diff --git a/hpvm/test/regressionTests/ClearDFG/ThreeLevel.x86.ll b/hpvm/test/regressionTests/ClearDFG/ThreeLevel.CPU.ll
similarity index 93%
rename from hpvm/test/regressionTests/ClearDFG/ThreeLevel.x86.ll
rename to hpvm/test/regressionTests/ClearDFG/ThreeLevel.CPU.ll
index 2bd337e8ab4ad4f36e65473f7a9d4d15c7d03e85..460dd15b6b1f6dd38483a18e899f0d96b68cac08 100644
--- a/hpvm/test/regressionTests/ClearDFG/ThreeLevel.x86.ll
+++ b/hpvm/test/regressionTests/ClearDFG/ThreeLevel.CPU.ll
@@ -1,4 +1,4 @@
-; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -S -dfg2llvm-x86 -clearDFG <  %s | FileCheck %s
+; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_CPU.so -load LLVMClearDFG.so -S -dfg2llvm-cpu -clearDFG <  %s | FileCheck %s
 ; ModuleID = 'ThreeLevel.ll'
 source_filename = "ThreeLevel.c"
 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
@@ -13,31 +13,31 @@ target triple = "x86_64-unknown-linux-gnu"
 
 ; CHECK-LABEL: i32 @main(
 ; CHECK-NOT: call void @llvm.hpvm.init()
-; CHECK: call i8* @llvm_hpvm_x86_launch(i8* (i8*)* @LaunchDataflowGraph, i8*
+; CHECK: call i8* @llvm_hpvm_cpu_launch(i8* (i8*)* @LaunchDataflowGraph, i8*
 ; CHECK-NOT: call i8* @llvm.hpvm.launch(i8*
-; CHECK: call void @llvm_hpvm_x86_wait(i8*
+; CHECK: call void @llvm_hpvm_cpu_wait(i8*
 
 ; CHECK-LABEL: @Func1_cloned.1_cloned_cloned_cloned_cloned_cloned_cloned
-; CHECK: call i8* @llvm_hpvm_x86_argument_ptr(
+; CHECK: call i8* @llvm_hpvm_cpu_argument_ptr(
 
 ; CHECK-LABEL: @Func3_cloned.2_cloned_cloned_cloned_cloned_cloned_cloned(
 ; CHECK-LABEL: for.body1:
 ; CHECK: %index.y = phi i64 [ 0, %for.body ], [ %index.y.inc, %for.body1 ]
-; CHECK-NEXT: call void @llvm_hpvm_x86_dstack_push(
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_push(
 ; CHECK-NEXT: @Func1_cloned.1_cloned_cloned_cloned_cloned_cloned_cloned(
-; CHECK-NEXT: call void @llvm_hpvm_x86_dstack_pop()
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop()
 
 ; CHECK-LABEL: @Func2_cloned.3_cloned_cloned_cloned_cloned_cloned_cloned(
 ; CHECK-LABEL: for.body:
 ; CHECK-NEXT: %index.x = phi i64 [ 0, %entry ], [ %index.x.inc, %for.body ]
-; CHECK-NEXT: call void @llvm_hpvm_x86_dstack_push(
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_push(
 ; CHECK-NEXT: @Func3_cloned.2_cloned_cloned_cloned_cloned_cloned_cloned(
-; CHECK-NEXT: call void @llvm_hpvm_x86_dstack_pop()
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop()
 
 ; CHECK-LABEL: @PipeRoot_cloned.4(
-; CHECK: call void @llvm_hpvm_x86_dstack_push(
+; CHECK: call void @llvm_hpvm_cpu_dstack_push(
 ; CHECK-NEXT: @Func2_cloned.3_cloned_cloned_cloned_cloned_cloned_cloned(
-; CHECK-NEXT: call void @llvm_hpvm_x86_dstack_pop()
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop()
 
 ; CHECK-LABEL: @LaunchDataflowGraph(
 ; CHECK: call %struct.out.PipeRoot @PipeRoot_cloned.4(
@@ -189,9 +189,9 @@ declare void @llvm.hpvm.wait(i8*) #3
 ; Function Attrs: nounwind
 declare void @llvm.hpvm.cleanup() #3
 
-attributes #0 = { "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #0 = { "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
 attributes #1 = { argmemonly nounwind }
-attributes #2 = { nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #2 = { nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
 attributes #3 = { nounwind }
 
 !llvm.module.flags = !{!0}
diff --git a/hpvm/test/regressionTests/ClearDFG/ThreeLevel.nvptxTox86.ll b/hpvm/test/regressionTests/ClearDFG/ThreeLevel.OpenCLToCPU.ll
similarity index 95%
rename from hpvm/test/regressionTests/ClearDFG/ThreeLevel.nvptxTox86.ll
rename to hpvm/test/regressionTests/ClearDFG/ThreeLevel.OpenCLToCPU.ll
index 163c4fc93ce275feec8b8db732bb862e289c3dae..5fc96483927bb2328fa63517ff094de3ff3bbcee 100644
--- a/hpvm/test/regressionTests/ClearDFG/ThreeLevel.nvptxTox86.ll
+++ b/hpvm/test/regressionTests/ClearDFG/ThreeLevel.OpenCLToCPU.ll
@@ -1,4 +1,4 @@
-; RUN: opt -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_NVPTX.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -S -localmem -dfg2llvm-nvptx -dfg2llvm-x86 -clearDFG <  %s | FileCheck %s
+; RUN: opt -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_OpenCL.so -load LLVMDFG2LLVM_CPU.so -load LLVMClearDFG.so -S -localmem -dfg2llvm-opencl -dfg2llvm-cpu -clearDFG <  %s | FileCheck %s
 ; ModuleID = 'ThreeLevel.ll'
 source_filename = "ThreeLevel.c"
 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
@@ -37,9 +37,9 @@ declare dso_local i8* @__hpvm__createNodeND(i32, ...) local_unnamed_addr #0
 declare dso_local void @__hpvm__bindIn(i8*, i32, i32, i32) local_unnamed_addr #0
 
 ; CHECK-LABEL: @Launch(
-; CHECK: call i8* @llvm_hpvm_x86_launch(i8*
+; CHECK: call i8* @llvm_hpvm_cpu_launch(i8*
 ; CHECK-NOT: call i8* @llvm.hpvm.launch(i8*
-; CHECK: call void @llvm_hpvm_x86_wait(i8*
+; CHECK: call void @llvm_hpvm_cpu_wait(i8*
 
 ; Function Attrs: noinline nounwind uwtable
 define dso_local void @Launch() local_unnamed_addr #2 {
@@ -90,9 +90,9 @@ declare dso_local void @__hpvm__wait(i8*) local_unnamed_addr #0
 ; CHECK-NEXT: call void @llvm_hpvm_ocl_free(i8*
 
 ; CHECK-LABEL: @PipeRoot_cloned.3(
-; CHECK: call void @llvm_hpvm_x86_dstack_push(
+; CHECK: call void @llvm_hpvm_cpu_dstack_push(
 ; CHECK-NEXT: @Func2_cloned.2_cloned_cloned_cloned_cloned_cloned_clone
-; CHECK-NEXT: call void @llvm_hpvm_x86_dstack_pop(
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop(
 
 ; CHECK-LABEL: define i8* @LaunchDataflowGraph(i8*
 
@@ -219,11 +219,11 @@ declare void @llvm.hpvm.init() #6
 ; Function Attrs: nounwind
 declare void @llvm.hpvm.cleanup() #6
 
-attributes #0 = { "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #0 = { "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
 attributes #1 = { argmemonly nounwind }
-attributes #2 = { noinline nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
-attributes #3 = { nofree nounwind "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
-attributes #4 = { nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #2 = { noinline nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #3 = { nofree nounwind "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #4 = { nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
 attributes #5 = { nounwind readnone }
 attributes #6 = { nounwind }
 
diff --git a/hpvm/test/regressionTests/ClearDFG/ThreeLevel.constmem.nvptxTox86.ll b/hpvm/test/regressionTests/ClearDFG/ThreeLevel.constmem.OpenCLToCPU.ll
similarity index 95%
rename from hpvm/test/regressionTests/ClearDFG/ThreeLevel.constmem.nvptxTox86.ll
rename to hpvm/test/regressionTests/ClearDFG/ThreeLevel.constmem.OpenCLToCPU.ll
index f6f0af5433bc5e37627f85ccd5b3be3a69465daa..af3c2cf9ffef654c002f7d261a622db9f09d6a58 100644
--- a/hpvm/test/regressionTests/ClearDFG/ThreeLevel.constmem.nvptxTox86.ll
+++ b/hpvm/test/regressionTests/ClearDFG/ThreeLevel.constmem.OpenCLToCPU.ll
@@ -1,4 +1,4 @@
-; RUN: opt -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_NVPTX.so -load LLVMDFG2LLVM_X86.so -load LLVMClearDFG.so -S -localmem -dfg2llvm-nvptx -dfg2llvm-x86 -clearDFG <  %s | FileCheck %s
+; RUN: opt -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_OpenCL.so -load LLVMDFG2LLVM_CPU.so -load LLVMClearDFG.so -S -localmem -dfg2llvm-opencl -dfg2llvm-cpu -clearDFG <  %s | FileCheck %s
 ; ModuleID = 'ThreeLevel.opt.ll'
 source_filename = "ThreeLevel.opt.c"
 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
@@ -44,9 +44,9 @@ declare dso_local void @__hpvm__bindIn(i8*, i32, i32, i32) local_unnamed_addr #0
 declare dso_local i8* @__hpvm__edge(i8*, i8*, i32, i32, i32, i32) local_unnamed_addr #0
 
 ; CHECK-LABEL: @Launch(
-; CHECK: call i8* @llvm_hpvm_x86_launch(i8*
+; CHECK: call i8* @llvm_hpvm_cpu_launch(i8*
 ; CHECK-NOT: call i8* @llvm.hpvm.launch(i8*
-; CHECK: call void @llvm_hpvm_x86_wait(i8*
+; CHECK: call void @llvm_hpvm_cpu_wait(i8*
 
 ; Function Attrs: noinline nounwind uwtable
 define dso_local void @Launch() local_unnamed_addr #2 {
@@ -96,9 +96,9 @@ declare dso_local void @__hpvm__wait(i8*) local_unnamed_addr #0
 ; CHECK-NEXT: call void @llvm_hpvm_ocl_free(i8*
 
 ; CHECK-LABEL: @PipeRoot_cloned.4(
-; CHECK: call void @llvm_hpvm_x86_dstack_push(
+; CHECK: call void @llvm_hpvm_cpu_dstack_push(
 ; CHECK-NEXT: @Func2_cloned.3_cloned_cloned_cloned_cloned_cloned_clone
-; CHECK-NEXT: call void @llvm_hpvm_x86_dstack_pop(
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop(
 
 ; CHECK-LABEL: define i8* @LaunchDataflowGraph(i8*
 
@@ -239,11 +239,11 @@ declare void @llvm.hpvm.init() #6
 ; Function Attrs: nounwind
 declare void @llvm.hpvm.cleanup() #6
 
-attributes #0 = { "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #0 = { "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
 attributes #1 = { argmemonly nounwind }
-attributes #2 = { noinline nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
-attributes #3 = { nofree nounwind "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
-attributes #4 = { nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #2 = { noinline nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #3 = { nofree nounwind "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #4 = { nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
 attributes #5 = { nounwind readnone }
 attributes #6 = { nounwind }
 
