hide_mem_transfer_latency.ll | Explore in Territory

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --function-signature
; RUN: opt -S -passes=openmp-opt-cgscc -aa-pipeline=basic-aa -openmp-hide-memory-transfer-latency < %s | FileCheck %s
target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"

; CHECK: %struct.__tgt_async_info = type { ptr }

%struct.ident_t = type { i32, i32, i32, i32, ptr }
%struct.__tgt_offload_entry = type { ptr, ptr, i64, i32, i32 }

@.offload_maptypes = private unnamed_addr constant [1 x i64] [i64 35]
@.__omp_offloading_heavyComputation1.region_id = weak constant i8 0
@.offload_sizes.1 = private unnamed_addr constant [1 x i64] [i64 8]
@.offload_maptypes.2 = private unnamed_addr constant [1 x i64] [i64 800]

@.__omp_offloading_heavyComputation2.region_id = weak constant i8 0
@.offload_maptypes.3 = private unnamed_addr constant [2 x i64] [i64 35, i64 35]

@.__omp_offloading_heavyComputation3.region_id = weak constant i8 0
@.offload_sizes.2 = private unnamed_addr constant [2 x i64] [i64 4, i64 0]
@.offload_maptypes.4 = private unnamed_addr constant [2 x i64] [i64 800, i64 544]

@.offload_maptypes.5 = private unnamed_addr constant [1 x i64] [i64 33]

@0 = private unnamed_addr global %struct.ident_t { i32 0, i32 34, i32 0, i32 0, ptr @.str0 }, align 8
@.str0 = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00", align 1

;double heavyComputation1() {
;  double a = rand() % 777;
;  double random = rand();
;
;  //#pragma omp target data map(a)
;  ptr args[1];
;  args[0] = &a;
;  __tgt_target_data_begin(..., args, ...)
;
;  #pragma omp target teams
;  for (int i = 0; i < 1000; ++i) {
;    a *= i*i / 2;
;  }
;
;  return random + a;
;}
define dso_local double @heavyComputation1() {
; CHECK-LABEL: define {{[^@]+}}@heavyComputation1() {
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[A:%.*]] = alloca double, align 8
; CHECK-NEXT:    [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 8
; CHECK-NEXT:    [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 8
; CHECK-NEXT:    [[DOTOFFLOAD_BASEPTRS4:%.*]] = alloca [1 x ptr], align 8
; CHECK-NEXT:    [[DOTOFFLOAD_PTRS5:%.*]] = alloca [1 x ptr], align 8
; CHECK-NEXT:    [[CALL:%.*]] = tail call i32 (...) @rand()
; CHECK-NEXT:    [[REM:%.*]] = srem i32 [[CALL]], 777
; CHECK-NEXT:    [[CONV:%.*]] = sitofp i32 [[REM]] to double
; CHECK-NEXT:    store double [[CONV]], ptr [[A]], align 8
; CHECK-NEXT:    [[CALL1:%.*]] = tail call i32 (...) @rand()
; CHECK-NEXT:    store ptr [[A]], ptr [[DOTOFFLOAD_BASEPTRS]], align 8
; CHECK-NEXT:    store ptr [[A]], ptr [[DOTOFFLOAD_PTRS]], align 8
; CHECK-NEXT:    call void @__tgt_target_data_begin_mapper(ptr @[[GLOB0:[0-9]+]], i64 -1, i32 1, ptr nonnull [[DOTOFFLOAD_BASEPTRS]], ptr nonnull [[DOTOFFLOAD_PTRS]], ptr @.offload_sizes.1, ptr @.offload_maptypes, ptr null, ptr null)
; CHECK-NEXT:    [[TMP0:%.*]] = load i64, ptr [[A]], align 8
; CHECK-NEXT:    store i64 [[TMP0]], ptr [[DOTOFFLOAD_BASEPTRS4]], align 8
; CHECK-NEXT:    store i64 [[TMP0]], ptr [[DOTOFFLOAD_PTRS5]], align 8
; CHECK-NEXT:    [[TMP1:%.*]] = call i32 @__tgt_target_teams_mapper(ptr @[[GLOB0]], i64 -1, ptr nonnull @.__omp_offloading_heavyComputation1.region_id, i32 1, ptr nonnull [[DOTOFFLOAD_BASEPTRS4]], ptr nonnull [[DOTOFFLOAD_PTRS5]], ptr @.offload_sizes.1, ptr @.offload_maptypes.2, ptr null, ptr null, i32 0, i32 0)
; CHECK-NEXT:    [[DOTNOT:%.*]] = icmp eq i32 [[TMP1]], 0
; CHECK-NEXT:    br i1 [[DOTNOT]], label [[OMP_OFFLOAD_CONT:%.*]], label [[OMP_OFFLOAD_FAILED:%.*]]
; CHECK:       omp_offload.failed:
; CHECK-NEXT:    call void @heavyComputation1FallBack(i64 [[TMP0]])
; CHECK-NEXT:    br label [[OMP_OFFLOAD_CONT]]
; CHECK:       omp_offload.cont:
; CHECK-NEXT:    [[CONV2:%.*]] = sitofp i32 [[CALL1]] to double
; CHECK-NEXT:    call void @__tgt_target_data_end_mapper(ptr @[[GLOB0]], i64 -1, i32 1, ptr nonnull [[DOTOFFLOAD_BASEPTRS]], ptr nonnull [[DOTOFFLOAD_PTRS]], ptr @.offload_sizes.1, ptr @.offload_maptypes, ptr null, ptr null)
; CHECK-NEXT:    [[TMP2:%.*]] = load double, ptr [[A]], align 8
; CHECK-NEXT:    [[ADD:%.*]] = fadd double [[TMP2]], [[CONV2]]
; CHECK-NEXT:    ret double [[ADD]]
;






