    [[TMP46:%.*]] = load ptr, ptr [[TMP45]], align 8
 683 // CHECK-64-NEXT:    [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP4]], i64 0, i64 0
 684 // CHECK-64-NEXT:    [[TMP48:%.*]] = load ptr, ptr [[TMP47]], align 8
 685 // CHECK-64-NEXT:    [[TMP49:%.*]] = load i32, ptr [[TMP46]], align 4
 686 // CHECK-64-NEXT:    store i32 [[TMP49]], ptr [[TMP48]], align 4
 687 // CHECK-64-NEXT:    [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 1
 688 // CHECK-64-NEXT:    [[TMP51:%.*]] = load ptr, ptr [[TMP50]], align 8
 689 // CHECK-64-NEXT:    [[TMP52:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP4]], i64 0, i64 1
 690 // CHECK-64-NEXT:    [[TMP53:%.*]] = load ptr, ptr [[TMP52]], align 8
 691 // CHECK-64-NEXT:    [[TMP54:%.*]] = load i16, ptr [[TMP51]], align 2
 692 // CHECK-64-NEXT:    store i16 [[TMP54]], ptr [[TMP53]], align 2
 693 // CHECK-64-NEXT:    br label [[IFCONT7:%.*]]
 694 // CHECK-64:       else6:
 695 // CHECK-64-NEXT:    br label [[IFCONT7]]
 696 // CHECK-64:       ifcont7:
 697 // CHECK-64-NEXT:    ret void
 698 //
 699 //
 700 // CHECK-64-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func4
 701 // CHECK-64-SAME: (ptr noundef [[TMP0:%.*]], i32 noundef [[TMP1:%.*]]) #[[ATTR2]] {
 702 // CHECK-64-NEXT:  entry:
 703 // CHECK-64-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 704 // CHECK-64-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
 705 // CHECK-64-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 706 // CHECK-64-NEXT:    store i32 [[TMP1]], ptr [[DOTADDR1]], align 4
 707 // CHECK-64-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
 708 // CHECK-64-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
 709 // CHECK-64-NEXT:    [[NVPTX_LANE_ID:%.*]] = and i32 [[TMP4]], 31
 710 // CHECK-64-NEXT:    [[TMP5:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
 711 // CHECK-64-NEXT:    [[NVPTX_WARP_ID:%.*]] = ashr i32 [[TMP5]], 5
 712 // CHECK-64-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 713 // CHECK-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
 714 // CHECK-64-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP2]])
 715 // CHECK-64-NEXT:    [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0
 716 // CHECK-64-NEXT:    br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]]
 717 // CHECK-64:       then:
 718 // CHECK-64-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP6]], i64 0, i64 0
 719 // CHECK-64-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[TMP7]], align 8
 720 // CHECK-64-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]]
 721 // CHECK-64-NEXT:    [[TMP10:%.*]] = load i32, ptr [[TMP8]], align 4
 722 // CHECK-64-NEXT:    store volatile i32 [[TMP10]], ptr addrspace(3) [[TMP9]], align 4
 723 // CHECK-64-NEXT:    br label [[IFCONT:%.*]]
 724 // CHECK-64:       else:
 725 // CHECK-64-NEXT:    br label [[IFCONT]]
 726 // CHECK-64:       ifcont:
 727 // CHECK-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
 728 // CHECK-64-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP2]])
 729 // CHECK-64-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTADDR1]], align 4
 730 // CHECK-64-NEXT:    [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[TMP3]], [[TMP11]]
 731 // CHECK-64-NEXT:    br i1 [[IS_ACTIVE_THREAD]], label [[THEN2:%.*]], label [[ELSE3:%.*]]
 732 // CHECK-64:       then3:
 733 // CHECK-64-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[TMP3]]
 734 // CHECK-64-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP6]], i64 0, i64 0
 735 // CHECK-64-NEXT:    [[TMP14:%.*]] = load ptr, ptr [[TMP13]], align 8
 736 // CHECK-64-NEXT:    [[TMP15:%.*]] = load volatile i32, ptr addrspace(3) [[TMP12]], align 4
 737 // CHECK-64-NEXT:    store i32 [[TMP15]], ptr [[TMP14]], align 4
 738 // CHECK-64-NEXT:    br label [[IFCONT4:%.*]]
 739 // CHECK-64:       else4:
 740 // CHECK-64-NEXT:    br label [[IFCONT4]]
 741 // CHECK-64:       ifcont5:
 742 // CHECK-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
 743 // CHECK-64-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP2]])
 744 // CHECK-64-NEXT:    [[WARP_MASTER5:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0
 745 // CHECK-64-NEXT:    br i1 [[WARP_MASTER5]], label [[THEN6:%.*]], label [[ELSE7:%.*]]
 746 // CHECK-64:       then8:
 747 // CHECK-64-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP6]], i64 0, i64 1
 748 // CHECK-64-NEXT:    [[TMP17:%.*]] = load ptr, ptr [[TMP16]], align 8
 749 // CHECK-64-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]]
 750 // CHECK-64-NEXT:    [[TMP19:%.*]] = load i16, ptr [[TMP17]], align 2
 751 // CHECK-64-NEXT:    store volatile i16 [[TMP19]], ptr addrspace(3) [[TMP18]], align 2
 752 // CHECK-64-NEXT:    br label [[IFCONT8:%.*]]
 753 // CHECK-64:       else9:
 754 // CHECK-64-NEXT:    br label [[IFCONT8]]
 755 // CHECK-64:       ifcont10:
 756 // CHECK-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
 757 // CHECK-64-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP2]])
 758 // CHECK-64-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTADDR1]], align 4
 759 // CHECK-64-NEXT:    [[IS_ACTIVE_THREAD9:%.*]] = icmp ult i32 [[TMP3]], [[TMP20]]
 760 // CHECK-64-NEXT:    br i1 [[IS_ACTIVE_THREAD9]], label [[THEN10:%.*]], label [[ELSE11:%.*]]
 761 // CHECK-64:       then13:
 762 // CHECK-64-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[TMP3]]
 763 // CHECK-64-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP6]], i64 0, i64 1
 764 // CHECK-64-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP22]], align 8
 765 // CHECK-64-NEXT:    [[TMP24:%.*]] = load volatile i16, ptr addrspace(3) [[TMP21]], align 2
 766 // CHECK-64-NEXT:    store i16 [[TMP24]], ptr [[TMP23]], align 2
 767 // CHECK-64-NEXT:    br label [[IFCONT12:%.*]]
 768 // CHECK-64:       else14:
 769 // CHECK-64-NEXT:    br label [[IFCONT12]]
 770 // CHECK-64:       ifcont15:
 771 // CHECK-64-NEXT:    ret void
 772 //
 773 //
 774 // CHECK-32-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l24
 775 // CHECK-32-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[E:%.*]]) #[[ATTR0:[0-9]+]] {
 776 // CHECK-32-NEXT:  entry:
 777 // CHECK-32-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 4
 778 // CHECK-32-NEXT:    [[E_ADDR:%.*]] = alloca ptr, align 4
 779 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x ptr], align 4
 780 // CHECK-32-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 4
 781 // CHECK-32-NEXT:    store ptr [[E]], ptr [[E_ADDR]], align 4
 782 // CHECK-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[E_ADDR]], align 4
 783 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l24_kernel_environment, ptr [[DYN_PTR]])
 784 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 785 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 786 // CHECK-32:       user_code.entry:
 787 // CHECK-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 788 // CHECK-32-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
 789 // CHECK-32-NEXT:    store ptr [[TMP0]], ptr [[TMP3]], align 4
 790 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l24_omp_outlined, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 1)
 791 // CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 792 // CHECK-32-NEXT:    ret void
 793 // CHECK-32:       worker.exit:
 794 // CHECK-32-NEXT:    ret void
 795 //
 796 //
 797 // CHECK-32-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l24_omp_outlined
 798 // CHECK-32-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[E:%.*]]) #[[ATTR1:[0-9]+]] {
 799 // CHECK-32-NEXT:  entry:
 800 // CHECK-32-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 801 // CHECK-32-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 802 // CHECK-32-NEXT:    [[E_ADDR:%.*]] = alloca ptr, align 4
 803 // CHECK-32-NEXT:    [[E1:%.*]] = alloca double, align 8
 804 // CHECK-32-NEXT:    [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x ptr], align 4
 805 // CHECK-32-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 806 // CHECK-32-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 807 // CHECK-32-NEXT:    store ptr [[E]], ptr [[E_ADDR]], align 4
 808 // CHECK-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[E_ADDR]], align 4
 809 // CHECK-32-NEXT:    store double 0.000000e+00, ptr [[E1]], align 8
 810 // CHECK-32-NEXT:    [[TMP1:%.*]] = load double, ptr [[E1]], align 8
 811 // CHECK-32-NEXT:    [[ADD:%.*]] = fadd double [[TMP1]], 5.000000e+00
 812 // CHECK-32-NEXT:    store double [[ADD]], ptr [[E1]], align 8
 813 // CHECK-32-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0
 814 // CHECK-32-NEXT:    store ptr [[E1]], ptr [[TMP2]], align 4
 815 // CHECK-32-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_nvptx_parallel_reduce_nowait_v2(ptr @[[GLOB1]], i64 8, ptr [[DOTOMP_REDUCTION_RED_LIST]], ptr @_omp_reduction_shuffle_and_reduce_func, ptr @_omp_reduction_inter_warp_copy_func)
 816 // CHECK-32-NEXT:    [[TMP4:%.*]] = icmp eq i32 [[TMP3]], 1
 817 // CHECK-32-NEXT:    br i1 [[TMP4]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]]
 818 // CHECK-32:       .omp.reduction.then:
 819 // CHECK-32-NEXT:    [[TMP5:%.*]] = load double, ptr [[TMP0]], align 8
 820 // CHECK-32-NEXT:    [[TMP6:%.*]] = load double, ptr [[E1]], align 8
 821 // CHECK-32-NEXT:    [[ADD2:%.*]] = fadd double [[TMP5]], [[TMP6]]
 822 // CHECK-32-NEXT:    store double [[ADD2]], ptr [[TMP0]], align 8
 823 // CHECK-32-NEXT:    br label [[DOTOMP_REDUCTION_DONE]]
 824 // CHECK-32:       .omp.reduction.done:
 825 // CHECK-32-NEXT:    ret void
 826 //
 827 //
 828 // CHECK-32-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func
 829 // CHECK-32-SAME: (ptr noundef [[TMP0:%.*]], i16 noundef signext [[TMP1:%.*]], i16 noundef signext [[TMP2:%.*]], i16 noundef signext [[TMP3:%.*]]) #[[ATTR2:[0-9]+]] {
 830 // CHECK-32-NEXT:  entry:
 831 // CHECK-32-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 4
 832 // CHECK-32-NEXT:    [[DOTADDR1:%.*]] = alloca i16, align 2
 833 // CHECK-32-NEXT:    [[DOTADDR2:%.*]] = alloca i16, align 2
 834 // CHECK-32-NEXT:    [[DOTADDR3:%.*]] = alloca i16, align 2
 835 // CHECK-32-NEXT:    [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [1 x ptr], align 4
 836 // CHECK-32-NEXT:    [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca double, align 8
 837 // CHECK-32-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 4
 838 // CHECK-32-NEXT:    store i16 [[TMP1]], ptr [[DOTADDR1]], align 2
 839 // CHECK-32-NEXT:    store i16 [[TMP2]], ptr [[DOTADDR2]], align 2
 840 // CHECK-32-NEXT:    store i16 [[TMP3]], ptr [[DOTADDR3]], align 2
 841 // CHECK-32-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR]], align 4
 842 // CHECK-32-NEXT:    [[TMP5:%.*]] = load i16, ptr [[DOTADDR1]], align 2
 843 // CHECK-32-NEXT:    [[TMP6:%.*]] = load i16, ptr [[DOTADDR2]], align 2
 844 // CHECK-32-NEXT:    [[TMP7:%.*]] = load i16, ptr [[DOTADDR3]], align 2
 845 // CHECK-32-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP4]], i32 0, i32 0
 846 // CHECK-32-NEXT:    [[TMP9:%.*]] = load ptr, ptr [[TMP8]], align 4
 847 // CHECK-32-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0
 848 // CHECK-32-NEXT:    [[TMP11:%.*]] = getelementptr double, ptr [[TMP9]], i32 1
 849 // CHECK-32-NEXT:    [[TMP12:%.*]] = load i64, ptr [[TMP9]], align 8
 850 // CHECK-32-NEXT:    [[TMP13:%.*]] = call i32 @__kmpc_get_warp_size()
 851 // CHECK-32-NEXT:    [[TMP14:%.*]] = trunc i32 [[TMP13]] to i16
 852 // CHECK-32-NEXT:    [[TMP15:%.*]] = call i64 @__kmpc_shuffle_int64(i64 [[TMP12]], i16 [[TMP6]], i16 [[TMP14]])
 853 // CHECK-32-NEXT:    store i64 [[TMP15]], ptr [[DOTOMP_REDUCTION_ELEMENT]], align 8
 854 // CHECK-32-NEXT:    [[TMP16:%.*]] = getelementptr i64, ptr [[TMP9]], i32 1
 855 // CHECK-32-NEXT:    [[TMP17:%.*]] = getelementptr i64, ptr [[DOTOMP_REDUCTION_ELEMENT]], i32 1
 856 // CHECK-32-NEXT:    store ptr [[DOTOMP_REDUCTION_ELEMENT]], ptr [[TMP10]], align 4
 857 // CHECK-32-NEXT:    [[TMP18:%.*]] = icmp eq i16 [[TMP7]], 0
 858 // CHECK-32-NEXT:    [[TMP19:%.*]] = icmp eq i16 [[TMP7]], 1
 859 // CHECK-32-NEXT:    [[TMP20:%.*]] = icmp ult i16 [[TMP5]], [[TMP6]]
 860 // CHECK-32-NEXT:    [[TMP21:%.*]] = and i1 [[TMP19]], [[TMP20]]
 861 // CHECK-32-NEXT:    [[TMP22:%.*]] = icmp eq i16 [[TMP7]], 2
 862 // CHECK-32-NEXT:    [[TMP23:%.*]] = and i16 [[TMP5]], 1
 863 // CHECK-32-NEXT:    [[TMP24:%.*]] = icmp eq i16 [[TMP23]], 0
 864 // CHECK-32-NEXT:    [[TMP25:%.*]] = and i1 [[TMP22]], [[TMP24]]
 865 // CHECK-32-NEXT:    [[TMP26:%.*]] = icmp sgt i16 [[TMP6]], 0
 866 // CHECK-32-NEXT:    [[TMP27:%.*]] = and i1 [[TMP25]], [[TMP26]]
 867 // CHECK-32-NEXT:    [[TMP28:%.*]] = or i1 [[TMP18]], [[TMP21]]
