Add two patches from upstream to support llvm37. And switch to using llvm37

so beignet is now using the same llvm version as the Mesa ports.

Obtained from:	beignet upstream

4 files changed, 2204 insertions, 3 deletions

diff --git a/lang/beignet/Makefile b/lang/beignet/Makefile
index 71c047487213..9360bec693bd 100644
--- a/lang/beignet/Makefile
+++ b/lang/beignet/Makefile
@@ -3,6 +3,7 @@
 
 PORTNAME=	beignet
 PORTVERSION=	1.1.1
+PORTREVISION=	1
 CATEGORIES=	lang
 MASTER_SITES=	https://01.org/sites/default/files/
 DISTVERSIONSUFFIX=	-source
@@ -10,11 +11,12 @@ DISTVERSIONSUFFIX=	-source
 MAINTAINER=	x11@FreeBSD.org
 COMMENT=	OpenCL library for Intel GPUs
 
-BUILD_DEPENDS=	clang${LLVMVER}:lang/clang${LLVMVER} \
+BUILD_DEPENDS=	clang${LLVMVER}:devel/llvm${LLVMVER} \
 		opencl>=0:devel/opencl
 LIB_DEPENDS=	libOpenCL.so:devel/ocl-icd \
 		libdrm.so:graphics/libdrm
-RUN_DEPENDS=	opencl>=0:devel/opencl
+RUN_DEPENDS=	opencl>=0:devel/opencl \
+		clang${LLVMVER}:devel/llvm${LLVMVER}
 
 WRKSRC=		${WRKDIR}/Beignet-${PORTVERSION}-Source
 
@@ -22,7 +24,7 @@ USES=		cmake gmake pkgconfig shebangfix
 USE_XORG=	sm ice x11 xext xfixes
 USE_GL=		gl
 SHEBANG_FILES=	src/git_sha1.sh backend/kernels/compile.sh
-LLVMVER=	36
+LLVMVER=	37
 
 CMAKE_ARGS+=	-DLLVM_CONFIG_EXECUTABLE=${LOCALBASE}/bin/llvm-config${LLVMVER}
 
@@ -35,6 +37,9 @@ BROKEN_FreeBSD_9=	Beignet is only supported on FreeBSD 10.1 and newer
 #OPTIONS_DEFINE=	TESTS
 #TESTS_DESC=	Build and run the test suite
 
+EXTRA_PATCHES=	${PATCHDIR}/llvm37-68b5180:-p1 \
+		${PATCHDIR}/llvm37-27522f9:-p1		
+
 .include <bsd.port.options.mk>
 
 .if ${OPSYS} == FreeBSD && ${OSVERSION} >= 1000000 && ${OSVERSION} < 1002000
diff --git a/lang/beignet/files/llvm37-27522f9 b/lang/beignet/files/llvm37-27522f9
new file mode 100644
index 000000000000..505c51e4be02
--- /dev/null
+++ b/lang/beignet/files/llvm37-27522f9
@@ -0,0 +1,1185 @@
+From 27522f9c83303078be7d927a23f2a43c587efc9a Mon Sep 17 00:00:00 2001
+From: Yang Rong <rong.r.yang@intel.com>
+Date: Wed, 16 Sep 2015 16:49:35 +0800
+Subject: GBE: use opencl c to implement llvm.memset and llvm.memcpy.
+
+llvm 3.7 change to llvm IR, need two copies if still use the llvm IR
+to implement llvm.memset and llvm.memcpy. And opencl c is more clearly.
+
+Signed-off-by: Yang Rong <rong.r.yang@intel.com>
+Reviewed-by: Ruiling Song <ruiling.song@intel.com>
+Reviewed-by: Igor Gnatenko <i.gnatenko.brain@gmail.com>
+
+diff --git a/backend/src/libocl/CMakeLists.txt b/backend/src/libocl/CMakeLists.txt
+index 0cd1eef..0fffd9b 100644
+--- a/backend/src/libocl/CMakeLists.txt
++++ b/backend/src/libocl/CMakeLists.txt
+@@ -52,7 +52,8 @@ FOREACH(M ${OCL_COPY_HEADERS})
+     COPY_THE_HEADER(${M})
+ ENDFOREACH(M) 
+ 
+-SET (OCL_COPY_MODULES ocl_workitem ocl_atom ocl_async ocl_sync ocl_misc ocl_vload ocl_geometric ocl_image)
++SET (OCL_COPY_MODULES ocl_workitem ocl_atom ocl_async ocl_sync ocl_memcpy
++                      ocl_memset ocl_misc ocl_vload ocl_geometric ocl_image)
+ FOREACH(M ${OCL_COPY_MODULES})
+     COPY_THE_HEADER(${M})
+     COPY_THE_SOURCE(${M})
+@@ -181,7 +182,7 @@ MACRO(ADD_LL_TO_BC_TARGET M)
+ 	)
+ ENDMACRO(ADD_LL_TO_BC_TARGET)
+ 
+-SET (OCL_LL_MODULES ocl_barrier ocl_memcpy ocl_memset ocl_clz)
++SET (OCL_LL_MODULES ocl_barrier ocl_clz)
+ FOREACH(f ${OCL_LL_MODULES})
+     COPY_THE_LL(${f})
+     ADD_LL_TO_BC_TARGET(${f})
+diff --git a/backend/src/libocl/include/ocl.h b/backend/src/libocl/include/ocl.h
+index a4af4aa..7897567 100644
+--- a/backend/src/libocl/include/ocl.h
++++ b/backend/src/libocl/include/ocl.h
+@@ -30,6 +30,7 @@
+ #include "ocl_image.h"
+ #include "ocl_integer.h"
+ #include "ocl_math.h"
++#include "ocl_memcpy.h"
+ #include "ocl_misc.h"
+ #include "ocl_printf.h"
+ #include "ocl_relational.h"
+diff --git a/backend/src/libocl/include/ocl_memcpy.h b/backend/src/libocl/include/ocl_memcpy.h
+new file mode 100644
+index 0000000..2672298
+--- /dev/null
++++ b/backend/src/libocl/include/ocl_memcpy.h
+@@ -0,0 +1,51 @@
++/*
++ * Copyright © 2012 - 2014 Intel Corporation
++ *
++ * This library is free software; you can redistribute it and/or
++ * modify it under the terms of the GNU Lesser General Public
++ * License as published by the Free Software Foundation; either
++ * version 2.1 of the License, or (at your option) any later version.
++ *
++ * This library is distributed in the hope that it will be useful,
++ * but WITHOUT ANY WARRANTY; without even the implied warranty of
++ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
++ * Lesser General Public License for more details.
++ *
++ * You should have received a copy of the GNU Lesser General Public
++ * License along with this library. If not, see <http://www.gnu.org/licenses/>.
++ *
++ */
++#ifndef __OCL_MEMCPY_H__
++#define __OCL_MEMCPY_H__
++#include "ocl_types.h"
++
++/////////////////////////////////////////////////////////////////////////////
++// memcopy functions
++/////////////////////////////////////////////////////////////////////////////
++void __gen_memcpy_gg_align(__global uchar* dst, __global uchar* src, size_t size);
++void __gen_memcpy_gp_align(__global uchar* dst, __private uchar* src, size_t size);
++void __gen_memcpy_gl_align(__global uchar* dst, __local uchar* src, size_t size);
++void __gen_memcpy_gc_align(__global uchar* dst, __constant uchar* src, size_t size);
++void __gen_memcpy_pg_align(__private uchar* dst, __global uchar* src, size_t size);
++void __gen_memcpy_pp_align(__private uchar* dst, __private uchar* src, size_t size);
++void __gen_memcpy_pl_align(__private uchar* dst, __local uchar* src, size_t size);
++void __gen_memcpy_pc_align(__private uchar* dst, __constant uchar* src, size_t size);
++void __gen_memcpy_lg_align(__local uchar* dst, __global uchar* src, size_t size);
++void __gen_memcpy_lp_align(__local uchar* dst, __private uchar* src, size_t size);
++void __gen_memcpy_ll_align(__local uchar* dst, __local uchar* src, size_t size);
++void __gen_memcpy_lc_align(__local uchar* dst, __constant uchar* src, size_t size);
++
++void __gen_memcpy_gg(__global uchar* dst, __global uchar* src, size_t size);
++void __gen_memcpy_gp(__global uchar* dst, __private uchar* src, size_t size);
++void __gen_memcpy_gl(__global uchar* dst, __local uchar* src, size_t size);
++void __gen_memcpy_gc(__global uchar* dst, __constant uchar* src, size_t size);
++void __gen_memcpy_pg(__private uchar* dst, __global uchar* src, size_t size);
++void __gen_memcpy_pp(__private uchar* dst, __private uchar* src, size_t size);
++void __gen_memcpy_pl(__private uchar* dst, __local uchar* src, size_t size);
++void __gen_memcpy_pc(__private uchar* dst, __constant uchar* src, size_t size);
++void __gen_memcpy_lg(__local uchar* dst, __global uchar* src, size_t size);
++void __gen_memcpy_lp(__local uchar* dst, __private uchar* src, size_t size);
++void __gen_memcpy_ll(__local uchar* dst, __local uchar* src, size_t size);
++void __gen_memcpy_lc(__local uchar* dst, __constant uchar* src, size_t size);
++
++#endif  /* __OCL_MEMCPY_H__ */
+diff --git a/backend/src/libocl/include/ocl_memset.h b/backend/src/libocl/include/ocl_memset.h
+new file mode 100644
+index 0000000..2d444ad
+--- /dev/null
++++ b/backend/src/libocl/include/ocl_memset.h
+@@ -0,0 +1,33 @@
++/*
++ * Copyright © 2012 - 2014 Intel Corporation
++ *
++ * This library is free software; you can redistribute it and/or
++ * modify it under the terms of the GNU Lesser General Public
++ * License as published by the Free Software Foundation; either
++ * version 2.1 of the License, or (at your option) any later version.
++ *
++ * This library is distributed in the hope that it will be useful,
++ * but WITHOUT ANY WARRANTY; without even the implied warranty of
++ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
++ * Lesser General Public License for more details.
++ *
++ * You should have received a copy of the GNU Lesser General Public
++ * License along with this library. If not, see <http://www.gnu.org/licenses/>.
++ *
++ */
++#ifndef __OCL_MEMSET_H__
++#define __OCL_MEMSET_H__
++#include "ocl_types.h"
++
++/////////////////////////////////////////////////////////////////////////////
++// memcopy functions
++/////////////////////////////////////////////////////////////////////////////
++void __gen_memset_g_align(__global uchar* dst, uchar val, size_t size);
++void __gen_memset_p_align(__private uchar* dst, uchar val, size_t size);
++void __gen_memset_l_align(__local uchar* dst, uchar val, size_t size);
++
++void __gen_memset_g(__global uchar* dst, uchar val, size_t size);
++void __gen_memset_p(__private uchar* dst, uchar val, size_t size);
++void __gen_memset_l(__local uchar* dst, uchar val, size_t size);
++
++#endif  /* __OCL_MEMSET_H__ */
+diff --git a/backend/src/libocl/src/ocl_memcpy.cl b/backend/src/libocl/src/ocl_memcpy.cl
+new file mode 100644
+index 0000000..85f490f
+--- /dev/null
++++ b/backend/src/libocl/src/ocl_memcpy.cl
+@@ -0,0 +1,49 @@
++/*
++ * Copyright © 2012 - 2014 Intel Corporation
++ *
++ * This library is free software; you can redistribute it and/or
++ * modify it under the terms of the GNU Lesser General Public
++ * License as published by the Free Software Foundation; either
++ * version 2.1 of the License, or (at your option) any later version.