diff --git a/hpvm/test/regressionTests/DFG2LLVM_CPU/OneLevel.codeGen.ll b/hpvm/test/regressionTests/DFG2LLVM_CPU/OneLevel.codeGen.ll
new file mode 100644
index 0000000000000000000000000000000000000000..85dedaaaab83edf19b96efca97dbed58c4403731
--- /dev/null
+++ b/hpvm/test/regressionTests/DFG2LLVM_CPU/OneLevel.codeGen.ll
@@ -0,0 +1,182 @@
+; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_CPU.so -S -dfg2llvm-cpu <  %s | FileCheck %s
+; ModuleID = 'CreateNode.ll'
+source_filename = "CreateNode.c"
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+%struct.Root = type { i32*, i64, i32*, i64, i32*, i64 }
+%struct.out.Func = type <{ i32* }>
+%struct.out.PipeRoot = type <{ i32* }>
+
+; CHECK-LABEL: i32 @main(
+; CHECK: call void @llvm.hpvm.init()
+; CHECK: call i8* @llvm_hpvm_cpu_launch(i8* (i8*)* @LaunchDataflowGraph, i8*
+; CHECK-NEXT: call i8* @llvm.hpvm.launch(i8*
+; CHECK-NEXT: call void @llvm_hpvm_cpu_wait(i8*
+
+; CHECK-LABEL: @PipeRoot_cloned(
+; CHECK: call i8* @llvm.hpvm.createNode(
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.output(i8* %Func_cloned.node
+
+; CHECK-LABEL: @Func_cloned.1_cloned_cloned_cloned_cloned_cloned_cloned
+; CHECK: call i8* @llvm_hpvm_cpu_argument_ptr(
+
+; CHECK-LABEL: @PipeRoot_cloned.2(
+; CHECK: call void @llvm_hpvm_cpu_dstack_push(
+; CHECK-NEXT: @Func_cloned.1_cloned_cloned_cloned_cloned_cloned_cloned(
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop()
+
+; CHECK-LABEL: @LaunchDataflowGraph(i8*
+; call %struct.out.PipeRoot @PipeRoot_cloned.2(
+
+
+declare dso_local void @__hpvm__hint(i32) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__attributes(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__return(i32, ...) local_unnamed_addr #0
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture) #1
+
+declare dso_local i8* @__hpvm__createNodeND(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__bindIn(i8*, i32, i32, i32) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__bindOut(i8*, i32, i32, i32) local_unnamed_addr #0
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture) #1
+
+; Function Attrs: nounwind uwtable
+define dso_local i32 @main() local_unnamed_addr #2 {
+entry:
+  %In1 = alloca i32, align 4
+  %In2 = alloca i32, align 4
+  %Out = alloca i32, align 4
+  %RootArgs = alloca %struct.Root, align 8
+  %0 = bitcast i32* %In1 to i8*
+  call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %0) #3
+  store i32 1, i32* %In1, align 4, !tbaa !4
+  %1 = bitcast i32* %In2 to i8*
+  call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %1) #3
+  store i32 2, i32* %In2, align 4, !tbaa !4
+  %2 = bitcast i32* %Out to i8*
+  call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %2) #3
+  store i32 0, i32* %Out, align 4, !tbaa !4
+  %3 = bitcast %struct.Root* %RootArgs to i8*
+  call void @llvm.lifetime.start.p0i8(i64 48, i8* nonnull %3) #3
+  %input1 = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 0
+  store i32* %In1, i32** %input1, align 8, !tbaa !8
+  %Insize1 = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 1
+  store i64 32, i64* %Insize1, align 8, !tbaa !12
+  %input2 = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 2
+  store i32* %In2, i32** %input2, align 8, !tbaa !13
+  %Insize2 = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 3
+  store i64 32, i64* %Insize2, align 8, !tbaa !14
+  %output = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 4
+  store i32* %Out, i32** %output, align 8, !tbaa !15
+  %Outsize = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 5
+  store i64 32, i64* %Outsize, align 8, !tbaa !16
+  call void @llvm.hpvm.init()
+  %4 = bitcast %struct.Root* %RootArgs to i8*
+  %graphID = call i8* @llvm.hpvm.launch(i8* bitcast (%struct.out.PipeRoot (i32*, i64, i32*, i64, i32*, i64)* @PipeRoot_cloned to i8*), i8* %4, i1 false)
+  call void @llvm.hpvm.wait(i8* %graphID)
+  call void @llvm.hpvm.cleanup()
+  call void @llvm.lifetime.end.p0i8(i64 48, i8* nonnull %3) #3
+  call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %2) #3
+  call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %1) #3
+  call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %0) #3
+  ret i32 0
+}
+
+declare dso_local void @__hpvm__init(...) local_unnamed_addr #0
+
+declare dso_local i8* @__hpvm__launch(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__wait(i8*) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__cleanup(...) local_unnamed_addr #0
+
+declare i8* @llvm_hpvm_initializeTimerSet()
+
+declare void @llvm_hpvm_switchToTimer(i8**, i32)
+
+declare void @llvm_hpvm_printTimerSet(i8**, i8*)
+
+; Function Attrs: nounwind uwtable
+define dso_local %struct.out.Func @Func_cloned(i32* in %In, i64 %Insize, i32* out %Out, i64 %Outsize) #2 {
+entry:
+  %returnStruct = insertvalue %struct.out.Func undef, i32* %Out, 0
+  ret %struct.out.Func %returnStruct
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.createNode(i8*) #3
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.bind.input(i8*, i32, i32, i1) #3
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.bind.output(i8*, i32, i32, i1) #3
+
+; Function Attrs: nounwind uwtable
+define dso_local %struct.out.PipeRoot @PipeRoot_cloned(i32* in %In1, i64 %Insize1, i32* in %In2, i64 %InSize2, i32* out %Out, i64 %Outsize) #2 {
+entry:
+  %Func_cloned.node = call i8* @llvm.hpvm.createNode(i8* bitcast (%struct.out.Func (i32*, i64, i32*, i64)* @Func_cloned to i8*))
+  call void @llvm.hpvm.bind.input(i8* %Func_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func_cloned.node, i32 3, i32 3, i1 false)
+  call void @llvm.hpvm.bind.output(i8* %Func_cloned.node, i32 0, i32 0, i1 false)
+  ret %struct.out.PipeRoot undef
+}
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.init() #3
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.launch(i8*, i8*, i1) #3
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.wait(i8*) #3
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.cleanup() #3
+
+attributes #0 = { "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #1 = { argmemonly nounwind }
+attributes #2 = { nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #3 = { nounwind }
+
+!llvm.module.flags = !{!0}
+!llvm.ident = !{!1}
+!hpvm_hint_cpu = !{!2, !3}
+!hpvm_hint_gpu = !{}
+!hpvm_hint_spir = !{}
+!hpvm_hint_cudnn = !{}
+!hpvm_hint_promise = !{}
+!hpvm_hint_cpu_gpu = !{}
+!hpvm_hint_cpu_spir = !{}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{!"clang version 9.0.0 (https://gitlab.engr.illinois.edu/llvm/hpvm.git 6690f9e7e8b46b96aea222d3e85315cd63545953)"}
+!2 = !{%struct.out.Func (i32*, i64, i32*, i64)* @Func_cloned}
+!3 = !{%struct.out.PipeRoot (i32*, i64, i32*, i64, i32*, i64)* @PipeRoot_cloned}
+!4 = !{!5, !5, i64 0}
+!5 = !{!"int", !6, i64 0}
+!6 = !{!"omnipotent char", !7, i64 0}
+!7 = !{!"Simple C/C++ TBAA"}
+!8 = !{!9, !10, i64 0}
+!9 = !{!"Root", !10, i64 0, !11, i64 8, !10, i64 16, !11, i64 24, !10, i64 32, !11, i64 40}
+!10 = !{!"any pointer", !6, i64 0}
+!11 = !{!"long", !6, i64 0}
+!12 = !{!9, !11, i64 8}
+!13 = !{!9, !10, i64 16}
+!14 = !{!9, !11, i64 24}
+!15 = !{!9, !10, i64 32}
+!16 = !{!9, !11, i64 40}
diff --git a/hpvm/test/regressionTests/DFG2LLVM_CPU/OneRootBasic.ll b/hpvm/test/regressionTests/DFG2LLVM_CPU/OneRootBasic.ll
new file mode 100644
index 0000000000000000000000000000000000000000..5ff1dffba26f6a92029e914016bd7a07e0619159
--- /dev/null
+++ b/hpvm/test/regressionTests/DFG2LLVM_CPU/OneRootBasic.ll
@@ -0,0 +1,121 @@
+; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_CPU.so -S -dfg2llvm-cpu <  %s | FileCheck %s
+; ModuleID = 'oneLaunchAlloca.ll'
+source_filename = "oneLaunchAlloca.c"
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+%struct.Root = type { i32*, i64, i32*, i64 }
+%emptyStruct = type <{}>
+
+declare dso_local void @__hpvm__hint(i32) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__attributes(i32, ...) local_unnamed_addr #0
+
+; CHECK-LABEL: i32 @main(
+; CHECK: call void @llvm.hpvm.init()
+; CHECK: call i8* @llvm_hpvm_cpu_launch(i8* (i8*)* @LaunchDataflowGraph, i8*
+; CHECK-NEXT: call i8* @llvm.hpvm.launch(i8*
+; CHECK-NEXT: call void @llvm_hpvm_cpu_wait(i8*
+
+; CHECK-LABEL: @PipeRoot_cloned.1(
+
+; CHECK-LABEL: @LaunchDataflowGraph(i8*
+; call %struct.out.PipeRoot @PipeRoot_cloned.1(
+
+
+; Function Attrs: nounwind uwtable
+define dso_local i32 @main() local_unnamed_addr #1 {
+entry:
+  %In = alloca i32, align 4
+  %Out = alloca i32, align 4
+  %RootArgs = alloca %struct.Root, align 8
+  %0 = bitcast i32* %In to i8*
+  call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %0) #3
+  %1 = bitcast i32* %Out to i8*
+  call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %1) #3
+  %2 = bitcast %struct.Root* %RootArgs to i8*
+  call void @llvm.lifetime.start.p0i8(i64 32, i8* nonnull %2) #3
+  %input = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 0
+  store i32* %In, i32** %input, align 8, !tbaa !3
+  %insize = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 1
+  store i64 32, i64* %insize, align 8, !tbaa !9
+  %output = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 2
+  store i32* %Out, i32** %output, align 8, !tbaa !10
+  %outsize = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 3
+  store i64 32, i64* %outsize, align 8, !tbaa !11
+  call void @llvm.hpvm.init()
+  %3 = bitcast %struct.Root* %RootArgs to i8*
+  %graphID = call i8* @llvm.hpvm.launch(i8* bitcast (%emptyStruct (i32*, i64, i32*, i64)* @PipeRoot_cloned to i8*), i8* %3, i1 false)
+  call void @llvm.hpvm.wait(i8* %graphID)
+  call void @llvm.hpvm.cleanup()
+  call void @llvm.lifetime.end.p0i8(i64 32, i8* nonnull %2) #3
+  call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %1) #3
+  call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %0) #3
+  ret i32 0
+}
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture) #2
+
+declare dso_local void @__hpvm__init(...) local_unnamed_addr #0
+
+declare dso_local i8* @__hpvm__launch(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__wait(i8*) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__cleanup(...) local_unnamed_addr #0
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture) #2
+
+declare i8* @llvm_hpvm_initializeTimerSet()
+
+declare void @llvm_hpvm_switchToTimer(i8**, i32)
+
+declare void @llvm_hpvm_printTimerSet(i8**, i8*)
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.init() #3
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.launch(i8*, i8*, i1) #3
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct @PipeRoot_cloned(i32* in %In, i64 %insize, i32* out %Out, i64 %outsize) #1 {
+entry:
+  ret %emptyStruct undef
+}
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.wait(i8*) #3
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.cleanup() #3
+
+attributes #0 = { "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #1 = { nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #2 = { argmemonly nounwind }
+attributes #3 = { nounwind }
+
+!llvm.module.flags = !{!0}
+!llvm.ident = !{!1}
+!hpvm_hint_cpu = !{!2}
+!hpvm_hint_gpu = !{}
+!hpvm_hint_spir = !{}
+!hpvm_hint_cudnn = !{}
+!hpvm_hint_promise = !{}
+!hpvm_hint_cpu_gpu = !{}
+!hpvm_hint_cpu_spir = !{}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{!"clang version 9.0.0 (https://gitlab.engr.illinois.edu/llvm/hpvm.git 6690f9e7e8b46b96aea222d3e85315cd63545953)"}
+!2 = !{%emptyStruct (i32*, i64, i32*, i64)* @PipeRoot_cloned}
+!3 = !{!4, !5, i64 0}
+!4 = !{!"Root", !5, i64 0, !8, i64 8, !5, i64 16, !8, i64 24}
+!5 = !{!"any pointer", !6, i64 0}
+!6 = !{!"omnipotent char", !7, i64 0}
+!7 = !{!"Simple C/C++ TBAA"}
+!8 = !{!"long", !6, i64 0}
+!9 = !{!4, !8, i64 8}
+!10 = !{!4, !5, i64 16}
+!11 = !{!4, !8, i64 24}
diff --git a/hpvm/test/regressionTests/DFG2LLVM_CPU/ThreeLevel.OpenCL.ll b/hpvm/test/regressionTests/DFG2LLVM_CPU/ThreeLevel.OpenCL.ll
new file mode 100644
index 0000000000000000000000000000000000000000..aea2545e44b6ffaad40f28a1b0f1bad570fae775
--- /dev/null
+++ b/hpvm/test/regressionTests/DFG2LLVM_CPU/ThreeLevel.OpenCL.ll
@@ -0,0 +1,250 @@
+; RUN: opt -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_OpenCL.so -load LLVMDFG2LLVM_CPU.so -S -localmem -dfg2llvm-opencl -dfg2llvm-cpu <  %s | FileCheck %s
+; ModuleID = 'ThreeLevel.ll'
+source_filename = "ThreeLevel.c"
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+%struct.Root = type { i32*, i64, i32*, i64 }
+%emptyStruct = type <{}>
+%emptyStruct.0 = type <{}>
+%emptyStruct.1 = type <{}>
+%emptyStruct.2 = type <{}>
+
+declare dso_local void @__hpvm__hint(i32) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__attributes(i32, ...) local_unnamed_addr #0
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture) #1
+
+declare dso_local i8* @__hpvm__getNode(...) local_unnamed_addr #0
+
+declare dso_local i8* @__hpvm__getParentNode(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNodeInstanceID_x(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNodeInstanceID_y(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNumNodeInstances_x(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNumNodeInstances_y(i8*) local_unnamed_addr #0
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture) #1
+
+declare dso_local i8* @__hpvm__createNodeND(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__bindIn(i8*, i32, i32, i32) local_unnamed_addr #0
+
+; CHECK-LABEL: @Launch(
+; CHECK: call i8* @llvm_hpvm_cpu_launch(i8*
+; CHECK-NEXT: call i8* @llvm.hpvm.launch(i8*
+; CHECK-NEXT: call void @llvm_hpvm_cpu_wait(i8*
+
+; Function Attrs: noinline nounwind uwtable
+define dso_local void @Launch() local_unnamed_addr #2 {
+entry:
+  %RootArgs = alloca %struct.Root, align 8
+  %0 = bitcast %struct.Root* %RootArgs to i8*
+  call void @llvm.lifetime.start.p0i8(i64 32, i8* nonnull %0) #6
+  %call = tail call noalias i8* @malloc(i64 1024) #6
+  %1 = bitcast %struct.Root* %RootArgs to i8**
+  store i8* %call, i8** %1, align 8, !tbaa !6
+  %Insize = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 1
+  store i64 1024, i64* %Insize, align 8, !tbaa !12
+  %output = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 2
+  %call1 = tail call noalias i8* @malloc(i64 1024) #6
+  %2 = bitcast i32** %output to i8**
+  store i8* %call1, i8** %2, align 8, !tbaa !13
+  %Outsize = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 3
+  store i64 1024, i64* %Outsize, align 8, !tbaa !14
+  %3 = bitcast %struct.Root* %RootArgs to i8*
+  %graphID = call i8* @llvm.hpvm.launch(i8* bitcast (%emptyStruct.2 (i32*, i64, i32*, i64)* @PipeRoot_cloned to i8*), i8* %3, i1 false)
+  call void @llvm.hpvm.wait(i8* %graphID)
+  call void @llvm.lifetime.end.p0i8(i64 32, i8* nonnull %0) #6
+  ret void
+}
+
+; Function Attrs: nofree nounwind
+declare dso_local noalias i8* @malloc(i64) local_unnamed_addr #3
+
+declare dso_local i8* @__hpvm__launch(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__wait(i8*) local_unnamed_addr #0
+
+; CHECK-LABEL: @main(
+; CHECK: call i8* @llvm_hpvm_ocl_initContext(i32
+; CHECK: call i8* @llvm_hpvm_ocl_launch(i8*
+; CHECK: call void @llvm_hpvm_ocl_clearContext(i8*
+
+; CHECK-LABEL: @Func2_cloned.2_cloned_cloned_cloned_cloned_cloned_cloned(
+; CHECK: call i8* @llvm_hpvm_ocl_argument_ptr(i8*
+; CHECK: call void @llvm_hpvm_ocl_argument_scalar(i8*
+; CHECK: call i8* @llvm_hpvm_ocl_argument_ptr(i8*
+; CHECK: call void @llvm_hpvm_ocl_argument_scalar(i8*
+; CHECK: call i8* @llvm_hpvm_ocl_executeNode(i8*
+; CHECK-NEXT: call void @llvm_hpvm_ocl_wait(i8*
+; CHECK-NEXT: call void @llvm_hpvm_ocl_free(i8*
+; CHECK-NEXT: call void @llvm_hpvm_ocl_free(i8*
+
+; CHECK-LABEL: @PipeRoot_cloned.3(
+; CHECK: call void @llvm_hpvm_cpu_dstack_push(
+; CHECK-NEXT: @Func2_cloned.2_cloned_cloned_cloned_cloned_cloned_clone
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop(
+
+; CHECK-LABEL: define i8* @LaunchDataflowGraph(i8*
+
+; Function Attrs: nounwind uwtable
+define dso_local i32 @main() local_unnamed_addr #4 {
+entry:
+  call void @llvm.hpvm.init()
+  tail call void @Launch()
+  call void @llvm.hpvm.cleanup()
+  ret i32 0
+}
+
+declare dso_local void @__hpvm__init(...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__cleanup(...) local_unnamed_addr #0
+
+declare i8* @llvm_hpvm_initializeTimerSet()
+
+declare void @llvm_hpvm_switchToTimer(i8**, i32)
+
+declare void @llvm_hpvm_printTimerSet(i8**, i8*)
+
+; Function Attrs: nounwind readnone
+declare i8* @llvm.hpvm.getNode() #5
+
+; Function Attrs: nounwind readnone
+declare i8* @llvm.hpvm.getParentNode(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNodeInstanceID.x(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNodeInstanceID.y(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNumNodeInstances.x(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNumNodeInstances.y(i8*) #5
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.createNode2D(i8*, i64, i64) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct @Func1_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %call4 = call i8* @llvm.hpvm.getNode()
