1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=aarch64-unknown-unknown -mcpu=cyclone -mattr=+slow-misaligned-128store | FileCheck %s --check-prefixes=CHECK,SPLITTING
3 ; RUN: llc < %s -mtriple=aarch64 -mattr=-slow-misaligned-128store | FileCheck %s --check-prefixes=CHECK,MISALIGNED
5 @g0 = external dso_local global <3 x float>, align 16
6 @g1 = external dso_local global <3 x float>, align 4
9 ; SPLITTING-LABEL: blam:
10 ; SPLITTING: // %bb.0:
11 ; SPLITTING-NEXT: adrp x8, g1
12 ; SPLITTING-NEXT: add x8, x8, :lo12:g1
13 ; SPLITTING-NEXT: adrp x9, g0
14 ; SPLITTING-NEXT: ldr q0, [x9, :lo12:g0]
15 ; SPLITTING-NEXT: str d0, [x8]
18 ; MISALIGNED-LABEL: blam:
19 ; MISALIGNED: // %bb.0:
20 ; MISALIGNED-NEXT: adrp x8, g0
21 ; MISALIGNED-NEXT: ldr q0, [x8, :lo12:g0]
22 ; MISALIGNED-NEXT: adrp x8, g1
23 ; MISALIGNED-NEXT: add x8, x8, :lo12:g1
24 ; MISALIGNED-NEXT: str d0, [x8]
25 ; MISALIGNED-NEXT: ret
26 %tmp4 = getelementptr inbounds <3 x float>, ptr @g1, i64 0, i64 0
27 %tmp5 = load <3 x float>, ptr @g0, align 16
28 %tmp6 = extractelement <3 x float> %tmp5, i64 0
29 store float %tmp6, ptr %tmp4
30 %tmp7 = getelementptr inbounds float, ptr %tmp4, i64 1
31 %tmp8 = load <3 x float>, ptr @g0, align 16
32 %tmp9 = extractelement <3 x float> %tmp8, i64 1
33 store float %tmp9, ptr %tmp7
38 ; PR21711 - Merge vector stores into wider vector stores.
40 ; On Cyclone, the stores should not get merged into a 16-byte store because
41 ; unaligned 16-byte stores are slow. This test would infinite loop when
42 ; the fastness of unaligned accesses was not specified correctly.
44 define void @merge_vec_extract_stores(<4 x float> %v1, ptr %ptr) {
45 ; SPLITTING-LABEL: merge_vec_extract_stores:
46 ; SPLITTING: // %bb.0:
47 ; SPLITTING-NEXT: ext v1.16b, v0.16b, v0.16b, #8
48 ; SPLITTING-NEXT: stp d0, d1, [x0, #24]
51 ; MISALIGNED-LABEL: merge_vec_extract_stores:
52 ; MISALIGNED: // %bb.0:
53 ; MISALIGNED-NEXT: stur q0, [x0, #24]
54 ; MISALIGNED-NEXT: ret
55 %idx0 = getelementptr inbounds <2 x float>, ptr %ptr, i64 3
56 %idx1 = getelementptr inbounds <2 x float>, ptr %ptr, i64 4
58 %shuffle0 = shufflevector <4 x float> %v1, <4 x float> undef, <2 x i32> <i32 0, i32 1>
59 %shuffle1 = shufflevector <4 x float> %v1, <4 x float> undef, <2 x i32> <i32 2, i32 3>
61 store <2 x float> %shuffle0, ptr %idx0, align 8
62 store <2 x float> %shuffle1, ptr %idx1, align 8
65 ;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line: