Add memcpy, memset and barrier bitcode files into libocl

Signed-off-by: Junyan He <junyan.he@linux.intel.com>
Reviewed-by: Zhigang Gong <zhigang.gong@intel.com>

3 files changed, 502 insertions, 0 deletions

diff --git a/backend/src/libocl/src/ocl_barrier.ll b/backend/src/libocl/src/ocl_barrier.ll
new file mode 100644
index 00000000..4e55fcb7
--- /dev/null
+++ b/backend/src/libocl/src/ocl_barrier.ll
@@ -0,0 +1,39 @@
+;XXX FIXME as llvm can't use macros, we hardcoded 3, 1, 2
+;here, we may need to use a more grace way to handle this type
+;of values latter.
+;#define CLK_LOCAL_MEM_FENCE  (1 << 0)
+;#define CLK_GLOBAL_MEM_FENCE (1 << 1)
+
+declare i32 @_get_local_mem_fence() nounwind alwaysinline
+declare i32 @_get_global_mem_fence() nounwind alwaysinline
+declare void @__gen_ocl_barrier_local() nounwind alwaysinline noduplicate
+declare void @__gen_ocl_barrier_global() nounwind alwaysinline noduplicate
+declare void @__gen_ocl_barrier_local_and_global() nounwind alwaysinline noduplicate
+
+define void @barrier(i32 %flags) nounwind noduplicate alwaysinline {
+  %1 = icmp eq i32 %flags, 3
+  br i1 %1, label %barrier_local_global, label %barrier_local_check
+
+barrier_local_global:
+  call void @__gen_ocl_barrier_local_and_global()
+  br label %done
+
+barrier_local_check:
+  %2 = icmp eq i32 %flags, 1
+  br i1 %2, label %barrier_local, label %barrier_global_check
+
+barrier_local:
+  call void @__gen_ocl_barrier_local()
+  br label %done
+
+barrier_global_check:
+  %3 = icmp eq i32 %flags, 2
+  br i1 %3, label %barrier_global, label %done
+
+barrier_global:
+  call void @__gen_ocl_barrier_global()
+  br label %done
+
+done:
+  ret void
+}
diff --git a/backend/src/libocl/src/ocl_memcpy.ll b/backend/src/libocl/src/ocl_memcpy.ll
new file mode 100644
index 00000000..476033e0
--- /dev/null
+++ b/backend/src/libocl/src/ocl_memcpy.ll
@@ -0,0 +1,336 @@
+;The memcpy's source code.
+; INLINE_OVERLOADABLE void __gen_memcpy(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(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 ult 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(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 ult 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(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 ult 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(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 ult 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(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 ult 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(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 ult 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(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 ult 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(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 ult 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(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 ult 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
+}
diff --git a/backend/src/libocl/src/ocl_memset.ll b/backend/src/libocl/src/ocl_memset.ll
new file mode 100644
index 00000000..addf9f55
--- /dev/null
+++ b/backend/src/libocl/src/ocl_memset.ll
@@ -0,0 +1,127 @@
+;The memset's source code.
+; INLINE_OVERLOADABLE void __gen_memset(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(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 ult 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(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 ult 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(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 ult 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
+}