 868 // CHECK-32-NEXT:    [[TMP29:%.*]] = or i1 [[TMP28]], [[TMP27]]
 869 // CHECK-32-NEXT:    br i1 [[TMP29]], label [[THEN:%.*]], label [[ELSE:%.*]]
 870 // CHECK-32:       then:
 871 // CHECK-32-NEXT:    call void @"{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l24_omp_outlined_omp$reduction$reduction_func"(ptr [[TMP4]], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]]) #[[ATTR3:[0-9]+]]
 872 // CHECK-32-NEXT:    br label [[IFCONT:%.*]]
 873 // CHECK-32:       else:
 874 // CHECK-32-NEXT:    br label [[IFCONT]]
 875 // CHECK-32:       ifcont:
 876 // CHECK-32-NEXT:    [[TMP30:%.*]] = icmp eq i16 [[TMP7]], 1
 877 // CHECK-32-NEXT:    [[TMP31:%.*]] = icmp uge i16 [[TMP5]], [[TMP6]]
 878 // CHECK-32-NEXT:    [[TMP32:%.*]] = and i1 [[TMP30]], [[TMP31]]
 879 // CHECK-32-NEXT:    br i1 [[TMP32]], label [[THEN4:%.*]], label [[ELSE5:%.*]]
 880 // CHECK-32:       then4:
 881 // CHECK-32-NEXT:    [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0
 882 // CHECK-32-NEXT:    [[TMP34:%.*]] = load ptr, ptr [[TMP33]], align 4
 883 // CHECK-32-NEXT:    [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP4]], i32 0, i32 0
 884 // CHECK-32-NEXT:    [[TMP36:%.*]] = load ptr, ptr [[TMP35]], align 4
 885 // CHECK-32-NEXT:    [[TMP37:%.*]] = load double, ptr [[TMP34]], align 8
 886 // CHECK-32-NEXT:    store double [[TMP37]], ptr [[TMP36]], align 8
 887 // CHECK-32-NEXT:    br label [[IFCONT6:%.*]]
 888 // CHECK-32:       else5:
 889 // CHECK-32-NEXT:    br label [[IFCONT6]]
 890 // CHECK-32:       ifcont6:
 891 // CHECK-32-NEXT:    ret void
 892 //
 893 //
 894 // CHECK-32-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func
 895 // CHECK-32-SAME: (ptr noundef [[TMP0:%.*]], i32 noundef [[TMP1:%.*]]) #[[ATTR2]] {
 896 // CHECK-32-NEXT:  entry:
 897 // CHECK-32-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 4
 898 // CHECK-32-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
 899 // CHECK-32-NEXT:    [[DOTCNT_ADDR:%.*]] = alloca i32, align 4
 900 // CHECK-32-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 4
 901 // CHECK-32-NEXT:    store i32 [[TMP1]], ptr [[DOTADDR1]], align 4
 902 // CHECK-32-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
 903 // CHECK-32-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
 904 // CHECK-32-NEXT:    [[NVPTX_LANE_ID:%.*]] = and i32 [[TMP4]], 31
 905 // CHECK-32-NEXT:    [[TMP5:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
 906 // CHECK-32-NEXT:    [[NVPTX_WARP_ID:%.*]] = ashr i32 [[TMP5]], 5
 907 // CHECK-32-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[DOTADDR]], align 4
 908 // CHECK-32-NEXT:    store i32 0, ptr [[DOTCNT_ADDR]], align 4
 909 // CHECK-32-NEXT:    br label [[PRECOND:%.*]]
 910 // CHECK-32:       precond:
 911 // CHECK-32-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCNT_ADDR]], align 4
 912 // CHECK-32-NEXT:    [[TMP8:%.*]] = icmp ult i32 [[TMP7]], 2
 913 // CHECK-32-NEXT:    br i1 [[TMP8]], label [[BODY:%.*]], label [[EXIT:%.*]]
 914 // CHECK-32:       body:
 915 // CHECK-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
 916 // CHECK-32-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2:[0-9]+]], i32 [[TMP2]])
 917 // CHECK-32-NEXT:    [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0
 918 // CHECK-32-NEXT:    br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]]
 919 // CHECK-32:       then:
 920 // CHECK-32-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP6]], i32 0, i32 0
 921 // CHECK-32-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[TMP9]], align 4
 922 // CHECK-32-NEXT:    [[TMP11:%.*]] = getelementptr i32, ptr [[TMP10]], i32 [[TMP7]]
 923 // CHECK-32-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]]
 924 // CHECK-32-NEXT:    [[TMP13:%.*]] = load i32, ptr [[TMP11]], align 4
 925 // CHECK-32-NEXT:    store volatile i32 [[TMP13]], ptr addrspace(3) [[TMP12]], align 4
 926 // CHECK-32-NEXT:    br label [[IFCONT:%.*]]
 927 // CHECK-32:       else:
 928 // CHECK-32-NEXT:    br label [[IFCONT]]
 929 // CHECK-32:       ifcont:
 930 // CHECK-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
 931 // CHECK-32-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP2]])
 932 // CHECK-32-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTADDR1]], align 4
 933 // CHECK-32-NEXT:    [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[TMP3]], [[TMP14]]
 934 // CHECK-32-NEXT:    br i1 [[IS_ACTIVE_THREAD]], label [[THEN2:%.*]], label [[ELSE3:%.*]]
 935 // CHECK-32:       then3:
 936 // CHECK-32-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[TMP3]]
 937 // CHECK-32-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP6]], i32 0, i32 0
 938 // CHECK-32-NEXT:    [[TMP17:%.*]] = load ptr, ptr [[TMP16]], align 4
 939 // CHECK-32-NEXT:    [[TMP18:%.*]] = getelementptr i32, ptr [[TMP17]], i32 [[TMP7]]
 940 // CHECK-32-NEXT:    [[TMP19:%.*]] = load volatile i32, ptr addrspace(3) [[TMP15]], align 4
 941 // CHECK-32-NEXT:    store i32 [[TMP19]], ptr [[TMP18]], align 4
 942 // CHECK-32-NEXT:    br label [[IFCONT4:%.*]]
 943 // CHECK-32:       else4:
 944 // CHECK-32-NEXT:    br label [[IFCONT4]]
 945 // CHECK-32:       ifcont5:
 946 // CHECK-32-NEXT:    [[TMP20:%.*]] = add nsw i32 [[TMP7]], 1
 947 // CHECK-32-NEXT:    store i32 [[TMP20]], ptr [[DOTCNT_ADDR]], align 4
 948 // CHECK-32-NEXT:    br label [[PRECOND]]
 949 // CHECK-32:       exit:
 950 // CHECK-32-NEXT:    ret void
 951 //
 952 //
 953 // CHECK-32-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29
 954 // CHECK-32-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], ptr noundef nonnull align 1 dereferenceable(1) [[C:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[D:%.*]]) #[[ATTR0]] {
 955 // CHECK-32-NEXT:  entry:
 956 // CHECK-32-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 4
 957 // CHECK-32-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 958 // CHECK-32-NEXT:    [[D_ADDR:%.*]] = alloca ptr, align 4
 959 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [2 x ptr], align 4
 960 // CHECK-32-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 4
 961 // CHECK-32-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 962 // CHECK-32-NEXT:    store ptr [[D]], ptr [[D_ADDR]], align 4
 963 // CHECK-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 964 // CHECK-32-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[D_ADDR]], align 4
 965 // CHECK-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29_kernel_environment, ptr [[DYN_PTR]])
 966 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP2]], -1
 967 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 968 // CHECK-32:       user_code.entry:
 969 // CHECK-32-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
 970 // CHECK-32-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [2 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
 971 // CHECK-32-NEXT:    store ptr [[TMP0]], ptr [[TMP4]], align 4
 972 // CHECK-32-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [2 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
 973 // CHECK-32-NEXT:    store ptr [[TMP1]], ptr [[TMP5]], align 4
 974 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP3]], i32 1, i32 -1, i32 -1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29_omp_outlined, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 2)
 975 // CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 976 // CHECK-32-NEXT:    ret void
 977 // CHECK-32:       worker.exit:
 978 // CHECK-32-NEXT:    ret void
 979 //
 980 //
 981 // CHECK-32-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29_omp_outlined
 982 // CHECK-32-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 1 dereferenceable(1) [[C:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[D:%.*]]) #[[ATTR1]] {
 983 // CHECK-32-NEXT:  entry:
 984 // CHECK-32-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 985 // CHECK-32-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 986 // CHECK-32-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 987 // CHECK-32-NEXT:    [[D_ADDR:%.*]] = alloca ptr, align 4
 988 // CHECK-32-NEXT:    [[C1:%.*]] = alloca i8, align 1
 989 // CHECK-32-NEXT:    [[D2:%.*]] = alloca float, align 4
 990 // CHECK-32-NEXT:    [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x ptr], align 4
 991 // CHECK-32-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 992 // CHECK-32-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 993 // CHECK-32-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 994 // CHECK-32-NEXT:    store ptr [[D]], ptr [[D_ADDR]], align 4
 995 // CHECK-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 996 // CHECK-32-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[D_ADDR]], align 4
 997 // CHECK-32-NEXT:    store i8 0, ptr [[C1]], align 1
 998 // CHECK-32-NEXT:    store float 1.000000e+00, ptr [[D2]], align 4
 999 // CHECK-32-NEXT:    [[TMP2:%.*]] = load i8, ptr [[C1]], align 1
1000 // CHECK-32-NEXT:    [[CONV:%.*]] = sext i8 [[TMP2]] to i32
1001 // CHECK-32-NEXT:    [[XOR:%.*]] = xor i32 [[CONV]], 2
1002 // CHECK-32-NEXT:    [[CONV3:%.*]] = trunc i32 [[XOR]] to i8
1003 // CHECK-32-NEXT:    store i8 [[CONV3]], ptr [[C1]], align 1
1004 // CHECK-32-NEXT:    [[TMP3:%.*]] = load float, ptr [[D2]], align 4
1005 // CHECK-32-NEXT:    [[MUL:%.*]] = fmul float [[TMP3]], 3.300000e+01
1006 // CHECK-32-NEXT:    store float [[MUL]], ptr [[D2]], align 4
1007 // CHECK-32-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0
1008 // CHECK-32-NEXT:    store ptr [[C1]], ptr [[TMP4]], align 4
1009 // CHECK-32-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1
1010 // CHECK-32-NEXT:    store ptr [[D2]], ptr [[TMP5]], align 4
1011 // CHECK-32-NEXT:    [[TMP6:%.*]] = call i32 @__kmpc_nvptx_parallel_reduce_nowait_v2(ptr @[[GLOB1]], i64 8, ptr [[DOTOMP_REDUCTION_RED_LIST]], ptr @_omp_reduction_shuffle_and_reduce_func1, ptr @_omp_reduction_inter_warp_copy_func2)
1012 // CHECK-32-NEXT:    [[TMP7:%.*]] = icmp eq i32 [[TMP6]], 1
1013 // CHECK-32-NEXT:    br i1 [[TMP7]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]]
1014 // CHECK-32:       .omp.reduction.then:
1015 // CHECK-32-NEXT:    [[TMP8:%.*]] = load i8, ptr [[TMP0]], align 1
1016 // CHECK-32-NEXT:    [[CONV4:%.*]] = sext i8 [[TMP8]] to i32
1017 // CHECK-32-NEXT:    [[TMP9:%.*]] = load i8, ptr [[C1]], align 1
1018 // CHECK-32-NEXT:    [[CONV5:%.*]] = sext i8 [[TMP9]] to i32
1019 // CHECK-32-NEXT:    [[XOR6:%.*]] = xor i32 [[CONV4]], [[CONV5]]
1020 // CHECK-32-NEXT:    [[CONV7:%.*]] = trunc i32 [[XOR6]] to i8
1021 // CHECK-32-NEXT:    store i8 [[CONV7]], ptr [[TMP0]], align 1
1022 // CHECK-32-NEXT:    [[TMP10:%.*]] = load float, ptr [[TMP1]], align 4
1023 // CHECK-32-NEXT:    [[TMP11:%.*]] = load float, ptr [[D2]], align 4
1024 // CHECK-32-NEXT:    [[MUL8:%.*]] = fmul float [[TMP10]], [[TMP11]]
1025 // CHECK-32-NEXT:    store float [[MUL8]], ptr [[TMP1]], align 4
1026 // CHECK-32-NEXT:    br label [[DOTOMP_REDUCTION_DONE]]
1027 // CHECK-32:       .omp.reduction.done:
1028 // CHECK-32-NEXT:    ret void
1029 //
1030 //
1031 // CHECK-32-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func1
1032 // CHECK-32-SAME: (ptr noundef [[TMP0:%.*]], i16 noundef signext [[TMP1:%.*]], i16 noundef signext [[TMP2:%.*]], i16 noundef signext [[TMP3:%.*]]) #[[ATTR2]] {
1033 // CHECK-32-NEXT:  entry:
1034 // CHECK-32-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 4
1035 // CHECK-32-NEXT:    [[DOTADDR1:%.*]] = alloca i16, align 2
1036 // CHECK-32-NEXT:    [[DOTADDR2:%.*]] = alloca i16, align 2
1037 // CHECK-32-NEXT:    [[DOTADDR3:%.*]] = alloca i16, align 2
1038 // CHECK-32-NEXT:    [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [2 x ptr], align 4
1039 // CHECK-32-NEXT:    [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca i8, align 1
1040 // CHECK-32-NEXT:    [[DOTOMP_REDUCTION_ELEMENT4:%.*]] = alloca float, align 4
1041 // CHECK-32-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 4
1042 // CHECK-32-NEXT:    store i16 [[TMP1]], ptr [[DOTADDR1]], align 2
1043 // CHECK-32-NEXT:    store i16 [[TMP2]], ptr [[DOTADDR2]], align 2
1044 // CHECK-32-NEXT:    store i16 [[TMP3]], ptr [[DOTADDR3]], align 2
1045 // CHECK-32-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR]], align 4
1046 // CHECK-32-NEXT:    [[TMP5:%.*]] = load i16, ptr [[DOTADDR1]], align 2
1047 // CHECK-32-NEXT:    [[TMP6:%.*]] = load i16, ptr [[DOTADDR2]], align 2
1048 // CHECK-32-NEXT:    [[TMP7:%.*]] = load i16, ptr [[DOTADDR3]], align 2
1049 // CHECK-32-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP4]], i32 0, i32 0
1050 // CHECK-32-NEXT:    [[TMP9:%.*]] = load ptr, ptr [[TMP8]], align 4
1051 // CHECK-32-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0