++ *
++ * This library is distributed in the hope that it will be useful,
++ * but WITHOUT ANY WARRANTY; without even the implied warranty of
++ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
++ * Lesser General Public License for more details.
++ *
++ * You should have received a copy of the GNU Lesser General Public
++ * License along with this library. If not, see <http://www.gnu.org/licenses/>.
++ *
++ */
++#include "ocl_memcpy.h"
++
++#define DECL_TWO_SPACE_MEMCOPY_FN(NAME, DST_SPACE, SRC_SPACE) \
++void __gen_memcpy_ ##NAME## _align (DST_SPACE uchar* dst, SRC_SPACE uchar* src, size_t size) { \
++  size_t index = 0; \
++  while((index + 4) <= size) { \
++    *((DST_SPACE uint *)(dst + index)) = *((SRC_SPACE uint *)(src + index)); \
++    index += 4; \
++  } \
++  while(index < size) { \
++    dst[index] = src[index]; \
++    index++; \
++  } \
++} \
++void __gen_memcpy_ ##NAME (DST_SPACE uchar* dst, SRC_SPACE uchar* src, size_t size) { \
++  size_t index = 0; \
++  while(index < size) { \
++    dst[index] = src[index]; \
++    index++; \
++  } \
++}
++
++#define DECL_ONE_SPACE_MEMCOPY_FN(NAME, DST_SPACE) \
++  DECL_TWO_SPACE_MEMCOPY_FN( NAME## g, DST_SPACE, __global) \
++  DECL_TWO_SPACE_MEMCOPY_FN( NAME## l, DST_SPACE, __local) \
++  DECL_TWO_SPACE_MEMCOPY_FN( NAME## p, DST_SPACE, __private) \
++  DECL_TWO_SPACE_MEMCOPY_FN( NAME## c, DST_SPACE, __constant)
++
++DECL_ONE_SPACE_MEMCOPY_FN(g, __global)
++DECL_ONE_SPACE_MEMCOPY_FN(l, __local)
++DECL_ONE_SPACE_MEMCOPY_FN(p, __private)
++
+diff --git a/backend/src/libocl/src/ocl_memcpy.ll b/backend/src/libocl/src/ocl_memcpy.ll
+deleted file mode 100644
+index b3fadb2..0000000
+--- a/backend/src/libocl/src/ocl_memcpy.ll
++++ /dev/null
+@@ -1,729 +0,0 @@
+-;The memcpy's source code.
+-; INLINE_OVERLOADABLE void __gen_memcpy_align(uchar* dst, uchar* src, size_t size) {
+-;   size_t index = 0;
+-;   while((index + 4) <= size) {
+-;     *((uint *)(dst + index)) = *((uint *)(src + index));
+-;     index += 4;
+-;   }
+-;   while(index < size) {
+-;     dst[index] = src[index];
+-;     index++;
+-;   }
+-; }
+-
+-define void @__gen_memcpy_gg_align(i8 addrspace(1)* %dst, i8 addrspace(1)* %src, i32 %size) nounwind alwaysinline {
+-entry:
+-  br label %while.cond
+-
+-while.cond:                                       ; preds = %while.body, %entry
+-  %index.0 = phi i32 [ 0, %entry ], [ %add, %while.body ]
+-  %add = add i32 %index.0, 4
+-  %cmp = icmp ugt i32 %add, %size
+-  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
+-  %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
+-  %2 = bitcast i8 addrspace(1)* %add.ptr1 to i32 addrspace(1)*
+-  store i32 %1, i32 addrspace(1)* %2, align 4
+-  br label %while.cond
+-
+-while.cond3:                                      ; preds = %while.cond, %while.body5
+-  %index.1 = phi i32 [ %index.0, %while.cond ], [ %inc, %while.body5 ]
+-  %cmp4 = icmp ult i32 %index.1, %size
+-  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
+-  store i8 %3, i8 addrspace(1)* %arrayidx6, align 1
+-  %inc = add i32 %index.1, 1
+-  br label %while.cond3
+-
+-while.end7:                                       ; preds = %while.cond3
+-  ret void
+-}
+-
+-define void @__gen_memcpy_gp_align(i8 addrspace(1)* %dst, i8 addrspace(0)* %src, i32 %size) nounwind alwaysinline {
+-entry:
+-  br label %while.cond
+-
+-while.cond:                                       ; preds = %while.body, %entry
+-  %index.0 = phi i32 [ 0, %entry ], [ %add, %while.body ]
+-  %add = add i32 %index.0, 4
+-  %cmp = icmp ugt i32 %add, %size
+-  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
+-  %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
+-  %2 = bitcast i8 addrspace(1)* %add.ptr1 to i32 addrspace(1)*
+-  store i32 %1, i32 addrspace(1)* %2, align 4
+-  br label %while.cond
+-
+-while.cond3:                                      ; preds = %while.cond, %while.body5
+-  %index.1 = phi i32 [ %index.0, %while.cond ], [ %inc, %while.body5 ]
+-  %cmp4 = icmp ult i32 %index.1, %size
+-  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
+-  store i8 %3, i8 addrspace(1)* %arrayidx6, align 1
+-  %inc = add i32 %index.1, 1
+-  br label %while.cond3
+-
+-while.end7:                                       ; preds = %while.cond3
+-  ret void
+-}
+-
+-define void @__gen_memcpy_gl_align(i8 addrspace(1)* %dst, i8 addrspace(3)* %src, i32 %size) nounwind alwaysinline {
+-entry:
+-  br label %while.cond
+-
+-while.cond:                                       ; preds = %while.body, %entry
+-  %index.0 = phi i32 [ 0, %entry ], [ %add, %while.body ]
+-  %add = add i32 %index.0, 4
+-  %cmp = icmp ugt i32 %add, %size
+-  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
+-  %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
+-  %2 = bitcast i8 addrspace(1)* %add.ptr1 to i32 addrspace(1)*
+-  store i32 %1, i32 addrspace(1)* %2, align 4
+-  br label %while.cond
+-
+-while.cond3:                                      ; preds = %while.cond, %while.body5
+-  %index.1 = phi i32 [ %index.0, %while.cond ], [ %inc, %while.body5 ]
+-  %cmp4 = icmp ult i32 %index.1, %size
+-  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
+-  store i8 %3, i8 addrspace(1)* %arrayidx6, align 1
+-  %inc = add i32 %index.1, 1
+-  br label %while.cond3
+-
+-while.end7:                                       ; preds = %while.cond3
+-  ret void
+-}
+-
+-define void @__gen_memcpy_pg_align(i8 addrspace(0)* %dst, i8 addrspace(1)* %src, i32 %size) nounwind alwaysinline {
+-entry:
+-  br label %while.cond
+-
+-while.cond:                                       ; preds = %while.body, %entry
+-  %index.0 = phi i32 [ 0, %entry ], [ %add, %while.body ]
+-  %add = add i32 %index.0, 4
+-  %cmp = icmp ugt i32 %add, %size
+-  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
+-  %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
+-  %2 = bitcast i8 addrspace(0)* %add.ptr1 to i32 addrspace(0)*
+-  store i32 %1, i32 addrspace(0)* %2, align 4
+-  br label %while.cond
+-
+-while.cond3:                                      ; preds = %while.cond, %while.body5
+-  %index.1 = phi i32 [ %index.0, %while.cond ], [ %inc, %while.body5 ]
+-  %cmp4 = icmp ult i32 %index.1, %size
+-  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
+-  store i8 %3, i8 addrspace(0)* %arrayidx6, align 1
+-  %inc = add i32 %index.1, 1
+-  br label %while.cond3
+-
+-while.end7:                                       ; preds = %while.cond3
+-  ret void
+-}
+-
+-define void @__gen_memcpy_pp_align(i8 addrspace(0)* %dst, i8 addrspace(0)* %src, i32 %size) nounwind alwaysinline {
+-entry:
+-  br label %while.cond
+-
+-while.cond:                                       ; preds = %while.body, %entry
+-  %index.0 = phi i32 [ 0, %entry ], [ %add, %while.body ]
+-  %add = add i32 %index.0, 4
+-  %cmp = icmp ugt i32 %add, %size
+-  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
+-  %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
+-  %2 = bitcast i8 addrspace(0)* %add.ptr1 to i32 addrspace(0)*
+-  store i32 %1, i32 addrspace(0)* %2, align 4
+-  br label %while.cond
+-
+-while.cond3:                                      ; preds = %while.cond, %while.body5
+-  %index.1 = phi i32 [ %index.0, %while.cond ], [ %inc, %while.body5 ]
+-  %cmp4 = icmp ult i32 %index.1, %size
+-  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
+-  store i8 %3, i8 addrspace(0)* %arrayidx6, align 1
+-  %inc = add i32 %index.1, 1
+-  br label %while.cond3
+-
+-while.end7:                                       ; preds = %while.cond3
+-  ret void
+-}
+-
+-define void @__gen_memcpy_pl_align(i8 addrspace(0)* %dst, i8 addrspace(3)* %src, i32 %size) nounwind alwaysinline {
+-entry:
+-  br label %while.cond
+-
+-while.cond:                                       ; preds = %while.body, %entry
+-  %index.0 = phi i32 [ 0, %entry ], [ %add, %while.body ]
+-  %add = add i32 %index.0, 4
+-  %cmp = icmp ugt i32 %add, %size
+-  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
+-  %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
+-  %2 = bitcast i8 addrspace(0)* %add.ptr1 to i32 addrspace(0)*
+-  store i32 %1, i32 addrspace(0)* %2, align 4
+-  br label %while.cond
+-
+-while.cond3:                                      ; preds = %while.cond, %while.body5
+-  %index.1 = phi i32 [ %index.0, %while.cond ], [ %inc, %while.body5 ]
+-  %cmp4 = icmp ult i32 %index.1, %size
+-  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
+-  store i8 %3, i8 addrspace(0)* %arrayidx6, align 1
+-  %inc = add i32 %index.1, 1
+-  br label %while.cond3
+-
+-while.end7:                                       ; preds = %while.cond3
+-  ret void
+-}
+-
+-define void @__gen_memcpy_lg_align(i8 addrspace(3)* %dst, i8 addrspace(1)* %src, i32 %size) nounwind alwaysinline {
+-entry:
+-  br label %while.cond
+-
+-while.cond:                                       ; preds = %while.body, %entry
+-  %index.0 = phi i32 [ 0, %entry ], [ %add, %while.body ]
+-  %add = add i32 %index.0, 4
+-  %cmp = icmp ugt i32 %add, %size
+-  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
+-  %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
+-  %2 = bitcast i8 addrspace(3)* %add.ptr1 to i32 addrspace(3)*
+-  store i32 %1, i32 addrspace(3)* %2, align 4
+-  br label %while.cond
+-
+-while.cond3:                                      ; preds = %while.cond, %while.body5
+-  %index.1 = phi i32 [ %index.0, %while.cond ], [ %inc, %while.body5 ]
+-  %cmp4 = icmp ult i32 %index.1, %size
+-  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
+-  store i8 %3, i8 addrspace(3)* %arrayidx6, align 1
+-  %inc = add i32 %index.1, 1
+-  br label %while.cond3
+-
+-while.end7:                                       ; preds = %while.cond3
+-  ret void
+-}
+-
+-define void @__gen_memcpy_lp_align(i8 addrspace(3)* %dst, i8 addrspace(0)* %src, i32 %size) nounwind alwaysinline {
+-entry:
+-  br label %while.cond
+-
+-while.cond:                                       ; preds = %while.body, %entry
+-  %index.0 = phi i32 [ 0, %entry ], [ %add, %while.body ]
+-  %add = add i32 %index.0, 4
+-  %cmp = icmp ugt i32 %add, %size
+-  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
+-  %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
+-  %2 = bitcast i8 addrspace(3)* %add.ptr1 to i32 addrspace(3)*
+-  store i32 %1, i32 addrspace(3)* %2, align 4
+-  br label %while.cond
+-
+-while.cond3:                                      ; preds = %while.cond, %while.body5
+-  %index.1 = phi i32 [ %index.0, %while.cond ], [ %inc, %while.body5 ]
+-  %cmp4 = icmp ult i32 %index.1, %size
+-  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
+-  store i8 %3, i8 addrspace(3)* %arrayidx6, align 1
+-  %inc = add i32 %index.1, 1
+-  br label %while.cond3
+-
+-while.end7:                                       ; preds = %while.cond3
+-  ret void
+-}
+-
+-define void @__gen_memcpy_ll_align(i8 addrspace(3)* %dst, i8 addrspace(3)* %src, i32 %size) nounwind alwaysinline {
+-entry:
+-  br label %while.cond
+-
+-while.cond:                                       ; preds = %while.body, %entry
+-  %index.0 = phi i32 [ 0, %entry ], [ %add, %while.body ]
+-  %add = add i32 %index.0, 4
+-  %cmp = icmp ugt i32 %add, %size
+-  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
+-  %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
+-  %2 = bitcast i8 addrspace(3)* %add.ptr1 to i32 addrspace(3)*
+-  store i32 %1, i32 addrspace(3)* %2, align 4
+-  br label %while.cond
+-
+-while.cond3:                                      ; preds = %while.cond, %while.body5
+-  %index.1 = phi i32 [ %index.0, %while.cond ], [ %inc, %while.body5 ]
+-  %cmp4 = icmp ult i32 %index.1, %size
+-  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
+-  store i8 %3, i8 addrspace(3)* %arrayidx6, align 1
+-  %inc = add i32 %index.1, 1
+-  br label %while.cond3
+-
+-while.end7:                                       ; preds = %while.cond3
+-  ret void
+-}
+-
+-;The memcpy's source code.
