Index: llvm/test/Transforms/OpenMP/mem_transfer_hiding.ll
===================================================================
--- /dev/null
+++ llvm/test/Transforms/OpenMP/mem_transfer_hiding.ll
@@ -0,0 +1,297 @@
+; RUN: opt -S -openmpopt -stats < %s 2>&1
+; REQUIRES: asserts
+
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+
+; FIXME: This struct should be generated after splitting one of the runtime calls.
+; %struct.__tgt_async_info = type { i8* }
+
+@.offload_sizes = private unnamed_addr constant [1 x i64] [i64 8]
+@.offload_maptypes = private unnamed_addr constant [1 x i64] [i64 35]
+@.__omp_offloading_heavyComputation1.region_id = weak constant i8 0
+@.offload_sizes.2 = private unnamed_addr constant [1 x i64] [i64 8]
+@.offload_maptypes.2 = private unnamed_addr constant [1 x i64] [i64 800]
+
+@.offload_maptypes.3 = private unnamed_addr constant [2 x i64] [i64 35, i64 35]
+@.offload_sizes.4 = 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]
+@.__omp_offloading_heavyComputation2.region_id = weak constant i8 0
+
+;double heavyComputation1() {
+;  double random = rand();
+;  double a = rand() % 777;
+;
+;  #pragma omp target data map(a)
+;  #pragma omp target teams
+;  for (int i = 0; i < 1000; ++i) {
+;    a *= i*i / 2;
+;  }
+;
+;  return random + a;
+;}
+define dso_local double @heavyComputation1() {
+entry:
+  %random = alloca double, align 8
+  %a = alloca double, align 8
+  %.offload_baseptrs = alloca [1 x i8*], align 8
+  %.offload_ptrs = alloca [1 x i8*], align 8
+  %a.casted = alloca i64, align 8
+  %.offload_baseptrs2 = alloca [1 x i8*], align 8
+  %.offload_ptrs2 = alloca [1 x i8*], align 8
+
+  ; FIXME: Should have after splitting the runtime call __tgt_target_data_begin.
+  ; %device_id1 = alloca i64, align 8
+  ; %async_info1 = alloca %struct.__tgt_async_info, align 8
+
+  ; FIXME: This is the argument setup and call to the runtime call __tgt_target_data_begin_issue,
+  ;        and should be moved here.
+  ; %call1 = call i32 @rand()
+  ; %rem = srem i32 %call1, 777
+  ; %conv2 = sitofp i32 %rem to double
+  ; store double %conv2, double* %a, align 8
+  ; %0 = getelementptr inbounds [1 x i8*], [1 x i8*]* %.offload_baseptrs, i32 0, i32 0
+  ; %1 = bitcast i8** %0 to double**
+  ; store double* %a, double** %1, align 8
+  ; %2 = getelementptr inbounds [1 x i8*], [1 x i8*]* %.offload_ptrs, i32 0, i32 0
+  ; %3 = bitcast i8** %2 to double**
+  ; store double* %a, double** %3, align 8
+  ; %4 = getelementptr inbounds [1 x i8*], [1 x i8*]* %.offload_baseptrs, i32 0, i32 0
+  ; %5 = getelementptr inbounds [1 x i8*], [1 x i8*]* %.offload_ptrs, i32 0, i32 0
+  ; store i64 -1, i64* %device_id1, align 8
+  ; %handle1 = call i8* @__tgt_target_data_begin_issue(i64* dereferenceable(8) %device_id1, i32 1, i8** %4, i8** %5, i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_sizes, i32 0, i32 0), i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_maptypes, i32 0, i32 0))
+
+  ; FIXME: The initialization of the'random' variable should end up between the issue and wait calls.
+  ;        Note that this might be more code, but for simplicity only this variable is used here.
+  %call = call i32 @rand()
+  %conv = sitofp i32 %call to double
+  store double %conv, double* %random, align 8
+
+  ; FIXME: This argument setup and call to __tgt_target_data_begin should be moved above,
+  ;        changing __tgt_target_data_begin for __tgt_target_data_begin_issue.
+  %call1 = call i32 @rand()
+  %rem = srem i32 %call1, 777
+  %conv2 = sitofp i32 %rem to double
+  store double %conv2, double* %a, align 8
+  %0 = getelementptr inbounds [1 x i8*], [1 x i8*]* %.offload_baseptrs, i32 0, i32 0
+  %1 = bitcast i8** %0 to double**
+  store double* %a, double** %1, align 8
+  %2 = getelementptr inbounds [1 x i8*], [1 x i8*]* %.offload_ptrs, i32 0, i32 0
+  %3 = bitcast i8** %2 to double**
+  store double* %a, double** %3, align 8
+  %4 = getelementptr inbounds [1 x i8*], [1 x i8*]* %.offload_baseptrs, i32 0, i32 0
+  %5 = getelementptr inbounds [1 x i8*], [1 x i8*]* %.offload_ptrs, i32 0, i32 0
+  call void @__tgt_target_data_begin(i64 -1, i32 1, i8** %4, i8** %5, i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_sizes, i32 0, i32 0), i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_maptypes, i32 0, i32 0))
+
+  ; FIXME: The initialization of the 'random' variable should end up here.