+  %call15 = call i8* @llvm.hpvm.getParentNode(i8* %call4)
+  %call26 = call i64 @llvm.hpvm.getNodeInstanceID.x(i8* %call4)
+  %call37 = call i64 @llvm.hpvm.getNodeInstanceID.y(i8* %call4)
+  %call58 = call i64 @llvm.hpvm.getNodeInstanceID.x(i8* %call15)
+  %call79 = call i64 @llvm.hpvm.getNodeInstanceID.y(i8* %call15)
+  %call910 = call i64 @llvm.hpvm.getNumNodeInstances.x(i8* %call4)
+  %call1111 = call i64 @llvm.hpvm.getNumNodeInstances.y(i8* %call4)
+  %mul = mul i64 %call910, %call58
+  %add = add i64 %mul, %call26
+  %mul13 = mul i64 %call1111, %call79
+  %add14 = add i64 %mul13, %call37
+  %sext = shl i64 %add14, 32
+  %idxprom = ashr exact i64 %sext, 32
+  %arrayidx = getelementptr inbounds i32, i32* %In, i64 %idxprom
+  %0 = load i32, i32* %arrayidx, align 4, !tbaa !15
+  %sext36 = shl i64 %add, 32
+  %idxprom15 = ashr exact i64 %sext36, 32
+  %arrayidx16 = getelementptr inbounds i32, i32* %Out, i64 %idxprom15
+  %1 = load i32, i32* %arrayidx16, align 4, !tbaa !15
+  %add17 = add nsw i32 %1, %0
+  store i32 %add17, i32* %arrayidx16, align 4, !tbaa !15
+  ret %emptyStruct undef
+}
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.bind.input(i8*, i32, i32, i1) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct.0 @Func3_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %Func1_cloned.node = call i8* @llvm.hpvm.createNode2D(i8* bitcast (%emptyStruct (i32*, i64, i32*, i64)* @Func1_cloned to i8*), i64 3, i64 5)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 3, i32 3, i1 false)
+  ret %emptyStruct.0 undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.createNode(i8*) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct.1 @Func2_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %Func3_cloned.node = call i8* @llvm.hpvm.createNode2D(i8* bitcast (%emptyStruct.0 (i32*, i64, i32*, i64)* @Func3_cloned to i8*), i64 3, i64 5)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 3, i32 3, i1 false)
+  ret %emptyStruct.1 undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.launch(i8*, i8*, i1) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct.2 @PipeRoot_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %Func2_cloned.node = call i8* @llvm.hpvm.createNode(i8* bitcast (%emptyStruct.1 (i32*, i64, i32*, i64)* @Func2_cloned to i8*))
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 3, i32 3, i1 false)
+  ret %emptyStruct.2 undef
+}
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.wait(i8*) #6
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.init() #6
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.cleanup() #6
+
+attributes #0 = { "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #1 = { argmemonly nounwind }
+attributes #2 = { noinline nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #3 = { nofree nounwind "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #4 = { nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #5 = { nounwind readnone }
+attributes #6 = { nounwind }
+
+!llvm.module.flags = !{!0}
+!llvm.ident = !{!1}
+!hpvm_hint_gpu = !{!2}
+!hpvm_hint_cpu = !{!3, !4, !5}
+!hpvm_hint_spir = !{}
+!hpvm_hint_cudnn = !{}
+!hpvm_hint_promise = !{}
+!hpvm_hint_cpu_gpu = !{}
+!hpvm_hint_cpu_spir = !{}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{!"clang version 9.0.0 (https://gitlab.engr.illinois.edu/llvm/hpvm.git 6690f9e7e8b46b96aea222d3e85315cd63545953)"}
+!2 = !{%emptyStruct (i32*, i64, i32*, i64)* @Func1_cloned}
+!3 = !{%emptyStruct.0 (i32*, i64, i32*, i64)* @Func3_cloned}
+!4 = !{%emptyStruct.1 (i32*, i64, i32*, i64)* @Func2_cloned}
+!5 = !{%emptyStruct.2 (i32*, i64, i32*, i64)* @PipeRoot_cloned}
+!6 = !{!7, !8, i64 0}
+!7 = !{!"Root", !8, i64 0, !11, i64 8, !8, i64 16, !11, i64 24}
+!8 = !{!"any pointer", !9, i64 0}
+!9 = !{!"omnipotent char", !10, i64 0}
+!10 = !{!"Simple C/C++ TBAA"}
+!11 = !{!"long", !9, i64 0}
+!12 = !{!7, !11, i64 8}
+!13 = !{!7, !8, i64 16}
+!14 = !{!7, !11, i64 24}
+!15 = !{!16, !16, i64 0}
+!16 = !{!"int", !9, i64 0}
diff --git a/hpvm/test/regressionTests/DFG2LLVM_CPU/ThreeLevel.codeGen.ll b/hpvm/test/regressionTests/DFG2LLVM_CPU/ThreeLevel.codeGen.ll
new file mode 100644
index 0000000000000000000000000000000000000000..457d0c2bc3538e7e8cd94c1481e513da07b48602
--- /dev/null
+++ b/hpvm/test/regressionTests/DFG2LLVM_CPU/ThreeLevel.codeGen.ll
@@ -0,0 +1,246 @@
+; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_CPU.so -S -dfg2llvm-cpu <  %s | FileCheck %s
+; ModuleID = 'ThreeLevel.ll'
+source_filename = "ThreeLevel.c"
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+%struct.Root = type { i32*, i64, i32*, i64, i32*, i64 }
+%struct.out.Func1 = type <{ i32* }>
+%struct.out.Func3 = type <{ i32* }>
+%struct.out.Func2 = type <{ i32* }>
+%struct.out.PipeRoot = type <{ i32* }>
+
+
+; CHECK-LABEL: i32 @main(
+; CHECK: call void @llvm.hpvm.init()
+; CHECK: call i8* @llvm_hpvm_cpu_launch(i8* (i8*)* @LaunchDataflowGraph, i8*
+; CHECK-NEXT: call i8* @llvm.hpvm.launch(i8*
+; CHECK-NEXT: call void @llvm_hpvm_cpu_wait(i8*
+
+; CHECK-LABEL: @Func3_cloned(
+; CHECK: call i8* @llvm.hpvm.createNode2D(
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.output(i8* %Func1_cloned.node
+
+; CHECK-LABEL: @Func2_cloned(
+; CHECK: call i8* @llvm.hpvm.createNode1D(
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.output(i8* %Func3_cloned.node
+
+; CHECK-LABEL: @PipeRoot_cloned(
+; CHECK: call i8* @llvm.hpvm.createNode(
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.output(i8* %Func2_cloned.node
+
+; CHECK-LABEL: @Func1_cloned.1_cloned_cloned_cloned_cloned_cloned_cloned
+; CHECK: call i8* @llvm_hpvm_cpu_argument_ptr(
+
+; CHECK-LABEL: @Func3_cloned.2_cloned_cloned_cloned_cloned_cloned_cloned(
+; CHECK-LABEL: for.body1:
+; CHECK: %index.y = phi i64 [ 0, %for.body ], [ %index.y.inc, %for.body1 ]
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_push(
+; CHECK-NEXT: @Func1_cloned.1_cloned_cloned_cloned_cloned_cloned_cloned(
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop()
+
+; CHECK-LABEL: @Func2_cloned.3_cloned_cloned_cloned_cloned_cloned_cloned(
+; CHECK-LABEL: for.body:
+; CHECK-NEXT: %index.x = phi i64 [ 0, %entry ], [ %index.x.inc, %for.body ]
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_push(
+; CHECK-NEXT: @Func3_cloned.2_cloned_cloned_cloned_cloned_cloned_cloned(
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop()
+
+; CHECK-LABEL: @PipeRoot_cloned.4(
+; CHECK: call void @llvm_hpvm_cpu_dstack_push(
+; CHECK-NEXT: @Func2_cloned.3_cloned_cloned_cloned_cloned_cloned_cloned(
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop()
+
+; CHECK-LABEL: @LaunchDataflowGraph(
+; CHECK: call %struct.out.PipeRoot @PipeRoot_cloned.4(
+
+declare dso_local void @__hpvm__hint(i32) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__attributes(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__return(i32, ...) local_unnamed_addr #0
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture) #1
+
+declare dso_local i8* @__hpvm__createNodeND(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__bindIn(i8*, i32, i32, i32) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__bindOut(i8*, i32, i32, i32) local_unnamed_addr #0
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture) #1
+
+; Function Attrs: nounwind uwtable
+define dso_local i32 @main() local_unnamed_addr #2 {
+entry:
+  %In1 = alloca i32, align 4
+  %In2 = alloca i32, align 4
+  %Out = alloca i32, align 4
+  %RootArgs = alloca %struct.Root, align 8
+  %0 = bitcast i32* %In1 to i8*
+  call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %0) #3
+  store i32 1, i32* %In1, align 4, !tbaa !6
+  %1 = bitcast i32* %In2 to i8*
+  call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %1) #3
+  store i32 2, i32* %In2, align 4, !tbaa !6
+  %2 = bitcast i32* %Out to i8*
+  call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %2) #3
+  store i32 0, i32* %Out, align 4, !tbaa !6
+  %3 = bitcast %struct.Root* %RootArgs to i8*
+  call void @llvm.lifetime.start.p0i8(i64 48, i8* nonnull %3) #3
+  %input1 = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 0
+  store i32* %In1, i32** %input1, align 8, !tbaa !10
+  %Insize1 = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 1
+  store i64 32, i64* %Insize1, align 8, !tbaa !14
+  %input2 = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 2
+  store i32* %In2, i32** %input2, align 8, !tbaa !15
+  %Insize2 = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 3
+  store i64 32, i64* %Insize2, align 8, !tbaa !16
+  %output = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 4
+  store i32* %Out, i32** %output, align 8, !tbaa !17
+  %Outsize = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 5
+  store i64 32, i64* %Outsize, align 8, !tbaa !18
+  call void @llvm.hpvm.init()
+  %4 = bitcast %struct.Root* %RootArgs to i8*
+  %graphID = call i8* @llvm.hpvm.launch(i8* bitcast (%struct.out.PipeRoot (i32*, i64, i32*, i64, i32*, i64)* @PipeRoot_cloned to i8*), i8* %4, i1 false)
+  call void @llvm.hpvm.wait(i8* %graphID)
+  call void @llvm.hpvm.cleanup()
+  call void @llvm.lifetime.end.p0i8(i64 48, i8* nonnull %3) #3
+  call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %2) #3
+  call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %1) #3
+  call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %0) #3
+  ret i32 0
+}
+
+declare dso_local void @__hpvm__init(...) local_unnamed_addr #0
+
+declare dso_local i8* @__hpvm__launch(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__wait(i8*) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__cleanup(...) local_unnamed_addr #0
+
+declare i8* @llvm_hpvm_initializeTimerSet()
+
+declare void @llvm_hpvm_switchToTimer(i8**, i32)
+
+declare void @llvm_hpvm_printTimerSet(i8**, i8*)
+
+; Function Attrs: nounwind uwtable
+define dso_local %struct.out.Func1 @Func1_cloned(i32* in %In, i64 %Insize, i32* out %Out, i64 %Outsize) #2 {
+entry:
+  %returnStruct = insertvalue %struct.out.Func1 undef, i32* %Out, 0
+  ret %struct.out.Func1 %returnStruct
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.createNode2D(i8*, i64, i64) #3
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.bind.input(i8*, i32, i32, i1) #3
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.bind.output(i8*, i32, i32, i1) #3
+
+; Function Attrs: nounwind uwtable
+define dso_local %struct.out.Func3 @Func3_cloned(i32* in %In, i64 %Insize, i32* out %Out, i64 %Outsize) #2 {
+entry:
+  %Func1_cloned.node = call i8* @llvm.hpvm.createNode2D(i8* bitcast (%struct.out.Func1 (i32*, i64, i32*, i64)* @Func1_cloned to i8*), i64 3, i64 5)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 3, i32 3, i1 false)
+  call void @llvm.hpvm.bind.output(i8* %Func1_cloned.node, i32 0, i32 0, i1 false)
+  ret %struct.out.Func3 undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.createNode1D(i8*, i64) #3
+
+; Function Attrs: nounwind uwtable
+define dso_local %struct.out.Func2 @Func2_cloned(i32* in %In, i64 %Insize, i32* out %Out, i64 %Outsize) #2 {
+entry:
+  %Func3_cloned.node = call i8* @llvm.hpvm.createNode1D(i8* bitcast (%struct.out.Func3 (i32*, i64, i32*, i64)* @Func3_cloned to i8*), i64 3)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 3, i32 3, i1 false)
+  call void @llvm.hpvm.bind.output(i8* %Func3_cloned.node, i32 0, i32 0, i1 false)
+  ret %struct.out.Func2 undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.createNode(i8*) #3
+
+; Function Attrs: nounwind uwtable
+define dso_local %struct.out.PipeRoot @PipeRoot_cloned(i32* in %In1, i64 %Insize1, i32* in %In2, i64 %InSize2, i32* out %Out, i64 %Outsize) #2 {
+entry:
+  %Func2_cloned.node = call i8* @llvm.hpvm.createNode(i8* bitcast (%struct.out.Func2 (i32*, i64, i32*, i64)* @Func2_cloned to i8*))
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 3, i32 3, i1 false)
+  call void @llvm.hpvm.bind.output(i8* %Func2_cloned.node, i32 0, i32 0, i1 false)
+  ret %struct.out.PipeRoot undef
+}
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.init() #3
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.launch(i8*, i8*, i1) #3
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.wait(i8*) #3
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.cleanup() #3
+
+attributes #0 = { "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #1 = { argmemonly nounwind }
+attributes #2 = { nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #3 = { nounwind }
+
+!llvm.module.flags = !{!0}
+!llvm.ident = !{!1}
+!hpvm_hint_cpu = !{!2, !3, !4, !5}
+!hpvm_hint_gpu = !{}
+!hpvm_hint_spir = !{}
+!hpvm_hint_cudnn = !{}
+!hpvm_hint_promise = !{}
+!hpvm_hint_cpu_gpu = !{}
+!hpvm_hint_cpu_spir = !{}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{!"clang version 9.0.0 (https://gitlab.engr.illinois.edu/llvm/hpvm.git 6690f9e7e8b46b96aea222d3e85315cd63545953)"}
+!2 = !{%struct.out.Func1 (i32*, i64, i32*, i64)* @Func1_cloned}
+!3 = !{%struct.out.Func3 (i32*, i64, i32*, i64)* @Func3_cloned}
+!4 = !{%struct.out.Func2 (i32*, i64, i32*, i64)* @Func2_cloned}
+!5 = !{%struct.out.PipeRoot (i32*, i64, i32*, i64, i32*, i64)* @PipeRoot_cloned}
+!6 = !{!7, !7, i64 0}
+!7 = !{!"int", !8, i64 0}
+!8 = !{!"omnipotent char", !9, i64 0}
+!9 = !{!"Simple C/C++ TBAA"}
+!10 = !{!11, !12, i64 0}
+!11 = !{!"Root", !12, i64 0, !13, i64 8, !12, i64 16, !13, i64 24, !12, i64 32, !13, i64 40}
+!12 = !{!"any pointer", !8, i64 0}
+!13 = !{!"long", !8, i64 0}
+!14 = !{!11, !13, i64 8}
+!15 = !{!11, !12, i64 16}
+!16 = !{!11, !13, i64 24}
+!17 = !{!11, !12, i64 32}
+!18 = !{!11, !13, i64 40}
diff --git a/hpvm/test/regressionTests/DFG2LLVM_CPU/ThreeLevel.cond.ll b/hpvm/test/regressionTests/DFG2LLVM_CPU/ThreeLevel.cond.ll
new file mode 100644
index 0000000000000000000000000000000000000000..a96357bac630d04ead246e059cc170647fa7a043
--- /dev/null
+++ b/hpvm/test/regressionTests/DFG2LLVM_CPU/ThreeLevel.cond.ll
@@ -0,0 +1,300 @@
+; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_CPU.so -S -dfg2llvm-cpu <  %s | FileCheck %s
+; ModuleID = 'ThreeLevel.cond.ll'
+source_filename = "ThreeLevel.cond.c"
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+%struct.Root = type { i32*, i64, i32*, i64 }
+%emptyStruct = type <{}>
+%emptyStruct.0 = type <{}>
+%emptyStruct.1 = type <{}>
+%emptyStruct.2 = type <{}>
+
+; CHECK-LABEL: @Launch(
+; CHECK: call i8* @llvm_hpvm_cpu_launch(i8* (i8*)* @LaunchDataflowGraph, i8*
+; CHECK-NEXT: call i8* @llvm.hpvm.launch(i8*
+; CHECK-NEXT: call void @llvm_hpvm_cpu_wait(i8*
+
+; CHECK-LABEL: i32 @main(
+; CHECK: call void @llvm.hpvm.init()
+; CHECK: call void @llvm.hpvm.cleanup()
+
+; CHECK-LABEL: @Func3_cloned(
+; CHECK: call i8* @llvm.hpvm.createNode2D(
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node
+
+; CHECK-LABEL: @Func2_cloned(
+; CHECK: call i8* @llvm.hpvm.createNode2D(
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node
+
+; CHECK-LABEL: @PipeRoot_cloned(
+; CHECK: call i8* @llvm.hpvm.createNode(
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node
+
+; CHECK-LABEL: @Func1_cloned.1_cloned_cloned_cloned_cloned_cloned_cloned
+; CHECK: call i8* @llvm_hpvm_cpu_argument_ptr(
+; CHECK: call i8* @llvm_hpvm_cpu_argument_ptr(
+; CHECK: call i64 @llvm_hpvm_cpu_getDimInstance(
+; CHECK: call i64 @llvm_hpvm_cpu_getDimInstance(
+
+
+; CHECK-LABEL: @Func3_cloned.2_cloned_cloned_cloned_cloned_cloned_cloned(
+; CHECK: call void @llvm_hpvm_cpu_dstack_push(
+; CHECK-NEXT: @Func1_cloned.1_cloned_cloned_cloned_cloned_cloned_cloned(
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop()
+; CHECK: br i1 %cond.y, label %for.body1, label %for.end2
+; CHECK-LABEL: for.end2:
+; CHECK:  br i1 %cond.x, label %for.body, label %for.end
+; CHECK-LABEL: for.end:
+
+; CHECK-LABEL: @Func2_cloned.3_cloned_cloned_cloned_cloned_cloned_cloned(
+; CHECK-LABEL: for.body:
+; CHECK-NEXT: %index.x = phi i64 [ 0, %entry ], [ %index.x.inc, %for.end2 ]
+; CHECK-LABEL: for.body1:
+; CHECK-NEXT: %index.y = phi i64 [ 0, %for.body ], [ %index.y.inc, %for.body1 ]
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_push(
+; CHECK-NEXT: @Func3_cloned.2_cloned_cloned_cloned_cloned_cloned_cloned(
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop()
+; CHECK: br i1 %cond.y, label %for.body1, label %for.end2
+; CHECK-LABEL: for.end2:
+; CHECK: br i1 %cond.x, label %for.body, label %for.end
+; CHECK-LABEL: for.end:
+
+; CHECK-LABEL: @PipeRoot_cloned.4(
+; CHECK: call void @llvm_hpvm_cpu_dstack_push(
+; CHECK-NEXT: @Func2_cloned.3_cloned_cloned_cloned_cloned_cloned_cloned(
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop()
+
+; CHECK-LABEL: @LaunchDataflowGraph(
+; CHECK: call %emptyStruct.2 @PipeRoot_cloned.4(
+
+declare dso_local void @__hpvm__hint(i32) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__attributes(i32, ...) local_unnamed_addr #0
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture) #1
+
+declare dso_local i8* @__hpvm__getNode(...) local_unnamed_addr #0
