|
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
|
|
|
4
|
|
|
5 --- |
|
|
|
6 define amdgpu_kernel void @check_vcc() #0 {
|
|
|
7 ret void
|
|
|
8 }
|
|
|
9
|
|
|
10 attributes #0 = { "frame-pointer"="all" }
|
|
|
11 ...
|
|
|
12 ---
|
|
|
13 name: check_vcc
|
|
|
14 tracksRegLiveness: true
|
|
|
15 liveins:
|
|
|
16 - { reg: '$sgpr4_sgpr5' }
|
|
|
17 - { reg: '$sgpr6_sgpr7' }
|
|
|
18 - { reg: '$sgpr8' }
|
|
|
19 frameInfo:
|
|
|
20 maxAlignment: 4
|
|
|
21 stack:
|
|
|
22 - { id: 0, type: spill-slot, size: 8, alignment: 4 }
|
|
|
23 machineFunctionInfo:
|
|
|
24 isEntryFunction: true
|
|
|
25 waveLimiter: true
|
|
|
26 scratchRSrcReg: '$sgpr96_sgpr97_sgpr98_sgpr99'
|
|
|
27 stackPtrOffsetReg: '$sgpr32'
|
|
|
28 frameOffsetReg: '$sgpr33'
|
|
|
29 argumentInfo:
|
|
|
30 privateSegmentBuffer: { reg: '$sgpr0_sgpr1_sgpr2_sgpr3' }
|
|
|
31 dispatchPtr: { reg: '$sgpr4_sgpr5' }
|
|
|
32 kernargSegmentPtr: { reg: '$sgpr6_sgpr7' }
|
|
|
33 workGroupIDX: { reg: '$sgpr8' }
|
|
|
34 privateSegmentWaveByteOffset: { reg: '$sgpr9' }
|
|
|
35 body: |
|
|
|
36 bb.0:
|
|
|
37 liveins: $sgpr8, $sgpr4_sgpr5, $sgpr6_sgpr7
|
|
|
38
|
|
|
39 ; GFX9-LABEL: name: check_vcc
|
|
|
40 ; GFX9: liveins: $sgpr8, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr9
|
|
|
41 ; GFX9: $sgpr33 = S_MOV_B32 0
|
|
|
42 ; GFX9: $sgpr12 = S_MOV_B32 &SCRATCH_RSRC_DWORD0, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15
|
|
|
43 ; GFX9: $sgpr13 = S_MOV_B32 &SCRATCH_RSRC_DWORD1, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15
|
|
|
44 ; GFX9: $sgpr14 = S_MOV_B32 4294967295, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15
|
|
|
45 ; GFX9: $sgpr15 = S_MOV_B32 14680064, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15
|
|
|
46 ; GFX9: $sgpr12 = S_ADD_U32 $sgpr12, $sgpr9, implicit-def $scc, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15
|
|
|
47 ; GFX9: $sgpr13 = S_ADDC_U32 $sgpr13, 0, implicit-def $scc, implicit $scc, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15
|
|
|
48 ; GFX9: $vcc = IMPLICIT_DEF
|
|
|
49 ; GFX9: $sgpr0_sgpr1 = S_MOV_B64 $exec
|
|
|
50 ; GFX9: $exec = S_MOV_B64 3, implicit-def $vgpr0
|
|
223
|
51 ; GFX9: 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)
|
|
221
|
52 ; GFX9: $vgpr0 = V_WRITELANE_B32 $vcc_lo, 0, undef $vgpr0, implicit $vcc
|
|
|
53 ; GFX9: $vgpr0 = V_WRITELANE_B32 $vcc_hi, 1, $vgpr0, implicit $vcc
|
|
223
|
54 ; GFX9: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr12_sgpr13_sgpr14_sgpr15, $sgpr33, 4, 0, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
|
|
|
55 ; GFX9: $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)
|
|
221
|
56 ; GFX9: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
|
|
|
57 ; GFX9: $vcc = IMPLICIT_DEF
|
|
|
58 ; GFX9: $sgpr0_sgpr1 = S_MOV_B64 $exec
|
|
|
59 ; GFX9: $exec = S_MOV_B64 3, implicit-def $vgpr0
|
|
223
|
60 ; GFX9: 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)
|
|
221
|
61 ; GFX9: $vgpr0 = V_WRITELANE_B32 $vcc_lo, 0, undef $vgpr0, implicit $vcc
|
|
|
62 ; GFX9: $vgpr0 = V_WRITELANE_B32 $vcc_hi, 1, $vgpr0, implicit killed $vcc
|
|
223
|
63 ; GFX9: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr12_sgpr13_sgpr14_sgpr15, $sgpr33, 4, 0, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
|
|
|
64 ; GFX9: $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)
|
|
221
|
65 ; GFX9: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
|
|
|
66 ; GFX9: $sgpr0_sgpr1 = S_MOV_B64 $exec
|
|
|
67 ; GFX9: $exec = S_MOV_B64 3, implicit-def $vgpr0
|
|
223
|
68 ; GFX9: 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)
|
|
|