entry:
  %a = alloca double, align 8
  %.offload_baseptrs = alloca [1 x ptr], align 8
  %.offload_ptrs = alloca [1 x ptr], align 8
  %.offload_baseptrs4 = alloca [1 x ptr], align 8
  %.offload_ptrs5 = alloca [1 x ptr], align 8

  %call = tail call i32 (...) @rand()
  %rem = srem i32 %call, 777
  %conv = sitofp i32 %rem to double
  store double %conv, ptr %a, align 8

  ; FIXME: call to @__tgt_target_data_begin_mapper_issue(ptr @0, ...) should be moved here.
  %call1 = tail call i32 (...) @rand()

  store ptr %a, ptr %.offload_baseptrs, align 8
  store ptr %a, ptr %.offload_ptrs, align 8
  call void @__tgt_target_data_begin_mapper(ptr @0, i64 -1, i32 1, ptr nonnull %.offload_baseptrs, ptr nonnull %.offload_ptrs, ptr @.offload_sizes.1, ptr @.offload_maptypes, ptr null, ptr null)

  %0 = load i64, ptr %a, align 8
  store i64 %0, ptr %.offload_baseptrs4, align 8
  store i64 %0, ptr %.offload_ptrs5, align 8

  ; FIXME: call to @__tgt_target_data_begin_mapper_wait(...) should be moved here.
  %1 = call i32 @__tgt_target_teams_mapper(ptr @0, i64 -1, ptr nonnull @.__omp_offloading_heavyComputation1.region_id, i32 1, ptr nonnull %.offload_baseptrs4, ptr nonnull %.offload_ptrs5, ptr @.offload_sizes.1, ptr @.offload_maptypes.2, ptr null, ptr null, i32 0, i32 0)
  %.not = icmp eq i32 %1, 0
  br i1 %.not, label %omp_offload.cont, label %omp_offload.failed

omp_offload.failed:                               ; preds = %entry
  call void @heavyComputation1FallBack(i64 %0)
  br label %omp_offload.cont

