1; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
2; RUN: llc -march=amdgcn -mcpu=fiji -O0 -stop-after=irtranslator -global-isel -verify-machineinstrs -o - %s | FileCheck %s
3
4; TODO: Could potentially insert it here
5define void @arg_align_8(i8 addrspace(1)* align 8 %arg0) {
6  ; CHECK-LABEL: name: arg_align_8
7  ; CHECK: bb.1 (%ir-block.0):
8  ; CHECK-NEXT:   liveins: $vgpr0, $vgpr1
9  ; CHECK-NEXT: {{  $}}
10  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
11  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
12  ; CHECK-NEXT:   [[MV:%[0-9]+]]:_(p1) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
13  ; CHECK-NEXT:   [[C:%[0-9]+]]:_(s8) = G_CONSTANT i8 0
14  ; CHECK-NEXT:   G_STORE [[C]](s8), [[MV]](p1) :: (store (s8) into %ir.arg0, align 8, addrspace 1)
15  ; CHECK-NEXT:   SI_RETURN
16  store i8 0, i8 addrspace(1)* %arg0, align 8
17  ret void
18}
19
20declare i8 addrspace(1)* @returns_ptr()
21declare align 8 i8 addrspace(1)* @returns_ptr_align8()
22
23define void @call_result_align_1() {
24  ; CHECK-LABEL: name: call_result_align_1
25  ; CHECK: bb.1 (%ir-block.0):
26  ; CHECK-NEXT:   liveins: $sgpr12, $sgpr13, $sgpr14, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9, $sgpr10_sgpr11
27  ; CHECK-NEXT: {{  $}}
28  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr31
29  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr14
30  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr13
31  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr12
32  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:sgpr_64 = COPY $sgpr10_sgpr11
33  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:sgpr_64 = COPY $sgpr8_sgpr9
34  ; CHECK-NEXT:   [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr6_sgpr7
35  ; CHECK-NEXT:   [[COPY7:%[0-9]+]]:sgpr_64 = COPY $sgpr4_sgpr5
36  ; CHECK-NEXT:   [[C:%[0-9]+]]:_(s8) = G_CONSTANT i8 0
37  ; CHECK-NEXT:   ADJCALLSTACKUP 0, 0, implicit-def $scc
38  ; CHECK-NEXT:   [[GV:%[0-9]+]]:_(p0) = G_GLOBAL_VALUE @returns_ptr
39  ; CHECK-NEXT:   [[COPY8:%[0-9]+]]:_(p4) = COPY [[COPY7]]
40  ; CHECK-NEXT:   [[COPY9:%[0-9]+]]:_(p4) = COPY [[COPY6]]
41  ; CHECK-NEXT:   [[COPY10:%[0-9]+]]:_(p4) = COPY [[COPY5]]
42  ; CHECK-NEXT:   [[COPY11:%[0-9]+]]:_(s64) = COPY [[COPY4]]
43  ; CHECK-NEXT:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY3]]
44  ; CHECK-NEXT:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY2]]
45  ; CHECK-NEXT:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY1]]
46  ; CHECK-NEXT:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
47  ; CHECK-NEXT:   [[COPY16:%[0-9]+]]:_(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
48  ; CHECK-NEXT:   $sgpr0_sgpr1_sgpr2_sgpr3 = COPY [[COPY16]](<4 x s32>)
49  ; CHECK-NEXT:   $sgpr4_sgpr5 = COPY [[COPY8]](p4)
50  ; CHECK-NEXT:   $sgpr6_sgpr7 = COPY [[COPY9]](p4)
51  ; CHECK-NEXT:   $sgpr8_sgpr9 = COPY [[COPY10]](p4)
52  ; CHECK-NEXT:   $sgpr10_sgpr11 = COPY [[COPY11]](s64)
53  ; CHECK-NEXT:   $sgpr12 = COPY [[COPY12]](s32)
54  ; CHECK-NEXT:   $sgpr13 = COPY [[COPY13]](s32)
55  ; CHECK-NEXT:   $sgpr14 = COPY [[COPY14]](s32)
56  ; CHECK-NEXT:   $vgpr31 = COPY [[COPY15]](s32)
