diff --git a/backend/src/backend/gen_program.cpp b/backend/src/backend/gen_program.cpp
index f4c74f8..f1b9961 100644
--- a/backend/src/backend/gen_program.cpp
+++ b/backend/src/backend/gen_program.cpp
@@ -393,7 +393,11 @@ namespace gbe {
       llvm::Module* src = (llvm::Module*)((GenProgram*)src_program)->module;
       llvm::Module* dst = (llvm::Module*)((GenProgram*)dst_program)->module;
 
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+      if (LLVMLinkModules(wrap(dst), wrap(src), LLVMLinkerPreserveSource_Removed, &errMsg)) {
+#else
       if (LLVMLinkModules(wrap(dst), wrap(src), LLVMLinkerPreserveSource, &errMsg)) {
+#endif
         if (err != NULL && errSize != NULL && stringSize > 0u) {
           strncpy(err, errMsg, stringSize-1);
           err[stringSize-1] = '\0';
diff --git a/backend/src/libocl/src/ocl_memcpy.ll b/backend/src/libocl/src/ocl_memcpy.ll
index b3fadb2..dae3166 100644
--- a/backend/src/libocl/src/ocl_memcpy.ll
+++ b/backend/src/libocl/src/ocl_memcpy.ll
@@ -22,10 +22,10 @@ while.cond:                                       ; preds = %while.body, %entry
   br i1 %cmp, label %while.cond3, label %while.body
 
 while.body:                                       ; preds = %while.cond
-  %add.ptr = getelementptr inbounds i8 addrspace(1)* %src, i32 %index.0
+  %add.ptr = getelementptr inbounds i8, i8 addrspace(1)* %src, i32 %index.0
   %0 = bitcast i8 addrspace(1)* %add.ptr to i32 addrspace(1)*
-  %1 = load i32 addrspace(1)* %0, align 4
-  %add.ptr1 = getelementptr inbounds i8 addrspace(1)* %dst, i32 %index.0
+  %1 = load i32, i32 addrspace(1)* %0, align 4
+  %add.ptr1 = getelementptr inbounds i8, i8 addrspace(1)* %dst, i32 %index.0
   %2 = bitcast i8 addrspace(1)* %add.ptr1 to i32 addrspace(1)*
   store i32 %1, i32 addrspace(1)* %2, align 4
   br label %while.cond
@@ -36,9 +36,9 @@ while.cond3:                                      ; preds = %while.cond, %while.
   br i1 %cmp4, label %while.body5, label %while.end7
 
 while.body5:                                      ; preds = %while.cond3
-  %arrayidx = getelementptr inbounds i8 addrspace(1)* %src, i32 %index.1
-  %3 = load i8 addrspace(1)* %arrayidx, align 1
-  %arrayidx6 = getelementptr inbounds i8 addrspace(1)* %dst, i32 %index.1
+  %arrayidx = getelementptr inbounds i8, i8 addrspace(1)* %src, i32 %index.1
+  %3 = load i8, i8 addrspace(1)* %arrayidx, align 1
+  %arrayidx6 = getelementptr inbounds i8, i8 addrspace(1)* %dst, i32 %index.1
   store i8 %3, i8 addrspace(1)* %arrayidx6, align 1
   %inc = add i32 %index.1, 1
   br label %while.cond3
@@ -58,10 +58,10 @@ while.cond:                                       ; preds = %while.body, %entry
   br i1 %cmp, label %while.cond3, label %while.body
 
 while.body:                                       ; preds = %while.cond
-  %add.ptr = getelementptr inbounds i8 addrspace(0)* %src, i32 %index.0
+  %add.ptr = getelementptr inbounds i8, i8 addrspace(0)* %src, i32 %index.0
   %0 = bitcast i8 addrspace(0)* %add.ptr to i32 addrspace(0)*
-  %1 = load i32 addrspace(0)* %0, align 4
-  %add.ptr1 = getelementptr inbounds i8 addrspace(1)* %dst, i32 %index.0
+  %1 = load i32, i32 addrspace(0)* %0, align 4
+  %add.ptr1 = getelementptr inbounds i8, i8 addrspace(1)* %dst, i32 %index.0
   %2 = bitcast i8 addrspace(1)* %add.ptr1 to i32 addrspace(1)*
   store i32 %1, i32 addrspace(1)* %2, align 4
   br label %while.cond
@@ -72,9 +72,9 @@ while.cond3:                                      ; preds = %while.cond, %while.
   br i1 %cmp4, label %while.body5, label %while.end7
 
 while.body5:                                      ; preds = %while.cond3
-  %arrayidx = getelementptr inbounds i8 addrspace(0)* %src, i32 %index.1
-  %3 = load i8 addrspace(0)* %arrayidx, align 1
-  %arrayidx6 = getelementptr inbounds i8 addrspace(1)* %dst, i32 %index.1
+  %arrayidx = getelementptr inbounds i8, i8 addrspace(0)* %src, i32 %index.1
+  %3 = load i8, i8 addrspace(0)* %arrayidx, align 1
+  %arrayidx6 = getelementptr inbounds i8, i8 addrspace(1)* %dst, i32 %index.1
   store i8 %3, i8 addrspace(1)* %arrayidx6, align 1
   %inc = add i32 %index.1, 1
   br label %while.cond3
@@ -94,10 +94,10 @@ while.cond:                                       ; preds = %while.body, %entry
   br i1 %cmp, label %while.cond3, label %while.body
 
 while.body:                                       ; preds = %while.cond
-  %add.ptr = getelementptr inbounds i8 addrspace(3)* %src, i32 %index.0
+  %add.ptr = getelementptr inbounds i8, i8 addrspace(3)* %src, i32 %index.0
   %0 = bitcast i8 addrspace(3)* %add.ptr to i32 addrspace(3)*
-  %1 = load i32 addrspace(3)* %0, align 4
-  %add.ptr1 = getelementptr inbounds i8 addrspace(1)* %dst, i32 %index.0
+  %1 = load i32, i32 addrspace(3)* %0, align 4
+  %add.ptr1 = getelementptr inbounds i8, i8 addrspace(1)* %dst, i32 %index.0
   %2 = bitcast i8 addrspace(1)* %add.ptr1 to i32 addrspace(1)*
   store i32 %1, i32 addrspace(1)* %2, align 4
   br label %while.cond
@@ -108,9 +108,9 @@ while.cond3:                                      ; preds = %while.cond, %while.
   br i1 %cmp4, label %while.body5, label %while.end7
 