+
+declare dso_local i8* @__hpvm__getParentNode(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNodeInstanceID_x(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNodeInstanceID_y(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNumNodeInstances_x(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNumNodeInstances_y(i8*) local_unnamed_addr #0
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture) #1
+
+declare dso_local i8* @__hpvm__createNodeND(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__bindIn(i8*, i32, i32, i32) local_unnamed_addr #0
+
+; Function Attrs: noinline nounwind uwtable
+define dso_local void @Launch() local_unnamed_addr #2 {
+entry:
+  %RootArgs = alloca %struct.Root, align 8
+  %0 = bitcast %struct.Root* %RootArgs to i8*
+  call void @llvm.lifetime.start.p0i8(i64 32, i8* nonnull %0) #6
+  %call = tail call noalias i8* @malloc(i64 1024) #6
+  %1 = bitcast %struct.Root* %RootArgs to i8**
+  store i8* %call, i8** %1, align 8, !tbaa !6
+  %Insize = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 1
+  store i64 1024, i64* %Insize, align 8, !tbaa !12
+  %output = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 2
+  %call1 = tail call noalias i8* @malloc(i64 1024) #6
+  %2 = bitcast i32** %output to i8**
+  store i8* %call1, i8** %2, align 8, !tbaa !13
+  %Outsize = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 3
+  store i64 1024, i64* %Outsize, align 8, !tbaa !14
+  %3 = bitcast %struct.Root* %RootArgs to i8*
+  %graphID = call i8* @llvm.hpvm.launch(i8* bitcast (%emptyStruct.2 (i32*, i64, i32*, i64)* @PipeRoot_cloned to i8*), i8* %3, i1 false)
+  call void @llvm.hpvm.wait(i8* %graphID)
+  call void @llvm.lifetime.end.p0i8(i64 32, i8* nonnull %0) #6
+  ret void
+}
+
+; Function Attrs: nofree nounwind
+declare dso_local noalias i8* @malloc(i64) local_unnamed_addr #3
+
+declare dso_local i8* @__hpvm__launch(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__wait(i8*) local_unnamed_addr #0
+
+; Function Attrs: nounwind uwtable
+define dso_local i32 @main() local_unnamed_addr #4 {
+entry:
+  call void @llvm.hpvm.init()
+  tail call void @Launch()
+  call void @llvm.hpvm.cleanup()
+  ret i32 0
+}
+
+declare dso_local void @__hpvm__init(...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__cleanup(...) local_unnamed_addr #0
+
+declare i8* @llvm_hpvm_initializeTimerSet()
+
+declare void @llvm_hpvm_switchToTimer(i8**, i32)
+
+declare void @llvm_hpvm_printTimerSet(i8**, i8*)
+
+; Function Attrs: nounwind readnone
+declare i8* @llvm.hpvm.getNode() #5
+
+; Function Attrs: nounwind readnone
+declare i8* @llvm.hpvm.getParentNode(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNodeInstanceID.x(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNodeInstanceID.y(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNumNodeInstances.x(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNumNodeInstances.y(i8*) #5
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.createNode2D(i8*, i64, i64) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct @Func1_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %call4 = call i8* @llvm.hpvm.getNode()
+  %call15 = call i8* @llvm.hpvm.getParentNode(i8* %call4)
+  %call26 = call i64 @llvm.hpvm.getNodeInstanceID.x(i8* %call4)
+  %conv = trunc i64 %call26 to i32
+  %call37 = call i64 @llvm.hpvm.getNodeInstanceID.y(i8* %call4)
+  %conv4 = trunc i64 %call37 to i32
+  %call58 = call i64 @llvm.hpvm.getNodeInstanceID.x(i8* %call15)
+  %conv6 = trunc i64 %call58 to i32
+  %call79 = call i64 @llvm.hpvm.getNodeInstanceID.y(i8* %call15)
+  %conv8 = trunc i64 %call79 to i32
+  %call910 = call i64 @llvm.hpvm.getNumNodeInstances.x(i8* %call4)
+  %conv10 = trunc i64 %call910 to i32
+  %call1111 = call i64 @llvm.hpvm.getNumNodeInstances.y(i8* %call4)
+  %conv12 = trunc i64 %call1111 to i32
+  %mul = mul nsw i32 %conv10, %conv6
+  %add = add nsw i32 %mul, %conv
+  %mul13 = mul nsw i32 %conv12, %conv8
+  %add14 = add nsw i32 %mul13, %conv4
+  %cmp = icmp eq i32 %add, %add14
+  br i1 %cmp, label %if.end, label %if.then
+
+if.then:                                          ; preds = %entry
+  %arrayidx = getelementptr inbounds i32, i32* %In, i64 3
+  %0 = load i32, i32* %arrayidx, align 4, !tbaa !15
+  %idxprom = sext i32 %add to i64
+  %arrayidx16 = getelementptr inbounds i32, i32* %Out, i64 %idxprom
+  %1 = load i32, i32* %arrayidx16, align 4, !tbaa !15
+  %add17 = add nsw i32 %1, %0
+  store i32 %add17, i32* %arrayidx16, align 4, !tbaa !15
+  br label %if.end
+
+if.end:                                           ; preds = %if.then, %entry
+  ret %emptyStruct undef
+}
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.bind.input(i8*, i32, i32, i1) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct.0 @Func3_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %Func1_cloned.node = call i8* @llvm.hpvm.createNode2D(i8* bitcast (%emptyStruct (i32*, i64, i32*, i64)* @Func1_cloned to i8*), i64 3, i64 5)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 3, i32 3, i1 false)
+  ret %emptyStruct.0 undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.createNode(i8*) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct.1 @Func2_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %Func3_cloned.node = call i8* @llvm.hpvm.createNode2D(i8* bitcast (%emptyStruct.0 (i32*, i64, i32*, i64)* @Func3_cloned to i8*), i64 3, i64 5)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 3, i32 3, i1 false)
+  ret %emptyStruct.1 undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.launch(i8*, i8*, i1) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct.2 @PipeRoot_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %Func2_cloned.node = call i8* @llvm.hpvm.createNode(i8* bitcast (%emptyStruct.1 (i32*, i64, i32*, i64)* @Func2_cloned to i8*))
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 3, i32 3, i1 false)
+  ret %emptyStruct.2 undef
+}
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.wait(i8*) #6
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.init() #6
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.cleanup() #6
+
+attributes #0 = { "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #1 = { argmemonly nounwind }
+attributes #2 = { noinline nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #3 = { nofree nounwind "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #4 = { nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #5 = { nounwind readnone }
+attributes #6 = { nounwind }
+
+!llvm.module.flags = !{!0}
+!llvm.ident = !{!1}
+!hpvm_hint_cpu = !{!2, !3, !4, !5}
+!hpvm_hint_gpu = !{}
+!hpvm_hint_spir = !{}
+!hpvm_hint_cudnn = !{}
+!hpvm_hint_promise = !{}
+!hpvm_hint_cpu_gpu = !{}
+!hpvm_hint_cpu_spir = !{}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{!"clang version 9.0.0 (https://gitlab.engr.illinois.edu/llvm/hpvm.git 3551132592a00cab6c966df508ab511598269f78)"}
+!2 = !{%emptyStruct (i32*, i64, i32*, i64)* @Func1_cloned}
+!3 = !{%emptyStruct.0 (i32*, i64, i32*, i64)* @Func3_cloned}
+!4 = !{%emptyStruct.1 (i32*, i64, i32*, i64)* @Func2_cloned}
+!5 = !{%emptyStruct.2 (i32*, i64, i32*, i64)* @PipeRoot_cloned}
+!6 = !{!7, !8, i64 0}
+!7 = !{!"Root", !8, i64 0, !11, i64 8, !8, i64 16, !11, i64 24}
+!8 = !{!"any pointer", !9, i64 0}
+!9 = !{!"omnipotent char", !10, i64 0}
+!10 = !{!"Simple C/C++ TBAA"}
+!11 = !{!"long", !9, i64 0}
+!12 = !{!7, !11, i64 8}
+!13 = !{!7, !8, i64 16}
+!14 = !{!7, !11, i64 24}
+!15 = !{!16, !16, i64 0}
+!16 = !{!"int", !9, i64 0}
diff --git a/hpvm/test/regressionTests/DFG2LLVM_CPU/ThreeLevel.constmem.OpenCL.ll b/hpvm/test/regressionTests/DFG2LLVM_CPU/ThreeLevel.constmem.OpenCL.ll
new file mode 100644
index 0000000000000000000000000000000000000000..c3fd32fd5f9c6ccee12834871d11212e39f70bc7
--- /dev/null
+++ b/hpvm/test/regressionTests/DFG2LLVM_CPU/ThreeLevel.constmem.OpenCL.ll
@@ -0,0 +1,270 @@
+; RUN: opt -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_OpenCL.so -load LLVMDFG2LLVM_CPU.so -S -localmem -dfg2llvm-opencl -dfg2llvm-cpu <  %s | FileCheck %s
+; ModuleID = 'ThreeLevel.opt.ll'
+source_filename = "ThreeLevel.opt.c"
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+%struct.Root = type { i32*, i64, i32*, i64 }
+%struct.out.Allocation = type <{ i8*, i64 }>
+%emptyStruct = type <{}>
+%emptyStruct.0 = type <{}>
+%emptyStruct.1 = type <{}>
+%emptyStruct.2 = type <{}>
+
+declare dso_local void @__hpvm__hint(i32) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__attributes(i32, ...) local_unnamed_addr #0
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture) #1
+
+declare dso_local i8* @__hpvm__getNode(...) local_unnamed_addr #0
+
+declare dso_local i8* @__hpvm__getParentNode(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNodeInstanceID_x(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNodeInstanceID_y(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNumNodeInstances_x(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNumNodeInstances_y(i8*) local_unnamed_addr #0
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture) #1
+
+declare dso_local i8* @__hpvm__malloc(i64) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__return(i32, ...) local_unnamed_addr #0
+
+declare dso_local i8* @__hpvm__createNodeND(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__bindIn(i8*, i32, i32, i32) local_unnamed_addr #0
+
+declare dso_local i8* @__hpvm__edge(i8*, i8*, i32, i32, i32, i32) local_unnamed_addr #0
+
+; CHECK-LABEL: @Launch(
+; CHECK: call i8* @llvm_hpvm_cpu_launch(i8*
+; CHECK-NEXT: call i8* @llvm.hpvm.launch(i8*
+; CHECK-NEXT: call void @llvm_hpvm_cpu_wait(i8*
+
+; Function Attrs: noinline nounwind uwtable
+define dso_local void @Launch() local_unnamed_addr #2 {
+entry:
+  %RootArgs = alloca %struct.Root, align 8
+  %0 = bitcast %struct.Root* %RootArgs to i8*
+  call void @llvm.lifetime.start.p0i8(i64 32, i8* nonnull %0) #6
+  %call = tail call noalias i8* @malloc(i64 1024) #6
+  %1 = bitcast %struct.Root* %RootArgs to i8**
+  store i8* %call, i8** %1, align 8, !tbaa !6
+  %Insize = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 1
+  store i64 1024, i64* %Insize, align 8, !tbaa !12
+  %output = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 2
+  %call1 = tail call noalias i8* @malloc(i64 1024) #6
+  %2 = bitcast i32** %output to i8**
+  store i8* %call1, i8** %2, align 8, !tbaa !13
+  %Outsize = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 3
+  store i64 1024, i64* %Outsize, align 8, !tbaa !14
+  %3 = bitcast %struct.Root* %RootArgs to i8*
+  %graphID = call i8* @llvm.hpvm.launch(i8* bitcast (%emptyStruct.2 (i32*, i64, i32*, i64)* @PipeRoot_cloned to i8*), i8* %3, i1 false)
+  call void @llvm.hpvm.wait(i8* %graphID)
+  call void @llvm.lifetime.end.p0i8(i64 32, i8* nonnull %0) #6
+  ret void
+}
+
+; Function Attrs: nofree nounwind
+declare dso_local noalias i8* @malloc(i64) local_unnamed_addr #3
+
+declare dso_local i8* @__hpvm__launch(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__wait(i8*) local_unnamed_addr #0
+
+; CHECK-LABEL: @main(
+; CHECK: call i8* @llvm_hpvm_ocl_initContext(i32
+; CHECK: call i8* @llvm_hpvm_ocl_launch(i8*
+; CHECK: call void @llvm_hpvm_ocl_clearContext(i8*
+
+; CHECK-LABEL: @Func2_cloned.3_cloned_cloned_cloned_cloned_cloned_cloned(
+; CHECK: call i8* @llvm_hpvm_ocl_argument_ptr(i8*
+; CHECK: call void @llvm_hpvm_ocl_argument_scalar(i8*
+; CHECK: call void @llvm_hpvm_ocl_argument_shared(i8*
+; CHECK: call void @llvm_hpvm_ocl_argument_scalar(i8*
+; CHECK: call i8* @llvm_hpvm_ocl_executeNode(i8*
+; CHECK-NEXT: call void @llvm_hpvm_ocl_wait(i8*
+; CHECK-NEXT: call void @llvm_hpvm_ocl_free(i8*
+
+; CHECK-LABEL: @PipeRoot_cloned.4(
+; CHECK: call void @llvm_hpvm_cpu_dstack_push(
+; CHECK-NEXT: @Func2_cloned.3_cloned_cloned_cloned_cloned_cloned_clone
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop(
+
+; CHECK-LABEL: define i8* @LaunchDataflowGraph(i8*
+
+
+; Function Attrs: nounwind uwtable
+define dso_local i32 @main() local_unnamed_addr #4 {
+entry:
+  call void @llvm.hpvm.init()
+  tail call void @Launch()
+  call void @llvm.hpvm.cleanup()
+  ret i32 0
+}
+
+declare dso_local void @__hpvm__init(...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__cleanup(...) local_unnamed_addr #0
+
+declare i8* @llvm_hpvm_initializeTimerSet()
+
+declare void @llvm_hpvm_switchToTimer(i8**, i32)
+
+declare void @llvm_hpvm_printTimerSet(i8**, i8*)
+
+; Function Attrs: nounwind readnone
+declare i8* @llvm.hpvm.getNode() #5
+
+; Function Attrs: nounwind readnone
+declare i8* @llvm.hpvm.getParentNode(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNodeInstanceID.x(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNodeInstanceID.y(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNumNodeInstances.x(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNumNodeInstances.y(i8*) #5
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.malloc(i64) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %struct.out.Allocation @Allocation_cloned(i64 %block) #4 {
+entry:
+  %call1 = call i8* @llvm.hpvm.malloc(i64 %block)
+  %returnStruct = insertvalue %struct.out.Allocation undef, i8* %call1, 0
+  %returnStruct2 = insertvalue %struct.out.Allocation %returnStruct, i64 %block, 1
+  ret %struct.out.Allocation %returnStruct2
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.createNode2D(i8*, i64, i64) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct @Func1_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %call4 = call i8* @llvm.hpvm.getNode()
+  %call15 = call i8* @llvm.hpvm.getParentNode(i8* %call4)
+  %call26 = call i64 @llvm.hpvm.getNodeInstanceID.x(i8* %call4)
+  %call37 = call i64 @llvm.hpvm.getNodeInstanceID.y(i8* %call4)
+  %call58 = call i64 @llvm.hpvm.getNodeInstanceID.x(i8* %call15)
+  %call79 = call i64 @llvm.hpvm.getNodeInstanceID.y(i8* %call15)
+  %call910 = call i64 @llvm.hpvm.getNumNodeInstances.x(i8* %call4)
+  %call1111 = call i64 @llvm.hpvm.getNumNodeInstances.y(i8* %call4)
+  %mul = mul i64 %call910, %call58
+  %add = add i64 %mul, %call26
+  %arrayidx = getelementptr inbounds i32, i32* %In, i64 3
+  %0 = load i32, i32* %arrayidx, align 4, !tbaa !15
+  %sext = shl i64 %add, 32
+  %idxprom = ashr exact i64 %sext, 32
+  %arrayidx15 = getelementptr inbounds i32, i32* %Out, i64 %idxprom
+  %1 = load i32, i32* %arrayidx15, align 4, !tbaa !15
+  %add16 = add nsw i32 %1, %0
+  store i32 %add16, i32* %arrayidx15, align 4, !tbaa !15
+  ret %emptyStruct undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.createNode(i8*) #6
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.bind.input(i8*, i32, i32, i1) #6
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.createEdge(i8*, i8*, i1, i32, i32, i1) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct.0 @Func3_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %Func1_cloned.node = call i8* @llvm.hpvm.createNode2D(i8* bitcast (%emptyStruct (i32*, i64, i32*, i64)* @Func1_cloned to i8*), i64 3, i64 5)
+  %Allocation_cloned.node = call i8* @llvm.hpvm.createNode(i8* bitcast (%struct.out.Allocation (i64)* @Allocation_cloned to i8*))
+  call void @llvm.hpvm.bind.input(i8* %Allocation_cloned.node, i32 1, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 3, i32 3, i1 false)
+  %output = call i8* @llvm.hpvm.createEdge(i8* %Allocation_cloned.node, i8* %Func1_cloned.node, i1 true, i32 0, i32 0, i1 false)
+  %output1 = call i8* @llvm.hpvm.createEdge(i8* %Allocation_cloned.node, i8* %Func1_cloned.node, i1 true, i32 1, i32 1, i1 false)
+  ret %emptyStruct.0 undef
+}
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct.1 @Func2_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %Func3_cloned.node = call i8* @llvm.hpvm.createNode2D(i8* bitcast (%emptyStruct.0 (i32*, i64, i32*, i64)* @Func3_cloned to i8*), i64 3, i64 5)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 3, i32 3, i1 false)
+  ret %emptyStruct.1 undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.launch(i8*, i8*, i1) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct.2 @PipeRoot_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %Func2_cloned.node = call i8* @llvm.hpvm.createNode(i8* bitcast (%emptyStruct.1 (i32*, i64, i32*, i64)* @Func2_cloned to i8*))
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 3, i32 3, i1 false)
+  ret %emptyStruct.2 undef
+}
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.wait(i8*) #6
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.init() #6
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.cleanup() #6
+
+attributes #0 = { "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #1 = { argmemonly nounwind }
+attributes #2 = { noinline nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #3 = { nofree nounwind "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #4 = { nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #5 = { nounwind readnone }