57  ; CHECK-NEXT:   $sgpr30_sgpr31 = G_SI_CALL [[GV]](p0), @returns_ptr, csr_amdgpu, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr4_sgpr5, implicit $sgpr6_sgpr7, implicit $sgpr8_sgpr9, implicit $sgpr10_sgpr11, implicit $sgpr12, implicit $sgpr13, implicit $sgpr14, implicit $vgpr31, implicit-def $vgpr0, implicit-def $vgpr1
58  ; CHECK-NEXT:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr0
59  ; CHECK-NEXT:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr1
60  ; CHECK-NEXT:   [[MV:%[0-9]+]]:_(p1) = G_MERGE_VALUES [[COPY17]](s32), [[COPY18]](s32)
61  ; CHECK-NEXT:   ADJCALLSTACKDOWN 0, 0, implicit-def $scc
62  ; CHECK-NEXT:   G_STORE [[C]](s8), [[MV]](p1) :: (store (s8) into %ir.ptr, addrspace 1)
63  ; CHECK-NEXT:   SI_RETURN
64  %ptr = call align 1 i8 addrspace(1)* @returns_ptr()
65  store i8 0, i8 addrspace(1)* %ptr, align 1
66  ret void
67}
68
69define void @call_result_align_8() {
70  ; CHECK-LABEL: name: call_result_align_8
71  ; CHECK: bb.1 (%ir-block.0):
72  ; CHECK-NEXT:   liveins: $sgpr12, $sgpr13, $sgpr14, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9, $sgpr10_sgpr11
73  ; CHECK-NEXT: {{  $}}
74  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr31
75  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr14
76  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr13
77  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr12
78  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:sgpr_64 = COPY $sgpr10_sgpr11
79  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:sgpr_64 = COPY $sgpr8_sgpr9
80  ; CHECK-NEXT:   [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr6_sgpr7
81  ; CHECK-NEXT:   [[COPY7:%[0-9]+]]:sgpr_64 = COPY $sgpr4_sgpr5
82  ; CHECK-NEXT:   [[C:%[0-9]+]]:_(s8) = G_CONSTANT i8 0
83  ; CHECK-NEXT:   ADJCALLSTACKUP 0, 0, implicit-def $scc
84  ; CHECK-NEXT:   [[GV:%[0-9]+]]:_(p0) = G_GLOBAL_VALUE @returns_ptr
85  ; CHECK-NEXT:   [[COPY8:%[0-9]+]]:_(p4) = COPY [[COPY7]]
86  ; CHECK-NEXT:   [[COPY9:%[0-9]+]]:_(p4) = COPY [[COPY6]]
87  ; CHECK-NEXT:   [[COPY10:%[0-9]+]]:_(p4) = COPY [[COPY5]]
88  ; CHECK-NEXT:   [[COPY11:%[0-9]+]]:_(s64) = COPY [[COPY4]]
89  ; CHECK-NEXT:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY3]]
90  ; CHECK-NEXT:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY2]]
91  ; CHECK-NEXT:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY1]]
92  ; CHECK-NEXT:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
93  ; CHECK-NEXT:   [[COPY16:%[0-9]+]]:_(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
94  ; CHECK-NEXT:   $sgpr0_sgpr1_sgpr2_sgpr3 = COPY [[COPY16]](<4 x s32>)
95  ; CHECK-NEXT:   $sgpr4_sgpr5 = COPY [[COPY8]](p4)
96  ; CHECK-NEXT:   $sgpr6_sgpr7 = COPY [[COPY9]](p4)
97  ; CHECK-NEXT:   $sgpr8_sgpr9 = COPY [[COPY10]](p4)
98  ; CHECK-NEXT:   $sgpr10_sgpr11 = COPY [[COPY11]](s64)
99  ; CHECK-NEXT:   $sgpr12 = COPY [[COPY12]](s32)
100  ; CHECK-NEXT:   $sgpr13 = COPY [[COPY13]](s32)
101  ; CHECK-NEXT:   $sgpr14 = COPY [[COPY14]](s32)
102  ; CHECK-NEXT:   $vgpr31 = COPY [[COPY15]](s32)
103  ; CHECK-NEXT:   $sgpr30_sgpr31 = G_SI_CALL [[GV]](p0), @returns_ptr, csr_amdgpu, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr4_sgpr5, implicit $sgpr6_sgpr7, implicit $sgpr8_sgpr9, implicit $sgpr10_sgpr11, implicit $sgpr12, implicit $sgpr13, implicit $sgpr14, implicit $vgpr31, implicit-def $vgpr0, implicit-def $vgpr1