 while.body5:                                      ; preds = %while.cond3
-  %arrayidx = getelementptr inbounds i8 addrspace(3)* %src, i32 %index.1
-  %3 = load i8 addrspace(3)* %arrayidx, align 1
-  %arrayidx6 = getelementptr inbounds i8 addrspace(1)* %dst, i32 %index.1
+  %arrayidx = getelementptr inbounds i8, i8 addrspace(3)* %src, i32 %index.1
+  %3 = load i8, i8 addrspace(3)* %arrayidx, align 1
+  %arrayidx6 = getelementptr inbounds i8, i8 addrspace(1)* %dst, i32 %index.1
   store i8 %3, i8 addrspace(1)* %arrayidx6, align 1
   %inc = add i32 %index.1, 1
   br label %while.cond3
@@ -130,10 +130,10 @@ while.cond:                                       ; preds = %while.body, %entry
   br i1 %cmp, label %while.cond3, label %while.body
 
 while.body:                                       ; preds = %while.cond
-  %add.ptr = getelementptr inbounds i8 addrspace(1)* %src, i32 %index.0
+  %add.ptr = getelementptr inbounds i8, i8 addrspace(1)* %src, i32 %index.0
   %0 = bitcast i8 addrspace(1)* %add.ptr to i32 addrspace(1)*
-  %1 = load i32 addrspace(1)* %0, align 4
-  %add.ptr1 = getelementptr inbounds i8 addrspace(0)* %dst, i32 %index.0
+  %1 = load i32, i32 addrspace(1)* %0, align 4
+  %add.ptr1 = getelementptr inbounds i8, i8 addrspace(0)* %dst, i32 %index.0
   %2 = bitcast i8 addrspace(0)* %add.ptr1 to i32 addrspace(0)*
   store i32 %1, i32 addrspace(0)* %2, align 4
   br label %while.cond
@@ -144,9 +144,9 @@ while.cond3:                                      ; preds = %while.cond, %while.
   br i1 %cmp4, label %while.body5, label %while.end7
 
 while.body5:                                      ; preds = %while.cond3
-  %arrayidx = getelementptr inbounds i8 addrspace(1)* %src, i32 %index.1
-  %3 = load i8 addrspace(1)* %arrayidx, align 1
-  %arrayidx6 = getelementptr inbounds i8 addrspace(0)* %dst, i32 %index.1
+  %arrayidx = getelementptr inbounds i8, i8 addrspace(1)* %src, i32 %index.1
+  %3 = load i8, i8 addrspace(1)* %arrayidx, align 1
+  %arrayidx6 = getelementptr inbounds i8, i8 addrspace(0)* %dst, i32 %index.1
   store i8 %3, i8 addrspace(0)* %arrayidx6, align 1
   %inc = add i32 %index.1, 1
   br label %while.cond3
@@ -166,10 +166,10 @@ while.cond:                                       ; preds = %while.body, %entry
   br i1 %cmp, label %while.cond3, label %while.body
 
 while.body:                                       ; preds = %while.cond
-  %add.ptr = getelementptr inbounds i8 addrspace(0)* %src, i32 %index.0
+  %add.ptr = getelementptr inbounds i8, i8 addrspace(0)* %src, i32 %index.0
   %0 = bitcast i8 addrspace(0)* %add.ptr to i32 addrspace(0)*
-  %1 = load i32 addrspace(0)* %0, align 4
-  %add.ptr1 = getelementptr inbounds i8 addrspace(0)* %dst, i32 %index.0
+  %1 = load i32, i32 addrspace(0)* %0, align 4
+  %add.ptr1 = getelementptr inbounds i8, i8 addrspace(0)* %dst, i32 %index.0
   %2 = bitcast i8 addrspace(0)* %add.ptr1 to i32 addrspace(0)*
   store i32 %1, i32 addrspace(0)* %2, align 4
   br label %while.cond
@@ -180,9 +180,9 @@ while.cond3:                                      ; preds = %while.cond, %while.
   br i1 %cmp4, label %while.body5, label %while.end7
 
 while.body5:                                      ; preds = %while.cond3
-  %arrayidx = getelementptr inbounds i8 addrspace(0)* %src, i32 %index.1
-  %3 = load i8 addrspace(0)* %arrayidx, align 1
-  %arrayidx6 = getelementptr inbounds i8 addrspace(0)* %dst, i32 %index.1
+  %arrayidx = getelementptr inbounds i8, i8 addrspace(0)* %src, i32 %index.1
+  %3 = load i8, i8 addrspace(0)* %arrayidx, align 1
+  %arrayidx6 = getelementptr inbounds i8, i8 addrspace(0)* %dst, i32 %index.1
   store i8 %3, i8 addrspace(0)* %arrayidx6, align 1
   %inc = add i32 %index.1, 1
   br label %while.cond3
@@ -202,10 +202,10 @@ while.cond:                                       ; preds = %while.body, %entry
   br i1 %cmp, label %while.cond3, label %while.body
 
 while.body:                                       ; preds = %while.cond
-  %add.ptr = getelementptr inbounds i8 addrspace(3)* %src, i32 %index.0
+  %add.ptr = getelementptr inbounds i8, i8 addrspace(3)* %src, i32 %index.0
   %0 = bitcast i8 addrspace(3)* %add.ptr to i32 addrspace(3)*
-  %1 = load i32 addrspace(3)* %0, align 4
-  %add.ptr1 = getelementptr inbounds i8 addrspace(0)* %dst, i32 %index.0
+  %1 = load i32, i32 addrspace(3)* %0, align 4
+  %add.ptr1 = getelementptr inbounds i8, i8 addrspace(0)* %dst, i32 %index.0
   %2 = bitcast i8 addrspace(0)* %add.ptr1 to i32 addrspace(0)*
   store i32 %1, i32 addrspace(0)* %2, align 4
   br label %while.cond
@@ -216,9 +216,9 @@ while.cond3:                                      ; preds = %while.cond, %while.
   br i1 %cmp4, label %while.body5, label %while.end7
 
 while.body5:                                      ; preds = %while.cond3
-  %arrayidx = getelementptr inbounds i8 addrspace(3)* %src, i32 %index.1
-  %3 = load i8 addrspace(3)* %arrayidx, align 1
-  %arrayidx6 = getelementptr inbounds i8 addrspace(0)* %dst, i32 %index.1
+  %arrayidx = getelementptr inbounds i8, i8 addrspace(3)* %src, i32 %index.1
+  %3 = load i8, i8 addrspace(3)* %arrayidx, align 1
+  %arrayidx6 = getelementptr inbounds i8, i8 addrspace(0)* %dst, i32 %index.1
   store i8 %3, i8 addrspace(0)* %arrayidx6, align 1
   %inc = add i32 %index.1, 1
   br label %while.cond3
@@ -238,10 +238,10 @@ while.cond:                                       ; preds = %while.body, %entry
   br i1 %cmp, label %while.cond3, label %while.body
 