+attributes #6 = { nounwind }
+
+!llvm.module.flags = !{!0}
+!llvm.ident = !{!1}
+!hpvm_hint_gpu = !{!2}
+!hpvm_hint_cpu = !{!3, !4, !5}
+!hpvm_hint_spir = !{}
+!hpvm_hint_cudnn = !{}
+!hpvm_hint_promise = !{}
+!hpvm_hint_cpu_gpu = !{}
+!hpvm_hint_cpu_spir = !{}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{!"clang version 9.0.0 (https://gitlab.engr.illinois.edu/llvm/hpvm.git 6690f9e7e8b46b96aea222d3e85315cd63545953)"}
+!2 = !{%emptyStruct (i32*, i64, i32*, i64)* @Func1_cloned}
+!3 = !{%emptyStruct.0 (i32*, i64, i32*, i64)* @Func3_cloned}
+!4 = !{%emptyStruct.1 (i32*, i64, i32*, i64)* @Func2_cloned}
+!5 = !{%emptyStruct.2 (i32*, i64, i32*, i64)* @PipeRoot_cloned}
+!6 = !{!7, !8, i64 0}
+!7 = !{!"Root", !8, i64 0, !11, i64 8, !8, i64 16, !11, i64 24}
+!8 = !{!"any pointer", !9, i64 0}
+!9 = !{!"omnipotent char", !10, i64 0}
+!10 = !{!"Simple C/C++ TBAA"}
+!11 = !{!"long", !9, i64 0}
+!12 = !{!7, !11, i64 8}
+!13 = !{!7, !8, i64 16}
+!14 = !{!7, !11, i64 24}
+!15 = !{!16, !16, i64 0}
+!16 = !{!"int", !9, i64 0}
diff --git a/hpvm/test/regressionTests/DFG2LLVM_CPU/TwoLevel.codeGen.ll b/hpvm/test/regressionTests/DFG2LLVM_CPU/TwoLevel.codeGen.ll
new file mode 100644
index 0000000000000000000000000000000000000000..f8eed46c26eec0b383695ba6da9d81426f2abec7
--- /dev/null
+++ b/hpvm/test/regressionTests/DFG2LLVM_CPU/TwoLevel.codeGen.ll
@@ -0,0 +1,213 @@
+; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_CPU.so -S -dfg2llvm-cpu <  %s | FileCheck %s
+; ModuleID = 'TwoLevel.ll'
+source_filename = "TwoLevel.c"
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+%struct.Root = type { i32*, i64, i32*, i64, i32*, i64 }
+%struct.out.Func1 = type <{ i32* }>
+%struct.out.Func2 = type <{ i32* }>
+%struct.out.PipeRoot = type <{ i32* }>
+
+; CHECK-LABEL: i32 @main(
+; CHECK: call void @llvm.hpvm.init()
+; CHECK: call i8* @llvm_hpvm_cpu_launch(i8* (i8*)* @LaunchDataflowGraph, i8*
+; CHECK-NEXT: call i8* @llvm.hpvm.launch(i8* 
+; CHECK-NEXT: call void @llvm_hpvm_cpu_wait(i8*
+
+; CHECK-LABEL: @Func2_cloned(
+; CHECK: call i8* @llvm.hpvm.createNode1D(
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.output(i8* %Func1_cloned.node
+
+; CHECK-LABEL: @PipeRoot_cloned(
+; CHECK: call i8* @llvm.hpvm.createNode(
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.output(i8* %Func2_cloned.node
+
+; CHECK-LABEL: @Func1_cloned.1_cloned_cloned_cloned_cloned_cloned_cloned(
+; CHECK: call i8* @llvm_hpvm_cpu_argument_ptr(
+
+; CHECK-LABEL: @Func2_cloned.2_cloned_cloned_cloned_cloned_cloned_cloned(
+; CHECK-LABEL: for.body
+; CHECK: %index.x = phi i64 [ 0, %entry ], [ %index.x.inc, %for.body ]
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_push(
+; CHECK-NEXT: @Func1_cloned.1_cloned_cloned_cloned_cloned_cloned_cloned(
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop()
+
+; CHECK-LABEL: @PipeRoot_cloned.3(
+; CHECK: call void @llvm_hpvm_cpu_dstack_push(
+; CHECK-NEXT: @Func2_cloned.2_cloned_cloned_cloned_cloned_cloned_cloned(
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop()
+
+; CHECK-LABEL: @LaunchDataflowGraph(i8*
+; call %struct.out.PipeRoot @PipeRoot_cloned.3(
+
+declare dso_local void @__hpvm__hint(i32) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__attributes(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__return(i32, ...) local_unnamed_addr #0
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture) #1
+
+declare dso_local i8* @__hpvm__createNodeND(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__bindIn(i8*, i32, i32, i32) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__bindOut(i8*, i32, i32, i32) local_unnamed_addr #0
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture) #1
+
+; Function Attrs: nounwind uwtable
+define dso_local i32 @main() local_unnamed_addr #2 {
+entry:
+  %In1 = alloca i32, align 4
+  %In2 = alloca i32, align 4
+  %Out = alloca i32, align 4
+  %RootArgs = alloca %struct.Root, align 8
+  %0 = bitcast i32* %In1 to i8*
+  call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %0) #3
+  store i32 1, i32* %In1, align 4, !tbaa !5
+  %1 = bitcast i32* %In2 to i8*
+  call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %1) #3
+  store i32 2, i32* %In2, align 4, !tbaa !5
+  %2 = bitcast i32* %Out to i8*
+  call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %2) #3
+  store i32 0, i32* %Out, align 4, !tbaa !5
+  %3 = bitcast %struct.Root* %RootArgs to i8*
+  call void @llvm.lifetime.start.p0i8(i64 48, i8* nonnull %3) #3
+  %input1 = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 0
+  store i32* %In1, i32** %input1, align 8, !tbaa !9
+  %Insize1 = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 1
+  store i64 32, i64* %Insize1, align 8, !tbaa !13
+  %input2 = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 2
+  store i32* %In2, i32** %input2, align 8, !tbaa !14
+  %Insize2 = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 3
+  store i64 32, i64* %Insize2, align 8, !tbaa !15
+  %output = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 4
+  store i32* %Out, i32** %output, align 8, !tbaa !16
+  %Outsize = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 5
+  store i64 32, i64* %Outsize, align 8, !tbaa !17
+  call void @llvm.hpvm.init()
+  %4 = bitcast %struct.Root* %RootArgs to i8*
+  %graphID = call i8* @llvm.hpvm.launch(i8* bitcast (%struct.out.PipeRoot (i32*, i64, i32*, i64, i32*, i64)* @PipeRoot_cloned to i8*), i8* %4, i1 false)
+  call void @llvm.hpvm.wait(i8* %graphID)
+  call void @llvm.hpvm.cleanup()
+  call void @llvm.lifetime.end.p0i8(i64 48, i8* nonnull %3) #3
+  call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %2) #3
+  call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %1) #3
+  call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %0) #3
+  ret i32 0
+}
+
+declare dso_local void @__hpvm__init(...) local_unnamed_addr #0
+
+declare dso_local i8* @__hpvm__launch(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__wait(i8*) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__cleanup(...) local_unnamed_addr #0
+
+declare i8* @llvm_hpvm_initializeTimerSet()
+
+declare void @llvm_hpvm_switchToTimer(i8**, i32)
+
+declare void @llvm_hpvm_printTimerSet(i8**, i8*)
+
+; Function Attrs: nounwind uwtable
+define dso_local %struct.out.Func1 @Func1_cloned(i32* in %In, i64 %Insize, i32* out %Out, i64 %Outsize) #2 {
+entry:
+  %returnStruct = insertvalue %struct.out.Func1 undef, i32* %Out, 0
+  ret %struct.out.Func1 %returnStruct
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.createNode1D(i8*, i64) #3
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.bind.input(i8*, i32, i32, i1) #3
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.bind.output(i8*, i32, i32, i1) #3
+
+; Function Attrs: nounwind uwtable
+define dso_local %struct.out.Func2 @Func2_cloned(i32* in %In, i64 %Insize, i32* out %Out, i64 %Outsize) #2 {
+entry:
+  %Func1_cloned.node = call i8* @llvm.hpvm.createNode1D(i8* bitcast (%struct.out.Func1 (i32*, i64, i32*, i64)* @Func1_cloned to i8*), i64 3)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 3, i32 3, i1 false)
+  call void @llvm.hpvm.bind.output(i8* %Func1_cloned.node, i32 0, i32 0, i1 false)
+  ret %struct.out.Func2 undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.createNode(i8*) #3
+
+; Function Attrs: nounwind uwtable
+define dso_local %struct.out.PipeRoot @PipeRoot_cloned(i32* in %In1, i64 %Insize1, i32* in %In2, i64 %InSize2, i32* out %Out, i64 %Outsize) #2 {
+entry:
+  %Func2_cloned.node = call i8* @llvm.hpvm.createNode(i8* bitcast (%struct.out.Func2 (i32*, i64, i32*, i64)* @Func2_cloned to i8*))
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 3, i32 3, i1 false)
+  call void @llvm.hpvm.bind.output(i8* %Func2_cloned.node, i32 0, i32 0, i1 false)
+  ret %struct.out.PipeRoot undef
+}
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.init() #3
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.launch(i8*, i8*, i1) #3
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.wait(i8*) #3
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.cleanup() #3
+
+attributes #0 = { "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #1 = { argmemonly nounwind }
+attributes #2 = { nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #3 = { nounwind }
+
+!llvm.module.flags = !{!0}
+!llvm.ident = !{!1}
+!hpvm_hint_cpu = !{!2, !3, !4}
+!hpvm_hint_gpu = !{}
+!hpvm_hint_spir = !{}
+!hpvm_hint_cudnn = !{}
+!hpvm_hint_promise = !{}
+!hpvm_hint_cpu_gpu = !{}
+!hpvm_hint_cpu_spir = !{}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{!"clang version 9.0.0 (https://gitlab.engr.illinois.edu/llvm/hpvm.git 6690f9e7e8b46b96aea222d3e85315cd63545953)"}
+!2 = !{%struct.out.Func1 (i32*, i64, i32*, i64)* @Func1_cloned}
+!3 = !{%struct.out.Func2 (i32*, i64, i32*, i64)* @Func2_cloned}
+!4 = !{%struct.out.PipeRoot (i32*, i64, i32*, i64, i32*, i64)* @PipeRoot_cloned}
+!5 = !{!6, !6, i64 0}
+!6 = !{!"int", !7, i64 0}
+!7 = !{!"omnipotent char", !8, i64 0}
+!8 = !{!"Simple C/C++ TBAA"}
+!9 = !{!10, !11, i64 0}
+!10 = !{!"Root", !11, i64 0, !12, i64 8, !11, i64 16, !12, i64 24, !11, i64 32, !12, i64 40}
+!11 = !{!"any pointer", !7, i64 0}
+!12 = !{!"long", !7, i64 0}
+!13 = !{!10, !12, i64 8}
+!14 = !{!10, !11, i64 16}
+!15 = !{!10, !12, i64 24}
+!16 = !{!10, !11, i64 32}
+!17 = !{!10, !12, i64 40}
diff --git a/hpvm/test/regressionTests/DFG2LLVM_CPU/TwoRoot.ll b/hpvm/test/regressionTests/DFG2LLVM_CPU/TwoRoot.ll
new file mode 100644
index 0000000000000000000000000000000000000000..c562ab141ea7276977a10fadf37c1bc49da9288c
--- /dev/null
+++ b/hpvm/test/regressionTests/DFG2LLVM_CPU/TwoRoot.ll
@@ -0,0 +1,230 @@
+; RUN: opt -load LLVMBuildDFG.so -load LLVMDFG2LLVM_CPU.so -S -dfg2llvm-cpu <  %s | FileCheck %s
+; ModuleID = 'TwoLaunch.ll'
+source_filename = "TwoLaunch.c"
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+%struct.Root = type { i32*, i64, i32*, i64, i32*, i64 }
+%struct.out.Func1 = type <{ i32* }>
+%struct.out.Func2 = type <{ i32* }>
+%struct.out.PipeRoot = type <{ i32* }>
+
+; CHECK-LABEL: i32 @main(
+; CHECK: call void @llvm.hpvm.init()
+; CHECK: @llvm_hpvm_cpu_launch(i8* (i8*)* @LaunchDataflowGraph, i8*
+; CHECK-NEXT: call i8* @llvm.hpvm.launch(i8*
+; CHECK: @llvm_hpvm_cpu_launch(i8* (i8*)* @LaunchDataflowGraph.7, i8*
+; CHECK-NEXT: call i8* @llvm.hpvm.launch(i8*
+; CHECK-NEXT: call void @llvm_hpvm_cpu_wait(i8*
+; CHECK-NEXT: call void @llvm_hpvm_cpu_wait(i8*
+
+; CHECK-LABEL: @Func2_cloned(
+; CHECK: call i8* @llvm.hpvm.createNode1D(
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.output(i8* %Func1_cloned.node
+
+; CHECK-LABEL: @PipeRoot_cloned(
+; CHECK: call i8* @llvm.hpvm.createNode(
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node
+; CHECK-NEXT: call void @llvm.hpvm.bind.output(i8* %Func2_cloned.node
+
+; CHECK-LABEL: @Func1_cloned.1_cloned_cloned_cloned_cloned_cloned_cloned(
+; CHECK: call i8* @llvm_hpvm_cpu_argument_ptr(
+
+; CHECK-LABEL: @Func2_cloned.2_cloned_cloned_cloned_cloned_cloned_cloned(
+; CHECK: %index.x = phi i64 [ 0, %entry ], [ %index.x.inc, %for.body ]
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_push(
+; CHECK-NEXT: @Func1_cloned.1_cloned_cloned_cloned_cloned_cloned_cloned(
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop()
+
+; CHECK-LABEL: @PipeRoot_cloned.3(
+; CHECK: call void @llvm_hpvm_cpu_dstack_push(
+; CHECK-NEXT: @Func2_cloned.2_cloned_cloned_cloned_cloned_cloned_cloned(
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop()
+
+; CHECK-LABEL: @LaunchDataflowGraph(i8*
+; CHECK: call %struct.out.PipeRoot @PipeRoot_cloned.3(
+
+; CHECK-LABEL: @Func1_cloned.4_cloned_cloned_cloned_cloned_cloned_cloned(
+; CHECK: @llvm_hpvm_cpu_argument_ptr(
+
+; CHECK-LABEL: @Func2_cloned.5_cloned_cloned_cloned_cloned_cloned_cloned(
+; CHECK: %index.x = phi i64 [ 0, %entry ], [ %index.x.inc, %for.body ]
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_push(
+; CHECK-NEXT: @Func1_cloned.4_cloned_cloned_cloned_cloned_cloned_cloned(
+; CHECK-NEXT: call void @llvm_hpvm_cpu_dstack_pop()
+
+; CHECK-LABEL: @LaunchDataflowGraph.7(i8*
+; call %struct.out.PipeRoot @PipeRoot_cloned.6(
+
+declare dso_local void @__hpvm__hint(i32) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__attributes(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__return(i32, ...) local_unnamed_addr #0
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture) #1
+
+declare dso_local i8* @__hpvm__createNodeND(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__bindIn(i8*, i32, i32, i32) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__bindOut(i8*, i32, i32, i32) local_unnamed_addr #0
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture) #1
+
+; Function Attrs: nounwind uwtable
+define dso_local i32 @main() local_unnamed_addr #2 {
+entry:
+  %In1 = alloca i32, align 4
+  %In2 = alloca i32, align 4
+  %Out = alloca i32, align 4
+  %RootArgs = alloca %struct.Root, align 8
+  %0 = bitcast i32* %In1 to i8*
+  call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %0) #3
+  store i32 1, i32* %In1, align 4, !tbaa !5
+  %1 = bitcast i32* %In2 to i8*
+  call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %1) #3
+  store i32 2, i32* %In2, align 4, !tbaa !5
+  %2 = bitcast i32* %Out to i8*
+  call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %2) #3
+  store i32 0, i32* %Out, align 4, !tbaa !5
+  %3 = bitcast %struct.Root* %RootArgs to i8*
+  call void @llvm.lifetime.start.p0i8(i64 48, i8* nonnull %3) #3
+  %input1 = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 0
+  store i32* %In1, i32** %input1, align 8, !tbaa !9
+  %Insize1 = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 1
+  store i64 32, i64* %Insize1, align 8, !tbaa !13
+  %input2 = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 2
+  store i32* %In2, i32** %input2, align 8, !tbaa !14
+  %Insize2 = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 3
+  store i64 32, i64* %Insize2, align 8, !tbaa !15
+  %output = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 4
+  store i32* %Out, i32** %output, align 8, !tbaa !16
+  %Outsize = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 5
+  store i64 32, i64* %Outsize, align 8, !tbaa !17
+  call void @llvm.hpvm.init()
+  %4 = bitcast %struct.Root* %RootArgs to i8*
+  %graphID = call i8* @llvm.hpvm.launch(i8* bitcast (%struct.out.PipeRoot (i32*, i64, i32*, i64, i32*, i64)* @PipeRoot_cloned to i8*), i8* %4, i1 false)
+  %5 = bitcast %struct.Root* %RootArgs to i8*
+  %graphID1 = call i8* @llvm.hpvm.launch(i8* bitcast (%struct.out.PipeRoot (i32*, i64, i32*, i64, i32*, i64)* @PipeRoot_cloned to i8*), i8* %5, i1 false)
+  call void @llvm.hpvm.wait(i8* %graphID)
+  call void @llvm.hpvm.wait(i8* %graphID1)
+  call void @llvm.hpvm.cleanup()
+  call void @llvm.lifetime.end.p0i8(i64 48, i8* nonnull %3) #3
+  call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %2) #3
+  call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %1) #3
+  call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %0) #3
+  ret i32 0
+}
+
+declare dso_local void @__hpvm__init(...) local_unnamed_addr #0
+
+declare dso_local i8* @__hpvm__launch(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__wait(i8*) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__cleanup(...) local_unnamed_addr #0
+
+declare i8* @llvm_hpvm_initializeTimerSet()
+
+declare void @llvm_hpvm_switchToTimer(i8**, i32)
+
+declare void @llvm_hpvm_printTimerSet(i8**, i8*)
+
+; Function Attrs: nounwind uwtable
+define dso_local %struct.out.Func1 @Func1_cloned(i32* in %In, i64 %Insize, i32* out %Out, i64 %Outsize) #2 {
+entry:
+  %returnStruct = insertvalue %struct.out.Func1 undef, i32* %Out, 0
+  ret %struct.out.Func1 %returnStruct
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.createNode1D(i8*, i64) #3
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.bind.input(i8*, i32, i32, i1) #3
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.bind.output(i8*, i32, i32, i1) #3
+
+; Function Attrs: nounwind uwtable
+define dso_local %struct.out.Func2 @Func2_cloned(i32* in %In, i64 %Insize, i32* out %Out, i64 %Outsize) #2 {
+entry:
+  %Func1_cloned.node = call i8* @llvm.hpvm.createNode1D(i8* bitcast (%struct.out.Func1 (i32*, i64, i32*, i64)* @Func1_cloned to i8*), i64 3)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 3, i32 3, i1 false)
+  call void @llvm.hpvm.bind.output(i8* %Func1_cloned.node, i32 0, i32 0, i1 false)
+  ret %struct.out.Func2 undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.createNode(i8*) #3
+
+; Function Attrs: nounwind uwtable
+define dso_local %struct.out.PipeRoot @PipeRoot_cloned(i32* in %In1, i64 %Insize1, i32* in %In2, i64 %InSize2, i32* out %Out, i64 %Outsize) #2 {
+entry:
+  %Func2_cloned.node = call i8* @llvm.hpvm.createNode(i8* bitcast (%struct.out.Func2 (i32*, i64, i32*, i64)* @Func2_cloned to i8*))
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 3, i32 3, i1 false)