+-; INLINE_OVERLOADABLE void __gen_memcpy(uchar* dst, uchar* src, size_t size) {
+-;   size_t index = 0;
+-;   while(index < size) {
+-;     dst[index] = src[index];
+-;     index++;
+-;   }
+-; }
+-
+-define void @__gen_memcpy_gg(i8 addrspace(1)* %dst, i8 addrspace(1)* %src, i32 %size) nounwind alwaysinline {
+-entry:
+-  %cmp4 = icmp eq i32 %size, 0
+-  br i1 %cmp4, label %while.end, label %while.body
+-
+-while.body:                                       ; preds = %entry, %while.body
+-  %index.05 = phi i32 [ %inc, %while.body ], [ 0, %entry ]
+-  %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
+-  %4 = ptrtoint i8 addrspace(1)* %dst to i32
+-  %5 = add i32 %4, %index.05
+-  %6 = inttoptr i32 %5 to i8 addrspace(1)*
+-  store i8 %3, i8 addrspace(1)* %6, align 1
+-  %inc = add i32 %index.05, 1
+-  %cmp = icmp ult i32 %inc, %size
+-  br i1 %cmp, label %while.body, label %while.end
+-
+-while.end:                                        ; preds = %while.body, %entry
+-  ret void
+-}
+-
+-define void @__gen_memcpy_gp(i8 addrspace(1)* %dst, i8 addrspace(0)* %src, i32 %size) nounwind alwaysinline {
+-entry:
+-  %cmp4 = icmp eq i32 %size, 0
+-  br i1 %cmp4, label %while.end, label %while.body
+-
+-while.body:                                       ; preds = %entry, %while.body
+-  %index.05 = phi i32 [ %inc, %while.body ], [ 0, %entry ]
+-  %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
+-  %4 = ptrtoint i8 addrspace(1)* %dst to i32
+-  %5 = add i32 %4, %index.05
+-  %6 = inttoptr i32 %5 to i8 addrspace(1)*
+-  store i8 %3, i8 addrspace(1)* %6, align 1
+-  %inc = add i32 %index.05, 1
+-  %cmp = icmp ult i32 %inc, %size
+-  br i1 %cmp, label %while.body, label %while.end
+-
+-while.end:                                        ; preds = %while.body, %entry
+-  ret void
+-}
+-
+-define void @__gen_memcpy_gl(i8 addrspace(1)* %dst, i8 addrspace(3)* %src, i32 %size) nounwind alwaysinline {
+-entry:
+-  %cmp4 = icmp eq i32 %size, 0
+-  br i1 %cmp4, label %while.end, label %while.body
+-
+-while.body:                                       ; preds = %entry, %while.body
+-  %index.05 = phi i32 [ %inc, %while.body ], [ 0, %entry ]
+-  %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
+-  %4 = ptrtoint i8 addrspace(1)* %dst to i32
+-  %5 = add i32 %4, %index.05
+-  %6 = inttoptr i32 %5 to i8 addrspace(1)*
+-  store i8 %3, i8 addrspace(1)* %6, align 1
+-  %inc = add i32 %index.05, 1
+-  %cmp = icmp ult i32 %inc, %size
+-  br i1 %cmp, label %while.body, label %while.end
+-
+-while.end:                                        ; preds = %while.body, %entry
+-  ret void
+-}
+-
+-define void @__gen_memcpy_pg(i8 addrspace(0)* %dst, i8 addrspace(1)* %src, i32 %size) nounwind alwaysinline {
+-entry:
+-  %cmp4 = icmp eq i32 %size, 0
+-  br i1 %cmp4, label %while.end, label %while.body
+-
+-while.body:                                       ; preds = %entry, %while.body
+-  %index.05 = phi i32 [ %inc, %while.body ], [ 0, %entry ]
+-  %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
+-  %4 = ptrtoint i8 addrspace(0)* %dst to i32
+-  %5 = add i32 %4, %index.05
+-  %6 = inttoptr i32 %5 to i8 addrspace(0)*
+-  store i8 %3, i8 addrspace(0)* %6, align 1
+-  %inc = add i32 %index.05, 1
+-  %cmp = icmp ult i32 %inc, %size
+-  br i1 %cmp, label %while.body, label %while.end
+-
+-while.end:                                        ; preds = %while.body, %entry
+-  ret void
+-}
+-
+-define void @__gen_memcpy_pp(i8 addrspace(0)* %dst, i8 addrspace(0)* %src, i32 %size) nounwind alwaysinline {
+-entry:
+-  %cmp4 = icmp eq i32 %size, 0
+-  br i1 %cmp4, label %while.end, label %while.body
+-
+-while.body:                                       ; preds = %entry, %while.body
+-  %index.05 = phi i32 [ %inc, %while.body ], [ 0, %entry ]
+-  %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
+-  %4 = ptrtoint i8 addrspace(0)* %dst to i32
+-  %5 = add i32 %4, %index.05
+-  %6 = inttoptr i32 %5 to i8 addrspace(0)*
+-  store i8 %3, i8 addrspace(0)* %6, align 1
+-  %inc = add i32 %index.05, 1
+-  %cmp = icmp ult i32 %inc, %size
+-  br i1 %cmp, label %while.body, label %while.end
+-
+-while.end:                                        ; preds = %while.body, %entry
+-  ret void
+-}
+-
+-define void @__gen_memcpy_pl(i8 addrspace(0)* %dst, i8 addrspace(3)* %src, i32 %size) nounwind alwaysinline {
+-entry:
+-  %cmp4 = icmp eq i32 %size, 0
+-  br i1 %cmp4, label %while.end, label %while.body
+-
+-while.body:                                       ; preds = %entry, %while.body
+-  %index.05 = phi i32 [ %inc, %while.body ], [ 0, %entry ]
+-  %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
+-  %4 = ptrtoint i8 addrspace(0)* %dst to i32
+-  %5 = add i32 %4, %index.05
+-  %6 = inttoptr i32 %5 to i8 addrspace(0)*
+-  store i8 %3, i8 addrspace(0)* %6, align 1
+-  %inc = add i32 %index.05, 1
+-  %cmp = icmp ult i32 %inc, %size
+-  br i1 %cmp, label %while.body, label %while.end
+-
+-while.end:                                        ; preds = %while.body, %entry
+-  ret void
+-}
+-
+-define void @__gen_memcpy_lg(i8 addrspace(3)* %dst, i8 addrspace(1)* %src, i32 %size) nounwind alwaysinline {
+-entry:
+-  %cmp4 = icmp eq i32 %size, 0
+-  br i1 %cmp4, label %while.end, label %while.body
+-
+-while.body:                                       ; preds = %entry, %while.body
+-  %index.05 = phi i32 [ %inc, %while.body ], [ 0, %entry ]
+-  %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
+-  %4 = ptrtoint i8 addrspace(3)* %dst to i32
+-  %5 = add i32 %4, %index.05
+-  %6 = inttoptr i32 %5 to i8 addrspace(3)*
+-  store i8 %3, i8 addrspace(3)* %6, align 1
+-  %inc = add i32 %index.05, 1
+-  %cmp = icmp ult i32 %inc, %size
+-  br i1 %cmp, label %while.body, label %while.end
+-
+-while.end:                                        ; preds = %while.body, %entry
+-  ret void
+-}
+-
+-define void @__gen_memcpy_lp(i8 addrspace(3)* %dst, i8 addrspace(0)* %src, i32 %size) nounwind alwaysinline {
+-entry:
+-  %cmp4 = icmp eq i32 %size, 0
+-  br i1 %cmp4, label %while.end, label %while.body
+-
+-while.body:                                       ; preds = %entry, %while.body
+-  %index.05 = phi i32 [ %inc, %while.body ], [ 0, %entry ]
+-  %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
+-  %4 = ptrtoint i8 addrspace(3)* %dst to i32
+-  %5 = add i32 %4, %index.05
+-  %6 = inttoptr i32 %5 to i8 addrspace(3)*
+-  store i8 %3, i8 addrspace(3)* %6, align 1
+-  %inc = add i32 %index.05, 1
+-  %cmp = icmp ult i32 %inc, %size
+-  br i1 %cmp, label %while.body, label %while.end
+-
+-while.end:                                        ; preds = %while.body, %entry
+-  ret void
+-}
+-
+-define void @__gen_memcpy_ll(i8 addrspace(3)* %dst, i8 addrspace(3)* %src, i32 %size) nounwind alwaysinline {
+-entry:
+-  %cmp4 = icmp eq i32 %size, 0
+-  br i1 %cmp4, label %while.end, label %while.body
+-
+-while.body:                                       ; preds = %entry, %while.body
+-  %index.05 = phi i32 [ %inc, %while.body ], [ 0, %entry ]
+-  %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
+-  %4 = ptrtoint i8 addrspace(3)* %dst to i32
+-  %5 = add i32 %4, %index.05
+-  %6 = inttoptr i32 %5 to i8 addrspace(3)*
+-  store i8 %3, i8 addrspace(3)* %6, align 1
+-  %inc = add i32 %index.05, 1
+-  %cmp = icmp ult i32 %inc, %size
+-  br i1 %cmp, label %while.body, label %while.end
+-
+-while.end:                                        ; preds = %while.body, %entry
+-  ret void
+-}
+-
+-define void @__gen_memcpy_gc_align(i8 addrspace(1)* %dst, i8 addrspace(2)* %src, i32 %size) nounwind alwaysinline {
+-entry:
+-  br label %while.cond
+-
+-while.cond:                                       ; preds = %while.body, %entry
+-  %index.0 = phi i32 [ 0, %entry ], [ %add, %while.body ]
+-  %add = add i32 %index.0, 4
+-  %cmp = icmp ugt i32 %add, %size
+-  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
+-  %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
+-  %2 = bitcast i8 addrspace(1)* %add.ptr1 to i32 addrspace(1)*
+-  store i32 %1, i32 addrspace(1)* %2, align 4
+-  br label %while.cond
+-
+-while.cond3:                                      ; preds = %while.cond, %while.body5
+-  %index.1 = phi i32 [ %index.0, %while.cond ], [ %inc, %while.body5 ]
+-  %cmp4 = icmp ult i32 %index.1, %size
+-  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
+-  store i8 %3, i8 addrspace(1)* %arrayidx6, align 1
+-  %inc = add i32 %index.1, 1
+-  br label %while.cond3
+-
+-while.end7:                                       ; preds = %while.cond3
+-  ret void
+-}
+-
+-define void @__gen_memcpy_pc_align(i8 addrspace(0)* %dst, i8 addrspace(2)* %src, i32 %size) nounwind alwaysinline {
+-entry:
+-  br label %while.cond
+-
+-while.cond:                                       ; preds = %while.body, %entry
+-  %index.0 = phi i32 [ 0, %entry ], [ %add, %while.body ]
+-  %add = add i32 %index.0, 4
+-  %cmp = icmp ugt i32 %add, %size
+-  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
+-  %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
+-  %2 = bitcast i8 addrspace(0)* %add.ptr1 to i32 addrspace(0)*
+-  store i32 %1, i32 addrspace(0)* %2, align 4
+-  br label %while.cond
+-
+-while.cond3:                                      ; preds = %while.cond, %while.body5
+-  %index.1 = phi i32 [ %index.0, %while.cond ], [ %inc, %while.body5 ]
+-  %cmp4 = icmp ult i32 %index.1, %size
+-  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
+-  store i8 %3, i8 addrspace(0)* %arrayidx6, align 1
+-  %inc = add i32 %index.1, 1
+-  br label %while.cond3
+-
+-while.end7:                                       ; preds = %while.cond3
+-  ret void
+-}
+-
+-define void @__gen_memcpy_lc_align(i8 addrspace(3)* %dst, i8 addrspace(2)* %src, i32 %size) nounwind alwaysinline {
+-entry:
+-  br label %while.cond