 while.body:                                       ; preds = %while.cond
-  %add.ptr = getelementptr inbounds i8 addrspace(1)* %src, i32 %index.0
+  %add.ptr = getelementptr inbounds i8, i8 addrspace(1)* %src, i32 %index.0
   %0 = bitcast i8 addrspace(1)* %add.ptr to i32 addrspace(1)*
-  %1 = load i32 addrspace(1)* %0, align 4
-  %add.ptr1 = getelementptr inbounds i8 addrspace(3)* %dst, i32 %index.0
+  %1 = load i32, i32 addrspace(1)* %0, align 4
+  %add.ptr1 = getelementptr inbounds i8, i8 addrspace(3)* %dst, i32 %index.0
   %2 = bitcast i8 addrspace(3)* %add.ptr1 to i32 addrspace(3)*
   store i32 %1, i32 addrspace(3)* %2, align 4
   br label %while.cond
@@ -252,9 +252,9 @@ while.cond3:                                      ; preds = %while.cond, %while.
   br i1 %cmp4, label %while.body5, label %while.end7
 
 while.body5:                                      ; preds = %while.cond3
-  %arrayidx = getelementptr inbounds i8 addrspace(1)* %src, i32 %index.1
-  %3 = load i8 addrspace(1)* %arrayidx, align 1
-  %arrayidx6 = getelementptr inbounds i8 addrspace(3)* %dst, i32 %index.1
+  %arrayidx = getelementptr inbounds i8, i8 addrspace(1)* %src, i32 %index.1
+  %3 = load i8, i8 addrspace(1)* %arrayidx, align 1
+  %arrayidx6 = getelementptr inbounds i8, i8 addrspace(3)* %dst, i32 %index.1
   store i8 %3, i8 addrspace(3)* %arrayidx6, align 1
   %inc = add i32 %index.1, 1
   br label %while.cond3
@@ -274,10 +274,10 @@ while.cond:                                       ; preds = %while.body, %entry
   br i1 %cmp, label %while.cond3, label %while.body
 
 while.body:                                       ; preds = %while.cond
-  %add.ptr = getelementptr inbounds i8 addrspace(0)* %src, i32 %index.0
+  %add.ptr = getelementptr inbounds i8, i8 addrspace(0)* %src, i32 %index.0
   %0 = bitcast i8 addrspace(0)* %add.ptr to i32 addrspace(0)*
-  %1 = load i32 addrspace(0)* %0, align 4
-  %add.ptr1 = getelementptr inbounds i8 addrspace(3)* %dst, i32 %index.0
+  %1 = load i32, i32 addrspace(0)* %0, align 4
+  %add.ptr1 = getelementptr inbounds i8, i8 addrspace(3)* %dst, i32 %index.0
   %2 = bitcast i8 addrspace(3)* %add.ptr1 to i32 addrspace(3)*
   store i32 %1, i32 addrspace(3)* %2, align 4
   br label %while.cond
@@ -288,9 +288,9 @@ while.cond3:                                      ; preds = %while.cond, %while.
   br i1 %cmp4, label %while.body5, label %while.end7
 
 while.body5:                                      ; preds = %while.cond3
-  %arrayidx = getelementptr inbounds i8 addrspace(0)* %src, i32 %index.1
-  %3 = load i8 addrspace(0)* %arrayidx, align 1
-  %arrayidx6 = getelementptr inbounds i8 addrspace(3)* %dst, i32 %index.1
+  %arrayidx = getelementptr inbounds i8, i8 addrspace(0)* %src, i32 %index.1
+  %3 = load i8, i8 addrspace(0)* %arrayidx, align 1
+  %arrayidx6 = getelementptr inbounds i8, i8 addrspace(3)* %dst, i32 %index.1
   store i8 %3, i8 addrspace(3)* %arrayidx6, align 1
   %inc = add i32 %index.1, 1
   br label %while.cond3
@@ -310,10 +310,10 @@ while.cond:                                       ; preds = %while.body, %entry
   br i1 %cmp, label %while.cond3, label %while.body
 
 while.body:                                       ; preds = %while.cond
-  %add.ptr = getelementptr inbounds i8 addrspace(3)* %src, i32 %index.0
+  %add.ptr = getelementptr inbounds i8, i8 addrspace(3)* %src, i32 %index.0
   %0 = bitcast i8 addrspace(3)* %add.ptr to i32 addrspace(3)*
-  %1 = load i32 addrspace(3)* %0, align 4
-  %add.ptr1 = getelementptr inbounds i8 addrspace(3)* %dst, i32 %index.0
+  %1 = load i32, i32 addrspace(3)* %0, align 4
+  %add.ptr1 = getelementptr inbounds i8, i8 addrspace(3)* %dst, i32 %index.0
   %2 = bitcast i8 addrspace(3)* %add.ptr1 to i32 addrspace(3)*
   store i32 %1, i32 addrspace(3)* %2, align 4
   br label %while.cond
@@ -324,9 +324,9 @@ while.cond3:                                      ; preds = %while.cond, %while.
   br i1 %cmp4, label %while.body5, label %while.end7
 