+  call void @llvm.hpvm.bind.output(i8* %Func2_cloned.node, i32 0, i32 0, i1 false)
+  ret %struct.out.PipeRoot undef
+}
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.init() #3
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.launch(i8*, i8*, i1) #3
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.wait(i8*) #3
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.cleanup() #3
+
+attributes #0 = { "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #1 = { argmemonly nounwind }
+attributes #2 = { nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="cpu-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #3 = { nounwind }
+
+!llvm.module.flags = !{!0}
+!llvm.ident = !{!1}
+!hpvm_hint_cpu = !{!2, !3, !4}
+!hpvm_hint_gpu = !{}
+!hpvm_hint_spir = !{}
+!hpvm_hint_cudnn = !{}
+!hpvm_hint_promise = !{}
+!hpvm_hint_cpu_gpu = !{}
+!hpvm_hint_cpu_spir = !{}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{!"clang version 9.0.0 (https://gitlab.engr.illinois.edu/llvm/hpvm.git 6690f9e7e8b46b96aea222d3e85315cd63545953)"}
+!2 = !{%struct.out.Func1 (i32*, i64, i32*, i64)* @Func1_cloned}
+!3 = !{%struct.out.Func2 (i32*, i64, i32*, i64)* @Func2_cloned}
+!4 = !{%struct.out.PipeRoot (i32*, i64, i32*, i64, i32*, i64)* @PipeRoot_cloned}
+!5 = !{!6, !6, i64 0}
+!6 = !{!"int", !7, i64 0}
+!7 = !{!"omnipotent char", !8, i64 0}
+!8 = !{!"Simple C/C++ TBAA"}
+!9 = !{!10, !11, i64 0}
+!10 = !{!"Root", !11, i64 0, !12, i64 8, !11, i64 16, !12, i64 24, !11, i64 32, !12, i64 40}
+!11 = !{!"any pointer", !7, i64 0}
+!12 = !{!"long", !7, i64 0}
+!13 = !{!10, !12, i64 8}
+!14 = !{!10, !11, i64 16}
+!15 = !{!10, !12, i64 24}
+!16 = !{!10, !11, i64 32}
+!17 = !{!10, !12, i64 40}
diff --git a/hpvm/test/regressionTests/DFG2LLVM_OPENCL/ThreeLevel.atomic.ll b/hpvm/test/regressionTests/DFG2LLVM_OPENCL/ThreeLevel.atomic.ll
new file mode 100644
index 0000000000000000000000000000000000000000..ea6ec14d1069d45fcf7b434cfd08308c8f7c9158
--- /dev/null
+++ b/hpvm/test/regressionTests/DFG2LLVM_OPENCL/ThreeLevel.atomic.ll
@@ -0,0 +1,280 @@
+; RUN: opt -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_OpenCL.so -S -localmem -dfg2llvm-opencl <  %s | FileCheck %s
+; ModuleID = 'ThreeLevel.atomic.ll'
+source_filename = "ThreeLevel.constmem.c"
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+%struct.Root = type { i32*, i64, i32*, i64 }
+%struct.out.Allocation = type <{ i8*, i64 }>
+%emptyStruct = type <{}>
+%emptyStruct.0 = type <{}>
+%emptyStruct.1 = type <{}>
+%emptyStruct.2 = type <{}>
+
+declare dso_local void @__hpvm__hint(i32) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__attributes(i32, ...) local_unnamed_addr #0
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture) #1
+
+declare dso_local i8* @__hpvm__getNode(...) local_unnamed_addr #0
+
+declare dso_local i8* @__hpvm__getParentNode(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNodeInstanceID_x(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNodeInstanceID_y(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNumNodeInstances_x(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNumNodeInstances_y(i8*) local_unnamed_addr #0
+
+declare dso_local i32 @__hpvm__atomic_add(i32*, i32) local_unnamed_addr #0
+
+declare dso_local i32 @__hpvm__atomic_sub(i32*, i32) local_unnamed_addr #0
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture) #1
+
+declare dso_local i8* @__hpvm__malloc(i64) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__return(i32, ...) local_unnamed_addr #0
+
+declare dso_local i8* @__hpvm__createNodeND(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__bindIn(i8*, i32, i32, i32) local_unnamed_addr #0
+
+declare dso_local i8* @__hpvm__edge(i8*, i8*, i32, i32, i32, i32) local_unnamed_addr #0
+
+; Function Attrs: noinline nounwind uwtable
+define dso_local void @Launch() local_unnamed_addr #2 {
+entry:
+  %RootArgs = alloca %struct.Root, align 8
+  %0 = bitcast %struct.Root* %RootArgs to i8*
+  call void @llvm.lifetime.start.p0i8(i64 32, i8* nonnull %0) #6
+  %call = tail call noalias i8* @malloc(i64 1024) #6
+  %1 = bitcast %struct.Root* %RootArgs to i8**
+  store i8* %call, i8** %1, align 8, !tbaa !6
+  %Insize = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 1
+  store i64 1024, i64* %Insize, align 8, !tbaa !12
+  %output = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 2
+  %call1 = tail call noalias i8* @malloc(i64 1024) #6
+  %2 = bitcast i32** %output to i8**
+  store i8* %call1, i8** %2, align 8, !tbaa !13
+  %Outsize = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 3
+  store i64 1024, i64* %Outsize, align 8, !tbaa !14
+  %3 = bitcast %struct.Root* %RootArgs to i8*
+  %graphID = call i8* @llvm.hpvm.launch(i8* bitcast (%emptyStruct.2 (i32*, i64, i32*, i64)* @PipeRoot_cloned to i8*), i8* %3, i1 false)
+  call void @llvm.hpvm.wait(i8* %graphID)
+  call void @llvm.lifetime.end.p0i8(i64 32, i8* nonnull %0) #6
+  ret void
+}
+
+; Function Attrs: nofree nounwind
+declare dso_local noalias i8* @malloc(i64) local_unnamed_addr #3
+
+declare dso_local i8* @__hpvm__launch(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__wait(i8*) local_unnamed_addr #0
+
+; CHECK-LABEL: @main(
+; CHECK: call i8* @llvm_hpvm_ocl_initContext(i32
+; CHECK: call i8* @llvm_hpvm_ocl_launch(i8*
+; CHECK: call void @llvm_hpvm_ocl_clearContext(i8*
+
+; CHECK-LABEL: @Func1_cloned(
+; CHECK: call i32 @llvm.hpvm.atomic.add
+; CHECK: call i32 @llvm.hpvm.atomic.sub
+
+; CHECK-LABEL: @Func2_cloned.3_cloned_cloned_cloned_cloned_cloned_cloned
+; CHECK: call i8* @llvm_hpvm_ocl_argument_ptr(i8*
+; CHECK: call void @llvm_hpvm_ocl_argument_scalar(i8*
+; CHECK: call void @llvm_hpvm_ocl_argument_shared(i8*
+; CHECK: call void @llvm_hpvm_ocl_argument_scalar(i8*
+; CHECK: call i8* @llvm_hpvm_ocl_executeNode(i8*
+; CHECK-NEXT: call void @llvm_hpvm_ocl_wait(i8*
+; CHECK-NEXT: call void @llvm_hpvm_ocl_free(i8*
+
+
+; Function Attrs: nounwind uwtable
+define dso_local i32 @main() local_unnamed_addr #4 {
+entry:
+  call void @llvm.hpvm.init()
+  tail call void @Launch()
+  call void @llvm.hpvm.cleanup()
+  ret i32 0
+}
+
+declare dso_local void @__hpvm__init(...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__cleanup(...) local_unnamed_addr #0
+
+declare i8* @llvm_hpvm_initializeTimerSet()
+
+declare void @llvm_hpvm_switchToTimer(i8**, i32)
+
+declare void @llvm_hpvm_printTimerSet(i8**, i8*)
+
+; Function Attrs: nounwind readnone
+declare i8* @llvm.hpvm.getNode() #5
+
+; Function Attrs: nounwind readnone
+declare i8* @llvm.hpvm.getParentNode(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNodeInstanceID.x(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNodeInstanceID.y(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNumNodeInstances.x(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNumNodeInstances.y(i8*) #5
+
+; Function Attrs: nounwind
+declare i32 @llvm.hpvm.atomic.add(i8*, i32) #6
+
+; Function Attrs: nounwind
+declare i32 @llvm.hpvm.atomic.sub(i8*, i32) #6
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.malloc(i64) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %struct.out.Allocation @Allocation_cloned(i64 %block) #4 {
+entry:
+  %call1 = call i8* @llvm.hpvm.malloc(i64 %block)
+  %returnStruct = insertvalue %struct.out.Allocation undef, i8* %call1, 0
+  %returnStruct2 = insertvalue %struct.out.Allocation %returnStruct, i64 %block, 1
+  ret %struct.out.Allocation %returnStruct2
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.createNode2D(i8*, i64, i64) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct @Func1_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %call4 = call i8* @llvm.hpvm.getNode()
+  %call16 = call i8* @llvm.hpvm.getParentNode(i8* %call4)
+  %call27 = call i64 @llvm.hpvm.getNodeInstanceID.x(i8* %call4)
+  %call38 = call i64 @llvm.hpvm.getNodeInstanceID.y(i8* %call4)
+  %call59 = call i64 @llvm.hpvm.getNodeInstanceID.x(i8* %call16)
+  %call710 = call i64 @llvm.hpvm.getNodeInstanceID.y(i8* %call16)
+  %call911 = call i64 @llvm.hpvm.getNumNodeInstances.x(i8* %call4)
+  %call1112 = call i64 @llvm.hpvm.getNumNodeInstances.y(i8* %call4)
+  %mul = mul i64 %call911, %call59
+  %add = add i64 %mul, %call27
+  %mul13 = mul i64 %call1112, %call710
+  %add14 = add i64 %mul13, %call38
+  %sext = shl i64 %add14, 32
+  %idxprom = ashr exact i64 %sext, 32
+  %arrayidx = getelementptr inbounds i32, i32* %Out, i64 %idxprom
+  %0 = load i32, i32* %arrayidx, align 4, !tbaa !15
+  %1 = bitcast i32* %In to i8*
+  %call1513 = call i32 @llvm.hpvm.atomic.add(i8* %1, i32 %0)
+  %sext47 = shl i64 %add, 32
+  %idxprom16 = ashr exact i64 %sext47, 32
+  %arrayidx17 = getelementptr inbounds i32, i32* %Out, i64 %idxprom16
+  store i32 %call1513, i32* %arrayidx17, align 4, !tbaa !15
+  %2 = load i32, i32* %arrayidx, align 4, !tbaa !15
+  %3 = bitcast i32* %Out to i8*
+  %call2014 = call i32 @llvm.hpvm.atomic.sub(i8* %3, i32 %2)
+  store i32 %call2014, i32* %arrayidx17, align 4, !tbaa !15
+  ret %emptyStruct undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.createNode(i8*) #6
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.bind.input(i8*, i32, i32, i1) #6
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.createEdge(i8*, i8*, i1, i32, i32, i1) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct.0 @Func3_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %Func1_cloned.node = call i8* @llvm.hpvm.createNode2D(i8* bitcast (%emptyStruct (i32*, i64, i32*, i64)* @Func1_cloned to i8*), i64 3, i64 5)
+  %Allocation_cloned.node = call i8* @llvm.hpvm.createNode(i8* bitcast (%struct.out.Allocation (i64)* @Allocation_cloned to i8*))
+  call void @llvm.hpvm.bind.input(i8* %Allocation_cloned.node, i32 1, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 3, i32 3, i1 false)
+  %output = call i8* @llvm.hpvm.createEdge(i8* %Allocation_cloned.node, i8* %Func1_cloned.node, i1 true, i32 0, i32 0, i1 false)
+  %output1 = call i8* @llvm.hpvm.createEdge(i8* %Allocation_cloned.node, i8* %Func1_cloned.node, i1 true, i32 1, i32 1, i1 false)
+  ret %emptyStruct.0 undef
+}
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct.1 @Func2_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %Func3_cloned.node = call i8* @llvm.hpvm.createNode2D(i8* bitcast (%emptyStruct.0 (i32*, i64, i32*, i64)* @Func3_cloned to i8*), i64 3, i64 5)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 3, i32 3, i1 false)
+  ret %emptyStruct.1 undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.launch(i8*, i8*, i1) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct.2 @PipeRoot_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %Func2_cloned.node = call i8* @llvm.hpvm.createNode(i8* bitcast (%emptyStruct.1 (i32*, i64, i32*, i64)* @Func2_cloned to i8*))
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 3, i32 3, i1 false)
+  ret %emptyStruct.2 undef
+}
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.wait(i8*) #6
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.init() #6
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.cleanup() #6
+
+attributes #0 = { "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #1 = { argmemonly nounwind }
+attributes #2 = { noinline nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #3 = { nofree nounwind "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #4 = { nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #5 = { nounwind readnone }
+attributes #6 = { nounwind }
+
+!llvm.module.flags = !{!0}
+!llvm.ident = !{!1}
+!hpvm_hint_gpu = !{!2}
+!hpvm_hint_cpu = !{!3, !4, !5}
+!hpvm_hint_spir = !{}
+!hpvm_hint_cudnn = !{}
+!hpvm_hint_promise = !{}
+!hpvm_hint_cpu_gpu = !{}
+!hpvm_hint_cpu_spir = !{}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{!"clang version 9.0.0 (https://gitlab.engr.illinois.edu/llvm/hpvm.git 3551132592a00cab6c966df508ab511598269f78)"}
+!2 = !{%emptyStruct (i32*, i64, i32*, i64)* @Func1_cloned}
+!3 = !{%emptyStruct.0 (i32*, i64, i32*, i64)* @Func3_cloned}
+!4 = !{%emptyStruct.1 (i32*, i64, i32*, i64)* @Func2_cloned}
+!5 = !{%emptyStruct.2 (i32*, i64, i32*, i64)* @PipeRoot_cloned}
+!6 = !{!7, !8, i64 0}
+!7 = !{!"Root", !8, i64 0, !11, i64 8, !8, i64 16, !11, i64 24}
+!8 = !{!"any pointer", !9, i64 0}
+!9 = !{!"omnipotent char", !10, i64 0}
+!10 = !{!"Simple C/C++ TBAA"}
+!11 = !{!"long", !9, i64 0}
+!12 = !{!7, !11, i64 8}
+!13 = !{!7, !8, i64 16}
+!14 = !{!7, !11, i64 24}
+!15 = !{!16, !16, i64 0}
+!16 = !{!"int", !9, i64 0}
diff --git a/hpvm/test/regressionTests/DFG2LLVM_OPENCL/ThreeLevel.cond.const.ll b/hpvm/test/regressionTests/DFG2LLVM_OPENCL/ThreeLevel.cond.const.ll
new file mode 100644
index 0000000000000000000000000000000000000000..04fd2ca0b5787990bb866151672e472ff009118c
--- /dev/null
+++ b/hpvm/test/regressionTests/DFG2LLVM_OPENCL/ThreeLevel.cond.const.ll
@@ -0,0 +1,259 @@
+; RUN: opt -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_OpenCL.so -S -localmem -dfg2llvm-opencl <  %s | FileCheck %s
+; ModuleID = 'ThreeLevel.cond.genhpvm.gpu.ll'
+source_filename = "ThreeLevel.cond.c"
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+%struct.Root = type { i32*, i64, i32*, i64 }
+%emptyStruct = type <{}>
+%emptyStruct.0 = type <{}>
+%emptyStruct.1 = type <{}>
+%emptyStruct.2 = type <{}>
+
+declare dso_local void @__hpvm__hint(i32) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__attributes(i32, ...) local_unnamed_addr #0
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture) #1
+
+declare dso_local i8* @__hpvm__getNode(...) local_unnamed_addr #0
+
+declare dso_local i8* @__hpvm__getParentNode(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNodeInstanceID_x(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNodeInstanceID_y(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNumNodeInstances_x(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNumNodeInstances_y(i8*) local_unnamed_addr #0
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture) #1
+
+declare dso_local i8* @__hpvm__createNodeND(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__bindIn(i8*, i32, i32, i32) local_unnamed_addr #0
+
+
+; Function Attrs: noinline nounwind uwtable
+define dso_local void @Launch() local_unnamed_addr #2 {
+entry:
+  %RootArgs = alloca %struct.Root, align 8
+  %0 = bitcast %struct.Root* %RootArgs to i8*
+  call void @llvm.lifetime.start.p0i8(i64 32, i8* nonnull %0) #6
+  %call = tail call noalias i8* @malloc(i64 1024) #6
+  %1 = bitcast %struct.Root* %RootArgs to i8**
+  store i8* %call, i8** %1, align 8, !tbaa !6
+  %Insize = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 1
+  store i64 1024, i64* %Insize, align 8, !tbaa !12
+  %output = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 2
+  %call1 = tail call noalias i8* @malloc(i64 1024) #6
+  %2 = bitcast i32** %output to i8**
+  store i8* %call1, i8** %2, align 8, !tbaa !13
+  %Outsize = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 3
+  store i64 1024, i64* %Outsize, align 8, !tbaa !14
+  %3 = bitcast %struct.Root* %RootArgs to i8*
+  %graphID = call i8* @llvm.hpvm.launch(i8* bitcast (%emptyStruct.2 (i32*, i64, i32*, i64)* @PipeRoot_cloned to i8*), i8* %3, i1 false)
+  call void @llvm.hpvm.wait(i8* %graphID)
+  call void @llvm.lifetime.end.p0i8(i64 32, i8* nonnull %0) #6
+  ret void
+}
+
+; Function Attrs: nofree nounwind
+declare dso_local noalias i8* @malloc(i64) local_unnamed_addr #3
+
+declare dso_local i8* @__hpvm__launch(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__wait(i8*) local_unnamed_addr #0
+
+; CHECK-LABEL: @main(
+; CHECK: call i8* @llvm_hpvm_ocl_initContext(i32
+; CHECK: call i8* @llvm_hpvm_ocl_launch(i8*
+; CHECK: call void @llvm.hpvm.init(
+; CHECK: call void @llvm_hpvm_ocl_clearContext(i8*
+; CHECK: call void @llvm.hpvm.cleanup(
+
+; CHECK-LABEL: @Func1_cloned(
+; CHECK: br i1 %cmp, label %if.end, label %if.then
+; CHECK-LABEL: if.then:
+; CHECK-LABEL: if.end:
+
+; CHECK-LABEL: @Func2_cloned.2_cloned_cloned_cloned_cloned_cloned_cloned
+; CHECK-NOT: call void @llvm_hpvm_ocl_argument_shared(i8*
+; CHECK: call i8* @llvm_hpvm_ocl_argument_ptr(i8*
+; CHECK: call void @llvm_hpvm_ocl_argument_scalar(i8*
+; CHECK-NOT: call void @llvm_hpvm_ocl_argument_shared(i8*
+; CHECK: call i8* @llvm_hpvm_ocl_argument_ptr(i8*
+; CHECK: call void @llvm_hpvm_ocl_argument_scalar(i8*
+; CHECK: call i8* @llvm_hpvm_ocl_executeNode(i8*
+; CHECK-NEXT: call void @llvm_hpvm_ocl_wait(i8*
+; CHECK-NEXT: call void @llvm_hpvm_ocl_free(i8*
+; CHECK-NEXT: call void @llvm_hpvm_ocl_free(i8*
+
+
+
+define dso_local i32 @main() local_unnamed_addr #4 {
+entry:
+  call void @llvm.hpvm.init()
+  tail call void @Launch()
+  call void @llvm.hpvm.cleanup()
+  ret i32 0
+}
+
+declare dso_local void @__hpvm__init(...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__cleanup(...) local_unnamed_addr #0
+
+declare i8* @llvm_hpvm_initializeTimerSet()
+
+declare void @llvm_hpvm_switchToTimer(i8**, i32)
+
+declare void @llvm_hpvm_printTimerSet(i8**, i8*)
+
+; Function Attrs: nounwind readnone
