|
221
|
1 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
|
|
|
2 # RUN: llc -mtriple=amdgcn -mcpu=gfx900 -verify-machineinstrs -run-pass=prologepilog %s -o - | FileCheck --check-prefix=GFX9 %s
|
|
|
3 # RUN: llc -mtriple=amdgcn -mcpu=gfx1010 -mattr=-wavefrontsize32,+wavefrontsize64 -verify-machineinstrs -run-pass=prologepilog %s -o - | FileCheck --check-prefix=GFX10 %s
|
|
236
|
4 # RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=-wavefrontsize32,+wavefrontsize64 -verify-machineinstrs -run-pass=prologepilog %s -o - | FileCheck --check-prefix=GFX11 %s
|
|
221
|
5
|
|
|
6 --- |
|
|
|
7 define amdgpu_kernel void @check_vcc() #0 {
|
|
|
8 ret void
|
|
|
9 }
|
|
|
10
|
|
|
11 attributes #0 = { "frame-pointer"="all" }
|
|
|
12 ...
|
|
|
13 ---
|
|
|
14 name: check_vcc
|
|
|
15 tracksRegLiveness: true
|
|
|
16 liveins:
|
|
|
17 - { reg: '$sgpr4_sgpr5' }
|
|
|
18 - { reg: '$sgpr6_sgpr7' }
|
|
|
19 - { reg: '$sgpr8' }
|
|
|
20 frameInfo:
|
|
|
21 maxAlignment: 4
|
|
|
22 stack:
|
|
|
23 - { id: 0, type: spill-slot, size: 8, alignment: 4 }
|
|
|
24 machineFunctionInfo:
|
|
|
25 isEntryFunction: true
|
|
|
26 waveLimiter: true
|
|
|
27 scratchRSrcReg: '$sgpr96_sgpr97_sgpr98_sgpr99'
|
|
|
28 stackPtrOffsetReg: '$sgpr32'
|
|
|
29 frameOffsetReg: '$sgpr33'
|
|
|
30 argumentInfo:
|
|
|
31 privateSegmentBuffer: { reg: '$sgpr0_sgpr1_sgpr2_sgpr3' }
|
|
|
32 dispatchPtr: { reg: '$sgpr4_sgpr5' }
|
|
|
33 kernargSegmentPtr: { reg: '$sgpr6_sgpr7' }
|
|
|
34 workGroupIDX: { reg: '$sgpr8' }
|
|
|
35 privateSegmentWaveByteOffset: { reg: '$sgpr9' }
|
|
|
36 body: |
|
|
|
37 bb.0:
|
|
|
38 liveins: $sgpr8, $sgpr4_sgpr5, $sgpr6_sgpr7
|
|
|
39
|
|
|
40 ; GFX9-LABEL: name: check_vcc
|
|
|
41 ; GFX9: liveins: $sgpr8, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr9
|
|
236
|
42 ; GFX9-NEXT: {{ $}}
|
|
|
43 ; GFX9-NEXT: $sgpr33 = S_MOV_B32 0
|
|
|
44 ; GFX9-NEXT: $sgpr12 = S_MOV_B32 &SCRATCH_RSRC_DWORD0, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15
|
|
|
45 ; GFX9-NEXT: $sgpr13 = S_MOV_B32 &SCRATCH_RSRC_DWORD1, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15
|
|
|
46 ; GFX9-NEXT: $sgpr14 = S_MOV_B32 4294967295, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15
|
|
|
47 ; GFX9-NEXT: $sgpr15 = S_MOV_B32 14680064, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15
|
|
|
48 ; GFX9-NEXT: $sgpr12 = S_ADD_U32 $sgpr12, $sgpr9, implicit-def $scc, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15
|
|
|
49 ; GFX9-NEXT: $sgpr13 = S_ADDC_U32 $sgpr13, 0, implicit-def dead $scc, implicit $scc, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15
|
|
|
50 ; GFX9-NEXT: $vcc = IMPLICIT_DEF
|
|
|
51 ; GFX9-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
|
|
|
52 ; GFX9-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr0
|
|
|
53 ; GFX9-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr12_sgpr13_sgpr14_sgpr15, $sgpr33, 0, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
|
|
|
54 ; GFX9-NEXT: $vgpr0 = V_WRITELANE_B32 $vcc_lo, 0, undef $vgpr0, implicit $vcc
|
|
|
55 ; GFX9-NEXT: $vgpr0 = V_WRITELANE_B32 $vcc_hi, 1, $vgpr0, implicit $vcc
|
|
|
56 ; GFX9-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr12_sgpr13_sgpr14_sgpr15, $sgpr33, 4, 0, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
|
|
|
57 ; GFX9-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr12_sgpr13_sgpr14_sgpr15, $sgpr33, 0, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
|
|
|
58 ; GFX9-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
|
|
|
59 ; GFX9-NEXT: $vcc = IMPLICIT_DEF
|
|
|
60 ; GFX9-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