69 ; GFX9: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr12_sgpr13_sgpr14_sgpr15, $sgpr33, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
|
|
221
|
70 ; GFX9: $vcc_lo = V_READLANE_B32 $vgpr0, 0, implicit-def $vcc
|
|
|
71 ; GFX9: $vcc_hi = V_READLANE_B32 killed $vgpr0, 1
|
|
223
|
72 ; GFX9: $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)
|
|
221
|
73 ; GFX9: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
|
|
|
74 ; GFX10-LABEL: name: check_vcc
|
|
|
75 ; GFX10: liveins: $sgpr8, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr9
|
|
|
76 ; GFX10: $sgpr33 = S_MOV_B32 0
|
|
|
77 ; GFX10: $sgpr96 = S_MOV_B32 &SCRATCH_RSRC_DWORD0, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
|
|
|
78 ; GFX10: $sgpr97 = S_MOV_B32 &SCRATCH_RSRC_DWORD1, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
|
|
|
79 ; GFX10: $sgpr98 = S_MOV_B32 4294967295, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
|
|
|
80 ; GFX10: $sgpr99 = S_MOV_B32 836853760, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
|
|
|
81 ; GFX10: $sgpr96 = S_ADD_U32 $sgpr96, $sgpr9, implicit-def $scc, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
|
|
|
82 ; GFX10: $sgpr97 = S_ADDC_U32 $sgpr97, 0, implicit-def $scc, implicit $scc, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
|
|
|
83 ; GFX10: $vcc = IMPLICIT_DEF
|
|
|
84 ; GFX10: $sgpr0_sgpr1 = S_MOV_B64 $exec
|
|
|
85 ; GFX10: $exec = S_MOV_B64 3, implicit-def $vgpr0
|
|
223
|
86 ; GFX10: 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)
|
|
221
|
87 ; GFX10: $vgpr0 = V_WRITELANE_B32 $vcc_lo, 0, undef $vgpr0, implicit $vcc
|
|
|
88 ; GFX10: $vgpr0 = V_WRITELANE_B32 $vcc_hi, 1, $vgpr0, implicit $vcc
|
|
223
|
89 ; GFX10: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 4, 0, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
|
|
|
90 ; GFX10: $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)
|
|
221
|
91 ; GFX10: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
|
|
|
92 ; GFX10: $vcc = IMPLICIT_DEF
|
|
|
93 ; GFX10: $sgpr0_sgpr1 = S_MOV_B64 $exec
|
|
|
94 ; GFX10: $exec = S_MOV_B64 3, implicit-def $vgpr0
|
|
223
|
95 ; GFX10: 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)
|
|
221
|
96 ; GFX10: $vgpr0 = V_WRITELANE_B32 $vcc_lo, 0, undef $vgpr0, implicit $vcc
|
|
|
97 ; GFX10: $vgpr0 = V_WRITELANE_B32 $vcc_hi, 1, $vgpr0, implicit killed $vcc
|
|
223
|
98 ; GFX10: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 4, 0, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
|
|
|
99 ; GFX10: $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)
|
|
221
|
100 ; GFX10: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
|
|
|
101 ; GFX10: $sgpr0_sgpr1 = S_MOV_B64 $exec
|
|
|
102 ; GFX10: $exec = S_MOV_B64 3, implicit-def $vgpr0
|
|
223
|
103 ; GFX10: 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)
|
|
|
104 ; GFX10: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 4, 0, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
|
|
221
|
105 ; GFX10: $vcc_lo = V_READLANE_B32 $vgpr0, 0, implicit-def $vcc
|
|
|
106 ; GFX10: $vcc_hi = V_READLANE_B32 killed $vgpr0, 1
|
|
223
|
107 ; GFX10: $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)
|
|
221
|
108 ; GFX10: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
|
|
|
109 $vcc = IMPLICIT_DEF
|
|
|
110 SI_SPILL_S64_SAVE $vcc, %stack.0, implicit $exec, implicit $sgpr96_sgpr97_sgpr98_sgpr99, implicit $sgpr32
|
|
|
111
|
|
|
112 $vcc = IMPLICIT_DEF
|
|
|
113 SI_SPILL_S64_SAVE killed $vcc, %stack.0, implicit $exec, implicit $sgpr96_sgpr97_sgpr98_sgpr99, implicit $sgpr32
|
|
|
114
|
|
|
115 $vcc = SI_SPILL_S64_RESTORE %stack.0, implicit $exec, implicit $sgpr96_sgpr97_sgpr98_sgpr99, implicit $sgpr32
|
|
|
116 ...
|