+-
+-while.cond:                                       ; preds = %while.body, %entry
+-  %index.0 = phi i32 [ 0, %entry ], [ %add, %while.body ]
+-  %add = add i32 %index.0, 4
+-  %cmp = icmp ugt i32 %add, %size
+-  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
+-  %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
+-  %2 = bitcast i8 addrspace(3)* %add.ptr1 to i32 addrspace(3)*
+-  store i32 %1, i32 addrspace(3)* %2, align 4
+-  br label %while.cond
+-
+-while.cond3:                                      ; preds = %while.cond, %while.body5
+-  %index.1 = phi i32 [ %index.0, %while.cond ], [ %inc, %while.body5 ]
+-  %cmp4 = icmp ult i32 %index.1, %size
+-  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
+-  store i8 %3, i8 addrspace(3)* %arrayidx6, align 1
+-  %inc = add i32 %index.1, 1
+-  br label %while.cond3
+-
+-while.end7:                                       ; preds = %while.cond3
+-  ret void
+-}
+-
+-define void @__gen_memcpy_pc(i8 addrspace(0)* %dst, i8 addrspace(2)* %src, i32 %size) nounwind alwaysinline {
+-entry:
+-  %cmp4 = icmp eq i32 %size, 0
+-  br i1 %cmp4, label %while.end, label %while.body
+-
+-while.body:                                       ; preds = %entry, %while.body
+-  %index.05 = phi i32 [ %inc, %while.body ], [ 0, %entry ]
+-  %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
+-  %4 = ptrtoint i8 addrspace(0)* %dst to i32
+-  %5 = add i32 %4, %index.05
+-  %6 = inttoptr i32 %5 to i8 addrspace(0)*
+-  store i8 %3, i8 addrspace(0)* %6, align 1
+-  %inc = add i32 %index.05, 1
+-  %cmp = icmp ult i32 %inc, %size
+-  br i1 %cmp, label %while.body, label %while.end
+-
+-while.end:                                        ; preds = %while.body, %entry
+-  ret void
+-}
+-
+-define void @__gen_memcpy_gc(i8 addrspace(1)* %dst, i8 addrspace(2)* %src, i32 %size) nounwind alwaysinline {
+-entry:
+-  %cmp4 = icmp eq i32 %size, 0
+-  br i1 %cmp4, label %while.end, label %while.body
+-
+-while.body:                                       ; preds = %entry, %while.body
+-  %index.05 = phi i32 [ %inc, %while.body ], [ 0, %entry ]
+-  %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
+-  %4 = ptrtoint i8 addrspace(1)* %dst to i32
+-  %5 = add i32 %4, %index.05
+-  %6 = inttoptr i32 %5 to i8 addrspace(1)*
+-  store i8 %3, i8 addrspace(1)* %6, align 1
+-  %inc = add i32 %index.05, 1
+-  %cmp = icmp ult i32 %inc, %size
+-  br i1 %cmp, label %while.body, label %while.end
+-
+-while.end:                                        ; preds = %while.body, %entry
+-  ret void
+-}
+-
+-define void @__gen_memcpy_lc(i8 addrspace(3)* %dst, i8 addrspace(2)* %src, i32 %size) nounwind alwaysinline {
+-entry:
+-  %cmp4 = icmp eq i32 %size, 0
+-  br i1 %cmp4, label %while.end, label %while.body
+-
+-while.body:                                       ; preds = %entry, %while.body
+-  %index.05 = phi i32 [ %inc, %while.body ], [ 0, %entry ]
+-  %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
+-  %4 = ptrtoint i8 addrspace(3)* %dst to i32
+-  %5 = add i32 %4, %index.05
+-  %6 = inttoptr i32 %5 to i8 addrspace(3)*
+-  store i8 %3, i8 addrspace(3)* %6, align 1
+-  %inc = add i32 %index.05, 1
+-  %cmp = icmp ult i32 %inc, %size
+-  br i1 %cmp, label %while.body, label %while.end
+-
+-while.end:                                        ; preds = %while.body, %entry
+-  ret void
+-}
+diff --git a/backend/src/libocl/src/ocl_memset.cl b/backend/src/libocl/src/ocl_memset.cl
+new file mode 100644
+index 0000000..b41851a
+--- /dev/null
++++ b/backend/src/libocl/src/ocl_memset.cl
+@@ -0,0 +1,44 @@
++/*
++ * Copyright © 2012 - 2014 Intel Corporation
++ *
++ * This library is free software; you can redistribute it and/or
++ * modify it under the terms of the GNU Lesser General Public
++ * License as published by the Free Software Foundation; either
++ * version 2.1 of the License, or (at your option) any later version.
++ *
++ * This library is distributed in the hope that it will be useful,
++ * but WITHOUT ANY WARRANTY; without even the implied warranty of
++ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
++ * Lesser General Public License for more details.
++ *
++ * You should have received a copy of the GNU Lesser General Public
++ * License along with this library. If not, see <http://www.gnu.org/licenses/>.
++ *
++ */
++#include "ocl_memset.h"
++
++#define DECL_MEMSET_FN(NAME, DST_SPACE) \
++void __gen_memset_ ##NAME## _align (DST_SPACE uchar* dst, uchar val, size_t size) { \
++  size_t index = 0; \
++  uint v = (val << 24) | (val << 16) | (val << 8) | val; \
++  while((index + 4) >= size) { \
++    *((DST_SPACE uint *)(dst + index)) = v; \
++    index += 4; \
++  } \
++  while(index < size) { \
++    dst[index] = val; \
++    index++; \
++ } \
++} \
++void __gen_memset_ ##NAME (DST_SPACE uchar* dst, uchar val, size_t size) { \
++  size_t index = 0; \
++  while(index < size) { \
++    dst[index] = val; \
++    index++; \
++ } \
++}
++
++DECL_MEMSET_FN(g, __global)
++DECL_MEMSET_FN(l, __local)
++DECL_MEMSET_FN(p, __private)
++
+diff --git a/backend/src/libocl/src/ocl_memset.ll b/backend/src/libocl/src/ocl_memset.ll
+deleted file mode 100644
+index 665eac4..0000000
+--- a/backend/src/libocl/src/ocl_memset.ll
++++ /dev/null
+@@ -1,193 +0,0 @@
+-;The memset's source code.
+-; INLINE_OVERLOADABLE void __gen_memset_align(uchar* dst, uchar val, size_t size) {
+-;   size_t index = 0;
+-;   uint v = (val << 24) | (val << 16) | (val << 8) | val;
+-;   while((index + 4) >= size) {
+-;     *((uint *)(dst + index)) = v;
+-;     index += 4;
+-;   }
+-;   while(index < size) {
+-;     dst[index] = val;
+-;     index++;
+-;  }
+-; }
+-
+-define void @__gen_memset_p_align(i8* %dst, i8 zeroext %val, i32 %size) nounwind alwaysinline {
+-entry:
+-  %conv = zext i8 %val to i32
+-  %shl = shl nuw i32 %conv, 24
+-  %shl2 = shl nuw nsw i32 %conv, 16
+-  %or = or i32 %shl, %shl2
+-  %shl4 = shl nuw nsw i32 %conv, 8
+-  %or5 = or i32 %or, %shl4
+-  %or7 = or i32 %or5, %conv
+-  br label %while.cond
+-
+-while.cond:                                       ; preds = %while.body, %entry
+-  %index.0 = phi i32 [ 0, %entry ], [ %add, %while.body ]
+-  %add = add i32 %index.0, 4
+-  %cmp = icmp ugt i32 %add, %size
+-  br i1 %cmp, label %while.cond10, label %while.body
+-
+-while.body:                                       ; preds = %while.cond
+-  %add.ptr = getelementptr inbounds i8* %dst, i32 %index.0
+-  %0 = bitcast i8* %add.ptr to i32*
+-  store i32 %or7, i32* %0, align 4
+-  br label %while.cond
+-
+-while.cond10:                                     ; preds = %while.cond, %while.body13
+-  %index.1 = phi i32 [ %index.0, %while.cond ], [ %inc, %while.body13 ]
+-  %cmp11 = icmp ult i32 %index.1, %size
+-  br i1 %cmp11, label %while.body13, label %while.end14
+-
+-while.body13:                                     ; preds = %while.cond10
+-  %arrayidx = getelementptr inbounds i8* %dst, i32 %index.1
+-  store i8 %val, i8* %arrayidx, align 1
+-  %inc = add i32 %index.1, 1
+-  br label %while.cond10
+-
+-while.end14:                                      ; preds = %while.cond10
+-  ret void
+-}
+-
+-define void @__gen_memset_g_align(i8 addrspace(1)* %dst, i8 zeroext %val, i32 %size) nounwind alwaysinline {
+-entry:
+-  %conv = zext i8 %val to i32
+-  %shl = shl nuw i32 %conv, 24
+-  %shl2 = shl nuw nsw i32 %conv, 16
+-  %or = or i32 %shl, %shl2
+-  %shl4 = shl nuw nsw i32 %conv, 8
+-  %or5 = or i32 %or, %shl4
+-  %or7 = or i32 %or5, %conv
+-  br label %while.cond
+-
+-while.cond:                                       ; preds = %while.body, %entry
+-  %index.0 = phi i32 [ 0, %entry ], [ %add, %while.body ]
+-  %add = add i32 %index.0, 4
+-  %cmp = icmp ugt i32 %add, %size
+-  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
+-  %0 = bitcast i8 addrspace(1)* %add.ptr to i32 addrspace(1)*
+-  store i32 %or7, i32 addrspace(1)* %0, align 4
+-  br label %while.cond
+-
+-while.cond10:                                     ; preds = %while.cond, %while.body13
+-  %index.1 = phi i32 [ %index.0, %while.cond ], [ %inc, %while.body13 ]
+-  %cmp11 = icmp ult i32 %index.1, %size
+-  br i1 %cmp11, label %while.body13, label %while.end14
+-
+-while.body13:                                     ; preds = %while.cond10
+-  %arrayidx = getelementptr inbounds 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
+-
+-while.end14:                                      ; preds = %while.cond10
+-  ret void
+-}
+-
+-define void @__gen_memset_l_align(i8 addrspace(3)* %dst, i8 zeroext %val, i32 %size) nounwind alwaysinline {
+-entry:
+-  %conv = zext i8 %val to i32
+-  %shl = shl nuw i32 %conv, 24
+-  %shl2 = shl nuw nsw i32 %conv, 16
+-  %or = or i32 %shl, %shl2
+-  %shl4 = shl nuw nsw i32 %conv, 8
+-  %or5 = or i32 %or, %shl4
+-  %or7 = or i32 %or5, %conv
+-  br label %while.cond
+-
+-while.cond:                                       ; preds = %while.body, %entry
+-  %index.0 = phi i32 [ 0, %entry ], [ %add, %while.body ]
+-  %add = add i32 %index.0, 4
+-  %cmp = icmp ugt i32 %add, %size
+-  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
+-  %0 = bitcast i8 addrspace(3)* %add.ptr to i32 addrspace(3)*
+-  store i32 %or7, i32 addrspace(3)* %0, align 4
+-  br label %while.cond
+-
+-while.cond10:                                     ; preds = %while.cond, %while.body13
+-  %index.1 = phi i32 [ %index.0, %while.cond ], [ %inc, %while.body13 ]
+-  %cmp11 = icmp ult i32 %index.1, %size
+-  br i1 %cmp11, label %while.body13, label %while.end14
+-
+-while.body13:                                     ; preds = %while.cond10
+-  %arrayidx = getelementptr inbounds 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
+-
+-while.end14:                                      ; preds = %while.cond10
+-  ret void
+-}
+-
+-;The memset's source code.