omp_offload.cont:                                 ; preds = %omp_offload.failed, %entry
  %conv2 = sitofp i32 %call1 to double
  call void @__tgt_target_data_end_mapper(ptr @0, i64 -1, i32 1, ptr nonnull %.offload_baseptrs, ptr nonnull %.offload_ptrs, ptr @.offload_sizes.1, ptr @.offload_maptypes, ptr null, ptr null)
  %2 = load double, ptr %a, align 8
  %add = fadd double %2, %conv2
  ret double %add
}

define internal void @heavyComputation1FallBack(i64 %a) {
; CHECK-LABEL: define {{[^@]+}}@heavyComputation1FallBack
; CHECK-SAME: (i64 [[A:%.*]]) {
; CHECK-NEXT:  entry:
; CHECK-NEXT:    ret void
;
entry:
  ; Fallback for offloading function heavyComputation1.
  ret void
}

;int heavyComputation2(ptr a, unsigned size) {
;  int random = rand() % 7;
;
;  //#pragma omp target data map(a[0:size], size)
;  ptr args[2];
;  args[0] = &a;
;  args[1] = &size;
;  __tgt_target_data_begin(..., args, ...)
;
;  #pragma omp target teams
;  for (int i = 0; i < size; ++i) {
;    a[i] = ++aptr 3.141624;
;  }
;
;  return random;
;}
define dso_local i32 @heavyComputation2(ptr %a, i32 %size) {
; CHECK-LABEL: define {{[^@]+}}@heavyComputation2
; CHECK-SAME: (ptr [[A:%.*]], i32 [[SIZE:%.*]]) {
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[SIZE_ADDR:%.*]] = alloca i32, align 4
; CHECK-NEXT:    [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [2 x ptr], align 8
; CHECK-NEXT:    [[DOTOFFLOAD_PTRS:%.*]] = alloca [2 x ptr], align 8
; CHECK-NEXT:    [[DOTOFFLOAD_SIZES:%.*]] = alloca [2 x i64], align 8
; CHECK-NEXT:    [[DOTOFFLOAD_BASEPTRS2:%.*]] = alloca [2 x ptr], align 8
; CHECK-NEXT:    [[DOTOFFLOAD_PTRS3:%.*]] = alloca [2 x ptr], align 8
; CHECK-NEXT:    store i32 [[SIZE]], ptr [[SIZE_ADDR]], align 4
; CHECK-NEXT:    [[CALL:%.*]] = tail call i32 (...) @rand()
; CHECK-NEXT:    [[CONV:%.*]] = zext i32 [[SIZE]] to i64
; CHECK-NEXT:    [[TMP0:%.*]] = shl nuw nsw i64 [[CONV]], 3
; CHECK-NEXT:    store ptr [[A]], ptr [[DOTOFFLOAD_BASEPTRS]], align 8
; CHECK-NEXT:    store ptr [[A]], ptr [[DOTOFFLOAD_PTRS]], align 8
; CHECK-NEXT:    store i64 [[TMP0]], ptr [[DOTOFFLOAD_SIZES]], align 8
; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i64 0, i64 1
; CHECK-NEXT:    store ptr [[SIZE_ADDR]], ptr [[TMP1]], align 8
; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS]], i64 0, i64 1
; CHECK-NEXT:    store ptr [[SIZE_ADDR]], ptr [[TMP2]], align 8
; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [2 x i64], ptr [[DOTOFFLOAD_SIZES]], i64 0, i64 1
; CHECK-NEXT:    store i64 4, ptr [[TMP3]], align 8
; CHECK-NEXT:    call void @__tgt_target_data_begin_mapper(ptr @[[GLOB0]], i64 -1, i32 2, ptr nonnull [[DOTOFFLOAD_BASEPTRS]], ptr nonnull [[DOTOFFLOAD_PTRS]], ptr nonnull [[DOTOFFLOAD_SIZES]], ptr @.offload_maptypes.3, ptr null, ptr null)
; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr [[SIZE_ADDR]], align 4
; CHECK-NEXT:    [[SIZE_CASTED:%.*]] = zext i32 [[TMP4]] to i64
; CHECK-NEXT:    store i64 [[SIZE_CASTED]], ptr [[DOTOFFLOAD_BASEPTRS2]], align 8
; CHECK-NEXT:    store i64 [[SIZE_CASTED]], ptr [[DOTOFFLOAD_PTRS3]], align 8
; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i64 0, i64 1
; CHECK-NEXT:    store ptr [[A]], ptr [[TMP5]], align 8
; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i64 0, i64 1
; CHECK-NEXT:    store ptr [[A]], ptr [[TMP6]], align 8
; CHECK-NEXT:    [[TMP7:%.*]] = call i32 @__tgt_target_teams_mapper(ptr @[[GLOB0]], i64 -1, ptr nonnull @.__omp_offloading_heavyComputation2.region_id, i32 2, ptr nonnull [[DOTOFFLOAD_BASEPTRS2]], ptr nonnull [[DOTOFFLOAD_PTRS3]], ptr @.offload_sizes.2, ptr @.offload_maptypes.4, ptr null, ptr null, i32 0, i32 0)
; CHECK-NEXT:    [[DOTNOT:%.*]] = icmp eq i32 [[TMP7]], 0
; CHECK-NEXT:    br i1 [[DOTNOT]], label [[OMP_OFFLOAD_CONT:%.*]], label [[OMP_OFFLOAD_FAILED:%.*]]
; CHECK:       omp_offload.failed:
; CHECK-NEXT:    call void @heavyComputation2FallBack(i64 [[SIZE_CASTED]], ptr [[A]])
; CHECK-NEXT:    br label [[OMP_OFFLOAD_CONT]]
; CHECK:       omp_offload.cont:
; CHECK-NEXT:    [[REM:%.*]] = srem i32 [[CALL]], 7
; CHECK-NEXT:    call void @__tgt_target_data_end_mapper(ptr @[[GLOB0]], i64 -1, i32 2, ptr nonnull [[DOTOFFLOAD_BASEPTRS]], ptr nonnull [[DOTOFFLOAD_PTRS]], ptr nonnull [[DOTOFFLOAD_SIZES]], ptr @.offload_maptypes.3, ptr null, ptr null)
; CHECK-NEXT:    ret i32 [[REM]]
;


