1; RUN: llc < %s -mtriple=armv8-linux-gnueabi -verify-machineinstrs \
2; RUN:     -asm-verbose=false | FileCheck %s
3
4%struct.uint32x2x2_t = type { <2 x i32>, <2 x i32> }
5%struct.uint32x2x3_t = type { <2 x i32>, <2 x i32>, <2 x i32> }
6%struct.uint32x2x4_t = type { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> }
7
8declare %struct.uint32x2x2_t @llvm.arm.neon.vld2dup.v2i32.p0i8(i8*, i32)
9declare %struct.uint32x2x3_t @llvm.arm.neon.vld3dup.v2i32.p0i8(i8*, i32)
10declare %struct.uint32x2x4_t @llvm.arm.neon.vld4dup.v2i32.p0i8(i8*, i32)
11
12; CHECK-LABEL: test_vld2_dup_update
13; CHECK: vld2.32 {d16[], d17[]}, {{\[}}[[SRC_R:r[0-9]+]]]
14; CHECK: add {{r[0-9]+|lr}}, [[SRC_R]], #4
15define i8* @test_vld2_dup_update(%struct.uint32x2x2_t* %dest, i8* %src) {
16entry:
17  %tmp = tail call %struct.uint32x2x2_t @llvm.arm.neon.vld2dup.v2i32.p0i8(i8* %src, i32 4)
18  store %struct.uint32x2x2_t %tmp, %struct.uint32x2x2_t* %dest, align 8
19  %updated_src = getelementptr inbounds i8, i8* %src, i32 4
20  ret i8* %updated_src
21}
22
23; CHECK-LABEL: test_vld3_dup_update
24; CHECK: vld3.32 {d16[], d17[], d18[]}, {{\[}}[[SRC_R:r[0-9]+]]]
25; CHECK: add {{r[0-9]+|lr}}, [[SRC_R]], #4
26define i8* @test_vld3_dup_update(%struct.uint32x2x3_t* %dest, i8* %src) {
27entry:
28  %tmp = tail call %struct.uint32x2x3_t @llvm.arm.neon.vld3dup.v2i32.p0i8(i8* %src, i32 4)
29  store %struct.uint32x2x3_t %tmp, %struct.uint32x2x3_t* %dest, align 8
30  %updated_src = getelementptr inbounds i8, i8* %src, i32 4
31  ret i8* %updated_src
32}
33
34; CHECK-LABEL: test_vld4_dup_update
35; CHECK: vld4.32 {d16[], d17[], d18[], d19[]}, {{\[}}[[SRC_R:r[0-9]+]]]
36; CHECK: add {{r[0-9]+|lr}}, [[SRC_R]], #4
37define i8* @test_vld4_dup_update(%struct.uint32x2x4_t* %dest, i8* %src) {
38entry:
39  %tmp = tail call %struct.uint32x2x4_t @llvm.arm.neon.vld4dup.v2i32.p0i8(i8* %src, i32 4)
40  store %struct.uint32x2x4_t %tmp, %struct.uint32x2x4_t* %dest, align 8
41  %updated_src = getelementptr inbounds i8, i8* %src, i32 4
42  ret i8* %updated_src
43}
44