|
|
|
61 ; GFX9-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr0
|
|
|
62 ; GFX9-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr12_sgpr13_sgpr14_sgpr15, $sgpr33, 0, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
|
|
|
63 ; GFX9-NEXT: $vgpr0 = V_WRITELANE_B32 $vcc_lo, 0, undef $vgpr0, implicit $vcc
|
|
|
64 ; GFX9-NEXT: $vgpr0 = V_WRITELANE_B32 $vcc_hi, 1, $vgpr0, implicit killed $vcc
|
|
|
65 ; GFX9-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr12_sgpr13_sgpr14_sgpr15, $sgpr33, 4, 0, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
|
|
|
66 ; GFX9-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr12_sgpr13_sgpr14_sgpr15, $sgpr33, 0, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
|
|
|
67 ; GFX9-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
|
|
|
68 ; GFX9-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
|
|
|
69 ; GFX9-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr0
|
|
|
70 ; GFX9-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr12_sgpr13_sgpr14_sgpr15, $sgpr33, 0, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
|
|
|
71 ; GFX9-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr12_sgpr13_sgpr14_sgpr15, $sgpr33, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
|
|
|
72 ; GFX9-NEXT: $vcc_lo = V_READLANE_B32 $vgpr0, 0, implicit-def $vcc
|
|
|
73 ; GFX9-NEXT: $vcc_hi = V_READLANE_B32 killed $vgpr0, 1
|
|
|
74 ; GFX9-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr12_sgpr13_sgpr14_sgpr15, $sgpr33, 0, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
|
|
|
75 ; GFX9-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
|
|
221
|
76 ; GFX10-LABEL: name: check_vcc
|
|
|
77 ; GFX10: liveins: $sgpr8, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr9
|
|
236
|
78 ; GFX10-NEXT: {{ $}}
|
|
|
79 ; GFX10-NEXT: $sgpr33 = S_MOV_B32 0
|
|
|
80 ; GFX10-NEXT: $sgpr96 = S_MOV_B32 &SCRATCH_RSRC_DWORD0, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
|
|
|
81 ; GFX10-NEXT: $sgpr97 = S_MOV_B32 &SCRATCH_RSRC_DWORD1, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
|
|
|
82 ; GFX10-NEXT: $sgpr98 = S_MOV_B32 4294967295, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
|
|
|
83 ; GFX10-NEXT: $sgpr99 = S_MOV_B32 836853760, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
|
|
|
84 ; GFX10-NEXT: $sgpr96 = S_ADD_U32 $sgpr96, $sgpr9, implicit-def $scc, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
|
|
|
85 ; GFX10-NEXT: $sgpr97 = S_ADDC_U32 $sgpr97, 0, implicit-def dead $scc, implicit $scc, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
|
|
|
86 ; GFX10-NEXT: $vcc = IMPLICIT_DEF
|
|
|
87 ; GFX10-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
|
|
|
88 ; GFX10-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr0
|
|
|
89 ; GFX10-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
|
|
|
90 ; GFX10-NEXT: $vgpr0 = V_WRITELANE_B32 $vcc_lo, 0, undef $vgpr0, implicit $vcc
|
|
|
91 ; GFX10-NEXT: $vgpr0 = V_WRITELANE_B32 $vcc_hi, 1, $vgpr0, implicit $vcc
|
|
|
92 ; GFX10-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 4, 0, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
|
|
|
93 ; GFX10-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
|
|
|
94 ; GFX10-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
|
|
|
95 ; GFX10-NEXT: $vcc = IMPLICIT_DEF
|
|
|
96 ; GFX10-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
|
|
|
97 ; GFX10-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr0
|
|
|
98 ; GFX10-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
|
|
|
99 ; GFX10-NEXT: $vgpr0 = V_WRITELANE_B32 $vcc_lo, 0, undef $vgpr0, implicit $vcc