entry:
  %size.addr = alloca i32, align 4
  %.offload_baseptrs = alloca [2 x ptr], align 8
  %.offload_ptrs = alloca [2 x ptr], align 8
  %.offload_sizes = alloca [2 x i64], align 8
  %.offload_baseptrs2 = alloca [2 x ptr], align 8
  %.offload_ptrs3 = alloca [2 x ptr], align 8

  store i32 %size, ptr %size.addr, align 4
  %call = tail call i32 (...) @rand()

  %conv = zext i32 %size to i64
  %0 = shl nuw nsw i64 %conv, 3
  store ptr %a, ptr %.offload_baseptrs, align 8
  store ptr %a, ptr %.offload_ptrs, align 8
  store i64 %0, ptr %.offload_sizes, align 8
  %1 = getelementptr inbounds [2 x ptr], ptr %.offload_baseptrs, i64 0, i64 1
  store ptr %size.addr, ptr %1, align 8
  %2 = getelementptr inbounds [2 x ptr], ptr %.offload_ptrs, i64 0, i64 1
  store ptr %size.addr, ptr %2, align 8
  %3 = getelementptr inbounds [2 x i64], ptr %.offload_sizes, i64 0, i64 1
  store i64 4, ptr %3, align 8
  call void @__tgt_target_data_begin_mapper(ptr @0, i64 -1, i32 2, ptr nonnull %.offload_baseptrs, ptr nonnull %.offload_ptrs, ptr nonnull %.offload_sizes, ptr @.offload_maptypes.3, ptr null, ptr null)

  %4 = load i32, ptr %size.addr, align 4
  %size.casted = zext i32 %4 to i64
  store i64 %size.casted, ptr %.offload_baseptrs2, align 8
  store i64 %size.casted, ptr %.offload_ptrs3, align 8
  %5 = getelementptr inbounds [2 x ptr], ptr %.offload_baseptrs2, i64 0, i64 1
  store ptr %a, ptr %5, align 8
  %6 = getelementptr inbounds [2 x ptr], ptr %.offload_ptrs3, i64 0, i64 1
  store ptr %a, ptr %6, align 8