+  ; %call = call i32 @rand()
+  ; %conv = sitofp i32 %call to double
+  ; store double %conv, double* %random, align 8
+
+  %6 = load double, double* %a, align 8
+  %conv3 = bitcast i64* %a.casted to double*
+  store double %6, double* %conv3, align 8
+  %7 = load i64, i64* %a.casted, align 8
+  %8 = getelementptr inbounds [1 x i8*], [1 x i8*]* %.offload_baseptrs2, i32 0, i32 0
+  %9 = bitcast i8** %8 to i64*
+  store i64 %7, i64* %9, align 8
+  %10 = getelementptr inbounds [1 x i8*], [1 x i8*]* %.offload_ptrs2, i32 0, i32 0
+  %11 = bitcast i8** %10 to i64*
+  store i64 %7, i64* %11, align 8
+  %12 = getelementptr inbounds [1 x i8*], [1 x i8*]* %.offload_baseptrs2, i32 0, i32 0
+  %13 = getelementptr inbounds [1 x i8*], [1 x i8*]* %.offload_ptrs2, i32 0, i32 0
+
+  ; FIXME: A fter splitting the runtime call __tgt_target_data_begin, this argument setup and call to
+  ;        __tgt_target_data_wait should be moved here, right before the kernel execution.
+  ; %device_id1_copy = load i64, i64* %device_id1, align 8 ; device_id
+  ; %queue1 = getelementptr inbounds %struct.__tgt_async_info, %struct.__tgt_async_info* %async_info1, i32 0, i32 0
+  ; store i8* %handle1, i8** %queue1, align 8
+  ; call void @__tgt_target_data_begin_wait(i64 %device_id1_copy, %struct.__tgt_async_info* dereferenceable(8) %async_info1)
+
+  %14 = call i32 @__tgt_target_teams(i64 -1, i8* @.__omp_offloading_heavyComputation1.region_id, i32 1, i8** %12, i8** %13, i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_sizes.2, i32 0, i32 0), i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_maptypes.2, i32 0, i32 0), i32 0, i32 0)
+  %15 = icmp ne i32 %14, 0
+  br i1 %15, label %omp_offload.failed, label %omp_offload.cont
+
+omp_offload.failed:                               ; preds = %entry
+  call void @heavyComputation1FallBack(i64 %7)
+  br label %omp_offload.cont
+
+omp_offload.cont:                                 ; preds = %omp_offload.failed, %entry
+  %16 = getelementptr inbounds [1 x i8*], [1 x i8*]* %.offload_baseptrs, i32 0, i32 0
+  %17 = getelementptr inbounds [1 x i8*], [1 x i8*]* %.offload_ptrs, i32 0, i32 0
+  call void @__tgt_target_data_end(i64 -1, i32 1, i8** %16, i8** %17, i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_sizes, i32 0, i32 0), i64* getelementptr inbounds ([1 x i64], [1 x i64]* @.offload_maptypes, i32 0, i32 0))
+  %18 = load double, double* %random, align 8
+  %19 = load double, double* %a, align 8
+  %add = fadd double %18, %19
+  ret double %add
+}
+
+define internal void @heavyComputation1FallBack(i64 %a) {
+entry:
+  ; Fallback for offloading function heavyComputation1.
+  ret void
+}
+
+;int heavyComputation2(double* a, unsigned size) {
+;  int random = rand() % 7;
+;
+;  #pragma omp target data map(a[0:size], size)
+;  #pragma omp target teams
+;  for (int i = 0; i < size; ++i) {
+;    a[i] = ++a[i] * 3.141624;
+;  }
+
+;  return random;
+;}
+define dso_local i32 @heavyComputation2(double* %a, i32 %size) {
+entry:
+  %a.addr = alloca double*, align 8
+  %size.addr = alloca i32, align 4
+  %random = alloca i32, align 4
+  %.offload_baseptrs = alloca [2 x i8*], align 8
+  %.offload_ptrs = alloca [2 x i8*], align 8
+  %.offload_sizes = alloca [2 x i64], align 8
+  %size.casted = alloca i64, align 8
+  %.offload_baseptrs2 = alloca [2 x i8*], align 8
+  %.offload_ptrs3 = alloca [2 x i8*], align 8
+
+  ; FIXME: Should have after splitting the runtime call __tgt_target_data_begin.