 while.body5:                                      ; preds = %while.cond3
-  %arrayidx = getelementptr inbounds i8 addrspace(3)* %src, i32 %index.1
-  %3 = load i8 addrspace(3)* %arrayidx, align 1
-  %arrayidx6 = getelementptr inbounds i8 addrspace(3)* %dst, i32 %index.1
+  %arrayidx = getelementptr inbounds i8, i8 addrspace(3)* %src, i32 %index.1
+  %3 = load i8, i8 addrspace(3)* %arrayidx, align 1
+  %arrayidx6 = getelementptr inbounds i8, i8 addrspace(3)* %dst, i32 %index.1
   store i8 %3, i8 addrspace(3)* %arrayidx6, align 1
   %inc = add i32 %index.1, 1
   br label %while.cond3
@@ -354,7 +354,7 @@ while.body:                                       ; preds = %entry, %while.body
   %0 = ptrtoint i8 addrspace(1)* %src to i32
   %1 = add i32 %0, %index.05
   %2 = inttoptr i32 %1 to i8 addrspace(1)*
-  %3 = load i8 addrspace(1)* %2, align 1
+  %3 = load i8, i8 addrspace(1)* %2, align 1
   %4 = ptrtoint i8 addrspace(1)* %dst to i32
   %5 = add i32 %4, %index.05
   %6 = inttoptr i32 %5 to i8 addrspace(1)*
@@ -377,7 +377,7 @@ while.body:                                       ; preds = %entry, %while.body
   %0 = ptrtoint i8 addrspace(0)* %src to i32
   %1 = add i32 %0, %index.05
   %2 = inttoptr i32 %1 to i8 addrspace(0)*
-  %3 = load i8 addrspace(0)* %2, align 1
+  %3 = load i8, i8 addrspace(0)* %2, align 1
   %4 = ptrtoint i8 addrspace(1)* %dst to i32
   %5 = add i32 %4, %index.05
   %6 = inttoptr i32 %5 to i8 addrspace(1)*
@@ -400,7 +400,7 @@ while.body:                                       ; preds = %entry, %while.body
   %0 = ptrtoint i8 addrspace(3)* %src to i32
   %1 = add i32 %0, %index.05
   %2 = inttoptr i32 %1 to i8 addrspace(3)*
-  %3 = load i8 addrspace(3)* %2, align 1
+  %3 = load i8, i8 addrspace(3)* %2, align 1
   %4 = ptrtoint i8 addrspace(1)* %dst to i32
   %5 = add i32 %4, %index.05
   %6 = inttoptr i32 %5 to i8 addrspace(1)*
@@ -423,7 +423,7 @@ while.body:                                       ; preds = %entry, %while.body
   %0 = ptrtoint i8 addrspace(1)* %src to i32
   %1 = add i32 %0, %index.05
   %2 = inttoptr i32 %1 to i8 addrspace(1)*
-  %3 = load i8 addrspace(1)* %2, align 1
+  %3 = load i8, i8 addrspace(1)* %2, align 1
   %4 = ptrtoint i8 addrspace(0)* %dst to i32
   %5 = add i32 %4, %index.05
   %6 = inttoptr i32 %5 to i8 addrspace(0)*
@@ -446,7 +446,7 @@ while.body:                                       ; preds = %entry, %while.body
   %0 = ptrtoint i8 addrspace(0)* %src to i32
   %1 = add i32 %0, %index.05
   %2 = inttoptr i32 %1 to i8 addrspace(0)*
-  %3 = load i8 addrspace(0)* %2, align 1
+  %3 = load i8, i8 addrspace(0)* %2, align 1
   %4 = ptrtoint i8 addrspace(0)* %dst to i32
   %5 = add i32 %4, %index.05
   %6 = inttoptr i32 %5 to i8 addrspace(0)*
@@ -469,7 +469,7 @@ while.body:                                       ; preds = %entry, %while.body
   %0 = ptrtoint i8 addrspace(3)* %src to i32
   %1 = add i32 %0, %index.05
   %2 = inttoptr i32 %1 to i8 addrspace(3)*
-  %3 = load i8 addrspace(3)* %2, align 1
+  %3 = load i8, i8 addrspace(3)* %2, align 1
   %4 = ptrtoint i8 addrspace(0)* %dst to i32
   %5 = add i32 %4, %index.05
   %6 = inttoptr i32 %5 to i8 addrspace(0)*
@@ -492,7 +492,7 @@ while.body:                                       ; preds = %entry, %while.body
   %0 = ptrtoint i8 addrspace(1)* %src to i32
   %1 = add i32 %0, %index.05
   %2 = inttoptr i32 %1 to i8 addrspace(1)*
-  %3 = load i8 addrspace(1)* %2, align 1
+  %3 = load i8, i8 addrspace(1)* %2, align 1
   %4 = ptrtoint i8 addrspace(3)* %dst to i32
   %5 = add i32 %4, %index.05
   %6 = inttoptr i32 %5 to i8 addrspace(3)*
@@ -515,7 +515,7 @@ while.body:                                       ; preds = %entry, %while.body
   %0 = ptrtoint i8 addrspace(0)* %src to i32
   %1 = add i32 %0, %index.05
   %2 = inttoptr i32 %1 to i8 addrspace(0)*
-  %3 = load i8 addrspace(0)* %2, align 1
+  %3 = load i8, i8 addrspace(0)* %2, align 1
   %4 = ptrtoint i8 addrspace(3)* %dst to i32
   %5 = add i32 %4, %index.05
   %6 = inttoptr i32 %5 to i8 addrspace(3)*
@@ -538,7 +538,7 @@ while.body:                                       ; preds = %entry, %while.body
   %0 = ptrtoint i8 addrspace(3)* %src to i32
   %1 = add i32 %0, %index.05
   %2 = inttoptr i32 %1 to i8 addrspace(3)*
-  %3 = load i8 addrspace(3)* %2, align 1
+  %3 = load i8, i8 addrspace(3)* %2, align 1
   %4 = ptrtoint i8 addrspace(3)* %dst to i32
   %5 = add i32 %4, %index.05
   %6 = inttoptr i32 %5 to i8 addrspace(3)*
@@ -562,10 +562,10 @@ while.cond:                                       ; preds = %while.body, %entry
   br i1 %cmp, label %while.cond3, label %while.body
 
 while.body:                                       ; preds = %while.cond
-  %add.ptr = getelementptr inbounds i8 addrspace(2)* %src, i32 %index.0
+  %add.ptr = getelementptr inbounds i8, i8 addrspace(2)* %src, i32 %index.0
   %0 = bitcast i8 addrspace(2)* %add.ptr to i32 addrspace(2)*
-  %1 = load i32 addrspace(2)* %0, align 4
-  %add.ptr1 = getelementptr inbounds i8 addrspace(1)* %dst, i32 %index.0
+  %1 = load i32, i32 addrspace(2)* %0, align 4
+  %add.ptr1 = getelementptr inbounds i8, i8 addrspace(1)* %dst, i32 %index.0
   %2 = bitcast i8 addrspace(1)* %add.ptr1 to i32 addrspace(1)*
   store i32 %1, i32 addrspace(1)* %2, align 4
   br label %while.cond
@@ -576,9 +576,9 @@ while.cond3:                                      ; preds = %while.cond, %while.
   br i1 %cmp4, label %while.body5, label %while.end7
 
 while.body5:                                      ; preds = %while.cond3
-  %arrayidx = getelementptr inbounds i8 addrspace(2)* %src, i32 %index.1
-  %3 = load i8 addrspace(2)* %arrayidx, align 1
-  %arrayidx6 = getelementptr inbounds i8 addrspace(1)* %dst, i32 %index.1
+  %arrayidx = getelementptr inbounds i8, i8 addrspace(2)* %src, i32 %index.1
+  %3 = load i8, i8 addrspace(2)* %arrayidx, align 1
+  %arrayidx6 = getelementptr inbounds i8, i8 addrspace(1)* %dst, i32 %index.1
   store i8 %3, i8 addrspace(1)* %arrayidx6, align 1
   %inc = add i32 %index.1, 1
   br label %while.cond3
@@ -598,10 +598,10 @@ while.cond:                                       ; preds = %while.body, %entry
   br i1 %cmp, label %while.cond3, label %while.body
 