  ; FIXME: call to @__tgt_target_data_begin_mapper_wait(...) should be moved here.
  %7 = call i32 @__tgt_target_teams_mapper(ptr @0, i64 -1, ptr nonnull @.__omp_offloading_heavyComputation2.region_id, i32 2, ptr nonnull %.offload_baseptrs2, ptr nonnull %.offload_ptrs3, ptr @.offload_sizes.2, ptr @.offload_maptypes.4, ptr null, ptr null, i32 0, i32 0)
  %.not = icmp eq i32 %7, 0
  br i1 %.not, label %omp_offload.cont, label %omp_offload.failed

omp_offload.failed:                               ; preds = %entry
  call void @heavyComputation2FallBack(i64 %size.casted, ptr %a)
  br label %omp_offload.cont

omp_offload.cont:                                 ; preds = %omp_offload.failed, %entry
  %rem = srem i32 %call, 7
  call void @__tgt_target_data_end_mapper(ptr @0, i64 -1, i32 2, ptr nonnull %.offload_baseptrs, ptr nonnull %.offload_ptrs, ptr nonnull %.offload_sizes, ptr @.offload_maptypes.3, ptr null, ptr null)
  ret i32 %rem
}

define internal void @heavyComputation2FallBack(i64 %size, ptr %a) {
; CHECK-LABEL: define {{[^@]+}}@heavyComputation2FallBack
; CHECK-SAME: (i64 [[SIZE:%.*]], ptr [[A:%.*]]) {
; CHECK-NEXT:  entry:
; CHECK-NEXT:    ret void
;
entry:
  ; Fallback for offloading function heavyComputation2.
  ret void
}