104  ; CHECK-NEXT:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr0
105  ; CHECK-NEXT:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr1
106  ; CHECK-NEXT:   [[MV:%[0-9]+]]:_(p1) = G_MERGE_VALUES [[COPY17]](s32), [[COPY18]](s32)
107  ; CHECK-NEXT:   ADJCALLSTACKDOWN 0, 0, implicit-def $scc
108  ; CHECK-NEXT:   [[ASSERT_ALIGN:%[0-9]+]]:_(p1) = G_ASSERT_ALIGN [[MV]], 8
109  ; CHECK-NEXT:   G_STORE [[C]](s8), [[ASSERT_ALIGN]](p1) :: (store (s8) into %ir.ptr, align 8, addrspace 1)
110  ; CHECK-NEXT:   SI_RETURN
111  %ptr = call align 8 i8 addrspace(1)* @returns_ptr()
112  store i8 0, i8 addrspace(1)* %ptr, align 8
113  ret void
114}
115
116define void @declaration_result_align_8() {
117  ; CHECK-LABEL: name: declaration_result_align_8
118  ; CHECK: bb.1 (%ir-block.0):
119  ; CHECK-NEXT:   liveins: $sgpr12, $sgpr13, $sgpr14, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9, $sgpr10_sgpr11
120  ; CHECK-NEXT: {{  $}}
121  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr31
122  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr14
123  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr13
124  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr12
125  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:sgpr_64 = COPY $sgpr10_sgpr11
126  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:sgpr_64 = COPY $sgpr8_sgpr9
127  ; CHECK-NEXT:   [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr6_sgpr7
128  ; CHECK-NEXT:   [[COPY7:%[0-9]+]]:sgpr_64 = COPY $sgpr4_sgpr5
129  ; CHECK-NEXT:   [[C:%[0-9]+]]:_(s8) = G_CONSTANT i8 0
130  ; CHECK-NEXT:   ADJCALLSTACKUP 0, 0, implicit-def $scc
131  ; CHECK-NEXT:   [[GV:%[0-9]+]]:_(p0) = G_GLOBAL_VALUE @returns_ptr_align8
132  ; CHECK-NEXT:   [[COPY8:%[0-9]+]]:_(p4) = COPY [[COPY7]]
133  ; CHECK-NEXT:   [[COPY9:%[0-9]+]]:_(p4) = COPY [[COPY6]]
134  ; CHECK-NEXT:   [[COPY10:%[0-9]+]]:_(p4) = COPY [[COPY5]]
135  ; CHECK-NEXT:   [[COPY11:%[0-9]+]]:_(s64) = COPY [[COPY4]]
136  ; CHECK-NEXT:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY3]]
137  ; CHECK-NEXT:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY2]]
138  ; CHECK-NEXT:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY1]]
139  ; CHECK-NEXT:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
140  ; CHECK-NEXT:   [[COPY16:%[0-9]+]]:_(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
141  ; CHECK-NEXT:   $sgpr0_sgpr1_sgpr2_sgpr3 = COPY [[COPY16]](<4 x s32>)
142  ; CHECK-NEXT:   $sgpr4_sgpr5 = COPY [[COPY8]](p4)
143  ; CHECK-NEXT:   $sgpr6_sgpr7 = COPY [[COPY9]](p4)
144  ; CHECK-NEXT:   $sgpr8_sgpr9 = COPY [[COPY10]](p4)
145  ; CHECK-NEXT:   $sgpr10_sgpr11 = COPY [[COPY11]](s64)
146  ; CHECK-NEXT:   $sgpr12 = COPY [[COPY12]](s32)
147  ; CHECK-NEXT:   $sgpr13 = COPY [[COPY13]](s32)
148  ; CHECK-NEXT:   $sgpr14 = COPY [[COPY14]](s32)
149  ; CHECK-NEXT:   $vgpr31 = COPY [[COPY15]](s32)
150  ; CHECK-NEXT:   $sgpr30_sgpr31 = G_SI_CALL [[GV]](p0), @returns_ptr_align8, csr_amdgpu, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr4_sgpr5, implicit $sgpr6_sgpr7, implicit $sgpr8_sgpr9, implicit $sgpr10_sgpr11, implicit $sgpr12, implicit $sgpr13, implicit $sgpr14, implicit $vgpr31, implicit-def $vgpr0, implicit-def $vgpr1