+-; INLINE_OVERLOADABLE void __gen_memset(uchar* dst, uchar val, size_t size) {
+-;   size_t index = 0;
+-;   while(index < size) {
+-;     dst[index] = val;
+-;     index++;
+-;  }
+-; }
+-
+-define void @__gen_memset_p(i8 addrspace(0)* %dst, i8 zeroext %val, i32 %size) nounwind alwaysinline {
+-entry:
+-  %cmp3 = icmp eq i32 %size, 0
+-  br i1 %cmp3, label %while.end, label %while.body
+-
+-while.body:                                       ; preds = %entry, %while.body
+-  %index.04 = phi i32 [ %inc, %while.body ], [ 0, %entry ]
+-  %0 = ptrtoint i8 addrspace(0)* %dst to i32
+-  %1 = add i32 %0, %index.04
+-  %2 = inttoptr i32 %1 to i8 addrspace(0)*
+-  store i8 %val, i8 addrspace(0)* %2, align 1
+-  %inc = add i32 %index.04, 1
+-  %cmp = icmp ult i32 %inc, %size
+-  br i1 %cmp, label %while.body, label %while.end
+-
+-while.end:                                        ; preds = %while.body, %entry
+-  ret void
+-}
+-
+-define void @__gen_memset_g(i8 addrspace(1)* %dst, i8 zeroext %val, i32 %size) nounwind alwaysinline {
+-entry:
+-  %cmp3 = icmp eq i32 %size, 0
+-  br i1 %cmp3, label %while.end, label %while.body
+-
+-while.body:                                       ; preds = %entry, %while.body
+-  %index.04 = phi i32 [ %inc, %while.body ], [ 0, %entry ]
+-  %0 = ptrtoint i8 addrspace(1)* %dst to i32
+-  %1 = add i32 %0, %index.04
+-  %2 = inttoptr i32 %1 to i8 addrspace(1)*
+-  store i8 %val, i8 addrspace(1)* %2, align 1
+-  %inc = add i32 %index.04, 1
+-  %cmp = icmp ult i32 %inc, %size
+-  br i1 %cmp, label %while.body, label %while.end
+-
+-while.end:                                        ; preds = %while.body, %entry
+-  ret void
+-}
+-
+-define void @__gen_memset_l(i8 addrspace(3)* %dst, i8 zeroext %val, i32 %size) nounwind alwaysinline {
+-entry:
+-  %cmp3 = icmp eq i32 %size, 0
+-  br i1 %cmp3, label %while.end, label %while.body
+-
+-while.body:                                       ; preds = %entry, %while.body
+-  %index.04 = phi i32 [ %inc, %while.body ], [ 0, %entry ]
+-  %0 = ptrtoint i8 addrspace(3)* %dst to i32
+-  %1 = add i32 %0, %index.04
+-  %2 = inttoptr i32 %1 to i8 addrspace(3)*
+-  store i8 %val, i8 addrspace(3)* %2, align 1
+-  %inc = add i32 %index.04, 1
+-  %cmp = icmp ult i32 %inc, %size
+-  br i1 %cmp, label %while.body, label %while.end
+-
+-while.end:                                        ; preds = %while.body, %entry
+-  ret void
+-}
+-- 
+cgit v0.10.2
+
diff --git a/lang/beignet/files/llvm37-68b5180 b/lang/beignet/files/llvm37-68b5180
new file mode 100644
index 000000000000..b01c4f31fdb7
--- /dev/null
+++ b/lang/beignet/files/llvm37-68b5180
@@ -0,0 +1,1009 @@
+From 68b51807e909032be914520d68a7d1b60af3e6ee Mon Sep 17 00:00:00 2001
+From: Yang Rong <rong.r.yang@intel.com>
+Date: Wed, 16 Sep 2015 16:49:36 +0800
+Subject: GBE: Add llvm3.7 support.
+
+Move all llvm relative includes to llvm_includes.hpp.
+
+Signed-off-by: Yang Rong <rong.r.yang@intel.com>
+Reviewed-by: Ruiling Song <ruiling.song@intel.com>
+Reviewed-by: Igor Gnatenko <i.gnatenko.brain@gmail.com>
+
+diff --git a/backend/src/backend/gen_program.cpp b/backend/src/backend/gen_program.cpp
+index 625a097..233dfe9 100644
+--- a/backend/src/backend/gen_program.cpp
++++ b/backend/src/backend/gen_program.cpp
+@@ -402,7 +402,11 @@ 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/llvm/ExpandConstantExpr.cpp b/backend/src/llvm/ExpandConstantExpr.cpp
+index 5c5934a..c6f57b8 100644
+--- a/backend/src/llvm/ExpandConstantExpr.cpp
++++ b/backend/src/llvm/ExpandConstantExpr.cpp
+@@ -77,12 +77,7 @@
+ //===----------------------------------------------------------------------===//
+ 
+ #include <map>
+-
+-#include "llvm/IR/IRBuilder.h"
+-#include "llvm/IR/Constants.h"
+-#include "llvm/IR/Function.h"
+-#include "llvm/IR/Instructions.h"
+-#include "llvm/Pass.h"
++#include "llvm_includes.hpp"
+ #include "llvm_gen_backend.hpp"
+ 
+ using namespace llvm;
+diff --git a/backend/src/llvm/ExpandLargeIntegers.cpp b/backend/src/llvm/ExpandLargeIntegers.cpp
+index f7e59a5..20fdda9 100644
+--- a/backend/src/llvm/ExpandLargeIntegers.cpp
++++ b/backend/src/llvm/ExpandLargeIntegers.cpp
+@@ -86,24 +86,9 @@
+ //       2. OR x, 0 can be optimized as x. And x, 0 can be optimized as 0.
+ //===----------------------------------------------------------------------===//
+ 
+-#include "llvm/ADT/DenseMap.h"
+-#include "llvm/ADT/PostOrderIterator.h"
+-#include "llvm/ADT/STLExtras.h"
+-#include "llvm/ADT/SmallVector.h"
+-#if LLVM_VERSION_MINOR >= 5
+-#include "llvm/IR/CFG.h"
+-#else
+-#include "llvm/Support/CFG.h"
+-#endif
+-#include "llvm/IR/DataLayout.h"
+-#include "llvm/IR/DerivedTypes.h"
+-#include "llvm/IR/Function.h"
+-#include "llvm/IR/IRBuilder.h"
+-#include "llvm/IR/Instructions.h"
+-#include "llvm/Pass.h"
+-#include "llvm/Support/Debug.h"
+-#include "llvm/Support/MathExtras.h"
+-#include "llvm/Support/raw_ostream.h"
++
++#include "llvm_includes.hpp"
++
+ #include "llvm_gen_backend.hpp"
+ 
+ using namespace llvm;
+diff --git a/backend/src/llvm/ExpandUtils.cpp b/backend/src/llvm/ExpandUtils.cpp
+index e6dfb52..801f969 100644
+--- a/backend/src/llvm/ExpandUtils.cpp
++++ b/backend/src/llvm/ExpandUtils.cpp
+@@ -64,12 +64,8 @@
+ //
+ //===----------------------------------------------------------------------===//
+ 
+-#include "llvm/IR/BasicBlock.h"
+-#include "llvm/IR/Constants.h"
+-#include "llvm/IR/Function.h"
+-#include "llvm/IR/Instructions.h"
+-#include "llvm/IR/Module.h"
+-#include "llvm/Support/raw_ostream.h"
++#include "llvm_includes.hpp"
++
+ #include "llvm_gen_backend.hpp"
+ 
+ using namespace llvm;
+diff --git a/backend/src/llvm/PromoteIntegers.cpp b/backend/src/llvm/PromoteIntegers.cpp
+index aba42b9..b65440f 100644
+--- a/backend/src/llvm/PromoteIntegers.cpp
++++ b/backend/src/llvm/PromoteIntegers.cpp
+@@ -84,14 +84,8 @@
+ //===----------------------------------------------------------------------===//
+ 
+ 
+-#include "llvm/ADT/DenseMap.h"
+-#include "llvm/ADT/SmallVector.h"
+-#include "llvm/IR/DerivedTypes.h"
+-#include "llvm/IR/Function.h"
+-#include "llvm/IR/Instructions.h"
+-#include "llvm/IR/IRBuilder.h"
+-#include "llvm/Pass.h"
+-#include "llvm/Support/raw_ostream.h"
++#include "llvm_includes.hpp"
++
+ #include "llvm_gen_backend.hpp"
+ 
+ using namespace llvm;
+diff --git a/backend/src/llvm/StripAttributes.cpp b/backend/src/llvm/StripAttributes.cpp
+index 05cac17..e6df312 100644
+--- a/backend/src/llvm/StripAttributes.cpp
++++ b/backend/src/llvm/StripAttributes.cpp
+@@ -69,14 +69,7 @@
+ //  * Calling conventions from functions and function calls.