;int heavyComputation3(ptr restrict a, unsigned size) {
;  int random = rand() % 7;
;
;  //#pragma omp target data map(a[0:size], size)
;  ptr args[2];
;  args[0] = &a;
;  args[1] = &size;
;  __tgt_target_data_begin(..., args, ...)
;
;  #pragma omp target teams
;  for (int i = 0; i < size; ++i) {
;    a[i] = ++aptr 3.141624;
;  }
;
;  return random;
;}
define dso_local i32 @heavyComputation3(ptr noalias %a, i32 %size) {
; CHECK-LABEL: define {{[^@]+}}@heavyComputation3
; CHECK-SAME: (ptr noalias [[A:%.*]], i32 [[SIZE:%.*]]) {
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[SIZE_ADDR:%.*]] = alloca i32, align 4
; CHECK-NEXT:    [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [2 x ptr], align 8
; CHECK-NEXT:    [[DOTOFFLOAD_PTRS:%.*]] = alloca [2 x ptr], align 8
; CHECK-NEXT:    [[DOTOFFLOAD_SIZES:%.*]] = alloca [2 x i64], align 8
; CHECK-NEXT:    [[DOTOFFLOAD_BASEPTRS2:%.*]] = alloca [2 x ptr], align 8
; CHECK-NEXT:    [[DOTOFFLOAD_PTRS3:%.*]] = alloca [2 x ptr], align 8
; CHECK-NEXT:    store i32 [[SIZE]], ptr [[SIZE_ADDR]], align 4
; CHECK-NEXT:    [[CALL:%.*]] = tail call i32 (...) @rand()
; CHECK-NEXT:    [[CONV:%.*]] = zext i32 [[SIZE]] to i64
; CHECK-NEXT:    [[TMP0:%.*]] = shl nuw nsw i64 [[CONV]], 3
; CHECK-NEXT:    store ptr [[A]], ptr [[DOTOFFLOAD_BASEPTRS]], align 8
; CHECK-NEXT:    store ptr [[A]], ptr [[DOTOFFLOAD_PTRS]], align 8
; CHECK-NEXT:    store i64 [[TMP0]], ptr [[DOTOFFLOAD_SIZES]], align 8
; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i64 0, i64 1
; CHECK-NEXT:    store ptr [[SIZE_ADDR]], ptr [[TMP1]], align 8
; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS]], i64 0, i64 1
; CHECK-NEXT:    store ptr [[SIZE_ADDR]], ptr [[TMP2]], align 8
; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [2 x i64], ptr [[DOTOFFLOAD_SIZES]], i64 0, i64 1
; CHECK-NEXT:    store i64 4, ptr [[TMP3]], align 8
; CHECK-NEXT:    call void @__tgt_target_data_begin_mapper(ptr @[[GLOB0]], i64 -1, i32 2, ptr nonnull [[DOTOFFLOAD_BASEPTRS]], ptr nonnull [[DOTOFFLOAD_PTRS]], ptr nonnull [[DOTOFFLOAD_SIZES]], ptr @.offload_maptypes.3, ptr null, ptr null)
; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr [[SIZE_ADDR]], align 4
; CHECK-NEXT:    [[SIZE_CASTED:%.*]] = zext i32 [[TMP4]] to i64
; CHECK-NEXT:    store i64 [[SIZE_CASTED]], ptr [[DOTOFFLOAD_BASEPTRS2]], align 8
; CHECK-NEXT:    store i64 [[SIZE_CASTED]], ptr [[DOTOFFLOAD_PTRS3]], align 8
; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS2]], i64 0, i64 1
; CHECK-NEXT:    store ptr [[A]], ptr [[TMP5]], align 8
; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS3]], i64 0, i64 1
; CHECK-NEXT:    store ptr [[A]], ptr [[TMP6]], align 8
; CHECK-NEXT:    [[TMP7:%.*]] = call i32 @__tgt_target_teams_mapper(ptr @[[GLOB0]], i64 -1, ptr nonnull @.__omp_offloading_heavyComputation3.region_id, i32 2, ptr nonnull [[DOTOFFLOAD_BASEPTRS2]], ptr nonnull [[DOTOFFLOAD_PTRS3]], ptr @.offload_sizes.2, ptr @.offload_maptypes.4, ptr null, ptr null, i32 0, i32 0)
; CHECK-NEXT:    [[DOTNOT:%.*]] = icmp eq i32 [[TMP7]], 0
; CHECK-NEXT:    br i1 [[DOTNOT]], label [[OMP_OFFLOAD_CONT:%.*]], label [[OMP_OFFLOAD_FAILED:%.*]]
; CHECK:       omp_offload.failed:
; CHECK-NEXT:    call void @heavyComputation3FallBack(i64 [[SIZE_CASTED]], ptr [[A]])
; CHECK-NEXT:    br label [[OMP_OFFLOAD_CONT]]
; CHECK:       omp_offload.cont:
; CHECK-NEXT:    [[REM:%.*]] = srem i32 [[CALL]], 7
; CHECK-NEXT:    call void @__tgt_target_data_end_mapper(ptr @[[GLOB0]], i64 -1, i32 2, ptr nonnull [[DOTOFFLOAD_BASEPTRS]], ptr nonnull [[DOTOFFLOAD_PTRS]], ptr nonnull [[DOTOFFLOAD_SIZES]], ptr @.offload_maptypes.3, ptr null, ptr null)
; CHECK-NEXT:    ret i32 [[REM]]
;


entry:
  %size.addr = alloca i32, align 4
  %.offload_baseptrs = alloca [2 x ptr], align 8
  %.offload_ptrs = alloca [2 x ptr], align 8
  %.offload_sizes = alloca [2 x i64], align 8
  %.offload_baseptrs2 = alloca [2 x ptr], align 8
  %.offload_ptrs3 = alloca [2 x ptr], align 8
  store i32 %size, ptr %size.addr, align 4

  ; FIXME: call to @__tgt_target_data_begin_mapper_issue(ptr @0, ...) should be moved here.
  %call = tail call i32 (...) @rand()