+declare i8* @llvm.hpvm.getNode() #5
+
+; Function Attrs: nounwind readnone
+declare i8* @llvm.hpvm.getParentNode(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNodeInstanceID.x(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNodeInstanceID.y(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNumNodeInstances.x(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNumNodeInstances.y(i8*) #5
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.createNode2D(i8*, i64, i64) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct @Func1_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %call4 = call i8* @llvm.hpvm.getNode()
+  %call15 = call i8* @llvm.hpvm.getParentNode(i8* %call4)
+  %call26 = call i64 @llvm.hpvm.getNodeInstanceID.x(i8* %call4)
+  %conv = trunc i64 %call26 to i32
+  %call37 = call i64 @llvm.hpvm.getNodeInstanceID.y(i8* %call4)
+  %conv4 = trunc i64 %call37 to i32
+  %call58 = call i64 @llvm.hpvm.getNodeInstanceID.x(i8* %call15)
+  %conv6 = trunc i64 %call58 to i32
+  %call79 = call i64 @llvm.hpvm.getNodeInstanceID.y(i8* %call15)
+  %conv8 = trunc i64 %call79 to i32
+  %call910 = call i64 @llvm.hpvm.getNumNodeInstances.x(i8* %call4)
+  %conv10 = trunc i64 %call910 to i32
+  %call1111 = call i64 @llvm.hpvm.getNumNodeInstances.y(i8* %call4)
+  %conv12 = trunc i64 %call1111 to i32
+  %mul = mul nsw i32 %conv10, %conv6
+  %add = add nsw i32 %mul, %conv
+  %mul13 = mul nsw i32 %conv12, %conv8
+  %add14 = add nsw i32 %mul13, %conv4
+  %cmp = icmp eq i32 %add, %add14
+  br i1 %cmp, label %if.end, label %if.then
+
+if.then:                                          ; preds = %entry
+  %arrayidx = getelementptr inbounds i32, i32* %In, i64 3
+  %0 = load i32, i32* %arrayidx, align 4, !tbaa !15
+  %idxprom = sext i32 %add to i64
+  %arrayidx16 = getelementptr inbounds i32, i32* %Out, i64 %idxprom
+  %1 = load i32, i32* %arrayidx16, align 4, !tbaa !15
+  %add17 = add nsw i32 %1, %0
+  store i32 %add17, i32* %arrayidx16, align 4, !tbaa !15
+  br label %if.end
+
+if.end:                                           ; preds = %if.then, %entry
+  ret %emptyStruct undef
+}
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.bind.input(i8*, i32, i32, i1) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct.0 @Func3_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %Func1_cloned.node = call i8* @llvm.hpvm.createNode2D(i8* bitcast (%emptyStruct (i32*, i64, i32*, i64)* @Func1_cloned to i8*), i64 3, i64 5)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 3, i32 3, i1 false)
+  ret %emptyStruct.0 undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.createNode(i8*) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct.1 @Func2_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %Func3_cloned.node = call i8* @llvm.hpvm.createNode2D(i8* bitcast (%emptyStruct.0 (i32*, i64, i32*, i64)* @Func3_cloned to i8*), i64 3, i64 5)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 3, i32 3, i1 false)
+  ret %emptyStruct.1 undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.launch(i8*, i8*, i1) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct.2 @PipeRoot_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %Func2_cloned.node = call i8* @llvm.hpvm.createNode(i8* bitcast (%emptyStruct.1 (i32*, i64, i32*, i64)* @Func2_cloned to i8*))
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 3, i32 3, i1 false)
+  ret %emptyStruct.2 undef
+}
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.wait(i8*) #6
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.init() #6
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.cleanup() #6
+
+attributes #0 = { "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #1 = { argmemonly nounwind }
+attributes #2 = { noinline nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #3 = { nofree nounwind "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #4 = { nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #5 = { nounwind readnone }
+attributes #6 = { nounwind }
+
+!llvm.module.flags = !{!0}
+!llvm.ident = !{!1}
+!hpvm_hint_gpu = !{!2}
+!hpvm_hint_cpu = !{!3, !4, !5}
+!hpvm_hint_spir = !{}
+!hpvm_hint_cudnn = !{}
+!hpvm_hint_promise = !{}
+!hpvm_hint_cpu_gpu = !{}
+!hpvm_hint_cpu_spir = !{}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{!"clang version 9.0.0 (https://gitlab.engr.illinois.edu/llvm/hpvm.git 3551132592a00cab6c966df508ab511598269f78)"}
+!2 = !{%emptyStruct (i32*, i64, i32*, i64)* @Func1_cloned}
+!3 = !{%emptyStruct.0 (i32*, i64, i32*, i64)* @Func3_cloned}
+!4 = !{%emptyStruct.1 (i32*, i64, i32*, i64)* @Func2_cloned}
+!5 = !{%emptyStruct.2 (i32*, i64, i32*, i64)* @PipeRoot_cloned}
+!6 = !{!7, !8, i64 0}
+!7 = !{!"Root", !8, i64 0, !11, i64 8, !8, i64 16, !11, i64 24}
+!8 = !{!"any pointer", !9, i64 0}
+!9 = !{!"omnipotent char", !10, i64 0}
+!10 = !{!"Simple C/C++ TBAA"}
+!11 = !{!"long", !9, i64 0}
+!12 = !{!7, !11, i64 8}
+!13 = !{!7, !8, i64 16}
+!14 = !{!7, !11, i64 24}
+!15 = !{!16, !16, i64 0}
+!16 = !{!"int", !9, i64 0}
diff --git a/hpvm/test/regressionTests/DFG2LLVM_OPENCL/ThreeLevel.cond.ll b/hpvm/test/regressionTests/DFG2LLVM_OPENCL/ThreeLevel.cond.ll
new file mode 100644
index 0000000000000000000000000000000000000000..1b1c589a4b0bdd9e1893d4fe51b37a03f6ff95b7
--- /dev/null
+++ b/hpvm/test/regressionTests/DFG2LLVM_OPENCL/ThreeLevel.cond.ll
@@ -0,0 +1,258 @@
+; RUN: opt -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_OpenCL.so -S -localmem -dfg2llvm-opencl <  %s | FileCheck %s
+; ModuleID = 'ThreeLevel.cond.ll'
+source_filename = "ThreeLevel.cond.c"
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+%struct.Root = type { i32*, i64, i32*, i64 }
+%emptyStruct = type <{}>
+%emptyStruct.0 = type <{}>
+%emptyStruct.1 = type <{}>
+%emptyStruct.2 = type <{}>
+
+declare dso_local void @__hpvm__hint(i32) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__attributes(i32, ...) local_unnamed_addr #0
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture) #1
+
+declare dso_local i8* @__hpvm__getNode(...) local_unnamed_addr #0
+
+declare dso_local i8* @__hpvm__getParentNode(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNodeInstanceID_x(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNodeInstanceID_y(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNumNodeInstances_x(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNumNodeInstances_y(i8*) local_unnamed_addr #0
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture) #1
+
+declare dso_local i8* @__hpvm__createNodeND(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__bindIn(i8*, i32, i32, i32) local_unnamed_addr #0
+
+; Function Attrs: noinline nounwind uwtable
+define dso_local void @Launch() local_unnamed_addr #2 {
+entry:
+  %RootArgs = alloca %struct.Root, align 8
+  %0 = bitcast %struct.Root* %RootArgs to i8*
+  call void @llvm.lifetime.start.p0i8(i64 32, i8* nonnull %0) #6
+  %call = tail call noalias i8* @malloc(i64 1024) #6
+  %1 = bitcast %struct.Root* %RootArgs to i8**
+  store i8* %call, i8** %1, align 8, !tbaa !6
+  %Insize = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 1
+  store i64 1024, i64* %Insize, align 8, !tbaa !12
+  %output = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 2
+  %call1 = tail call noalias i8* @malloc(i64 1024) #6
+  %2 = bitcast i32** %output to i8**
+  store i8* %call1, i8** %2, align 8, !tbaa !13
+  %Outsize = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 3
+  store i64 1024, i64* %Outsize, align 8, !tbaa !14
+  %3 = bitcast %struct.Root* %RootArgs to i8*
+  %graphID = call i8* @llvm.hpvm.launch(i8* bitcast (%emptyStruct.2 (i32*, i64, i32*, i64)* @PipeRoot_cloned to i8*), i8* %3, i1 false)
+  call void @llvm.hpvm.wait(i8* %graphID)
+  call void @llvm.lifetime.end.p0i8(i64 32, i8* nonnull %0) #6
+  ret void
+}
+
+; Function Attrs: nofree nounwind
+declare dso_local noalias i8* @malloc(i64) local_unnamed_addr #3
+
+declare dso_local i8* @__hpvm__launch(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__wait(i8*) local_unnamed_addr #0
+
+; CHECK-LABEL: @main(
+; CHECK: call i8* @llvm_hpvm_ocl_initContext(i32
+; CHECK: call i8* @llvm_hpvm_ocl_launch(i8*
+; CHECK: call void @llvm.hpvm.init(
+; CHECK: call void @llvm_hpvm_ocl_clearContext(i8*
+; CHECK: call void @llvm.hpvm.cleanup(
+
+; CHECK-LABEL: @Func1_cloned(
+; CHECK: br i1 %cmp, label %if.end, label %if.then
+; CHECK-LABEL: if.then:
+; CHECK-LABEL: if.end:
+
+; CHECK-LABEL: @Func2_cloned.2_cloned_cloned_cloned_cloned_cloned_cloned
+; CHECK-NOT: call void @llvm_hpvm_ocl_argument_shared(i8*
+; CHECK: call i8* @llvm_hpvm_ocl_argument_ptr(i8*
+; CHECK: call void @llvm_hpvm_ocl_argument_scalar(i8*
+; CHECK-NOT: call void @llvm_hpvm_ocl_argument_shared(i8*
+; CHECK: call i8* @llvm_hpvm_ocl_argument_ptr(i8*
+; CHECK: call void @llvm_hpvm_ocl_argument_scalar(i8*
+; CHECK: call i8* @llvm_hpvm_ocl_executeNode(i8*
+; CHECK-NEXT: call void @llvm_hpvm_ocl_wait(i8*
+; CHECK-NEXT: call void @llvm_hpvm_ocl_free(i8*
+; CHECK-NEXT: call void @llvm_hpvm_ocl_free(i8*
+
+; Function Attrs: nounwind uwtable
+define dso_local i32 @main() local_unnamed_addr #4 {
+entry:
+  call void @llvm.hpvm.init()
+  tail call void @Launch()
+  call void @llvm.hpvm.cleanup()
+  ret i32 0
+}
+
+declare dso_local void @__hpvm__init(...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__cleanup(...) local_unnamed_addr #0
+
+declare i8* @llvm_hpvm_initializeTimerSet()
+
+declare void @llvm_hpvm_switchToTimer(i8**, i32)
+
+declare void @llvm_hpvm_printTimerSet(i8**, i8*)
+
+; Function Attrs: nounwind readnone
+declare i8* @llvm.hpvm.getNode() #5
+
+; Function Attrs: nounwind readnone
+declare i8* @llvm.hpvm.getParentNode(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNodeInstanceID.x(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNodeInstanceID.y(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNumNodeInstances.x(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNumNodeInstances.y(i8*) #5
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.createNode2D(i8*, i64, i64) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct @Func1_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %call4 = call i8* @llvm.hpvm.getNode()
+  %call15 = call i8* @llvm.hpvm.getParentNode(i8* %call4)
+  %call26 = call i64 @llvm.hpvm.getNodeInstanceID.x(i8* %call4)
+  %conv = trunc i64 %call26 to i32
+  %call37 = call i64 @llvm.hpvm.getNodeInstanceID.y(i8* %call4)
+  %conv4 = trunc i64 %call37 to i32
+  %call58 = call i64 @llvm.hpvm.getNodeInstanceID.x(i8* %call15)
+  %conv6 = trunc i64 %call58 to i32
+  %call79 = call i64 @llvm.hpvm.getNodeInstanceID.y(i8* %call15)
+  %conv8 = trunc i64 %call79 to i32
+  %call910 = call i64 @llvm.hpvm.getNumNodeInstances.x(i8* %call4)
+  %conv10 = trunc i64 %call910 to i32
+  %call1111 = call i64 @llvm.hpvm.getNumNodeInstances.y(i8* %call4)
+  %conv12 = trunc i64 %call1111 to i32
+  %mul = mul nsw i32 %conv10, %conv6
+  %add = add nsw i32 %mul, %conv
+  %mul13 = mul nsw i32 %conv12, %conv8
+  %add14 = add nsw i32 %mul13, %conv4
+  %cmp = icmp eq i32 %add, %add14
+  br i1 %cmp, label %if.end, label %if.then
+
+if.then:                                          ; preds = %entry
+  %idxprom = sext i32 %add14 to i64
+  %arrayidx = getelementptr inbounds i32, i32* %In, i64 %idxprom
+  %0 = load i32, i32* %arrayidx, align 4, !tbaa !15
+  %idxprom16 = sext i32 %add to i64
+  %arrayidx17 = getelementptr inbounds i32, i32* %Out, i64 %idxprom16
+  %1 = load i32, i32* %arrayidx17, align 4, !tbaa !15
+  %add18 = add nsw i32 %1, %0
+  store i32 %add18, i32* %arrayidx17, align 4, !tbaa !15
+  br label %if.end
+
+if.end:                                           ; preds = %if.then, %entry
+  ret %emptyStruct undef
+}
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.bind.input(i8*, i32, i32, i1) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct.0 @Func3_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %Func1_cloned.node = call i8* @llvm.hpvm.createNode2D(i8* bitcast (%emptyStruct (i32*, i64, i32*, i64)* @Func1_cloned to i8*), i64 3, i64 5)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 3, i32 3, i1 false)
+  ret %emptyStruct.0 undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.createNode(i8*) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct.1 @Func2_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %Func3_cloned.node = call i8* @llvm.hpvm.createNode2D(i8* bitcast (%emptyStruct.0 (i32*, i64, i32*, i64)* @Func3_cloned to i8*), i64 3, i64 5)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 3, i32 3, i1 false)
+  ret %emptyStruct.1 undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.launch(i8*, i8*, i1) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct.2 @PipeRoot_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %Func2_cloned.node = call i8* @llvm.hpvm.createNode(i8* bitcast (%emptyStruct.1 (i32*, i64, i32*, i64)* @Func2_cloned to i8*))
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 3, i32 3, i1 false)
+  ret %emptyStruct.2 undef
+}
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.wait(i8*) #6
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.init() #6
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.cleanup() #6
+
+attributes #0 = { "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #1 = { argmemonly nounwind }
+attributes #2 = { noinline nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #3 = { nofree nounwind "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #4 = { nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #5 = { nounwind readnone }
+attributes #6 = { nounwind }
+
+!llvm.module.flags = !{!0}
+!llvm.ident = !{!1}
+!hpvm_hint_gpu = !{!2}
+!hpvm_hint_cpu = !{!3, !4, !5}
+!hpvm_hint_spir = !{}
+!hpvm_hint_cudnn = !{}
+!hpvm_hint_promise = !{}
+!hpvm_hint_cpu_gpu = !{}
+!hpvm_hint_cpu_spir = !{}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{!"clang version 9.0.0 (https://gitlab.engr.illinois.edu/llvm/hpvm.git 3551132592a00cab6c966df508ab511598269f78)"}
+!2 = !{%emptyStruct (i32*, i64, i32*, i64)* @Func1_cloned}
+!3 = !{%emptyStruct.0 (i32*, i64, i32*, i64)* @Func3_cloned}
+!4 = !{%emptyStruct.1 (i32*, i64, i32*, i64)* @Func2_cloned}
+!5 = !{%emptyStruct.2 (i32*, i64, i32*, i64)* @PipeRoot_cloned}
+!6 = !{!7, !8, i64 0}
+!7 = !{!"Root", !8, i64 0, !11, i64 8, !8, i64 16, !11, i64 24}
+!8 = !{!"any pointer", !9, i64 0}
+!9 = !{!"omnipotent char", !10, i64 0}
+!10 = !{!"Simple C/C++ TBAA"}
+!11 = !{!"long", !9, i64 0}
+!12 = !{!7, !11, i64 8}
+!13 = !{!7, !8, i64 16}
+!14 = !{!7, !11, i64 24}
+!15 = !{!16, !16, i64 0}
+!16 = !{!"int", !9, i64 0}
diff --git a/hpvm/test/regressionTests/DFG2LLVM_OPENCL/ThreeLevel.constmem.ll b/hpvm/test/regressionTests/DFG2LLVM_OPENCL/ThreeLevel.constmem.ll
new file mode 100644
index 0000000000000000000000000000000000000000..5de9fd4e33adf8bfa819c8d4c2eb2a52f7a3a15c
--- /dev/null
+++ b/hpvm/test/regressionTests/DFG2LLVM_OPENCL/ThreeLevel.constmem.ll
@@ -0,0 +1,261 @@
+; RUN: opt -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_OpenCL.so -S -localmem -dfg2llvm-opencl <  %s | FileCheck %s
+; ModuleID = 'ThreeLevel.opt.ll'
+source_filename = "ThreeLevel.opt.c"
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+%struct.Root = type { i32*, i64, i32*, i64 }
+%struct.out.Allocation = type <{ i8*, i64 }>
+%emptyStruct = type <{}>
+%emptyStruct.0 = type <{}>
+%emptyStruct.1 = type <{}>
+%emptyStruct.2 = type <{}>
+
+declare dso_local void @__hpvm__hint(i32) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__attributes(i32, ...) local_unnamed_addr #0
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture) #1
+
+declare dso_local i8* @__hpvm__getNode(...) local_unnamed_addr #0
+
+declare dso_local i8* @__hpvm__getParentNode(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNodeInstanceID_x(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNodeInstanceID_y(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNumNodeInstances_x(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNumNodeInstances_y(i8*) local_unnamed_addr #0
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture) #1
+
+declare dso_local i8* @__hpvm__malloc(i64) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__return(i32, ...) local_unnamed_addr #0
+
+declare dso_local i8* @__hpvm__createNodeND(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__bindIn(i8*, i32, i32, i32) local_unnamed_addr #0
+
+declare dso_local i8* @__hpvm__edge(i8*, i8*, i32, i32, i32, i32) local_unnamed_addr #0
+
+; Function Attrs: noinline nounwind uwtable
+define dso_local void @Launch() local_unnamed_addr #2 {
+entry:
+  %RootArgs = alloca %struct.Root, align 8
+  %0 = bitcast %struct.Root* %RootArgs to i8*
+  call void @llvm.lifetime.start.p0i8(i64 32, i8* nonnull %0) #6
+  %call = tail call noalias i8* @malloc(i64 1024) #6
+  %1 = bitcast %struct.Root* %RootArgs to i8**
+  store i8* %call, i8** %1, align 8, !tbaa !6
+  %Insize = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 1