 while.body:                                       ; preds = %while.cond
-  %add.ptr = getelementptr inbounds i8 addrspace(2)* %src, i32 %index.0
+  %add.ptr = getelementptr inbounds i8, i8 addrspace(2)* %src, i32 %index.0
   %0 = bitcast i8 addrspace(2)* %add.ptr to i32 addrspace(2)*
-  %1 = load i32 addrspace(2)* %0, align 4
-  %add.ptr1 = getelementptr inbounds i8 addrspace(0)* %dst, i32 %index.0
+  %1 = load i32, i32 addrspace(2)* %0, align 4
+  %add.ptr1 = getelementptr inbounds i8, i8 addrspace(0)* %dst, i32 %index.0
   %2 = bitcast i8 addrspace(0)* %add.ptr1 to i32 addrspace(0)*
   store i32 %1, i32 addrspace(0)* %2, align 4
   br label %while.cond
@@ -612,9 +612,9 @@ while.cond3:                                      ; preds = %while.cond, %while.
   br i1 %cmp4, label %while.body5, label %while.end7
 
 while.body5:                                      ; preds = %while.cond3
-  %arrayidx = getelementptr inbounds i8 addrspace(2)* %src, i32 %index.1
-  %3 = load i8 addrspace(2)* %arrayidx, align 1
-  %arrayidx6 = getelementptr inbounds i8 addrspace(0)* %dst, i32 %index.1
+  %arrayidx = getelementptr inbounds i8, i8 addrspace(2)* %src, i32 %index.1
+  %3 = load i8, i8 addrspace(2)* %arrayidx, align 1
+  %arrayidx6 = getelementptr inbounds i8, i8 addrspace(0)* %dst, i32 %index.1
   store i8 %3, i8 addrspace(0)* %arrayidx6, align 1
   %inc = add i32 %index.1, 1
   br label %while.cond3
@@ -634,10 +634,10 @@ while.cond:                                       ; preds = %while.body, %entry
   br i1 %cmp, label %while.cond3, label %while.body
 
 while.body:                                       ; preds = %while.cond
-  %add.ptr = getelementptr inbounds i8 addrspace(2)* %src, i32 %index.0
+  %add.ptr = getelementptr inbounds i8, i8 addrspace(2)* %src, i32 %index.0
   %0 = bitcast i8 addrspace(2)* %add.ptr to i32 addrspace(2)*
-  %1 = load i32 addrspace(2)* %0, align 4
-  %add.ptr1 = getelementptr inbounds i8 addrspace(3)* %dst, i32 %index.0
+  %1 = load i32, i32 addrspace(2)* %0, align 4
+  %add.ptr1 = getelementptr inbounds i8, i8 addrspace(3)* %dst, i32 %index.0
   %2 = bitcast i8 addrspace(3)* %add.ptr1 to i32 addrspace(3)*
   store i32 %1, i32 addrspace(3)* %2, align 4
   br label %while.cond
@@ -648,9 +648,9 @@ while.cond3:                                      ; preds = %while.cond, %while.
   br i1 %cmp4, label %while.body5, label %while.end7
 
 while.body5:                                      ; preds = %while.cond3
-  %arrayidx = getelementptr inbounds i8 addrspace(2)* %src, i32 %index.1
-  %3 = load i8 addrspace(2)* %arrayidx, align 1
-  %arrayidx6 = getelementptr inbounds i8 addrspace(3)* %dst, i32 %index.1
+  %arrayidx = getelementptr inbounds i8, i8 addrspace(2)* %src, i32 %index.1
+  %3 = load i8, i8 addrspace(2)* %arrayidx, align 1
+  %arrayidx6 = getelementptr inbounds i8, i8 addrspace(3)* %dst, i32 %index.1
   store i8 %3, i8 addrspace(3)* %arrayidx6, align 1
   %inc = add i32 %index.1, 1
   br label %while.cond3
@@ -669,7 +669,7 @@ while.body:                                       ; preds = %entry, %while.body
   %0 = ptrtoint i8 addrspace(2)* %src to i32
   %1 = add i32 %0, %index.05
   %2 = inttoptr i32 %1 to i8 addrspace(2)*
-  %3 = load i8 addrspace(2)* %2, align 1
+  %3 = load i8, i8 addrspace(2)* %2, align 1
   %4 = ptrtoint i8 addrspace(0)* %dst to i32
   %5 = add i32 %4, %index.05
   %6 = inttoptr i32 %5 to i8 addrspace(0)*
@@ -692,7 +692,7 @@ while.body:                                       ; preds = %entry, %while.body
   %0 = ptrtoint i8 addrspace(2)* %src to i32
   %1 = add i32 %0, %index.05
   %2 = inttoptr i32 %1 to i8 addrspace(2)*
-  %3 = load i8 addrspace(2)* %2, align 1
+  %3 = load i8, i8 addrspace(2)* %2, align 1
   %4 = ptrtoint i8 addrspace(1)* %dst to i32
   %5 = add i32 %4, %index.05
   %6 = inttoptr i32 %5 to i8 addrspace(1)*
@@ -715,7 +715,7 @@ while.body:                                       ; preds = %entry, %while.body
   %0 = ptrtoint i8 addrspace(2)* %src to i32
   %1 = add i32 %0, %index.05
   %2 = inttoptr i32 %1 to i8 addrspace(2)*
-  %3 = load i8 addrspace(2)* %2, align 1
+  %3 = load i8, i8 addrspace(2)* %2, align 1
   %4 = ptrtoint i8 addrspace(3)* %dst to i32
   %5 = add i32 %4, %index.05
   %6 = inttoptr i32 %5 to i8 addrspace(3)*
diff --git a/backend/src/libocl/src/ocl_memset.ll b/backend/src/libocl/src/ocl_memset.ll
index 665eac4..6506a69 100644
--- a/backend/src/libocl/src/ocl_memset.ll
+++ b/backend/src/libocl/src/ocl_memset.ll
@@ -30,7 +30,7 @@ while.cond:                                       ; preds = %while.body, %entry
   br i1 %cmp, label %while.cond10, label %while.body
 