  %conv = zext i32 %size to i64
  %0 = shl nuw nsw i64 %conv, 3
  store ptr %a, ptr %.offload_baseptrs, align 8
  store ptr %a, ptr %.offload_ptrs, align 8
  store i64 %0, ptr %.offload_sizes, align 8
  %1 = getelementptr inbounds [2 x ptr], ptr %.offload_baseptrs, i64 0, i64 1
  store ptr %size.addr, ptr %1, align 8
  %2 = getelementptr inbounds [2 x ptr], ptr %.offload_ptrs, i64 0, i64 1
  store ptr %size.addr, ptr %2, align 8
  %3 = getelementptr inbounds [2 x i64], ptr %.offload_sizes, i64 0, i64 1
  store i64 4, ptr %3, align 8
  call void @__tgt_target_data_begin_mapper(ptr @0, i64 -1, i32 2, ptr nonnull %.offload_baseptrs, ptr nonnull %.offload_ptrs, ptr nonnull %.offload_sizes, ptr @.offload_maptypes.3, ptr null, ptr null)

  %4 = load i32, ptr %size.addr, align 4
  %size.casted = zext i32 %4 to i64
  store i64 %size.casted, ptr %.offload_baseptrs2, align 8
  store i64 %size.casted, ptr %.offload_ptrs3, align 8
  %5 = getelementptr inbounds [2 x ptr], ptr %.offload_baseptrs2, i64 0, i64 1
  store ptr %a, ptr %5, align 8
  %6 = getelementptr inbounds [2 x ptr], ptr %.offload_ptrs3, i64 0, i64 1
  store ptr %a, ptr %6, align 8

  ; FIXME: call to @__tgt_target_data_begin_mapper_wait(...) should be moved here.
  %7 = call i32 @__tgt_target_teams_mapper(ptr @0, i64 -1, ptr nonnull @.__omp_offloading_heavyComputation3.region_id, i32 2, ptr nonnull %.offload_baseptrs2, ptr nonnull %.offload_ptrs3, ptr @.offload_sizes.2, ptr @.offload_maptypes.4, ptr null, ptr null, i32 0, i32 0)
  %.not = icmp eq i32 %7, 0
  br i1 %.not, label %omp_offload.cont, label %omp_offload.failed

omp_offload.failed:                               ; preds = %entry
  call void @heavyComputation3FallBack(i64 %size.casted, ptr %a)
  br label %omp_offload.cont

omp_offload.cont:                                 ; preds = %omp_offload.failed, %entry
  %rem = srem i32 %call, 7
  call void @__tgt_target_data_end_mapper(ptr @0, i64 -1, i32 2, ptr nonnull %.offload_baseptrs, ptr nonnull %.offload_ptrs, ptr nonnull %.offload_sizes, ptr @.offload_maptypes.3, ptr null, ptr null)
  ret i32 %rem
}

define internal void @heavyComputation3FallBack(i64 %size, ptr %a) {
; CHECK-LABEL: define {{[^@]+}}@heavyComputation3FallBack
; CHECK-SAME: (i64 [[SIZE:%.*]], ptr [[A:%.*]]) {
; CHECK-NEXT:  entry:
; CHECK-NEXT:    ret void
;
entry:
  ; Fallback for offloading function heavyComputation3.
  ret void
}