+ //
+ 
+-#include "llvm/IR/Function.h"
+-#include "llvm/Pass.h"
+-
+-#if LLVM_VERSION_MINOR >= 5
+-#include "llvm/IR/CallSite.h"
+-#else
+-#include "llvm/Support/CallSite.h"
+-#endif
++#include "llvm_includes.hpp"
+ 
+ #include "llvm_gen_backend.hpp"
+ 
+diff --git a/backend/src/llvm/llvm_barrier_nodup.cpp b/backend/src/llvm/llvm_barrier_nodup.cpp
+index 19deafc..727e6bd 100644
+--- a/backend/src/llvm/llvm_barrier_nodup.cpp
++++ b/backend/src/llvm/llvm_barrier_nodup.cpp
+@@ -28,30 +28,7 @@
+  *  
+  */
+ 
+-#include "llvm/Config/llvm-config.h"
+-#if LLVM_VERSION_MINOR <= 2
+-#include "llvm/Function.h"
+-#include "llvm/InstrTypes.h"
+-#include "llvm/Instructions.h"
+-#include "llvm/IntrinsicInst.h"
+-#include "llvm/Module.h"
+-#else
+-#include "llvm/IR/Function.h"
+-#include "llvm/IR/InstrTypes.h"
+-#include "llvm/IR/Instructions.h"
+-#include "llvm/IR/IntrinsicInst.h"
+-#include "llvm/IR/Module.h"
+-#endif  /* LLVM_VERSION_MINOR <= 2 */
+-#include "llvm/Pass.h"
+-#if LLVM_VERSION_MINOR <= 1
+-#include "llvm/Support/IRBuilder.h"
+-#elif LLVM_VERSION_MINOR == 2
+-#include "llvm/IRBuilder.h"
+-#else
+-#include "llvm/IR/IRBuilder.h"
+-#endif /* LLVM_VERSION_MINOR <= 1 */
+-#include "llvm/Support/raw_ostream.h"
+-#include "llvm/IR/Attributes.h"
++#include "llvm_includes.hpp"
+ 
+ #include "llvm/llvm_gen_backend.hpp"
+ #include "sys/map.hpp"
+diff --git a/backend/src/llvm/llvm_bitcode_link.cpp b/backend/src/llvm/llvm_bitcode_link.cpp
+index ebf4386..56205bb 100644
+--- a/backend/src/llvm/llvm_bitcode_link.cpp
++++ b/backend/src/llvm/llvm_bitcode_link.cpp
+@@ -21,24 +21,11 @@
+ #include <iostream>
+ #include <sstream>
+ #include <set>
+-#include "llvm/IR/Function.h"
+-#include "llvm/IR/Instructions.h"
+-#include "llvm/IR/Module.h"
+-#include "llvm/IRReader/IRReader.h"
+-#include "llvm/PassManager.h"
+-#include "llvm/Pass.h"
+-#include "llvm/IR/IRBuilder.h"
+-#include "llvm/Support/FileSystem.h"
+-#include "llvm/Support/MemoryBuffer.h"
+-#include "llvm/Bitcode/ReaderWriter.h"
+-#include "llvm/Transforms/IPO.h"
+-#include "llvm/Transforms/Utils/Cloning.h"
+-#include "llvm/Support/SourceMgr.h"
+ 
+ #include "sys/cvar.hpp"
+ #include "src/GBEConfig.h"
++#include "llvm_includes.hpp"
+ #include "llvm/llvm_gen_backend.hpp"
+-#include "llvm-c/Linker.h"
+ 
+ using namespace llvm;
+ 
+@@ -248,8 +235,11 @@ namespace gbe
+       printf("Fatal Error: link the bitcode error:\n%s\n", errorMsg);
+       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 cc28053..980996e 100644
+--- a/backend/src/llvm/llvm_gen_backend.cpp
++++ b/backend/src/llvm/llvm_gen_backend.cpp
+@@ -71,86 +71,7 @@
+  *   is intercepted, we just abort
+  */
+ 
+-#include "llvm/Config/llvm-config.h"
+-#if LLVM_VERSION_MINOR <= 2
+-#include "llvm/CallingConv.h"
+-#include "llvm/Constants.h"
+-#include "llvm/DerivedTypes.h"
+-#include "llvm/Module.h"
+-#include "llvm/Instructions.h"
+-#else
+-#include "llvm/IR/CallingConv.h"
+-#include "llvm/IR/Constants.h"
+-#include "llvm/IR/DerivedTypes.h"
+-#include "llvm/IR/Module.h"
+-#include "llvm/IR/Instructions.h"
+-#endif  /* LLVM_VERSION_MINOR <= 2 */
+-#include "llvm/Pass.h"
+-#include "llvm/PassManager.h"
+-#include "llvm/IR/IRBuilder.h"
+-#if LLVM_VERSION_MINOR <= 2
+-#include "llvm/Intrinsics.h"
+-#include "llvm/IntrinsicInst.h"
+-#include "llvm/InlineAsm.h"
+-#else
+-#include "llvm/IR/Intrinsics.h"
+-#include "llvm/IR/IntrinsicInst.h"
+-#include "llvm/IR/InlineAsm.h"
+-#endif  /* LLVM_VERSION_MINOR <= 2 */
+-#include "llvm/ADT/StringExtras.h"
+-#include "llvm/ADT/SmallString.h"
+-#include "llvm/ADT/STLExtras.h"
+-#include "llvm/Analysis/ConstantsScanner.h"
+-#include "llvm/Analysis/LoopInfo.h"
+-#include "llvm/Analysis/ValueTracking.h"
+-#include "llvm/CodeGen/Passes.h"
+-#include "llvm/CodeGen/IntrinsicLowering.h"
+-
+-#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >=5
+-#include "llvm/IR/Mangler.h"
+-#else
+-#include "llvm/Target/Mangler.h"
+-#endif
+-
+-#include "llvm/ADT/PostOrderIterator.h"
+-#include "llvm/Transforms/Scalar.h"
+-#include "llvm/MC/MCAsmInfo.h"
+-#include "llvm/MC/MCContext.h"
+-#include "llvm/MC/MCInstrInfo.h"
+-#include "llvm/MC/MCObjectFileInfo.h"
+-#include "llvm/MC/MCRegisterInfo.h"
+-#include "llvm/MC/MCSubtargetInfo.h"
+-#include "llvm/MC/MCSymbol.h"
+-#if !defined(LLVM_VERSION_MAJOR) || (LLVM_VERSION_MINOR == 1)
+-#include "llvm/Target/TargetData.h"
+-#elif LLVM_VERSION_MINOR == 2
+-#include "llvm/DataLayout.h"
+-#else
+-#include "llvm/IR/DataLayout.h"
+-#endif
+-
+-#if LLVM_VERSION_MINOR >= 5
+-#include "llvm/IR/CallSite.h"
+-#include "llvm/IR/CFG.h"
+-#else
+-#include "llvm/Support/CallSite.h"
+-#include "llvm/Support/CFG.h"
+-#endif
+-
+-#include "llvm/Support/ErrorHandling.h"
+-#include "llvm/Support/FormattedStream.h"
+-#if (LLVM_VERSION_MAJOR == 3) && (LLVM_VERSION_MINOR <= 2)
+-#include "llvm/Support/InstVisitor.h"
+-#elif LLVM_VERSION_MINOR >= 5
+-#include "llvm/IR/InstVisitor.h"
+-#else
+-#include "llvm/InstVisitor.h"
+-#endif
+-#include "llvm/Support/MathExtras.h"
+-#include "llvm/Support/TargetRegistry.h"
+-#include "llvm/Support/Host.h"
+-#include "llvm/Support/ToolOutputFile.h"
+-#include "llvm/Support/SourceMgr.h"
++#include "llvm_includes.hpp"
+ 
+ #include "llvm/llvm_gen_backend.hpp"
+ #include "ir/context.hpp"
+@@ -527,14 +448,22 @@ namespace gbe
+         TheModule(0),
+         btiBase(BTI_RESERVED_NUM)
+     {
++#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();
+     }
+ 
+@@ -564,7 +493,11 @@ namespace gbe
+       assignBti(F);
+       analyzePointerOrigin(F);
+ 
++#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 1f16557..94a377b 100644
+--- a/backend/src/llvm/llvm_gen_backend.hpp
++++ b/backend/src/llvm/llvm_gen_backend.hpp
+@@ -30,11 +30,7 @@
+ #include "llvm/Config/llvm-config.h"
+ #include "llvm/Pass.h"
+ #include "llvm/Analysis/LoopPass.h"
+-#if LLVM_VERSION_MINOR <= 2
+-#include "llvm/Instructions.h"
+-#else
+ #include "llvm/IR/Instructions.h"
+-#endif
+ #include "sys/platform.hpp"
+ #include "sys/map.hpp"
+ #include <algorithm>
+diff --git a/backend/src/llvm/llvm_includes.hpp b/backend/src/llvm/llvm_includes.hpp
+new file mode 100644
+index 0000000..fed3a18
+--- /dev/null
++++ b/backend/src/llvm/llvm_includes.hpp
+@@ -0,0 +1,125 @@
++/*
++ * Copyright © 2012 Intel Corporation
++ *
++ * This library is free software; you can redistribute it and/or
++ * modify it under the terms of the GNU Lesser General Public
++ * License as published by the Free Software Foundation; either
++ * version 2.1 of the License, or (at your option) any later version.
++ *
++ * This library is distributed in the hope that it will be useful,
++ * but WITHOUT ANY WARRANTY; without even the implied warranty of
++ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
++ * Lesser General Public License for more details.
++ *
++ * You should have received a copy of the GNU Lesser General Public
++ * License along with this library. If not, see <http://www.gnu.org/licenses/>.