+  ; %device_id1 = alloca i64, align 8
+  ; %async_info1 = alloca %struct.__tgt_async_info, align 8
+
+  ; FIXME: This is the argument setup and call to the runtime call __tgt_target_data_begin_issue,
+  ;        and should be moved here.
+  ; store double* %a, double** %a.addr, align 8
+  ; store i32 %size, i32* %size.addr, align 4
+  ; %0 = load double*, double** %a.addr, align 8
+  ; %1 = load double*, double** %a.addr, align 8
+  ; %arrayidx = getelementptr inbounds double, double* %1, i64 0
+  ; %2 = load i32, i32* %size.addr, align 4
+  ; %conv = zext i32 %2 to i64
+  ; %3 = mul nuw i64 %conv, 8
+  ; %4 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_baseptrs, i32 0, i32 0
+  ; %5 = bitcast i8** %4 to double**
+  ; store double* %0, double** %5, align 8
+  ; %6 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_ptrs, i32 0, i32 0
+  ; %7 = bitcast i8** %6 to double**
+  ; store double* %arrayidx, double** %7, align 8
+  ; %8 = getelementptr inbounds [2 x i64], [2 x i64]* %.offload_sizes, i32 0, i32 0
+  ; store i64 %3, i64* %8, align 8
+  ; %9 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_baseptrs, i32 0, i32 1
+  ; %10 = bitcast i8** %9 to i32**
+  ; store i32* %size.addr, i32** %10, align 8
+  ; %11 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_ptrs, i32 0, i32 1
+  ; %12 = bitcast i8** %11 to i32**
+  ; store i32* %size.addr, i32** %12, align 8
+  ; %13 = getelementptr inbounds [2 x i64], [2 x i64]* %.offload_sizes, i32 0, i32 1
+  ; store i64 4, i64* %13, align 8
+  ; %14 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_baseptrs, i32 0, i32 0
+  ; %15 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_ptrs, i32 0, i32 0
+  ; %16 = getelementptr inbounds [2 x i64], [2 x i64]* %.offload_sizes, i32 0, i32 0
+  ; store i64 -1, i64* %device_id1, align 8
+  ; %handle1 = call i8* @__tgt_target_data_begin_issue(i64* dereferenceable(8) %device_id1, i32 2, i8** %14, i8** %15, i64* %16, i64* getelementptr inbounds ([2 x i64], [2 x i64]* @.offload_maptypes.3, i32 0, i32 0))
+
+  store double* %a, double** %a.addr, align 8
+  store i32 %size, i32* %size.addr, align 4
+
+  ; FIXME: The initialization of the'random' variable should end up between the issue and wait calls.
+  ;        Note that this might be more code, but for simplicity only this variable is used here.
+  %call = call i32 @rand()
+  %rem = srem i32 %call, 7
+  store i32 %rem, i32* %random, align 4
+
+  ; FIXME: This argument setup and call to __tgt_target_data_begin should be moved above,
+  ;        changing __tgt_target_data_begin for to __tgt_target_data_begin_issue.
+  %0 = load double*, double** %a.addr, align 8
+  %1 = load double*, double** %a.addr, align 8
+  %arrayidx = getelementptr inbounds double, double* %1, i64 0
+  %2 = load i32, i32* %size.addr, align 4
+  %conv = zext i32 %2 to i64
+  %3 = mul nuw i64 %conv, 8
+  %4 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_baseptrs, i32 0, i32 0
+  %5 = bitcast i8** %4 to double**
+  store double* %0, double** %5, align 8
+  %6 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_ptrs, i32 0, i32 0
+  %7 = bitcast i8** %6 to double**
+  store double* %arrayidx, double** %7, align 8
+  %8 = getelementptr inbounds [2 x i64], [2 x i64]* %.offload_sizes, i32 0, i32 0
+  store i64 %3, i64* %8, align 8
+  %9 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_baseptrs, i32 0, i32 1
+  %10 = bitcast i8** %9 to i32**
+  store i32* %size.addr, i32** %10, align 8
+  %11 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_ptrs, i32 0, i32 1
+  %12 = bitcast i8** %11 to i32**
+  store i32* %size.addr, i32** %12, align 8
+  %13 = getelementptr inbounds [2 x i64], [2 x i64]* %.offload_sizes, i32 0, i32 1
+  store i64 4, i64* %13, align 8
+  %14 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_baseptrs, i32 0, i32 0
+  %15 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_ptrs, i32 0, i32 0
+  %16 = getelementptr inbounds [2 x i64], [2 x i64]* %.offload_sizes, i32 0, i32 0
+  call void @__tgt_target_data_begin(i64 -1, i32 2, i8** %14, i8** %15, i64* %16, i64* getelementptr inbounds ([2 x i64], [2 x i64]* @.offload_maptypes.3, i32 0, i32 0))
+
+  ; FIXME: The initialization of the 'random' variable should end up here.
