1; RUN: llc -march=hexagon -disable-hcp < %s | FileCheck %s
2; Make sure we build the constant vector <7, 7, 7, 7> with a vsplatb.
3; CHECK: vsplatb
4@B = common global [400 x i8] zeroinitializer, align 8
5@A = common global [400 x i8] zeroinitializer, align 8
6@C = common global [400 x i8] zeroinitializer, align 8
7
8define void @run(i8 %v) nounwind {
9entry:
10  br label %polly.loop_body
11
12polly.loop_after:                                 ; preds = %polly.loop_body
13  ret void
14
15polly.loop_body:                                  ; preds = %entry, %polly.loop_body
16  %polly.loopiv25 = phi i32 [ 0, %entry ], [ %polly.next_loopiv, %polly.loop_body ]
17  %polly.next_loopiv = add i32 %polly.loopiv25, 4
18  %p_arrayidx1 = getelementptr [400 x i8], [400 x i8]* @A, i32 0, i32 %polly.loopiv25
19  %p_arrayidx = getelementptr [400 x i8], [400 x i8]* @B, i32 0, i32 %polly.loopiv25
20  %vector_ptr = bitcast i8* %p_arrayidx to <4 x i8>*
21  %_p_vec_full = load <4 x i8>, <4 x i8>* %vector_ptr, align 8
22  %vec0 = insertelement <4 x i8> undef, i8 %v, i32 0
23  %vec1 = insertelement <4 x i8> %vec0, i8 %v, i32 1
24  %vec2 = insertelement <4 x i8> %vec1, i8 %v, i32 2
25  %vec3 = insertelement <4 x i8> %vec2, i8 %v, i32 3
26  %mulp_vec = mul <4 x i8> %_p_vec_full, %vec3
27  %vector_ptr14 = bitcast i8* %p_arrayidx1 to <4 x i8>*
28  %_p_vec_full15 = load <4 x i8>, <4 x i8>* %vector_ptr14, align 8
29  %addp_vec = add <4 x i8> %_p_vec_full15, %mulp_vec
30  store <4 x i8> %addp_vec, <4 x i8>* %vector_ptr14, align 8
31  %0 = icmp slt i32 %polly.next_loopiv, 400
32  br i1 %0, label %polly.loop_body, label %polly.loop_after
33}
34