;int dataTransferOnly1(ptr restrict a, unsigned size) {
;  // Random computation.
;  int random = rand();
;
;  //#pragma omp target data map(to:a[0:size])
;  ptr args[1];
;  args[0] = &a;
;  __tgt_target_data_begin(..., args, ...)
;
;  // Random computation.
;  random %= size;
;  return random;
;}
define dso_local i32 @dataTransferOnly1(ptr noalias %a, i32 %size) {
; CHECK-LABEL: define {{[^@]+}}@dataTransferOnly1
; CHECK-SAME: (ptr noalias [[A:%.*]], i32 [[SIZE:%.*]]) {
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 8
; CHECK-NEXT:    [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 8
; CHECK-NEXT:    [[DOTOFFLOAD_SIZES:%.*]] = alloca [1 x i64], align 8
; CHECK-NEXT:    [[HANDLE:%.*]] = alloca [[STRUCT___TGT_ASYNC_INFO:%.*]], align 8
; CHECK-NEXT:    [[CALL:%.*]] = tail call i32 (...) @rand()
; CHECK-NEXT:    [[CONV:%.*]] = zext i32 [[SIZE]] to i64
; CHECK-NEXT:    [[TMP0:%.*]] = shl nuw nsw i64 [[CONV]], 3
; CHECK-NEXT:    store ptr [[A]], ptr [[DOTOFFLOAD_BASEPTRS]], align 8
; CHECK-NEXT:    store ptr [[A]], ptr [[DOTOFFLOAD_PTRS]], align 8
; CHECK-NEXT:    store i64 [[TMP0]], ptr [[DOTOFFLOAD_SIZES]], align 8
; CHECK-NEXT:    call void @__tgt_target_data_begin_mapper_issue(ptr @[[GLOB0]], i64 -1, i32 1, ptr [[DOTOFFLOAD_BASEPTRS]], ptr [[DOTOFFLOAD_PTRS]], ptr [[DOTOFFLOAD_SIZES]], ptr @.offload_maptypes.5, ptr null, ptr null, ptr [[HANDLE]])
; CHECK-NEXT:    [[REM:%.*]] = urem i32 [[CALL]], [[SIZE]]
; CHECK-NEXT:    call void @__tgt_target_data_begin_mapper_wait(i64 -1, ptr [[HANDLE]])
; CHECK-NEXT:    call void @__tgt_target_data_end_mapper(ptr @[[GLOB0]], i64 -1, i32 1, ptr nonnull [[DOTOFFLOAD_BASEPTRS]], ptr nonnull [[DOTOFFLOAD_PTRS]], ptr nonnull [[DOTOFFLOAD_SIZES]], ptr @.offload_maptypes.5, ptr null, ptr null)
; CHECK-NEXT:    ret i32 [[REM]]
;






entry:
  %.offload_baseptrs = alloca [1 x ptr], align 8
  %.offload_ptrs = alloca [1 x ptr], align 8
  %.offload_sizes = alloca [1 x i64], align 8

  ; FIXME: call to @__tgt_target_data_begin_issue_mapper(...) should be moved here.
  %call = tail call i32 (...) @rand()

  %conv = zext i32 %size to i64
  %0 = shl nuw nsw i64 %conv, 3
  store ptr %a, ptr %.offload_baseptrs, align 8
  store ptr %a, ptr %.offload_ptrs, align 8
  store i64 %0, ptr %.offload_sizes, align 8
  call void @__tgt_target_data_begin_mapper(ptr @0, i64 -1, i32 1, ptr nonnull %.offload_baseptrs, ptr nonnull %.offload_ptrs, ptr nonnull %.offload_sizes, ptr @.offload_maptypes.5, ptr null, ptr null)

  %rem = urem i32 %call, %size

  call void @__tgt_target_data_end_mapper(ptr @0, i64 -1, i32 1, ptr nonnull %.offload_baseptrs, ptr nonnull %.offload_ptrs, ptr nonnull %.offload_sizes, ptr @.offload_maptypes.5, ptr null, ptr null)
  ret i32 %rem
}

declare void @__tgt_target_data_begin_mapper(ptr, i64, i32, ptr, ptr, ptr, ptr, ptr, ptr)
declare i32 @__tgt_target_teams_mapper(ptr, i64, ptr, i32, ptr, ptr, ptr, ptr, ptr, ptr, i32, i32)
declare void @__tgt_target_data_end_mapper(ptr, i64, i32, ptr, ptr, ptr, ptr, ptr, ptr)

declare dso_local i32 @rand(...)


!llvm.module.flags = !{!0}

!0 = !{i32 7, !"openmp", i32 50}
llvm/llvm/test/Transforms/OpenMP/hide_mem_transfer_latency.ll
