File: si-lower-sgpr-spills-vgpr-lanes-usage.mir

package info (click to toggle)
llvm-toolchain-19 1%3A19.1.7-3
  • links: PTS, VCS
  • area: main
  • in suites: forky, sid, trixie
  • size: 1,998,520 kB
  • sloc: cpp: 6,951,680; ansic: 1,486,157; asm: 913,598; python: 232,024; f90: 80,126; objc: 75,281; lisp: 37,276; pascal: 16,990; sh: 10,009; ml: 5,058; perl: 4,724; awk: 3,523; makefile: 3,167; javascript: 2,504; xml: 892; fortran: 664; cs: 573
file content (205 lines) | stat: -rw-r--r-- 14,514 bytes parent folder | download | duplicates (3)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx908 -verify-machineinstrs -stress-regalloc=3 -start-before=greedy,0 -stop-after=si-lower-sgpr-spills -o - %s | FileCheck -check-prefix=SGPR_SPILLED %s

# INFO : The test starts from the sgpr-regalloc pipeline.

# This file contains 3 test cases to observe the optimal stack slot usage for SGPR spills utilizing Stack Slot Coloring pass.
# @stack-slot-share-equal-sized-spills : In this, the stack slot indices is shared among the spill stack objects of equal size.
# @stack-slot-share-unequal-sized-spills-with-large-spill-first AND
# @stack-slot-share-unequal-sized-spills-with-small-spill-first :
# In the remaining 2 test cases mentioned in just above 2 lines, the stack slot indices is shared among the spill stack objects of
# unequal size, with spill slot having the size of the largest of the stack objects sharing the common stack indices.

---
name:            stack-slot-share-equal-sized-spills
tracksRegLiveness: true
frameInfo:
  adjustsStack:    true
  hasCalls:        true
machineFunctionInfo:
  scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
  frameOffsetReg: $sgpr32
  stackPtrOffsetReg: $sgpr32
body:             |
  bb.0:
    ; SGPR_SPILLED-LABEL: name: stack-slot-share-equal-sized-spills
    ; SGPR_SPILLED: liveins: $sgpr30, $sgpr31, $vgpr62
    ; SGPR_SPILLED-NEXT: {{  $}}
    ; SGPR_SPILLED-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
    ; SGPR_SPILLED-NEXT: $vgpr62 = SI_SPILL_S32_TO_VGPR killed $sgpr30, 0, $vgpr62
    ; SGPR_SPILLED-NEXT: $vgpr62 = SI_SPILL_S32_TO_VGPR killed $sgpr31, 1, $vgpr62
    ; SGPR_SPILLED-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR $sgpr32, 0, [[DEF]]
    ; SGPR_SPILLED-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR $sgpr0, 1, [[DEF]], implicit-def $sgpr0_sgpr1, implicit $sgpr0_sgpr1
    ; SGPR_SPILLED-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR $sgpr1, 2, [[DEF]], implicit $sgpr0_sgpr1
    ; SGPR_SPILLED-NEXT: [[DEF1:%[0-9]+]]:vreg_64 = IMPLICIT_DEF
    ; SGPR_SPILLED-NEXT: [[FLAT_LOAD_DWORD:%[0-9]+]]:vgpr_32 = FLAT_LOAD_DWORD [[DEF1]], 0, 0, implicit $exec, implicit $flat_scr
    ; SGPR_SPILLED-NEXT: renamable $sgpr4_sgpr5 = SI_PC_ADD_REL_OFFSET target-flags(amdgpu-rel32-lo) 4, target-flags(amdgpu-rel32-hi) 4, implicit-def dead $scc
    ; SGPR_SPILLED-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR $sgpr4, 3, [[DEF]], implicit-def $sgpr4_sgpr5, implicit $sgpr4_sgpr5
    ; SGPR_SPILLED-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr5, 4, [[DEF]], implicit killed $sgpr4_sgpr5
    ; SGPR_SPILLED-NEXT: ADJCALLSTACKUP 0, 0, implicit-def $scc, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr32
    ; SGPR_SPILLED-NEXT: $sgpr4 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 3, implicit-def $sgpr4_sgpr5
    ; SGPR_SPILLED-NEXT: $sgpr5 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 4
    ; SGPR_SPILLED-NEXT: dead $sgpr30_sgpr31 = SI_CALL killed renamable $sgpr4_sgpr5, 0, csr_amdgpu, implicit undef $vgpr0
    ; SGPR_SPILLED-NEXT: $sgpr32 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 0
    ; SGPR_SPILLED-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR $sgpr32, 5, [[DEF]]
    ; SGPR_SPILLED-NEXT: $sgpr0 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 1, implicit-def $sgpr0_sgpr1
    ; SGPR_SPILLED-NEXT: $sgpr1 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 2
    ; SGPR_SPILLED-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR $sgpr2, 6, [[DEF]], implicit-def $sgpr2_sgpr3, implicit $sgpr2_sgpr3
    ; SGPR_SPILLED-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR $sgpr3, 7, [[DEF]], implicit $sgpr2_sgpr3
    ; SGPR_SPILLED-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def $scc, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr32
    ; SGPR_SPILLED-NEXT: ADJCALLSTACKUP 0, 0, implicit-def $scc, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr32
    ; SGPR_SPILLED-NEXT: $vgpr0 = COPY [[FLAT_LOAD_DWORD]]
    ; SGPR_SPILLED-NEXT: $sgpr4 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 3, implicit-def $sgpr4_sgpr5
    ; SGPR_SPILLED-NEXT: $sgpr5 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 4
    ; SGPR_SPILLED-NEXT: dead $sgpr30_sgpr31 = SI_CALL killed renamable $sgpr4_sgpr5, 0, csr_amdgpu, implicit killed $vgpr0
    ; SGPR_SPILLED-NEXT: $sgpr32 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 5
    ; SGPR_SPILLED-NEXT: $sgpr2 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 6, implicit-def $sgpr2_sgpr3
    ; SGPR_SPILLED-NEXT: $sgpr3 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 7
    ; SGPR_SPILLED-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def $scc, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr32
    %0:sreg_32_xm0 = COPY $sgpr32
    %5:sreg_64 = COPY $sgpr0_sgpr1
    %1:vreg_64 = IMPLICIT_DEF
    %2:vgpr_32 = FLAT_LOAD_DWORD %1, 0, 0, implicit $exec, implicit $flat_scr
    %3:sreg_64 = SI_PC_ADD_REL_OFFSET target-flags(amdgpu-rel32-lo) 4, target-flags(amdgpu-rel32-hi) 4, implicit-def dead $scc
    ADJCALLSTACKUP 0, 0, implicit-def $scc, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr32
    dead $sgpr30_sgpr31 = SI_CALL %3, 0, csr_amdgpu, implicit undef $vgpr0
    $sgpr32 = COPY %0
    %4:sreg_32_xm0 = COPY $sgpr32
    $sgpr0_sgpr1 = COPY %5
    %6:sreg_64 = COPY $sgpr2_sgpr3
    ADJCALLSTACKDOWN 0, 0, implicit-def $scc, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr32
    ADJCALLSTACKUP 0, 0, implicit-def $scc, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr32
    $vgpr0 = COPY %2
    dead $sgpr30_sgpr31 = SI_CALL %3, 0, csr_amdgpu, implicit killed $vgpr0
    $sgpr32 = COPY %4
    $sgpr2_sgpr3 = COPY %6
    ADJCALLSTACKDOWN 0, 0, implicit-def $scc, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr32
...

---
name:            stack-slot-share-unequal-sized-spills-with-large-spill-first
tracksRegLiveness: true
frameInfo:
  adjustsStack:    true
  hasCalls:        true
machineFunctionInfo:
  scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
  frameOffsetReg: $sgpr32
  stackPtrOffsetReg: $sgpr32
body:             |
  bb.0:
    ; SGPR_SPILLED-LABEL: name: stack-slot-share-unequal-sized-spills-with-large-spill-first
    ; SGPR_SPILLED: liveins: $sgpr30, $sgpr31, $vgpr62
    ; SGPR_SPILLED-NEXT: {{  $}}
    ; SGPR_SPILLED-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
    ; SGPR_SPILLED-NEXT: $vgpr62 = SI_SPILL_S32_TO_VGPR killed $sgpr30, 0, $vgpr62
    ; SGPR_SPILLED-NEXT: $vgpr62 = SI_SPILL_S32_TO_VGPR killed $sgpr31, 1, $vgpr62
    ; SGPR_SPILLED-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR $sgpr32, 0, [[DEF]]
    ; SGPR_SPILLED-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR $sgpr2, 1, [[DEF]], implicit-def $sgpr2_sgpr3, implicit $sgpr2_sgpr3
    ; SGPR_SPILLED-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR $sgpr3, 2, [[DEF]], implicit $sgpr2_sgpr3
    ; SGPR_SPILLED-NEXT: [[DEF1:%[0-9]+]]:vreg_64 = IMPLICIT_DEF
    ; SGPR_SPILLED-NEXT: [[FLAT_LOAD_DWORD:%[0-9]+]]:vgpr_32 = FLAT_LOAD_DWORD [[DEF1]], 0, 0, implicit $exec, implicit $flat_scr
    ; SGPR_SPILLED-NEXT: renamable $sgpr4_sgpr5 = SI_PC_ADD_REL_OFFSET target-flags(amdgpu-rel32-lo) 4, target-flags(amdgpu-rel32-hi) 4, implicit-def dead $scc
    ; SGPR_SPILLED-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR $sgpr4, 3, [[DEF]], implicit-def $sgpr4_sgpr5, implicit $sgpr4_sgpr5
    ; SGPR_SPILLED-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr5, 4, [[DEF]], implicit killed $sgpr4_sgpr5
    ; SGPR_SPILLED-NEXT: ADJCALLSTACKUP 0, 0, implicit-def $scc, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr32
    ; SGPR_SPILLED-NEXT: $sgpr4 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 3, implicit-def $sgpr4_sgpr5
    ; SGPR_SPILLED-NEXT: $sgpr5 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 4
    ; SGPR_SPILLED-NEXT: dead $sgpr30_sgpr31 = SI_CALL killed renamable $sgpr4_sgpr5, 0, csr_amdgpu, implicit undef $vgpr0
    ; SGPR_SPILLED-NEXT: $sgpr32 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 0
    ; SGPR_SPILLED-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR $sgpr32, 5, [[DEF]]
    ; SGPR_SPILLED-NEXT: $sgpr2 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 1, implicit-def $sgpr2_sgpr3
    ; SGPR_SPILLED-NEXT: $sgpr3 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 2
    ; SGPR_SPILLED-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR $sgpr2, 6, [[DEF]]
    ; SGPR_SPILLED-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def $scc, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr32
    ; SGPR_SPILLED-NEXT: ADJCALLSTACKUP 0, 0, implicit-def $scc, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr32
    ; SGPR_SPILLED-NEXT: $vgpr0 = COPY [[FLAT_LOAD_DWORD]]
    ; SGPR_SPILLED-NEXT: $sgpr4 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 3, implicit-def $sgpr4_sgpr5
    ; SGPR_SPILLED-NEXT: $sgpr5 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 4
    ; SGPR_SPILLED-NEXT: dead $sgpr30_sgpr31 = SI_CALL killed renamable $sgpr4_sgpr5, 0, csr_amdgpu, implicit killed $vgpr0
    ; SGPR_SPILLED-NEXT: $sgpr32 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 5
    ; SGPR_SPILLED-NEXT: $sgpr2 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 6
    ; SGPR_SPILLED-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def $scc, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr32
    %0:sreg_32_xm0 = COPY $sgpr32
    %5:sreg_64 = COPY $sgpr2_sgpr3
    %1:vreg_64 = IMPLICIT_DEF
    %2:vgpr_32 = FLAT_LOAD_DWORD %1, 0, 0, implicit $exec, implicit $flat_scr
    %3:sreg_64 = SI_PC_ADD_REL_OFFSET target-flags(amdgpu-rel32-lo) 4, target-flags(amdgpu-rel32-hi) 4, implicit-def dead $scc
    ADJCALLSTACKUP 0, 0, implicit-def $scc, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr32
    dead $sgpr30_sgpr31 = SI_CALL %3, 0, csr_amdgpu, implicit undef $vgpr0
    $sgpr32 = COPY %0
    %4:sreg_32_xm0 = COPY $sgpr32
    $sgpr2_sgpr3 = COPY %5
    %6:sreg_32 = COPY $sgpr2
    ADJCALLSTACKDOWN 0, 0, implicit-def $scc, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr32
    ADJCALLSTACKUP 0, 0, implicit-def $scc, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr32
    $vgpr0 = COPY %2
    dead $sgpr30_sgpr31 = SI_CALL %3, 0, csr_amdgpu, implicit killed $vgpr0
    $sgpr32 = COPY %4
    $sgpr2 = COPY %6
    ADJCALLSTACKDOWN 0, 0, implicit-def $scc, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr32
...

---
name:            stack-slot-share-unequal-sized-spills-with-small-spill-first
tracksRegLiveness: true
frameInfo:
  adjustsStack:    true
  hasCalls:        true
machineFunctionInfo:
  scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
  frameOffsetReg: $sgpr32
  stackPtrOffsetReg: $sgpr32
body:             |
  bb.0:
    ; SGPR_SPILLED-LABEL: name: stack-slot-share-unequal-sized-spills-with-small-spill-first
    ; SGPR_SPILLED: liveins: $sgpr30, $sgpr31, $vgpr62
    ; SGPR_SPILLED-NEXT: {{  $}}
    ; SGPR_SPILLED-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
    ; SGPR_SPILLED-NEXT: $vgpr62 = SI_SPILL_S32_TO_VGPR killed $sgpr30, 0, $vgpr62
    ; SGPR_SPILLED-NEXT: $vgpr62 = SI_SPILL_S32_TO_VGPR killed $sgpr31, 1, $vgpr62
    ; SGPR_SPILLED-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR $sgpr32, 0, [[DEF]]
    ; SGPR_SPILLED-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR $sgpr0, 1, [[DEF]]
    ; SGPR_SPILLED-NEXT: [[DEF1:%[0-9]+]]:vreg_64 = IMPLICIT_DEF
    ; SGPR_SPILLED-NEXT: [[FLAT_LOAD_DWORD:%[0-9]+]]:vgpr_32 = FLAT_LOAD_DWORD [[DEF1]], 0, 0, implicit $exec, implicit $flat_scr
    ; SGPR_SPILLED-NEXT: renamable $sgpr4_sgpr5 = SI_PC_ADD_REL_OFFSET target-flags(amdgpu-rel32-lo) 4, target-flags(amdgpu-rel32-hi) 4, implicit-def dead $scc
    ; SGPR_SPILLED-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR $sgpr4, 2, [[DEF]], implicit-def $sgpr4_sgpr5, implicit $sgpr4_sgpr5
    ; SGPR_SPILLED-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR killed $sgpr5, 3, [[DEF]], implicit killed $sgpr4_sgpr5
    ; SGPR_SPILLED-NEXT: ADJCALLSTACKUP 0, 0, implicit-def $scc, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr32
    ; SGPR_SPILLED-NEXT: $sgpr4 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 2, implicit-def $sgpr4_sgpr5
    ; SGPR_SPILLED-NEXT: $sgpr5 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 3
    ; SGPR_SPILLED-NEXT: dead $sgpr30_sgpr31 = SI_CALL killed renamable $sgpr4_sgpr5, 0, csr_amdgpu, implicit undef $vgpr0
    ; SGPR_SPILLED-NEXT: $sgpr32 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 0
    ; SGPR_SPILLED-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR $sgpr32, 4, [[DEF]]
    ; SGPR_SPILLED-NEXT: $sgpr0 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 1
    ; SGPR_SPILLED-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR $sgpr2, 5, [[DEF]], implicit-def $sgpr2_sgpr3, implicit $sgpr2_sgpr3
    ; SGPR_SPILLED-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = SI_SPILL_S32_TO_VGPR $sgpr3, 6, [[DEF]], implicit $sgpr2_sgpr3
    ; SGPR_SPILLED-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def $scc, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr32
    ; SGPR_SPILLED-NEXT: ADJCALLSTACKUP 0, 0, implicit-def $scc, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr32
    ; SGPR_SPILLED-NEXT: $vgpr0 = COPY [[FLAT_LOAD_DWORD]]
    ; SGPR_SPILLED-NEXT: $sgpr4 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 2, implicit-def $sgpr4_sgpr5
    ; SGPR_SPILLED-NEXT: $sgpr5 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 3
    ; SGPR_SPILLED-NEXT: dead $sgpr30_sgpr31 = SI_CALL killed renamable $sgpr4_sgpr5, 0, csr_amdgpu, implicit killed $vgpr0
    ; SGPR_SPILLED-NEXT: $sgpr32 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 4
    ; SGPR_SPILLED-NEXT: $sgpr2 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 5, implicit-def $sgpr2_sgpr3
    ; SGPR_SPILLED-NEXT: $sgpr3 = SI_RESTORE_S32_FROM_VGPR [[DEF]], 6
    ; SGPR_SPILLED-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def $scc, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr32
    %0:sreg_32_xm0 = COPY $sgpr32
    %5:sreg_32 = COPY $sgpr0
    %1:vreg_64 = IMPLICIT_DEF
    %2:vgpr_32 = FLAT_LOAD_DWORD %1, 0, 0, implicit $exec, implicit $flat_scr
    %3:sreg_64 = SI_PC_ADD_REL_OFFSET target-flags(amdgpu-rel32-lo) 4, target-flags(amdgpu-rel32-hi) 4, implicit-def dead $scc
    ADJCALLSTACKUP 0, 0, implicit-def $scc, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr32
    dead $sgpr30_sgpr31 = SI_CALL %3, 0, csr_amdgpu, implicit undef $vgpr0
    $sgpr32 = COPY %0
    %4:sreg_32_xm0 = COPY $sgpr32
    $sgpr0 = COPY %5
    %6:sreg_64 = COPY $sgpr2_sgpr3
    ADJCALLSTACKDOWN 0, 0, implicit-def $scc, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr32
    ADJCALLSTACKUP 0, 0, implicit-def $scc, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr32
    $vgpr0 = COPY %2
    dead $sgpr30_sgpr31 = SI_CALL %3, 0, csr_amdgpu, implicit killed $vgpr0
    $sgpr32 = COPY %4
    $sgpr2_sgpr3 = COPY %6
    ADJCALLSTACKDOWN 0, 0, implicit-def $scc, implicit-def $sgpr32, implicit $sgpr32, implicit $sgpr32
...