|
|
|
100 ; GFX10-NEXT: $vgpr0 = V_WRITELANE_B32 $vcc_hi, 1, $vgpr0, implicit killed $vcc
|
|
|
101 ; GFX10-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 4, 0, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
|
|
|
102 ; GFX10-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
|
|
|
103 ; GFX10-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
|
|
|
104 ; GFX10-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
|
|
|
105 ; GFX10-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr0
|
|
|
106 ; GFX10-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
|
|
|
107 ; GFX10-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
|
|
|
108 ; GFX10-NEXT: $vcc_lo = V_READLANE_B32 $vgpr0, 0, implicit-def $vcc
|
|
|
109 ; GFX10-NEXT: $vcc_hi = V_READLANE_B32 killed $vgpr0, 1
|
|
|
110 ; GFX10-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
|
|
|
111 ; GFX10-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
|
|
|
112 ; GFX11-LABEL: name: check_vcc
|
|
|
113 ; GFX11: liveins: $sgpr8, $sgpr4_sgpr5, $sgpr6_sgpr7
|
|
|
114 ; GFX11-NEXT: {{ $}}
|
|
|
115 ; GFX11-NEXT: $sgpr33 = S_MOV_B32 0
|
|
|
116 ; GFX11-NEXT: $vcc = IMPLICIT_DEF
|
|
|
117 ; GFX11-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
|
|
|
118 ; GFX11-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr0
|
|
|
119 ; GFX11-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
|
|
|
120 ; GFX11-NEXT: $vgpr0 = V_WRITELANE_B32 $vcc_lo, 0, undef $vgpr0, implicit $vcc
|
|
|
121 ; GFX11-NEXT: $vgpr0 = V_WRITELANE_B32 $vcc_hi, 1, $vgpr0, implicit $vcc
|
|
|
122 ; GFX11-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 4, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.0, addrspace 5)
|
|
|
123 ; GFX11-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
|
|
|
124 ; GFX11-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
|
|
|
125 ; GFX11-NEXT: $vcc = IMPLICIT_DEF
|
|
|
126 ; GFX11-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
|
|
|
127 ; GFX11-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr0
|
|
|
128 ; GFX11-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
|
|
|
129 ; GFX11-NEXT: $vgpr0 = V_WRITELANE_B32 $vcc_lo, 0, undef $vgpr0, implicit $vcc
|
|
|
130 ; GFX11-NEXT: $vgpr0 = V_WRITELANE_B32 $vcc_hi, 1, $vgpr0, implicit killed $vcc
|
|
|
131 ; GFX11-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 4, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.0, addrspace 5)
|
|
|
132 ; GFX11-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
|
|
|
133 ; GFX11-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
|
|
|
134 ; GFX11-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
|
|
|
135 ; GFX11-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr0
|
|
|
136 ; GFX11-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
|
|
|
137 ; GFX11-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
|
|
|
138 ; GFX11-NEXT: $vcc_lo = V_READLANE_B32 $vgpr0, 0, implicit-def $vcc
|
|
|
139 ; GFX11-NEXT: $vcc_hi = V_READLANE_B32 killed $vgpr0, 1
|
|
|
140 ; GFX11-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
|
|
|
141 ; GFX11-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
|
|
221
|
142 $vcc = IMPLICIT_DEF
|
|
|
143 SI_SPILL_S64_SAVE $vcc, %stack.0, implicit $exec, implicit $sgpr96_sgpr97_sgpr98_sgpr99, implicit $sgpr32
|
|
|
144
|
|
|
145 $vcc = IMPLICIT_DEF
|
|
|
146 SI_SPILL_S64_SAVE killed $vcc, %stack.0, implicit $exec, implicit $sgpr96_sgpr97_sgpr98_sgpr99, implicit $sgpr32
|
|
|
147
|
|
|
148 $vcc = SI_SPILL_S64_RESTORE %stack.0, implicit $exec, implicit $sgpr96_sgpr97_sgpr98_sgpr99, implicit $sgpr32
|
|
|
149 ...
|