+  ; %call = call i32 @rand()
+  ; %rem = srem i32 %call, 7
+  ; store i32 %rem, i32* %random, align 4
+
+  %17 = load i32, i32* %size.addr, align 4
+  %conv1 = bitcast i64* %size.casted to i32*
+  store i32 %17, i32* %conv1, align 4
+  %18 = load i64, i64* %size.casted, align 8
+  %19 = load double*, double** %a.addr, align 8
+  %20 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_baseptrs2, i32 0, i32 0
+  %21 = bitcast i8** %20 to i64*
+  store i64 %18, i64* %21, align 8
+  %22 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_ptrs3, i32 0, i32 0
+  %23 = bitcast i8** %22 to i64*
+  store i64 %18, i64* %23, align 8
+  %24 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_baseptrs2, i32 0, i32 1
+  %25 = bitcast i8** %24 to double**
+  store double* %19, double** %25, align 8
+  %26 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_ptrs3, i32 0, i32 1
+  %27 = bitcast i8** %26 to double**
+  store double* %19, double** %27, align 8
+  %28 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_baseptrs2, i32 0, i32 0
+  %29 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_ptrs3, i32 0, i32 0
+
+  ; FIXME: A fter splitting the runtime call __tgt_target_data_begin, this argument setup and call to
+  ;        __tgt_target_data_wait should be moved here, right before the kernel execution.
+  ; %device_id1_copy = load i64, i64* %device_id1, align 8 ; device_id
+  ; %queue1 = getelementptr inbounds %struct.__tgt_async_info, %struct.__tgt_async_info* %async_info1, i32 0, i32 0
+  ; store i8* %handle1, i8** %queue1, align 8
+  ; call void @__tgt_target_data_begin_wait(i64 %device_id1_copy, %struct.__tgt_async_info* dereferenceable(8) %async_info1)
+
+  %30 = call i32 @__tgt_target_teams(i64 -1, i8* @.__omp_offloading_heavyComputation2.region_id, i32 2, i8** %28, i8** %29, i64* getelementptr inbounds ([2 x i64], [2 x i64]* @.offload_sizes.4, i32 0, i32 0), i64* getelementptr inbounds ([2 x i64], [2 x i64]* @.offload_maptypes.4, i32 0, i32 0), i32 0, i32 0)
+  %31 = icmp ne i32 %30, 0
+  br i1 %31, label %omp_offload.failed, label %omp_offload.cont
+
+omp_offload.failed:                               ; preds = %entry
+  call void @heavyComputation2FallBack(i64 %18, double* %19)
+  br label %omp_offload.cont
+
+omp_offload.cont:                                 ; preds = %omp_offload.failed, %entry
+  %32 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_baseptrs, i32 0, i32 0
+  %33 = getelementptr inbounds [2 x i8*], [2 x i8*]* %.offload_ptrs, i32 0, i32 0
+  %34 = getelementptr inbounds [2 x i64], [2 x i64]* %.offload_sizes, i32 0, i32 0
+  call void @__tgt_target_data_end(i64 -1, i32 2, i8** %32, i8** %33, i64* %34, i64* getelementptr inbounds ([2 x i64], [2 x i64]* @.offload_maptypes.3, i32 0, i32 0))
+  %35 = load i32, i32* %random, align 4
+  ret i32 %35
+}
+
+define internal void @heavyComputation2FallBack(i64 %size, double* %a) {
+entry:
+  ; Fallback for offloading function heavyComputation1.
+  ret void
+}
+
+declare dso_local void @__tgt_target_data_begin(i64, i32, i8**, i8**, i64*, i64*)
+
+declare dso_local i32 @__tgt_target_teams(i64, i8*, i32, i8**, i8**, i64*, i64*, i32, i32)
+
+declare dso_local void @__tgt_target_data_end(i64, i32, i8**, i8**, i64*, i64*)
+
+declare dso_local i32 @rand()
+
+; FIXME: These two function declarations must be generated after splitting the runtime function
+;        __tgt_target_data_begin.
+; declare dso_local i8* @__tgt_target_data_begin_issue(i64* dereferenceable(8), i32, i8**, i8**, i64*, i64*)
+; declare dso_local void @__tgt_target_data_begin_wait(i64, %struct.__tgt_async_info* dereferenceable(8))
\ No newline at end of file