 while.body:                                       ; preds = %while.cond
-  %add.ptr = getelementptr inbounds i8* %dst, i32 %index.0
+  %add.ptr = getelementptr inbounds i8, i8* %dst, i32 %index.0
   %0 = bitcast i8* %add.ptr to i32*
   store i32 %or7, i32* %0, align 4
   br label %while.cond
@@ -41,7 +41,7 @@ while.cond10:                                     ; preds = %while.cond, %while.
   br i1 %cmp11, label %while.body13, label %while.end14
 
 while.body13:                                     ; preds = %while.cond10
-  %arrayidx = getelementptr inbounds i8* %dst, i32 %index.1
+  %arrayidx = getelementptr inbounds i8, i8* %dst, i32 %index.1
   store i8 %val, i8* %arrayidx, align 1
   %inc = add i32 %index.1, 1
   br label %while.cond10
@@ -68,7 +68,7 @@ while.cond:                                       ; preds = %while.body, %entry
   br i1 %cmp, label %while.cond10, label %while.body
 
 while.body:                                       ; preds = %while.cond
-  %add.ptr = getelementptr inbounds i8 addrspace(1)* %dst, i32 %index.0
+  %add.ptr = getelementptr inbounds i8, i8 addrspace(1)* %dst, i32 %index.0
   %0 = bitcast i8 addrspace(1)* %add.ptr to i32 addrspace(1)*
   store i32 %or7, i32 addrspace(1)* %0, align 4
   br label %while.cond
@@ -79,7 +79,7 @@ while.cond10:                                     ; preds = %while.cond, %while.
   br i1 %cmp11, label %while.body13, label %while.end14
 
 while.body13:                                     ; preds = %while.cond10
-  %arrayidx = getelementptr inbounds i8 addrspace(1)* %dst, i32 %index.1
+  %arrayidx = getelementptr inbounds i8, i8 addrspace(1)* %dst, i32 %index.1
   store i8 %val, i8 addrspace(1)* %arrayidx, align 1
   %inc = add i32 %index.1, 1
   br label %while.cond10
@@ -106,7 +106,7 @@ while.cond:                                       ; preds = %while.body, %entry
   br i1 %cmp, label %while.cond10, label %while.body
 
 while.body:                                       ; preds = %while.cond
-  %add.ptr = getelementptr inbounds i8 addrspace(3)* %dst, i32 %index.0
+  %add.ptr = getelementptr inbounds i8, i8 addrspace(3)* %dst, i32 %index.0
   %0 = bitcast i8 addrspace(3)* %add.ptr to i32 addrspace(3)*
   store i32 %or7, i32 addrspace(3)* %0, align 4
   br label %while.cond
@@ -117,7 +117,7 @@ while.cond10:                                     ; preds = %while.cond, %while.
   br i1 %cmp11, label %while.body13, label %while.end14
 
 while.body13:                                     ; preds = %while.cond10
-  %arrayidx = getelementptr inbounds i8 addrspace(3)* %dst, i32 %index.1
+  %arrayidx = getelementptr inbounds i8, i8 addrspace(3)* %dst, i32 %index.1
   store i8 %val, i8 addrspace(3)* %arrayidx, align 1
   %inc = add i32 %index.1, 1
   br label %while.cond10
diff --git a/backend/src/llvm/llvm_bitcode_link.cpp b/backend/src/llvm/llvm_bitcode_link.cpp
index ebf4386..40040f9 100644
--- a/backend/src/llvm/llvm_bitcode_link.cpp
+++ b/backend/src/llvm/llvm_bitcode_link.cpp
@@ -25,7 +25,11 @@
 #include "llvm/IR/Instructions.h"
 #include "llvm/IR/Module.h"
 #include "llvm/IRReader/IRReader.h"
-#include "llvm/PassManager.h"
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+#include "llvm/IR/LegacyPassManager.h"
+#else
+#include "llvm/PassManager"
+#endif
 #include "llvm/Pass.h"
 #include "llvm/IR/IRBuilder.h"
 #include "llvm/Support/FileSystem.h"
@@ -249,7 +253,11 @@ namespace gbe
       return NULL;
     }
 
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+    llvm::legacy::PassManager passes;
+#else
     llvm::PassManager passes;
+#endif
 
     passes.add(createInternalizePass(kernels));
     passes.add(createGlobalDCEPass());
diff --git a/backend/src/llvm/llvm_gen_backend.cpp b/backend/src/llvm/llvm_gen_backend.cpp
index bf03a13..fec8bd4 100644
--- a/backend/src/llvm/llvm_gen_backend.cpp
+++ b/backend/src/llvm/llvm_gen_backend.cpp
@@ -86,7 +86,12 @@
 #include "llvm/IR/Instructions.h"
 #endif  /* LLVM_VERSION_MINOR <= 2 */
 #include "llvm/Pass.h"
+
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+#include "llvm/IR/LegacyPassManager.h"
+#else
 #include "llvm/PassManager.h"
+#endif
 #if LLVM_VERSION_MINOR <= 2
 #include "llvm/Intrinsics.h"
 #include "llvm/IntrinsicInst.h"
@@ -99,7 +104,11 @@
 #include "llvm/ADT/StringExtras.h"
 #include "llvm/ADT/SmallString.h"
 #include "llvm/ADT/STLExtras.h"
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+#include "llvm/IR/InstIterator.h"
+#else
 #include "llvm/Analysis/ConstantsScanner.h"
+#endif
 #include "llvm/Analysis/LoopInfo.h"
 #include "llvm/Analysis/ValueTracking.h"
 #include "llvm/CodeGen/Passes.h"
@@ -501,14 +510,22 @@ namespace gbe
         TheModule(0),
         btiBase(BTI_RESERVED_NUM)
     {
-      initializeLoopInfoPass(*PassRegistry::getPassRegistry());
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+        initializeLoopInfoWrapperPassPass(*PassRegistry::getPassRegistry());
+#else
+        initializeLoopInfoPass(*PassRegistry::getPassRegistry());
+#endif
       pass = PASS_EMIT_REGISTERS;
     }
 
     virtual const char *getPassName() const { return "Gen Back-End"; }
 
     void getAnalysisUsage(AnalysisUsage &AU) const {
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+      AU.addRequired<LoopInfoWrapperPass>();
+#else
       AU.addRequired<LoopInfo>();
+#endif
       AU.setPreservesAll();
     }
 
