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>

8 files changed, 181 insertions, 924 deletions

diff --git a/backend/src/libocl/CMakeLists.txt b/backend/src/libocl/CMakeLists.txt
index 0cd1eefd..0fffd9b1 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 a4af4aa0..7897567f 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 00000000..2672298f
--- /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 00000000..2d444ad9
--- /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 00000000..85f490fe
--- /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 b3fadb27..00000000
--- 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 00000000..b41851aa
--- /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 665eac40..00000000
--- 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
-}