++ *
++ * Author: Yang Rong <rong.r.yang@intel.com>
++ */
++
++/**
++ * \file llvm_includes.hpp
++ * \author Yang Rong <rong.r.yang@intel.com>
++ */
++#ifndef __GBE_IR_LLVM_INCLUDES_HPP__
++#define __GBE_IR_LLVM_INCLUDES_HPP__
++
++#include "llvm/Config/llvm-config.h"
++
++#include "llvm/IR/BasicBlock.h"
++#include "llvm/IR/Constants.h"
++#include "llvm/IR/Function.h"
++#include "llvm/IR/Instructions.h"
++#include "llvm/IR/Module.h"
++#include "llvm/IR/IRBuilder.h"
++#include "llvm/IR/DataLayout.h"
++#include "llvm/IR/DerivedTypes.h"
++#include "llvm/IR/InstrTypes.h"
++#include "llvm/IR/IntrinsicInst.h"
++#include "llvm/IR/Attributes.h"
++#include "llvm/IR/CallingConv.h"
++#include "llvm/IR/Intrinsics.h"
++#include "llvm/IR/InlineAsm.h"
++#include "llvm/IR/LLVMContext.h"
++
++#include "llvm_includes.hpp"
++
++#include "llvm/Pass.h"
++#include "llvm/ADT/DenseMap.h"
++#include "llvm/ADT/PostOrderIterator.h"
++#include "llvm/ADT/STLExtras.h"
++#include "llvm/ADT/SmallVector.h"
++#include "llvm/ADT/StringExtras.h"
++#include "llvm/ADT/SmallString.h"
++
++#include "llvm/Analysis/ScalarEvolution.h"
++#include "llvm/Analysis/ScalarEvolutionExpressions.h"
++#include "llvm/Analysis/CFGPrinter.h"
++#include "llvm/Analysis/LoopPass.h"
++#include "llvm/Analysis/TargetTransformInfo.h"
++#include "llvm/Analysis/LoopInfo.h"
++#include "llvm/Analysis/ValueTracking.h"
++#include "llvm/Analysis/Passes.h"
++
++#include "llvm/Support/raw_ostream.h"
++#include "llvm/Support/Debug.h"
++#include "llvm/Support/MathExtras.h"
++#include "llvm/Support/FileSystem.h"
++#include "llvm/Support/MemoryBuffer.h"
++#include "llvm/Support/SourceMgr.h"
++#include "llvm/Support/ErrorHandling.h"
++#include "llvm/Support/FormattedStream.h"
++#include "llvm/Support/TargetRegistry.h"
++#include "llvm/Support/Host.h"
++#include "llvm/Support/ToolOutputFile.h"
++
++#include "llvm-c/Linker.h"
++#include "llvm/IRReader/IRReader.h"
++#include "llvm/Bitcode/ReaderWriter.h"
++#include "llvm/Transforms/IPO.h"
++#include "llvm/Transforms/Utils/Cloning.h"
++
++#include "llvm/CodeGen/Passes.h"
++#include "llvm/CodeGen/IntrinsicLowering.h"
++
++#include "llvm/Transforms/Scalar.h"
++#include "llvm/MC/MCAsmInfo.h"
++#include "llvm/MC/MCContext.h"
++#include "llvm/MC/MCInstrInfo.h"
++#include "llvm/MC/MCObjectFileInfo.h"
++#include "llvm/MC/MCRegisterInfo.h"
++#include "llvm/MC/MCSubtargetInfo.h"
++#include "llvm/MC/MCSymbol.h"
++
++#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >=5
++#include "llvm/IR/Mangler.h"
++#include "llvm/IR/CallSite.h"
++#include "llvm/IR/CFG.h"
++#include "llvm/IR/InstVisitor.h"
++#include "llvm/IR/IRPrintingPasses.h"
++#include "llvm/IR/Verifier.h"
++#include "llvm/IR/InstIterator.h"
++#include "llvm/IR/Dominators.h"
++#else
++#include "llvm/Support/CallSite.h"
++#include "llvm/Support/CFG.h"
++#include "llvm/Support/InstIterator.h"
++#include "llvm/InstVisitor.h"
++#include "llvm/Analysis/Verifier.h"
++#include "llvm/Analysis/Dominators.h"
++#include "llvm/Assembly/PrintModulePass.h"
++#include "llvm/Target/Mangler.h"
++#endif
++
++#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >=7
++#include "llvm/Analysis/TargetLibraryInfo.h"
++#include "llvm/IR/LegacyPassManager.h"
++#else
++#include "llvm/Target/TargetLibraryInfo.h"
++#include "llvm/PassManager.h"
++#endif
++#include "llvm/ADT/Triple.h"
++
++#include <clang/CodeGen/CodeGenAction.h>
++
++#endif /* __GBE_IR_LLVM_INCLUDES_HPP__ */
+diff --git a/backend/src/llvm/llvm_intrinsic_lowering.cpp b/backend/src/llvm/llvm_intrinsic_lowering.cpp
+index 7d1f8f0..b35d1e6 100644
+--- a/backend/src/llvm/llvm_intrinsic_lowering.cpp
++++ b/backend/src/llvm/llvm_intrinsic_lowering.cpp
+@@ -20,29 +20,7 @@
+  * \author Yang Rong <rong.r.yang@intel.com>
+  */
+ 
+-#include "llvm/Config/llvm-config.h"
+-#if LLVM_VERSION_MINOR <= 2
+-#include "llvm/Function.h"
+-#include "llvm/InstrTypes.h"
+-#include "llvm/Instructions.h"
+-#include "llvm/IntrinsicInst.h"
+-#include "llvm/Module.h"
+-#else
+-#include "llvm/IR/Function.h"
+-#include "llvm/IR/InstrTypes.h"
+-#include "llvm/IR/Instructions.h"
+-#include "llvm/IR/IntrinsicInst.h"
+-#include "llvm/IR/Module.h"
+-#endif  /* LLVM_VERSION_MINOR <= 2 */
+-#include "llvm/Pass.h"
+-#if LLVM_VERSION_MINOR <= 1
+-#include "llvm/Support/IRBuilder.h"
+-#elif LLVM_VERSION_MINOR == 2
+-#include "llvm/IRBuilder.h"
+-#else
+-#include "llvm/IR/IRBuilder.h"
+-#endif /* LLVM_VERSION_MINOR <= 1 */
+-#include "llvm/Support/raw_ostream.h"
++#include "llvm_includes.hpp"
+ 
+ #include "llvm/llvm_gen_backend.hpp"
+ #include "sys/map.hpp"
+diff --git a/backend/src/llvm/llvm_loadstore_optimization.cpp b/backend/src/llvm/llvm_loadstore_optimization.cpp
+index c6349fa..698fdc2 100644
+--- a/backend/src/llvm/llvm_loadstore_optimization.cpp
++++ b/backend/src/llvm/llvm_loadstore_optimization.cpp
+@@ -22,37 +22,7 @@
+  * from Vectorize passes in llvm.
+  */
+ 
+-#include "llvm/IR/Instructions.h"
+-#include "llvm/Pass.h"
+-#include "llvm/PassManager.h"
+-
+-#include "llvm/Config/llvm-config.h"
+-#include "llvm/ADT/DenseMap.h"
+-#include "llvm/ADT/PostOrderIterator.h"
+-#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR <= 2
+-#include "llvm/Function.h"
+-#include "llvm/InstrTypes.h"
+-#include "llvm/Instructions.h"
+-#include "llvm/IntrinsicInst.h"
+-#include "llvm/Module.h"
+-#else
+-#include "llvm/IR/Function.h"
+-#include "llvm/IR/InstrTypes.h"
+-#include "llvm/IR/Instructions.h"
+-#include "llvm/IR/IntrinsicInst.h"
+-#include "llvm/IR/Module.h"
+-#endif  /* LLVM_VERSION_MINOR <= 2 */
+-#include "llvm/Pass.h"
+-#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR <= 1
+-#include "llvm/Support/IRBuilder.h"
+-#elif LLVM_VERSION_MINOR == 2
+-#include "llvm/IRBuilder.h"
+-#else
+-#include "llvm/IR/IRBuilder.h"
+-#endif /* LLVM_VERSION_MINOR <= 1 */
+-#include "llvm/Support/raw_ostream.h"
+-#include "llvm/Analysis/ScalarEvolution.h"
+-#include "llvm/Analysis/ScalarEvolutionExpressions.h"
++#include "llvm_includes.hpp"
+ 
+ using namespace llvm;
+ namespace gbe {
+@@ -72,7 +42,9 @@ namespace gbe {
+ 
+     virtual bool runOnBasicBlock(BasicBlock &BB) {
+       SE = &getAnalysis<ScalarEvolution>();
+-      #if LLVM_VERSION_MINOR >= 5
++      #if LLVM_VERSION_MINOR >= 7
++        TD = &BB.getModule()->getDataLayout();
++      #elif LLVM_VERSION_MINOR >= 5
+         DataLayoutPass *DLP = getAnalysisIfAvailable<DataLayoutPass>();
+         TD = DLP ? &DLP->getDataLayout() : nullptr;
+       #else
+diff --git a/backend/src/llvm/llvm_passes.cpp b/backend/src/llvm/llvm_passes.cpp
+index 223f61b..d5d965b 100644
+--- a/backend/src/llvm/llvm_passes.cpp
++++ b/backend/src/llvm/llvm_passes.cpp
+@@ -30,75 +30,7 @@
+  * Segovia) the right to use another license for it (MIT here)
+  */
+ 
+-#include "llvm/Config/llvm-config.h"
+-#if LLVM_VERSION_MINOR <= 2
+-#include "llvm/CallingConv.h"
+-#include "llvm/Constants.h"
+-#include "llvm/DerivedTypes.h"
+-#include "llvm/Module.h"
+-#include "llvm/Instructions.h"
+-#else
+-#include "llvm/IR/CallingConv.h"
+-#include "llvm/IR/Constants.h"
+-#include "llvm/IR/DerivedTypes.h"
+-#include "llvm/IR/Module.h"
+-#include "llvm/IR/Instructions.h"
+-#endif  /* LLVM_VERSION_MINOR <= 2 */
+-#include "llvm/Pass.h"
+-#include "llvm/PassManager.h"
+-#if LLVM_VERSION_MINOR <= 2
+-#include "llvm/Intrinsics.h"
+-#include "llvm/IntrinsicInst.h"
+-#include "llvm/InlineAsm.h"
+-#else
+-#include "llvm/IR/Intrinsics.h"
+-#include "llvm/IR/IntrinsicInst.h"
+-#include "llvm/IR/InlineAsm.h"
+-#endif  /* LLVM_VERSION_MINOR <= 2 */
+-#include "llvm/ADT/StringExtras.h"
+-#include "llvm/ADT/SmallString.h"
+-#include "llvm/ADT/STLExtras.h"
+-#include "llvm/Analysis/ConstantsScanner.h"
+-#include "llvm/Analysis/LoopInfo.h"
+-#include "llvm/Analysis/ValueTracking.h"
+-#include "llvm/CodeGen/Passes.h"
+-#include "llvm/CodeGen/IntrinsicLowering.h"
+-
+-#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >=5
+-#include "llvm/IR/Mangler.h"
+-#else
+-#include "llvm/Target/Mangler.h"
+-#endif
+-
+-#include "llvm/Transforms/Scalar.h"
+-#include "llvm/MC/MCAsmInfo.h"
+-#include "llvm/MC/MCContext.h"
+-#include "llvm/MC/MCInstrInfo.h"
+-#include "llvm/MC/MCObjectFileInfo.h"
+-#include "llvm/MC/MCRegisterInfo.h"
+-#include "llvm/MC/MCSubtargetInfo.h"
+-#include "llvm/MC/MCSymbol.h"
+-#if !defined(LLVM_VERSION_MAJOR) || (LLVM_VERSION_MINOR == 1)
+-#include "llvm/Target/TargetData.h"
+-#elif LLVM_VERSION_MINOR == 2
+-#include "llvm/DataLayout.h"
+-#else
+-#include "llvm/IR/DataLayout.h"
+-#endif
+-#include "llvm/Support/ErrorHandling.h"
+-#include "llvm/Support/FormattedStream.h"
+-#if (LLVM_VERSION_MAJOR == 3) && (LLVM_VERSION_MINOR <= 2)
+-#include "llvm/Support/InstVisitor.h"
+-#elif LLVM_VERSION_MINOR >= 5
+-#include "llvm/IR/InstVisitor.h"
+-#else
+-#include "llvm/InstVisitor.h"
+-#endif
+-#include "llvm/Support/MathExtras.h"
+-#include "llvm/Support/TargetRegistry.h"
+-#include "llvm/Support/Host.h"
+-#include "llvm/Support/ToolOutputFile.h"
+-#include "llvm/Support/SourceMgr.h"
++#include "llvm_includes.hpp"
+ 
+ #include "llvm/llvm_gen_backend.hpp"
+ #include "ir/unit.hpp"
+diff --git a/backend/src/llvm/llvm_printf_parser.cpp b/backend/src/llvm/llvm_printf_parser.cpp
+index 3d84457..1e8427c 100644
+--- a/backend/src/llvm/llvm_printf_parser.cpp
++++ b/backend/src/llvm/llvm_printf_parser.cpp
+@@ -33,39 +33,7 @@
+ #include <stdio.h>
+ #include <stdlib.h>
+ 
+-#include "llvm/Config/llvm-config.h"
+-#if LLVM_VERSION_MINOR <= 2
+-#include "llvm/Function.h"
+-#include "llvm/InstrTypes.h"
+-#include "llvm/Instructions.h"
+-#include "llvm/IntrinsicInst.h"
+-#include "llvm/Module.h"
+-#else
+-#include "llvm/IR/Function.h"
+-#include "llvm/IR/InstrTypes.h"
+-#include "llvm/IR/Instructions.h"
+-#include "llvm/IR/IntrinsicInst.h"
+-#include "llvm/IR/Module.h"
+-#endif  /* LLVM_VERSION_MINOR <= 2 */
+-#include "llvm/Pass.h"
+-#if LLVM_VERSION_MINOR <= 1
+-#include "llvm/Support/IRBuilder.h"
+-#elif LLVM_VERSION_MINOR == 2
+-#include "llvm/IRBuilder.h"
+-#else
+-#include "llvm/IR/IRBuilder.h"
+-#endif /* LLVM_VERSION_MINOR <= 1 */
+-
+-#if LLVM_VERSION_MINOR >= 5
+-#include "llvm/IR/CallSite.h"
+-#include "llvm/IR/CFG.h"
+-#else
+-#include "llvm/Support/CallSite.h"
+-#include "llvm/Support/CFG.h"
+-#endif
+-
+-#include "llvm/Support/raw_ostream.h"
+-#include "llvm/IR/Attributes.h"
++#include "llvm_includes.hpp"
+ 
+ #include "llvm/llvm_gen_backend.hpp"
+ #include "sys/map.hpp"
+diff --git a/backend/src/llvm/llvm_sampler_fix.cpp b/backend/src/llvm/llvm_sampler_fix.cpp
+index 8c76324..01db8fe 100644
+--- a/backend/src/llvm/llvm_sampler_fix.cpp
++++ b/backend/src/llvm/llvm_sampler_fix.cpp
+@@ -20,27 +20,8 @@
+  * make sure to get correct pixel value. But for some other
+  * sampler, we don't need those work around code.