151  ; CHECK-NEXT:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr0
152  ; CHECK-NEXT:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr1
153  ; CHECK-NEXT:   [[MV:%[0-9]+]]:_(p1) = G_MERGE_VALUES [[COPY17]](s32), [[COPY18]](s32)
154  ; CHECK-NEXT:   ADJCALLSTACKDOWN 0, 0, implicit-def $scc
155  ; CHECK-NEXT:   [[ASSERT_ALIGN:%[0-9]+]]:_(p1) = G_ASSERT_ALIGN [[MV]], 8
156  ; CHECK-NEXT:   G_STORE [[C]](s8), [[ASSERT_ALIGN]](p1) :: (store (s8) into %ir.ptr, align 8, addrspace 1)
157  ; CHECK-NEXT:   SI_RETURN
158  %ptr = call i8 addrspace(1)* @returns_ptr_align8()
159  store i8 0, i8 addrspace(1)* %ptr, align 8
160  ret void
161}
162
163define i8 addrspace(1)* @tail_call_assert_align() {
164  ; CHECK-LABEL: name: tail_call_assert_align
165  ; CHECK: bb.1.entry:
166  ; CHECK-NEXT:   liveins: $sgpr12, $sgpr13, $sgpr14, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9, $sgpr10_sgpr11
167  ; CHECK-NEXT: {{  $}}
168  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr31
169  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr14
170  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr13
171  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr12
172  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:sgpr_64 = COPY $sgpr10_sgpr11
173  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:sgpr_64 = COPY $sgpr8_sgpr9
174  ; CHECK-NEXT:   [[COPY6:%[0-9]+]]:sgpr_64 = COPY $sgpr6_sgpr7
175  ; CHECK-NEXT:   [[COPY7:%[0-9]+]]:sgpr_64 = COPY $sgpr4_sgpr5
176  ; CHECK-NEXT:   [[GV:%[0-9]+]]:sreg_64(p0) = G_GLOBAL_VALUE @returns_ptr_align8
177  ; CHECK-NEXT:   [[COPY8:%[0-9]+]]:_(p4) = COPY [[COPY7]]
178  ; CHECK-NEXT:   [[COPY9:%[0-9]+]]:_(p4) = COPY [[COPY6]]
179  ; CHECK-NEXT:   [[COPY10:%[0-9]+]]:_(p4) = COPY [[COPY5]]
180  ; CHECK-NEXT:   [[COPY11:%[0-9]+]]:_(s64) = COPY [[COPY4]]
181  ; CHECK-NEXT:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY3]]
182  ; CHECK-NEXT:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY2]]
183  ; CHECK-NEXT:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY1]]
184  ; CHECK-NEXT:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
185  ; CHECK-NEXT:   [[COPY16:%[0-9]+]]:_(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
186  ; CHECK-NEXT:   $sgpr0_sgpr1_sgpr2_sgpr3 = COPY [[COPY16]](<4 x s32>)
187  ; CHECK-NEXT:   $sgpr4_sgpr5 = COPY [[COPY8]](p4)
188  ; CHECK-NEXT:   $sgpr6_sgpr7 = COPY [[COPY9]](p4)
189  ; CHECK-NEXT:   $sgpr8_sgpr9 = COPY [[COPY10]](p4)
190  ; CHECK-NEXT:   $sgpr10_sgpr11 = COPY [[COPY11]](s64)
191  ; CHECK-NEXT:   $sgpr12 = COPY [[COPY12]](s32)
192  ; CHECK-NEXT:   $sgpr13 = COPY [[COPY13]](s32)
193  ; CHECK-NEXT:   $sgpr14 = COPY [[COPY14]](s32)
194  ; CHECK-NEXT:   $vgpr31 = COPY [[COPY15]](s32)
195  ; CHECK-NEXT:   SI_TCRETURN [[GV]](p0), @returns_ptr_align8, 0, csr_amdgpu, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr4_sgpr5, implicit $sgpr6_sgpr7, implicit $sgpr8_sgpr9, implicit $sgpr10_sgpr11, implicit $sgpr12, implicit $sgpr13, implicit $sgpr14, implicit $vgpr31
196entry:
197  %call = tail call i8 addrspace(1)* @returns_ptr_align8()
198  ret i8 addrspace(1)* %call
199}
200