1052 // CHECK-32-NEXT:    [[TMP11:%.*]] = getelementptr i8, ptr [[TMP9]], i32 1
1053 // CHECK-32-NEXT:    [[TMP12:%.*]] = load i8, ptr [[TMP9]], align 1
1054 // CHECK-32-NEXT:    [[TMP13:%.*]] = sext i8 [[TMP12]] to i32
1055 // CHECK-32-NEXT:    [[TMP14:%.*]] = call i32 @__kmpc_get_warp_size()
1056 // CHECK-32-NEXT:    [[TMP15:%.*]] = trunc i32 [[TMP14]] to i16
1057 // CHECK-32-NEXT:    [[TMP16:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP13]], i16 [[TMP6]], i16 [[TMP15]])
1058 // CHECK-32-NEXT:    [[TMP17:%.*]] = trunc i32 [[TMP16]] to i8
1059 // CHECK-32-NEXT:    store i8 [[TMP17]], ptr [[DOTOMP_REDUCTION_ELEMENT]], align 1
1060 // CHECK-32-NEXT:    [[TMP18:%.*]] = getelementptr i8, ptr [[TMP9]], i32 1
1061 // CHECK-32-NEXT:    [[TMP19:%.*]] = getelementptr i8, ptr [[DOTOMP_REDUCTION_ELEMENT]], i32 1
1062 // CHECK-32-NEXT:    store ptr [[DOTOMP_REDUCTION_ELEMENT]], ptr [[TMP10]], align 4
1063 // CHECK-32-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP4]], i32 0, i32 1
1064 // CHECK-32-NEXT:    [[TMP21:%.*]] = load ptr, ptr [[TMP20]], align 4
1065 // CHECK-32-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1
1066 // CHECK-32-NEXT:    [[TMP23:%.*]] = getelementptr float, ptr [[TMP21]], i32 1
1067 // CHECK-32-NEXT:    [[TMP24:%.*]] = load i32, ptr [[TMP21]], align 4
1068 // CHECK-32-NEXT:    [[TMP25:%.*]] = call i32 @__kmpc_get_warp_size()
1069 // CHECK-32-NEXT:    [[TMP26:%.*]] = trunc i32 [[TMP25]] to i16
1070 // CHECK-32-NEXT:    [[TMP27:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP24]], i16 [[TMP6]], i16 [[TMP26]])
1071 // CHECK-32-NEXT:    store i32 [[TMP27]], ptr [[DOTOMP_REDUCTION_ELEMENT4]], align 4
1072 // CHECK-32-NEXT:    [[TMP28:%.*]] = getelementptr i32, ptr [[TMP21]], i32 1
1073 // CHECK-32-NEXT:    [[TMP29:%.*]] = getelementptr i32, ptr [[DOTOMP_REDUCTION_ELEMENT4]], i32 1
1074 // CHECK-32-NEXT:    store ptr [[DOTOMP_REDUCTION_ELEMENT4]], ptr [[TMP22]], align 4
1075 // CHECK-32-NEXT:    [[TMP30:%.*]] = icmp eq i16 [[TMP7]], 0
1076 // CHECK-32-NEXT:    [[TMP31:%.*]] = icmp eq i16 [[TMP7]], 1
1077 // CHECK-32-NEXT:    [[TMP32:%.*]] = icmp ult i16 [[TMP5]], [[TMP6]]
1078 // CHECK-32-NEXT:    [[TMP33:%.*]] = and i1 [[TMP31]], [[TMP32]]
1079 // CHECK-32-NEXT:    [[TMP34:%.*]] = icmp eq i16 [[TMP7]], 2
1080 // CHECK-32-NEXT:    [[TMP35:%.*]] = and i16 [[TMP5]], 1
1081 // CHECK-32-NEXT:    [[TMP36:%.*]] = icmp eq i16 [[TMP35]], 0
1082 // CHECK-32-NEXT:    [[TMP37:%.*]] = and i1 [[TMP34]], [[TMP36]]
1083 // CHECK-32-NEXT:    [[TMP38:%.*]] = icmp sgt i16 [[TMP6]], 0
1084 // CHECK-32-NEXT:    [[TMP39:%.*]] = and i1 [[TMP37]], [[TMP38]]
1085 // CHECK-32-NEXT:    [[TMP40:%.*]] = or i1 [[TMP30]], [[TMP33]]
1086 // CHECK-32-NEXT:    [[TMP41:%.*]] = or i1 [[TMP40]], [[TMP39]]
1087 // CHECK-32-NEXT:    br i1 [[TMP41]], label [[THEN:%.*]], label [[ELSE:%.*]]
1088 // CHECK-32:       then:
1089 // CHECK-32-NEXT:    call void @"{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29_omp_outlined_omp$reduction$reduction_func"(ptr [[TMP4]], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]]) #[[ATTR3]]
1090 // CHECK-32-NEXT:    br label [[IFCONT:%.*]]
1091 // CHECK-32:       else:
1092 // CHECK-32-NEXT:    br label [[IFCONT]]
1093 // CHECK-32:       ifcont:
1094 // CHECK-32-NEXT:    [[TMP42:%.*]] = icmp eq i16 [[TMP7]], 1
1095 // CHECK-32-NEXT:    [[TMP43:%.*]] = icmp uge i16 [[TMP5]], [[TMP6]]
1096 // CHECK-32-NEXT:    [[TMP44:%.*]] = and i1 [[TMP42]], [[TMP43]]
1097 // CHECK-32-NEXT:    br i1 [[TMP44]], label [[THEN5:%.*]], label [[ELSE6:%.*]]
1098 // CHECK-32:       then5:
1099 // CHECK-32-NEXT:    [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0
1100 // CHECK-32-NEXT:    [[TMP46:%.*]] = load ptr, ptr [[TMP45]], align 4
1101 // CHECK-32-NEXT:    [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP4]], i32 0, i32 0
1102 // CHECK-32-NEXT:    [[TMP48:%.*]] = load ptr, ptr [[TMP47]], align 4
1103 // CHECK-32-NEXT:    [[TMP49:%.*]] = load i8, ptr [[TMP46]], align 1
1104 // CHECK-32-NEXT:    store i8 [[TMP49]], ptr [[TMP48]], align 1
1105 // CHECK-32-NEXT:    [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1
1106 // CHECK-32-NEXT:    [[TMP51:%.*]] = load ptr, ptr [[TMP50]], align 4
1107 // CHECK-32-NEXT:    [[TMP52:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP4]], i32 0, i32 1
1108 // CHECK-32-NEXT:    [[TMP53:%.*]] = load ptr, ptr [[TMP52]], align 4
1109 // CHECK-32-NEXT:    [[TMP54:%.*]] = load float, ptr [[TMP51]], align 4
1110 // CHECK-32-NEXT:    store float [[TMP54]], ptr [[TMP53]], align 4
1111 // CHECK-32-NEXT:    br label [[IFCONT7:%.*]]
1112 // CHECK-32:       else6:
1113 // CHECK-32-NEXT:    br label [[IFCONT7]]
1114 // CHECK-32:       ifcont7:
1115 // CHECK-32-NEXT:    ret void
1116 //
1117 //
1118 // CHECK-32-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func2
1119 // CHECK-32-SAME: (ptr noundef [[TMP0:%.*]], i32 noundef [[TMP1:%.*]]) #[[ATTR2]] {
1120 // CHECK-32-NEXT:  entry:
1121 // CHECK-32-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 4
1122 // CHECK-32-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
1123 // CHECK-32-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 4
1124 // CHECK-32-NEXT:    store i32 [[TMP1]], ptr [[DOTADDR1]], align 4
1125 // CHECK-32-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
1126 // CHECK-32-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
1127 // CHECK-32-NEXT:    [[NVPTX_LANE_ID:%.*]] = and i32 [[TMP4]], 31
1128 // CHECK-32-NEXT:    [[TMP5:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
1129 // CHECK-32-NEXT:    [[NVPTX_WARP_ID:%.*]] = ashr i32 [[TMP5]], 5
1130 // CHECK-32-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[DOTADDR]], align 4
1131 // CHECK-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
1132 // CHECK-32-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP2]])
1133 // CHECK-32-NEXT:    [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0
1134 // CHECK-32-NEXT:    br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]]
1135 // CHECK-32:       then:
1136 // CHECK-32-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP6]], i32 0, i32 0
1137 // CHECK-32-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[TMP7]], align 4
1138 // CHECK-32-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]]
1139 // CHECK-32-NEXT:    [[TMP10:%.*]] = load i8, ptr [[TMP8]], align 1
1140 // CHECK-32-NEXT:    store volatile i8 [[TMP10]], ptr addrspace(3) [[TMP9]], align 1
1141 // CHECK-32-NEXT:    br label [[IFCONT:%.*]]
1142 // CHECK-32:       else:
1143 // CHECK-32-NEXT:    br label [[IFCONT]]
1144 // CHECK-32:       ifcont:
1145 // CHECK-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
1146 // CHECK-32-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP2]])
1147 // CHECK-32-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTADDR1]], align 4
1148 // CHECK-32-NEXT:    [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[TMP3]], [[TMP11]]
1149 // CHECK-32-NEXT:    br i1 [[IS_ACTIVE_THREAD]], label [[THEN2:%.*]], label [[ELSE3:%.*]]
1150 // CHECK-32:       then3:
1151 // CHECK-32-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[TMP3]]
1152 // CHECK-32-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP6]], i32 0, i32 0
1153 // CHECK-32-NEXT:    [[TMP14:%.*]] = load ptr, ptr [[TMP13]], align 4
1154 // CHECK-32-NEXT:    [[TMP15:%.*]] = load volatile i8, ptr addrspace(3) [[TMP12]], align 1
1155 // CHECK-32-NEXT:    store i8 [[TMP15]], ptr [[TMP14]], align 1
1156 // CHECK-32-NEXT:    br label [[IFCONT4:%.*]]
1157 // CHECK-32:       else4:
1158 // CHECK-32-NEXT:    br label [[IFCONT4]]
1159 // CHECK-32:       ifcont5:
1160 // CHECK-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
1161 // CHECK-32-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP2]])
1162 // CHECK-32-NEXT:    [[WARP_MASTER5:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0
1163 // CHECK-32-NEXT:    br i1 [[WARP_MASTER5]], label [[THEN6:%.*]], label [[ELSE7:%.*]]
1164 // CHECK-32:       then8:
1165 // CHECK-32-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP6]], i32 0, i32 1
1166 // CHECK-32-NEXT:    [[TMP17:%.*]] = load ptr, ptr [[TMP16]], align 4
1167 // CHECK-32-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]]
1168 // CHECK-32-NEXT:    [[TMP19:%.*]] = load i32, ptr [[TMP17]], align 4
1169 // CHECK-32-NEXT:    store volatile i32 [[TMP19]], ptr addrspace(3) [[TMP18]], align 4
1170 // CHECK-32-NEXT:    br label [[IFCONT8:%.*]]
1171 // CHECK-32:       else9:
1172 // CHECK-32-NEXT:    br label [[IFCONT8]]
1173 // CHECK-32:       ifcont10:
1174 // CHECK-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
1175 // CHECK-32-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP2]])
1176 // CHECK-32-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTADDR1]], align 4
1177 // CHECK-32-NEXT:    [[IS_ACTIVE_THREAD9:%.*]] = icmp ult i32 [[TMP3]], [[TMP20]]
1178 // CHECK-32-NEXT:    br i1 [[IS_ACTIVE_THREAD9]], label [[THEN10:%.*]], label [[ELSE11:%.*]]
1179 // CHECK-32:       then13:
1180 // CHECK-32-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[TMP3]]
1181 // CHECK-32-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP6]], i32 0, i32 1
1182 // CHECK-32-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP22]], align 4
1183 // CHECK-32-NEXT:    [[TMP24:%.*]] = load volatile i32, ptr addrspace(3) [[TMP21]], align 4
1184 // CHECK-32-NEXT:    store i32 [[TMP24]], ptr [[TMP23]], align 4
1185 // CHECK-32-NEXT:    br label [[IFCONT12:%.*]]
1186 // CHECK-32:       else14:
1187 // CHECK-32-NEXT:    br label [[IFCONT12]]
1188 // CHECK-32:       ifcont15:
1189 // CHECK-32-NEXT:    ret void
1190 //
1191 //
1192 // CHECK-32-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l35
1193 // CHECK-32-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 2 dereferenceable(2) [[B:%.*]]) #[[ATTR0]] {
1194 // CHECK-32-NEXT:  entry:
1195 // CHECK-32-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 4
1196 // CHECK-32-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
1197 // CHECK-32-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
1198 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [2 x ptr], align 4
1199 // CHECK-32-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 4
1200 // CHECK-32-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
1201 // CHECK-32-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
1202 // CHECK-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
1203 // CHECK-32-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[B_ADDR]], align 4
1204 // CHECK-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l35_kernel_environment, ptr [[DYN_PTR]])
1205 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP2]], -1
1206 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
1207 // CHECK-32:       user_code.entry:
1208 // CHECK-32-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
1209 // CHECK-32-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [2 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
1210 // CHECK-32-NEXT:    store ptr [[TMP0]], ptr [[TMP4]], align 4
1211 // CHECK-32-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [2 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
1212 // CHECK-32-NEXT:    store ptr [[TMP1]], ptr [[TMP5]], align 4
1213 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP3]], i32 1, i32 -1, i32 -1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l35_omp_outlined, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 2)
1214 // CHECK-32-NEXT:    call void @__kmpc_target_deinit()
1215 // CHECK-32-NEXT:    ret void
1216 // CHECK-32:       worker.exit:
1217 // CHECK-32-NEXT:    ret void
1218 //
1219 //
1220 // CHECK-32-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l35_omp_outlined
1221 // CHECK-32-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 2 dereferenceable(2) [[B:%.*]]) #[[ATTR1]] {
1222 // CHECK-32-NEXT:  entry:
1223 // CHECK-32-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
1224 // CHECK-32-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
1225 // CHECK-32-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
1226 // CHECK-32-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
1227 // CHECK-32-NEXT:    [[A1:%.*]] = alloca i32, align 4
1228 // CHECK-32-NEXT:    [[B2:%.*]] = alloca i16, align 2
1229 // CHECK-32-NEXT:    [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x ptr], align 4
1230 // CHECK-32-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
1231 // CHECK-32-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
1232 // CHECK-32-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
1233 // CHECK-32-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
1234 // CHECK-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
1235 // CHECK-32-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[B_ADDR]], align 4
1236 // CHECK-32-NEXT:    store i32 0, ptr [[A1]], align 4
1237 // CHECK-32-NEXT:    store i16 -32768, ptr [[B2]], align 2
1238 // CHECK-32-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A1]], align 4
1239 // CHECK-32-NEXT:    [[OR:%.*]] = or i32 [[TMP2]], 1
1240 // CHECK-32-NEXT:    store i32 [[OR]], ptr [[A1]], align 4
1241 // CHECK-32-NEXT:    [[TMP3:%.*]] = load i16, ptr [[B2]], align 2
1242 // CHECK-32-NEXT:    [[CONV:%.*]] = sext i16 [[TMP3]] to i32
1243 // CHECK-32-NEXT:    [[CMP:%.*]] = icmp sgt i32 99, [[CONV]]
1244 // CHECK-32-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
1245 // CHECK-32:       cond.true:
1246 // CHECK-32-NEXT:    br label [[COND_END:%.*]]
1247 // CHECK-32:       cond.false:
1248 // CHECK-32-NEXT:    [[TMP4:%.*]] = load i16, ptr [[B2]], align 2
1249 // CHECK-32-NEXT:    [[CONV3:%.*]] = sext i16 [[TMP4]] to i32
1250 // CHECK-32-NEXT:    br label [[COND_END]]
1251 // CHECK-32:       cond.end:
1252 // CHECK-32-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[CONV3]], [[COND_FALSE]] ]
1253 // CHECK-32-NEXT:    [[CONV4:%.*]] = trunc i32 [[COND]] to i16
1254 // CHECK-32-NEXT:    store i16 [[CONV4]], ptr [[B2]], align 2
1255 // CHECK-32-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0
1256 // CHECK-32-NEXT:    store ptr [[A1]], ptr [[TMP5]], align 4
1257 // CHECK-32-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1
1258 // CHECK-32-NEXT:    store ptr [[B2]], ptr [[TMP6]], align 4
1259 // CHECK-32-NEXT:    [[TMP7:%.*]] = call i32 @__kmpc_nvptx_parallel_reduce_nowait_v2(ptr @[[GLOB1]], i64 8, ptr [[DOTOMP_REDUCTION_RED_LIST]], ptr @_omp_reduction_shuffle_and_reduce_func3, ptr @_omp_reduction_inter_warp_copy_func4)
1260 // CHECK-32-NEXT:    [[TMP8:%.*]] = icmp eq i32 [[TMP7]], 1
1261 // CHECK-32-NEXT:    br i1 [[TMP8]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]]
1262 // CHECK-32:       .omp.reduction.then:
1263 // CHECK-32-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP0]], align 4
1264 // CHECK-32-NEXT:    [[TMP10:%.*]] = load i32, ptr [[A1]], align 4
1265 // CHECK-32-NEXT:    [[OR5:%.*]] = or i32 [[TMP9]], [[TMP10]]
1266 // CHECK-32-NEXT:    store i32 [[OR5]], ptr [[TMP0]], align 4
1267 // CHECK-32-NEXT:    [[TMP11:%.*]] = load i16, ptr [[TMP1]], align 2
1268 // CHECK-32-NEXT:    [[CONV6:%.*]] = sext i16 [[TMP11]] to i32
1269 // CHECK-32-NEXT:    [[TMP12:%.*]] = load i16, ptr [[B2]], align 2
1270 // CHECK-32-NEXT:    [[CONV7:%.*]] = sext i16 [[TMP12]] to i32
1271 // CHECK-32-NEXT:    [[CMP8:%.*]] = icmp sgt i32 [[CONV6]], [[CONV7]]
1272 // CHECK-32-NEXT:    br i1 [[CMP8]], label [[COND_TRUE9:%.*]], label [[COND_FALSE10:%.*]]
1273 // CHECK-32:       cond.true9:
1274 // CHECK-32-NEXT:    [[TMP13:%.*]] = load i16, ptr [[TMP1]], align 2
1275 // CHECK-32-NEXT:    br label [[COND_END11:%.*]]
1276 // CHECK-32:       cond.false10:
1277 // CHECK-32-NEXT:    [[TMP14:%.*]] = load i16, ptr [[B2]], align 2
1278 // CHECK-32-NEXT:    br label [[COND_END11]]
1279 // CHECK-32:       cond.end11:
1280 // CHECK-32-NEXT:    [[COND12:%.*]] = phi i16 [ [[TMP13]], [[COND_TRUE9]] ], [ [[TMP14]], [[COND_FALSE10]] ]
1281 // CHECK-32-NEXT:    store i16 [[COND12]], ptr [[TMP1]], align 2
1282 // CHECK-32-NEXT:    br label [[DOTOMP_REDUCTION_DONE]]
1283 // CHECK-32:       .omp.reduction.done:
1284 // CHECK-32-NEXT:    ret void
1285 //
1286 //
1287 // CHECK-32-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func3
1288 // CHECK-32-SAME: (ptr noundef [[TMP0:%.*]], i16 noundef signext [[TMP1:%.*]], i16 noundef signext [[TMP2:%.*]], i16 noundef signext [[TMP3:%.*]]) #[[ATTR2]] {
1289 // CHECK-32-NEXT:  entry:
1290 // CHECK-32-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 4
1291 // CHECK-32-NEXT:    [[DOTADDR1:%.*]] = alloca i16, align 2
1292 // CHECK-32-NEXT:    [[DOTADDR2:%.*]] = alloca i16, align 2
1293 // CHECK-32-NEXT:    [[DOTADDR3:%.*]] = alloca i16, align 2
1294 // CHECK-32-NEXT:    [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [2 x ptr], align 4
1295 // CHECK-32-NEXT:    [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca i32, align 4
1296 // CHECK-32-NEXT:    [[DOTOMP_REDUCTION_ELEMENT4:%.*]] = alloca i16, align 2
1297 // CHECK-32-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 4
1298 // CHECK-32-NEXT:    store i16 [[TMP1]], ptr [[DOTADDR1]], align 2
1299 // CHECK-32-NEXT:    store i16 [[TMP2]], ptr [[DOTADDR2]], align 2
1300 // CHECK-32-NEXT:    store i16 [[TMP3]], ptr [[DOTADDR3]], align 2
1301 // CHECK-32-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR]], align 4
1302 // CHECK-32-NEXT:    [[TMP5:%.*]] = load i16, ptr [[DOTADDR1]], align 2
1303 // CHECK-32-NEXT:    [[TMP6:%.*]] = load i16, ptr [[DOTADDR2]], align 2
1304 // CHECK-32-NEXT:    [[TMP7:%.*]] = load i16, ptr [[DOTADDR3]], align 2
1305 // CHECK-32-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP4]], i32 0, i32 0
1306 // CHECK-32-NEXT:    [[TMP9:%.*]] = load ptr, ptr [[TMP8]], align 4
1307 // CHECK-32-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0
1308 // CHECK-32-NEXT:    [[TMP11:%.*]] = getelementptr i32, ptr [[TMP9]], i32 1
1309 // CHECK-32-NEXT:    [[TMP12:%.*]] = load i32, ptr [[TMP9]], align 4
1310 // CHECK-32-NEXT:    [[TMP13:%.*]] = call i32 @__kmpc_get_warp_size()
1311 // CHECK-32-NEXT:    [[TMP14:%.*]] = trunc i32 [[TMP13]] to i16
1312 // CHECK-32-NEXT:    [[TMP15:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP12]], i16 [[TMP6]], i16 [[TMP14]])
1313 // CHECK-32-NEXT:    store i32 [[TMP15]], ptr [[DOTOMP_REDUCTION_ELEMENT]], align 4
1314 // CHECK-32-NEXT:    [[TMP16:%.*]] = getelementptr i32, ptr [[TMP9]], i32 1
1315 // CHECK-32-NEXT:    [[TMP17:%.*]] = getelementptr i32, ptr [[DOTOMP_REDUCTION_ELEMENT]], i32 1
1316 // CHECK-32-NEXT:    store ptr [[DOTOMP_REDUCTION_ELEMENT]], ptr [[TMP10]], align 4
1317 // CHECK-32-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP4]], i32 0, i32 1
1318 // CHECK-32-NEXT:    [[TMP19:%.*]] = load ptr, ptr [[TMP18]], align 4
1319 // CHECK-32-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1
1320 // CHECK-32-NEXT:    [[TMP21:%.*]] = getelementptr i16, ptr [[TMP19]], i32 1
1321 // CHECK-32-NEXT:    [[TMP22:%.*]] = load i16, ptr [[TMP19]], align 2
1322 // CHECK-32-NEXT:    [[TMP23:%.*]] = sext i16 [[TMP22]] to i32
1323 // CHECK-32-NEXT:    [[TMP24:%.*]] = call i32 @__kmpc_get_warp_size()
1324 // CHECK-32-NEXT:    [[TMP25:%.*]] = trunc i32 [[TMP24]] to i16
1325 // CHECK-32-NEXT:    [[TMP26:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP23]], i16 [[TMP6]], i16 [[TMP25]])
1326 // CHECK-32-NEXT:    [[TMP27:%.*]] = trunc i32 [[TMP26]] to i16
1327 // CHECK-32-NEXT:    store i16 [[TMP27]], ptr [[DOTOMP_REDUCTION_ELEMENT4]], align 2
1328 // CHECK-32-NEXT:    [[TMP28:%.*]] = getelementptr i16, ptr [[TMP19]], i32 1
1329 // CHECK-32-NEXT:    [[TMP29:%.*]] = getelementptr i16, ptr [[DOTOMP_REDUCTION_ELEMENT4]], i32 1
1330 // CHECK-32-NEXT:    store ptr [[DOTOMP_REDUCTION_ELEMENT4]], ptr [[TMP20]], align 4
1331 // CHECK-32-NEXT:    [[TMP30:%.*]] = icmp eq i16 [[TMP7]], 0
1332 // CHECK-32-NEXT:    [[TMP31:%.*]] = icmp eq i16 [[TMP7]], 1
1333 // CHECK-32-NEXT:    [[TMP32:%.*]] = icmp ult i16 [[TMP5]], [[TMP6]]
1334 // CHECK-32-NEXT:    [[TMP33:%.*]] = and i1 [[TMP31]], [[TMP32]]
1335 // CHECK-32-NEXT:    [[TMP34:%.*]] = icmp eq i16 [[TMP7]], 2
1336 // CHECK-32-NEXT:    [[TMP35:%.*]] = and i16 [[TMP5]], 1
1337 // CHECK-32-NEXT:    [[TMP36:%.*]] = icmp eq i16 [[TMP35]], 0
1338 // CHECK-32-NEXT:    [[TMP37:%.*]] = and i1 [[TMP34]], [[TMP36]]
1339 // CHECK-32-NEXT:    [[TMP38:%.*]] = icmp sgt i16 [[TMP6]], 0
1340 // CHECK-32-NEXT:    [[TMP39:%.*]] = and i1 [[TMP37]], [[TMP38]]
1341 // CHECK-32-NEXT:    [[TMP40:%.*]] = or i1 [[TMP30]], [[TMP33]]
1342 // CHECK-32-NEXT:    [[TMP41:%.*]] = or i1 [[TMP40]], [[TMP39]]
1343 // CHECK-32-NEXT:    br i1 [[TMP41]], label [[THEN:%.*]], label [[ELSE:%.*]]
1344 // CHECK-32:       then:
1345 // CHECK-32-NEXT:    call void @"{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l35_omp_outlined_omp$reduction$reduction_func"(ptr [[TMP4]], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]]) #[[ATTR3]]
1346 // CHECK-32-NEXT:    br label [[IFCONT:%.*]]
1347 // CHECK-32:       else:
1348 // CHECK-32-NEXT:    br label [[IFCONT]]
1349 // CHECK-32:       ifcont:
1350 // CHECK-32-NEXT:    [[TMP42:%.*]] = icmp eq i16 [[TMP7]], 1
1351 // CHECK-32-NEXT:    [[TMP43:%.*]] = icmp uge i16 [[TMP5]], [[TMP6]]
1352 // CHECK-32-NEXT:    [[TMP44:%.*]] = and i1 [[TMP42]], [[TMP43]]
1353 // CHECK-32-NEXT:    br i1 [[TMP44]], label [[THEN5:%.*]], label [[ELSE6:%.*]]
1354 // CHECK-32:       then5:
1355 // CHECK-32-NEXT:    [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0
1356 // CHECK-32-NEXT:    [[TMP46:%.*]] = load ptr, ptr [[TMP45]], align 4
1357 // CHECK-32-NEXT:    [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP4]], i32 0, i32 0
1358 // CHECK-32-NEXT:    [[TMP48:%.*]] = load ptr, ptr [[TMP47]], align 4
1359 // CHECK-32-NEXT:    [[TMP49:%.*]] = load i32, ptr [[TMP46]], align 4
1360 // CHECK-32-NEXT:    store i32 [[TMP49]], ptr [[TMP48]], align 4
1361 // CHECK-32-NEXT:    [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1
1362 // CHECK-32-NEXT:    [[TMP51:%.*]] = load ptr, ptr [[TMP50]], align 4
1363 // CHECK-32-NEXT:    [[TMP52:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP4]], i32 0, i32 1
1364 // CHECK-32-NEXT:    [[TMP53:%.*]] = load ptr, ptr [[TMP52]], align 4
1365 // CHECK-32-NEXT:    [[TMP54:%.*]] = load i16, ptr [[TMP51]], align 2
1366 // CHECK-32-NEXT:    store i16 [[TMP54]], ptr [[TMP53]], align 2
1367 // CHECK-32-NEXT:    br label [[IFCONT7:%.*]]
1368 // CHECK-32:       else6:
1369 // CHECK-32-NEXT:    br label [[IFCONT7]]
1370 // CHECK-32:       ifcont7:
1371 // CHECK-32-NEXT:    ret void
1372 //
1373 //
1374 // CHECK-32-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func4
1375 // CHECK-32-SAME: (ptr noundef [[TMP0:%.*]], i32 noundef [[TMP1:%.*]]) #[[ATTR2]] {
1376 // CHECK-32-NEXT:  entry:
1377 // CHECK-32-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 4
1378 // CHECK-32-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
1379 // CHECK-32-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 4
1380 // CHECK-32-NEXT:    store i32 [[TMP1]], ptr [[DOTADDR1]], align 4
1381 // CHECK-32-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
1382 // CHECK-32-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
1383 // CHECK-32-NEXT:    [[NVPTX_LANE_ID:%.*]] = and i32 [[TMP4]], 31
1384 // CHECK-32-NEXT:    [[TMP5:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
1385 // CHECK-32-NEXT:    [[NVPTX_WARP_ID:%.*]] = ashr i32 [[TMP5]], 5
1386 // CHECK-32-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[DOTADDR]], align 4
1387 // CHECK-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
1388 // CHECK-32-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP2]])
1389 // CHECK-32-NEXT:    [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0
1390 // CHECK-32-NEXT:    br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]]
1391 // CHECK-32:       then:
1392 // CHECK-32-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP6]], i32 0, i32 0
1393 // CHECK-32-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[TMP7]], align 4
1394 // CHECK-32-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]]
1395 // CHECK-32-NEXT:    [[TMP10:%.*]] = load i32, ptr [[TMP8]], align 4
1396 // CHECK-32-NEXT:    store volatile i32 [[TMP10]], ptr addrspace(3) [[TMP9]], align 4
1397 // CHECK-32-NEXT:    br label [[IFCONT:%.*]]
1398 // CHECK-32:       else:
1399 // CHECK-32-NEXT:    br label [[IFCONT]]
1400 // CHECK-32:       ifcont:
1401 // CHECK-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
1402 // CHECK-32-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP2]])
1403 // CHECK-32-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTADDR1]], align 4
1404 // CHECK-32-NEXT:    [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[TMP3]], [[TMP11]]
1405 // CHECK-32-NEXT:    br i1 [[IS_ACTIVE_THREAD]], label [[THEN2:%.*]], label [[ELSE3:%.*]]
1406 // CHECK-32:       then3:
1407 // CHECK-32-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[TMP3]]
1408 // CHECK-32-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP6]], i32 0, i32 0
1409 // CHECK-32-NEXT:    [[TMP14:%.*]] = load ptr, ptr [[TMP13]], align 4
1410 // CHECK-32-NEXT:    [[TMP15:%.*]] = load volatile i32, ptr addrspace(3) [[TMP12]], align 4
1411 // CHECK-32-NEXT:    store i32 [[TMP15]], ptr [[TMP14]], align 4
1412 // CHECK-32-NEXT:    br label [[IFCONT4:%.*]]
1413 // CHECK-32:       else4:
1414 // CHECK-32-NEXT:    br label [[IFCONT4]]
1415 // CHECK-32:       ifcont5:
1416 // CHECK-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
1417 // CHECK-32-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP2]])
1418 // CHECK-32-NEXT:    [[WARP_MASTER5:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0
1419 // CHECK-32-NEXT:    br i1 [[WARP_MASTER5]], label [[THEN6:%.*]], label [[ELSE7:%.*]]
1420 // CHECK-32:       then8:
1421 // CHECK-32-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP6]], i32 0, i32 1
1422 // CHECK-32-NEXT:    [[TMP17:%.*]] = load ptr, ptr [[TMP16]], align 4
1423 // CHECK-32-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]]
1424 // CHECK-32-NEXT:    [[TMP19:%.*]] = load i16, ptr [[TMP17]], align 2
1425 // CHECK-32-NEXT:    store volatile i16 [[TMP19]], ptr addrspace(3) [[TMP18]], align 2
1426 // CHECK-32-NEXT:    br label [[IFCONT8:%.*]]
1427 // CHECK-32:       else9:
1428 // CHECK-32-NEXT:    br label [[IFCONT8]]
1429 // CHECK-32:       ifcont10:
1430 // CHECK-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
1431 // CHECK-32-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP2]])
1432 // CHECK-32-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTADDR1]], align 4
1433 // CHECK-32-NEXT:    [[IS_ACTIVE_THREAD9:%.*]] = icmp ult i32 [[TMP3]], [[TMP20]]
1434 // CHECK-32-NEXT:    br i1 [[IS_ACTIVE_THREAD9]], label [[THEN10:%.*]], label [[ELSE11:%.*]]
1435 // CHECK-32:       then13:
1436 // CHECK-32-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[TMP3]]
1437 // CHECK-32-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP6]], i32 0, i32 1
1438 // CHECK-32-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP22]], align 4
1439 // CHECK-32-NEXT:    [[TMP24:%.*]] = load volatile i16, ptr addrspace(3) [[TMP21]], align 2
1440 // CHECK-32-NEXT:    store i16 [[TMP24]], ptr [[TMP23]], align 2
1441 // CHECK-32-NEXT:    br label [[IFCONT12:%.*]]
1442 // CHECK-32:       else14:
1443 // CHECK-32-NEXT:    br label [[IFCONT12]]
1444 // CHECK-32:       ifcont15:
1445 // CHECK-32-NEXT:    ret void
1446 //
1447 //
1448 // CHECK-32-EX-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l24
1449 // CHECK-32-EX-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[E:%.*]]) #[[ATTR0:[0-9]+]] {
1450 // CHECK-32-EX-NEXT:  entry:
1451 // CHECK-32-EX-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 4
1452 // CHECK-32-EX-NEXT:    [[E_ADDR:%.*]] = alloca ptr, align 4
1453 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x ptr], align 4
1454 // CHECK-32-EX-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 4
1455 // CHECK-32-EX-NEXT:    store ptr [[E]], ptr [[E_ADDR]], align 4
1456 // CHECK-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[E_ADDR]], align 4
1457 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l24_kernel_environment, ptr [[DYN_PTR]])
1458 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
1459 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
1460 // CHECK-32-EX:       user_code.entry:
1461 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
1462 // CHECK-32-EX-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
1463 // CHECK-32-EX-NEXT:    store ptr [[TMP0]], ptr [[TMP3]], align 4
1464 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l24_omp_outlined, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 1)
1465 // CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
1466 // CHECK-32-EX-NEXT:    ret void
1467 // CHECK-32-EX:       worker.exit:
1468 // CHECK-32-EX-NEXT:    ret void
1469 //
1470 //
1471 // CHECK-32-EX-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l24_omp_outlined
1472 // CHECK-32-EX-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[E:%.*]]) #[[ATTR1:[0-9]+]] {
1473 // CHECK-32-EX-NEXT:  entry:
1474 // CHECK-32-EX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
1475 // CHECK-32-EX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
1476 // CHECK-32-EX-NEXT:    [[E_ADDR:%.*]] = alloca ptr, align 4
1477 // CHECK-32-EX-NEXT:    [[E1:%.*]] = alloca double, align 8
1478 // CHECK-32-EX-NEXT:    [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x ptr], align 4
1479 // CHECK-32-EX-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
1480 // CHECK-32-EX-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
1481 // CHECK-32-EX-NEXT:    store ptr [[E]], ptr [[E_ADDR]], align 4
1482 // CHECK-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[E_ADDR]], align 4
1483 // CHECK-32-EX-NEXT:    store double 0.000000e+00, ptr [[E1]], align 8
1484 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = load double, ptr [[E1]], align 8
1485 // CHECK-32-EX-NEXT:    [[ADD:%.*]] = fadd double [[TMP1]], 5.000000e+00
1486 // CHECK-32-EX-NEXT:    store double [[ADD]], ptr [[E1]], align 8
1487 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0
1488 // CHECK-32-EX-NEXT:    store ptr [[E1]], ptr [[TMP2]], align 4
1489 // CHECK-32-EX-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_nvptx_parallel_reduce_nowait_v2(ptr @[[GLOB1]], i64 8, ptr [[DOTOMP_REDUCTION_RED_LIST]], ptr @_omp_reduction_shuffle_and_reduce_func, ptr @_omp_reduction_inter_warp_copy_func)
1490 // CHECK-32-EX-NEXT:    [[TMP4:%.*]] = icmp eq i32 [[TMP3]], 1
1491 // CHECK-32-EX-NEXT:    br i1 [[TMP4]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]]
1492 // CHECK-32-EX:       .omp.reduction.then:
1493 // CHECK-32-EX-NEXT:    [[TMP5:%.*]] = load double, ptr [[TMP0]], align 8
1494 // CHECK-32-EX-NEXT:    [[TMP6:%.*]] = load double, ptr [[E1]], align 8
1495 // CHECK-32-EX-NEXT:    [[ADD2:%.*]] = fadd double [[TMP5]], [[TMP6]]
1496 // CHECK-32-EX-NEXT:    store double [[ADD2]], ptr [[TMP0]], align 8
1497 // CHECK-32-EX-NEXT:    br label [[DOTOMP_REDUCTION_DONE]]
1498 // CHECK-32-EX:       .omp.reduction.done:
1499 // CHECK-32-EX-NEXT:    ret void
1500 //
1501 //
1502 // CHECK-32-EX-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func
1503 // CHECK-32-EX-SAME: (ptr noundef [[TMP0:%.*]], i16 noundef signext [[TMP1:%.*]], i16 noundef signext [[TMP2:%.*]], i16 noundef signext [[TMP3:%.*]]) #[[ATTR2:[0-9]+]] {
1504 // CHECK-32-EX-NEXT:  entry:
1505 // CHECK-32-EX-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 4
1506 // CHECK-32-EX-NEXT:    [[DOTADDR1:%.*]] = alloca i16, align 2
1507 // CHECK-32-EX-NEXT:    [[DOTADDR2:%.*]] = alloca i16, align 2
1508 // CHECK-32-EX-NEXT:    [[DOTADDR3:%.*]] = alloca i16, align 2
1509 // CHECK-32-EX-NEXT:    [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [1 x ptr], align 4
1510 // CHECK-32-EX-NEXT:    [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca double, align 8
1511 // CHECK-32-EX-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 4
1512 // CHECK-32-EX-NEXT:    store i16 [[TMP1]], ptr [[DOTADDR1]], align 2
1513 // CHECK-32-EX-NEXT:    store i16 [[TMP2]], ptr [[DOTADDR2]], align 2
1514 // CHECK-32-EX-NEXT:    store i16 [[TMP3]], ptr [[DOTADDR3]], align 2
1515 // CHECK-32-EX-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR]], align 4
1516 // CHECK-32-EX-NEXT:    [[TMP5:%.*]] = load i16, ptr [[DOTADDR1]], align 2
1517 // CHECK-32-EX-NEXT:    [[TMP6:%.*]] = load i16, ptr [[DOTADDR2]], align 2
1518 // CHECK-32-EX-NEXT:    [[TMP7:%.*]] = load i16, ptr [[DOTADDR3]], align 2
1519 // CHECK-32-EX-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP4]], i32 0, i32 0
1520 // CHECK-32-EX-NEXT:    [[TMP9:%.*]] = load ptr, ptr [[TMP8]], align 4
1521 // CHECK-32-EX-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0
1522 // CHECK-32-EX-NEXT:    [[TMP11:%.*]] = getelementptr double, ptr [[TMP9]], i32 1
1523 // CHECK-32-EX-NEXT:    [[TMP12:%.*]] = load i64, ptr [[TMP9]], align 8
1524 // CHECK-32-EX-NEXT:    [[TMP13:%.*]] = call i32 @__kmpc_get_warp_size()
1525 // CHECK-32-EX-NEXT:    [[TMP14:%.*]] = trunc i32 [[TMP13]] to i16
1526 // CHECK-32-EX-NEXT:    [[TMP15:%.*]] = call i64 @__kmpc_shuffle_int64(i64 [[TMP12]], i16 [[TMP6]], i16 [[TMP14]])
1527 // CHECK-32-EX-NEXT:    store i64 [[TMP15]], ptr [[DOTOMP_REDUCTION_ELEMENT]], align 8
1528 // CHECK-32-EX-NEXT:    [[TMP16:%.*]] = getelementptr i64, ptr [[TMP9]], i32 1
1529 // CHECK-32-EX-NEXT:    [[TMP17:%.*]] = getelementptr i64, ptr [[DOTOMP_REDUCTION_ELEMENT]], i32 1
1530 // CHECK-32-EX-NEXT:    store ptr [[DOTOMP_REDUCTION_ELEMENT]], ptr [[TMP10]], align 4
1531 // CHECK-32-EX-NEXT:    [[TMP18:%.*]] = icmp eq i16 [[TMP7]], 0
1532 // CHECK-32-EX-NEXT:    [[TMP19:%.*]] = icmp eq i16 [[TMP7]], 1
1533 // CHECK-32-EX-NEXT:    [[TMP20:%.*]] = icmp ult i16 [[TMP5]], [[TMP6]]
1534 // CHECK-32-EX-NEXT:    [[TMP21:%.*]] = and i1 [[TMP19]], [[TMP20]]
1535 // CHECK-32-EX-NEXT:    [[TMP22:%.*]] = icmp eq i16 [[TMP7]], 2
1536 // CHECK-32-EX-NEXT:    [[TMP23:%.*]] = and i16 [[TMP5]], 1
1537 // CHECK-32-EX-NEXT:    [[TMP24:%.*]] = icmp eq i16 [[TMP23]], 0
1538 // CHECK-32-EX-NEXT:    [[TMP25:%.*]] = and i1 [[TMP22]], [[TMP24]]
1539 // CHECK-32-EX-NEXT:    [[TMP26:%.*]] = icmp sgt i16 [[TMP6]], 0
1540 // CHECK-32-EX-NEXT:    [[TMP27:%.*]] = and i1 [[TMP25]], [[TMP26]]
1541 // CHECK-32-EX-NEXT:    [[TMP28:%.*]] = or i1 [[TMP18]], [[TMP21]]
1542 // CHECK-32-EX-NEXT:    [[TMP29:%.*]] = or i1 [[TMP28]], [[TMP27]]
1543 // CHECK-32-EX-NEXT:    br i1 [[TMP29]], label [[THEN:%.*]], label [[ELSE:%.*]]
1544 // CHECK-32-EX:       then:
1545 // CHECK-32-EX-NEXT:    call void @"{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l24_omp_outlined_omp$reduction$reduction_func"(ptr [[TMP4]], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]]) #[[ATTR3:[0-9]+]]
1546 // CHECK-32-EX-NEXT:    br label [[IFCONT:%.*]]
1547 // CHECK-32-EX:       else:
1548 // CHECK-32-EX-NEXT:    br label [[IFCONT]]
1549 // CHECK-32-EX:       ifcont:
1550 // CHECK-32-EX-NEXT:    [[TMP30:%.*]] = icmp eq i16 [[TMP7]], 1
1551 // CHECK-32-EX-NEXT:    [[TMP31:%.*]] = icmp uge i16 [[TMP5]], [[TMP6]]
1552 // CHECK-32-EX-NEXT:    [[TMP32:%.*]] = and i1 [[TMP30]], [[TMP31]]
1553 // CHECK-32-EX-NEXT:    br i1 [[TMP32]], label [[THEN4:%.*]], label [[ELSE5:%.*]]
1554 // CHECK-32-EX:       then4:
1555 // CHECK-32-EX-NEXT:    [[TMP33:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0
1556 // CHECK-32-EX-NEXT:    [[TMP34:%.*]] = load ptr, ptr [[TMP33]], align 4
1557 // CHECK-32-EX-NEXT:    [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP4]], i32 0, i32 0
1558 // CHECK-32-EX-NEXT:    [[TMP36:%.*]] = load ptr, ptr [[TMP35]], align 4
1559 // CHECK-32-EX-NEXT:    [[TMP37:%.*]] = load double, ptr [[TMP34]], align 8
1560 // CHECK-32-EX-NEXT:    store double [[TMP37]], ptr [[TMP36]], align 8
1561 // CHECK-32-EX-NEXT:    br label [[IFCONT6:%.*]]
1562 // CHECK-32-EX:       else5:
1563 // CHECK-32-EX-NEXT:    br label [[IFCONT6]]
1564 // CHECK-32-EX:       ifcont6:
1565 // CHECK-32-EX-NEXT:    ret void
1566 //
1567 //
1568 // CHECK-32-EX-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func
1569 // CHECK-32-EX-SAME: (ptr noundef [[TMP0:%.*]], i32 noundef [[TMP1:%.*]]) #[[ATTR2]] {
1570 // CHECK-32-EX-NEXT:  entry:
1571 // CHECK-32-EX-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 4
1572 // CHECK-32-EX-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
1573 // CHECK-32-EX-NEXT:    [[DOTCNT_ADDR:%.*]] = alloca i32, align 4
1574 // CHECK-32-EX-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 4
1575 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], ptr [[DOTADDR1]], align 4
1576 // CHECK-32-EX-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
1577 // CHECK-32-EX-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
1578 // CHECK-32-EX-NEXT:    [[NVPTX_LANE_ID:%.*]] = and i32 [[TMP4]], 31
1579 // CHECK-32-EX-NEXT:    [[TMP5:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
1580 // CHECK-32-EX-NEXT:    [[NVPTX_WARP_ID:%.*]] = ashr i32 [[TMP5]], 5
1581 // CHECK-32-EX-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[DOTADDR]], align 4
1582 // CHECK-32-EX-NEXT:    store i32 0, ptr [[DOTCNT_ADDR]], align 4
1583 // CHECK-32-EX-NEXT:    br label [[PRECOND:%.*]]
1584 // CHECK-32-EX:       precond:
1585 // CHECK-32-EX-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCNT_ADDR]], align 4
1586 // CHECK-32-EX-NEXT:    [[TMP8:%.*]] = icmp ult i32 [[TMP7]], 2
1587 // CHECK-32-EX-NEXT:    br i1 [[TMP8]], label [[BODY:%.*]], label [[EXIT:%.*]]
1588 // CHECK-32-EX:       body:
1589 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
1590 // CHECK-32-EX-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2:[0-9]+]], i32 [[TMP2]])
1591 // CHECK-32-EX-NEXT:    [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0
1592 // CHECK-32-EX-NEXT:    br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]]
1593 // CHECK-32-EX:       then:
1594 // CHECK-32-EX-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP6]], i32 0, i32 0
1595 // CHECK-32-EX-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[TMP9]], align 4
1596 // CHECK-32-EX-NEXT:    [[TMP11:%.*]] = getelementptr i32, ptr [[TMP10]], i32 [[TMP7]]
1597 // CHECK-32-EX-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]]
1598 // CHECK-32-EX-NEXT:    [[TMP13:%.*]] = load i32, ptr [[TMP11]], align 4
1599 // CHECK-32-EX-NEXT:    store volatile i32 [[TMP13]], ptr addrspace(3) [[TMP12]], align 4
1600 // CHECK-32-EX-NEXT:    br label [[IFCONT:%.*]]
1601 // CHECK-32-EX:       else:
1602 // CHECK-32-EX-NEXT:    br label [[IFCONT]]
1603 // CHECK-32-EX:       ifcont:
1604 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
1605 // CHECK-32-EX-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP2]])
1606 // CHECK-32-EX-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTADDR1]], align 4
1607 // CHECK-32-EX-NEXT:    [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[TMP3]], [[TMP14]]
1608 // CHECK-32-EX-NEXT:    br i1 [[IS_ACTIVE_THREAD]], label [[THEN2:%.*]], label [[ELSE3:%.*]]
1609 // CHECK-32-EX:       then3:
1610 // CHECK-32-EX-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[TMP3]]
1611 // CHECK-32-EX-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP6]], i32 0, i32 0
1612 // CHECK-32-EX-NEXT:    [[TMP17:%.*]] = load ptr, ptr [[TMP16]], align 4
1613 // CHECK-32-EX-NEXT:    [[TMP18:%.*]] = getelementptr i32, ptr [[TMP17]], i32 [[TMP7]]
1614 // CHECK-32-EX-NEXT:    [[TMP19:%.*]] = load volatile i32, ptr addrspace(3) [[TMP15]], align 4
1615 // CHECK-32-EX-NEXT:    store i32 [[TMP19]], ptr [[TMP18]], align 4
1616 // CHECK-32-EX-NEXT:    br label [[IFCONT4:%.*]]
1617 // CHECK-32-EX:       else4:
1618 // CHECK-32-EX-NEXT:    br label [[IFCONT4]]
1619 // CHECK-32-EX:       ifcont5:
1620 // CHECK-32-EX-NEXT:    [[TMP20:%.*]] = add nsw i32 [[TMP7]], 1
1621 // CHECK-32-EX-NEXT:    store i32 [[TMP20]], ptr [[DOTCNT_ADDR]], align 4
1622 // CHECK-32-EX-NEXT:    br label [[PRECOND]]
1623 // CHECK-32-EX:       exit:
1624 // CHECK-32-EX-NEXT:    ret void
1625 //
1626 //
1627 // CHECK-32-EX-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29
1628 // CHECK-32-EX-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], ptr noundef nonnull align 1 dereferenceable(1) [[C:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[D:%.*]]) #[[ATTR0]] {
1629 // CHECK-32-EX-NEXT:  entry:
1630 // CHECK-32-EX-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 4
1631 // CHECK-32-EX-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
1632 // CHECK-32-EX-NEXT:    [[D_ADDR:%.*]] = alloca ptr, align 4
1633 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [2 x ptr], align 4
1634 // CHECK-32-EX-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 4
1635 // CHECK-32-EX-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
1636 // CHECK-32-EX-NEXT:    store ptr [[D]], ptr [[D_ADDR]], align 4
1637 // CHECK-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 4
1638 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[D_ADDR]], align 4
1639 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29_kernel_environment, ptr [[DYN_PTR]])
1640 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP2]], -1
1641 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
1642 // CHECK-32-EX:       user_code.entry:
1643 // CHECK-32-EX-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
1644 // CHECK-32-EX-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [2 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
1645 // CHECK-32-EX-NEXT:    store ptr [[TMP0]], ptr [[TMP4]], align 4
1646 // CHECK-32-EX-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [2 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
1647 // CHECK-32-EX-NEXT:    store ptr [[TMP1]], ptr [[TMP5]], align 4
1648 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP3]], i32 1, i32 -1, i32 -1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29_omp_outlined, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 2)
1649 // CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
1650 // CHECK-32-EX-NEXT:    ret void
1651 // CHECK-32-EX:       worker.exit:
1652 // CHECK-32-EX-NEXT:    ret void
1653 //
1654 //
1655 // CHECK-32-EX-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29_omp_outlined
1656 // CHECK-32-EX-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 1 dereferenceable(1) [[C:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[D:%.*]]) #[[ATTR1]] {
1657 // CHECK-32-EX-NEXT:  entry:
1658 // CHECK-32-EX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
1659 // CHECK-32-EX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
1660 // CHECK-32-EX-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
1661 // CHECK-32-EX-NEXT:    [[D_ADDR:%.*]] = alloca ptr, align 4
1662 // CHECK-32-EX-NEXT:    [[C1:%.*]] = alloca i8, align 1
1663 // CHECK-32-EX-NEXT:    [[D2:%.*]] = alloca float, align 4
1664 // CHECK-32-EX-NEXT:    [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x ptr], align 4
1665 // CHECK-32-EX-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
1666 // CHECK-32-EX-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
1667 // CHECK-32-EX-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
1668 // CHECK-32-EX-NEXT:    store ptr [[D]], ptr [[D_ADDR]], align 4
1669 // CHECK-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 4
1670 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[D_ADDR]], align 4
1671 // CHECK-32-EX-NEXT:    store i8 0, ptr [[C1]], align 1
1672 // CHECK-32-EX-NEXT:    store float 1.000000e+00, ptr [[D2]], align 4
1673 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = load i8, ptr [[C1]], align 1
1674 // CHECK-32-EX-NEXT:    [[CONV:%.*]] = sext i8 [[TMP2]] to i32
1675 // CHECK-32-EX-NEXT:    [[XOR:%.*]] = xor i32 [[CONV]], 2
1676 // CHECK-32-EX-NEXT:    [[CONV3:%.*]] = trunc i32 [[XOR]] to i8
1677 // CHECK-32-EX-NEXT:    store i8 [[CONV3]], ptr [[C1]], align 1
1678 // CHECK-32-EX-NEXT:    [[TMP3:%.*]] = load float, ptr [[D2]], align 4
1679 // CHECK-32-EX-NEXT:    [[MUL:%.*]] = fmul float [[TMP3]], 3.300000e+01
1680 // CHECK-32-EX-NEXT:    store float [[MUL]], ptr [[D2]], align 4
1681 // CHECK-32-EX-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0
1682 // CHECK-32-EX-NEXT:    store ptr [[C1]], ptr [[TMP4]], align 4
1683 // CHECK-32-EX-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1
1684 // CHECK-32-EX-NEXT:    store ptr [[D2]], ptr [[TMP5]], align 4
1685 // CHECK-32-EX-NEXT:    [[TMP6:%.*]] = call i32 @__kmpc_nvptx_parallel_reduce_nowait_v2(ptr @[[GLOB1]], i64 8, ptr [[DOTOMP_REDUCTION_RED_LIST]], ptr @_omp_reduction_shuffle_and_reduce_func1, ptr @_omp_reduction_inter_warp_copy_func2)
1686 // CHECK-32-EX-NEXT:    [[TMP7:%.*]] = icmp eq i32 [[TMP6]], 1
1687 // CHECK-32-EX-NEXT:    br i1 [[TMP7]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]]
1688 // CHECK-32-EX:       .omp.reduction.then:
1689 // CHECK-32-EX-NEXT:    [[TMP8:%.*]] = load i8, ptr [[TMP0]], align 1
1690 // CHECK-32-EX-NEXT:    [[CONV4:%.*]] = sext i8 [[TMP8]] to i32
1691 // CHECK-32-EX-NEXT:    [[TMP9:%.*]] = load i8, ptr [[C1]], align 1
1692 // CHECK-32-EX-NEXT:    [[CONV5:%.*]] = sext i8 [[TMP9]] to i32
1693 // CHECK-32-EX-NEXT:    [[XOR6:%.*]] = xor i32 [[CONV4]], [[CONV5]]
1694 // CHECK-32-EX-NEXT:    [[CONV7:%.*]] = trunc i32 [[XOR6]] to i8
1695 // CHECK-32-EX-NEXT:    store i8 [[CONV7]], ptr [[TMP0]], align 1
1696 // CHECK-32-EX-NEXT:    [[TMP10:%.*]] = load float, ptr [[TMP1]], align 4
1697 // CHECK-32-EX-NEXT:    [[TMP11:%.*]] = load float, ptr [[D2]], align 4
1698 // CHECK-32-EX-NEXT:    [[MUL8:%.*]] = fmul float [[TMP10]], [[TMP11]]
1699 // CHECK-32-EX-NEXT:    store float [[MUL8]], ptr [[TMP1]], align 4
1700 // CHECK-32-EX-NEXT:    br label [[DOTOMP_REDUCTION_DONE]]
1701 // CHECK-32-EX:       .omp.reduction.done:
1702 // CHECK-32-EX-NEXT:    ret void
1703 //
1704 //
1705 // CHECK-32-EX-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func1
1706 // CHECK-32-EX-SAME: (ptr noundef [[TMP0:%.*]], i16 noundef signext [[TMP1:%.*]], i16 noundef signext [[TMP2:%.*]], i16 noundef signext [[TMP3:%.*]]) #[[ATTR2]] {
1707 // CHECK-32-EX-NEXT:  entry:
1708 // CHECK-32-EX-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 4
1709 // CHECK-32-EX-NEXT:    [[DOTADDR1:%.*]] = alloca i16, align 2
1710 // CHECK-32-EX-NEXT:    [[DOTADDR2:%.*]] = alloca i16, align 2
1711 // CHECK-32-EX-NEXT:    [[DOTADDR3:%.*]] = alloca i16, align 2
1712 // CHECK-32-EX-NEXT:    [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [2 x ptr], align 4
1713 // CHECK-32-EX-NEXT:    [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca i8, align 1
1714 // CHECK-32-EX-NEXT:    [[DOTOMP_REDUCTION_ELEMENT4:%.*]] = alloca float, align 4
1715 // CHECK-32-EX-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 4
1716 // CHECK-32-EX-NEXT:    store i16 [[TMP1]], ptr [[DOTADDR1]], align 2
1717 // CHECK-32-EX-NEXT:    store i16 [[TMP2]], ptr [[DOTADDR2]], align 2
1718 // CHECK-32-EX-NEXT:    store i16 [[TMP3]], ptr [[DOTADDR3]], align 2
1719 // CHECK-32-EX-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR]], align 4
1720 // CHECK-32-EX-NEXT:    [[TMP5:%.*]] = load i16, ptr [[DOTADDR1]], align 2
1721 // CHECK-32-EX-NEXT:    [[TMP6:%.*]] = load i16, ptr [[DOTADDR2]], align 2
1722 // CHECK-32-EX-NEXT:    [[TMP7:%.*]] = load i16, ptr [[DOTADDR3]], align 2
1723 // CHECK-32-EX-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP4]], i32 0, i32 0
1724 // CHECK-32-EX-NEXT:    [[TMP9:%.*]] = load ptr, ptr [[TMP8]], align 4
1725 // CHECK-32-EX-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0
1726 // CHECK-32-EX-NEXT:    [[TMP11:%.*]] = getelementptr i8, ptr [[TMP9]], i32 1
1727 // CHECK-32-EX-NEXT:    [[TMP12:%.*]] = load i8, ptr [[TMP9]], align 1
1728 // CHECK-32-EX-NEXT:    [[TMP13:%.*]] = sext i8 [[TMP12]] to i32
1729 // CHECK-32-EX-NEXT:    [[TMP14:%.*]] = call i32 @__kmpc_get_warp_size()
1730 // CHECK-32-EX-NEXT:    [[TMP15:%.*]] = trunc i32 [[TMP14]] to i16
1731 // CHECK-32-EX-NEXT:    [[TMP16:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP13]], i16 [[TMP6]], i16 [[TMP15]])
1732 // CHECK-32-EX-NEXT:    [[TMP17:%.*]] = trunc i32 [[TMP16]] to i8
1733 // CHECK-32-EX-NEXT:    store i8 [[TMP17]], ptr [[DOTOMP_REDUCTION_ELEMENT]], align 1
1734 // CHECK-32-EX-NEXT:    [[TMP18:%.*]] = getelementptr i8, ptr [[TMP9]], i32 1
1735 // CHECK-32-EX-NEXT:    [[TMP19:%.*]] = getelementptr i8, ptr [[DOTOMP_REDUCTION_ELEMENT]], i32 1
1736 // CHECK-32-EX-NEXT:    store ptr [[DOTOMP_REDUCTION_ELEMENT]], ptr [[TMP10]], align 4
1737 // CHECK-32-EX-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP4]], i32 0, i32 1
1738 // CHECK-32-EX-NEXT:    [[TMP21:%.*]] = load ptr, ptr [[TMP20]], align 4
1739 // CHECK-32-EX-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1
1740 // CHECK-32-EX-NEXT:    [[TMP23:%.*]] = getelementptr float, ptr [[TMP21]], i32 1
1741 // CHECK-32-EX-NEXT:    [[TMP24:%.*]] = load i32, ptr [[TMP21]], align 4
1742 // CHECK-32-EX-NEXT:    [[TMP25:%.*]] = call i32 @__kmpc_get_warp_size()
1743 // CHECK-32-EX-NEXT:    [[TMP26:%.*]] = trunc i32 [[TMP25]] to i16
1744 // CHECK-32-EX-NEXT:    [[TMP27:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP24]], i16 [[TMP6]], i16 [[TMP26]])
1745 // CHECK-32-EX-NEXT:    store i32 [[TMP27]], ptr [[DOTOMP_REDUCTION_ELEMENT4]], align 4
1746 // CHECK-32-EX-NEXT:    [[TMP28:%.*]] = getelementptr i32, ptr [[TMP21]], i32 1
1747 // CHECK-32-EX-NEXT:    [[TMP29:%.*]] = getelementptr i32, ptr [[DOTOMP_REDUCTION_ELEMENT4]], i32 1
1748 // CHECK-32-EX-NEXT:    store ptr [[DOTOMP_REDUCTION_ELEMENT4]], ptr [[TMP22]], align 4
1749 // CHECK-32-EX-NEXT:    [[TMP30:%.*]] = icmp eq i16 [[TMP7]], 0
1750 // CHECK-32-EX-NEXT:    [[TMP31:%.*]] = icmp eq i16 [[TMP7]], 1
1751 // CHECK-32-EX-NEXT:    [[TMP32:%.*]] = icmp ult i16 [[TMP5]], [[TMP6]]
1752 // CHECK-32-EX-NEXT:    [[TMP33:%.*]] = and i1 [[TMP31]], [[TMP32]]
1753 // CHECK-32-EX-NEXT:    [[TMP34:%.*]] = icmp eq i16 [[TMP7]], 2
1754 // CHECK-32-EX-NEXT:    [[TMP35:%.*]] = and i16 [[TMP5]], 1
1755 // CHECK-32-EX-NEXT:    [[TMP36:%.*]] = icmp eq i16 [[TMP35]], 0
1756 // CHECK-32-EX-NEXT:    [[TMP37:%.*]] = and i1 [[TMP34]], [[TMP36]]
1757 // CHECK-32-EX-NEXT:    [[TMP38:%.*]] = icmp sgt i16 [[TMP6]], 0
1758 // CHECK-32-EX-NEXT:    [[TMP39:%.*]] = and i1 [[TMP37]], [[TMP38]]
1759 // CHECK-32-EX-NEXT:    [[TMP40:%.*]] = or i1 [[TMP30]], [[TMP33]]
1760 // CHECK-32-EX-NEXT:    [[TMP41:%.*]] = or i1 [[TMP40]], [[TMP39]]
1761 // CHECK-32-EX-NEXT:    br i1 [[TMP41]], label [[THEN:%.*]], label [[ELSE:%.*]]
1762 // CHECK-32-EX:       then:
1763 // CHECK-32-EX-NEXT:    call void @"{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29_omp_outlined_omp$reduction$reduction_func"(ptr [[TMP4]], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]]) #[[ATTR3]]
1764 // CHECK-32-EX-NEXT:    br label [[IFCONT:%.*]]
1765 // CHECK-32-EX:       else:
1766 // CHECK-32-EX-NEXT:    br label [[IFCONT]]
1767 // CHECK-32-EX:       ifcont:
1768 // CHECK-32-EX-NEXT:    [[TMP42:%.*]] = icmp eq i16 [[TMP7]], 1
1769 // CHECK-32-EX-NEXT:    [[TMP43:%.*]] = icmp uge i16 [[TMP5]], [[TMP6]]
1770 // CHECK-32-EX-NEXT:    [[TMP44:%.*]] = and i1 [[TMP42]], [[TMP43]]
1771 // CHECK-32-EX-NEXT:    br i1 [[TMP44]], label [[THEN5:%.*]], label [[ELSE6:%.*]]
1772 // CHECK-32-EX:       then5:
1773 // CHECK-32-EX-NEXT:    [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0
1774 // CHECK-32-EX-NEXT:    [[TMP46:%.*]] = load ptr, ptr [[TMP45]], align 4
1775 // CHECK-32-EX-NEXT:    [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP4]], i32 0, i32 0
1776 // CHECK-32-EX-NEXT:    [[TMP48:%.*]] = load ptr, ptr [[TMP47]], align 4
1777 // CHECK-32-EX-NEXT:    [[TMP49:%.*]] = load i8, ptr [[TMP46]], align 1
1778 // CHECK-32-EX-NEXT:    store i8 [[TMP49]], ptr [[TMP48]], align 1
1779 // CHECK-32-EX-NEXT:    [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1
1780 // CHECK-32-EX-NEXT:    [[TMP51:%.*]] = load ptr, ptr [[TMP50]], align 4
1781 // CHECK-32-EX-NEXT:    [[TMP52:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP4]], i32 0, i32 1
1782 // CHECK-32-EX-NEXT:    [[TMP53:%.*]] = load ptr, ptr [[TMP52]], align 4
1783 // CHECK-32-EX-NEXT:    [[TMP54:%.*]] = load float, ptr [[TMP51]], align 4
1784 // CHECK-32-EX-NEXT:    store float [[TMP54]], ptr [[TMP53]], align 4
1785 // CHECK-32-EX-NEXT:    br label [[IFCONT7:%.*]]
1786 // CHECK-32-EX:       else6:
1787 // CHECK-32-EX-NEXT:    br label [[IFCONT7]]
1788 // CHECK-32-EX:       ifcont7:
1789 // CHECK-32-EX-NEXT:    ret void
1790 //
1791 //
1792 // CHECK-32-EX-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func2
1793 // CHECK-32-EX-SAME: (ptr noundef [[TMP0:%.*]], i32 noundef [[TMP1:%.*]]) #[[ATTR2]] {
1794 // CHECK-32-EX-NEXT:  entry:
1795 // CHECK-32-EX-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 4
1796 // CHECK-32-EX-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
1797 // CHECK-32-EX-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 4
1798 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], ptr [[DOTADDR1]], align 4
1799 // CHECK-32-EX-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
1800 // CHECK-32-EX-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
1801 // CHECK-32-EX-NEXT:    [[NVPTX_LANE_ID:%.*]] = and i32 [[TMP4]], 31
1802 // CHECK-32-EX-NEXT:    [[TMP5:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
1803 // CHECK-32-EX-NEXT:    [[NVPTX_WARP_ID:%.*]] = ashr i32 [[TMP5]], 5
1804 // CHECK-32-EX-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[DOTADDR]], align 4
1805 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
1806 // CHECK-32-EX-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP2]])
1807 // CHECK-32-EX-NEXT:    [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0
1808 // CHECK-32-EX-NEXT:    br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]]
1809 // CHECK-32-EX:       then:
1810 // CHECK-32-EX-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP6]], i32 0, i32 0
1811 // CHECK-32-EX-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[TMP7]], align 4
1812 // CHECK-32-EX-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]]
1813 // CHECK-32-EX-NEXT:    [[TMP10:%.*]] = load i8, ptr [[TMP8]], align 1
1814 // CHECK-32-EX-NEXT:    store volatile i8 [[TMP10]], ptr addrspace(3) [[TMP9]], align 1
1815 // CHECK-32-EX-NEXT:    br label [[IFCONT:%.*]]
1816 // CHECK-32-EX:       else:
1817 // CHECK-32-EX-NEXT:    br label [[IFCONT]]
1818 // CHECK-32-EX:       ifcont:
1819 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
1820 // CHECK-32-EX-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP2]])
1821 // CHECK-32-EX-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTADDR1]], align 4
1822 // CHECK-32-EX-NEXT:    [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[TMP3]], [[TMP11]]
1823 // CHECK-32-EX-NEXT:    br i1 [[IS_ACTIVE_THREAD]], label [[THEN2:%.*]], label [[ELSE3:%.*]]
1824 // CHECK-32-EX:       then3:
1825 // CHECK-32-EX-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[TMP3]]
1826 // CHECK-32-EX-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP6]], i32 0, i32 0
1827 // CHECK-32-EX-NEXT:    [[TMP14:%.*]] = load ptr, ptr [[TMP13]], align 4
1828 // CHECK-32-EX-NEXT:    [[TMP15:%.*]] = load volatile i8, ptr addrspace(3) [[TMP12]], align 1
1829 // CHECK-32-EX-NEXT:    store i8 [[TMP15]], ptr [[TMP14]], align 1
1830 // CHECK-32-EX-NEXT:    br label [[IFCONT4:%.*]]
1831 // CHECK-32-EX:       else4:
1832 // CHECK-32-EX-NEXT:    br label [[IFCONT4]]
1833 // CHECK-32-EX:       ifcont5:
1834 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
1835 // CHECK-32-EX-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP2]])
1836 // CHECK-32-EX-NEXT:    [[WARP_MASTER5:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0
1837 // CHECK-32-EX-NEXT:    br i1 [[WARP_MASTER5]], label [[THEN6:%.*]], label [[ELSE7:%.*]]
1838 // CHECK-32-EX:       then8:
1839 // CHECK-32-EX-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP6]], i32 0, i32 1
1840 // CHECK-32-EX-NEXT:    [[TMP17:%.*]] = load ptr, ptr [[TMP16]], align 4
1841 // CHECK-32-EX-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]]
1842 // CHECK-32-EX-NEXT:    [[TMP19:%.*]] = load i32, ptr [[TMP17]], align 4
1843 // CHECK-32-EX-NEXT:    store volatile i32 [[TMP19]], ptr addrspace(3) [[TMP18]], align 4
1844 // CHECK-32-EX-NEXT:    br label [[IFCONT8:%.*]]
1845 // CHECK-32-EX:       else9:
1846 // CHECK-32-EX-NEXT:    br label [[IFCONT8]]
1847 // CHECK-32-EX:       ifcont10:
1848 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
1849 // CHECK-32-EX-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP2]])
1850 // CHECK-32-EX-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTADDR1]], align 4
1851 // CHECK-32-EX-NEXT:    [[IS_ACTIVE_THREAD9:%.*]] = icmp ult i32 [[TMP3]], [[TMP20]]
1852 // CHECK-32-EX-NEXT:    br i1 [[IS_ACTIVE_THREAD9]], label [[THEN10:%.*]], label [[ELSE11:%.*]]
1853 // CHECK-32-EX:       then13:
1854 // CHECK-32-EX-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[TMP3]]
1855 // CHECK-32-EX-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP6]], i32 0, i32 1
1856 // CHECK-32-EX-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP22]], align 4
1857 // CHECK-32-EX-NEXT:    [[TMP24:%.*]] = load volatile i32, ptr addrspace(3) [[TMP21]], align 4
1858 // CHECK-32-EX-NEXT:    store i32 [[TMP24]], ptr [[TMP23]], align 4
1859 // CHECK-32-EX-NEXT:    br label [[IFCONT12:%.*]]
1860 // CHECK-32-EX:       else14:
1861 // CHECK-32-EX-NEXT:    br label [[IFCONT12]]
1862 // CHECK-32-EX:       ifcont15:
1863 // CHECK-32-EX-NEXT:    ret void
1864 //
1865 //
1866 // CHECK-32-EX-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l35
1867 // CHECK-32-EX-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 2 dereferenceable(2) [[B:%.*]]) #[[ATTR0]] {
1868 // CHECK-32-EX-NEXT:  entry:
1869 // CHECK-32-EX-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 4
1870 // CHECK-32-EX-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
1871 // CHECK-32-EX-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
1872 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [2 x ptr], align 4
1873 // CHECK-32-EX-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 4
1874 // CHECK-32-EX-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
1875 // CHECK-32-EX-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
1876 // CHECK-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
1877 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[B_ADDR]], align 4
1878 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l35_kernel_environment, ptr [[DYN_PTR]])
1879 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP2]], -1
1880 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
1881 // CHECK-32-EX:       user_code.entry:
1882 // CHECK-32-EX-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
1883 // CHECK-32-EX-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [2 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
1884 // CHECK-32-EX-NEXT:    store ptr [[TMP0]], ptr [[TMP4]], align 4
1885 // CHECK-32-EX-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [2 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
1886 // CHECK-32-EX-NEXT:    store ptr [[TMP1]], ptr [[TMP5]], align 4
1887 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP3]], i32 1, i32 -1, i32 -1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l35_omp_outlined, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 2)
1888 // CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
1889 // CHECK-32-EX-NEXT:    ret void
1890 // CHECK-32-EX:       worker.exit:
1891 // CHECK-32-EX-NEXT:    ret void
1892 //
1893 //
1894 // CHECK-32-EX-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l35_omp_outlined
1895 // CHECK-32-EX-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 2 dereferenceable(2) [[B:%.*]]) #[[ATTR1]] {
1896 // CHECK-32-EX-NEXT:  entry:
1897 // CHECK-32-EX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
1898 // CHECK-32-EX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
1899 // CHECK-32-EX-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
1900 // CHECK-32-EX-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
1901 // CHECK-32-EX-NEXT:    [[A1:%.*]] = alloca i32, align 4
1902 // CHECK-32-EX-NEXT:    [[B2:%.*]] = alloca i16, align 2
1903 // CHECK-32-EX-NEXT:    [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [2 x ptr], align 4
1904 // CHECK-32-EX-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
1905 // CHECK-32-EX-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
1906 // CHECK-32-EX-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
1907 // CHECK-32-EX-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
1908 // CHECK-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
1909 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[B_ADDR]], align 4
1910 // CHECK-32-EX-NEXT:    store i32 0, ptr [[A1]], align 4
1911 // CHECK-32-EX-NEXT:    store i16 -32768, ptr [[B2]], align 2
1912 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A1]], align 4
1913 // CHECK-32-EX-NEXT:    [[OR:%.*]] = or i32 [[TMP2]], 1
1914 // CHECK-32-EX-NEXT:    store i32 [[OR]], ptr [[A1]], align 4
1915 // CHECK-32-EX-NEXT:    [[TMP3:%.*]] = load i16, ptr [[B2]], align 2
1916 // CHECK-32-EX-NEXT:    [[CONV:%.*]] = sext i16 [[TMP3]] to i32
1917 // CHECK-32-EX-NEXT:    [[CMP:%.*]] = icmp sgt i32 99, [[CONV]]
1918 // CHECK-32-EX-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
1919 // CHECK-32-EX:       cond.true:
1920 // CHECK-32-EX-NEXT:    br label [[COND_END:%.*]]
1921 // CHECK-32-EX:       cond.false:
1922 // CHECK-32-EX-NEXT:    [[TMP4:%.*]] = load i16, ptr [[B2]], align 2
1923 // CHECK-32-EX-NEXT:    [[CONV3:%.*]] = sext i16 [[TMP4]] to i32
1924 // CHECK-32-EX-NEXT:    br label [[COND_END]]
1925 // CHECK-32-EX:       cond.end:
1926 // CHECK-32-EX-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[CONV3]], [[COND_FALSE]] ]
1927 // CHECK-32-EX-NEXT:    [[CONV4:%.*]] = trunc i32 [[COND]] to i16
1928 // CHECK-32-EX-NEXT:    store i16 [[CONV4]], ptr [[B2]], align 2
1929 // CHECK-32-EX-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0
1930 // CHECK-32-EX-NEXT:    store ptr [[A1]], ptr [[TMP5]], align 4
1931 // CHECK-32-EX-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 1
1932 // CHECK-32-EX-NEXT:    store ptr [[B2]], ptr [[TMP6]], align 4
1933 // CHECK-32-EX-NEXT:    [[TMP7:%.*]] = call i32 @__kmpc_nvptx_parallel_reduce_nowait_v2(ptr @[[GLOB1]], i64 8, ptr [[DOTOMP_REDUCTION_RED_LIST]], ptr @_omp_reduction_shuffle_and_reduce_func3, ptr @_omp_reduction_inter_warp_copy_func4)
1934 // CHECK-32-EX-NEXT:    [[TMP8:%.*]] = icmp eq i32 [[TMP7]], 1
1935 // CHECK-32-EX-NEXT:    br i1 [[TMP8]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]]
1936 // CHECK-32-EX:       .omp.reduction.then:
1937 // CHECK-32-EX-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP0]], align 4
1938 // CHECK-32-EX-NEXT:    [[TMP10:%.*]] = load i32, ptr [[A1]], align 4
1939 // CHECK-32-EX-NEXT:    [[OR5:%.*]] = or i32 [[TMP9]], [[TMP10]]
1940 // CHECK-32-EX-NEXT:    store i32 [[OR5]], ptr [[TMP0]], align 4
1941 // CHECK-32-EX-NEXT:    [[TMP11:%.*]] = load i16, ptr [[TMP1]], align 2
1942 // CHECK-32-EX-NEXT:    [[CONV6:%.*]] = sext i16 [[TMP11]] to i32
1943 // CHECK-32-EX-NEXT:    [[TMP12:%.*]] = load i16, ptr [[B2]], align 2
1944 // CHECK-32-EX-NEXT:    [[CONV7:%.*]] = sext i16 [[TMP12]] to i32
1945 // CHECK-32-EX-NEXT:    [[CMP8:%.*]] = icmp sgt i32 [[CONV6]], [[CONV7]]
1946 // CHECK-32-EX-NEXT:    br i1 [[CMP8]], label [[COND_TRUE9:%.*]], label [[COND_FALSE10:%.*]]
1947 // CHECK-32-EX:       cond.true9:
1948 // CHECK-32-EX-NEXT:    [[TMP13:%.*]] = load i16, ptr [[TMP1]], align 2
1949 // CHECK-32-EX-NEXT:    br label [[COND_END11:%.*]]
1950 // CHECK-32-EX:       cond.false10:
1951 // CHECK-32-EX-NEXT:    [[TMP14:%.*]] = load i16, ptr [[B2]], align 2
1952 // CHECK-32-EX-NEXT:    br label [[COND_END11]]
1953 // CHECK-32-EX:       cond.end11:
1954 // CHECK-32-EX-NEXT:    [[COND12:%.*]] = phi i16 [ [[TMP13]], [[COND_TRUE9]] ], [ [[TMP14]], [[COND_FALSE10]] ]
1955 // CHECK-32-EX-NEXT:    store i16 [[COND12]], ptr [[TMP1]], align 2
1956 // CHECK-32-EX-NEXT:    br label [[DOTOMP_REDUCTION_DONE]]
1957 // CHECK-32-EX:       .omp.reduction.done:
1958 // CHECK-32-EX-NEXT:    ret void
1959 //
1960 //
1961 // CHECK-32-EX-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func3
1962 // CHECK-32-EX-SAME: (ptr noundef [[TMP0:%.*]], i16 noundef signext [[TMP1:%.*]], i16 noundef signext [[TMP2:%.*]], i16 noundef signext [[TMP3:%.*]]) #[[ATTR2]] {
1963 // CHECK-32-EX-NEXT:  entry:
1964 // CHECK-32-EX-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 4
1965 // CHECK-32-EX-NEXT:    [[DOTADDR1:%.*]] = alloca i16, align 2
1966 // CHECK-32-EX-NEXT:    [[DOTADDR2:%.*]] = alloca i16, align 2
1967 // CHECK-32-EX-NEXT:    [[DOTADDR3:%.*]] = alloca i16, align 2
1968 // CHECK-32-EX-NEXT:    [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [2 x ptr], align 4
1969 // CHECK-32-EX-NEXT:    [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca i32, align 4
1970 // CHECK-32-EX-NEXT:    [[DOTOMP_REDUCTION_ELEMENT4:%.*]] = alloca i16, align 2
1971 // CHECK-32-EX-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 4
1972 // CHECK-32-EX-NEXT:    store i16 [[TMP1]], ptr [[DOTADDR1]], align 2
1973 // CHECK-32-EX-NEXT:    store i16 [[TMP2]], ptr [[DOTADDR2]], align 2
1974 // CHECK-32-EX-NEXT:    store i16 [[TMP3]], ptr [[DOTADDR3]], align 2
1975 // CHECK-32-EX-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR]], align 4
1976 // CHECK-32-EX-NEXT:    [[TMP5:%.*]] = load i16, ptr [[DOTADDR1]], align 2
1977 // CHECK-32-EX-NEXT:    [[TMP6:%.*]] = load i16, ptr [[DOTADDR2]], align 2
1978 // CHECK-32-EX-NEXT:    [[TMP7:%.*]] = load i16, ptr [[DOTADDR3]], align 2
1979 // CHECK-32-EX-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP4]], i32 0, i32 0
1980 // CHECK-32-EX-NEXT:    [[TMP9:%.*]] = load ptr, ptr [[TMP8]], align 4
1981 // CHECK-32-EX-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0
1982 // CHECK-32-EX-NEXT:    [[TMP11:%.*]] = getelementptr i32, ptr [[TMP9]], i32 1
1983 // CHECK-32-EX-NEXT:    [[TMP12:%.*]] = load i32, ptr [[TMP9]], align 4
1984 // CHECK-32-EX-NEXT:    [[TMP13:%.*]] = call i32 @__kmpc_get_warp_size()
1985 // CHECK-32-EX-NEXT:    [[TMP14:%.*]] = trunc i32 [[TMP13]] to i16
1986 // CHECK-32-EX-NEXT:    [[TMP15:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP12]], i16 [[TMP6]], i16 [[TMP14]])
1987 // CHECK-32-EX-NEXT:    store i32 [[TMP15]], ptr [[DOTOMP_REDUCTION_ELEMENT]], align 4
1988 // CHECK-32-EX-NEXT:    [[TMP16:%.*]] = getelementptr i32, ptr [[TMP9]], i32 1
1989 // CHECK-32-EX-NEXT:    [[TMP17:%.*]] = getelementptr i32, ptr [[DOTOMP_REDUCTION_ELEMENT]], i32 1
1990 // CHECK-32-EX-NEXT:    store ptr [[DOTOMP_REDUCTION_ELEMENT]], ptr [[TMP10]], align 4
1991 // CHECK-32-EX-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP4]], i32 0, i32 1
1992 // CHECK-32-EX-NEXT:    [[TMP19:%.*]] = load ptr, ptr [[TMP18]], align 4
1993 // CHECK-32-EX-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1
1994 // CHECK-32-EX-NEXT:    [[TMP21:%.*]] = getelementptr i16, ptr [[TMP19]], i32 1
1995 // CHECK-32-EX-NEXT:    [[TMP22:%.*]] = load i16, ptr [[TMP19]], align 2
1996 // CHECK-32-EX-NEXT:    [[TMP23:%.*]] = sext i16 [[TMP22]] to i32
1997 // CHECK-32-EX-NEXT:    [[TMP24:%.*]] = call i32 @__kmpc_get_warp_size()
1998 // CHECK-32-EX-NEXT:    [[TMP25:%.*]] = trunc i32 [[TMP24]] to i16
1999 // CHECK-32-EX-NEXT:    [[TMP26:%.*]] = call i32 @__kmpc_shuffle_int32(i32 [[TMP23]], i16 [[TMP6]], i16 [[TMP25]])
2000 // CHECK-32-EX-NEXT:    [[TMP27:%.*]] = trunc i32 [[TMP26]] to i16
2001 // CHECK-32-EX-NEXT:    store i16 [[TMP27]], ptr [[DOTOMP_REDUCTION_ELEMENT4]], align 2
2002 // CHECK-32-EX-NEXT:    [[TMP28:%.*]] = getelementptr i16, ptr [[TMP19]], i32 1
2003 // CHECK-32-EX-NEXT:    [[TMP29:%.*]] = getelementptr i16, ptr [[DOTOMP_REDUCTION_ELEMENT4]], i32 1
2004 // CHECK-32-EX-NEXT:    store ptr [[DOTOMP_REDUCTION_ELEMENT4]], ptr [[TMP20]], align 4
2005 // CHECK-32-EX-NEXT:    [[TMP30:%.*]] = icmp eq i16 [[TMP7]], 0
2006 // CHECK-32-EX-NEXT:    [[TMP31:%.*]] = icmp eq i16 [[TMP7]], 1
2007 // CHECK-32-EX-NEXT:    [[TMP32:%.*]] = icmp ult i16 [[TMP5]], [[TMP6]]
2008 // CHECK-32-EX-NEXT:    [[TMP33:%.*]] = and i1 [[TMP31]], [[TMP32]]
2009 // CHECK-32-EX-NEXT:    [[TMP34:%.*]] = icmp eq i16 [[TMP7]], 2
2010 // CHECK-32-EX-NEXT:    [[TMP35:%.*]] = and i16 [[TMP5]], 1
2011 // CHECK-32-EX-NEXT:    [[TMP36:%.*]] = icmp eq i16 [[TMP35]], 0
2012 // CHECK-32-EX-NEXT:    [[TMP37:%.*]] = and i1 [[TMP34]], [[TMP36]]
2013 // CHECK-32-EX-NEXT:    [[TMP38:%.*]] = icmp sgt i16 [[TMP6]], 0
2014 // CHECK-32-EX-NEXT:    [[TMP39:%.*]] = and i1 [[TMP37]], [[TMP38]]
2015 // CHECK-32-EX-NEXT:    [[TMP40:%.*]] = or i1 [[TMP30]], [[TMP33]]
2016 // CHECK-32-EX-NEXT:    [[TMP41:%.*]] = or i1 [[TMP40]], [[TMP39]]
2017 // CHECK-32-EX-NEXT:    br i1 [[TMP41]], label [[THEN:%.*]], label [[ELSE:%.*]]
2018 // CHECK-32-EX:       then:
2019 // CHECK-32-EX-NEXT:    call void @"{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l35_omp_outlined_omp$reduction$reduction_func"(ptr [[TMP4]], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]]) #[[ATTR3]]
2020 // CHECK-32-EX-NEXT:    br label [[IFCONT:%.*]]
2021 // CHECK-32-EX:       else:
2022 // CHECK-32-EX-NEXT:    br label [[IFCONT]]
2023 // CHECK-32-EX:       ifcont:
2024 // CHECK-32-EX-NEXT:    [[TMP42:%.*]] = icmp eq i16 [[TMP7]], 1
2025 // CHECK-32-EX-NEXT:    [[TMP43:%.*]] = icmp uge i16 [[TMP5]], [[TMP6]]
2026 // CHECK-32-EX-NEXT:    [[TMP44:%.*]] = and i1 [[TMP42]], [[TMP43]]
2027 // CHECK-32-EX-NEXT:    br i1 [[TMP44]], label [[THEN5:%.*]], label [[ELSE6:%.*]]
2028 // CHECK-32-EX:       then5:
2029 // CHECK-32-EX-NEXT:    [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 0
2030 // CHECK-32-EX-NEXT:    [[TMP46:%.*]] = load ptr, ptr [[TMP45]], align 4
2031 // CHECK-32-EX-NEXT:    [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP4]], i32 0, i32 0
2032 // CHECK-32-EX-NEXT:    [[TMP48:%.*]] = load ptr, ptr [[TMP47]], align 4
2033 // CHECK-32-EX-NEXT:    [[TMP49:%.*]] = load i32, ptr [[TMP46]], align 4
2034 // CHECK-32-EX-NEXT:    store i32 [[TMP49]], ptr [[TMP48]], align 4
2035 // CHECK-32-EX-NEXT:    [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i32 0, i32 1
2036 // CHECK-32-EX-NEXT:    [[TMP51:%.*]] = load ptr, ptr [[TMP50]], align 4
2037 // CHECK-32-EX-NEXT:    [[TMP52:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP4]], i32 0, i32 1
2038 // CHECK-32-EX-NEXT:    [[TMP53:%.*]] = load ptr, ptr [[TMP52]], align 4
2039 // CHECK-32-EX-NEXT:    [[TMP54:%.*]] = load i16, ptr [[TMP51]], align 2
2040 // CHECK-32-EX-NEXT:    store i16 [[TMP54]], ptr [[TMP53]], align 2
2041 // CHECK-32-EX-NEXT:    br label [[IFCONT7:%.*]]
2042 // CHECK-32-EX:       else6:
2043 // CHECK-32-EX-NEXT:    br label [[IFCONT7]]
2044 // CHECK-32-EX:       ifcont7:
2045 // CHECK-32-EX-NEXT:    ret void
2046 //
2047 //
2048 // CHECK-32-EX-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func4
2049 // CHECK-32-EX-SAME: (ptr noundef [[TMP0:%.*]], i32 noundef [[TMP1:%.*]]) #[[ATTR2]] {
2050 // CHECK-32-EX-NEXT:  entry:
2051 // CHECK-32-EX-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 4
2052 // CHECK-32-EX-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
2053 // CHECK-32-EX-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 4
2054 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], ptr [[DOTADDR1]], align 4
2055 // CHECK-32-EX-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
2056 // CHECK-32-EX-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
2057 // CHECK-32-EX-NEXT:    [[NVPTX_LANE_ID:%.*]] = and i32 [[TMP4]], 31
2058 // CHECK-32-EX-NEXT:    [[TMP5:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
2059 // CHECK-32-EX-NEXT:    [[NVPTX_WARP_ID:%.*]] = ashr i32 [[TMP5]], 5
2060 // CHECK-32-EX-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[DOTADDR]], align 4
2061 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
2062 // CHECK-32-EX-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP2]])
2063 // CHECK-32-EX-NEXT:    [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0
2064 // CHECK-32-EX-NEXT:    br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]]
2065 // CHECK-32-EX:       then:
2066 // CHECK-32-EX-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP6]], i32 0, i32 0
2067 // CHECK-32-EX-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[TMP7]], align 4
2068 // CHECK-32-EX-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]]
2069 // CHECK-32-EX-NEXT:    [[TMP10:%.*]] = load i32, ptr [[TMP8]], align 4
2070 // CHECK-32-EX-NEXT:    store volatile i32 [[TMP10]], ptr addrspace(3) [[TMP9]], align 4
2071 // CHECK-32-EX-NEXT:    br label [[IFCONT:%.*]]
2072 // CHECK-32-EX:       else:
2073 // CHECK-32-EX-NEXT:    br label [[IFCONT]]
2074 // CHECK-32-EX:       ifcont:
2075 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
2076 // CHECK-32-EX-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP2]])
2077 // CHECK-32-EX-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTADDR1]], align 4
2078 // CHECK-32-EX-NEXT:    [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[TMP3]], [[TMP11]]
2079 // CHECK-32-EX-NEXT:    br i1 [[IS_ACTIVE_THREAD]], label [[THEN2:%.*]], label [[ELSE3:%.*]]
2080 // CHECK-32-EX:       then3:
2081 // CHECK-32-EX-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[TMP3]]
2082 // CHECK-32-EX-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP6]], i32 0, i32 0
2083 // CHECK-32-EX-NEXT:    [[TMP14:%.*]] = load ptr, ptr [[TMP13]], align 4
2084 // CHECK-32-EX-NEXT:    [[TMP15:%.*]] = load volatile i32, ptr addrspace(3) [[TMP12]], align 4
2085 // CHECK-32-EX-NEXT:    store i32 [[TMP15]], ptr [[TMP14]], align 4
2086 // CHECK-32-EX-NEXT:    br label [[IFCONT4:%.*]]
2087 // CHECK-32-EX:       else4:
2088 // CHECK-32-EX-NEXT:    br label [[IFCONT4]]
2089 // CHECK-32-EX:       ifcont5:
2090 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
2091 // CHECK-32-EX-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP2]])
2092 // CHECK-32-EX-NEXT:    [[WARP_MASTER5:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0
2093 // CHECK-32-EX-NEXT:    br i1 [[WARP_MASTER5]], label [[THEN6:%.*]], label [[ELSE7:%.*]]
2094 // CHECK-32-EX:       then8:
2095 // CHECK-32-EX-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP6]], i32 0, i32 1
2096 // CHECK-32-EX-NEXT:    [[TMP17:%.*]] = load ptr, ptr [[TMP16]], align 4
2097 // CHECK-32-EX-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]]
2098 // CHECK-32-EX-NEXT:    [[TMP19:%.*]] = load i16, ptr [[TMP17]], align 2
2099 // CHECK-32-EX-NEXT:    store volatile i16 [[TMP19]], ptr addrspace(3) [[TMP18]], align 2
2100 // CHECK-32-EX-NEXT:    br label [[IFCONT8:%.*]]
2101 // CHECK-32-EX:       else9:
2102 // CHECK-32-EX-NEXT:    br label [[IFCONT8]]
2103 // CHECK-32-EX:       ifcont10:
2104 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
2105 // CHECK-32-EX-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP2]])
2106 // CHECK-32-EX-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTADDR1]], align 4
2107 // CHECK-32-EX-NEXT:    [[IS_ACTIVE_THREAD9:%.*]] = icmp ult i32 [[TMP3]], [[TMP20]]
2108 // CHECK-32-EX-NEXT:    br i1 [[IS_ACTIVE_THREAD9]], label [[THEN10:%.*]], label [[ELSE11:%.*]]
2109 // CHECK-32-EX:       then13:
2110 // CHECK-32-EX-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [32 x i32], ptr addrspace(3) @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[TMP3]]
2111 // CHECK-32-EX-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [2 x ptr], ptr [[TMP6]], i32 0, i32 1
2112 // CHECK-32-EX-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP22]], align 4
2113 // CHECK-32-EX-NEXT:    [[TMP24:%.*]] = load volatile i16, ptr addrspace(3) [[TMP21]], align 2
2114 // CHECK-32-EX-NEXT:    store i16 [[TMP24]], ptr [[TMP23]], align 2
2115 // CHECK-32-EX-NEXT:    br label [[IFCONT12:%.*]]
2116 // CHECK-32-EX:       else14:
2117 // CHECK-32-EX-NEXT:    br label [[IFCONT12]]
2118 // CHECK-32-EX:       ifcont15:
2119 // CHECK-32-EX-NEXT:    ret void
2120 //