@@ -535,7 +552,11 @@ namespace gbe
       if(!bKernel) return false;
 
       analyzePointerOrigin(F);
-      LI = &getAnalysis<LoopInfo>();
+      #if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+        LI = &getAnalysis<LoopInfoWrapperPass>().getLoopInfo();
+      #else
+        LI = &getAnalysis<LoopInfo>();
+      #endif
       emitFunction(F);
       phiMap.clear();
       globalPointer.clear();
diff --git a/backend/src/llvm/llvm_gen_backend.hpp b/backend/src/llvm/llvm_gen_backend.hpp
index b21e804..d78fb12 100644
--- a/backend/src/llvm/llvm_gen_backend.hpp
+++ b/backend/src/llvm/llvm_gen_backend.hpp
@@ -131,7 +131,11 @@ namespace gbe
   llvm::BasicBlockPass *createRemoveGEPPass(const ir::Unit &unit);
 
   /*! Merge load/store if possible */
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+  llvm::BasicBlockPass *createLoadStoreOptimizationPass(const llvm::DataLayout* DL);
+#else
   llvm::BasicBlockPass *createLoadStoreOptimizationPass();
+#endif
 
   /*! Scalarize all vector op instructions */
   llvm::FunctionPass* createScalarizePass();
diff --git a/backend/src/llvm/llvm_loadstore_optimization.cpp b/backend/src/llvm/llvm_loadstore_optimization.cpp
index c6349fa..a258f55 100644
--- a/backend/src/llvm/llvm_loadstore_optimization.cpp
+++ b/backend/src/llvm/llvm_loadstore_optimization.cpp
@@ -24,7 +24,11 @@
 
 #include "llvm/IR/Instructions.h"
 #include "llvm/Pass.h"
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+#include "llvm/IR/LegacyPassManager.h"
+#else
 #include "llvm/PassManager.h"
+#endif
 
 #include "llvm/Config/llvm-config.h"
 #include "llvm/ADT/DenseMap.h"
@@ -62,7 +66,11 @@ namespace gbe {
     static char ID;
     ScalarEvolution *SE;
     const DataLayout *TD;
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+    GenLoadStoreOptimization(const DataLayout* DL) : BasicBlockPass(ID), TD(DL) {}
+#else
     GenLoadStoreOptimization() : BasicBlockPass(ID) {}
+#endif
 
     void getAnalysisUsage(AnalysisUsage &AU) const {
       AU.addRequired<ScalarEvolution>();
@@ -72,7 +80,8 @@ namespace gbe {
 
     virtual bool runOnBasicBlock(BasicBlock &BB) {
       SE = &getAnalysis<ScalarEvolution>();
-      #if LLVM_VERSION_MINOR >= 5
+      #if LLVM_VERSION_MINOR == 7
+      #elif LLVM_VERSION_MINOR >= 5
         DataLayoutPass *DLP = getAnalysisIfAvailable<DataLayoutPass>();
         TD = DLP ? &DLP->getDataLayout() : nullptr;
       #else
@@ -278,8 +287,13 @@ namespace gbe {
     return changed;
   }
 
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+  BasicBlockPass *createLoadStoreOptimizationPass(const DataLayout* DL) {
+    return new GenLoadStoreOptimization(DL);
+#else
   BasicBlockPass *createLoadStoreOptimizationPass() {
     return new GenLoadStoreOptimization();
+#endif
   }
 };
 
diff --git a/backend/src/llvm/llvm_passes.cpp b/backend/src/llvm/llvm_passes.cpp
index 223f61b..5b3d22f 100644
--- a/backend/src/llvm/llvm_passes.cpp
+++ b/backend/src/llvm/llvm_passes.cpp
@@ -45,7 +45,11 @@
 #include "llvm/IR/Instructions.h"
 #endif  /* LLVM_VERSION_MINOR <= 2 */
 #include "llvm/Pass.h"
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+#include "llvm/IR/LegacyPassManager.h"
+#else
 #include "llvm/PassManager.h"
+#endif
 #if LLVM_VERSION_MINOR <= 2
 #include "llvm/Intrinsics.h"
 #include "llvm/IntrinsicInst.h"
@@ -58,7 +62,11 @@
 #include "llvm/ADT/StringExtras.h"
 #include "llvm/ADT/SmallString.h"
 #include "llvm/ADT/STLExtras.h"
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+#include "llvm/IR/InstIterator.h"
+#else
 #include "llvm/Analysis/ConstantsScanner.h"
+#endif
 #include "llvm/Analysis/LoopInfo.h"
 #include "llvm/Analysis/ValueTracking.h"
 #include "llvm/CodeGen/Passes.h"
diff --git a/backend/src/llvm/llvm_sampler_fix.cpp b/backend/src/llvm/llvm_sampler_fix.cpp
index 8c76324..a2a57f4 100644
--- a/backend/src/llvm/llvm_sampler_fix.cpp
+++ b/backend/src/llvm/llvm_sampler_fix.cpp
@@ -22,7 +22,11 @@
  */
 #include "llvm/IR/Instructions.h"
 #include "llvm/Pass.h"
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+#include "llvm/IR/LegacyPassManager.h"
+#else
 #include "llvm/PassManager.h"
+#endif
 
 #include "llvm/Config/llvm-config.h"
 #include "llvm/ADT/DenseMap.h"
@@ -40,7 +44,11 @@
 #include "llvm/Support/CFG.h"
 #endif
 
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+#include "llvm/IR/InstIterator.h"
+#else
 #include "llvm/Analysis/ConstantsScanner.h"
+#endif
 
 #include "llvm_gen_backend.hpp"
 #include "ocl_common_defines.h"
diff --git a/backend/src/llvm/llvm_scalarize.cpp b/backend/src/llvm/llvm_scalarize.cpp
index 15309de..846282d 100644
--- a/backend/src/llvm/llvm_scalarize.cpp
+++ b/backend/src/llvm/llvm_scalarize.cpp
@@ -128,7 +128,11 @@ namespace gbe {
 
     Scalarize() : FunctionPass(ID)
     {
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+      initializeLoopInfoWrapperPassPass(*PassRegistry::getPassRegistry());
+#else
       initializeLoopInfoPass(*PassRegistry::getPassRegistry());
+#endif
 #if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 5
       initializeDominatorTreeWrapperPassPass(*PassRegistry::getPassRegistry());
 #else
diff --git a/backend/src/llvm/llvm_to_gen.cpp b/backend/src/llvm/llvm_to_gen.cpp
index b1dc686..74650e9 100644
--- a/backend/src/llvm/llvm_to_gen.cpp
+++ b/backend/src/llvm/llvm_to_gen.cpp
@@ -32,11 +32,19 @@
 #include "llvm/IR/Module.h"
 #include "llvm/IR/DataLayout.h"
 #endif  /* LLVM_VERSION_MINOR <= 2 */
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+#include "llvm/IR/LegacyPassManager.h"
+#else
 #include "llvm/PassManager.h"
+#endif
 #include "llvm/Pass.h"
 #include "llvm/Analysis/Passes.h"
 #include "llvm/Transforms/IPO.h"
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+#include "llvm/Analysis/TargetLibraryInfo.h"
+#else
 #include "llvm/Target/TargetLibraryInfo.h"
+#endif
 #include "llvm/ADT/Triple.h"
 #if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR <= 2
 #include "llvm/Support/IRReader.h"
@@ -77,12 +85,21 @@ namespace gbe
   BVAR(OCL_OUTPUT_CFG_GEN_IR, false);
   using namespace llvm;
 