+  store i64 1024, i64* %Insize, align 8, !tbaa !12
+  %output = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 2
+  %call1 = tail call noalias i8* @malloc(i64 1024) #6
+  %2 = bitcast i32** %output to i8**
+  store i8* %call1, i8** %2, align 8, !tbaa !13
+  %Outsize = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 3
+  store i64 1024, i64* %Outsize, align 8, !tbaa !14
+  %3 = bitcast %struct.Root* %RootArgs to i8*
+  %graphID = call i8* @llvm.hpvm.launch(i8* bitcast (%emptyStruct.2 (i32*, i64, i32*, i64)* @PipeRoot_cloned to i8*), i8* %3, i1 false)
+  call void @llvm.hpvm.wait(i8* %graphID)
+  call void @llvm.lifetime.end.p0i8(i64 32, i8* nonnull %0) #6
+  ret void
+}
+
+; Function Attrs: nofree nounwind
+declare dso_local noalias i8* @malloc(i64) local_unnamed_addr #3
+
+declare dso_local i8* @__hpvm__launch(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__wait(i8*) local_unnamed_addr #0
+
+; CHECK-LABEL: @main(
+; CHECK: call i8* @llvm_hpvm_ocl_initContext(i32
+; CHECK: call i8* @llvm_hpvm_ocl_launch(i8*
+; CHECK: call void @llvm_hpvm_ocl_clearContext(i8*
+
+; CHECK-LABEL: @Allocation_cloned.2
+
+; CHECK-LABEL: @Func2_cloned.3_cloned_cloned_cloned_cloned_cloned_cloned
+; CHECK: call i8* @llvm_hpvm_ocl_argument_ptr(i8*
+; CHECK: call void @llvm_hpvm_ocl_argument_scalar(i8*
+; CHECK: @Allocation_cloned.2(
+; CHECK: call void @llvm_hpvm_ocl_argument_shared(i8*
+; CHECK: call void @llvm_hpvm_ocl_argument_scalar(i8*
+; CHECK: call i8* @llvm_hpvm_ocl_executeNode(i8*
+; CHECK-NEXT: call void @llvm_hpvm_ocl_wait(i8*
+; CHECK-NEXT: call void @llvm_hpvm_ocl_free(i8*
+
+
+; Function Attrs: nounwind uwtable
+define dso_local i32 @main() local_unnamed_addr #4 {
+entry:
+  call void @llvm.hpvm.init()
+  tail call void @Launch()
+  call void @llvm.hpvm.cleanup()
+  ret i32 0
+}
+
+declare dso_local void @__hpvm__init(...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__cleanup(...) local_unnamed_addr #0
+
+declare i8* @llvm_hpvm_initializeTimerSet()
+
+declare void @llvm_hpvm_switchToTimer(i8**, i32)
+
+declare void @llvm_hpvm_printTimerSet(i8**, i8*)
+
+; Function Attrs: nounwind readnone
+declare i8* @llvm.hpvm.getNode() #5
+
+; Function Attrs: nounwind readnone
+declare i8* @llvm.hpvm.getParentNode(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNodeInstanceID.x(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNodeInstanceID.y(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNumNodeInstances.x(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNumNodeInstances.y(i8*) #5
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.malloc(i64) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %struct.out.Allocation @Allocation_cloned(i64 %block) #4 {
+entry:
+  %call1 = call i8* @llvm.hpvm.malloc(i64 %block)
+  %returnStruct = insertvalue %struct.out.Allocation undef, i8* %call1, 0
+  %returnStruct2 = insertvalue %struct.out.Allocation %returnStruct, i64 %block, 1
+  ret %struct.out.Allocation %returnStruct2
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.createNode2D(i8*, i64, i64) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct @Func1_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %call4 = call i8* @llvm.hpvm.getNode()
+  %call15 = call i8* @llvm.hpvm.getParentNode(i8* %call4)
+  %call26 = call i64 @llvm.hpvm.getNodeInstanceID.x(i8* %call4)
+  %call37 = call i64 @llvm.hpvm.getNodeInstanceID.y(i8* %call4)
+  %call58 = call i64 @llvm.hpvm.getNodeInstanceID.x(i8* %call15)
+  %call79 = call i64 @llvm.hpvm.getNodeInstanceID.y(i8* %call15)
+  %call910 = call i64 @llvm.hpvm.getNumNodeInstances.x(i8* %call4)
+  %call1111 = call i64 @llvm.hpvm.getNumNodeInstances.y(i8* %call4)
+  %mul = mul i64 %call910, %call58
+  %add = add i64 %mul, %call26
+  %arrayidx = getelementptr inbounds i32, i32* %In, i64 3
+  %0 = load i32, i32* %arrayidx, align 4, !tbaa !15
+  %sext = shl i64 %add, 32
+  %idxprom = ashr exact i64 %sext, 32
+  %arrayidx15 = getelementptr inbounds i32, i32* %Out, i64 %idxprom
+  %1 = load i32, i32* %arrayidx15, align 4, !tbaa !15
+  %add16 = add nsw i32 %1, %0
+  store i32 %add16, i32* %arrayidx15, align 4, !tbaa !15
+  ret %emptyStruct undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.createNode(i8*) #6
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.bind.input(i8*, i32, i32, i1) #6
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.createEdge(i8*, i8*, i1, i32, i32, i1) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct.0 @Func3_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %Func1_cloned.node = call i8* @llvm.hpvm.createNode2D(i8* bitcast (%emptyStruct (i32*, i64, i32*, i64)* @Func1_cloned to i8*), i64 3, i64 5)
+  %Allocation_cloned.node = call i8* @llvm.hpvm.createNode(i8* bitcast (%struct.out.Allocation (i64)* @Allocation_cloned to i8*))
+  call void @llvm.hpvm.bind.input(i8* %Allocation_cloned.node, i32 1, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 3, i32 3, i1 false)
+  %output = call i8* @llvm.hpvm.createEdge(i8* %Allocation_cloned.node, i8* %Func1_cloned.node, i1 true, i32 0, i32 0, i1 false)
+  %output1 = call i8* @llvm.hpvm.createEdge(i8* %Allocation_cloned.node, i8* %Func1_cloned.node, i1 true, i32 1, i32 1, i1 false)
+  ret %emptyStruct.0 undef
+}
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct.1 @Func2_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %Func3_cloned.node = call i8* @llvm.hpvm.createNode2D(i8* bitcast (%emptyStruct.0 (i32*, i64, i32*, i64)* @Func3_cloned to i8*), i64 3, i64 5)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 3, i32 3, i1 false)
+  ret %emptyStruct.1 undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.launch(i8*, i8*, i1) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct.2 @PipeRoot_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %Func2_cloned.node = call i8* @llvm.hpvm.createNode(i8* bitcast (%emptyStruct.1 (i32*, i64, i32*, i64)* @Func2_cloned to i8*))
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 3, i32 3, i1 false)
+  ret %emptyStruct.2 undef
+}
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.wait(i8*) #6
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.init() #6
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.cleanup() #6
+
+attributes #0 = { "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #1 = { argmemonly nounwind }
+attributes #2 = { noinline nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #3 = { nofree nounwind "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #4 = { nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #5 = { nounwind readnone }
+attributes #6 = { nounwind }
+
+!llvm.module.flags = !{!0}
+!llvm.ident = !{!1}
+!hpvm_hint_gpu = !{!2}
+!hpvm_hint_cpu = !{!3, !4, !5}
+!hpvm_hint_spir = !{}
+!hpvm_hint_cudnn = !{}
+!hpvm_hint_promise = !{}
+!hpvm_hint_cpu_gpu = !{}
+!hpvm_hint_cpu_spir = !{}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{!"clang version 9.0.0 (https://gitlab.engr.illinois.edu/llvm/hpvm.git 6690f9e7e8b46b96aea222d3e85315cd63545953)"}
+!2 = !{%emptyStruct (i32*, i64, i32*, i64)* @Func1_cloned}
+!3 = !{%emptyStruct.0 (i32*, i64, i32*, i64)* @Func3_cloned}
+!4 = !{%emptyStruct.1 (i32*, i64, i32*, i64)* @Func2_cloned}
+!5 = !{%emptyStruct.2 (i32*, i64, i32*, i64)* @PipeRoot_cloned}
+!6 = !{!7, !8, i64 0}
+!7 = !{!"Root", !8, i64 0, !11, i64 8, !8, i64 16, !11, i64 24}
+!8 = !{!"any pointer", !9, i64 0}
+!9 = !{!"omnipotent char", !10, i64 0}
+!10 = !{!"Simple C/C++ TBAA"}
+!11 = !{!"long", !9, i64 0}
+!12 = !{!7, !11, i64 8}
+!13 = !{!7, !8, i64 16}
+!14 = !{!7, !11, i64 24}
+!15 = !{!16, !16, i64 0}
+!16 = !{!"int", !9, i64 0}
diff --git a/hpvm/test/regressionTests/DFG2LLVM_OPENCL/ThreeLevel.ll b/hpvm/test/regressionTests/DFG2LLVM_OPENCL/ThreeLevel.ll
new file mode 100644
index 0000000000000000000000000000000000000000..9af2e48674ae0867e74eb46da2a399d3e1d6df44
--- /dev/null
+++ b/hpvm/test/regressionTests/DFG2LLVM_OPENCL/ThreeLevel.ll
@@ -0,0 +1,239 @@
+; RUN: opt -load LLVMBuildDFG.so -load LLVMLocalMem.so -load LLVMDFG2LLVM_OpenCL.so -S -localmem -dfg2llvm-opencl <  %s | FileCheck %s
+; ModuleID = 'ThreeLevel.ll'
+source_filename = "ThreeLevel.c"
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+%struct.Root = type { i32*, i64, i32*, i64 }
+%emptyStruct = type <{}>
+%emptyStruct.0 = type <{}>
+%emptyStruct.1 = type <{}>
+%emptyStruct.2 = type <{}>
+
+declare dso_local void @__hpvm__hint(i32) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__attributes(i32, ...) local_unnamed_addr #0
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture) #1
+
+declare dso_local i8* @__hpvm__getNode(...) local_unnamed_addr #0
+
+declare dso_local i8* @__hpvm__getParentNode(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNodeInstanceID_x(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNodeInstanceID_y(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNumNodeInstances_x(i8*) local_unnamed_addr #0
+
+declare dso_local i64 @__hpvm__getNumNodeInstances_y(i8*) local_unnamed_addr #0
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture) #1
+
+declare dso_local i8* @__hpvm__createNodeND(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__bindIn(i8*, i32, i32, i32) local_unnamed_addr #0
+
+; Function Attrs: noinline nounwind uwtable
+define dso_local void @Launch() local_unnamed_addr #2 {
+entry:
+  %RootArgs = alloca %struct.Root, align 8
+  %0 = bitcast %struct.Root* %RootArgs to i8*
+  call void @llvm.lifetime.start.p0i8(i64 32, i8* nonnull %0) #6
+  %call = tail call noalias i8* @malloc(i64 1024) #6
+  %1 = bitcast %struct.Root* %RootArgs to i8**
+  store i8* %call, i8** %1, align 8, !tbaa !6
+  %Insize = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 1
+  store i64 1024, i64* %Insize, align 8, !tbaa !12
+  %output = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 2
+  %call1 = tail call noalias i8* @malloc(i64 1024) #6
+  %2 = bitcast i32** %output to i8**
+  store i8* %call1, i8** %2, align 8, !tbaa !13
+  %Outsize = getelementptr inbounds %struct.Root, %struct.Root* %RootArgs, i64 0, i32 3
+  store i64 1024, i64* %Outsize, align 8, !tbaa !14
+  %3 = bitcast %struct.Root* %RootArgs to i8*
+  %graphID = call i8* @llvm.hpvm.launch(i8* bitcast (%emptyStruct.2 (i32*, i64, i32*, i64)* @PipeRoot_cloned to i8*), i8* %3, i1 false)
+  call void @llvm.hpvm.wait(i8* %graphID)
+  call void @llvm.lifetime.end.p0i8(i64 32, i8* nonnull %0) #6
+  ret void
+}
+
+; Function Attrs: nofree nounwind
+declare dso_local noalias i8* @malloc(i64) local_unnamed_addr #3
+
+declare dso_local i8* @__hpvm__launch(i32, ...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__wait(i8*) local_unnamed_addr #0
+
+; CHECK-LABEL: @main(
+; CHECK: call i8* @llvm_hpvm_ocl_initContext(i32
+; CHECK: call i8* @llvm_hpvm_ocl_launch(i8*
+; CHECK: call void @llvm_hpvm_ocl_clearContext(i8*
+
+; CHECK-LABEL: @Func2_cloned.2_cloned_cloned_cloned_cloned_cloned_cloned
+; CHECK: call i8* @llvm_hpvm_ocl_argument_ptr(i8*
+; CHECK: call void @llvm_hpvm_ocl_argument_scalar(i8*
+; CHECK: call i8* @llvm_hpvm_ocl_argument_ptr(i8*
+; CHECK: call void @llvm_hpvm_ocl_argument_scalar(i8*
+; CHECK: call i8* @llvm_hpvm_ocl_executeNode(i8*
+; CHECK-NEXT: call void @llvm_hpvm_ocl_wait(i8*
+; CHECK-NEXT: call void @llvm_hpvm_ocl_free(i8*
+; CHECK-NEXT: call void @llvm_hpvm_ocl_free(i8*
+
+
+; Function Attrs: nounwind uwtable
+define dso_local i32 @main() local_unnamed_addr #4 {
+entry:
+  call void @llvm.hpvm.init()
+  tail call void @Launch()
+  call void @llvm.hpvm.cleanup()
+  ret i32 0
+}
+
+declare dso_local void @__hpvm__init(...) local_unnamed_addr #0
+
+declare dso_local void @__hpvm__cleanup(...) local_unnamed_addr #0
+
+declare i8* @llvm_hpvm_initializeTimerSet()
+
+declare void @llvm_hpvm_switchToTimer(i8**, i32)
+
+declare void @llvm_hpvm_printTimerSet(i8**, i8*)
+
+; Function Attrs: nounwind readnone
+declare i8* @llvm.hpvm.getNode() #5
+
+; Function Attrs: nounwind readnone
+declare i8* @llvm.hpvm.getParentNode(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNodeInstanceID.x(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNodeInstanceID.y(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNumNodeInstances.x(i8*) #5
+
+; Function Attrs: nounwind readnone
+declare i64 @llvm.hpvm.getNumNodeInstances.y(i8*) #5
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.createNode2D(i8*, i64, i64) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct @Func1_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %call4 = call i8* @llvm.hpvm.getNode()
+  %call15 = call i8* @llvm.hpvm.getParentNode(i8* %call4)
+  %call26 = call i64 @llvm.hpvm.getNodeInstanceID.x(i8* %call4)
+  %call37 = call i64 @llvm.hpvm.getNodeInstanceID.y(i8* %call4)
+  %call58 = call i64 @llvm.hpvm.getNodeInstanceID.x(i8* %call15)
+  %call79 = call i64 @llvm.hpvm.getNodeInstanceID.y(i8* %call15)
+  %call910 = call i64 @llvm.hpvm.getNumNodeInstances.x(i8* %call4)
+  %call1111 = call i64 @llvm.hpvm.getNumNodeInstances.y(i8* %call4)
+  %mul = mul i64 %call910, %call58
+  %add = add i64 %mul, %call26
+  %mul13 = mul i64 %call1111, %call79
+  %add14 = add i64 %mul13, %call37
+  %sext = shl i64 %add14, 32
+  %idxprom = ashr exact i64 %sext, 32
+  %arrayidx = getelementptr inbounds i32, i32* %In, i64 %idxprom
+  %0 = load i32, i32* %arrayidx, align 4, !tbaa !15
+  %sext36 = shl i64 %add, 32
+  %idxprom15 = ashr exact i64 %sext36, 32
+  %arrayidx16 = getelementptr inbounds i32, i32* %Out, i64 %idxprom15
+  %1 = load i32, i32* %arrayidx16, align 4, !tbaa !15
+  %add17 = add nsw i32 %1, %0
+  store i32 %add17, i32* %arrayidx16, align 4, !tbaa !15
+  ret %emptyStruct undef
+}
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.bind.input(i8*, i32, i32, i1) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct.0 @Func3_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %Func1_cloned.node = call i8* @llvm.hpvm.createNode2D(i8* bitcast (%emptyStruct (i32*, i64, i32*, i64)* @Func1_cloned to i8*), i64 3, i64 5)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func1_cloned.node, i32 3, i32 3, i1 false)
+  ret %emptyStruct.0 undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.createNode(i8*) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct.1 @Func2_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %Func3_cloned.node = call i8* @llvm.hpvm.createNode2D(i8* bitcast (%emptyStruct.0 (i32*, i64, i32*, i64)* @Func3_cloned to i8*), i64 3, i64 5)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func3_cloned.node, i32 3, i32 3, i1 false)
+  ret %emptyStruct.1 undef
+}
+
+; Function Attrs: nounwind
+declare i8* @llvm.hpvm.launch(i8*, i8*, i1) #6
+
+; Function Attrs: nounwind uwtable
+define dso_local %emptyStruct.2 @PipeRoot_cloned(i32* in %In, i64 %Insize, i32* in out %Out, i64 %Outsize) #4 {
+entry:
+  %Func2_cloned.node = call i8* @llvm.hpvm.createNode(i8* bitcast (%emptyStruct.1 (i32*, i64, i32*, i64)* @Func2_cloned to i8*))
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 0, i32 0, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 1, i32 1, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 2, i32 2, i1 false)
+  call void @llvm.hpvm.bind.input(i8* %Func2_cloned.node, i32 3, i32 3, i1 false)
+  ret %emptyStruct.2 undef
+}
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.wait(i8*) #6
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.init() #6
+
+; Function Attrs: nounwind
+declare void @llvm.hpvm.cleanup() #6
+
+attributes #0 = { "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #1 = { argmemonly nounwind }
+attributes #2 = { noinline nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #3 = { nofree nounwind "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #4 = { nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="false" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="true" "use-soft-float"="false" }
+attributes #5 = { nounwind readnone }
+attributes #6 = { nounwind }
+
+!llvm.module.flags = !{!0}
+!llvm.ident = !{!1}
+!hpvm_hint_gpu = !{!2}
+!hpvm_hint_cpu = !{!3, !4, !5}
+!hpvm_hint_spir = !{}
+!hpvm_hint_cudnn = !{}
+!hpvm_hint_promise = !{}
+!hpvm_hint_cpu_gpu = !{}
+!hpvm_hint_cpu_spir = !{}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{!"clang version 9.0.0 (https://gitlab.engr.illinois.edu/llvm/hpvm.git 6690f9e7e8b46b96aea222d3e85315cd63545953)"}
+!2 = !{%emptyStruct (i32*, i64, i32*, i64)* @Func1_cloned}
+!3 = !{%emptyStruct.0 (i32*, i64, i32*, i64)* @Func3_cloned}
+!4 = !{%emptyStruct.1 (i32*, i64, i32*, i64)* @Func2_cloned}
+!5 = !{%emptyStruct.2 (i32*, i64, i32*, i64)* @PipeRoot_cloned}
+!6 = !{!7, !8, i64 0}
+!7 = !{!"Root", !8, i64 0, !11, i64 8, !8, i64 16, !11, i64 24}
+!8 = !{!"any pointer", !9, i64 0}
+!9 = !{!"omnipotent char", !10, i64 0}
+!10 = !{!"Simple C/C++ TBAA"}
+!11 = !{!"long", !9, i64 0}
+!12 = !{!7, !11, i64 8}
+!13 = !{!7, !8, i64 16}
+!14 = !{!7, !11, i64 24}
+!15 = !{!16, !16, i64 0}
+!16 = !{!"int", !9, i64 0}