+  */
+-#include "llvm/IR/Instructions.h"
+-#include "llvm/Pass.h"
+-#include "llvm/PassManager.h"
+ 
+-#include "llvm/Config/llvm-config.h"
+-#include "llvm/ADT/DenseMap.h"
+-#include "llvm/ADT/PostOrderIterator.h"
+-#include "llvm/IR/Function.h"
+-#include "llvm/IR/InstrTypes.h"
+-#include "llvm/IR/Instructions.h"
+-#include "llvm/IR/IntrinsicInst.h"
+-#include "llvm/IR/Module.h"
+-#include "llvm/Pass.h"
+-#include "llvm/IR/IRBuilder.h"
+-#if LLVM_VERSION_MINOR >= 5
+-#include "llvm/IR/CFG.h"
+-#else
+-#include "llvm/Support/CFG.h"
+-#endif
+-
+-#include "llvm/Analysis/ConstantsScanner.h"
++#include "llvm_includes.hpp"
+ 
+ #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 bc985c6..7ee5259 100644
+--- a/backend/src/llvm/llvm_scalarize.cpp
++++ b/backend/src/llvm/llvm_scalarize.cpp
+@@ -59,39 +59,7 @@
+ //
+ //===----------------------------------------------------------------------===//
+ 
+-#include "llvm/Config/llvm-config.h"
+-#include "llvm/ADT/DenseMap.h"
+-#include "llvm/ADT/PostOrderIterator.h"
+-#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR <= 2
+-#include "llvm/Function.h"
+-#include "llvm/InstrTypes.h"
+-#include "llvm/Instructions.h"
+-#include "llvm/IntrinsicInst.h"
+-#include "llvm/Module.h"
+-#else
+-#include "llvm/IR/Function.h"
+-#include "llvm/IR/InstrTypes.h"
+-#include "llvm/IR/Instructions.h"
+-#include "llvm/IR/IntrinsicInst.h"
+-#include "llvm/IR/Module.h"
+-#endif  /* LLVM_VERSION_MINOR <= 2 */
+-#include "llvm/Pass.h"
+-#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR <= 1
+-#include "llvm/Support/IRBuilder.h"
+-#elif LLVM_VERSION_MINOR == 2
+-#include "llvm/IRBuilder.h"
+-#else
+-#include "llvm/IR/IRBuilder.h"
+-#endif /* LLVM_VERSION_MINOR <= 1 */
+-
+-#if LLVM_VERSION_MINOR >= 5
+-#include "llvm/IR/CallSite.h"
+-#include "llvm/IR/CFG.h"
+-#else
+-#include "llvm/Support/CallSite.h"
+-#include "llvm/Support/CFG.h"
+-#endif
+-#include "llvm/Support/raw_ostream.h"
++#include "llvm_includes.hpp"
+ 
+ #include "llvm/llvm_gen_backend.hpp"
+ #include "sys/map.hpp"
+@@ -128,7 +96,6 @@ namespace gbe {
+ 
+     Scalarize() : FunctionPass(ID)
+     {
+-      initializeLoopInfoPass(*PassRegistry::getPassRegistry());
+ #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 891f2a1..538d1c5 100644
+--- a/backend/src/llvm/llvm_to_gen.cpp
++++ b/backend/src/llvm/llvm_to_gen.cpp
+@@ -22,40 +22,8 @@
+  * \author Benjamin Segovia <benjamin.segovia@intel.com>
+  */
+ 
+-#include "llvm/Config/llvm-config.h"
+-#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR <= 2
+-#include "llvm/LLVMContext.h"
+-#include "llvm/Module.h"
+-#include "llvm/DataLayout.h"
+-#else
+-#include "llvm/IR/LLVMContext.h"
+-#include "llvm/IR/Module.h"
+-#include "llvm/IR/DataLayout.h"
+-#endif  /* LLVM_VERSION_MINOR <= 2 */
+-#include "llvm/PassManager.h"
+-#include "llvm/Pass.h"
+-#include "llvm/Analysis/Passes.h"
+-#include "llvm/Transforms/IPO.h"
+-#include "llvm/Target/TargetLibraryInfo.h"
+-#include "llvm/ADT/Triple.h"
+-#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR <= 2
+-#include "llvm/Support/IRReader.h"
+-#else
+-#include "llvm/IRReader/IRReader.h"
+-#include "llvm/Support/SourceMgr.h"
+-#endif  /* LLVM_VERSION_MINOR <= 2 */
+-#include "llvm/Support/raw_ostream.h"
+-#include "llvm/Transforms/Scalar.h"
+-
+-#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >=5
+-#include "llvm/IR/IRPrintingPasses.h"
+-#include "llvm/IR/Verifier.h"
+-#else
+-#include "llvm/Analysis/Verifier.h"
+-#include "llvm/Assembly/PrintModulePass.h"
+-#endif
++#include "llvm_includes.hpp"
+ 
+-#include "llvm/Analysis/CFGPrinter.h"
+ #include "llvm/llvm_gen_backend.hpp"
+ #include "llvm/llvm_to_gen.hpp"
+ #include "sys/cvar.hpp"
+@@ -64,8 +32,6 @@
+ #include "ir/function.hpp"
+ #include "ir/structurizer.hpp"
+ 
+-#include <clang/CodeGen/CodeGenAction.h>
+-
+ #include <sys/types.h>
+ #include <sys/stat.h>
+ #include <fcntl.h>
+@@ -78,11 +44,19 @@ 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
++  using namespace llvm::legacy;
++  #define TARGETLIBRARY  TargetLibraryInfoImpl
++#else
++  #define TARGETLIBRARY  TargetLibraryInfo
++#endif
++
++  void runFuntionPass(Module &mod, TARGETLIBRARY *libraryInfo, const DataLayout &DL)
+   {
+     FunctionPassManager FPM(&mod);
+ 
+-#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
+     FPM.add(new DataLayoutPass());
+ #elif LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR == 5
+     FPM.add(new DataLayoutPass(DL));
+@@ -95,7 +69,11 @@ 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());
+@@ -111,18 +89,24 @@ namespace gbe
+     FPM.doFinalization();
+   }
+ 
+-  void runModulePass(Module &mod, TargetLibraryInfo *libraryInfo, const DataLayout &DL, int optLevel, bool strictMath)
++  void runModulePass(Module &mod, TARGETLIBRARY *libraryInfo, const DataLayout &DL, int optLevel, bool strictMath)
+   {
+-    llvm::PassManager MPM;
++    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());
+@@ -202,7 +186,7 @@ namespace gbe
+ 
+ #if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 5
+ #define OUTPUT_BITCODE(STAGE, MOD)  do {         \
+-   llvm::PassManager passes__;                   \
++   PassManager passes__;           \
+    if (OCL_OUTPUT_LLVM_##STAGE) {                \
+      passes__.add(createPrintModulePass(*o));    \
+      passes__.run(MOD);                          \
+@@ -210,7 +194,7 @@ namespace gbe
+  }while(0)
+ #else
+ #define OUTPUT_BITCODE(STAGE, MOD)  do {         \
+-   llvm::PassManager passes__;                   \
++   PassManager passes__;           \
+    if (OCL_OUTPUT_LLVM_##STAGE) {                \
+      passes__.add(createPrintModulePass(&*o));   \
+      passes__.run(MOD);                          \
+@@ -260,16 +244,20 @@ namespace gbe
+     Module &mod = *M.get();
+     DataLayout DL(&mod);
+ 
++#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 7
++    mod.setDataLayout(DL);
++#endif
+     Triple TargetTriple(mod.getTargetTriple());
+-    TargetLibraryInfo *libraryInfo = new TargetLibraryInfo(TargetTriple);
++    TARGETLIBRARY *libraryInfo = new TARGETLIBRARY(TargetTriple);
+     libraryInfo->disableAllFunctions();
+ 
+     OUTPUT_BITCODE(AFTER_LINK, mod);
+ 
+     runFuntionPass(mod, libraryInfo, DL);
+     runModulePass(mod, libraryInfo, DL, optLevel, strictMath);
+-    llvm::PassManager passes;
+-#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 6
++    PassManager passes;
++#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));
+diff --git a/backend/src/llvm/llvm_unroll.cpp b/backend/src/llvm/llvm_unroll.cpp
+index 5d3fad8..6990e39 100644
+--- a/backend/src/llvm/llvm_unroll.cpp
++++ b/backend/src/llvm/llvm_unroll.cpp
+@@ -18,34 +18,9 @@
+ #include "llvm/Config/llvm-config.h"
+ #if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR >= 5
+ #include <set>
+-#if LLVM_VERSION_MINOR <= 2
+-#include "llvm/Function.h"
+-#include "llvm/InstrTypes.h"
+-#include "llvm/Instructions.h"
+-#include "llvm/IntrinsicInst.h"
+-#include "llvm/Module.h"
+-#else
+-#include "llvm/IR/Function.h"
+-#include "llvm/IR/InstrTypes.h"
+-#include "llvm/IR/Instructions.h"
+-#include "llvm/IR/IntrinsicInst.h"
+-#include "llvm/IR/Module.h"
+-#endif  /* LLVM_VERSION_MINOR <= 2 */
+-#include "llvm/Pass.h"
+-#if LLVM_VERSION_MINOR <= 1
+-#include "llvm/Support/IRBuilder.h"
+-#elif LLVM_VERSION_MINOR == 2
+-#include "llvm/IRBuilder.h"
+-#else
+-#include "llvm/IR/IRBuilder.h"
+-#endif /* LLVM_VERSION_MINOR <= 1 */
+-#include "llvm/Support/raw_ostream.h"
+-#include "llvm/PassManager.h"
+-#include "llvm/Transforms/Scalar.h"
+-#include "llvm/Analysis/ScalarEvolution.h"
+-#include "llvm/Analysis/LoopPass.h"
+-#include "llvm/Analysis/TargetTransformInfo.h"
+-#include "llvm/IR/Dominators.h"
++
++#include "llvm_includes.hpp"
++
+ #include "llvm/llvm_gen_backend.hpp"
+ #include "sys/map.hpp"
+ 
+@@ -61,8 +36,13 @@ 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);
+-- 
+cgit v0.10.2
+
diff --git a/lang/beignet/pkg-plist b/lang/beignet/pkg-plist
index 3188fe039312..8ac22f9dcd38 100644
--- a/lang/beignet/pkg-plist
+++ b/lang/beignet/pkg-plist
@@ -13,6 +13,8 @@ lib/beignet/include/ocl_geometric.h
 lib/beignet/include/ocl_image.h
 lib/beignet/include/ocl_integer.h
 lib/beignet/include/ocl_math.h
+lib/beignet/include/ocl_memcpy.h
+lib/beignet/include/ocl_memset.h
 lib/beignet/include/ocl_misc.h
 lib/beignet/include/ocl_printf.h
 lib/beignet/include/ocl_relational.h