-  void runFuntionPass(Module &mod, TargetLibraryInfo *libraryInfo, const DataLayout &DL)
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+  void runFunctionPass(Module &mod, TargetLibraryInfoImpl *libraryInfo, const DataLayout &DL)
+#else
+  void runFunctionPass(Module &mod, TargetLibraryInfo *libraryInfo, const DataLayout &DL)
+#endif
   {
-    FunctionPassManager FPM(&mod);
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+      llvm::legacy::FunctionPassManager FPM(&mod);
+#else
+      FunctionPassManager FPM(&mod);
+#endif
 
-#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 6
-    FPM.add(new DataLayoutPass());
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+#elif LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR == 6
+    FPM.add(new DataLayout(DL));
 #elif LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR == 5
     FPM.add(new DataLayoutPass(DL));
 #else
@@ -94,7 +111,12 @@ namespace gbe
 #else
     FPM.add(createVerifierPass());
 #endif
+
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+    FPM.add(new TargetLibraryInfoWrapperPass(*libraryInfo));
+#else
     FPM.add(new TargetLibraryInfo(*libraryInfo));
+#endif
     FPM.add(createTypeBasedAliasAnalysisPass());
     FPM.add(createBasicAliasAnalysisPass());
     FPM.add(createCFGSimplificationPass());
@@ -110,18 +132,27 @@ namespace gbe
     FPM.doFinalization();
   }
 
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+  void runModulePass(Module &mod, TargetLibraryInfoImpl *libraryInfo, const DataLayout &DL, int optLevel, bool strictMath)
+#else
   void runModulePass(Module &mod, TargetLibraryInfo *libraryInfo, const DataLayout &DL, int optLevel, bool strictMath)
+#endif
   {
-    llvm::PassManager MPM;
+    llvm::legacy::PassManager MPM;
 
-#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 6
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+#elif LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR == 6
     MPM.add(new DataLayoutPass());
 #elif LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR == 5
     MPM.add(new DataLayoutPass(DL));
 #else
     MPM.add(new DataLayout(DL));
 #endif
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+    MPM.add(new TargetLibraryInfoWrapperPass(*libraryInfo));
+#else
     MPM.add(new TargetLibraryInfo(*libraryInfo));
+#endif
     MPM.add(createTypeBasedAliasAnalysisPass());
     MPM.add(createBasicAliasAnalysisPass());
     MPM.add(createIntrinsicLoweringPass());
@@ -198,7 +229,15 @@ namespace gbe
   }
 
 
-#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 5
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+#define OUTPUT_BITCODE(STAGE, MOD)  do {         \
+   llvm::legacy::PassManager passes__;                   \
+   if (OCL_OUTPUT_LLVM_##STAGE) {                \
+     passes__.add(createPrintModulePass(*o));    \
+     passes__.run(MOD);                          \
+   }                                             \
+ }while(0)
+#elif LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 5
 #define OUTPUT_BITCODE(STAGE, MOD)  do {         \
    llvm::PassManager passes__;                   \
    if (OCL_OUTPUT_LLVM_##STAGE) {                \
@@ -259,15 +298,26 @@ namespace gbe
     DataLayout DL(&mod);
 
     Triple TargetTriple(mod.getTargetTriple());
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+    TargetLibraryInfoImpl *libraryInfo = new TargetLibraryInfoImpl(TargetTriple);
+#else
     TargetLibraryInfo *libraryInfo = new TargetLibraryInfo(TargetTriple);
+#endif
     libraryInfo->disableAllFunctions();
 
     OUTPUT_BITCODE(AFTER_LINK, mod);
 
-    runFuntionPass(mod, libraryInfo, DL);
+    runFunctionPass(mod, libraryInfo, DL);
     runModulePass(mod, libraryInfo, DL, optLevel, strictMath);
+
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR == 7
+    llvm::legacy::PassManager passes;
+#else
     llvm::PassManager passes;
-#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 6
+#endif
+
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR == 7 
+#elif LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR == 6
     passes.add(new DataLayoutPass());
 #elif LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR == 5
     passes.add(new DataLayoutPass(DL));
@@ -277,8 +327,12 @@ namespace gbe
     // Print the code before further optimizations
     passes.add(createIntrinsicLoweringPass());
     passes.add(createFunctionInliningPass(200000));
-    passes.add(createScalarReplAggregatesPass(64, true, -1, -1, 64));
+    passes.add(createScalarReplAggregatesPass(64, true, -1, -1, 64)); 
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR == 7
+    passes.add(createLoadStoreOptimizationPass(&DL));
+#else
     passes.add(createLoadStoreOptimizationPass());
+#endif
     passes.add(createConstantPropagationPass());
     passes.add(createPromoteMemoryToRegisterPass());
     if(optLevel > 0)
diff --git a/backend/src/llvm/llvm_unroll.cpp b/backend/src/llvm/llvm_unroll.cpp
index 5d3fad8..685c297 100644
--- a/backend/src/llvm/llvm_unroll.cpp
+++ b/backend/src/llvm/llvm_unroll.cpp
@@ -40,7 +40,11 @@
 #include "llvm/IR/IRBuilder.h"
 #endif /* LLVM_VERSION_MINOR <= 1 */
 #include "llvm/Support/raw_ostream.h"
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+#include "llvm/IR/LegacyPassManager.h"
+#else
 #include "llvm/PassManager.h"
+#endif
 #include "llvm/Transforms/Scalar.h"
 #include "llvm/Analysis/ScalarEvolution.h"
 #include "llvm/Analysis/LoopPass.h"
@@ -61,8 +65,13 @@ namespace gbe {
        LoopPass(ID) {}
 
       void getAnalysisUsage(AnalysisUsage &AU) const {
+#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
+        AU.addRequired<LoopInfoWrapperPass>();
+        AU.addPreserved<LoopInfoWrapperPass>();
+#else
         AU.addRequired<LoopInfo>();
         AU.addPreserved<LoopInfo>();
+#endif
         AU.addRequiredID(LoopSimplifyID);
         AU.addPreservedID(LoopSimplifyID);
         AU.addRequiredID(LCSSAID);